Discovery and Validation of Protein Abundance Differences Between Follicular Thyroid Neoplasms

Research Article

Discovery and Validation of Protein Abundance Differences between Follicular Thyroid Neoplasms

Romana T. Netea-Maier,1 Stephen W. Hunsucker,3,6 Brigiet M. Hoevenaars,2 Steve M. Helmke,4 Pieter J. Slootweg,2 Ad R. Hermus,1 Bryan R. Haugen,4,5,6 and Mark W. Duncan3,4,6

Departments of 1Endocrinology and 2Pathology, Radboud University Nijmegen Medical Centre, Nijmegen, the Netherlands and 3Section of Pulmonary Medicine, Department of Pediatrics, 4Division of Endocrinology, Diabetes and Metabolism, Department of Medicine, 5Department of Pathology, and 6University of Colorado Cancer Center, School of Medicine at the University of Colorado at Denver and Health Sciences Center, Aurora, Colorado

Abstract solitary or within a multinodular goiter. Approximately 5% to 10% Distinguishing between benign follicular thyroid adenoma of adults have palpable thyroid nodules and 30% to 50% have (FTA) and malignant follicular thyroid carcinoma (FTC) by nodules identified by ultrasound. Although the majority of these cytologic features alone is not possible. Molecular markers are benign, approximately 5% to 7% of thyroid nodules are may aid distinguishing FTA from FTC in patients with malignant (1). Fine-needle aspiration biopsy (FNAB) is the most indeterminate cytology. The aim of this study is to define important diagnostic test in the initial evaluation of a patient with protein abundance differences between FTC from FTA through a thyroid nodule and offers a diagnostic accuracy of between f a discovery (proteomics) and validation (immunohistochem- 70% and 97% in experienced centers (2). Typically, 70% of FNAB istry) approach. Difference gel electrophoresis (DIGE) and are classified as benign, 4% are classified as malignant [predom- peptide mass fingerprinting were performed on protein inantly papillary thyroid carcinomas (PTC)], 2% to 10% supply extracts from five patients with FTC and compared with six insufficient sample, and the remainder are classified as either patients with FTA. Individual gel comparisons (i.e., each FTC indeterminate or suspicious (5–23%; refs. 1, 2). Typically, patients extract versus FTA pool) were also performed for the five FTC returning either indeterminate or suspicious results undergo patients. Immunohistochemical validation studies were per- diagnostic hemithyroidectomy or complete thyroidectomy to formed on three of the identified proteins. Based on DIGE exclude malignancy. images, 680 protein spots were matched on individual gels. It is particularly challenging to distinguish between thyroid Of these, 102 spots showed statistically significant differences neoplasms of the follicular type [i.e., benign follicular thyroid in abundance between FTC and FTA in the individual gel adenoma (FTA), malignant follicular thyroid carcinoma (FTC), analyses and were therefore studied further. Matrix-assisted and follicular variant of papillary carcinoma] based on cytologic laser desorption/ionization time-of-flight mass spectrometry examination alone. All these tumors have similar cytologic features was used to identify 54 of these protein spots. Three candidates and surgery is usually required to obtain a definitive tissue sample. involved in protein folding (heat shock protein gp96, protein However, because only 5% to 7% of the clinically identified nodules disulfide isomerase A3, and calreticulin) were studied by prove to be malignant, the indeterminate findings subject most immunohistochemistry. Moderate calreticulin immunohisto- patients to unnecessary surgery, potential risks, and, occasionally, chemical staining was the best single marker with a high irreversible complications. Improving the diagnostic accuracy of negative predictive value (88%); combining all three markers FNAB is therefore of crucial clinical importance. (any marker less than moderate staining) had the best positive Differentiated epithelial thyroid tumors represent a spectrum predictive value (75%) while still retaining a good negative of morphologically and biologically diverse neoplasms and the predictive value (68%). With DIGE, we identified 54 proteins molecular etiology and pathogenesis of thyroid carcinoma, differentially abundant between FTC and FTA. Three of these especially of the follicular type, is unknown (3, 4). Thyroid cancer were validated by immunohistochemistry. These findings is believed to result from the accumulation of oncogene mutations provide further insights into the diagnosis, prognosis, and or rearrangements (RAS, BRAF, RET, NTRK1, and MET) and pathophysiology of follicular-derived thyroid neoplasms. silencing of tumor suppressor genes (p53, RASSF1A, PTEN, PPARc, [Cancer Res 2008;68(5):1572–80] and CDK inhibitors; ref. 3). Recent data suggest that the so-called atypical FTA, which is characterized by high cellular density, Introduction mitoses, and a less regular cytologic pattern, may share genetic features with both FTCand PTC(5), but the progression of thyroid Thyroid cancer is the most common endocrine malignancy and adenoma to carcinoma has not been clearly shown. Therefore, its most frequent clinical presentation is as a thyroid nodule, either defining the differences in protein levels that distinguish between FTA and FTCwill provide additional insight in the earliest steps of follicular neoplasia transformation and might also deliver a Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/). clinical tool that could improve the diagnostic accuracy of FNAB B.R. Haugen and M.W. Duncan contributed equally to this work. in patients with indeterminate cytology. Requests for reprints: Bryan R. Haugen, Division of Endocrinology, Diabetes and Metabolism, Department of Medicine, University of Colorado Cancer Center, The aim of the present study was to define protein abundance University of Colorado at Denver and Health Sciences Center, MS 8106, P. O. Box differences between FTA and FTCtissue. We sought changes at the 6511, Aurora, CO 80045. Phone: 303-724-3921; Fax: 303-724-3920. E-mail: bryan. protein level for several reasons. First, many cellular processes are [email protected]. I2008 American Association for Cancer Research. regulated posttranscriptionally and mRNA studies are incapable doi:10.1158/0008-5472.CAN-07-5020 of determining some differences that may affect tumor biology.

