Clinical Chemistry

June 1 2007, Volume 53, Issue 6,pp.999- 1180

Editorials

Andrew McCaddon and Peter R. Hudson Methylation and Phosphorylation: A Tangled Relationship? Clin Chem 2007 53: 999-1000. Bob Palais Quantitative Heteroduplex Analysis Clin Chem 2007 53: 1001-1003. Eleftherios P. Diamandis Oncopeptidomics: A Useful Approach for Cancer Diagnosis? Clin Chem 2007 53: 1004-1006. Roger D. Klein The Pain Protective Haplotype: Introducing the Modern Genetic Test Clin Chem 2007 53: 1007-1009. Molecular Diagnostics and Genetics

Jörn Lötsch, Inna Belfer, Anja Kirchhof, Bikash K. Mishra, Mitchell B. Max, Alexandra Doehring, Michael Costigan, Clifford J. Woolf, Gerd Geisslinger, and Irmgard Tegeder Reliable Screening for a Pain-Protective Haplotype in the GTP Cyclohydrolase 1 Gene (GCH1) Through the Use of 3 or Fewer Single Nucleotide Polymorphisms Clin Chem 2007 53: 1010-1015. Published online March 15, 2007; 10.1373/clinchem.2006.082883 Kerstin L. Edlefsen, Jonathan F. Tait, Mark H. Wener, and Michael Astion Utilization and Diagnostic Yield of Neurogenetic Testing at a Tertiary Care Facility Clin Chem 2007 53: 1016-1022. Published online April 19, 2007; 10.1373/clinchem.2006.083360

Sara Bremer, Helge Rootwelt, and Stein Bergan Real-Time PCR Determination of IMPDH1 and IMPDH2 Expression in Blood Cells Clin Chem 2007 53: 1023-1029. Published online April 26, 2007; 10.1373/clinchem.2006.081968 Elizabeth Herness Peters, Sandra Rojas-Caro, Mitchell G. Brigell, Robert J. Zahorchak, Shelley Ann des Etages, Patricia L. Ruppel, Charles R. Knight, Bradley Austermiller, Myrna C. Graham, Steve Wowk, Sean Banks, Lakshmi V. Madabusi, Patrick Turk, Donna Wilder, Carole Kempfer, Terry W. Osborn, and James C. Willey Quality-Controlled Measurement Methods for Quantification of Variations in Transcript Abundance in Whole Blood Samples from Healthy Volunteers Clin Chem 2007 53: 1030-1037. Published online April 13, 2007; 10.1373/clinchem.2006.078154 Sung Jae Kim, David J. Dix, Kary E. Thompson, Rachel N. Murrell, Judith E. Schmid, Jane E. Gallagher, and John C. Rockett Effects of Storage, RNA Extraction, Genechip Type, and Donor Sex on Gene Expression Profiling of Human Whole Blood Clin Chem 2007 53: 1038-1045. Published online April 13, 2007; 10.1373/clinchem.2006.078436 Kok Seong Lim, Robert K. Naviaux, and Richard H. Haas Quantitative Mitochondrial DNA Mutation Analysis by Denaturing HPLC Clin Chem 2007 53: 1046-1052. Published online April 19, 2007; 10.1373/clinchem.2006.083303 Hemostasis and Thrombosis

Isabella Russo, Paola Del Mese, Gabriella Doronzo, Alessandro De Salve, Mariantonietta Secchi, Mariella Trovati, and Giovanni Anfossi Platelet Resistance to the Antiaggregatory Cyclic Nucleotides in Central Obesity Involves Reduced Phosphorylation of Vasodilator-Stimulated Phosphoprotein Clin Chem 2007 53: 1053-1060. Published online April 26, 2007; 10.1373/clinchem.2006.076208 Proteomics and Protein Markers

Yuan-Yuan Li, Chi Zhang, Bo-Sheng Li, Li-Fan Zhao, Xiao-bo Li, Wen-Jie Yang, and Shun-Qing Xu Ultrasensitive Densitometry Detection of Cytokines with Nanoparticle- Modified Aptamers Clin Chem 2007 53: 1061-1066. Published online April 19, 2007; 10.1373/clinchem.2006.082271

Mary F. Lopez, Alvydas Mikulskis, Scott Kuzdzal, Eva Golenko, Emanuel F. Petricoin, III, Lance A. Liotta, Wayne F. Patton, Gordon R. Whiteley, Kevin Rosenblatt, Prem Gurnani, Animesh Nandi, Samuel Neill, Stuart Cullen, Martin O’Gorman, David Sarracino, Christopher Lynch, Andrew Johnson, William Mckenzie, and David Fishman A Novel, High-Throughput Workflow for Discovery and Identification of Serum Carrier Protein-Bound Peptide Biomarker Candidates in Ovarian Cancer Samples Clin Chem 2007 53: 1067-1074. Published online April 26, 2007; 10.1373/clinchem.2006.080721 Winfried März, Beate Tiran, Ursula Seelhorst, Britta Wellnitz, Johann Bauersachs, Bernhard R. Winkelmann, and Bernhard O. Boehm N-Terminal Pro-B-Type Natriuretic Peptide Predicts Total and Cardiovascular Mortality in Individuals with or without Stable Coronary Artery Disease: The Ludwigshafen Risk and Cardiovascular Health Study Clin Chem 2007 53: 1075-1083. Published online April 19, 2007; 10.1373/clinchem.2006.075929 Cancer Diagnostics

Maureen Cronin, Chithra Sangli, Mei-Lan Liu, Mylan Pho, Debjani Dutta, Anhthu Nguyen, Jennie Jeong, Jenny Wu, Kim Clark Langone, and Drew Watson Analytical Validation of the Oncotype DX Genomic Diagnostic Test for Recurrence Prognosis and Therapeutic Response Prediction in Node- Negative, Estrogen Receptor–Positive Breast Cancer Clin Chem 2007 53: 1084-1091. Published online April 26, 2007; 10.1373/clinchem.2006.076497 Lipids, Lipoproteins, and Cardiovascular Risk Factors

Ji-Sun Lim, Duk-Hee Lee, Joo-Yun Park, Soo-Hee Jin, and David R. Jacobs, Jr A Strong Interaction between Serum -Glutamyltransferase and Obesity on the Risk of Prevalent Type 2 Diabetes: Results from the Third National Health and Nutrition Examination Survey Clin Chem 2007 53: 1092-1098. Published online May 3, 2007; 10.1373/clinchem.2006.079814 Hematology

Christine L.H. Snozek, Amy K. Saenger, Philip R. Greipp, Sandra C. Bryant, Robert A. Kyle, S. Vincent Rajkumar, and Jerry A. Katzmann Comparison of Bromcresol Green and Agarose Protein Electrophoresis for Quantitation of Serum Albumin in Multiple Myeloma Clin Chem 2007 53: 1099-1103. Published online April 26, 2007; 10.1373/clinchem.2007.088252 Endocrinology and Metabolism

Isolde Seiden-Long and Reinhold Vieth Evaluation of a 1,25-Dihydroxyvitamin D Enzyme Immunoassay Clin Chem 2007 53: 1104-1108. Published online April 13, 2007; 10.1373/clinchem.2006.077560 Tim Cundy, Anne Horne, Mark Bolland, Greg Gamble, and James Davidson Bone Formation Markers in Adults with Mild Osteogenesis Imperfecta Clin Chem 2007 53: 1109-1114. Published online April 13, 2007; 10.1373/clinchem.2006.083055 Automation and Analytical Techniques

Joris R. Delanghe, Anders Helander, Jos P.M. Wielders, J. Maurits Pekelharing, Heinz J. Roth, François Schellenberg, Catherine Born, Eray Yagmur, Wolfgang Gentzer, and Harald Althaus Development and Multicenter Evaluation of the N Latex CDT Direct Immunonephelometric Assay for Serum Carbohydrate-Deficient Transferrin Clin Chem 2007 53: 1115-1121. Published online April 5, 2007; 10.1373/clinchem.2006.084459 General Clinical Chemistry

John Mahoney and John Ellison Assessing the Quality of Glucose Monitor Studies: A Critical Evaluation of Published Reports Clin Chem 2007 53: 1122-1128. Published online May 3, 2007; 10.1373/clinchem.2006.083493 Rima Obeid, Mariz Kasoha, Jean-Pierre Knapp, Panagiotis Kostopoulos, George Becker, Klaus Fassbender, and Wolfgang Herrmann Folate and Methylation Status in Relation to Phosphorylated Tau Protein(181P) and ß-Amyloid(1–42) in Cerebrospinal Fluid Clin Chem 2007 53: 1129-1136. Published online March 23, 2007; 10.1373/clinchem.2006.085241 Clinical Immunology

Jochen Metzger, Philipp von Landenberg, Marcus Kehrel, Alexander Buhl, Karl J. Lackner, and Peter B. Luppa Biosensor Analysis of ß2-Glycoprotein I–Reactive Autoantibodies: Evidence for Isotype-Specific Binding and Differentiation of Pathogenic from Infection-Induced Antibodies Clin Chem 2007 53: 1137-1143. Published online April 13, 2007; 10.1373/clinchem.2006.079632 Nutrition

Dolores Corella, Donna K. Arnett, Michael Y. Tsai, Edmond K. Kabagambe, James M. Peacock, James E. Hixson, Robert J. Straka, Michael Province, Chao-Qiang Lai, Laurence D. Parnell, Ingrid Borecki, and Jose M. Ordovas The –256T>C Polymorphism in the Apolipoprotein A-II Gene Promoter Is Associated with Body Mass Index and Food Intake in the Genetics of Lipid Lowering Drugs and Diet Network Study Clin Chem 2007 53: 1144-1152. Published online April 19, 2007; 10.1373/clinchem.2006.084863 Technical Briefs

Dihua Shangguan, Zehui Charles Cao, Ying Li, and Weihong Tan Aptamers Evolved from Cultured Cancer Cells Reveal Molecular Differences of Cancer Cells in Patient Samples Clin Chem 2007 53: 1153-1155. Published online April 26, 2007; 10.1373/clinchem.2006.083246 Joseph Skulan, Thomas Bullen, Ariel D. Anbar, J. Edward Puzas, Linda Shackelford, Adrian LeBlanc, and Scott M. Smith Natural Calcium Isotopic Composition of Urine as a Marker of Bone Mineral Balance Clin Chem 2007 53: 1155-1158. Published online April 26, 2007; 10.1373/clinchem.2006.080143 Letters to the Editor

Giovanni Targher, Massimo Franchini, Gian Cesare Guidi, Michele Muggeo, and Giuseppe Lippi Alanine Aminotransferase as an Independent Predictor of Incident Nonalcoholic Fatty Liver Disease Clin Chem 2007 53: 1159. Yoosoo Chang, Seungho Ryu, Eunju Sung, and Yumi Jang Reply to Alanine Aminotransferase as an Independent Predictor of Incident Nonalcoholic Fatty Liver Disease Clin Chem 2007 53: 1159-1161. Mads V. Hollegaard, Karina M. Sørensen, Høgni K. Petersen, Maria B. Arnardottir, Bent Nørgaard-Pedersen, Poul Thorsen, and David M. Hougaard Whole Genome Amplification and Genetic Analysis after Extraction of Proteins from Dried Blood Spots Clin Chem 2007 53: 1161-1162. Francisco J. Morón, Nicolás Mendoza, Francisco Quereda, Francisco Vázquez, Reposo Ramírez-Lorca, Juan Velasco, Jose L. Gallo, Ana Salinas, Txantón Martínez-Astorquiza, Rafael Sánchez-Borrego, Maria E. Sáez, and Agustín Ruiz Pyrosequencing Technology for Automated Detection of the BMP15 A180T Variant in Spanish Postmenopausal Women Clin Chem 2007 53: 1162-1164. David Friedecky, Jana Tomkova, and Tomas Adam Determination of ITPase Activity by Capillary Electrophoresis Clin Chem 2007 53: 1164-1165. Vivienne M. Homer and Peter M. George Novel Mutation (c.G1124A) in Exon 9 of the APOB Gene Causes Aberrant Splicing and Familial Hypobetalipoproteinemia Clin Chem 2007 53: 1165-1167. Kiyonori Miura, Kentaro Yamasaki, Shoko Miura, Koh-ichiro Yoshiura, Takako Shimada, Daisuke Nakayama, Norio Niikawa, and Hideaki Masuzaki Circulating Cell-Free Placental mRNA in the Maternal Plasma as a Predictive Marker for Twin-Twin Transfusion Syndrome Clin Chem 2007 53: 1167-1168. Simone G. van Breda, Janneke G. Hogervorst, Leo J. Schouten, Ad M. Knaapen, Joost H. van Delft, R. Alexandra Goldbohm, Frederik J. van Schooten, and Piet A. van den Brandt Toenails: An Easily Accessible and Long-Term Stable Source of DNA for Genetic Analyses in Large-Scale Epidemiological Studies Clin Chem 2007 53: 1168-1170. Book, Software, and Web Site Reviews

Sarah D. de Ferranti Review of The Metabolic Syndrome. Christopher D. Byrne and Sarah H. Wild, eds. Chichester, West Sussex, England: John Wiley & Sons Ltd., 2005, 432 pp., $110.00, hardcover. ISBN 0-470-02511-5. Clin Chem 2007 53: 1171. Sverre Sandberg Clinical Diagnostic Technology-The Total Testing Process, Volume 3: The Postanalytical Phase. Kory M. Ward-Cook, Craig A. Lehmann, Larry E. Schoeff, and Robert H. Williams, eds. Washington, DC: AACC Press, 2006, 216 pp., $65 ($52.00 AACC members), softcover. ISBN 978-1-59425-055-2. Clin Chem 2007 53: 1171-1172. Johannes Mair Natriuretic Peptides: The Hormones of the Heart. Aldo Clerico and Michele Emdin, eds. Trento, Italy: Springer-Verlag Italia, 2006, 184 pp., $149.00, hardcover. ISBN 978-88-470-0497-9. Clin Chem 2007 53: 1172-1173. Geza Bodor Quick Guide to Coagulation Testing. Marisa B. Marques and George A. Fritsma, Washington, DC: AACC Press, 2006, 62 pp., $15 ($12 AACC members), softcover. ISBN 1-59425-049-9. Clin Chem 2007 53: 1173-1174. Corrections

Correction Clin Chem 2007 53: 1170. The Clinical Chemist

David E. Bruns The Clinical Chemist Clin Chem 2007 53: 1175-1180.

Editorial

Methylation and Phosphorylation: A Tangled Relationship?

Many neurodegenerative diseases are associated with mocysteine is derived from dietary methionine, and in- distinctive brain lesions. One particular group of hetero- creased concentrations are signs of disrupted cellular geneous dementias and movement disorders is character- methylation. Homocysteine can be remethylated to me- ized by intracellular accumulations of abnormal filaments thionine by the vitamin B12–dependent enzyme methio- called neurofibrillary tangles (NFT), formed by the micro- nine synthase, with methyl-folate providing the methyl tubule-associated protein tau. These neurodegenerative group. The activated methionine derivative S-adenosyl- tauopathies include Pick disease and Alzheimer disease methionine (SAM) is essential for many cellular methyl- (AD), but the term also encompasses a range of other ation reactions, and is converted to S-adenosylhomocys- clinical conditions that share a common end-point: neu- teine (SAH) in the process. SAH hydrolase converts SAH rodegeneration with pathological tau accumulation (1). to homocysteine in a reversible reaction, the kinetics of Tau occurs predominantly in neuronal axons, where it which favor condensation of homocysteine and adeno- binds to microtubules and regulates their length and sine. Clearance of homocysteine by methionine synthase “treadmilling” dynamics. Tight regulation of microtubule therefore maintains a favorable SAM:SAH ratio, an index activity is critical to cell viability, and fine regulation of of cellular methylation potential (10). tau is likely to be equally important (2). In this issue of Clinical Chemistry, Obeid et al. (11) Tau activity is modulated by phosphorylation, and the provide in vivo evidence for an association between ability of tau to bind to and stabilize microtubules corre- disrupted methylation and P-tau accumulation. These lates inversely with its degree of phosphorylation. This authors report a strong negative correlation, in the cere- relationship has led to the suggestion of a role for tau in brospinal fluid, of P-tau concentrations and the SAM:SAH the adaptive response of neurons to stress (3). Tau phos- ratio among 182 patients with various neurological disor- phorylation may represent a physiological and reversible ders, including AD. Cerebrospinal fluid folate concentra- process integral to the stress response system. For exam- tions also correlated inversely with P-tau. Aging was ple, in animal models, tau phosphorylation occurs in associated with higher concentrations of homocysteine response to ether anesthesia, cold-water stress, and star- and SAH in cerebrospinal fluid, and with lower concen- vation (4). trations of folate and a lower SAM:SAH ratio. Neverthe- Tau is highly phosphorylated in several neurodegen- less, the striking association between SAH and higher erative diseases associated with NFT formation. Disor- concentrations of P-tau was observed across 3 separate dered phosphorylation disrupts the normal colocalization age groups (Ͻ41, 41–60, and Ͼ60 years). Given the of tau with microtubules, leading to further phosphory- elegant and plausible hypothesis that PP2A hypomethylation at fibrillogenic sites and/or cleavage by caspases. lation can lead to tau hyperphosphorylation, it is tempt- This process increases the probability of tau-tau interac- ing to view these findings as providing strong evidence tions leading to the formation of paired helical filaments, for a direct effect of impaired methylation on P-tau and their subsequent aggregation into NFTs (5). accumulation, but is there an alternative explanation? What is the origin of tau hyperphosphorylation associ- AD is a typical tauopathy, in which NFT formation and ated with the tauopathies? Protein phosphorylation is amyloid beta peptide deposition are both key pathologi- governed by the competing effects of kinases and phos- cal features. Chronic inflammation, however, is now a phatases. Among several potential kinases, attention has recognized additional component of AD, and contributes focused on glycogen synthase kinase 3␤ and cyclin- to disease progression (6). Various neuroinflammatory dependent kinase 5, both of which are associated with mediators, including complement activators and inhibi- NFTs in the brains of AD patients. The signaling mecha- tors, chemokines, cytokines, radical oxygen species, and nisms responsible for the regulation of these kinases are inflammatory enzymes are generated in the disease by complex, but may be modulated, inter alia, by inflamma- microglia, astrocytes, and neurons. Although amyloid tion (6). Phosphoseryl/phosphothreonyl protein phos- beta peptide probably plays a central role in the neuro- phatase-2A (PP2A), which is found in association with tau degenerative process, inflammatory mediators also stim- and microtubules in the brain, appears to be the most ulate its deposition, thereby establishing a vicious cycle of active enzyme in dephosphorylating abnormal tau to a inflammation (12). normal-like state (7). It is likely that similar inflammatory cascades are a The processes regulating PP2A activity are still emerg- common feature of the neurodegenerative tauopathies. ing, but activation via methylation seems to be an impor- Neuroinflammation-associated redox changes might tant step. Vafai and Stock (8) highlighted the importance therefore independently drive both P-tau generation and of such methylation for PP2A activity, suggesting that de- a methylation disturbance. The peptidyl-prolyl isomerase creased methylation could lead to tau hyperphosphorylation. enzyme Pin1 regulates the function and processing of tau The effect of decreased methylation on tau is of partic- and amyloid precursor protein. In particular, it ensures ular interest in light of accumulating data demonstrating that P-tau is in the correct conformation for dephosphor- a relationship between increased plasma homocysteine ylation. Pin1 activity is downregulated under conditions and neurodegenerative diseases, including AD (9). Ho- of oxidative stress (13). Folate depletion may also occur as

Clinical Chemistry 53, No. 6, 2007 999 1000 McCaddon et al.: Relationship of Methylation and Phosphorylation

a consequence of oxidative stress associated with chronic 3. Feijoo C, Campbell DG, Jakes R, Goedert M, Cuenda A. Evidence that phosphorylation of the microtubule-associated protein Tau by SAPK4/ inflammation. Tetrahydrofolate is very susceptible to ox- p38delta at Thr50 promotes microtubule assembly. J Cell Sci 2005;118: idation; a resulting increase in homocysteine concentra- 397–408. tions may become relevant under such conditions (14). 4. Okawa Y, Ishiguro K, Fujita SC. Stress-induced hyperphosphorylation of tau in the mouse brain. FEBS Lett 2003;535:183–9. Furthermore, methionine synthase activity is also exquisi- 5. Stoothoff WH, Johnson GV. Tau phosphorylation: physiological and patho- tively sensitive to cellular redox status, which affects logical consequences. Biochim Biophys Acta 2005;1739:280–97. 6. Arnaud L, Robakis NK, Figueiredo-Pereira ME. It may take inflammation, oxidation of the cob(I)alamin form of its vitamin B12 phosphorylation and ubiquitination to ‘tangle’ in Alzheimer’s disease. Neu- cofactor (15). rodegener Dis 2006;3:313–9. Chronic neuroinflammation, with its associated oxida- 7. Wang JZ, Grundke-Iqbal I, Iqbal K. Kinases and phosphatases and tau sites tive stress, may therefore be an alternative explanation for involved in Alzheimer neurofibrillary degeneration. Eur J Neurosci 2007;25: both the observed reduction in methylation capacity and 59–68. 8. Vafai SB, Stock JB. Protein phosphatase 2A methylation: a link between also the increase in P-tau; hypomethylation and hyper- elevated plasma homocysteine and Alzheimer’s Disease. FEBS Lett 2002; phosphorylation may not necessarily be directly causally 518:1–4. related. 9. McCaddon A. Homocysteine and cognition: a historical perspective. J Alzheimers Dis 2006;9:361–80. Obeid et al. (11) cautiously speculate that restoration of 10. Zou CG, Banerjee R. Homocysteine and redox signaling. Antioxid Redox methylation might exert a neuroprotective effect by pre- Signal 2005;7:547–59. venting P-tau accumulation. In a folate-depleted and 11. Obeid R, Kasoha M, Knapp J-P, Kostopoulos P, Becker G, Fassbender K, et al. Folate and methylation status in relation to phosphorylated tau oxidatively-stressed mouse model, however, dietary sup- protein(181P) and [beta]-amyloid(1–42) in cerebrospinal fluid. Clin Chem plementation with the methyl-donor SAM alleviated an 2007;1129–36. increase in non–phospho-tau, but failed to attenuate in- 12. Heneka MT, O’Banion MK. Inflammatory processes in Alzheimer’s disease. J Neuroimmunol 2007;184:69–91. creased P-tau (16). This finding lends support to the 13. Balastik M, Lim J, Pastorino L, Lu KP. Pin1 in Alzheimer’s disease: Multiple alternative suggestion that oxidative stress plays an addi- substrates, one regulatory mechanism? Biochim Biophys Acta 2007;1772: tional contributory role. Similarly, B-vitamin supplemen- 422–9. 14. Widner B, Enzinger C, Laich A, Wirleitner B, Fuchs D. Hyperhomocysteine- tation can decrease serum homocysteine without neces- mia, pteridines and oxidative stress. Curr Drug Metab 2002;3:225–32. sarily affecting any underlying inflammation; treatment 15. McCaddon A, Regland B, Hudson P, Davies G. Functional vitamin B(12) with B-vitamins reduces serum homocysteine in AD pa- deficiency and Alzheimer disease. Neurology 2002;58:1395–9. tients, but this reduction is not accompanied by a reduc- 16. Chan A, Shea TB. Dietary and genetically-induced oxidative stress alter tau phosphorylation: influence of folate and apolipoprotein E deficiency. tion in markers of immune activation or inflammation J Alzheimers Dis 2006;9:399–405. (17). 17. Frick B, Gruber B, Schroecksnadel K, Leblhuber F, Fuchs D. Homocysteine The findings of Obeid et al. (11) are certainly thought- but not neopterin declines in demented patients on B vitamins. J Neural Transm 2006;113:1815–9. provoking, and represent an important bridge between clinical observations of hyperhomocysteinemia-related Andrew McCaddon1*† neurological dysfunction and pathological observations Peter R. Hudson2‡ of P-tau accumulation. In addition, given the increasing realization of the role of chronic inflammation in neuro- 1 Cardiff School of Medicine degenerative disease, and of the contribution of redox Wrexham, United Kingdom regulation to homocysteine metabolism (10), these findings perhaps hint that oxidative stress is an important 2 Department of Medicinal Biochemistry underlying common denominator. The challenge now is Maelor Hospital to determine exactly how disrupted methylation and Wrexham, United Kingdom aberrant phosphorylation are temporally related, and whether modification of these processes can make any † Honorary Senior Research Fellow, Cardiff School of meaningful impact in treating these neurodegenerative Medicine, Gardden Road Surgery, Rhosllanerchrugog, disorders. Wrexham, United Kingdom

‡ Principal Biochemist, Department of Medical Biochemistry, Financial disclosures: AM and PH are shareholders in Maelor Hospital, Croesnewydd Road, Wrexham, COBALZ Limited, a private company developing combined United Kingdom vitamin B12 and antioxidant supplements.

References * Address correspondence to this author at: Cardiff School of 1. Williams DR. Tauopathies: classification and clinical update on neurodegen- Medicine, Gardden Road Surgery, Rhosllanerchrugog, Wrexham erative diseases associated with microtubule-associated protein tau. Intern LL14 2EN, United Kingdom. Fax ϩ44-1978-845782; e-mail andrew. Med J 2006;36:652–60. [email protected]. 2. Bunker JM, Wilson L, Jordan MA, Feinstein SC. Modulation of microtubule dynamics by tau in living cells: implications for development and neurode- DOI: 10.1373/clinchem.2007.086579 generation. Mol Biol Cell 2004;15:2720–8. Editorial

Quantitative Heteroduplex Analysis

General DNA analysis includes detection of targets, as gradient capillary electrophoresis (TGCE) (14). Closed- well as identification and quantification of specific tube methods without separation have also been used variants. Genomic DNA exists in the form of homodu- to detect heteroduplexes by changes in the shape of plexes, with all corresponding base pairs being comple- high-resolution melting curves with fluorescently la- mentary, A:T and C:G. We call a double-stranded DNA beled primers (15) or a saturating DNA dye (16). Direct (dsDNA) molecule a heteroduplex when it contains any sequencing has also been used to identify and quantify noncomplementary base pairs. Conformational and heteroduplexes in gene dosage studies (17). Other thermodynamic changes produced by mismatched studies (18–20) have analyzed the effects of heterodu- bases facilitate the detection of heteroduplexes. Dena- plex formation on molecular diagnostic tests including turation of samples followed by hybridization to pro- quantitative PCR. mote heteroduplex formation has been used to screen In quantitative heteroduplex analysis, the heterodu- diploid DNA for heterozygous variations. This form of plex content of a product is indicated by the magnitude heteroduplex analysis has focused on detection, and of the changes detected by these methods after PCR. although it may involve a quantitative threshold, the Then the proportions of dsDNA genotypes in an orig- result is essentially binary. Recent work in this area (1) inal sample before PCR can be determined. This anal- and elsewhere (2) has explored new implementations ysis involves 2 primary factors, isolating the effect of and applications for a more truly quantitative form of the heteroduplex content on the measured quantity heteroduplex analysis. (such as the melting curve or elution trace), and estab- One setting in which heteroduplexes commonly arise is lishing the relationship of the measured heteroduplex during PCR. Although the extension process of PCR content to the duplex proportions in the original sam- replicates genomic homoduplexes with high accuracy, if ple. The methods for systematically isolating and quan- the original sample is heterozygous, heteroduplexes arise tifying the portion of the signal that depends on het- during the plateau stage when dsDNA is more likely to be eroduplex content are somewhat platform specific. formed by hybridization than extension. Denaturation These methods typically involve separation of a back- and hybridization associates complementary and near- ground component, followed by identification of mea- complementary strands almost randomly. Any of various surable quantities such as heights or areas of particular methods (listed below) that can detect the presence of peaks in the normalized data that are directly propor- heteroduplexes can then be used to detect heterozygous tional to heteroduplex content. For dHPLC, it has been sequence variations. shown (1) that temperature conditions can be opti- If heteroduplexes are not detected, variations, if any, mized by use of predictive software to improve the are known to be homozygous. Homozygous variants separation of heteroduplex and homoduplex peaks and may sometimes be distinguished by differences in achieve high-resolution results. For high-resolution melting temperature, but these differences may be near melting, the procedure has been carried out (2) using a the limit of detection even with high-resolution tech- novel exponential fit to 2 slopes of the raw curve, at niques. Indeed, 4% of human single nucleotide poly- temperatures below and above its transition, and re- morphisms (SNPs) exhibit a symmetry that makes ported elsewhere (21). As an independent confirma- homozygous variants thermodynamically identical up tion, quantitative heteroduplex analysis was performed on to the nearest-neighbor approximation (3). In this case, TGCE traces taken from the same PCR products in (2). a 2-stage mixing strategy allows heteroduplex analysis The reports by Lim et al. (1) and Palais et al. (2) both to be used instead of melting temperature analysis for confirm that the mathematical model relating the pro- the detection of homozygous changes. Samples that are portions of 2 different types of homozygous DNA in a identified as not heterozygous in the first analysis are sample before PCR to the heteroduplex content after mixed with known wild-type DNA, denatured, and PCR is simple yet accurate. This model describes the reannealed. Detection of the heteroduplexes in the heteroduplex proportion after PCR is performed on a mixed sample then indicates the presence of a distinct mixture that before amplification consists of 2 types of homozygous variant of the wild-type DNA. The 2 dsDNA, type A:AЈ in proportion x and type B:BЈ in analyses can be performed in parallel, but then all proportion 1 Ϫ x. These reports support the assumption samples must be prepared mixed and unmixed, then that different strands are amplified in proportion to amplified, denatured, hybridized, and analyzed. Alter- their initial concentrations, a theory that has been natively, the mixing can be performed after determin- confirmed in multiple studies (5–7). Also, thermody- ing which samples are not heterozygous. namic analyses predict that for small mutations, the The specific methods that have been employed for strands A and B are sufficiently similar that they detection of heteroduplexes include postPCR separa- hybridize randomly with their own and each other’s tion techniques such as conventional gel electrophore- complements, again in proportion to their concentra- sis (4–9), electron microscopy and electrophoresis (10), tions. Based on these considerations, after PCR, the denaturing HPLC (dHPLC) (11–13), and temperature- total heteroduplex proportion will be the sum of the

Clinical Chemistry 53, No. 6, 2007 1001 1002 Palais: Quantitative Heteroduplex Analysis

contributions from A:BЈ and B:AЈ,or2x(1 Ϫ x). This to other technologies and problems. Better understanding model is quadratic and symmetric with respect to of differential amplification and hybridization of distinct interchanging the 2 components of the mixture about its strands based on different PCR conditions and initial central maximum, which represents the equal mixture concentrations can also lead to improvements in existing present in a natural heterozygote. Except at this point, methods, such as including initial mixtures of more than 2 possible mixtures always exist that can give rise to the 2 types of dsDNA, refining the models of the effect of same heteroduplex proportion. Inversion to find the duplex proportions on the cumulative signal, and devel- initial mixture proportion from the heteroduplex pro- oping superior techniques to resolve these signals from portion is least accurate when the mixture components background and from each other for eventual quantification. are almost equal in proportion, because the slope of the model is near zero. The greatest accuracy occurs when one component of the mixture is much smaller than the other and the slope is steepest, the conditions under Grant funding/support: None declared. which, in the context of the report in this issue (1), Financial disclosures: None declared. quantitative heteroduplex analysis can provide simpli- fication over existing techniques. Heretofore the determination of mutation load by measuring mitochondrial References DNA heteroplasmy Ͻ10% has required more involved 1. Lim K, Naviaux R, Haas R. Quantitative DNA mutation analysis by DHPLC. procedures, such as cloning and sequencing. Clin Chem 2007;1046–52. 2. Palais R, Liew M, Wittwer CT. Quantitative heteroduplex analysis for single In (2), this model is taken a step further to enable nucleotide polymorphism genotyping. Anal Biochem 2005;346:167–75. complete genotyping of SNPs, both heterozygous and 3. Liew M, Pryor R, Palais R, Meadows C, Erali M, Lyon E, et al. Genotyping of homozygous, with a single heteroduplex assay. The single-nucleotide polymorphisms by high-resolution melting of small ampli- model is extended to include the case in which the cons. Clin Chem 2004;50:1156–64. component of the mixture that is present in proportion 4. Highsmith W Jr, Jin Q, Nataraj A, O’Connor J, Burland V, Baubonis W, et al. Ϫ Use of a DNA toolbox for the characterization of mutation scanning methods. 1 x is heterozygous, sharing one allele with the I. Construction of the toolbox and evaluation of heteroduplex analysis. homozygous reference DNA, so that half of this pro- Electrophoresis 1999;20:1186–94. portion contributes to each type of homoduplex ini- 5. Ruano G, Kidd K. Modeling of heteroduplex formation during PCR from tially present in the previous analysis. The result of this mixtures of DNA templates. PCR Methods Appl 1992;2:112–6. 6. Jensen MA, Straus N. Effect of PCR conditions on the formation of change is that the total heteroduplex content after PCR heteroduplex and single-stranded DNA products in the amplification of of a mixture of homozygous reference DNA of type bacterial ribosomal DNA spacer regions. PCR Methods Appl 1993;3:186– A/A of initial proportion x, with heterozygous type 94. 1 2 7. Ruano G, Deinard A, Tishkoff S, Kidd K. Detection of DNA sequence variation A/B DNA having initial proportion 1-x,is ⁄2 (1 Ϫ x ). via deliberate heteroduplex formation from genomic DNAs amplified en Geometrically, this model is a transformation of the masse in ‘‘population tubes.“ PCR Methods Appl 1994;4:225–31. original: The maximum heteroduplex content that oc- 8. Thomas G, Williams D, Soper S. Capillary electrophoresis-based heterodu- curred in the first model when the proportions were plex analysis with a universal heteroduplex generator for detection of point mutations associated with rifampin resistance in tuberculosis. Clin Chem equal (x ϭ 1⁄2), creating an artificial heterozygote, has ϭ 2001;47:1195–1203. been stretched horizontally to x 0, where the un- 9. Weber J, Barbier V, Pages-Berhouet S, Caux-Moncoutier V, Stoppa-Lyonnet mixed sample is already heterozygous. When perform- D, Viovy J-L. A high-throughput mutation detection method based on hetero- ing a repeat heteroduplex analysis of known homozy- duplex analysis using graft copolymer matrixes: application to Brca1 and gous DNA as described above, it is reasonable to mix Brca2 analysis. Anal Chem 2004;76:4839–48. 10. Gonda M, Kaminchick J, Oliff A, Menke J, Nagashima K, Scolnick E. the target and wild type in these equal proportions to Heteroduplex analysis of molecular clones of the pathogenic Friend virus obtain this maximal heteroduplex content, but this complex: Friend murine leukemia virus, Friend mink cell focus-forming virus, value is far from effective in a single test. It is shown in and the polycythemia- and anemia-inducing strains of Friend spleen focus- (2), by use of the 2 quantitative heteroduplex propor- forming virus. J Virol 1984;51:306–14. 11. Xiao W, Oefner P. Denaturing high-performance liquid chromatography: a tion models above (given that no heteroduplex forma- review. Hum Mutat 2001;17:439–74. tion occurs when the sample has the same genotype as 12. Shinka T, Naroda T, Tamura T, Sasahara K, Nakahori Y. A rapid and simple the reference), that mixing all samples before PCR with method for sex identification by heteroduplex analysis, using denaturing the optimal fraction x ϭ 1/7 of homozygous reference high-performance liquid chromatography (DHPLC). J Hum Genet 2001;46: 263–6. DNA makes it possible to distinguish all 3 genotypes of 13. Jones A, Austin J, Hansen N, Hoogendoorn B, Oefner P, Cheadle, J, et al. a nearest-neighbor symmetric SNP in a single closed- Optimal temperature selection for mutation detection by denaturing HPLC tube assay. This value, according to the model, equal- and comparison to single-stranded conformation polymorphism and hetero- izes the separations in heteroduplex content among the duplex analysis. Clin Chem 1999;45:1133–40. 14. Li Q, Liu Z, Monroe H, Culiat C. Integrated platform for detection of DNA mixtures with the 3 genotypes. When mixed in these sequence variants using capillary array electrophoresis. Electrophoresis proportions, the heteroduplex content of the natural 2002;23:1499–1511. heterozygous genotype is reduced only by ϳ2%, so 15. Gundry C, Vandersteen J, Reed G, Pryor R, Chen J, Wittwer CT. Amplicon almost no resolution is sacrificed compared with the melting analysis with labeled primers: a closed-tube method for differentiating homozygotes and heterozygotes. Clin Chem 2003;49:396–406. original 2-step test. 16. Wittwer CT, Reed G, Gundry C, Vandersteen J, Pryor R. High-resolution Beyond their immediate results, these reports suggest genotyping by amplicon melting analysis using LCGreen. Clin Chem 2003; that quantitative heteroduplex analysis will be applicable 49:853–60. Clinical Chemistry 53, No. 6, 2007 1003

17. Ogino S, Leonard D, Rennert H, Gao S, Wilson R. Heteroduplex formation in Bob Palais SMN gene dosage analysis. J Mol Diagn 2001;3:150. 18. Henley W, Schuebel K, Nielsen D. Limitations imposed by heteroduplex formation on quantitative RT-PCR. Biochem Biophys Res Commun 1996; University of Utah 226:113–7. Math Department 19. Boer P, Ramamoorthy J. How to correct for errors in mRNA quantitation by competitive PCR due to heteroduplex formation of amplification products. Salt Lake City, UT 84112 Cell Mol Biol 1997;43:841–50. Fax 801-585-7664 20. Poirier J, Ohshima K, Pandolfo M. Heteroduplexes may confuse the interpretation of PCR-based molecular tests for the Friedreich ataxia GAA triplet E-mail [email protected] repeat [Letter]. Hum Mutat 1999;13:328–30. 21. Montgomery J, Wittwer CT, Palais R, Zhou L. Simultaneous mutation scanning and genotyping by high-resolution DNA melting analysis. Nat DOI: 10.1373/clinchem.2007.087072 Protoc 2007; 2:59–66. Editorial

Oncopeptidomics: A Useful Approach for Cancer Diagnosis?

In this issue of Clinical Chemistry, Mary Lopez and col- and processing; (b) peptide extraction, chromatographic leagues (1) describe novel methods for isolation of pro- separation, and analytical detection; (c) appropriate bioin- tein-bound peptides from serum and their characteriza- formatic tools for multiparametric analysis of potentially tion by mass spectrometry. Lopez et al. used selected thousands of informative peptides and methods for com- peptide combinations to develop a new profiling method paring differential peptide patterns from healthy individ- for ovarian cancer diagnosis. To put this advance into uals and patients with various diseases; and (d) methods perspective, I will briefly summarize relevant previous for positive peptide identification through sequence de- literature on diagnostic applications of serum proteomic termination to derive clues about their possible biological and peptidomic profiling by mass spectrometry. function. Approximately 5 years ago, a new approach for diag- In early reports on peptidomics for diagnostics, Mar- nosing ovarian cancer, by use of SELDI-TOF mass spec- shall et al. (14) claimed that peptides from the sera of trometry, was proposed by the coauthors of the article healthy individuals and of patients who suffered myocar- under discussion (2). It was then hypothesized that pro- dial infarction can produce MALDI-TOF patterns useful teins or protein fragments released by tumor cells or their for diagnosis. Lopez and collaborators have previously microenvironment may enter the general circulation. By demonstrated the use of peptidomic analysis for diagno- the use of a SELDI chip, proteins or peptides could be sis of Alzheimer disease and ovarian carcinoma (15, 16). extracted from crude serum and used for diagnostic More recently, Villanueva et al. (17) used peptidomic purposes with the aid of mass spectrometry and a math- analysis to diagnose breast, prostate, and ovarian cancers. ematical algorithm. Similar methods have subsequently The limitations of the latter approach have been summa- been used to diagnose numerous other malignancies, such rized recently (18). The method of Villanueva et al. differs as breast, prostate, bladder, pancreatic, head and neck, from other diagnostic peptidomic methods in that it lung, liver, and nasopharyngeal cancers, as well as glio- apparently uses the enzymatic activities of the coagula- mas and melanomas, with impressive diagnostic sensitiv- tion and complement cascades for ex vivo generation of ities and specificities. This method has enjoyed ample informative peptides. This approach should await inde- coverage in scientific journals, the media, and interna- pendent reproduction before any definitive conclusions tional conferences (3). on its validity are drawn. This author and others have criticized this diagnostic Previously (19), I questioned whether the serum pep- approach for methodological shortcomings and bioin- tidome actually exists. Indeed, Koomen et al. (20) re- formatic artifacts (4–9). During the last 5 years, healthy ported more than 250 peptides, Villanueva et al. (17) more debates have been conducted in journals (including this than 650 peptides, and Lowenthal et al. (16) more than one) and at conferences (10, 11). An independent valida- 1200 peptides in serum. It is clear that a large number of tion study would be the best test for this technology. As peptides exist in serum and plasma and that these can be yet, however, no published validation is available to extracted with reliable techniques (1). A caveat for this confirm that these methodologies are working. The pro- large peptide load in serum, however, as shown by ponents of the serum proteomic profiling methods for Koomen et al. (20), is that most of these peptides, includ- cancer diagnostics have now turned their attention to ing the high abundance ones, are generated by a surpris- another possibility, proposing to interrogate the serum ingly small number of proteins, owing to proteolytic peptidome (the collective peptide population of serum) as digestion of high abundance proteins by common en- a source of putative novel cancer biomarkers (12). Others zymes such as thrombin, plasmin, and complement pro- have also postulated that the serum peptidome has po- teins, followed by aminopeptidase and carboxypeptidase tential for diagnostics (13). Significant literature is now processing. We are thus coming down to the critical forming around this principle, which is known as pep- question on peptidomics for diagnostics: Can peptides tidomics, or when applied specifically to cancer diagnos- originating as products of exoproteolytic and endoproteo- tics, as oncopeptidomics. In clinical chemistry practice, we lytic digestion of high abundance proteins have value as already use a number of serum peptides for diagnostics, biomarkers of disease? Given that the parent proteins including insulin and C-peptide for diabetes; parathyroid originate mostly from the liver and/or in the process of hormone, calcitonin, and collagen fragments for osteopo- acute-phase reactions, would these putative biomarkers rosis; probrain-type naturetic peptide for congestive heart have the necessary sensitivity and specificity for early failure; progastrin-releasing peptide for small cell lung detection of cancer and other diseases? Obviously, these carcinoma, ␤-amyloid 1–42 for Alzheimer disease, and questions can be answered only by well-designed, angiotensin 2 for hypertension. The theory that peptides blinded, bias-free and, preferably, prospective investiga- carry important diagnostic information seems reasonable tions in which sample collection and storage are highly because they can act as surrogate endpoints of protein standardized and well documented. synthesis, processing, and degradation. Interrogation of Let us now turn our attention to the recent paper by peptides for diagnostics requires a related array of tech- Lopez et al. (1). Their contribution has several distinct nologies to be in place: (a) optimized sample collection merits. The described method for biomarker discovery is

1004 Clinical Chemistry 53, No. 6, 2007 Clinical Chemistry 53, No. 6, 2007 1005

unbiased and is not restricted by any prior hypothesis. It sis (21). More recently, I made similar proposals for is also a multiparametric approach, which renders itself to peptidomics (19). Here, I wish to reemphasize some multiplexing, and, it is hoped, to better diagnostic sensi- points. Because most of the previous proteomic methods tivity and specificity. The methods of Lopez et al. for (2), as well as genomic methods, for cancer diagnosis and sample preparation and processing are high throughput, subclassification did not pass validation (23). Future allowing large numbers of samples to be analyzed simul- publications, even if describing highly promising data, taneously. No immunologic or other specific reagents are should be viewed with caution (24) until independent necessary. The adoption of high-resolution MALDI-Tan- validations are in place. It is fortunate that organizations dem mass spectrometry for sequence determination of such as the Early Detection Research Network (EDRN; identified peptides is a major asset. Lopez et al. developed edrn.nci.nih.gov) and the Ontario Cancer Biomarker Net- and used software that allowed identification of discrim- work (OBCN; www.obcn.ca) have shown interest in val- inating peptides by comparing spectra from normal and idating in a blinded fashion promising technologies such cancer patients. Following a strict sample collection pro- as the one reported by Lopez et al. Studies on promising tocol to avoid biases as much as possible, the authors technologies that attract media attention should be pub- identified 162 peptides that might be useful as cancer lished so that laboratorians and clinicians, as well as the biomarkers. Their biomarker panels, consisting of Ͼ10 public, receive information that otherwise might fail to peptides per panel, produced sensitivities and specifici- reach patients because of validation problems. ties of ϳ90%. These are good numbers compared with I will conclude with a cautionary note. The reservations CA125, but not yet good enough for screening the general outlined above are not targeting mass spectrometry and population. its potential in diagnostics (25, 26). The published data so The reported approach is also associated with a number far have created euphoria, as well as confusion. It is our of shortcomings. The sample preparation method isolates collective responsibility to find out where we stand now, peptides that are bound to albumin and other high- so that we can better plan for the future. abundance proteins. It is almost certain that non–protein- bound peptides with diagnostic value are lost during this process. Grant/funding support: None declared. The concentration differences identified for peptides Financial disclosures: None declared. used to differentiate between cancer and noncancer patients are relatively small (up to 3.6-fold increases and up References 1. Lopez MF, Mikulskis A, Kuzdzal S, Golenko E, Petricoin III EF, Liotta LA, et al. to 2.6-fold decreases) compared with the range of concen- “Top-down” discovery and identification of serum carrier protein-bound trations seen in the best classical biomarkers (10- to peptide biomarker candidates in ovarian cancer samples using a novel, 1000-fold increases in patients with cancer). Another high-throughput workflow. Clin Chem 2007;53:1067–74. 2. Petricoin III EF, Ardekani AM, Hitt BA, Levine PJ, Fusaro VA, Steinberg SM, concern regarding concentrations is that we do not yet et al. 2002. Use of proteomic patterns in serum to identify ovarian cancer. have good hypotheses for explaining why a putative Lancet 2002;359:572–7. biomarker will decrease, rather than increase, in serum of 3. Check E. Proteomics and cancer: running before we can walk? Nature 2004;429:496–7. cancer patients (19, 21). 4. Diamandis EP. Analysis of serum proteomic patterns for early cancer One question requiring further investigation is whether diagnosis: drawing attention to potential problems. J Natl Cancer Inst the discriminatory peptides, which originated mainly 2004;96:353–6. 5. Diamandis EP. Mass spectrometry as a diagnostic and a cancer biomarker from high abundance proteins and the coagulation cas- discovery tool: opportunities and potential limitations. Mol Cell Proteomics cade, are specific to cancer patients and thus useful for 2004;3:367–78. 6. Diamandis EP, van der Merwe D-E. Plasma protein profiling by mass differentiating them from patients without cancer and spectrometry for cancer diagnosis: opportunities and limitations. Clin Can- whether these peptides are cancer-type specific and can cer Res 2005;11:963–5. be used to differentiate patients with various types of 7. Baggerly KA, Morris JS, Edmonson SR, Coombes KR. Signal in noise: evaluating reported reproducibility of serum proteomic tests for ovarian cancer. As mentioned earlier, these high abundance pro- cancer. J Natl Cancer Inst 2005;97:307–9. teins, usually synthesized by the liver, are likely altered 8. Banks RE, Stanley AJ, Cairns DA, Barrett JH, Clarke P, Thompson D, et al. due to generalized effects such as cancer cachexia, mal- Influences of blood sample processing on low-molecular-weight proteome identified by surface-enhanced laser desorption/ionization mass spectrom- nutrition, and inflammation. These changes are unlikely etry. Clin Chem 2005;51:1637–49. to be specific for cancer and may represent epiphenom- 9. Karsan A, Eigl, BJ, Flibotte S, Gelmon K, Switzer P, Hassell P, et al. Analytical and preanalytical biases in serum proteomic pattern analysis for breast ena. For example, more than 40 years ago, it was reported cancer diagnosis. Clin Chem 2005;51:1525–8. (22) that the coagulation cascade might be defective in 10. Diamandis EP. Point: Proteomic patterns in biological fluids: do they cancer patients. In my opinion, it is unlikely that alter- represent the future of cancer diagnostics? Clin Chem 2003;49:1272–8. 11. Petricoin E 3rd, Liotta LA. Counterpoint: The vision of a new diagnostic ations in nonspecific proteins and their fragmentation paradigm. Clin Chem 2003;49:1276–8. patterns will produce robust algorithms for early cancer 12. Liotta LA, Ferrari M, Petricoin E. Clinical proteomics: written in blood. Nature detection. The relatively small changes in concentration 2003;425:905. 13. Schulte I, Tammen H, Selle H, Schulz-Knappe P. Peptides in body fluids and seen with these biomarkers suggest that the derived tissues as markers of disease. Expert Rev Mol Diagn 2005;5:145–57. algorithms may not be sufficiently reproducible if tested 14. Marshall J, Kupchak P, Zhu W, Yantha J, Vrees T, Furesz S, et al. Processing in an independent series of samples. of serum proteins underlies the mass spectral fingerprinting of myocardial infarction. J Proteome Res 2003;2:361–72. In the past, I made specific recommendations for pub- 15. Lopez MF, Mikulskis A, Kuzdzal S, Bennett DA, Kelly J, Golenko E, et al. lishing future serum proteomic profiling data for diagno- High-resolution serum proteomic profiling of Alzheimer disease samples 1006 Diamandis: Oncopeptidomics for Cancer Diagnosis

reveals disease-specific, carrier-protein-bound mass signatures. Clin Chem 26. Hortin GL. The MALDI-TOF mass spectrometric view of the plasma proteome 2005;51:1946–54. and peptidome. Clin Chem 2006;52:1223–37. 16. Lowenthal MS, Mehta AI, Frogale K, Bandle RW, Araujo RP, Hood BL, et al. Analysis of albumin-associated peptides and proteins from ovarian cancer patients. Clin Chem 2005;51:1933–45. Eleftherios P. Diamandis 17. Villanueva J, Shaffer DR, Philip J, Chaparro CA, Erdjument-Bromage H, Olshen AB, et al. Differential exoprotease activities confer tumor-specific Department of Pathology serum peptidome patterns. J Clin Invest 2006;116:271–84. 18. Diamandis EP, Kulasingam V, Sardana G. Letter to the Editor about and Laboratory Medicine differential exprotease activities confer tumor-specific serum peptidome. Mount Sinai Hospital J Clin Invest 2006;464 http://www.jci.org/cgi/eletters/116/1/271 (accessed December 2006). 600 University Avenue 19. Diamandis EP. Peptidomics for cancer diagnostics: present and future. J Toronto, ONT M5G 1X5 Canada Proteome Res 2006;5:2079–82. 20. Koomen JM, Li D, Xiao L-C, Liu TC, Coombes KR, Abbruzzese J, et al. Direct tandem mass spectrometry reveals limitations in protein profiling experi- Department of Laboratory Medicine ments for plasma biomarker discovery. J Proteome Res 2005;4:972–81. and Pathobiology 21. Diamandis EP. Serum proteomic profiling by matrix-assisted laser desorption-ionization time-of-flight mass spectrometry for cancer diagnosis: Next University of Toronto steps. Cancer Res 2006;66:5540–1. Toronto, ONT 22. S. Carlsson, Fibrinogen degradation products in serum from patients with Canada cancer, Acta Chir Scand 1973;139:499–502. 23. Michiels S, Koscielny S, Hill C. Prediction of cancer outcome with microar- E-mail [email protected] rays: a multiple random validation strategy. Lancet 2005;365:488–92. 24. Diamandis EP. Quality of the scientific literature: all that glitters is not gold. Clin Biochem 2006;39:1109–11. 25. Hortin GL, Jortani SA, Ritchie JC Jr, Valdes R Jr, Chan DW. Proteomics: a new DOI: 10.1373/clinchem.2006.082552 diagnostic frontier. Clin Chem 2006;52:1218–22. Editorial

The Pain Protective Haplotype: Introducing the Modern Genetic Test

Chronic pain is a major health problem, affecting 2%– decrease the number of SNPs needed to identify the 46.5% of adult populations (1, 2). In the United States, haplotype from 15 to Յ3. This work greatly facilitates its common pain conditions account for approximately $61.2 laboratory detection. billion in lost workplace time (3). The human sensation of A simple, inexpensive assay to detect the pain-protec- pain is affected by an individual’s past experiences, health tive haplotype, whether using Pyrosequencing or another status, psychological state, cultural and ethnic back- SNP detection method, has obvious utility in research grounds, pain-coping skills, age, sex, and even pending settings. At some point, such an assay may form the basis litigation (4, 5). Emotional variables profoundly influence of a useful clinical diagnostic test. This latter prospect the perception of painful stimuli (6). In addition, genetic captures the central dilemma presented by the recent contributions to the experience of pain have increasingly flurry of literature reporting on possible genotype-pheno- been recognized (7). However, the complexity of pain type associations with yet undetermined medical rele- processing, the multiplicity of influences on it, and the vance. Tools and techniques that allow for the quick and sensation’s inherent subjectivity and protean manifesta- easy design, validation, and implementation of DNA- tions make the elucidation of relevant heritable factors based tests are now available. However, the acquisition of extremely challenging. knowledge about the clinical significance and utility of Neuropathic pain is caused by injury to, or dysfunction genetic information is laborious, and is proceeding far of, the peripheral or central nervous system (8).Itis more slowly than are technical advances in molecular particularly troublesome because of its prevalence, sever- diagnostics. ity, chronicity, and resistance to therapy (9). Patients with The report of the discovery of a pain-protective haplo- conditions as diverse as diabetes mellitus, alcoholism, type is fascinating and potentially groundbreaking, but HIV, multiple sclerosis, postherpetic neuralgia, and spinal questions about the underlying research remain. The radiculopathies may suffer from chronic neuropathic pain-protective haplotype was identified by haplotyping pain. Radiculopathic pain is among the most frequently enrollees in the Maine Lumbar Spine Study (MLSS) encountered neuropathic pain syndromes (8, 9). (14, 15). This outcomes study compared medical and Sciatica associated with intervertebral disc herniation is surgical treatment for sciatica associated with low back the most common type of radicular leg pain in adult pain. It was organized to address geographically related working populations (10). Although many patients with variations in the rate of back surgery, a likely reflection of herniated discs have favorable outcomes with medical professional uncertainty about the relative benefits of therapy, individuals who have ongoing or severe pain surgical and nonsurgical treatment. The MLSS was non- often undergo lumbar discectomy. Recently, Tegeder et al. randomized and observational. Its enrollment percentage (11) described a putative association between an appar- was low. Consequently, potential biases in patient recruit- ently common haplotype (15.4% of alleles in the study ment are of concern. In addition, enrollment was based on population) of the GTP cyclohydrolase gene (dopa-re- sciatica symptoms without other neurologic confirmation sponsive dystonia, GCH1) and lower degrees of persistent of nerve injury, and follow-up information was obtained pain after back surgery. by mailed survey rather than during face-to-face inter- GTP cyclohydrolase is the rate-limiting enzyme in views by trained professionals (14, 15). tetrahydrobiopterin (BH4) synthesis. BH4 serves as an Low back pain is caused by a combination of neuro- essential cofactor in the production of catecholamines, pathic, skeletal, and myofacial mechanisms. No generally serotonin, and nitric oxide, as well as in phenylalanine accepted diagnostic criteria exist to isolate its neuropathic metabolism. Inactivating mutations in GCH1 have been components. Outcomes of patients with low back pain shown to cause dopa-responsive dystonia, and an atypical and sciatica secondary to disc herniation vary, and knowl- hyperphenylalaninemia characterized by mental retarda- edge of causative factors is incomplete. In the MLSS, as tion, seizures, hyperthermia, and abnormalities of muscle in other studies, surgical patients tended to display tone (12). Tegeder et al. (11) postulate a relationship more severe symptoms than those receiving only medical between the pain protective haplotype and reduction in treatment. At 10 years, increased numbers of operative the upregulation of GCH1 that normally occurs in re- patients reported complete pain relief and greater satis- sponse to pain-inducing stimuli. According to this hy- faction, but surgical treatment did not independently pothesis, patients carrying the haplotype have lower than predict improvement in patients’ predominant symptom, expected BH4 synthesis and diminished catecholamine whether back pain or sciatica. Other variables such as and nitric oxide production. receipt of workers’ compensation and better physical and In this issue of Clinical Chemistry, Lotsch et al. (13) mental health status were associated with worse and describe the development of sequence-based single nucle- better outcomes, respectively. otide polymorphism (SNP) identification assays for the Given the controversy over appropriate management of detection of the pain-protective haplotype. The authors low back pain/sciatica and the lack of correlation of performed discriminant analysis, together with in silico symptom severity with objective findings, the strong haplotyping using the Bayesian algorithm PHASE, to statistical relationship between the haplotype and persis-

Clinical Chemistry 53, No. 6, 2007 1007 1008 Klein: Pain Protective Haplotype

tent pain is surprising. This is especially true because the Once we have established satisfactory test utility in majority of haplotype-positive patients (42 of 46) were defined populations in controlled settings, we must eval- heterozygotes, in whom one might expect limited effects. uate the extent to which utility has been preserved with Given the apparent strength of association of the haplo- broader clinical use. We must compare the sensitivity, type with pain predisposition, the large number of hap- specificity, and positive and negative predictive values in lotype-positive patients who underwent an elective the field with those obtained under more rigidly defined surgical procedure for symptomatic relief (46 of 162) is study conditions. To illustrate the complexity of this troublesome and unexplained. Moreover, only 4 of the process one need only look to the emerging area of limited number of study participants were homozygotes. pharmacogenetics. Fifty years of research has demon- Finally, the authors provided insufficient clinical informa- strated that genetics contributes to individual differences tion about the enrollees to allow us to assess potentially in the pharmacokinetics, efficacy, and toxicity of certain confounding variables. This research now awaits confir- medications. Yet, there remains a paucity of specific data mation by other investigators. that instructs us on how to actually apply this knowledge How do we decide whether and when to transfer the in clinical practice. Moreover, there is little demonstration pain-protective haplotype test, and others similar to it, that acquisition of such information positively impacts into clinical use? At what point do we have enough data? patient outcomes. While achieving a consensus may be difficult, the number Confirmation of the existence of the pain-protective of reported genotype-phenotype correlations ensures that haplotype by other investigators, assuming it will occur, we will encounter such conundrums with increasing will be insufficient to inform us of ways to properly use frequency. Comprehensive data collection often takes the resulting information. The era in which we can rapidly years. In the meantime, our clinical colleagues are eager to design and analytically validate SNP detection assays is make use of information that may help them better care here. The responsibility will increasingly be on us to for their patients. ensure that the clinical utility of these tests, and guidelines The issues that must be addressed before the introduc- for their appropriate use, are established before their tion of DNA-based assays do not necessarily differ in unrestricted introduction into clinical practice. principle from those associated with other diagnostic testing. However, it is the volume of novel findings arising from the abundance of genomic research, coupled Grant/funding support: None declared. with widely varying and often indistinct phenotypes, that Financial disclosures: None declared. mandates special attention to the area. Standard concepts of analytical and clinical sensitivity References 1. Verhaak PFM, Kerssens JJ, Dekker J, Sorbi MJ, Bensing JM. Prevalence of and specificity, positive and negative predictive values, chronic benign pain disorder among adults: a review of the literature. Pain and test utility are as applicable to assays that detect 1998;77:231–9. phenotypically relevant SNPs as they are to routine 2. Elliott AM, Smith BH, Penny KI, Smith WC, Chambers WA. The epidemiology clinical chemistry tests. In today’s world, analytical sen- of chronic pain in the community. Lancet 1999;354:1248–52. 3. Stewart WF, Ricci JA, Chee E, Morganstein D, Lipton R. Lost productive time sitivity and specificity of DNA-based testing usually and cost due to common pain conditions in the US workforce. JAMA approach 100%. Clinical sensitivity and specificity are 2003;290:2443–54. much harder to assess, particularly when considering 4. Fillingim RB. Individual differences in pain responses. Curr Rheumatol Rep 2005;7:342–7. amorphous phenotypes affected by many associated vari- 5. Deyo RA. Pain and public policy. N Engl J Med 2000;342:1211–3. ables. Therefore, we must first establish through rigor- 6. Klossika I, Flor H, Kamping S, Bleichhardt G, Trautmann N, Treede R-D, et al. ously performed prospective validation studies that re- Emotional modulation of pain: a clinical perspective Pain 2006;124:264–8. 7. Edwards RR. Genetic predictors of acute and chronic pain. Curr Rheumatol ported associations are indeed real. Rep 2006;8:411–7. Even if a proposed correlation has been confirmed, we 8. Horowitz SH. The diagnostic workup of patients with neuropathic pain. Med must ask ourselves whether an assay provides sufficient Clin North Am 2007;91:21–30. 9. Gilron I, Watson PN, Cahill CM, Moulin DE. Neuropathic pain: a practical information to improve the care of our patients. Will it guide for the clinical. CMAJ 2006;175:265–75. help direct the workup or management of a clinical 10. Frymoyer JW. Back pain and sciatica. N Engl J Med 1988;318:291–300. problem? Does it aid in identifying candidates for specific 11. Tegeder I, Costigan M, Griffin RS, Abele A, Belfer I, Schmidt H, et al. GTP therapeutic or preventive interventions? Can it guide us cyclohydrolase and tetrahydrobiopterin regulate pain sensitivity and persistence. Nat Med 2006;12:1269–77. toward a mechanistic basis for symptomatic therapy? 12. Ichinose H, Ohye T, Takahashi E, Seki N, Hori T, Segawa M, et al. Hereditary Does it provide useful diagnostic or prognostic informa- progressive dystonia with marked diurnal fluctuation caused by mutations in tion? For example, will the presence or absence of the the GTP cyclohydrolase I gene. Nat Genet 1994;8:236–42. 13. Lotsch J, Belfer I, Kirchhof A, Mishra B, Max M, Doehring A, et al. Reliable pain-protective haplotype help us determine which can- screening for a pain-protective haplotype in the GTP cyclohydrolase 1 gene didates are most likely to benefit from, or be harmed by, (GCH1) through the use of 3 or fewer single nucleotide polymorphisms. Clin elective spinal surgery? Will it help classify patients with Chem 2007;1010–5. 14. Keller RB, Atlas SJ, Singer DE, Chapin AM, Mooney NA, Patrick DL, et al. The diabetes into those more or less predisposed to the Maine Lumbar Spine Study, Part I: background and concepts. Spine development of painful neuropathy? Should patients who 1996;21:1769–76. do not carry the haplotype avoid certain elective proce- 15. Atlas SJ, Keller RB, Wu YA, Deyo RA, Singer DE. Long-term outcomes of surgical and nonsurgical management of sciatica secondary to a lumbar disc dures? Should these patients eschew activities that place herniation: 10 year results from the Maine Lumbar Spine Study. Spine them at risk for neuropathic pain–causing injuries? 2005;30:927–35. Clinical Chemistry 53, No. 6, 2007 1009

Roger D. Klein Fax 216-896-0568 E-mail [email protected] Yale University School of Medicine Department of Genetics DOI: 10.1373/clinchem.2007.087056 333 Cedar St., New Haven, CT 06510 Clinical Chemistry 53:6 1010–1015 (2007) Molecular Diagnostics and Genetics

Reliable Screening for a Pain-Protective Haplotype in the GTP Cyclohydrolase 1 Gene (GCH1) Through the Use of 3 or Fewer Single Nucleotide Polymorphisms

Jo¨rnLo¨ tsch,1* Inna Belfer,2 Anja Kirchhof,1 Bikash K. Mishra,2 Mitchell B. Max,2 Alexandra Doehring,1 Michael Costigan,3 Clifford J. Woolf,3 Gerd Geisslinger,1 and Irmgard Tegeder1

,Background: A haplotype in the GTP cyclohydrolase 1 (dbSNP rs8007267G>A) in the 5؅ untranslated region ,dopa-responsive dystonia) gene (GCH1) is associated c.343 ؉ 8900A>T (dbSNP rs3783641A>T) in intron 1) -with decreased persistent pain. The aim of the present and c.*4279 (dbSNP rs10483639C>G) in the 3؅ untrans study was to develop a screening method for the pain- lated region. Test sensitivity and specificity were still protective haplotype. >95% with 2 or even just 1 of these GCH1 DNA Methods: Complete genetic information for all 15 GCH1 positions. DNA positions constituting the pain-protective GCH1 Conclusions: In silico analysis of complex GCH1 gene haplotype was available from 278 patients. In silico haplotypes reduced the requisite number of tested DNA analyses, including discriminant analysis of the most positions from 15 to 3 while maintaining the reliability, frequent haplotypes, identified distinctive DNA posi- specificity, and sensitivity of the genetic diagnosis. This tions that allow detection of the pain-protective haplo- screening method could reduce laboratory diagnostic type at high sensitivity and specificity with the smallest efforts and facilitate investigations of the pain-protec- ௢ possible number of DNA positions. Pyrosequencing tive GCH1 haplotype. assays were subsequently developed for these DNA © 2007 American Association for Clinical Chemistry positions, established with 662 DNA samples from healthy volunteers, and prospectively validated with a A haplotype in the GTP cyclohydrolase 1 (dopa-respon- random selection of DNA samples genotyped for all 15 sive dystonia) gene (GCH1)4 [(1), Fig. 1B] has recently DNA positions. been associated with decreased persistent pain (2). GTP Results: Diagnosis of the pain-protective GCH1 haplo- cyclohydrolase 1 is upregulated in primary sensory neu- type was possible with 100% sensitivity and specificity rons after peripheral nerve injury. Its inhibition reduced by screening for just 3 GCH1 genetic variants that span nociceptive responses in various rodent models of neuro- the entire DNA range of the haplotype: c.–9610G>A pathic and inflammatory pain, whereas injection of tetrahydrobiopterin produces pain (2). GTP cyclohydrolase 1 is the rate limiting enzyme for the synthesis of tetrahydrobiopterin, an essential cofactor for nitric oxide, cate- 1 pharmazentrum frankfurt/ZAFES, Institute of Clinical Pharmacology, Jo- cholamine, and serotonin synthesis (Fig. 1A). Leukocytes hann Wolfgang Goethe-University, Frankfurt am Main, Germany. 2 National Institute of Dental and Craniofacial Research, National Insti- in carriers of the pain-protective haplotype show reduced tutes of Health, Department of Health and Human Services, Bethesda, MD. up-regulation of tetrahydrobiopterin synthesis following 3 Neural Plasticity Research Group, Department of Anesthesia and Critical stimulation. The functional consequence of the pain- Care, Massachusetts General Hospital and Harvard Medical School, Charles- town, MA. protective haplotype at the molecular level therefore is * Address correspondence to this author at: pharmazentrum frankfurt/ expected to be a prevention of excessive injury-evoked de ZAFES, Institute of Clinical Pharmacology, Johann Wolfgang Goethe-Univer- novo tetrahydrobiopterin synthesis (2). sity, Theodor Stern Kai 7, 60590 Frankfurt am Main, Germany. Fax 49-69-6301- 7636; e-mail: [email protected]. Received November 9, 2006; accepted February 12, 2007. Previously published online at DOI: 10.1373/clinchem.2006.082883 4 Human genes: GCH1, GTP cyclohydrolase 1 (dopa-responsive dystonia).

1010 Clinical Chemistry 53, No. 6, 2007 1011

Fig. 1. Tetrahydrobiopterin pathway and schema of the GCH1 gene. (A), tetrahydrobiopterin pathway with the rate-limiting enzyme, GTP cyclohydrolase 1. (B), schema of the GCH1 gene showing the distribution of the 15 SNPs that characterize the pain-protective haplotype (2). The SNPs are first given in the notation suggested in http://www.hgvs.org/mutnomen (4). Nucleotide 1 is the A of the ATG translation-initiation codon, nucleotides 5Ј of the ATG translation-initiation codon have negative numbers, and the nucleotide 3Ј of the translation stop codon is *1 (the next nucleotide is *2, and so on). Intronic nucleotides for a coding-DNA reference sequence are named as follows: Nucleotides at the beginning of the intron are named with the number of the last nucleotide of the preceding exon, a plus sign, and the position of the nucleotide in the intron (e.g., c.77 ϩ 1G, c.77 ϩ 2T, and so on). Nucleotides at the end of the intron are named with the number of the first nucleotide of the following exon, a minus sign, and the position of the nucleotide upstream in the intron (e.g., c.78–1G). In addition, the nomenclature based on the dbSNP accession number is given for each SNP. The positions of the SNPs along the GCH1 gene are given below the gene schema. The SNPs included in the screening assay are presented in larger letters and italicized. (C), the 15 GCH1 SNPs, their haplotypes, and their allelic frequencies as found in the 388 DNA samples genotyped for all 15 GCH1 SNPs. The pain-protective haplotype is underlined.†, 95% binomial CI.

Identification of the pain-protective GCH1 haplotype Materials and Methods (Fig. 1, B and C) adds to the few known human genetic The sequences of the GCH1 gene on chromosome 14q22.1- variants with evidence for modulation of pain (3), and q22.2 were obtained from the Ensembl Gene ID this haplotype is the first with an apparent association ENSG00000131979 databases at http://www.ensembl. with neuropathic pain to be described. In the context of org/Homo_sapiens/geneview?;geneϭENSG00000131979. pain research and therapy, this variant is therefore of As in the original description of the pain-protective high interest for drug target selection as well as for GCH1 haplotype (2), we have named SNPs according to individualized prognosis determination and analgesic the notation suggested in http://www.hgvs.org/ treatment. Because the haplotype consists of specific mutnomen (4) (Fig. 1). SNPs are also denoted according nucleotides at 15 DNA positions, a decrease in labora- to the nomenclature of the National Center for Biotech- tory efforts required for its identification would facili- nology Information SNP database at http://www. tate wider scientific and clinical application. Through in ncbi.nlm.nih.gov/SNP (i.e., “dbSNP” followed by the silico selection of indicative single nucleotide polymor- accession number; Fig. 1 is cited whenever needed to phisms (SNPs), we show that identifying 3 or even facilitate SNP and sequence identification). fewer SNPs enables reliable detection of the entire haplotype, and we prospectively demonstrate the sen- selection of snps for gch1 genetic screening sitivity and specificity of this approach with Pyrose- We investigated 278 DNA samples available from the ௢ quencing screening assays. original data set (2), and we evaluated an additional 110 1012 Lo¨tsch et al.: GCH1 Pain-Protective Haplotype Screening

DNA samples from randomly chosen individuals of the healthy individuals [age, 27.1 (5.5) years] of white ethnic- cohort from which the 15 GCH1 SNPs forming the pain- ity who had consented to genotyping. The procedure was protective GCH1 haplotype (Fig. 1) had been identified approved by the Medical Faculty Ethics Committee of the with the 5Ј-exonuclease method (5). Johann Wolfgang Goethe University of Frankfurt. Blood The selection of GCH1 SNPs for screening assays was samples were drawn into tubes containing ammonium based on their representation in the pain-protective GCH1 heparin. Genomic DNA was extracted from 200 ␮L blood haplotype. All 15 SNP haplotype positions found at allelic on a BioRobot EZ1 workstation by means of the blood and frequencies of Ͼ1% (constituting ϳ95% of all haplotypes; body fluid spin protocol provided in the EZ1 DNA Blood see Fig. 1) were submitted for discriminant analysis (SPSS 200-␮L Kit (Qiagen). version 12.0.2, SPSS). This analysis showed that an ade- nine at the 1st nucleotide position (c.–9610GϾA) and a assay design thymine at the 4th nucleotide position (c.343 ϩ 8900AϾT) In the Pyrosequencing technique (8, 9), a short oligonu- of the GCH1 haplotype revealed in this analysis identified cleotide (the sequencing primer) binds to single-stranded the pain-protective haplotype with a classification perfor- DNA close to the mutation site and is elongated via mance of 100%. To span the full range of nucleosides sequential dispensing of deoxynucleoside triphosphates. involved in the haplotype, we added the guanine at the If the dispensed deoxynucleoside triphosphate matches last position (c.*4279CϾG) of the haplotype to the screen- the next nucleotide of the DNA sequence, it is incorpo- ing assay. rated into the oligonucleotide, and pyrophosphate is The sensitivity and specificity of the screening assay released. Pyrophosphate is converted to ATP, which (Fig. 2) for the pain-protective haplotype were subse- drives a luciferin-to-oxiluciferin conversion, which pro- quently assessed as follows: sensitivity ϭ correctly posi- duces light that is recorded as a peak with a height tive/(correctly positive ϩ false negative); specificity ϭ proportional to the number of nucleotides incorporated in correctly negative/(correctly negative ϩ false positive). the so-called pyrograms (Fig. 3). We compared the results obtained with the selected SNPs The PCR primers required for amplifying the GCH1 with the “true” assignment of the pain-protective haplo- gene segments of interest (Fig. 1; see Table 1 in the Data type based on the identification of all 15 SNPs (Fig. 1, B Supplement that accompanies the online version of this and C). article at http://www.clinchem.org/content/vol53/ GCH1 haplotypes were identified by means of in silico issue5) and the sequencing primers were designed with haplotyping with PHASE software (http://www.stat. Pyrosequencing Assay Design Software (version 1.0.6; washington.edu/stephens/software.html) (6, 7). Linkage Biotage AB). The specificities of the GCH1 gene primers disequilibrium between SNPs was assessed with EMLD were verified by alignment (http://www.ncbi.nlm. software (Qiqing Huang, University of Texas, Houston, nih.gov/BLAST/). In addition, the software defined the TX; https://cge.mdanderson.org/ϳqhuang/Software/pub. deoxynucleoside triphosphate dispensation orders (Fig. 3) htm; accessed November 3, 2006). for detecting the 3 SNPs.

pyrosequencing screening assays pcr amplification DNA extraction. DNA samples for the development of PCRs were performed in a 25-␮L total assay volume (see screening assays were obtained from 662 unrelated Table 1 in the online Data Supplement) on a Mastercycler ep gradient S instrument (Eppendorf). After PCR amplification, several samples were evaluated by electrophoresis on ethidium bromide–stained agarose gels; the sizes of the PCR products were 321 bp for SNP c.–9610GϾA, 216 bp for c.343 ϩ 8900AϾT, and 161 bp for c.*4279CϾG.

pyrosequencing analysis A volume of each of the 25-␮L PCR templates (biotinylated and nonbiotinylated single strands) was pipetted into a well containing 3 ␮L streptavidin-coated Sepharose beads (Amersham Pharmacia Biotech), 37 ␮L binding buffer [10 mmol/L Tris(hydroxymethyl)-aminomethan, 2 mol/L NaCl, 1 mmol/L EDTA and 0.1% polyoxyethyl- ␮ Fig. 2. Assignment of the pain-protective haplotype (#3, Fig. 1C) on the enesorbitan monolaureate (Tween 20), pH 7.6, and 15 L basis of SNP genotype. HPLC-purified water]. This mixture was incubated for 5 If condition 1 is true, the pain-protective haplotype cannot be present. If min at room temperature (shaker speed, 800/min) to form condition 2 is true, the pain-protective haplotype is present homozygously. If specific complexes of streptavidin-coated Sepharose neither condition 1 nor condition 2 is true, the pain-protective haplotype is present heterozygously. When fewer than 3 SNPs are used, lines of condition 3 beads and biotinylated single strands. These complexes redundant with condition 2 must be removed. were separated from the nonbiotinylated single strands Clinical Chemistry 53, No. 6, 2007 1013

Fig. 3. Designs of forward simplex assays with expected and observed pyrograms for detecting GCH1 SNPs. The relevant positions for genotype identifications are framed. The DNA positions on the reverse DNA strand following the attached sequencing primer is the sequence to be analyzed and determines the nucleotide-dispensation order. The 9 pyrograms denote the 3 possible genotypes for each DNA position. Note that the initial part corresponding to the dispensation of enzyme and substrates has been omitted. For conventional GCH1 sequence analysis, only the fragments corresponding to the pyrograms and relevant for identifying the SNPs (marked with frames) are displayed (nucleotides are coded by different line styles). Note that in the National Center for Biotech- nology Information SNP database (http:// www.ncbi.nlm.nih.gov/SNP/) sequences of some SNPs are given in the reverse strand, according to the sequence from Ensembl Gene ID ENSG00000131979.

on a PyroMark Vacuum Prep Worktable (Biotage). After Plate Low (Biotage) prefilled with 0.16 ␮L of 100 ␮mol/L removal of all liquid by suction, the specific complexes sequencing primer and 39.84 ␮L annealing buffer [20 were captured, transferred into 70% ethanol for 5 s, mmol/L Tris(hydroxymethyl)-aminomethan and 2 denatured in 0.2 mol/L NaOH for 5 s, and washed with mmol/L magnesium acetate tetrahydrate, pH 7.6]. The Tris(hydroxymethyl)-aminomethan (10 mmol/L in water) plate was subsequently heated at 80 °C for 2 min in a PSQ for 5 s. The complexes were then transferred to a PSQ 96 96 Sample Prep Thermoplate Low (Biotage) and cooled to 1014 Lo¨tsch et al.: GCH1 Pain-Protective Haplotype Screening

room temperature. Sequencing took place on a Biotage 20.9%) for c.343 ϩ 8900T, and 20.2% (18.1%–22.5%) for PSQ 96MA System with Biotage enzymes, substrate, and c.*4279G. The DЈ and r2 values ranging between 0.90–0.83 nucleotides (Pyro Gold Reagents reagent set for SNP and 0.74–0.60, respectively, indicated that the variants genotyping and mutation analysis). were in linkage disequilibrium. Among the 662 healthy volunteers, the numbers of homozygous carriers, het- validation of the pyrosequencing assays erozygous carriers, and noncarriers of the 3 SNPs were in To verify the correctness of the genetic diagnoses pro- accord with the Hardy–Weinberg equilibrium (c.–9610A: vided by the 3 assays, we used conventional means ␹2 (2) ϭ 0.05; P ϭ 0.97; c.343 ϩ 8900T: ␹2 (2) ϭ 0.93; P ϭ (AGOWA) to sequence 15 samples of homozygous and 0.63; c.*4279G: ␹2 (2) ϭ 1.38; P ϭ 0.5). The same was true heterozygous carriers of each allele and used these sam- for the numbers of 18 homozygous carriers, 159 heterozy- ples as positive controls during Pyrosequencing. gous carriers, and 485 noncarriers of the pain-protective DNA samples from all 18 individuals who were estab- haplotype as assigned from the genotypes of the 3 SNPs lished by screening assays to be homozygous carriers of (␹2 (2) ϭ 1.27; P ϭ 0.53). The allele frequency of the the pain-protective haplotype and DNA samples from a pain-protective haplotype in 662 DNA samples was 14.7% random selection of 14 heterozygous carriers and 14 (95% binomial CI, 12.9%–16.8%). noncarriers were genetically diagnosed for all 15 GCH1 The screening diagnoses for the pain-protective haplo- SNPs (Fig. 1) by the 5Ј-exonuclease method initially used type with the 3 SNPs were verified for all 18 homozygous to identify the haplotype (5). carriers, 14 heterozygous carriers, and 14 noncarriers of the pain-protective haplotype. These results correspond Results to a prospective test sensitivity and specificity of 100% for in silico prediction of the pain-protective detecting the number of copies of the pain-protective gch1 haplotype with 3 snps haplotype. We used 15 DNA positions to identify 71 heterozygous, 5 homozygous, and 202 noncarriers of the pain-protective Discussion haplotype (haplotype no. 3, framed in Fig. 1C) in the Diagnosis of the pain-protective haplotype by evaluating original haplotyping of 278 DNA patient samples (see Fig. all 15 positions in the GCH1 gene (2) would require 1C for allele frequencies). The sensitivity and specificity of considerable laboratory effort, including both the identi- the screening test for the pain-protective haplotype for fication of the 15 DNA positions and in silico haplotype various combinations of the 3 SNPs were always Ͼ95% assignment. Our analysis of 278 DNA samples genotyped (Table 1). In addition, we were also able to reliably for all 15 positions shows that the pain-protective haplo- diagnose the pain-protective haplotype from the genotype can be reliably identified by assessing the genotypes typic assignment of single SNPs (Fig. 2). of only 3 GCH1 DNA positions, or even fewer. This is possible because the GCH1 SNPs are in linkage disequi- validation of the pyrosequencing screening librium, as we and others (2) have observed. Our analysis assay of these 278 DNA samples suggests that the use of just 2 The Pyrosequencing assays reliably genotyped all 662 SNPs has a sensitivity and a specificity that are suffi- DNA samples for the selected GCH1 variants (c.– ciently high for a screening assay; even an analysis of only 9610GϾA, c.343 ϩ 8900AϾT, and c.*4279CϾG), and the 1 position may be acceptable. Thus, our intention of results of conventional sequencing analyses were in com- designing a screening assay that substantially simplifies plete agreement with those of Pyrosequencing (Fig. 3). the genetic diagnosis of the pain-protective haplotype has In the 662 DNA samples genotyped for the selected been achieved, with a reduction of the GCH1 SNPs to be GCH1 variants, allelic frequencies (95% binomial CIs) assayed from 15 to only 2 or 3. The Pyrosequencing assays were 17.5% (15.4%–19.6%) for c.–9610A, 18.7% (16.7%– that we have developed shorten assay times and contrib-

Table 1. Sensitivity and specificity of screening tests for detecting the GCH1 pain-protective haplotype with 3 or fewer SNPs.a % ,c.–9610G>A c0.343 ؉ 8900A>T c.*4279C>G Sensitivity, % Specificity X X X 100 100 X X – 100 98.6 X – X 100 99.6 X X 100 98.8 X – – 100 97.2 X – 98.5 97.0 – X 100 97.4 a Haplotype assignment based on 278 DNA samples in which the “true” haplotype was established by genotyping all 15 GCH1 SNPs. X, SNP used for haplotype assignment; –, SNP not used for haplotype assignment. Clinical Chemistry 53, No. 6, 2007 1015

ute to a reliable diagnosis. The present analysis suggests ing confirmation of the original haplotype association that other assay methods (e.g., real-time PCR, signal- through additional studies and the acquisition of greater amplification methods, or even restriction enzyme di- knowledge of the assay’s clinical utility. gests) could also be used for SNP diagnosis. This pain-protective GCH1 haplotype is not associated with any overt neurologic dysfunction or other pathology Grant/funding support: This work was supported in part by (10) known to be associated with rare GCH1 mutations National Institutes of Health intramural grant Z01 DE00366 that affect the coding regions of the gene. The SNPs (to M.B.M.). forming the pain-protective GCH1 haplotype are all local- Financial disclosures: None declared. ized in noncoding regions. Their most likely molecular consequence is to decrease GCH1 transcription or RNA References stability, as is suggested by the lower concentrations of 1. Milstien S, Jaffe H, Kowlessur D, Bonner TI. Purification and GCH1 mRNA and protein in forskolin-stimulated leuko- cloning of the GTP cyclohydrolase I feedback regulatory protein, cytes from carriers of the pain-protective GCH1 haplotype GFRP. J Biol Chem 1996;271:19743–51. than in leukocytes from noncarrier controls (2). Thus far, 2. Tegeder I, Costigan M, Griffin RS, Abele A, Belfer I, Schmidt H, et al. GTP cyclohydrolase and tetrahydrobiopterin regulate pain sen- evidence for a pain-protective functional association is sitivity and persistence. Nat Med 2006;12:1269–77. available from a cohort of 162 patients (including 4 3. Kim H, Dionne RA. Genetics, pain, and analgesia. Pain: Clinical homozygous carriers of the pain-protective haplotype) Updates (International Association for the Study of Pain) 2005;8. with chronic lumbar root pain and from 2 different 4. den Dunnen JT, Antonarakis SE. Mutation nomenclature exten- cohorts totaling 547 healthy volunteers (including 10 sions and suggestions to describe complex mutations: a discus- homozygous carriers) who participated in pain assess- sion. Hum Mutat 2000;15:7–12. ments that used several experimental pain models (2). 5. Shi L, Ho J, Norling LA, Roy M, Xu Y. A real time quantitative This limited evidence indicates the need for reproducing PCR-based method for the detection and quantification of simian virus 40. Biologicals 1999;27:241–52. the results and for further research on the importance of 6. Stephens M, Smith NJ, Donnelly P. A new statistical method for the pain-protective GCH1 haplotype for various pain haplotype reconstruction from population data. Am J Hum Genet states. Such studies might enable an assessment of the role 2001;68:978–89. of this genetic variant in the risk for developing pain and 7. Stephens M, Donnelly P. A comparison of bayesian methods for the development of personalized analgesic therapies. haplotype reconstruction from population genotype data. Am J Hum Genet 2003;73:1162–9. In summary, we have developed high-throughput screen- 8. Nyren P, Lundin A. Enzymatic method for continuous monitoring of ing assays for a pain-protective GCH1 haplotype. The inorganic pyrophosphate synthesis. Anal Biochem 1985;151: number of DNA positions that need to be genotyped can 504–9. be reduced from the 15 that originally defined the haplo- 9. Ronaghi M, Karamohamed S, Pettersson B, Uhlen M, Nyren P. Real-time DNA sequencing using detection of pyrophosphate type to 3 or fewer. This reduction provides a substantial release. Anal Biochem 1996;242:84–9. decrease in the laboratory effort required for the identifi- 10. Ichinose H, Ohye T, Takahashi E, Seki N, Hori T, Segawa M, et al. cation of the pain-protective haplotype. The assay we Hereditary progressive dystonia with marked diurnal fluctuation have described therefore could facilitate future investiga- caused by mutations in the GTP cyclohydrolase I gene. Nat Genet tions and possibly clinical diagnostic applications, pend- 1994;8:236–42. Clinical Chemistry 53:6 1016–1022 (2007) Molecular Diagnostics and Genetics

Utilization and Diagnostic Yield of Neurogenetic Testing at a Tertiary Care Facility

Kerstin L. Edlefsen,1,2 Jonathan F. Tait,1,2,3 Mark H. Wener,1,3 and Michael Astion1*

Background: Institutions face increasing charges re- As new genetic tests become available, institutions face lated to molecular genetic testing for neurological dis- increasing charges related to their use. Some forms of eases. The literature contains little information on the genetic testing, such as testing for familial cancer syn- utilization and performance of these tests. dromes, have been well examined by medical societies Methods: A retrospective utilization review was per- (1), insurance companies (2), and others (3, 4). In contrast, formed to determine the diagnostic yield of neuroge- there has been little examination of the clinical use of netic tests ordered during calendar year 2005 at a large neurogenetic tests, although this testing is rapidly ex- academic medical center in the western United States. panding as new disease-causing mutations are identified Results: Overall, a relevant mutation was identified in and new clinical laboratory tests are made available. 30.2% of the 162 patients tested and in 21.5% of the 121 Neurogenetic disorders are individually rare, with a probands, defined as patients for whom no mutation typical prevalence of Ͻ1/10 000. Genetic testing methods has been previously identified in a family member. are complex and are often subject to intellectual property -had a mutation restrictions. Often there is only 1 CLIA-certified labora (65 ؍ Patients with muscle weakness (n detected in 26.2% of all patients and 23.5% of probands tory (or at most a few) offering an individual genetic test, -with an estimated testing cost per positive and charges range from several hundred to several thou ,(51 ؍ n) sand dollars per test. In addition, these disorders are (36 ؍ result of $3190. Patients tested for neuropathy (n had a mutation detected in 27.8% of patients and 22.6% complicated and may require evaluation of multiple ge- with an estimated cost per positive netic loci to confirm or exclude a diagnosis. To address ,(31 ؍ of probands (n had a this issue, some laboratories offer neurogenetic tests in (25 ؍ result of $5955. Patients with chorea (n positive result obtained in 68% of patients and 71.4% of panels; however, the use of panels may lead to less .the estimated cost per positive test efficient testing strategies and increased testing charges ;(7 ؍ probands (n was $440. Other diagnostic categories evaluated include In this context, medical centers are seeking ways to -yield, 11.1%; $7620 per positive), familial encourage clinically efficient and cost-effective test utili ;18 ؍ ataxias (n yield, 12.5%; $6760 zation, but limited data are available to guide clinical ;8 ؍ stroke or dementia syndromes (n decision-making. We analyzed the use of neurogenic per positive), and multisystem mitochondrial disorders testing, including the diagnostic yield and cost per posi- .(yield, 20%; $6485 per positive ;10 ؍ n) tive result, at a university-based tertiary care center. Conclusions: Expert clinicians at a tertiary care center who ordered neurogenetic tests obtained a positive Materials and Methods result in 21.5% of patients without previously identified The study took place at a tertiary care center in the familial mutations. These results can be used for com- western United States. The institutional review board of parison and to help establish utilization guidelines for the host medical center approved the study. neurogenetic testing. We conducted a review of laboratory databases to © 2007 American Association for Clinical Chemistry obtain a complete list of patients who had 1 or more neurogenetic tests ordered during calendar year 2005. The databases included the laboratory information system, the

Departments of 1 Laboratory Medicine, 2 Pathology, and 3 Medicine, Uni- laboratory billing system, and a database within the versity of Washington School of Medicine, Seattle, WA. molecular diagnostics laboratory. Neurogenetic testing * Address correspondence to this author at: University of Washington, was defined as molecular testing relating to the diagnosis Department of Laboratory Medicine, Box 357100, Seattle, WA 98195-7110. Fax of familial neuropathies, muscular dystrophies (including 206-598-6189; e-mail [email protected]. Received November 27, 2006; accepted March 22, 2007. muscular dystonias and atrophies), ataxias, choreas, and Previously published online at DOI: 10.1373/clinchem.2006.083360 familial stroke or dementia syndromes. Tests for mito-

1016 Clinical Chemistry 53, No. 6, 2007 1017

chondrial disorders, including multisystem mitochondrial presenting with muscle weakness, and all other tests disorders, were also included for review as long as the ordered along with DYSF also fell into the “muscle individual patient’s symptoms included neurological disorder” category. The exceptions were tests for Hun- symptoms, because there is some clinical overlap between tington disease and for mitochondrial disorders. Tests for mitochondrial disorders and the other neurogenetic dis- these 2 entities were ordered on patients with a variety of orders considered. presenting symptoms and were therefore categorized Once cases were identified, one author (K.L.E.) re- according to the patient’s primary presenting complaint. viewed medical and laboratory records to determine the For the initial analysis, tests on asymptomatic patients results and clinical context of testing. Problematic cases, were grouped into the most appropriate symptom cate- such as those for which classification of results was gory. For example, an asymptomatic patient tested for reported as “indeterminate”, were reviewed by a 2nd Huntington disease was included in the chorea group, author (J.F.T.). The following information was recorded: and an asymptomatic patient tested for myotonic dystrophy was included in the muscular disorder group. • Provider characteristics, including ordering provider, The results were deidentified and summarized using provider type, and ordering location. SPSS, with the percentage of patients testing positive • Patient characteristics, including year of birth, sex, (diagnostic yield) used as the main outcome measure. The presence or absence of symptoms, presence or absence data were sorted in a variety of ways for evaluation. The of a gene previously identified in a family member, and initial analysis included all 162 patients for whom a presence or absence of nongenetic disorders in the neurogenetic test had been ordered during 2005, to sum- differential diagnosis. marize the aggregate utilization and performance of these • Testing characteristics, including test results, total num- tests as ordered at the host institution. This analysis ber of genetic loci ordered at the host institution, and included some patients for whom a specific genetic mu- sendout laboratory or laboratories performing testing. tation had previously been identified in a family member. The majority of the individual genetic tests were ordered A 2nd analysis included only the 121 patients who were infrequently, so tests were grouped into related diagnostic probands, defined as patients for whom no mutation has categories to facilitate interpretation of ordering patterns been previously identified in a family member, and and diagnostic yields (see Table 1). For clarity, multitest excluded all patients for whom a specific mutation had panels were broken down into the individual genetic loci previously been identified in a family member. This ordered and were predominantly evaluated as such. Be- included those tested presymptomatically (for whom the cause individual tests were at times sent to different pretest probability would be a simple calculation based laboratories because of clinician and laboratory director on the inheritance pattern of the gene and the patient’s preferences and because laboratory charges varied over family history) and those presenting with concerning time, billing records indicated variable test charges for symptoms (for whom the pretest probability would be individual test orders. For the sake of simplicity, a typical even higher). This 2nd analysis was performed to better cost was assigned to each genetic locus tested, and this reflect the diagnostic acumen of the providers ordering estimated cost was used to generate aggregate testing the tests. costs within diagnostic groups. Estimated costs were based on typical charges in the billing records for tests Results sent out during 2005, and estimated costs for in-house Of the 162 patients, 121 (74.7%) were probands, and 41 tests were based on the host institution’s published test (25.3%) had specific genetic mutations previously docu- charges during 2005. Cost of testing did not include any mented in family members. Ninety-nine patients (61.1%) costs of initial or follow-up clinic visits. were male, and patients ranged in age from newborn to 79 All tests ordered on an individual patient usually fell years [mean (SD), 45.3 (16.6) years]. The medical center into a single diagnostic category, and most individual studied primarily sees adult patients, although there are a tests were ordered on patients with only 1 type of few clinics and inpatient units that see infants and chil- presenting symptom. For example, testing for the dysfer- dren. Only 8 patients (4.9%) were born after 1987 and thus lin (DYSF)4 gene for a subtype of limb girdle muscular were younger than 18 in 2005. dystrophy (LGMD)5 was performed only on patients The overall results are presented in Table 2. A relevant mutation was identified in 49 (30.2%) of the 162 patients tested. Retrospective chart review indicated that a total of 4 Human genes: DYSF, dysferlin, limb girdle muscular dystrophy 2B 282 tests for individual genetic loci on these 162 patients (autosomal recessive); PMP22, peripheral myelin protein 22; GJB2, gap junction protein, beta 2, 26kDa; HD, huntington (Huntington disease); BRCA1, originated from the study institution. Forty-nine (17.4%) breast cancer 1, early onset; BRCA2, breast cancer 2, early onset. of the 282 genetic tests were positive. No patient had more 5 Nonstandard abbreviations: LGMD, limb girdle muscular dystrophy; than 1 positive result. The positive rate included 2 pa- SCA, spinocerebellar ataxia; MELAS, mitochondrial encephalopathy, lactic tients for whom test results were reported as indetermi- acidosis and stroke-like episodes; MERRF, myoclonic epilepsy with ragged-red fibers; NARP, neurogenic muscle weakness, ataxia and retinitis pigmentosa nate by the performing laboratory. These results were syndrome; CMT, Charcot-Marie-Tooth. considered molecular and clinical true positives on the 1018 Edlefsen et al.: Utilization and Yield of Neurogenetic Testing

Table 1. List of the 282 neurogenetic tests ordered on the 162 patients included in the study. No. of tests No. of tests Disorder (gene) ordered Disorder (gene) ordered Patients with neuropathy Patients with ataxia CMT1A/HNPPa (PMP22 duplication/deletion) 27 SCA 1, 2, 3, 6, 7 (ATXN1, ATXN2, ATXN3, CACNAIA, ATXN7)10 CMT1A (PMP22 sequencing) 4 SCA 3 (ATXN3)1 CMT1B (MPZ) 11 SCA7(ATXN7)1 CMT1C (LITAF) 4 SCA8(ATXN8OS)2 CMT1D/4E (EGR2) 3 SCA 10 (ATXN10)3 CMT2A (MFN2) 11 SCA 12 (PPP2R2B)3 CMT2K/4A (GDAP1) 7 SCA 14 (PRKCG)2 CMTX1 (GJB1) 15 SCA 17 (TBP)5 CMT4F (PRX) 4 Dentatorubral-pallidoluysian atrophy (ATN1) 1 CMT 2E/1F (NEFL) 5 Ataxia with oculomotor apraxia 1 (APTX)2 Amyloidosis, familial (TTR) 4 Ataxia with oculomotor apraxia 2 (SETX)1 MELAS/MERRF/NARPb 1 Friedreich ataxia (FXN)5 Spastic paraplegia 2 (PLP1)1 Patients with muscle weakness Spastic paraplegia 4 (SPAST)1 Becker/Duchenne MD (DMD) 10 Fragile X associated tremor/ataxia (FMR1)3 Fascioscapulohumeral MD (FRG1) 19 Huntington disease (HD)1 Oculopharyngeal MD (PABPN1) 1 MERRFb 1 LGMD 2A (CAPN3) 3 Leigh syndromeb 1 LGMD 2B (DYSF)7 LGMD 2C (SGCA) 1 Patients with chorea LGMD 2I (FKRP) 7 Huntington disease (HD)25 Emery-Dreyfus/LGMD 1B (LMNA)4 Emery-Dreyfus (EMD) 1 Patients with stroke or dementia Myotonic dystrophy 1 (DMPK) 14 CADASIL (NOTCH3)7 Myotonic dystrophy 2 (CNBP) 8 Frontotemporal dementia with parkinsonism (MAPT)1 Myotonia congenital (CLCN1)1 Myoclonus dystonia (SGCE) 2 Patients with multisystemic symptoms Torsion dystonia (TOR1A) 1 MELAS/MERRF/NARP/KSS/LHONb 1 Familial amyotrophic lateral sclerosis (SOD1) 4 MELAS/MERRF/NARPb 2 Spinal muscular atrophy (SMN1) 3 MELASb 2 X-linked spinal and bulbar muscular atrophy (AR) 9 LHONb 5 Huntington disease (HD) 1 MELASb 1 KSSb 1 DNA polymerase gammab 1 a HNPP, Hereditary neuropathy with liability to pressure palsies; MD, muscular dystrophy; KSS, Kearns-Sayre syndrome; LHON, Leber hereditary optic neuropathy; CADASIL, cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy. b Mitochondrial disorder. basis of an amino acid change affecting a start codon in 1 $440. Patients evaluated for neuropathy (n ϭ 36) had a case and the presence of a frameshift mutation in the mutation detected in 27.8% of patients and 22.6% of other. Seven additional indeterminate results were re- probands. Eight of the 10 positive results were for PMP22 garded as clinical negatives, because the testing did not duplication/deletion tests (charge, $575/test). A total of provide a definitive molecular diagnosis. Of the 121 96 tests (mean, 2.7 tests per patient) were ordered in this probands, 26 (21.5%) had positive results, and of the 20 category. Test costs ranged from $525 to $1700 per locus, patients tested presymptomatically, 7 (35%) had positive results. A more detailed analysis of the data is presented in Table 2. Summary of results of neurogenetic testing in the Table 3. The highest diagnostic yield was in patients 162 patients who had a total of 282 tests ordered. presenting with chorea (n ϭ 25), with a positive result Test results Test results ؍ ؍ obtained in 68.0% of all patients tested and 5 of the 7 (n 162 patients) (n 282 loci) probands (71.4%). The only test ordered in this category Positive 49 (30.2%) 49 (17.4%) Negative 113 (69.8%) 233 (82.6%) was for Huntington disease, which has a charge of $300 Total 162 282 per test. The estimated total cost per positive test was Clinical Chemistry 53, No. 6, 2007 1019

Table 3. Detailed analysis of neurogenetic testing of 162 patients. (121 ؍ Test results in probandsa (n (162 ؍ Test results from all patients (n

No. of patients No. positive % Positive No. of patients No. positive % Positive Total 162 49 30.2% 121 26 21.5% Diagnostic class Choreas 25 17 68.0% 7 5 71.4% Neuropathies 36 10 27.8% 31 7 22.6% Muscular dystrophies, dystonias, or atrophies 65 17 26.2% 51 12 23.5% Mitochondrial 10 2 20.0% 9 1 11.1% Ataxias 18 2 11.1% 17 1 5.9% Stroke/dementia 8 1 12.5% 6 0 0% Provider typeb Genetics (including neurogenetics) 85 32 37.6% 51 15 29.4% Neurology 56 12 21.4% 51 8 15.7% Rehabilitation medicine 10 4 40.0% 9 3 33.3% Other 8 1 12.5% 7 0 0% Nongenetic disorders in differential diagnosisc No 97 43 44.3% 58 21 36.2% Yes 60 5 8.3% 58 4 6.9% Total number of loci tested 1 103 43 41.7% 62 20 32.3% 2 22 2 9.1% 22 2 9.1% 3ϩ 37 4 10.8% 37 4 10.8% a Proband is defined as a patient for whom no molecular genetic defect has previously been identified in a family member, and excludes those presenting for presymptomatic testing as well as those with symptoms and a known familial mutation. b Information was not available on 3 of the 162 patients. c Information was not available on 5 of the 162 patients. resulting in an estimated cost per positive result of $5955. The 2 most commonly ordered specific tests, PMP22 Patients presenting with symptoms suggesting a muscle duplication/deletion and Huntington disease, each per- dystrophy or other muscular disorder (n ϭ 65) had a formed well regarding their diagnostic yield. The test- mutation detected in 26.2% of all patients and 23.5% of specific diagnostic yield for PMP22 duplication/deletion probands, with an estimated cost per positive result of testing (n ϭ 27) was 29.6%. The estimated cost per $3190. A total of 99 individual tests (1.5 tests per patient) positive test for this individual test was $1940. Two were ordered in this diagnostic category, and test costs Huntington disease tests were ordered in patients pre- ranged from $240 to $1850. senting with symptoms primarily suggesting a different Eighteen patients presented with ataxia. Only 2 posi- diagnostic category (1 ataxia and 1 muscle disorder). tive results (11.1%) were obtained in this diagnostic When these 2 patients were included, the test-specific category, including a result for 1 patient with a known diagnostic yield for Huntington testing (n ϭ 27) was 63%, family history of a specific spinocerebellar ataxia (SCA) with an estimated cost per positive result of $475. subtype. The estimated cost per positive result was $7620 The diagnostic yield of neurogenetic testing with re- (individual test costs ranged from $225 to $2335). Simi- sults stratified by the type of ordering provider, the larly, only 1 (12.5%) of the 8 patients tested for a familial presence or absence of nongenetic disorders in the differ- stroke or dementia syndromes had a positive result, and ential diagnosis, and the number of genetic loci tested are that was for a known familial mutation. The estimated shown in Table 3. A total of 19 providers ordered tests on cost per positive result in this patient group was $6760 the 162 patients. Eighty-five (52.5%) of the 162 patients (individual test costs ranged from $342 to $917). Two had tests ordered by geneticists, some of whom subspe- (20.0%) of the 10 patients presenting with symptoms cialized in neurogenetics. The diagnostic yield for the 85 suggesting a multisystem mitochondrial disorder (n ϭ 10) patients whose tests were ordered by geneticists was had a positive result, 1 of whom had a known familial 37.6%, compared with 21.4% for the 56 patients whose mutation. The estimated cost per positive result was $3950 tests were ordered by neurologists, 40.0% for the 10 (tests ranged from $225 to $1365). When the 6 additional patients whose tests were ordered by physicians special- mitochondrial tests that were ordered on patients present- izing in rehabilitation medicine, and 12.5% for 8 patients ing predominantly with neuropathy, muscle weakness, or whose tests were ordered by other physicians (n ϭ 8), ataxia were included, the total diagnostic yield for mito- including pediatricians and ophthalmologists. A diagno- chondrial testing was 12.5%, with an estimated cost per sis was obtained for 44.3% of the 97 patients for whom positive result of $6485. only genetic disorders were considered in their differen- 1020 Edlefsen et al.: Utilization and Yield of Neurogenetic Testing

tial diagnosis, compared with 8.3% of the 60 patients for when diagnostic yields are similar to those obtained in whom nongenetic disorders were considered. A positive our study. result was obtained in 41.7% of patients for whom only a Our belief is that the relatively high diagnostic yields single genetic locus was tested, compared with 9.1% of obtained in this study reflect the overall expertise of patients for whom 2 loci were tested and 10.8% of patients providers in the host institution. Although a breakdown for whom 3 or more loci were tested. Similar patterns for by provider type is shown in Table 3, the intention is to diagnostic yield held for probands. show that nearly all groups of providers had reasonable In some cases in which multiple loci were tested, the diagnostic yields and not to draw conclusions regarding tests were ordered in panels. In total, 22 panels were the relative performance of various provider types. In ordered on 21 patients. Ten of these were orders for a addition, as a major referral center with subspecialized panel consisting of SCA 1, 2, 3, 6, and 7. One patient had providers in the field of neurogenetics, the host institution a broader ataxia panel performed at an outside labora- is likely to see a higher proportion of patients for whom a tory, which included testing for all commercially available mutation has already been identified in a family member SCA types, as well as DRPLA. Four patients had mito- than would be the case in most practice settings. When chondrial panels including at least MELAS (mitochon- these patients present for presymptomatic testing, deter- drial encephalopathy, lactic acidosis and stroke-like epi- mination of the pretest probability is a simple calculation sodes), MERRF (myoclonic epilepsy with ragged-red based on the inheritance pattern of the gene and the fibers), and NARP (neurogenic muscle weakness, ataxia family history of the patient, and therefore the overall and retinitis pigmentosa syndrome). One patient had a diagnostic yield obtained in this study reflects the patient panel performed for LGMD, and 6 patients had a variety population evaluated as much as the diagnostic acumen of panels related to the diagnosis of Charcot-Marie-Tooth of the ordering providers. For comparisons between in- (CMT). Of the 22 panels on 21 patients, 2 panels were stitutions the more relevant number is the positive result positive, resulting in a diagnostic yield of 9.5% per patient obtained in 21.5% of probands, patients who had not had tested. a mutation identified previously in a family member. Although our study was primarily a utilization review Discussion and was not designed to provide a comprehensive anal- In this study, we determined an overall diagnostic yield of ysis of variables affecting the diagnostic yield of neuro- 30.2% for neurogenetic testing performed at a large ter- genetic tests, the results suggest some general principles tiary care facility during 2005. Because the diseases for that may be helpful when weighing a decision to test. A which tests were performed have low prevalence (gener- large proportion (84%) of the positive results were ob- ally Ͻ1 case in 10 000), this diagnostic yield suggests tained among probands for whom only genetic disorders enormous enrichment of the tested population by selec- were considered in their differential list, with a similar tion of patients with high pretest probabilities. proportion (77%) among patients for whom only a single The percentage of positive results (diagnostic yield) genetic locus was tested. This finding suggests that posi- was selected as the main outcome measure because it is tive results may be most likely when the clinical presen- easily calculated and should be readily comparable be- tation suggests a very specific genetic diagnosis. It is also tween institutions, although neurogenetic testing is infor- likely that experienced clinicians use a testing strategy mative whether the results are positive or negative. Diag- that begins with the most prevalent disease-causing mu- nostic yield is also better suited to the evaluation of tation. A specific example would be testing first for neurogenetic testing than quality-adjusted life years PMP22 duplications and deletions in a patient with neu- saved (4, 5), which has been used to evaluate cancer ropathy and an appropriate clinical presentation, before syndrome testing and newborn screening, because neuro- proceeding with testing for less prevalent mutations that genetic disorders typically do not affect longevity and/or can also cause hereditary neuropathy. do not have effective treatments. Likewise, although panels were infrequently ordered The process of establishing a definitive molecular di- among our providers, in some cases such orders were agnosis for suspected neurogenetic disorders has many associated with inefficient testing strategies, particularly benefits. These include genetic counseling and family among less experienced clinicians. When used, panels planning, patient access to appropriate services, and lim- should be constructed to provide those tests with the iting further diagnostic work-up (6). In fact, incorporation highest diagnostic yields; otherwise a reflexive testing of similar molecular genetic testing for connexin 26 mu- strategy should be provided that limits the ordering of tations [gap junction beta-2 gene (GJB2)] in selected pa- tests for very rare molecular subtypes. For example, the tients early in the diagnostic evaluation of childhood clinical presentations of the various SCA subtypes are idiopathic sensorineural hearing loss was actually pro- essentially indistinguishable, so molecular tests for these jected to improve cost-effectiveness in a recent evaluation disorders are typically offered by laboratories and or- (7). The diagnostic yields obtained in the study of con- dered by clinicians in batteries. The practice in many nexin 26 were 18% overall and 38% in selected subgroups, laboratories is to perform testing for the most common suggesting that genetic testing can be used cost-effectively subtypes in an initial battery (SCA 1, 2, 3, 6, and 7), Clinical Chemistry 53, No. 6, 2007 1021

followed by testing for less common mutations in a 2nd nostic entities included in this review, tests for 12 were round of testing. We found this to be the most common available in-house at the host institution. The majority of practice, with specific subtypes tested only in the setting the remainder were available from 2 or more clinical of a previously identified familial mutation. laboratories. For 16 of the entities, however, testing was In contrast, the bundling on test requisition forms of available only from a single laboratory in the United genetic tests for other entities, such as LGMD and CMT, States. Many of these exclusive tests fell within the likely encourages the ordering of these tests in panels by neuropathy category, but there were other tests within ordering providers. We observed a few instances in which this category that were offered by multiple laboratories. panel ordering led to frankly inappropriate testing, such Our study sample was insufficient to allow a broad as testing for genes with an X-linked inheritance pattern analysis of the effect on the cost of testing of test meth- when the patient’s family history excluded this possibil- odology or the number of laboratories offering the test. ity. The host institution has a program of active review of Much discussion has centered around the effect of unusual sendout requests, including most requests for exclusive licensing of genetic testing for familial cancer genetic testing in panels. This protocol likely led to the syndromes such as BRCA1 and BRCA2 and the impact of relatively few orders for panels of this type (only 1 for the associated high testing charges on healthcare systems LGMD and 6 for CMT) in this review, but it may also (8, 9). Little attention has been focused on the same issue represent a larger area of inefficiency in facilities in which as it relates to neurogenetic testing, however. The patent- there is less oversight of test ordering. ing and exclusive licensing of genetic tests restricts the The diagnostic yield and cost per positive result did availability of testing and reduces competition, and it can vary widely between the different diagnostic groups, also promote undesirable over-bundling of related tests. partly reflecting differences in the biology of the genetic Some exclusive licenses applicable to neurogenetic tests entities reviewed. For example, Huntington disease has a are incompletely enforced. Universal enforcement of ex- highly characteristic clinical presentation and is autoso- clusive testing licenses or limitation of the majority of mal dominant and highly penetrant. In addition, testing tests to a sole provider would substantially increase typically requires evaluation of only a single genetic locus typical test charges, comparable to those seen with BRCA1 for trinucleotide expansion of the HD gene (although it testing. For example, the cost per positive result for ataxia should be noted that additional testing for other loci may testing would increase nearly 7-fold, to Ͼ$50 000, if all be indicated in some patients). These characteristics re- tests had been obtained from the laboratory that claims sulted in a high diagnostic yield and, when combined exclusive patent rights for many of these tests. This with the relatively low cost of Huntington genetic testing, increase reflects both higher per-test cost and test pack- a relatively low cost of less than $500 per positive result. aging that encourages the ordering of larger panels of In contrast, each positive result obtained in patients with tests. Thus, policymakers should be aware that many of neuropathy cost more than $5000. Tests in this category the costs per positive result found in this study may be do include some more esoteric and expensive entities; greatly increased in the future because of intellectual however, the high cost per positive result also reflects the property restrictions. need to test multiple genetic loci in many patients, be- The results of this study should be useful to other cause hereditary peripheral neuropathies can be associ- institutions to help establish utilization guidelines for ated with multiple specific genetic mutations with multi- neurogenetic testing and should also contribute to a ple inheritance patterns. Ataxia also had a much higher broader dialogue about the appropriate and cost-effective cost of $7620 per positive result, associated with a diag- use of this important type of molecular genetic testing. nostic yield that was low relative to other diagnostic groups, even among these expert clinicians. To our knowledge, this study is the first to address the Grant/funding support: None declared. cost per positive result of neurogenetic testing. In this Financial disclosures: None declared. study, cost estimates were limited to laboratory charges alone and did not include costs associated with clinic visits and other diagnostic testing, such as imaging or References biopsy. Individual test charges ranged from $225 to up to 1. American Society of Clinical Oncology policy statement update: $2335 per individual genetic locus tested. Most test genetic testing for cancer susceptibility. J Clin Oncol 2003;21: charges ranged from $300 to $700 per locus, with those 2397–406. requiring gene sequencing tending to be at the more 2. Aetna. Aetna clinical policy bulletin on BRCA testing, prophylactic mastectomy, and prophylactic oophorectomy for persons at risk expensive end of the range. A few disorders, including for breast and ovarian cancer. http://www.aetna.com/cpb/medical/ CMT 4F, myotonia congenita, and spastic paraplegia type data/200 299/0227.html (accessed September 15, 2006). ϳ 4, were associated with costs of $2000. For each of these 3. Nelson HD, Huffman LH, Fu R, Harris EL. Genetic risk assessment very rare disorders, diagnostic tests require gene sequenc- and BRCA mutation testing for breast and ovarian cancer suscep- ing, and each test is offered by only a single clinical tibility: systematic evidence review for the U.S. Preventive Ser- laboratory in the United States. Of the 55 distinct diag- vices Task Force. Ann Intern Med 2005;143:362–79. 1022 Edlefsen et al.: Utilization and Yield of Neurogenetic Testing

4. Griffith GL, Edwards RT, Gray J. Cancer genetics services: a 7. Preciado DA, Lim LHY, Cohen AP, Madden C, Myer D, Ngo C, et al. systematic review of the economic evidence and issues. Br J A diagnostic paradigm for childhood idiopathic sensorineural Cancer 2004;90:1697–703. hearing loss. Otolaryngol Head Neck Surg 2004;131:804–9. 5. Carlson JJ, Henrikson NB, Veenstra DL, Ramsey SD. Economic 8. Matthijs G. The European opposition against the BRCA gene analyses of human genetics services: a systematic review. Genet patents. Fam Cancer 2006;5:95–102. Med 2005;7:519–23. 9. Caulfield TA, Knoppers BM, Gold ER, Sheremeta LE, Bridge 6. Bird TD. Risks and benefits of DNA testing for neurogenetic PJ. Genetic technologies, health care policy and the patent disorders. Semin Neurol 1999;19:253–9. bargain. Clin Genet 2003;63:15–8. Clinical Chemistry 53:6 1023–1029 (2007) Molecular Diagnostics and Genetics

Real-Time PCR Determination of IMPDH1 and IMPDH2 Expression in Blood Cells

Sara Bremer,1,2 Helge Rootwelt,1,2 and Stein Bergan*1,2

Background: Inosine monophosphate dehydrogenase ing strategies for treatment effect and disease activity. (IMPDH) catalyzes the rate-limiting step in de novo © 2007 American Association for Clinical Chemistry guanine nucleotide synthesis and is implicated in cell cycle control. Inhibition of this enzyme is associated Inosine monophosphate dehydrogenase (IMPDH)3 is the with immunosuppressive, antiviral, and antitumor ac- rate-limiting enzyme in de novo synthesis of guanine tivity. IMPDH basal activity increases after initiation of nucleotides. Two closely related human IMPDH isoforms, immunosuppressive therapy. types 1 and 2, have been identified, each consisting of 514 Methods: A real-time reverse-transcription PCR assay amino acids with 84% sequence identity (1). The isoforms 4 was developed and validated for mRNA quantification are encoded by 2 distinct genes, IMPDH1 and IMPDH2 of the 2 human IMPDH isoforms. Target gene expres- (inosine monophosphate dehydrogenase 1 and 2), located sions were normalized to the geometric mean of 3 at 7q31.3–q32 and 3p21.2–p24.2, respectively (2, 3). Gene housekeeping genes. Assay utility was tested by analyz- expression of the 2 isoforms is differently regulated in ing patient samples and cultured cells exposed to im- various tissues and cell populations, and they are not munosuppressive drugs such as the IMPDH inhibitor mutually redundant (4, 5). Inhibitors of IMPDH lead to depletion of guanine nucleotide pools, resulting in vari- mycophenolic acid. ous immunosuppressive, antiviral, and antitumor effects. Results: The assay was linear over 6 logs of cDNA input Proliferating lymphocytes are particularly sensitive to and demonstrated specific quantification of IMPDH1 IMPDH inhibition owing to their strong dependency on and IMPDH2 expression in cultured cells and patient de novo guanine nucleotide synthesis (6). Mycophenolic samples. Limits of detection and quantification were 10 acid (MPA), a noncompetitive inhibitor of IMPDH, is and 103 copies of cDNA per reaction, respectively. included in immunosuppressive drug regimens to pre- < Within-run and total between-day CVs were 15% for vent graft rejection after organ transplantation (7). normalized expression. Changes in IMPDH1 and 2 ex- Studies in yeast and human cell lines indicate IMPDH pression were observed in patient samples after initia- gene expression feedback regulation by guanine nucleo- tion of an immunosuppressive regimen that included tides (8, 9). Baseline IMPDH enzyme activity is upregu- calcineurin inhibitors, mycophenolate mofetil, and lated in erythrocytes and whole blood after transplanta- steroids. tion and initiation of immunosuppressive therapy (10– Conclusions: This assay can be used to study the 12). Methods for IMPDH1 and 2 mRNA quantification regulation of IMPDH expression and the involvement of rely mainly on Northern blotting assays. Real-time re- the enzymes in immunological and malignant prolifer- verse transcription (RT)-PCR is the most sensitive method ative conditions. This may contribute to the processes of to measure changes in mRNA concentrations. Published drug development and to the establishment of monitor- real-time RT-PCR assays include a method for IMPDH2

1 Department of Medical Biochemistry, Rikshospitalet-Radiumhospitalet 3 Nonstandard abbreviations: IMPDH, inosine monophosphate dehydro- Medical Centre, Oslo, Norway. genase; MPA, mycophenolic acid; RT, reverse transcription; MMF, mycophe- 2 Institute of Clinical Biochemistry, Faculty Division Rikshospitalet, Uni- nolate mofetil; RIN, RNA integrity number; Cp, crossing point. versity of Oslo, Norway. 4 Human genes: IMPDH1 and IMPDH2, inosine monophosphate dehydro- * Address correspondence to this author at: Department of Medical Bio- genase 1 and 2; CFLAR, CASP8 and FADD-like apoptosis regulator; ALAS1, chemistry, Rikshospitalet-Radiumhospitalet Medical Centre, N-0027 Oslo, ␦-aminolevulinate synthase 1; B2M, ␤2-microglobulin; G6PD, glucose-6-phos- Norway. Fax 0047-23071080; e-mail [email protected]. phate dehydrogenase; HPRT1, hypoxanthine phosphoribosyltransferase 1; and Received October 25, 2006; accepted March 15, 2007. HMBS, hydroxymethylbilane synthase (formerly PBGD, porphobilinogen Previously published online at DOI: 10.1373/clinchem.2006.081968 deaminase).