Cancer Res 2008; 68: (5). March 1, 2008 1572 www.aacrjournals.org

Consequently, proteomics is seen as an essential tool to enhance Labeling reactions were carried out as previously described (7, 8). After our understanding of disease processes (6). Second, peptides and labeling, the samples were combined (e.g., one sample labeled with Cy5, one proteins can be measured by well-established methods with high sample labeled with Cy3, and the internal standard labeled with Cy2) and A sensitivity, precision, and accuracy. Any changes we observe have the mixture was taken to a final volume of 450 L with reaction buffer, hydroxyethyl disulfide (0.1 mol/L, 5.4 AL, Destreak, GE Healthcare), 1% the potential to form the basis of a sensitive and specific protein- broad range Pharmalytes 3-10 NL (GE Healthcare), and bromphenol blue based diagnostic test for follicular-derived thyroid neoplasms. (0.003%). Our studies used difference gel electrophoresis (DIGE) in After resuspension in the rehydration buffer, protein samples were combination with matrix-assisted laser desorption/ionization passively rehydrated into 24-cm immobilized pH gradient strips (IPG 3-10 time-of-flight mass spectrometry (MALDI-TOF MS). The DIGE NL, GE Healthcare) for 24 h and then focused (IPGphor System, GE component of this study allows the precise quantitative compar- Healthcare) for 66,000 Vh (analytic gels) or 133,000 Vh (preparative gels). ison of thousands of distinct proteins; MALDI-TOF MS provides Cysteine side chains were reduced and alkylated by incubating the focused identity of the proteins after they are excised from the gel (7). A strips (10 min, room temperature) in equilibration solution [6 mol/L urea, similar approach has been adopted by us and others in studies that 100 mmol/L Tris (pH 8.8), 30% glycerol, 2% SDS, 0.25% saturated aqueous aim to identify biomarkers (8, 9). bromphenol blue] containing 0.5% DTT followed by incubation in equilibration solution with 4.5% iodoacetamide (10 min, room temperature). Gel electrophoresis was performed on precast 8% to 16% acrylamide Materials and Methods gradient gels (Jule, Inc.) as previously described (8). Voltage and current Tissue samples for DIGE analysis. Snap-frozen tumor tissue samples were continuously monitored throughout all runs for quality control. from 11 patients undergoing surgery for follicular neoplasms were obtained Gels were scanned on a Typhoon 9400 Variable Mode Laser Imager through the Cooperative Human Tissue Network. At final histopathologic (GE Healthcare) at 100 Am resolution. Laser and filter settings for each of diagnosis, five were identified as FTC(three women, ages 31–75 years, the dyes were as follows: Cy3 (excitation, 532 nm; emission, 580 nm; tumor sizes of 3.5–8 cm) and six were FTA (all women, ages 29–58 years, bandpass, 30 nm), Cy5 (excitation, 633 nm; emission, 670 nm; bandpass, tumor sizes of 1.3–5.5 cm). In all FTCpatients, tumor capsular invasion was 30 nm), Cy2 (excitation, 468 nm; emission, 520 nm; bandpass, 40 nm), and present, and in two of these, the capsular invasion was extensive. In three Deep Purple (excitation, 532 nm; emission, 610 nm; bandpass, 30 nm). FTCpatients, vascular invasion was present, and in one of these, the DeCyder software (version 5.0; GE Healthcare) was used for spot vascular invasion was extensive. detection and relative quantification of protein spots on the fluorescence Preparation of tissue protein extracts. Protein was extracted from images. For each gel image, the DeCyder Differential In-gel Analysis each fresh-frozen tissue sample (50 mg) as previously described (8). software module was initially adjusted to detect an estimated number of Proteins were precipitated with methanol/chloroform (10), dried in a 2,500 spots. Individual spots at the extreme edges of the gel, extremely low SpeedVac, and then rehydrated overnight in 400 AL of reaction buffer intensity spots, and dust particles (i.e., those spots with a high slope) were [7 mol/L urea, 2 mol/L thiourea, 4% (w/v) CHAPS]. Each sample was excluded. Volumes were measured for each protein spot in the three then supplemented with 10 mmol/L DTT (20 AL of 200 mmol/L DTT in fluorescent channels (i.e., Cy3, Cy5, and Cy2). Individual DIGE gels were reaction buffer), homogenized with a small pellet pestle (Kimble Kontes), matched using the Biological Variation Analysis (BVA) software module and incubated for 2 h. Samples were thoroughly mixed and centrifuged (GE Healthcare). Spots matched on at least four of the five individual gels (16,000 g, 15 min, room temperature), and the solubilized protein were subjected to statistical analysis in BVA. Spot volumes of the Cy2 supernatants were collected. An aliquot was diluted 50-fold with water internal standard were used to calculate standardized volume ratios for the immediately before protein assay by the method of Bradford (11). Based on Cy5- and Cy3-labeled FTC and FTA protein spots. A Student’s t test was these findings, each sample was diluted to 5 mg/mL protein with reaction used to compare the differences in protein spot volumes between the FTC buffer containing 10 mmol/L DTT. Samples were flash frozen with liquid and the pooled FTA samples in the individual gel analysis. Statistical

N2 and stored at 80jCuntil analysis. significance was defined as P < 0.05 (two sided). DIGE experiment. Each analytic DIGE gel was composed of the Spots that showed a statistically significant difference in abundance following: 50 Ag of total protein isolated from an individual FTCsample between FTCand FTA were used to generate a list of candidate spots for (e.g., labeled with Cy5), 50 Ag of total protein from a pool prepared from all identification. These spots were matched on the preparative gel, excised, FTA samples (e.g., labeled with Cy3), and 50 Ag of total protein from a and subjected to in-gel enzymatic digestion and identification by MALDI- pooled internal standard. The FTA pool was created by combining equal TOF MS. Additional protein spots were also processed to serve as internal amounts of total protein isolated from individual FTA tissue samples. We molecular weight (MW) and isoelectric point (pI) markers. The positions of had a limited number of well-defined (histopathologically) snap-frozen FTA these markers were used to generate calibration curves for protein MW samples with some yielding limited amounts of total protein. We therefore (cubic spline) and pI (log linear) and to determine the observed pI and MW decided to pool protein from these tissues and compare the pool against for each protein spot. The measured MW and pI reported in Table 1 have an individual FTCsamples (a more heterogeneous group) rather than omit one approximate error of F20% of the predicted values and deviations larger FTA sample and randomly compare one FTA sample with one FTCsample. than this are likely the result of posttranslational modification. Predicted The internal standard, composed of an equal amount of total protein protein MW and pI were derived from the Swiss-Prot database7 using the isolated from all tissue samples (five FTCplus six FTA), was always labeled mature protein form (chain) when available. with Cy2 and included on every gel to improve quantitative precision and Protein identification by mass spectrometry. Protein spot excision enhance spot matching (12). The labeling of FTCand FTA samples was and in-gel enzymatic digestion were performed automatically by the Ettan reversed on alternate gels to minimize any dye bias. Spot Picker and Ettan Spot Digester (GE Healthcare) as previously Differentially abundant proteins were identified from preparative gels described (8). All digests were analyzed by MALDI-TOF MS (Voyager DE- containing 50 Ag of the pooled internal standard labeled with Cy2 and PRO, Applied Biosystems), again as described previously (13). Spectra were 950 Ag of unlabeled pooled internal standard. The inclusion of the Cy2- collected over the range m/z 500 to 5,000. Peptide mass fingerprints were labeled proteins is necessary to facilitate spot matching between analytic internally calibrated to monoisotopic trypsin peaks (i.e., m/z 515.33, 842.51, and preparative gels. Labeled and unlabeled proteins can have slightly 1,045.56, and 2,211.10). Spectra were processed using ProTS Data (Efeckta different migration behavior resulting from dye conjugation, and therefore, Technologies) to generate a peak list that was then submitted to Mascot gels were also poststained with Deep Purple (GE Healthcare) to visualize the corresponding unlabeled protein spots (see Supplementary Fig. S1). This method ensures that the unlabeled and labeled proteins are correctly matched and the desired protein spot is picked. 7 http://us.expasy.org/sprot/ www.aacrjournals.org 1573 Cancer Res 2008; 68: (5). March 1, 2008

Table 1. Proteins differentially abundant in FTC versus FTA tissue samples as determined by DIGE

c c Protein name Swiss- Spot Average Range of MW pI DMW from DpI from % Coverage Mascot Prot no. no.* intensity intensity (kDa) predicted predicted score ratio ratios (kDa)