1023 1024 Bremer et al.: PCR Determination of IMPDH1 and 2 in Blood Cells

quantification (13) and an assay for IMPDH1 and 2 MagNA Pure instrument (Roche). RNA was stored at expression in sheep (14). Both assays normalize target Ϫ70 °C until further processing. gene expression to a single housekeeping gene. Our goal was to study changes in IMPDH1 and 2 RNA analysis. We measured RNA concentration and pu- expression in different blood cell populations to obtain rity spectrophotometrically on the NanoDrop ND-1000 further insight into mechanisms underlying these activity instrument (NanoDrop Technologies). We assessed RNA changes and provide further information on the contribu- quality by electrophoresis using the RNA 6000 Pico Lab- tion of each isoform. We report here an RT-PCR–based Chip Kit on the Agilent Bioanalyzer 2100 (both Agilent assay combining the determination of expression of both Technologies). The RNA integrity number (RIN) algo- human IMPDH isoenzymes based on normalization to a rithm (Agilent 2100 expert software) and calculations of housekeeping gene index. the 28S-to-18S rRNA band intensity ratio were used to evaluate RNA integrity. Materials and Methods patients and samples reverse transcription We collected whole blood specimens from healthy volun- We used MS2 RNA (Roche) as carrier RNA in all steps at ϭ teers (n 8) for assay development. To evaluate the a final concentration of 10 mg/L. The reverse transcrip- utility of the assay we also obtained specimens from tion reaction was performed in a thermal cycler (PTC- individuals enrolled in 2 ongoing studies focusing on the 0150; MJ Research). We compared 2 different reverse impact of immunosuppressive drugs on IMPDH expres- transcription enzymes, Transcriptor (Roche) and Super- sion. One study included renal allograft recipients, from Script III (Invitrogen), and 2 different priming strategies, whom samples were drawn predose on 11 occasions random hexamers (Roche and Invitrogen), and oligo(dT) before they received transplants and during the 1st 2 primers (Invitrogen). We tested the reverse transcription weeks posttransplantation. Routine immunosuppression linearity by running RT-PCR on serial dilutions of isolated included cyclosporine, mycophenolate mofetil (MMF), RNA and RNA stock solutions (Roche). and steroids. The 2nd study included healthy volunteers The optimized reverse transcription reaction involved receiving a single dose of MMF (0.25–1.0 g), followed by preincubation of a 13-␮L mixture of 5 ␮L total RNA, 0.08 IMPDH expression measurement in 9 samples drawn A random primers p(dN) , and H O (PCR grade) predose and the first 24 h after dose. The Regional 260 6 2 before adding Roche reaction buffer, 20 U Roche RNase Committee for Medical Research Ethics approved both inhibitor, Roche dNTP-Mix (1 mmol/L each), and 10 U studies, and informed consent was obtained from all Transcriptor Reverse Transcriptase to a total volume of 20 participants. The assay was also adapted to measure ␮L. Reaction parameters are listed (see Table 1 in the IMPDH expression in cultured lymphoblasts (MOLT-4; online Data Supplement that accompanies this article at European Collection of Animal Cell Cultures). www.clinchem.org/content/vol53/issue6). We stored all To assess the expression profile in whole blood cells, Ϫ we collected 2.5 mL blood in PAXgene Blood RNA tubes cDNA samples at 20 °C until analysis. ϩ (PreAnalytiX). We isolated CD4 cells from 3 mL EDTA blood using paramagnetic beads coated with CD4 anti- quantitative real-time pcr bodies (Dynal Biotech) according to the manufacturer’s Real-time PCR. We performed real-time PCR on the Roche instructions, except for a shortened incubation time of 15 LightCycler 1.0 instrument. SYBR Green I (Roche) was min and 2 washes with PBS. We isolated erythrocytes used to monitor the amplification during the optimizing from 2 mL EDTA blood by filtration through a column stages, and hybridization probes were used for containing ␣-cellulose and microcrystalline cellulose (15). quantification. We lysed isolated blood cells with lysis/binding buffer We titrated template concentrations to identify poten- (Roche) and froze all prepared samples at Ϫ70 °C until tial PCR inhibitors. The optimized reaction mixture con- ␮ RNA isolation. tained 2 L cDNA, 3.5 and 4.5 mmol/L MgCl2 for IMPDH1 and 2, respectively, 0.5 ␮mol/L sense primer rna isolation and 1 ␮mol/L antisense primer for both target genes, After thawing the PAXgene tubes, we centrifuged a 1-mL 0.225 ␮mol/L donor and acceptor probe for both target aliquot (corresponding to 0.3 mL whole blood) at 3500g genes, LightCycler Fast Start DNA Master Hybridization for 10 min. The pellet was suspended in lysis/binding Probes reagent, and PCR-grade H2O to a final volume of buffer and transferred to sample cartridges. We trans- 20 ␮L. The reaction parameters are provided in Table 1 of ϩ ferred lysates of erythrocytes, CD4 cells (4–9 ϫ 105), and the online Data Supplement. MOLT-4 cells (2–8 ϫ 105) to sample cartridges after The crossing point (Cp), defined as the PCR cycle thawing and vortex mixing. Total RNA was extracted where the fluorescence rises above background noise, was from 720-␮L aliquots and eluted in 50 ␮L buffer using the calculated automatically by the 2nd-derivative maximum MagNA Pure HP RNA extraction reagent set on the method (LightCycler Software version 3.5; Roche). Clinical Chemistry 53, No. 6, 2007 1025

Oligonucleotide design. Sequences for IMPDH1 and 2,2 rectly after amplification. We sequenced IMPDH1 and 2 known pseudogenes, and CFLAR (CASP8 and FADD-like amplification products using BigDye Terminator Chemis- apoptosis regulator) were obtained from GenBank. Gene- try on an ABI Prism 3100 genetic analyzer (Applied specific primers were designed using Oligo 6.60 software Biosystems). Reverse transcriptase negative controls and (Molecular Biology Insights); hybridization probe design template negative controls were included in the PCRs. additionally involved the LightCycler Probe Design software (Roche) combined with manual evaluation of alter- Calibration curves. We constructed calibration curves to native sequences. We also used the software to assess the determine reaction efficiencies, linearity, detection and probability of primer dimer or hairpin formation and to quantification limits, and to relate Cp values to template estimate melting temperatures. We checked the primer concentrations. PCR products were purified by agarose and probe target sequences for possible homologies with gel (1.2%) electrophoresis, excised, and eluted from the other sequences by applying Basic Local Alignment gel using a Montage DNA Gel Extraction Device (Milli- Search Tool searches. Primers and probes (see Table 2 in pore). An equimolar quantity of each gene-specific PCR the online Data Supplement) were synthesized by TIB product was serially diluted, and 4–6 replicates of each Molbiol. concentration (106–10 copies/reaction) were amplified. Primers were designed to give amplification products We created calibration curves by plotting the log number spanning different exons to selectively amplify cDNA and of cDNA copies corresponding to each calibrator vs their not genomic DNA. All samples were treated with DNase corresponding Cp value. To assess a possible effect of I (Roche) during RNA purification. matrix components on the PCR efficiency, we amplified a dilution series of heterogeneous cDNA samples. The PCR Quantification. We evaluated 5 housekeeping genes efficiency was calculated from the slope of the calibration Ϫ (housekeeping gene selection set; Roche): ␦-aminolevuli- curve according to the equation E ϭ 10( 1/slope) (16).We ␤ nate synthase 1 (ALAS1), 2-microglobulin (B2M), glu- constructed calibration curves on 5 independent occasions cose-6-phosphate dehydrogenase (G6PD), hypoxanthine and determined the repeatability of the slope and phosphoribosyltransferase 1 (HPRT1), and hydroxymeth- intercept. ylbilane synthase (HMBS; formerly PBGD, porphobilinogen deaminase), as potential references for relative Imprecision. We measured assay imprecision by calculat- quantification. The first 3 genes were selected for normal- ing the CV of relative concentrations and normalized ϩ ization. In the CD4 cells, however, only ALAS1 and B2M ratios of gene expression. We assessed the variability of were selected, because there was a tendency toward the reverse transcription step by synthesizing cDNA on 3 coregulation of G6PD and IMPDH1. We assayed all days and amplifying these cDNAs in a single PCR run. samples and a cDNA calibrator, included in each ampli- Imprecision of the post–reverse transcription steps was fication reaction, in triplicate. Quantification of target calculated for the entire dynamic range based on the gene expression was relative to the geometric mean replicates that generated the calibration curves. Calibrator expression of the selected housekeeping genes in the same stability and the imprecision of the post–reverse transcrip- sample. This ratio was normalized to the corresponding tion steps were determined in 20 calibrator samples run ratio calculated for the calibrator. Gene-specific amplifi- repeatedly over a 2-year period. To assess within-run cation efficiency values were used in these calculations to variability of the complete RT-PCR assay, we measured adjust for efficiency differences. An included gene-spe- relative IMPDH1 and 2 expression in 5 replicates origi- cific calibrator (104 copies) served as a positive control and nating from the same whole blood sample. This was allowed estimation of absolute template concentrations by repeated on 3 occasions to measure the total RT-PCR use of preformed calibration curves. between-run imprecision. data analysis Results We analyzed results of the RT-PCR assay with LightCy- rt-pcr cler Software version 3.5. Relative gene expression calcu- RNA isolation. Isolation of total RNA resulted in concen- lations, including housekeeping indexes, were performed trations of 7–28, 4–26, and 2–13 mg/L with PAXgene, ϩ using LightCycler Relative Quantification software ver- CD4 cell, and reticulocyte samples, respectively. Yields sion 1.0 (Roche) and REST-384 software version 1 (http:// were donor dependent, probably resulting from fluctua- www.gene-quantification.de). We evaluated gene expres- tions in the number of blood cells. The mean 260/280 nm sion changes using the Wilcoxon signed-rank test (SPSS absorption ratios were Ͼ1.9 for all sample types. The version 13.0), and defined P values Ͻ0.05 as significant. mean RIN values were 8.3, 9.5, and 6.7 for PAXgene, ϩ CD4 cell, and reticulocyte samples, respectively. The assay validation 28S:18S rRNA band intensity ratio ranged from 0.4 to 1.9. Specificity. PCR products were run on 1% agarose gels The shapes of the amplification and melting curves were (ethidium bromide, ultraviolet detection), and melting similar for samples and positive controls, indicating the curve analyses were performed on the LightCycler di- presence of intact RNA. 1026 Bremer et al.: PCR Determination of IMPDH1 and 2 in Blood Cells

Reverse transcription. Random hexamer priming was more Fig. 1 in the online Data Supplement). Repeated genera- efficient than oligo(dT) priming for IMPDH1 and the tion of the calibration curves resulted in CVs Ͻ1.7% and selected housekeeping genes, decreasing Cp by 0.3 to 2.0 Ͻ0.8% for slope and intercept, respectively. Assay sensi- cycles. For IMPDH2, oligo(dT) priming decreased Cp tivity was confirmed by the slopes (absolute values 3.04– values by 0.4 cycles. Transcriptor was preferred over 3.26) of the linear regression curves used to measure the SuperScript III, since it resulted in a slightly greater yield dynamic range (see Fig. 1 in the online Data Supplement). and a shortened incubation time of 15 min. The reverse The detection limit was 10 templates per reaction, defined ϫ 2 transcription reaction proved to be linear from 5 10 to as the lowest starting template concentration resulting in ϫ 6 5 10 RNA copies, and a constant volume of total RNA an amplification curve. The lower limit of quantification was used in the reverse transcription reaction for further was set to 103 templates per reaction, since lower concen- applications. trations resulted in a significant increase in imprecision, from CV Յ12.0% to Ն25.7% (Fig. 1). Amplification. PCR efficiencies for IMPDH1 and 2 were The total between-day RT-PCR CVs were 10.4% and 2.03 and 1.99, whereas the efficiencies for ALAS1, B2M, 13.4% (normalized values) for IMPDH1 and 2, respec- and G6PD were 1.99, 1.99, and 1.96, respectively. Ampli- tively. Further details of assay imprecision are given in fication of a heterogeneous cDNA pool showed PCR efficiencies similar to those of the purified cDNA, and Table 1 and Fig. 1. The between-day CVs specific for the therefore calibration curves for further applications were based on the latter material. Depending on the gene, material, yield, and patient status, the Cp values ranged Table 1. Imprecision in reference normalized target gene from 19 to 38 cycles. Reticulocyte and PAXgene samples expression. resulted in the highest (IMPDH1 30–38; IMPDH2 31–36; IMPDH1, IMPDH2, n ϭ 30, day 1) and lowest (IMPDH1 26–30; IMPDH2 CV % CV % ϭ 26–31; n ϭ 30, day 1) Cp values, respectively. Calibrator-normalized Within-run, n 5 6.4 12.8 Total, n ϭ 15a 10.4 13.4 ϭ b expression analysis Not calibrator-normalized Total, n 20 10.6 14.0 a All tested housekeeping genes were detectable in the Five replicates were analyzed on 3 separate days. b Determined over a 2-year period and based on the relative gene expression relevant samples. HPRT1 was excluded because of its role in the calibrator. in purine metabolism, suggesting coregulation with IM- PDH (4). HMBS expression was lower compared with the target genes (in a representative series, n ϭ 4 samples, HMBS Cp ranged from 36 to 39, compared with IMPDH1 30–32 and IMPDH2 30–33). Expression and reaction kinetics were similar for the selected housekeeping genes, ALAS1, B2M, and G6PD, and the target genes. Expression ϩ results obtained from CD4 cell samples indicated potential coregulation of G6PD and IMPDH1. With the excep- ϩ tion of G6PD in CD4 cells, results obtained after separate normalization against each of the housekeeping genes showed the same trends as the results after normalization to the geometric mean of the expression. The target and reference gene expression results showed the following tendencies in the tested sample types: B2M Ͼ G6PD Ͼ IMPDH1 Ն IMPDH2 Ͼ ALAS1. The differences between ϩ IMPDH1 and 2 expressions were largest in the CD4 cells. assay validation PCR product sequencing demonstrated specific amplification of IMPDH1 and 2. The melting temperatures were 90.1 and 86.3 °C for the IMPDH1 and 2 amplicons, and no contaminating products or primer dimers were observed. 6 Gel electrophoresis resulted in single bands correspond- Fig. 1. Within-run imprecision in the range 10 to 10 templates per reaction. ing to the expected length of ϳ190 bp. Target and housekeeping gene CVs with respect to relative concentrations. Calibration was linear over a dynamic range of at least Calculations were based on the replicates that generated the calibration curves 6 orders of magnitude, corresponding to starting concen- and included all steps in the assay after reverse transcription. The horizontal line 6 at 25% represents the upper level of imprecision that we accept for this type of trations of 10 –10 templates per reaction; Pearson corre- assay. Because imprecision increased at lower template concentrations, the lation coefficients were Ϫ0.99 for both IMPDH1 and 2 (see lower limit of quantification was set to 103 templates per reaction. Clinical Chemistry 53, No. 6, 2007 1027

reverse transcription reaction were 2.0% and 5.5% for IMPDH1 and 2, respectively. application We applied the assay to a limited number of samples from a renal transplant recipient. Initiation of immunosuppressive therapy after transplantation was associated with fluctuations in IMPDH1 and 2 expression (Fig. 2). Within the 1st 1–4 h after a single dose of MMF, healthy volunteers demonstrated downregulation of IMPDH1 and upregulation of IMPDH2. Samples drawn on 4 occasions from a control individual not receiving immunosuppressive drugs demonstrated no significant changes of IM- PDH1 or 2 expression (relative expression range, n ϭ 4; ϩ IMPDH1 PAXgene 0.98–1.06, CD4 cells 0.63–0.69; IM- ϩ PDH2 PAXgene 0.36–0.41, CD4 cells 1.94–2.18; i.e., all ranges well within the CV of the assay). Cultured cells incubated with 0.05–0.5 ␮mol/L MPA for 72 h showed an increase in IMPDH enzyme activity, measured by HPLC, which was reflected by similar changes in IMPDH1 and 2 expression. The maximum increase of ϳ60% in IMPDH basal activity (from a mean of 39 pmol XMP/106 cells/ min), as well as IMPDH2 expression (increase by median factor of 2.03, range 2.0–2.4, n ϭ 3 experiments), was observed after incubation with 0.15 ␮mol/L MPA.

Discussion We developed and validated an assay for IMPDH1 and 2 ϩ mRNA quantification in whole blood, CD4 cells, and reticulocytes. The focus of our studies was the impact of the IMPDH inhibitor MPA, and concomitantly administered drugs such as steroids, on the expression of the IMPDH isoforms. We analyzed whole blood cells to study IMPDH1 and 2 expression in relation to the IMPDH enzyme activity measured in whole blood. Lymphocytes, ϩ and CD4 cells in particular, play a pivotal role in graft rejection, and MPA exerts its effects primarily by affecting ϩ lymphocytes. We therefore specifically studied CD4 cells. Erythrocytes constitute the largest population of blood cells and probably influence the whole blood IM- PDH enzyme activity. Expression in reticulocytes, which contain some preformed mRNA, could provide a sensitive indicator for changes in the IMPDH expression and activity of erythrocytes, since reticulocytes represent the Fig. 2. Application of the RT-PCR assay; investigating IMPDH1 and 2 most newly formed population. expression in 1 patient under the influence of immunosuppressive The results with this assay revealed IMPDH1 and 2 therapy. IMPDH1 (A) and IMPDH2 (B) expression was measured relative to the geometric gene regulation after initiation of immunosuppressive mean of ALAS1, B2M, and G6PD (not used for CD4ϩ cells) expression. Whole therapy. This regulation was observed in patient samples blood, CD4ϩ cells, and reticulocytes were obtained from a renal transplant patient before (n ϭ 3, median) and the first 2 weeks after (n ϭ 6) transplanta- as well as in cultured cells, and both steroids and MPA tion. Immunosuppressive therapy (cyclosporine, MMF, and steroids) was initi- seem to be involved (Fig. 2). In single-dose studies of ated at the time of transplantation. Baseline expression is shown at time 0. ϩ healthy volunteers, the preliminary data indicate a qual- IMPDH1 was transiently upregulated in whole blood, CD4 cells, and reticulocytes in the first days after transplantation (A). Reticulocyte and whole blood itatively different response. Although these data so far are samples showed significant IMPDH2 downregulation after transplantation (B). sparse, they could support the hypothesis that steroids (not administered to the healthy volunteers) may be the most important factor regulating IMPDH expression early risk of rejection after transplantation. We will follow up after transplantation. The IMPDH expression changes on these findings in the continuation of the mentioned might influence immunosuppressive response and the studies. 1028 Bremer et al.: PCR Determination of IMPDH1 and 2 in Blood Cells

PAXgene tubes offer a convenient and efficient way to ␤-actin (13, 14). There is, however, no known ubiquitous, ϩ stabilize RNA (17). RNA from CD4 cells, reticulocytes, stably expressed reference gene. To strengthen the accu- and MOLT-4 cells was stabilized in guanidine thiocyanate racy of this assay, the target gene expressions were buffer within an hour after sampling. The RIN values normalized to the geometric mean expression of 3 refer- ϩ showed isolation of intact RNA from PAXgene and CD4 ence genes, a housekeeping gene index. Reference genes cell samples. Reticulocyte samples resulted in lower but were selected among ubiquitously expressed genes from acceptable RIN values, probably because of lower RNA different functional classes to reduce the influence of content and longer processing time before RNA stabiliza- coregulation. tion. The 28S:18S ribosomal RNA band ratio was dependent on RNA input concentration and showed a larger clinical applications degree of variability. This assay includes gene sequences The association between increased IMPDH activity and shorter than 200 nucleotides, which usually can be suc- cellular proliferation and malignant transformation cessfully reverse transcribed and amplified even when makes this enzyme a potential target for treatment in RNA is partially degraded. organ transplantation and for malignant disorders. The 2 This is the first RT-PCR assay using hybridization IMPDH isoforms are associated with different functions probes for specific quantification of both human IMPDH1 and gene regulation, and proliferation and neoplastic and 2. Variations in cell numbers and populations occur- transformation are reported to be linked to increased ring because of drug treatment or immunological condi- IMPDH2 expression (18, 19). Assessment of the relative tions might contribute to the observed variation in yield importance and contribution of the IMPDH isoforms to and expression. Single-template detection is theoretically intracellular processes in various disease states will pro- feasible in real-time RT-PCR, but distribution statistics at vide information to establish whether both enzymes are low template concentrations influence the actual lower valid therapeutic targets. Elucidation of the association limit of detection, as illustrated by the increase in CVs between the IMPDH isoforms’ gene expression and im- with decreasing template concentrations shown in Fig. 1. munosuppression or malignancy might also provide an Low-template reactions may also lead to underestimation opportunity for the monitoring of treatment effect or of the copy number due to relatively higher unspecific disease activity. The presented IMPDH1 and 2 expression background amplification competing for reagents. The assay will be a useful tool for such applications. log–linear inverse relationship between concentrations and Cp values indicates that CVs based on relative concentrations will be nominally higher than CVs based Grant funding/support: The study was performed without on Cp values. We calculated CVs in terms of relative any external financial support. concentrations and the final normalized ratios. These CVs Financial disclosures: There are no conflicts of interest for are directly applicable and relevant for the interpretation any of the authors. of results presented in terms of relative concentrations. Acknowledgements: The collaboration with Ingrid Rasmus- Within-run and total imprecision were Ͻ25%, which may sen, Nils Tore Vethe, and Randeep Mandla in the clinical be an acceptable limit in this type of assay. Repeated studies is greatly acknowledged. We also thank May Ellen analysis of the calibrator sample, already included in each Lauritsen, Thai Tran, and colleagues for their skillful techni- run, demonstrated the stability of the cDNA material over cal assistance. a 2-year period (Table 1). We determined gene-specific PCR efficiencies and References used them in the expression calculations to correct for 1. Natsumeda Y, Ohno S, Kawasaki H, Konno Y, Weber G, Suzuki K. gene- and assay-specific factors. The linear calibration Two distinct cDNAs for human IMP dehydrogenase. J Biol Chem curves indicate constant amplification efficiencies over 1990;265:5292–5. the concentration range studied, confirming minimal in- 2. Gu JJ, Kaiser-Rogers K, Rao K, Mitchell BS. Assignment of the human type I IMP dehydrogenase gene (IMPDH1) to chromosome fluence of PCR inhibitors. 7q31.3-q32. Genomics 1994;24:179–81. Relative expression results may be influenced by inter- 3. Glesne D, Collart F, Varkony T, Drabkin H, Huberman E. Chromo- individual differences in housekeeping gene expression. somal localization and structure of the human type II IMP dehy- When fluctuations over time are studied in individuals, drogenase gene (IMPDH2). Genomics 1993;16:274–7. such variability will be of less importance. Normalization 4. Gu JJ, Stegmann S, Gathy K, Murray R, Laliberte J, Ayscue L, et al. of the target gene expression to an expression ratio of Inhibition of T lymphocyte activation in mice heterozygous for loss several housekeeping genes further compensates for this, of the IMPDH II gene. J Clin Invest 2000;106:599–606. and for variations in the sample amount, RNA recovery, 5. Gu JJ, Tolin AK, Jain J, Huang H, Santiago L, Mitchell BS. Targeted disruption of the inosine 5Ј-monophosphate dehydrogenase type I RNA integrity, efficiency of cDNA synthesis, presence of gene in mice. Mol Cell Biol 2003;23:6702–12. inhibitors, and differences in the overall transcriptional 6. Eugui EM, Almquist SJ, Muller CD, Allison AC. Lymphocyte-selec- activity of the tissues or cells analyzed. Earlier published tive cytostatic and immunosuppressive effects of mycophenolic methods use normalization to a single housekeeping gene acid in vitro: role of deoxyguanosine nucleotide depletion. Scand such as glyceraldehyde-3-phosphate dehydrogenase or J Immunol 1991;33:161–73. Clinical Chemistry 53, No. 6, 2007 1029

7. Allison AC, Eugui EM. Mycophenolate mofetil and its mechanisms phosphate dehydrogenase II in peripheral mononuclear cells from of action. Immunopharmacology 2000;47:85–118. liver-transplant recipients. Transplant Proc 2004;36:2787–90. 8. Escobar-Henriques M, Daignan-Fornier B. Transcriptional regula- 14. Dzidic A, Prgomet C, Mohr A, Meyer K, Bauer J, Meyer HH, et al. tion of the yeast GMP synthesis pathway by its end products. J Biol Effects of mycophenolic acid on inosine monophosphate dehydro- Chem 2001;276:1523–30. genase I and II mRNA expression in white blood cells and various 9. Glesne DA, Collart FR, Huberman E. Regulation of IMP dehydroge- tissues in sheep. J Vet Med A Physiol Pathol Clin Med 2006;53: nase gene expression by its end products, guanine nucleotides. 163–9. Mol Cell Biol 1991;11:5417–25. 15. Beutler E, West C, Blume KG. The removal of leukocytes and 10. Weigel G, Griesmacher A, Zuckermann AO, Laufer G, Mueller MM. platelets from whole blood. J Lab Clin Med 1976;88:328–33. Effect of mycophenolate mofetil therapy on inosine monophos- 16. Rasmussen R. Quantification on the LightCycler. In: Meuer S, phate dehydrogenase induction in red blood cells of heart trans- Wittwer C, Nakagawara K, eds. Rapid Cycle Real-Time PCR, plant recipients. Clin Pharmacol Ther 2001;69:137–44. Methods and Applications. Heidelberg: Springer Press, 2001:21– 11. Vethe NT, Mandla R, Line PD, Midtvedt K, Hartmann A, Bergan S. 34. Inosine monophosphate dehydrogenase activity in renal allograft 17. Rainen L, Oelmueller U, Jurgensen S, Wyrich R, Ballas C, Schram recipients during mycophenolate treatment. Scand J Clin Lab J, et al. Stabilization of mRNA expression in whole blood samples. Invest 2006;66:31–44. Clin Chem 2002;48:1883–90. 12. Sanquer S, Breil M, Baron C, Dhamane D, Astier A, Lang P. 18. Zimmermann AG, Spychala J, Mitchell BS. Characterization of the Induction of inosine monophosphate dehydrogenase activity after human inosine-5Ј-monophosphate dehydrogenase type II gene. long-term treatment with mycophenolate mofetil. Clin Pharmacol J Biol Chem 1995;270:6808–14. Ther 1999;65:640–8. 19. Gu JJ, Spychala J, Mitchell BS. Regulation of the human inosine 13. Vannozzi F, Filipponi F, Di Paolo A, Danesi R, Urbani L, Bocci G, et monophosphate dehydrogenase type I gene: utilization of alterna- al. An exploratory study on pharmacogenetics of inosine-mono- tive promoters. J Biol Chem 1997;272:4458–66. Clinical Chemistry 53:6 1030–1037 (2007) Molecular Diagnostics and Genetics

Quality-Controlled Measurement Methods for Quantification of Variations in Transcript Abundance in Whole Blood Samples from Healthy Volunteers

Elizabeth Herness Peters,1 Sandra Rojas-Caro,2 Mitchell G. Brigell,2 Robert J. Zahorchak,1 Shelley Ann des Etages,2 Patricia L. Ruppel,3 Charles R. Knight,4 Bradley Austermiller,4 Myrna C. Graham,1 Steve Wowk,5 Sean Banks,5 Lakshmi V. Madabusi,5 Patrick Turk,6 Donna Wilder,6 Carole Kempfer,6 Terry W. Osborn,1 and James C. Willey4*

Background: Transcript abundance (TA) measurement tests, the distribution of measured values was gaussian in whole blood frequently is conducted to identify (Shapiro–Wilks test). potential biomarkers for disease risk and to predict or Conclusion: For control or diseased population groups monitor drug response. Potential biomarkers discovered with variation rates as low as those observed in this in this way must be validated by quantitative technol- control group, 17 individuals per group would be re- ogy. In this study we assessed the use of standardized quired to detect 1 SD change with 80% power with a ؍ ␣ reverse transcription PCR (StaRT-PCR™) to validate 2-sided 0.05 statistical test for mean differences. potential biomarkers discovered through whole blood © 2007 American Association for Clinical Chemistry TA profiling. Methods: For each of 15 healthy volunteers, 6 blood Pharmacogenomic technologies increase opportunities to determine the best pharmacologic modality to treat each samples were obtained, including 3 samples at each of 2 individual with a particular disease. The completion of separate visits. Total variation in TA for each gene was the human genome sequencing project (1–3) has made it partitioned into replicate, sample, visit, study partici- feasible to measure transcript abundance (TA)7 of any pant, and residual components. gene in relationship to a clinically important phenotype. Results: Variation originating from technical processing Moreover, several recent studies attempting to develop was <5% of total combined variation and was primarily valid biomarkers through TA measurements in peripheral preanalytical. Interindividual biological sample varia- blood samples yielded encouraging results (4–7). How- tion was larger than technical variation. For 12 of 19 ever, for a TA measurement technology to be successfully used in diagnostic test or pharmaceutical development, it is necessary to demonstrate that the technology generates data suitable for regulatory decision-making. 1 Gene Express, Inc., Toledo, OH. Analytical performance characteristics that must be 2 Pfizer Global Research and Development, Ann Arbor, MI. optimized for such technologies have been detailed by the 3 Innovative Analytics, Inc., Kalamazoo, MI. 4 Division of Pulmonary and Critical Care Medicine, Departments of Food and Drug Administration and the CDC (8–11) and Medicine and Pathology, University of Toledo Health Sciences Campus, include analytical sensitivity and assay limits, impreci- Toledo, OH. sion, analytical specificity (selectivity), interference, and 5 Asuragen, Austin, TX. 6 Radiant Research, Lincoln, NE. * Address correspondence to this author at: Division of Pulmonary and Critical Care Medicine, Department of Medicine, Rm. 0012, Ruppert Health Bldg., University of Toledo Health Sciences Campus, 3000 Arlington Ave., 7 Nonstandard abbreviations: TA, transcript abundance; StaRT-PCR, stan- Toledo, OH 43614. Fax 419-383-6244; e-mail: [email protected]. dardized reverse transcription PCR; MAQC, MicroArray Quality Control Received August 10, 2006; accepted February 13, 2007. Consortium; IS, internal standard; SMIS, Standardized Mixture of Internal Previously published online at DOI: 10.1373/clinchem.2006.078154 Standards; ACTB, ␤-actin.

1030 Clinical Chemistry 53, No. 6, 2007 1031

QC. Standardized reverse transcription PCR (StaRT- chemistry panel included a comprehensive metabolic PCR™) was developed with the goal of optimizing TA profile and cholesterol, iron, and hepatitis B and C testing. measurement with respect to each of these criteria. Suc- Hematology analyses included complete blood count, cess in this effort was documented in the recent Micro- differential, C-reactive protein, and immunology assess- Array Quality Control Consortium (MAQC) project, ment of HIV-1 ϩ 2. Other tests included urinalysis, urine sponsored by the Food and Drug Administration (12). drug abuse screen, electrocardiogram, and breathalyzer Specifically, StaRT-PCR assays had an optimal lower alcohol and carbon monoxide tests. detection threshold (Ͻ10 molecules/assay) and signal-to- analyte response (100%), high precision (mean CV across collection of peripheral blood samples Ն all genes was 6%, and for 6000 molecules it was 3.2%), The sample collection protocol can be found in the Ͼ and linear dynamic range ( 6 orders of magnitude, the Study Operations Manual (see online Data Supplement full range of TA in the MAQC samples) (12). The combi- B). Venipuncture to obtain blood samples was per- nation of high signal-to-analyte response and high repro- formed on each study participant on 2 separate visits ducibility routinely enabled the detection of differences as (Radiant Research). Blood was collected directly into small as 20%. Specificity of StaRT-PCR is ensured through PreAnalytiX PAXgene™ Blood RNA Tubes (Qiagen) quality-controlled design and production of reagents, and according to the manufacturer’s recommendations. On cross-contamination is minimized by use of Good Labo- each visit, 12 tubes of blood (2.5 mL each) were ratory Practices procedures (13). The performance characteristics achieved in the MAQC sequentially collected into serially numbered tubes, study were due to the presence of internal standards (ISs) mixed thoroughly, incubated for2hatroom tempera- Ϫ within a Standardized Mixture of Internal Standards ture, and then stored at 80 °C. PAXgene blood tubes (SMIS™) in every measurement. Use of the SMIS controls were then shipped on dry ice to Asuragen via express for all known sources of variation, including intersample courier service. Before collection in the PAXgene tubes variation in loading due to pipetting, interfering sub- for RNA extraction, 2 additional tubes of blood were stances such as PCR inhibitors, and intergene variation in collected at each visit and sent for serum chemistry and amplification efficiency. The presence of an IS in each hematology analysis. measurement controls for analytical false positives and false negatives. Recent reports have described the success- isolation and quality assessment of total rna ful use of StaRT-PCR to measure the TA values of several Total RNA isolations were carried out by Asuragen with promising biomarkers in samples of blood (14) or other the PAXgene Blood RNA reagent set (Qiagen) protocol tissues (15–18). with a modification to the DNase I treatment. In brief, Development of a drug or clinical diagnostic test instead of the on-column DNase I treatment, the RNA was requires that variation in the technical process and bio- eluted with nuclease-free water from the PAXgene Blood logical variation within test groups both be sufficiently RNA System spin columns and subjected to a DNase I limited to ensure a cost-to-benefit advantage for the treatment in solution. RNA was subsequently purified by proposed test. In this study, to explore multiple potential organic extraction. After individual tubes of blood were sources of variation, triplicate TA measurements were extracted, the RNA was combined into 3 groups of 4 conducted on triplicate blood samples obtained at each of tubes; tubes 1 through 4 became sample A, tubes 5 2 visits from 15 study participants, and each detected through 8 became sample B, and tubes 9 through 12 source of variance was then quantified. became sample C. This procedure enabled analysis of triplicate samples (each replicate sample was processed Materials and Methods separately for collection, RNA extraction, and reverse study population transcription) from each individual for each visit. Any The study protocol (see Data Supplement A that accom- residual DNA contamination was determined by PCR panies the online version of this article at http://www. with a TBP TaqMan® assay. The quality of the RNA was clinchem.org/content/vol53/issue6) was approved by Schulman Associates Institutional Review Board. Partici- spectrophotometrically assessed by determining the A260: pants in this 2-visit study were prescreened and recruited A280 ratio and by determining the RIN (RNA integrity by Radiant Research. All study participants provided number) on a 2100 Bioanalyzer (Agilent). Total RNA informed consent before study enrollment. Selection on samples were checked for residual nuclease contamina- the basis of healthy normals is found in the Study tion and RNA stability by incubating an aliquot of each Operations Manual (see online Data Supplement B). Dur- sample overnight at 37 °C and comparing the Agilent ing the first visit, volunteers were screened by their 2100 Bioanalyzer RNA electropherogram with that ob- medical history, a questionnaire (see online Data Supple- tained from the freshly analyzed sample. After extraction ment C), physical exam, and series of clinical tests. Vol- and quality assessment, all RNA samples were shipped unteers found to be free of apparent illness were invited frozen on dry ice to Gene Express via express courier back for visit 2, scheduled 3 to 7 days after visit 1. Serum service. 1032 Peters et al.: Standardized PCR Measurement in Human Whole Blood

reverse transcription information to calculate the initial number of native Total RNA was reverse transcribed with Moloney murine template molecules in the PCR for each target gene by leukemia virus, reverse transcriptase, deoxynucleoside mathematics of ratios. Following these calculations, the triphosphates, buffer, and oligo(dT) primers according to software determined for each gene whether the initial the manufacturer’s protocol (Invitrogen) and as described native template molecule was in balance (ratio, Ͼ0.1 and previously (15, 16). Ͻ10) with E SMIS or determined which other SMIS was most appropriate in the subsequent high-throughput step. StaRT-PCR analysis The software calculated appropriate SMIS for high- TA for each of 19 genes (see Table 1 in the online Data throughput analysis Ͼ95% of the time. Supplement)8 was determined by StaRT-PCR (Gene Ex- In the high-throughput step, PCRs were prepared with press) at the Standardized Expression Measurement the SMIS that was calculated by HT S-GEM Suite software Center™ under standard operating procedures (12, 13, to be the most appropriate, and triplicate TA measure- 15, 16). ments were obtained for each gene. Results were reported In each assay the TA of a gene was measured relative as target gene cDNA molecules/106 ACTB cDNA to its respective IS within the SMIS. Analysis included 3 molecules. steps; calibration, range finding, and high-throughput analysis. In the calibration step each cDNA sample was statistical analysis diluted so that the ␤-actin (ACTB) cDNA concentration Statistical analyses were performed by Innovative Analyt- (600 000 molecules of ACTB transcript/␮L) was equiva- ics using SAS V8. Clinical assessments were summarized lent to the ACTB IS concentration in the SMIS (600 000 with descriptive statistics, including frequencies for cate- ACTB IS molecules/␮L). gorical measures and means, SDs, and median, minimum, For each subsequent StaRT-PCR experiment a master and maximum for continuous measures. We used a mixture was prepared containing appropriate amounts of random-effects model to categorize total variation into SMIS, calibrated cDNA sample (600 000 molecules of between-individual and within-individual variance com- ACTB transcript/␮l), and all PCR reagents except for ponents. The within-individual variation was further cat- primers; 18-␮L aliquots of the mixture were transferred egorized into intrasample replicate, intravisit sample, and into 96-well plates containing 2 ␮L of gene-specific for- intervisit sample terms, all 2-way and 3-way interactions, ward and reverse primers. PCRs were preheated (95 °C, and residual error. We performed these analyses using 5-min) then amplified for 35 cycles (94 °C, 30 s; 58 °C, 45 s; SAS Proc Varcomp with method ϭ REML. Percentage of 72 °C, 120 s). The Caliper AMS 90 automated, microfluidic total variation for each component was estimated by capillary electrophoresis instrument (Caliper Life Sci- calculating the variance of each component as a percent- ences) with HT DNA 5000 SE30 LabChip was used to age of the total variance. Correlations among intrasample separate and quantify the PCR products. During electro- replicates, intravisit sample replicates, and intervisit sam- phoresis, each IS and/or native template peak was ple replicates, respectively, were evaluated with the Pear- stained with intercalator dye, and the HT S-GEM Suite son correlation coefficient. Bland–Altman analysis for software (Gene Express) was used to automatically quan- intervisit sample variation was also performed to evaluate tify base-pair size and number of PCR product molecules. whether significant bias (visit 1 Ϫ visit 2) existed and In the range-finding step, an aliquot of each calibrated whether it was related to the level of expression. Technical sample was mixed with an equal volume of E SMIS process variation included preanalytical and analytical (600 000 molecules of ACTB IS and 600 molecules of target components. The only definable variation component gene IS/␮L) and PCR reagents. PCR amplification, detec- attributable to replicate measurements of a single sample tion, and analysis were performed as described above. was analytical. Likewise, the definable variation compo- The number of IS molecules before PCR was known for nents attributable to interindividual comparisons were each reaction, and the HT S-GEM Suite software used this biological. In contrast, both technical and biological components contributed to intravisit and intervisit sample variation. Distributions of TA values across individuals were 8 ␤ Human genes: ARHGDIB, Rho GDP dissociation inhibitor (GDI) ; BCL2, characterized by use of descriptive statistics and histo- B-cell CLL/lymphoma 2 (transcript variant ␣); CDC20, CDC20 cell division cycle 20 homolog (S. cerevisiae); CNN2, calponin 2; FANCG, Fanconi anemia, grams, primarily to assess normality. complementation group G; FCGR3A, Fc fragment of IgG, low-affinity IIIa, receptor (CD16a); GMFG, glia maturation factor, ␥; IFNG, interferon, ␥; IL1B, Results interleukin 1, ␤; IL8, interleukin 8; LGALS9, lectin, galactoside-binding, soluble, 9 (galectin 9, transcript variant long); MMP16, matrix metalloproteinase 16 All 15 participants (8 men, 7 women) enrolled in the study (membrane-inserted, transcript variants 1 and 2); PMS2L3, postmeiotic segre- completed the protocol in its entirety. Volunteers were gation increased 2-like 3; PTGS2, prostaglandin-endoperoxide synthase 2 between the ages of 19 and 47 years (mean, 28; median, (prostaglandin G/H synthase and cyclooxygenase); RAG1, recombination 25); 4 were Hispanic/Latino, 3 were Black/African Amer- activating gene 1; SLC31A2, solute carrier family 31 (copper transporters), member 2; TIMP2, TIMP metallopeptidase inhibitor 2; TNF, tumor necrosis ican, 1 was Asian, and 7 were white. Study participants factor (TNF superfamily, member 2); TPBG, trophoblast glycoprotein. were apparently healthy and selected based on the inclu- Clinical Chemistry 53, No. 6, 2007 1033

sion and exclusion criteria described in Materials and as nondetectable. Genes for which some or all of the 267 Methods. measurements were nondetectable due to values Ͻ10 After RNA was isolated from the 12 tubes of blood molecules/106 ACTB molecules included MMP16, TPBG, collected at each visit, sets of 4 tubes were combined, LGALS9, CDC20, RAG1, and IFNG. As reported below, creating 3 RNA samples, A, B, and C. Having 3 samples variations in detection and in measured TA were primar- enabled analysis of variation in TA measured relative to ily related to interindividual biological variation, not variation in RNA quantity and quality in each sample or analytical variation. variation in effects of reverse transcription of each sample. Total variation in TA observation for each gene was Each volunteer provided a total of 6 RNA samples, partitioned into between-individual and within-individ- including the 3 samples obtained at each of 2 visits, except ual variation. Within-individual sample variation was for study participant 23, for whom 1 sample (sample C, further categorized into intrasample replicate variation, visit 1) was lost. RNA yield per PAXgene tube was intravisit sample variation, intervisit sample variation, 4.5–14.4 ␮g [mean, 8.2 ␮g (per sample range, 17.9–57.8 ␮g; ␮ and residual biological variation (Fig. 1). The between- mean, 32.8 g)]. individual variation and the within-individual residual The indicators of sample RNA quality were the A 260:280 variation equally accounted for almost all of the TA ratios, with values of 1.81–2.03, and the 28S:18S ribosomal variation for each of the genes. The technical process RNA ratios, which were 0.5–2.2. Most of the triplicate variation was Ͻ5% and was primarily due to the preana- samples from any individual at any visit displayed simi- lytical components (intravisit or intervisit sample varia- lar 28S:18S ratios. In a few instances, however, this tion). The analytical component, intrasample replicate parameter was quite variable even among samples taken assay variation, was negligible. from the same individual during the same visit. For When intrasample triplicate measurements were eval- example, the 3 samples from study participant 10 taken at Յ visit 2 had 28S:18S ratios of 1.5, 0.7, and 0.7. After an uated, 95% of CVs were 32%, with a median CV of 4.9% overnight incubation (37 °C) of an aliquot of each sample, and mean of 8.9%, indicating very good reproducibility of the 28S:18S ratios remained relatively the same: 1.4, 0.6, the analytical measurement (Fig. 2). There was higher and 0.7, respectively, indicating that samples were stable variation for genes with low expression, such as RAG1. and free of significant nuclease contamination. This effect was probably due to increased stochastic We observed large biological variation in expression effects on random sampling at low transcript numbers. among the 19 genes measured in each individual, and Consistent with this conclusion, comparison of StaRT- large biological variation in expression of each gene PCR with Taqman demonstrated that increased replicate among the 15 participants in the study. The TA measured variation in association with lower gene expression was for most genes in most samples was above the lower limit platform independent (12). of detection of 10 molecules/106 ACTB molecules and With respect to intravisit sample comparison, the CV spanned a broad range from ϳ10 to Ͼ105. Because of was Ͻ30% for 75% of the sample sets (median, 18%; mean, increasing stochastic effects on sampling, a value of Ͻ10 22%; Fig. 3). Among all genes the sample-to-sample molecules/106 ACTB molecules was established a priori variance component was ϳ2%–3% of total variation. Thus

Fig. 1. Percentage of total TA measurement variability attributed to each component. The total variation in a TA observation for each gene was partitioned into between- individual and within-individual sample variation. Within-individual sample variation was further decomposed into visit, sample, assay replicate, and residual variation. The total overall estimated CV for all biological, analytical, and random variation for all 19 genes in this 2-visit study was 54%. 1034 Peters et al.: Standardized PCR Measurement in Human Whole Blood

Fig. 2. Assay replicate variability for each of the 19 genes assayed. Scatter plots showing the variation of intrasample replicate measurements from the mean of the replicates. For each gene, 89 intrasample replicate measurement pairs were plotted, including 1 for each of 3 samples at each of 2 visits for each of 15 individuals, minus values for study participant 23 (sample C, visit 1). The dotted lines reflect a 2-fold difference from the mean of the replicates (solid lines). (A), TA measurement replicate 1 vs TA measurement replicate 2. (B), TA measurement replicate 1 vs TA measurement replicate 3.

sample-to-sample variation was greater than assay repli- Descriptive statistics for the TA distribution for the cate variation (which was Ͻ1% of total variation). mean of samples A, B, and C for each gene across Evaluation of the intervisit sample variation of TA participants and visits are provided in Table 1. Gaussian measurement of each gene showed that 75% of the CVs distribution was detected by the Shapiro–Wilks test for 12 were Ͻ33% (median, 19%; mean, 22%; Fig. 4). Among all of the 19 genes (63%). Genes with several nondetectable genes the visit-to-visit variance component was ϳ2%–3% observations—CDC20, LGALS9, MMP16, and TPBG— of total variation. For CDC20, FANCG, IFNG, and TPNG were more skewed. Lack of gaussian distribution for these the visit-to-visit variance component was somewhat genes was due to a combination of biological interindi- higher, 5%–10%. vidual variation and platform-independent stochastic ef-

Fig. 3. Sample-to-sample variability for each of the 19 genes assayed. Scatter plots showing the sample-sample variability of each sample from the mean of the samples. The dotted lines reflect a 2-fold difference from the mean (solid lines). (A), TA measurement sample A vs TA measurement sample B. (B), TA measurement sample A vs TA measurement sample C. Clinical Chemistry 53, No. 6, 2007 1035

The CV of the mean indicates the assay sensitivity to detect differences in group mean TA measurement differences. Table 2 details the sensitivity of the group mean for detecting shifts with different individual group sizes. Although the SD of an observation appears high, the CV for the mean is very reasonable, Յ25%, with 20 observations. The CV would still be Ͻ30% with only 15 individuals, which indicates fairly good power to detect shifts in mean for different groups. A change of 1 SD in group mean response is often taken as a clinically relevant change. Traditional sample size calculations show that differentiating group means in a study would require 17 participants per group to detect a 1 SD change with 80% power with a 2-sided ␣ ϭ 0.05 statistical test for mean differences, assuming that the disease groups will exhibit low variation similar to what has been observed in healthy volunteers.

Discussion Partly because of the methods used in this study, most of Fig. 4. Visit–retest variability for each of the 19 genes assayed. the variations in gene expression were attributable to Scatter plots showing the visit–visit variability. For each gene, 15 visit pairs were plotted, 1 for each individual. The dotted lines reflect a 2-fold difference from the interindividual biological variation, with little variation mean (solid line). The data are shown as TA measurement visit 1 vs TA due to technical process components. In addition, inter- measurement visit 2. individual biological variation in expression of the selected genes was relatively small in these healthy persons. fects. FCGR3A and PMS2L3 also appeared nongaussian, Under these conditions, it is feasible to use data from a primarily because of biological interindividual variation relatively small number of individuals to define TA in expression. TA distributions for most genes appeared reference intervals in whole blood, which are key to fairly gaussian for these small numbers of individuals/ enabling identification of clinically important groups observations. through TA profile comparison. Furthermore, if the low variation observed within the control group is also ob- Table 1. Distribution of TA measurement for all genes served within a diseased group, only 17 individuals per evaluated in 30 samples.a group will be required to detect a 1 SD difference in group Gene Meanb SD means with 80% power with a 2-sided ␣ ϭ 0.05 statistical ARHGDIB 123 472 33 871 test. BCL2 1024 362 Low analytical variation often increases the power of CDC20 19 17 statistical tests, which translates into relatively low fre- CNN2 14 593 6402 quency of clinical false negatives. In our study analytical FANCG 594 163 variation among intrasample replicates was low, with FCGR3A 51 890 44 651 mean CV of 8.9%. Variation was higher among replicates GMFG 21 422 5695 for genes with low expression (e.g., RAG1, LGALS9). IFNG 163 80 Based on the MAQC study, the increased replicate varia- IL1B 4998 2444 tion at low expression levels was a consequence of natural IL8 250 170 phenomena, specifically platform-independent stochastic LGALS9 56 60 random sampling error (12). If the cause of increased MMP16 Ͻ10 1 PMS2L3 243 95 analytical variation at low signal levels is known to be PTGS2 879 361 stochastic sampling error, increasing the number of rep- RAG1 14 5 licate measurements will decrease mean CVs, illustrating SLC31A2 2516 751 an important reason for numerical quantification. MMP16 TIMP2 8934 2203 concentrations were below the limit of detection in all TNF 287 75 samples; because MMP16 is not expressed in peripheral TPBG 14 7 blood cells, this measurement was included to test the QC a Distribution of TA measurement for all genes evaluated. For each gene there of the methods used (19, 20). were 30 observations including the sample mean (of samples A, B, and C) for Some samples exhibited low 28S:18S ribosomal RNA each of 2 visits across 15 participants. ratios, indicating RNA degradation. Variation in the 28S: b Reported in units of target gene cDNA molecules per 106 ACTB cDNA 18S ratio in 1 sample compared with the others in a molecules. 3-sample set was not associated with higher CVs. For 1036 Peters et al.: Standardized PCR Measurement in Human Whole Blood

Table 2. Imprecision of group mean for various group sizes.a CV of CV of mean CV of mean CV of mean % ,15 ؍ with n % ,20 ؍ with n % ,30 ؍ Gene Meanb SD observations, % with n ARHGDIB 123 472 33 871 27 5 6 7 BCL2 1024 362 35 7 8 9 CDC20 19 17 89 16 20 23 CNN2 14 593 6402 44 8 10 11 FANCG 594 163 27 5 6 7 FCGR3A 51 890 44 651 86 16 19 22 GMFG 21 422 5695 27 5 6 7 IFNG 163 80 49 9 11 13 IL1B 4998 2444 49 9 11 13 IL8 250 170 68 12 15 18 LGALS9 56 60 107 20 24 28 MMP16 Ͻ10 PMS2L3 243 95 39 7 9 10 PTGS2 879 361 41 8 9 11 RAG1 14536789 SLC31A2 2516 751 30 5 7 8 TIMP2 8934 2203 25 5 6 6 TNF 287 75 26 5 6 7 TPBG 14 7 50 9 11 13 a Imprecision of group mean for various group sizes. The CV of mean is represented for all data. The CV is the variability associated with the mean statistic from a sample of observations, based on the standard error, and indicates the utility for detecting shifts in reasonable participant group sizes. The CVs of the mean values for groups with n ϭ 30, n ϭ 20, or n ϭ 15 are presented. b Reported in units of target gene cDNA molecules per 106 ACTB cDNA molecules. instance, samples B and C (visit 2) from study participant the expected variability in acquisition. The quality-con- 23 produced neither a 28S nor 18S band that was quanti- trolled, standardized data reported here will effectively fiable, indicating a high degree of degradation. Neverthe- contribute to defining reference intervals of TA measure- less, the results for all 19 genes in the sample were ment values in peripheral blood. The clinical utility of the comparable to those in the other 2 samples, and the CV data collected is enhanced by the ability to compare and was no greater than the average among all sets. Although combine data across institutions and experiments. Addi- there clearly is a limit to how degraded a sample may be tional TA data collected in future experiments with the before it affects TA measurement with StaRT-PCR re- same cDNA samples or samples from additional healthy, agents, it is well below the threshold associated with diseased, and/or treated individuals will be directly com- complete loss of both ribosomal bands. Thus, a QC step parable to the data presented here and will serve to build that remains to be established is a marker for the mini- the knowledge base regarding measurements of this re- mum quality that an RNA sample must have for reliable sponse and an understanding of how to interpret them. measurement. The 2nd step is to assess the utility of the measure for Intrasample replicate, intravisit sample, and intervisit either detecting change within an individual or for detect- sample variation were identifiable preanalytical and ana- ing group differences such as healthy vs diseased persons. lytical components of variation. Of these, the intravisit Such detection is possible only if the group or change sample variation component was greater than either the difference is greater than the “noise” in the assessment. intrasample replicate or intervisit sample component. The Clinical assessments with low technical variability and intravisit sample component of variation likely was asso- manageable biological variability have a greater potential ciated with variation in RNA extraction and/or reverse to enable accurate detection of treatment effect or disease transcription. Although measurement of 3 different diagnosis. In conclusion, by using methods that minimize cDNA samples derived from the same RNA sample analytical variation, we obtained results indicating that contributed more variation than replicate measurement of clinical studies with high power to detect effects may be the same cDNA sample, all of these technical process achieved with a relatively small number of peripheral Ͻ variation components accounted for 5% of the total blood samples in each group. combined variation. Validation of any clinical assessment requires a 2-step process. The 1st step is to establish the frame of reference by characterization in the normal population, including Grant/funding support: Dr. Willey was supported by Na- distribution of the magnitude of the measured values and tional Institutes of Health/National Cancer Institute Grants Clinical Chemistry 53, No. 6, 2007 1037

CA 85147, CA 95806, and CA 103594, and the George Isaac 10. Ehrmeyer, SS. US Department of Health and Human Services. Cancer Research Fund. Method validation: the regulations. http://www.westgard.com/ Financial disclosures: Dr. Willey has significant equity inter- guest11.htm (accessed 1992). 11. US Department of Health and Human Services, Food and Drug est in Gene Express, Inc. and serves as a consultant to Gene Administration, Center for Devices and Radiological Health. Guid- Express, Inc. ance for industry and FDA staff class II special controls guidance document: instrumentation for clinical multiplex test systems. References http://www.fda.gov/cdrh/oivd/guidance/1546.html (accessed 1. Lander ES, Linton LM, Birren B, Nusbaum C, Zody MC, Baldwin J, March 2005). et al. International Human Genome Sequencing Consortium: initial 12. Canales RD, Luo Y, Willey JC, Austermiller B, Barbacioru CC, sequencing and analysis of the human genome. Nature 2001; Boysen C, et al. Evaluation of DNA microarray results with 409:860–921. quantitative gene expression platforms. Nat Biotechnol 2006;9: 1115–22. 2. McPherson JD, Marra M, Hillier L, Waterston RH, Chinwalla A, 13. Willey JC, Crawford EL, Knight CR, Warner KA, Motten CA, Herness Wallis J, et al. International Human Genome Mapping Consortium: EA, et al. Standardized RT-PCR and the standardized expression a physical map of the human genome. Nature 2001;409:934–41. measurement center, Methods Mol Biol 2004;258:13–41. 3. Venter JC, Adams MD, Myers EW, Li PW, Mural RJ, Sutton GG, et 14. Rots MG, Willey JC, Jansen G, Van Zantwijk CH, Noordhuis P, al. The sequence of the human genome. Science 2001;291: DeMuth JP, et al. mRNA expression levels of methotrexate 1304–51. resistance-related proteins in childhood leukemia as determined 4. Whitney AR, Diehn M, Popper SJ, Alizadeh AA, Boldrick JC, Relman by a standardized competitive template-based RT-PCR method. DA, et al. Individuality and variation in gene expression patterns in Leukemia 2000;14:2166–75. human blood. Proc Natl Acad Sci U S A 2003;100:1896–901. 15. Crawford EL, Khuder SA, Durham SJ, Frampton, M, Utell, MA, Thilly 5. Cobb JP, Mindrinos MN, Miller-Graziano C, Calvano SE, Baker HV, WG, et al. Normal bronchial epithelial cell expression of glutathi- Xiao W, et al. Application of genome-wide expression analysis to one transferase P1, glutathione transferase M3, and glutathione human health and disease. Proc Natl Acad Sci U S A 2005;102: peroxidase is low in subjects with bronchogenic carcinoma. 4801–6. Cancer Res 2000;60:1609–18. 6. McLoughlin K, Turteltaub K, Bankaitis-Davis D, Gerren R, Siconolfi 16. Warner KA, Crawford EL, Zaher A, Coombs RJ, Elsamaloty H, L, Storm K, et al. Limited dynamic range of immune response gene Roshong-Denk SL, et al. The c-myc x E2F-1/p21 interactive gene expression observed in healthy blood donors using RT-PCR. Mol expression index augments cytomorphologic diagnosis of lung Med 2006;12:185–95. cancer in fine-needle aspirate specimens. J Mol Diagn 2003;5: 176–83. 7. Palmer C, Diehn M, Alizadeh AA, Brown PO. Cell-type specific gene 17. Vondracek M, Weaver DA, Sarang Z, Hedberg JJ, Willey JC, expression profiles of leukocytes in human peripheral blood. BMC Warngard L, et al. Transcript profiling of enzymes involved in Genomics 2006;7:115. detoxification of xenobiotics and reactive oxygen in human normal 8. US Department of Health and Human Services, Food and Drug and simian virus 40 T antigen-immortalized oral keratinocytes. Int Administration, Center for Drug Evaluation and Research, Center J Cancer 2002;99:776–82. for Biologics Evaluation and Research, Center for Devices and 18. Mitra AP, Almal AA, George B, Fry DW, Lenehan PF, Pagliarulo V, et Radiological Health. Guidance for industry: pharmacogenomic al. The use of genetic programming in the analysis of quantitative data submissions. http://www.fda.gov/OHRMS/DOCKETS/98fr/ gene expression profiles for identification of nodal status in 2003d-0497-gdl0002.pdf (accessed March 2005). bladder cancer. BMC Cancer 2006;6:159. 9. US Department of Health and Human Services, Food and Drug 19. Kevorkian L, Young DA, Darrah C, Donell ST, Shepstone L, Porter Administration, Center for Devices and Radiological Health. S, et al. Expression profiling of metalloproteinases and their Guidance for industry and FDA staff class II special controls inhibitors in cartilage. Arthritis Rheum 2004;50:131–41. guidance document: drug metabolizing enzyme genotyping sys- 20. Ali A, Langdon J, Stern P, Partridge M. The pattern of expression tem. http://www.fda.gov/cdrh/oivd/guidance/1551.html#7 (ac- of the 5T4 oncofoetal antigen on normal, dysplastic and malignant cessed March 2005). oral mucosa. Oral Oncol 2001;37:57–64. Clinical Chemistry 53:6 1038–1045 (2007) Molecular Diagnostics and Genetics

Effects of Storage, RNA Extraction, Genechip Type, and Donor Sex on Gene Expression Profiling of Human Whole Blood

Sung Jae Kim,1,2 David J. Dix,1*† Kary E. Thompson,1‡ Rachel N. Murrell,1,3§ Judith E. Schmid,1 Jane E. Gallagher,1 and John C. Rockett1¶

Background: Gene expression profiling of whole blood vidual differences. Regression analysis was used to may be useful for monitoring toxicological exposure assess the relative impact of the studied variables. and for diagnosis and monitoring of various diseases. Results: Storage of blood samples for >1 week at 4 °C Several methods are available that can be used to diminished subsequent RNA quality. Interindividual transport, store, and extract RNA from whole blood, but GEP differences were seen, but larger effects were it is not clear which procedures alter results. In addition, observed related to RNA extraction technique, Gene- characterization of interindividual and sex-based varia- Chip, and donor sex. The relative importance of the tion in gene expression is needed to understand sources variables was as follows: storage < genechip < extrac- and extent of variability. tion technique < donor sex. Methods: Whole blood was obtained from adult male Conclusion: Sample storage and extraction methods and -and stored at various interindividual differences, particularly donor sex, af (42 ؍ and female volunteers (n temperatures for various lengths of time. RNA was fect GEP of human whole blood. isolated and RNA quality analyzed. Affymetrix Gene- © 2007 American Association for Clinical Chemistry were used to characterize gene expression (23 ؍ Chips (n profiles (GEPs) and to determine the effects on GEP of Whole blood is increasingly considered a valuable source 4 storage conditions, extraction techniques, types of Ge- of RNA for gene expression profiling (GEP) because neChip, or donor sex. Hierarchical clustering and prin- blood GEPs may be used to determine whether toxicolog- cipal component analysis were used to assess interindi- ical or disease-related events have occurred in inaccessible target tissues (1). Studies in rats have demonstrated differing gene expression changes in blood after ischemic stroke, hemorrhagic stroke, sham surgery, induced sei- 1 National Health and Environmental Effects Research Laboratory, Office zures, hypoxia, and insulin-induced hypoglycemia (2) of Research and Development, US Environmental Protection Agency, Research and similar changes in gene expression after estradiol Triangle Park, NC. exposure in blood and uterine tissue (3). In humans, 2 Curriculum in Toxicology, University of North Carolina at Chapel Hill, Chapel Hill, NC. Twine et al. found disease-associated expression profiles 3 Department of Environmental and Molecular Toxicology, North Carolina in peripheral blood mononuclear cells from patients with State University, Raleigh, NC. renal cell carcinoma (4). * Address correspondence to this author at: National Center for Compu- tational Toxicology (D343-03), Office of Research and Development, US Investigation of the effects of biological and technolog- Environmental Protection Agency, Research Triangle Park, NC 27711. Fax ical variables is essential for confident use of whole blood 919-541-1194; e-mail [email protected]. GEP in research and clinical studies and has been the aim † Current affiliation: National Center for Computational Toxicology, Office of several studies (Table 1). Whitney et al. (5) used cDNA of Research and Development, Research Triangle Park, NC. ‡ Current affiliation: Reproductive Toxicology, Pharmaceutical Research arrays to analyze blood from healthy volunteers and Institute, Bristol-Myers Squibb, New Brunswick, NJ. found that interindividual sample variation was associ- § Current affiliation: Department of Pharmacology and Toxicology, Brody School of Medicine, East Carolina University, Greenville, NC. ¶ Current affiliation: Rosetta Inpharmatics LLC (a wholly owned subsid- iary of Merck & Co., Inc.), Seattle, WA. Received August 14, 2006; accepted March 14, 2007. 4 Nonstandard abbreviations: GEP, gene expression profile; RIN, RNA Previously published online at DOI: 10.1373/clinchem.2006.078436 integrity number.

1038 Clinical Chemistry 53, No. 6, 2007 1039

Table 1. Studies examining sources of biological and technical variation in gene expression profiles of human whole blood. Number of Blood collection Reference subjects method Storage conditions RNA preparation method Expression profiling method Rainen et al. (8) 4 EDTA Vacutainer 0 to 7 days at Trizol and QiaAmp Northern blot (2002) 22 °C PAXgene Vacutainer Up to 90 days at PAXgene blood RNA isolation Quantitative RT-PCR 4°CorϪ20 °C system Tanner et al. (9) 1 EDTA Vacutainer Up to 24 h at room ABI Prism 6100 Nucleic Acid 5Ј Nuclease PCR assay (2002) temperature or PrepStation 4°C Pahl & and Brune Not given EDTA Vacutainer Up to 7 days at Acidic phenol/guanidine TaqMan RT-qPCR (18) (2002) room temperature isothiocyanate EDTA Vacutainer PAXgene blood RNA isolation system Whitney et al. (5) 75 Sodium citrate Trizol at Ϫ80 °C Trizol cDNA microarray (2003) Vacutainer Thach et al. (10) 1 PAXgene Vacutainer 2 h at room PAXgene blood RNA isolation Affymetrix U133A GeneChip (2003) temperature system 9 h at room temperature to froze at Ϫ20 °C for 6 days Radich et al. (6) 15 Preservative-free None EL Lysis buffer/RNEasy Agilent Hu25K microarray (2004) heparin tubes Debey et al. (12) 29 Sodium citrate Trizol at Ϫ80 °C Trizol (PBMC fraction) Affymetrix U133A GeneChip (2004) Vacutainer Up to 24 h at room PAXgene blood RNA isolation temperature system (whole blood) Up to 24 h at room QiaAmp (whole blood) temperature Vu et al. (13) 4 PAXgene Vacutainer Up to 10 days at PAXgene blood RNA isolation TaqMan RT-qPCR (2004) room temperature system or 4 °C Atlas glass trial cDNA microarray Wang et al. (11) 2 PAXgene Vacutainer 2 h at room PAXgene blood RNA isolation Affymetrix U133A GeneChip (2004) temperature system 4 h at room temperature Eady et al. (7) 18 Sodium citrate None EL Lysis buffer/RNeasy Operon human 14K microarray (2005) Vacutainer (present study) 10 PAXgene Vacutainer Ͻ4 h at room PAXgene blood RNA isolation Affymetrix Human Genome Focus temperature for system and Hu133 ϩ 2.0 GeneChips 98 days at Ϫ20 °C ZR buffer Ͻ4 h at room ZR whole-blood total RNA temperature system R/T ϭ Room Temperature. ated with donor sex and age, the time of day the sample 2 Affymetrix GeneChips were compared, as was biologi- was taken, and the proportions of blood cell subsets. cal variation based on sex. Interindividual variation has been demonstrated, but significant variation in repeated sampling of the same indi- Materials and Methods viduals was not reported (6, 7). Other studies (8–16) The research was carried out in accordance with the examined how GEP of blood samples is affected by principles of the current version of the Helsinki Declara- storage conditions and different RNA extraction and tion. After we obtained protocol approval from the Uni- amplification techniques and have reported that both versity of North Carolina at Chapel Hill Biomedical these factors can affect GEP. Institutional Review Board, we obtained blood samples In the current study, sources of technical and biological from 42 fully informed, consenting adult volunteers, 21 variation were characterized with respect to microarray- men (mean age, 32 years; range, 20–40 years) and 21 based GEP of human whole blood. Specifically, the effects women (mean age, 25 years; range, 19–40 years). All of storage of whole blood on overall RNA quality and samples were collected at the same location, a clinical GEPs were examined by comparison with freshly pre- office at the US Environmental Protection Agency facility pared RNA. In addition, 2 RNA isolation techniques and in Chapel Hill, NC. 1040 Kim et al.: Gene Expression in Human Whole Blood

sample collection, transportation, and Software). Mean (SE) values for RIN and 28S:18S rRNA storage ratio were calculated, and a mixed-effects linear model Blood was collected and processed according to the was used to look for differences among all storage tem- manufacturers’ instructions in PAXgene Vacutainer tubes perature by day groups. The model was a one-way (PreAnalytix/Qiagen; 2.5 mL blood per tube) and 1.5-mL ANOVA to which was added a random predictor for each Eppendorf tubes containing 0.6 mL ZR buffer (Zymo study participant. This adjusted the covariance matrix for Research; 0.2 mL blood per tube). Two tubes of each type having 2 observations from each study participant. Pair- were collected from each study participant. The PAXgene wise t-tests were calculated between each temperature by tubes were stored at ambient temperature, and were day group and the fresh sample group. An additional transported within4hofcollection (also at ambient analysis was run for data for all time points pooled within temperature) from the clinical office to the analytical each storage temperature group. laboratory, a journey time of ϳ30 min. ZR tubes were stored on ice and within4hofcollection transported (also probe preparation and hybridization on ice) from the clinical office to the analytical laboratory. Probe preparation and GeneChip hybridization were con- On reaching the laboratory the tubes were either imme- ducted by Expression Analysis. Each RNA sample se- diately processed to extract RNA or else stored at Ϫ20 °C, lected for microarray hybridization (see Table 1 in the 4 °C, or room temperature. Data Supplement that accompanies the online version of this article at http://www.clinchem.org/content/vol53/ rna isolation and quality analysis issue6 for details of specific samples used in microarray RNA was isolated with the PAXgene Blood RNA Isolation hybridizations) was amplified with the MessageAmp Kit or the ZR Whole-Blood Total RNA Kit according to the aRNA reagent set (Ambion), according to manufacturer’s manufacturers’ instructions. The RNA from the PAXgene instructions. The amplified RNA was used to synthesize tubes was extracted on the day of collection, after 1–2 cDNA. In vitro transcription was performed to produce days of storage at room temperature, after 1–40 days of biotin-labeled cRNA according to the manufacturer’s in- storage at 4 °C, or after 98–194 days of storage at Ϫ20 °C. structions. Biotin-labeled cRNA was purified with an The RNA from the ZR tubes was extracted either imme- RNeasy reagent set (Qiagen). Labeled cRNA was frag- diately or after 1–4 days of storage at 4 °C. All extracted mented and hybridized overnight to Affymetrix Human RNA samples were quantified by absorbance readings at Genome Focus (HG-Focus; 8400 genes) or Human Ge- 260 and 280 nm performed with a GeneQuant spectro- nome U133 Plus 2.0 (HG-U133 Plus 2.0; 39 000 genes) photometer (Pharmacia Biotech) and then stored at GeneChips (Affymetrix) in an Affymetrix Fluidics Station Ϫ80 °C. 400. After being washed, the arrays were stained with PCR analysis of each RNA sample was conducted to phycoerythrin-conjugated streptavidin, amplified by bio- ensure absence of contaminating genomic DNA. The PCR tinylated antistreptavidin, and then scanned in a GCS3000 contained 12.5 ␮L2ϫ PCR Master Mix (Promega), 1 ␮L (Affymetrix). QC metrics for probe preparation and hy- RNA template (30 ng), and 1 ␮L human ␤-actin primers bridization are included in Tables 2 and 3 in the online (10 ␮mol/L; Promega) in a final 25-␮L reaction volume. Data Supplement. Cycling was as follows: 2 min at 94 °C as an initial denaturation step; 40 cycles of 94 °C for 30 s, 65 °C for 1 microarray data analysis min, and 68 °C for 2 min; and then a final extension step Affymetrix Microarray Suite 5.0 (MAS 5.0) was used for at 68 °C for 7 min. PCRs were run on a 2% agarose gel to calculation of the signal and determination of the identify those samples containing genomic DNA. All “present” call. These data will be deposited in the Na- samples were treated with DNase (DNA-free reagent set; tional Center for Biotechnology Information’s Gene Ex- Ambion) for 30 min at 37 °C followed by a clean-up step, pression Omnibus database (http://www.ncbi.nlm.nih. according to the manufacturer’s instructions. After DNase gov/projects/geo/). The data were transferred to treatment, RNA was recovered by ethanol precipitation: GeneSpring 6.1 (Silicon Genetics) for further analysis. By 1/10th volume of sodium acetate (pH 5.2) and 2 volumes use of normalization options in GeneSpring, the raw data of 100% ethanol were added to the sample, and the were globally normalized to the median of each array, mixture incubated for1hatϪ20 °C. RNA was then and each gene to its median. For comparison between recovered by centrifugation (20 817g for 30 min at 4 °C). HG-focus and HG-U133Plus2.0 genechips, only the data The RNA pellet was washed with 75% ethanol and for genes common to both chips were transferred to resuspended in 20 ␮L RNase-free water. GeneSpring. We analyzed 1 ␮L of each RNA sample with an RNA 6000 Nano Lapchip reagent set (Agilent Technologies) effect of extraction technique, storage, donor with a 2100 Bioanalyzer (Agilent Technologies) according sex, and genechip on gep to the manufacturer’s instructions. RNA integrity was Five sample RNAs from male donors were isolated with determined by 28S:18S ribosomal RNA ratio and RNA each of the 2 extraction techniques and used to determine integrity number (RIN; Agilent 2100 RIN Beta Version the effect of extraction technique on GEPs. Samples from Clinical Chemistry 53, No. 6, 2007 1041

5 female donors were used to determine the effect of ribosomal bands of intact RNA extracted from most long-term frozen storage. To determine the effect of donor biological samples should be ϳ2.0. The 28S:18S ribosomal sex, expression profiles of the RNA freshly isolated from band ratios for 80 RNA samples (4 volunteers could not PAXgene samples from the 5 male and 5 female individ- provide a 2nd tube of blood) were Ն2.0 for 40 (50%) of uals were compared. To determine whether there was any samples, 1.8–2.0 for 14 (17.5%), and Ͻ1.8 for 26 (32.5%; see genechip effect, RNA from 3 blood samples that had been Table 1 in the online Data Supplement). The mean 28S:18S isolated within 24 h of collection was hybridized to both rRNA ratios were 2.54 for the freshly isolated samples, HG-focus and HG-U133 Plus 2.0 genechips. Group com- 2.23 for the samples stored at 4 °C (all storage times), and parisons were performed in GeneSpring, using two-way 1.49 for the samples stored at Ϫ20 °C (all storage times). ANOVA to test for effects of storage, extraction method, The mean (range) total RNA yields from each ZR tube or genechip while adjusting for any donor effect. A (0.2 mL blood) were 1.0 (0.3–3.4) ␮g (see Table 4 in the one-way ANOVA was used to test for effects of donor sex. online Data Supplement). The 28S:18S ribosomal band The ANOVA options selected in GeneSpring were para- ratios were Ն2.0 for 1 (4.5%) of the 22 samples, 1.8–2.0 for metric test, not assuming equal variances, and a false 1 (4.5%), and Ͻ1.8 for 20 (91%). The mean 28S:18S rRNA discovery rate of 0.05. Genes with Affymetrix present calls ratios were 1.52 for samples (n ϭ 15) extracted up to a day in at least 5 of 10 arrays (for storage, extraction, and donor after blood collection, and 1.47 for samples (n ϭ 7) sex analyses) or 3 of 6 arrays (for chip analyses), and with extracted 2–4 days after blood collection [1.29 if sample P Ͻ0.01 for the effect of interest were selected as differ- 10M(1) was treated as an outlier]. Statistical analysis entially expressed genes. comparing all samples showed no significant differences We performed an additional ANOVA analysis using in quality (as determined by ribosomal ratio) between the SAS software (SAS Institute). MAS5 values were log (base freshly isolated RNA (0–1 days) and RNA from stored 2) transformed, and a global normalization was used, samples (2–4 days). However, if sample 10M(1), which normalizing each of the 23 unique data files to its median. appeared to be an outlier, was removed from the 2–4 day A total of 4299 genes were identified as being present in at group, the mean 28:18S ratio of this group was signifi- least 5 of 10 female chips, 5 of 10 male chips, or 3 of 6 of cantly lower than the mean of the 0–1 day group (P the HG-focus and HG-U133Plus2.0 genechips. For each Ͻ0.05). We used the Agilent Bioanalyzer to assess RNA gene, an ANOVA was performed with SAS Proc GLM, for integrity (17). This system uses an RIN scale on which an which normalized log intensity was predicted by sex, RIN of 1 is the most degraded and 10 is the most intact. donor (nested within sex), genechip, extraction, and stor- The RIN was calculated for all PAXgene samples. The age. The mean variance due to each factor was examined statistical analysis of RNA quality by temperature and by comparing the mean square values. This analysis was storage days is summarized in Table 2. Results indicated repeated with quantile rather than median normalization. that RNA in samples in PAXgene tubes were stable for up The “Find the similar samples” function in GeneSpring to 1 day of storage at room temperature and up to 4 days calculates pairwise correlation coefficients between a tar- at 4 °C and moderately stable for up to 194 days at get sample and each member of a list of other samples. Ϫ20 °C. This function was used to calculate correlation coefficients between all pairs of arrays. Means of these coefficients effect of rna extraction technique, storage, were calculated to find mean inter- and intragroup corre- donor sex, and genechip on expression profile lations. Principal component analyses were performed in Group comparisons of 8500 genes identified 264 genes Genespring on those genes that were present in at least 5 differentially expressed between RNAs extracted by the 2 of 10 arrays (PAXgene vs ZR, fresh vs stored, male vs female) or at least 3 of 6 arrays (HG-Focus vs HG-U133 Table 2. Statistical analysis of RNA quality by temperature Plus 2.0). and storage days. RIN blood transcriptome The blood transcriptome was characterized with data Temperature Days n Mean SE obtained from the HG-Focus genechips after hybridiza- Not applicable 0 21 8.93 0.16 tion with RNA from the 10 samples (5 male, 5 female) that Room temperature 1 2 9.00 0.58 a had been isolated by use of PAXgene within 24 h of blood Room temperature 2 2 7.96 0.45 collection. To identify genes expressed across all samples, 4 °C 1 7 8.81 0.30 the FLAGS filter of GeneSpring was used. 4 °C 3 6 8.95 0.34 4 °C 4 4 9.28 0.41 a Results 4 °C 7 7 7.90 0.30 4 °C 8 4 7.95a 0.41 effect of storage on rna quality 4 °C 40 2 6.95a 0.49 The mean (range) total RNA yields from each PAXgene Ϫ20 °C 98–194 19 8.14 0.20 tube (2.5 mL blood) was 5.40 (0.01–21.90) ␮g (see Table 1 a Ͻ in the online Data Supplement). The ratio of 28S:18S Significant reduction in RNA integrity (P 0.05). 1042 Kim et al.: Gene Expression in Human Whole Blood

Table 3. Genes with >6-fold differential expression between sexes. Gene symbol Gene namea Fold change Gene ontologyb SGKL Serum/glucocorticoid regulated kinase-like 33.2 Signal transduction XIST X (inactive)-specific transcript 31.3 X inactivation CPNE3 Copine III 12.8 Lipid metabolism ECGF1 Endothelial cell growth factor 1 (platelet-derived) 12.2 Growth factor STX16 Syntaxin 16 9.6 Protein transport DICER1 Dicer1, Dcr-1 homolog (Drosophila) 8.2 RNA processing CASP8AP2 CASP8 associated protein 2 7.8 Signal transduction SAS10 Disrupter of silencing 10 7.5 (Not yet classified) ADD3 Adducin 3 (gamma) 6 Cytoskeleton EIF1AY Eukaryotic translation initiation factor 1A, Y-linked Ϫ14.1 Translation initiation RPS4Y1 Ribosomal protein S4, Y-linked 1 Ϫ31.2 Protein biosynthesis DDX3Y DEAD (Asp-Glu-Ala-Asp) box polypeptide 3, Y-linked Ϫ208.3 ATP binding a Genes for which P Ͻ0.05 by t-test, present calls in at least 5 of 10 arrays, and Ն6-fold difference between sexes. b NetAffx Gene Ontology mining tool used for analysis (http://www.affymetrix.com/analysis/query/go_analysis.affx).

different methods. A total of 67 genes were identified as partition of variance and the number of genes that were being differentially expressed in the group of frozen significantly different between groups indicated stor- samples compared with the group of their respective age Ͻ sample donor Ͻ genechip Ͻ extraction Ͻ donor freshly extracted samples. Group comparisons identified sex. 748 genes differentially expressed between sexes; 600 of these showed higher expression in samples from female correlation analysis donors than in samples from male doors, whereas 148 Inter- and intragroup correlation coefficients (R values) showed lower expression in samples from females vs were calculated to determine the overall congruency males. To identify which genes with differential expres- among the samples in each of the tested groups. Correla- sion related to sample-donor sex were likely to have a tion analysis showed variation due to storage Ͻ gene- sex-specific role, we selected transcripts with a mean chip Ͻ extraction technique Ͻ donor sex. Intergroup expression difference of 6-fold or greater between the variability was greater than intragroup variability except male and female sample groups, a total of 12 genes, of for the frozen group, demonstrating more effects than which 9 had higher expression and 3 had lower expres- interindividual differences. Mean values for sex were 0.74 sion in samples from female than from samples in male between groups, and 0.838 (male) and 0.887 (female) 5 donors (Table 3). Biotin-labeled amplification products within groups. For extraction, the mean was 0.816 be- from 3 samples (36F, 37F, and 38F) were each split and tween groups, and 0.838 (PAX) and 0.870 (ZR) within hybridized to both the HG-Focus chip and the HG-U133 groups. For storage, the mean was 0.867 intergroup, and Plus 2.0 chip. Group comparisons identified 435 genes 0.887 (fresh) and 0.835 (frozen) intragroup. And finally, differentially expressed between the chips. for genechip, the mean was 0.844 intergroup, and 0.903 An additional ANOVA analysis, using median normal- (HG-Focus) and 0.916 (HG-U133 Plus 2.0) intragroup. ization, showed the mean square due to storage (mean, 0.256; median, 0.084) was the smallest, and donor sex cluster analysis (mean, 1.378; median, 0.391) the largest. Genechip (mean, In unsupervised principal component analyses, GEPs 0.751; median, 0.190), extraction (mean, 0.684; median, from the paired fresh and frozen samples showed parti- 0.191), and sample donor (mean, 0.331; median, 0.194) had tioning in part according to individual sample donor, similar median mean square values, but could be sepa- although the results were mixed (Fig. 1A). GEP partition- Ͻ rated according to their mean squares as sample donor ing for the paired PAXgene and ZR samples occurred Ͻ extraction genechip. We also performed ANOVA anal- sharply according to method rather than individual sam- ysis after quantile normalization, for which both the ple donor (Fig. 1B); male and female donor profiles also partitioned sharply from one another (Fig. 1C), as did profiles generated on the different genechips (Fig. 1D).

5 Human genes: SGKL, serum/glucocorticoid regulated kinase family, member 3; XIST, X (inactive)-specific transcript; CPNE3, copine III; ECGF1, blood transcriptome endothelial cell growth factor 1 (platelet-derived); STX16, syntaxin 16; DICER1, Results for gene expression were 2558) genes expressed in Dicer1, Dcr-1 homolog (Drosophila); CASP8AP2, CASP8-associated protein 2; all 10 samples, 2629 genes expressed in all 5 samples from SAS10, disrupter of silencing 10; ADD3, adducin 3 (␥); EIF1AY, eukaryotic translation initiation factor 1A, Y-linked; RPS4Y1, ribosomal protein S4, male donors, 3343 expressed in all 5 samples from female Y-linked 1; DDX3Y, DEAD (Asp-Glu-Ala-Asp) box polypeptide 3, Y-linked. donors, and 4013 unique genes (from a total of 8500 Clinical Chemistry 53, No. 6, 2007 1043

Fig. 1. Principal component analysis. (A), effects of long-term storage. A 1st F denotes sample from a female donor. S denotes RNA isolated after long-term storage. A 2nd F denotes RNA isolated within 1 day (“fresh”). Principal component analysis used the 3996 genes that had present calls in at least 5 of the 10 microarrays. (B), comparison of 2 extraction techniques, PAX and ZR. Principal component analysis used 3473 genes that had present calls in at least 5 of the 10 microarrays. (C), effect of donor sex. Principal component analysis used 3789 genes that had present calls in at least 5 of the 10 microarrays. (D), comparison of Affymetrix GeneChips. Principal component analysis used 4086 genes that had present calls in at least 3 of the 6 microarrays, and consensus genes between HG-Focus and HG-U133 Plus 2.0 arrays. unique probe sets ϭ 8400 unique Entrez IDs) expressed in samples eventually leads to decreased RNA quality, as at least 1 of the 10 individual samples hybridized to the indicated by decreased 28S:18S rRNA ratio and RINs. The HG-Focus Arrays. recommendation is that storage of whole blood in PAX- gene tubes for microarray studies should not exceed 1 day Discussion at room temperature, 4 days at 4 °C, and probably 3 The current study demonstrates that an extended storage months at Ϫ20 °C. In contrast, previous reverse transcrip- period before RNA extraction does not significantly affect tion PCR–based studies have demonstrated that blood the yield and/or quality of RNA extracted from whole can be stored in PAXgene tubes at ambient temperature blood samples. Not surprisingly, however, RNA quality for at least 7 days without significant effects on gene does decrease over time, the rate being dependent on the expression (18). These results are not surprising; it is storage system and temperature. The data generated generally accepted that microarray-based profiling re- herein suggest that blood collected into ZR buffer should quires higher quality RNA than reverse transcription be processed as soon as possible, preferably within 24 h. PCR–based methods. Integrity of RNA is maintained more effectively in blood Even without incorporating a globin decrease protocol stored in PAXgene tubes. Nevertheless, even in PAXgene (19), a present call level in the 36%–48% range was tubes longer-term storage and freeze/thawing of blood obtained from both PAXgene and ZR samples after gene- 1044 Kim et al.: Gene Expression in Human Whole Blood

chip hybridization, a result that is close to previously human blood under “normal” conditions is well charac- reported values (12). RNA amplification was needed for terized will it be possible to identify robust genomic the limited amount of blood RNA obtained from both biomarkers of environmental exposure or disease extraction methods, especially from the ZR method, for progression. which the small volume of blood used (0.2 mL) generally yielded Ͻ1 ␮g of total RNA. Thus, 2 rounds of RNA amplification were required. Amplified RNA probes may Grant/funding support: S.J.K. was supported by the Envi- undergo truncation during reverse transcription and T7- ronmental Protection Agency/University of North Carolina mediated transcription, resulting in short-labeled prod- Cooperative Training Agreement T829472, and R.N.M. was ucts (20). This effect can be detected in the GAPDH 3Ј:5Ј supported by Environmental Protection Agency/North signal ratio number (21), and GAPDH ratios of the female Carolina State University Cooperative Training Agreement samples in this study were near the upper end of this CT826512010. range after 2 rounds of amplification. Ratios for the male Financial disclosures: None declared. samples were higher, however, indicating that the 2 Acknowledgements: We thank Maryann Bassett, Debbie Le- rounds of amplification produced truncated transcripts. vine, and Tracey Mantilla (US Environmental Protection The percentage present calls for the male samples were Agency) for assistance in sample collection. We thank Dr. lower than for the female samples, a result at least Don Graff (US Environmental Protection Agency) for criti- partially attributable to truncation of transcripts in the 2 cally reviewing this work before submission. The informa- rounds of amplification. The reasons underlying this tion in this document has been funded wholly by the US sex-based difference remain to be determined. Environmental Protection Agency. It has been subjected to In terms of numbers of genes for which expression was review by the National Health and Environmental Effects significantly different between groups, frozen storage Research Laboratory and approved for publication. Ap- produced the smallest effect, and donor sex produced the proval does not signify that the contents reflect the views of greatest effect. When median normalization was used, the the Agency, nor does mention of trade names or commercial partition of variance indicated that variance due to stor- products constitute endorsement or recommendation for use. age Ͻ extraction Ͻ genechip Ͻ donor sex. When quantile normalization was used rather than normalization of each References chip to the median, both the partition of variance and the 1. Rockett JC, Burczynski ME, Fornace AJ, Herrmann PC, Krawetz SA, number of genes that were significantly different between Dix DJ. Surrogate tissue analysis: monitoring toxicant exposure groups indicated storage Ͻ genechip Ͻ extraction Ͻ and health status of inaccessible tissues through the analysis of accessible tissues and cells. Toxicol Appl Pharmacol 2004;194: donor sex. This result agreed with the correlation analysis 189–99. and can be explained by different distribution of values 2. Tang Y, Lu A, Aronow BJ, Sharp FR. Blood genomic responses for the 3 pairs of arrays with different chips for the same differ after stroke, seizures, hypoglycemia, and hypoxia: blood samples. With quantile normalization, all 6 chips are genomic fingerprints of disease. Ann Neurol 2001;50:699–707. forced to have the same distribution, so the high values in 3. Rockett JC, Kavlock RJ, Lambright C, Parks LG, Schmid JE, Wilson the HG-U133 chips are brought in line with the high VS, et al. DNA arrays to monitor gene expression in rat blood and ␤ values in the HG-U133 Plus 2.0 chips, decreasing the uterus following 17- -estradiol exposure: biomonitoring environmental effects using surrogate tissues. Toxicol Sci 2002;69:49– number of significant changes between the 2 while pre- 59. serving the correlation. Not surprisingly, several of the 4. Twine NC, Stover JA, Marshall B, Dukart G, Hidalgo M, Stadler W, genes differently expressed between sexes were sex chro- et al. Disease-associated expression profiles in peripheral blood mosome–linked: XIST, EIF1AY, RPS4Y1, and DDX3Y. The mononuclear cells from patients with advanced renal cell carci- present study confirmed previous findings that biological noma. Cancer Res 2003;63:6069–75. variation in gene expression exceeds variation produced 5. Whitney AR, Diehn M, Popper SJ, Alizadeh AA, Boldrick JC, Relman by technical factors. DA, et al. Individuality and variation in gene expression patterns in The results of this study demonstrate that microarray- human blood. Proc Natl Acad Sci U S A 2003;100:1896–901. 6. Radich JP, Mao M, Stepaniants S, Biery M, Castle J, Ward T, et al. based transcript profiling in clinical studies will require Individual-specific variation of gene expression in peripheral blood standardized protocols for collection, transportation, and leukocytes. Genomics 2004;83:980–8. storage of biospecimens; RNA extraction and amplifica- 7. Eady JJ, Wortley GM, Wormstone YM, Hughes JC, Astley SB, Foxall tion; microarray probe labeling and hybridization; and RJ, et al. Variation in gene expression profiles of peripheral blood statistical analysis of results. Without this standardiza- mononuclear cells from healthy volunteers. Physiol Genomics tion, results will not be reproducible or comparable. 2005;22:402–11. Standardization will afford better understanding of how 8. Rainen L, Oelmueller U, Jurgensen S, Wyrich R, Ballas C, Schram J, et al. Stabilization of mRNA expression in whole blood samples. technical and biological factors affect blood GEP, and Clin Chem 2002;48:1883–90. comparisons of standardized studies will generate the 9. Tanner MA, Berk LS, Felten DL, Blidy AD, Bit SL, Ruff DW. power to determine significant sources of variation of Substantial changes in gene expression level due to the storage gene expression in blood samples. Only when robust temperature and storage duration of human whole blood. Clin Lab standardized procedures are used and gene expression in Haematol 2002;24:337–41. Clinical Chemistry 53, No. 6, 2007 1045

10. Thach DC, Lin B, Walter E, Kruzelock R, Rowley RK, Tibbetts C, et target preparation methods for hybridization to high-density oligo- al. Assessment of two methods for handling blood in collection nucleotide microarrays. BMC Genomics 2004;5:2. tubes with RNA stabilizing agent for surveillance of gene expres- 16. Li L, Roden J, Shapiro BE, Wold BJ, Bhatia S, Forman SJ, et al. sion profiles with high density microarrays. J Immunol Methods Reproducibility, fidelity, and discriminant validity of mRNA ampli- 2003;283:269–79. fication for microarray analysis from primary hematopoietic cells. 11. Wang J, Robinson JF, Khan HM, Carter DE, McKinney J, Miskie BA, J Mol Diagn 2005;7:48–56. et al. Optimizing RNA extraction yield from whole blood for 17. Mueller O, Lightfoot S, Schroeder A. RNA integrity number (RIN): microarray gene expression analysis. Clin Biochem 2004;37: standardization of RNA quality control. Agilent Technologies 741–4. Application Note, 2004. http://www.chem.agilent.com/temp/ 12. Debey S, Schoenbeck U, Hellmich M, Gathof BS, Pillai R, Zander radA423C/00047692.pdf (accessed December 7, 2005). T, et al. Comparison of different isolation techniques prior gene 18. Pahl A, Brune K. Gene expression changes in blood after phlebot- expression profiling of blood derived cells: impact on physiological omy: implications for gene expression profiling. Blood 2002;100: responses, on overall expression and the role of different cell 1094–5. types. Pharmacogenomics J 2004;4:193–207. 19. Affymetrix. Globin reduction protocol: a method for processing 13. Vu NT, Zhu H, Owuor ED, Huggins ME, White VL, Charturvedi AK, for whole blood RNA samples for improved array results. et al. Isolation of RNA from peripheral blood cells: a validation http://www.affymetrix.com/support/technical/technotes/blood2_ study for molecular diagnostics by microarray and kinetic RT-PCR technote.pdf (accessed February 2, 2006). assays: application in aerospace medicine. Office of Aerospace 20. Dumur CI, Garrett CT, Archer KJ, Nasim S, Wilkinson DS, Ferreira- Medicine, US Department of Transportation, Federal Aviation Gonzalez A. Evaluation of a linear amplification method for small Administration Report No. DOT/FAA/AM-04/1. 2004. samples used on high-density oligonucleotide microarray analy- 14. Baugh LR, Hill AA, Brown EL, Hunter CP. Quantitative analysis of sis. Anal Biochem 2004;331:314–21. mRNA amplification by in vitro transcription. Nucleic Acids Res 21. Affymetrix. What 3Ј/5Ј ratio for control genes, for example GAPDH 2001;29:E29. and Actin, should I anticipate to obtain on GeneChip probe arrays? 15. Gold D, Coombes K, Medhane D, Ramaswamy A, Ju Z, Strong L, http://www.affymetrix.com/support/help/faqs/ge_assays/faq_17. et al. A comparative analysis of data generated using two different jsp (accessed February 2, 2006). Clinical Chemistry 53:6 1046–1052 (2007) Molecular Diagnostics and Genetics

Quantitative Mitochondrial DNA Mutation Analysis by Denaturing HPLC

Kok Seong Lim,1* Robert K. Naviaux,2,3 and Richard H. Haas1,3

Background: In recent years, denaturing HPLC (DHPLC) been developed to study mutations include denaturing has been widely used to screen the whole mitochondrial HPLC (DHPLC) (1–6), single-strand conformation poly- genome or specific regions of the genome for DNA morphism (7–9), denaturing gradient gel electrophoresis mutations. The quantification and mathematical model- (10–12), temperature gradient gel electrophoresis (13), ing of DHPLC results is, however, underexplored. temporal temperature gradient gel electrophoresis (14– Methods: We generated site-directed mutants contain- 16), and pyrosequencing (2, 17). A mutation detection ing some common mutations in the mitochondrial DNA method recently used in the identification of hetero- (mtDNA) tRNA(leu) region with different mutation plasmy relies on the use of Surveyor nuclease, which loads and used PCR to amplify the gene segment of cleaves DNA at sites of base-substitution mismatch and interest in these mutants. We then performed restriction short insertion/deletion (18, 19). Highly sensitive but digestion followed by slow reannealing to induce het- time-consuming methods include ligation-mediated PCR eroduplex formation and analyzed the samples by use (20), real-time fluorescent PCR (21, 22), and peptide nu- of DHPLC. cleic acid binding assays (23). A recently developed Results: We observed a quadratic relationship between method is based on reversed-phase HPLC coupled to elec- the heteroduplex peak areas and mutant loads, consis- trospray ionization time-of-flight mass spectrometry (24). tent with the kinetics of heteroduplex formation re- DHPLC has become a popular technology in the study ported by others. This was modeled mathematically and of mutation detection. It relies on the use of a hydropho- used to quantify mtDNA mutation load. The method bic column based on reversed-phase liquid chromatogra- was able to detect a mutation present in a concentration phy for the separation of heteroduplex and homoduplex as low as 1% and gave reproducible measurements of DNA at specific optimized temperatures and has recently the mutations in the range of 2.5%–97.5%. been used by several research groups to study mitochon- Conclusion: The quantitative DHPLC assay is well drial DNA (mtDNA) mutation (1–6). In these studies, suited for simultaneous detection and quantification of DHPLC has been used either to screen the whole mito- DNA mutations. © 2007 American Association for Clinical Chemistry chondrial genome (1, 2, 4–6) or to study specific regions of the genome for mutation (3, 4, 25), followed by DNA Most methods for detecting point mutations are either sequencing to identify the mutation sites. Several of these simple with low sensitivity or laborious and overly sen- studies have also measured the percentage of hetero- sitive. Relatively simple, inexpensive, and conventional plasmy in the mtDNA by use of PCR-RFLP-PAGE (4) and methods include PCR-restriction fragment length poly- pyrosequencing (2). morphism (RFLP)4 analysis and PCR/allele-specific oli- The use of DHPLC in quantitative analysis of DNA gonucleotide dot blot analysis. Radiolabeled PCR-RFLP is mutation, however, has not been popular. Earlier studies more sensitive but radioactive. Other methods that have on heteroduplex analysis suggested that the kinetics of heteroduplex formation would fit a quadratic model (26, 27). Using DHPLC technology, we investigated the Departments of 1 Neurosciences, 2 Medicine, and 3 Pediatrics, School of relationship between heteroduplex peak area (which is Medicine, University of California San Diego, La Jolla, CA. more commonly used in quantitative chromatographic *Address correspondence to this author at: University of California San Diego School of Medicine, 9500 Gilman Dr. B112, La Jolla, CA 92093-0935. Fax analysis) and level of heteroplasmy, finding a parabolic 619-543-7868; e-mail [email protected]. rather than a linear relationship between peak area and Received November 17, 2006; accepted March 27, 2007. mutant load. We developed a method to accurately quan- Previously published online at DOI: 10.1373/clinchem.2006.083303 4 Nonstandard abbreviations: RFLP, restriction fragment length polymor- tify the mutant load in mtDNA; using this quantitative phism; DHPLC, denaturing HPLC; mtDNA, mitochondrial DNA. DHPLC analysis method, we measured the levels of

1046 Clinical Chemistry 53, No. 6, 2007 1047

heteroplasmy in several DNA samples with known mu- CCAACACAGG-3Ј (2415–2436 bp) and 5Ј-TGTGTTGT- tation loads. GATAAGGGTGGAGAG-3Ј (3790–3812 bp) for detection of A3243G, C3256T, A3260G, T3271C, T3291C, and Materials and Methods T3308C. These primers can be obtained from the Mito- cell culture Screen reagent set (Transgenomic) and have been used in We cultured 143B osteosarcoma rho-0 cells (gift from Dr. a published study (1). PCRs were performed in 70 ␮L Keshav Singh of Roswell Park Cancer Institute, Buffalo, Optimase reaction buffer (Transgenomic) containing 50 NY) in DMEM with 4.5 g/L d-glucose and l-glutamine ng plasmid DNA or 150 ng total DNA, 200 ␮mol/L of (Invitrogen) containing 10% (vol/vol) fetal bovine serum, each dNTP (Transgenomic), 21 pmol of the forward and 1% (vol/vol) penicillin/streptomycin (Invitrogen), 200 reverse primer, and 3.5 units Optimase DNA polymerase ␮ mol/L uridine, and 1 mmol/L pyruvate (Sigma-Al- (Transgenomic). We performed PCR by use of the iCycler drich) in 5% CO2 at 37 °C. We maintained the cells in Thermal Cycler (Bio-Rad). The conditions for PCR were as logarithmic growth phase by routine passage every 1–2 follows: 95 °C for 2 min; 35 cycles of 95 °C for 30 s, 56 °C days. Uridine and pyruvate added to the media were for 30 s, and 72 °C for 3 min; and a final extension step of always from fresh stock solution. 72 °C for 5 min.

dna extraction restriction enzyme digestion and heteroduplex We used a blood DNA sample from a healthy individual formation as the template for the generation of site-directed mutants PCR products generated were digested with 1 unit of DdeI and 143B osteosarcoma rho-0 cells as the source of pure (New England Biolabs) at 37 °C for 6 h. Digested PCR nuclear DNA. In 1 experiment involving the investigation products were denatured at 95 °C for 5 min and then of the effect of DNA amount on heteroplasmy, we used slowly cooled to 25 °C at a rate of 1 °C/min to induce patient blood DNA; the sample was obtained under a heteroduplex formation before DHPLC analysis. These University of California San Diego Institutional Review conditions produce maximal heteroduplex formation— Board–approved protocol, and informed consent was ϳ50% heteroduplex (see Results) in samples containing obtained from the patient. We extracted total DNA from 50% heteroplasmy, indicating that the slow reannealing whole blood samples and nuclear DNA from rho-0 cells procedure is optimal. We confirmed stability of these with the Puregene DNA Purification reagent set (Gentra complexes by comparison of samples injected immedi- Systems) according to the manufacturer’s instructions. ately after reannealing and 2 days later, finding no change Briefly, we collected the leukocytes from whole blood by in heteroduplex peak areas (data not shown). centrifugation after removal of erythrocytes by mild lysis, whereas we collected rho-0 cells at 80%–90% confluence. dhplc analysis The cells were lysed and incubated with RNase A, and We analyzed samples by use of WAVE 3500 equipped protein was removed by high-salt precipitation. DNA was with a Hitachi D-7000 Interface, L-7100 Pump, L-7200 precipitated by use of isopropanol. Autosampler, L-7300 Oven, and L-7400 UV Detector (Transgenomic). We separated nucleic acids by use of a site-directed mutagenesis DNASep cartridge (4.6 mm ϫ 50 mm; Transgenomic). We All mutant clones used in this study have the genome predicted temperatures for sample analysis by use of segment containing mitochondrial tRNA Leu (UUR; Navigator software (version 1.6.2) (28) (Transgenomic) 2415–3812 bp), which is inserted into pCRII (Invitrogen). and confirmed them experimentally to be 59 °C. Alterna- We generated point mutations by use of the QuikChange tively, prediction of melting temperatures could be per- XL Site-Directed Mutagenesis reagent set (Stratagene) formed by use of the DHPLC Melt program created by according to the manufacturer’s instructions. Forward Stanford Genome Technology (29). The gradient mobile and reverse primers used to create point mutations are phase consists of buffer A [0.1 mol/L triethylammonium listed in Table 1 in the Data Supplement that accompanies acetate, pH 7 (Transgenomic)] and buffer B [0.1 mol/L the online version of this article at http://www.clinche- triethylammonium acetate, pH 7, and 25% acetonitrile m.org/content/vol53/issue6. We confirmed target muta- (Chromasolv; Sigma-Aldrich)]. Samples (5 ␮L) were in- Ј tions by sequencing with 5 -CTA CTT CAA ATT CCT jected for analysis. Fragments were eluted with a linear Ј Ј CCC TGT AC-3 (3104–3126 bp) and 5 -CATTAGGAAT- acetonitrile gradient of 2%/min from 45% to 67% buffer B Ј GCCATTGC-3 (3352–3369 bp). at a flow rate of 0.9 mL/min. UV detection was set at 260 nm. After each run, we washed the column with 75% pcr primers and conditions acetonitrile for 1 min and equilibrated it for 1.5 min before Primers (Invitrogen) used for the amplification of the the next sample injection. The peak areas were deter- region of interest in the mitochondrial genome (J01415.2) mined by use of Navigator software, and quadratic curve before DHPLC analysis were 5Ј-CTCACTGTCAAC- fitting was performed with Microsoft Excel. 1048 Lim et al.: Quantitative DHPLC Analysis of mtDNA Mutation

Results restriction enzyme digestion Incubation of the PCR products with restriction enzyme DdeI results in the generation of 5 fragments of different sizes: 442, 342, 278, 210, and 126 bp (1). The mutation sites of interest lie in the 342-bp fragment containing the tRNA (leu1) sequence. We studied the amount of DNA and incubation time required for complete restriction enzyme digestion of PCR products. Various amounts of the total and plasmid DNA were used separately in the PCR, and amplicons were analyzed by DHPLC at 50 °C after restriction enzyme digestion (see Fig. 1 in the online Data Supplement). PCR products generated from plasmid DNA were more abundant than those from total DNA, suggesting a difference in kinetics of DNA amplification between these 2 types of DNA templates. The concentrations of both types of PCR products began to reach steady state after use of ϳ25 ng (for plasmid DNA) and 50 ng (for total DNA; see Fig. 1 in the online Data Supplement). In the time course experiment, PCR products were digested by DdeI for 6, 12, and 24 h and analyzed by DHPLC at 50 °C. Neither an increase in restriction fragment peak area nor appearance of additional fragment peaks occurred, suggesting that the restriction enzyme digestion was complete (data not shown). This was also confirmed by gel electrophoresis of the digestion mixture, which showed the absence of undigested PCR products after6hofrestriction enzyme digestion (data not shown). detection of mutation by dhplc We experimentally confirmed the DHPLC temperature predicted by the Navigator software by analyzing the 50% heteroplasmic A3243G mutant samples at various temperatures (57–61 °C). We found the optimal temperature (the temperature that results in the best resolution of the heteroduplex and homoduplex peak) to be 59 °C (data not shown). We generated serial dilutions of the 100% mutant with 100% wild-type PCR products, resulting in samples with mutation present at levels of 0%, 1%, 2.5%, 5%, 10%, 25%, 40%, 50%, 60%, 75%, 90%, 95%, 97.5%, 99%, and 100%. We subjected these samples to reannealing to induce heteroduplex formation before DHPLC analysis. We determined peak areas of both heteroduplex and homoduplex peaks (Fig. 1A). The detection limit in DH- PLC analysis with a signal:noise ratio Ͼ1.3 for the mutants studied was 1%, whereas the quantification limit with a signal:noise ratio Ͼ2 was 2.5% (Fig. 1B). Optimase is a proofreading thermostable polymerase that has been shown to effectively minimize the misincor- Fig. 1. Detection of A3243G mutations by DHPLC in a series of poration of nucleotides during DNA amplification by samples with increasing mutant loads. (A), various amounts of PCR products generated from mutant and wild-type DNA PCR compared with other DNA polymerase, e.g., Ampli- were mixed to produce samples of different mutant loads ranging from 0% to Taq and Pfu polymerase (5, 30). However, we consis- 100%, and these samples were subjected to heteroduplex formation. DHPLC tently observed a small peak that precedes the homodu- analysis at 59 °C showed that the heteroduplex species (eluted earlier) resulting from the mutation at A3243G were well separated from the homoduplex species plex peak in all the 100% wild-type and mutant DNA (Fig. (eluted later). (B), enlarged image of the heteroduplex peaks in samples with 1B), a finding similar to what has been reported in the 2 mutant loads from 0% to 5% A3243G. The small peak preceding the homoduplex peak in the 0% mutant is labeled peak X. The detection limit in DHPLC analysis abovementioned studies. Such a low-level peak with a with a signal:noise ratio Ͼ1.3 for the A3243G mutants studied was 1%. consistent retention time (referred to as peak X in the 0% Clinical Chemistry 53, No. 6, 2007 1049

mutant in Fig. 1B) can affect the correct identification of determination of exact percentage of real mutations, especially in DNA with low levels of mutation mutation. We investigated whether peak X is affected by Given the parabolic relationship between the measured the amount of DNA used in the PCR. We observed no proportion of heteroduplex and levels of heteroplasmy, increase in the ratio of peak X area to total peak area, and only the vertex of the parabola at 50% heteroduplex has a this ratio remained constant at ϳ3% or less over the range single, unique solution for the level of heteroplasmy, all of DNA concentrations studied (see Fig. 2 in the online other proportions of heteroduplex formation having 2 Data Supplement). Further increase in the total DNA positive solutions by virtue of the quadratic function that amount to 600 ng did not increase the ratio measured (not governs DNA reannealing. For example, a measured shown). Such a background level may vary from column proportion of 30% heteroduplex would suggest 2 possible heteroplasmy levels at ϳ20% and 80% (Fig. 2). By use of to column (but the maximum we have observed is 3%), PCR products with known mutant loads (10%, 25%, 40%, and we always used a wild-type DNA (DNA from a 60%, 75%, and 90%), we determined the correct levels by healthy volunteer that has been confirmed by sequencing mixing these samples with equal concentrations of 100% to be 100% identical in the region studied to the published wild-type PCR products and subjected them to additional mitochondrial genome sequence J01415.2) in the DHPLC slow reannealing process before DHPLC analysis. Analy- analysis as a negative control. sis of the proportions of measured heteroduplex peak area and the level of heteroplasmy gave a quadratic plot that quantitative analysis of mutation by dhplc peaks at 50% when a 100% mutant DNA is mixed with We measured peak areas of both heteroduplex and ho- 100% wild-type DNA (see Fig. 4A in the online Data moduplex for samples containing different levels of Supplement). This quadratic equation allows confirma- A3243G mutation (Fig. 1A). Regression analysis of the tion of the heteroplasmy level in mtDNA. Such a mixing measured proportion of heteroduplex (ratio of heterodu- experiment produces samples that have a measured pro- plex peak area to total peak area, %) and the level of portion of heteroduplex that may be higher or lower than heteroplasmy (%) yielded a quadratic plot with a correla- the measured proportion before mixing, depending on tion coefficient of ϳ0.99 (Fig. 2). Such a relationship the level of heteroplasmy. appeared to be consistent with the kinetics of heteroduplex formation (27). To confirm this, we generated 5 other modeling of quantitative dhplc analysis site-directed mutants and performed the regression anal- Because all 6 mutants that we studied showed a similar ysis with different levels of mutations. The plots gener- quadratic relationship between the measured proportion ated on the basis of these mutants confirmed our finding of heteroduplex and level of heteroplasmy in plasmid with A3243G (see Fig. 3 in the online Data Supplement), DNA, and because an independent group also reported a and thus such a calibration curve could be used for the similar observation with genomic DNA (31), we specu- measurement of level of heteroplasmy. lated that there might be a universal equation that is applicable to most if not all types of quantitative heteroduplex analysis—hence quantitative DHPLC analysis. Theoretically, the proportion of a given heteroduplex or homoduplex can be calculated from the product of the stoichiometric fraction of each DNA species (27). Let m ϭ fraction of mutant DNA species; 1 Ϫ m ϭ fraction of wild-type DNA species; І after heteroduplex formation:

Proportion of mutant homoduplex ϭ m2 (Eq. 1)

Proportion of wild-type homoduplex ϭ ͑1 Ϫ m͒2 (Eq. 2)

Proportion of each type of heteroduplex ϭ m͑1 Ϫ m͒ (Eq. 3)

On the basis of Eqs. 1–3, we calculated the predicted proportion (in percentage) of the heteroduplexes formed after slow reannealing for DNA with mutant load of 0%, Fig. 2. Parabolic relationships between proportion of measured peak 1%, 2.5%, 5%, 10%, 25%, 40%, 50%, 60%, 75%, 90%, 95%, areas of heteroduplexes and levels of heteroplasmy. 97.5%, 99%, and 100% and plotted a graph on the basis of PCR products generated from mutant and wild-type plasmids were mixed in different proportions to give different levels of heteroplasmy, and the mixtures these predicted values (see Fig. 4B in the online Data were subjected to heteroduplex formation and DHPLC analysis. A parabolic Supplement). We observed a quadratic relationship (Eq. relationship with a correlation coefficient of ϳ0.99 was observed in the A3234G mutant. Each data point represents mean Ϯ SD for 3 separate samples. Graphs 4) between the predicted proportions of heteroduplex (%) for other mutants can be found in Fig. 3 in the online Data Supplement. and the level of heteroplasmy, consistent with our exper- 1050 Lim et al.: Quantitative DHPLC Analysis of mtDNA Mutation

imental data (Fig. 2; see Fig. 3 in the online Data analyzed them by DHPLC, and determined the mutant Supplement). load with these equations. Measured levels of hetero- Let x ϭ level of heteroplasmy (%); y ϭ predicted/ plasmy based on this equation were highly consistent measured proportion of heteroduplex (%): with actual levels of heteroplasmy (Fig. 3). ϭ Ϫ 2 ϩ y 0.02x 2x (Eq. 4) effect of the dna amount and the presence of Solving x in a quadratic equation: nuclear dna Because the calibration curves that we generated by use of ϭ Ϫ 2 ϩ Ϫ y ax bx c plasmid DNA fitted our parabolic equation, we examined whether the presence of nuclear DNA from rho-0 cells has ax2 Ϫ bx ϩ y ϩ c ϭ 0 any effect on the measured level of heteroplasmy. Fig. 4A b Ϯ ͱb2 Ϫ 4a͑ y ϩ c͒ shows a parabola that is highly consistent between the Іx ϭ (Eq. 5) measurements obtained from samples amplified in the 2a presence and absence of nuclear DNA. Therefore, the It follows from Eqs. 4 and 5 that: presence of nuclear DNA does not affect the quantification of heteroplasmy. We also measured the level of І Ϸ Ϯ ͱ Ϫ x 50 2500 50y (Eq. 6) heteroplasmy with various amounts of DNA (obtained In an experiment, y can be obtained with Eq. 7: from a single patient with a known level of hetero- ϭ plasmy). We observed a similar level of mutation with Let AHET measured peak area of heteroduplex peak; A ϭ measured peak area of homoduplex peak differing DNA amounts in analysis, concluding that HOM quantification by DHPLC is robust with various DNA A ϭ HET ϫ amounts used in the analysis, at least in the interval of 10 y ϩ 100% (Eq. 7) AHET AHOM to 300 ng (Fig. 4B). Eq. 6 showed that there would be 2 values for each x. Discussion As mentioned earlier, we can confirm the levels by mixing DHPLC exploits the differential melting properties of samples with 100% wild-type samples. Using Eqs. 1–3, we homoduplex and heteroduplex DNA to achieve the sep- calculated the predicted proportions of the heterodu- aration of these 2 species. Since its introduction in 1995 plexes (in percentage) formed after slow reannealing for (32), DHPLC has gained tremendous popularity in the DNA (with initial mutant load of 0%, 1%, 2.5%, 5%, 10%, field of DNA mutation detection, especially over the last 25%, 40%, 50%, 60%, 75%, 90%, 95%, 97.5%, 99%, and 5 years, owing to its high degree of automation and high 100%) that are later mixed in equal concentrations with sensitivity (33). It has been used widely as a mutation 100% wild-type PCR products. Fig. 4C in the online Data screening tool, but little has been done to explore its use as Supplement shows a graph plotted on the basis of these a quantitative tool. In this study, by analyzing plasmid predicted values. We also obtained a quadratic equation DNA of various mutant loads, we demonstrated a qua- (Eq. 8) consistent with our experimental data (see Fig. 4A dratic relationship between the proportion of the peak in the online Data Supplement). area of heteroduplex and the level of heteroplasmy. Such ϭ Let y1 predicted/measured proportion of heteroduplex (%) after mixing with 100% wild-type: ϭ Ϫ 2 ϩ y1 0.005x x (Eq. 8) І Ϸ Ϯ ͱ Ϫ x 100 10 000 200y1 (Eq. 9) As heteroplasmy must be Ͻ100%: І Ϸ Ϫ ͱ Ϫ x 100 10 000 200y1 (Eq. 10) Therefore, we hypothesized that Eqs. 6 and 10 can be applied to the generic problem of DHPLC analysis to determine the correct level of heteroplasmy in mtDNA. It is important to note that if y or y1 in these equations has a value Ͼ50 (which is likely to occur during the analysis because of variations in the measurements of DNAs with Fig. 3. Measurement of levels of heteroplasmy in plasmid DNA samples with known mutant loads. mutant loads at ϳ50%), x or x will have undefined 1 Digested PCR products generated from wild-type plasmid DNAs and A3243G values. In these cases, the level of heteroplasmy can be mutant plasmid DNA were mixed to generate samples of various levels of estimated to be ϳ50% Ϯ 10%. heteroplasmy (0%–100%), subjected to heteroduplex formation, and analyzed by DHPLC. Measured levels consistently corresponded to the actual levels in We tested this model by generating several samples samples tested, hence validating the quantitative method used in our analysis. with various degrees of mutation in A3243G (0%–100%), Each level represents the mean of 3 separate samples with SD. Clinical Chemistry 53, No. 6, 2007 1051

eroplasmy level in several plasmid DNA samples of known mutant loads. We found levels of heteroplasmy in DNA to be highly consistent with the expected mutant load of DNA. Sequencing can be carried out to identify the sites and types of these mutations. One main caveat in quantitative DHPLC analysis, however, which is common to all quantitative chromatographic analyses (e.g., gas chromatography–mass spectrometry, liquid chromatography–mass spectrometry, HPLC/UV), is that the resolution of the heteroduplex and homoduplex peaks is critical to allow accurate quantification. Good peak separation can be achieved with the selection of optimal temperature and acetonitrile gradient. Theoretically, 4 peaks are resolved in a reannealed heteroplasmic DNA mixture, which produces 2 homoduplex peaks and 2 heteroduplex peaks. Unfortunately, in most cases the homoduplex peaks elute at almost the same time, preventing quantification of heteroplasmy level from the homoduplex peaks (data not shown). Several reports have used DHPLC for mutation detection but have relied on alternative assays to measure the level of heteroplasmy (2, 4). This report is the first validated quantitative method based on DHPLC, and with the increasingly widespread applications of DHPLC in the study of mtDNA mutation, it will be useful to rely on DHPLC for both mutation screening and quantification in the study of mtDNA. Similarly, the quantitative DHPLC Fig. 4. Effects of nuclear DNA and DNA amount on the quantitative method can also be applied to the study of nuclear DNA DHPLC analysis. mutation. For instance, cancerous tissues are heteroge- (A), parabolic relationships between measured proportion of heteroduplexes and neous in nature (tumor cells are often aneuploidy and levels of heteroplasmy in the presence of nuclear DNA. Samples of different levels of heteroplasmy were subjected to PCR in the presence or absence of surrounded by normal cells), and the tissue specimens nuclear DNA isolated from rho-0 cells and then analyzed by DHPLC after obtained may contain both wild-type and mutant DNAs. heteroduplex formation. Each data point represents the mean Ϯ SD for 3 separate samples. (B), determination of mutant loads in various amounts of DNA Determination of the mutant loads in such tissues may be isolated from whole blood of 1 patient. Increasing amounts of total DNA (10–300 useful in the genetic characterization of cancer. ng) were subjected to PCR, restriction enzyme digestion, and analysis by DHPLC after heteroduplex formation. The level of heteroplasmy measured was independent of the sample amount used and remained at ϳ34%. Each data point represents the mean Ϯ SD for 3 separate samples. Grant/funding support: We are grateful to the United Mito- chondrial Disease Foundation for a grant to R.H.H. in sup- a relationship agrees well with the kinetics of heterodu- port of this project and to the Wright Family Foundation and plex formation (see Fig. 4, B and C, in the online Data Rita and Steven Achard for generous gifts in support of this Supplement) and may well be applicable to all types of work. R.K.N. was supported by the University of California heteroduplex-based assays. Consistent with this observa- San Diego Foundation Christini Fund and by generous gifts tion is a recent report by Palais et al. (31), who observed from the Scott Pawlowski Memorial Fund and Betty Gleeson. a similar quadratic relationship in genomic DNA by Financial disclosures: None declared. high-resolution melting analysis. In an attempt to identify Acknowledgements: We thank Scott Wong for technical the optimal proportion of reference homozygous DNA to assistance with instrumentation and cloning and Dr. Keshav be added to a DNA sample to differentiate heterozygous, Singh for the gift of the 143B osteosarcoma rho-0 cells. homozygous, and wild-type genotypes, Palais et al. (31) studied the dependence of heteroduplex proportion on genotype and the fraction of added wild-type reference References DNA by the use of a series of samples containing different 1. Bayat A, Walter J, Lamb H, Marino M, Ferguson MW, Ollier WE. Mitochondrial mutation detection using enhanced multiplex dena- fractions of reference DNA. They concluded that optimal turing high-performance liquid chromatography. Int J Immunogenet mixing with reference DNA permits genotyping of all 2005;32:199–205. single nucleotide polymorphisms (31). 2. Biggin A, Henke R, Bennetts B, Thorburn DR, Christodoulou J. On the basis of these observations, we developed a Mutation screening of the mitochondrial genome using denaturing quantitative DHPLC method to study mtDNA mutation high-performance liquid chromatography. Mol Genet Metab 2005; load. We validated this method by determining the het- 84:61–74. 1052 Lim et al.: Quantitative DHPLC Analysis of mtDNA Mutation

3. Liu MR, Pan KF, Li ZF, Wang Y, Deng DJ, Zhang L, et al. Rapid mitochondrial DNA mutations in patients with respiratory chain screening mitochondrial DNA mutation by using denaturing high- defects. Hum Mutat 2005;25:575–82. performance liquid chromatography. World J Gastroenterol 2002; 19. Qiu P, Shandilya H, D’Alessio JM, O’Connor K, Durocher J, Gerard 8:426–30. GF. Mutation detection using Surveyor nuclease. Biotechniques 4. van Den Bosch BJ, de Coo RF, Scholte HR, Nijland JG, van Den 2004;36:702–7. Bogaard R, de Visser M, et al. Mutation analysis of the entire 20. Urata M, Wakiyama M, Iwase M, Yoneda M, Kinoshita S, Ha- mitochondrial genome using denaturing high performance liquid masaki N, et al. New sensitive method for the detection of the chromatography. Nucleic Acids Res 2000;28:E89. A3243G mutation of human mitochondrial deoxyribonucleic acid 5. Meierhofer D, Mayr JA, Ebner S, Sperl W, Kofler B. Rapid screening in diabetes mellitus patients by ligation-mediated polymerase of the entire mitochondrial DNA for low-level heteroplasmic muta- chain reaction. Clin Chem 1998;44:2088–93. tions. Mitochondrion 2005;5:282–96. 21. Szuhai K, Ouweland J, Dirks R, Lemaitre M, Truffert J, Janssen G, 6. Wulfert M, Tapprich C, Gattermann N. Optimized PCR fragments et al. Simultaneous A8344G heteroplasmy and mitochondrial DNA for heteroduplex analysis of the whole human mitochondrial copy number quantification in myoclonus epilepsy and ragged-red genome with denaturing HPLC. J Chromatogr B Analyt Technol fibers (MERRF) syndrome by a multiplex molecular beacon based Biomed Life Sci 2006;831:236–47. real-time fluorescence PCR. Nucleic Acids Res 2001;29:E13. 7. Suomalainen A, Ciafaloni E, Koga Y, Peltonen L, DiMauro S, Schon 22. Bai RK, Wong LJ. Detection and quantification of heteroplasmic EA. Use of single strand conformation polymorphism analysis to mutant mitochondrial DNA by real-time amplification refractory detect point mutations in human mitochondrial DNA. J Neurol Sci mutation system quantitative PCR analysis: a single-step ap- 1992;111:222–6. proach. Clin Chem 2004;50:996–1001. 8. Mashima Y, Saga M, Hiida Y, Oguchi Y, Wakakura M, Kudoh J, et 23. Urata M, Wada Y, Kim SH, Chumpia W, Kayamori Y, Hamasaki N, al. Quantitative determination of heteroplasmy in Leber’s heredi- et al. High-sensitivity detection of the A3243G mutation of mito- tary optic neuropathy by single-strand conformation polymor- chondrial DNA by a combination of allele-specific PCR and peptide phism. Invest Ophthalmol Vis Sci 1995;36:1714–20. nucleic acid-directed PCR clamping. Clin Chem 2004;50:2045– 9. Tanno Y, Yoneda M, Tanaka K, Tanaka H, Yamazaki M, Nishizawa 51. M, et al. Quantitation of heteroplasmy of mitochondrial tRNA(L- 24. Oberacher H, Niederstatter H, Huber CG, Parson W. Accurate eu(UUR)) gene using PCR-SSCP. Muscle Nerve 1995;18:1390–7. determination of allelic frequencies in mitochondrial DNA mixtures by electrospray ionization time-of-flight mass spectrometry. Anal 10. Tully LA, Parsons TJ, Steighner RJ, Holland MM, Marino MA, Bioanal Chem 2006;384:1155–63. Prenger VL. A sensitive denaturing gradient-gel electrophoresis 25. Conley YP, Brockway H, Beatty M, Kerr ME. Qualitative and assay reveals a high frequency of heteroplasmy in hypervariable quantitative detection of mitochondrial heteroplasmy in cerebro- region 1 of the human mtDNA control region. Am J Hum Genet spinal fluid using denaturing high-performance liquid chromatog- 2000;67:432–43. raphy. Brain Res Protoc 2003;12:99–103. 11. Sternberg D, Danan C, Lombes A, Laforet P, Girodon E, Goossens 26. Ruano G, Deinard AS, Tishkoff S, Kidd KK. Detection of DNA M, et al. Exhaustive scanning approach to screen all the mito- sequence variation via deliberate heteroduplex formation from chondrial tRNA genes for mutations and its application to the genomic DNAs amplified en masse in “population tubes.” PCR investigation of 35 independent patients with mitochondrial dis- Methods Appl 1994;3:225–31. orders. Hum Mol Genet 1998;7:33–42. 27. Ruano G, Kidd KK. Modeling of heteroduplex formation during PCR 12. Michikawa Y, Hofhaus G, Lerman LS, Attardi G. Comprehensive, from mixtures of DNA templates. PCR Methods Appl 1992;2: rapid and sensitive detection of sequence variants of human 112–6. mitochondrial tRNA genes. Nucleic Acids Res 1997;25:2455–63. 28. Gjerde DT, Hanna CP, Hornby D. DNA Chromatography. 1st ed. 13. Wartell RM, Hosseini SH, Moran CP Jr. Detecting base pair Weinheim: Wiley-VCH Verlag GmbH, 2002:83–91. substitutions in DNA fragments by temperature-gradient gel elec- 29. Jones AC, Austin J, Hansen N, Hoogendoorn B, Oefner PJ, Cheadle trophoresis. Nucleic Acids Res 1990;18:2699–705. JP, et al. Optimal temperature selection for mutation detection by 14. Chen TJ, Boles RG, Wong LJ. Detection of mitochondrial DNA denaturing HPLC and comparison to single-stranded conformation mutations by temporal temperature gradient gel electrophoresis. polymorphism and heteroduplex analysis. Clin Chem 1999;45: Clin Chem 1999;45:1162–7. 1133–40. 15. Higashimoto T, Yano S, Ito M, Clark-Sheehan NC, Cowan L, Boles 30. Bayat A, Walter JM, Lamb H, Ferguson MW, Ollier WE. Rapid RG. Rapid detection of FGFR mutations in syndromic craniosynos- denaturing high-performance liquid chromatography (DHPLC) for tosis by temporal temperature gradient gel electrophoresis. Clin mutation scanning of the transforming growth factor beta3 gene Chem 1999;45:2005–6. using a novel proof-reading polymerase. Eur J Immunogenet 16. Wong LJ, Chen TJ, Tan DJ. Detection of mitochondrial DNA 2003;30:335–40. mutations using temporal temperature gradient gel electrophore- 31. Palais RA, Liew MA, Wittwer CT. Quantitative heteroduplex analysis. Electrophoresis 2004;25:2602–10. sis for single nucleotide polymorphism genotyping. Anal Biochem 17. White HE, Durston VJ, Seller A, Fratter C, Harvey JF, Cross NC. 2005;346:167–75. Accurate detection and quantitation of heteroplasmic mitochon- 32. Oefner PJ, Underhill PA. Comparative DNA sequencing by denatur- drial point mutations by pyrosequencing. Genet Test 2005;9: ing high-performance liquid chromatography (DHPLC). Am J Hum 190–9. Genet 1995;57:A266. 18. Bannwarth S, Procaccio V, Paquis-Flucklinger V. Surveyor nucle- 33. Xiao W, Oefner PJ. Denaturing high-performance liquid chromatog- ase: a new strategy for a rapid identification of heteroplasmic raphy: a review. Hum Mutat 2001;17:439–74. Clinical Chemistry 53:6 1053–1060 (2007) Hemostasis and Thrombosis

Platelet Resistance to the Antiaggregatory Cyclic Nucleotides in Central Obesity Involves Reduced Phosphorylation of Vasodilator-Stimulated Phosphoprotein

Isabella Russo, Paola Del Mese, Gabriella Doronzo, Alessandro De Salve, Mariantonietta Secchi, Mariella Trovati, and Giovanni Anfossi*

Background: Impairment of platelet response to antiag- Conclusions: In central obesity the reduced ability of gregatory agents is seen in individuals with central cyclic nucleotides to inhibit platelet aggregation is as- obesity and may play a role in the increased cardiovas- sociated with reduced activation of their specific ki- cular risk associated with obesity. In this study we nases. Because cyclic nucleotides help regulate platelet evaluated whether this impairment involves the antiag- antiaggregation, alteration of this ability is consistent gregatory pathways regulated by cAMP and cGMP. with platelet hyperactivity in obesity. Methods: We obtained platelet-rich plasma from 12 © 2007 American Association for Clinical Chemistry obese individuals and 12 controls. We investigated the effects of the cyclic nucleotide analogs 8-pCPT-cAMP Central obesity is a major risk factor for early cardiovas- (10–500 ␮mol/L) and 8-pCPT-cGMP (10–500 ␮mol/L) on cular events and other atherothrombotic diseases (1, 2).It ADP-induced platelet aggregation as assessed by de- is associated with insulin resistance, lipid disorders, oxi- creased light scattering. We assessed the activation of dative stress, and low-grade chronic inflammation as well cAMP- and cGMP-dependent protein kinases by mea- as with a prothrombotic state (1, 2). For this reason, suring phosphorylation of the vasodilator-stimulated central obesity has been considered in the recent classifi- phosphoprotein (VASP) at Ser157 and Ser239. cation of the International Diabetes Federation as a nec- Results: The antiaggregatory effect of both cyclic nucle- essary component of the metabolic syndrome (3), a cluster otide analogs was impaired in obese individuals com- of cardiovascular risk factors. pared to controls, with mean (SE) half-maximal inhibi- One possible mechanism contributing to enhanced tory concentrations (IC50) (after 20-min incubation) of thrombotic risk in obese patients is platelet hyperactiva- 123 (33) ␮mol/L vs 5 (1) ␮mol/L, respectively, for tion, which is involved in the different steps of the 8-pCPT-cAMP (P <0.01) and of 172 (43) ␮mol/L vs 17 (8) atherosclerotic process (4, 5). ␮mol/L, respectively, for 8-pCPT-cGMP (P <0.01). The Our previous studies of platelets of obese persons Homeostasis Model Assessment Index of Insulin Resis- identified multiple defects in the sensitivity to the antiag- tance was independently correlated with cyclic nucleo- gregatory mediators. In particular, we observed reduction tide analog IC50. In obese individuals, VASP phosphor- of the antiaggregatory effects of insulin and organic ylation at Ser157 and Ser239 in response to cyclic nitrates, which activate the cGMP pathway (6–9), and of nucleotides was significantly lower than in controls. adenosine and prostacyclin, which activate the cAMP pathway (10, 11). We also observed reduced sensitivity to the antiaggregatory effects of cGMP and cAMP them-

Diabetes Unit, Department of Clinical and Biological Sciences of the University of Turin, San Luigi Gonzaga Hospital, Orbassano (Turin), Italy. * Address correspondence to this author at: Diabetes Unit, Department of 1 Nonstandard abbreviations: PKG, protein kinase G; PKA, protein kinase Clinical and Biological Sciences of the University of Turin, San Luigi Gonzaga A; VASP, vasodilator-stimulated phosphoprotein; GP IIb/IIIa, glycoprotein Hospital, 10043 Orbassano (Turin), Italy. Fax 39-011-9038639; e-mail IIb/IIIa; 8-pCPT-cAMP, 8-(4-Chlorophenylthio)-cAMP; 8-pCPT-cGMP, 8-(4- [email protected]. Chlorophenylthio)-cGMP; HOMA IR, Homeostasis Model Assessment Index Received July 22, 2006; accepted March 16, 2007. of Insulin Resistance; PRP, platelet-rich plasma; PPP, platelet-poor plasma;

Previously published online at DOI: 10.1373/clinchem.2006.076208 IC50, half-maximal inhibitory concentration.

1053 1054 Russo et al.: Cyclic Nucleotides and VASP Phosphorylation in Obesity

selves, which are the main intracellular messengers re- pressure value Ͻ140/90 mm Hg); and had fasting and sponsible for inhibition of platelet responses elicited by postchallenge plasma glucose concentrations within ref- the large majority of platelet agonists (11). erence intervals [fasting plasma glucose Յ6.105 mmol/L Platelet response to cyclic nucleotides is complex and (110 mg/dL) and plasma glucose Ͻ7.77 mmol/L (140 depends on nucleotide concentrations and the time- mg/dL) 2 h after a 75-g oral glucose load]. Biochemical course of platelet exposure. Most authors agree that the variables were measured as described below. main effect of cyclic nucleotides is inhibitory and is exerted through activation of the corresponding cyclic study design nucleotide-dependent protein kinases, i.e., protein kinase In previous investigations we used 8-bromo analogs of G (PKG)1 for cGMP and protein kinase A (PKA) for cAMP cyclic nucleotides, which because of their hydrophilicity (12–16), which are involved in the regulation of basic are poorly permeable through cell membranes (11). In this mechanisms of platelet activation, such as agonist-in- study we used the more lipophilic molecules 8-(4-Chloro- duced increases of cytosolic calcium (12, 15–17), fibrino- phenylthio)-cAMP (8-pCPT-cAMP) and 8-(4-Chlorophe- gen binding (18), and cytoskeleton protein contraction nylthio)-cGMP (8-pCPT-cGMP), which are highly effec- (19). Recent evidence indicates, however, that increased tive in PKA and PKG activation (28) and do not interfere concentrations of platelet cGMP are associated with en- with cyclic nucleotide phosphodiesterases (28); because hanced platelet function (20, 21). these analogs have not been previously used in studies of A relevant target of both cyclic nucleotide-regulated platelets from obese individuals, we also evaluated protein kinases is the focal adhesion protein vasodilator- whether their antiaggregatory effect is decreased in cen- stimulated phosphoprotein (VASP) (22, 23), which is stra- tral obesity. tegically involved in platelet inhibitory pathways. VASP For platelets from both lean and obese study partici- phosphorylation closely correlates with inhibition of fi- pants, we investigated (a) sensitivity to the antiaggrega- brinogen binding to glycoprotein IIb/IIIa (GP IIb/IIIa) tory effects of the cyclic nucleotide analogs 8-pCPT-cAMP (18, 24), and it affects initial sequences of platelet adhe- and 8-pCPT-cGMP on ADP-induced platelet aggregation; sion and activation by modulating interactions of platelet (b) PKA and PKG concentrations; and (c) concentrations actin filaments (4, 19). VASP is considered a reliable of total VASP and of VASP phosphorylation in response mediator of cyclic nucleotide action (12, 24, 25). to 8-pCPT-cAMP and 8-pCPT-cGMP. The reduced platelet antiaggregatory activity exerted by both cyclic nucleotides (11) does not necessarily lead to biochemical variables impaired activation of cyclic nucleotide/specific kinase/ Fasting venous plasma glucose, serum cholesterol, HDL VASP pathways. Platelets from individuals with insulin cholesterol, and triglycerides were measured by auto- resistance have higher free cytoplasmic calcium concen- mated chemical analyses in the central laboratory of our trations than platelets from controls (26); thus, calcium hospital. Fasting plasma insulin was measured by RIA fluxes may present a primitive resistance that is also with a reagent set from Biochem Immuno System S.p.A.; inhibited by efficient cyclic nucleotide/specific kinase/ the cross-reactivity was 100% for human insulin, 14% for VASP pathways. The aim of these studies was to clarify human proinsulin, and 0.0002% for C-peptide and gluca- whether cyclic nucleotides activate downstream path- gon. Fasting C-peptide was measured by RIA with a ways in individuals with central obesity, as they do in reagent set from Biochem Immuno System S.p.A.; the lean controls. cross-reactivity was 100% for human C peptide, 3.2% for human proinsulin, and absent for glucagon. Insulin sen- Patients and Methods sitivity in the fasting state was estimated with the Ho- study participants meostasis Model Assessment Index of Insulin Resistance The study participants were 12 healthy volunteers [6 men (HOMA IR) according to the following formula: fasting and 6 women, mean (SE) age 34.7 (1.9) years] and 12 plasma glucose (mmol/L) ϫ fasting serum insulin (␮U/ individuals with central obesity [6 men and 6 women, age mL) divided by 22.5 (29). HOMA IR is commonly used in 35.4 (2.1) years]. The criterion for central obesity was a clinical studies as a marker of insulin resistance (30, 31); waist circumference, measured at its smallest point with high HOMA IR scores denote low insulin sensitivity. the abdomen relaxed, Ͼ88 cm in women or Ͼ102 cm in men (27). All study participants gave informed consent platelet aggregation studies before investigation and the Ethics Committee of our Blood samples were collected after study participants had Department approved the study design. None of the fasted overnight. A venous blood sample was collected study participants had smoked or taken medications that without stasis and anticoagulated with 1 volume of so- could influence platelet function during the previous 4 dium citrate, 38 g/L, pH 7.4, to 9 volumes of blood. weeks. Obese participants were otherwise healthy on the Platelet-rich plasma (PRP) was obtained from citrated basis of medical history, physical examination, and stan- whole blood by 20-min centrifugation at 100g at room dard diagnostic procedures; had no family history of temperature; platelet-poor plasma (PPP) was prepared by diabetes mellitus; were normotensive (i.e., arterial blood further centrifugation at 2000g for 10 min. Platelet counts Clinical Chemistry 53, No. 6, 2007 1055

Table 1. Clinical characteristics of the study participants.a Obese Characteristics Controls individuals Significance No. 12 12 Men/women (n/n) 6/6 6/6 Age, years 34.7 (1.9) 35.4 (2.1) NS BMI,b kg/m2 21.5 (0.4) 32.0 (0.7) P Ͻ0.0001 Waist circumference, cm 78.5 (3.0) 106.5 (2.1) P Ͻ0.0001 Glucose, mmol/L 4.83 (0.14) 5.18 (0.37) NS Insulin, pmol/L 33.7 (3.3) 217.8 (40.8) P Ͻ0.002 HOMA IR index 1.7 (0.2) 4.8 (0.6) P Ͻ0.0001 Fasting C-peptide, ng/mL 2.17 (0.18) 3.69 (0.21) P Ͻ0.0001 Systolic blood pressure, mmHg 118 (2.4) 125 (2.0) P Ͻ0.05 Diastolic blood pressure, mmHg 78 (1.3) 82 (2.1) NS Triglycerides, mmol/L 1.1 (0.1) 1.7 (0.2) P Ͻ0.03 Total cholesterol, mmol/L 4.8 (0.3) 5.3 (0.3) NS HDL cholesterol, mmol/L 1.6 (0.1) 1.1 (0.1) P Ͻ0.005 Platelet count in PRP, 109/L 268 (25) 248 (18) NS a Values are expressed as mean (SE). b BMI, Body mass index; NS, not significant.

were determined on an S-plus Coulter Counter (Coulter gation at 30 000g for 60 min, 30 ␮g protein from platelet Electronics). Mean (SE) platelet counts in PRP (109/L) lysates was subjected to 8% sodium dodecyl sulfate- were 268 (25) in controls and 248 (18) in obese individuals polyacrylamide gel electrophoresis and then transferred (not significant). Because the study design consisted of to polyvinylidenedifluoride membrane (Millipore). Mem- measurement of platelet responses in samples from the branes were incubated at room temperature for 1 h with same PRP after addition of buffer solutions or different rabbit polyclonal antibodies against PKA 1␣/1␤ (Santa substances for each study participant, platelet numbers Cruz Biotechnology; 1:3000), or rabbit polyclonal antibod- were not adjusted. Platelet aggregation was carried out by ies against PKG 1␣/1␤ (Calbiochem; 1:300). Then, mem- following light-scattering changes as originally described branes were washed 3 times for 10 min each with PBS (136 by Born (32), using a model 500 Chrono Log aggregome- mmol/L NaCl, 2.7 mmol/L KCl, 10 mmol/L NaHPO4, 1.8 ter at a constant stirring speed of 900 rpm. mmol/L KH2PO4)/0.1% Tween-20 (PBS-Tween) and in- Platelet aggregation in response to ADP was reported cubated with antirabbit horseradish peroxidase-conju- ϫ as maximal aggregation, calculated as: 100 [(initial gated secondary antibody (1:10 000) for 45 min. After 3 Ϫ absorbance absorbance after addition of ADP)/(initial final washes (10 min each) in PBS-Tween, membranes absorbance)], with ADP added at a final concentration of were subjected to chemiluminescence (Amersham Life ␮ 4 mol/L. Half-maximal inhibitory concentration (IC50) Sciences) for detection of the specific antigen. Density of values of the 2 cyclic nucleotide analogs were determined bands in Western blots was analyzed with Kodak 1D at the different incubation times, when possible. Image Analysis software.

protein content of pka and pkg protein content of vasp and vasp PKA and PKG concentrations were determined by West- phosphorylation in response to analogs of ern blotting as previously described by Dey et al. (33). camp and cgmp Experiments were carried out in 50-mL blood samples PKA preferentially induces VASP phophorylation at antiaggregated with acid citrate-dextrose solution (vol/ Ser157 causing an upward shift in the apparent molecular vol:1/6). ACD-anticoagulated PRP, obtained by centrifu- weight from 46 kDa to 50 kDa in sodium dodecyl sulfate- gation at 100g for 20 min, underwent further centrifuga- polyacrylamide gel electrophoresis, whereas PKG prefer- tion at 2000g for 10 min. The pellet was washed 2 times at entially induces VASP phosphorylation at Ser239 without 37 °C in HEPES-Na buffer (10 mmol/L HEPES Na, 140 any change in molecular mass (22, 23).

mmol/L NaCl, 2.1 mmol/L MgSO4, 10 mmol/L D-glu- Experiments were carried out in blood samples antico- cose, pH 7.4); 500 ␮L of washed platelets (2.5 ϫ 109 agulated with ACD and washed platelets prepared as platelets/mL) were sedimented by centrifugation at 2000g previously described; 500 ␮L of washed platelets contain- for 10 min and solubilized by lysis buffer (1% SDS, 0.1% ing 2.5 ϫ 109 platelets/mL were preincubated in the Triton X-100, 10 mmol/L Tris-HCl, pH 7.4), supple- presence of 8-pCPT-cAMP (10–500 ␮mol/L), 8-pCPT- mented with protease inhibitors (Sigma). After centrifu- cGMP (10–500 ␮mol/L), or buffer solution for 10 min. 1056 Russo et al.: Cyclic Nucleotides and VASP Phosphorylation in Obesity

Platelets were then sedimented by centrifugation at 2000g for 10 min, solubilized, and subjected to Western blot analysis as described previously. The incubation of membranes was carried out with the following antibodies (all obtained from Calbiochem): rabbit antihuman VASP protein (1:15 000); monoclonal antibody recognizing VASP phosphorylated at Ser157 (1:1000); and monoclonal antibody recognizing VASP phosphorylated at Ser239 (1:1000). After 3 washes in PBS-Tween, membranes were incubated for 45 min with monoclonal antirabbit horseradish peroxidase-conjugated secondary antibody (1:10 000) for VASP protein detection or with horseradish peroxidase- conjugated rabbit antimouse IgG (1:50 000) for phosphorylated VASP detection. After additional washes, protein expression was visualized as described previously. chemicals ADP, 8-pCPT-cAMP, 8-pCPT-cGMP, PBS, and Tween-20 were obtained from Sigma. The source of the specific antibodies for Western blotting has been previously indicated. statistical analysis Data in the text and in the Figs. are expressed as mean (SE). Statistical analyses were performed with ANOVA for repeated measurements, and, when appropriate, by the Student t-test for unpaired data. IC50 values were determined by probit analysis. Furthermore, simple and multiple regression analyses were carried out by use of the Stat View Software for Macintosh.

Results Characteristics of the study participants are summarized in Table 1. The 2 groups differed significantly in body mass index (P Ͻ0.0001), waist circumference (P Ͻ0.0001), fasting insulin (P Ͻ0.002), fasting C-peptide (P Ͻ0.0001), HOMA IR index (P Ͻ0.0001), systolic blood pressure (P Ͻ0.05), triglycerides (P Ͻ0.03), and HDL cholesterol (P Ͻ0.005). Although all of the obese study participants presented with central obesity, none fulfilled the criteria for the diagnosis of the metabolic syndrome according to the Third Report of the National Cholesterol Education Fig. 1. (A), concentration- and time-dependent effects of the cell Program Expert Panel on Detection, Evaluation, and permeable cAMP analog 8-pCPT-cAMP (10–500 ␮mol/L) on ADP- Treatment of High Blood Cholesterol in Adults (ATP III) induced platelet aggregation in controls (n ϭ 12) and obese individuals [i.e., the simultaneous presence of 3 of the following (n ϭ 12) after 3-, 10-, and 20-min incubation. variables: waist circumference Ͼ88 cm in women, Ͼ102 Data [mean (SE), n ϭ 12] are expressed as percentage of the aggregation values cm in men; HDL-C Ͻ1.036 mmol/L (40 mg/dL) in men, in absence of the cyclic nucleotide analog. Inhibition of platelet response was significantly lower in obese individuals, as shown in the Fig. (*P Ͻ0.05, **P Ͻ Ͻ Ͻ ϭ 1.295 mmol/L (50 mg/dL) in women; fasting triglycer- 0.01, and ***P 0.0001). (B), IC50 values [mean (SE), n 12] of 8-pCPT- ides Ͼ1.695 mmol/L (150 mg/dL); blood pressure cAMP (␮mol/L) for ADP-induced aggregation after 20-min incubation in controls (n ϭ 12) and obese individuals (n ϭ 12). Significance in the Fig.: **P Ͻ0.01. Ͼ130/85 mmHg; fasting glucose Ͼ6.105 mmol/L (110 mg/dL)] (27). response to 4 ␮mol/L ADP in obese and lean individuals platelet aggregation were 53% (3%) and 51% (4%), respectively (not Platelet numbers were the same for obese and lean significant). individuals, and platelet response to ADP was similar. In Platelet aggregation in response to ADP was reduced particular, mean (SE) maximal aggregation values in by 8-pCPT-cAMP (3–20 min incubation; 10–500 ␮mol/L) Clinical Chemistry 53, No. 6, 2007 1057

in a concentration-dependent manner both in lean con- protein content of pka and pkg trols and in obese individuals (Fig. 1A; ANOVA for Protein content of PKA and PKG was similar in platelets repeated measurements: P Ͻ0.0001 for both groups). The from obese individuals and from lean controls (Fig. 3A). antiaggregatory effects of 8-pCPT-cAMP, however, were greater in controls for each concentration of the cyclic phosphorylation of vasp nucleotide analog at all experimental times (P Ͻ0.05– The total VASP protein content of resting platelets was 0.0001). In particular, the mean (SE) IC50 values were similar in lean and obese individuals (Fig. 3A). In both lower in controls than in obese-individuals at 10 min [24 lean and obese individuals 8-pCPT-cAMP increased (7) vs 252 (41) ␮mol/L; P Ͻ0.0001] and at 20 min [5 (1) vs VASP phosphorylation at Ser157 (ANOVA for repeated ␮ Ͻ 123 (33) mol/L; P 0.01] (Fig. 1B). In individuals with measurements: P Ͻ0.0001 for both groups) (Fig. 3B); the central obesity, no 8-pCPT-cAMP concentration inhibited increase was smaller in obese than in lean individuals platelet aggregation to ADP by 50%, so the 8-pCPT-cAMP (significance vs lean individuals: P Ͻ0.0001 with 10 IC50 for a 3 min platelet exposure could not be calculated. ␮mol/L, P Ͻ0.01 with 100 ␮mol/L, and P Ͻ0.05 with 500 With simple regression analysis, the IC50 value of 8-pCPT- ␮mol/L). cAMP at 20 min was positively correlated with waist Platelet exposure to 8-pCPT-cGMP increased VASP circumference, HOMA IR, HDL cholesterol, triglycerides, phosporylation at Ser239 in both lean and obese individ- and systolic and diastolic blood pressure (Table 2). With uals (ANOVA for repeated measurements: P Ͻ0.001 for multiple regression analysis, however, only HOMA IR both groups; Fig. 4); the increase was smaller in obese was significantly correlated with 8-pCPT-cAMP IC . 50 than in lean individuals (significance vs lean individuals: In both lean controls and in obese individuals, 8-pCPT- P Ͻ0.0001 with 10 ␮mol/L, P Ͻ0.001 with 100 ␮mol/L, cGMP (3–20 min incubation; 10–500 ␮mol/L) decreased and P Ͻ0.01 with 500 ␮mol/L). ADP-induced platelet aggregation in a concentration- dependent manner (ANOVA: P Ͻ0.0001) (Fig. 2A). The Discussion 8-pCPT-cGMP antiaggregatory effects, however, were greater in controls for each concentration of the cyclic This study showed that decreased antiaggregatory action nucleotide analog at all investigated times (P Ͻ0.05– of the cyclic nucleotides cAMP and cGMP in platelets of individuals with central obesity was associated with de- 0.0001). Furthermore, mean (SE) 8-pCPT-cGMP IC50 values were lower in the control participants than in obese creased phosphorylation of VASP at specific sites, reflect- individuals with exposure for 20 min [17 (8) ␮mol/L vs ing impaired activation of PKA and PKG. 172 (43) ␮mol/L; P Ͻ0.01] (Fig. 2B). In individuals with This result strengthens our previous observation that the antiaggregatory action of cAMP and cGMP is reduced central obesity, it was impossible to calculate the IC50 value with shorter 8-pCPT-cGMP exposure, no concentra- in patients with central obesity (11). In fact, in the present tion inhibited ADP-induced aggregation by 50%. When study we used highly lipophilic, permeable cyclic nucle- simple regression analysis was used, the IC50 value of otide analogs (i.e., 8-pCPT-cAMP and 8-pCPT-cGMP), 8-pCPT-cGMP at 20 min was positively correlated with which, unlike those previously used (11), do not interfere waist circumference, HOMA IR, HDL cholesterol, triglyc- with phosphodiesterases and are very effective in the erides, and systolic and diastolic blood pressure (Table 2). activation of specific protein kinases (28). With multiple regression analysis, however, only HOMA In obese individuals the inhibitory effect of these

IR was significantly correlated with 8-pCPT-cGMP IC50. analogs on platelet aggregation induced by ADP was

Table 2. Simple and multiple regression analysis concerning the relation between cyclic nucleotide IC50 values and different variables measured in controls and obese individuals.

IC50 value of 8-pCPT-cAMP IC50 value of 8-pCPT-cGMP

Regression analysis Simple Multiple Simple Multiple

rP␤ tPr P ␤ tP Waist circumference 0.796 0.001 Ϫ0.035 Ϫ0.108 0.917 0.786 0.001 Ϫ0.109 Ϫ0.269 0.797 BMIa 0.858 0.0001 Ϫ0.197 Ϫ0.525 0.619 0.839 0.0001 Ϫ0.362 Ϫ0.779 0.466 HOMA IR 0.944 0.0001 1.157 4.079 0.007 0.929 0.0001 1.203 3.416 0.014 HDL cholesterol Ϫ0.531 0.051 0.258 1.789 0.124 Ϫ0.534 0.049 0.215 1.203 0.274 Triglycerides 0.747 0.002 0.149 0.584 0.581 0.744 0.002 0.341 1.077 0.323 Systolic blood pressure 0.611 0.020 Ϫ0.014 Ϫ0.059 0.955 0.605 0.022 0.141 0.472 0.654 Diastolic blood pressure 0.667 0.009 0.128 0.626 0.554 0.628 0.016 Ϫ0.057 Ϫ0.224 0.830 R2 0.947 0.918 a BMI, Body mass index. 1058 Russo et al.: Cyclic Nucleotides and VASP Phosphorylation in Obesity

Fig. 3. (A), expression of PKG, PKA, and VASP in platelet lysates from both controls and obese individuals. The results, obtained by Western blot with antibodies against the specific proteins, are representative of duplicate experiments carried out in 9 individuals. By densitometric analysis, no difference in protein expression was shown in platelets from obese individuals in comparison with lean controls. (B), expression of the protein VASP phosphorylated at Ser157 (PKA-dependent) in platelet lysates from both controls (n ϭ 12) and obese individuals (n ϭ 12) after 10-min incubation with the cell permeable cAMP analog 8-pCPT-cAMP (10–500 ␮mol/L). The densitometric analysis carried out on the bands obtained by Western blot showed a reduced phosphorylation in platelets from obese individuals in comparison with lean controls. Data are expressed as mean (SE), n ϭ 12. In both lean and obese individuals 8-pCPT-cAMP induced a significant concentration- dependent VASP phosphorylation (ANOVA Ͻ0.0001); the effect in obese individuals, however, was lower than in lean individuals as shown in the Fig. (*P Ͻ0.05, **P Ͻ0.01, and ***P Ͻ0.0001).

HOMA IR and with other variables such as HDL choles- Fig. 2. (A), concentration- and time-dependent effects of the cell permeable cGMP analog 8-pCPT-cGMP (10–500 ␮mol/L) on ADP- terol, triglycerides, and systolic and diastolic blood pres- induced platelet aggregation in controls (n ϭ 12) and obese individuals sure, whose alterations are considered for the diagnosis of (n ϭ 12) after 3-, 10-, and 20-min incubation. metabolic syndrome (27). It should be emphasized, how- Data [mean (SE), n ϭ 12] are expressed as percentage of the aggregation values ever, that none of the obese individuals met criteria for the in absence of the cyclic nucleotide analog. Inhibition of platelet response was diagnosis of metabolic syndrome, because each of them significantly lower in obese individuals as shown in the Fig. (*P Ͻ0.05, **P Ͻ Ͻ ϭ 0.01, and ***P 0.0001). (B), IC50 values [mean (SE), n 12] of 8-pCPT- presented at the most only 1 diagnostic feature in addition cGMP (␮mol/L) for ADP-induced aggregation after 20-min incubation in controls to central obesity. (n ϭ 12) and obese individuals (n ϭ 12). Significance in the Fig.: **P Ͻ0.01. When all parameters were pooled together, multiple

regression analysis showed that only IC50 values of both much reduced, as shown by the fact that their IC50 values cyclic nucleotide analogs remained significantly corre- at 20 min were ϳ10-fold higher than in controls. lated with HOMA IR, thus suggesting that the molecular

Univariate regression analyses showed that IC50 values defect involved in the resistance to cyclic nucleotides is of both cyclic nucleotide analogs were correlated with directly related to insulin resistance. Clinical Chemistry 53, No. 6, 2007 1059

organic nitrates, and prostacyclin, which act through activation of cyclic nucleotide/protein kinase pathways (10, 11), may be explained, at least in part, by an impaired action of the cyclic nucleotides on their specific kinases. VASP phosphorylation induced by PKA and PKG is known to be relevant to the inhibition of aggregation through modulation of actin polymerization and inhibition of fibrinogen binding to the platelet integrin GP IIb/IIIa (18, 24). The phosphorylation state of VASP in intact cells is regulated to a major extent by serine/ threonine protein phosphatases (37); therefore further studies are needed to evaluate whether increased activity of these phosphatases may play a role in the reduced content of phosphorylated VASP in platelets from obese individuals in response to cyclic nucleotide analogs.

In conclusion, the findings of this study further elucidate the complex picture of platelet alterations in obese Fig. 4. Expression of the protein VASP phosphorylated at Ser239 individuals with insulin resistance. Our previous studies (PKG-dependent) in platelet lysates from both controls (n ϭ 12) and showed resistance to antiaggregatory action of agents due ϭ obese individuals (n 12) after 10-min incubation with the cell to impaired ability to increase cyclic nucleotide synthesis permeable cGMP analog 8-pCPT-cGMP (10–500 ␮mol/L). (6–11). The present results also show impaired ability of The densitometric analysis carried out on the bands obtained by Western blot showed a reduced phosphorylation in platelets from obese individuals in cyclic nucleotides themselves to activate downstream comparison with lean controls. Data are expressed as mean (SE), n ϭ 12. In both steps of antiaggregation, such as those related to VASP lean and obese individuals 8-pCPT-cGMP induced a significant concentration- dependent VASP phosphorylation (ANOVA Ͻ0.001); the effect in obese individ- phosphorylation. uals, however, was lower than in lean individuals as shown in the Fig. (*P Ͻ0.01, These defects, which are closely linked to insulin Ͻ Ͻ **P 0.001, and ***P 0.0001). resistance, could contribute to the pathogenesis of the prothrombotic state described in the insulin resistance Our observations may be of interest in the current syndrome and justify, at least in part, the increased debate on the causal relationship between insulin resis- cardiovascular risk attributed to this syndrome tance and metabolic syndrome (34, 35), because they (1, 38–40). suggest that insulin resistance per se is directly involved in the pathogenesis of some platelet abnormalities occurring in central obesity, which is a classical component of the metabolic syndrome (27). Grant/funding support: This study was supported by a Mechanistically, the present study clearly showed im- research grant from Italian Ministero dell’Istruzione, Univer- paired activation of the cyclic nucleotide/specific kinase/ sita` e Ricerca (MIUR) COFIN 2004 within the project “The VASP pathways in individuals with central obesity. Plate- molecular basis of insulin resistance and their importance in lets from obese individuals showed a significant decrease the pathogenesis of the alterations of the vessel wall”, Local in VASP phosphorylation both at Ser157 after exposure to Coordinator: Prof. Giovanni Anfossi, National Coordinator: 8-pCPT-cAMP and at Ser239 after exposure to 8-pCPT- Prof. Amalia Bosia and by a research grant from Regione cGMP, despite a similar platelet content of PKA, PKG, Piemonte (to G.A.). and VASP proteins. To the best of our knowledge, our Financial disclosures: None declared. study provides the 1st demonstration of an impaired Acknowledgements: We thank Mrs. Anna Baker for her activation of both cyclic nucleotide-dependent protein linguistic assistance. kinases in obese patients. Abnormalities of these kinases have been previously observed only in patients with neuro-psychiatric disorders (36). References This study, therefore, identified a novel feature of 1. Grundy SM. Obesity, metabolic syndrome, and coronary athero- platelet dysfunction occurring in central obesity. Because sclerosis. Circulation 2002;105:2696–8. the molecular defects observed in obese individuals are 2. McGill HC, McMahan CA, Herderick EE, Zieske AW, Malcolm GT, Tracy RE, et al. Pathobiological Determinants of Atherosclerosis in involved in crucial steps of the control of platelet function Youth (PDAY) Research Group. Obesity accelerates the progres- (12–16), the present results may be relevant in explaining sion of coronary atherosclerosis in young men. Circulation 2002; resistance not only to the antiaggregatory effects of the 105:2712–8. cyclic nucleotide analogs but also to other antiaggregatory 3. Alberti KG, Zimmet P, Shaw J. IDF Epidemiology Task Force mediators, which we investigated in previous studies Consensus Group. The metabolic syndrome: a new worldwide (6, 9, 11). The reduced antiaggregatory effects of insulin, definition. Lancet 2005;366:1059–62. 1060 Russo et al.: Cyclic Nucleotides and VASP Phosphorylation in Obesity

4. Massberg S, Brand K, Gruner S, Page S, Muller E, Muller I, et al. 22. Waldmann R, Nieberding M, Walter U. Vasodilator-stimulated A critical role of platelet adhesion in the initiation of atheroscle- protein phosphorylation in platelets is mediated by cAMP- and rotic lesion formation. J Exp Med 2002;196:887–96. cGMP-dependent protein kinases. Eur J Biochem 1987;167: 5. Ruggeri ZM. Platelets in atherothrombosis. Nat Med 2002;8: 441–8. 1227–34. 23. Butt E, Abel K, Krieger M, Palm D, Hoppe V, Walter U. cAMP- and 6. Trovati M, Mularoni E, Burzacca S, Ponziani MC, Massucco P, cGMP-dependent protein kinase phosphorylation sites of the focal Mattiello L, et al. Impaired insulin-induced platelet anti-aggregat- adhesion vasodilator-stimulated phosphoprotein (VASP) in vitro ing effect in obesity and in obese NIDDM patients. Diabetes and in intact human platelets. J Biol Chem 1994;269:14509–17. 1995;44:1318–22. 24. Hauser W, Knobeloch KP, Eigenthaler M, Gambaryan S, Krenn V, 7. Trovati M, Anfossi G. Insulin, insulin-resistance, and platelet Geiger J, et al. Megakaryocyte hyperplasia and enhanced agonist- function: similarities with insulin effects on cultured smooth induced platelet activation in vasodilator-stimulated phosphopro- muscle cells. Diabetologia 1998;41:609–22. tein knockout mice. Proc Natl Acad Sci U S A 1999;96:8120–5. 8. Trovati M, Anfossi G. Influence of insulin and of insulin resistance 25. Nolte C, Eigenthaler M, Horstrup K, Honig-Liedl P, Walter U. on platelet and vascular smooth muscle cell function. J Diab Synergistic phosphorylation of the focal adhesion-associated va- Complicat 2002;16:35–40. sodilator-stimulated phosphoprotein in intact human platelets in 9. Anfossi G, Mularoni EM, Burzacca S, Ponziani MC, Massucco P. response to cGMP- and cAMP-elevating platelet inhibitors Bio- Mattiello L, et al. Platelet resistance to nitrates in obesity and chem Pharmacol 1994;48:1569–75. obese NIDDM, and normal platelet sensitivity to both insulin and 26. Resnick LM. Ionic basis of hypertension, insulin resistance, nitrates in lean NIDDM. Diabetes Care 1998;21:121–6. vascular disease, and related disorders: the mechanism of “Syn- 10. Anfossi G, Mularoni E, Burzacca S, Ponziani MC, Massucco P, drome X”. Am J Hypertens 1993;6:123S–34S. Mattiello L, et al. Impaired platelet sensitivity to the anti-aggregat- 27. National Cholesterol Education Program (NCEP) Expert Panel on ing effects of adenosine in obesity and obese non-insulin-depen- Detection, Evaluation, and Treatment of High Blood Cholesterol in dent diabetes mellitus. In: Belfiore F, Lorenzi M, Molinatti GM, Adults (Adult Treatment Panel III). Third report of the National Porta M eds. Molecular and Cell Biology of Type 2 Diabetes and its Cholesterol Education Program (NCEP) Expert Panel on Detection, Complications, Frontiers in Diabetes Vol. 14, Basel: Karger, Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult 1998:184–7. Treatment Panel III): final report. Circulation 2002;106:3143–421. 28. Geiger J, Nolte C, Butt E, Sage SO, Walter U. Role of cGMP and 11. Anfossi G, Russo I, Massucco P, Mattiello L, Doronzo G, De Salve cGMP-dependent protein kinase in nitrovasodilator inhibition of A, et al. Impaired synthesis and action of antiaggregating cyclic agonist-evoked calcium elevation in human platelets. Proc Natl nucleotides in platelets from obese subjects: possible role in Acad Sci U S A 1992;89:1031–5. platelet hyperactivation in obesity. Eur J Clin Invest 2004;34: 29. Matthews DR, Hosker JP, Rudenski AS, Naylor BA, Treacher DF, 482–9. Turner RC. Homeostasis model assessment: insulin resistance 12. Pfeifer A, Ruth P. Structure and function of cGMP-dependent and ␤-cell function from fasting plasma glucose and insulin protein kinases. Rev Physiol Biochem Pharmacol 1999;135:105– concentration in man. Diabetologia 1985;28:412–9. 49. 30. Haffner SM, Miettinen H, Stern MP. Homeostasis model in the 13. Schwarz UR, Walter U, Eigenthaler M. Taming platelets with cyclic San Antonio Heart Study. Diabetes Care 1997;20:1087–92. nucleotides. Biochem Pharmacol 2001;62:1153–61. 31. Ferrannini E, Mari A. How to measure insulin sensitivity. J Hyper- 14. Jensen BO, Selheim F, Doskeland SO, Gear ARL, Holmsen H. tens 1998;16:895–906. Protein kinase A mediates inhibition of the thrombin-induced 32. Born GVR. Aggregation of blood platelets by adenosine diphos- platelet shape change by nitric oxide. Blood 2004;104:2775–82. phate and its reversal. Nature 1962;194:927–9. 15. Antl M, von Bruhl ML, Eiglsperger C, Werner M, Konrad I, Kocher 33. Dey NB, Boerth NJ, Murphy-Ullrich JE, Chang PL, Prince CW, T, et al. IRAG mediates NO/cGMP-dependent inhibition of platelet Lincoln TM. Cyclic GMP-dependent protein kinase inhibits os- aggregation and thrombus formation. Blood 2007;109:552–9. teopontin and thrombospondin production in rat aortic smooth 16. Yamanishi J, Kawahara Y, Fukuzaki H. Effect of cyclic AMP on muscle cells. Circ Res 1998;82:139–46. cytoplasmic free calcium in human platelets stimulated by throm- 34. Kahn R, Buse J, Ferrannini E, Stern M; American Diabetes Associa- bin: direct measurement with quin-2. Thromb Res 1983;32: tion; European Association for the Study of Diabetes. The Metabolic 183–8. Syndrome: time for a critical appraisal: Joint statement from the 17. Kawahara Y, Yamanishi J, Fukuzaki H. Inhibitory action of American Diabetes Association and the European Association for the guanosine 3Ј,5Ј monophosphate on thrombin-induced calcium Study of Diabetes. Diabetes Care 2005;28:2289–304. mobilization in human platelets. Thromb Res 1984;33:203–9. 35. Kim SH, Reaven GM. The metabolic syndrome: one step forward, 18. Horstrup K, Jablonka B, Honig-Liedl P, Just M, Kochsiek K, Walter two steps back. Diab Vasc Dis Res 2004;1:68–75. U. Phosphorylation of focal adhesion vasodilator-stimulated phos- 36. Tardito D, Maina G, Tura GB, Bogetto F, Pioli R, Ravizza L, et al. phoprotein at Ser157 in intact human platelets correlates with The cAMP-dependent protein kinase substrate Rap1 in platelets fibrinogen receptor inhibition. Eur J Biochem 1994;22:21–7. from patients with obsessive compulsive disorder or schizophre- 19. Bearer EL, Prakash JM, Manchester RD, Allen PG. VASP protects nia. Eur Neuropsychopharmacol 2001;11:221–5. actin filaments from gelsolin: an in vitro study with implications for 37. Abel K, Mieskes G, Walter U. Dephosphorylation of the focal platelet actin reorganizations. Cell Motil Cytoskeleton 2000;47: adhesion protein VASP in vitro and in intact human platelets. FEBS 351–64. Lett 1995;370:184–8. 20. Li Z, Xi X, Gu M, Feil R, Ye RD, Eigenthaler M, et al. A stimulatory 38. Isomaa B, Almgren P, Tuomi T, Forsen B, Lahti K, Nissen M, et al. role for cGMP-dependent protein kinase in platelet activation. Cell Cardiovascular morbidity and mortality associated with the meta- 2003;112:77–86. bolic syndrome. Diabetes Care 2001;24:683–9. 21. Li Z, Zhang G, Feil R, Han J, Du X. Sequential activation of p38 and 39. Vinik AI, Erbas T, Park TS, Nolan R, Pittenger GL. Platelet ERK pathways by cGMP-dependent protein kinase leading to dysfunction in type 2 diabetes. Diab Care 2001;24:1476–85. activation of the platelet integrin alphaIIb beta3. Blood 2006;107: 40. Colwell JA, Nesto RW. The platelet in diabetes: focus on preven- 965–72. tion of ischemic events. Diab Care 2003;26:2181–8. Clinical Chemistry 53:6 1061–1066 (2007) Proteomics and Protein Markers

Ultrasensitive Densitometry Detection of Cytokines with Nanoparticle-Modified Aptamers

Yuan-Yuan Li, Chi Zhang, Bo-Sheng Li, Li-Fan Zhao, Xiao-bo Li, Wen-Jie Yang, and Shun-Qing Xu*

Background: Aptamers mimic properties of antibodies or RNA) that have been selected in in vitro selection and sometimes turn out to be even better than antibod- experiments via the systematic evolution of ligands by an ies as reagents for assays. We describe the establishment exponential enrichment (SELEX)1 process to act as ligands of an ultrasensitive densitometry method for cytokine with high-affinity binding to their targets (7). Once the detection by nanoparticle (NP)-modified aptamers. sequences of DNA aptamers have been identified, their Methods: The assay simultaneously uses a gold NP– straightforward synthesis makes them particularly desir- modified aptamer and a biotin-modified aptamer to able. Although aptamers are different from antibodies, bind to the target protein, forming a sandwich complex. they mimic the properties of antibodies in a variety of The absorbance signal generated by the aptamer-protein diagnostic formats (6). The demand for diagnostic assays complex is amplified and detected with a microplate to assist in the management of existing and emerging reader. diseases is increasing, and aptamers have the potential to Results: The assay for platelet-derived growth factor fulfill the requirement for molecular recognition in such B-chain homodimer (PDGF-BB) was linear from 1 assays (8). The analytical Several aptamer-based methods for protein detection .(0.9869 ؍ fmol/L to 100 pmol/L (R2 detection limit was 83 amol/L. The intraassay and inter- have been developed. Fluorescence anisotropy has been assay imprecision (CVs) was <7.5%. Serum concentra- used to measure platelet-derived growth factor (PDGF) tions of PDGF-BB determined with the gold NP–modi- with a fluorescein-labeled single-stranded DNA aptamer fied aptamer assay and with ELISA were not (9), and molecular aptamer beacons have been designed significantly different. for the detection of proteins (10–13). Fredriksson et al. Conclusions: The gold NP–modified aptamer assay pro- (14) and Wang et al. (15) have established highly sensitive vides a fast, convenient method for cytokine detection detection methods by combining aptamer recognition and improves the detection range and the detection with the sensitivity of real-time quantitative PCR. Al- limit compared with ELISA. though real-time PCR–based methods can detect zepto- © 2007 American Association for Clinical Chemistry mole quantities of protein, real-time PCR instruments are still required. Huang et al. developed a colorimetric assay Current methods for the analysis of cytokines and growth that uses gold nanoparticles (NPs) to measure PDGFs and factors are mostly immunologic assays, and antibodies are their receptors (16). The assay was not useful for clinical the most commonly used affinity reagents (1–5). Aptam- monitoring, which requires measurements in the nano- ers can rival antibodies as affinity ligands and have been molar range. used in numerous studies of the detection of proteins (6). In the present study, we developed an ultrasensitive Aptamers are short synthetic nucleotide sequences (DNA densitometric cytokine-detection method that uses gold NP–modified aptamers. The assay relies on aptamer pairs that bind to 2 distinct sites on the target protein or to the same epitope of a homodimer. Binding of the aptamers to MOE Key Laboratory of Environment and Health, School of Public Health, the target protein forms a sandwich complex, which is Tongji Medical College, Huazhong University of Science and Technology, Wuhan, People’s Republic of China. immobilized onto a microplate via avidin. After silver * Address correspondence to this author at: MOE Key Laboratory of Environment and Health, School of Public Health, Tongji Medical College, Huazhong University of Science and Technology, Wuhan 430030, People’s Republic of China. Fax 86-27-83657705; e-mail: [email protected]. edu.cn. 1 Nonstandard abbreviations: SELEX, systematic evolution of ligands by Received October 31, 2006; accepted March 14, 2007. exponential enrichment; PDGF, platelet-derived growth factor; NP, nanopar- Previously published online at DOI: 10.1373/clinchem.2006.082271 ticle; BB, B-chain homodimer.

1061 1062 Li et al.: Cytokine Detection with NP-Modified Aptamers

enhancement, the signal is amplified, and a microplate temperature for 20 min with 10 ␮L of 10-fold serial reader measures the absorbance. dilutions of PDGF-BB (1 amol/L to 10 nmol/L) containing

137 mmol/L NaCl, 10.1 mmol/L Na2HPO4 (pH 7.4), 2.7 Materials and Methods mmol/L KCl, 1 mmol/L MgCl2, and 10 g/L BSA. The materials solution was transferred into the microplate wells and Oligonucleotides modified with a thiol group and biotin incubated at 37 °C for 30 min. The plates were then were synthesized by Invitrogen Biotech and used without washed with phosphate-buffered saline (0.2 mol/L NaCl, Ј further purification (sequence 1, 5 -SH-(CH2)-AGTTCTT- 20mM Na2HPO4/NaH2PO4, pH 7.0) for 5 min and twice ϫ ACTCAGGGCACTTGCAAGCAATTGTGGTCCCAATG- with a 2 buffer containing 0.3 mol/L NaNO3 and 10 Ј Ј GGCTGAGTAT-3 ; sequence 2, 5 -Bio-GCAGTTACTCA- mmol/L Na2HPO4/NaH2PO4, pH 7.0, for 5 min. GGGCACTTGCAAGCAATTGTGGTCCCAATGGGCTG- Signal enhancement was carried out by incubating the Ј ␮ AGTAT-3 ). The underlined sequences correspond to the wells in 100 L enhancer solution (0.5 g AgNO3 in2mL aptamer sequences, which Green et al. (17, 18) identified H2O, 1.7 g hydroquinone in 30 mL H2O, 2.55 g citric acid, ϫ 14 via the SELEX process from a library of 3 10 molecules and 2.35 g trisodium citrate in 10 mL H2O, all mixed (500 pmol) of single-stranded DNA randomized at 40 together simultaneously immediately before use) at room contiguous positions. PDGF A-chain homodimer (AA), temperature (21). The enhancement time was ϳ60–80 s. PDGF A-chain and B-chain heterodimer (AB), and -B- The reaction was terminated by immersing the micro- chain homodimer (BB) were purchased from Sigma-Al- plates in doubly distilled water. drich. The RayBio Human PDGF-BB ELISA Kit was Although most of the enhancer can be used in daylight, purchased from RayBiotech. Gold NPs, 15 nm in diame- covering the microplates or working under subdued light ter, and avidin were provided by SABC. AgNO3, citric conditions is recommended because high light intensities acid, trisodium citrate, and hydroquinone were pur- promote the self-nucleation of silver ions, which may chased from ShengGong. Enzyme immunoassay micro- confound the results. plates (2592) were obtained from Corning Costar. A microplate reader (BioTek ELx800; Cole-Parmer) was detection of pdgf-bb by elisa used to measure the absorbance at 630 nm. Ten-fold serial dilutions of PDGF-BB (1 amol/L to 10 nmol/L) were detected by ELISA with the Human preparation of gold np–modified PDGF-BB ELISA Kit according to the manufacturer’s 5Ј-alkanethiol–capped aptamer and instructions. In brief, 100 ␮L of PDGF-BB samples were avidin-coated microplates pipetted into wells and incubated for 2.5 h at room Gold NP–modified aptamers were prepared as described temperature. We washed the wells 4 times, added 100 ␮L by Mirkin et al. (19) and modified by Zhang et al. (20) at biotinylated antihuman PDGF-BB antibody, and incu- the storage step at 4 °C. Fifty microliters of Tris-EDTA bated the wells for1hatroom temperature. We then (pH 7.4, containing 10 mmol/L Tris-HCl and 1 mmol/L removed unbound biotinylated antibody with 4 washes, EDTA) buffer containing 300 nmol/L thiol-modified pipetted 100 ␮L horseradish peroxidase–conjugated aptamer probe (sequence 1) was added to the precipitated streptavidin into the wells, and incubated the plates at gold NPs of 500 ␮L gold NPs. We stored the mixture at room temperature for 45 min. The wells were washed 5 4 °C for 12 h, added another 50 ␮L Tris-EDTA buffer times, and 100 ␮L tetramethylbenzidine substrate solu- containing 0.2 mol/L NaCl, and stored the aptamer tion was added to the wells. After incubation for 30 min at mixture at 4 °C for another 24 h. We then purified the room temperature in the dark, 50 ␮L of stop solution was NP-modified aptamer probe by centrifugation at 13 600 g added to each well, and the intensity of the color was for 10 min at 4 °C, washing with ultrapure water, and measured immediately at 450 nm. precipitating the probe twice again to eliminate free oligonucleotides. Finally, the gold NP–modified aptamer Results probe was dissolved in 100 ␮L Tris-EDTA buffer contain- design strategy of the np–modified aptamer ing 0.1 pmol/L NaCl. assay The avidin-coated microplates were prepared by dilut- Fig. 1 illustrates the scheme of the NP-modified aptamer ing avidin in sodium carbonate buffer (0.16 g Na2CO3, assay. It relies on the aptamer pairs binding 2 distinct sites 0.29 g NaHCO3, diluted in ultrapure water to a final on the target protein or to the same epitope of a ho- volume of 100 mL, pH 9.6) to a final concentration of 5 modimer. The gold NP–modified aptamer and the biotin- mg/L, immobilizing the avidin on microplates, and incu- labeled aptamer were incubated with the target protein to bating the plates overnight at 4 °C. form a sandwich complex, which was immobilized onto the microplate via avidin. To facilitate the visualization of detection of pdgf-bb with np-modified aptamer gold NP signals, we designed a signal-amplification pro- We incubated 20 ␮L of 10 nmol/L biotin-labeled aptamer cedure in which silver ions were reduced by hydroqui- probe (sequence 2) and 20 ␮L of an 8 nmol/L solution of none to silver metal at the surface of the gold NPs; this gold NP–modified aptamer probe (sequence 1) at room step enabled visualizing NPs, even at very low concen- Clinical Chemistry 53, No. 6, 2007 1063

fmol/L and 100 pmol/L. The calibration curve was described with the equation: y ϭ kx ϩ b, where y is the absorbance, x is the logarithm of the PDGF-BB concentration, k is the slope, and b is the y intercept. In this study, the equation was: y ϭ 0.1943x ϩ 3.1019, with a correlation coefficient (r2) of 0.9896.

silver enhancement in the np–modified aptamer assay We used silver-enhancement technology to amplify the signals of gold NPs. An image of a microplate containing Fig. 1. Schematic illustration of the NP-modified aptamer assay. different PDGF-BB concentrations in this assay is shown The 1st step in the preparation of gold NP–modified aptamers is the binding of thiol-modified aptamers to gold NPs. The gold NP–modified aptamer and the in Fig. 3A. The background gray scale of the silver- biotin-modified aptamer then bind simultaneously to the target protein to form a enhancer solution increased with time (Fig. 3A). The NPs, sandwich complex, which is immobilized to a microwell with avidin. Finally, the signal is amplified by silver enhancement and recorded with a microplate reader. which serve as nucleation sites for catalysis of silver-ion reduction (22), significantly accelerate the deposition of silver particles. In other words, the photodensity of trations. Deposition of silver on the gold NPs amplifies NP-based silver enhancement was produced in a concen- the absorbance signal, which is recorded densitometri- tration- and time-dependent manner. To determine the cally (Fig. 1). optimal silver-deposition time, we measured the background absorbance with a microplate reader after varying quantification of pdgf-bb with the the silver-deposition time from 0 to 100 s. Blank signals np–modified aptamer assay were low for silver-deposition times of 20–80 s (Fig. 3B); To verify the assay experimentally, we used 10-fold serial however, the background increased rapidly when the dilutions of PDGF-BB (1 amol/L to 10 nmol/L) and an silver-deposition time exceeded 80 s. We thus chose a time 80-s silver-enhancement time to generate a calibration of 60–80 s for silver enhancement. curve. The absorbance at 630 nm was plotted against known PDGF-BB concentrations (Fig. 2); each point rep- reproducibility, recovery, and specificity resents the mean of duplicate measurements of 3 separate We evaluated the method’s intraassay imprecision by experiments. The analytical detection limit, calculated as consecutively analyzing the same samples 5 times and the mean background signal ϩ 2 SD, was 83 amol/L in the assessed interassay imprecision by analyzing the same 50-␮L reaction system. The absorbance at 630 nm was samples on 5 consecutive days. The results are shown in linearly related to PDGF-BB concentration between 1 Table 1. Various known PDGF-BB concentrations were added to 2 biological samples (fetal calf serum and cerebrospinal fluid) and a culture medium (Eagle’s minimal essential medium). Mean (range) recoveries were 98% (86%– 115%). PDGF-AA homodimer and PDGF-AB heterodimer were used to evaluate the specificity of this assay.

Fig. 2. Quantification of a dilution series of PDGF-BB with the NP– modified aptamer assay. The absorbance at 630 nm was plotted against PDGF-BB concentration. Data are Fig. 3. Silver enhancement in the NP–modified aptamer assay. presented as the mean (SD) of triplicate (interassay) and duplicate (intraassay) (A), a microplate image illustrating the detection of different PDGF-BB concen- determinations (inset, calibration curve for PDGF-BB concentration as deter- trations in the NP–modified aptamer assay. (B), time-absorbance relationship in mined with the NP–modified aptamer assay). silver enhancement for various PDGF-BB concentrations. 1064 Li et al.: Cytokine Detection with NP-Modified Aptamers

Table 1. Intraassay and interassay imprecision for the Table 2. PDGF-BB concentrations in sera. NP-modified aptamer assay. NP-modified Coefficient of variations, %a aptamer Samples assay, ng/La ELISA, ng/La P PDGF-BB concentration Intraassay Interassay Healthy controls 23 0.0008 (0.0001) 0.001 (0.0002) Ͼ0.2 100 amol/L 4.8 7.5 Hepatic cirrhosis 35 89.7 (13.7) 90.2 (15.6) Ͼ0.2 100 fmol/L 3.5 5.8 without ascites 100 pmol/L 3.3 4.7 Hepatic cirrhosis 33 180.4 (43.4) 178.6 (60.8) Ͼ0.1 with ascites a n ϭ 5. a Data are presented as the mean (SD).

PDGF-AA and PDGF-AB are 2 other PDGF isoforms. They were each incubated with the PDGF-BB aptamer ascites. The results with the NP–modified aptamer assay and processed in the same way as PDGF-BB. Neither were consistent with those obtained with ELISA. The CVs PDGF-AA nor PDGF-AB was detected in the assay (see of the NP–modified aptamer assay was 12.5%–24.06%, Fig. 1 in the Data Supplement that accompanies the online whereas that of ELISA was 17.29%–34.04%. version of this article at http://www.clinchem.org/ Discussion content/vol53/issue6). Cytokines, products of cells of the immune system, may comparison of the np–modified aptamer assay stimulate the immune response and cause the regression and elisa of some cancers. Cytokine detection and quantification The concentration–response relationships of the NP–mod- play essential roles in clinical practice, as well as in basic ified aptamer assay and ELISA for detecting PDGF-BB are research. Methods currently used for cytokine analysis shown in Fig. 4. The detection limit of the NP–modified include ELISA (5), membrane protein arrays (2), immu- aptamer assay was 83 amol/L, ϳ1000-fold lower than that nohistochemistry (3), electrochemiluminescence (23), of ELISA. The linear range for the NP–modified aptamer sandwich immunoassays with microspheres (4), and flow assay extends over a 106-fold concentration range (1 cytometry (1). Antibodies are currently the most com- fmol/L to 100 pmol/L), whereas the ELISA spans 4 orders monly used affinity reagents for these immunologic as- of magnitude (10 pmol/L to 10 nmol/L). In addition, the says. Molecular probes based on nucleic acid platforms entire NP aptamer assay procedure requires only 1 h, are emerging as complements to antibody-based affinity whereas the ELISA requires ϳ5h. ligands. DNA and RNA sequences that recognize specific target analytes are known as aptamers (24, 25). Although detection of pdgf-bb in serum aptamers are structurally different from antibodies, they We analyzed different serum samples from patients with mimic properties of antibodies in a variety of diagnostic hepatic cirrhosis with both the NP–modified aptamer formats (6). Metallic NPs (e.g., gold NPs) possess various assay and ELISA. The measured PDGF-BB concentrations characteristics (including no radioactivity, high electron are shown in Table 2. PDGF-BB concentrations in the density, and high sensitivity) that are extensively ex- hepatic cirrhosis patients with and without ascites were ploited in different bioanalytical applications (26–28). significantly higher than in the healthy controls, and Taking advantage of these properties of aptamers and PDGF-BB concentrations in the hepatic cirrhosis patients gold NPs, we developed a novel densitometric assay for with ascites were much higher than in those without cytokines that is based on NP-modified aptamers. The signals of gold NPs are amplified through the use of silver-enhancement technology, which is based on the use of hydroquinone as the reducing agent. Hydroquinone facilitates the colloidal gold–catalyzed reduction of silver ions (from silver lactate or silver acetate) to metallic silver on the particle surface (29). Enhancement occurs via the deposition of silver from the solution onto the surface of gold NPs. Each colloidal gold particle thus acts as a catalytic site, and the shell of metallic silver that forms around the gold particles also catalyzes the reaction, producing multiple layers of silver atoms around the gold NPs. Furthermore, silver deposition is proportional to the concentration of the gold NPs, which in turn is proportional to the concentration of the target protein. The Fig. 4. Comparison of the PDGF-BB concentration–response curves for power of the silver-enhancement effect is such that a low the NP–modified aptamer assay and ELISA. degree of binding to NPs becomes detectable, greatly Data are presented as the mean (SD) of 3 independent determinations. promoting the assay’s sensitivity (29). Use of silver- Clinical Chemistry 53, No. 6, 2007 1065

enhancement technology enables the detection of concen- basic fibroblast growth factor (36), l-selectin (37), inter- trations as low as 83 amol/L PDGF-BB, and the detection feron-␥ (38), and others. We believe this method has a range of the assay can reach 7 orders of magnitude. promising future in the field of protein detection. The sensitivity of this method is superior to that of the reported aptamer-functionalized gold-NP assay for the In conclusion, this approach has attractive performance optical detection of thrombin with a quartz crystal mi- features, such as simplicity and high sensitivity. The crobalance (detection limit, 2 nmol/L) (30) and is close to method may contribute to the growing number of appli- that of 2 other assays that use amplification approaches: cations of nanotechnology and may find a use in diagnos- multiplexed protein profiling on microarrays by rolling- tics and therapeutics. circle amplification (31) and colorimetric bio-barcode amplification (32). In addition, the absorbance of NP-based silver en- Grant/funding support: National Natural Science Foun- hancement depends on dose and time. Prolonging the dation of China Grants 20377017 and 20677018; Hi-Tech silver-deposition time increases the background of the Research and Development Program of China Grant gray scale of the silver-enhancer solution; thus, the silver- 2006AA06Z401. deposition time is a critical factor for this method. Financial disclosures: None declared. This method has a sensitivity comparable to the sensitivities reported for PCR-based methods (14, 15); how- References 1. Ao X, Sellati TJ, Stenken JA. Enhanced microdialysis relative ever, our assay is more convenient and does not require recovery of inflammatory cytokines using antibody-coated micro- complicated instrumentation. The entire procedure is spheres analyzed by flow cytometry. Anal Chem 2004;76:3777– similar to ELISA, and the prerequisite instrument, a 84. microplate reader, is available in almost all laboratories. 2. Copeland S, Siddiqui J, Remick D. Direct comparison of traditional Compared with immunologic methods for cytokine de- ELISAs and membrane protein arrays for detection and quantifi- tection such as ELISA, our assay with an NP-modified cation of human cytokines. J Immunol Methods 2004;284:99– aptamer possesses significantly increased sensitivity. 106. Moreover, it uses aptamers as affinity ligands, which are 3. Nagashima G, Suzuki R, Asai J, Fujimoto T. Immunohistochemical analysis of reactive astrocytes around glioblastoma: an immuno- simpler and more flexible to design, prepare, and apply. histochemical study of postmortem glioblastoma cases. Clin Compared with the batch-to-batch variability of antibody Neurol Neurosurg 2002;104:125–31. production, which relies on animals or cell cultures, 4. Prabhakar U, Eirikis E, Miller BE, Davis HM. Multiplexed cytokine aptamers are produced by chemical synthesis with very sandwich immunoassays: clinical applications. Methods Mol Med high accuracy and reproducibility (6). These synthetic 2005;114:223–32. oligonucleotides have greater temperature stabilities and 5. Triolo G, Accardo-Palumbo A, Dieli F, Ciccia F, Ferrante A, Giardina longer shelf lives than most antibodies (33). Aptamer E, et al. Humoral and cell mediated immune response to cow’s milk proteins in Behcet’s disease. Ann Rheum Dis 2002;61:459– isolation can be automated and is readily scaled up (34). 62. Moreover, aptamer probes can be shared within the 6. Jayasena SD. Aptamers: an emerging class of molecules that rival research community via publication of the nucleotide antibodies in diagnostics. Clin Chem 1999;45:1628–50. sequence, thereby facilitating the standardization of pro- 7. Hermann T, Patel DJ. Adaptive recognition by nucleic acid aptam- tein assays between laboratories. ers. Science 2000;287:820–5. On the other hand, several limitations also exist. Light 8. German I, Buchanan DD, Kennedy RT. Aptamers as ligands in is an interference factor in silver enhancement, because affinity probe capillary electrophoresis. Anal Chem 1998;70: high light intensity will promote the self-nucleation of 4540–5. silver ions, which may obscure the results. We therefore 9. Fang X, Cao Z, Beck T, Tan W. Molecular aptamer for real-time oncoprotein platelet-derived growth factor monitoring by fluores- recommend that microplates be covered or experiments cence anisotropy. Anal Chem 2001;73:5752–7. be carried out under a red safe light. In addition, higher 10. Fang X, Mi Y, Li JJ, Beck T, Schuster S, Tan W. Molecular beacons: temperatures and movement of the microplates during fluorogenic probes for living cell study. Cell Biochem Biophys incubation in the enhancer solution may speed up silver 2002;37:71–81. deposition on the gold surface. Another limitation resides 11. Hamaguchi N, Ellington A, Stanton M. Aptamer beacons for the in the aptamers themselves. Currently, high-affinity direct detection of proteins. Anal Biochem 2001;294:126–31. aptamers are not as widely available as antibodies. The 12. Li JJ, Fang X, Tan W. Molecular aptamer beacons for real-time SELEX process is evolving, however, and new selection protein recognition. Biochem Biophys Res Commun 2002;292: protocols will facilitate more efficient aptamer selection 31–40. 13. Yamamoto R, Baba T, Kumar PK. Molecular beacon aptamer and create a more diverse aptamer pool, both of which fluoresces in the presence of Tat protein of HIV-1. Genes Cells will greatly extend the application of this new method. To 2000;5:389–96. date, several aptamers that have good affinity and good 14. Fredriksson S, Gullberg M, Jarvius J, Olsson C, Pietras K, Gustafs- specificity have been characterized for a number of cyto- dottir SM, et al. Protein detection using proximity-dependent DNA kines, including vascular endothelial growth factor (35), ligation assays. Nat Biotechnol 2002;20:473–7. 1066 Li et al.: Cytokine Detection with NP-Modified Aptamers

15. Wang XL, Li F, Su YH, Sun X, Li XB, Schluesener HJ, et al. 27. Stoeva SI, Lee JS, Smith JE, Rosen ST, Mirkin CA. Multiplexed Ultrasensitive detection of protein using an aptamer-based exo- detection of protein cancer markers with biobarcoded nanopar- nuclease protection assay. Anal Chem 2004;76:5605–10. ticle probes. J Am Chem Soc 2006;128:8378–9. 16. Huang CC, Huang YF, Cao Z, Tan W, Chang HT. Aptamer- 28. Rosi NL, Mirkin CA. Nanostructures in biodiagnostics. Chem Rev modified gold nanoparticles for colorimetric determination of 2005;105:1547–62. platelet-derived growth factors and their receptors. Anal Chem 29. Hernandez-Santos D, Gonzalez-Garcia MB, Costa-Garcia A. Elec- 2005;77:5735–41. trochemical determination of gold nanoparticles in colloidal solu- 17. Green LS, Jellinek D, Jenison R, Ostman A, Heldin CH, Janjic N. tions. Electrochim Acta 2000;46:607–15. Inhibitory DNA ligands to platelet-derived growth factor B-chain. 30. Pavlov V, Xiao Y, Shlyahovsky B, Willner I. Aptamer-functionalized Biochemistry 1996;35:14413–24. Au nanoparticles for the amplified optical detection of thrombin. 18. Green LS, Bell C, Janjic N. Aptamers as reagents for high- J Am Chem Soc 2004;126:11768–9. throughput screening. Biotechniques 2001;30:1094–6, 1098, 31. Nam JM, Wise AR, Groves JT. Colorimetric bio-barcode amplifica- 1100 passim. tion assay for cytokines. Anal Chem 2005;77:6985–8. 19. Mirkin CA, Letsinger RL, Mucic RC, Storhoff JJ. A DNA-based 32. Schweitzer B, Roberts S, Grimwade B, Shao W, Wang M, Fu Q, et method for rationally assembling nanoparticles into macroscopic al. Multiplexed protein profiling on microarrays by rolling-circle materials. Nature 1996;382:607–9. amplification. Nat Biotechnol 2002;20:359–65. 33. O’Sullivan CK. Aptasensors: the future of biosensing? Anal Bio- 20. Zhang ZL, Pang DW, Yuan H, Cai RX, Abruna HD. Electrochemical anal Chem 2002;372:44–8. DNA sensing based on gold nanoparticle amplification. Anal 34. Drolet DW, Jenison RD, Smith DE, Pratt D, Hicke BJ. A high Bioanal Chem 2005;381:833–8. throughput platform for systematic evolution of ligands by expo- 21. Nie L, Tang J, Guo H, Chen H, Xiao P, He N. Colorimetric detection nential enrichment (SELEX). Comb Chem High Throughput Screen of polynucleotides on polypropylene slices. Anal Sci 2004;20: 1999;2:271–8. 461–3. 35. Ng EW, Shima DT, Calias P, Cunningham ET Jr, Guyer DR, Adamis 22. Su X, Li SFY, O’Shea SJ. Au nanoparticle- and silver-enhancement AP. Pegaptanib, a targeted anti-VEGF aptamer for ocular vascular reaction-amplified microgravimetric biosensor. Chem Commun disease. Nat Rev Drug Discov 2006;5:123–32. 2001:755–6. 36. Jellinek D, Lynott CK, Rifkin DB, Janjic N. High-affinity RNA ligands 23. Sennikov SV, Krysov SV, Injelevskaya TV, Silkov AN, Grishina LV, to basic fibroblast growth factor inhibit receptor binding. Proc Natl Kozlov VA. Quantitative analysis of human immunoregulatory Acad Sci U S A 1993;90:11227–31. cytokines by electrochemiluminescence method. J Immunol Meth- 37. Hicke BJ, Watson SR, Koenig A, Lynott CK, Bargatze RF, Chang YF, ods 2003;275:81–8. et al. DNA aptamers block L-selectin function in vivo. Inhibition of 24. Joyce GF. Amplification, mutation and selection of catalytic RNA. human lymphocyte trafficking in SCID mice. J Clin Invest 1996; Gene 1989;82:83–7. 98:2688–92. 25. Ellington AD, Szostak JW. In vitro selection of RNA molecules that 38. Ramanathan M, Lantz M, MacGregor RD, Huey B, Tam S, Li Y, bind specific ligands. Nature 1990;346:818–22. et al. Inhibition of interferon-gamma-induced major histocompati- 26. Niemeyer CM, Ceyhan B. DNA-directed functionalization of colloi- bility complex class I expression by certain oligodeoxynucleotides. dal gold with proteins. Angew Chem Int Ed Engl 2001;40:3685–8. Transplantation 1994;57:612–5. Clinical Chemistry 53:6 1067–1074 (2007) Proteomics and Protein Markers

A Novel, High-Throughput Workflow for Discovery and Identification of Serum Carrier Protein-Bound Peptide Biomarker Candidates in Ovarian Cancer Samples

Mary F. Lopez,1* Alvydas Mikulskis,1 Scott Kuzdzal,1 Eva Golenko,1 Emanuel F. Petricoin III,2 Lance A. Liotta,2 Wayne F. Patton,1 Gordon R. Whiteley,3 Kevin Rosenblatt,4 Prem Gurnani,4 Animesh Nandi,4 Samuel Neill,5 Stuart Cullen,5 Martin O’Gorman,5 David Sarracino,6 Christopher Lynch,1 Andrew Johnson,1 William Mckenzie,1 and David Fishman6

Background: Most cases of ovarian cancer are detected ovarian cancer, with positive results in >93% of samples at later stages when the 5-year survival is ϳ15%, but from patients with disease-negative results and in 97% 5-year survival approaches 90% when the cancer is of disease-free controls. The carrier protein–based ap- detected early (stage I). To use mass spectrometry (MS) proach identified additional protein fragments, many of serum proteins for early detection, a seamless work- from low-abundance proteins or proteins not previously flow is needed that provides an opportunity for rapid seen in serum. profiling along with direct identification of the under- Conclusions: This workflow system using a highly pinning ions. reproducible, high-resolution MALDI-TOF platform en- Methods: We used carrier protein–bound affinity enables rapid enrichment and profiling of large numbers richment of serum samples directly coupled with of clinical samples for discovery of ion signatures and MALDI orthagonal TOF MS profiling to rapidly integration of direct sequencing and identification of search for potential ion signatures that contained the ions without need for additional offline, time- discriminatory power. These ions were subsequent- consuming purification strategies. ly directly subjected to tandem MS for sequence © 2007 American Association for Clinical Chemistry identification. Results: We discovered several biomarker panels that The need for development of early screening methods for enabled differentiation of stage I ovarian cancer from epithelial ovarian carcinoma is particularly urgent. Ovar- unaffected (age-matched) patients with no evidence of ian cancer is the 4th leading cause of cancer-related deaths among women in the United States (1, 2). Regrettably, 70%–75% of new cases are diagnosed in stage III or IV, ϳ 1 PerkinElmer Life and Analytical Sciences, Wellesley, MA. with a predicted 5-year survival of 15% (3). The 5-year 2 Center for Applied Proteomics and Molecular Medicine, George Mason survival approaches 90%, however, if cancer is detected University, Manassas, VA. when confined to the ovary (stage I) (3). More sensitive 3 Clinical Proteomics Reference Laboratory, Gaithersburg, MD. and specific tests for earlier detection may improve pa- 4 Department of Pathology, Division of Translational Pathology, Univer- sity of Texas Southwestern Medical Center, Dallas, TX. tient survival rates by facilitating early treatments such as 5 Nonlinear Dynamics, Cuthbert House, All Saints, Newcastle upon Tyne, surgical intervention (4–8). United Kingdom. Recently, new methods of disease detection based on 6 Harvard Partners, Cambridge, MA. 7 New York University Medical School, Division of Obstetrics and Gyne- discriminant mass spectral analysis (serum pattern profil- cology, New York, NY. ing) have been proposed (9–14). The power of this * Address correspondence to this author at: PerkinElmer Life and Analyt- approach is 4-fold: (a) it is unbiased and does not presup- ical Sciences, 45 Williams St., Wellesley, MA 02481-4008. Fax 617-574-9864; pose any particular disease mechanism, (b) multiple dif- e-mail [email protected]. Received September 29, 2006; accepted March 27, 2007. ferences, i.e., multiple putative disease markers, are often Previously published online at DOI: 10.1373/clinchem.2006.080721 discovered and combinations of markers are likely to be

1067 1068 Lopez et al.: Serum Profiling and Sequencing of Cancer Biomarkers

more powerful discriminators than single markers, (c) of diseased and normal tissues, and the low molecular large numbers of samples (usually blood sera) from ap- weight peptide archive can be viewed as a direct reflec- propriate cohorts can be analyzed quickly for discovery tion of the ongoing pathophysiology that can facilitate a and subsequent validation of putative marker sets, and (d) true systems biology approach to biomarker discovery. the method does not require a priori antibody develop- Thus, adsorption to albumin serendipitously provides an ment for success. endogenous and very efficient enrichment process for Serum peptide pattern profiling studies have shown rare or low-abundance protein fragments. Unfortunately, promise, especially in clinical research (13, 14). The ap- many potentially interesting biomarkers are likely to be proach has, however, generated controversy centered inadvertently eliminated during commonly used strate- primarily on 2 issues: the relative importance of obtaining gies for the removal of albumin and other highly abun- definitive sequence identification for differentiating dant proteins in blood serum, and sample fractionation masses and the likelihood that peptides or protein frag- protocols that omit the capture of carrier protein-bound ments, as opposed to intact proteins, can be useful as peptides may yield preparations consisting almost exclu- disease biomarkers (14–17). The clinical relevance of the sively of peptides derived from blood coagulation cas- serum peptidome has been vigorously debated, but recent cades (30, 31). publications have confirmed that specific protein frag- To develop a new workflow for biomarker candidate ments are correlated with disease stages (18, 19). A major identification, we evaluated high-throughput carrier pro- area of controversy has been the lack of data consistency tein-bound affinity enrichment of serum samples coupled and reproducibility across the various published studies with high-resolution MALDI orthogonal TOF (OTOF) MS, (17), although with proper attention to stringent experi- discriminant analysis of the resulting mass spectral pat- mental design and protocols, these issues can be ad- terns, and sequence identification of the discriminating dressed in future studies (20). High resolution and mass ions to search for putative early protein/peptide biomar- accuracy are required for accurate comparison of mass kers in ovarian cancer serum samples. spectral data across a broad mass range. Sample preparation is another area of critical importance. Reduction of Materials and Methods sample complexity is an essential first step for all blood– clinical serum samples borne biomarker discovery methods because of the large Serum samples were obtained from study participants dynamic range of protein concentrations (21). Low- with full consent and institutional review board approval throughput methods such as 2-dimensional gels coupled and collected before physical evaluation, diagnosis, and with mass spectrometry (MS)8 or shotgun proteomics treatment. The study set samples are shown in Table 1. have been used to mine the serum proteome for biomar- Serum samples from cancer patients were obtained from kers (22–25), but these methods have limited clinical the National Ovarian Cancer Early Detection Program value because they do not allow the simultaneous screen- and the gynecologic oncology clinic at Northwestern ing of adequate (large) numbers of samples. Numerous University (Chicago, IL). Cancer samples were stage I–IV. depletion strategies have been developed to remove the The apparently healthy (nondisease) serum samples more common, abundant proteins in sera or plasma (26). were collected from Northwestern University; Innsbruck However, many of these proteins (in part because of their Medical University, Innsbruck, Austria; and Johns Hop- high abundance in the blood), act as carrier proteins and kins University, Baltimore, MD, from age-matched bind a vast assortment of peptides and protein fragments women without any evidence of cancer for 5 years before (27). With an albumin blood concentration Ͼ600 ␮mol/L, sample collection. All samples were collected from 1999 to the probability is Ͼ98% that even molecules with rela- 2002 and stored until analysis in 2005. The mean age for tively low binding affinities will be complexed with all study participants (cancer patients and healthy indi- albumin (19). These carrier protein-bound protein frag- viduals) was ϳ40 years for all groups. To minimize ments and peptides may provide potential diagnostic institutional bias attributable to sample handling tech- information for many diseases (18, 28, 29). Peptides and niques (15), all samples were collected and handled with protein fragments from proteins catabolized by proteo- the same standard operating procedures. Ovarian cancer lytic cascades in tissues diffuse into the blood and are samples were obtained from patients before therapy and bound to highly abundant proteins such as albumin. This surgery; all patients were surgically staged. There was a process effectively prolongs the bloodstream half-life of predominance of serous cases (54.9%) with 7.1% clear cell low molecular weight peptides and protein fragments that otherwise would be eliminated by the kidneys. The Table 1. Clinical samples.a collection of peptides and protein fragments bound to carrier proteins therefore provides a metabolic snapshot Healthy Total cancer High CA125 Low CA125 Samples 110 453 214 239 Spectra 330 1359 642 717 a Samples were classified as high CA125 if the concentrations were 8 Nonstandard abbreviations: MS, mass spectrometry; OTOF, orthogonal Ͼ22 kIU/L, and low CA125 if Ͻ22 kIU/L (see Materials and Methods). TOF. Clinical Chemistry 53, No. 6, 2007 1069

and the remaining 38.0% listed only as adenocarcinoma. ultrahigh resolution tandem ms Each sample was accompanied by a verified pathologic Pools (6) of either ovarian cancer samples or healthy diagnosis. samples were dissolved in 50 ␮L of 5% acetonitrile 0.1% All serum samples were processed from blood drawn formic acid/water, and transferred to an MS plate and under strict National Cancer Institute/US Food and Drug lyophilized. Samples in 5% acetonitrile 0.1% formic acid Administration Proteomics Program standard operating were injected with a Famos Autosampler onto a 75 ␮m ϫ guidelines as follows: Specimens were collected in red-top 18 cm fused silica capillary column packed with C18 or C8 Vacutainer tubes and allowed to clot for 1 h on ice, media, in a 250-␮L/min gradient of 5% acetonitrile 0.1% followed by centrifugation at 4 °C for 10 min at 2000g. The formic acid to 50% acetonitrile 0.1% formic acid over the serum supernatant was divided into aliquots and stored course of 100 min with a total run length of 150 min. For at Ϫ80 °C until needed. The serum samples were assayed the 240-min runs, 25-cm columns were used to achieve for CA125 by use of the Elecsys CA125 reagent set (Roche higher chromatographic resolution and loading capacity. Pharmaceuticals). Samples with a CA125 concentration The LTQ-FT ultrahybrid mass spectrometer (Thermo Ͼ22 kIU/L were classified as high CA125 and those with Fisher Scientific) was run in a top 4 configuration at 200 K ϩ a concentration Ͻ22 kIU/L as low CA125. We selected resolution for a full scan. Ions that were 1 or undefined 22 kIU/L because it was the mean of the samples in the in charge states were rejected for MS 2 analysis. Dynamic study; CA125 Ͼ35 kIU/L is generally considered to be the exclusion was set to 1 with a limit of 180 s, with early cutoff indicating likely disease recurrence. CA125 was expiration set to 6 full scans. Peptide identification was below the limit of detection in the healthy controls. performed with Sequest through the Bioworks Browser 3.2 EF2 (Thermo Scientific). Database searches were made sample processing with a no-enzyme indexed version of the National Center Samples from cancer patients and healthy controls were for Biotechnology Information RefSeqhuman/reversed processed in a random order to account for any systematic Refseqhuman database using differential oxidized methi- onines at a tolerance of 10 ppm. Peptide score cutoff errors and variations from experiment to experiment. values were chosen at Xcorr of 1.8 for singly charged ions, Locations of all samples were random on each MALDI 2.0 for doubly charged ions, and 2.5 for triply charged plate. ions, along with deltaCN values of Ն0.1, and rank score Serum samples were processed using prototype preliminary values of Ͻ10 with a peptide P value of 1eϪ3 ProXPRESSION௢ biomarker enrichment reagent sets or better. The small mass tolerance of the search ensured (PerkinElmer) as described in (28). The Cibachron blue that only relevant peptides were matched. The cross- dye affinity chromatography-based technology is de- correlation values chosen for each peptide assured a high signed to capture high-abundance carrier proteins in confidence match for the different charge states, and the blood (such as albumin) and dramatically enriches the deltaCN cutoff insured the uniqueness of the peptide hit. peptide and protein fragments bound to the carrier pro- ௢ The P value is a probability score for a random hit teins. ZipPlates and a vacuum manifold were purchased peptide. Typically, multiple peptide hits were obtained from Millipore. Millipore also provided custom-fitting for any identified protein, for example, more than 28 adapters for direct spotting of samples on single use separate hits were obtained for plasma kallikrein-sensi- ௢ MALDIchip prOTOF Target plates (PerkinElmer). Pre- tive glycoprotein (data not shown). However, there were mixed alpha-cyano-4-hydroxycinnamic acid matrix was also a number of proteins that were identified by single from PerkinElmer. peptide hits.

maldi-otof ms processing and analysis of spectral profiles ௢ ௢ Mass spectra were acquired on a prOTOF 2000 MALDI- Progenesis PG600 software (NonLinear Dynamics) was ௢ OTOF Mass Spectrometer interfaced with TOFWorks used to process and analyze the OTOF mass spectral data. software (PerkinElmer/SCIEX). Because of the orthogonal Raw spectra from the OTOF were directly loaded into the design, a single external mass calibrant was used to PG600 program using the prOTOF loader program. Bin- achieve better than 5 ppm mass accuracy over an entire ning was set at 4. Analyses were performed to find sample plate (up to 384 samples). In this study, a 2-point discriminant markers between the following groups: external calibration of the prOTOF instrument was per- healthy vs all cancer, healthy vs high CA125, healthy vs formed before acquiring the spectra in a batch mode from low CA125, and healthy vs stage I cancer. For the initial 96 samples. MALDI-OTOF-MS can collect data over a analysis, the stringency parameters for biomarker selec- wide range of mass values (300 kDa) in a single acquisition were set to include peaks with an mean quantity tion. Typical resolution for peptides and proteins up to threshold of Ն75 (higher intensity peaks, to facilitate 10 kDa was Ͼ12 000 full width at half maximum. subsequent sequence identification by tandem MS) and P The raw mass spectral data used in this study are Յ0.01. Subsequent analyses were performed with peak accessible without restriction upon request to the corre- intensity stringency of Ͻ50 or 0 and a P Յ0.01. These sponding authors. parameters ensured the detection of differently expressed 1070 Lopez et al.: Serum Profiling and Sequencing of Cancer Biomarkers

peaks that were highly significant. Once the putative are shown in Fig. 1. All samples were processed in a peaks were detected, classification models were devel- high-throughput, parallel manner to obtain the informa- oped using flexible discriminant analysis and the R station-rich mass spectra. Spectra from the various groups tistical package (32). The flexible discriminant analysis were compared and analyzed, and the discriminant pep- algorithm determined nonlinear decision boundaries that tide masses were identified. Subsequently, pooled disease were better able to separate classes, resulting in a classi- and healthy serum samples were processed for peptide fication technique that was more powerful for high- enrichment and then submitted for de novo sequence dimensional data with complex interrelationships. We analysis by ultrahigh resolution tandem-MS. This proce- used independent stratified balanced random sampling to dure was efficient and allowed rapid (hours to days) split the data into a training set and a test set. The training discovery and identification of putative disease markers set was used to build a classifier model, and this model from large numbers of samples. Two advantages to this was evaluated on the test set. The classifier classified test approach were as follows: (i) a large number of samples cases as healthy or diseased, and these data were then could be analyzed simultaneously, lending statistical rel- used to create ROC curves. Monte Carlo cross-validation evance to the putative markers, and (ii) sequence identi- of training and test sets was used (33, 34). For this fication was directly obtained from the same samples, validation, the results of 100 runs of the sampling and increasing confidence in marker identification accuracy classifier modeling procedure were averaged together to and dispelling the need for further purification by gels or create the final ROC curve. other methods.