Underabundant in FTC Cytokeratin 8 P05787 1 4.63 10.6 to 1.9 37.8 4.6 15.7 0.8 45 224 HSP gp96/endoplasmin P14625 2 3.61 19.5 to 2.21 98.5 4.6 8.3 0 18 192 78-kDa glucose-regulated protein (BiP; ER P11021 3 2.57 11.8 to 1.2 70.4 5.0 0 0 37 255 luminal Ca2+-binding protein grp78) Calreticulin P27797 4 2.26 9.6 to 1.6 46.4 4.2 0 0 36 148 Annexin A3 (lipocortin III) P12429 5 2.23 3.8 to 1.4 36.2 5.6 0 0 40 150 h-Actin P60709 6 2.05 6.4 to 1.2 37.8 5.3 3.7 0 32 125 PDI A3 P30101 7 1.99 7.3 to 1.25 51.0 5.7 3.1 0.1 31 165 PDI A3 P30101 8 1.96 6.9 to 1.4 50.7 5.5 3.5 0 24 141 Hexokinase-1 P19367 9 1.87 2.2 to 1.4 102.4 6.3 0 0 11 111 h-Actin P60709 10 1.79 4.2 to 1.2 37.8 5.1 3.7 0.1 24 82 PDI A3 P30101 11 1.75 6.1 to 1.4 50.5 5.4 3.6 0.1 35 196 Cathepsin B P07858 12 1.73 2.3 to 1.2 24.8 4.9 2.9 0.2 30 107 HSP gp96/endoplasmin P14625 13 1.66 2.5 to 1.1 97.9 4.8 7.7 0 17 96 Histone H2B P62807 14 1.64 3.9 to 1.0 13.6 10.3 0 0 52 85 Glucosidase 2 h subunit P14314 15 1.63 2.5 to 1.0 88.9 4.3 31.0 0 19 130 Macrophage capping protein P40121 16 1.59 2.8 to 1.0 37.8 6.0 0.6 0.1 21 76 Aminoacylase-1 Q03154 17 1.57 1.9 to 1.4 38.2 5.9 7.6 0.1 31 181 Annexin A5 (lipocortin V) P08758 18 1.57 2.1 to 1.0 34.0 4.7 1.7 0.1 20 101 HSP90h P08238 19 1.5 1.6 to 1.3 91.2 4.8 8.0 0.1 29 132 Cytosolic nonspecific dipeptidase Q96KP4 20 1.49 1.6 to 1.1 45.6 5.7 7.1 0.1 33 181 (glutamate carboxypeptidase-like protein 1) Proliferation-inducing gene 4 protein (mitofilin) Q16891 21 1.43 1.8 to 1.0 83.6 6.0 0 0 45 282 ER-associated HSP40 co-chaperone Q9UBS4 22 1.43 2.4 to 1.0 37.8 6.2 0.2 0.4 35 168 (DnaJ homologue subfamily B member 11) + Alcohol dehydrogenase [NADP ] P14550 23 1.38 1.9 to 1.1 37.8 6.9 1.4 0.6 45 176 (aldehyde reductase) Lamin A/C P02545 24 1.33 1.7 to 1.1 74.1 6.5 0 0 41 215 TCP-1-u (T-complex protein 1 subunit u) P50990 25 1.3 1.7 to 1.0 54.0 5.3 5.3 0 30 153 p100 coactivator (staphylococcal nuclease Q7KZF4 26 1.28 1.6 to 1.1 102.8 7.3 0.8 0.5 13 127 domain-containing protein 1) 26S proteasome non-ATPase regulatory subunit Q9UNM6 27 1.28 1.5–1.0 37.8 5.6 5.0 0.1 17 72 13 (26S proteasome regulatory subunit S1) Overabundant in FTC 14-3-3 protein g P61981 28 1.2 1.0–1.4 28.2 4.5 0 0.2 36 130 Tubulin h-1 chain (h-tubulin isotype I) P69893 29 1.21 1.0–1.3 40.2 6.5 9.3 1.7 19 65 h-Actin P60709 30 1.21 1.0–1.4 37.8 5.0 3.7 0.2 32 174 Peptidyl-prolyl cis-trans isomerase A P62937 31 1.21 1.1–1.4 16.9 8.1 0.9 0.2 35 86 (rotamase A) Pyridoxine-5¶-phosphate oxidase Q9NVS9 32 1.25 1.0–1.4 26.6 6.2 3.3 0.3 30 84 h-Actin P60709 33 1.32 1.0–1.5 37.9 5.1 3.6 0.1 34 164 Actin-related protein 2/3 complex subunit 2 O15144 34 1.36 1.0–1.8 33.5 7.0 0.8 0.2 21 80 Peroxiredoxin-2 (thioredoxin peroxidase 1) P32119 35 1.51 1.1–2.2 22.5 5.0 0.7 0.5 34 85 Nucleoside-diphosphate kinase 1 isoform b P15531 36 1.56 1.0–2.9 17.1 5.8 0 0 53 84 Dodecenoyl-CoA isomerase P42126 37 1.57 1.0–2.8 28.6 5.8 0.1 0.1 16 75 (3,2-trans-enoyl-CoA isomerase) Dihydrolipoamide succinyltransferase componentP36957 38 1.7 1.0–2.3 46.3 5.8 4.9 0 23 102 of 2-oxoglutarate dehydrogenaseb complex Cytokeratin 18 (424 AA) P05783 39 1.85 1.4–2.3 38.6 5.2 9.2 0 36 264 Cytokeratin 8 P05787 40 1.93 1.0–2.7 44.8 5.5 8.7 0 35 251 Cytokeratin 8 P05787 41 2.32 1.6–2.9 44.8 5.4 8.7 0.1 35 192 Histone H2B P62807 42 2.5 1.3–4.6 <13.0 4.8 ND 5.4 56 106 Cytokeratin 7 P08729 43 2.62 1.3–4.4 44.4 5.3 6.8 0.1 34 190

NOTE: MW or pI differences >20% from predicted are shown in bold. DMW and DpI are the differences between measured and calculated values. Abbreviation: ND, not determined. *Corresponds to the spot annotations recorded on the gel images in Fig. 1. cIndicates measured values. bComparisons of experimental and theoretical MW and pI and MS sequence coverage suggest that this is a fragment of cytokeratin 18.