Results analysis and identification of discriminant development of study design and workflow peptide masses and putative biomarkers When processed with the carrier protein-based biomarker An initial set of 9 discriminating peptides resulted from enrichment protocol, serum samples routinely generated the initial analysis comparing the spectral profiles from 4 highly reproducible peptide profiles with intensity CVs of different groups: healthy vs all cancer (low CA125 ϩ high 5%–10%, on average (28). The workflow and study design CA125), healthy vs low CA125, healthy vs high CA125,

Fig. 1. Serum based biomarker discovery workflow. Serum is processed through albumin capture ProXPRESSION 96-well plates. Fragments bound to albumin are eluted, concentrated through ZipPlate, and analyzed by MALDI OTOF MS and discriminant analysis. The differentially expressed fragments are identified by sequence analysis using tandem MS. Clinical Chemistry 53, No. 6, 2007 1071

and healthy vs stage I cancer (Table 2). The identities of peptides, or peptides not related to coagulation, we low- these peptides included multiple peptide hits from com- ered the intensity stringency of the analysis to 0, keeping plement component 3 and interalpha (globulin) inhibitor the P values at 0.01 or better. Five additional discriminat- H4, and single peptides from complement component 4A, ing masses resulted from this analysis (Tables 2 and 3). transthyretin, and fibrinogen (Table 3). Because the anal- None of the identified proteins in this set are related to ysis parameters were purposefully set to screen out low- coagulation, and all are correlated with or involved in intensity masses (to improve our success rate with se- cellular oncogenenesis [casein kinase 2, transgelin (35– quence identification), it is not surprising that these 37)], proliferation [keratin 2, LARGE (38, 39)], or detoxi- fragments are derived from relatively abundant serum fication of ROS [diamine oxidase (40)]. A model created proteins. Although these peptides were derived from with these 5 markers plus 2 additional markers from the highly abundant resident proteins, they are not necessar- 4-marker set classified the healthy and low CA125 sam- ily nonspecific or themselves highly abundant. Examples ples with 77% sensitivity and 85% specificity (Table 2, of overlaid disease and healthy spectra are shown in 7-marker model). Fig. 2; putative marker expression ranged from an ϳ3.6- fold increase to a 2.6-fold decrease in cancer samples additional protein fragments detected in (Table 2). To test the discriminating power of the 9-marker cancer sera set, a flexible discriminant analysis classification model In addition to the discriminating protein fragments iden- was built and tested. The discriminating power of the tified above, additional sequence identities were obtained 9-marker model was quite good; 93% of samples from patients with disease were positive and 93% of samples (162 total) for other peptides/proteins in the ovarian from disease-free controls were negative (Table 2). cancer serum samples (see Table 1 in the Data Supple- To expand the analysis, we reduced the stringency ment that accompanies the online version of this article at parameter slightly to allow the inclusion of lower inten- http://www.clinchem.org/content/vol53/issue6). Many sity discriminating masses. The results of this analysis of these fragments were not detected in the healthy sera. yielded the set of 4 markers shown in Tables 2 and 3. Two We could not be certain that the fragments were exclusive of the peptides at m/z 1739.9 and 2582.35 were also in the to the cancer sera, however, because we extensively initial set of 9, and the remaining 2 masses at m/z 2659.27 sequenced peptides from only a limited number of pooled and 2989.49 remained unidentified. The 4-marker model samples. Further sequencing experiments may determine delivered equivalent diagnostic sensitivity (93%) and bet- the exclusivity of these peptides to disease or healthy ter specificity (97%) than the 9-marker model (Table 2). In samples and their potential as putative disease biomark- an effort to discover lower abundance discriminating ers. Many of the proteins from which these peptides are

Table 2. Healthy vs cancer.a Diagnostic Diagnostic % Fold sensitivity, specificity, Area under Prediction Model m/z expression P % (SD) % (SD) ROC curve error 9 1690.94 Ϫ2.06 0.0000000 93 (2.5) 93 (1.7) 0.98 (0.01) 7.0 (1.3) 1739.93 Ϫ2.62 0.0000000 1777.97 Ϫ2.03 0.0000000 1865.01 Ϫ2.03 0.0000000 2021.11 Ϫ1.51 0.0000000 2582.35 3.38 0.0000000 2898.54 2.01 0.0000000 3027.57 2.59 0.0000000 3239.55 2.10 0.0000000 4 1739.93 Ϫ2.62 0.0000000 93 (2.0) 97 (0.8) 0.98 (0.01) 4.3 (0.7) 2582.35 3.38 0.0000000 2659.27 2.00 0.0000000 2989.49 2.23 0.0000000 7 1739.93 Ϫ2.62 0.0000000 77 (4.3) 85 (3.0) 0.87 (0.02) 16.9 (1.9) 2582.35 3.38 0.0000000 1966.91 Ϫ1.21 0.0000000 1041.68 1.34 0.0000000 2115.05 Ϫ1.11 0.0137930 1224.68 1.87 0.0000000 2345.19 2.07 0.0000000 a Marker model expression and classification (healthy vs low CA125 samples). 1072 Lopez et al.: Serum Profiling and Sequencing of Cancer Biomarkers

Table 3. Sequence identifications of putative ovarian cancer peptide biomarkers. m/z Ppeptide sequence P (pep) Iidentity 1041.68 L.NVKVDPEIQ.N 4.1876E-05 gi 47132620 ref NP_000414.2 keratin 2a ͓Homo sapiens͔͓MASS ϭ 65432͔ 1224.68 L.KPRVSWIPNK.H 2.54E-04 gi 33285008 ref NP_689525.2 glycosyltransferase-like 1B ͓Homo sapiens͔ 1690.94 S.KITHRIHWESASLL.R 5.40E-10 gi 4557385 ref NP_000055.1 complement component 3 precursor ͓Homo sapiens͔ 1739.93 R.NGFKSHALQLNNRQI.R 9.96E-09 gi 67190748 ref NP_009224.2 complement component 4A preproprotein ͓Homo sapiens͔ 1777.97 S.SKITHRIHWESASLL.R 3.07E-11 gi 4557385 ref NP_000055.1 complement component 3 precursor ͓Homo sapiens͔ 1865.01 R.SSKITHRIHWESASLL.R 1.35E-09 gi 4557385 ref NP_000055.1 complement component 3 precursor ͓Homo sapiens͔ 1966.91 T.DVNTHRPREYWDYES.H 3.0179E-05 gi 29570791 ref NP_808227.1 casein kinase II alpha 1 subunit isoform a ͓Homo sapiens͔ 2021.11 R.SSKITHRIHWESASLLR.S 1.13E-10 gi 4557385 ref NP_000055.1 complement component 3 precursor ͓Homo sapiens͔ 2115.05 A.REGADVIVNCTGVWAGALQR.D 0.00061097 gi 21536470 ref NP_001908.2 D-amino-acid oxidase ͓Homo sapiens͔͓MASS ϭ 39496͔ 2345.19 Q.M*GTNRGASQAGM*TGYGM*PRQIL.- 1.9783E-05 gi 4507357 ref NP_003555.1 transgelin 2 ͓Homo sapiens͔͓MASS ϭ 22391͔ 2582.35 R.NVHSGSTFFKYYLQGAKIPKPEA.S 3.9553E-06 gi 31542984 ref NP_002209.2 inter-alpha (globulin) inhibitor H4 (plasma Kallikrein- sensitive glyco 2898.54 G.PRRYTIAALLSPYSYSTTAVVTNPKE.- 0.00173243 gi 4507725 ref NP_000362.1 transthyretin ͓Homo sapiens͔͓MASS ϭ 15887͔ 3027.57 N.FRPGVLSSRQLGLPGPPDVPDHAAYHPF.R 2.9701E-06 gi 31542984 ref NP_002209.2 inter-alpha (globulin) inhibitor H4 (plasma Kallikrein- sensitive glyco 3239.55 K.SYKMADEAGSEADHEGTHSTKRGHAKSRPV.R 0.00140644 gi 11761629 ref NP_068657.1 fibrinogen, alpha chain isoform alpha preproprotein ͓Homo sapiens͔ derived are present in very low abundance or have not dence supports the association between activation of been identified previously in blood (41). blood coagulation and progression of cancer. Recent studies from several laboratories have linked malignant trans- Discussion formation (oncogenesis), tumor angiogenesis, and metas- The primary aim of this study was to investigate the tasis to the generation of clotting intermediates, clotting or utility of a new workflow method for biomarker discov- platelet function inhibitors, or fibrinolysis inhibitors (42). ery. As a means to this end, we used a large and Many researchers have published putative markers for well-characterized set of serum samples that reflected an cancer, and ovarian cancer in particular, that are related to urgent clinical need. The new workflow we describe for coagulation or inflammation (43). Interestingly, when we discovery of potentially clinically-relevant biomarkers lowered the stringency of our analysis to allow the from large, statistically powered study sets incorporates inclusion of very low intensity discriminating signals, we MS profiling with an extremely stable MALDI-TOF plat- identified a further 5 peptides not related to either coag- form coupled to an automated process for albumin-bound ulation or inflammation pathways. Among these lower peptide enrichment and direct sequencing and identifica- intensity peptides, casein kinase 2 is oncogenic and up- tion of ion peaks by high-resolution tandem MS without regulated in tumors (35), and trangelin has been reported the need for additional chromatography or enrichment. previously as a putative marker for ovarian and endome- Thus the discriminating ions are directly identified and trial cancer in other studies using widely different discov- sequenced. ery methods including LC-MS and cDNA-representa- The carrier protein–based approach yielded a number tional difference analysis (36, 37). The other 3, keratin 2, of discriminating peptides, and we built flexible discrimi- glycosyl transferase (LARGE), and diamino oxidase are nant models with multiple marker sets (9, 4, and 7 also associated with processes related to cancer (38–40). markers). These models enabled classification with high In addition to the discriminating peptides described specificity and sensitivity of samples from cancer patients above, a rich trove of protein fragments, many from and healthy controls. Perhaps not surprisingly, peptide low-abundance proteins or proteins not previously seen fragments associated with the coagulation cascade pro- in serum (41), were recovered from the ovarian cancer vided the highest classification power. Substantial evi- sera. These results are consistent with those reported by Clinical Chemistry 53, No. 6, 2007 1073

their functions are described in Table 2 of the online Data Supplement. Several proteins identified in this study (e.g. transthyretin) have also been reported by others as putative biomarkers for ovarian cancer (10, 45). Interestingly, the transthyretin fragment reported herein is a different unique mass than that reported in a previous serum- based ovarian cancer study (10). The proteins we identified are involved in cellular inflammation, differentiation, signaling, apoptosis, transcriptional regulation, and other regulatory mechanisms. It is remarkable that this rich variety of low-abundance species is so well represented in the fraction bound to serum albumin.

In summary, in a period of 2–3 weeks we identified ϳ162 proteins from peptides and protein fragments bound to carrier proteins from ovarian cancer patient serum samples. Within this study, 3 sets of the discriminating carrier-protein bound fragments differentiated samples from patients with ovarian cancer and from apparently healthy controls with sensitivities and specificities of up to 93% and 97%, respectively. These values compare very favorably with published mean sensitivities and specificities of ϳ50% for CA125, the current gold standard biomarker for ovarian cancer (4). Thus, this new high- throughput, top-down approach to biomarker discovery provides a clear path for the rapid detection of potential markers for early disease detection.

Grant/funding support: None declared. Financial disclosures: M.F.L., A.M., S.K., E.G., W.F.P., C.L., A.J., and W.M. are employees of PerkinElmer Life and Analytical Sciences, the vendor of ProXPRESSIONS reagent sets and the prOTOF 2000 orthogonal MALDI-TOF mass spectrometer described in this manuscript. L.A.L. and E.F.P. have US government- and University-assigned patents that cover certain aspects of the technology discussed. No other financial interests declared.

References Fig. 2. Differential expression of fragments. 1. Bast RC. Status of tumor markers in ovarian cancer screening. J Clin Oncol 2003;21:200–5. (A), transthyretin fragment at m/z 2898.54, zoom of m/z 2890 to 2915 area. (B), plasma kallikrein-sensitive glycoprotein fragment at m/z 2582.35, zoom of 2. Boyce EA, Kohn EC. Ovarian cancer in the proteomics era: m/z 2550 to 2610 area. (C), complement component precursor 3 fragment at diagnosis, prognosis, and therapeutics targets. Int J Gynecol m/z 1690.94, zoom of m/z 1685 to 1700 area. Cancer 2005;15(Suppl 3):266–73. Screen capture from PG600 analysis. Overlaid mass spectral traces from 3. Bhoola S, Hoskins WJ. Diagnosis and management of epithelial representative ovarian cancer and healthy samples. Red ϭ disease, green ϭ healthy. ovarian cancer. Obstet Gynecol 2006;107:1399–410. 4. Moss EL, Hollingworth J, Reynolds TM. The role of CA125 in clinical practice. J Clin Pathol 2005;58:308–12. Lowenthal et al. (19). In this study, we identified a 5. Hortin GL, Jortani SA, Ritchee, JC Jr, Valdes R Jr, Chan DW. number of proteins associated with cellular proliferation, Proteomics: a new diagnostic frontier. Clin Chem 2006;52:1218– 22. cancer, and cancer signaling pathways in the ovarian 6. Wulfkuhle JD, Liotta LA, Petricoin EF. Proteomic applications for cancer samples. Although we could not be certain that the early detection of cancer. Nat Rev Cancer 2003;3:267–75. these peptides/proteins were exclusive to the cancer sera, 7. Ardekani AM, Liotta LA, Petricoin EF III. Clinical potential of many of the peptides were not found in the pooled proteomics in the diagnosis of ovarian cancer. Expert Rev Mol healthy serum samples. A sampling of these proteins and Diagn 2002;2:312–20. 1074 Lopez et al.: Serum Profiling and Sequencing of Cancer Biomarkers

8. Srinivas PR, S. Srivastava S, Hanash S, Wright GL, Proteomics in 27. Mehta AI, Ross S, Lowenthal MS, Fusaro V, Fishman DA, Petricoin early detection of cancer. Clin Chem 2001;47:1901–11. EF III, et al. Biomarker amplification by serum carrier protein 9. Rosenblatt KP, Bryant-Greenwood P, Killian JK, Mehta A, Geho D, binding. Dis Markers 2003;19:1–10. Espina V, et al. Serum proteomics in cancer diagnosis and 28. Lopez MF, Mikulskis A, Kuzdzal S, Bennett D, Kelly J, Golenko E, management. Annu Rev Med 2004;55:97–112. et al. High-resolution serum proteomic profiling of Alzheimer’s 10. Zhang Z, Bast RC Jr, Yu Y, Li J, Sokoll LJ, Rai AJ, et al. Three disease samples reveals disease-specific, carrier-protein-bound biomarkers identified from serum proteomic analysis for the mass signatures. Clin Chem 2005;51:1946–54. detection of early stage ovarian cancer. Cancer Res 2004;64: 29. Zhou M, Lucas DA, Chan KC, Issaq HJ, Petricoin EF III, Liotta LA, 5882–90. et al. An investigation into the human serum “interactome”. 11. Li J, Zhang Z, Rosenzweig J, Wang YY, Chan DW. Proteomics and Electrophoresis 2004;25:1289–98. bioinformatics approaches for identification of serum biomarkers 30. Villanueva J, Shaffer DR, Philip J, Chaparro CA, Erdjument- to detect breast cancer. Clin Chem 2002;48:1296–1304. Bromage H, Olshen AB, et al. Differential exoprotease activities confer tumor-specific serum peptidome patterns. J Clin Invest 12. Petricoin EF, Ardekani AM, Hitt BA, Levine PJ, Fusaro VA, Steinberg 2006;116:271–84. SM, et al. Use of proteomic patterns in serum to identify ovarian 31. Koomen JM, Li D, Xiao LC, Liu TC, Coombes KR, Abbruzzese J, et cancer. Lancet 2002;359:572–7. al. Direct tandem mass spectrometry reveals limitations in protein 13. Hortin GL. The MALDI-TOF mass spectrometric view of the plasma profiling experiments for plasma biomarker discovery. J Proteome proteome and peptidome. Clin Chem 2006;52:1223–37. Res 2005;4:972–81. 14. Whiteley GR. Proteomic patterns for cancer diagnosis-promise and 32. Sing T, Sander O, Beerenwinkel N, Lengauer T. ROCR: An R challenges. Mol BioSyst 2006;2:358–63. Package for visualizing the performance of scoring classifiers. 15. Baggerly KA, Morris JS, Edmonson SR, Coombes KR. Signal in http://rocr.bioinf.mpi-sb.mpg.de (accessed January 2006). noise: evaluating reported reproducibility of serum proteomic 33. Hastie TJ, Tibshirani R, Buja A. Flexible Discriminant Analysis by tests for ovarian cancer. J Natl Cancer Inst 2005;97:307–9. Optimal Scoring. J Am Stat Assoc 1994;89:1255–70. 16. Liotta LA, Lowenthal M, Mehta A, Conrads TP, Veenstra TD, 34. Molinaro AM, Simon R, Pfeiffer RM. Prediction error estimation: a Fishman DA, et al. Importance of communication between produc- comparison of resampling methods. Bioinformatics 2005;21: ers and consumers of publicly available experimental data. J Natl 3294–3300. Cancer Inst 2005;97:310–14. 35. Scaglioni PP, Yung TM, Cai LF, Erdjument-Bromage H, Kaufman 17. Diamandis EP. Point: proteomic patterns in biological fluids: do AJ, Singh B, et al. A CK2-dependent mechanism for degradation of they represent the future of cancer diagnostics? Clin Chem the PML tumor suppressor. Cell 2006;126:269–83. 2003;49:1272–5. 36. DeSouza L, Diehl G, Rodrigues MJ, Guo J, Romaschin AD, Colgan 18. Liotta LA, Petricoin EF. Serum peptidome for cancer detection: TJ, et al. Search for cancer markers from endometrial tissues spinning biologic trash into diagnostic gold. J Clin Invest 2006; using differentially labeled tags iTRAQ and cICAT with multidimen- 116:26–30. sional liquid chromatography and tandem mass spectrometry. J 19. Lowenthal MS, Mehta AI, Frogale K, Bandle RW, Araujo RP, Hood Proteome Res 2005;4:377–86. BL, et al. Analysis of albumin-associated peptides and proteins 37. Chen H, Wang M, Wang XY, Gao S, Wang J, Guan XM. Identifica- from ovarian cancer patients. Clin Chem 2005;51:1933–45. tion of differential genes in ovarian cancer using representational 20. Diamandis EP. Peptidomics for cancer diagnosis: present and difference analysis of cDNA. Chin Med Sci J 2005;3:185–9. future. J Proteome Res 2006;9:2079–82. 38. Sieja K, Stanosz S, von Mach-Szczypinski J, Olewniczak S, Stanosz M. Concentration of histamine in serum and tissues of 21. Tirumalai RS, Chan KC, Prieto DA, Issaq HJ, Conrads TP, Veenstra the primary ductal breast cancers in women. Breast 2005;14: TD. Characterization of the low molecular weight human serum 236–41. proteome. Mol Cell Proteomics 2003;2:1096–103. 39. Bloor BK, Tidman N, Leigh IM, Odell E, Dogan B, Wollina U, et al. 22. Liu H, Sadygov RG, Yates JR. A model for random sampling and Expression of keratin K2e in cutaneous and oral lesions: associ- estimation of relative protein abundance in shotgun proteomics. ation with keratinocyte activation, proliferation, and keratinization. Anal Chem 2004;76:4193–201. Am J Pathol 2003;162:963–75. 23. Pieper R, Gatlin CL, Makusky AJ, Russo PS, Schatz CR, Miller SS, 40. Juang CM, Yang YH, Lo WH, Lai CR, Hsieh SL, Yuan CC. Altered et al. The human serum proteome: display of nearly 3700 mRNA expressions of sialyltransferases in ovarian cancers.Wang chromatographically separated protein spots on two-dimensional PH, Lee WL, Gynecol Oncol 2005;99:631–9. electrophoresis gels and identification of 325 distinct proteins. 41. Anderson NL, Polanski M, Pieper R, Gatlin T, Tirumalai RS, Proteomics 2003;3:1345–64. Contrads TP, et al. The human plasma proteome. Molec Cell 24. Adkins JN, Varnum SM, Auberry KJ, Moore RJ, Angell NH, Smith Proteom 2004;3:311–26. RD, et al. Toward a human blood serum proteome: analysis by 42. Rickles FR. Mechanisms of cancer-induced thrombosis in cancer. multidimensional separation coupled with mass spectrometry. Pathophysiol Haemost Thromb 2006;35:103–10. Mol Cell Proteomics 2002;1:947–55. 43. Bast RC Jr, Badgwell D, Lu Z, Marquez R, Rosen D, Liu J, et al. 25. Washburn MP, Wolters D, Yates JR III. Large-scale analysis of the New tumor markers: CA125 and beyond. Int J Gynecol Cancer. yeast proteome by multidimensional protein identification technol- 2005;15(Suppl 3):274–81. ogy. Nat Biotechnol 2001;19:242–7. 44. Kumar R, Hung M-C. Signaling intricacies take center stage in 26. Wang YY, Cheng P, Chan DW. A simple affinity spin tube filter cancer cells. Cancer Res 2005;65:2511–5. method for removing high-abundant common proteins or enriching 45. Kozak KR, Su F, Whitelegge JP, Faull K, Reddy S, Farias-Eisner R. low-abundant biomarkers for serum proteomic analysis. Proteom- Characterization of serum biomarkers for detection of early stage ics 2003;3:243–8. ovarian cancer. Proteomics 2005;17:4589–96. Clinical Chemistry 53:6 1075–1083 (2007) Proteomics and Protein Markers

N-Terminal Pro-B-Type Natriuretic Peptide Predicts Total and Cardiovascular Mortality in Individuals with or without Stable Coronary Artery Disease: The Ludwigshafen Risk and Cardiovascular Health Study

Winfried Ma¨ rz,1* Beate Tiran,2 Ursula Seelhorst,3 Britta Wellnitz,3 Johann Bauersachs,4 Bernhard R. Winkelmann,5 and Bernhard O. Boehm6

Background: Measurement of N-terminal pro-B-type 16.5 (9.2–29.8), respectively, compared with concentra- natriuretic peptide (NT-pro-BNP) measurement can be tions <100 ng/L. Hazard ratios (CI) for death from used to predict mortality in patients with acute coronary cardiovascular causes were 3.8 (1.8–8.2), 9. 3 (4.4–19.5), syndromes. Information on the value of NT-pro-BNP in and 22.2 (10.2–48.4). NT-pro-BNP remained predictive of clinically stable persons scheduled for angiography is total and cardiovascular mortality after accounting for limited. age, sex, diabetes mellitus, body mass index, smoking, Methods: We used Cox proportional hazards regression hypertension, dyslipidemia, glomerular filtration rate, to examine the effect of NT-pro-BNP on total and presence or absence of CAD on angiography, cardiovas- cardiovascular mortality in 1135 with and 506 individu- cular medication, revascularization at baseline, clinical als without stable coronary artery disease (CAD). signs of heart failure, LV systolic function, and C-reac- Results: NT-pro-BNP was associated with New York tive protein. Heart Association functional class, left ventricular (LV) Conclusions: NT-pro-BNP is predictive of all-cause and systolic function, and LV end-diastolic pressure. NT- cardiovascular mortality in individuals with or without pro-BNP was positively related to age, female sex, stable angiographic CAD independently of other car- hypertension, and former and current smoking and diovascular risk factors, coronary atherosclerosis, and negatively related to body mass index and glomerular cardiac function. filtration rate. During a median follow-up of 5.45 years, © 2007 American Association for Clinical Chemistry NT-pro-BNP concentrations of 100–399, 400-1999, or >2000 ng/L resulted in unadjusted hazard ratios (95% B-type natriuretic peptide (BNP)7 is released by the car- CI) for all-cause death of 3.2 (1.8–5.6), 6.63 (3.8–11.6), and diac ventricles in response to increased wall tension. The major biological actions of BNP include vasodilation and increased renal output of sodium and water. BNP is 1 Synlab Center of Laboratory Diagnostics, Heidelberg, Germany. synthesized as a prohormone that is cleaved into active 2 Clinical Institute of Medical and Chemical Laboratory Diagnostics, Med- BNP and an inactive N-terminal fragment (NT-pro-BNP) ical University of Graz, Graz, Austria. 3 Ludwigshafen Risk and Cardiovascular Health Study gGmbH, Freiburg, (1). Both BNP and NT-pro-BNP are well-established bi- Germany. omarkers used for diagnosis of heart failure (2). Both BNP 4 Division of Cardiology, Department of Medicine, University of Wu¨ rz- and NT-pro-BNP have been shown to predict mortality in burg, Wu¨ rzburg, Germany. 5 Cardiology Group, Frankfurt-Sachsenhausen, Germany. 6 Division of Endocrinology, Department of Medicine, University Hospital, Ulm, Germany. * Address correspondence to this author at: Synlab Center of Laboratory 7 Nonstandard abbreviations: BNP, B-type natriuretic peptide; NT-pro- Diagnostics Heidelberg, PO Box 10 47 80, D-69037 Heidelberg, Germany. Fax BNP, N-terminal pro-B-type natriuretic peptide; CAD, coronary artery disease; 49-6221-793-111; e-mail [email protected]. LV, left ventricular; LURIC, Ludwigshafen Risk and Cardiovascular Health; Received July 10, 2006; accepted March 14, 2007. LVEF, LV ejection fraction; CRP, C-reactive protein; GFR, glomerular filtration Previously published online at DOI: 10.1373/clinchem.2006.075929 rate; NYHA, New York Heart Association.

1075 1076 Ma¨rz et al.: NT-pro-BNP Predicts Total and Cardiovascular Mortality

patients presenting with acute coronary syndromes (3–7) Of 3279 individuals with coronary angiograms, 1638 and in asymptomatic individuals in the general popula- presented with unstable angina, non–ST-elevation myo- tion (8, 9). cardial infarction (troponin T Ն 0.1 ␮g/L, which was the Fewer studies have investigated whether BNP or NT- diagnostic cutoff at the time of recruitment of our study pro-BNP provides prognostic information in patients with population), ST-elevation myocardial infarction (troponin stable coronary artery disease (CAD) (10–13) and how T Ն 0.1 ␮g/L), or valve disease. These patients were this information relates to the clinical assessment of left excluded from the current evaluation. Thus, we finally ventricular (LV) function. included 1641 individuals. In these patients, LV function We investigated the importance of NT-pro-BNP as a was graded semiquantitatively by contrast ventriculogra- prognostic factor for long-term mortality in clinically phy into “normal” or “minimally”, “moderately”, or stable individuals scheduled for elective coronary angiog- “severely” impaired. LV end diastolic pressure was avail- Ͼ raphy. We also assessed whether NT-pro-BNP was re- able for 1488 patients; it was considered increased at 12 lated to mortality in individuals without LV systolic mmHg. Among the 1641 individuals included in the dysfunction or clinical signs of heart failure. analysis, LV ejection fraction (LVEF), calculated from the right anterior oblique view, was available in 658 patients Materials and Methods (40.1%). In this subgroup, the semiquantitative assess- study design and participants ment of LV function correlated well with the LVEF ϭ Ͻ To examine the relationships between NT-pro-BNP and (Spearman correlation coefficient 0.696, P 0.001). mortality, we studied participants of the Ludwigshafen Information on vital status was obtained from local Risk and Cardiovascular Health (LURIC) Study. LURIC is person registries. The median time of follow-up was 5.45 an ongoing prospective study of environmental, biochem- years. No patients were lost to follow-up. Among the 1641 ical, and genetic risk factors for CAD in a hospital-based patients included in this examination, 187 deaths (11.4%) cohort of white individuals (14). Between June 1997 and had occurred. A death certificate was obtained for 186 of the decedents and was missing for 1 decedent, who was January 2000, 3297 German patients who had undergone included in the total mortality analysis but excluded from coronary angiography at the Ludwigshafen General Hos- the cardiovascular mortality analysis. Cardiovascular pital and did not suffer from major noncardiovascular death included the following categories: sudden death, disease were included. The study was approved by the fatal myocardial infarction, death due to congestive heart institutional review board at the “A¨ rztekammer Rhein- failure, death immediately after intervention to treat land-Pfalz”. Informed written consent was obtained from CAD, fatal stroke, and other causes of death due to CAD. each of the participants. Individuals were classified as having diabetes mellitus laboratory procedures if plasma glucose was Ͼ1.25 g/L in the fasting state or The standard laboratory methods used have previously Ͼ2.00 g/L 2 h after the oral glucose load (performed in been described in detail (14). Lipoproteins were separated individuals without previous diabetes mellitus diagnosis) by a combined ultracentrifugation-precipitation method. (15), or if individuals were receiving oral antidiabetics or NT-pro-BNP was measured by electrochemiluminescence insulin. Hypertension was diagnosed if the systolic on an Elecsys 2010 (Roche Diagnostics). Interassay CVs and/or diastolic blood pressure exceeded 140 and/or 90 were 3.2% and 2.0% at mean values of 157 and 5125 ng/L. mmHg or if there was a significant history of hyperten- Sensitive C-reactive protein (CRP) was measured by sion, mainly evident by the use of antihypertensive med- immunonephelometry on a Behring Nephelometer II ication. Dyslipidemia was defined by categories proposed (N High Sensitivity CRP, Dade Behring). In this assay, by the National Cholesterol Education Program Adult the limit of detection for CRP is 0.17 mg/L; it is linear Ͻ Treatment Panel III (16) as HDL cholesterol (HDL-C) 1.0 up to 500 mg/L. Glomerular filtration rate (GFR) was Ϫ Ϫ mmol/L (40 mg/dL) and/or LDL cholesterol (LDL-C) calculated as GFR [mL ⅐ min 1 ⅐ (1.73 m2) 1] ϭ Ͼ Ͼ Ϫ Ϫ Ϫ 4.1 mmol/L (160 mg/dL), and/or triglycerides 2.4 186 ⅐ creatinine 1.154 ⅐ age 0.203 and GFR (mL ⅐ min 1 ⅐ [1.73 Ϫ Ϫ Ϫ (200 mg/dL). m2) 1] ϭ 138 ⅐ creatinine 1.154 ⅐ age 0.203 in males and in CAD was assessed by angiography using the maxi- females, respectively (18). mum luminal narrowing estimated by visual analysis. In LURIC, clinically relevant CAD was defined as the occur- statistical analysis rence of at least 1 stenosis Ն20% in at least 1 of 15 NT-pro-BNP and CRP were transformed logarithmically coronary segments (17). Individuals with stenoses Ͻ20% before being used in parametric statistical procedures. were considered as controls. To examine the impact of Study participants were divided into 4 groups with con- other definitions of CAD on the current analysis, we centrations of NT-pro-BNP Ͻ100, 100–399, 400–1999, or provisionally used the presence of 1 stenosis Ն50% as a Ն2000 ng/L. Clinical and anthropometric characteristics criterion. Furthermore, we stratified our cohort into indi- of individuals grouped according to NT-pro-BNP are viduals according to the number of vessels (none, 1, 2, or presented as percentages for categorical variables and as 3) with stenoses Ն50%. means and SD for continuous variables. Associations of Clinical Chemistry 53, No. 6, 2007 1077

categorical and continuous variables with strata of NT- drugs (beta-blockers, angiotensin converting enzyme in- pro-BNP were analyzed by ␹2 test and ANOVA with hibitors, angiotensin type 1 receptor antagonists, calcium covariables as indicated, respectively (Table 1). channel blockers, diuretics, antiplatelet drugs, and/or We studied the effect of the angiographic CAD status, lipid lowering agents), revascularization at baseline, and sex, age, and risk factors (body mass index, diabetes semiquantitative LV function (Tables 3 and 4). All statis- mellitus, hypertension, smoking, dyslipidemia, and GFR) tical tests were 2-sided. P Ͻ0.05 was considered statisti- on NT-pro-BNP with an ANOVA model in which we cally significant. The SPSS 11.0 statistical package (SPSS, included those factors not under examination and semi- SA) was used for all analyses. quantitative LV function as covariables (Table 2). To examine the effect of NT-pro-BNP on mortality Results from all causes and from cardiovascular causes, we cal- clinical characteristics according to culated hazard ratios and 95% CIs by the Cox propor- nt-pro-bnp tional hazards regression model. As indicated by log Individuals with high NT-pro-BNP concentrations were, minus log diagnostic plots, the proportional hazards on average, older than those with low concentrations. In assumption was met throughout. Multivariable adjust- this unadjusted analysis, NT-pro-BNP was associated ment was carried out for sex, age, cardiovascular risk with augmenting prevalence rates of diabetes mellitus, factors (body mass index, diabetes mellitus, hypertension, dyslipidemia, systemic hypertension, past smoking, the smoking, dyslipidemia, and GFR), angiographic CAD, prevalence and severity of angiographic CAD, previous previous myocardial infarction, use of cardiovascular myocardial infarction, peripheral vascular disease, and

Table 1. Clinical and biochemical characteristics associated with NT-pro-BNP. NT-pro-BNP, ng/L

<100 100–399 400–1999 >2000 Pa (n ϭ 498) (n ϭ 641) (n ϭ 389) (n ϭ 112) Age, years, mean (SD) 56 (10) 62 (10) 65 (10) 65 (12) Ͻ0.001 Male sex, % 74.9 61.8b 71.2 81.3b Ͻ0.001 Body mass index, kg/m2, mean (SD) 28 (4) 28 (4) 27 (4) 26 (4) Ͻ0.001 Diabetes mellitus, % 19.9 25.2 35.5 38.4 Ͻ0.001 Dyslipidemia, % 66.3 60.9 69.4 73.2 0.009 Systemic hypertension, % 64.7 72.9 79.2 75.9 Ͻ0.001 Current smoking, % 21.0 16.5 15.1 23.2 Ͻ0.001 Systolic blood pressure, mmHg mean (SD) 138 (21) 144 (22)b 147 (26)b 140 (25) Ͻ0.001c Glomerular filtration rate, ml/min 90 (15) 85 (17) 80 (18) 71 (25) Ͻ0.001 CRP, mg/L 1.8 2.4 3.1 5.5 Ͻ0.001d Median, 25th and 75th percentile (0.9–4.3) (1.0–5.5) (1.2–7.5) (1.8–11.1) Angiographic CAD (Ն20 percent stenosis), % 58.2 70.4 82.0 75.0 Ͻ0.001 Angiographic CAD (Ն50 stenosis), % 44.6 59.2 73.3 65.2 Ͻ0.001 Percutaneous intervention at baseline, % 9.3 13.6b 12.3b 6.3 0.036 Previous myocardial infarction, % 17.3 31.3 47.0 51.8 Ͻ0.001 Peripheral vascular disease, % 7.0 7.0 10.5 13.4 0.031 Cerebrovascular disease, % 2.4 7.2 9.3 8.9 Ͻ0.001 Cardiovascular medication Beta-blockers, % 49.4 64.8 65.3 56.3 Ͻ0.001 ACE inhibitors, % 29.9 43.9 59.6 70.5 Ͻ0.001 AT1 receptor antagonists, % 2.2 5.9 4.4 8.0 0.007 Calcium channel blockers, % 11.0 19.5 16.2 15.2 0.002 Diuretics, % 11.0 21.0 36.5 60.7 Ͻ0.001 Aspirin and/or other antiplatelet agents, % 64.3 70.6 69.9 61.6 Ͻ0.001 Lipid lowering drugs, % 34.9 47.4 47.8 34.8 Ͻ0.001 NYHA functional class IV, % 1.7 3.0 Ͻ0.001 LV function ЉseverelyЉ impaired, % 0.4 1.2 8.7 36.6 Ͻ0.001 LVEDP Ͼ12 mmHg, % 29.1 37.0 46.4 75.0 Ͻ0.001 LVEF Ͻ40, % 2.6 5.3 25.3 55.4 Ͻ0.001 a ANOVA, adjusted for age and sex, or ␹2 test. b Significant vs NT-pro-BNP Ͻ100 ng/L. c Adjusted for use of beta-blockers, ACE inhibitors, AT1 receptor antagonists, calcium channel blockers, and diuretics. d ANOVA of logarithmically transformed values, adjusted for age and sex. 1078 Ma¨rz et al.: NT-pro-BNP Predicts Total and Cardiovascular Mortality

Ͼ Table 2. Effect of cardiovascular risk factors and pro-BNP 100 ng/L. As expected, NT-pro-BNP was angiographic status on NT-pro-BNP. associated with the New York Heart Association (NYHA) functional class, semiquantitative LV function, the likeli- n NT-pro-BNP, ng/La P hood of increased LV end-diastolic pressure, and the Men 1138 193.1 (1.0) LVEF. Women 503 256.7 (1.1) Ͻ0.001 Age, years Ͻ60 691 146.3 (1.0) association of nt-pro-bnp with cardiovascular 60–70 597 235.6 (1.0) risk factors and cad status Ͼ70 353 356.0 (1.1) Ͻ0.001 We examined in more detail the effect of sex, age, risk BMI, kg/m2 factors (body mass index, diabetes mellitus, hypertension, 26 or 27b 754 233.0 (1.0) smoking), and angiographic status on NT-pro-BNP in a Ͼ26 or 27b 887 193.6 (1.0) 0.001 general linear model in which we included those factors Diabetes not under examination as covariables. We additionally No 1199 209.3 (1.0) adjusted for semiquantitative LV function. NT-pro-BNP Yes 442 214.9 (1.1) 0.678 was significantly influenced by sex, age, body mass index, Hypertension hypertension, smoking status, and GFR. Patients with No 458 189.2 (1.1) angiographic CAD had, on average, higher NT-pro-BNP Yes 1183 219.9 (1.0) 0.018 than those without CAD. Diabetes mellitus and dyslipi- Smoking demia had no significant effect on NT-pro-BNP. Never 599 186.8 (1.0) Former 746 216.4 (1.0) nt-pro-bnp and mortality from all causes Current 296 252.1 (1.1) 0.001 Among the 1641 patients included in this examination, Dyslipidemia 187 deaths (11.4%) had occurred during a median obser- No 568 196.2 (1.0) vation time of 5.45 years. Compared with individuals Yes 1073 219.0 (1.0) 0.059 with NT-pro-BNP Ͻ100 ng/L, the unadjusted hazard GFR, mL/min ratios for death of patients with NT-pro-BNP of 100–399 Ͻ30 628 1811.7 (1.4) ng/L, 400–1999 ng/L, and Ն2000 were 3.16 (P Ͻ0.001), 30–59 890 340.0 (1.1) Ͻ Ͻ 60–89 114 214.6 (1.0) 6.65 (P 0.001), and 16.53 (P 0.001), respectively (Table Ն90 9 182.9 (1.0) Ͻ0.001 3, Model 1, Fig. 1). Inclusion of age and sex as covariables Ͻ CAD, Ն20 stenosis decreased these estimates to 2.59 (P 0.001), 4.55 (P Ͻ Ͻ No 506 174.5 (1.1) 0.001), and 10.24 (P 0.001), respectively (Table 3, Yes 1135 229.3 (1.0) Ͻ0.001 Model 2). Although hazard ratios progressively de- a Estimated marginal means (SE) obtained in a general linear model (ANOVA), creased, NT-pro-BNP retained sizeable prognostic impor- adjusted for each of the other factors and, in addition, semiquantitative LV tance after adjustment for cardiovascular risk factors, the function, whereby body mass index and age were used as continuous rather than presence or absence of CAD, and semiquantitative LV categorial covariables, geometric means are reported for NT-pro-BNP. function (Table 3, Model 3). Accounting in addition for b 2 Thresholds of 26 and 27 kg/m apply to males and females, respectively. CRP did not materially change hazard ratios (Table 3, Model 4). cerebrovascular disease (Table 1). The proportion of men We further analyzed the prognostic value of NT-pro- with NT-pro-BNP concentrations between 100 and 399 BNP according to quartiles at baseline. Interquartile ng/L was significantly lower than the proportion with ranges were 5–81, 82–194, 195–521, and 522–35 000 ng/L, concentrations Ͻ100 ng/L; it was, however, significantly respectively. Compared with the lowest quartile, crude higher in the small group with NT-pro-BNP Ն2000 ng/L hazard ratios for death at NT-pro-BNP concentrations in (P ϭ 0.001 and P ϭ 0.026 by logistic regression, respec- the 2nd, 3rd, and 4th quartiles were 1.73 (P ϭ 0.102), 3.36 tively). The rates of percutaneous coronary interventions (P Ͻ0.001), and 8.69 (P Ͻ0.001), respectively (Table 3, were higher at NT-pro-BNP concentrations between 100 Model 1). As expected, these estimates decreased stepwise and 1999 ng/L than concentrations outside this range of with inclusion in the model, of age, sex (Table 3, Model 2), values. Systolic blood pressure was significantly higher in cardiovascular risk factors, angiographic status, and patients with NT-pro-BNP between 100 and 1999 ng/L semiquantitative LV function (Table 3, Model 3). Finally, than in those with NT-pro-BNP Ͻ100 ng/L. In the sub- considering CRP as a covariable had no appreciable group with the highest NT-pro-BNP (Ն2000 ng/L), mean influence. systolic blood pressure approximately came down to the To study whether NT-pro-BNP added prognostic in- mean value seen in the subgroup with the lowest NT-pro- formation beyond clinical assessment, we performed mul- BNP (Ͻ100 ng/L). There was no association of NT-pro- tivariate Cox regression in 1211 individuals in whom LV BNP with diastolic blood pressure. Body mass index and function was graded “normal”. Among these, 93 (7.7%) GFR decreased as NT-pro-BNP increased. The use of all had deceased during the follow-up period. NT-pro-BNP cardiovascular medications was more prevalent at NT- remained significantly predictive of total mortality; ad- Clinical Chemistry 53, No. 6, 2007 1079

Table 3. Hazard ratios for mortality from all causes according to NT-pro-BNP. NT-pro-BNP, ng/L, hazard ratio (95% CI)a

Deaths, n (%) Model 1 Model 2 Model 3 Model 4 All subjects (n ϭ 1641) Ͻ100 (n ϭ 498) 15 (3.0) 1.0 1.0 1.0 1.0 100–399 (n ϭ 642) 59 (9.2) 3.16 (1.80–5.58)b 2.59 (1.46–4.61)b 3.18 (1.39–7.29)c 3.13 (1.37–7.17)c 400–1999 (n ϭ 389) 71 (18.3) 6.63 (3.80–11.60)b 4.54 (2.56–8.07)b 5.07 (2.18–11.80)b 4.96 (2.13–11.56)b Ն2000 (n ϭ 112) 42 (37.5) 16.53 (9.16–29.82)b 10.24 (5.56–18.88)b 8.11 (3.16–20.83)b 7.83 (3.04–20.13)b 5–81 (n ϭ 410) 14 (3.4) 1.0 1.0 1.0 1.0 82–194 (n ϭ 411) 24 (5.8) 1.73 (0.90–3.35) 1.53 (0.79–2.97) 1.55 (0.78–3.01) 1.52 (0.76–3.03) 195–521 (n ϭ 411) 45 (10.9) 3.35 (1.83–6.10)b 2.55 (1.38–4.71)d 2.56 (1.34–4.91)d 2.52 (1.32–4.82)d 522–35 000 (n ϭ 409) 104 (25.4) 8.69 (4.97–15.19)b 5.60 (3.14–10.01)b 3.80 (1.99–7.28)b 3.64 (1.90–6.97)b LV function “normal” (n ϭ 1211) Ͻ100 (n ϭ 465) 13 (2.8) 1.0 1.0 1.0 1.0 100–399 (n ϭ 517) 37 (7.2) 2.62 (1.39–4.93)d 1.88 (0.98–3.61) 2.44 (0.95–6.25)e 2.44 (0.96–6.24)e Ն400 (n ϭ 229) 43 (18.8) 7.50 (4.03–13.95)b 4.06 (2.09–7.88)b 5.15 (1.92–13.80)e,b 5.12 (1.91–13.73)b,e LV function “normal” and NYHA class I (n ϭ 602) Ͻ100 (n ϭ 258) 6 (2.3) 1.0 1.0 1.0 1.0 100–399 (n ϭ 258) 12 (4.7) 1.99 (0.75–5.31) 1.66 (0.60–4.58) 1.60 (0.32–8.00)e 1.48 (0.29–7.58)e Ն400 (n ϭ 86) 15 (17.4) 8.12 (3.15–20.93)b 5.30 (1.87–15.04)d 15.39 (3.04–77.91)b,e 9.79 (3.13–30.61)b,e No angiographic CAD (n ϭ 506) Ͻ100 (n ϭ 218) 5 (2.3) 1.0 1.0 1.0 1.0 100–399 (n ϭ 190) 9 (4.7) 2.12 (0.71–6.32) 1.90 (0.62–5.83) 1.38 (0.43–4.45)f 1.35 (0.42–4.40)f Ն400 (n ϭ 98) 13 (13.3) 6.24 (2.22–17.53)b 4.24 (1.45–12.45)c 1.84 (0.52–6.43)f 1.88 (0.53–6.64)f Angiographic CAD (n ϭ 1135) Ͻ100 (n ϭ 280) 10 (3.6) 1.0 1.0 1.0 1.0 100–399 (n ϭ 452) 50 (11.1) 3.23 (1.64–6.37)b 2.76 (1.39–5.48)d 3.61 (1.68–7.75)b,f 3.53 (1.65–7.57)b,f Ն400 (n ϭ 403) 100 (24.8) 8.02 (4.19–15.37)b 5.97 (3.07–11.64)b 6.47 (3.02–13.84)b,f 6.20 (2.90–13.28)b,f a Model 1, unadjusted. Model 2, adjusted for age and sex. Model 3, in addition, adjusted for type 2 diabetes, body mass index, smoking status, hypertension, dyslipidemia, GFR, presence or absence of CAD on angiography, previous myocardial infarction, use of beta-blockers, angiotensin converting enzyme inhibitors, angiotensin type 1 receptor antagonists, calcium channel blockers, diuretics, antiplatelet drugs, lipid lowering agents, revascularization at baseline, and semiquantitative LV function. Model 4, in addition, adjusted for CRP. b P Յ0.001. c P Ͻ0.05. d P Յ0.005. e Semiquantitative LV function not included as a covariable. f Absence or presence of angiographic CAD not included as a covariable.

justed and unadjusted hazard ratios were lower than and CRP (Table 3). Finally, when we used CAD diag- those for the entire study population. nosed as the presence of at least 1 stenosis Ն50% or when We also separately investigated study participants we used the number stenosed of vessels as a covariable in with “normal” LV function, also classified as NYHA stage the Cox models instead of CAD diagnosed as at least 1 I(nϭ 602). Once more, NT-pro-BNP Ն400 ng/L turned stenosis Ն20%, the results were materially unchanged. out predictive of all-cause mortality in each of the statistical models. nt-pro-bnp and mortality from Separate analysis of patients with or without angio- cardiovascular causes graphic CAD did not substantially alter our findings Because a death certificate was not available from 1 (Table 3). In patients with angiographic CAD, NT-pro- deceased person, the analysis for cardiovascular mortality BNP remained associated with mortality from all causes included a total of 1640 individuals. Among these, 129 in the unadjusted model and in all adjusted models. In (69.4%) died from cardiovascular causes. Hazard ratios individuals without angiographic CAD, the association for death from cardiovascular causes according to NT- was significant in the unadjusted model and in the model pro-BNP were consistent with those obtained for mortal- adjusted for age and sex but turned out insignificant in ity from all causes in all models (Table 4, Fig. 1), including the models adjusted for risk factors, clinical confounders those that used CAD diagnosed at 1 stenosis Ն50% or the 1080 Ma¨rz et al.: NT-pro-BNP Predicts Total and Cardiovascular Mortality

Table 4. Hazard ratios for cardiovascular death according to NT-pro-BNP. NT-pro-BNP, ng/L, hazard ratio (95% CI)a

Deaths, n (%) Model 1 Model 2 Model 3 Model 4 All subjects (n ϭ 1640) Ͻ100 (n ϭ 498) 8 (1.6) 1.0 1.0 1.0 1.0 100–399 (n ϭ 642) 38 (5.9) 3.82 (1.78–8.18)b 3.26 (1.51–7.01)c 2.63 (1.21–5.72)d 2.59 (1.19–5.64)d 400–1999 (n ϭ 389) 53 (13.6) 9.27 (4.41–19.50)b 6.78 (3.16–14.53)b 4.01 (1.82–8.86)b 3.89 (1.76–8.60)b Ն2000 (n ϭ 111) 30 (27.0) 22.19 (10.17–48.43)b 14.89 (6.67–33.25)b 6.21 (2.53–15.27)b 5.86 (2.38–14.39)b 5–81 (n ϭ 410) 8 (2.0) 1.0 1.0 1.0 1.0 82–194 (n ϭ 411) 13 (3.2) 1.64 (0.68–3.96) 1.49 (0.62–3.63) 1.35 (0.56–3.28) 1.33 (0.55–3.24) 195–521 (n ϭ 411) 30 (7.3) 3.90 (1.79–8.51)b 3.16 (1.42–7.00)c 2.35 (1.05–5.28)d 2.34 (1.04–5.25)d 522–35 000 (n ϭ 408) 78 (19.1) 11.40 (5.51–26.60)b 7.98 (3.76–16.93)b 4.05 (1.82–9.02)b 3.92 (1.76–8.74)b LV function “normal” (n ϭ 1211) Ͻ100 (n ϭ 465) 7 (1.5) 1.0 1.0 1.0 1.0 100–399 (n ϭ 517) 21 (4.1) 2.76 (1.17–6.48)d 2.04 (0.85–4.90) 1.88 (0.78–4.55)e 1.89 (0.78–4.55)e Ն400 (n ϭ 229) 26 (11.4) 8.41 (3.65–19.38)b 4.79 (1.97–11.67)b 4.01 (1.61–10.01)c,e 3.96 (1.86–9.87)c,e LV function “normal” and NYHA class I (n ϭ 602) Ͻ100 (n ϭ 258) 3 (1.2) 1.0 1.0 1.0 1.0 100–399 (n ϭ 258) 4 (1.6) 1.33 (0.30–5.93) 1.21 (0.26–5.68) 1.28 (0.28–5.95)e 1.27 (0.27–5.90)e Ն400 (n ϭ 86) 10 (11.6) 10.78 (2.70–39.19)b 8.03 (1.94–33.31)c 8.27 (1.96–34.81)c,e 8.81 (2.07–37.55)c,e No angiographic CAD (n ϭ 506) Ͻ100 (n ϭ 218) 1 (0.5) 1.0 1.0 1.0 1.0 100–399 (n ϭ 190) 4 (2.1) 4.70 (0.53–42.06) 4.05 (0.44–37.01) 2.01 (0.20–20.37)f 1.99 (0.20–20.16)f Ն400 (n ϭ 98) 11 (11.2) 26.10 (3.37–202.19)c 19.49 (2.43–156.52)c 8.96 (0.98–82.15)f 8.93 (0.97–82.28)f Angiographic CAD (n ϭ 1134) Ͻ100 (n ϭ 280) 7 (2.5) 1.0 1.0 1.0 1.0 100–399 (n ϭ 452) 34 (7.5) 3.14 (1.39–7.07)d 2.80 (1.23–6.38)d 3.52 (1.47–8.55)d,f 3.45 (1.42–8.38)d,f Ն400 (n ϭ 402) 72 (17.9) 8.23 (3.79–17.89)b 6.57 (2.96–14.53)b 7.04 (2.92–17.00)b,f 6.80 (2.81–16.46)b,f a Model 1, unadjusted. Model 2, adjusted for age and sex. Model 3, in addition, adjusted for type 2 diabetes, body mass index, smoking status, hypertension, dyslipidemia, GFR, presence or absence of CAD on angiography (at least one stenosis Ն20%, previous myocardial infarction, use of beta-blockers, angiotensin converting enzyme inhibitors, angiotensin type 1 receptor antagonists, calcium channel blockers, diuretics, antiplatelet drugs, lipid lowering agents, revascularization at baseline, and semiquantitative LV function. Model 4, in addition, adjusted for CRP. b P Յ0.001. c P Յ0.005. d P Ͻ0.05. e Semiquantitative LV function not included as a covariable. f Absence or presence of angiographic CAD not included as a covariable.

number of vessels with stenoses Ն50% as covariables sistently, we found NT-pro-BNP associated with semi- (data not shown). quantitative LV function, NYHA functional class, LVEF, and the likelihood of increased LV end-diastolic pressure. Discussion Patients with angiographic CAD had higher average This study has 2 key findings. First, NT-pro-BNP is a NT-pro-BNP than those without, and there was a positive strong predictor of all-cause mortality and cardiovascular relationship between the severity of angiographic CAD mortality in clinically stable patients undergoing coronary and NT-pro-BNP. NT-pro-BNP was also influenced by a angiography, independent of well-established risk factors series of traditional cardiovascular risk factors. As shown and the presence of CAD. Second, NT-pro-BNP is also a previously, after controlling for other confounding fac- strong predictor in individuals with no clinical evidence tors, women had significantly higher concentrations of of heart failure or impaired LV function. NT-pro-BNP than men, and NT-pro-BNP was positively Abundant information has become available with re- related to age (21). NT-pro-BNP was lower in obese than gard to the utility of BNP and NT-pro-BNP as diagnostic in lean persons, whereas it was higher in hypertensive and prognostic markers in heart failure (2, 19, 20). Con- than in normotensive persons. Former and current smok- Clinical Chemistry 53, No. 6, 2007 1081

Kragelund et al. found interquartile concentration ranges of 5–63, 64–169, 170–455, and 456–13 889 ng/L (10); ours were 5–81, 82–194, 195–521, and 522–35 000 ng/L. Thus, the 2 study populations appear to have had similar LV function, but in our study the atherosclerotic burden seems to have been much lower. When we used the criterion of at least 1 stenosis Ն20%, the prevalence of angiographic CAD was 69% in our study population; using lumen narrowing of Ն70% (or Ն50% in the left main coronary artery) as the cutoff, Kragelund et al. (10) classified ϳ84% of their study participants as having CAD. To directly compare our results with those of Krage- lund et al. (10), who analyzed mortality from all causes but not cardiovascular mortality, we examined quartiles of NT-pro-BNP at baseline and also in relation to all-cause mortality. Basically, we found an identical predictive value of NT-pro-BNP, the crude hazard ratio for death in the highest quartile of NT-pro-BNP being 8.69 compared with 6.1 in their study. We suggest, however, that the use of cutoff values of 100, 400, and 2000 ng/L NT-pro-BNP will enable better differentiation than the use of quartiles, because hazard ratios for death in subgroups defined by these cutoff values were roughly 2-fold those found in groups defined by quartiles (Tables 3 and 4). Schnabel et al. (12) examined the association of quartiles of NT-pro-BNP and BNP with cardiovascular death and nonfatal myocardial infarction in 570 and 334 individuals with stable and unstable CAD, respectively. In both groups, they saw a significant association of NT-pro- BNP with future cardiovascular events only in their highest quartile of NT-pro-BNP. Similar results were reported when the same investigators measured BNP rather than NT-pro-BNP in 1085 patients with stable angina (13). At variance to these studies, we consistently detected a continuous relationship between NT-pro-BNP and mortality, both from any cause and from cardiovas- Fig. 1. Cumulative survival functions according to NT-pro-BNP in cular causes, once NT-pro-BNP exceeded 100 ng/L. individuals with or without stable angiographic CAD. Remarkably, the association of NT-pro-BNP with mor- (A), total mortality; (B), cardiovascular mortality. For hazard ratios and CIs, see tality from all causes and from cardiovascular causes Tables 3 and 4. remained after we adjusted for cardiovascular risk factors, the presence or absence and severity (number of vessels ers had increased NT-pro-BNP concentrations. Similar to with stenoses Ն50%) of angiographic CAD, and semi- BNP, NT-pro-BNP increased as GFR decreased (22). quantitative LV function. The severity of CAD thus did Many studies have dealt with the role of NT-pro-BNP not affect the prognostic value of NT-pro-BNP. Further- in acute coronary syndromes (3–7), and 4 investigations more, in individuals with “normal” LV function, and in have addressed the prognostic value of NT-pro-BNP (or those with both “normal” LV function and NYHA func- BNP) in stable patients undergoing coronary angiography tional class I, NT-pro-BNP remained predictive, even in (10–13). In the study by Kragelund et al. (10), the propor- fully adjusted models. NT-pro-BNP, therefore, predicts tions of individuals with LVEF Ն60% were 80%, 61%, mortality independently, at least in part, from its role as a 34%, and 25% as NT-pro-BNP concentrations increased by marker of LV function. The mechanism underlying this quartiles. These proportional increments resemble quite observation remains unknown, however. BNP and NT- closely those in our study participants found to have LV pro-BNP correlate well with the extent of inducible myo- function graded as normal according to NT-pro-BNP cardial ischemia on stress single-photon emission com- concentrations, which were 93%, 81%, 53%, and 21%, puted tomography (23, 24). BNP concentrations increase respectively, in our 4 groups (Table 1). Furthermore, the after short-term coronary occlusion during coronary an- distributions of NT-pro-BNP concentrations were similar. gioplasty, but not after coronary angiography (25). To- 1082 Ma¨rz et al.: NT-pro-BNP Predicts Total and Cardiovascular Mortality

gether, these findings may suggest that these peptides References reflect chronically recurring myocardial ischemia, but our 1. Levin ER, Gardner DG, Samson WK. Natriuretic peptides. N Engl observation that NT-pro-BNP also predicted death in J Med 1998;339:321–8. persons with coronary stenoses Ͻ20% does not support 2. Doust JA, Glasziou PP, Pietrzak E, Dobson AJ. A systematic review of the diagnostic accuracy of natriuretic peptides for heart failure. this explanation. It is possible, however, that NT-pro-BNP Arch Intern Med 2004;164:1978–84. reflects LV dysfunction not detected by the clinical means 3. Omland T, Persson A, Ng L, O’Brien R, Karlsson T, Herlitz J, et al. used here. N-terminal pro-B-type natriuretic peptide and long-term mortality in The utility of measuring BNP and NT-pro-BNP in acute coronary syndromes. Circulation 2002;106:2913–8. patients at high risk of death, namely those with chronic 4. de Lemos JA, Morrow DA, Bentley JH, Omland T, Sabatine MS, McCabe CH, et al. The prognostic value of B-type natriuretic heart failure (2, 19, 20) or acute coronary syndromes (3– peptide in patients with acute coronary syndromes. N Engl J Med 7), has been firmly established, but information on the use 2001;345:1014–21. of these markers in other patient groups is limited. Wang 5. Mega JL, Morrow DA, De Lemos JA, Sabatine MS, Murphy SA, Rifai et al. (8) demonstrated that BNP was a predictor of N, et al. B-type natriuretic peptide at presentation and prognosis in all-cause mortality in asymptomatic participants in the patients with ST-segment elevation myocardial infarction: an EN- Framingham Heart Study. Only recently, in an older TIRE-TIMI-23 substudy. J Am Coll Cardiol 2004;44:335–9. 6. James SK, Lindahl B, Siegbahn A, Stridsberg M, Venge P, Arm- community-based population, NT-pro-BNP was a strong strong P, et al. N-terminal pro-brain natriuretic peptide and other risk biomarker for cardiovascular disease and death but risk markers for the separate prediction of mortality and subse- CRP was not (9). quent myocardial infarction in patients with unstable coronary The current study is the largest to address the clinical artery disease: a Global Utilization of Strategies To Open Occluded value of NT-pro-BNP in clinically stable individuals un- Arteries (GUSTO)-IV substudy. Circulation 2003;108:275–81. dergoing angiography. By and large our results are con- 7. Jernberg T, James S, Lindahl B, Stridsberg M, Venge P, Wallentin L. NT-proBNP in unstable coronary artery disease: experiences sistent with those of smaller previous investigations (10– from the FAST, GUSTO IV and FRISC II trials. Eur J Heart Fail 13), all suggesting that the measurement of NT-pro-BNP 2004;6:319–25. may be helpful in individuals without clinical indications 8. Wang TJ, Larson MG, Levy D, Benjamin EJ, Leip EP, Omland T, et of heart failure. Furthermore, because NT-pro-BNP may al. Plasma natriuretic peptide levels and the risk of cardiovascular be associated with mortality in individuals without rele- events and death. N Engl J Med 2004;350:655–63. vant angiographic CAD, we speculate that screening of 9. Kistorp C, Raymond I, Pedersen F, Gustafsson F, Faber J, Hilde- brandt P. N-terminal pro-brain natriuretic peptide, C-reactive pro- presymptomatic individuals with cardiovascular risk fac- tein, and urinary albumin levels as predictors of mortality and tors might refine risk stratification in the future. Further cardiovascular events in older adults. JAMA 2005;293:1609–16. cohort studies addressing this issue in populations at low 10. Kragelund C, Gronning B, Kober L, Hildebrandt P, Steffensen R. or intermediate cardiovascular risk are clearly warranted. N-terminal pro-B-type natriuretic peptide and long-term mortality in Because natriuretic peptide measurement may not be stable coronary heart disease. N Engl J Med 2005;352:666–75. applicable to screening large populations, our results lend 11. Omland T, Richards AM, Wergeland R, Vik-Mo H. B-type natriuretic peptide and long-term survival in patients with stable coronary further clarity regarding whether screening would be artery disease. Am J Cardiol 2005;95:24–8. useful in an at-risk population. 12. Schnabel R, Rupprecht HJ, Lackner KJ, Lubos E, Bickel C, Meyer J, et al. Analysis of N-terminal-pro-brain natriuretic peptide and C-reactive protein for risk stratification in stable and unstable coronary artery disease: results from the AtheroGene study. Eur Grant/funding support: Roche Diagnostics provided unre- Heart J 2005;26:241–9. stricted financial support and assay reagent sets free of 13. Schnabel R, Lubos E, Rupprecht HJ, Espinola-Klein C, Bickel C, Lackner KJ, et al. B-type natriuretic peptide and the risk of charge but did not assume any other role in the conduct of cardiovascular events and death in patients with stable angina: the study. results from the AtheroGene study. J Am Coll Cardiol 2006;47: Financial disclosures: Winfried Ma¨rz, Bernhard R. Winkel- 552–8. mann, and Bernhard Boehm are shareholders of the not-for- 14. Winkelmann BR, Ma¨rz W, Boehm BO, Zotz R, Hager J, Hellstern P, profit LURIC Study company (LURIC gemeinnu¨ tzige et al. Rationale and design of the LURIC study: a resource for GmbH). functional genomics, pharmacogenomics and long-term prognosis of cardiovascular disease. Pharmacogenomics 2001;2:S1–73. Acknowledgements: We extend our appreciation to the par- 15. American Diabetes Association. Diagnosis and classification of ticipants of the LURIC Study; without their collaboration this diabetes mellitus. Diabetes Care 2006;29(Suppl 1):S43–8. article would not have been written. The technical assistance 16. Third Report of the National Cholesterol Education Program of Sabine Kern is gratefully acknowledged. We thank all (NCEP) Expert Panel on Detection, Evaluation, and Treatment of members of the LURIC study team who were either tempo- High Blood Cholesterol in Adults (Adult Treatment Panel III) final rarily or permanently involved in patient recruitment and report. Circulation 2002;106:3143–421. 17. Austen WG, Edwards JE, Frye RL, Gensini GG, Gott VL, Griffith LS, sample and data handling; the laboratory staff at the Lud- et al. A reporting system on patients evaluated for coronary artery wigshafen General Hospital and the Universities of Freiburg disease: Report of the Ad Hoc Committee for Grading of Coronary and Ulm, Germany; and the German registration offices and Artery Disease, Council on Cardiovascular Surgery, American local public health departments for their assistance. Heart Association. Circulation 1975;51(Suppl 4):5–40. Clinical Chemistry 53, No. 6, 2007 1083

18. Levey AS, Bosch JP, Lewis JB, Greene T, Rogers N, Roth D. A more adults: effects of age and sex. Clin Chem Lab Med 2002;40: accurate method to estimate glomerular filtration rate from serum 371–7. creatinine: a new prediction equation: Modification of Diet in 22. Wallen T, Landahl S, Hedner T, Saito Y, Masuda I, Nakao K. Brain Renal Disease Study Group. Ann Intern Med 1999;130:461–70. natriuretic peptide in an elderly population. J Intern Med 1997; 19. Tsutamoto T, Wada A, Maeda K, Hisanaga T, Maeda Y, Fukai D, et 242:307–11. al. Attenuation of compensation of endogenous cardiac natriuretic 23. Palumbo B, Siepi D, Lupattelli G, Sinzinger H, Fiorucci G, Annibo- peptide system in chronic heart failure: prognostic role of plasma letti PF, et al. Usefulness of brain natriuretic peptide levels to brain natriuretic peptide concentration in patients with chronic discriminate patients with stable angina pectoris without and with electrocardiographic myocardial ischemia and patients with symptomatic left ventricular dysfunction. Circulation 1997;96: healed myocardial infarction. Am J Cardiol 2004;94:780–3. 509–16. 24. Sabatine MS, Morrow DA, de Lemos JA, Omland T, Desai MY, 20. Gardner RS, Ozalp F, Murday AJ, Robb SD, McDonagh TA. N- Tanasijevic M, et al. Acute changes in circulating natriuretic terminal pro-brain natriuretic peptide: a new gold standard in peptide levels in relation to myocardial ischemia. J Am Coll Cardiol predicting mortality in patients with advanced heart failure. Eur 2004;44:1988–95. Heart J 2003;24:1735–43. 25. Tateishi J, Masutani M, Ohyanagi M, Iwasaki T. Transient increase 21. Clerico A, Del Ry S, Maffei S, Prontera C, Emdin M, Giannessi D. in plasma brain (B-type) natriuretic peptide after percutaneous The circulating levels of cardiac natriuretic hormones in healthy transluminal coronary angioplasty. Clin Cardiol 2000;23:776–80. Clinical Chemistry 53:6 1084–1091 (2007) Cancer Diagnostics

Analytical Validation of the Oncotype DX Genomic Diagnostic Test for Recurrence Prognosis and Therapeutic Response Prediction in Node-Negative, Estrogen Receptor–Positive Breast Cancer

Maureen Cronin,* Chithra Sangli, Mei-Lan Liu, Mylan Pho, Debjani Dutta, Anhthu Nguyen, Jennie Jeong, Jenny Wu, Kim Clark Langone, and Drew Watson

Background: Oncotype DX™ is a clinically validated, contributed by instrument, operator, reagent, and day-to- high-complexity, multianalyte reverse transcription– day baseline variation was low, with SDs of <0.5 CT. PCR genomic test that predicts the likelihood of breast Conclusion: The analytical and operational perfor- cancer recurrence in early-stage, node-negative, estrogen mance specifications defined for the Oncotype DX assay receptor–positive breast cancer. The Recurrence Score™ allow the reporting of quantitative RS values for indi- (RS) provides a more accurate, reproducible measure of vidual patients with an SD within 2 RS units on a breast cancer aggressiveness and therapeutic respon- 100-unit scale. siveness than standard measures. Individualized patient © 2007 American Association for Clinical Chemistry management requires strict performance criteria for clinical laboratory tests. We therefore investigated the Women with early-stage breast cancer have many treat- analytical performance of the assay. ment options, but diagnostic methods derived from tra- Methods: Assays used a pooled RNA sample from fixed ditional anatomic-pathology assessment of tumor are of paraffin-embedded tissues to evaluate the analytical limited value for predicting treatment outcomes (1).Itis performance of a 21-gene panel with respect to amplifi- clear that some women are undertreated for their disease cation efficiency, precision, linearity, and dynamic and others are overtreated. Emerging molecular genomic range, as well as limits of detection and quantification. methods, such as molecular characterization of tumor Performance variables were estimated from assays car- tissue, are advancing clinical oncology diagnostics and ried out with sample dilutions. In addition, individual improving patient outcomes by providing a rational basis patient samples were used to test the optimized assay for individualizing oncology treatment (2, 3). for reproducibility and sources of imprecision. Molecular-biomarker tests must be rigorously validated Results: Assay results defined acceptable operational per- both analytically and clinically. Tests that simply dichoto- formance ranges, including an estimated maximum devi- mize patient populations into broad categories do not trans- late easily into clinically useful tools. Clinically relevant tests ation from linearity of <1 cycle threshold (CT) units over a >2000-fold range of RNA concentrations, with a mean provide precise, quantitative results that can be used to quantification bias of 0.3% and CVs of 3.2%–5.7%. An gauge the probabilities of success with available alternative analysis of study design showed that assay imprecision courses of disease treatment (4). We have adapted methods commonly used to validate single-analyte laboratory tests to validate a complex multianalyte genomic diagnostic test, the Oncotype DX Breast Cancer Assay. Genomic Health, Inc., Redwood City, CA. * Address correspondence to this author at: Genomic Health, Inc., 301 Materials and Methods Penobscot Dr., Redwood City, CA 94063. Fax 650-556-1132; e-mail mcronin@ gene selection for the oncotype dx assay genomichealth.com. Received July 14, 2006; accepted March 22, 2007. Review of published breast cancer biomarkers studies led Previously published online at DOI: 10.1373/clinchem.2006.076497 to the identification of 250 gene candidates, which were

1084 Clinical Chemistry 53, No. 6, 2007 1085

Table 1. Oncotype DX assay panel of 16 cancer-related and other variables unrelated to gene expression and genes and 5 reference genes.a patient outcome, we used the mean expression of the 5 reference genes to normalize the expression of the 16 Common Official gene gene name Sequence ID symbol cancer-related genes before calculating the Recurrence 2 BAG1 NM_004323 BAG1 Score™ (RS) (8–12). Bcl2 NM_000633 BCL2 CCNB1 NM_031966 CCNB1 samples CD68 NM_001251 CD68 We defined the histopathologic characteristics of samples CEGP1 NM_020974 SCUBE2 acceptable for Oncotype DX assay analysis during 3 pre- CTSL2 NM_001333 CTSL2 liminary clinical-association studies. Guidelines for select- EstR1 NM_000125 ESR1 ing fixed paraffin-embedded tissue (FPET) samples for GRB7 NM_005310 GRB7 the Oncotype DX assay are consistent with those for GSTM1 NM_000561 GSTM1 selecting optimal tissue blocks for standard immuno- HER2 NM_004448 ERBB2 histochemistry assays. Essentially, the block containing Ki-67 NM_002417 MKI67 the greatest amount of invasive breast carcinoma that is MYBL2 NM_002466 MYBL2 morphologically consistent with the submitting diagnosis PR NM_000926 PGR and having the least amount of noninvasive mammary STK15 NM_003600 AURKA epithelium is selected. Samples with metabolically active STMY3 NM_005940 MMP11 nontumor elements constituting Ͼ50% of the tissue have SURV NM_001168 BIRC5 those elements dissected out before sample extraction. ␤-Actin NM_001101 ACTB The goal of analytical validation was to provide an GAPDH NM_002046 GAPDH GUS NM_000181 GUSB assay process with a fully optimized and documented RPLPO NM_001002 RPLP0 standard operating procedure. Our studies focused on TFRC NM_003234 TFRC characterizing individual genes to define assay amplification efficiencies, linearity, dynamic range, reproducibility, a Genes annotated with NCBI LocusLink sequence identification numbers and official gene symbols of the HUGO Gene Nomenclature Committee (http:// and limits of detection and quantification. To measure the www.gene.ucl.ac.uk/cgi-bin/nomenclature/searchgenes.pl). realistic limits of assay performance uncomplicated by biological variability contributed by individual samples, we created a single pooled test sample that represented a evaluated in 3 clinical-association studies to find biomar- range of high and low expression values across the (5–7) kers related to the risk of breast cancer recurrence . 21-gene panel (details in the online Data Supplement). Statistical analysis and modeling of these study results led The final set of analytical validation studies tested the type to the selection of the 21 genes constituting the Onco performance limits of the RS with individual patient DX assay panel (Table 1; for primers and probes for all 21 samples to evaluate comprehensive assay performance in gene expression assays, see Table 1 in the Data Supple- a way that realistically reflected patient sample testing. ment that accompanies the online version of this article at http://www.clinchem.org/content/vol53/issue6). We fpet extraction chose 16 cancer-related genes for their consistent univar- RNA was extracted from 3 sections (10 ␮m thick) of each iate association with clinical outcome in all 3 clinical FPET block. Paraffin was removed by xylene extraction association studies, and we selected 5 reference genes 1 followed by ethanol washing. RNA was isolated from (ACTB, GAPDH, GUSB, RPLP0, and TFRC) TM for their deparaffinized tissue with the MasterPure Purification consistently low variation in expression and lack of asso- Kit (Epicentre Biotechnologies) with DNase I treatment, ciation with clinical outcome in the same clinical studies. as previously described (13). To compensate for sample variation in extracted RNA attributable to tissue-fixation processes, specimen age, rna quantification Extracted RNA was quantitated with the RiboGreen® fluorescence method (Molecular Probes/Invitrogen) as 1 Human genes: BAG1, BCL2-associated athanogene; BCL2, B-cell CLL/ described previously (13). lymphoma 2; CCNB1, cyclin B1; CD68, CD68 molecule; SCUBE2, signal peptide, CUB domain, EGF-like 2; CTSL2, cathepsin L2; ESR1, estrogen receptor 1; GRB7, growth factor receptor-bound protein 7; GSTM1, glutathione residual genomic dna S-transferase M1; ERBB2, v-erb-b2 erythroblastic leukemia viral oncogene Extracts were evaluated for residual genomic DNA with a homolog 2, neuro/glioblastoma derived oncogene homolog (avian); MKI67, ® ␤ antigen identified by monoclonal antibody Ki-67; MYBL2, v-myb myeloblas- TaqMan quantitative PCR assay for -actin DNA (13). tosis viral oncogene homolog (avian)-like 2; PGR, progesterone receptor; Samples exceeding the DNA threshold were retreated AURKA, aurora kinase A; MMP11, matrix metallopeptidase 11 (stromelysin 3); BIRC5, baculoviral IAP repeat-containing 5 (survivin); ACTB, actin, beta; GAPDH, glyceraldehyde-3-phosphate dehydrogenase; GUSB, glucuronidase, beta; RPLP0, ribosomal protein, large, P0; TFRC, transferrin receptor (p90, 2 Nonstandard abbreviations: RS, Recurrence Score; FPET, fixed paraffin-

CD71). embedded tissue; RT, reverse transcription; CT, cycle threshold. 1086 Cronin et al.: Analytical Validation of a Genomic Diagnostic Test

Ϫ with DNase I. Nine of the Oncotype DX gene primer sets relationship: amplification efficiency Ϸ (2 1/Slope Ϫ 1) ϫ (for BAG1, CCNB1, SCUBE2, GRB7, ERBB2, MKI67, 100%, where the slope is estimated from the simple linear

MYBL2, PGR, and TFRC genes) span intron sequences and regression of CT measurements vs log2 RNA concentra- so do not detect genomic DNA; the 12 remaining On- tion (15, 16). Each of 15 sample dilutions underwent cotype DX genes were tested for sensitivity to genomic RT-PCR analysis twice with triplicate assays. Mean CT DNA added into an FPET RNA pool. A threshold was set scores from both runs were averaged to assign a final at 3 SDs below the lowest value of genomic DNA detected expression value for each RNA concentration measure- for this assay to function as a reverse transcription (RT) ment used in the linear regression analysis. negative control. measurement of assay linearity rt For each gene in the assay panel, the range of linear assay RT of the purified FPET RNA was carried out as previ- response was assessed by comparing the proportionality TM ously described with the Omniscript RT Kit (Qiagen) of CT values relative to the input RNA concentrations and combined random hexamer and gene-specific prim- calculated for each of the 15 pooled test sample dilutions ing (13). used to measure amplification efficiency (see the online Data Supplement for sample dilution details). For this pcr analysis, the polynomial method originally proposed by Quantitative RT-PCR analysis was done in 384-well plates Krouwer et al. (17) and recommended Clinical and Lab- in a 10-␮L volume with cDNA equivalent to 2 ng RNA. oratory Standards Institute guidelines (18) were applied. The only exception was in the linearity study, in which Specifically, orthogonal polynomial regression was used Ϫ sample input varied over RNA equivalents between 2 10 to obtain coefficients and associated tests of significance and 23 ng/reaction. Plate cycling was carried out with ABI for the 1st (linear), 2nd (quadratic), and 3rd (cubic) order ® PRISM 7900HT instruments (Applied Biosystems) ac- polynomials. Given that the log of the variance in CT cording to the manufacturer’s instructions, as described measurement is proportional to the mean CT, error vari- previously (details in the online Data Supplement) (13). ance was modeled by use of a log-linear variance model. Model fits were obtained by use of the PROC Mixed instrumentation and reagent calibration procedure in SAS version 8.02. We calibrated liquid-handling robots to an independent In accordance with CLSI guidelines, any degree of calibration system. The performance of the ABI PRISM nonlinearity in signal response was assessed by examin- 7900HT instruments was assessed by calculating a within- ing the SE of the regression and selecting the higher-order plate CV with a standardized ribonuclease P assay (Ap- (nonlinear) polynomial model with the best fit. Specifi- plied Biosystems). cally, at each input RNA concentration, the deviation from linearity (DL) was calculated as follows: data reference normalization and rs calculation DL ϭ p͑ x ͒ Ϫ ͑b ϩ b x ͒, To make RT-PCR measurements comparable for clinical i i 0 1 i interpretation, the Oncotype DX assay normalizes gene where p(xi) is the value of the best-fitting polynomial at expression measurements to the mean expression of 5 RNA concentration xi. Consequently, DLi is a measure of reference genes (ACTB, GAPDH, GUSB, RPLP0, and the difference between the nonlinear model and the TFRC). We tested combinations of these genes for normal- best-fit straight line at each of the calculated RNA ization performance before selecting the mean of all 5 concentrations. genes for our standard reference normalization method. For each sample, normalized expression measurements measurement of quantitative bias and are calculated as the mean cycle threshold (CT) for the 5 precision reference genes minus the mean CT of triplicate measure- Assessments of analytical bias are typically obtained by ments for each individual gene. Normalized expression determining how much observed gene expression mea- measurements are scaled from 0 to 15 units, where 1 unit surements differ from expected expression values derived reflects an ϳ2-fold change in RNA quantity. from standard reference RNAs and definitive methods of After normalization, a single quantitative RS is calcu- analysis. Universal standard reference RNAs are not lated with expression values for 16 cancer-related genes, available for the 21 genes, nor are there universally as has been described previously (14). The RS, scaled accepted definitive analysis methods. Consequently, the from 0 through 100, expresses the likelihood of distant quantitative bias and imprecision of predicted RNA con- breast cancer recurrence and is specific to the tumor of centrations relative to the calculated input RNA amounts each individual patient. in the series of pooled sample dilutions was estimated for

each of the 21 genes. Specifically, for every CT measure- measurement of assay amplification efficiencies ment, an inverse prediction of RNA concentration was Amplification efficiency for each gene was approximated derived from the best fitting (nonlinear) polynomial cali- from RNA serial-dilution experiments recognizing the bration model derived during the linearity analysis (19). Clinical Chemistry 53, No. 6, 2007 1087

An estimate of the quantitative bias of the assay is given instrument, liquid-handling robot, and operator. All by the mean percent bias in prediction at each RNA plates within days were processed in randomized order. concentration k (k ϭ 1 . . . 15); namely, by: Mixed-effect ANOVA was used to divide calculated

total variability in observed CT measurement and RS into 2 3 components of variance due to days, plates within days, ͸ ͸ Ϫ xˆ ijk xk and within plates by treating operator, 7900HT instru- ϭ ϭ ϭ i 1 j 1 ment, and liquid-handling robot as fixed effects. For each Biask , 6xk of the 21 Oncotype DX genes and RS, restricted maximum likelihood estimates of the components of assay variance where xk is the calculated input RNA amount for the kth were obtained. level and xˆijk is the predicted RNA amount obtained from inverse prediction for the ith plate (i ϭ 1, 2), jth well (j ϭ Results 1, 2, 3), and kth RNA concentration value. amplification efficiency As a measure of the quantitative precision of the assay, Amplification efficiencies for individual genes must be for each RNA concentration level, ANOVA was used to similar for Oncotype DX test reference normalization and separate the total variability in the difference between RS calculation procedures to be valid (15, 16).Inan individual predicted and calculated RNA concentrations optimized PCR mixture without inhibitors or reagent into components of variance due to plate and well within limitations, amplification efficiency is an inherent charac- plate, treating plates and wells within plates as random. teristic of the primers and probes, so under a standard set For this purpose, a random effects model of the following of conditions, assay sets can be directly compared for form was applied to the individual differences between relative amplification efficiencies. Although individual ϭ ␮ predicted and calculated RNA concentrations: Ydiff ij patient samples may affect amplification efficiency, the ϩ ␣ ϩ ␧ j i(j), where Ydiff ij is the difference between the influence is on all of the transcripts, leaving the relation- predicted and actual RNA concentrations for the ith (i ϭ ships among the genes unaffected. Amplification efficien- 1, 2, 3) observation on the jth plate (j ϭ 1, 2), ␮ is the cies for the 16 cancer-related genes range from ϳ75% to ␣ ϳiid ␴2 ϳ overall mean difference, j N(0, ␣) is the effect of the 112%, whereas mean efficiency was 88% for the refer- ϭ ␧ ϳiid ␴2 jth plate (i 1, 2), and i(j) N(0, ␧) is random error. ence genes (Table 2). Some values exceeded 100% because We further assume that ␣s, ␤s, and ␧s distribute indepen- of inherent assay error throughout the cumulative analy- dently and that the distribution of error terms vary as a sis of the sample-dilution series. The 95% lower confi- function of calculated RNA concentration level (e.g., vari- dence bounds on the amplification efficiencies reflect ance increases near the limit of quantification of the model error as well as inherent assay error. The results assay). This information was used to obtain an estimate of within-plate CV in predicted RNA concentration at the kth RNA concentration level, namely: Table 2. Summary of amplification efficiency results for all 21 genes in the Oncotype DX assay. 2 Official gene Amplification 95% CI lower ͱ␴ˆ ␧/3 %CV ϭ , symbol efficiency, % boundary, % xk BAG1 112.10 94.80 BCL2 105.50 98.50 where xk is the calculated input RNA concentration for the CCNB1 89.40 81.30 ␴2 kth level and ˆ ␧ is the estimate of error variance derived CD68 84.90 79.00 from the ANOVA. Limit of detection and limit of quan- SCUBE2 96.20 89.80 tification values were calculated for each gene as well CTSL2 89.70 78.90 (details in the online Data Supplement). ESR1 97.80 93.50 GRB7 98.50 95.00 measurement of assay reproducibility GSTM1 93.50 86.90 Reproducibility in RS and in the measurement of the ERBB2 97.80 94.60 expression of individual genes ensures that results remain MKI67 109.00 100.90 comparable for patients over time and for different sub- MYBL2 75.40 68.20 mitting pathology laboratories. Individual gene and RS PGR 103.40 97.40 reproducibility were measured by performing repeat AURKA 87.10 79.50 analyses across multiple days, operators, RT-PCR plates, MMP11 101.40 98.00 7900HT instruments, and liquid-handling robots. Two BIRC5 101.20 93.10 ACTB 91.30 88.60 operators obtained replicate CT measurements on 2 aliquots of a single RNA sample over the course of 5 days GAPDH 75.30 73.70 with 3 7900HT instruments and 2 liquid-handling robots. GUSB 81.00 75.00 RPLP0 94.10 91.30 The study design has a G-efficiency of Ͼ50% (20), allow- TFRC 101.20 96.50 ing estimation of all main fixed effects, including 7900HT 1088 Cronin et al.: Analytical Validation of a Genomic Diagnostic Test

were within the experimental error expected for this Table 4. Quantitative bias and imprecision of predicted study design and were consistent among all 21 gene RNA concentration for all 21 genes in the Oncotype DX assays. assay at the standard assay RNA input of 2 ng/reaction. Deviation from assay linearity and dynamic range Official gene expected value CV at For each gene, assay linearity was assessed by comparing symbol at 2 ng RNA, % 2 ng RNA, % BAG1 Ϫ10.0 15.8 the proportionality of CT measurements with respect to calculated RNA concentrations over the series of sample BCL2 Ϫ2.7 5.5 dilutions used to measure amplification efficiency. The CCNB1 Ϫ6.3 7.2 quadratic and cubic polynomial terms were nonsignifi- CD68 Ϫ3.3 3.9 cant (P Ͼ0.05) for 6 genes (ACTB, BAG1, CD68, ESR1, SCUBE2 4.3 4.7 RPLP0, and TFRC), indicating linear performance over the CTSL2 6.0 7.3 Ϫ entire RNA range tested (2 10 to 8 ng). Another 6 genes ESR1 2.1 2.6 GRB7 2.8 2.0 (CCNB1, CTSL2, GSTM1, GUSB, MKI67, and PGR) were GSTM1 0.9 4.0 linear over more restricted RNA concentration ranges. For ERBB2 2.2 1.6 the remaining genes, the deviation from the linear model MKI67 2.6 2.7 was estimated to be within the prespecified acceptance MYBL2 Ϫ3.8 3.0 criterion of 1 CT unit. Based on the prespecified CLSI PGR Ϫ0.9 3.3 criterion, each of the 21 genes in the breast cancer– AURKA 3.6 4.1 recurrence gene panel had an estimated maximum devi- MMP11 0.6 1.9 ation from linearity of 1 CT over at least an 11-log2 BIRC5 Ϫ3.2 4.5 (Ͼ2000-fold) concentration range (Table 3). ACTB 3.3 2.5 At the highest RNA concentration tested in the series, GAPDH 1.5 0.8 8 ng RNA/reaction, CT values for the 21 genes varied over GUSB Ϫ1.5 6.0 an ϳ8-fold range (256-fold concentration range). Conse- RPLP0 Ϫ0.4 1.0 quently, the lowest-expressing genes (GUSB and CTSL2) TFRC 0.7 2.9 were studied over a more limited concentration range than high-expressing genes (Table 3). Low-expressing onstrate linear performances similar to highly expressed genes in this pooled sample would be expected to dem- genes at higher RNA concentrations. A linear assay response was seen for all genes over the expression range typically experienced in clinical samples. Table 3. Assay linearity and dynamic range results for each of the 21 genes in the Oncotype DX assay. assay quantitative bias and precision Official gene Linear RNA Minimum linear C ,8ng T The quantitative nature of the Oncotype DX RS depends symbol range, ng range RNA directly on analytical accuracy and precision of measure- BAG1 2Ϫ10 to 23 16 384-fold 27.4 Ϫ ment for each component of the 21-gene assay. Analytical BCL2a 2 10 to 23 16 384-fold 27.9 CCNB1 2Ϫ8 to 23 4096-fold 29.0 accuracy could not be assessed in the absence of standard CD68 2Ϫ10 to 23 16 384-fold 26.5 reference materials for each of the 21 analytes, so quanti- SCUBE2a 2Ϫ10 to 23 16 384-fold 27.7 tative bias was measured by comparing predicted RNA CTSL2 2Ϫ7 to 23 2048-fold 30.0 concentrations for each gene to expected RNA concentra- ESR1 2Ϫ10 to 23 16 384-fold 26.6 tions calculated for a range of sample dilutions. At every a Ϫ10 3 GRB7 2 to 2 16 384-fold 26.2 CT measurement, an inverse prediction of RNA concen- GSTM1 2Ϫ8 to 23 4096-fold 28.5 tration was derived from the best-fitting polynomial cal- ERBB2a 2Ϫ10 to 23 16 384-fold 23.7 ibration model derived from the linearity study (19). MKI67 2Ϫ9 to 23 8192-fold 28.5 Assay bias is expressed as mean percentage deviation MYBL2a 2Ϫ8 to 23 4096-fold 28.8 from the calculated value in prediction at each RNA PGR 2Ϫ9 to 23 8192-fold 28.1 concentration. At the 2-ng/well value used for the On- AURKAa 2Ϫ8 to 23 4096-fold 29.1 cotype DX assay, the estimated deviation from the ex- MMP11a 2Ϫ10 to 23 16 384-fold 24.8 pected value in the predicted RNA concentration ranged Ϫ BIRC5a 2 8 to 23 4096-fold 29.3 from Ϫ10% to 6%, with an estimated mean deviation from Ϫ10 3 ACTB 2 to 2 16 384-fold 21.9 the expected value of Ϫ0.3% for the 16 cancer-related a Ϫ10 3 GAPDH 2 to 2 16 384-fold 24.0 genes. For the reference genes, the mean percentage Ϫ7 3 GUSB 2 to 2 2048-fold 29.7 deviation from the expected value was 0.7%, indicating RPLP0 2Ϫ10 to 23 16 384-fold 22.8 Ͼ Ϫ10 3 99% mean quantitative correctness at this assay condi- TFRC 2 to 2 16 384-fold 27.2 tion (Table 4). a In accordance with a prespecified criterion, these genes exhibited a maximal Each gene was analyzed for analytical imprecision of deviation from linearity less than 1 C over the stated ЉlinearЉ range. T measurement. ANOVA was used at each RNA concentra- Clinical Chemistry 53, No. 6, 2007 1089

tion to estimate the total variance in predicted RNA liquid-handling robots and 7900HT instruments, were Ͻ concentration derived from the inverse calibration model 0.5 CT units for each of the 21 Oncotype DX genes (data vs the actual RNA concentration. This information was not shown). used to estimate the CV in the predicted RNA concentration at each known RNA concentration. For the standard monitoring and controlling assay reaction, the imprecision of measurement for the 16 performance in the clinical reference cancer-related genes on the RNA concentration scale had laboratory a CV of 5.7%, whereas the reference genes had a CV of Clinical validation of the Oncotype DX Breast Cancer 3.2%. All values were well within the prespecified accep- Assay was conducted after validating assay analytical tance limit of 20% (Table 4). performance with quality-control measures established from the results of analytical validation studies. The day-to-day reproducibility entire assay process, including process controls with ϳ SDs in CT measurements varied from 0.06 to 0.15 CT associated performance-acceptance limits, was docu- units for each of the 21 genes, and the upper bounds on mented as a series of standard operating procedures 2-sided 95% CIs for the CVs were all within 10%, indicat- that provide the basis for the current reference labora- ing a high degree of precision and reproducibility in the tory operation. assay (Table 5). These SD and CV values are for the A standard RNA control sample is assayed at least estimates of total variance. The between-day SD values once per batch of patient samples (Յ46 samples), and PCR were close to 0 for all 21 genes. A maximum SD of 0.15 at controls are run in every assay plate to verify that the aCT of 30 translates to a CV of 0.5%. At this level of process and reagents continue to perform within specified precision it is possible to reliably distinguish a 15% ranges. RT-PCR failures, identified by analyzing the am- change in expression for specific genes. plification curve from every assay well, are excluded from Additionally, pairwise differences in (least-squares) analysis. Expression values are assigned when at least 2 of mean C values between operators, liquid-handling ro- T 3 assay wells provide acceptable RT-PCR results. All 21 bots, and 7900HT instruments were calculated. The larg- genes must have an expression value assigned for an RS est differences between operators, as well as between to be calculated and reported. Process monitoring in our clinical reference laboratory shows that RS reproducibility remains very high. Table 5. Analytical reproducibility for normalized expression Repeat testing with deidentified patient samples shows measurements and restricted maximum likelihood (REML) a cumulative SD of Ͻ2 RS units on a 100-unit scale, estimates of the variance components for all 21 genes and which represents all sources of process variation RS in the Oncotype DX assay. (Table 6). REML estimates, CT units

Official gene Between-day Between-plate Within-plate Total Discussion symbol Wells SD SD SD SD The Oncotype DX RS, developed to quantify likelihood of ACTB 114 0.009 0.000 0.057 0.057 distant breast cancer recurrence for individual patients, BAG1 114 0.053 0.000 0.119 0.130 has been clinically validated in multiple studies (14, 21, BCL2 114 0.000 0.090 0.079 0.120 22). Each clinical study successfully met prospectively CCNB1 114 0.018 0.047 0.095 0.108 defined statistical endpoints after analytical validation CD68 114 0.001 0.000 0.125 0.125 studies allowed standardization of the underlying RT- SCUBE2 114 0.000 0.000 0.069 0.069 PCR–based laboratory test (4, 23). CTSL2 113 0.000 0.026 0.147 0.150 ESR1 113 0.035 0.051 0.076 0.098 This genomic diagnostic test is highly complex com- GAPDH 114 0.048 0.056 0.059 0.094 pared with more traditional clinical chemistry tests. Be- GRB7 114 0.000 0.000 0.088 0.088 cause it requires 21 quantitative measurements rather GSTM1 114 0.030 0.049 0.111 0.125 than the measurement of a single analyte, the assay does GUSB 114 0.000 0.000 0.103 0.103 not conform to standard assay-validation formats. Never- ERBB2 114 0.018 0.019 0.057 0.062 theless, successful analytical validation was critical to our MKI67 114 0.055 0.048 0.094 0.119 goal of reporting quantitative results for individual pa- MYBL2 114 0.026 0.007 0.092 0.096 tients. We defined a validation process analogous to PGR 114 0.040 0.025 0.078 0.091 validation methods for single analyte clinical assays to RPLP0 114 0.000 0.000 0.057 0.057 characterize assay amplification efficiency, linearity, AURKA 114 0.000 0.000 0.087 0.087 quantification limitations, dynamic range, analytical pre- MMP11 114 0.033 0.000 0.073 0.080 cision, and reproducibility performance of the RT-PCR BIRC5 114 0.000 0.000 0.079 0.079 process that underlies the Oncotype DX assay for individ- TFRC 114 0.000 0.021 0.081 0.084 ual genes and for the assay as a whole as reflected by the RS 38 0.000 0.000 0.792 0.792 RS (24). 1090 Cronin et al.: Analytical Validation of a Genomic Diagnostic Test

Table 6. SD of the normalized expression measurement for places a patient precisely within a defined continuum of the Oncotype DX assay, representing cumulative sources clinical outcomes on the basis of the results of clinical- of process imprecision shown for each gene and the RS. validation studies. Maintaining consistent, predictable RT-PCR assay per- Official gene symbol SD, reference normalized C T formance for individual genes throughout clinical valida- ACTB 0.01 tion allowed assay-performance effects to be reliably BAG1 0.03 BCL2 0.09 differentiated from true patient variability. Consequently, CCNB1 0.09 each biomarker was repeatedly confirmed as important in CD68 0.10 predicting patient outcome while a foundation was con- SCUBE2 0.11 currently established for continuous monitoring of assay CTSL2 0.11 quality in the clinical reference laboratory setting. Because ESR1 0.05 key assay-performance metrics are quantitatively moni- GAPDH 0.25 tored, patients can be assured that their clinical reports GRB7 0.22 are reliable and remain consistent with clinical-validation GSTM1 0.04 experience. This approach to analytical validation for a GUSB 0.03 high-complexity genomic diagnostic test has been dem- ERBB2 0.07 onstrated to be useful by more than 3 years of successful MKI67 0.06 operation in the clinical reference laboratory setting. Rou- MYBL2 0.30 tine monitoring of the ongoing performance of the test PGR 0.08 shows that it continues to perform within the originally RPLP0 0.05 defined analytical conditions and to give reliable results AURKA 0.10 of breast cancer risk to patients and their physicians. MMP11 0.02 BIRC5 0.05 TFRC 0.04 We gratefully acknowledge assistance from Jay Snable, Julie RS 1.53 Ballard, and Heidi Cheng in carrying out these studies; Jerry Hurst for helpful discussions; Stephanie Butler and Raymond Diagnostic gene expression assays fall into 2 principal Chao for assistance with documentation; and Joff Baker and categories (25). Both types test multiple analytes but Steve Shak for their critical review of the manuscript. diverge from one another in the quantitative analytical resolution they are able to achieve. One type of diagnostic test yields gene expression profiles for thousands or tens Grant/funding support: None declared. of thousands of candidate analytes by means of technol- Financial disclosures: All authors are employees of Genomic ogies such as hybridization microarrays, bead arrays, and Health, Inc. protein mass spectrometry (26–32). Initially large analyte sets ultimately may be reduced to smaller sets that classify References patients nearly as well as the original large test panel. 1. Hayes DF. Prognostic and predictive factors for breast cancer: Results obtained thus far indicate the challenge in vali- translating technology to oncology [Editorial]. J Clin Oncol 2005; dating this type of test for clinical utility, because such 23:1596–7. 2. Abramovitz M, Leyland-Jones B. A systems approach to clinical complex sets of targets—each with a relatively low pre- oncology: focus on breast cancer. Proteome Sci 2006;4:5. cision and correctness of measurement—lend themselves 3. Duffy MJ. Predictive markers in breast and other cancers: a review. to overfitting in clinical-development and -validation Clin Chem 2005;51:494–503. studies (33, 34). Reproducibly measuring large panels is 4. Simon R. Roadmap for developing and validating therapeutically difficult, and such measurements typically produce high relevant genomic classifiers. J Clin Oncol 2005;23:7332–41. CVs. This imprecision erodes the quality of patient clas- 5. Cobleigh MA, Tabesh B, Bitterman P, Baker J, Cronin M, Liu ML, sifications, thereby limiting the value of these assays in et al. Tumor gene expression and prognosis in breast cancer patients with 10 or more positive lymph nodes. Clin Cancer Res clinical decision-making (35–37). 2005;11:8623–31. The Oncotype DX assay is a prototype for an alternative 6. Esteban J, Baker J, Cronin M, Liu M-L, Llamas MG, Walker MG, et type of genomic diagnostic test. From hundreds of candi- al. Tumor gene expression and prognosis in breast cancer: date genes quantitatively evaluated for association with multi-gene RT-PCR assay of paraffin-embedded tissue [Abstract]. the clinical outcomes of individual patients during a Proc Am Soc Clin Oncol 2003;22:850. series of clinical-development studies, we selected an 7. Paik S, Shak S, Tang G, Kim C, Baker J, Cronin M, et al. Multi-gene optimal set of RT-PCR assays for high overall perfor- RT-PCR assay for predicting recurrence in node negative breast cancer patients: NSABP studies B-20 and B-14 [Abstract]. Breast mance in predicting the outcome of individual patients. In Cancer Res Treat 2003;82:A16. contrast to tests that dichotomize patients into general 8. Andersen CL, Jensen JL, Orntoft TF. Normalization of real-time populations by matching classification profiles, the out- quantitative reverse transcription-PCR data: a model-based vari- put from this test is individualized; a quantitative value ance estimation approach to identify genes suited for normaliza- Clinical Chemistry 53, No. 6, 2007 1091

tion, applied to bladder and colon cancer data sets. Cancer Res 24. Bustin SA. Absolute quantification of mRNA using real-time re- 2004;64:5245–50. verse transcription polymerase chain reaction assays. J Mol 9. Simon R. Development and validation of therapeutically relevant Endocrinol 2000;25:169–93. multi-gene biomarker classifiers. J Natl Cancer Inst 2005;97: 25. van’t Veer LJ, Paik S, Hayes DF. Gene expression profiling of 866–7. breast cancer: a new tumor marker. J Clin Oncol 2005;23: 10. Szabo A, Perou CM, Karaca M, Perreard L, Quackenbush JF, 1631–5. Bernard PS. Statistical modeling for selecting housekeeper 26. Huang E, Cheng SH, Dressman H, Pittman J, Tsou MH, Horng CF, genes. Genome Biol 2004;5:R59. et al. Gene expression predictors of breast cancer outcomes. 11. Thellin O, Zorzi W, Lakaye B, De Borman B, Coumans B, Hennen Lancet 2003;361:1590–6. G, et al. Housekeeping genes as internal standards: use and 27. Ma XJ, Salunga R, Tuggle JT, Gaudet J, Enright E, McQuary P, et al. limits. J Biotechnol 1999;75:291–5. Gene expression profiles of human breast cancer progression. 12. Vandesompele J, De Preter K, Pattyn F, Poppe B, Van Roy N, De Proc Natl Acad Sci U S A 2003;100:5974–9. Paepe A, et al. Accurate normalization of real-time quantitative 28. Perou CM, Jeffrey SS, van de Rijn M, Rees CA, Eisen MB, Ross DT, RT-PCR data by geometric averaging of multiple internal control et al. Distinctive gene expression patterns in human mammary genes. Genome Biol 2002;3:RESEARCH0034. epithelial cells and breast cancers. Proc Natl Acad Sci U S A 13. Cronin M, Pho M, Dutta D, Stephans JC, Shak S, Kiefer MC, et al. 1999;96:9212–7. Measurement of gene expression in archival paraffin-embedded 29. Ramaswamy S, Ross KN, Lander ES, Golub TR. A molecular tissues: development and performance of a 92-gene reverse signature of metastasis in primary solid tumors. Nat Genet transcriptase-polymerase chain reaction assay. Am J Pathol 2003;33:49–54. 2004;164:35–42. 30. van de Vijver MJ, He YD, van’t Veer LJ, Dai H, Hart AA, Voskuil DW, 14. Paik S, Shak S, Tang G, Kim C, Baker J, Cronin M, et al. A et al. A gene-expression signature as a predictor of survival in multigene assay to predict recurrence of tamoxifen-treated, node- breast cancer. N Engl J Med 2002;347:1999–2009. negative breast cancer. N Engl J Med 2004;351:2817–26. 31. Foekens JA, Atkins D, Zhang Y, Sweep FC, Harbeck N, Paradiso A, 15. Applied Biosystems. User bulletin #2: relative quantitation of et al. Multicenter validation of a gene expression-based prognos- gene expression. Foster City: Applied Biosystems, 2001:36pp. tic signature in lymph node-negative primary breast cancer. J Clin 16. Pfaffl MW. A new mathematical model for relative quantification in Oncol 2006;24:1665–71. real-time RT-PCR. Nucleic Acids Res 2001;29:e45. 32. Chemnitz JM, Driesen J, Classen S, Riley JL, Debey S, Beyer M, et 17. Krouwer JS, Schlain B. A method to quantify deviations from assay al. Prostaglandin E2 impairs CD4ϩ T cell activation by inhibition of linearity [published erratum appears in Clin Chem 1994;40: lck: implications in Hodgkin’s lymphoma. Cancer Res 2006;66: 1781]. Clin Chem 1993;39:1689–93. 1114–22. 18. NCCLS. Evaluation of the linearity of quantitative measurement 33. Brenton JD, Carey LA, Ahmed AA, Caldas C. Molecular classifica- procedures: a statistical approach. Approved guideline NCCLS tion and molecular forecasting of breast cancer: ready for clinical document EP6-A (ISBN 1-56238-498-8). Wayne, PA: NCCLS; application? J Clin Oncol 2005;23:7350–60. 2003:47pp. 34. Simon R, Radmacher MD, Dobbin K, McShane LM. Pitfalls in the 19. Brown P. Measurement, Regression and Calibration. Oxford, use of DNA microarray data for diagnostic and prognostic classi- United Kingdom: Oxford University Press, 1993:210pp. fication. J Natl Cancer Inst 2003;95:14–8. 20. Atkinson AC, Donev AN. Optimum Experimental Designs. New 35. Daly TM, Dumaual CM, Dotson CA, Farmen MW, Kadam SK, York: Oxford University Press, 1992:528pp. Hockett RD. Precision profiling and components of variability 21. Habel LA, Shak S, Jacobs MK, Capra A, Alexander C, Pho M, et al. analysis for Affymetrix microarray assays run in a clinical context. A population-based study of tumor gene expression and risk of J Mol Diagn 2005;7:404–12. breast cancer death among lymph node-negative patients. Breast 36. Stec J, Wang J, Coombes K, Ayers M, Hoersch S, Gold DL, et al. Cancer Res 2006;8:R25. Comparison of the predictive accuracy of DNA array-based multi- 22. Paik S, Tang G, Shak S, Kim C, Baker J, Kim W, et al. Gene gene classifiers across cDNA arrays and Affymetrix GeneChips. J expression and benefit of chemotherapy in women with node- Mol Diagn 2005;7:357–67. negative, estrogen receptor-positive breast cancer. J Clin Oncol 37. Ma C, Lyons-Weiler M, Liang W, LaFramboise W, Gilbertson JR, 2006;24:3726–34. Becich MJ, et al. In vitro transcription amplification and labeling 23. Ransohoff DF. Rules of evidence for cancer molecular-marker methods contribute to the variability of gene expression profiling discovery and validation. Nat Rev Cancer 2004;4:309–14. with DNA microarrays. J Mol Diagn 2006;8:183–92. Clinical Chemistry 53:6 Lipids, Lipoproteins, 1092–1098 (2007) and Cardiovascular Risk Factors

A Strong Interaction between Serum ␥-Glutamyltransferase and Obesity on the Risk of Prevalent Type 2 Diabetes: Results from the Third National Health and Nutrition Examination Survey

Ji-Sun Lim,1 Duk-Hee Lee,1* Joo-Yun Park,1 Soo-Hee Jin,1 and David R. Jacobs, Jr.2,3

Background: Some studies have found an association of that obesity itself may not be a sufficient risk factor for obesity with type 2 diabetes only among individuals type 2 diabetes. Practically, this interaction can be useful with high normal serum ␥-glutamyltransferase (GGT) in clinical settings to identify individuals at high risk activity, not in those with low serum GGT. If this for type 2 diabetes. interaction reflected pathophysiology, it would have © 2007 American Association for Clinical Chemistry scientific and clinical importance. The findings failed to reach statistical significance, however, and no articles Although increased serum ␥-glutamyltransferase (GGT)4 have focused on the topic. We investigated possible has been regarded as a biomarker of hepatobiliary disease interactions between serum GGT and body mass index and alcohol consumption, cellular GGT is an ectoplasmic (BMI) and their effects on the risk of prevalent type 2 enzyme responsible for the extracellular catabolism of diabetes and homeostasis model assessment (HOMA) glutathione and is widely distributed in various cells with insulin resistance. high secretory or absorptive activities (1). Methods: We analyzed 4011 adults >40 years old who Recent population-based epidemiological studies have participated in the 3rd US National Health and Nutri- shown a strong association of serum GGT activities within tion Examination Survey. the reference interval with many cardiovascular disease Results: BMI was associated with prevalent diabetes risk factors or components of metabolic syndrome (2–14). only among persons with high normal serum GGT In addition, in prospective studies, baseline serum GGT ,In the highest serum activity predicted future diabetes, hypertension, stroke .(0.002 ؍ activity (P for interaction GGT quartile, adjusted odds ratios for BMI 25–29.9, and myocardial infarction (3–14). Among these diseases, 30–34.5, and >35 kg/m2 compared with BMI<25 kg/m2 serum GGT within the reference interval most strongly were 3.1, 5.1, and 6.2, respectively (P for trend <0.001). In predicted incident type 2 diabetes (5, 8–13). the lowest serum GGT quartile, BMI was not associated Recently, the prevalence and incidence of type 2 dia- with diabetes; corresponding adjusted odds ratios were betes have increased alarmingly worldwide and across all After prevalent age, sex, and race/ethnic groups (15). Although obesity is .(0.551 ؍ and 0.8 (P for trend ,1.8 ,0.9 ,1.0 diabetes was excluded, there was a parallel interaction undoubtedly an important risk factor for type 2 diabetes with HOMA levels (P for interaction <0.001). (16), some prospective cohort studies (8–10, 13) [with the Conclusions: BMI was not associated with prevalent exception of male participants in study (13)] found that type 2 diabetes when GGT was low normal, suggesting the association of obesity with the risk of type 2 diabetes may be weak to nonexistent in individuals with low normal serum GGT but strong in those with high normal 1 Department of Preventive Medicine and Health Promotion Research serum GGT. For example, in Korean men (8), incidence Center, School of Medicine, Kyungpook National University, Daegu, Korea. rates of type 2 diabetes were 0.4%, 0.8%, and 2.6% across 2 Department of Epidemiology, School of Public Health, University of Minnesota, Minneapolis, MN. 3 Department of Nutrition, University of Oslo, Oslo, Norway. * Address correspondence to this author at: Department of Preventive Medicine, School of Medicine, Kyungpook University, 101 Dongin-dong, 4 Nonstandard abbreviations: GGT, ␥-glutamyltransferase; BMI, body Jung-gu, Daegu 700-422, Korea. Fax 82-53-425-2447; e-mail [email protected]. mass index; HOMA, homeostasis model assessment; NHANES, National Received September 8, 2006; accepted April 3, 2007. Health and Nutrition Examination Survey; ALT, alanine aminotransferase; Previously published online at DOI: 10.1373/clinchem.2006.079814 POP, persistent organic pollutant.