Cancer Res 2008; 68: (5). March 1, 2008 1574 www.aacrjournals.org

(Matrix Science Ltd.) for database searching. Spectral preprocessing 16 spots, corresponding to 14 distinct proteins entities, were more included defining the baseline, noise, and signal-to-noise ratio as well as abundant (average fold changes, 1.20–2.62) in FTCversus FTA. monoisotopic peak selection. A signal-to-noise ratio in ProTS Data of >4 Figure 2 shows that most of the more abundant proteins in the was required for inclusion in the peak list. Database searches were FTCtumors are involved in cytoskeletal structure and cell conducted using the mammalian subset of the nonredundant protein organization, whereas nearly half of the proteins underabundant database (National Center for Biotechnology Information, database release 05/07/2006 with 446,224 mammalian sequences) and the Swiss-Prot in FTCfunction in protein synthesis and folding. For some of the h database (release 49.6 with 193,477 mammalian sequences). Other settings proteins (e.g., HSP gp96, PDI A3, cytokeratin 8, and -actin), several in ProTS included the following: peak amplitude, 100; peak width, 250; and distinct isoforms were shown to change in the same direction. chemical noise factor, 1.5. Settings in Mascot were as follows: peptide mass In three cases (cytokeratin 8, h-actin, and histone H2B), some tolerance of F100 ppm, fixed modification of carbamidomethylation of isoforms showed a lower abundance in the FTCsample, whereas cysteine side chains, and trypsin selected as the enzyme with one missed other isoforms showed a higher abundance in the FTCcompared cleavage accepted. Searches were not constrained by pI or MW. Minimum with the FTA samples (Table 1). requirements for positive protein identification were described previously Immunohistochemistry. Immunohistochemical validation (13) and peptide and protein assignments were made according to recently studies were performed on independent paraffin-embedded tissue published guidelines (14). samples from patients with benign and malignant follicular thyroid Immunohistochemistry. In accordance to the requirements of the local Medical Ethical Committee, all specimens used in this phase of the work tumors using antibodies against three of the identified proteins: were stripped of linked patient identifiers. HSP gp96, calreticulin, and PDI A3. We chose to study these We retrospectively selected archival tissue blocks from 16 patients with proteins based on their abundance, large volume ratio difference, FTC(5 widely invasive) and 18 patients with FTA who underwent thyroid absence of prior studies reporting their association with follicular surgery at the Radboud University Nijmegen Medical Centre (Nijmegen, the thyroid neoplasia, and the availability of commercial antibodies. Netherlands). Of the patients with FTC, preoperative FNAB was inconclu- Figure 3 shows the immunostaining of HSP gp96, calreticulin, sive in 10 patients (follicular cell proliferation) and suspect for carcinoma and PDI A3 on 18 FTA and 16 FTCtissue samples. The staining in 4 patients. In the remaining 2 patients, the FTCwas found incidentally intensity scores for the individual samples are presented in Table 2. after the patients had their goiter removed because of mechanical All three putative markers were underabundant in FTCbased on complaints. Four-micrometer-thick sections of the paraffin-embedded DIGE analysis. An optimal marker (or combination of markers) tissue samples were deparaffinized in xylene and rehydrated. Antigen retrieval was performed in 20 mmol/L citrate buffer (pH 6.0) following should identify most or all malignancies (high sensitivity/negative heating in a household microwave oven (10 min at 95jCfollowed by cooling predictive value), especially all widely invasive carcinomas, while down to room temperature) and brief washing in PBS. Endogenous minimizing the number of ‘‘benign’’ follicular adenomas subjected peroxidase blocking was performed in the PT Module (Lab Vision) using to surgery (high specificity/positive predictive value). Sensitivity

H2O2 in methanol for 10 min and rinsing the slides thrice in PBS (pH 7.4). analysis for these markers is shown in Table 3. Calreticulin Immunohistochemistry was performed on an Autostainer (Lab Vision). (staining V3+) was the best single marker with a high negative Following incubation with the primary antibody [protein disulfide predictive value, whereas combining the three markers (any marker isomerase A3 (PDI A3) monoclonal antibody (clone RL 77), Abcam; V2+) had the best positive predictive value while still retaining a calreticulin monoclonal antibody (clone FMC75), Abcam; heat shock fairly high negative predictive value. protein (HSP) gp96 polyclonal antibody (clone ZMD 287), Zymed Laboratories, Invitrogen Immunodetection] for 60 min at a dilution of 1:1600 (PDI A3), 1:400 (calreticulin), and 1:200 (anti–HSP gp96), slides were Discussion reacted with an immunoperoxidase detection system (poly-HRP-ant Ms/Rb/Ra IgG, Immunologic). The slides were then rinsed in PBS (pH 7.4) We have used discovery (proteomics) and validation (immuno- thrice and localization of the staining was performed for 5 min with 3,3¶- histochemistry) tools to identify and confirm novel molecular diaminobenzidine tetrahydrochloride (DAB+, Power DAB, Immunologic). markers that distinguish between FTA and FTCtissue. These After rinsing in PBS, the slides were finally counterstained with Mayer’s protein identifications provide insight into the pathogenesis of hematoxylin, dehydrated in ethanol and xylene, and coverslipped using a follicular thyroid neoplasia and a subset of these biomarkers may nonaqueous mounting medium. Cytoplasmatic and nuclear staining was serve as sensitive and specific markers that differentiate between considered as a positive reaction and intensity of staining was measured. benign and malignant form of follicular-derived thyroid neoplasia. A pathologist (B.M. Hoevenaars) was blinded to the histologic diagnosis and We specifically selected an analytic strategy measuring intact reported the results in a semiquantitative fashion: that is, no staining (0), faint (+1), low (+2), moderate (+3), and intense (+4) staining. proteins (two-dimensional gels) because this allowed us to detect and quantify changes in specific isoforms. Genomic research has shown several genetic alterations Results associated with follicular neoplasia (4, 15–17), but these alterations DIGE image analysis and protein identification. More than have only been documented in a small subset of tumors. Further, 1,500 protein spots were detected on each analytic DIGE gel and a the utility of these findings is limited because the level of mRNA total of 680 of these were matched on four of five individual gels expression frequently does not reflect the amount of protein in the and the pooled sample gel. Fifty-four of the protein spots that cell, in part because gene sequences cannot predict posttransla- showed statistically significant (P < 0.05) abundance differences tional modifications nor reflect dynamic cellular processes. Clearly, between individual FTCsamples and the FTA pool were identified thyroid tumorigenesis is a complex process and the additional (Fig. 1). Eleven proteins were excluded from further analysis: that is, quantitative and qualitative information intrinsic to the proteomic albumin, h-globin, thyroglobulin (four distinct spots), and five data is critical to understanding this complex pathophysiologic spots that were identified as a mixture of several proteins. This process. left 43 protein spots for further consideration (Table 1). Of these, Although used extensively in other forms of malignancy, the 27 spots, corresponding to 23 distinct protein entities, were less proteomic approach has had limited application in studies of abundant (average fold changes, 1.28–4.63) in FTCversus FTA; thyroid cancer. Berger et al. (18) performed a quantitative www.aacrjournals.org 1575 Cancer Res 2008; 68: (5). March 1, 2008