1092 Clinical Chemistry 53, No. 6, 2007 1093

categories of body mass index (BMI) among study partic- Plasma glucose concentration was measured with an ipants with serum GGT Ͻ20 U/L, whereas corresponding enzymatic reaction, and insulin was measured by RIA. rates were 0%, 5.4%, and 22.7% among those with serum GGT Ͼ40 U/L. Like BMI, age was a strong risk factor for exclusion diabetes only among individuals with high normal GGT Fasting blood specimens were obtained from 11 448 par- in a prospective cohort study of Asian men (8), but this ticipants Ն40 years old. We excluded 4467 participants pattern of findings was not replicated in other prospective with missing data for serum GGT, glucose, insulin, BMI, cohort studies (9, 10). cigarette smoking, alcohol consumption, or leisure time Although the reported interactions of obesity and GGT physical activity. We also excluded 2334 participants who received little attention, perhaps because they failed to had fasted fewer than 8 h, and 3 women who were reach statistical significance, the interaction of obesity and pregnant. The sample for our analysis included 4011 GGT is scientifically important if it indicates that obesity participants. may not be a sufficient cause for developing type 2 diabetes. Confirmation of this finding could simplify statistical analysis diabetes screening, because overweight and obese people Participants were considered to have diabetes mellitus if with low normal GGT would no longer be considered at their fasting plasma glucose was 7.0 mmol/L, they were high risk. taking insulin or a hypoglycemic agent, or they reported Given the potential scientific importance of an interac- a history of diabetes diagnosed by a physician. To en- tion between obesity and GGT in predicting diabetes, hance our ability to remove confounding due to life- replication and careful analysis focused on the interaction style changes typically made after a diagnosis of dia- itself is needed. Therefore, we investigated whether the betes, including weight loss, we studied newly recognized associations of obesity or age with type 2 diabetes differed diabetes as 1 endpoint (that is, we deleted previously known diabetics from the analysis). The HOMA estimate according to serum GGT activities. In addition, we exam- of insulin resistance was calculated as follows: fasting ined the interactions with the homeostasis model assess- plasma insulin (mU/L) ϫ fasting plasma glucose ment (HOMA) estimate of insulin resistance after exclud- (mmol/L)/22.5. ing prevalent diabetes cases. Insulin resistance, often a We used multiple logistic regression to examine the diabetes precursor, was included as a 2nd dependent interaction relationships with prevalent diabetes as the variable as a marker of consistency of the findings. dependent variable, predicted from either BMI (Ͻ25, Ն 2 Materials and Methods 25–29.9, 30–34.9, and 35 kg/m ) or age (40–49, 50–59, 60–69, and Ն70 years) within 4 quartiles of GGT (Ͻ16, The National Health and Nutrition Examination Survey 17–22, 23–35, and Ն36 U/L). Next, we used linear regres- (NHANES) was conducted in the US from 1988 to 1994 by sion to examine interaction models with HOMA as the the National Center for Health Statistics of the CDC. The dependent variable and the same independent variables study population comprised complex, multistage, clus- used for logistic regressions, after excluding the 578 tered samples of civilian, noninstitutionalized individuals prevalent diabetes cases. Adjusting variables were race/ at least 2 months old. A detailed description of survey ethnicity, sex, poverty income ratio, smoking status methods and data collection procedures has been pub- (never smoker, exsmoker, and current smoker), alcohol lished (17). intake (frequency of drinking beer, wine, or liquor per month), and leisure time physical activity (created by measurements summing the products of the frequency of participation The NHANES data collection included a standardized by metabolic equivalent levels for each reported activity). home interview followed by a detailed physical examina- All statistical analyses were performed with SAS 9.1 and tion in a mobile evaluation clinic or in the participant’s SUDAAN 9.0 and accounted for correlations within pri- home. Information on a wide variety of sociodemo- mary sampling units. Our stratified models were adjusted graphic, health behavior, medical, nutritional, and family for age, race/ethnicity, and poverty income ratio, adjust- history questions, including self-reported age, race/eth- ments regarded as a good compromise between efficiency nicity, sex, history of smoking, alcohol consumption, and and bias (18, 19); therefore we did not use sampling physical activity, was obtained during the home inter- weights. view. The study protocol was reviewed and approved by the CDC institutional review board; additionally, in- Results formed written consent was obtained from all individuals The study population included 578 patients with preva- before they took part in the study. lent diabetes and 205 with newly recognized diabetes. Venous serum and plasma were collected into EDTA Although the marginal association of BMI with prevalent tubes and shipped weekly at Ϫ20 °C. Serum GGT activity diabetes was strong, the association varied by serum GGT was assayed with a Hitachi 737 Analyzer (Boehringer- activity (Table 1). As serum GGT activity increased, the Mannheim Diagnostics) at White Sands Research Center. association between BMI and prevalent diabetes strength- 1094 Lim et al.: Interaction between GGT and Obesity

Table 1. Adjusteda odds ratios (ORs) and 95% CI of prevalent diabetes by category of BMI after stratification by quartile of .(0.004 ؍ serum GGT (P for multiplicative interaction BMI, kg/m2

<25 25–29.9 30–34.9 >35 P All study participants Case/no. of participants 105/1316 224/1540 158/770 91/385 Prevalence, % 8.0% 14.6% 20.5% 23.6% Adjusted OR Reference 2.0 3.4 4.5 Ͻ0.001 (95% CI) (1.6–2.6) (2.6–4.5) (3.0–6.8) Stratified by serum GGT Ͻ16 U/L Case/no. of participants 26/455 17/341 11/119 2/49 Prevalence, % 5.7% 5.0% 9.2% 4.1% Adjusted OR Reference 0.9 1.8 0.8 0.551 (95% CI) (0.5–1.7) (0.8–4.6) (0.2–4.1) 17–22 U/L Case/no. of participants 28/337 51/385 21/178 12/102 Prevalence, % 8.3% 13.3% 11.8% 11.8% Adjusted OR Reference 1.6 1.6 1.7 0.139 (95% CI) (0.9–2.9) (0.9–3.1) (0.7–4.2) 23–35 U/L Case/no. of participants 27/268 65/417 59/254 34/113 Prevalence, % 10.1% 15.6% 23.2% 30.1% Adjusted OR Reference 1.7 3.2 4.7 Ͻ0.001 (95% CI) (1.1–2.6) (1.8–5.8) (2.2–10.3) Ն36 Case/no. of participants 24/256 91/397 67/219 43/121 Prevalence, % 9.4% 22.9% 30.6% 35.5% Adjusted OR Reference 3.1 5.1 6.2 Ͻ0.001 (95% CI) (1.8–5.3) (2.8–9.1) (3.0–12.9) a Multiple logistic regression, adjusted for age, sex, race/ethnicity, poverty income ratio, cigarette smoking, leisure time physical activity, and alcohol consumption. ened (P ϭ 0.002 for multiplicative interaction). For exam- BMI in prediction of diabetes; the nonsignificant estimate ple, within the lowest quartile of serum GGT, BMI was of interaction was further attenuated after adjustment for not associated with prevalent diabetes, in contrast to the serum GGT (data not shown). After adjustment for serum highest quartile of serum GGT, wherein adjusted odds ALT (data not shown), however, the interaction of BMI ratios for prevalent diabetes were 1.0, 3.1, 5.1, and 6.2 (P and serum GGT that enabled estimation of prevalent for trend Ͻ0.001). The interaction was more clearly shown diabetes persisted. with newly recognized diabetes cases (Fig. 1; note that The association of age with diabetes was not indicated quartiles 1 and 2 of serum GGT were merged given small by differences in serum GGT activity (Table 2; P for numbers of newly recognized events). The pattern of multiplicative interaction ϭ 0.12). Similarly, no GGT event rates in Table 1 and Fig. 1 also indicated that the interaction was seen for age in relation to either newly association between serum GGT and diabetes became recognized diabetes or HOMA (data not shown). stronger as BMI increased and that serum GGT was In exploratory analyses to detect interactions that may positively associated with diabetes even among patients enable prediction of prevalent or newly recognized dia- with BMI Ͻ25 kg/m2. When we analyzed waist circum- betes, we examined possible interactions of BMI with ference instead of BMI, a similar interaction with serum other clinical variables, including serum antioxidant vita- GGT was observed (data not shown). mins (vitamin C or carotenoid), inflammation markers After prevalent diabetes was excluded, an interaction (C-reactive protein), or serum lipids (LDL-cholesterol or that affected HOMA levels also occurred between BMI triglyceride). No interaction was identified (data not and serum GGT activity (Fig. 2). The association between shown). BMI and HOMA became stronger at higher serum GGT activity (P for additive interaction Ͻ0.001). Discussion Although serum GGT had a high correlation of 0.51 This study found that the association of obesity with the with another liver enzyme, serum alanine aminotransfer- prevalence of type 2 diabetes varied with serum GGT ase (ALT), serum ALT did not show any interaction with activity; BMI was associated with prevalent diabetes only Clinical Chemistry 53, No. 6, 2007 1095

Fig. 1. Adjusted odds ratios (ORs) and 95% CIs of newly recognized diabetes (205 cases) by category of BMI and quartiles (Q) of serum GGT. Numbers in the table are the numbers of cases and individuals at risk in each category. ORs were adjusted for age, sex, race/ ethnicity, poverty income ratio, cigarette smoking, leisure time physical activity, and alcohol consumption; all ORs were calculated with the reference group of subjects with lowest category of both BMI and serum GGT. First and 2nd quartiles of serum GGT were combined because of small numbers of cases of newly recognized diabetes.

when GGT was in its high normal range. In patients with previous cohort studies, and in fact, an interaction model low normal serum GGT activity, BMI was not associated appears to be a better fit to the data than does the simple with diabetes. Although these data were cross-sectional, adjustment model. this study was performed to further investigate and An interaction between BMI and GGT on type 2 confirm the limited findings concerning this interaction in diabetes could be important both scientifically and clini- prospective cohort studies (8–10, 13). The consistency of cally. The most interesting scientific finding was that BMI the finding for related outcome variables (newly recog- was not associated with type 2 diabetes among individu- nized diabetes and HOMA in patients without prevalent als with low normal GGT. This finding suggests that diabetes) adds inferential strength, because participants obesity itself may not be a sufficient risk factor for type 2 who do not know they have diabetes cannot take lifestyle diabetes. Rather, in this view, to be a risk factor for measures (such as weight loss, increased physical activity, diabetes, obesity must be coupled with other factors, such or diet change) in response to diabetes. as those associated with increased serum GGT. In this All epidemiological studies to date on serum GGT context, one suggestion has been that fatty liver disease, have focused on the main associations of serum GGT with such as nonalcoholic steatohepatitis, can explain the cur- various disease outcomes, including diabetes (3–7, 11, rent finding because fatty liver may be related to obesity, 12, 14). Three prospective cohort studies (8–10), however, serum GGT, and type 2 diabetes (20). Serum ALT is a investigated BMI subgroups and found an interaction, not more sensitive marker of fatty liver than is GGT (21), statistically significant, in which obesity was weakly as- however, and we observed no BMI and serum ALT sociated with incident diabetes in people with low normal interaction associated with prevalent diabetes. The inter- serum GGT but strongly associated in those with high action of diabetes prevalence and BMI with GGT was normal serum GGT. One possible interpretation of this specific; no interaction was seen with age or several other interaction, if it did not arise by chance, is that obese clinical variables representing antioxidants, inflamma- individuals with high normal GGT have already suffered tion, and serum lipids. subclinical pathological changes attributable to obesity, Interestingly, in our recent study of the NHANES whereas obese individuals with low normal GGT are at an dataset in which we observed a striking dose–response earlier stage of pathogenesis. In this interpretation, serum relation between serum concentrations of persistent or- GGT activity is an intervening factor in the association ganic pollutants (POPs; endocrine disruptors) and preva- between obesity and diabetes. If this were true, adjust- lence of diabetes (22), we also observed an interaction ment for GGT should have substantially attenuated the between POPs and BMI on the risk of diabetes, parallel to association between BMI and type 2 diabetes; however, the findings of current study. The primary source for adjustment for GGT (ignoring interaction) did not mate- exposure in the general population is through the diet, rially change the association between BMI and the risk of especially fatty animal foods. In that study, the associa- type 2 diabetes in either the present cross-sectional or tion between obesity and diabetes became stronger as 1096 Lim et al.: Interaction between GGT and Obesity

Fig. 2. Adjusted geometric means of HOMA by category of BMI and quartiles (Q) of serum GGT after excluding prevalent diabetes (P for interaction Ͻ0.001). Numbers in the table are the numbers of individuals in each category. Linear regression was adjusted for age, sex, race/ethnicity, poverty income ratio, cigarette smoking, leisure time physical activity, and alcohol consumption.

serum concentrations of POPs increased. Obesity was not logic mechanism underlies the interaction between BMI associated with diabetes among individuals with very and GGT and its association with diabetes. Characteriza- low concentrations of POPs, in whom the prevalence of tion of the interaction between BMI and GGT may help to diabetes itself was quite low. We interpreted these obser- define a high-risk target population in which timely vations as suggesting that POPs contained in the adipose intervention may prevent development of type 2 diabetes tissue, not obesity itself, may play a key role in the while allowing for less screening attention applied to pathogenesis of diabetes. A related important finding is a obese people who have low normal GGT. Such a clinical dose–response relation between serum concentrations of application may become more useful as the proportion of POPs and serum GGT in the same NHANES dataset (23); obese persons increases among the total population. as serum concentrations of POPs increased, serum GGT This study has several limitations. The current findings also increased. In that study (23), we interpreted serum must be interpreted with caution because of the cross- GGT within its reference range as a general marker of sectional study design. The possibility of reverse causality exposure to various environmental xenobiotics that are is lessened, however, because similar findings were metabolized by glutathione conjugation, including both shown in several prospective studies and the findings of diabetes-related xenobiotics, such as POPs, and other the present study were confirmed by the outcomes for compounds unrelated to diabetes (23). It is possible that newly recognized diabetes and HOMA. Another limita- recently reported associations between serum GGT and tion is that the NHANES dataset did not allow us to various health outcomes (2–14) may be explained by differentiate type 1 from type 2 diabetes. Because we increases in serum GGT due to the exposure to various restricted the study to participants Ͼ40 years old, how- environmental pollutants. Taken together, findings link- ever, the majority of diabetic cases were likely to be type ing serum GGT, POPs, obesity, and diabetes could be 2. When we further restricted the study to participants interpreted as indicating that obesity can increase the risk Ͼ60 years old or combined the 2 categories of BMI (Ͻ25 of type 2 diabetes only in the presence of diabetogenic and 25–29.9), similar relations were observed. In addition, environmental xenobiotics, which are associated with in people Ͼ40 years old, the number of individuals increased serum GGT. belonging to the category of low normal serum GGT who On the other hand, the current findings may be useful did not show a significant relation between obesity and in the clinical or research setting, whatever pathophysio- diabetes was ϳ50% the total study participants, and the Clinical Chemistry 53, No. 6, 2007 1097

Table 2. Adjusteda odds ratios (ORs) and 95% CI of prevalent diabetes by category of age after stratification by quartiles of .(0.100 ؍ serum GGT (P for interaction Age, years

40–49 50–59 60–69 >70 P trend All study participants Case/no. of participants 83/1128 107/765 169/908 219/1210 Prevalence, % 7.4% 14.0% 18.6% 18.1% Adjusted OR Reference 2.2 3.2 4.0 Ͻ0.001 (95% CI) (1.6–2.9) (2.3–4.5) (3.0–5.2) Stratified analysis by serum GGT Ͻ16 U/L Case/no. of participants 5/280 3/157 16/183 32/344 Prevalence, % 1.8% 1.9% 8.7% 9.3% Adjusted OR Reference 1.2 5.5 6.9 Ͻ0.001 (95% CI) (0.2–6.6) (1.5–19.8) (2.6–18.2) 17–22 U/L Case/no. of participants 16/233 11/187 24/215 61/367 Prevalence, % 6.9% 5.9% 11.2% 16.6% Adjusted OR Reference 0.9 1.7 2.8 Ͻ0.001 (95% CI) (0.4–2.0) (1.0–3.1) (1.6–5.1) 23–35 U/L Case/no. of participants 21/294 40/206 56/257 68/295 Prevalence, % 7.1% 19.4% 21.8% 23.1% Adjusted OR Reference 3.2 4.2 5.2 Ͻ0.001 (95% CI) (1.9–5.5) (2.2–8.0) (3.2–8.6) Ն36 Case/no. of participants 41/321 53/215 73/253 58/204 Prevalence, % 12.8% 24.7% 28.9% 28.4% Adjusted OR Reference 2.3 2.9 3.6 Ͻ0.001 (95% CI) (1.4–3.7) (1.8–4.7) (2.2–5.8) a Logistic Logistic regression adjusted for age, sex, race/ethnicity, poverty income ratio, cigarette smoking, leisure time physical activity, and alcohol consumption. number of diabetes cases was 168. Thus, it seems quanti- the interaction between BMI and serum GGT may be tatively improbable that the lack of association between useful for detecting a high-risk subpopulation of type 2 obesity and diabetes among those with low normal serum diabetes. GGT is attributable to a few patients with type 1 diabetes who were not obese or insulin resistant. The final limitation to be addressed is that patients in the newly recog- Grant/funding support: This work was partly supported by nized diabetes group included those who had had hyper- Project BK21. glycemia for several years before it was discovered or Financial disclosures: None declared. diagnosed. When we examine the association between BMI and diabetes, however, whether patients knew that they had diabetes or not may be important because References knowing about diabetes can lead to changes in behaviors 1. Whitfield JB. Gamma glutamyl transferase. Crit Rev Clin Lab Sci and body weight. To guard against this type of bias, 2001;38:263–355. common in cross-sectional analyses, we analyzed the 3 2. Nilssen O, Forde OH, Brenn T. The Tromso Study: distribution and outcomes of all diabetes, newly recognized diabetes, and population determinants of gamma-glutamyltransferase. Am J HOMA-IR; all of them were consistent. Epidemiol 1990;132:318–26. 3. Wannamethee G, Ebrahim S, Shaper AG. Gamma-glutamyltrans- In summary, our analysis of the NHANES III study ferase: determinants and association with mortality from ischemic heart disease and all causes. Am J Epidemiol 1995;142:699– participants revealed an interaction between BMI and 708. serum GGT that was associated with the risk of prevalent 4. Brenner H, Rothenbacher D, Arndt V, Schuberth S, Fraisse E, diabetes. These findings indicate that obesity itself may Fliedner TM. Distribution, determinants, and prognostic value of not be a sufficient risk factor for type 2 diabetes; obesity gamma-glutamyltransferase for all-cause mortality in a cohort of may enhance the risk of type 2 diabetes in the presence of construction workers from southern Germany. Prev Med 1997;26: diabetogenic xenobiotics. Clinically or in case findings, 305–10. 1098 Lim et al.: Interaction between GGT and Obesity

5. Perry IJ, Wannamethee SG, Shaper AG. Prospective study of gamma-glutamyltransferase as a risk factor for cardiovascular serum gamma-glutamyltransferase and risk of NIDDM. Diabetes disease mortality: an epidemiological investigation in a cohort of Care 1998;21:732–7. 163,944 Austrian adults. Circulation 2005;112:2130–7. 6. Miura K, Nakagawa H, Nakamura H, Tabata M, Nagase H, Yoshida 15. King, H, Aubert RE, Herman WH. Global burden of diabetes, M, et al. Serum gamma-glutamyl transferase level in predicting 1995–2025. Diabetes Care 1998;21:1414–31. hypertension among male drinkers. J Hum Hypertens 1994;8: 16. Mensah GA, Mokdad AH, Ford E, Narayan KM, Giles WH, Vinicor F, 445–9. et al. Obesity, metabolic syndrome, and type 2 diabetes: emerging 7. Jousilahti P, Rastenyte D, Tuomilehto J. Serum gamma-glutamyl epidemics and their cardiovascular implications. Cardiol Clin transferase, self-reported alcohol drinking, and the risk of stroke. 2004;22:485–504. Stroke 2000;31:1851–5. 8. Lee DH, Ha MH, Kim JH, Christiani DC, Gross MD, Steffes M, et al. 17. National Center for Health Statistics, US Department of Health Gamma-glutamyltransferase and diabetes-A 4 year follow-up and Human Services. Third National Health and Nutrition Exami- study. Diabetologia 2003;46:359–64. nation Survey, 1988–1994. References, manuals and reports 9. Lee DH, Jacobs DR, Gross M, Kiefe CI, Roseman J, Lewis CE, et (CD-ROM). Hyattsville, MD: Centers for Disease Control and al. Gamma glutamyltransferase is a predictor of incident diabetes Prevention, 1996. and hypertension: the CARDIA Study. Clin Chem 2003;49:1358– 18. Korn EL, Graubard BI. Epidemiologic studies utilizing surveys: 66. accounting for the sampling design. Am J Public Health 1991;81: 10. Lee DH, Silventoinen K, Jacobs DR Jr, Jousilahti P, Tuomileto J. 1166–73. Gamma-glutamyltransferase, obesity, and the risk of type 2 19. Graubard BI, Korn EL. Analyzing health surveys for cancer-related diabetes: observational cohort study among 20,158 middle-aged objectives. J Natl Cancer Inst 1999;91:1005–16. men and women. J Clin Endocrinol Metab 2004;89:5410–4. 20. Brunt EM. Nonalcoholic steatohepatitis. Semin Liver Dis 2004; 11. Andre P, Balkau B, Born C, Charles MA, Eschwege E, D.E.S.I.R. 24:3–20. Study Group. Three-year increase of gamma-glutamyltransferase 21. Yu AS, Keeffe EB. Elevated AST or ALT to nonalcoholic fatty liver level and development of type 2 diabetes in middle-aged men and disease: accurate predictor of disease prevalence? Am J Gastro- women: the D.E.S.I.R. cohort. Diabetologia 2006;49:2599–603. enterol 2003;98:955–6. 12. Nannipieri M, Gonzales C, Baldi S, Posadas R, Williams K, Haffner SM, et al. Liver enzymes, the metabolic syndrome, and incident 22. Lee DH, Lee IK, Song KE, Steffes M, Toscano W, Baker BA, et al. diabetes: the Mexico City diabetes study. Diabetes Care 2005; A strong dose-response relation between serum concentrations of 28:1757–62. Persistent Organic Pollutants and diabetes: results from the 13. Meisinger C, Lowel H, Heier M, Schneider A, Thorand B. KORA National Health and Examination Survey. Diabetes Care 2006;29: Study Group: serum gamma-glutamyltransferase and risk of type 2 1638–44. diabetes mellitus in men and women from the general population. 23. Lee DH, Jacobs DR. Association between serum concentrations of J Intern Med 2005;258:527–35. persistent organic pollutants and gamma glutamyltransferase: 14. Ruttmann E, Brant LJ, Concin H, Diem G, Rapp K, Ulmer H. results from the National Health and Examination Survey 1999– Vorarlberg Health Monitoring and Promotion Program Study Group: 2002. Clin Chem 2006;52:1825–7. Clinical Chemistry 53:6 1099–1103 (2007) Hematology

Comparison of Bromcresol Green and Agarose Protein Electrophoresis for Quantitation of Serum Albumin in Multiple Myeloma

Christine L.H. Snozek,1 Amy K. Saenger,1 Philip R. Greipp,2 Sandra C. Bryant,3 Robert A. Kyle,2 S. Vincent Rajkumar,2 and Jerry A. Katzmann1*

Background: The International Staging System for mul- Conclusions: Both BCG and PEL correlate well to neph- tiple myeloma has increased the importance of accurate elometry in sera with zero/low M-spikes. In the pres- measurement of serum albumin. Two common albumin ence of larger M-spikes, PEL correlates poorly to neph- assays, bromcresol green (BCG) and agarose gel protein elometry or BCG, whereas BCG compares well with electrophoresis (PEL), frequently yield discordant re- nephelometry regardless of M-spike. Thus, albumin sults, creating confusion regarding which assay is supe- measurement can be performed reliably in myeloma rior for use in myeloma. patient sera by use of inexpensive, automated BCG assays. Methods: We measured albumin by BCG on a Roche © 2007 American Association for Clinical Chemistry Modular system, by PEL with a Helena SPIFE SPE Vis agarose gel, and by immunonephelometry performed on Discordant results in the measurement of serum albumin a Dade Behring BNII nephelometer. BCG and PEL were concentrations have long been recognized with various used to measure albumin in 5777 patient samples, and assays, including agarose gel protein electrophoresis 4 all 3 methods were used in an additional 252 samples. (PEL), bromcresol green (BCG), and bromcresol purple The clinical impact was assessed on 698 myeloma pa- (BCP) (1, 2). This issue has recently increased in impor- tient samples. tance with development of the International Staging Sys- tem (ISS) for multiple myeloma, which estimates progno- Results: For sera with zero/low monoclonal immuno- ␤ ␤ globulin protein (M)-spike (0 to <15 g/L), results for sis according to serum 2-microglobulin (S 2M) and albumin concentrations (3). Stage I (longest median sur- both BCG and PEL correlated well to nephelometry, vival) is indicated by S␤ M Ͻ3.5 mg/L and albumin Ն35 although median PEL results were 8 g/L lower than 2 g/L and stage III (shortest survival) by S␤ M Ն5.5 mg/L corresponding BCG measurements. Correlation be- 2 irrespective of albumin concentrations. Stage II (interme- tween PEL and nephelometry or BCG diminished with diate survival) is indicated by concentrations that differ increasing M-spike, with PEL eventually overestimating from those characterizing stages I and III. Although albumin compared with both other assays. IgG and IgA clinical progression for myeloma patients is still measured M-spikes showed significantly different effects on al- by the size of the monoclonal immunoglobulin protein bumin discordance. For 35% of myeloma patients, dis- ␤ (M)-spike produced by the malignant cells, S 2M and crepancy between BCG and PEL had a potentially clin- albumin independently provide an initial measure of ically significant effect on staging, but no difference in prognosis. group survival was found. Three methods for measuring albumin are used in our clinical laboratories: automated analysis of BCG binding, PEL quantified by densitometry, and immunological quantification by nephelometry. Only PEL and BCG mea- Departments of 1 Laboratory Medicine and Pathology, 2 Internal Medi- cine/Division of Hematology, and 3 Division of Biostatistics, Mayo Clinic surements of serum albumin are routinely used in clinical College of Medicine, Rochester, MN. * Address correspondence to this author at: Department of Laboratory Medicine and Pathology, Hilton Bldg. 210, Mayo Clinic, 200 First St. SW, Rochester, MN 55905. Fax 507-266-4088; e-mail [email protected]. 4 Nonstandard abbreviations: PEL, protein electrophoresis; BCG, brom- ␤ Received February 28, 2007; accepted March 29, 2007. cresol green; BCP, bromcresol purple; ISS, International Staging System; S 2M, ␤ Previously published online at DOI: 10.1373/clinchem.2007.088252 serum 2-microglobulin; M, immunoglobulin protein.

1099 1100 Snozek et al.: Serum Albumin Quantitation and Multiple Myeloma

practice. Immunonephelometric quantification is usually Mayo Clinic Division of Biostatistics. Recursive partition- reserved for cerebrospinal fluid, which has lower albumin ing [r-part analysis, (6)] was used to determine optimal concentrations (4). Although nephelometry is arguably M-spike values for partitioning differences in PEL and the most selective and therefore most likely to accurately BCG albumin measurements (i.e., points where the differ- quantify serum albumin, the relative expense and the ence PEL Ϫ BCG changes as a function of M-spike), with requirement for additional analytical equipment often M-spike and PEL Ϫ BCG as the independent and depen- make nephelometry impractical. dent variables, respectively. Standard linear and Deming To assess whether a particular albumin assay is pref- regression analyses were used to assess the relationships erable for use in multiple myeloma staging, we analyzed among albumin assays. For the myeloma patient cohort, the relationship between PEL and BCG analyses of albu- survival was defined as time from diagnosis of multiple min in a large number (n ϭ 5777) of patient samples both myeloma to death or last follow-up. Estimates were with and without a monoclonal immunoglobulin. In ad- generated by use of the method of Kaplan and Meier, and dition, we compared albumin concentrations from all 3 survival curves were compared by use of log-rank tests. available methods (PEL, BCG, and nephelometry) in an BCG measurement of albumin was performed with independent group of patient samples (n ϭ 252) over a Roche reagents on a Roche P-modular automated ana- range of M-spike values. Finally, we examined the effect lyzer according to the manufacturer’s procedure. of PEL and BCG albumin concentrations on staging and Electrophoretic quantitations of both M-spike and al- estimation of survival in a cohort of multiple myeloma bumin were performed on a Helena SPIFE 3000 electro- patients to determine the clinical impact of the discrepan- phoresis unit by use of a SPIFE SPE Vis agarose gel, which cies in albumin measurement. was stained with Acid Blue to visualize proteins and scanned with a Helena Quick Scan 2000 densitometer. The Materials and Methods relative amount of each protein fraction was determined This study followed a protocol approved by the Mayo by multiplying the relative amount by the total serum Institutional Review Board. Albumin results were re- protein concentration. Total protein was measured on a trieved retrospectively on 5777 consecutive patient sera Hitachi 912 automated analyzer, by use of a Roche color- analyzed from August 1, 2005, to August 1, 2006; all imetric biuret assay. patients studied had given research authorization. The Nephelometry was performed on a Dade Behring BNII only inclusion criterion was that both BCG and PEL immunonephelometer. Albumin was measured by use of assays had been performed on each sample. Of 404 Dade Behring antibodies specific to albumin; after anti- samples containing an M-spike, 280 (69.3%) were IgG, 47 body binding, light scatter at 840 nm was used to quantify (11.6%) IgA, 59 (14.6%) IgM, 1 (0.2%) IgD, 4 (1.0%) light the target protein. In addition, for IgG M-spikes Ͼ30 g/L, chain only, 11 (2.7%) biclonal with 2 different isotypes, antiserum to the gamma chain was used to accurately and 2 (0.5%) did not have the isotype determined. quantify the monoclonal protein, because IgG M-spikes of An additional 252 sera were collected prospectively Ͼ30 g/L may saturate the electrophoresis gel (7, 8). and analyzed by all 3 albumin methods; consecutive samples were selected such that roughly half had little or Results no M-spike (0 to Ͻ15 g/L, n ϭ 118) and half had an To determine the relationship between PEL and BCG M-spike Ն15 g/L (n ϭ 134). Of 166 samples containing an measurements of serum albumin, we retrieved informa- M-spike, 116 (69.9%) were IgG, 25 (15.1%) IgA, 21 (12.7%) tion on all patient samples that had been subjected to both IgM, and 4 (2.4%) biclonal with 2 different isotypes. assays within a 1-year interval. Of the 5777 samples Data from a cohort of 1027 multiple myeloma patients analyzed, 5373 were from patients with no quantifiable were retrieved retrospectively; this patient population is M-spike; M-spikes in the 404 remaining samples were 1 to well-defined and has been previously described (5). All 79 g/L. PEL and BCG values for albumin were compared serum measurements were made within 30 days of diag- for all 5777 patients. Recursive partitioning analysis sug- nosis; the 698 patients described herein were chosen for gested that the relationship between PEL and BCG having had both PEL and BCG measurements of albumin changed with increasing M-spikes, with cutpoints at performed on the initial serum sample. The isotype dis- 13.5, 24.5, and 40.5 g/L M-spikes. We therefore catego- tribution in these samples was as follows: 359 (51.4%) IgG, rized the patient samples into 4 groups according to the 152 (21.8%) IgA, 2 (0.3%) IgM, 11 (1.6%) IgD, 128 (18.3%) concentration of M-spike (Table 1). The mean difference light chain only, 36 (5.2%) nonsecretory, and 10 (1.4%) between BCG and PEL values changed progressively with biclonal with 2 different isotypes. Sera whose M-spike increasing M-protein concentrations, from Ϫ8 to 4 g/L. isotype was not determined, sera with biclonal (2 iso- This change in the relationship between BCG and PEL types) M-spikes, and sera from nonsecretory myeloma was statistically significant (P Ͻ0.0001) and is shown in were excluded from the analysis of the influence of Fig. 1A. isotype on albumin discordance. To assess the relative accuracy of the BCG and PEL Statistical analyses were performed in SAS version 9.1 albumin values, an additional 253 consecutive sera with (SAS Institute) and Splus version 7.0 (Insightful) by the M-spikes of 0–55 g/L were analyzed by BCG and PEL Clinical Chemistry 53, No. 6, 2007 1101

Table 1. Relationship between BCG and PEL measurements contrast, BCG analysis showed a stable relationship to of serum albumin. nephelometry across all M-spike values (Fig. 1C). The overall correlation between BCG and nephelometry was Median difference, :M-spike (g/L) n PEL ؊ BCG, g/L (range) good (R ϭ 0.96); Deming regression showed the following Zero/low (0 to Ͻ15)a 5617 Ϫ8.0 (Ϫ29.0, 15.0) ϭ ϫ ϩ Low-intermediate (15 to Ͻ25) 79 Ϫ5.0 (Ϫ20.0, 8.0) (BCG) 0.81 (Nephelometry) 7.17. High-intermediate (25 to Ͻ40) 52 Ϫ2.0 (Ϫ13.0, 17.0) High (Ն40) 29 4.0 (Ϫ11.0, 14.0) To address the impact of disagreement between PEL Total 5777 and BCG measurements of albumin in multiple myeloma, a A value of 0 may represent the lack of a monoclonal protein or the presence we analyzed data from a well-defined cohort of 1027 of a monoclonal protein that is too small to quantify. myeloma patients (5). Of these, 698 had PEL and BCG assays run on the same serum sample within 30 days of diagnosis; these results were used to assess whether the and immunonephelometry as an independent 3rd immunoglobulin isotype of the M-spike could affect the method. Notably, these 252 samples included 81 patients disparity between PEL and BCG results. To obtain suffi- Ͻ with albumin values 35 g/L as measured by nephelom- cient numbers, we combined the 698 myeloma patient etry; 35 g/L is the lower reference limit for albumin in our results with the M-spike–containing samples from the laboratory, as well as the ISS cutoff between stages I and 5777- and 252-patient groups. There were too few samples Ͻ II. For patients with zero/low M-spikes ( 15 g/L), both containing only IgM, IgD, or light chain to obtain mean- BCG and PEL demonstrated good correlation to nephe- ingful statistical analyses for the largest M-spike catego- 2 ϭ ϭ lometry (R 0.92 and 0.91, respectively; n 118), ries, but when these isotypes were compared with the rest although PEL showed a negative bias relative to nephe- of the data set in the smaller M-spike ranges, no signifi- lometry. On inclusion of samples with larger M-spikes, cant differences were seen. In contrast, IgG-containing however, PEL analysis showed poor correlation to neph- sera showed significantly greater discordance in albumin ϭ elometry (R 0.67); Deming regression showed the concentrations in the highest and lowest M-spike catego- following: ries than did IgA-containing sera (Table 2). (PEL) ϭ 0.48 ϫ (Nephelometry) ϩ 15.98. Albumin concentration of 35 g/L is an important cutoff for defining ISS stage. We therefore determined how This poor correlation was largely attributable to a many of the 698 myeloma patient samples had discordant progression from underestimation of albumin at low PEL and BCG values across this cutoff, i.e., 1 albumin M-spikes to overestimation at large M-spikes (Fig. 1B). In measurement Ն35 g/L and the other Ͻ35 g/L. Approxi-

Fig. 1. Variance in reported albumin concentrations by M-spike. Note that an M-spike of 0 may represent the lack of a monoclonal protein or the presence of a monoclonal protein that is too small to quantify. (A), difference (in g/L) between PEL and BCG albumin values for 5777 patient sera according to M-spike. Linear regression [(PEL Ϫ BCG) ϭ 0.19 ϫ (mean) Ϫ 7.9]; solid line) and 95% confidence limits (dotted lines) are shown. (B), difference (in g/L) between electrophoresis (PEL) and nephelometry (neph) albumin values for 252 patient sera according to M- spike. (C), difference (in g/L) between BCG and nephelometry (neph) albumin values for 252 patient sera according to M-spike. 1102 Snozek et al.: Serum Albumin Quantitation and Multiple Myeloma

Table 2. Influence of M-spike isotype on albumin samples to assess the correlation of PEL or BCG values assay disagreement. with patient outcome. For both assays, albumin concentration Ն35 g/L was predictive of longer survival than IgG IgA albumin Ͻ35 g/L (P Ͻ0.001). Neither assay performed Median PEL ؊BCG, Median PEL ؊ BCG, better than the other; Kaplan-Meier plots of survival in M-spike, g/L n g/L (range) n g/L (range) patients with albumin Ն35 g/L or Ͻ35 g/L are indistin- 0toϽ15a 218 Ϫ7.0 (Ϫ29.0, 6.0) 55 Ϫ6.0 (Ϫ14.0, 3.4) guishable for PEL and BCG (Fig. 3). Similarly, the hazard 15 to Ͻ25 135 Ϫ5.0 (Ϫ20.0, 6.0) 50 Ϫ4.0 (Ϫ11.0, 8.0) ratios for albumin Ͻ35 g/L are identical at 1.3 (95% 25 to Ͻ40 212 Ϫ0.3 (Ϫ10.0, 17.0) 57 Ϫ1.0 (Ϫ13.0, 7.0) confidence limits 1.1, 1.6 and 1.1, 1.5 for BCG and PEL, Ն40b 190 5.0 (Ϫ4.3, 27.1) 62 3.0 (Ϫ13.0, 11.5) respectively), indicating that PEL and BCG albumin con- a ϭ P 0.019, IgG vs IgA. centrations are equally good predictors of patient survival b P Ͻ0.0001, IgG vs IgA. in multiple myeloma. mately one third (35.0%) of samples showed discordance Discussion across the 35 g/L cutoff (Fig. 2). In the presence of Measurement of serum albumin in multiple myeloma zero/low or intermediate M-spike, BCG is almost invari- patients has increased in importance since the recent ably the higher of the albumin measurements, whereas in introduction of the ISS, which stratifies prognosis accord- ␤ the high M-spike category, PEL provides consistently ing to concentrations of albumin and S 2M (3). We used higher albumin values. This pattern mirrors the findings immunonephelometry as an independent 3rd method to described in Table 1 and Fig. 1A and could affect staging clarify the discordance between BCG and PEL measure- of such patients by the ISS. Interestingly, myeloma patient ments of albumin, long recognized to have poor correla- sera containing IgA and IgG had similar numbers of tion (1, 2), and determined the impact of discrepant discordant samples over the full M-spike range: 31.6% of albumin concentrations on staging and survival of pa- IgA-containing samples and 36.2% of IgG-containing tients from a well-defined multiple myeloma cohort (5). samples, but the distributions of discordant albumin A retrospective look at 5777 serum samples analyzed results were quite different between the 2 isotypes (data by both BCG and PEL confirmed lack of agreement not shown). In the lowest M-spike category, discordance between the 2 assays and indicated that the relationship occurred more frequently in samples containing IgA than between them was unstable with increasing M-spikes. IgG [7 of 18 (38.9%) vs 10 of 34 (29.4%) for IgA and IgG, Although PEL underestimates albumin relative to BCG respectively]. In the highest M-spike category this was when M-spikes are low, this difference gradually de- reversed, and more IgG-containing samples were discor- creases until, at M-spikes Ն40 g/L, the PEL albumin value dant [15 of 49 (30.6%) vs 74 of 159 (46.5%) for IgA and IgG, is on average 4 g/L higher than the corresponding BCG respectively]. value. The overall slope of this statistically significant Because discordant measurements of albumin could change is 0.19, increasing from 0.07 for zero/low M- influence ISS staging, we further analyzed the 698 patient spikes to 0.21 in the high M-spike group. In an additional 252 serum samples, the BCG assay showed good overall correlation to nephelometry (R ϭ 0.96), with a consistent relationship across the full M- spike range. Interestingly, although PEL has previously been suggested to be a more accurate measurement of albumin (1, 2), we found the overall correlation of PEL to nephelometry to be low (R ϭ 0.67) because PEL results

Fig. 3. PEL and BCG albumin assays are equally good predictors of Fig. 2. Discordance in albumin across the staging system cutoff. survival in multiple myeloma. Albumin values from both assays were compared for 698 multiple myeloma Albumin values from PEL and BCG were compared with survival in 698 multiple patient sera. Samples with albumin results discordant across the 35 g/L cutoff myeloma patients. A cutoff of 35 g/L was used (P Ͻ0.001, both assays); set by the ISS were categorized by the assay that reported the higher result. Data compared with the group with albumin Ն35 g/L, an albumin value Ͻ35 g/L are reported according to M-spike. showed a hazard ratio of 1.3 for both assays. Clinical Chemistry 53, No. 6, 2007 1103

showed underestimation of albumin at low M-spikes, is likely that BCP assays will behave similarly to BCG for shifting to overestimation of albumin at high M-spikes. predicting prognosis in multiple myeloma. BCG and BCP The latter effect is likely attributable to saturation of Acid tend to show acceptable linear correlation, with BCG Blue binding by large IgG M-spikes, leading to assign- albumin typically 3–5 g/L higher than the corresponding ment of a falsely high percentage of the total protein to BCP results (10). albumin (7, 8). For this reason our laboratory recom- Staging of multiple myeloma requires accurate deter- mends quantification of large (Ͼ30 g/L) IgG M-spikes by mination of albumin concentration in the presence of both nephelometry; our in-house PEL assay validation studies small and large M-spikes; as many as one third of patient of sera with large M-spikes suggest that dilution of the samples show discrepancies between BCG and PEL that samples does not provide adequate linearity for accurate may affect ISS staging. To our knowledge, this study is the measurement of serum proteins (J.A.K., unpublished first to address the utility of different albumin assays in data). Saturation of dense IgG bands is not observed on the presence of large M-spikes. The data presented here capillary zone electrophoresis (7). suggest that an automated assay with BCG methodology Analysis of the influence of M-spike isotype on albu- provides consistent measurement of albumin regardless min assay disagreement supports the concept that dye of the size of the M-spike; the PEL assay is less reliable saturation by large IgG M-spikes affects the accuracy of with large M-spikes and may report artifactually in- PEL measurements. Although the number of samples was creased albumin concentrations in these patients. insufficient for analysis of other isotypes (e.g., IgM, light chain only) across the full range of M-spike values, comparison of IgG and IgA showed significantly larger Grant/funding support: This work was supported in part by albumin discrepancies in IgG-containing sera for both the Research Grant CA62242 from the National Cancer Institute. lowest and highest M-spike categories. Interestingly, al- Financial disclosures: None declared. bumin measurements in 698 myeloma patients suggested Acknowledgements: We thank Karen Lockington, Thomas that almost half (46.5%) of all IgG-containing sera in the Milz, and Raynell Clark for technical assistance and Colin highest M-spike category have PEL and BCG albumin Colby and Robert Tarell for statistical analysis. values that are discordant across the 35 g/L ISS cutoff. Of 1027 previously described multiple myeloma pa- References tients, 698 had both PEL and BCG measurements of 1. Doumas BT, Peters T Jr. Serum and urine albumin: a progress albumin performed on the same serum sample within 30 report on their measurement and clinical significance. Clin Chim days of diagnosis. About one third of these samples had Acta 1997;258:3–20. discrepant albumin results that crossed the 35 g/L ISS 2. Lyon AW, Meinert P, Bruce GA, Laxdal VA, Salkie ML. Influence of methodology on the detection and diagnosis of congenital anal- cutoff, indicating that a substantial fraction of patients buminemia. Clin Chem 1998;44:2365–7. may be staged differentially by the 2 albumin assays, 3. Greipp PR, San Miguel J, Durie BG, Crowley JJ, Barlogie B, Blade ␤ depending on S 2M concentrations. Despite these differ- J, et al. International staging system for multiple myeloma. J Clin ences, PEL and BCG performed equally well as predictors Oncol 2005;23:3412–20. of patient survival, likely because the effect of albumin on 4. Seyfert S, Faulstich A. Is the blood-CSF barrier altered in disease? outcome is most strongly dictated by patients at the high Acta Neurol Scand 2003;108:252–6. and low extremes of the variable, and because patients 5. Kyle RA, Gertz MA, Witzig TE, Lust JA, Lacy MQ, Dispenzieri A, et with intermediate albumin concentrations (i.e., close to al. Review of 1027 patients with newly diagnosed multiple my- the 35 g/L cutoff) do not greatly impact the prognostic eloma. Mayo Clin Proc 2003;78:21–33. value of the test. This finding illustrates a point raised by 6. Therneau TM, Atkinson EJ. An Introduction to Recursive Partition- ing Using the RPART Routines. Rochester: Technical Report Series the ISS, that albumin alone has limited prognostic value No. 61, Mayo Clinic, 1997:52pp. and its use in prognosis is best when combined with other 7. Katzmann JA, Massey MA, Greipp PR, Clark RJ, Thompson CK, ␤ variables, such as S 2M (3). Lust JA, et al. Artifactually low IgG monoclonal protein (M-spike) For the majority of the general population, M-spikes quantitation on agarose gel electrophoresis: comparison of aga- will be absent, or present only in very low concentrations rose gel electrophoresis, capillary zone electrophoresis and neph- (9); notably, both BCG and PEL demonstrated good elometry [Abstract 655]. Am Assoc Clin Chem 52nd Annual correlation to nephelometry for M-spikes from 0 to Ͻ15 Meeting. Clin Chem 2000;46:A169. g/L. However, values from the 2 assays are often discrep- 8. Chang CY, Fritsche HA, Glassman AB, McClure KC, Liu FJ. Underestimation of monoclonal proteins by agarose serum protein ant even in the absence of an M-spike, because in normal electrophoresis. Ann Clin Lab Sci 1997;27:123–9. samples BCG typically overestimates albumin relative to 9. Kyle RA, Rajkumar SV. Monoclonal gammopathy of undetermined PEL. Measurements from the same patient should there- significance. Br J Haematol 2006;134:573–89. fore be performed by the same assay or interpreted with 10. Clase CM, St Pierre MW, Churchill DN. Conversion between this relationship in mind. Although this study did not bromcresol green- and bromcresol purple-measured albumin in address the utility of BCP in sera containing M-spikes, it renal disease. Nephrol Dial Transplant 2001;16:1925–9. Clinical Chemistry 53:6 1104–1108 (2007) Endocrinology and Metabolism

Evaluation of a 1,25-Dihydroxyvitamin D Enzyme Immunoassay

Isolde Seiden-Long1 and Reinhold Vieth1,2*

Background: Radioactive reagents are used in most the IDS EIA underestimates 1,25(OH)2D2 compared assays for measurement of 1,25-dihydroxyvitamin D with the D3 form. [1,25(OH)2D]. We evaluated a 1,25(OH)2D enzyme im- © 2007 American Association for Clinical Chemistry munoassay (EIA) from IDS Ltd. that uses solid-phase immunoextraction and colorimetric detection and com- Among the many reported metabolites of vitamin D (1),2 pared results to those of the thymus radioreceptor assay are measured to address clinical questions: 25-hydroxyvi- 3 (RRA) for 1,25(OH)2D. tamin D [25(OH)D, calcidiol] , representative of vitamin rep- D nutritional status, and 1,25-dihydroxyvitamin D (145 ؍ Methods: We collected serum samples (n resenting an even distribution (0–200 pmol/L) of [1,25(OH)2D, calcitriol], the bioavailable hormone in- 1,25(OH)2D concentrations and Vitamin D External volved in the regulation of calcium metabolism. Quality Assessment Scheme (DEQAS) proficiency sur- The measurement of 1,25(OH)2D presents several ana- and stored them lytical challenges. The compound circulates at picomole (15 ؍ vey samples from 2004 surveys (n -at ؊20 °C. We analyzed all samples with both EIA and per liter concentrations and is highly lipophilic. Fur RRA methods. We calculated imprecision using 5 QC thermore, the structurally similar metabolic precursor ,including 25(OH)D circulates at nanomole per liter concentrations ,(6 ؍ samples in quadruplicate in each run (n both pooled patient material used for QC with the RRA making assay specificity a constant analytical concern. and QC material included in the EIA reagent set. We Measurement of 1,25(OH)2D has undergone many im- evaluated calibration stability by analyzing calibrators provements (2). A radioreceptor assay (RRA) based on from different lots on the same plate and determining if the competitive binding of 1,25(OH)2D and tritiated tracer calculated sample values drifted significantly. to its nuclear receptor isolated from calf thymus was Results: Deming linear regression between IDS EIA introduced in the mid-1980s (3). The calf-thymus re- and RRA methods yielded slope 1.25 (95% CI 1.13–1.37), ceptor assay is attractive because it involves purification 2 DEQAS and binding to the natural receptor for 1,25(OH)2D and .0.74 ؍ y-intercept ؊3 (95% CI ؊18 to 12), R proficiency survey samples for 2004 were all within 30% adjustment for recovery. Commercial methods for of the all-methods-trimmed mean. Imprecision CVs 1,25(OH)2D do not adjust for recovery, and they involve were 12%–16% within-run and 15%–20% between-run. the use of antibodies. Our laboratory has continued to use Conclusions: We find no evidence of inferiority to the the calf-thymus receptor method, and is 1 of the last 2 classic calf-thymus receptor assay for 1,25(OH)2D and laboratories reporting data with the method to the Vita- no disadvantage in the results generated by the IDS EIA min D External Quality Assurance Scheme (DEQAS). using samples from the major proficiency survey for Most clinical laboratories who offer 1,25(OH)2D assays

1,25(OH)2D. According to the product insert, however, use commercially available methods (4). Currently, the most commonly used methods for 1,25(OH)2D quantification are competitive RIAs using a 125I tracer. The methods involve extraction, but they do not adjust for 1 Department of Laboratory Medicine and Pathobiology, University of between-sample variation in extraction efficiency. Two Toronto Pathology and Laboratory Medicine, Mount Sinai Hospital, Toronto, commercially available assays sold by DiaSorin and IDS Canada. 2 Department of Nutritional Sciences, University of Toronto, Toronto, Canada. * Address correspondence to this author at: Pathology and Laboratory Medicine, Mount Sinai Hospital, 600 University Ave., Toronto, Ontario, M5G 3 Nonstandard abbreviations: 25(OH)D, 25-hydroxyvitamin D;

1X5 Canada. Fax 416-586-8628; e-mail [email protected]. 1,25(OH)2D, 1,25-dihydroxyvitamin D; RRA, radioreceptor assay; DEQAS, Received August 1, 2006; accepted March 20, 2007. Vitamin D External Quality Assessment Scheme; EIA, enzyme immunoassay; Previously published online at DOI: 10.1373/clinchem.2006.077560 ALTM, all-methods-trimmed mean.

1104 Clinical Chemistry 53, No. 6, 2007 1105

dominate the market. They primarily differ in their re- Controls were run in duplicate on half plates and in spective 1,25(OH)2D extraction methods before assay quadruplicate on full plates and placed on the ends and and in sensitivity to 1,25(OH)2D2. The DiaSorin method middle of the plate. All samples were placed so that no uses a solid-phase extraction and silica purification replicate was ever placed in an immediately adjoining using organic solvents (acetonitrile, methanol, methylene well. The plate was incubated on an orbital shaker at chloride, hexane, and isopropanol) while the IDS method 18–25 °C for 90 min, then 1,25(OH)2D biotin solution was employs a solid phase immunoextraction, avoiding or- added to all wells except the substrate blanks, and the ganic solvents, but involving overnight incubation and a plate was incubated on an orbital shaker (500–750 rpm) at

2-day assay procedure. For the 1,25(OH)2D2 metabolite, 18–25 °C for 60 min. We washed the plate 3 times, added the DiaSorin and IDS RIA methods state 100% and 91% avidin linked to horseradish peroxidase to all wells except specificity, respectively. the substrate blanks, and incubated the plate at 18–25 °C A new enzyme immunoassay (EIA) for the measure- for 30 min. We repeated the washes and added tetra- ment of 1,25(OH)2D uses solid-phase immunoextraction methylbenzidine substrate to all wells including the sub- and colorimetric detection. The manufacturer’s claim that strate blanks. The reaction was developed at 18–25 °C for this method has only 39% reactivity of 1,25(OH)2D2 com- 30 min and stopped with 0.5 mol/L hydrochloric acid. pared with 1,25(OH)2D3 is a concern because vitamin D2 Plates were read at 450 nm (reference 630 nm) using a is commonly used clinically. We present our data for microplate reader within 30 min of adding the HCl comparison of the EIA to our in-house RRA, as well as an solution. evaluation of lot-to-lot calibration stability and potential We calculated percentage binding (B/B0) of each cali- compliance with external proficiency surveys. brator, control, and unknown sample as (mean absorbance Ϫ mean absorbance substrate blank) ϫ 100/(mean Materials and Methods absorbance for 0 calibrator Ϫ mean absorbance substrate thymus rra blank). We prepared a calibration curve by plotting

Calf-thymus receptor assay, involving purification of ana- B/B0% on the ordinate against log concentration of lyte on Bond Elut C18OH cartridges (Varian) and an 1,25(OH)2D on the abscissa. We fitted a 4-parameter internal tritiated calibrator to correct for losses during logistic model to the curve and interpolated sample data purification, has been described (5). Instead of an anti- using GraphPad software. body, this method is based on competition for the nuclear

1,25(OH)2D receptor extracted from calf thymus. samples Serum samples (n ϭ 145) for method comparison were ids eia previously assayed by thymus RRA in our laboratory and Reagent sets were donated by IDS Ltd., and we followed selected to represent an even distribution of typical (40– manufacturer instructions. We delipidated samples with 140 pmol/L), low (Ͻ40 pmol/L), and high (Ͼ140 pmol/L) dextran sulfate and magnesium chloride solution, centri- concentrations. We included 5 QC samples in each run fuged them to pellet debris, and extracted analytes using (Table 1) and pooled both patient material routinely used immunocapsules (100 ␮L sample/capsule, assayed in for QC with the in-house thymus RRA and the QC duplicate) containing monoclonal antibody to 1,25(OH)2D material from the manufacturer (lyophilized human se- linked to solid-phase particles in suspension with vitamin rum containing 1,25(OH)2D and 0.9 g/L sodium azide). D binding protein inhibitor. Immunocapsules were agi- DEQAS survey samples included tubes 141–155 from the tated on a rocker-shaker for 90 min at room temperature 2004 survey year. All samples were stored at Ϫ20 °C (18 to 25 °C), washed 3 times with deionized water, and between runs, and freeze-thaw cycles were minimized. extracted analyte was eluted with 3 successive applications of 150 ␮L ethanol. Eluates were evaporated under a statistical analyses gentle flow of nitrogen or by SpeedVac evaporator (Sa- All statistical analyses were performed using GraphPad vant). We reconstituted calibrators (lyophilized BSA- Prism software. To compare different lines (Figs. 1 and 2), phosphate buffer containing 1,25(OH)2D and 0.9 g/L we plotted the data as appropriate and determined sta- sodium azide) immediately before assay or thawed them from frozen. Dry, immunopurified samples were resuspended with assay buffer (BSA-phosphate buffer with 0.9 Table 1. QC imprecision values (% CV). g/L sodium azide). We added primary antibody solution Concentration, Within-run CV, Between-run CV, 6 ؍ n 8 ؍ pmol/L n (sheep anti–1,25(OH)2D in BSA-phosphate buffer with 0.9 g/L sodium azide) to all tubes, and they were incubated EIA control 1 40 14.0 19.3 EIA control 2 131 13.7 15.3 overnight (16–20 h) at 2–8 °C. The next day, we applied In-house level 1 46 15.8 19.7 samples to appropriate wells of the antisheep IgG–linked In-house level 2 121 13.1 20.4 96-well microplate. We selected the plate layout to mini- In-house level 3 237 12.4 19.2 mize sample positional bias. Calibrators were run in duplicate and placed on opposite ends of the plate. EIA controls were provided by the manufacturer. 1106 Seiden-Long et al.: Non-Radioactive 1,25(OH)2D Assay

Fig. 1. Comparison of the thymus RRA and IDS EIA assays. Deming (model II) linear regression analysis of 145 human serum samples measured for 1,25(OH)2D concentrations. (A), each sample was measured once by thymus RRA and in duplicate by IDS EIA (means shown). 95% CI of the slope ϭ 1.128– 1.370 and y-intercept ϭϪ18.19 to 12.07. (B), the same data plotted by lot number. The slopes of the 2 lines are significantly different (F-test, P Ͻ0.0001). 95% CI of the slope for lot 1 ϭ 1.2828–1.860 and lot 2 ϭ 1.067–1.309. 95% CI of the y-intercept for lot 1 ϭϪ29.81 to 30.65 and lot 2 ϭ Ϫ19.59 to 11.64.

tistical differences between lines or curves by F-test on the efforts eliminated factors such as mixing/reconstitution residuals of data points around the calculated line or problems and variability from different drying techniques curve. We determined the limit of detection by triplicate (SpeedVac concentrator vs nitrogen stream) and ulti- extraction and measurement of 4 serial 2-fold dilutions mately revealed a shift in calibration when calibrators using patient pool material at low concentration (i.e., Ͻ20 from different lots were run side-by-side on the same pmol/L). We calculated the limit of detection, defined as plate. The manufacturer subsequently informed us that the upper 95% CI for values at theoretical 0 concentration, they had changed their calibration between lots; the curve based on the SD of residuals (Sy͉x). fits in Fig. 2 demonstrate the curve shift. Additionally, we calculated patient data collected from the same plate Results using both calibration curves, and paired t-test demon- Method comparison by Deming linear regression of the strated a significant shift in sample values (P Ͻ0.0001). EIA and thymus RRA methods yielded slope 1.25 (95% CI We measured imprecision with 5 QC samples in qua- 1.13–1.37), y-intercept Ϫ3 (95% CI Ϫ18 to 12), R2 ϭ 0.74 druplicate in each run (n ϭ 6), including both pooled (Fig. 1A). We were initially supplied with 1 lot of EIA patient material used for QC with the RRA and QC reagent, which was divided into 2 runs and showed material included with the EIA reagent set. Samples were promising correlation with the thymus RRA method: analyzed once by RRA and twice by EIA in duplicate on slope 1.57 (95% CI 1.28–1.86), y-intercept 0.4 (95% CI Ϫ30 to 31), R2 ϭ 0.84 (Fig. 1B). With a 2nd lot of EIA reagents, the same plate. Duplicates for the EIA method were however, results for in-house QC material values were divided at the time of extraction, as opposed to pipetting lower, also corresponding to a significant shift in the slope duplicate wells from the same sample tube as is com- vs the RRA: slope 1.19 (95% CI 1.07–1.31), y-intercept 0.4 monly used with ELISA techniques not requiring an (95% CI Ϫ20 to 11), R2 ϭ 0.75 (Fig. 1B). Troubleshooting extraction step. Imprecision (CV) values were 12%–16% within-run and 15%–20% between-run (Table 1). The cumulative distribution of between-replicate variation (Fig. 3A) demonstrated a median between-replicate CV of 3.7% (25th percentile 1.8%, 75th percentile 7.3%, 95th percentile 18.6%). The distribution of between-replicate variation as a function of analyte concentration (Fig. 3B) demonstrates that variability is minimized within the

reference interval for 1,25(OH)2D (40–140 pmol/L). To calculate the relative contribution of extraction and EIA to the variability of the method, we repeated analysis of between-replicate variation using calibrators from each run (data not shown). Assuming that total error can be ␴2 ϭ ␴2 ϩ ␴2 expressed as total extraction EIA, and that we measure total error at mean between-replicate CV, then the total error would be Ϯ5.9%, the EIA assay error would be Ϯ2.4%, and the extraction error would be Ϯ5.4%. Thus, Fig. 2. Calibrator shift between lots. the extraction procedure contributes significantly more to Calibrators from both lots were analyzed in parallel on the same plate in the same run. Data are plotted using a 4-parameter logistic curve fit. The curves are the between-replicate error than the detection protocol of significantly different (F-test, P ϭ 0.0135). the EIA itself. Clinical Chemistry 53, No. 6, 2007 1107

Fig. 3. Analysis of between-replicate variability. CV values were calculated from mean difference between replicate measurements performed by IDS EIA. (A), cumulative frequency plot of between-replicate %CV (curve fit: LOWESS). (B), between-replicate %CV as a function of analyte concentration (curve fit: LOWESS).

The limit of detection, defined as the upper 95% CI for following criterion for 25(OH)D: 80% of samples to be values at theoretical 0 concentration, was determined by within 30% of the ALTM (4). Two methods dominate the use of replicates of extractions at low concentration. The calculation of the 1,25(OH)2D ALTM: Diasorin RIA (30% Syxvalue was 4.7 nmol/L. Thus, the limit of detection was of participants) and IDS RIA (59% of participants). IDS 9.4 pmol/L, consistent with our practice of reporting low EIA performance by single measurement (Fig. 4A) shows values as Ͻ10 pmol/L. that 14 of 15 samples for 2004 were within 30% of the DEQAS was established in 1989 for 25(OH)D and ALTM. The in-house RRA method also shows that 14 of extended in 1997 to include 1,25(OH)2D. DEQAS distrib- 15 samples were within 30% for 2004 (Fig. 4, B–D). EIA utes 5 serum samples 4 times per year to over 200 performance would be expected to improve if duplicates participants who measure 1,25(OH)2D) in 24 countries had been measured, although most laboratories routinely worldwide. The current analytical target for 25(OH)D is report clinical results based on singleton measurement. for 80% of samples to be within 30% of the all-methods- trimmed mean (ALTM) (4). In 2003, 59% of participants Discussion met the 25(OH)D target—a somewhat disappointing sta- The IDS EIA and our in-house thymus RRA compare well tistic. There are currently no analytical targets set for with other reports about the performance of 1,25(OH)2D 1,25(OH)2D assay performance by DEQAS, but for the assays (6–9). Although between-run imprecision seems purpose of comparison among methods, we used the high in the general context of clinical chemistry, it is

Fig. 4. Performance of IDS EIA method

for 1,25(OH)2D DEQAS data 2004. Values (141–155) on ordinate axis represent survey sample vial designations. Shaded area represents Ϯ30% from the ALTM. (A and B), we measured DEQAS

samples for 1,25(OH)2D in singleton using IDS EIA and our thymus RRA. (C and D), performance of method mean values for Diasorin RIA and IDS RIA methods as reported for the 2004 DEQAS sample set for

1,25(OH)2D. 1108 Seiden-Long et al.: Non-Radioactive 1,25(OH)2D Assay

equally within expected performance characteristics of In theory, the IDS EIA may underestimate 1,25(OH)2D2; assays of this analyte. While we observed between-run however, our data using clinical samples reveal no evi- CVs of 15%–20% in our evaluation, historical DEQAS data dence of inferiority to the classic calf-thymus receptor show that other methods are unlikely to provide better assay for 1,25(OH)2D, and no disadvantage in the results performance. Between-laboratory CVs in the 2004 DEQAS generated by the IDS EIA using samples from the major report were as follows: IDS RIA 11%–26%, Diasorin RIA proficiency survey for 1,25(OH)2D. 12%–30%, and overall between-method 15%–28%. The relative imprecision of this assay in general can in large part be attributed picomolar concentrations of analyte. Grant/funding support: None declared. However, we found that the extraction contributes more Financial disclosures: IDS Ltd. contributed assay reagent sets. variability than the immunoassay component of the Acknowledgements: We thank Flor Reyes, Syeda Hasani, procedure. and Shirley Siu for assay of serum samples by calf-thymus The shift in calibration was discussed with the IDS receptor assay. technical staff, and they have made assurances that this observation was attributable to an early preproduction lot References of reagents being distributed to us for evaluation. Never- 1. Bouillon R, Okamura WH, Norman AW. Structure-function relation- theless, it would be prudent for users to approach lot ships in the vitamin D endocrine system. Endocr Rev 1995;16: changes cautiously. This issue may be of less concern for 200–57. laboratories that intend to use a predetermined number of 2. Hollis BW. Detection of vitamin D and its major metabolites. In: Pike JW, Glorieux FH, Feldman D, eds. Vitamin D. San Diego: reagent sets for clinical trials that have a defined number Academic Press, 2004:931–50. of patients and may not require the long-term analytical 3. Reinhardt TA, Horst RL, Orf JW, Hollis BW. A microassay for stability of a clinical service laboratory. In those cases, 1,25-dihydroxyvitamin D not requiring high performance liquid reagents from a single lot for a given trial would be chromatography: application to clinical studies. J Clin Endocrinol optimal. It should be noted that calibrators are not ex- Metab 1984;58:91–8. tracted in the EIA method or the RRA method, but the 4. Carter GD, Carter CR, Gunter E, Jones J, Jones G, Makin HL, Sufi RRA method uses a tracer for monitoring extraction S. Measurement of Vitamin D metabolites: an international per- efficiency whereas the EIA method has no such adjust- spective on methodology and clinical interpretation. J Steroid Biochem Mol Biol 2004;89–90:467–71. ment. This process would putatively make the EIA 5. Hollis BW, Kilbo T. The assay of circulating 1,25(OH) D using method more difficult to troubleshoot should there be a 2 nonend-capped C18 silica: performance and validation. In: Nor- discrepancy between lots, as it would have to be deter- man AW, Schaefer K, Grigoleit H-G, von Herrath D, eds. Vitamin D: mined initially if the problem was with extraction or with Molecular, Cellular and Clinical Endocrinology. Berlin: W de- calibration. Gruyter, 1988:710–9. According to 2003 DEQAS participation data (4), 59% 6. De Leenheer AP, Bauwens RM. Comparison of a cytosol radiore- of participants currently use the IDS RIA method for ceptor assay with a radioimmunoassay for 1,25-dihydroxyvitamin D in serum or plasma. Clin Chim Acta 1985;152:143–54. 1,25(OH)2D, which uses the identical immunoextraction 7. Bertelloni S, Baroncelli GI, Benedetti U, Franchi G, Saggese G. method. Equally, the IDS RIA method for 1,25(OH)2D does not include calibrator extraction as part of the Commercial kits for 1,25-dihydroxyvitamin D compared with a liquid-chromatographic assay. Clin Chem 1993;39:1086–8. protocol. These results indicate that the laboratories cur- 8. Manolagas SC, Reitz R, Horst R, Haddad J, Deftos LJ. Multicentre rently using the IDS RIA and wishing to switch to a comparison of 1,25-dihydroxycholecalciferol measurements in hu- nonradioactive method could do so with relative ease. man serum. Lancet 1983;i:191–2. Our evaluation demonstrates that the IDS EIA would 9. Jongen MJ, Van Ginkel FC, van der Vijgh WJ, Kuiper S, Netelenbos easily meet the DEQAS goal of 80% of survey samples JC, Lips P. An international comparison of vitamin D metabolite within 30% of the all-methods mean (Fig. 4). measurements. Clin Chem 1984;30:399–403. Clinical Chemistry 53:6 1109–1114 (2007) Endocrinology and Metabolism

Bone Formation Markers in Adults with Mild Osteogenesis Imperfecta

Tim Cundy,1* Anne Horne,1 Mark Bolland,1 Greg Gamble,1 and James Davidson2

Background: Plasma concentrations of procollagen pep- utility in the diagnosis of other types of nondeforming tides are decreased in osteogenesis imperfecta (OI), OI. whereas other bone formation markers may be in- © 2007 American Association for Clinical Chemistry creased. We examined the utility of combining these 3 markers in the diagnosis of OI in adults. Osteogenesis imperfecta (OI) is often considered a diag- Methods: We measured plasma concentrations of pro- nostic possibility when severe osteoporosis presents rela- collagen-1 N-peptide (P1NP), osteocalcin, and bone al- tively early in life (1). Mild forms of OI can be difficult to kaline phosphatase in 24 patients with nondeforming diagnose, however; radiological findings are nonspecific, OI, 25 patients with low bone mass due to other causes, in some cases sclerae may be of indeterminate hue, there and 38 age- and sex-matched controls. The discriminant may be no obvious dental abnormalities, and family ability of various test combinations was assessed by the history is unhelpful (2). construction of ROC curves. The most commonly encountered forms of OI result from mutations in the COL1A14 or COL1A2 genes, which Results: The median (range) ratio of osteocalcin to P1NP encode the ␣1 and ␣2 chains of type 1 collagen. These was significantly greater in patients with type I OI [1.75 mutations result in the production of reduced amounts of (0.80–3.86)] than in controls [0.59 (0.34–0.90)] and pa- normal type 1 collagen and/or an abnormal type 1 tients with other causes of low bone mass [0.48 (0.05– collagen (3). During collagen synthesis, 1 ␣2 chain and 2 < 1.38); P 0.0001]. This ratio allowed nearly complete ␣1 chains assemble into a triple helix structure, after differentiation between healthy controls and patients which the nonhelical propeptides at both the C-terminal with type I OI, but not patients with type IV OI. With a and N-terminal ends of the trimer are cleaved. These cutoff of 0.97 for osteocalcin:P1NP, the sensitivity and propeptides—procollagen-1 C-terminal propeptide (P1CP) specificity were maximized at 95% (95% CI 76%–100%) and procollagen-1 N-terminal propeptide (P1NP)—can be and 88% (69%–97%), respectively, for patients with measured in plasma and are considered markers of bone other causes of low bone mass vs those with type I OI formation (4). only. For patients with other causes of low bone mass vs There is no specific, noninvasive, and inexpensive all OI patients, sensitivity and specificity were 83% laboratory test for OI; accurate diagnosis requires either (63%–95%) and 88% (69%–97%), respectively. The addi- sequencing of COL1A1 and COL1A2 or electrophoretic tion of bone alkaline phosphatase data did not improve studies on collagen secreted by cultured skin fibroblasts the discriminant ability of the osteocalcin:P1NP ratio. (5). Several studies investigating bone turnover markers Conclusions: The osteocalcin:P1NP ratio is a sensitive in OI have reported that, compared with normal individ- and specific test for type I OI in adults, but it has less uals, patients with OI have relatively low plasma concentrations of P1CP (6–9), which is believed to reflect the defective synthesis of type 1 collagen in OI. Several studies have suggested that the plasma concentrations of 1 Department of Medicine, Faculty of Medical and Health Sciences, Uni- other bone formation markers such as osteocalcin and versity of Auckland, Auckland, New Zealand. 2 Department of Clinical Biochemistry, Auckland City Hospital, Auckland, New Zealand. * Address correspondence to this author at: Department of Medicine, Faculty of Medical and Health Sciences, University of Auckland, Private Bag 3 Nonstandard abbreviations: OI, osteogenesis imperfecta; P1CP, procolla- 92019, Auckland, New Zealand. Fax 64-9367-7146; e-mail t.cundy@auckland. gen-1 C-terminal propeptide; P1NP, procollagen-1 N-terminal propeptide; ac.nz. bALP, bone-specific alkaline phosphatase; AUC, area under the curve. Received November 14, 2006; accepted March 14, 2007. 4 Human genes: COL1A1, collagen type 1, ␣ 1; COL1A2, collagen type 1, ␣ Previously published online at DOI: 10.1373/clinchem.2006.083055 2; LRP5, low-density lipoprotein receptor–related protein 5.

1109 1110 Cundy et al.: Formation Markers in OI

bone-specific alkaline phosphatase (bALP) may be in- undertook these studies on patients with type IV OI and creased in OI (7, 9, 10). These observations raise the dentinogenesis imperfecta. In brief, we cultured skin question that we addressed in this prospective study: fibroblasts and assessed the synthesis and secretion of could the measurement of bone formation markers, used type 1 procollagen relative to type 3 procollagen by in combination, be of diagnostic value in determining protein gel electrophoresis. In type 1 OI, normal procol- whether OI is the cause of severe osteoporosis presenting lagen is synthesized and secreted, but in reduced quan- in adulthood? tity. In other types of OI, 2 populations of type 1 procollagen are found, 1 normal and the other with abnormal Materials and Methods electrophoretic mobility (11). patients The study had local institutional review board ap- We measured plasma markers of bone formation and proval, and all participants gave informed consent. bone mineral density in the lumbar spine in 3 groups of adults: biochemical markers (a) Twenty-four patients with OI (from 15 different We measured the plasma concentrations of 3 bone forma- families) consecutively referred to our Bone Clinic for tion markers. In midafternoon, we drew blood samples evaluation. None had bone deformity and all were of into lithium-heparin bottles. We did not take account of normal stature (Ն3rd centile for height). Twenty-one of the stage of menstrual cycle or contraceptive use in the 24 had apparently normal teeth and sclerae of varying participating young women. The plasma was separated degrees of blue and thus had type I OI according to the and stored at Ϫ70 °C until analysis. The tests were carried Sillence classification (2). Three patients (2 with obvious out in fully accredited hospital laboratories, and the dentinogenesis imperfecta) had normal-colored sclerae analysts were masked from clinical information about the (type IV). In 18 cases, there was a clear family history of participants. dominant inheritance of the bone disease, but in 6 cases, We measured bALP by automated sandwich-type im- there was no other affected family member. Six of the 24 munoassay (Ostase) on a Beckman Access analyzer (Beck- patients were being treated with bisphosphonates (oral man Coulter). Interbatch imprecision (CV) was 8.4% at alendronate of Ͼ1 year duration) at the time blood 11.3 ␮g/L and 5.2% at 80.1 ␮g/L. The reference interval samples were drawn; in all the other cases, the samples quoted by the manufacturer is 4–22 ␮g/L for men and were taken before any specific antiosteoporotic medica- premenopausal women. We measured osteocalcin by tion was prescribed. The bone density measures reported sandwich immunoassay with electrochemiluminescence here were all made before any treatment was prescribed. detection on a Roche E170 analyzer (Roche Diagnostics). (b) Twenty-five patients (from 22 different families) This assay measures both the intact (1–49) osteocalcin with low bone mass due to causes other than OI, referred molecule and the N-Mid (1–43) fragment. Interbatch to our Bone Clinic for evaluation. The etiology of low imprecision was 1.6% at 7 ␮g/L and 1.5% at 78 ␮g/L. The bone mass in these patients was attributed to the follow- normal interval quoted by the manufacturer is 11–46 ing: idiopathic osteoporosis (n ϭ 6); familial osteoporosis ␮g/L for both men and premenopausal women. We (n ϭ 4); glucocorticoid use (n ϭ 3); loss-of-function measured P1NP by immunoassay on a Roche E170 ana- mutations in the LRP5 (low-density lipoprotein receptor– lyzer. Interbatch imprecision was 6.7% at 79 ␮g/L and related protein 5) gene (heterozygous, n ϭ 2; homozy- 2.9% at 795 ␮g/L. The normal interval quoted by the gous, n ϭ 1); postmenopausal osteoporosis (n ϭ 2); manufacturer is 20–80 ␮g/L for men and premenopausal juvenile Paget disease (n ϭ 2); postpregnancy osteoporo- women. sis (n ϭ 2); and anorexia, hypophosphatasia, and depot We measured bone density by dual-energy x-ray ab- medroxyprogesterone use (n ϭ 1 each). Three of the 25 sorptiometry (Lunar) and expressed the results as a patients were being treated with bisphosphonates at the standard deviation score relative to age- and sex-matched time the blood samples were drawn; in all other cases, the normal individuals (z score). samples were taken before any specific antiosteoporotic medication was prescribed. The bone density measures statistics reported here were all made before any treatment was We compared median values for the various biochemical prescribed. markers and the ratios between them by the Kruskal– (c) Thirty-eight normal individuals, selected without Wallis test (for 3-way comparisons) and the Mann–Whit- knowledge of their results for bone formation markers or ney U-test (for 2-way comparisons), as the values were bone density, to provide an age- and sex-matched com- not gaussian distributed. We calculated the specificity and parator group. These individuals were recruited as con- sensitivity of the various markers and their ratios for the trols for other studies in our unit, and their baseline data diagnosis of OI for various cutoff points and plotted them were used in this study. as ROC curves (12) calculated using Prism 4.03 for In 6 patients from groups a and b, because of uncer- Windows software (GraphPad Software). We calculated tainty about the diagnosis, we undertook in vitro studies the area under the curve (AUC) by a trapezoidal method of type 1 collagen on fibroblasts grown from skin. We also (with 95% CIs) and compared it against the area that Clinical Chemistry 53, No. 6, 2007 1111

would be expected by chance (0.5). We compared the OI patients (0.21, 0.31, and 0.67, respectively; Fig. 1). A AUCs for various models by use of a nonparametric similar pattern was seen when the ratios bALP:P1NP approach (13). and (osteocalcin ϩ bALP):P1NP were compared between the groups (Fig. 2 and see Fig. 1 in the Data Supplement Results that accompanies the online version of this article at bone formation markers http://www.clinchem.org/content/vol53/issue6). The patient groups were recruited over the 4-year period of 2003–2006. The 3 groups were similar in mean age, age roc curves range, and sex distribution. Summary data on the 3 bone The ROC curves for the individual bone formation mark- formation markers and the bone density measurements ers and the various ratios are summarized in Table 2. The are given in Table 1. In both the OI and other low bone ROC curves illustrating the ability of P1NP and the mass groups, there was considerable variability in the osteocalcin:P1NP, bALP:P1NP, and (osteocalcin ϩ bALP): results of individual markers compared with the rela- P1NP ratios to distinguish OI from other causes of low tively narrow range in controls. The median value for bone mass are shown in Fig. 3. Fig. 2 in the online Data P1NP was significantly lower in the OI group than in Supplement illustrates the ability of P1NP and the osteo- normal individuals and the group with other causes of calcin:P1NP, bALP:P1NP, and (osteocalcin ϩ bALP): low bone mass (P Ͻ0.01). The median value for osteocal- P1NP ratios to distinguish OI from normal individuals, as cin was significantly higher in the OI patients than in well as those with other causes of low bone mass. Al- controls (P Ͻ0.05). Median values for bALP were signifi- though the AUCs for P1NP and bALP alone were signif- cantly higher in both the OI and other low bone mass icantly different from the curve that would be produced groups than in controls (P Ͻ0.005; Table 1). by chance, the osteocalcin/P1NP, bALP:P1NP, and (os- We then compared the ratios of osteocalcin to P1NP teocalcin ϩ bALP):P1NP ratios all yielded CIs that span and bALP to P1NP in the 3 groups. There was a clear 0.8 to 1.0; values Ͼ0.8 are usually taken to indicate good difference among the groups (P Ͻ0.0001, Kruskal–Wallis diagnostic utility (14). The AUCs for the osteocalcin: test), with a much clearer distinction between the OI P1NP, bALP:P1NP, and (osteocalcin ϩ bALP):P1NP ratios patients and either of the other 2 groups. The ratio of were all significantly better than that for P1NP alone in osteocalcin to P1NP was significantly greater in patients distinguishing type I OI (P Ͻ0.05). with type I OI (median, 1.75; range, 0.80–3.86) than in The AUC values for distinguishing patients with OI controls (median, 0.59; range, 0.34–0.90) and patients from those with other causes of low bone mass were with other causes of low bone mass (median, 0.48; greatest for the osteocalcin:P1NP ratio. With a cutoff of range, 0.05–1.38). The median values were significantly 0.75 for osteocalcin:P1NP, the sensitivity and specificity greater for the type I OI patients than both the other were maximized at 100% (95% CI 84%–100%) and 97% groups (P Ͻ0.0001). However, the results from the 3 (86%–100%), respectively, for controls vs type I OI only; patients with type IV OI lay below the range of the type I for controls vs all OI patients, the sensitivity and specific-

Table 1. Bone densitometry and biochemical findings. OI Other causes of low bone mass Control n242538 Sex, male/female 11/13 7/18 13/25 Age, years Mean (SD) 37 (15) 37 (14) 38 (11) Median (range) 37 (19–64) 35 (17–76) 36 (21–61) Lumbar spine BMD, z score, mean (SD) Ϫ2.5 (1.3)a Ϫ3.1 (0.6)a 0.5 (1.3) P1NP, ␮g/L Mean (SD) 44.1 (64.5) 181.8 (419.7) 41.2 (13.3) Median (range) 23.7 (2.9–290.1)b,c 39.3 (4.5–1972) 38.9 (18.5–82.7) Osteocalcin, ␮g/L Mean (SD) 39.8 (31.9) 40.3 (45.4) 22.9 (7.2) Median (range) 35.2 (6.7–141.9)d 23.5 (6.2–178.2) 22.4 (9.2–39.5) bALP, ␮g/L Mean (SD) 23.0 (22.2) 29.8 (56.2) 9.2 (3.1) Median (range) 16.0 (6.0–102.8)e 13.5 (5.5–273.0)e 8.5 (4.4–20.9) a Statistically significant difference compared with controls: P Ͻ0.0001 (Student’s t-test). b Statistically significant difference compared with controls: P Ͻ0.01. c Statistically significant difference between OI patients and patients with low bone mass due to other causes: P Ͻ0.01 (Mann–Whitney U-test). d Statistically significant difference compared with controls: P Ͻ0.05. e Statistically significant difference compared with controls: P Ͻ0.005 (Mann–Whitney U-test). 1112 Cundy et al.: Formation Markers in OI

Five patients with type I OI were taking bisphosphonates when the biochemical tests were done. The mean (SD) osteocalcin:P1NP ratio in this group [1.94 (0.67)] was not significantly different (P ϭ 0.48, Student t-test) from that in the 16 type I OI patients not taking bisphosphonates [1.77 (0.73)]. Six individuals had skin collagen studies for diagnostic purposes. In 4, normal type 1 collagen was produced in normal quantities by cultured fibroblasts; these individuals were included in the “low bone mass due to other causes” group. In these patients, the osteocalcin:P1NP ratios were 0.40–0.84, with a mean value of 0.58 (results in controls were 0.34–0.90; mean, 0.57). The collagen studies indicated OI type IV in 1 case and type I in the other. The latter patient had no family history of OI, only slightly blue sclerae, and delayed puberty. He sustained his first fractures after considerable trauma at the age of 14. When assessed at the age of 19, his osteocalcin:P1NP ratio was 0.80 (results in the other type I OI patients were 1.00–3.86; Fig. 1. Ratio of plasma concentrations of osteocalcin to P1NP (both mean, 1.86). ␮g/L) in controls, patients with OI, and patients with other causes of low bone mass. Discussion The ratio was significantly higher in the OI patients than either of the other groups Although there was considerable overlap in the concen- (P Ͻ0.0001, Mann–Whitney U-test). The 3 patients with OI type IV are shown by the darker symbols. Note the logarithmic scale. trations of the 3 plasma bone formation markers in the 3 diagnostic groups, the osteocalcin:P1NP and bALP:P1NP ratios both provided excellent differentiation of type I OI ity were 91% (72%–99%) and 97% (86%–100%), respec- patients from normal individuals. These ratios did not tively. With a cutoff of 0.97 for osteocalcin:P1NP, the discriminate so well patients with OI from a heteroge- sensitivity and specificity were maximized at 95% (76%– neous group of patients with other causes of low bone 100%) and 88% (6%–97%), respectively, for patients with mass, but they still performed well, with areas under the other causes of low bone mass vs type I OI only; and vs all ROC curve ranging from 0.88 to 0.97. Incorporating both OI patients, 83% (63%–95%) and 88% (69%–97%), osteocalcin and bALP data in the ratio [(osteocalcin ϩ respectively. bALP):P1NP] did not add to the discriminant value, and of course added the imprecision of having 3 analytes rather than 2. In the patients with nondeforming type IV OI, we observed the opposite pattern to that seen in type I OI—their P1NP concentrations were relatively high in relation to osteocalcin. The likely explanation is that in type I OI the production of normal type 1 collagen is approximately halved (hence the low P1NP and P1CP) but no abnormal collagen is produced. In contrast, in type IV OI abnormal collagen is produced (often associated with obvious dentinogenesis imperfecta) but the total quantity (normal ϩ abnormal) is not necessarily decreased (3, 11). Our findings are consistent with those of both Lund et al. (6) and Brenner et al. (7), who found that plasma P1CP was proportionately lower in type I OI than in types III or IV. If our findings of differing ratios in type IV OI are confirmed in larger numbers of patients, then the diagnostic utility of the osteocalcin:P1NP ratio in differentiating type IV OI from other causes of low bone Fig. 2. Ratio of plasma concentrations of osteocalcin and bone alkaline mass is going to be limited, but the ratio could be of value phosphatase to P1NP (both ␮g/L) in controls, patients with OI, and in distinguishing type I from type IV OI. Nor is the patients with other causes of low bone mass. osteocalcin:P1NP ratio likely to be useful in the diagnosis The ratio was significantly higher in the OI patients than either of the other groups (P Ͻ0.0001, Mann–Whitney U-test). The 3 patients with OI type IV are shown by of other rare forms of OI not associated with abnormali- the darker symbols. Note the logarithmic scale. ties of type 1 collagen (15); however, type I OI is signifi- Clinical Chemistry 53, No. 6, 2007 1113

Table 2. ROC curve analyses for individual bone markers and their ratios in the diagnosis of OI.a Control vs Other causes of low bone mass vs

Biochemical test Type I OI All OI Type I OI All OI P1NP, ␮g/L 0.78 (0.63–0.93) 0.71 (0.55–0.87) 0.75 (0.60–0.89) 0.70 (0.55–0.85) P Ͻ0.0005 Ͻ0.01 Ͻ0.005 Ͻ0.02 Osteocalcin, ␮g/L 0.66 (0.48–0.84) 0.66 (0.49–0.83) 0.56 (0.38–0.52) 0.56 (0.40–0.73) P Ͻ0.05 Ͻ0.05 0.52 0.45 bALP, ␮g/L 0.84 (0.72–0.96) 0.83 (0.71–0.95) 0.76 (0.62–0.90) 0.72 (0.58–0.87) P Ͻ0.0001 Ͻ0.0001 Ͻ0.005 Ͻ0.01 bALP:P1NP 0.99 (0.99–1.00) 0.98 (0.95–1.00) 0.91 (0.82–1.00) 0.88 (0.78–0.98) P Ͻ0.0001 Ͻ0.0001 Ͻ0.0001 Ͻ0.0001 Osteocalcin:P1NP 0.99 (0.99–1.00) 0.92 (0.80–1.00) 0.97 (0.93–1.00) 0.90 (0.80–1.00) P Ͻ0.0001 Ͻ0.0001 Ͻ0.0001 Ͻ0.0001 (Osteocalcin ϩ bALP):P1NP 1.00 (1.00–1.00) 0.91 (0.80–1.00) 0.96 (0.93–1.00) 0.90 (0.80–1.00) P Ͻ0.0001 Ͻ0.0001 Ͻ0.0001 Ͻ0.0001 a Data are AUC (95% CI) and statistical significance for ROC curve analyses of patients with OI compared with controls and patients with other causes of low bone mass. cantly more prevalent than all the other types of OI clearly worthy of further investigation, because OI usu- combined (1, 2). We have not included in this report data ally presents in childhood and can be mistaken for non- on adult patients with deforming forms of OI (type III and accidental injury (17). the more severe forms of type IV), because these rarely Patients with osteoporosis of any cause are often pose a diagnostic dilemma. prescribed bisphosphonates or other antiresorptive agents We have not included data on children. It is quite that increase bone density predominantly by lowering probable that during growth the ratios between the vari- bone turnover. In patients with type I OI, we found no ous bone formation markers will differ at various ages difference in the mean osteocalcin:P1CP ratio between the and between children and adults (16). The limited infor- small number taking bisphosphonates and those not mation we have suggests that, compared with adults, taking bisphosphonates. If this observation is confirmed, osteocalcin:P1NP ratios are substantially lower in chil- then the clinical utility of the test would be enhanced, dren with type I OI [that is, the procollagen peptide is because it would not be necessary to stop antiresorptive increased relative to osteocalcin (data not shown)]. This therapy before testing. may explain why the osteocalcin/P1CP ratio did not Most of the earlier studies of procollagen peptides have clearly distinguish the 19-year-old with late-presenting focused on the peptide derived from the C-terminal end type I OI, who was not at skeletal maturity when assessed (P1CP) rather than P1NP (6–9), and there have been for this study. The possibility that the osteocalcin:P1NP relatively few data published concerning P1NP in OI. ratio could be of value in diagnosing OI in childhood is Lund et al. (6) reported that in adults with OI, plasma

Fig. 3. ROC curves for type I OI only (left) and for all OI patients (type I and type IV, right) compared with patients with other causes of low bone mass. AUCs with 95% CIs are given for P1NP alone; ratios are given for osteocalcin: P1NP, bALP:P1NP, and (osteocalcin ϩ bALP):P1NP. 1114 Cundy et al.: Formation Markers in OI

P1NP concentrations were low in children but normal in 4. Delmas PD. Biochemical markers for the assessment of bone adults. We chose to study P1NP because it has proven to turnover. In: Riggs BL, Melton LJ, eds. Osteoporosis: Etiology, be a robust and sensitive marker of bone formation (18). Diagnosis and Management, 2nd ed. Philadelphia, PA: Lippincott- Raven, 1995:319–33. Our results, that plasma concentrations of P1NP are low 5. Whyte MP. Osteogenesis imperfecta. In: Favus MJ, ed. Primer on in adults with type I OI, contrast with the findings of the Metabolic Bone Diseases and Disorders of Mineral Metabo- Lund et al. (6) but are more in keeping with what we lism, 5th ed. Washington, DC: American Society for Bone and understand of the abnormality in collagen metabolism in Mineral Research, 2003:470–3. this disorder. It is not clear why osteocalcin and bALP 6. Lund AM, Hansen M, Kollerup G, Juul A, Teisner B, Skovby F. concentrations are relatively increased in OI, but osteo- Collagen-derived markers of bone metabolism in osteogenesis blast numbers are increased in OI bone (19), and given the imperfecta. Acta Paediatr 1998;87:1131–7. 7. Brenner RE, Schiller B, Vetter U, Ittner J, Teller WM. Serum putative role of osteocalcin and bALP in mineralization it concentrations of procollagen I C-terminal propeptide, osteocal- is interesting to note that in patients with OI the bone cin, and insulin-like growth factor-I in patients with non-lethal tends to be hypermineralized (20), and in diaphyseal osteogenesis imperfecta. Acta Paediatr 1993;82:764–7. bone, volumetric bone density may be increased (21). 8. Cepollaro C, Gonnelli S, Pondrelli C, Montagnani A, Martini S, Osteocalcin is incorporated into bone matrix and released Bruni D, et al. Osteogenesis imperfecta: bone turnover, bone during resorption, so the increased plasma concentrations density, and ultrasound parameters. Calcif Tissue Int 1999;65: 129–32. may reflect increased bone turnover in OI. 9. Minisola S, Piccioni AL, Rosso R, Romagnoli E, Pacitti, MT, Bone markers are generally thought to have little use in Scarnecchia L, et al. Reduced serum levels of carboxy-terminal the specific diagnosis of particular metabolic bone dis- propeptide of human type I procollagen in a family with type I-A eases, because resorption and formation markers tend to osteogenesis imperfecta. Metabolism 1994;43:1261–5. change in parallel. There are a few rare circumstances 10. Braga V, Gatti D, Rossini, Colapietro F, Battaglia E, Viapiana O, et (such as Ehlers–Danlos syndrome type VIA) in which a al. Bone turnover markers in patients with osteogenesis imper- distinctive pattern of bone markers is of diagnostic value fecta. Bone 2004;34:1013–6. 11. Wenstrup RJ, Willing MC, Starman BJ, Byers PH. Distinct biochem- (22). The accurate diagnosis of OI requires either sequenc- ical phenotypes predict clinical severity in nonlethal variants of ing of COL1A1 and COL1A2 or electrophoretic studies on osteogenesis imperfecta. Am J Hum Genet 1990;46:975–82. collagen secreted by cultured skin fibroblasts, but these 12. Kramer MS. Clinical Epidemiology and Biostatistics: A Primer for tests are not always readily available. Our data suggest Clinical Investigators and Decision-Makers. Berlin, Germany: that the ratio of the plasma concentrations of osteocalcin Springer-Verlag, 1988:286pp. or bALP to P1NP could be helpful in the diagnosis of OI 13. DeLong ER, DeLong DM, Clarke-Paerson DL. Comparing the areas when definitive tests on type 1 collagen are not possible. under two or more correlated receiver operator curves: a nonparametric approach. Biometrics 1998;44:837–45. As with all preliminary case-control studies, the findings 14. Weinstein MC, Fineberg HV. Clinical Decision Analysis. Philadel- in this report need to be confirmed in other studies in phia, PA: WB Saunders, 1980:351pp. independent patient populations. 15. Rauch F, Glorieux FH. Osteogenesis imperfecta. Lancet 2004; 363:1377–85. 16. Szulc P, Seeman E, Delmas PD. Biochemical measurements of bone turnover in children and adolescents. Osteoporos Int 2000; Grant/funding support: Health Research Council of New 11:281–94. Zealand. 17. Paterson CR, McAllion SJ. Osteogenesis imperfecta in the differ- Financial disclosures: None declared. ential diagnosis of child abuse. BMJ 1989;299:1451–4. Acknowledgements: We are grateful to the Collagen Diag- 18. Reid IR, Davidson JS, Wattie D, Wu F, Lucas J, Gamble GD, et al. nostic Laboratory, University of Washington, Seattle, for the Comparative responses of bone turnover markers to bisphospho- nate therapy in Paget’s disease of bone. Bone 2004;35:224–30. skin collagen studies. 19. Rauch F, Travers R, Parfitt AM, Glorieux FH. Static and dynamic bone histomorphometry in children with osteogenesis imperfecta. References Bone 2000;26:581–9. 20. Boyde A, Travers R, Glorieux FH, Jones SJ. The mineralization 1. Shapiro JR. Osteogenesis imperfecta and other defects of bone density of iliac crest bone from children with osteogenesis imper- development as occasional causes of adult osteoporosis. In: fecta. Calcif Tissue Int 1999;64:185–90. Marcus R, Feldman D, Kelsey J, eds. Osteoporosis, 2nd ed. San 21. Rauch F, Land C, Cornibert S, Schoenau E, Glorieux FH. High and Diego, CA: Academic Press, 2001:271–301. low density in the same bone: a study on children and adolescents 2. Sillence DO, Senn A, Danks DM. Genetic heterogeneity in osteo- with mild osteogenesis imperfecta. Bone 2005;37:634–41. genesis imperfecta. J Med Genet 1979;16:101–16. 22. Steinmann B, Eyre DR, Shao P. Urinary pyridinoline cross-links in 3. Byers PH, Wallis GA, Willing MC. Osteogenesis imperfecta: trans- Ehlers-Danlos syndrome type VI. Am J Hum Genet 1995;57: lation of mutation to phenotype. J Med Genet 1991;28:433–42. 1505–8. Clinical Chemistry 53:6 1115–1121 (2007) Automation and Analytical Techniques

Development and Multicenter Evaluation of the N Latex CDT Direct Immunonephelometric Assay for Serum Carbohydrate-Deficient Transferrin

Joris R. Delanghe,1* Anders Helander,2 Jos P.M. Wielders,3 J. Maurits Pekelharing,4 Heinz J. Roth,5 Franc¸ois Schellenberg,6 Catherine Born,7 Eray Yagmur,8 Wolfgang Gentzer,9 and Harald Althaus9

Background: Carbohydrate-deficient transferrin (CDT) 1.01%–2.85%). No marked sex or age differences were is a promising biomarker of alcohol abuse. We describe noted. The 97.5th percentile was at 2.35%. Transferrin the development and multicenter evaluation of N Latex genetic variants did not interfere with measurements. CDT (Dade Behring), an automated, particle-enhanced, High transferrin concentrations did not falsely increase homogeneous immunonephelometric assay for directly %CDT values, but increased %CDT values were noted determining CDT. for some samples with transferrin concentrations <1.1 Methods: N Latex CDT uses a monoclonal antibody that g/L. N Latex CDT results correlated with those of a recognizes the structure of transferrin glycoforms lack- commercial CDT immunoassay involving column sepa- and an HPLC candidate reference (0.862 ؍ ing 1 or 2 complete N-glycans [i.e., disialo-, monosialo-, ration (r2 .(0.978 ؍ and asialotransferrins (CDT glycoforms)] in combina- method (r2 tion with a simultaneous assay for total transferrin. The Conclusion: N Latex CDT is the first direct immunoas- Dade Behring BN II™ and BN ProSpec® systems auto- say for quantifying %CDT in serum. The specificity of matically calculate the CDT value as a percentage of N Latex CDT for identifying alcohol abuse may be total transferrin (%CDT). No preanalytical sample treat- higher than for immunoassays that use column separa- ment is used. tion, because transferrin genetic variants do not inter- Results: Total imprecision values for serum pools con- fere with measurements. taining 1.8%–8.7% CDT were 3.4%–10.4% (mean, 6.8%). © 2007 American Association for Clinical Chemistry The mean (SD) %CDT for 561 serum samples from healthy control individuals was 1.76% (0.27%; range, Alcohol abuse causes physical injuries, neuropsychiatric defects, social problems, and substantial costs to society (1). Identifying individuals with alcohol-related problems 1 Department of Clinical Chemistry, Ghent University Hospital, Ghent, is important but is not easy. Conventional laboratory tests Belgium. such as the ␥-glutamyltransferase assay focus mainly on 2 Alcohol Laboratory, Karolinska Institute and Karolinska University Hos- identifying individuals engaged in long-term, chronic pital, Stockholm, Sweden. 3 Department of Clinical Chemistry, Meander Medical Center, Amersfoort, alcohol abuse. These tests indicate organ damage and The Netherlands. show low diagnostic sensitivity and specificity before the 4 Reinier de Graaf Groep, Diagnostic Center SSDZ, Delft, The Netherlands. clinical manifestations of alcohol abuse develop (2, 3). 5 Limbach Laboratories, Heidelberg, Germany. Carbohydrate-deficient transferrin (CDT)10 is consid- 6 Laboratory of Clinical Chemistry, Hopital Trousseau, Centre Hospitalier Re´gional Universitaire, Tours, Tours, France. ered the most accurate biomarker for identifying sus- 7 Institut Regional pour la Sante, La Riche, France. tained heavy alcohol consumption and for monitoring 8 Central Laboratory, University Hospital, Rheinisch Westfa¨lische Tech- abstinence (3, 4). Transferrin, which occurs at concentra- nische Hochschule, Aachen, Germany. 9 Research Laboratories, Dade Behring Marburg GmbH, Marburg, Ger- tions of 2.0–3.5 g/L in serum, exhibits a degree of many. * Address correspondence to this author at: Department of Clinical Chem- istry, De Pintelaan 185, B-9000 Ghent, Belgium. Fax 32-9-240-4985; e-mail [email protected]. 10 Nonstandard abbreviations: CDT, carbohydrate-deficient transferrin; Received December 11, 2006; accepted March 8, 2007. CDG, congenital disorders of glycosylation; mAb, monoclonal antibody; Previously published online at DOI: 10.1373/clinchem.2006.084459 %CDT, CDT as a percentage of total transferrin.