Downloaded from cancerres.aacrjournals.org on October 1, 2021. © 2008 American Association for Cancer Research. Cancer Research proteomic analysis in benign thyroid nodular disease and identified abundance between human PTCand matched normal thyroid several proteins showing abundance differences between benign tissue (8). This approach uncovered novel potential biomarkers and nodular tissue and matched normal thyroid tissue from the same confirmed several known biomarkers for PTC, illustrating the patients. Krause et al. (19) applied two-dimensional GE and MS advantages offered by this approach. to study the protein abundance differences between cold thyroid In the present report, we have identified a subset of 43 protein nodules and normal thyroid tissue and found up-regulation of spots (corresponding to 37 distinct proteins) that show statistically proteins involved in thyroglobulin folding and thyroid hormone significant differences in abundance between FTCand FTA tissue. synthesis as well as up-regulation of proteins that reflect increased Several of these proteins have previously been described in relation oxidative stress in the cold thyroid nodule tissue. Our studies have to thyroid or other cancers. Among the proteins underrepresented shown that proteins involved in protein synthesis and folding in FTCtissue are proteins involved in protein folding (e.g., HSP represent a large group of underabundant proteins in FTC gp96, PDI A3, calreticulin, HSP40, HSP90h, and BiP); proteins compared with FTA. Using two-dimensional DIGE and peptide involved in nuclear stability, chromatin structure, and gene mass fingerprinting by MALDI-TOF MS, we have previously expression (lamin A/C); and thyroglobin. Of the proteins over- investigated quantitative and qualitative differences in protein abundant in FTC, some are involved in cell stabilization against

Figure 1. Representative two-dimensional gel image of the Cy2-labeled proteins that comprise the internal standard. The grey outlines give the position of identified protein spots that had statistically different abundance (P < 0.05) between FTC and FTA. A, protein spots underabundant in FTC. B, protein spots overabundant in FTC. The spot numbers correspond to the spot numbers listed in Table 1.

Cancer Res 2008; 68: (5). March 1, 2008 1576 www.aacrjournals.org

nucleoside diphosphate kinase 1 isoform b). Our results also confirm the presence of other proteins previously associated with follicular-derived thyroid neoplasia, including nucleoside diphosphate kinase 1 (also known as nm23-H1), the nm23 metastatic suppressor gene product. Published mRNA and immunohistochemical studies suggest that the level of expression of nm23-H1 might be useful as a prognostic marker, especially for FTCand, less so, for PTC(20–22). Notably, because the present study compared two types of follicular neoplasia, we did not find galectin 3 nor cytokeratin 19, both known markers for PTC. We have also identified nine proteins that are novel to the thyroid neoplasm literature. One major advantage of the DIGE approach we used is the capability to run several different samples on a single gel. This leads to a dramatic improvement in quantitative precision and increases the likelihood of obtaining statistically meaningful results, even when the fold change is small. However, a limitation of our discovery study is the small sample size. When measuring many hundreds, even thousands of variables simultaneously, especially in a small population, there will be differences arising by chance alone and unrelated to any biochemical dissimilarity between the groups under investigation. There are statistical approaches to account for chance events, but these indiscriminate correction factors result in the loss of potentially important findings. Therefore, it is important to acknowledge that in this instance the sample size is too small to fully define the groups and Figure 2. Distribution of the identified proteins [underabundant (A) and some of the observed differences will likely be artifacts of the study overabundant (B) in FTC versus FTA] accordingto their cellular function. design. The situation is exacerbated by the heterogeneity of the follicular neoplasms (particularly the FTC). We selected the FTA mechanical stress (e.g., cytokeratins 7, 8, and 18 and tubulin), samples based on histology and exercised extreme caution to whereas others are linked to tumor invasiveness and metastatic ensure that these were indeed FTA and not minimally invasive potential in other malignancies, and kinase signaling (e.g., FTC. Therefore, we had limited snap-frozen FTA samples. We feel

Figure 3. Immunohistochemical analysis showinglower intensity stainingfor PDI A3 (PDI), calreticulin, and HSP gp96 in FTC versus FTA in the paraffin-embedded tissue samples. Magnification, 40.

www.aacrjournals.org 1577 Cancer Res 2008; 68: (5). March 1, 2008

Downloaded from cancerres.aacrjournals.org on October 1, 2021. © 2008 American Association for Cancer Research. Cancer Research that pooling the FTA samples was justified as we were able to Given the limitations of the study design, the data on proteins incorporate an additional sample and this group is more that were not further validated must be interpreted with caution homogeneous than the FTCgroup (minimally invasive, widely and any potential marker found in this initial discovery phase must invasive, and vascular invasion). We acknowledge that this is not an be verified in a larger, more comprehensive study. In the present optimal study design but feel the only other practical alternative study, we took this additional step and validated three of the (i.e., randomly omit one of the FTA sample and randomly pair identified proteins by immunohistochemistry in an independent individual FTA and FTCsamples) was also not without limitations. subset of paraffin-embedded tissue samples, showing that our

Table 2. Results of HSP gp96, PDI A3, and calreticulin staining in 16 patients with FTC and 18 patients with FTA

Sample no. Size (cm) Capsular invasion Vascular invasion HSP gp96* PDI A3* Calreticulin*