1115 1116 Delanghe et al.: %CDT by N Latex CDT

microheterogeneity that depends on iron saturation solution and purified the antibodies with Protein A (ϳ30%), amino acid sequence, and/or carbohydrate con- Sepharose Fast Flow (GE Healthcare/Amersham Bio- tent (5–7). Amino acid sequence variation is observed in sciences). The CDT mAb (98/84-011) with the highest individuals with genetic variants B, C, and D (8), whereas specificity for nonglycosylated transferrin but no affinity transferrin glycoforms with variable carbohydrate content for typical human transferrin was selected for assay and/or branching of the maximum 2 N-linked oligosac- development (see the Data Supplement that accompanies charide chains (N-glycans) are always present (4, 7, 9). the online version of this article at http://www. Typically, the major serum transferrin glycoform, tetra- clinchem.org/content/vol53/issue6). sialotransferrin, contains 2 disialylated biantennary gly- We diluted the selected CDT mAb to a concentration of cans. Other, less abundant glycoforms are pentasialo-, 80 mg/L in blocking buffer (10 g/L bovine serum albu- trisialo-, and disialotransferrins (10). Disialo- and asialo- min and 0.5 mL/L Tween 20 in Tris-buffered saline (0.02 transferrin fractions increase after sustained heavy drink- mol/L Tris, 0.5 mol/L NaCl, pH 7.5) and then added ing (10–12). These glycoforms, together with monosialo- alkaline phosphatase–linked secondary antibodies (Bio- transferrin (10), have collectively been referred to as CDT Rad Laboratories) in blocking buffer. p-Nitrotetrazolium (4). A regular intake of ϳ50–80 g ethanol/day for a blue and 5-bromo-4-chloro-3-indolyl phosphate (Sigma- minimum of ϳ1–2 weeks is required to increase the serum Aldrich) were added as substrates (31). We used sera CDT concentration in Ն80% of individuals (4, 13). The from 1 control individual and 1 alcoholic proband to half-life of the CDT marker is ϳ1–1.5 weeks, and a return compare the specificity of the CDT mAb to that of to the usual glycoform pattern requires Ͼ2 weeks of polyclonal antibodies directed against several transferrin abstinence (4, 14, 15). epitopes (Dade Behring). Early methods for assaying CDT used isoelectric focus- We further evaluated the specificity of the CDT mAb ing followed by immunofixation (16–18). Alternative by investigating its reaction with CDT, other transferrin procedures used chromatofocusing (19), HPLC (10, 14, glycoforms, and enzymatically modified transferrin. 20), fast protein liquid chromatography (21), and capil- Transferrin lacking the terminal sialic acid residues was lary electrophoresis (22–24). Immunoassays include an obtained by treating serum with neuraminidase (2.5 initial chromatographic separation of CDT glycoforms mU/mg transferrin; Dade Behring) in PBS (0.048 mol/L from non-CDT glycoforms on disposable minicolumns Na2HPO4, 0.02 mol/L KH2PO4, 0.145 mol/L NaCl, 0.015 (25–27). Drawbacks with this last approach are the labor mol/L NaN3, pH 7.2) for 18 h at 37 °C. We obtained involved and the fact that transferrin genetic variants may transferrin lacking entire N-glycan moieties by treating cause falsely high or falsely low results (28). We present transferrin with peptide-N-glycosidase F (500 mU/mg data on the development and multicenter evaluation of transferrin; Roche Diagnostics) for4hat37°CinPBS the 1st direct immunoassay for CDT (N Latex CDT; Dade (0.048 mol/L Na2HPO4, 0.02 mol/L KH2PO4, 0.145 Behring) and on possible interference by transferrin ge- mol/L NaCl, 0.015 mol/L NaN3, pH 7.2) containing 10 netic variants and congenital disorders of glycosylation mmol/L EDTA and 1 g/L sodium dodecyl sulfate. After (CDG) (29). this incubation, proteins were separated by sodium dodecyl sulfate–polyacrylamide gel electrophoresis. Immuno- Materials and Methods detection was performed with the CDT mAb and a development of a monoclonal antibody polyclonal antibody against human transferrin. We raised monoclonal antibodies (mAbs) against CDT by means of a recombinant nonglycosylated transferrin, cdt mAb–based immunoassay which was a gift from Anne B. Mason, Ph.D. (Department N Latex CDT is based on an mAb that specifically of Biochemistry, College of Medicine, University of Ver- recognizes transferrin glycoforms that lack one or both of mont, Burlington, VT) (30), in which the 2 asparagine the complete N-glycans [i.e., disialo-, monosialo-, and residues at the carbohydrate-linkage sites were mutated asialotransferrins (the CDT glycoforms)] in combination to aspartic acid (Asn413Asp and Asn611Asp). We immu- with a simultaneous assay for total transferrin (N Anti- nized BALB/c mice with recombinant transferrin in com- serum to Human Transferrin; Dade Behring). Polystyrene plete Freund adjuvant. A booster with an emulsion pre- particles coated with the CDT mAb are agglutinated by pared in incomplete adjuvant was given after 4 weeks, CDT-coated polystyrene particles. CDT inhibits this reac- and another booster without adjuvant was administered tion in a dose-dependent manner, allowing nephelometric after 8 weeks. During the final 3 days before the fusion, CDT quantification over 18 min on the Dade Behring BN we gave the mice daily intravenous boosters. II™ and BN ProSpec® systems. No sample pretreatment is After the mice were killed, we removed the spleens required. Because the degree of iron saturation of trans- and cloned the B-cells with myeloma cells. Single hybrid ferrin influences the binding affinity of the antibody, the cells that produced antibodies specific for CDT (i.e., transferrin-bound iron is stripped with a chelating agent binding to nonglycosylated transferrin but not to typical in the first incubation step. The simultaneous determina- transferrin) were cloned, and appropriate clones were tion of total transferrin allows an automatic calculation of expanded. After removing the cells, we concentrated the the amount of CDT as a percentage of total transferrin Clinical Chemistry 53, No. 6, 2007 1117

(%CDT). The measurement range is ϳ20–640 mg/L or Method Comparison 0.77%–25% CDT. We compared %CDT results obtained with the N Latex CDT assay with those obtained by turbidimetric chro- multicenter evaluation matographic separation followed by immunoassay (the We evaluated the N Latex CDT assay at 8 sites (A–H) with Axis-Shield %CDT or the Bio-Rad %CDT TIA assay) (27). 2 independent reagent lots (01 and 03). The reagent lots We also compared the N Latex CDT assay with an HPLC were distributed so that each lot was used at least once on candidate reference method (10) with 100 serum samples each analyzer. Imprecision was determined according to with percent disialotransferrin values within the reference Ͻ the Clinical and Laboratory Standards Institute EP5-2A interval of the HPLC method (97.5th percentile, 1.7%) Ͼ guideline. The 2 reagent set controls (N CDT Control SL/1 and 100 samples with increased values ( 2% disialotrans- and SL/2; Dade Behring) and 4 different human serum ferrin). Samples were selected to cover a wide range of pools were run in duplicate in 2 runs/day, over 20 days. percent disialotransferrin values (0.9%–22.2%). Dade Behring provided 2 serum pools (R1 and R2). R1 statistics contained samples with an increased %CDT, and R2 Results are expressed as the mean (SD). Differences contained samples with low and increased %CDT values. between samples that were saturated with iron before Two other serum pools (“low” and “high” %CDT pools) analysis and unaltered samples were evaluated by means were produced individually by each laboratory. We used of a paired t-test. Passing–Bablok regression and the the N Latex CDT assay to analyze a proficiency panel of Wilcoxon test were used to compare methods. The Stu- 29 serum samples that were provided in frozen aliquots at dent t-test and ANOVA were used to evaluate sex and age 6 of the sites. differences in the control population. We defined refer- We obtained informed consent and approval of the ence values as the 2.5%–97.5% interval in the distribution local ethics committee whenever required. To evaluate of values in the reference population. any relationship between total transferrin concentrations and %CDT values produced by N Latex CDT, we in- Results cluded serum samples from 113 patients with a wide evaluation of the cdt mAb range of transferrin concentrations (0.43–4.22 g/L; refer- Sodium dodecyl sulfate–polyacrylamide gel electrophore- ence interval, 2.0–3.6 g/L) but who had typical %CDT sis and Western blotting analysis with the CDT mAb values with the N Latex CDT method and no evidence of showed only transferrin molecules lacking one or both alcohol abuse. N-glycans in serum samples obtained from alcoholic individuals; no reaction with transferrin was obtained reference interval with control samples (Fig. 1). The CDT mAb also detected We collected serum samples from 561 apparently healthy transferrin in control samples after N-glycosidase F treat- adults (255 men, ages 20–70 years; 306 women, ages ment (Fig. 1), which yields transferrin molecules lacking 19–79 years). We included samples only from individuals entire N-glycans. Control serum samples incubated with who had no clinical indications or biochemical indications neuraminidase, which depletes only terminal sialic acid (as measured by ␥-glutamyltransferase activity and eryth- residues, were not detected by the CDT mAb (Fig. 1). rocyte mean corpuscular volume) of chronic alcohol Epitope mapping of the CDT mAb with overlapping abuse and consumed no more than 2 drinks/day (Ͻ25 g peptides corresponding to the human transferrin sequence identified 4 major binding sites (data not shown): ethanol/day). 1 site in the N-terminal domain and 3 sites in the For comparison, we obtained an additional 141 sam- C-terminal domain. Because peptide sequences at or near ples from children and adolescents (ages 11–18 years). the 2 N-glycan–binding sites (Asn413 and Asn611) were None of these patients exhibited any signs of liver or not detected, we concluded that the antibody is directed metabolic diseases or of alcohol consumption. To assess against a discontinuous structural epitope. This result alcohol consumption, we used both the section in the suggests differences in 3-dimensional structure between Kiddie–Sads–Present and Lifetime Version that is related transferrin molecules containing 2 N-glycans and those to alcohol abuse and the Alcohol Use Disorders Identifi- lacking 1 or both N-glycans (i.e., the CDT glycoforms). cation Test. Apparently, this structural change and the formation of the CDT-specific structural epitope(s) occur when 1 N- interference testing glycan is missing, and no major additional changes occur We evaluated potential interfering factors, such as trans- when the second N-glycan is also missing. ferrin saturation, iron deficiency, lipemia, transferrin genetic variants, and CDG. We obtained serum samples Method Imprecision from healthy white individuals who carried the BC (n ϭ The imprecision (CV) of the N Latex CDT assay was 3) and CD (n ϭ 4) transferrin variants and from 1 CDG determined for 4 serum pools at 6 laboratory sites (Table type Ic patient. 1). The R1 pool (mean %CDT, 3.92%; range, 3.8%–4.0%) 1118 Delanghe et al.: %CDT by N Latex CDT

Fig. 1. Comparison of the specificities of the polyclonal transferrin and monoclonal CDT antibodies. Immunoblots of transferrin glycoforms separated by sodium dodecyl sulfate–polyacrylamide gel electrophoresis with polyclonal antibodies against transferrin (anti-TF Ab; A) and the monoclonal CDT antibody (CDT mAb; B). The serum transferrin samples were obtained from a control proband (lane 1), an alcoholic individual (lane 2), a control proband after treatment with N-glycosidase F (lane 3), and a control proband after treatment with neuraminidase (lane 4). and the R2 pool (mean, 3.08%; range, 3.0%–3.2%) were mean %CDT value was 1.76% (0.26%), and the range was provided by Dade Behring, whereas the low pools (mean 1.01%–2.85%. The %CDT results for men [1.78% (0.27%)] %CDT, 1.95%; range, 1.8%–2.1%) and high pools (mean, and women [1.77 (0.25%)] were not significantly different 5.33%; range, 3.0%–8.7%) were produced at each site. The (P ϭ 0.538), and no significant age-related differences intraassay CVs were 1.9%–7.0% (mean, 4.3%), and total were found (data not shown). We proposed an upper CVs were 3.4%–10.4% (mean, 6.8%; Table 1). BN II sys- reference limit of 2.35% (97.5th percentile) for %CDT tems showed higher analytical imprecision (mean total values obtained with the N Latex CDT assay. The 141 CV, 8.2%) than BN ProSpec systems (mean total CV, serum samples collected from children and adolescents 5.3%). Imprecision results for the reagent set controls showed similar results, with a median %CDT of 1.91% were similar to those for the serum pools, with total CVs and 2.5th and 97.5th percentiles of 1.45% and 2.40%, of 3.8%–9.7% for Dade Behring N CDT Control SL/1 respectively. On the basis of these results, we proposed (mean concentration, 55.7 mg/L) and 3.7%–5.4% for N the same upper reference limit (2.35%, 97.5th percentile) CDT Control SL/2 (mean concentration, 163 mg/L). for %CDT obtained with the N Latex CDT assay (Table 2). reference interval for %cdt values interference testing We studied the distribution of N Latex CDT values with For the 113 serum samples with a wide range of trans- 561 serum samples from healthy nonalcoholic individu- ferrin values, we found no marked effect of transferrin als. Transferrin concentrations were 1.7–4.4 g/L (2.5th concentration on N Latex CDT results within the reference percentile, 2.0 g/L; 97.5th percentile, 3.8 g/L). The overall interval (2.0–3.6 g/L). At abnormally low concentrations,

Table 1. Multicenter analytical performance of the N Latex CDT assay. Imprecision (CV), % Imprecision (CV), %

Sample Within-run TotalMean %CDT Within-run Total Mean %CDT Site A: BN IIa Site B: BN ProSpeca R1b 5.0 8.3 4.0 2.6 3.8 4.0 R2b 5.8 8.6 3.1 3.8 4.9 3.1 Low poolc 6.5 8.4 1.8 3.7 5.3 1.9 High poolc 5.8 8.1 5.8 2.7 3.4 8.7 Site C: BN ProSpec Site D: BN II R1 1.9 4.9 3.9 5.7 8.7 3.8 R2 3.2 6.4 3.0 7.0 10.4 3.1 Low pool 2.9 6.0 2.1 6.2 9.1 2.0 High pool 2.5 5.5 4.8 5.2 7.3 3.0 Site F: BN ProSpec Site H: BN II R1 2.9 4.7 3.8 4.3 6.8 4.0 R2 5.2 7.3 3.0 5.2 8.7 3.2 Low pool 3.3 6.4 1.9 4.8 8.6 2.0 High pool 2.8 5.3 4.7 4.3 5.3 5.0 a The study sites used either the BN II or the BN ProSpec system (Dade Behring). b R1 and R2 were serum sample pools provided by Dade Behring. c The low and high pools were individual serum pools produced at each study site. Clinical Chemistry 53, No. 6, 2007 1119

Table 2. Distribution of %CDT values produced by the N of the original value. Highly lipemic samples can cause Latex CDT assay for serum samples collected from healthy problems, and we recommend avoiding such samples. individuals at 2 study sites.a Method Comparison %CDT %CDT values obtained with N Latex CDT were correlated 2.5th 97.5th 99th with those obtained with the Bio-Rad %CDT TIA (n ϭ Group n percentile Median percentile percentile 132; y ϭ 0.838x ϩ 0.354; r2 ϭ 0.862; S ϭ 0.58). Passing– Total 561 1.29 1.76 2.35 2.47 yx Bablok regression analysis revealed the N Latex CDT Men 255 1.27 1.77 2.36 2.47 results to be generally lower in the low %CDT range than Women 306 1.37 1.76 2.35 2.44 with the Bio-Rad %CDT TIA (see the Figure in the online a The samples were collected in The Netherlands and France. Data Supplement). %CDT values obtained with the N Latex CDT assay however, we noted increased %CDT values for some were also correlated with percent disialotransferrin val- samples. A %CDT value Ͼ2.35% was observed in 8 of 10 ues obtained by HPLC in an analysis of 100 serum samples with transferrin concentrations of Ͻ1.1 g/L. Of samples with typical values and 100 samples with in- the 113 samples, 25 serum samples with a median trans- creased percent disialotransferrin values (range, 0.9%– ferrin concentration of 1.0 g/L (23 of the samples Ͻ1.5 22.2%; Fig. 2; n ϭ 200; y ϭ 0.700x ϩ 0.970; r2 ϭ 0.978; ϭ g/L) had CDT values below the measurement range of Syx 0.49). We used ROC curve analysis and the 97.5th the method (Ͻ20 mg/L). %CDT values could not be percentile for percent disialotransferrin as determined calculated for these samples. with the HPLC method as a reference, along with these Because the N Latex CDT assay requires iron depletion 200 samples to calculate the agreement of N Latex CDT of serum transferrin before analysis, we investigated the results with reference method results. With the upper efficiency of the iron-chelating capacity. In patients with reference limit of 2.35% for %CDT obtained with N Latex pronounced iron overload (transferrin saturation Ͼ70%) CDT as a cutoff point, 97% of the results that showed but without signs of alcohol abuse, all %CDT values were increased %CDT in the HPLC analysis were increased in within the reference range (data not shown). N Latex CDT, and 94% of the results that were below the An abstinent student had a markedly increased %CDT cutoff point according to the HPLC method were also value of 8.2% according to the Axis-Shield photometric below the cutoff point in the N Latex CDT assay (Table 3). method. Measurement with the N Latex CDT method on the BN ProSpec system yielded a typical %CDT value of Discussion 2.1%. An HPLC analysis of this sample revealed a CD The important component of N Latex CDT is the mAb, phenotype, which was confirmed by genotyping. For a which specifically recognizes the transferrin glycoforms sample from an individual presenting with a B2C pheno- that lack one or both entire N-glycans (corresponding to type, a typical %CDT value of 2.2% with the Axis-Shield asialo-, monosialo- and disialotransferrins) (4). Disialo- method was obtained, whereas the N Latex CDT assay produced an increased %CDT value of 2.9%. The %CDT values obtained with the Axis-Shield CDT assay for an individual who had been abstinent for Ͼ1 year (2.8%–3.0%) did not show a return from increased values to typical values; however, N Latex CDT revealed a typical value of 1.7%. This patient had a highly increased trisialotransferrin fraction, which led to falsely increased results in the Axis-Shield assay. The trisialo glycoform normally accounts for Ͻ5% of the total transferrin. In our experience, an increased trisialo fraction occurs more often in Europeans than CD genetic variants (28). A 3-year-old boy with a type Ic CDG syndrome had an abnormally high %CDT value (17.1%) with the N Latex CDT method and a 27.4% value with the Axis-Shield assay. Two of 8 lipemic samples (serum triglycerides Ͼ3.5 mmol/L) showed 15% and 35% relative increases in CDT Fig. 2. Comparison of %CDT values obtained with N Latex CDT and in the low-normal CDT range after the samples were percent disialotransferrin values obtained by HPLC analysis. cleared by high-speed centrifugation. Two other lipemic Passing–Bablok regression line (r ϭ 0.989, P Ͻ0.001) with 95% CI for 200 serum samples with respect to nonincreased and increased percent disialotrans- samples showed relative decreases of 15% and 20% after ferrin values (range, 0.9%–22.2%) obtained with the candidate HPLC reference centrifugation, and 4 other samples remained within 3% method (10). Dotted line, y ϭ x. 1120 Delanghe et al.: %CDT by N Latex CDT

Table 3. Sensitivity and specificity for the N Latex CDT The %CDT results obtained with the N Latex CDT assay calculated at different %CDT thresholds. assay correlated well with those of a column-based %CDT immunoassay (27) and with the percent disialotransferrin %CDT threshold Specificity, %a Sensitivity, %a values obtained with an HPLC candidate reference 2.2 90 99 method (10); however, because these methods measure 2.3 93 96 2.4 97 94 different transferrin glycoforms as CDT, the values ob- 2.5 97 93 tained with the different methods are not interchangeable. 2.6 97 86 This fact highlights the need for standardization of CDT measurements. a The percent disialotransferrin results obtained with the HPLC method (10) were used as a reference (threshold, 1.7%).

Grant/funding support: The N Latex CDT assay and control and asialotransferrins are associated with sustained heavy materials used for the Multicenter evaluation, as described in drinking, whereas monosialotransferrin is correlated with Materials and Methods, were provided by Dade Behring. the amount of trisialotransferrin (28). Monosialotrans- Financial disclosures: The authors had complete indepen- ferrin probably represents no obstacle to %CDT testing dence in the interpretation of data and writing of the report. with N Latex CDT because it is usually present in very low concentrations (10). Trisialotransferrin has caused References some confusion in CDT testing (7, 28, 32), because this 1. Wallace P, Cutler S, Haines A. Randomised controlled trial of glycoform is always present at higher concentrations and general practitioner intervention in patients with excessive alcohol was (26, 33)—and still is (34)—included in the CDT consumption. BMJ 1988;297:663–8. 2. Conigrave KM, Saunders JB, Whitfield JB. Diagnostic tests for fraction of some methods. Immunoblotting analysis, how- alcohol consumption. Alcohol Alcohol 1995;30:13–26. ever, has demonstrated that the CDT mAb does not detect 3. Helander A. Biological markers in alcoholism [Review]. J Neural trisialotransferrin in the N Latex CDT assay. Transm Suppl 2003;15–32. Another advantage of N Latex CDT over the indirect 4. Stibler H. Carbohydrate-deficient transferrin in serum: a new column-based immunoassays is that the CDT mAb is not marker of potentially harmful alcohol consumption reviewed [Re- influenced by transferrin genetic variants. Genetic vari- view]. Clin Chem 1991;37:2029–37. ants, which are rare in white individuals but more com- 5. van Eijk HG, van Noort WL, Kroos MJ, van der Heul C. Analysis of mon in other populations (8, 35), may cause falsely low the iron-binding sites of transferrin by isoelectric focussing. J Clin Chem Clin Biochem 1978;16:557–60. and high CDT values with the column-based immunoas- 6. de Jong G, van Dijk JP, van Eijk HG. The biology of transferrin says (28). For example, trisialotransferrin D in samples [Review]. Clin Chim Acta 1990;190:1–46. from individuals with C and D genetic variants will 7. Arndt T. Carbohydrate-deficient transferrin as a marker of chronic coelute with disialotransferrin C and thereby cause over- alcohol abuse: a critical review of preanalysis, analysis, and estimation of CDT. Use of the N Latex CDT assay may interpretation [Review]. Clin Chem 2001;47:13–27. therefore decrease the need for confirmatory CDT testing 8. Kamboh MI, Ferrell RE. Human transferrin polymorphism [Review]. by HPLC or capillary electrophoresis (36). Additional Hum Hered 1987;37:65–81. studies are needed to confirm that other transferrin ge- 9. Fu D, van Halbeek H. N-glycosylation site mapping of human serotransferrin by serial lectin affinity chromatography, fast atom netic variants and samples with divergent N-glycan struc- bombardment-mass spectrometry, and 1H nuclear magnetic res- tures, such as those occurring in the CDG subtypes, do onance spectroscopy. Anal Biochem 1992;206:53–63. not interfere with the N Latex CDT assay (22, 37, 38). 10. Helander A, Husa A, Jeppsson JO. Improved HPLC method for Besides variations in amino acid sequence and carbo- carbohydrate-deficient transferrin in serum. Clin Chem 2003;49: hydrate content, the degree of transferrin microheteroge- 1881–90. neity also depends on the number of bound iron mole- 11. Landberg E, Påhlsson P, Lundblad A, Arnetorp A, Jeppsson J-O. cules (7). Under physiological conditions, serum Carbohydrate composition of serum transferrin isoforms from transferrin is ϳ30% saturated with iron. Four transferrin patients with high alcohol consumption. Biochem Biophys Res Commun 1995;210:267–74. glycoforms can be distinguished with respect to iron 12. Flahaut C, Michalski JC, Danel T, Humbert MH, Klein A. The content: apotransferrin, N-terminal and C-terminal mono- effects of ethanol on the glycosylation of human transferrin. ferric transferrins, and diferric transferrin. To exclude Glycobiology 2003;13:191–8. analytical interference due to variation in iron saturation, 13. Schellenberg F, Schwan R, Mennetrey L, Loiseaux MN, Pages JC, many CDT methods completely saturate the transferrin in Reynaud M. Dose-effect relation between daily ethanol intake in the sample with iron before analysis. The degree of iron the range 0–70 grams and %CDT value: validation of a cut-off saturation also influences the binding affinity of the CDT value. Alcohol Alcohol 2005;40:531–4. 14. Jeppsson JO, Kristensson H, Fimiani C. Carbohydrate-deficient mAb in the N Latex CDT assay, but this assay uses a transferrin quantified by HPLC to determine heavy consumption of chelating agent to completely deplete the iron from trans- alcohol. Clin Chem 1993;39:2115–20. ferrin before analysis. The reproducibility results indicate 15. Helander A, Carlsson S. Carbohydrate-deficient transferrin and that iron depletion is complete and stable during the gamma-glutamyl transferase levels during disulfiram therapy. Al- immunonephelometric analysis. cohol Clin Exp Res 1996;20:1202–5. Clinical Chemistry 53, No. 6, 2007 1121

16. Schellenberg F, Weill J. Serum desialotransferrin in the detection quantification in the identification of alcohol abuse. Clin Chem of alcohol abuse. Definition of a Tf index. Drug Alcohol Depend 2001;47:1225–33. 1987;19:181–91. 29. Jaeken J. Komrower Lecture: congenital disorders of glycosylation 17. Bean P, Peter JB. Allelic D variants of transferrin in evaluation of (CDG): it’s all in it [Review]! J Inherit Metab Dis 2003;26:99–118. alcohol abuse: differential diagnosis by isoelectric focusing-immu- 30. Mason AB, Miller MK, Funk WD, Banfield DK, Savage KJ, Oliver noblotting-laser densitometry. Clin Chem 1994;40:2078–83. RW, et al. Expression of glycosylated and nonglycosylated human 18. Stibler H, Blennow G, Kristiansson B, Lindehammer H, Hagberg B. transferrin in mammalian cells: characterization of the recombi- Carbohydrate-deficient glycoprotein syndrome: clinical expression nant proteins with comparison to three commercially available in adults with a new metabolic disease. J Neurol Neurosurg transferrins. Biochemistry 1993;32:5472–9. Psychiatry 1994;57:552–6. 19. Storey EL, Mack U, Powell LW, Halliday JW. Use of chromatofo- 31. Blake MS, Johnston KH, Russell-Jones GJ, Gotschlich EC. A rapid, cusing to detect a transferrin variant in serum of alcoholic sensitive method for detection of alkaline phosphatase-conju- subjects. Clin Chem 1985;31:1543–5. gated anti-antibody on Western blots. Anal Biochem 1984;136: 20. Helander A, Bergstrom JP. Determination of carbohydrate-defi- 175–9. cient transferrin in human serum using the Bio-Rad %CDT by HPLC 32. Lipkowski M, Dibbelt L, Seyfarth M. Is there an analytical or test. Clin Chim Acta 2006;371:187–90. diagnostic advantage from including trisialo transferrin into the 21. Sillanaukee P, Lof K, Harlin A, Martensson O, Brandt R, Seppa K. fraction of carbohydrate-deficient transferrin? Lessons from a Comparison of different methods for detecting carbohydrate- comparison of two commercial turbidimetric immunoassays deficient transferrin. Alcohol Clin Exp Res 1994;18:1150–5. with the carbohydrate-deficient transferrin determination by high- 22. Oda RP, Prasad R, Stout RL, Coffin D, Patton WP, Kraft DL, et al. performance liquid chromatography. Clin Biochem 2000;33:635– Capillary electrophoresis-based separation of transferrin sialo- 41. forms in patients with carbohydrate-deficient glycoprotein syn- 33. Wuyts B, Delanghe JR, Kasvosve I, Wauters A, Neels H, Janssens drome. Electrophoresis 1997;18:1819–26. J. Determination of carbohydrate-deficient transferrin using capil- 23. Prasad R, Stout RL, Coffin D, Smith J. Analysis of carbohydrate lary zone electrophoresis. Clin Chem 2001;47:247–55. deficient transferrin by capillary zone electrophoresis. Electro- 34. Alden A, Ohlson S, Pahlsson P, Ryden I. HPLC analysis of phoresis 1997;18:1814–8. carbohydrate deficient transferrin isoforms isolated by the Axis- 24. Bortolotti F, De Paoli G, Tagliaro F. Carbohydrate-deficient trans- Shield %CDT method. Clin Chim Acta 2005;356:143–6. ferrin (CDT) as a marker of alcohol abuse: a critical review of the 35. Kasvosve I, Delanghe JR, Gomo ZA, Gangaidzo IT, Khumalo H, literature 2001–2005. J Chromatogr B Analyt Technol Biomed Life Wuyts B, et al. Transferrin polymorphism influences iron status in Sci 2006;841:96–109. blacks. Clin Chem 2000;46:1535–9. 25. Stibler H, Borg S, Joustra M. A modified method for the assay of carbohydrate-deficient transferrin (CDT) in serum. Alcohol Alcohol 36. Helander A, Wielders JP, Te Stroet R, Bergstrom JP. Comparison Suppl 1991;1:451–4. of HPLC and capillary electrophoresis for confirmatory testing of 26. Helander A. Absolute or relative measurement of carbohydrate- the alcohol misuse marker carbohydrate-deficient transferrin. Clin deficient transferrin in serum? Experiences with three immunolog- Chem 2005;51:1528–31. ical assays. Clin Chem 1999;45:131–5. 37. Helander A, Bergstrom J, Freeze HH. Testing for congenital 27. Helander A, Fors M, Zakrisson B. Study of Axis-Shield new %CDT disorders of glycosylation by HPLC measurement of serum trans- immunoassay for quantification of carbohydrate-deficient transferrin glycoforms. Clin Chem 2004;50:954–8. ferrin (CDT) in serum. Alcohol Alcohol 2001;36:406–12. 38. Carchon HA, Chevigne R, Falmagne JB, Jaeken J. Diagnosis of 28. Helander A, Eriksson G, Stibler H, Jeppsson JO. Interference of congenital disorders of glycosylation by capillary zone electro- transferrin isoform types with carbohydrate-deficient transferrin phoresis of serum transferrin. Clin Chem 2004;50:101–11. Clinical Chemistry 53:6 1122–1128 (2007) General Clinical Chemistry

Assessing the Quality of Glucose Monitor Studies: A Critical Evaluation of Published Reports

John Mahoney1* and John Ellison2

Background: In recent years, a large number of studies Glucose monitors, when used in conjunction with appro- have been published on the performance of glucose priate interventional treatment, can effectively improve monitors. The quality of these reports is not known. glycemic control (1). The US Food and Drug Administra- Methods: We searched the PubMed database for perfor- tion (FDA)3 has cleared more than 200 glucose monitors mance evaluations of handheld glucose monitors pub- for home and institutional use (2). To ascertain whether or lished from August 2002 to November 2006. Relevant not a monitor is acceptable for its intended use, the FDA articles were compared to 20 recommendations from the carefully reviews clinical and laboratory evidence pro- Standards for Reporting Diagnostic Accuracy (STARD) vided by the device manufacturer (3). and 18 recommendations from the Clinical and Labora- Clinicians are advised to evaluate medical devices tory Standards Institute (CLSI). before initial use (4). Some glucose monitor evaluations published by clinicians have reported poor results and Results: A total of 52 reports met our inclusion criteria have concluded that data from glucose monitors are and were reviewed. None (0%) of the reports conformed unreliable (5), unsatisfactory (6), or show concentration to all 38 STARD and CLSI recommendations. The range dependency (7). Other clinicians have reported positive of compliance to these recommendations varied widely results—in some cases using the same monitor—and have (median 53%; range 21%–84%). Only 1 study of the 52 concluded that data from glucose monitors are accurate reported following a CLSI recommendation for check- and meet performance expectations (8, 9). This inconsis- ing reference test results. Fewer than half (42%) of the tency in the literature is problematic because it causes reports contained STARD-recommended statements re- confusion and may slow adoption of new indications for garding how and when comparative measurements glucose monitors (e.g., continuous glucose monitoring). were performed. Four potential sources of error must be considered in Conclusions: None of the glucose monitor reports from the evaluation of any analytical device: (a) analytical our review conformed to all STARD and CLSI recom- imprecision, (b) analytical bias, (c) protocol-specific bias, mendations. Our finding that the average rate of com- and (d) random patient interferences (10). Device manu- pliance to recommendations was low suggests that facturers are generally knowledgeable about these many of the researchers did not follow published rec- sources of error and carefully follow procedures to control ommendations for study design, methodology, and re- them. Bias and imprecision are controlled by testing porting and that study quality and conclusions may products that conform to specifications, protocol-specific have been affected. Future studies evaluating the per- bias by adherence to careful study design, and random formance of glucose monitoring systems should be patient interferences by inclusion and exclusion criteria carefully designed and follow published recommenda- for recruitment of study participants. tions for methodological and reporting quality. Clinicians performing evaluation studies also need to © 2007 American Association for Clinical Chemistry be cognizant of protocol-design factors and potential sources of error (11). Guidelines have been published in an attempt to educate clinicians on proper study methodology and reporting (12, 13). Although the Standards for

Departments of 1 Global Product Support and 2 Clinical Research, LifeScan, Inc. * Address correspondence to this author at: LifeScan, Inc., 1000 Gibraltar Dr., M/S 3I, Milpitas, CA 95035-6312. Fax 1-408-941-9892; e-mail address 3 Nonstandard abbreviations: FDA, Food and Drug Administration; [email protected]. STARD, Standards for Reporting Diagnostic Accuracy; CLSI, Clinical and Received November 21, 2006; accepted March 16, 2007. Laboratory Standards Institute; SKUP, Scandinavian Evaluation of Laboratory Previously published online at DOI: 10.1373/clinchem.2006.083493 Equipment for Primary Health Care.

1122 Clinical Chemistry 53, No. 6, 2007 1123

Table 1. Clinical glucose monitor studies and their conformance to 38 recommended study factors .38͔ ؍ (STARD (20) ؉ CLSI (18͓ Total number of conforming Study statements or data Percentage, % Kendall, 2005 (16) 32 84 Garg, 2004 (17) 32 84 Chen, 2003 (18) 30 79 Kristensen, 2004 (19) 28 74 Baum, 2006 (20) 28 74 Kilo, 2005 (21) 27 71 St-Louis, 2002 (22) 27 71 Buhling, 2003 (23) 26 68 Larbig, 2003 (24) 25 66 Michel, 2005 (25) 24 63 Chlup, 2005 (26) 24 63 Kilo, 2005 (27) 24 63 Püntmann, 2003 (28) 23 61 Tieszen, 2003 (29) 23 61 Lippi, 2006 (30) 23 61 Khan, 2006 (31) 23 61 Rivers, 2006 (32) 23 61 Hawkins, 2005 (33) 22 58 Dai, 2004 (34) 22 58 Cohen, 2005 (35) 22 58 Dhatt, 2004 (36) 22 58 Oyibo, 2002 (37) 22 58 Fig. 1. Flow chart of literature search and the number of studies Kiattimongkol, 2003 (38) 21 55 meeting search criteria. Mohan, 2004 (39) 20 53 Demers, 2003 (40) 20 53 Reporting Diagnostic Accuracy (STARD) guidelines are Miendje Deyi, 2002 (8) 20 53 intended for studies of diagnostic accuracy (13) rather Bohme, 2003 (41) 20 53 than for studies of analytical performance, many of the Savoca, 2006 (42) 20 53 items of the STARD checklist are important for the Greenhalgh, 2004 (43) 19 50 readers’ interpretation of either type of study. The pur- Aboezz, 2005 (44) 19 50 Corstjens, 2006 (45) 19 50 pose of our study was to compare recent reports on blood Kanji, 2005 (46) 19 50 glucose monitor performance to these guidelines. Rao, 2005 (47) 17 45 Martin, 2005 (7) 17 45 Materials and Methods Kavsak, 2004 (48) 17 45 search strategy and report criteria Ho, 2004 (6) 17 45 We searched the PubMed database for articles from Solnica, 2003 (49) 17 45 August 2002 to November 2006 using combinations of the DirecNet, 2005 (50) 17 45 words: blood glucose, performance, evaluation, accurate, accu- Elusiyan, 2006 (51) 17 45 racy, point-of-care, meter, glucometer, and monitor. The ref- Wehmeier, 2006 (52) 17 45 erence lists of the selected articles were also reviewed and DirecNet, 2003 (9) 16 42 personal files were hand searched for additional reports. Boyd, 2005 (53) 15 39 Studies selected for inclusion were published analytical Ajala, 2003 (54) 15 39 evaluations of marketed, handheld, blood glucose moni- Pavlicek, 2006 (55) 15 39 toring systems that used a laboratory method as a com- Solnica, 2005 (56) 14 37 parison method. We excluded studies that were not in Finkielman, 2005 (5) 14 37 English, studies of nonhuman blood samples, and studies Kulkarni, 2005 (57) 14 37 of continuous monitoring and noninvasive devices. Our Medina, 2003 (58) 14 37 PubMed search terms and details were as follows: (“blood Choubtum, 2002 (59) 14 37 glucose”[MeSH Terms] OR blood glucose[Text Word]) Meex, 2006 (60) 14 37 AND (performance[Text Word] OR evaluation[Text Apperloo, 2005 (61) 10 26 Word] OR accurate[Text Word] OR accuracy[Text Word] Nobels, 2004 (62) 821 OR point-of-care[Text Word] OR meter[Text Word] OR 1124 Mahoney et al.: The Quality of Glucose Monitor Studies

Table 2. CLSI quality recommendations for glucose monitor evaluation studies, and the percentage of 52 studies that were found to contain conforming statements or data pertaining to these CLSI recommendations. Topic CLSI (NCCLS) C30-A2 factors for glucose monitor evaluation studies Percentage of studies, % 1 Blood sample Blood sample type (e.g., venous, capillary) is appropriate for monitor method. 90 2 Blood hematocrit checked to be within monitor’s acceptable range. 33 3 Blood sample Appropriate anticoagulants, blood additives, or preservatives (if used). 87 4collection method Catheter is properly flushed of IV solution prior to sampling (if done). 79 5 Skin is cleaned and dried prior to puncture (if done). 29 6 Blood sample handling Monitor and reference method are both tested from the same sample. 56 7 Blood is tested (or centrifuged) within 5 min of monitor test. Centrifuged 13 plasma is tested with reference method within 60 min of monitor test. 8 Monitor method Operators are trained to manufacturer’s instructions. 58 9 Monitor is tested in duplicate. 46 10 Reference method Laboratory method checked for stability and for being within its QC control limits. 44 11 Laboratory method is tested in duplicate. 23 12 Laboratory method is verified with NIST standard reference materials (optional). 19 13 Laboratory duplicates are within 4% or 0.22 mmol/L (4 mg/dL) (or else excluded). 2 14 Statistics and Distribution of glucose in blood samples spans monitor’s measurement range. 92 15acceptance criteria Specimen sample size is Ն40 specimen. 85 16 For glucose Ͻ4.2 mmol/L (75 mg/dL), monitor result is accurate if within 25 Ϯ0.83 mmol/L (15 mg/dL) of laboratory average. 17 For glucose Ն4.2 mmol/L (75 mg/dL), monitor result is accurate if 37 within Ϯ20% of laboratory average. 18 Individual monitor results are compared to the mean of duplicate results 10 from laboratory analyzer. meters[Text Word] OR glucometer[Text Word] OR glu- (0-point) value to each recommendation on our checklists cometers[Text Word] OR monitor[Text Word] OR moni- depending on whether the authors had (1) included the tors[Text Word]) AND (“2002/08/01”[PDAT]: “2006/11/ recommended procedure in the report, (2) included sup- 01”[PDAT]) AND Englishͳ AND “humans”[MeSH porting data confirming the use of the recommended Terms]. procedure, or (3) acknowledged in the report that the recommended procedure had been considered. Differ- assessment ences in interpretation and discrepancies in ratings be- One reviewer (J.E.) screened the titles from the computer- tween the 2 reviewers were rare and were settled via based search to determine relevant articles for retrieval. If consensus after additional review of the report for sup- the title did not provide enough information to decide porting evidence. whether or not to include the study, the abstract was read. Each checklist item was given a numerical value of 1 The full article was retrieved if the abstract did not point. Possible points included 20 STARD (reporting) provide enough information. Studies were eliminated if items and 18 CLSI (methodological) items, for a possible both reviewers (J.M., J.E.) agreed that the report did not maximum of 38 points. Calculations were based on per- meet inclusion criteria. We obtained printed copies of all centages of 38 total points, 20 (STARD), and 18 (CLSI) articles meeting our inclusion criteria. To evaluate the points. Correlation with P ϭ 0.05 was considered quality of reporting, we chose the 25-item STARD check- significant. list (13, 14). However, because whole blood glucose monitors are not diagnostic devices, 5 STARD criteria Results (STARD checklist items 1, 9,12, 21, and 23) were deemed A total of 1407 titles/abstracts were retrieved, of which 93 not applicable and were not scored. Because study meth- were initially proposed (Fig. 1). On further review, 41 of odology should be evaluated independently of the quality these studies were found ineligible and were excluded. of the reporting (15), we developed an additional 18-item Exclusions were because of inappropriate (nonhuman method checklist based on Clinical and Laboratory Stan- fresh whole blood) test samples or the use of an inappro- dards Institute (CLSI) C30-A2 (12). priate reference method (e.g., methods not traceable to In analyzing published reports, we found that not all materials or methods of higher order). For the 52 selected STARD or CLSI factors were obvious or clearly reported. reports published between August 2002 and November In addition, the omission of procedural statements in the 2006, the scores ranged from a high of 32 points (84%) to report was considered to indicate only that the proce- a low of 8 points (21%) (Table 1). The average score of the dures were not reported, not that they were not per- glucose monitor reports was low (median score, 20 of 38 formed. Therefore, we assigned a yes (1-point) or no points or 53%). No published report incorporated 100% of Clinical Chemistry 53, No. 6, 2007 1125

Table 3. STARD recommendations applied to 52 published glucose monitor evaluation studies, and the percentage of these studies that were found to contain conforming statements or data regarding STARD recommendations. Percentage of Section or topic STARD factors for reporting diagnostic accuracy studies, % 1 Keywords Identify the article as a study of diagnostic accuracy (recommended MeSH heading “sensitivity N/A and specificity”). 2 Introduction State the research questions or study aims, such as estimating diagnostic accuracy or 100 comparing accuracy between tests or across participant groups. 3 Participants The study population: the inclusion and exclusion criteria, setting, and locations where the data 73 were collected. 4 Participant recruitment: Was recruitment based on presenting symptoms, results from previous 71 tests, or the fact that the participants had received the index tests or the reference standard? 5 Participant sampling: Was the study population a consecutive series of participants defined by 50 the selection criteria in item 3 and 4? If not, specify how participants were further selected. 6 Data collection: Was data collection planned before the index test and reference standard were 100 performed (prospective study) or after (retrospective study)? 7 Test methods The reference standard and its rationale. 94 8 Technical specifications of material and methods involved including how and when 42 measurements were taken, and/or cite references for index tests and reference standard. 9 Definition of and rationale for the units, cutoffs, and/or categories of the results of the index N/A tests and the reference standard. 10 The number, training, and expertise of the persons executing and reading the index tests and 56 the reference standard. 11 Whether or not the readers of the index tests and reference standard were blind (masked) to the 0 results of the other test and describe any other clinical information available to the readers. 12 Statistical methods Method for calculating or comparing measures of diagnostic accuracy, and the statistical N/A methods used to quantify uncertainty (e.g., 95% CIs). 13 Methods for calculating test reproducibility, if done. 75 14 Participants When the study was done, including beginning and ending dates of recruitment. 21 15 Clinical and demographic characteristics of the study population (e.g., age, sex, spectrum of 65 presenting symptoms, comorbidity, current treatments, recruitment centers). 16 The number of participants satisfying the criteria for inclusion who did or did not undergo the 88 index tests and/or the reference standard; describe why participants failed to receive either test (a flow diagram is strongly recommended). 17 Test results Time interval from the index tests to the reference standard, and any treatment administered 42 between. 18 Distribution of severity of disease (define criteria) in those with the target condition; other 23 diagnoses in participants without the target condition. 19 A cross tabulation of the results of the index tests (including indeterminate and missing results) 81 by the results of the reference standard; for continuous results, the distribution of test results by the results of the reference standard. 20 Any adverse events from performing the index tests or the reference standard. 4 21 Estimates Estimates of diagnostic accuracy and measures of statistical uncertainty (e.g., 95% CIs). N/A 22 How indeterminate results, missing responses, and outliers of the index tests were handled. 33 23 Estimates of variability of diagnostic accuracy between subgroups of participants, readers or N/A centers, if done. 24 Estimates of test reproducibility, if done. 73 25 Discussion Discuss the clinical applicability of the study findings. 96 NA, Not applicable. the quality factors recommended by STARD or CLSI. The Discussion CLSI checklist developed by the authors and the percent- Our study shows that the average glucose monitor report age of conforming reports to 18 CLSI recommendations used only ϳ50% of the combined CLSI and STARD (range 2%–92%) is shown in Table 2, and the STARD recommendations and that the overall quality of reports checklist and the percentage of reports that conformed to is low. Compliance to these recommendations varied 20 STARD recommendations (range 0%–100%) in Table 3. widely (range 21%–84%), and none of the 52 reports No significant trend was found when the report scores conformed to all recommendations. These findings sug- were grouped by journal type or assessed by date of gest that many investigators disagree with, are unaware publication (P ϭ 0.5). Neither the source journal nor dates of, or are neglectful of published CLSI and STARD of publication were found to be predictive of report recommendations for conducting and reporting glucose conformity to published recommendations. monitor evaluation studies. 1126 Mahoney et al.: The Quality of Glucose Monitor Studies

A report’s procedural statements, especially how and duplicate reference tests to ensure both method and when monitor and reference measurements are per- glycemic stability, and emphasized control of elapsed formed, provide important information regarding the time and glycolysis. Unfortunately, these SKUP reports quality and reproducibility of the study. We found that are not found in the PubMed database. only 42% of the studies reported this information (Table 3) To our knowledge, a single, published checklist that and only 13% reported following appropriate sample includes key reporting and methodological factors for timing and handling procedures (Table 2). Control of glucose monitor evaluations does not exist. We selected sampling time is important because after a carbohydrate the STARD and CLSI checklists because they are pub- load blood glucose can change rapidly at a sampling site lished and both contain important elements. Our study (63). Postcollection control of sample handling time is also shows that although a large number of glucose monitor important because glycolysis can cause rapid glycemic evaluation studies have been published over a 4-year change, depending on the hematocrit (64). If either of period, investigators did not address many of the vari- these circumstances is not controlled, observed differ- ables that can adversely impact internal and external ences in the data could be caused by glycemic concentra- validity. All glucose-monitoring systems have perfor- tion differences in the comparative samples instead of mance limitations [e.g., hematocrit extremes (70)] that are differences between the 2 methods. included in the published manufacturer labeling, yet we We observed that many investigators made a number found several studies in which the devices were evaluated of assumptions. Some assumed that the concentration of under off-label conditions. The availability and ease with glucose in capillary and venous blood is equivalent, which clinicians can perform evaluation studies using although equivalence cannot be assumed for individuals glucose monitors is relatively unique among in vitro tests. in the postprandial state (65). In addition, 29 (56%) of 52 With the growing incidence of diabetes and new technol- studies reported testing the same sample with both mon- ogies for measuring blood glucose on the horizon, it is itor and comparative methods (Table 2). Thus for the reasonable to believe that the number of such studies will other 23 reports, observed differences in the data may be continue to grow. We believe that a checklist that com- attributable to glycemic concentration differences in the bines key elements from the STARD and CLSI recommen- comparative samples. Most investigators also assumed dations, if published and used, would help to improve the that there is little error associated with their reference quality of monitor evaluation studies and could form the method; only 19% checked the bias of their reference with basis for future checklists applicable to continuous mon- traceable materials (Table 2), although reference glucose itoring and noninvasive devices. Such a tool has the methods can have a total error of up to 10% (66). Only 1 potential to improve the quality of future studies. of the studies reported that they followed CLSI advice to check that duplicate reference tests were stable and We conclude that none of the glucose monitor evaluation acceptable. reports in our review conform to all published quality Reports differed considerably in regard to the use of recommendations, and that the overall quality of reports appropriate acceptance criteria for glucose monitor per- is low. The range of conformance to STARD and CLSI formance. Many reports used expert opinion, medical recommendations varied widely, suggesting that many of society opinion, or their own acceptance criteria, whereas the researchers did not follow published recommenda- relatively few used CLSI acceptance criteria for glucose tions for study design and methodology, an omission that monitors (12) (Table 2), which are identical to acceptance may have adversely affected study quality. Future studies criteria published by the International Standards Organi- evaluating glucose monitoring systems should be care- zation (67). fully designed and should follow published recommen- One limitation of our study was our choice to limit our dations for methodological and reporting quality. search to English language reports, although we believe that inclusion of studies published in other languages would not alter our conclusions. In addition, glucose Grant/funding support: LifeScan, Inc., provided funding for monitor evaluation studies exist (not revealed by our this study. search) that are, in our opinion, of relatively high quality. Financial disclosures: Both authors are employees of Life- The Scandinavian Evaluation of Laboratory Equipment Scan, Inc., a Johnson & Johnson company, and both hold for Primary Health Care (SKUP) has performed a number equity interests in Johnson & Johnson. of monitor evaluation studies and has issued reports. A Acknowledgements: We thank Drs. David Horwitz and review of 2 reports found that SKUP followed 100% of David Price for their helpful suggestions and comments. A CLSI recommendations and 85% of STARD recommenda- portion of this work was presented in poster format at the tions (68), (69). These 2 reports emphasized monitor 2006 AACC Annual Meeting, Chicago, IL. training, performed a thorough reference method evaluation, tested duplicate monitor tests, tested the reference References method before and after the monitor testing, properly 1. Standards of Medical Care in Diabetes. Diabetes Care 2006; checked [within 4% or 0.22 mmol/L (4 mg/dL)] the 29(Suppl):S4–42. Clinical Chemistry 53, No. 6, 2007 1127

2. Gutman S, Bernhardt P, Pinkos A, Moxey-Mims M, Knott T, Cooper 21. Kilo C, Pinson M, Joynes JO, Joseph H, Monhaut N, Parkes JL, et J. Regulatory aspects of invasive glucose measurements. Diabe- al. Evaluation of a new blood glucose monitoring system with tes Technol Ther 2002;4:775–7. auto-calibration. Diabetes Technol Ther 2005;7:283–94. 3. Review Criteria Assessment of Portable Blood Glucose Monitoring 22. St-Louis P, Ethier J. An evaluation of three glucose meter systems In Vitro Diagnostic Devices Using Glucose Oxidase, Dehydroge- and their performance in relation to criteria of acceptability for nase, or Hexokinase Methodology. HHS Publication FDA-96-604. neonatal specimens. Clin Chim Acta 2002;322:139–48. Rockville, MD: US Department of Health and Human Services, 23. Buhling KJ, Henrich W, Kjos SL, Siebert G, Starr E, Dreweck C, et Public Health Service, Food and Drug Administration, Center for al. Comparison of point-of-care-testing glucose meters with stan- Devices and Radiological Health, Office of Device Evaluation, dard laboratory measurement of the 50g-glucose-challenge test Division of Clinical Laboratory Services; 1996. (GCT) during pregnancy. Clin Biochem 2003;36:333–7. 4. Nichols JH. Interpreting method evaluations. Diabetes Technol 24. Larbig M, Forst T, Mondok A, Forst S, Pfu¨tzner A. Investigation of Ther 2002;4:623–5. the accuracy of the blood glucose monitoring device Prestige IQ. 5. Finkielman JD, Oyen LJ, Afessa B. Agreement between bedside Diab Nutr Metab 2003;16:257–61. blood and plasma glucose measurement in the ICU setting. Chest 25. Michel A, Ku¨ster H, Krebs A, Kadow I, Paul W, Nauck M, et al. 2005;127:1749–51. Evaluation of the Glucometer Elite XL device for screening for 6. Ho HT, Yeung WKY, Young BWY. Evaluation of “point of care” neonatal hypoglycemia. Eur J Pediatr 2005;164:660–4. devices in the measurement of low blood glucose in neonatal 26. Chlup R, Payne M, Zapletalova J, Komenda S, Doubravova B, practice. Arch Dis Child Fetal Neonatal Ed 2004;89:F356–9. Reznickova M, et al. Results of self monitoring on gluocometer 7. Martin DD, Shephard MDS, Freeman H, Bulsara MK, Jones TW, systems Advance and Optium in daily routine. Biomed Pap Med Davis EA, et al. Point-of-care testing of HbA1c and blood glucose Fac Univ Palacky Olomouc Czech Repub 2005;149:127–39. in a remote Aboriginal Australian community. Med J Aust 2005; 27. Kilo C, Dickey WT, Joynes JO, Pinson MB, Baum JM, Parkes JL, et 182:524–7. al. Evaluation of a new blood glucose monitoring system with 8. Miendje Deyi VY, Philippe M, Alexandre KC, De Nayer P, Hermans auto-calibration for home and hospital bedside use. Diabetes Res MP. Performance evaluation of the Precision PCx point-of-care Clin Pract 2006;74:66–74. blood glucose analyzer using discriminant ratio methodology. Clin 28. Pu¨ntmann I, Wosniok W, Haeckel R. Comparison of several Chem Lab Med 2002;40:1052–5. point-of-care testing (POCT) glucometers with an established 9. The Diabetes Research in Children Network (DIRECNET) Study laboratory procedure for the diagnosis of Type 2 diabetes using Group. A multicenter study of the accuracy of the One Touch Ultra the discordance rate: a new statistical approach. Clin Chem Lab home glucose meter in children with type 1 diabetes. Diabetes Med 2003;41:809–20. Technol Ther 2003;5:933–41. 29. Tieszen KL, New JP. Alternate site blood glucose testing: do 10. Krouwer JS. How to improve total error modeling by accounting for patients prefer it? Diabet Med 2003;20:325–8. error sources beyond imprecision and bias [Letter]. Clin Chem 30. Lippi G, Salvagno GL, Guidi GC, Negri M, Rizzotti P. Evaluation of 2001;47:1329–30. four portable self-monitoring blood glucose meters. Ann Clin 11. Binette TM, Cembrowski GS. Diverse influences on blood glucose Biochem 2006;43:408–13. measurements in the ICU setting. Chest 2005;128:3084–5. 31. Khan AI, Vasquez Y, Gray J, Wians FH, Kroll MH. The variability of 12. NCCLS. Point-of-Care Blood Glucose Testing in Acute and Chronic results between point-of-care testing glucose meters and the Care Facilities; Approved Guideline – Second Edition. NCCLS central laboratory analyzer. Arch Pathol Lab Med 2006;130: document C30–A2 (ISBN 1-56238-471-6). NCCLS, 2002. 1527–32. 13. Bossuyt PM, Reitsma JB, Bruns DE, Gatsonis CA, Glasziou PP, 32. Rivers SM, Kane MP, Bakst G, Busch RS, Hamilton RA. Precision Irwig LM, et al. The STARD statement for reporting studies of and accuracy of two blood glucose meters: FreeStyle Flash versus diagnostic accuracy: explanation and elaboration. Clin Chem One Touch Ultra. Am J Health Syst Pharm 2006;63:1411–6. 2003;49:7–18. 14. Bossuyt PM, Reitsma JB, Bruns DE, Gatsonis CA, Glasziou PP, 33. Hawkins RC. Evaluation of Roche Accu-Chek Go and Medisense Irwig LM, et al. Towards complete and accurate reporting of Optium blood glucose meters. Clin Chim Acta 2005;353:127–31. studies of diagnostic accuracy: the STARD initiative. Clin Chem 34. Dai K, Tai D, Ho P, Chen C, Peng W, Chen S, et al. Accuracy of the 2003;49:1–6. EasyTouch blood glucose self-monitoring system: a study of 516 15. Huwiler-Mu¨ntener K, Ju¨ni P, Junker C, Egger M. Quality of reporting cases. Clin Chim Acta 2004;349:135–41. of randomized trials as a measure of methodologic quality. JAMA 35. Cohen M, Boyle E, Delaney C, Shaw J. A comparison of blood 2002;287:2801–4. glucose meters in Australia. Diabetes Res Clin Pract 2006;71: 16. Kendall DM, Kaplan RA, Paulson CF, Parkes JL, Tideman AM. 113–8. Accuracy and utility of a 10-test disk blood glucose meter. 36. Dhatt GS, Agarwal M, Bishawi B. Evaluation of a glucose meter Diabetes Res Clin Pract 2005;67:29–35. against analytical quality specifications for hospital use. Clin Chim 17. Garg SK, Carter JA, Mullen L, Folker AC, Parkes JL, Tideman AM. Acta 2004;343:217–21. The clinical performance and ease of use of a blood glucose meter 37. Oyibo SO, Pritchard GM, Mclay L, James E, Laing I, Gokal R, et al. that uses a 10-test disk. Diabetes Technol Ther 2004;6:495– Blood glucose overestimation in diabetic patients on continuous 502. ambulatory peritoneal dialysis for end-stage renal disease. Diabet 18. Chen ET, Nichols JH, Duh SH, Hortin G. Performance evaluation of Med 2002;19:693–6. blood glucose monitoring devices. Diabetes Technol Ther 2003; 38. Kiattimongkol W, Watanachai A, Suprasongsin C. Evaluation of 5:749–68. the bedside glucose monitoring system in neonatal units. J Med 19. Kristensen GBB, Nerhus K, Thue G, Sandberg S. Standardized Assoc Thai 2003;86:883–8. evaluation of instruments for self-monitoring of blood glucose by 39. Mohan V, Deepa R, Shefali AK, Poongothai S, Monica M, patients and a technologist. Clin Chem 2004;50:1068–71. Karkuzhali K. Evaluation of One Touch Horizon: a highly affordable 20. Baum JM, Monhaut NM, Parker DR, Price CP. Improving the quality glucose monitor. J Assoc Physicians India 2004;52:779–82. of self-monitoring blood glucose measurement: a study in reduc- 40. Demers J, Kane MP, Bakst G, Busch RS, Hamilton RA. Accuracy of ing calibration errors. Diabetes Technol Ther 2006;8:347–57. home blood glucose monitors using forearm blood samples: 1128 Mahoney et al.: The Quality of Glucose Monitor Studies

FreeStyle versus One Touch Ultra. Am J Health Syst Pharm 56. Solnica B, Naskalski JW. Quality control of SMBG in clinical 2003;60:1130–5. practice. Scand J Clin Lab Invest 2005;65(Suppl):80–6. 41. Bo¨hme P, Floriot M, Sirveaux MA, Durain D, Ziegler O, Drouin P, et 57. Kulkarni A, Saxena M, Price G, O’Leary MJ, Jacques T, Myburgh JA. al. Evolution of analytical performance in portable glucose meters Analysis of blood glucose measurements using capillary and in the last decade. Diabetes Care 2003;26:1170–5. arterial blood samples in intensive care patients. Intensive Care 42. Savoca R, Jaworek B, Huber AR. New “plasma referenced” POCT Med 2005;31:142–5. glucose monitoring systems: are they suitable for glucose moni- 58. Vela´zquez Medina D, Climent C. Comparison of outpatient point of toring and diagnosis of diabetes? Clin Chim Acta 2006;372:199– care glucose testing vs venous glucose in the clinical laboratory. 201. P R Health Sci J 2003;22:385–9. 43. Greenhalgh S, Bradshaw S, Hall CM, Price DA. Forearm blood 59. Choubtum L, Mahachoklertwattana P, Udomsubpayakul U, Pree- glucose testing in diabetes mellitus. Arch Dis Child 2004;89: yasombat C. Accuracy of glucose meters in measuring low blood 516–8. glucose levels. J Med Assoc Thai 2002;85:S1104–10. 44. Aboezz R, Miller DR. Accuracy of portable blood glucose monitors. 60. Meex C, Poncin J, Chapelle JP, Cavalier E. Analytical validation of J Am Pharm Assoc (Wash DC) 2005;45:514–6. the new plasma calibrated Accu-Chek® test strips (Roche Diag- 45. Corstjens AM, Ligtenberg JJM, van der Horst ICC, Spanjersberg R, nostics). Clin Chem Lab Med 2006;44:1376–8. Lind JSW, Tulleken JE, et al. Accuracy and feasibility of point-of- 61. Apperloo JJ, Vader HL. A quantitative appraisal of interference by care and continuous blood glucose analysis in critically ill ICU icodextrin metabolites in point-of-care glucose analyses. Clin patients. Crit Care 2006;10:R135. Chem Lab Med 2005;43:314–8. 46. Kanji S, Buffie J, Hutton B, Bunting PS, Singh A, McDonald K, et al. 62. Nobels F, Beckers F, Bailleul E, De Schrijver P, Sierens L, Van Reliability of point-of-care testing for glucose measurement in Crombrugge P. Feasibility of a quality assurance programme of critically ill adults. Crit Care Med 2005;33:2778–85. bedside blood glucose testing in a hospital setting: 7 years’ 47. Rao LV, Jakubiak F, Sidwell JS, Winkelman JW, Snyder ML. experience. Diabet Med 2004;21:1288–91. Accuracy evaluation of a new glucometer with automated hemat- 63. Ellison JM, Stegmann JM, Colner SL, Michael RH, Sharma MK, ocrit measurement and correction. Clin Chim Acta 2005;356: Ervin KR, et al. Rapid changes in postprandial blood glucose 178–83. produce concentration differences at finger, forearm, and thigh 48. Kavsak PA, Zielinski N, Li D, McNamara PJ, Adeli K. Challenges of sampling sites. Diabetes Care 2002;25:961–4. implementing point-of-care testing (POCT) glucose meters in a 64. Sidebottom RA, Williams PR, Kanarek KS. Glucose determina- pediatric acute care setting. Clin Biochem 2004;37:811–7. tions in plasma and serum: potential error related to increased 49. Solnica B, Naskalski JW, Sieradzki J. Analytical performance of hematocrit. Clin Chem 1982;28:190–2. glucometers used for routine glucose self-monitoring of diabetic patients. Clin Chim Acta 2003;331:29–35. 65. Voss EM, McNeill L, Cembrowski GS. Assessing the accuracy of 50. The Diabetes Research in Children Network (DIRECNET) Study your blood glucose meter. LifeScan Publication AW 055-272. Group. Accuracy of newer-generation home blood glucose meters Milpitas, CA: LifeScan, 2003. in a diabetes research in children network (DirecNet) inpatient 66. Ehrmeyer SS, Laessig RH, Leinweber JE, Oryall JJ. 1990 Medi- exercise study. Diabetes Technol Ther 2005;7:675–80. care/CLIA final rules for proficiency testing: minimum intralabora- 51. Elusiyan JBE, Adeodu OO. Adejuyigbe EA. Evaluating the validity of tory performance characteristics (CV and bias) needed to pass. a bedside method of detecting hypoglycemia in children. Pediatr Clin Chem 1990;36:1736–40. Emerg Care 2006;22:488–90. 67. ISO 15197. In vitro diagnostic test systems: requirements for 52. Wehmeier M, Arndt BT, Schumann G, Ku¨lpmann WR. Evaluation blood-glucose monitoring systems for self-testing in managing and quality assessment of glucose concentration measurement in diabetes mellitus. First edition dated 2003-05-01. ISO 15197: blood by point-of-care testing devices. Clin Chem Lab Med 2006; 2003(E) International Standards Organization, Geneva, Switzer- 44:888–93. land. 53. Boyd R, Leigh B, Stuart P. Capillary versus venous bedside blood 68. Scandinavian evaluation of laboratory equipment for primary glucose estimations. Emerg Med J 2005;22:177–9. health care, SKUP. Report on OneTouch Ultra, SKUP/2005/39. 54. Ajala MO, Oladipo OO, Fasanmade O, Adewole TA. Laboratory http://www.SKUP.nu (accessed February 2007). assessment of three glucometers. Afr J Med Med Sci 2003;32: 69. Scandinavian evaluation of laboratory equipment for primary 279–82. health care, SKUP. Report on OneTouch GlucoTouch, SKUP/ 55. Pavlicek V, Garzoni D, Urech P, Bra¨ndle M. Inaccurate self- 2005/40. http://www.SKUP.nu (accessed February 2007). monitoring of blood glucose readings in patients on chronic 70. Tang Z, Lee JH, Louie RF, Kost GJ. Effects of different hematocrit ambulatory peritoneal dialysis with icodextrin. Exp Clin Endocrinol levels on glucose measurements with handheld meters for point- Diabetes 2006;114:124–6. of-care testing. Arch Pathol Lab Med 2000;124:1135–40. Clinical Chemistry 53:6 1129–1136 (2007) General Clinical Chemistry

Folate and Methylation Status in Relation to Phosphorylated Tau Protein(181P) and ␤-Amyloid(1–42) in Cerebrospinal Fluid

Rima Obeid,1 Mariz Kasoha,1 Jean-Pierre Knapp,1 Panagiotis Kostopoulos,2 George Becker,2† Klaus Fassbender,2 and Wolfgang Herrmann1*

Background: Increased plasma total homocysteine tween hyperhomocysteinemia and neurodegeneration. (tHcy) is a risk factor for neurological diseases, but the Lowering tHcy and SAH might protect the brain by underlying pathophysiology has not been adequately preventing P-tau accumulation. explained. © 2007 American Association for Clinical Chemistry Methods: We evaluated concentrations of tHcy, S-adenosyl homocysteine (SAH), S-adenosyl methionine The importance of methyl group metabolism for brain

(SAM), folate, and vitamin B12 in cerebrospinal fluid function has been recognized (1). S-adenosyl methionine (CSF) and plasma or serum from 182 patients with (SAM)3 is generated from methionine metabolism and different neurological disorders. We measured concen- constitutes a very important methyl donor in the central trations of phosphorylated tau protein (P-tau)(181P) and nervous system. In a 1st step, methionine is activated by ␤-amyloid(1–42) in the CSF. ATP to form SAM. Subsequently, SAM is converted into Results: Aging was associated with higher concentra- S-adenosyl homocysteine (SAH) after donating its methyl tions of tHcy and SAH in the CSF, in addition to lower group to other cellular components. SAH, a potent com- concentrations of CSF folate and lower SAM:SAH ratio. petitive inhibitor for several methyl transferases, binds to Concentrations of CSF SAH and CSF folate correlated the enzymes with a greater affinity than SAM (2). SAH is ؊0.28, further hydrolyzed into homocysteine (Hcy) that can be؍ and r 0.46 ؍ significantly with those of P-tau (r respectively). Moreover, P-tau correlated negatively either remethylated into methionine or transsulfurated to -؊0.40, P <0.001). The associ- cysteine. Alternatively, Hcy can be converted via SAH؍ with SAM:SAH ratio (r ation between SAH and higher P-tau was observed in 3 hydrolase into SAH in a reversible reaction that favors the age groups (<41, 41–60, and >60 years). CSF tHcy was formation of SAH when concentrations of total Hcy -tHcy) increase. SAM-dependent methylation encom) ؍ ␤) predicted by concentrations of CSF cystathionine and age passes a wide range of biological reactions, including ,(0.349 ؍ ␤) ؊0.403), albumin؍ ␤) folate ,(0.478 -myelin synthesis; synthesis and catabolism of neurotrans .(0.298 ؍ ␤) Conclusions: tHcy concentration in the brain is related mitters and phospholipids; and methylation of proteins, to age, B vitamins, and CSF albumin. Increase of CSF RNA, and DNA (3). Several neurological disorders have SAH is related to increased CSF P-tau; decreased degra- been related to abnormalities in methyl group or 1-carbon dation of P-tau might be a plausible explanation. Dis- metabolisms (4, 5). turbed methyl group metabolism may be the link be- In addition to its major role as a methyl donor, SAM represents a metabolic link between the methylation and the transsulfuration pathways (6). SAM enhances the Departments of 1 Clinical Chemistry and Laboratory Medicine and 2 Neu- activity of cystathionine ␤ synthase, and thereby the flux rology, Faculty of Medicine, University Hospital of Saarland, Homburg/Saar, Germany. † Dr. Becker died on August 6, 2003. * Address correspondence to this author at: Department of Clinical Chem- istry and Laboratory Medicine, University Hospital of the Saarland, Kirrberger 3 Nonstandard abbreviations: SAM, S-adenosyl methionine; SAH, S-ad- Straße, Geba¨ude 57, 66421 Homburg, Germany. Fax 49-6841-1630703; e-mail enosylhomocysteine; Hcy, homocysteine; tHcy, total homocysteine; P-tau, [email protected]. hyperphosphorylated tau protein; CSF, cerebrospinal fluid; PP2A, protein Received January 2, 2007; accepted February 13, 2007. phosphatase 2A; Cys, cystathionine; MMA, methylmalonic acid; holoTC, Previously published online at DOI: 10.1373/clinchem.2006.085241 holotranscobalamin.

1129 1130 Obeid et al.: Folate and Methylation Status in CSF

of Hcy via the transsulfuration pathway, to produce more with peripheral neuropathy, and a control group of 65 cysteine, the precursor of glutathione. Because folate patients with various neurological diseases other than

(5-methyltetrahydrofolate) and vitamin B12 (methyl cobal- dementia, Parkinson disease, polyneuropathy, multiple amin) play important roles in recycling Hcy to methio- sclerosis, stroke, Alzheimer disease, or depression. The nine, these 2 vitamins are considered essential for the diagnosis of Alzheimer disease was by criteria of a availability of methyl groups in the brain. Substantial cognitive subscale score on the Alzheimer’s Disease As- evidence suggests that the alternative remethylation path- sessment Scale and Mini Mental Examination Scores. The way of Hcy via betaine-Hcy methyltransferase does not control group has been described (21). Exclusion criteria account for Hcy remethylation in the brain (7). included renal or liver dysfunction and alcoholism. Increased plasma tHcy is an independent risk factor for We collected nonfasting blood samples from all pa- neurodegenerative and psychiatric disorders (5, 8–13).A tients; serum and EDTA plasma were available, and CSF tHcy concentration Ͼ10 ␮mol/L in plasma has been samples were obtained during clinically indicated lumbar associated with brain atrophy in elderly people (14). punctures. Blood and CSF samples collected within 24 h Moreover, longitudinal studies demonstrated an associa- of each other were directly centrifuged and stored at Ϫ tion between baseline plasma tHcy concentrations and 80 °C until analysis. Aliquots of the EDTA plasma and some measures of cognitive or memory decline after CSF were immediately deproteinized using perchloric Ϫ several years of follow-up (15, 16). Increased plasma tHcy acid (10%). These samples were stored at 80 °C and used is associated with increased concentrations of tHcy and for SAM and SAH assays. CSF samples contaminated SAH in the brain (17). This may explain the link be- with peripheral blood or hemoglobin were not included tween hyperhomocysteinemia and neural damage, but in the study. The study was approved by the Ethics the underlying pathophysiology has not been adequately Committee at the Saarland University Hospital, and all explained. patients gave written informed consent. We measured concentrations of SAM and SAH by use Neurodegenerative diseases share a common feature, of a modified liquid chromatography–tandem mass spec- accumulation of misfolded proteins. Tau and ␤-amyloid trometry method according to Gellekink et al. (22). The are 2 examples of proteins that accumulate in brains of CVs for SAM and SAH assays were 4.8% and 8%, re- patients with dementia. Concentrations of ␤-amyloid(1– spectively, at concentrations of 103 nmol/L for SAM and 42) and hyperphosphorylated tau protein (P-tau) in (181P) 15.6 nmol/L for SAH. We measured concentrations of cerebrospinal fluid (CSF) predict the development of tHcy, cystathionine (Cys), and methylmalonic acid dementia (18). A positive association between plasma (MMA) in serum and CSF samples by use of gas chroma- concentrations of tHcy and ␤-amyloid(1–40) and ␤-amy- tography–mass spectrometry as described (23). The CV loid(1–42) has been reported in neurodegenerative dis- for tHcy was Ͻ5% (at 8.0 and 16.0 ␮mol/L) in serum and ease (11). Moreover, tHcy can increase the vulnerability Ͻ ␮ ␤ 10% in CSF (at 0.30 mol/L). The CV for MMA was of neurons to being damaged by -amyloid (19). Like- Ͻ6% in serum and CSF (at 290 and 540 nmol/L, respec- wise, hyperphosphorylation of tau influences its structure tively), and the CV for cystathionine was Ͻ8% in serum and association with the plasma membrane. Increased (at 300 nmol/L) and Ͻ10% in CSF (at 60 nmol/L). We P-tau may be related to a lower phosphatase activity or to used pool serum and pool CSF prepared in-house on each increased activity of kinases (see Fig. 1 in the Data run with the study samples to calculate CV. Supplement that accompanies the online version of this We measured the concentrations of total vitamin B12 article at http://www.clinchem.org/content/vol53/ and folate in serum and CSF by use of a chemilumines- issue5). Because the activation of protein phosphatase 2A cence immunoassay (ADVIA Centaur System); plasma (PP2A) is SAM dependent, a possible link between de- concentrations of vitamin B6 (pyridoxal-5-phosphate, mentia and reduced methylation has been hypothesized PLP) by use of HPLC connected with fluorescence detec- (20); this association has never been investigated in hu- tor using reagents from Immundiagnostik; concentrations man studies. Data about the distributions and determi- of holotranscobalamin (holoTC) in serum and CSF by use nants of biomarkers of B vitamins and methylation status of an RIA from Axis-Shield (24); concentrations in the ␤ in blood and CSF are limited. We investigated the rela- CSF of P-tau(181P) and -amyloid(1–42) by use of specific tionship of markers of neurodegeneration, methylation, monoclonal antibodies against these 2 proteins (Inno- and vitamin status in CSF samples from patients with Genetics); and serum concentrations of creatinine, choles- neurological disorders. terol, triglycerides, and high-density and low-density cholesterol by use of routine methods at our laboratory. patients and methods The study included 182 patients who were treated at the statistical analyses Department of Neurology, University Hospital of the Data analyses were performed with SPSS (version 12). All Saarland, from April 2002 to April 2004: 31 patients with continuous variables were skewed and were therefore multiple sclerosis, 19 patients with stroke, 31 patients log-transformed to approach gaussian distribution before with dementia (8 with Alzheimer disease), 36 patients applying tests that propose such a distribution of the data. Clinical Chemistry 53, No. 6, 2007 1131

One-way ANOVA was used for multiple comparisons. higher serum concentrations of holoTC in addition to The post hoc Tamhane test was performed to identify the lower median concentrations of CSF SAH and higher CSF significantly different group means when ANOVA was SAM:SAH ratios. Despite the marked age differences, significant. Correlations between different variables were patients with stroke showed no significant differences in examined by Spearman ␳ test. All tests were 2-sided; P any blood or CSF marker compared with control patients values Ͻ0.05 were considered statistically significant. (Table 1). The age of our patients seemed to affect concentrations Results of the metabolites in CSF. To test this, we pooled the data The main characteristics of the study population accord- from all study populations and divided our patients into ing to disease are presented in Table 1 in the online Data 4 groups according to quartiles of age (Table 3). Higher Supplement. Concentrations of plasma/serum and CSF age was associated with higher blood concentrations of vitamin biomarkers are shown in Tables 1 and 2. Patients tHcy, Cys, MMA, SAH, SAM, and creatinine. Lower with polyneuropathy were significantly older than con- concentrations of serum folate and holoTC were also trol patients and had higher ratios of CSF:serum albumin. related to advanced age. Concentrations of CSF tHcy and Concentrations of blood markers (tHcy, Cys, SAH, and CSF SAH increased with age, and concentrations of CSF SAM) were higher in patients with polyneuropathy than folate and ratios of SAM:SAH in CSF decreased with in control patients (Table 1). Nevertheless, these results increasing age (Table 3). Concentrations of CSF MMA seem to be related to the older age of patients with tended to decrease with age, and concentrations of CSF polyneuropathy, because the differences were no longer holoTC and CSF SAM did not differ significantly with age significant after adjusting for age and CSF:serum albumin (Table 3). ratio (Table 1). To investigate the association between concentrations Patients with dementia were much older than the of CSF SAH and other vitamin biomarkers, we divided control patients (see Table 1 in the online Data Supple- concentrations of CSF SAH into quartiles (data from all ment). Comparison between CSF and blood markers was patient groups; Table 4). Higher concentrations of CSF not performed, because adjustment for age was not pos- SAH were related to higher concentrations of plasma sible in this case. Patients with multiple sclerosis had SAH and SAM and slightly lower SAM:SAH ratios in the

Table 1. Concentrations of serum/plasma and CSF vitamin and methylation markers according to disease status.a All Control patients Multiple sclerosis Stroke Peripheral neuropathy n 182 65 31 19 36 Plasma/serum markers tHcy, ␮mol/L 10.7 (7.4–18.0) 9.4 (7.3–16.2) 10.7 (7.6–14.4) 10.2 (7.0–17.4) 11.9 (7.0–19.1)b,c Cys, nmol/L 372 (180–886) 329 (150–661) 355 (157–1146) 420 (154–874) 415 (199–1038)b,c MMA, nmol/L 197 (128–394) 186 (109–300) 197 (116–366) 176 (144–510) 189 (142–340)

Total vitamin B12, 249 (166–419) 242 (158–403) 266 (181–402) 253 (162–485) 270 (183–630) pmol/L Folate, nmol/L 19.6 (9.9–41.1) 19.3 (11.4–42.2) 21.1 (9.5–41.6) 20.3 (12.6–42.0) 22.3 (9.9–43.8)

Vitamin B6, nmol/L 36.9 (13.8–100.8) 37.9 (17.8–90.6) 10.6 (16.1–168.4) 33.3 (12.2–115.6) 42.0 (15.3–199.7) HoloTC, pmol/L 70 (32–136) 63 (32–128) 89 (35–156)b 64 (32–154) 81 (44–143) SAM, nmol/L 123 (89–218) 116 (87–171) 114 (74–236) 130 (93–206) 138 (88–228)b,c SAH, nmol/L 18.1 (10.0–37.9) 16.2 (9.3–27.4) 13.5 (9.9–51.5) 18.2 (10.9–48.2) 19.6 (11.0–37.2)b,c SAM:SAH ratio 7.0 (3.8–13.0) 6.8 (3.8–13.8) 8.2 (3.0–14.4) 7.7 (3.4–11.9) 6.7 (4.1–10.1) CSF markers tHcy, ␮mol/L 0.10 (0.06–0.17) 0.09 (0.06–0.16) 0.08 (0.05–0.16) 0.08 (0.06–0.15) 0.11 (0.07–0.18)b,c Cys, nmol/L 49 (22–98) 54 (17–108) 42 (22–105) 48 (18–105) 34 (18–75) MMA, nmol/L 359 (267–552) 359 (266–574) 367 (281–525) 426 (333–600) 326 (214–511) Folate, nmol/L 19.5 (13.9–26.6) 20.7 (14.1–27.7) 20.5 (14.6–29.5) 19.2 (14.2–26.4) 18.6 (12.9–23.5)b,c HoloTC, pmol/L 16 (4–27) 16 (6–23) 16 (3–26) 17 (4–73) 17 (7–39) SAM, nmol/L 267 (180–356) 268 (197–355) 305 (151–389) 273 (172–339) 268 (187–385) SAH, nmol/L 13.5 (8.4–24.1) 13.2 (7.7–24.0) 10.2 (7.4–16.7)b 13.5 (7.1–24.8) 15.5 (9.1–22.5)b,c SAM:SAH ratio 19 (10–34) 20 (11–39) 28 (13–42)b 22 (10–30) 18 (12–24) b P-tau(181P), ng/L 39 (20–79) 35 (18–63) 37 (22–68) 48 (19–107) 49 (26–101) ␤-Amyloid(1–42), ng/L 651 (374–1060) 739 (444–1092) 552 (340–737)b 587 (327–919) 606 (520–1000) a Data are median (10th–90th percentile). b P Ͻ0.05 vs control group (ANOVA and post hoc Tamhane test). c Differences no longer significant after adjusting for age and CSF:serum albumin. 1132 Obeid et al.: Folate and Methylation Status in CSF

Table 2. CSF and blood markers in patients with dementia observed. Thus, higher concentrations of CSF SAH were associated with higher CSF tHcy and Cys, lower CSF .(31 ؍ n) folate, and lower CSF SAM:SAH ratio. Median (10th–90th percentile) Plasma/serum markers One very important association was observed between tHcy, ␮mol/L 12.9 (8.2–28.2) concentrations of CSF SAH and CSF P-tau (Table 4)—a Cys, nmol/L 491 (247–1764) highly significant direct correlation was found (Fig. 1A). ϭ MMA, nmol/L 293 (149–681) CSF P-tau correlated negatively with SAM:SAH ratio (r Ϫ Ͻ Total vitamin B12, pmol/L 195 (145–364) 0.40, P 0.001). Furthermore, a negative correlation was Folate, nmol/L 16.1 (5.4–27.1) found between CSF P-tau and CSF folate (Fig. 1B). The

Vitamin B6, nmol/L 19.5 (8.0–45.4) correlations between CSF P-tau and either CSF SAH, CSF HoloTC, pmol/L 58 (20–133) SAM:SAH, or CSF folate remained significant after adjust- SAM, nmol/L 159 (95–369) ing for age. SAH, nmol/L 25.1 (11.1–61.7) Fig. 2 shows median and 25th–75th percentiles of CSF SAM:SAH ratio 6.9 (1.6–12.3) SAH according to concentration of P-tau and age (Ͻ41, CSF markers 41– 60, and Ͼ60 years). In addition to the influence of age ␮ tHcy, mol/L 0.10 (0.06–0.32) on concentrations of CSF SAH, we found among each age Cys, nmol/L 47 (25–125) group that patients who had higher concentrations of MMA, nmol/L 419 (267–746) P-tau also had higher concentrations of CSF SAH com- Folate, nmol/L 18.4 (12.3–27.4) HoloTC, pmol/L 11 (4–23) pared with patients of similar age and lower CSF P-tau. SAM, nmol/L 250 (180–356) We applied multiple backward regression analyses to SAH, nmol/L 15.2 (9.7–30.0) find factors that significantly predict CSF tHcy, CSF SAH, SAM:SAH ratio 17 (7–29) and CSF SAM. Concentrations of CSF tHcy were pre- ␤ ϭ ␤ ϭϪ P-tau(181P), ng/L 36 (21–95) dicted by CSF Cys ( 0.478), CSF folate ( 0.403), ␤-Amyloid(1–42), ng/L 666 (290–1090) CSF albumin (␤ ϭ 0.349), and age (␤ ϭ 0.298). Significant predictors of concentrations of SAH in the CSF were CSF tHcy (␤ ϭ 0.377, P Ͻ0.001) and age (␤ ϭ 0.195, P ϭ 0.002). Concentrations of CSF SAM were not predicted by any plasma. Moreover, higher CSF SAH was associated with marker estimated in this study. Moreover, concentrations lower concentrations of serum folate and higher tHcy in of Cys correlated with SAM in the plasma (r ϭ 0.510, P serum. Similar associations between CSF SAH and con- Ͻ0.001), and this correlation was weaker in CSF (r ϭ centrations of other vitamin biomarkers in the CSF were 0.174, P ϭ 0.041).

Table 3. Blood and CSF markers according to quartile of age.a Quartile 1, 17–40 years Quartile 2, 41–55 years Quartile 3, 56–68 years Quartile 4, 69–86 years P (ANOVA) Plasma/serum markers tHcy, ␮mol/L 9.0 (6.0–12.7) 10.2 (7.8–20.2)b 11.0 (7.9–15.0)b 14.0 (8.1–28.4)b Ͻ0.001 Cys, nmol/L 246 (129–472) 388 (188–844)b 362 (197–877)b 510 (283–1950)b Ͻ0.001 MMA, nmol/L 166 (90–294) 179 (133–299) 196 (147–370)b 253 (146–621)b Ͻ0.001

Total B12, pmol/L 253 (165–433) 259 (173–435) 243 (162–424) 246 (162–422) 0.699 Folate, nmol/L 18.8 (11.9–42.8) 21.9 (10.5–43.7) 21.3 (12.7–42.8) 15.2 (7.4–25.8)b Ͻ0.001 HoloTC, pmol/L 70 (26–128) 71 (41–128) 73 (39–163) 57 (26–139) 0.077 SAM, nmol/L 106 (73–143) 116 (89–179) 132 (97–219)b 164 (105–327)b Ͻ0.001 SAH, nmol/L 12.8 (8.6–27.0) 16.4 (10.4–29.2) 19.8 (11.8–37.3)b 25.6 (13.8–79.7)b Ͻ0.001 SAM:SAH ratio 8.3 (3.4–14.1) 7.3 (4.1–12.2) 7.5 (3.8–12.7) 6.1 (2.8–11.2) 0.079 Creatinine, ␮mol/L 66.3 (53.0–88.4) 70.7 (53.0–97.2) 79.6 (53.0–106.1) 88.4 (53.0–189.2)b Ͻ0.001 CSF markers tHcy, ␮mol/L 0.07 (0.05–0.12) 0.10 (0.06–0.17) 0.10 (0.06–0.14) 0.13 (0.09–0.24)b Ͻ0.001 Cys, nmol/L 53 (17–110) 42 (19–97) 33 (18–70) 65 (27–131) 0.002 MMA, nmol/L 392 (251–776) 359 (289–509) 348 (261–522) 359 (258–554) 0.188 Folate, nmol/L 21.0 (16.3–28.9) 19.8 (12.3–27.4) 19.7 (15.3–23.1) 16.9 (11.9–21.8)b Ͻ0.001 HoloTC, pmol/L 14 (5–25) 17 (3–23) 18 (6–38) 14 (4–29) 0.204 SAM, nmol/L 254 (174–332) 270 (154–375) 259 (181–356) 274 (190–374) 0.464 SAH, nmol/L 10.5 (6.5–14.4) 12.7 (7.9–20.2) 15.3 (9.1–25.4)b 16.7 (10.3–30.0)b Ͻ0.001 SAM:SAH ratio 26 (16–40) 19 (11–36) 16 (8–31)b 17 (8–29)b Ͻ0.001 a Data are median (10th–90th percentile). b P Ͻ0.05 according to post hoc Tamhane test vs quartile 1. Clinical Chemistry 53, No. 6, 2007 1133

Table 4. Concentrations of plasma/serum and CSF biomarkers of B vitamins according to quartile of CSF-SAH.a Quartile 1, Quartile 2, Quartile 3, Quartile 4, 4.7–10.1 nmol/L 10.2–15.4 nmol/L 15.5–17.4 nmol/L 17.5–39.3 nmol/L P (ANOVA) Age, years 41 (16) 44 (18) 57 (15)b 64 (13)b Ͻ0.001 Plasma/serum markers tHcy, ␮mol/L 10.2 (3.1) 10.0 (8.1) 12.1 (9.6) 12.6 (5.0)b 0.019 Cys, nmol/L 308 (320) 388 (686) 451 (644)b 491 (571)b 0.007 MMA, nmol/L 201 (117) 188 (94) 214 (138) 262 (256) 0.018

Total vitamin B12, pmol/L 279 (1159) 280 (506) 274 (1111) 254 (90) 0.851 Folate, nmol/L 22.3 (16.2) 21.4 (11.8) 17.8 (13.1) 16.5 (10.2)b 0.036 HoloTC, pmol/L 68 (37) 56 (40) 62 (44) 69 (42) 0.349 SAM, nmol/L 110 (34) 130 (128) 134 (62) 157 (101)b 0.002 SAH, nmol/L 15.2 (6.8) 17.0 (11.8) 22.2 (41.7)b 24.4 (39.9)b 0.001 SAM:SAH 7.3 (3.5) 7.7 (3.1) 6.1 (4.5) 6.4 (3.7) 0.162 Creatinine, ␮mol/L 69.8 (14.4) 73.3 (16.9) 97.0 (94.3) 93.9 (56.3) 0.032 CSF markers tHcy, ␮mol/L 0.08 (0.04) 0.10 (0.26)b 0.10 (0.03)b 0.14 (0.19)b Ͻ0.001 Cys, nmol/L 42 (27.5) 46 (33) 46 (22)b 49 (35)b 0.666 MMA, nmol/L 414 (318) 372 (130) 363 (111) 372 (127) 0.334 Folate, nmol/L 20.8 (4.2) 20.2 (5.3) 18.4 (4.4) 17.1 (4.8)b 0.003 HoloTC, pmol/L 11 (12) 13 (24) 13 (7) 14 (20) 0.431 SAM, nmol/L 247 (71) 270 (58) 260 (61) 258 (72) 0.450 SAM:SAH 29.6 (9.4) 23.3 (6.1)b 17.1 (4.3)b 11.1 (3.9)b Ͻ0.001 b b Ͻ P-tau(181P), ng/L 29.7 (12.4) 34.5 (12.6) 44.9 (26.8) 49.6 (34.4) 0.001 ␤-Amyloid(1–42), ng/L 606 (188) 566 (237) 630 (250) 702 (239) 0.135 a Data are geometric mean (SD). b P Ͻ0.05 vs quartile 1 (post hoc Tamhane test).

Discussion a plausible explanation for the association between Hcy Concentrations of tHcy or related biomarkers in CSF have metabolism and neurodegenerative diseases. been tested in only a few studies (18, 25). Human studies Previous reports documented higher serum or plasma demonstrated decreased SAM or increased tHcy or SAH concentrations of tHcy and MMA in patients with demen- in brain or CSF from patients with certain disorders of the tia compared with nondemented individuals (5). More- central nervous system (26, 27). The most remarkable over, lower concentrations of CSF folate were observed in finding in this study is that increased concentrations of late-onset dementia patients compared with nonde- P-tau in the brain may be related to disturbed Hcy mented patients (28). Our study has shown that concen- metabolism or methylation status. These findings may be trations of folate decrease and concentrations of tHcy

Fig. 1. Correlation between concentra-

tions of P-tau(181P) and CSF SAH and CSF folate (Spearman test). 1134 Obeid et al.: Folate and Methylation Status in CSF

higher concentration of SAH in CSF via decrease of CSF folate. The abnormal aggregation of P-tau into paired helical filaments takes several decades to develop and is considered one of the hallmarks of Alzheimer disease. Tau aggregation that takes place in the cytosol is toxic for the neurons. Our study showed that concentrations of P-tau in the CSF are associated with concentrations of folate and SAH in the CSF (Fig. 1). This association was independent of age and evident in each age group (Fig. 2). Previous results of experimental folate deficiency in neuronal cells support our results (30). Folate deprivation in neuroblas- toma cells induced a marked increase in tHcy concentration, in addition to an increase in the immunoreactivity of P-tau (30), and the increment of P-tau was reversible after adding folate to the cultures (30). Animal studies have shown that P-tau accumulates in cases of diet-induced (folate deficiency) or genetic (apolipoprotein E deficiency) oxidative stress (31). Whereas SAM treatment partly restored oxidative stress, it did not reduce P-tau in Fig. 2. Box plot showing concentrations of CSF SAH according to CSF apolipoprotein E–deficient animals (31). Altogether, these

P-tau(181P) and age. results suggest that P-tau accumulates in folate deficiency Ͻ Ն Concentrations of P-tau(181P) were divided into 2 groups ( or median value; and in cases of oxidative stress, but this phenotype is 38.5 ng/L) in the pooled data (P-tau data available from 141 patients). P values are according to the Mann–Whitney test. Median CSF SAH values are indicated probably reversible, at least in short-term deficiency. in the Fig. Because reducing P-tau might lead to improved memory function, independent of paired helical filaments (32), tHcy-lowering treatment may improve memory function increase in CSF with age (Table 3). CSF folate was a via reducing P-tau. This effect should be tested in folate- stronger predictor (␤ ϭϪ0.403) of CSF tHcy than age (␤ ϭ deficient individuals after vitamin supplementation. 0.298). The positive correlation between plasma and CSF There is a plausible biological mechanism by which concentrations of tHcy has been confirmed by a previous folate deprivation or increased SAH can cause increased study (29). Therefore, increase of tHcy in the blood may P-tau. The dephosphorylation of P-tau depends on the indicate its increase in the brain, where tHcy can have activity of PP2A (see Fig. 1 in the online Data Supple- many neurotoxic effects. ment). Activating this phosphatase is accomplished by a Disturbed methyl group metabolism in the brain might specific transmethylase, the phosphatase methyltrans- be closely related to age and to methyl group metabolism ferase 1, which is SAM dependent (33). Low concentra- in other parts of the body. As suggested by our results, tions of SAM or a low SAM:SAH ratio results in a lower increase of CSF SAH with age might be related to in- activity of PP2A and accumulation of P-tau (20). This is in creased concentrations of tHcy in the plasma or in the CSF line with our results, in which a lower SAM:SAH ratio (Table 3). Hcy is converted into SAH via SAH-hydrolase. was negatively related to concentrations of P-tau in the The reaction in the SAH direction is favored in case of CSF. Earlier studies demonstrated that the protein con- excess tHcy. This is in line with in vivo results showing centration of phosphatase methyltransferase 1 and that of that Hcy administration led to increased brain SAH (17). the methylated C subunit of PP2A were ϳ40% lower in Because SAH is a potent inhibitor of many transmeth- frontal and temporal extracts from Alzheimer disease ylases, the lower ratio of SAM:SAH in the CSF indicates a patients compared with control patients (34). In addition, hypomethylation state in the brain and may affect several PP2A activity and gene expression were markedly re- important biological pathways. duced in the brains of Alzheimer disease patients (35).As It is of interest that aging was related to higher CSF a whole, these data strongly suggest that alterations in concentrations of tHcy and SAH, in addition to lower methyl group metabolism may contribute to the etiology folate and lower ratio of SAM:SAH in the CSF (Table 3). of dementia by inhibiting dephosphorylation of P-tau. These metabolic changes with age may be very important In the brains of Alzheimer disease patients, the intra- factors that play a paramount role in the genesis of cerebral deposition of ␤-amyloid (mainly 1–42) is the age-related disorders. In addition to the role of age as a most important pathologic process leading to dementia. significant modulator of CSF SAH, our data demonstrated In contrast to ␤-amyloid(1–40), ␤-amyloid(1–42) is insol- that a lower folate status (in the circulation or in the CSF) uble and can accumulate in the plaques. Concentrations of is related to an increased concentration of SAH in the CSF ␤-amyloid(1–42) are lower in CSF from patients with (Table 4). It is plausible that aging is associated with a dementia compared with nondemented patients. The ac- Clinical Chemistry 53, No. 6, 2007 1135

cumulation of ␤-amyloid is a long-term process that is 5. Clarke R, Smith AD, Jobst KA, Refsum H, Sutton L, Ueland PM. thought to start at a young age. We found no association Folate, vitamin B12, and serum total homocysteine levels in between ␤-amyloid(1–42) and methylation markers, vita- confirmed Alzheimer disease. Arch Neurol 1998;55:1449–55. 6. Prudova A, Bauman Z, Braun A, Vitvitsky V, Lu SC, Banerjee R. mins, or tHcy in the whole group or in patients with S-Adenosylmethionine stabilizes cystathionine beta-synthase and dementia. Our patients, other than those with dementia, modulates redox capacity. Proc Natl Acad Sci U S A 2006;103: were not tested for cognitive performance, and we cannot 6489–94. exclude the possibility that some elderly people were 7. McKeever MP, Weir DG, Molloy A, Scott JM. Betaine-homocysteine slightly demented, which might affect concentrations of methyltransferase: organ distribution in man, pig and rat and CSF ␤-amyloid(1–42). Available in vitro evidence has subcellular distribution in the rat. Clin Sci (Lond) 1991;81:551–6. suggested that tHcy accelerates dementia by stimulating 8. Bottiglieri T, Laundy M, Crellin R, Toone BK, Carney MW, Reynolds ␤-amyloid deposition in the brain (36). In line with this, a EH. Homocysteine, folate, methylation, and monoamine metabolism in depression. J Neurol Neurosurg Psychiatry 2000;69:228– positive association between plasma concentrations of ␤ 32. tHcy and concentrations of -amyloid has been docu- 9. Ravaglia G, Forti P, Maioli F, Martelli M, Servadei L, Brunetti N, et mented in patients with neurodegenerative disease (11). al. Homocysteine and folate as risk factors for dementia and Another clinical study demonstrated that folate treatment Alzheimer disease. Am J Clin Nutr 2005;82:636–43. lowered concentrations of plasma ␤-amyloid(1–40) (37). 10. Wright CB, Paik MC, Brown TR, Stabler SP, Allen RH, Sacco RL, et The effect of SAM or B vitamin treatments on the insolu- al. Total homocysteine is associated with white matter hyperin- ble ␤-amyloid(1–42) protein in CSF has not been tested. tensity volume: the Northern Manhattan Study. Stroke 2005;36: SAM treatment protects the neurons from degenera- 1207–11. 11. Irizarry MC, Gurol ME, Raju S, az-Arrastia R, Locascio JJ, Tennis M, (38 39) tion by several mechanisms , . The protective effect et al. Association of homocysteine with plasma amyloid beta of SAM was suppressed by simultaneous administration protein in aging and neurodegenerative disease. Neurology 2005; of SAH (40). These results, in addition to those of our 65:1402–8. study, suggest that increased SAH, rather than reduced 12. Perry IJ. Homocysteine as a risk factor for cerebrovascular dis- SAM, may be a more important mediator in neurodegen- ease and stroke. In: Robinson K, ed. Homocysteine and Vascular eration. Because we demonstrated that CSF tHcy is an Disease. Dordrecht, The Netherlands: Kluwer Academic Publish- important determinant of CSF SAH (regression analysis), ers, 2000:151–72. keeping tHcy and SAH at low concentrations may be 13. Coull BM, Malinow MR, Beamer N, Sexton G, Nordt F, deGarmo P. Elevated plasma homocyst(e)ine concentration as a possible important for enhancing P-tau degradation. independent risk factor for stroke. Stroke 1990;21:572–6. 14. Sachdev PS, Valenzuela M, Wang XL, Looi JC, Brodaty H. Rela- In summary, Hcy metabolism in the circulation is closely tionship between plasma homocysteine levels and brain atrophy related to that in the brain. This relation seems to extend in healthy elderly individuals. Neurology 2002;58:1539–41. beyond a simple exchange of tHcy, SAH, SAM, or the 15. Seshadri S, Beiser A, Selhub J, Jacques PF, Rosenberg IH, vitamins across the blood–brain barrier. Higher SAH D’Agostino RB, et al. Plasma homocysteine as a risk factor for can cause P-tau accumulation, increasing the risk for dementia and Alzheimer’s disease. N Engl J Med 2002;346:476– 83. neurodegenerative diseases. The response of P-tau to B 16. Nurk E, Refsum H, Tell GS, Engedal K, Vollset SE, Ueland PM, et vitamin treatment in deficient individuals needs to be al. Plasma total homocysteine and memory in the elderly: the investigated. Hordaland Homocysteine study. Ann Neurol 2005;58:847–57. 17. Gharib A, Chabannes B, Sarda N, Pacheco H. In vivo elevation of mouse brain S-adenosyl-l-homocysteine after treatment with l- homocysteine. J Neurochem 1983;40:1110–2. Grant/funding support: The study was supported by a grant 18. Hansson O, Zetterberg H, Buchhave P, Londos E, Blennow K, from Karl and Lore Stiftung. Minthon L. Association between CSF biomarkers and incipient Financial disclosures: None declared. Alzheimer’s disease in patients with mild cognitive impairment: a follow-up study. Lancet Neurol 2006;5:228–34. 19. Ho PI, Collins SC, Dhitavat S, Ortiz D, Ashline D, Rogers E, et al. References Homocysteine potentiates beta-amyloid neurotoxicity: role of oxidative stress. J Neurochem 2001;78:249–53. 1. Mulder C, Schoonenboom NS, Jansen EE, Verhoeven NM, van 20. Vafai SB, Stock JB. Protein phosphatase 2A methylation: a link Kamp GJ, Jakobs C, et al. The transmethylation cycle in the brain between elevated plasma homocysteine and Alzheimer’s Dis- of Alzheimer patients. Neurosci Lett 2005;386:69–71. ease. FEBS Lett 2002;518:1–4. 2. Hoffman DR, Marion DW, Cornatzer WE, Duerre JA. S-Adenosylme- 21. Obeid R, Kostopoulos P, Knapp JP, Kasoha M, Becker G, Fass- thionine and S-adenosylhomocysteine metabolism in isolated bender K, et al. Biomarkers of folate and vitamin B12 are related liver. Effects of L-methionine, L-homocystein, and adenosine. in blood and cerebrospinal fluid. Clin Chem 2007;53:326–33. J Biol Chem 1980;255:10822–7. 22. Gellekink H, van Oppenraaij-Emmerzaal D, van Rooij A, Struys EA, 3. Chiang PK, Gordon RK, Tal J, Zeng GC, Doctor BP, Pardhasaradhi den Heijer M, Blom HJ. Stable-isotope dilution liquid chromatog- K, et al. S-Adenosylmethionine and methylation. FASEB J 1996; raphy-electrospray injection tandem mass spectrometry method 10:471–80. for fast, selective measurement of S-adenosylmethionine and 4. Weir DG, Scott JM. The biochemical basis of the neuropathy in S-adenosylhomocysteine in plasma. Clin Chem 2005;51:1487– cobalamin deficiency. Baillieres Clin Haematol 1995;8:479–97. 92. 1136 Obeid et al.: Folate and Methylation Status in CSF

23. Obeid R, Kuhlmann MK, Kohler H, Herrmann W. Response of 32. Santacruz K, Lewis J, Spires T, Paulson J, Kotilinek L, Ingelsson homocysteine, cystathionine, and methylmalonic acid to vitamin M, et al. Tau suppression in a neurodegenerative mouse model treatment in dialysis patients. Clin Chem 2005;51:196–201. improves memory function. Science 2005;309:476–81. 24. Ulleland M, Eilertsen I, Quadros EV, Rothenberg SP, Fedosov SN, 33. Tolstykh T, Lee J, Vafai S, Stock JB. Carboxyl methylation regu- Sundrehagen E, et al. Direct assay for cobalamin bound to lates phosphoprotein phosphatase 2A by controlling the associa- transcobalamin (holo-transcobalamin) in serum. Clin Chem 2002; tion of regulatory B subunits. EMBO J 2000;19:5682–91. 48:526–32. 34. Sontag E, Hladik C, Montgomery L, Luangpirom A, Mudrak I, Ogris 25. Regland B, Abrahamsson L, Blennow K, Grenfeldt B, Gottfries CG. E, et al. Downregulation of protein phosphatase 2A carboxyl CSF-methionine is elevated in psychotic patients. J Neural Transm methylation and methyltransferase may contribute to Alzheimer 2004;111:631–40. disease pathogenesis. J Neuropathol Exp Neurol 2004;63:1080– 26. Bottiglieri T, Godfrey P, Flynn T, Carney MW, Toone BK, Reynolds 91. EH. Cerebrospinal fluid S-adenosylmethionine in depression and 35. Vogelsberg-Ragaglia V, Schuck T, Trojanowski JQ, Lee VM. PP2A dementia: effects of treatment with parenteral and oral S-adeno- mRNA expression is quantitatively decreased in Alzheimer’s dis- sylmethionine. J Neurol Neurosurg Psychiatry 1990;53:1096–8. ease hippocampus. Exp Neurol 2001;168:402–12. 36. Fuso A, Seminara L, Cavallaro RA, D’Anselmi F, Scarpa S. 27. Surtees R, Hyland K. Cerebrospinal fluid concentrations of S- S-Adenosylmethionine/homocysteine cycle alterations modify adenosylmethionine, methionine, and 5-methyltetrahydrofolate in DNA methylation status with consequent deregulation of PS1 and a reference population: cerebrospinal fluid S-adenosylmethionine BACE and beta-amyloid production. Mol Cell Neurosci 2005;28: declines with age in humans. Biochem Med Metab Biol 1990;44: 195–204. 192–9. 37. Flicker L, Martins RN, Thomas J, Psych JA, Taddei K, Vasikaran 28. Serot JM, Christmann D, Dubost T, Bene MC, Faure GC. CSF-folate SD, et al. B-vitamins reduce plasma levels of beta amyloid. levels are decreased in late-onset AD patients. J Neural Transm Neurobiol Aging 2006 [Epub ahead of print]. 2001;108:93–9. 38. Rao AM, Baskaya MK, Maley ME, Kindy MS, Dempsey RJ. 29. Selley ML, Close DR, Stern SE. The effect of increased concen- Beneficial effects of S-adenosyl-l-methionine on blood-brain bar- trations of homocysteine on the concentration of (E)-4-hydroxy-2- rier breakdown and neuronal survival after transient cerebral nonenal in the plasma and cerebrospinal fluid of patients with ischemia in gerbils. Brain Res Mol Brain Res 1997;44:134–8. Alzheimer’s disease. Neurobiol Aging 2002;23:383–8. 39. Villalobos MA, De La Cruz JP, Cuerda MA, Ortiz P, Smith-Agreda 30. Ho PI, Ashline D, Dhitavat S, Ortiz D, Collins SC, Shea TB, et al. JM, Sanchez DLC. Effect of S-adenosyl-l-methionine on rat brain Folate deprivation induces neurodegeneration: roles of oxidative oxidative stress damage in a combined model of permanent focal stress and increased homocysteine. Neurobiol Dis 2003;14:32– ischemia and global ischemia-reperfusion. Brain Res 2000;883: 42. 31–40. 31. Chan A, Shea TB. Dietary and genetically-induced oxidative stress 40. Sato H, Hariyama H, Moriguchi K. S-Adenosyl-l-methionine pro- alter tau phosphorylation: influence of folate and apolipoprotein E tects the hippocampal CA1 neurons from the ischemic neuronal deficiency. J Alzheimers Dis 2006;9:399–405. death in rat. Biochem Biophys Res Commun 1988;150:491–6. Clinical Chemistry 53:6 1137–1143 (2007) Clinical Immunology

Biosensor Analysis of ␤2-Glycoprotein I–Reactive Autoantibodies: Evidence for Isotype-Specific Binding and Differentiation of Pathogenic from Infection-Induced Antibodies

Jochen Metzger,1† Philipp von Landenberg,2† Marcus Kehrel,1 Alexander Buhl,1 Karl J. Lackner,2 and Peter B. Luppa1*

Background: For the laboratory diagnosis of the an- different APS patient sera to the biosensor-immobilized tiphospholipid syndrome (APS) we developed a biosen- ␤2GPI. In contrast to APS patient samples, no signifi- sor with the ability to distinguish between disease- cant anti-␤2GPI binding (response levels <35 RU) was relevant anti-␤2-glycoprotein I (␤2GPI) autoantibodies observed in samples from healthy individuals or from (anti-␤2GPI) and pathogen-specific ␤2GPI cross-reac- patients suffering from SLE, syphilis, or parvovirus B19 tive antibodies that occur transiently during infections. infection. Methods: We used a surface plasmon resonance (SPR) Conclusions: The SPR biosensor system enables spe- biosensor device. For the detection of anti-␤2GPI in cific detection of APS-associated ␤2GPI-reactive APL serum samples, affinity-purified human ␤2GPI was and differentiation from ␤2GPI cross-reactive antibod- covalently attached to a functionalized n-alkanethiol ies that occur frequently during acute infections. The self-assembling monolayer on the biosensor chip. After established association/dissociation plot for anti-␤2GPI verifying the specificity of the biosensor system with a responses in APS patient sera gives additional informa- panel of monoclonal antibodies to ␤2GPI, we analyzed tion regarding the influence of anti-␤2GPI IgG and IgM sera from healthy donors and patients suffering from isotype distribution. APS, systemic lupus erythematosus (SLE), syphilis, or © 2007 American Association for Clinical Chemistry parvovirus B19 infections. The SPR results were compared with ␤2GPI-specific ELISA. The antiphospholipid syndrome (APS)3 is an autoimmune Results: Using the SPR biosensor, we recorded antigen disorder characterized by arterial and venous thrombo- binding curves with response levels in the range of ses, recurrent fetal loss, and increased serum titers of 50–500, resonance units (RU) for anti-␤2GPI ELISA- antiphospholipid antibodies (APLs), such as lupus anti- positive APS patient sera. The amplitudes of the an- coagulants and antibodies against cardiolipin (CL) (1). tiphospholipid antibody (APL) responses in the biosen- According to the Sapporo classification criteria for sor correlated with the overall IgG and IgM anti-␤2GPI APS, identification by ELISA of APLs in moderate to high ELISA titers with a correlation coefficient of 0.87. More- serum titers is a significant finding only when it persists over, we observed immunoglobulin isotype-specific as- for more than 6 weeks and is closely associated with sociation and dissociation profiles for APL binding of clinical features (2). The poor positive predictive value of APL detection is mainly attributed to the frequent presence of these autoantibodies in other autoimmune dis- 1 Institute of Clinical Chemistry and Pathobiochemistry, Klinikum Rechts eases, particularly systemic lupus erythematosus (SLE), der Isar der TU Mu¨ nchen, Mu¨ nchen, Germany. 2 Institute of Clinical Chemistry and Laboratory Medicine, Johannes Gutenberg-Universita¨t Mainz, Mainz, Germany. † These authors contributed equally to this work. 3 Nonstandard abbreviations: APS, antiphospholipid syndrome; APL, an- * Address correspondence to this author at: Institute of Clinical Chemistry tiphospholipid antibody; CL, cardiolipin; SLE, systemic lupus erythematosus; and Pathobiochemistry, Ismaninger Str. 22, D-81675 Munich, Germany. Fax ␤2GPI, ␤2-glycoprotein I; anti-␤2GPI, anti-␤2GPI antibodies; SPR, surface 49-89-4140-4875; e-mail [email protected]. plasmon resonance; RU, resonance units; VDRL, Venereal Disease Research Received September 6, 2006; accepted March 14, 2007. Laboratory; HBS, HEPES-buffered saline; SAM, self-assembling monolayer; Previously published online at DOI: 10.1373/clinchem.2006.079632 CMD, carboxymethyl dextran; mAb, monoclonal antibody.