FTA 1134 4 21.543 3 33.533 4 43.533 4 52.233 4 6234 3 7233 3 82.434 3 9223 4 10 3 43 3 11 4 33 3 12 NA 43 3 13 5 34 4 14 3 33 4 15 1.2 43 4 16 0.8 33 3 17 2.8 23 3 18 3 24 3 Mean score (SD) 3.1 (0.6) 3.3 (0.5) 3.4 (0.5) FTC Minimally invasive 14+ 43 3 2 2.6 + 33 3 33+ 22 3 4 2.5 + + 3 3 3 5 3.3 + + 3 3 4 65++233 7 3.5 + + 2 3 3 8NA++343 9NA++422 10 3.5 + + 3 3 3 11 2.5 + + 2 2 3 c b b Mean score (SD) 2.8 (0.8) 2.8 (0.6) 3.0 (0.4) Widely invasive 13++222 23++333 3NA++232 4NA++223 5 5.5 + + 2 2 2 Mean score (SD) 2.2 (0.4) 2.4 (0.5) 2.4 (0.5)x All FTC Mean score (SD) 2.6 (0.7)k 2.7 (0.6){ 2.8 (0.4){

Abbreviation: NA not available. *Intensity scores: 0 to 4 (0, no staining; 4, intense staining). cP = 0.37, for the difference between mean minimally invasive FTCand FTA scores. bP < 0.05, for the difference between mean minimally invasive FTCand FTA scores. x P < 0.05, for the difference between mean minimally and widely invasive FTCscores. kP = 0.07, for the difference between mean FTCand FTA scores. {P < 0.002, for the difference between mean FTCand FTA scores.

Cancer Res 2008; 68: (5). March 1, 2008 1578 www.aacrjournals.org

Table 3. Sensitivity analysis of immunohistochemical staining

Marker Sensitivity (%) Specificity (%) PPV (%) NPV (%) Widely invasive FTC (% positive)

V2+ (positive) HSP gp96 50 83 72 65 80 PDI A3 37 100 100 64 60 Calreticulin 25 100 100 60 60 Any V25675756880 All V2 12 100 100 56 40 PDI A3/calreticulin V2 18 100 100 58 40 V3+ (positive) HSP gp96 87 22 50 66 100 PDI A3 93 27 53 83 100 Calreticulin 93 44 60 88 100 All V3 75 77 75 77 100 PDI A3/calreticulin V3 87 61 66 84 100

Abbreviations: PPV, positive predictive value; NPV, negative predictive value.

DIGE approach was robust for these three identified proteins. thyroid neoplasms. However, we recognize that there is a complex Ultimately, this validation is the only way to establish the clinically and variable relationship between solubilized protein levels, as relevant markers within the identified protein candidates. identified in our discovery studies, and those measured in tissue In this study, several residents of the endoplasmic reticulum (ER) samples by immunohistochemistry. Nevertheless, the immunohis- were present at a lower levels in FTCthan in FTA tissue. These are tochemistry and DIGE findings were consistent for all three molecular chaperones that play an essential role in the quality proteins. In each instance, the intensity scores for immunohisto- control system that regulates folding and maturation of newly chemical staining correlated with disease severity (i.e., FTAs synthesized proteins as well as the transport of the nascent showed the highest scores, whereas the widely invasive FTCs proteins from the ER to other compartments of the secretory showed the lowest scores). All three proteins showed a high pathway. Among these, BiP, PDI A3, and HSP gp96, a constitutively sensitivity with respect to detection of widely invasive FTCs. An expressed ER molecular chaperone belonging to the HSP90 family, immunohistochemical staining intensity score of three or less were underrepresented in the FTCsamples. These proteins are for any of the proteins detected all of the widely invasive FTCs in involved in the maturation of thyroglobulin, possibly as a part of our series. However, given the complexity of the protein patterns a macromolecular process, and assist with glycosylation and and the anticipated broad range in protein abundances, folding of thyroglobulin monomers (23, 24). Calreticulin, another dependent on the stage of transformation and specific mutational ER protein identified in our study, plays a key role in the synthesis patterns, we believe it is essential to extend our investigations to of glycoproteins, including thyroperoxidase (25, 26). Moreover, in include more of the identified proteins and to examine a larger addition to their role in protein folding, calreticulin and HSP gp96 number of patients having both minimally invasive and widely may trigger an anticancer immune response (27) and improve the invasive FTC. efficiency of phagocytosis (28). In conclusion, we used discovery proteomics and a validation Krause et al. (19) reported higher levels of calreticulin, PDI A3, approach (immunohistochemistry) to identify potential novel and HSP90h in benign cold thyroid nodules compared with biomarkers that aid in distinguishing between FTCand FTA. In normal thyroid tissue, but they could not find any evidence of addition, these studies provide insights into the global pathophys- somatic thyroglobulin mutations. In contrast, Paron et al. iologic changes in thyroid carcinoma. Notably, we identified reported down-regulation of calreticulin in thyroid cell lines protein isoform differences and posttranslational modifications transformed by mutant p53 alleles (29). Our finding of lower that would likely be missed by genomic or other proteomic levels of these proteins in FTCversus FTA tissue may reflect a approaches. Carefully designed, controlled prospective studies are lesser degree of differentiation of the malignant transformed now required to establish the clinical utility of each of these thyrocyte in the FTC. markers. Based on the findings of our discovery study, and a review of the literature, we decided to pursue 3 of the 43 proteins found to be Acknowledgments present at different levels in FTCversus FTA (i.e., HSP gp96, PDI Received 8/23/2007; revised 10/30/2007; accepted 10/31/2007. A3, and calreticulin). These three proteins were abundant in the Grant support: Generations Cancer Foundation, Mary Rossick Kern and Jerome H. tissue samples and showed a large difference in volume ratios Kern Endowment in Endocrine Neoplasms Research, and Niels-Stensen Foundation between the FTCand FTA. (R.T. Netea-Maier). The costs of publication of this article were defrayed in part by the payment of page We chose immunohistochemistry for our verification studies charges. This article must therefore be hereby marked advertisement in accordance because it is performed at the tissue level and it allowed us to with 18 U.S.C. Section 1734 solely to indicate this fact. We thank L. Brown and A. Gemmink for their technical support and the assess our candidate markers in an independent subset of paraffin- Cooperative Human Tissue Network, which is funded by the National Cancer Institute, embedded tissue samples obtained from patients with follicular for providing thyroid tissue samples. www.aacrjournals.org 1579 Cancer Res 2008; 68: (5). March 1, 2008