1137 1138 Metzger et al.: Differentiation of Anti-␤2GPI

and in response to various infectious agents such as an increase of arbitrary resonance units (RU) and plotted Treponema pallidum or parvovirus B19. In the latter, occur- in the form of a sensorgram. The interaction curves were rence of serum APLs is usually not correlated with evaluated for total amounts of bound antibody and ap- thrombotic events. parent percentage of dissociation, the latter being a mea- Although APL was initially thought to bind directly to sure of immune complex stability. Our goal in analyzing the phospholipid CL, results of 3 independent studies both variables was to gain information on the overall revealed that ␤2-glycoprotein I (␤2GPI), a highly glycosy- avidity of the autoantibody response. lated plasma protein, is involved in immune complex formation (3–5). Since then, evidence has suggested that patients ␤2GPI reactivity is related to the pathogenic potential of We studied 30 patients with APS, 9 with SLE, 10 with APLs, leading to the classification of ␤2GPI-dependent positive results for the Venereal Disease Research Labo- autoimmune- and ␤2GPI-independent infection-associ- ratory (VDRL) test for syphilis, and 20 with parvovirus ated types of APLs (6, 7). B19 infection. All serum samples were leftover human Although a complex of CL and ␤2GPI was identified as specimens that were not individually identifiable, in ac- the main antigenic determinant, Arvieux et al. (8) re- cordance with Food and Drug Administration guidelines ported that most APLs also recognize ␤2GPI linked to a (18). The serum samples were accompanied by clinical solid support in the absence of any phospholipid. Other information, but the sample source was not identifiable to studies have revealed significant correlations between the investigator. All APS patients whose sera were used in APL reactivities against ␤2GPI and those against CL, this study satisfied the Sapporo classification criteria for providing additional evidence that ␤2GPI alone may be APS (19), and the SLE patients fulfilled Ͼ4 of the revised used as a specific marker for APS diagnosis (9, 10). Even American College of Rheumatology criteria for the clas- in the case of anti-␤2GPI antibodies (anti-␤2GPI), how- sification of SLE (20). We also obtained 20 samples from ever, patients exhibit heterogeneous polyclonal antibody apparently healthy controls recruited from the laboratory responses. The identification of heterogeneous responses staff. The controls were age matched and were assumed strongly associated with clinical symptoms subsequently on the basis of a medical and clinical chemistry examina- leads to the detection of infection-induced APLs not tion to be free of any acute or chronic disease. Written involved in the development of an APS. This process informed consent was obtained from these participants. makes testing at 2 time points necessary for differentiation Each blood sample was collected in 10-mL tubes with- of APS-associated from infection-associated APLs. Anti- out anticoagulant, and after clotting was centrifuged at ␤2GPI may occur as IgM, IgG, or IgA. The significance of 1500g for 15 min. Sera were collected and stored in IgM and IgA isotypes, however, is questionable (11–13). aliquots at Ϫ70 °C. The finding that particularly high affinity anti-␤2GPI IgGs are associated with clinical signs (14, 15) suggests that elisa antigen-induced affinity maturation of polyreactive natu- Concentrations of ␤2GPI-specific IgG and IgM APLs in ral IgM antibodies and IgG isotype switching both con- sera were determined with ELISA reagent sets manufac- tribute to the development of APS (16). Thus, avidity tured by Orgentec and Phadia, subsequently referred testing was considered an appropriate method for differ- to as ELISA-A and -B, respectively. Results were inter- entiating between pathogenic and nonpathogenic anti- preted according to the following cutoff values: ELISA-A, ␤2GPI (17). 8 kU/L; ELISA-B, 15 kU/L. ELISA does not allow reliable differentiation of affinity and avidity of antibody binding. Therefore, we applied a apparatus surface plasmon resonance (SPR) biosensor device that ␤2GPI immobilization and APL interaction analyses were allows resolution of association and dissociation phases to carried out in a BiaCoreX® SPR biosensor (Biacore AB) at monitor the reactivity of APLs in APS patient sera to 25 °C. For all measurements, SIA Kit Au biosensor chips surface-immobilized human ␤2GPI. from Biacore AB were used.

Materials and Methods removal of IgG and IgM The principles of SPR technology are described in the We used agarose beads coated with goat antibodies Data Supplement Text that accompanies the online ver- (Sigma) against the IgG Fc-fragment or the IgM ␮-chain. sion of this article at http://www.clinchem.org/content/ We added 20 ␮L of a 50% suspension of antibody agarose vol53/issue6. In a typical biosensor run, one of the in HEPES-buffered saline to 180 ␮L of 1:90 prediluted interaction partners is coupled to the sensor surface, serum. After a 30-min incubation, the suspensions were whereas the dissolved opponent passes over the surface centrifuged at 10 000g for 10 min. Supernatants were under continuous flow conditions. The SPR signal is collected and assayed immediately for ␤2GPI binding generated by a change in mass concentration over the activity in the SPR biosensor. biosensor chip as the soluble analyte binds to or dissoci- Details on the preparation of the self-assembling ates from the immobilized ligand. Binding is expressed in monolayer (SAM), biosensor measurements, kinetic anal- Clinical Chemistry 53, No. 6, 2007 1139

yses, and statistics, as well as data on reagents and ent first dissociation constant (KD1) by fitting 4 binding antibodies, are given in the online Data Supplement Text. curves of 5F10.F3, representing concentrations of 34, 17, 11, and 9 nmol/L, to the bivalent binding model of the

Results BIA evaluation program. This KD1 value, 9.52 nmol/L, immobilization of ␤2gpi was taken as a measure for the affinity of 1 antibody Preliminary experiments with ␤2GPI, linked to a carboxy- binding site. The quality of the fitting analysis was methyl dextran (CMD) hydrogel matrix of CM5 chips indicated by a ␹2 value of 7.8. For a rough approximation ␤ from Biacore AB, revealed that binding of 2GPI-reactive of avidity, we calculated the overall KD by applying the APLs on the SPR biosensor strongly depends on the 1:1 Langmuir model to these sensorgrams. We obtained a surface density of ␤2GPI (data not shown). To gain value of 2.27 nmol/L, considered reasonable because the surfaces with high ligand densities, we established planar calculated ␹2 value of 8.04 was in the same range as the carboxyl-terminated SAM surfaces, allowing ␤2GPI to be value for the bivalent binding mode. From the 2 human coupled in the range of 2600–3000 RU. Because 1 RU mAbs, the IgG HL5B depicted no ␤2GPI-specific reactiv- resembles a surface mass change of 1 pg protein/mm2, ity, whereas the IgM JGG9 showed a strong binding these response levels correspond to 2.6–3.0 ng/mm2 of signal (Fig. 1B). By kinetic analyses of 4 JGG9 sensor- immobilized protein. With the SAM-based immobiliza- grams, representing concentrations of 1.0, 0.4, 0.2, and 0.1 tion strategy, we obtained significantly higher sensor nmol/L, the apparent KD1 and KD values were calculated signals for ␤2GPI binding of the APLs than with the as 0.71 nmol/L (␹2 ϭ 3.5) and 0.13 nmol/L (␹2 ϭ 4.4), commonly used CMD matrix. Moreover, the SAM-cov- respectively. ered gold surface was characterized by extraordinarily high stability attributable to reorganization of alkyl chains biosensor analyses of patient sera to a semicrystalline structure (21). Reproducibility was Whereas sera of healthy donors showed no ␤2GPI-specific assessed on 3 separate biosensor chips by making multi- activity, for sera of syphilis patients under low-salt con- ple injections of 1 APS serum with ␤2GPI IgG ELISA titers ditions a low-to-moderate degree of binding was ob- of Ͼ100 kU/L. After 50 serum injections and rigid regen- served. This binding was completely blocked by the erations to remove the surface-bound antibodies, the addition of 300 mmol/L NaCl to the HEPES-based run- mean loss of ␤2GPI-specific activity obtained with the ning and dilution buffers. reference serum was at most 8%. The maximum binding signals after injection of diluted sera of healthy donors under appropriate experimental binding of control antibodies to the ␤2gpi- conditions (300 mmol/L NaCl) was as small as ϳ35 RU, coated surface whereas sera of anti-␤2GPI ELISA-positive APS patients We first tested the functionality of the antigen-coated showed values of 50–550 RU (Fig. 2A). surface with the mouse IgG1 monoclonal antibody (mAb) By calculating the mean RU value plus 3 SDs for the 5F10.F3 raised against purified human ␤2GPI, and the negative control sera, we obtained a cutoff value of 51 RU. mAbs HL5B and JGG9 originally derived from APS By use of this cutoff, positive and negative results for patients with ELISA-defined specificities against CL and specimen anti-␤2GPI activity were the same as those ␤2GPI (16, 22). 5F10.F3 bound specifically to ␤2GPI (Fig. measured with ELISA (see Table in the online Data 1A), with no reactivity against transferrin on the reference Supplement). On the basis of the biosensor and ELISA cell. In contrast, the human IgG1␬ control antibody, data presented in the Table in the online Data Supple- injected at the same concentration as 5F10.F3, showed no ment, the Spearman rank coefficients between the ampli- reactivity on either the transferrin- or the ␤2GPI-coated tudes of the APL biosensor responses and the sum of the biosensor surface. We determined the value of the appar- respective IgG and IgM anti-␤2GPI titers assessed with the 2 ELISA were calculated. In both cases the statistical test gave a coefficient r of 0.87 with P Ͻ0.0001. All sera of SLE patients without secondary APS and of patients with positive VDRL test or parvovirus B19 infection were negative for ␤2GPI-specific APLs (Fig. 2B).

kinetic analyses Because of the polyclonal nature of the APL response, sensorgrams recorded for diluted patient serum samples ␤ Fig. 1. Specificity analysis of the ␤2GPI-coated SPR biosensor surface are the sums of multiple antibody- 2GPI interactions. As with a panel of mAbs. a consequence, equilibrium and kinetic rate constants ␤ (A), binding profiles of the mouse antihuman 2GPI IgG1 F510.F3 (black) and a cannot be determined by nonlinear regression analysis. human myeloma-derived IgG␬ (gray) as irrelevant control. (B), binding profiles of Therefore, we calculated the relative binding affinity of the anti-␤2GPI ELISA-positive human APL JGG9 (black) and the anti-␤2GPI ␤ ELISA-negative human APL HL5B (gray). Each antibody was applied at a APLs to 2GPI by determining the amplitude of the SPR concentration of 25 mg/L. signal after 270 s of sample injection and the percentage 1140 Metzger et al.: Differentiation of Anti-␤2GPI

Fig. 2. SPR biosensor detection of anti-␤2GPI in serum. (A), APS patients (black lines) and healthy donors (gray lines). (B), non-APS patient groups tested positive in the VDRL test for syphilis (top), suffering from SLE without secondary APS (middle), or affected by parvovirus B19 infection (bottom).

decrease of the maximum binding signal after 300 s of molecular mass and the kinetic rate constants of the dissociation. Whereas the first value serves as an indicator ligand interaction. of the amount of antibody-antigen complex formation, the latter reflects mainly complex stability. A representative Discussion sensorgram, demonstrating the selection criteria of these 2 ELISA detection of anti-␤2GPI is considered to be more SPR-derived binding variables, is presented in Fig. 3A. specific for the diagnosis of APS than that of antiphos- We analyzed a series of 30 patient sera on 3 separate pholipid autoantibodies, including CL (23, 24). Some ␤2GPI biosensor chips and calculated the mean (SD) value studies, however, have indicated that antibodies against of both SPR-derived binding variables for each serum ␤2GPI can also be found in patients with parvovirus B19, sample. APS sera were divided into subgroups on the leptospirosis, syphilis, and other infections (25–27). Be- basis of their anti-␤2GPI IgG and IgM ELISA titer (Fig. cause these patients did not show clinical signs of APS, 3B). As shown in the association/dissociation plots in Fig. the specificity of anti-␤2GPI detection by ELISA is ques- 3C, APS sera with dissociation values of 40%–80% are tionable. Therefore, our aim was to evaluate ␤2GPI bind- characterized by high IgG and low to moderate IgM titers, ing of APS- and infection-associated anti-␤2GPI in an SPR whereas those with 10%–30% of dissociation contain high biosensor system. As demonstrated, our biosensor assay titers of ␤2GPI-specific IgM. Overall, it is evident that the allows reliable detection of ␤2GPI-reactive APLs in sera of position of a ␤2GPI-positive serum along the x axis in the APS patients. No antibody purification steps are required scatter diagram depends on the ratio of IgG and IgM before analysis. The excellent specificity of the biosensor isotype distribution. assay for APS-associated anti-␤2GPI was demonstrated Next, we investigated whether these differences in by the complete absence of ␤2GPI-specific responses in affinity and avidity between the 2 antibody isotypes are the control serum samples from patients with SLE with- also detectable in APS sera depicting increased titers of out secondary APS, syphilis, or parvovirus B19 infection. both anti-␤2GPI IgG and IgM. For separate examination, In accordance with ELISA, we observed that high the 2 antibody isotypes were selectively removed from 3 ␤2GPI surface densities are required for the sensitive SPR APS sera by preincubation with antihuman IgG- or IgM- detection of anti-␤2GPI in serum of APS patients. With specific agarose beads. As shown in Table 1, the original this finding it became clear that the analytical quality of dissociation profile of all 3 sera could be resolved into an the biosensor strongly depended on the composition of IgG-mediated fast and IgM-mediated slow component. the functionalized biolayer attached to the gold surface. Values for the SPR response and percentage of dissocia- By application of an SAM rather than a CMD hydrogel tion in Table 1 were derived from the SPR sensorgrams matrix, we achieved high ␤2GPI immobilization rates. presented in the Figure in the online Data Supplement. The assembly of ␤2GPI onto the planar and densely No correlation was found between the amount of IgG and packed rigid monolayer may also promote increased IgM binding to biosensor-immobilized ␤2GPI and the access by APS-associated anti-␤2GPI to their antigenic corresponding ELISA titers, which can be explained by sites. the fact that the amplitude of the SPR signal depends not We tested the functionality of the biosensor with only on the concentration of the analyte but also on its antibody controls of known specificities, including the Clinical Chemistry 53, No. 6, 2007 1141

Fig. 3. Representative sensorgram depicting the SPR binding variables.

(A), association levels are defined by the SPR response after 270 s of serum injection (response at t270s), whereas the amount of dissociation is expressed as percentage decrease of the SPR response at t270s after 300 s of buffer flow. (B), list of classification criteria for the division of APS sera into subgroups on the basis of anti-␤2GPI IgG and IgM antibody titers as determined by ELISA-B. Each group is marked with a unique symbol for better identification in the subsequent comparison of anti-␤2GPI IgG and IgM distribution and SPR binding variables. (C), association/dissociation plot with values derived from sensorgrams. The SPR binding variables are plotted as described in A, with the response at t270s on the y axis and the percentage of dissociation on the x axis. Data points are labeled according to the classification scheme presented in B and represent the mean of serum injections over 3 biosensor chips on which the same amount of ␤2GPI was immobilized. SDs are displayed as gray lines. Because the percentage of dissociation (reflecting the stability of the antibody-␤2GPI complex) is plotted in reverse order, serum samples with a fast dissociation from the ␤2GPI biosensor surface are located on the left, and those with a slow dissociation are located on the right. Sera with negative ELISA results are clustered together in the bottom right corner of the plot. As indicated by the different symbols, APS sera with dissociation values of 40%–80% are characterized by high IgG and low to moderate IgM titers, whereas those with 10%–30% dissociation contain high titers of ␤2GPI-specific IgM. monoclonal APLs JGG9 and HL5B. As expected from the describe binding affinity and avidity. We found that SPR ELISA measurements, JGG9 demonstrated strong ␤2GPI association levels correlate well with anti-␤2GPI ELISA binding, whereas HL5B showed no significant binding. A titers, whereas the degree of dissociation provided addi- kinetic evaluation of the sensorgrams resulted in the tional information concerning the binding strength of the ␤ ␤ determination of KD1 and KD values for the JGG9- 2GPI overall 2GPI-directed APL response. As a consequence, interaction of 0.71 and 0.13 nmol/L. By comparison with simultaneous acquisition of the amount and stability of the mouse antihuman mAb 5F10.F3, it became clear that immune complexes on a ␤2GPI-coated biosensor surface JGG9 can bind to ␤2GPI as potently as high-affinity provided information on the binding properties of APLs anti-␤2GPI IgG antibodies. Although the high affinity of that is not accessible through conventional ELISAs. In the JGG9 mAb can be in part explained by the occurrence future studies we will evaluate whether the availability of of somatic hypermutations (16), the unexpectedly low this information will allow a more accurate estimation of influence of avidity on the overall dissociation rate (indi- the pathogenicity of APLs in serum of APS patients. cated by a KD1:KD ratio of 5.46 between the values) may The pathogenic potential of autoantibodies has been arise from the inability of the biosensor to differentiate suggested to depend mainly on high-affinity and -avidity between mono- and multivalent binding if kinetics are binding to self-antigens. Evidence has recently accumu- very fast. Additional SPR measurements with Fab-frag- lated that effector functions associated with the Fc regions ments derived from the mAbs by papain digestion will be of the different isotypes are more likely to play the major required to elucidate this process. role (28). Because they function in an early stage of the Because the concentrations of ␤2GPI-reactive APLs in immune response, IgMs are usually polyreactive and of sera needed for the estimation of affinity constants are low affinity. Low antigen affinity, however, is compen- unknown, we had to use total response levels and the sated for by the pentameric structure of IgM, which percentage of dissociation after a defined period of time to promotes a relatively high avidity. Because the amount of 1142 Metzger et al.: Differentiation of Anti-␤2GPI

Grant/funding support: Stiftung fu¨ r Pathobiochemie und Table 1. Differentiation of the overall anti-␤2GPI Molekulare Diagnostik of the Deutsche Vereinte Gesellschaft dissociation profile into its IgG-mediated fast and fu¨ r Klinische Chemie und Laboratoriumsmedizin. IgM-mediated slow component as determined for 3 APS Financial disclosures: The authors declare that they have no patients with both increased anti-␤2GPI IgG and competing financial interests. IgM titers.a Acknowledgements: We thank Anita Schreiegg for excellent ELISA-B, kU/L SPR binding variables technical assistance. APS Response Dissociation, serum IgG IgM Isotype at t ,RU % 270s References Ͼ Ͼ ϩ 2 100 100 IgG IgM 447.2 33.6 1. Bertolaccini M, Khamashta MA. Laboratory diagnosis and man- IgG 283.6 59.7 agement challenges in the antiphospholipid syndrome. Lupus IgM 392.9 21.7 2006;15:172–8. 9 65.7 21.0 IgG ϩ IgM 192.0 54.4 2. Greaves M, Cohen H, Machin SJ, Mackie I. Guidelines on the IgG 138.4 91.4 investigation and management of the antiphospholipid syndrome. IgM 131.8 24.1 Br J Haematol 2000;109:704–15. 8 54.9 17.0 IgG ϩ IgM 230.3 71.4 3. McNeil HP, Simpson RJ, Chesterman CN, Krilis SA. Anti-phospho- IgG 239.3 82.4 lipid antibodies are directed against a complex antigen that IgM 94.9 45.7 includes a lipid-binding inhibitor of coagulation: beta2-glycoprotein I (apolipoprotein H). Proc Natl Acad Sci U S A 1990;87:4120–4. a Selective removal of IgG or IgM was achieved by preincubation of the sera 4. Galli M, Comfurius P, Maassen C, Hemker HC, De Baets MH, van with agarose beads coupled to goat antihuman IgG- or IgM-specific antibodies. Breda-Vriesman PJ, et al. Anticardiolipin antibodies (ACA) directed ␤ Values are given for anti- 2GPI IgM and IgG ELISA titers as measured with not to cardiolipin but to a plasma protein cofactor. Lancet 1990; ELISA-B as well as the SPR values for association and dissociation of untreated, 335:1544–7. IgM- and IgG-depleted APS serum preparations. The corresponding sensorgrams 5. Matsuura E, Igarashi Y, Fujimoto M, Ichikawa K, Koike T. Anticar- are shown in the Figure in the online Data Supplement. diolipin cofactor(s) and differential diagnosis of autoimmune disease. Lancet 1990;336:177–8. 6. Matsuda J, Saitoh N, Gohchi K, Tsukamoto M. Distinguishing APL binding correlates well with the density of immobi- beta2-glycoprotein I dependent (systemic lupus erythematosus lized ␤2GPI on the biosensor surface, enhanced avidity type) and independent (syphilis type) anticardiolipin antibody with Tween-20. Br J Haematol 1993;85:799–802. seems a likely reason for the low dissociation rates ob- 7. Petrovas C, Vlachoyiannopoulos PG, Kordossis T, Moutsopoulos served in our biosensor system for APS patient sera with HM. Anti-phospholipid antibodies in HIV infection and SLE with or high anti-␤2GPI IgM titers. The capacity of IgM to medi- without anti-phospholipid syndrome: comparisons of phospholipid ate high-avidity binding, however, does not necessarily specificity, avidity and reactivity with beta2-GPI. J Autoimmun implicate association with cellular responses that contrib- 1999;13:347–55. ute to clinical manifestations of autoimmune diseases. 8. Arvieux J, Roussel B, Jacob MC, Colomb MG. Measurement of antiphospholipid antibodies by ELISA using beta2-glycoprotein as Indeed, some authors consider the prevalence of IgM- an antigen. J Immunol Methods 1991;143:223–9. specific autoantigen responses to be protective against 9. Martinuzzo ME, Forastiero RR, Carreras LO. Anti-beta2-glycopro- disease progression (29, 30). On the other hand, fast tein I antibodies: detection and association with thrombosis. Br J dissociation, typically found in sera of APS patients with Haematol 1995;89:397–402. high anti-␤2GPI IgG titers and low or completely absent 10. Sebastiani GD, Galeazzi M, Tincani A, Piette JC, Font J, Allegri F, IgM, strongly suggests the pathogenicity of ␤2GPI-reac- et al. Anticardiolipin and anti-beta2-GPI antibodies in a large tive IgG not being attributed to a high affinity of the series of European patients with systemic lupus erythematosus: prevalence and clinical associations: European Concerted Action variable region but instead to the effector functions of the on the Immunogenetics of SLE. Scand J Rheumatol 1999;28: Ig heavy chain constant regions. This explanation is in 344–51. accordance with a study by Fulpius et al. (31), who found 11. Bertolaccini ML, Atsumi T, Escudero-Contreras A, Khamashta MA, that immune complex–mediated vasculitis was induced Hughes GR. The value of IgA antiphospholipid testing for diagnosis of antiphospholipid (Hughes) syndrome in systemic lupus by low-affinity anti-IgG2a rheumatoid factor mAb. erythematosus. J Rheumatol 2001;28:2637–43. 12. Carmo-Pereira S, Bertolaccini ML, Escudero-Contreras A, Kha- In summary, the presented SPR biosensor analyses dem- ␤ mashta MA, Hughes GR. Value of IgA anticardiolipin and anti- onstrate that covalent attachment of 2GPI through a beta2-glycoprotein I antibody testing in patients with pregnancy covalent amide linkage to the SAM-coated sensor chip is morbidity. Ann Rheum Dis 2003;62:540–3. a suitable method for obtaining highly reproducible APL 13. Samarkos M, Davies KA, Gordon C, Loizou S. Clinical significance measurements in patient sera without noteworthy loss of of IgA anticardiolipin and anti-beta2-GP1 antibodies in patients activity after more than 50 measurement cycles. Our novel with systemic lupus erythematosus and primary antiphospholipid biosensor-based analytical assay system for ␤2GPI-depen- syndrome. Clin Rheumatol 2006;25:199–204. 14. Cucnik S, Kveder T, Krizaj I, Rozman B, Bozic B. High avidity dent APL detection in sera of APS patients offers relevant anti-beta2-glycoprotein I antibodies in patients with antiphospho- diagnostic advantages compared with conventional lipid syndrome. Ann Rheum Dis 2004;63:1478–82. ELISA formats and readily lends itself to adaptation to the 15. Zoghlami-Rintelen C, Vormittag R, Sailer T, Lehr S, Quehenberger serological evaluation of other autoimmune diseases. P, Rumpold H, et al. The presence of IgG antibodies against Clinical Chemistry 53, No. 6, 2007 1143

beta2-glycoprotein I predicts the risk of thrombosis in patients potential marker for clinical features of antiphospholipid antibody with lupus anticoagulant. J Thromb Haemost 2005;3:1160–5. syndrome in patients with systemic lupus erythematosus. J Rheu- 16. Buschmann C, Fischer C, Ochsenhirt V, Neukirch C, Lackner KJ, matol 1998;25:2131–4. von Landenberg P. Generation and characterization of three 25. Loizou S, Cazabon JK, Walport MJ, Tait D, So AK. Similarities of monoclonal IgM antiphospholipid antibodies recognizing different specificity and cofactor dependence in serum antiphospholipid phospholipid antigens. Ann N Y Acad Sci 2005;1051:240–54. antibodies from patients with human parvovirus B19 infection and 17. Bozic B, Cucnik S, Kveder T, Rozman B. Avidity of anti-beta-2- from those with systemic lupus erythematosus. Arthritis Rheum glycoprotein I antibodies. Autoimmun Rev 2005;4:303–8. 1997;40:103–8. 18. Guidance for sponsors institutional review boards clinical investi- 26. Santiago M, Martinelli R, Ko A, Reis EA, Fontes RD, Nascimento gators and FDA staff. OMB control no. 0910-0582, issued 09/ EG, et al. Anti-beta2 glycoprotein I and anticardiolipin antibodies 25/2006. http://www.fda.gov/cdrh/oivd/guidance/1588.html in leptospirosis, syphilis and Kala-azar. Clin Exp Rheumatol 2001; (accessed April 13, 2007). 19:425–30. 19. Wilson WA, Gharavi AE, Koike T, Lockshin MD, Branch DW, Piette 27. Frauenknecht K, Lackner K, von Landenberg P. Antiphospholipid JC, et al. International consensus statement on preliminary antibodies in pediatric patients with prolonged activated partial classification criteria for definite antiphospholipid syndrome. Ar- thromboplastin time during infection. Immunobiology 2005;210: thritis Rheum 1999;42:1309–11. 799–805. 20. Tan EM, Cohen AS, Fries JF, Masi AT, McShane DJ, Rothfield NF, 28. Azeredo da Silveira S, Kikuchi S, Fossati-Jimack L, Moll T, Saito T, et al. The 1982 revised criteria for the classification of systemic Verbeek JS, et al. Complement activation selectively potentiates lupus erythematosus. Arthritis Rheum 1982;25:1271–7. the pathogenicity of the IgG2b and IgG3 isotypes of a high affinity 21. Ulman A. Formation and structure of self-assembled monolayers. anti-erythrocyte autoantibody. J Exp Med 2002;195:665–72. Chem Rev 1996;96:1533–54. 29. Rieben R, Roos A, Muizert Y, Tinguely C, Gerritsen AF, Daha MR. 22. von Landenberg C, Lackner KJ, von Landenberg P, Lang B, Immunoglobulin M-enriched human intravenous immunoglobulin Schmitz G. Isolation and characterization of two human monoclo- prevents complement activation in vitro and in vivo in a rat model nal anti-phospholipid IgG from patients with autoimmune disease. of acute inflammation. Blood 1999;93:942–51. J Autoimmun 1999;13:215–23. 30. Ito MR, Terasaki S, Kondo E, Shiwaku H, Fukuoka Y, Nose M. 23. Cabiedes J, Cabral AR, Alarcon-Segovia D. Clinical manifestations Experimental lupus nephritis in severe combined immunodeficient of the antiphospholipid syndrome in patients with systemic lupus (SCID) mice: remodelling of the glomerular lesions by bystander erythematosus associate more strongly with anti-beta 2-glycopro- IgM antibodies. Clin Exp Immunol 2000;119:340–5. tein-I than with antiphospholipid antibodies. J Rheumatol 1995; 31. Fulpius T, Spertini F, Reininger L, Izui S. Immunoglobulin heavy 22:1899–906. chain constant region determines the pathogenicity and the 24. Sanfilippo SS, Khamashta MA, Atsumi T, Amengual O, Bertolac- antigen-binding activity of rheumatoid factor. Proc Natl Acad Sci cini ML, D’Cruz D, et al. Antibodies to beta2-glycoprotein I: a U S A 1993;90:2345–9. Clinical Chemistry 53:6 1144–1152 (2007) Nutrition

The Ϫ256TϾC Polymorphism in the Apolipoprotein A-II Gene Promoter Is Associated with Body Mass Index and Food Intake in the Genetics of Lipid Lowering Drugs and Diet Network Study

Dolores Corella,1,2 Donna K. Arnett,3 Michael Y. Tsai,4 Edmond K. Kabagambe,3 James M. Peacock,5 James E. Hixson,6 Robert J. Straka,7 Michael Province,8 Chao-Qiang Lai,1 Laurence D. Parnell,1 Ingrid Borecki,8 and Jose M. Ordovas1*

Background: Apolipoprotein A-II (APOA2) plays an Results: We observed recessive effects for this polymor- ambiguous role in lipid metabolism, obesity, and phism that were homogeneous by sex. Individuals ho- atherosclerosis. mozygous for the ؊265C allele had statistically higher Methods: We studied the association between a func- body mass index (BMI) than did carriers of the T allele. tional APOA2 promoter polymorphism (؊265T>C) and Consistently, after multivariate adjustment, the odds plasma lipids (fasting and postprandial), anthropometric ratio for obesity in CC individuals compared with T .(0.039 ؍ variables, and food intake in 514 men and 564 women who allele carriers was 1.70 (95% CI 1.02–2.80, P participated in the Genetics of Lipid Lowering Drugs and Interestingly, total energy intake in CC individuals was Diet Network (GOLDN) study. We obtained fasting and statistically higher [mean (SE) 9371 (497) vs 8456 (413) than in T allele carriers. Likewise, total [0.005 ؍ postprandial (after consuming a high-fat meal) measures. kJ/d, P We measured lipoprotein particle concentrations by pro- fat and protein intakes (expressed in grams per day) 0.002 ؍ ton nuclear magnetic resonance spectroscopy and esti- were statistically higher in CC individuals (P -respectively). After adjustment for en ,0.005 ؍ mated dietary intake by use of a validated questionnaire. and P ergy, percentage of carbohydrate intake was statistically lower in CC individuals. These associations remained statistically significant even after adjustment for BMI. 1 Nutrition and Genomics Laboratory, Jean Mayer-US Department of We found no associations with fasting lipids and only Agriculture Human Nutrition Research Center on Aging, Tufts University, some associations with HDL subfraction distribution in Boston, MA. 2 Genetic and Molecular Epidemiology Unit and CIBER Fisiopatologı´a de the postprandial state. -la Obesidad y Nutricio´n, School of Medicine, University of Valencia, Valencia, Conclusions: The ؊265T>C polymorphism is consis Spain. tently associated with food consumption and obesity, 3 Department of Epidemiology, School of Public Health, and Clinical suggesting a new role for APOA2 in regulating dietary Nutrition Research Center, University of Alabama at Birmingham, AL. 4 Laboratory of Medicine and Pathology, University of Minnesota, Minne- intake. apolis, MN. © 2007 American Association for Clinical Chemistry 5 Division of Epidemiology and Community Health, School of Public Health, University of Minnesota, MN. 9 6 Human Genetics Center, University of Texas Health Science Center, Hous- Apolipoprotein A-II (APOA2) is the 2nd most abundant ton, TX. protein of HDL particles, but its function remains largely 7 Experimental and Clinical Pharmacology Department, College of Phar- unknown (1). Although some initial studies reported an macy, University of Minnesota, Minneapolis, MN. 8 Division of Biostatistics, Washington University School of Medicine, St. Louis, MO. * Address correspondence to this author at: Nutrition and Genomics Laboratory, JM-USDA Human Nutrition Research Center on Aging at Tufts 9 Nonstandard abbreviations: APOA2, apolipoprotein A-II; CVD, cardio- University, 711 Washington St., Boston, MA 02111-1524. Fax 617-556-3211; vascular disease; GOLDN, Genetics of Lipid Lowering Drugs and Diet e-mail [email protected]. Network; BMI, body mass index; DHQ, diet history questionnaire; PUFA, Received December 19, 2006; accepted March 29, 2007. polyunsaturated fat; SATFAT, saturated fat; IDL, intermediate-density li- Previously published online at DOI: 10.1373/clinchem.2006.084863 poprotein; OR, odds ratio; MUFA, monounsaturated fat.

1144 Clinical Chemistry 53, No. 6, 2007 1145

inverse relationship between plasma APOA2 concentra- ters (Minneapolis, MN, and Salt Lake City, UT) (22). tions and cardiovascular disease (CVD) risk (2, 3), find- Nearly all individuals were of European ancestry. Exclu- ings of subsequent studies did not reveal significant sion criteria were age Ͻ18 years; fasting triglycerides associations or even suggest a proatherogenic role (4–6). Ͼ16.5 mmol/L; recent history (6 months) of myocardial Overexpression of murine APOA210 in mice resulted in infarction; history of liver, kidney, pancreas, or gall blad- greater development of aortic fatty streak lesions (7). der disease; history of malabsorption of nutrients; current There are marked differences between murine and human use of insulin; abnormal renal or hepatic function; and APOA2, however, and overexpression of human APOA2 pregnancy or nursing in women. Individuals who re- in mice does not exert effects similar to those of murine ported current use of hypolipidemic drugs or dietary APOA2 on atherosclerosis (8–10). Relatively few studies supplements known to influence lipids were required to have examined the association between APOA2 polymor- consult their physician for approval to discontinue these phisms and phenotypic traits (11–14). Moreover, few lipid-lowering agents for 4 weeks before study participa- genetic variants have been identified in the APOA2 gene tion. Written informed consent was obtained from each (15). Interestingly, a TϾC transition at position Ϫ265 (rs participant. The protocol was approved by the Institu- no. 5082) affecting element D of the APOA2 promoter has tional Review Boards at the University of Alabama, the been reported to be functional in 2 independent studies, University of Minnesota, the University of Utah, and both demonstrating an ϳ30% drop in basal transcription Tufts University. activity (16, 17). In one of these studies, the Ϫ265TϾC The initially estimated sample size for the GOLDN polymorphism was associated with waist circumference study was ϳ1200 individuals. Complete fasting and post- in men (16). Another study (18) reported an association prandial data were obtained from 1118 individuals. Indi- between this polymorphism and abdominal fat depots in viduals with inconsistent dietary data (total daily energy women. Although Castellani et al. (19) found increased intake outside the range of 800–5500 kcal in men or body weight in mice overexpressing murine APOA2, 600–4500 in women) were excluded, resulting in a final the mechanism by which APOA2 may influence body sample size of 1078 individuals (514 men and 564 weight is largely unknown. APOA2 is a member of the women). apolipoprotein multigene superfamily, which includes genes encoding soluble apolipoproteins (e.g., APOA1 and interventions and clinic visits APOA4) that share genomic structure and several func- Participants were asked to fast for Ͼ12 h and abstain from tions. Although all these apolipoprotein genes have been using alcohol for Ͼ24 h before visiting the clinic. We found to be related to obesity in at least one epidemiolog- measured weight with a beam balance, hip circumference ical study (20), only APOA4 has been subscribed in at maximal hip girth, and waist circumference at the regulation of food intake, acting as a satiety signal (21). umbilicus. Body mass index (BMI) was calculated as Therefore, we hypothesized that APOA2 may also be weight (kg)/height (m)2, and obesity was defined as BMI involved in regulating food intake. Overall, our aims were Ն30 kg/m2. We administered clinical and lifestyle ques- to study the effect of the Ϫ65TϾC polymorphism in the tionnaires and created an interviewer-administered, di- APOA2 gene on body weight, food intake, and plasma rect data entry system for the diet history questionnaire lipid concentrations (fasting and postprandial) in a North (DHQ) developed by the National Cancer Institute. American population. dietary intake Materials and Methods We estimated dietary intake by use of the DHQ, a food study participants and study design frequency questionnaire developed by staff at the Risk The study sample consisted of 514 men and 564 women Factor Monitoring and Methods Branch. It consists of 124 who participated in the Genetics of Lipid Lowering Drugs food items and includes both portion size and dietary and Diet Network (GOLDN) study. GOLDN is part of the supplement questions. Two studies have been conducted Program for Genetic Interactions Network and is funded to assess its validity (23, 24). The food list and nutrient by the NIH through the University of Alabama at Bir- database used with the DHQ are based on national mingham and in collaboration with the University of dietary data [US Department of Agriculture (USDA) Utah, Washington University, Tufts University, Univer- 1994–96 Continuing Survey of Food Intakes by Individu- sity of Texas, University of Michigan, University of Min- als, available from the USDA Food Surveys Research nesota, and Fairview-University of Minnesota Medical Group]. Center. The majority of participants were re-recruited from 3-generational pedigrees from 2 National Heart, postprandial study fat challenge Lung, and Blood Institute Family Heart Study field cen- The postprandial study fat challenge consisted of a meal formulated according to the protocol of Patsch et al. (25). The meal, which participants were instructed to consume 2 10 Human genes: APOA2, apolipoprotein A-II; HTR2A, 5-hydroxy- within 15 min, had 700 calories/m body surface area 2 tryptamine (serotonin) receptor 2A. (2.93 MJ/m body surface area); 3% of calories were 1146 Corella et al.: APOA2 Variation, Obesity, and Food Intake

derived from protein, 14% from carbohydrate, and 83% for family relationships. To evaluate replication, we per- from fat sources. Cholesterol content was 240 mg and the formed statistical analyses for the whole sample and for ratio of polyunsaturated fat (PUFA) to saturated fat men and women separately. We also tested the statistical (SATFAT) was 0.06. The average person consumed 175 homogeneity of the effects by sex in the corresponding mL heavy whipping cream (39.5% fat) and 7.5 mL pow- regression model with interaction terms. Obesity was dered, instant, nonfat dry milk, blended with ice and 15 defined as BMI Ն30 kg/m2. We fitted logistic regres- mL chocolate- or strawberry-flavored syrup to increase sion models to estimate the odds ratio (OR) and 95% CI palatability. We drew blood samples immediately before of obesity associated with the APOA2 polymorphism (time 0) and 3.5 and 6 h after the high-fat meal. and to control for the effect of covariates and family relationships. biochemical analyses We performed ANOVA for repeated measures to ana- We drew venous blood after study participants had fasted lyze data obtained in the postprandial study. In this overnight. Plasma samples were stored and analyzed analysis, we studied the statistical effects of the APOA2 together. We measured triglycerides by glycerol-blanked genotype alone, the effect of time (change in the variable enzymatic method on the Roche COBAS FARA centrifu- after the high-fat load over the entire lipemic period), and gal analyzer (Roche Diagnostics). We measured choles- the effect of the interaction of the 2 factors (genotype and terol on the Hitachi 911 Automatic Analyzer (Roche time), which is indicative of the magnitude of the post- Diagnostics) using a cholesterol esterase, cholesterol oxi- prandial response in each group of individuals. We also dase reaction (Chol R1; Roche Diagnostics). We used the carried out multivariate adjustment for covariates includ- same reaction to measure HDL cholesterol after precipi- ing demographic, clinical, lifestyle, and family relation- tation of non-HDL cholesterol with magnesium/dextran. ships. We used routine regression diagnostic procedures We measured LDL cholesterol by use of a homogeneous to ensure the appropriateness of the models. Statistical direct method (LDL Direct Liquid Select௢ Cholesterol analyses were done using SPSS 14.0 software (SPSS). All Reagent; Equal Diagnostics) on the Hitachi 911. We mea- reported probability tests were 2-sided. Differences be- sured lipoprotein particle concentrations and size by tween means were considered significant at P Ͻ0.05. proton nuclear magnetic resonance spectroscopy (26, 27). Data were obtained from the measured amplitudes of bioinformatics analysis their spectroscopically distinct lipid methyl group nuclear Analysis of the genomic DNA sequence segment centered magnetic resonance signals. We measured the concentra- on the Ϫ265 variant was conducted with MAPPER (28) to tions of the following subclasses: small LDL (diameter identify potential transcription factor binding sites. 18.0–21.2 nm), large LDL (21.2–23.0 nm), intermediate- density lipoprotein (IDL; 23.0–27.0 nm), large HDL (8.8– Results 13.0 nm), medium HDL (8.2–8.8 nm), small HDL (7.3– Characteristics of the study participants are shown in 8.2 nm), large VLDL (Ͼ60 nm), medium VLDL (35.0– Table 1. APOA2 genotype frequencies did not deviate 60.0 nm), and small VLDL (27.0–35.0 nm). The small LDL from Hardy-Weinberg equilibrium expectations (P ϭ subclass encompassed both intermediate small (19.8– 0.941) and did not differ between men and women (P ϭ 21.2 nm) and very small (18.0–19.0 nm) particles. 0.670). We first examined the association between the APOA2 Ϫ265TϾC polymorphism and anthropometric dna isolation and apoa2 genotyping variables (Table 2). Recessive effects were observed, and DNA was isolated from blood samples using routine thus TT and CT individuals were grouped and compared DNA isolation sets (Qiagen). We performed genotyping with CC homozygotes. Men and women were analyzed of the Ϫ265CϾT polymorphism (rs5082) using a Taqman together because homogeneity of the effect was observed assay with allele-specific probes on the ABIPrism 7900HT by sex (P for interaction APOA2 genotype*gender Ͼ 0.05 Sequence Detection System (Applied Biosystems) accord- for all the anthropometric variables examined). In the ing to routine laboratory protocols. crude analysis, we detected statistically significant mean differences in weight, BMI, and waist and hip circumfer- statistical analysis ences between CC homozygotes and carriers of the com- Triglyceride concentrations were log transformed and mon allele. These differences remained statistically signif- IDL and VLDL were square-root transformed for statisti- icant even after multivariate adjustment for age, sex, cal testing. We used Pearson ␹2 and Fisher tests to test tobacco smoking, alcohol consumption, diabetes, and differences in percentages. We applied ANOVA and the CVD. After additional adjustment for family relation- Student t-test to compare crude means. We tested codomi- ships, weight and BMI remained statistically significant. nant, dominant, and recessive models for the APOA2 Furthermore, when we estimated the effect of the APOA2 polymorphism. We carried out multivariate adjustments polymorphism on obesity risk, we found a consistent and of the associations by analysis of covariance and esti- statistically significant association following a recessive mated adjusted means. First, we adjusted analyses for pattern. Homogeneity of the effect was detected in both demographic, clinical, and lifestyle variables and then men and women. We estimated ORs and 95% CIs for CC Clinical Chemistry 53, No. 6, 2007 1147

ϭ Table 1. General characteristics of the study population.a with the CC genotype was 1.45 (95% CI 1.03–2.04, P 0.032). After adjustment for sex, age, tobacco smoking, Men Women n 514 564 alcohol consumption, diabetes, and CVD (model 2), the Age, years 49.1 (16.1) 48.1 (16.3) higher risk of obesity associated with the CC genotype Weight, kgb 90.5 (16.4) 75.9 (17.1) remained statistically significant (OR 1.47, 95% CI 1.03– Height, mb 1.78 (0.72) 1.65 (0.68) 2.09, P ϭ 0.034). After additional adjustment for the BMI, kg/m2 28.5 (4.9) 28.0 (6.2) previous variables plus family relationships (model 3), the Waist circumference, mb 1.00 (0.14) 0.92 (0.75) OR was 1.70 (95% CI 1.02–2.80, P ϭ 0.039), revealing a Hip circumference, mb 1.05 (0.09) 1.08 (0.14) strong consistency in the estimations. Cholesterol, mmol/L 4.91 (0.97) 4.96 (1.04) We next examined if the Ϫ265TϾC APOA2 polymor- b LDL cholesterol, mmol/L 3.19 (0.79) 3.10 (0.83) phism may relate to food intake in this population. A b HDL cholesterol, mmol/L 1.08 (0.25) 1.35 (0.36) recessive model of association was also found. Homoge- b Triglycerides, mmol/L 1.70 (1.25) 1.41 (0.93) neity of the associations was detected in both men and VLDL size, nm 51.25 (7.32) 51.17 (7.50) women; results for the overall sample are presented LDL size, nmb 20.48 (0.78) 21.10 (0.87) (Table 3). Homozygous individuals for the CC allele had HDL size, nmb 8.65 (0.38) 9.02 (0.44) Fasting glucose, mmol/Lb 5.84 (1.10) 5.43 (0.88) a statistically higher mean of energy intake than carriers Energy intake, KJ/dayb 9994 (3858) 7261 (2684) of the T allele, which remained statistically significant Total fat, g/dayb 97.2 (43.5) 68.1 (30.4) even after multivariate adjustment for sex, age, tobacco SATFAT, g/dayb 33.0 (15.9) 22.6 (10.8) smoking, alcohol consumption, diabetes, CVD, and famil- MUFA, g/dayb 36.9 (16.8) 25.3 (11.5) ial relationships. Thus, daily energy intake was ϳ900 PUFA, g/dayb 19.9 (9.5) 15.2 (7.3) KJ/d (200 Kcal/d) higher in CC homozygotes than in T Proteins, g/dayb 94.4 (39.9) 68.1 (26.6) allele carriers (P ϭ 0.005). In terms of amount of food, CC Carbohydrates, g/dayb 279.8 (112.9) 218.4 (87.6) homozygotes had significantly higher absolute intake Current smokers, n (%) 39 (7.6) 42 (7.5) than T allele carriers, even after multivariate adjustment. a Past smokers, n (%) 135 (26.3) 100 (17.8) Total fat presented strong associations with the APOA2 Current drinkers, n (%) 254 (49.4) 291 (51.6) polymorphism. Means of total fat intake (g/day) in men Diabetes or high blood sugar, 34 (6.6) 52 (9.2) n (%) (Fig. 1A) and women (Fig. 1B) depending on the APOA2 Heart attack, n (%)b 25 (4.7) 5 (0.9) polymorphism exemplify the recessive genetic effects Stroke, n (%) 5 (1.0) 3 (0.5) and the internal replication of results. Moreover, when Obesity, n (%) 167 (32.5) 192 (34.0) SATFAT, monounsaturated fat (MUFA), and PUFA as APOA2 Ϫ265TϾC well as specific fatty acids were analyzed, significant polymorphism, n (%) differences were found even after multivariate adjust- TT 188 (36.6) 213 (37.8) ment. When macronutrients were expressed as percent- TC 251 (48.8) 261 (46.3) age of daily energy intake, carbohydrate intake was CC 75 (14.6) 90 (16.0) statistically lower in CC homozygotes than in T allele a Data are mean (SD) except where noted. carriers [mean (SE) 46.8% (1.7%) in TT ϩ CT vs 44.9% b Statistically significant differences between men and women. (1.9%) in CC individuals, P ϭ 0.012, in the multivariate adjusted model including sex, age, tobacco smoking, homozygotes compared with carriers of the Ϫ265T allele, alcohol consumption, diabetes, CVD, and family relation- when each indicated variable was included into the ships]. Likewise, after this multivariate adjustment, total logistic regression models (models 1 to 3). Model 1 was fat intake expressed as percentage of daily energy intake adjusted only for sex, and the OR for obesity associated was statistically higher in CC homozygotes than in T

Table 2. Association between the APOA2 ؊265T>C polymorphism and anthropometric variables in men and women.a (165 ؍ CC (n (913 ؍ TT ؉ TC (n (165 ؍ CC (n (913 ؍ TT ؉ TC (n Crude values Multivariate adjusted means Mean (SD) Mean (SD) P Mean (SE) Mean (SE) PPb Weight, kg 82.2 (18.0) 86.1 (19.2) 0.012 87.9 (3.3) 92.14 (3.5) 0.003 0.036 Height, m 1.71 (0.01) 1.71 (0.01) 0.551 1.71 (0.01) 1.72 (0.01) 0.067 0.534 BMI, kg/m2 28.0 (5.5) 29.2 (6.2) 0.016 29.9 (1.1) 30.9 (1.2) 0.019 0.044 Waist, m 0.95 (0.16) 0.99 (0.17) 0.009 1.01 (0.03) 1.04 (0.03) 0.005 0.057 Hip, m 1.07 (0.12) 1.09 (0.12) 0.010 1.10 (0.02) 1.12 (0.02) 0.013 0.160 a Data are unadjusted and multivariate adjusted means. Estimated means and P values were adjusted for sex, age, tobacco smoking, alcohol consumption, diabetes, and CVD. b Additionally adjusted for family relationships. 1148 Corella et al.: APOA2 Variation, Obesity, and Food Intake

Table 3. Association between the APOA2 ؊265T>C polymorphism and dietary intake in men and women.a (165 ؍ CC (n (913 ؍ TT ؉ TC (n (165 ؍ CC (n (913 ؍ TT ؉ TC (n Crude values Multivariate adjusted means Mean (SD) Mean (SD) P Mean (SE) Mean (SE) P b Total energy, KJ/day 8447 (3456) 9208 (4067) 0.012 8456 (413) 9371 (497) 0.005 Total fat, g/day 80.6 (39.8) 89.8 (44.2) 0.007 78.1 (4.7) 89.4 (5.6) 0.002 SATFAT, g/day 27.2 (15.9) 29.9 (15.8) 0.025 25.7 (1.7) 29.2 (2.0) 0.008 MUFA, g/day 30.2 (15.1) 34.1 (17.1) 0.004 29.7 (1.8) 34.3 (2.2) 0.001 PUFA, g/day 17.1 (8.5) 19.2 (9.7) 0.005 16.8 (1.1) 19.2 (1.3) 0.004 Protein, g/day 79.5 (35.2) 87.0 (40.1) 0.013 80.2 (4.2) 89.5 (5.0) 0.005 Carbohydrate, g/day 245.6 (101.2) 259.3 (123.3) 0.121 242.6 (12.5) 261.0 (15.1) 0.062 Dietary fiber, g/day 18.6 (8.1) 19.4 (9.2) 0.221 18.6 (1.0) 19.7 (1.3) 0.154 Fatty acids, g/day 16:0 14.6 (7.4) 16.2 (8.4) 0.016 14.1 (0.9) 16.0 (1.1) 0.004 18:0 6.7 (3.4) 7.4 (3.9) 0.025 6.4 (0.4) 7.3 (0.5) 0.009 16:1 1.34 (0.74) 1.50 (0.85) 0.017 1.3 (0.1) 1.5 (0.1) 0.004 18:1 28.2 (14.05) 31.8 (15.9) 0.003 27.7 (1.7) 32.0 (2.0) 0.001 20:1 0.16 (0.11) 0.20 (0.13) Ͻ0.001 0.16 (0.01) 0.20 (0.02) 0.002 22:1 0.021 (0.019) 0.025 (0.020) 0.005 0.020 (0.002) 0.026 (0.003) 0.001 18:2 15.3 (7.7) 17.2 (8.8) 0.005 15.1 (0.9) 17.3 (1.2) 0.004 18:3 1.50 (0.74) 1.64 (0.81) 0.026 1.4 (0.1) 1.6 (0.1) 0.025 20:4 0.11 (0.06) 0.12 (0.07) 0.015 0.11 (0.01) 0.13 (0.1) 0.012 22:5 0.010 (0.010) 0.013 (0.012) 0.006 0.010 (0.001) 0.013 (0.002) 0.006 22:6 0.053 (0.052) 0.061 (0.057) 0.076 0.047 (0.007) 0.059 (0.008) 0.034 Total trans fatty acids, g/day 4.87 (2.67) 5.29 (2.79) 0.072 4.85 (0.32) 5.50 (0.38) 0.010 Amount of food, g/day 2503 (1281) 2777 (1344) 0.012 2653 (148) 2905 (178) 0.031 a Data are unadjusted and multivariate adjusted means. Estimated means and b P values were adjusted for sex, age, tobacco smoking, alcohol consumption, diabetes, CVD, and family relationships.

allele carriers [37.2% (1.4%) in TT ϩ CT vs 38.6% (1.5%) in homozygotes over the entire postprandial period than in CC individuals; P ϭ 0.023]. carriers of the T allele (P ϭ 0.029) and reached its highest Taking into account that CC individuals had higher statistical significance (P ϭ 0.014) at 3.5 h after the fatty BMI, we performed additional adjustment for BMI of the meal. Conversely, the amount of medium-sized HDL associations presented in Table 3 (results not shown). This particles increased and was lower in CC homozygotes additional adjustment slightly modified the P values and over the entire period (P ϭ 0.049). The fat-loading test did not change the inference. increased postprandial large VLDL particles, which were We investigated the effect of the Ϫ265TϾC APOA2 lower in homozygous individuals for the Ϫ265C allele. polymorphism on fasting plasma lipids, lipoproteins, and These results suggested a faster clearance in CC homozy- particle size in both men and women. No heterogeneity gotes than in carriers of the T allele. by sex was found. Neither in the crude model nor in the The bioinformatic analysis of the APOA2 gene control multivariate-adjusted model (Table 4) were statistically (upstream) region indicates the potential for a CEBPA significant differences between APOA2 genotypes found. [CCAAT/enhancer binding protein (C/EBP) ␣] binding Considering that a high-fat intake increases APOA2 con- site that is not predicted with the minor C allele. centrations, a postprandial study of fat challenge was also performed in the 1078 participants. Blood samples were Discussion drawn before time 0 and at 3.5 and 6 h after consuming Since its discovery, APOA2 has been a protein in search of the high-fat meal. After multivariate adjustment, no sta- a function (29). In fact, it has often been stated that tistically significant genotype effects or genotype*time APOA2 has no major function in HDL metabolism interactions were observed for total cholesterol, LDL (1, 5, 14, 29). This view is now changing as results become cholesterol, HDL cholesterol, triglycerides, small LDL available of more recent findings proposing roles for this particles, large LDL particles, mean LDL particle size, apolipoprotein in insulin resistance, obesity, and even mean VLDL particle size, or mean HDL particle size cancer (16, 18, 19, 30, 31). In the current study, we found a (results not shown). Statistically significant genotype ef- strong association between the APOA2 Ϫ265TϾC poly- fects were observed for HDL and VLDL particles (Fig. morphism and anthropometric measures that was repli- 2A–F). The amount of small HDL particles, which de- cated in both men and women, adding evidence to creased after the fat load, was significantly higher in CC previous studies that reported a role for APOA2 in obesity Clinical Chemistry 53, No. 6, 2007 1149

135 Table 4. Association between the APOA2 ؊265T>C A) MEN 120 P =0.014 polymorphism and fasting plasma lipid and lipoprotein P*=0.007 a

) 105 concentrations and particle size.

y NS a ؉ d TT CT CC / 90 109

؍ ؍ g ( (n 913) (n 165) 95 95 e

k 75 Multivariate adjusted means a t

n Mean (SE) Mean (SE) P i

60 t a f

a 45 t Fasting glucose, mmol/L 6.50 (0.18) 6.48 (0.20) 0.827 o T 30 Total cholesterol, mmol/L 5.18 (0.19) 5.14 (0.20) 0.600 15 Triglycerides, mmol/L 2.06 (0.22) 1.83 (0.24) 0.205 LDL cholesterol, mmol/L 3.26 (0.16) 3.29 (0.17) 0.630 0 HDL cholesterol, mmol/L 1.16 (0.06) 1.17 (0.06) 0.827 TT (n=188) TC (n=251) CC (n=75) HDL particle size, nm 8.74 (0.09) 8.69 (0.09) 0.196 -265T>C APOA2 polymorphism LDL particle size, nm 20.65 (0.18) 20.65 (0.19) 0.924 90 VLDL particle size, nm 52.67 (1.78) 52.72 (1.91) 0.947 B) WOMEN P =0.044 Large HDL, g/L 0.16 (0.02) 0.16 (0.03) 0.860 75 P*=0.033 NS Medium HDL, g/L 0.04 (0.01) 0.03 (0.02) 0.117 ) y

a Small HDL, g/L 0.22 (0.01) 0.23 (0.01) 0.119

d 74 / 60 g 67 IDL, g/L 0.08 (0.01) 0.08 (0.02) 0.905

( 67

k Large LDL, g/L 0.44 (0.07) 0.46 (0.08) 0.544 a t 45 n

i Small LDL, g/L 0.67 (0.07) 0.65 (0.08) 0.683

t a

f Medium small LDL, g/L 0.27 (0.04) 0.27 (0.04) 0.842

a 30 t Very small LDL, g/L 0.39 (0.05) 0.38 (0.06) 0.476 o T Large VLDL, g/L 0.58 (0.10) 0.50 (0.13) 0.055 15 Medium VLDL, g/L 0.65 (0.11) 0.57 (0.13) 0.209 Small VLDL, g/L 0.17 (0.02) 0.17 (0.02) 0.691 0 a TT (n=213) TC (n=261) CC (n=90) Data are multivariate adjusted means. Estimated means and P values were adjusted for sex, age, tobacco smoking, alcohol consumption, diabetes, cardio- -265T>C APOA2 polymorphism vascular disease, cholesterol medication, family relationships, and BMI. Fig. 1. Total fat intake (g/day) in men (A) and women (B) depending on the Ϫ265TϾC polymorphism in the APOA2 gene promoter. Crude means and P values are displayed. Multivariate P values were adjusted for age, tobacco smoking, alcohol consumption, diabetes, CVD, and family relation- receptor 2A] gene. Herbeth et al. (36) found that the ships. Error bars: SE. Ϫ1438G/A polymorphism in the HTR2A gene was significantly associated with lower energy, total fat, MUFA, and SATFAT intakes in children and adolescents from the (16, 18). Moreover, we observed for the first time a Stanislas Family Study. They also found homogeneity by consistent association between the APOA2 Ϫ265TϾC sex. These results were consistent with the findings of a polymorphism and food consumption that also showed previous study of overweight individuals from France internal replication, suggesting a potential new role of (34). These genotypic effects can be easily explained by APOA2 in the regulation of human appetite. In both men the relevant role of the serotonergic system as a deter- and women, we found a recessive effect of this polymor- minant of food intake. In the case of APOA2, however, phism. Thus, homozygous individuals for the variant no previous studies show a role of this protein in deter- allele C had significantly higher daily energy intake than mining food intake—only a recent report linking ethanol T allele carriers. When other variables (sex, age, tobacco self-administration in monkeys with higher serum smoking, alcohol consumption, diabetes, CVD, and fam- APOA2 concentrations (37). Collaku et al. (38),ina ily relationships) were progressively added to the crude genomic scan in white sibling pairs, identified several model, only slight variation of the initially estimated P chromosomal regions that contribute to variations in food values was observed, revealing an independent signifi- intake. The strongest evidence of linkage for dietary cant association of this polymorphism. In analyzing ma- energy and fat intake was found on chromosome 1p21.2 cronutrient intake, we found that CC homozygotes had (P Ͻ0.001). Although the authors did not mention APOA2 higher intake of fat and proteins than T allele carriers. as a potential candidate gene in explaining such results, Although some twin and family studies have documented the APOA2 gene is in this locus. the genetic component of total energy and macronutrient Despite the scarcity of previous data supporting a role intake (32, 33), only a few reports (34–36) have shown of APOA2 in regulating food intake, copious experimen- that a gene polymorphism may partly determine energy tal evidence demonstrates a pivotal role of another apo- or macronutrient intake in humans. These studies have lipoprotein, APOA4, as a satiety signal (20, 21, 39, 40). focused on the HTR2A [5-hydroxytryptamine (serotonin) Fujimoto et al. (39) were the first to report that APOA4 is 1150 Corella et al.: APOA2 Variation, Obesity, and Food Intake

0.18 1.0 phism and anthropometric measures, with controversial A D 0.17 P time = 0.001 0.9 results (16, 18). In this American population, we have P genotype = 0.913 TT+TC P interaction = 0.896 found that in both men and women, homozygous indi- ) 0.16 ) 0.8 L L / / Ϫ g g *P = 0.049 viduals for the 265C allele had higher obesity risk than ( (

L L CC D D 0.15 0.7 L carriers of the T allele. Our findings are in agreement with

H CC V e e g r g

r Lara-Castro et al. (18), who reported statistically higher a TT+TC

0.14 a L 0.6 L P time = 0.016 amounts of visceral adipose tissue in white women carri- P genotype = 0.068 0.13 0.5 P interaction = 0.350 ers of the Ϫ265C allele. Neither of these studies examined Ϫ Ͼ 0 0 the influence of the 265T C polymorphism in deter- 0 3.5 6 1 3.5 6 mining food intake. Van’t Hooft et al. (16) studied the Time (h) Time (h) association between this polymorphism and fasting

0.8 0.12 plasma glucose, insulin, lipids, and lipoproteins and B E found no significant associations, in agreement with our

0.10 TT+TC results. Moreover, they also carried out an oral fat toler- ) )

L 0.7 / L /

g ance test and found that the overall response of plasma ( g *P = 0.047

(

0.08 L

L CC D

D triglycerides during the entire 6-h period was unassoci- L H TT+TC CC V

m Ϫ Ͼ m

u ated with the APOA2 265T C polymorphism. Interest- u i

0.06 i d d

e 0.6 P time = 0.001 e M

M ingly, postprandial APOB-100 concentrations in the Sf P genotype = 0.049 P time = 0.001 0.04 P interaction = 0.575 Ͼ P genotype = 0.442 60 of triglyceride-rich protein (a measure of large VLDL P interaction = 0.031 particles) were significantly lower in individuals homozy- 0 0 0 3.5 6 1 3.5 6 gous for the Ϫ265C allele, suggesting that the reduced

Time (h) Time (h) plasma APOA2 concentrations associated with the C

0.24 0.18 allele enhances the ability to remove large VLDL from C F 0.23 circulation during alimentary lipemia. In our study, we P time = 0.001 0.16 P genotype = 0.029 measured large VLDL particles and obtained results in 0.22 P interaction = 0.441 ) ) 0.14 L L P time = 0.255 / CC / agreement with these observations. In addition, we mea- g g ( ( P genotype = 0.138

0.21 L *P = 0.014 L P interaction = 0.205 sured HDL particle size in the postprandial period and D D 0.12 L H l V l 0.20 l a l observed a significant increase in the small HDL subfrac- a m m S 0.10 TT+TC Ϫ 0.19 S tion associated with the 265C allele. TT+TC

0.18 0.08 CC In conclusion, in this population study the Ϫ265TϾC 0 0 0 3.5 6 0 3.5 6 polymorphism has no significant role in determining Time (h) Time (h) fasting plasma lipids and particle size. Postprandial state Fig. 2. Postprandial concentrations of large HDL (A), intermediate HDL increases the effects of this polymorphism, and significant (B), small HDL (C), large VLDL (D), intermediate VLDL (E), and small differences were observed in HDL subfraction distribu- VLDL (F) after a high-fat meal depending on the Ϫ265TϾC polymor- tion as well as borderline effects on large VLDL particles. phism in both men and women (at 0, 3.5, and 6 h after the oral fat However, the most consistent effects of this polymor- tolerance test). phism were observed on anthropometric variables and Multivariate adjusted means were estimated from a repeated-measures ANOVA Ϫ model. Models were adjusted for sex, age, BMI, tobacco smoking, alcohol food intake. Homozygous individuals for the 265C consumption, diabetes, CVD, medications, and family relationships. Multivariate allele had higher obesity risk than carriers of the T allele. P values for time, genotype, and the interaction genotype*time were estimated. *Statistically significant P values at every specific time point. n ϭ 913 TT ϩ TC CC individuals also had statistically higher total energy and 165 CC. and fat intake. As these effects were consistently found in both men and women, we suggest a new role for APOA2 in regulating food intake. More population studies are a satiety factor secreted by the intestine after fat absorp- needed to replicate these findings. tion and that this function of APOA4 is not shared by gut APOA1. Unfortunately, Fujimoto et al. (39) did not test the effects of APOA2, and its potential effects remain to be elucidated. If APOA2 acts as a satiety signal, taking into Grant/funding support: This work was supported by Na- account that the Ϫ265TϾC polymorphism results in lower tional Heart, Lung, and Blood Institute Grant U 01 HL72524, APOA2 concentrations, individuals homozygous for Genetic and Environmental Determinants of Triglycerides Ϫ265C can have a higher food intake than carriers of the Grant HL-54776, by contracts 53-K06-5-10 and 58-1950-9-001 T allele. A direct consequence of the involvement of from the US Department of Agriculture Research Service, the APOA2 in regulating food intake is its influence in body Ministerio de Educacio´n (PR2006-0258), and CIBER CB06/ weight regulation. Only 2 population studies have exam- 03/0035 from the Instituto de Salud Carlos III, Spain. ined the association between the Ϫ265TϾC polymor- Financial disclosures: None declared. Clinical Chemistry 53, No. 6, 2007 1151

References tion and a variant in the apolipoprotein A-II promoter: molecular 1. Blanco-Vaca F, Escola-Gil JC, Martin-Campos JM, Julve J. Role of study in a 1135-member familial hypercholesterolemia kindred. apoA-II in lipid metabolism and atherosclerosis: advances in the J Hum Genet 2002;47:656–64. study of an enigmatic protein. J Lipid Res 2001;42:1727–39. 18. Lara-Castro C, Hunter GR, Lovejoy JC, Gower BA, Fernandez JR. 2. Fager G, Wiklund O, Olofsson SO, Wilhelmsen L, Bondjers G. Apolipoprotein A-II polymorphism and visceral adiposity in African- Multivariate analyses of serum apolipoproteins and risk factors in American and white women. Obes Res 2005;13:507–12. relation to acute myocardial infarction. Arteriosclerosis 1981;1: 19. Castellani LW, Goto AM, Lusis AJ. Studies with apolipoprotein A-II 273–9. transgenic mice indicate a role for HDLs in adiposity and insulin 3. Buring JE, O’Connor GT, Goldhaber SZ, Rosner B, Herbert PN, resistance. Diabetes 2001;50:643–51. Blum CB, et al. Decreased HDL2 and HDL3 cholesterol, Apo A-I 20. Rankinen T, Zuberi A, Chagnon YC, Weisnagel SJ, Argyropoulos G, and Apo A-II, and increased risk of myocardial infarction. Circula- Walts B, et al. The human obesity gene map: the 2005 update. tion 1992;85:22–9. Obesity (Silver Spring) 2006;14:529–644. 4. Ridker PM, Rifai N, Cook NR, Bradwin G, Buring JE. Non-HDL 21. Tso P, Liu M. Apolipoprotein A-IV, food intake, and obesity. Physiol cholesterol, apolipoproteins A-I and B100, standard lipid mea- Behav 2004;83:631–43. sures, lipid ratios, and CRP as risk factors for cardiovascular 22. Higgins M, Province M, Heiss G, Eckfeldt J, Ellison RC, Folsom AR, disease in women. JAMA 2005;294:326–33. et al. NHLBI Family Heart Study: objectives and design. Am J 5. Tailleux A, Duriez P, Fruchart JC, Clavey V. Apolipoprotein A-II, HDL Epidemiol 1996;143:1219–28. metabolism and atherosclerosis. Atherosclerosis 2002;164:1– 23. Thompson FE, Subar AF, Brown CC, Smith AF, Sharbaugh CO, Jobe 13. JB, et al. Cognitive research enhances accuracy of food frequency 6. Escola-Gil JC, Marzal-Casacuberta A, Julve-Gil J, Ishida BY, Or- questionnaire reports: results of an experimental validation study. donez-Llanos J, Chan L, et al. Human apolipoprotein A-II is a J Am Diet Assoc 2002;102:212–25. pro-atherogenic molecule when it is expressed in transgenic mice 24. Subar AF, Thompson FE, Kipnis V, Midthune D, Hurwitz P, McNutt at a level similar to that in humans: evidence of a potentially S, et al. Comparative validation of the Block, Willett, and National relevant species-specific interaction with diet. J Lipid Res 1998; Cancer Institute food frequency questionnaires: the Eating at 39:457–62. America’s Table Study. Am J Epidemiol 2001;154:1089–99. 7. Warden CH, Daluiski A, Bu X, Purcell-Huynh DA, De Meester C, et 25. Patsch JR, Miesenbock G, Hopferwieser T, Muhlberger V, Knapp al. Evidence for linkage of the apolipoprotein A-II locus to plasma E, Dunn JK, et al. Relation of triglyceride metabolism and coronary apolipoprotein A-II and free fatty acid levels in mice and humans. artery disease: studies in the postprandial state. Arterioscler Proc Natl Acad Sci U S A 1993;90:10886–90. Thromb 1992;12:1336–45. 8. Schultz JR, Gong EL, McCall MR, Nichols AV, Clift SM, Rubin EM. 26. Otvos JD. Measurement of lipoprotein subclass profiles by Expression of human apolipoprotein A-II and its effect on high nuclear magnetic resonance spectroscopy. Clin Lab 2002;48: density lipoproteins in transgenic mice. J Biol Chem 1992;267: 171–80. 21630–6. 27. Tsai MY, Georgopoulos A, Otvos JD, Ordovas JM, Hanson NQ, 9. Castellani LW, Navab M, Van Lenten BJ, Hedrick CC, Hama SY, Peacock JM, et al. Comparison of ultracentrifugation and nuclear Goto AM, et al. Overexpression of apolipoprotein AII in transgenic magnetic resonance spectroscopy in the quantification of triglyc- mice converts high density lipoproteins to proinflammatory parti- eride-rich lipoproteins after an oral fat load. Clin Chem 2004;50: cles. J Clin Invest 1997;100:464–74. 1201–4. 10. Tailleux A, Bouly M, Luc G, Castro G, Caillaud JM, Hennuyer N, et 28. Marinescu V D, Kohane IS, Riva A. MAPPER: a search engine for al. Decreased susceptibility to diet-induced atherosclerosis in the computational identification of putative transcription factor human apolipoprotein A-II transgenic mice. Arterioscler Thromb binding sites in multiple genomes. BMC Bioinformatics 2005;6: Vasc Biol 2000;20:2453–8. 79. 11. Scott J, Knott TJ, Priestley LM, Robertson ME, Mann DV, Kostner G, et al. High-density lipoprotein composition is altered by a 29. Hedrick CC, Lusis AJ. Apolipoprotein A-II: a protein in search of a common DNA polymorphism adjacent to apoprotein AII gene in function. Can J Cardiol 1994;10:453–9. man. Lancet 1985;1:771–3. 30. Kalopissis AD, Pastier D, Chambaz J. Apolipoprotein A-II: beyond 12. Ferns GA, Shelley CS, Stocks J, Rees A, Paul H, Baralle F, et al. A genetic associations with lipid disorders and insulin resistance. DNA polymorphism of the apoprotein AII gene in hypertriglyceri- Curr Opin Lipidol 2003;14:165–72. daemia. Hum Genet 1986;74:302–6. 31. Malik G, Ward MD, Gupta SK, Trosset MW, Grizzle WE, Adam BL, 13. Vohl MC, Lamarche B, Bergeron J, Moorjani S, Prud’homme D, et al. Serum levels of an isoform of apolipoprotein A-II as a Nadeau A, et al. The MspI polymorphism of the apolipoprotein A-II potential marker for prostate cancer. Clin Cancer Res 2005;11: gene as a modulator of the dyslipidemic state found in visceral 1073–85. obesity. Atherosclerosis 1997;128:183–90. 32. Hur YM, Bouchard TJ Jr, Eckert E. Genetic and environmental 14. Martin-Campos JM, Escola-Gil JC, Ribas V, Blanco-Vaca F. Apoli- influences on self-reported diet: a reared-apart twin study. Physiol poprotein A-II, genetic variation on chromosome 1q21–q24, and Behav 1998;64:629–36. disease susceptibility. Curr Opin Lipidol 2004;15:247–53. 33. Tholin S, Rasmussen F, Tynelius P, Karlsson J. Genetic and 15. Fullerton SM, Clark AG, Weiss KM, Taylor SL, Stengard JH, environmental influences on eating behavior: the Swedish Young Salomaa V, et al. Sequence polymorphism at the human apoli- Male Twins Study. Am J Clin Nutr 2005;81:564–9. poprotein AII gene (APOA2): unexpected deficit of variation in an 34. Aubert R, Betoulle D, Herbeth B, Siest G, Fumeron F. 5-HT2A African-American sample. Hum Genet 2002;111:75–87. receptor gene polymorphism is associated with food and alcohol 16. van ’t Hooft FM, Ruotolo G, Boquist S, de Faire U, Eggertsen G, intake in obese people. Int J Obes Relat Metab Disord 2000;24: Hamsten A. Human evidence that the apolipoprotein A-II gene is 920–4. implicated in visceral fat accumulation and metabolism of triglyc- 35. Fumeron F, Betoulle D, Aubert R, Herbeth B, Siest G, Rigaud D. eride-rich lipoproteins. Circulation 2001;104:1223–8. Association of a functional 5-HT transporter gene polymorphism 17. Takada D, Emi M, Ezura Y, Nobe Y, Kawamura K, Iino Y, et al. with anorexia nervosa and food intake. Mol Psychiatry 2001;6:9– Interaction between the LDL-receptor gene bearing a novel muta- 10. 1152 Corella et al.: APOA2 Variation, Obesity, and Food Intake

36. Herbeth B, Aubry E, Fumeron F, Aubert R, Cailotto F, Siest G, et al. netics (HERITAGE) Family Study. Am J Clin Nutr 2004;79: Polymorphism of the 5-HT2A receptor gene and food intakes in 881–6. children and adolescents: the Stanislas Family Study. Am J Clin 39. Fujimoto K, Cardelli JA, Tso P. Increased apolipoprotein A-IV in rat Nutr 2005;82:467–70. mesenteric lymph after lipid meal acts as a physiological signal for 37. Freeman WM, Gooch RS, Lull ME, Worst TJ, Walker SJ, Xu AS, et satiation. Am J Physiol Gastrointest Liver Physiol 1992;262: al. Apo-AII is an elevated biomarker of chronic non-human primate G1002–6. ethanol self-administration. Alcohol Alcohol 2006;41:300–5. 40. Shen L, Ma LY, Qin XF, Jandacek R, Sakai R, Liu M. Diurnal 38. Collaku A, Rankinen T, Rice T, Leon AS, Rao DC, Skinner JS, et changes in intestinal apolipoprotein A-IV and its relation to food al. A genome-wide linkage scan for dietary energy and nutrient intake and corticosterone in rats. Am J Physiol Gastrointest Liver intakes: the Health, Risk Factors, Exercise Training, and Ge- Physiol 2005;288:G48–53. Technical Briefs

Aptamers Evolved from Cultured Cancer Cells Reveal of a same tumor type, but for most diseases no reliable Molecular Differences of Cancer Cells in Patient molecular probes are specific enough to recognize these Samples, Dihua Shangguan,1,3 Zehui Charles Cao,1,3 Ying subtle molecular differences (1). A critically important Li,2,3 and Weihong Tan1,3* (Departments of 1 Chemistry task for molecular medicine is to identify these differences and 2 Pathology and 3Shands Cancer Center, University of and then use them to further characterize and understand Florida, Gainesville, FL; * address correspondence to this the molecular basis of diseases. Despite the variety of author at: Department of Chemistry and Shands Cancer clinical variables used to classify human malignancies, Center, University of Florida, Gainesville, FL 32611; fax most neoplastic diseases cannot be defined or classified 352-846-2410, e-mail [email protected]) according to abnormal molecular disease processes because of the lack of molecular probes that can be used Background: Molecular-level differentiation of neo- define these processes. Thus, patients receiving similar plastic cells is essential for accurate and early diagnosis, diagnoses can have markedly different clinical courses but effective molecular probes for molecular analysis and responses to treatments (2). and profiling of neoplastic cells are not yet available. Currently, the diagnosis of leukemia is commonly We recently developed a cell-based SELEX (systematic based on morphologic evaluation supplemented by im- evolution of ligands by exponential enrichment) strat- munophenotype analysis by flow cytometry with mono- egy to generate aptamers (designer DNA/RNA probes) clonal antibodies of CD antigens (3). These antigens are usually expressed on both normal and neoplastic cells, as molecular probes to recognize neoplastic cells. however, and thus cannot accurately reflect the molecular Methods: We tested 6 cell-SELEX–generated aptamers features of the cancer cells. Although many antibodies are with equilibrium dissociation constants in the nanomo- available for phenotyping leukemia, they were not devel- lar to subnanomolar range: sgd5, selected from Toledo oped to enable comprehensive recognition of molecular cells, a human diffuse large-cell lymphoma cell line features of specific disease cells but were individually (B-cell), and sgc8, sgc3, sgc4, sgd2, and sgd3 from CCRF- developed at different times for various purposes. Sys- CEM cells, a human precursor T cell acute lymphoblas- tematic production of a panel of antibodies for molecular tic leukemia (T-ALL) cell line. Aptamers were labeled differentiation of cancer cells would be very difficult with fluorescein isothiocyanate fluorophores and then because of the technical difficulties involved in systematic used to recognize, by flow cytometric analysis, neoplas- development of antibodies for unknown surface biomar- tic cells in cultured hematopoietic cell lines and clinical kers. Novel approaches are therefore needed to systemat- samples. ically generate panels of new probes that recognize mo- Results: Aptamer sgd5 recognized only its target cells. lecular signatures of cancers. Aptamers sgc3, sgd2, sgd3, sgc4, and sgc8, selected from We previously developed panels of DNA aptamers, a T-cell leukemia cell line, identified all of the cultured single-stranded oligonucleotides, directly from live tumor T-cell leukemia cell lines with relatively high fluores- cells with a process called SELEX (systematic evolution of cence intensity. Aptamers sgc8, sgc3, and sgd3 showed ligands by exponential enrichment) (4–6). These aptam- good selectivity toward T-ALL cells and almost no ers recognized surface targets of cancer cells with high binding to normal hematopoietic cells or lymphoma and affinity and specificity. Selected aptamers can bind to myeloma cells. Selected aptamers also detected targets target molecules by folding into well-defined 3-dimensional structures (5–7). Unlike antibodies, aptamers, once on the cell membranes of neoplastic cells in patient the sequence is known, can be synthesized reproducibly samples. by a DNA synthesizer at a very low cost (8–13).In Conclusions: Aptamers selected against cultured neo- addition, aptamers have low molecular weight, fast tissue plastic cells can effectively be used as molecular probes penetration, and low toxicity. They can be specifically for recognition of neoplastic cells in patient samples. labeled with radioscopic, fluorescent, or other reporters Cell-based aptamer selection can be used to generate for molecular recognition. Moreover, aptamers are stable aptamer probes to obtain molecular signatures of neo- during long-term storage, can be transported at ambient plastic cells in patient samples. temperature, and sustain reversible denaturation. Despite © 2007 American Association for Clinical Chemistry their advantages and unique properties as molecular probes, however, aptamers have been used sparingly for Complex genetic and proteomic alterations reveal the medical applications because of the limited number of molecular heterogeneity within individual cancer diag- available aptamers that have medical relevance. Acquisi- nostic categories. Identification and understanding of the tion of aptamers directly from diseased cells is expected to molecular basis of diseased cells should provide the most link aptamers more closely to real medical problems and reliable approach toward effective diagnosis and treat- to greatly reduce the time gap between laboratory re- ments. Given the complexity and diversity of cancers, search and clinical applications. even within similar categories, multiple cancer-specific We report a group of new aptamers selected directly molecular probes are needed to delineate unique finger- from cancer cells (4) for the recognition of molecular prints of tumor cells. Molecular-level differences exist differences among leukemia patient samples. These between any 2 given types of cells, even 2 individual cases aptamers have high affinity and specificity for surface

Clinical Chemistry 53, No. 6, 2007 1153 1154 Technical Briefs

targets of neoplastic cells in clinical samples, and bind samples or lymphoma and myeloma cells. Further inspec- with clinical samples to form distinct recognition patterns. tion showed that aptamers sgc4 and sgd2 recognized Thus, these aptamers may be useful for both disease many different cell samples, including some normal bone diagnosis and efficient personalized therapy for individ- marrow cells, indicating the presence of common binding ual patients. entities on these cells. Combination of selected aptamers Aptamers are usually selected for single target mole- produced distinct patterns for different tumor cells, sug- cules. In contrast, cell-SELEX elects aptamers by use of gesting that aptamers may be used to define molecular complex whole cells as targets (4). A counterselection signatures of tumors. strategy is used to isolate aptamer sequences that interact Selected aptamers also detected targets on the cell only with the target cells and not with the control cells. membranes of neoplastic cells in patient samples, includ- Through this process, a group of cell-specific aptamers ing T-ALL, B-cell ALL, acute myeloid leukemia (AML), can be selected in a relatively short period (4–8 weeks) and lymphomas (Table 1). All lymphoma samples with even if it is not known which target molecules are present mature T or B cells showed no or very low binding (see on the cell surface and which membrane molecules might Table 1 in the online Data Supplement), indicating that play the most important role in cancer development. This recognition was not due to nonspecific interactions, in feature is the most important difference between cell- agreement with the fact that most of the aptamers were SELEX and other current methods of molecular probe selected against a cultured precursor T-ALL cell line. As development, and enables cell-SELEX to generate multi- expected, the aptamers showed more binding with cells ple molecular probes to recognize biomarkers in their from T-ALL patients than with other cell samples. native states, producing molecular signatures of diseases. Aptamer binding patterns corresponded well with gen- We used the cell-SELEX to obtain many aptamers with eral categories of acute leukemia predefined by antibody high affinity and specificity to surface molecules on target immunophenotyping. cancer cells. We then chose 6 aptamers with equilibrium Despite the results showing that aptamers can selec- dissociation constants in the nanomolar to subnanomolar tively recognize cultured T-ALL cells, demonstrating the range: sgd5—selected from Toledo cells, a human diffuse specificity of selected aptamers (Table 1), individual cases large-cell lymphoma cell line (B-cell)—and sgc3, sgd2, of clinical specimens may have quite different patterns sgd3, sgc4, and sgc8 from CCRF-CEM cells, a human T even in the same disease category. Although the explana- precursor T cell acute lymphoblastic leukemia (T-ALL) tion for these differences remains unknown, they pre- cell line. cisely reflect the complex nature of the disease. In addi- The 6 selected aptamers were first conjugated with tion to general categorization of the leukemia suggested fluorescein isothiocyanate (FITC) for recognition of differ- by available antibodies, our aptamer analyses provide ent kinds of cells. We then used flow cytometric analysis to monitor the binding of aptamers to cells from 4 T-cell leukemia cell lines; 8 B-cell lymphoma, leukemia, or myeloma cell lines; and normal human bone marrow aspirates (see the methods and materials in the Data Supplement that accompanies the online version of this Technical Brief at www.clinchem.org/content/vol53/ issue6). Subpopulations of bone marrow cells were identified in the flow cytometric analysis by their side-scatter properties and the expression of CD3, CD7, CD10, CD19, and CD45. The following cell types were identified: mature B cells, immature B cells, CD3(ϩ) T cells, mono- cytes, granulocytes, and nucleated erythrocytes. The dot plot of a typical flow cytometry analysis is shown in Fig. 1. The FITC-labeled DNA library was measured as the background binding, and a threshold was determined based on the background fluorescence. Cells recognized by aptamers were expressed as percentage of cells with fluorescence higher than the background threshold. The results (see Table 1 in the online Data Supplement) showed that aptamer sgd5 recognized only its target cells. All of the cultured T-cell leukemia cell lines were identified with relatively high fluorescence intensity by aptamers sgc3, sgd2, sgd3, sgc4, and sgc8, which was expected Fig. 1. Molecular recognition of T-ALL cells in patient bone marrow because they were selected from a T-cell leukemia cell aspirates with FITC-labeled sgc8’sgc3, sgc4, and PerCP-labeled anti- line. Aptamers sgc8, sgc3, and sgd3 showed good selec- CD45 antibody. tivity toward T-ALL cells and almost no binding to The background was measured by using FITC-labeled unselected library. The light normal hematopoietic cells in the human bone marrow gray dots represent T-ALL cells. Clinical Chemistry 53, No. 6, 2007 1155

direct evidence for the subtle molecular differences Grant/funding support: This work was supported by Na- among the same type of cancers. It is well known that tional Institutes of Health grants and a National Science responses to specific treatments may differ among dis- Foundation Nanotechnology Interdisciplinary Research eases of the same category (1, 2), but confirmation of Team grant. dissimilarities at the molecular level has been hindered by Financial disclosures: None declared. technical difficulties and lack of specific molecular probes. The cell-SELEX method (4) may provide a simple, fast, References 1. Espina V, Geho D, Mehta AI, Petricoin EF III, Liotta LA, Rosenblatt KP. and low-cost way to generate panels of molecular probes Pathology of the future: molecular profiling for targeted therapy. Cancer and reveal subtle differences even before specific disease Invest 2005;23:36–46. biomarkers are known. 2. Alizadeh AA, Eisen MB, Davis RE, Ma C, Lossos IS, Rosenwald A, et al. Distinct types of diffuse large B-cell lymphoma identified by gene expression In conclusion, with aptamers directly evolved from profiling. Nature 2000;403:503–11. T-ALL cells we were able to identify leukemia cells in 3. Belov L, dela Vega O, dos Remedios CG, Mulligan SP, Christopherson R. Immunophenotyping of leukemias using a cluster of differentiation antibody patient samples and detect subtle molecular differences microarray. Cancer Res 2001;61:4483–9. among individual samples from leukemia patients in the 4. Shangguan D, Li Y, Tang Z, Cao Z, Chen H, Mallikratchy P, et al. Aptamers same category. Our results demonstrate that cell-based evolved from live cells as effective molecular probes for cancer study. Proc Natl Acad Sci U S A 2006;103:11838–43. aptamer selection can be a valuable approach for gener- 5. Tuerk C, Gold L. Systematic evolution of ligands by exponential enrichment: ating aptamer probes to obtain molecular signatures of RNA ligands to bacteriophage T4 DNA polymerase. Science 1990;249:505– individual patient samples. Although the molecular pro- 10. 6. Ellington AD, Szostak JW. In vitro selection of RNA molecules that bind file or signature may not necessarily indicate the detailed specific ligands. Nature 1990;346:818–22. molecular mechanism of a disease, it may be the first step 7. Breaker RR. Natural and engineered nucleic acids as tools to explore biology. Nature 2004;432:838–45. toward understanding the molecular basis of a disease. 8. Yang CJ, Jockusch S, Vicens M, Turro N, Tan W. Light-switching excimer Because of features such as chemical-synthesis–based probes for rapid protein monitoring in complex biological fluids. Proc Natl production, low molecular weight, easy modification, and Acad Sci U S A 2005;102:17278–83. 9. Nutiu R, Li Y. In vitro selection of structure-switching signaling aptamers. long-term stability after modification, aptamers selected Angew Chem Int Ed Engl 2005;44:1061–5. from cancer cells may be effective molecular probes for 10. Liu JW, Lu Y. Fast colorimetric sensing of adenosine and cocaine based on cancer diagnosis. a general sensor design involving aptamers and nanoparticles. Angew Chem Int Ed Engl 2005;45:90–4. 11. Blank M, Weinschenk T, Priemer M, Schluesener H. Systematic evolution of Table 1. Aptamer profiling of cancer cells in a DNA aptamer binding to rat brain tumor microvessels. J Biol Chem 2001;276:16464–8. a patients’ samples. 12. Daniels DA, Chen H, Hicke BJ, Swiderek KM, Gold L. A tenascin-C aptamer sgc8 sgc3 sgc4 sgd2 sgd3 sgd5 identified by tumor cell SELEX: systematic evolution of ligands by exponential enrichment. Proc Natl Acad Sci U S A 2003;100:15416–21. T ALL 1 ϩϩ ϩϩϩ ϩϩϩ ϩϩϩ ϩϩϩ ND 13. Wang C, Zhang M, Yang G, Zhang D, Ding H, Wang H, et al. Single-stranded T ALL 2 ϩϩ ϩ ϩϩϩ ϩϩ ϩ 0 DNA aptamers that bind differentiated but not parental cells: subtractive ϩ ϩ ϩϩϩϩ ϩϩϩ ϩ systematic evolution of ligands by exponential enrichment. J. Biotechnol T ALL 3 0 2003;102:15–22. T ALL 4 ϩ ϩ ϩϩ ϩϩϩ ϩ 0 ϩϩ ϩϩϩ ϩ T ALL 5 0 Previously published online at DOI: 10.1373/clinchem.2006.083246 T ALL 6 0 0 ϩϩ00 T ALL 7 0 0 ϩϩ ϩϩ 00 TALL 8 ϩ ϩ ϩϩ ϩϩ ϩ 0 TALL 9 ϩ 0 ϩϩ00 TALL10 0 ϩϩ 0 ϩ 0 B ALL 1 0 0 ϩϩ ϩϩ 00Natural Calcium Isotopic Composition of Urine as a 1* B ALL 2 0 0 ϩϩ ϩϩ 0 ϩ Marker of Bone Mineral Balance, Joseph Skulan, Thomas 2 3 4 B ALL 3 ϩϩ 0 ϩϩ ϩϩ 0 ϩ Bullen, Ariel D. Anbar, J. Edward Puzas, Linda Shackel- 5 6 5 1 B-ALL 4 0 0 ϩϩ00ford, Adrian LeBlanc, and Scott M. Smith ( Department of AML 1 ϩ ϩ ϩϩ ϩ 00Geology and Geophysics, University of Wisconsin-Madi- 2 AML 2 ϩ 0 ϩϩ ϩ 00son, Madison, WI; Branch of Regional Research, Water AML 3 ϩ 0 ϩϩ00Resources Discipline, US Geological Survey, Menlo Park, 3 AML 4 0 0 ϩϩϩϩ ϩϩϩϩ 00CA; School of Earth and Space Exploration and Depart- AML 5 0 0 ϩ 000ment of Chemistry and Biochemistry, Arizona State Uni- 4 AML 6 ϩ 00 000versity, Tempe, AZ; Department of Orthopaedics, Uni- AML 7 ϩ 00 000versity of Rochester School of Medicine and Dentistry, 5 AML 8 ϩ 0 ϩϩϩ ϩϩϩ 00Rochester, NY; Human Adaptation and Countermea-

a sures Division, NASA Johnson Space Center, Houston, In the flow cytometry analysis, a threshold based on fluorescence intensity of 6 FITC was chosen so that 99% percent of cells incubated with the FITC-labeled TX; Department of Medicine, Baylor College of Medi- * unselected DNA library would have fluorescence intensity below it. When cine, Houston, TX; address correspondence to this FITC-labeled aptamer was allowed to interact with the cells, the percentage of the author at: Department of Geology and Geophysics, Uni- cells with fluorescence above the set threshold was used to evaluate the binding versity of Wisconsin-Madison, 1215 W. Dayton St., Mad- capacity of the aptamer to the cells. 0, Ͻ10%; ϩ, 10%—35%; ϩϩ, 35%—60%; ison, WI 53706; fax 608-262-0693, e-mail jlskulan@ ϩϩϩ ϩϩϩϩ Ͼ , 60%—85%; , 85%. geology.wisc.edu) 1156 Technical Briefs

Background: We investigated whether changes in the and can be measured reliably with high-precision mass natural isotopic composition of calcium in human urine spectrometry. track changes in net bone mineral balance, as predicted Calcium has 6 naturally occurring isotopes (40Ca, 42Ca, by a model of calcium isotopic behavior in vertebrates. 43Ca, 44Ca, 46Ca, and 48Ca). Because natural variations in If so, isotopic analysis of natural urine or blood calcium calcium isotopic compositions are, with very rare excep- could be used to monitor short-term changes in bone tions (3), strictly dependent on the relative masses of the mineral balance that cannot be detected with other isotopes, it is possible to calculate the complete isotopic techniques. composition of a calcium sample from the ratio of any 2 Methods: Calcium isotopic compositions are expressed isotopes (4, 5). Calcium isotopic compositions are most as ␦44Ca, or the difference in parts per thousand be- commonly reported as ␦44Ca, or the difference in parts per tween the 44Ca/40Ca of a sample and the 44Ca/40Ca of a thousand between the 44Ca/40Ca of a sample and the 44 40 standard reference material. ␦44Ca was measured in Ca/ Ca of a standard reference material (in this case urine samples from 10 persons who participated in a seawater): study of the effectiveness of countermeasures to bone ␦44Ca ͑‰͒ loss in spaceflight, in which 17 weeks of bed rest was ϭ ͕͓͑44 40 ͒ ͑44 40 ͔͒ Ϫ ͖ ϫ used to induce bone loss. Study participants were as- Ca/ Ca sample/ Ca/ Castandard 1 1000 signed to 1 of 3 treatment groups: controls received no Natural processes can fractionate (that is, alter the relative treatment, one treatment group received alendronate, abundances of) calcium isotopes (1, 4, 6). In nature, ␦44Ca and another group performed resistive exercise. Mea- ranges from approximately Ϫ1.6‰ to Ϫ0.4‰ in nonbio- surements were made on urine samples collected before, genic minerals, and from approximately Ϫ4.0‰ to at 2 or 3 points during, and after bed rest. ϩ1.0‰ in biological materials. Within organisms, there Results: Urine ␦44Ca values during bed rest were lower typically is ϳ2‰–4‰ difference between the highest and in controls than in individuals treated with alendronate 44 lowest ␦ Ca values measured in different tissues (1). < < (P 0.05, ANOVA) or exercise (P 0.05), and lower than Fractionation mechanisms are not well understood, but < the control group baseline (P 0.05, t-test). Results were calcium isotope fractionation during abiological and bio- consistent with the model and with biochemical and logical mineralization is well documented, and probably bone mineral density data. accounts for much of the observed natural variation in Conclusion: Results confirm the predicted relationship ␦44Ca (1, 4, 6). Among a phylogenetically diverse group of between bone mineral balance and calcium isotopes, nonhuman vertebrates, ␦44Ca of bone is ϳ1.3‰ lower suggesting that calcium isotopic analysis of urine might than that of dietary and dissolved soft tissue calcium from be refined into a clinical and research tool. which bone mineral precipitates (1). This difference © 2007 American Association for Clinical Chemistry largely results from fractionation during bone formation. Bone resorption, however, being a bulk dissolution pro- An empirical model of the behavior of calcium isotopes in cess, does not fractionate calcium isotopes. As a result, vertebrates (1) predicts that the natural calcium isotope when the rates of bone formation and bone resorption are composition of human soft tissue and urine is affected by equal and dietary intake of ␦44Ca is constant, the differ- changes in the relative rates of bone formation and ence in ␦44Ca between bone and soft tissue is expected to resorption. Confirmation of this prediction would indi- be constant. cate that measurements of the calcium isotope composi- A simple model of calcium isotope behavior predicts tion of urine or blood might be a useful marker of net that positive bone mineral balance (during which the rate bone mineral balance. Such a marker, which would not of bone formation exceeds the rate of bone resorption) require accelerator mass spectrometry or the administra- will cause positive excursions in soft tissue ␦44Ca, and vice tion of isotopic tracers, could provide information that versa: cannot be gained from bone biochemical markers, and ␦ ϭ ⌬ ϩ ͓ ͑␦ Ϫ ␦ ͒ Ϫ ⌬ ͔ ͑ ϩ ͒ provide it more quickly than is possible with observed s b Vs b d Vb b / Vd Vs , ␦ ␦ ␦ ␦44 changes in bone mineral density (BMD). where s, d, and b are Ca of soft tissue, dietary Ca, Although measurements of calcium isotopes have a and bone, respectively; Vs,Vb, and Vd are the fluxes of wide range of applications in research on bone and calcium resorbed from bone, incorporated into bone, and ⌬ calcium metabolism, to date these applications have been absorbed from diet, respectively; and b is the Ca isotope limited to the use of artificially prepared calcium isotopic fractionation during bone formation (1). tracers such as 41Ca (2). The application described in this To test the hypothesized relationship between soft paper is fundamentally different from tracer studies in tissue ␦44Ca and bone mineral balance in humans, we that it is based on measurements of natural, biologically measured ␦44Ca of urine from a 10-person subset of induced variations in the calcium isotopic composition participants in a previously published human study of the of urine. These variations are too small to be of interest effectiveness of countermeasures to bone loss in space- in tracer studies, but they are biologically informative, flight (7), in which extended bed rest (17 weeks) was used Clinical Chemistry 53, No. 6, 2007 1157

to induce bone loss and simulate weightlessness. Study and with published interpretations of the BMD and participants were assigned to 1 of 3 treatment groups: no biochemical marker data (7). Tested against a null hy- treatment (controls; n ϭ 4), treatment with the antiresorp- pothesis of no change in calcium isotope composition, tive drug alendronate (n ϭ 3), or treatment with resistive mean urinary ␦44Ca values became positive in the alen- exercise (n ϭ 3). Study participants were fed identical dronate group (P Ͻ0.05, t-test), remained unchanged in diets; 2 samples of homogenized diet yielded the same the exercise group (P Ͼ0.50), and became negative in the ␦44Ca (Ϫ1.88‰ and Ϫ1.90‰), indicating that dietary control group (P Ͻ0.05), the group expected to have lost ␦44Ca was approximately constant during the study. Data bone during bed rest (Fig. 1A, Table 1). No significant were collected on BMD and biochemical markers of bone trends were observed in any group between weeks 4 and resorption (n-telopeptide, NTX) and bone formation 17 (P Ͼ0.50, ANOVA). (bone-specific alkaline phosphatase, BSAP). ␦44Ca was ␦44Ca was lower in the control group than in the measured for food samples and 1-week pooled urine alendronate (P Ͻ0.05, ANOVA) and exercise groups (P samples (a soft tissue proxy) collected at intervals before, Ͻ0.05). There was no significant difference between the during, and after bed rest. alendronate and exercise groups (P ϭ 0.220). Because Analytical procedures for BMD and biochemical mark- analytical uncertainty in ␦44Ca measurements is low, ers are given elsewhere (7). ␦44Ca measurements were interindividual variability within groups reflects real bi- conducted at the US Geological Survey in Menlo Park, ological differences between individuals. CA, using well-established and rigorously tested geo- It is difficult to calculate bone mineral balance from chemical techniques, including a calcium isotope biochemical markers of bone formation and resorption. “double-spike” that corrects for matrix effects and instru- These markers are of marginal clinical value (10, 11) and mental isotope fractionation (1, 5, 6, 8, 9). Reported ␦44Ca are not calibrated to the mass of bone gained or lost from values are based on a minimum of 60 isotope ratio the skeleton. In many cases bone formation and resorp- measurements, with an internal imprecision of 0.15‰ (2␴, tion increase or decrease together. For example, bone 95% CI) or better. Total procedural replicates of 15 urine fractures cause a large increase in the rate of bone samples collected for this study yielded an mean external (intraassay) imprecision of Ϯ0.154‰ (2␴, 95% CI); external imprecision was Ϯ0.040‰ to Ϯ0.154‰ on full procedural replicates on reference materials having ␦44Ca values of 0‰ to Ϫ2.01‰ (see the Technical Information in the Data Supplement that accompanies the online version of this article at http://www.clinchem.org/content/ vol53/issue6). No funding organization influenced the design or execution of these methods, or the presentation of our results. Results (Table 1 and Fig. 1) were consistent with the hypothesized association between changes in soft tissue (e.g., urine) ␦44Ca and changes in bone calcium balance,

Table 1. Summary of data.a Marker Unit Treatment group Mean change SD n ␦44Cab ‰ Alendronate 0.32 0.28 3 ␦44Ca ‰ Exercisec 0.12 0.30 3 ␦44Ca ‰ Control Ϫ0.48 0.63 4 NTXb nmol/d Alendronate Ϫ63.5 72 3 NTX nmol/d Exercise 122 196 3 NTX nmol/d Control 256 119 4 BSAPb U/L Alendronate Ϫ2.86 3.17 3 BSAP U/L Exercise 9.81 7.67 3 BSAP U/L Controld Ϫ0.32 1.39 4 Fig. 1. (A), changes in ␦44Ca during bed rest, by treatment group, at a Values are means and SD of differences between initial values and values weeks 4, 10, and 16 or 17, and after bed rest at week 19. during bed rest of all samples taken during bed rest, by group. Symbols show mean values, ends of whiskers show extreme values. * P Ͻ0.05. b Samples obtained from each study participant at weeks 0, 4, 10 or 12, and ‡ P Ͻ0.10. (B), mean change during bed rest from initial values of urinary ␦44Ca 16 or 17 of bed rest (data for some individuals are missing for some weeks). P and BMD of the femoral neck and lumbar spine, for alendronate (A), exercise (E), values for differences in ␦44Ca between treatment groups are all Ͻ0.05 except and control (C) groups. For ␦44Ca, dark lines inside boxes show 50th percentiles, for exercise vs alendronate. c P ϭ 0.222 for exercise vs alendronate. d P ϭ ends of boxes show 25th and 75th percentiles, and ends of whiskers show upper central bars ends of 0.143 for control vs alendronate. and lower extremes. For BMD data, show means and whiskers show extreme values. 1158 Technical Briefs

remodeling and in the concentration of formation and J.E.P. and A.D.A.) and by HIH P30 ES 01247-24 (Department resorption markers (12). In this study both NTX and of Orthopaedics, University of Rochester School of Medicine BSAP decreased in individuals treated with alendronate, and Dentistry, Rochester, NY). We also acknowledge support and both increased in the exercise group (Table 1). In from the National Research Program of the Water Resources contrast, ␦44Ca is directly related to net change in skeletal Discipline, US Geological Survey. Financial disclosures: None declared mass and not to the individual rates of bone formation Acknowledgements: We thank Richard Slaughter, Dale and resorption. Schoeller, and Aarthy Raman for help with statistical analy- ␦44 Large changes in Ca occur much more rapidly than ses, and Victoria Abrash, Jane Krauhs, and 2 anonymous ␦44 in BMD (Fig. 1B). In the control group Ca had become reviewers for their invaluable assistance in preparing and negative by the 4th week of bed rest (the earliest point in revising the manuscript. the study from which urine samples were analyzed) and was trending positive in the alendronate group (P Ͻ0.05 References and 0.058, respectively), whereas 17 weeks were required 1. Skulan J, DePaolo DJ. Calcium isotope fractionation between soft and mineralized tissues as a monitor of calcium use in vertebrates. Proc Natl to produce detectable changes in some BMD measures. In Acad Sci U S A 1999;96:13709–13. 2. Fitzgerald RL, Hillegonds DJ, Burton DW, Griffin TL, Mullaney S, Vogel JS, et fact, extensive studies of calcium isotopic tracers (13–15) 41 ␦44 al. Ca and accelerator mass spectrometry to monitor calcium metabolism suggest that Ca responds to changes in bone mineral in end stage renal disease patients. Clin Chem 2005;51:2095–102. balance in much less than 4 weeks. The residence time of 3. Marshall BD, DePaolo DJ. Precise age determinations and petrogenic calcium in different nonskeletal pools varies, so a week or studies using the K-Ca method. Geochim Cosmochim Acta 1982;46:2537– 45. more may be required to achieve full equilibration be- 4. Skulan J, DePaolo DJ, Owens TL. Biological control of calcium isotopic tween bone mineral balance and urine ␦44Ca (16, 17), but abundances in the global calcium cycle. Geochim Cosmochim Acta 1997; 61:2205–10. changes in bone mineral balance should be detectable in 5. Russell WA, Papanastassiou DA, Tombrello TA. Ca isotope fractionation on 44 urinary ␦ Ca within a few days. Long-term changes in the earth and other solar system materials. Geochim Cosmochim Acta bone mineral balance could be monitored by periodic 1978;42:1075–90. 6. Zhu P, Macdougall JD. Calcium isotopes in the marine environment and the sampling of urine calcium. oceanic calcium cycle. Geochim Cosmochim Acta 1998;62:1691–8. The calcium isotope data reported here constitute 7. Shackelford LC, LeBlanc AD, Driscoll TB, Evans HJ, Rianon NJ, Smith SM, et al. Resistance exercise as a countermeasure to disuse induced bone loss. virtually all of the information we have on the natural J Appl Physiol 2004;97:119–29. behavior of calcium isotopes in humans, and more data 8. DePaolo DJ. Calcium isotopic variations produced by biological, kinetic, are needed to validate the use of natural calcium isotopes radiogenic and nucleosynthetic processes. Rev Mineral Geochem 2004;55: 255–88. as a biomarker. Although analytical variability in calcium 9. Perakis SS, Maguire DA, Bullen TD, Cromack K, Waring RH, Boyle JR. isotope measurements is very low, we have no informa- Coupled nitrogen and calcium cycles in forests of the Oregon Coast Range. ␦44 Ecosystems 2006;9:63–74. tion on the background variation in urinary Ca in 10. Gillett MJ, Vasikaran SD. Urinary NTX results rarely alter the clinical healthy humans outside of conditions designed to induce management of patients with osteoporosis in the tertiary hospital. Pathology bone loss, and thus cannot accurately assess preanalytical 2006;38:49–52. 11. NIH Consensus Development Panel on Osteoporosis Prevention, Diagnosis, variability (18) or establish a reference change value (19) and Therapy. Osteoporosis prevention, diagnosis, and therapy. JAMA 2001; in calcium isotope composition. Processes other than bone 285:785–95. 12. Watts NB. Clinical utility of biochemical markers of bone remodeling. Clin formation may fractionate calcium and produce isotopic Chem 1999;45:1359–68. differences between soft tissue calcium compartments. 13. Heaney RP, Recker RR. Estimation of true calcium absorption. Ann Intern For example, ␦44Ca of all urine samples was higher than Med 1985;103:516–21. ␦44 14. Freeman SPHT, Beck B, Bierman JM, Caffee MW, Heaney RP, Holloway L, et dietary Ca (see the online Data Supplement). Differ- al. The study of skeletal calcium metabolism with 41Ca and 45Ca. Nucl ences in calcium isotopic composition between soft tissue Instrum Methods Phys Res B 2000;172:930–3. compartments must be investigated and explained. 15. Weaver CM. Use of calcium tracers and biomarkers to determine calcium kinetics and bone turnover. Bone 1998;22(Suppl):103S–104S. Future research will focus on gaining a comprehensive 16. Smith SM, Wastney ME, Nyquist LE, Shih CY, Wiesmann H, Nillen JL, et al. understanding of the distribution of calcium isotopes in Calcium kinetics with microgram stable isotope doses and saliva sampling. J Mass Spectrom 1996;31:1265–70. the human body. This research is warranted by the 17. Smith SM, Wastney ME, O’Brien KO, Morukov BV, Larina IM, Abrams SA, et 44 potential for measurement of ␦ Ca to be an important al. Bone markers, calcium metabolism, and calcium kinetics during extend- addition to the tools currently available for diagnosing, ed-duration space flight on the Mir space station. J Bone Miner Res 2005;20:208–18. studying, and assessing treatments for metabolic bone 18. Vesper HW. Analytical and preanalytical issues in measurement of biochem- disease in humans. ical bone markers. Lab Med 2005;36:424–9. 19. Worsfold M, Powell DE, Jones TJ, Davie MW. Assessment of urinary bone markers for monitoring treatment of osteoporosis. Clin Chem 2004;50: 2263–70. Grant/funding support: This research was supported by a grant from the Environmental Health Sciences Center of the Previously published online at DOI: 10.1373/clinchem.2006.080143 University of Rochester School of Medicine and Dentistry (to Letters

Alanine Aminotransferase as an Interpretation of the results of the who were free of NAFLD at base- Independent Predictor of Incident Chang et al. study are limited by line (5). Nonalcoholic Fatty Liver Disease another major caveat (1), also recognized by the authors, that the diag- Grant funding/support: None de- To the Editor: nosis or exclusion of NAFLD, both at baseline and follow-up, was based clared. We read with interest the recent ar- Financial disclosures: None declared. ticle by Chang et al. (1) reporting on liver enzymes and ultrasound imaging, but was not confirmed by that higher serum alanine amino- References liver biopsy. Indeed, it is known that transferase (ALT) concentrations, 1. Chang Y, Ryu S, Sung E, Jang Y. Higher concen- within the reference interval, inde- liver enzymes may be within the trations of alanine aminotransferase within the reference interval in up to 70% of reference interval predict nonalcoholic fatty liver pendently predicted the incidence of disease. Clin Chem 2007;53:686–92. patients with diagnosed NAFLD and nonalcoholic fatty liver disease 2. Adams LA, Angulo P, Lindor KD. Nonalcoholic that the full histopathological spec- fatty liver disease. CMAJ 2005;172:899–905. (NAFLD) during a mean follow-up trum of NAFLD may be present in 3. Targher G. Non-alcoholic fatty liver disease, the of 2.5 years in a large cohort of ap- patients with normal liver enzymes, metabolic syndrome and the risk of cardiovascu- parently healthy Korean men. lar disease: the plot thickens [Mini-Review]. Dia- which therefore cannot be reliably bet Med 2007;24:1–6. Several prospective studies have used to exclude the presence of 4. Saadeh S, Younossi ZM, Remer EM, Gramlich T, previously shown that increased NAFLD (2). Moreover, although Ong JP, Hurley M, et al. The utility of radiological ALT concentrations, even within the imaging in nonalcoholic fatty liver disease. Gas- liver ultrasonography is widely used troenterology 2002;123:745–50. reference interval, also predict the for diagnosing NAFLD, this imaging 5. Suzuki A, Angulo P, Lymp J, St Sauver J, Muto A, future development of type 2 diabe- Okada T, et al. Chronological development of method has good sensitivity and tes (2) and cardiovascular events (3) elevated aminotransferases in a nonalcoholic specificity only for detection of mod- population. Hepatology 2005;41:64–71. independently of other known risk erate and severe hepatic steatosis, factors. In all of these studies, how- but its sensitivity is reduced when Giovanni Targher*1 ever, increased ALT concentrations hepatic fat infiltration on liver biopsy Massimo Franchini3 have been used as a surrogate is Ͻ33% (4). Only liver biopsy can be Gian Cesare Guidi2 marker of NAFLD. Indeed, increased used for diagnosing NAFLD and ac- Michele Muggeo1 liver enzymes are usually thought to curately determining the histological Giuseppe Lippi2 be a consequence (and not a cause) of severity and prognosis of liver dam- liver injury in NAFLD and can pos- age (2). Thus, although this limita- 1 Section of Endocrinology sibly be used as predictors of tion, if present, would probably tend Department of Biomedical and NAFLD progression (2). to reduce (at least partly) the Surgical Sciences and We think the association of the 2 strength of the association between conditions, increased ALT concentra- ALT concentrations and incident 2 Section of Clinical Chemistry tions and incident NAFLD, does not NAFLD shown by Chang et al. (1), Department of Biomedical and necessarily prove causation. The we think some nondifferential mis- Morphological Sciences greater incidence of NAFLD—as di- classification of NAFLD on the basis University Hospital of Verona agnosed by ultrasound—among of liver enzymes and ultrasonogra- Verona, Italy those with slightly increased ALT phy is likely. That is, some of the concentrations at baseline is most study participants may not have 3 Service of Immunohematology and likely attributable to an underlying NAFLD despite ultrasound detection Transfusion, Civil Hospital common mechanism, i.e., more se- of fatty liver and some may have Verona, Italy vere insulin resistance in those with underlying NAFLD with normal higher than in those with lower ALT liver enzymes and negative ultra- concentrations. That the association * Address correspondence to this au- sound findings. The latter situation thor at: University Hospital, Piazzale Ste- between increasing serum ALT con- may have been partly confirmed fani, 1, 37126 Verona, Italy. Fax 39-045- centrations and incident NAFLD re- during the short follow-up study (Ͻ3 917374; e-mail [email protected]. mained statistically significant even years) by the unexpectedly high inci- DOI: 10.1373/clinchem.2007.087304 after adjustment for the homeostasis dence of NAFLD detected in the 5237 model assessment (HOMA)-esti- healthy participants with a mean age mated insulin resistance may be due of 36 years (984 incident NAFLD simply to the fact that the HOMA cases, i.e., ϳ75 new cases per 1000 score is not a good proxy measure of person-years). In contrast, the inci- Reply to Alanine Aminotransferase as insulin resistance. Thus, we wonder dence of high aminotransferase con- an Independent Predictor of Incident how different the results would have centrations, as surrogate markers of Nonalcoholic Fatty Liver Disease been if the euglycemic clamp tech- NAFLD, was recently reported to be nique or methods that are more ac- ϳ30 new incident cases per 1000 per- To the Editor: curate had been used to measure son-years in a cohort of Japanese First, the relationship between ala- insulin resistance. healthy individuals (age 35 years) nine aminotransferase (ALT) and

Clinical Chemistry 53, No. 6, 2007 1159 1160 Letters

nonalcoholic fatty liver disease resistance (2). Furthermore, a recent infiltration (Ͻ33%), and the limited (NAFLD) has not yet been clearly study using sophisticated techniques ability, as we also discussed in our established, although increased liver such as MRI and the oral glucose article, to differentiate simple steato- enzymes are usually used as a surro- tolerance test has shown that only sis from inflammation or fibrosis (4). gate marker of NAFLD. Tagher et al. half of the patients with MRI- Whether the degree of misclassifica- have also noted that liver enzymes assessed fatty liver disease had in- tion of fatty liver diagnoses based on may be within the reference interval creased ALT, whereas even modest US varies according to the ALT con- in up to 70% of patients diagnosed ALT increases within the reference centration is presently unknown. As with NAFLD and that the full histo- interval were associated with deteri- Tagher et al. note, nondifferential logical spectrum of NAFLD may be oration in glucose and lipid metabo- misclassification, if present, would present in patients with normal liver lism as well as insulin resistance (3). tend to reduce the strength of the enzymes. Our observations were As we noted in our article, even in associations between ALT and therefore novel and have potential participants without any features of NAFLD. importance in understanding the re- the metabolic syndrome, any in- Finally, Tagher and colleagues lationship between ALT and crease of serum ALT, even that found differences in the incidence of NAFLD. Furthermore, several issues within the reference interval, contin- NAFLD in our study and a previ- need to be clarified about ALT, in- ued to predict the incidence of ously reported study by Suzuki et al. cluding the isoforms of ALT and the NAFLD (4). To date, although the (5). These differences may have been role of different ALT isoforms in relationship between ALT and in part attributable to differences in humans (1). NAFLD remains to be elucidated, both the study populations and the Second, regarding the adjustment our observation supports the idea outcome measures. In the study by of confounders, such as insulin resis- that ALT is not only a consequence Suzuki et al. (5), the study partici- tance, we did not perform this study but also a predictor of NAFLD. We pants were 20–59 years old, were using the euglycemic clamp and can- agree with Tagher and colleagues 26.8% female, and the incidence of not comment on Tagher’s concerns that insulin resistance as assessed by increased transaminasemia was about the interrelation between ALT, sophisticated techniques such as the highest among males 20–39 years old NAFLD, and insulin resistance as euglycemic clamp method could be (5). Our study population, however, assessed by the euglycemic clamp helpful in understanding the rela- was composed of only males, and the technique. Indeed, performance of tionship between ALT and NAFLD majority (70.8%) were 30–39 years studies with large samples are not through a pathway that is distinct old. Outcome measurements in the feasible with the glucose clamp from systemic insulin resistance. Suzuki et al. (5) study were based on methods. Fortunately, other investi- Third, regarding the determina- increased ALT whereas ours were gators have reported that the ho- tion of NAFLD, Tagher et al. raised a based on US. For exploratory pur- meostasis model assessment of insu- concern regarding the ultrasound poses, we performed ancillary anal- lin resistance (HOMA-IR) is strongly (US) measurements used in our yses to estimate the incidence of in- related to clamp-measured insulin study, the sensitivity of early fatty creased ALT as an outcome measure.

Table 1. Adjusted hazard ratios of incidence of nonalcoholic fatty liver disease in relation to ALT concentrations within the reference interval after excluding the participants who developed NAFLD within the first 1 or 2 years. ALT (U/L) I(<16) II (16ϳ18) III (19ϳ21) IV (22ϳ25) V (26ϳ34) P for trend Analyses restricted to the participants after excluding the participants who developed NAFLD within the first 1 year (n ϭ 4826) Cases 64 86 115 135 173 Person-years 3200.8 2588.1 2552.1 2326.5 2313.3 ID, (per 1000 person-years) 20.0 33.2 45.1 58.0 74.8 aHR, (95% CI) Model 1a 1.00 1.65 (1.19–2.28) 2.25 (1.66–3.06) 2.90 (2.16–3.91) 3.84 (2.88–5.12) Ͻ0.001 Model 2b 1.00 1.34 (0.97–1.88) 1.55 (1.13–2.14) 1.95 (1.43–2.65) 2.09 (1.54–2.84) Ͻ0.001 Analyses restricted to the participants after excluding the participants who developed NAFLD within the first 2 year (n ϭ 4502) Cases 36 40 58 57 58 Person-years 3165.2 2530.0 2482.9 2231.9 2174.7 ID, (per 1000 person-years) 11.4 15.8 23.6 25.5 26.7 aHR, (95% CI) Model 1a 1.00 1.35 (0.86–2.12) 2.05 (1.35–3.11) 2.21 (1.45–3.35) 2.44 (1.61–3.70) Ͻ0.001 Model 2b 1.00 1.26 (0.79–2.02) 1.68 (1.09–2.60) 1.74 (1.12–2.69) 1.65 (1.06–2.57) 0.011

a b Model 1: adjustment for age; model 2: model 1 plus adjustment for weight change, fasting serum glucose, loge triglyceride, HDL-cholesterol, body mass index, systolic blood pressure, smoking, exercise, alcohol intake, HOMA-IR, C-reactive protein, and incident diabetes. ID, Incidence density; aHR, adjusted hazard ratios. Clinical Chemistry 53, No. 6, 2007 1161

Whereas 984 incident cases of US- Yoosoo Chang1 say, based on Luminex xMAP tech- based NAFLD developed during Seungho Ryu2* nology that uses two 3-mm punches 13 882.4 person-years of follow-up Eunju Sung3 from each DBSS, for simultaneous [incidence density (ID), 74.1 per 1000 Yumi Jang4 determination of 25 inflammatory person-years; 95%, CI 69.5–78.7], 700 markers and neurotrophins (4).We incident cases of increased ALT, de- 1 Health Screening Center now report that after extraction of fined as serum ALT Ն35 U/L, devel- proteins for the immunoassays, there oped during 13 822.4 person-years of 2 Department of Occupational is sufficient genomic DNA (gDNA) follow-up (ID, 50.6 per 1000 person- Medicine and on the used DBSS disks to perform years; 95% CI, 46.9–54.4). To mini- several WGA reactions, each produc- mize the effect of existing fatty infil- 3 Department of Family Medicine ing enough whole–genome-ampli- tration on the baseline ALT, Kangbuk Samsung Hospital fied DNA (wgaDNA) for numerous additional analyses were performed Sungkyunkwan University reliable genotypings. by excluding participants with inci- School of Medicine Blood from 20 anonymous volun- dent NAFLD that occurred during Seoul, Korea teers was spotted on filter paper and the first 1 or 2 years of follow-up air dried, and six 3-mm disks were (Table 1). Although the magnitude of 4 Department of Radiology punched out from each DBSS. Half the association between ALT and Asan Medical Center the disks were subjected to protein NAFLD was slightly reduced, this University of Ulsan College of extraction (4). The 2 groups of disks analysis did not qualitatively change Medicine from each individual were then fur- any of the observed associations (Ta- Ulsan, Korea ther split into 2 subgroups with, re- ble 1). Future research will shed light spectively, 1 and 2 disks in each, and on the question of what underlies the gDNA was extracted in a volume of ௢ relationship between ALT and an * Address correspondence to this au- 200 ␮L using the Extract-N-Amp increased risk of NAFLD. thor at: Kangbuk Samsung Hospital, 108 reagent set (Sigma-Aldrich). The con- We appreciate the opportunity to Pyung dong, Jongro-Gu, Seoul, Korea centration of gDNA (0.2 mg/L, mea- 110-746. Fax 82-2-2001-2626; e-mail ௢ ® clarify our findings. The letter by [email protected]. sured by the Quant-IT PicoGreen Tagher and colleagues raises several dsDNA Reagent) was not correlated important issues in the interpretation DOI: 10.1373/clinchem.2007.089300 to the use of 1 or 2 disks or to prior of our findings. protein extraction. We then amplified extracted gDNA with 2 different WGA reagent sets: the GenomePlex Whole Genome Amplification Kit Whole Genome Amplification and Grant funding/support: None de- (Sigma-Aldrich), which is based on Genetic Analysis after Extraction of clared. the Omniplex method, and the Am- Financial disclosures: None declared. Proteins from Dried Blood Spots pliQ Genomic Amplifier Kit (Ampli- qon), which uses the multiple-dis- To the Editor: placement amplification approach. References Programs to screen newborns for ␮ 1. Schindhelm RK, Diamant M, Dekker JM, Tush- Allowed input volumes are 10 L for uizen ME, Teerlink T, Heine RJ. Alanine amino- congenital disorders are based on the GenomePlex (ϳ2.0 ng gDNA) transferase as a marker of non-alcoholic fatty analysis of dried blood spot samples and 8 ␮L for the AmpliQ, (ϳ1.6 ng liver disease in relation to type 2 diabetes mellitus and cardiovascular disease. Diabetes (DBSS), which have proven to be gDNA). Both reagent sets use less Metab Res Rev 2006;22:437–43. robust and convenient for collection, input gDNA than the recommended 2. Bonora E, Targher G, Alberiche M, Bonadonna transport, and storage. Because 10 ng, a feature that may be critical RC, Saggiani F, Zenere MB, et al. Homeostasis blood samples are collected with no model assessment closely mirrors the glucose for the genotyping performance of clamp technique in the assessment of insulin selection, and coverage of the popu- the resulting wgaDNA (5), and thus sensitivity: studies in subjects with various lation is essentially universal, resid- direct comparisons cannot be made. degrees of glucose tolerance and insulin sensitivity. Diabetes Care 2000;23:57–63. ual DBSS combined with patient reg- The yield of wgaDNA was 2.6–3.8 ␮ 3. Burgert TS, Taksali SE, Dziura J, Goodman TR, isters are a valuable resource for g, independent of whether protein Yeckel CW, Papademetris X, et al. Alanine epidemiological research (1). The extraction was performed and aminotransferase levels and fatty liver in childhood obesity: associations with insulin resis- usefulness of DBSS is limited by the whether 1 or 2 disks were used. tance, adiponectin, and visceral fat. J Clin small amount of blood available, The quality of wgaDNA was Endocrinol Metab 2006;91:4287–94. however, and methods that optimize assessed by TaqMan® single-nucle- 4. Chang Y, Ryu S, Sung E, Jang Y. Higher concen- the use of the sample are required, otide polymorphism (SNP) geno- trations of alanine aminotransferase within the reference interval predict nonalcoholic fatty such as various whole genome am- typing of 27 SNPs in 27 genes dis- liver disease. Clin Chem 2007;53:686–92. plification (WGA) methods used in tributed on 13 chromosomes. Refer- 5. Suzuki A, Angulo P, Lymp J, St Sauver J, Muto genetic epidemiological studies ence gDNA was extracted from 200 A, Okada T, et al. Chronological development of ␮ elevated aminotransferases in a nonalcoholic (2, 3). We previously described a L of whole blood from the same 20 population. Hepatology 2005;41:64–71. high-capacity multiplex immunoas- individuals. Genotype calls were 1162 Letters

2 NANEA at Department of Table 1. TaqMan call and error rates of wgaDNA from DBSS disks. Epidemiology Call rate % Error rate % Institute of Public Health Pre-extraction No pre-extraction Pre-extraction No pre-extraction Aarhus University, Aarhus, Denmark WGA kit 1 disk 2 disks 1 disk 2 disks 1 disk 2 disks 1 disk 2 disks GenomePlex 98.8 99.0 98.0 94.1 0.4 0.2 0.4 1.0 AmpliQ 96.4 99.0 94.6 88.8 1.3 0.6 2.1 1.7 *Address correspondence to this author at: Department of Clinical Biochem- Pre-extracted disks were subjected to protein extraction prior to DNA extraction. istry, Statens Serum Institut, Artillerivej 5, DK-2300 Copenhagen S, Denmark. Fax 45-3268-3878; e-mail [email protected]. done independently by 3 persons. niplex-based method producing DOI: 10.1373/clinchem.2006.082313 The call and error rates are shown in short wgaDNA of 400–500 bp and Table 1. Despite the suboptimal the multiple-displacement amplifica- amount of input gDNA, the best call tion-based method producing long and error rates were nearly optimal. wgaDNA of 10–20 kbp are well The most common errors, results in- suited for the amplification. In our Pyrosequencing Technology for dicating that heterozygous individu- setup, the amount of template gDNA Automated Detection of the BMP15 als were homozygous, were attribut- from each extraction was sufficient ϳ A180T Variant in Spanish able to allele dropouts most probably for 20 amplifications, and the yield Postmenopausal Women due to an insufficient amount or low of wgaDNA from 1 amplification quality of template gDNA. Details of was enough for ϳ300 TaqMan geno- To the Editor: sample flow and call and error rates typings. for each individual gene can be seen Germline mutations in different in the Data Supplement that accom- genes are associated with premature panies the online version of this arti- ovarian failure (POF, OMIM 311360), Grant/funding support: None declared. cle at http://www.clinchem.org/ defined as premature menopause Financial disclosures: None declared. content/vol53/issue5. with amenorrhea occurring before When extraction of proteins from the age of 40 years along with in- disks was performed before extrac- References creased gonadotropin concentrations tion of DNA, the resulting wgaDNA 1. Norgaard-Pedersen B, Hougaard DM. The Danish [follicle-stimulating hormone (FSH) Newborn Screening Biobank in Routine and Re- Ͼ generally showed higher TaqMan search. New Revised Biobank Regulations. In: 40 IU/L]. A new candidate gene, genotype call rates and lower error Knoppers BM. Koninklijke Brill NV Leiden, eds. bone morphogenetic protein 15 rates than wgaDNA from disks that Genomics and Public Health. 2007;97–110. (BMP15) has been investigated in 2. Hannelius U, Lindgren CM, Melen E, Malmberg had not undergone pre-extraction of A, von Dobeln U, Kere J. Phenylketonuria screen- POF. In a family affected by hyper- proteins (Table 1). This result sug- ing registry as a resource for population genetic gonadotropic ovarian failure, a mu- gests that the pre-extraction proce- studies. J Med Genet 2005;42:e60. tation in the pro-region of the BMP15 3. Sun G, Kaushal R, Pal P, Wolujewicz M, Smelser gene (Y235C) was found in 2 affected dure removes compounds from the D, Cheng H, et al. Whole-genome amplification: filter paper disks that may interfere relative efficiencies of current methods. Leg sisters (1), and 3 linked markers with the WGA reaction. This expla- Med [Tokyo] 2005;7:279–86. within the BMP15 gene (Ϫ673CϾT, 4. Skogstrand K, Thorsen P, Norgaard-Pedersen B, Ϫ9CϾG and IVS1 ϩ 905AϾG) are nation is in accordance with the ob- Schendel DE, Sorensen LC, Hougaard DM. Si- servation that the use of 2 disks that multaneous measurement of 25 inflammatory associated with high follicle produc- had not undergone protein pre-ex- markers and neurotropins in neonatal dried tion in women undergoing recombi- blood spots by immunoassay with xMAP technol- traction gave inferior results than did ogy. Clin Chem 2005;51:1854–66. nant FSH stimulation (2). the use of only 1 disk, whereas the 5. Bergen AW, Qi Y, Haque KA, Welch RA, Chanock Several heterozygous variations opposite was the case for the pre- SJ. Effects of DNA mass on multiple displace- affecting the pro-region and mature ment whole genome amplification and genotyp- peptide of the BMP15 gene have extracted disks, for which the use of ing performance. BMC Biotechnology 2005;5: 2 disks instead of 1 yielded superior 24–35. been identified in women with POF results (Table 1). The DBSS used in (3–5), but A180T was the only vari- this study were stored for only a Mads V. Hollegaard1,2 ant found in all reported studies, limited time, and we do not know Karina M. Sørensen1 occurring with relatively high fre- the effect on the quality of wgaDNA Høgni K. Petersen1 quency in POF women (8 of 502, of prolonged storage at room tem- Maria B. Arnardottir1 1.6%) but not at all in control groups. perature or at Ϫ20 °C. Bent Nørgaard-Pedersen1 To clarify the role of the A180T Our results demonstrate that DBSS Poul Thorsen2 allele in early menopause and ovar- disks previously used for multiplex David M. Hougaard1* ian failure, we used a pyrosequenc- protein measurements are reliable ing protocol (Biotage) to evaluate the sources of gDNA that is suitable for 1 Department of Clinical Biochemistry A180T variant. This technology al- WGA and SNP genotyping. The Statens Serum Institut lows an easy 96-well typing format. study also shows that both the Om- Copenhagen, Denmark The selected primers for pyrose- Clinical Chemistry 53, No. 6, 2007 1163

Table 1. Clinical profile of patients carrying BMP15 A180T variant. Individual number carrying BMP15 A180T variant

Phenotypic Features 83 363 410 477 838 1019 1107 Age, y 54 68 49 45 55 59 68 Age at menarche, y 14 14 12 13 13 10 12 Age of menopause, y 46 49 44 35 51 47 35 Cause of menopause Natural Natural Natural Endometriosis Natural Natural Natural Reproductive period, ya 32 35 32 22 38 37 23 Pregnancies/miscarriages/ 1/0/0/1 3/0/0/3 4/0/0/4 0/0/0/0 2/0/0/2 3/1/0/2 2/0/0/2 preterm/live births Body mass index 27.0 29.0 24.2 21.1 28.1 37.1 31.2 Body fat distribution None specially Gynecoid None specially None specially None specially None specially None specially Smoke habit No No No Yes No Yes No Alcohol consumption No Yes No Yes No No No Related comorbidity High cholesterol Hypertension Osteopenia Endometriosis High cholesterol Hypertension Hypertension Osteopenia Diabetes II Diabetes II Osteopenia Osteoporosis Osteoporosis a Reproductive period: time comprised between menarche and menopause. quencing analysis and PCR condi- pants, and the study protocol was finding was not statistically signifi- tions were forward: 5Ј-ACC GCC approved by the referral center Eth- cant. ATC ATC TCC AAC TAA-3Ј, re- ics Committees and Neocodex. Research to date has detected the verse: 5Ј-biotine-CCT GTG TCC CTT Genotyping revealed the A180T A180T variant in only 10 women. Of GTT ATT CCA-3Ј, and sequencing: variant only in 7 unrelated individu- these, 5 of 166 were Italian women 5Ј-AAC CTT CCC TGA TGT CT-3Ј. als, suggesting a variant allele fre- with POF (3);1of133and2of60 We used reextraction and conven- quency of 0.3% in the Spanish popu- Indian women with POF or primary tional resequencing of a random set lation. The clinical data and related amenorrhea, respectively (4); and 2 of samples (10%) with the CEQ Dye morbidity of the 7 women who were of 203 European women with POF Terminator Cycle Sequencing Quick heterozygous carriers of A180T vari- (5). Furthermore, no control individ- Start Kit (Beckman) to assure the qual- ant are reported in Table 1. Among uals analyzed to date have been car- ϭ ity of our genotyping protocol (see these women, 1 patient had meno- riers the A180T allele (n 462, Figure 1 in the Data Supplement that pause caused by a surgical interven- merged sample size). In contrast accompanies the online version of this tion for endometriosis (group C). with these observations, we detected letter at http://www.clinchem.org/ the A180T allele in 5 women with Only 1 patient displayed the SA phe- Ͼ content/vol53/issue4). notype (group B). The other 5 natural menopause occurring at 40 We investigated the presence of years of age. Our results indicate that women had proven fertility and a the A180T allele and also reproduc- the A180T allele is a rare variant that natural menopause occurring at Ͼ40 tive, gynecological, and other impor- is not sufficient to generate SA in years (group A). tant traits in 1157 white Spanish humans. That the allele has not been To analyze relevant variables in- postmenopausal women divided found in control groups to date does volved in age at menopause, such as into 3 clinical groups. Group A in- not exclude the possible presence of cluded women whose age at natural age at menarche, reproductive pe- this rare variant in unaffected indi- menopause was in the normal range riod, pregnancies, and body mass viduals, because the number of con- Ն ϭ index, we further studied patients ( 40 years, n 852), group B in- ϭ trols used in studies to date may cluded women with secondary am- with a natural menopause (n 898) have been insufficient to detect very enorrhea (SA; age at natural meno- by dividing them into 2 genotypic low frequency alleles. However, our pause Ͻ40 years, n ϭ 46), and group groups (presence/absence of A180T data also support the finding that C included women with menopause variant). Our results indicated that genetic variation in the BMP15 gene induced by surgery for treatment of the groups did not differ in any clin- may contribute to variation in age at Ͼ benign or malignant gynecological ical variable studied (P 0.13). Nev- menopause in a complex manner, pathologies (n ϭ 249). The observed ertheless, women carrying the A180T but this interpretation should be mean (SD) ages at menopause and variant seemed to have a lower mean viewed with caution. menarche [48.4 (5.0) and 12.9 (1.56) (SD) age at menopause [45.3 (5.6) vs To elucidate the role of BMP15 in years, respectively] were strictly con- 48.5 (5.0)], and they had a shorter ovulation rate, POF, and SA, exhaus- cordant with others previously re- reproductive period [32.8 (5.4) vs tive analysis of this gene in indepen- ported. Written informed consent 35.6 (5.1) years], although because of dent large populations would be was obtained from all study partici- the small number of individuals, this required. 1164 Letters

We are deeply grateful to the post- 2 Servicio de Ginecologıá violet detection was performed at menopausal women who participated y Obstetricia 250 nm. in this study. We are very grateful to Hospital Universitario We prepared buffer containing Eva Molero, Antonio Gońzalez, and Virgen de las Nieves citric acid (40 mmol/L) and cetyltri- Rocıó Pascual for their collaboration Granada, Spain methylammonium bromide (0.8 during this work. The authors F.J. Mo- mmol/L), adjusted to pH 4.4 with ron, M.E. Saez, and A. Ruiz, have de- 3 Servicio de Ginecologıá ␥-aminobutyric acid (2), filtered and clared that conflicts of interest exist. y Obstetricia sonicated for 0.5 min before use. At Some of the work described here is Hospital San Juan de Alicante the beginning of each working day, subject to patent filings for diagnostics Alicante, Spain the capillary was washed for 2 min purposes. Neocodex has been partially with water and separation buffer, funded by the Ministerio de Educacioń 4 CEOGA and also washed between runs for y Ciencia of Spain (PTQ2003-0546, Clıńica de Ginecologıá 1 min with separation buffer. We PTQ2003-0549, PTQ2003-0783) and the ␮ European Commission (Gendisrupt Lugo, Spain prepared a 100 mol/L aqueous project: QLK4-CT-2002-02403). mixture of inosine monophosphate 5 Servicio de Ginecologıá (IMP) and inosine triphosphate Clıńica Sanatorio Bilbaıńo (ITP). The compounds could be sep- Bilbao, Spain arated within 0.8 min with separa- Grant/funding support: None declared. tion efficiency up to 1 200 000 theo- Financial disclosures: None declared. 6 Clıńica Diatros Gava´ retical plates/m in a mixture and Barcelona, Spain 300 000 theoretical plates/m in biological samples (Fig. 1). References We collected blood samples in 1. Di Pasquale E, Beck-Peccoz P, Persani L. Hy- *Address correspondence to this au- EDTA tubes from healthy blood do- pergonadotropic ovarian failure associated with thor at: Departamento de Geno´mica Es- an inherited mutation of human bone morpho- nors and patients undergoing aza- genetic protein-15 (BMP15) gene. Am J Hum tructural. Neocodex. C/ Charles Darwin thioprine therapy (who gave in- Genet 2004;75:106–11. no.6, Acc. A. Parque Tecnolo´gico Isla de formed consent). Erythrocytes were la Cartuja, 41092-Sevilla, Spain. Fax 34- 2. Moron FJ, de Castro F, Royo JL, Montoro L, Mira separated by centrifugation (1200g, E, Saez ME, et al. Bone morphogenetic protein 955-047325; e-mail [email protected]. 15 (BMP15) alleles predict overresponse to re- 5 min) and washed twice with combinant follicle stimulation hormone and iatro- 3 volumes of NaCl, 9 g/L. We lysed genic ovarian hyperstimulation syndrome (OHSS). DOI: 10.1373/clinchem.2006.081307 ␮ Pharmacogenet Genomics 2006;16:485–95. 200 L of erythrocytes with 1 mL of 3. Di Pasquale E, Rossetti R, Marozzi A, Bodega ice-cold distilled water, centrifuged B, Borgato S, Cavallo L, et al. Identification of the lysates at 5000g for 10 min, and new variants of human bmp15 gene in a large ␮ cohort of women with premature ovarian failure. mixed 25 L of supernatant with 100 J Clin Endocrinol Metab 2006;91:1976–9. Determination of ITPase Activity by mmol/L Tris buffer (150␮L, pH 9.0), 4. Dixit H, Rao LK, Padmalatha VV, Kanakavalli M, Capillary Electrophoresis 10 mmol/L dithiothreitol (10 ␮L), Deenadayal M, Gupta N, et al. Missense muta- and 1 mol/L MgCl (10 ␮L). The tions in the BMP15 gene are associated with 2 ovarian failure. Hum Genet 2006;119:408–15. To the Editor: mixture was preincubated for 5 min 5. Laissue P, Christin-Maitre S, Touraine P, Shipkova et al. (1) recently reported at 37 °C, after which 40 mmol/L ITP Kuttenn F, Ritvos O, Aittomaki K, et al. Mu- (10 ␮L) was added and incubated for in Clinical Chemistry that liquid chro- tations and sequence variants in GDF9 and 15 min at 37 °C (2). The samples BMP15 in patients with premature ovarian fail- matography can be used to deter- ure. Eur J Endocrinol 2006;154:739–44. were deproteinized with 20 ␮Lof mine the activity of inosine triphos- trichloroacetic acid, 1 mol/L, soni- phate pyrophosphohydrolase (ITPase) Francisco J. Moroń1 cated (30 s), and centrifuged at 5000g 2 in erythrocytes. We find that ITPase Nicola´s Mendoza for 1 min. The supernatant was in- Francisco Quereda3 can also be measured by capillary jected into the capillary or stored at Francisco Va´zquez4 electrophoresis. Ϫ50 °C. We measured hemoglobin Reposo Ramı´rez-Lorca1 We performed capillary electro- (Hb) in the lysate with a Radiometer Juan Velasco1 phoresis on a Beckman P/ACE 5510 ABL 725 (Diamond Diagnostics). Jose L. Gallo2 with a diode array detector. Electro- The signal-to-noise ratio was Ͼ6at Ana Salinas1 phoretic separations were carried out 2.0 nkat/g Hb [7.2 ␮mol of IMP/(g Txantoń Martıńez-Astorquiza5 in an uncoated silica capillary (20 cm Hb ⅐ h)]. The calibration curve was Rafael Sańchez-Borrego6 effective/27 cm total length, 50 ␮m linear from 0.01 to 10 mmol/L (y ϭ Maria E. Saéz1 internal diameter; Polymicro CE and 11.3x Ϫ 0.6 nkat/g Hb; r ϭ 0.9974). Agustıń Ruiz1* CEC Technologies) at a constant volt- We evaluated recovery and impreci- age of Ϫ30 kV (1111 V/cm). We set sion by assaying erythrocytes with 1 Departamento de the data rate of the detector at 16 Hz. the added mixture of IMP. Recover- Geno´mica Estructural Neocodex Samples were loaded by a low-pres- ies were 85%, 83%, 76%, 75%, and Sevilla, Spain sure injection (0.5 psi, 6 s). Ultra- 80% for 0.06, 0.25, 0.54, 1.80, and 3.00 Clinical Chemistry 53, No. 6, 2007 1165

method for nucleotide analysis in cells: application on inherited metabolic disorders. Electro- phoresis 2007;28:373–80. 3. Marinaki AM, Ansari A, Duley JA, Arenas M, Sumi S, Lewis CM, et al. Adverse drug reactions to azathioprine therapy are associated with polymorphism in the gene encoding inosine triphosphate pyrophosphatase (ITPase). Pharmacoge- netics 2004;14:181–7. 4. Duley JA, Simmonds HA, Hopkinson DA, Levin- sky RJ. Inosine triphosphate pyrophosphohydrolase deficiency in a kindred with adenosine deaminase deficiency. Clin Chim Acta 1990; 188:243–52.

David Friedecky1* Jana Tomkova2 Tomas Adam1

1 Laboratory for Inherited Metabolic Disorders Department of Clinical Biochemistry University Hospital and Palacky University Olomouc Olomouc, Czech Republic

2 Department of Analytical Chemistry Palacky University Olomouc Olomouc, Czech Republic

*Address correspondence to this author at: Laboratory for Inherited Meta- Fig. 1. Capillary electrophoretic analyses of IMP. bolic Disorders, University Hospital, I. P. Pavlova 6, 775 20 Olomouc, Czech Re- (a), control sample from a healthy donor (b), and a sample from a patient with ITPase deficiency (c). *, impurity from ITP. public. Fax 420-5-88442509; e-mail david. [email protected]. mmol/L added IMP (n ϭ 6), respec- nkat/g Hb, which agrees with the DOI: 10.1373/clinchem.2007.086058 tively. The lower recoveries reflect previously published data (3). coprecipitation of IMP with proteins, Availability of alternative methods which agrees with previously pub- is important because available ana- Novel Mutation (c.G1124A) in Exon 9 lished findings (1). Imprecision val- lytical equipment varies among lab- ues (as CV, n ϭ 10) were 2.1%, 1.2%, oratories. Measurements of ITPase of the APOB Gene Causes Aberrant and 1.0% (within-day CV) and 4.2%, are important because ITPase defi- Splicing and Familial 3.2%, and 2.4% (between-day CV) for ciency may alter 6-mercaptopurine Hypobetalipoproteinemia 0.06, 0.54, and 3.00 mmol/L addi- (azathioprine) metabolism, leading tions of IMP, respectively. The repro- to adverse reactions (3), and the de- To the Editor: ducibility values (CV) of migration ficiency affects Ͼ10% of the popula- Familial hypobetalipoproteinemia times for 10 samples from healthy tion (4). (FHBL) is commonly caused by mu- volunteers were 0.92%, 2.8%, and tations in the apolipoprotein B gene 2.5% for run-to-run, sample-to-sam- (APOB). The APOB gene encodes 2 Grant/funding support: This study ple, and between-day measurements proteins, apolipoprotein (apo) B-48 was supported by Grant MSM (n ϭ 10), respectively. Because of the and apo B-100. Apo B-48 is formed in 6198959205 from the Ministry of Edu- use of acidic separation medium, we cation, Youth and Sports (Czech Re- the intestine and is essential for the observed no interference during the public). formation and recognition of dietary analysis of samples from 80 healthy Financial disclosures: None declared. derived chylomicrons, and apo B-100 blood donors and 20 patients under- is found in VLDLs and LDLs of he- going azathioprine therapy. References patic origin and is involved in the With this simple capillary electro- 1. Shipkova M, Lorenz K, Oellerich M, Wieland E, endogenous transport of triglycer- von Ahsen N. Measurement of erythrocyte ino- phoresis method, we estimated a ref- sine triphosphate pyrophosphohydrolase (ITPA) ides, cholesterol, and fat-soluble vita- erence interval (n ϭ 80, 38 males and activity by HPLC and correlation of ITPA geno- mins. A number of abnormally trun- 42 females) for healthy white indi- type-phenotype in a Caucasian population. Clin cated apo B proteins have been Chem 2006;52:240–7. viduals of 13.3–112.2 nkat/g Hb 2. Friedecky D, Tomkova J, Maier V, Janostakova A, described, and by convention are re- (5%–95%), with a median of 57.8 Prochazka M, Adam T. Capillary electrophoretic ferred to by a centile system reflect- 1166 Letters

ing their apparent Mr in relation to Indeed, the programs SpliceView 650-bp product was visualized in the apo B-100 (1). (http://l25.itba.mi.cnr.it/ϳwebgene/ wild-type, and a larger product of Truncations shorter than apo B-27 wwwspliceview.html) and NNSplice 690 bp was observed in the mutant are not expressed in lipoproteins, (http://www.fruitfly.org/seq_tools/ (Fig. 1B). DNA sequencing of the and those shorter than apo B-75 are splice.html) predicted that the G3A 690-bp and 650-bp bands revealed underrepresented in LDL (2–4). mutation would abolish splicing at that the increase in size of the mutant Consequently homozygous muta- the normal donor splice site of intron product reflected the inclusion of the tions in the N-terminal third of 9, and activate a cryptic donor site 40 first 40 bp of intron 9. A cryptic APOB result in the virtual absence of bp into the intron. Gene Splicer (http: donor splice site between c.1124 ϩ 40 both apo B-48 and apo B-100 and //www.tigr.org/tdb/GeneSplicer/ and c.1124 ϩ 41 was activated in the their corresponding lipoproteins, gene_spl.html) also predicted abol- mutant construct, and the normal and thus very low concentrations of ishment of the normal donor site but intron 9 acceptor site was used. Pre- plasma triglycerides, cholesterol, and did not predict the usage of a cryptic dictably, this message results in a the fat-soluble vitamins. This condi- splice site. frame shift in the translated pro- tion is known as FHBL and is char- To confirm these in silico predic- tein, a substitution of serine 348 to acterized clinically by failure to tions we performed minigene ex- lysine, and the insertion of 92 new thrive, steatorrhea, and eventually pression studies. A minigene con- amino acids before a premature stop both central and peripheral neuro- struct spanning exons 8–11 and the is encountered at residue 440 logical abnormalities (1). Heterozy- intervening sequences was cloned (Ser348LysfsX93). The resulting mu- gotes usually experience a milder into the pcDNA3.1/V5-His TOPO tant protein, a truncated apo B-9.7, phenotype or are asymptomatic. TA vector and then transfected into would not be viable for lipoprotein We report a novel APOB mutation, COS-7 cells. After 48 h the mRNA formation. In vivo the majority of the identified in a family with low total was isolated and reverse transcrip- transcripts would be expected to use cholesterol and apo B concentrations tion PCR was performed. The cDNA the cryptic splice site in intron 9, in plasma. The proband, a 64-year- was amplified using primers within creating the truncated apo B-9.7 and old man, had an LDL cholesterol exon 8 and exon 11. The expected causing the observed FHBL. concentration of 1.4 mmol/L and an apo B concentration of 0.39 g/L, and his 2 daughters both had LDL cholesterol concentrations Յ0.5 mmol/L, and apo B concentrations Ͻ0.35 g/L. In the mother, the concentrations of these analytes were within reference intervals. Western blotting of plasma from all 4 individuals showed no apo B truncations. DNA sequencing of the exons and exon/intron boundaries of the APOB gene revealed a novel heterozygous c.G1124A mutation in the proband and his 2 daughters, which was not present in the mother. No other APOB gene mutations were identified. The c.G1124A mutation predicts a p.Ser348Asn substitution in the ␤␣1 domain, which is essential for lipoprotein assembly. The p.Ser348Asn substitution may affect the structure or function of this domain but is predicted to be benign, according to Polyphen (http://www.polyphen. com), with a position-specific independent counts difference score of 0.675. Alternatively the mutation at the ultimate nucleotide of exon 9 could affect splicing at the adjacent Fig. 1. Pre-mRNA splicing of the G1124A minigene construct. intron 9 donor splice site, with vari- (A), diagram of the minigene construct and the potential splicing outcomes from the G1124A mutation. (B), ous potential splicing outcomes 2% agarose gel showing the wild-type (WT) and mutant (MUT) cDNA products arising from splicing of the (Fig. 1A). minigene construct in COS-7 cells. Two different-sized bands are visible at 650 and 690 bp, respectively. Clinical Chemistry 53, No. 6, 2007 1167

From this analysis, we have shown Circulating Cell-Free Placental mRNA cases of MCDA-T were complicated that the novel c.G1124A mutation in the Maternal Plasma as a by TTTS at the time of blood sam- causes FHBL by disrupting splicing. Predictive Marker for Twin-Twin pling, TTTS subsequently developed We identified 3 family members who Transfusion Syndrome in 5 cases (TTTS group), but not in were heterozygous for this mutation the remaining 12 cases (no-TTTS but were largely asymptomatic be- To the Editor: group). Gestational ages at diagnosis cause each still had 1 normal APOB Twin-twin transfusion syndrome of TTTS were 15–25 weeks. The 3 allele. (TTTS), which is a serious complica- groups had no significant differences This case highlights the difficulty tion in monochorionic diamniotic in population characteristics, includ- of interpreting novel mutations iden- twins (MCDA-T), involves unequal ing the maternal age, the number of tified in diagnostic laboratories and blood flow via the placental vascular nulliparous women, and the gesta- the need for a clear strategy to deter- anastomoses from the donor to the tional age at the time of sampling mine their significance. If sufficient recipient twin. Although the placen- (data not shown). family members are not available, tal anastomoses are present in all The blood samples (8 mL) from linkage analysis may be uninfor- MCDA-T and both fetuses are genet- each woman were collected into mative and functional analysis is ically identical, TTTS occurs in only an EDTA tube, and the plasma essential. 15% of MCDA-T, and much of the sample was stored at Ϫ20 °C until pathophysiological basis of TTTS re- use. After cf-mRNA was extracted from maternal plasma, a quantitative Grant/funding support: National mains poorly understood. Clinically, Heart Foundation of New Zealand and a staging system based on the ultra- 1-step real-time RT-PCR assay was Foundation for Research, Science and sound features of TTTS is widely performed using an ABI 7900T Se- Technology Contract grant CNTX0501 used for the management (1) but not quence Detector (Perkin-Elmer) as (V.M.H.). for the prediction of TTTS. In addi- described previously (4). Primer sets Financial disclosures: None declared. tion, the known predictive findings and TaqMan probes for each gene observable by ultrasonographic ex- and single-strand, and synthetic References amination are detectable only in a DNA oligonucleotides from each 1. Havel RJ, Kane JP. Disorders of the biogenesis and secretion of lipoproteins containing the B small portion of TTTS cases (2). New amplicon used for a calibration curve apolipoproteins. In: Scriver CR, Beaudet AL, Sly predictive markers are therefore de- were prepared as described previ- WS, Valle D, eds. The Metabolic Basis of Inher- sirable for the early detection and ously (4). Then, plasma concentra- ited Disease. New York: McGraw-Hill Book Co., 1995:1853–86. prevention of TTTS. Recently, pla- tions of cf-mRNA for human PL and 2. Talmud PJ, Krul ES, Pessah M, Gay G, Schonfeld cental mRNAs, such as human pla- for glyceraldehyde-3-phosphate de- G, Humphries SE, et al. Donor splice mutation cental lactogen (PL) and some other hydrogenase (GAPDH) were mea- generates a lipid-associated apolipoprotein B-27.6 in a patient with homozygous hypobeta- hormones were detected in maternal sured and converted into multiples lipoproteinemia. J Lipid Res 1994;35:468–77. plasma, and concentrations of each of the median (MoM) of the controls 3. Young SG, Hubl ST, Smith RS, Snyder SM, marker were measured with quanti- adjusted for gestational age, as de- Terdiman JF. Familial hypobetalipoproteinemia caused by a mutation in the apolipoprotein B tative real-time reverse transcription scribed previously (5). The differ- gene that results in a truncated species of (RT)-PCR (3, 4). Thus, circulating ences between the TTTS and the no- apolipoprotein B (B-31): a unique mutation that cell-free mRNA (cf-mRNA) in mater- TTTS groups were evaluated with helps to define the portion of the apolipoprotein B molecule required for the formation of buoyant nal plasma has become an attractive the Mann–Whitney U-test. Signifi- triglyceride-rich lipoproteins. J Clin Invest 1990; target for the noninvasive monitor- cant difference was defined as a P 85:933–42. ing of pregnancy disorders (3, 5). value Ͻ0.05. 4. Welty FK, Hubl ST, Pierotti VR, Young SG. A truncated species of apolipoprotein B (B67) in a The purpose of the present study The median (minimum–maxi- kindred with familial hypobetalipoproteinemia. was to investigate the use of cf- mum) cf-PL mRNA MoM values J Clin Invest 1991;87:1748–54. mRNA concentration in maternal were 1.80 (0.89–3.81) in the TTTS- plasma as a predictive marker of group, 1.14 (0.77–1.35) in the no- Vivienne M. Homer* later TTTS. The study participants TTTS group, and 1.00 (0.82–2.05) in Peter M. George included 17 pregnant women who the control group, respectively. At Molecular Pathology Unit visited the Obstetrics Clinic of Na- adjusted gestational age the cf-PL Canterbury Health Laboratories gasaki University Hospital at 12–21 mRNA concentration was signifi- Christchurch, New Zealand weeks of gestation for management cantly higher in the TTTS group than of their pregnancy with MCDA-T. in the no-TTTS group (Mann–Whit- Included as a control group were 135 ney U-test,Pϭ 0.035), whereas there * Address correspondence to this au- singleton pregnant women without was no significant difference of cf-PL thor at: Canterbury Health Laboratories, medical complications at similar ges- mRNA concentration between the Cnr Tuam Street and Hagley Avenue, Christchurch 8001, New Zealand. Fax tational age. All of the participants no-TTTS group and the control 0064 33640 545; e-mail: Vivienne.homer@ gave written informed consent, and group (P ϭ 0.41; Fig. 1). In addition, cdhb.govt.nz. the study was approved by the Re- the median cf-GAPDH mRNA MoM search Ethics Committee of Nagasaki value in the maternal plasma was DOI: 10.1373/clinchem.2007.086496 University. Although none of the 17 significantly higher in the TTTS 1168 Letters

3 Solution Oriented Research for Science and Technology, Fig. 1. Box and whis- Japan Science and Technology kers plots of cf-PL Agency, Kawaguchi, Japan MoM distribution in the TTTS group, no- TTTS group, and * Address correspondence to this au- control group. thor at: Department of Obstetrics and Gynecology, Nagasaki University Gradu- The median (minimum– ate School of Biomedical Sciences, 1-7-1 maximum) cf-PL mRNA MoM values were 1.80 Sakamoto, Nagasaki 852-8501, Japan. (0.89–3.81) in the Fax 81-95-849-7365; e-mail kiyonori@ TTTS group, 1.14 nagasaki-u.ac.jp. (0.77–1.35) in the no- TTTS group, and 1.00 DOI: 10.1373/clinchem.2007.087890 (0.82–2.05) in the control group. * P ϭ 0.035, ** P ϭ 0.41.

Toenails: An Easily Accessible and Long-Term Stable Source of DNA for Genetic Analyses in Large-Scale group (2.20; range 1.30–2.68) than in ence and Technology from the Japan Epidemiological Studies the no-TTTS group (1.09; range 0.68– Science and Technology Agency. 3.25; P ϭ 0.045). Our results sug- Financial disclosures: None declared. gested the possibility that unappar- Acknowledgements: We thank Drs. Ta- To the Editor: ent pathophysiological changes had dayuki Ishimaru, Joseph Wagstaff, Yo- Molecular tools are increasingly ap- already occurred in the women who shisada Shibata, Akira Fujishita, and plied in epidemiological studies to subsequently developed TTTS, al- Makoto Murakami for their help and unravel the relationship between en- though which specific conditions led valuable advice. vironmental exposures and disease to the increased mRNA in the mater- (1). DNA is required for analyses of nal plasma in the TTTS group remain References genetic factors, such as polymor- 1. Quintero RA, Morales WJ, Allen MH, Bornick PW, unknown. Johnson PK, Kruger M. Staging of twin-twin phisms, but currently used speci- In conclusion, a quantitative aber- transfusion syndrome. J Perinatol 1999;19: mens, such as lymphocytes and buc- ration of both the cf-PL and cf- 550–5. cal cells, have disadvantages related 2. Jain V, Fisk NM. The twin-twin transfusion syn- to collection, transport, storage, and GAPDH mRNA in maternal circula- drome. Clin Obstet Gynecol 2004;47:181–202. tion may be a novel predictive 3. Dennis Lo YM, Chiu RW. Prenatal diagnosis: processing of samples. A relatively marker for TTTS, although both sta- progress through plasma nucleic acids. Nat Rev infrequently used source of DNA tistical differences were small and Genet 2007;8:71–7. that may overcome these problems is 4. Ng EK, Tsui NB, Lau TK, Leung TN, Chiu RW, the sample size was too small to give Panesar NS, et al. mRNA of placental origin is nail material. Human toenails have sufficient strength to the analysis. readily detectable in maternal plasma. Proc Natl been collected in several epidemio- Therefore, a combination of several Acad Sci U S A 2003;100:4748–53. logical studies, predominantly for 5. Purwosunu Y, Sekizawa A, Koide K, Farina A, determination of trace elements as cell-free placental mRNA markers Wibowo N, Wiknjosastro GH, et al. Cell-free could be effective for the prediction mRNA concentrations of plasminogen activator biomarkers for the intake of these of TTTS, similar to the situation for inhibitor-1 and tissue-type plasminogen activator compounds (2). Until now, none of are increased in the plasma of pregnant women tumor markers. Further study to with preeclampsia. Clin Chem 2007;53:399– these epidemiological studies have identify gene transcripts that are ex- 404. applied human toenails as a source pressed only in the placenta and not of DNA. 1* in blood cells may help to both pre- Kiyonori Miura We investigated whether toenail 1 dict and prevent TTTS and also may Kentaro Yamasaki material collected 20 years ago in the 1,3 further elucidate the pathophysiol- Shoko Miura Netherlands Cohort Study on Diet 2,3 ϭ ogy of this serious complication. Koh-ichiro Yoshiura and Cancer (NLCS) (3) (n 120 852) Takako Shimada1 could be used as a source of DNA for Daisuke Nakayama1 analyses of multiple genetic poly- Norio Niikawa2,3 morphisms. Approximately 90 000 Grant/funding support: K.M. and 1 N.N. were supported in part by Hideaki Masuzaki participants provided toenail clip- Grants-in-Aid for Scientific Research pings (on average, 80 mg per partic- 1 (Nos. 19791155 and 17019055, respec- Departments of Obstetrics and ipant) (2). We optimized a protocol 2 tively) from the Ministry of Education, Gynecology and Human Genetics for DNA isolation from ϳ10 mg toe- Sports, Culture, Science and Technol- Nagasaki University Graduate School nail material, based on the method of ogy of Japan, and N.N. was supported of Biomedical Sciences Cline et al. (4), and tested the suit- by Solution Oriented Research for Sci- Nagasaki, Japan ability of this DNA in 2 PCR-based Clinical Chemistry 53, No. 6, 2007 1169

Table 1. DNA yield and multiplex genotyping results. DNA yield, ng/10 mg toenail Success rate of Corresponding buccal Genotype Toenails multiplex genotypingb swabs availablec identicald Materiala n Mean (SD) Range n (%) n n (%) Fresh material, Ͻ3 months old 11 2102 (669) 1147–3114 11 (100%) NAe NLCS material, 20 years old Irradiated 24 2013 (1167) 1156–7037 23 (96%) 23 22 (96%) Not irradiated 33 2173 (1623) 950–8929 31 (94%) 12 12 (100%) Total NLCS 57 2106 (1439) 950–8929 54 (95%) 35 34 (97%) a Material: NLCS material, toenail material collected in the Netherlands Cohort study on Diet and Cancer; Irradiated: toenail material irradiated with neutrons to determine selenium content (Instrumental Neutron Activation Analysis). b Samples in which all 10 polymorphisms could be identified were classified as successful. c Number of toenail samples for which a comparison with buccal swab samples was feasible. d Individuals for whom all 10 polymorphisms in the toenail sample were identical to those in the buccal swab sample. e NA, Not applicable.

assays in a subgroup of the cohort A summary of quantitative results 596-bp amplicon was unsuccessful (n ϭ 57) for which buccal DNA was of the multiplex genotyping assays for 60% of the samples, indicating also available. In the 1st assay, 10 on the DNA samples is provided in increased fragmentation of the single nucleotide polymorphisms Table 1. A genotype profile of 10 DNA. Irradiation probably causes were amplified in a multiplex PCR polymorphisms was successfully fragmentation of the DNA, an effect reaction and subsequently geno- generated for 100%, 96%, and 94% of that must be taken into account typed by means of single base exten- the DNA samples isolated from when the available nail material sion using primer extension and au- fresh, 20-year-old irradiated, and 20- has previously been used for trace tomated capillary gel electrophoresis year-old not irradiated toenail mate- element analyses, which require ir- as described by Knaapen et al. (5).A rial, respectively (amplicon sizes 92– radiation. 2nd PCR-based test was used to in- 148 bp). The success rate using the In conclusion, we showed that 20- vestigate the maximum length of buccal swab DNA samples was year-old nonirradiated and irradi- fragments that could be amplified. A 100%. Because DNA isolation was ated toenails can be a source of DNA portion of the toenail samples had successful for only 90% of the buccal for state of the art high-throughput been irradiated with neutrons for swab material in the NLCS samples, genetic analyses of polymorphisms. analyses of selenium content (not ir- the use of nail material as a source of For existing large-scale epidemiolog- radiated: n ϭ 33; irradiated: n ϭ 24). DNA resulted in a higher rate of ical studies, our results demonstrate To assess possible effects of age of successful outcomes. For 35 individ- that toenail material can be used for the toenails on the quality of the uals, we compared the outcome of genetic analyses in cohorts for which DNA, freshly harvested toenail ma- the genotyping assay for both their no other source of DNA is available. terial from healthy non-NLCS volun- buccal swab DNA and toenail mate- The use of toenails as source of DNA teers was investigated (n ϭ 11) (see rial DNA. Surprisingly, for 1 person, may be of considerable relevance in the Data Supplement that accompa- the 2 sources of DNA resulted in future molecular epidemiological nies the online version of this letter at different genetic profiles. This result studies, because toenail clippings can http://www.clinchem.org/content/ could not be related to technical er- be stored for long periods at low vol53/issue6 for a detailed descrip- rors, but may be from the switching costs while DNA quality remains tion of material and methods). or mislabeling of a sample during the constant. DNA was successfully isolated collection process 20 years ago. from all toenail samples. On average, The NLCS study received ap- 2 ␮g DNA could be extracted from 10 proval by the Medical Ethical Com- Grant/funding support: This study mg toenail material (Table 1). There mittees of the University Hospital was supported by the Network of Ex- was no significant difference in DNA Maastricht and TNO Quality of Life, cellence “Environmental Cancer Risk, quantity between the various groups Zeist, The Netherlands. Study partic- Nutrition and Individual Susceptibil- of toenail samples. The large range in ipants gave informed consent. ity” (ECNIS), which operates in the DNA yield was mainly due to 2 Results from the 2nd PCR test context of the 6th European Union outliers. The minimum amount iso- showed that DNA isolated from 20- Framework Program for Research and ␮ lated was almost 1 g DNA/10 mg year-old (not irradiated) toenail ma- Development (FP6). A.M.K. is sup- toenail material, which is sufficient terial or from fresh toenail material ported by a postdoctoral fellowship for more than 10 multiplex genotyp- could be amplified to Յ596 bp. With from the Netherlands Organization for ing analyses (80 ng of DNA is suffi- DNA from 20-year-old irradiated Scientific Research (NWO, VENI-Grant cient per analysis). nail material, generation of the 916.46.092). 1170 Letters

Financial disclosures: None declared. scale prospective cohort study on diet and can- 1 Department of Health Risk Analysis Acknowledgments: We thank Lucy cer in The Netherlands. J Clin Epidemiol 1990; 43:285–95. and Toxicology and van de Vijver and Erik Busink from 4. Cline RE, Laurent NM, Foran DR. The fingernails 2 TNO Quality of Life for the collection of Mary Sullivan: developing reliable methods for Department of Epidemiology of buccal swabs; Kim Wouters from the selectively isolating endogenous and exogenous NUTRIM, Department of Pathology of the Uni- DNA from evidence. J Forensic Sci 2003;48: 328–33. Maastricht University, Maastricht, versity Hospital Maastricht; and Prof. The Netherlands Harry Ostrer from the Departments of 5. Knaapen AM, Ketelslegers HB, Gottschalk RW, Janssen RG, Paulussen AD, Smeets HJ, et al. Pediatrics (Genetics) and Pathology Simultaneous genotyping of nine polymorphisms 3 Department of Food and Chemical and Medicine (Genetics) of the New in xenobiotic-metabolizing enzymes by multiplex York University Medical Center for PCR amplification and single base extension. Risk Analysis, TNO Quality of Life, DNA isolation from the buccal swabs. Clin Chem 2004;50:1664–8. Zeist, The Netherlands Simone G. van Breda1 References 2 1. Perera FP, Weinstein IB. Molecular epidemiol- Janneke G. Hogervorst * Address correspondence to this au- ogy: recent advances and future directions. Car- Leo J. Schouten2 thor at: Department of Epidemiology, Maastricht University, PO Box 616, 6200 cinogenesis 2000;21:517–24. Ad M. Knaapen1 2. van den Brandt PA, Goldbohm RA, van ’t Veer P, 1 MD Maastricht, The Netherlands. Fax 31- Bode P, Dorant E, Hermus RJ, et al. A prospec- Joost H. van Delft 43-3884128; e-mail PA.vandenBrandt@ tive cohort study on selenium status and the risk R. Alexandra Goldbohm3 EPID.unimaas.nl. of lung cancer. Cancer Res 1993;53:4860–5. Frederik J. van Schooten1 3. van den Brandt PA, Goldbohm RA, van ’t Veer P, DOI: 10.1373/clinchem.2006.085043 Volovics A, Hermus RJ, Sturmans F. A large- Piet A. van den Brandt2*

Correction The article by John Middleton and Jeffrey E. Vaks entitled “Evaluation of Assigned Value Uncertainty for Complex Calibrator Value Assignment Processes: A Prealbumin Example” (Clin Chem 2007;53:735–41), contains several errors. In the first section of Results, page 739, the text should be changed from “The IWC value assignment results are shown in Table 1” to “The IWC level 8 results are given in Table 1”. Also, in Table 2, for IWC5 to IWC8, the decimal point in the uncertainty SD’s should be shifted one digit to the left. Additionally, in Table 3, footnote b, the reference interval should be changed from “. . . based on width of 20 mg/L of the. . . 18–38 mg/L” to “. . . based on width of 200 mg/L of the...180–380 mg/L.” Also, in Table 3, the superscript “a” should be removed from the 180 and 380 mg/L table entries. Finally, in the 4th paragraph in Results (page 739), the text should be changed from “. . . we made 48 measurements (4 instruments, 12 replicates per sample)...to“...wemade 96 measurements (4 instruments, 24 replicates per sample)....”Theauthors regret the errors. DOI: 10.1373/clinchem.2007.090571 Book, Software, and Web Site Reviews

Review of The Metabolic Syn- and explanation of the pathophysiol- epidemiologists who are investigat- drome. Christopher D. Byrne and ogy is quite strong, and the chapters ing the metabolic syndrome and Sarah H. Wild, eds. Chichester, West on adipocytokines and inflammation seeking to better understand the ep- Sussex, England: John Wiley & Sons are clear and well written, including idemiology and pathophysiology of Ltd., 2005, 432 pp., $110.00, hard- the description of the interaction the syndrome that puts so many in- cover. ISBN 0-470-02511-5. between lipids and inflammation, dividuals at risk for cardiovascular which followed previous descrip- and endocrine consequences. This book ambitiously reviews a sub- tions by Peter Libby. The fetal origins ject that is undergoing shifts in pop- theory is well explained and the ularity at a rate equaled only by the chapter on genetic predisposition is Grant funding/support: Dr. de Fer- speed of identification of relevant accessible to the nongeneticist. The adipocytokines and inflammatory ranti is supported by a grant from the reader may find, as I did, that the Sandra A. Daugherty Foundation and markers. Nevertheless, the editors diagrams, presented in most chap- tackle the controversial topic with by a Children’s Hospital Faculty Ca- ters to clarify the relationships be- reer Development Fellowship Award. ease and competence. The collection tween elements of and causal factors Financial disclosures: None declared. highlights the underpinning ele- for metabolic syndrome, do not con- ments of the metabolic syndrome— tribute to understanding of interac- Sarah D. de Ferranti excess visceral adipose tissue foster- tions that may be too complicated to ing poor responses to oxidative be easily summarized in a schematic. Department of Cardiology stress and inflammation, which This book will best serve research- Children’s Hospital Boston and worsen and are worsened by insulin ers wishing to gain a deeper under- Harvard University resistance and lead to vascular dys- standing of the metabolic syndrome, School of Medicine function. Although the clinician eas- rather than clinicians. The chapter Boston, MA ily recognizes the phenotype of devoted exclusively to treatment, abdominal obesity with its accompa- however, does give useful advice DOI: 10.1373/clinchem.2006.079640 nying complications, the details of about pharmacotherapy, including the pathophysiology are exceedingly the broad recommendation to use complex, and the text serves to clar- medications that address more than Clinical Diagnostic Technology- ify underlying themes, if not to ren- one aspect of the metabolic syn- The Total Testing Process, Volume der them simple, which they are not. drome but do not worsen other facets 3: The Postanalytical Phase. Kory M. Cardiovascular disease and diabetes (e.g., avoiding beta blockers in favor Ward-Cook, Craig A. Lehmann, are the tip of the iceberg of cardiom- of ace inhibition). The sections ad- Larry E. Schoeff, and Robert H. Wil- etabolic disease—exemplified in the dressing dietary treatment of meta- liams, eds. Washington, DC: AACC cover art—and the high and increas- bolic syndrome now seem outdated Press, 2006, 216 pp., $65 ($52.00 ing prevalence of metabolic syn- because they do not include newer AACC members), softcover. ISBN drome predicts the burgeoning of information, published subsequent 978-1-59425-055-2. cardiovascular disease. to this book. For example, recent The epidemiology of metabolic studies such as the Women’s Health This book, volume 3 of a series, ad- syndrome is well presented in the Initiative have shown that low-fat dresses the postanalytical phase of first few chapters, including the in- diets may not be superior to other the testing process. Volumes 1 and 2 fluence of ethnicity and the global diets in reducing cardiovascular discover the preanalytical and analytical burden of metabolic syndrome, ease. Another shortcoming of the phases, respectively. All 8 chapters which the editors estimate at 10%– book is the unfortunate dearth of of this last volume are interesting in 23% of adults worldwide, based on consideration of the difficult but es- principle, but strictly speaking only published prevalence data and in- sential issue of implementing life- chapters 2 and 4 and parts of chap- formed estimations for which data style changes so crucial to reversing ters 1 and 3 deal with the postana- are not available. Metabolic syn- or stabilizing the syndrome and pre- lytical testing phase. Because of the drome in adolescence is only tan- venting its consequences. Exercise as focus on US laboratories found in gentially referred to, despite the a therapeutic intervention is not very most of the chapters, parts of the frightening rise in pediatric obesity. well covered. book are less relevant to readers out- Although the definition of the meta- Several books on the metabolic side of the US. bolic syndrome and the surrounding syndrome have been or will soon be The 1st chapter provides an over- controversy is nicely covered in the published, including clinical hand- view of the postanalytical phase, first chapter, the question of whether books and several edited texts quality goals, and quality control. the syndrome exists at all (i.e., (Reaven 1999, Moller 1993, Levine Quality systems and general man- whether the collection of disorders 2006). The Byrne and Wild collection, agement are treated in more detail. connotes a risk greater than the sum however, clearly summarizes the rel- This chapter provides important ref- of the individual abnormalities) re- evant research and is an excellent erences for the reader who wishes to ceives little attention. Consideration resource for clinical researchers and further pursue the material. Chapter

Clinical Chemistry 53, No. 6, 2007 1171 1172 Book, Software, and Web Site Reviews

2, one of the better chapters in the in the 21st century and “what has and therapeutic aspects of cardiac book, deals with verification and au- changed the last 3 years” (since the and systemic diseases. toverification of test results, detailing publication of Volume 1 of this book Chapter 1 summarizes the histori- the use of delta-checks in a very series). This chapter describes the cal background of the discovery of educational way, and ends by de- major health challenges faced by the the cardiac natriuretic peptide sys- scribing the possibilities for using US in this century and some of their tem. Chapter 2 summarizes current middleware. Chapter 3 deals with potential consequences for labora- concepts regarding the physiology of evidence-based laboratory medicine tory medicine. the heart and may be too detailed for (EBLM), of which some parts are In conclusion, parts of this book many laboratorians. Chapter 3 does postanalytical. In well-written short are very interesting. Unfortunately, an excellent job of presenting the paragraphs, this chapter gives an however, the chapters do not fit very complete physiology, pathophysiol- overview of the different steps in well together and only a few of them ogy, biochemistry, and molecular bi- EBLM from the preanalytical to the strictly deal with the “postanalytical ology of the natriuretic peptide fam- postanalytical phase, and it lists phase”. ily, including the noncardiac family valuable web resources in 2 tables. I members C-type natriuretic peptide think, however, that the difficult sub- Sverre Sandberg and urodilatin. Chapter 4 is of par- ject of EBLM implementation should ticular interest for the clinical chem- have been dealt with more exten- Laboratory of Clinical Biochemistry ist and is written very well. This sively. Chapter 4 addresses distribu- Haukeland University Hospital chapter reflects the great experience tion of laboratory results, discussing Bergen, Norway of chapter coauthor Mauro Pan- among other things the speed of de- teghini, who is a member of several livery and advantages and disadvan- DOI: 10.1373/clinchem.2006.080499 international marker standardization tages of printed, electronic, and tele- committees. It reviews the method- phone reports. The section on ological aspects of natriuretic pep- laboratory information systems is too tide testing with a focus on the com- Natriuretic Peptides: The Hormones short to give much valuable informa- mercially available B-type natriuretic of the Heart. Aldo Clerico and tion, but this shortcoming is compen- peptide (BNP) and N-terminal Michele Emdin, eds. Trento, Italy: sated for by the last paragraph, proBNP (NT-proBNP) assays. The Springer-Verlag Italia, 2006, 184 pp., which lists many resources on labo- recently described glycosylation of $149.00, hardcover. ISBN 978-88-470- ratory informatics, most of them in NT-proBNP is not mentioned, how- 0497-9. the US. Chapter 5 deals with auto- ever. Chapters 5 and 6 nicely review mated digital cell morphology. It is This is an excellent book by one of clinical considerations and applica- puzzling that the book includes a the major groups currently involved tions of natriuretic peptide testing in whole chapter on this topic, and that in cardiac natriuretic peptide re- cardiac and noncardiac diseases, in- the chapter deals almost solely with search. This book is directed to all cluding the effects of age and sex, as CellaVision. Alternative solutions readers seeking a deeper under- well as obesity, on reference limits. should have been discussed, such as standing of the cardiac natriuretic The official recommendations of car- different digital cell morphology peptide system (physiologists, cardi- diological and clinical chemistry so- EQA systems used in clinical pathol- ologists, and clinical chemists). The cieties on routine natriuretic peptide ogy and hematology. Chapter 6 deals backgrounds of the editors (clinical testing are included. Chapter 7, with the important topic of electronic physiology, cardiology, and endocri- aimed at clinicians, addresses the health informatics networks. The nology) strongly influence the con- cardiac natriuretic hormone system chapter is not focused on laboratory cept of this book; the principal aim as a target for therapy, including the medicine, however, and is not spe- is to demonstrate that the heart is a fact that clinical trials on the use of cific to the postanalytical phase, as multifunctional and interactive or- natriuretic peptides to treat heart or would be natural in this book. Chap- gan that exchanges information with renal failure and on drugs targeting ter 7 deals with biosensors, minia- the nervous, endocrine, and immune natriuretic peptide metabolism have turization, and noninvasive tech- systems. Evidence for the endocrine so far been somewhat disappointing. niques. It starts with a short section function of the heart revolutionized The book ends with “inconclusive” on method comparison, including the traditional mechanical concep- remarks on the past, present, and among other things Bland-Altman tion of this organ and unveiled new future of natriuretic peptides. plots and Deming regression. This and fascinating perspectives that Overall the editors have succeeded section is too short to be of much include the area of laboratory diag- in compiling an excellent book that value for the reader. Thereafter, prin- nosis. This book reviews current would be an important addition to ciples of biosensors are described in knowledge on the physiology and the libraries of those who wish to an educational way, and this section pathophysiology of cardiac endo- gain a deeper understanding of na- is well worth reading. The last chap- crine function and provides updated triuretic peptides and the endocrine ter deals with clinical laboratories information on diagnostic, clinical, function of the heart. It also summa- Clinical Chemistry 53, No. 6, 2007 1173

rizes and reviews all relevant aspects ment of a coagulation disorder be- its covers. I was so encouraged by for the clinical laboratorian. fore the final diagnosis is established, my initial good impressions that I can interfere with results of coagula- decided to submit the Quick Guide to Johannes Mair tion testing to the extent that correct a limited, real-life trial. I have asked diagnosis may not be achieved even pathology residents to use the book Department of Internal Medicine with appropriate test selection. It is in their daily practice and refer to it Clinical Division of Cardiology not uncommon for clinicians to be when they provide consultation for Innsbruck Medical University confused about ordering coagulation coagulation testing. This “trial by Anichstrasse 35 tests and interpreting results. At our fire” revealed that the Quick Guide to A-6020 Innsbruck, Austria teaching hospital we have had to Coagulation Testing is a convenient resort to regularly reviewing coagu- source of information that can expe- DOI: 10.1373/clinchem.2006.082578 lation test requisitions for appropri- dite patient care and accelerate learn- ateness, and we routinely provide ing. This small volume provides in- guidance in the form of consults ini- formation that may not be easily tiated by the pathology department accessed at one place during routine Quick Guide to Coagulation Test- to guide clinicians with test selection clinical pathology consultations. ing. Marisa B. Marques and George as well as anticoagulant therapy Even after more extended evalua- A. Fritsma, Washington, DC: AACC monitoring. Cheat sheets, diagrams, tions I found very few reasons to Press, 2006, 62 pp., $15 ($12 AACC and abbreviated procedures are of- criticize this book. There are the oc- members), softcover. ISBN 1-59425- ten used in such consultations, and casional omissions such as when the 049-9. the need for easy access to coagula- text on “Hematocrit adjustment for tion test interpretation has become hemostasis specimens” recommends In the past decades, significant ad- mandatory. I have been searching for removing sodium citrate solution vances have been made in under- a publication that can be used as a from the collection tube, but does not standing coagulation disorders, in- resource with such consultation, but explain how this can be done while cluding thrombophilias and bleed- I have been unable to find one until maintaining sterility or without ing diatheses. Anticoagulant therapy recently. breaking the vacuum. Other facts has advanced just as rapidly, leading Quick Guide to Coagulation Testing that I would have liked to be in- to the development of several new is the right book for such a purpose. cluded are a more detailed descrip- classes of drugs that may be used to It is a spiral-bound booklet that con- tion of prothrombin or partial throm- treat those coagulation disorders. tains only 60 small pages, and it is boplastin time mixing studies and Our theoretical understanding of co- intended to be carried in a coat interpretation of their results, or a agulopathies has been paralleled by packet while making rounds or see- table listing the various factor VIII– analytical developments to help es- ing patients in the clinic. It can also containing preparations that lists tablish accurate clinical diagnoses be used by laboratorians providing those also containing factor IX. These and to aid monitoring of new thera- coagulation consults to clinicians. additions could have helped the peutic interventions. The book is divided into 7 main reader select appropriate treatment Clinical practitioners have recog- chapters, describing specimen collec- for hemophilia A or B from a long list nized these coagulopathies and have tion, therapeutic ranges, reference in- of similar products. On the other been administering the new antico- tervals, and hemostasis and coagula- hand, parts of the book cover topics agulants, thus driving the need for tion test menus as well as guidelines thoroughly, such as the short chapter more complex coagulation testing in for thrombophilia testing, anticoagu- on von Willebrand Disease (vWD) clinical laboratories. Unfortunately, lant therapy monitoring, manage- that includes 3 tables comparing clin- the clinicians’ understanding of se- ment of bleeding patients, and man- ical manifestations and laboratory lection and interpretation of the ap- agement of platelet disorders. The profiles of vWD as well as expected propriate coagulation tests has not book has been written with a very concentrations of vWF in various progressed as rapidly. The principal practical purpose in mind; to provide ABO blood groups. Another advan- reason may be that diverse coagula- instantaneous reference to the most tage is that the book contains recom- tion disorders present with only a commonly encountered questions in mendations that are in synch with limited number of apparently similar coagulation testing. Essential refer- the most recent practice recommen- clinical symptoms. Often the exact ences are also provided at the end of dations, such as when describing diagnosis can be elucidated by test the book; however, the list is not management of warfarin overdose or panels only, or by sequential appli- extensive. The reader should per- discussing the role of ADAMTS-13 (a cation of multiple tests during which form his or her own literature search disintegrin and metalloproteinase each subsequent test is selected if more in-depth references are with a thrombospondin type 1 motif, based on results of the preceding needed. member 13) testing for the diagnosis analyses. The presence or absence of I have found the book to fulfill its of vWD. comorbid conditions, such as an promise. It is well organized, and If I could make a wish regarding acute illness, pregnancy, or the treat- information is easy to locate within this book I would ask the authors to 1174 Book, Software, and Web Site Reviews

include in the next edition a table starting anticoagulant therapy, not book in their everyday practice when describing how various anticoagu- realizing that the acute condition and information is needed quickly. I also lants, such as warfarin and heparin, therapy may render results uninter- feel that this book is an excellent or various clinical conditions, such as pretable. A comprehensive list of as- teaching aid for faculty and should recent thrombosis and pulmonary say interferences could be very valu- be an essential resource for residents embolism, interfere with the individ- able in deciding whether to perform in a clinical pathology rotation. ual coagulation tests. Further break- or defer testing in these circum- down by method (such as activity stances. assay vs genetic analysis) could be In conclusion, I can very highly Geza Bodor very useful to provide alternatives in recommend the Quick Guide to Coag- clinical situations when rapid testing ulation Testing to practicing clinical University of Colorado Health may be required. In my experience it pathologists, pathology residents, Sciences Center and Denver Health is very common for a clinician who laboratory professionals, and clini- Medical Center, Denver, CO 80204 suspects a hypercoagulation disor- cians who see patients with coagula- der to order all sorts of laboratory tion disorders. Even specialists in co- DOI: 10.1373/clinchem.2006.080515 investigations immediately after agulation can benefit from using this The Clinical Chemist Compiled by David E. Bruns, Editor ([email protected])

AACC Award Recipients, 2007 Area Committee for Clinical Chem- and Preyer Children’s Hospital. Dr. istry and Toxicology for 5 years. Mu¨ ller’s major research interests are AACC Award for Outstanding Dr. Miller also serves as chair of related to purine metabolism and Contributions to Clinical Chemistry, the Laboratory Working Group of clinical and applied biochemistry. Sponsored by Siemens Medical Solutions Diagnostics the National Institutes of Health/ His research has covered a broad National Kidney Disease Education scope of subjects, from early studies W. Greg Miller Jr., PhD, DABCC, Program. He is active on commit- of various enzymes, to studies of FACB, is professor of pathology, di- tees of the College of American murine metabolism in skeletal mus- rector of clinical chemistry, and di- Pathologists, American Diabetes cle and endothelial cells, to his recent rector of pathology information sys- Association, National Glycohemo- explorations of laboratory diagnosis tems at Virginia Commonwealth globin Standardization Program, in transplantation medicine and the University in Richmond. and IFCC. He has contributed to use of tumor markers. He has pub- training 26 graduate students and lished some 300 scientific papers, is postdoctoral fellows in clinical author or co-author of more than 250 chemistry and has published 100 abstracts, and has edited or co-edited papers, book chapters, and mono- 7 books and proceedings. He serves graphs in areas of analytical mea- on the editorial boards of several surement procedures and inter- leading journals. He has served the laboratory standardization and Austrian Society of Clinical Chemis- harmonization. try as secretary, treasurer, vice president, and president. He is currently AACC Award for Outstanding president of the Austrian Society of Contributions through Service to the Quality Assurance and Standardiza- Profession of Clinical Chemistry, tion. He was general secretary of the Sponsored by Beckman Coulter, Inc. organization for 11 years, and thus Mathias M. Mu¨ ller, MD, is pro- was the main person responsible for fessor of medical chemistry at the the development of the Austrian pro- University of Vienna, Austria. ficiency testing program covering the whole field of clinical diagnostics. He served as president of the European Society for Study of Purine and Py- rimidine Metabolism in Man. He has served the International Federation W. Greg Miller Jr. of Clinical Chemistry and Laboratory Medicine (IFCC) in many capacities, He has served the AACC in sev- including as vice president and pres- eral capacities. He is currently a ident. He is currently past-president. member of the board of editors of He negotiated and signed the collab- Clinical Chemistry and is the AACC oration agreement between the Clin- liaison to the International Organiza- ical and Laboratory Standards Insti- tion for Standardization Technical tute and the IFCC. He initiated the Committee 212 for Clinical Labora- global IFCC campaign for disease tory Testing and In Vitro Diagnostic management on diabetes mellitus Test Systems. He has chaired the and other programs. Lipids and Lipoproteins Division and the Standards Committee. He AACC Award for Outstanding has been on the faculty of the Re- Contributions in Education, Sponsored view Course in Clinical Chemistry by Quest Diagnostics Nichols Institute since its inception in 1990. Dr. Miller has also been active in several other Craig A. Lehmann, PhD, professional organizations. He is Mathias M. Mu¨ller C(NRCC), FACB, is dean and a pro- currently treasurer of the Clinical fessor at the School of Health Tech- and Laboratory Standards Institute He recently retired as director of nology and Management at Stony (CLSI) and a member of several of its the Institute of Laboratory Diagnos- Brook University in Stony Brook, subcommittees. He chaired the CLSI tics at Kaiser Franz Josef Hospital N.Y. In addition to these duties, he is

Clinical Chemistry 53, No. 6, 2007 1175 1176 The Clinical Chemist

cation, workflow analysis, laboratory genase (SCHAD), another enzyme of economics, e-health, point-of-care test- fatty acid oxidation. Dr. Bennett has ing, and disease management. also devoted 25 years to studying a group of untreatable and devastating AACC Award for Outstanding inherited neurodegenerative diseases Contributions to Clinical Chemistry in known collectively as Batten disease. a Selected Area of Research, He recently identified a novel anti- Sponsored by Olympus America Inc., Diagnostic Systems Group neuronal apoptosis pathway in one of these diseases, which may lead to Michael J. Bennett, PhD, FRC- the 1st rational therapeutic interven- Path, DABCC, FACB, is professor of tion. He has published more than 225 pathology and laboratory medicine peer-reviewed scientific papers, re- at the University of Pennsylvania views, and book chapters and has and director of the metabolic disease been involved with the organization of laboratory at The Children’s Hospi- numerous national and international tal of Philadelphia. congresses. Dr. Bennett reviews manu- scripts for more than 30 journals and is currently on the editorial boards of Clinical Chemistry Journal and Molecu- Craig A. Lehmann lar Genetics and Metabolism.

AACC Award for Outstanding currently interim executive dean for Scientific Achievements by a Young the Health Sciences Center. Investigator, Sponsored by Roche His 40-year affiliation with Stony Diagnostics Brook University began in 1967 as a Loralie J. Langman, PhD, FCACB, member of the chemistry department DABCC, DABFT, is director of the and includes a 9-year stint chairing drug/toxicology laboratory at the the clinical laboratory services divi- Mayo Clinic in Rochester, MN. sion. Among his many duties, his philanthropic endeavors this year alone have generated $1 million for Michael J. Bennett student scholarships. Dr. Lehmann’s teaching has been recognized by He also holds the Evelyn Willing many awards, including the State Bromley Endowed Chair in Clinical University of New York’s coveted Laboratories and Pathology at The Chancellor’s Award for Excellence in Children’s Hospital of Philadelphia. Teaching, the President’s Award for The main focus of Dr. Bennett’s Excellence in Teaching, and the Pro- research has been the investigation vost’s Award for Exceptional Ser- of inborn errors of mitochondrial vice to Undergraduate Education. energy metabolism with a special Dr. Lehmann served for 3 years on emphasis on disorders of fatty acid the editorial board of the AACC’s metabolism. He was among the Strategies and Solutions. He has been a first to describe the fatal clinical phe- member of the editorial board of notype and the first to identify neo- Clinical Laboratory Sciences since 1987. natal metabolite abnormalities in Loralie J. Langman In addition to authoring more than medium-chain acyl-CoA dehydroge- 65 journal articles, Dr. Lehmann is nase (MCAD) deficiency. These ob- She has produced more than 30 the author, editor, or co-editor of 5 servations led to the expansion of publications and 40 abstracts/pre- clinical laboratory science textbooks. newborn screening by tandem mass sentations at national and interna- Dr. Lehmann has also produced a spectrometry, in which most new- tional meetings. She has many areas 4-part educational video entitled, borns are now screened for MCAD of expertise in toxicology, particu- “Automation in the Clinical Labora- deficiency and a number of other in- larly in antipsychotic drugs, drugs of tory.” He has delivered more than 125 born errors of metabolism. He is cur- abuse (particularly amphetamine- presentations worldwide on clinical rently studying the hyperinsulinism type stimulants), post-mortem toxi- analysis by computer-assisted infrared associated with deficiency of short- cology, and pharmacogenetics. Her spectroscopy, clinical laboratory edu- chain L-3-hydroxyacyl-CoA dehydro- current research interests include Clinical Chemistry 53, No. 6, 2007 1177

pharmacogenomics of amphetamine- on the national board, and as chair of cist in charge” from 1986 to 1995. A type stimulants, and genotype–phe- several committees. He was also a University of Tennessee professor notype relationships of psychoactive member of the organizing committee since 1983, he chaired the depart- medications. She is currently the only of the joint national meeting of CLAS ment of clinical pharmacy from individual to have achieved diplomate and the National Academy of Clini- 1983 to 1991. For the past 30 years, status in all three disciplines offered by cal Biochemistry in 1998. He is a life his research at St. Jude has focused the American Board of Clinical Chem- member of the Pakistan Society of on the pharmacokinetics, pharma- istry: clinical chemistry, molecular di- Chemical Pathologists. Dr. Syed codynamics, and pharmacogenom- agnostics, and toxicological chemistry. plans to travel to his native Pakistan ics of anticancer agents in children. Dr. Langman serves on committees for to visit public and private clinical For this research, he has received 3 several professional organizations, in- laboratories in both urban and rural consecutive National Institutes of cluding the AACC, Society of Forensic settings, to assess the quality of test- Health MERIT Awards from the Toxicologists, American Academy of ing, and to promote the concepts of National Cancer Institute. The ma- Forensic Sciences, National Academy total quality management. He will jor thrust of his pharmacogenomics of Clinical Biochemistry, Canadian So- also explore the possibility of starting research is acute lymphoblastic leu- ciety of Clinical Chemists, Interna- a postdoctoral fellowship program in kemia in children. Dr. Evans has tional Association of Therapeutic Drug clinical chemistry at the University of authored more than 300 articles and Monitoring and Clinical Toxicology, Health Sciences at Lahore. book chapters, has edited several and New York Academy of Sciences. textbooks and scientific journals, The AACC Lectureship Award, and has received several national AACC International Travel Fellowship, Sponsored by an educational grant awards for his research. He was Sponsored by BD Diagnostics - from Siemens Medical Solutions elected to the Institute of Medicine Diagnostics Preanalytical Systems of the National Academy of Sci- Danyal B. Syed, PhD, DABCC, William E. Evans, PharmD, has ences in 2002. FACB, is a consultant and laboratory been director and chief executive of- director at William F. Ryan Com- ficer of St. Jude Children’s Research AACC Award for Outstanding Hospital in Memphis, TN, since 2004. Contribution for a Publication in the munity Health Center in New York, International Journal Clinical NY. Chemistry, Sponsored by Siemens Medical Solutions Diagnostics

Yuk-Ming Dennis Lo, MD, is professor of chemical pathology and the Dr. Li Ka Shing Professor of Medi- cine at The Chinese University of Hong Kong.

Danyal B. Syed

A member of the AACC for the past 28 years, Dr. Syed has served the New York Metro Section as chair, William E. Evans member of the executive board, and chair of the education committee. He He holds the St. Jude Endowed served on the organizing committee Chair at the University of Tennessee of the Northeast regional alliance Colleges of Pharmacy and Medicine. meetings (LAB MED) in 2004 and He has held many positions at each 2005. He has served the Clinical Li- institution. From 1986 to 2002, he gand Assay Society (CLAS) as presi- chaired the pharmaceutical depart- dent of the New York Metro Chapter, ment at St. Jude and was “pharma- Yuk-Ming Dennis Lo 1178 The Clinical Chemist

He is also the associate dean for Leung, Tze K. Lau, Stephen S.C. cancer initiation and progression. He research of the faculty of medicine, Chim, Grace T.Y. Chung, Kypros H. is also conducting research to vali- the director of the Li Ka Shing Insti- Nicolaides, and Y.M. Dennis Lo. Clin date multiparametric panels for early tute of Health Sciences, and associate Chem 2006; 52:12, 2211–8. ovarian, breast, and prostate cancer director of the state laboratory in diagnosis. Dr. Diamandis serves on oncology in South China. Dr. Lo’s The Morton K. Schwartz Award for the boards of 25 journals. He has main research interests lie in the bi- Significant Contributions in Cancer published more than 400 original pa- Research Diagnostics, Sponsored by pers and holds 13 patents, with an- ology and diagnostic applications of Siemens Medical Solutions cell-free DNA and RNA in plasma. Diagnostics other 20 pending. He co-authored a In 1997, Dr. Lo discovered the pres- recent textbook, Tumor Markers. For ence of cell-free fetal DNA in the Eleftherios P. Diamandis, MD, many years, he has run workshops plasma of pregnant women. The PhD, FRCPC, DABCC, FACB, is di- on tumor markers and proteomic work that led to this award-winning vision head of clinical biochemistry technologies at the AACC annual paper originated from his demon- in the department of pathology and meetings. stration that epigenetic markers can laboratory medicine at Mount Sinai be used for the detection of fetal Hospital; biochemist-in-chief at the The Edwin F. Ullman Award, DNA in maternal plasma. Unlike ge- University Health Network and To- Sponsored by Dade Behring Inc. ronto Medical Laboratories; and di- netic markers, in which no single Pratap Singh, PhD, is a research vision head of clinical biochemistry marker can be informative in all fellow and principal scientist in the in the department of laboratory med- pregnancies, epigenetic markers assay development group of a Dade icine and pathobiology at the Univer- have the potential to be used as uni- Behring facility in Glasgow, Dela- sity of Toronto in Ontario, Canada. versal fetal DNA markers. In 2005, ware. Dr. Lo developed the first such marker, SERPINB. In his award-winning paper, Dr. Lo demonstrated that the tumor suppressor gene RASSF1A exhibits a pattern of methylation which is reverse to that of SERPINB5, namely hypermethylated in the placenta but hypomethylated in maternal blood cells. This pattern of methylation allowed Dr. Lo to develop a simple methylation-sensitive restriction-enzyme–mediated real-time polymerase chain reaction assay for the detection of the hypomethylated RASSF1A gene. This RASSF1A assay could be implemented as a fetal DNA control for the performance of fetal RhD genotyping from maternal Eleftherios P. Diamandis plasma. The concepts explored in this paper also have implications for the development of other fetal epige- Dr. Diamandis has been active in Pratap Singh netic markers for noninvasive prena- the field of cancer diagnostics over tal diagnosis. Dr. Lo has pioneered a the past 20 years. He currently chairs number of other important applica- the National Academy of Clinical He has been on the staff of Dade tions of plasma nucleic acids; he has Biochemistry effort to develop Behring and its predecessor firms published more than 230 articles in guidelines for the clinical use of tu- since 1984. His research interests in- international journals. The winning mor markers. His main research include bio-conjugation, protein modi- paper is: “Hypermethylated RASS- terests are tumor markers, especially fication and characterization, and or- FIA in Maternal Plasma: A Universal a group of enzymes called human ganic synthesis. In the early 1990s, he Fetal DNA Marker that Improves the tissue kallikreins. His most recent conceived the use of dendrimers, Reliability of Noninvasive Prenatal research focuses on proteomic meth- well-defined nanoscopic synthetic Diagnosis.” K.C. Allen Chan, Chun- odologies for identifying novel can- polymers, in controlling the orienta- ming Ding, Ageliki Gerovassili, Sze cer biomarkers and the physiology of tion of immobilized immunore- W. Yeung, Rossa W.K. Chiu, Tse N. kallikrein enzymes, as they relate to agents for assays requiring very low Clinical Chemistry 53, No. 6, 2007 1179

detection limits and excellent preci- Broomfield, CO., that provides con- responsible for the prenatal and new- sion. He and his team demonstrated sulting services for health-care qual- born screening testing programs of the feasibility of this concept using ity management. For the past decade the state of California in Richmond, polyamidoamine dendrimers and she has provided laboratory clients CA. showed it was possible to reproduc- and professional organizations in ibly achieve highly specific and sensi- Europe, Asia, and North, Central, tive assays for important analytes such and South America with training in as creatine kinase MB subunit, thyrox- quality management systems, assis- ine, and human thyroid-stimulating tance in quality management imple- hormone. This work led to the devel- mentation, development and control opment of the Stratus CS, an auto- of laboratory documents, and devel- mated clinical analyzer able to analyze opment of training and competence 6 key analytes, including key cardiac assessment programs. She has 15 markers such as cardiac troponin I and years experience as an American As- N-terminal pro-brain natriuretic pep- sociation of Blood Banks (AABB) in- tide. These and other sensitive assays spector. She was a charter member of on the Stratus CS have found great the AABB committee that developed utility in emergency room and critical- the “Quality System Essentials” that care settings, where quick decisions have become the backbone of the for cardiac and other patients are AABB’s standards and accreditation needed. Dr. Singh is a co-inventor of 7 programs. She served as workshop patents relating to clinical diagnostics faculty and audioconference speaker and a contributing author of a chapter for numerous continuing education in a book on dendrimers. He has pub- programs and is a member of the John E. Sherwin lished more than 20 research articles in AABB Quality Management Sub- the fields of clinical and organic chem- committee of the Standards Com- His previous positions include istry. mittee. She serves on the CLMA Pa- directing laboratories at hospitals tient Safety and Quality Committee, in Chicago and Fresno, CA. He has AACC Award for Outstanding Clinical Laboratory Contributions to Patient which recently developed the “Good held positions as technical director Safety, Sponsored by Ortho-Clinical Laboratory Practices” website, where and chief operating officer of multi- Diagnostics laboratories can contribute and ob- million-dollar reference laboratories tain ideas for improving patient and been a consultant to a variety Lucia M. Berte, MA, MT(ASCP) safety. Ms. Berte chaired and partic- of businesses and physician prac- SBB, DLM; CQA, CMQ/OE(ASQ), is ipated in several subcommittees and tices. He has more than 50 publi- the president of Laboratories Made work groups of the Clinical and Lab- cations to his credit, primarily in his Better! P.C., a company based in oratory Standards Institute, with main research area of pediatric la- particular attention to guidelines for boratory medicine. He has served implementing a laboratory quality the AACC in many capacities with management system. She was a both the Chicago and Northern Cal- charter faculty member of the Amer- ifornia local sections. Among his ican Society of Clinical Pathology’s many contributions at the national (ASCP) teleconference program in level, he has chaired the Commis- the early 1980s, and achieved high sion for Education and Scientific Af- scores as a faculty member of ASCP’s fairs, the Commission on Publica- Workshops for Laboratory Profes- tions, and the Public Relations Pro- sionals. She continues to contribute gram. He considers his greatest articles, chapters, and books about accomplishment as AACC president quality management and patient to be the establishment, with the safety. generous support of Gopal Savjani, of the Past Presidents’ Scholarship AACC Past President’s Award, fund, which is designed to help Sponsored by Cardinal Health, Scientific Products Distribution young clinical chemists receive the quality postdoctoral training that John E. Sherwin, PhD, DABCC, will keep the AACC and the field FACB, is currently the acting chief of of laboratory medicine vibrant for Lucia M. Berte the Genetic Disease Branch that is years to come. He also successfully 1180 The Clinical Chemist

completed the merger of the Na- Meeting Notice Fax: 248-551-3694; email: dcrisan@ tional Academy of Clinical Biochem- beaumont.edu) or visit the website at William Beaumont Hospital’s 16th istry with the AACC. He has served www.beaumont.edu/dnasymposium. Annual Symposium on Molecular Pa- many other organizations as well, For registration materials, contact: thology, DNA Technology in the including being president of the Na- Clinical Laboratory, will be held Sep- Sherrie Ode (Phone: 248-551-0429; Fax: tional Academy of Clinical Biochem- tember 26–28, 2007, at the Somerset 248-551-1163; Email: sode@beaumont. istry. He is also active in the Associ- Inn in Troy, Michigan. This 3-day edu) or register at www.beaumont. ation of Public Health Laboratories symposium begins Wednesday morn- edu/dnasymposium. and is a member of the governing ing (September 26). CME credit is council of the International Society available. Information: Contact Dr. DOI: 10.1373/clinchem.2007.091397 for Newborn Screening. Domnita Crisan (Phone: 248-551-7261;