References 12. Alban A, David SO, Bjorkesten L, et al. A novel diphosphate kinase in human thyroid carcinomas: lack experimental design for comparative two-dimensional of correlation between its expression and lymph node 1. Mazzaferri EL. Management of a solitary thyroid gel analysis: two-dimensional difference gel electropho- metastasis. Thyroid 1993;3:105–9. nodule. N Engl J Med 1993;328:553–9. resis incorporating a pooled internal standard. Proteo- 22. Zou M, Shi Y, al-Sedairy S, Farid NR. High levels of 2. Haugen BR, Woodmansee WW, McDermott MT. mics 2003;3:36–44. Nm23 gene expression in advanced stage of thyroid Towards improving the utility of fine-needle aspiration 13. Nicolls MR, D’Antonio JM, Hutton JC, Gill RG, carcinomas. Br J Cancer 1993;68:385–8. biopsy for the diagnosis of thyroid tumours. Clin Czwornog JL, Duncan MW. Proteomics as a tool for 23. Delom F, Mallet B, Carayon P, Lejeune PJ. Role of Endocrinol (Oxf) 2002;56:281–90. discovery: proteins implicated in Alzheimer’s disease are extracellular molecular chaperones in the folding of 3. Bojunga J, Zeuzem S. Molecular detection of thyroid highly expressed in normal pancreatic islets. J Proteome oxidized proteins. Refolding of colloidal thyroglobulin cancer: an update. Clin Endocrinol (Oxf) 2004;61:523–30. Res 2003;2:199–205. by protein disulfide isomerase and immunoglobulin 4. Segev DL, Umbricht C, Zeiger MA. Molecular patho- 14. Bradshaw RA, Burlingame AL, Carr S, Aebersold R. heavy chain-binding protein. J Biol Chem 2001;276: genesis of thyroid cancer. Surg Oncol 2003;12:69–90. Reporting protein identification data: the next genera- 21337–42. 5. Sarquis MS, Weber F, Shen L, et al. High frequency of tion of guidelines. Mol Cell Proteomics 2006;5:787–8. 24. Mezghrani A, Courageot J, Mani JC, Pugniere M, loss of heterozygosity in imprinted, compared with 15. Melillo RM, Castellone MD, Guarino V, et al. The Bastiani P, Miquelis R. Protein-disulfide isomerase (PDI) nonimprinted, genomic regions in follicular thyroid RET/PTC-RAS-BRAF linear signaling cascade mediates in FRTL5 cells. pH-dependent thyroglobulin/PDI inter- carcinomas and atypical adenomas. J Clin Endocrinol the motile and mitogenic phenotype of thyroid cancer actions determine a novel PDI function in the post- Metab 2006;91:262–9. cells. J Clin Invest 2005;115:1068–81. endoplasmic reticulum of thyrocytes. J Biol Chem 2000; 6. Duncan MW, Hunsucker SW. Proteomics as a tool for 16. Cerutti JM, Delcelo R, Amadei MJ, et al. A 275:1920–9. clinically relevant biomarker discovery and validation. preoperative diagnostic test that distinguishes benign 25. Le Fourn V, Siffroi-Fernandez S, Ferrand M, Franc Exp Biol Med (Maywood) 2005;230:808–17. from malignant thyroid carcinoma based on gene JL. Competition between calnexin and BiP in the 7. Unlu M, Morgan ME, Minden JS. Difference gel expression. J Clin Invest 2004;113:1234–42. endoplasmic reticulum can lead to the folding or electrophoresis: a single gel method for detecting 17. Chevillard S, Ugolin N, Vielh P, et al. Gene degradation of human thyroperoxidase. Biochemistry changes in protein extracts. Electrophoresis 1997;18: expression profiling of differentiated thyroid neo- 2006;45:7380–8. 2071–7. plasms: diagnostic and clinical implications. Clin 26. Fayadat L, Siffroi-Fernandez S, Lanet J, Franc JL. 8. Brown LM, Helmke SM, Hunsucker SW, et al. Cancer Res 2004;10:6586–97. Calnexin and calreticulin binding to human thyro- Quantitative and qualitative differences in protein 18. Berger K, Wissmann D, Ihling C, et al. Quantitative peroxidase is required for its first folding step(s) but is expression between papillary thyroid carcinoma and proteome analysis in benign thyroid nodular disease not sufficient to promote efficient cell surface expres- normal thyroid tissue. Mol Carcinog 2006;45:613–26. using the fluorescent ruthenium II tris(bathophenan- sion. Endocrinology 2000;141:959–66. 9. Friedman DB, Hill S, Keller JW, et al. Proteome analysis throline disulfonate) stain. Mol Cell Endocrinol 2004; 27. Nair S, Wearsch PA, Mitchell DA, Wassenberg JJ, of human colon cancer by two-dimensional difference 227:21–30. Gilboa E, Nicchitta CV. Calreticulin displays in vivo gel electrophoresis and mass spectrometry. Proteomics 19. Krause K, Karger S, Schierhorn A, Poncin S, Many peptide-binding activity and can elicit CTL responses 2004;4:793–811. MC, Fuhrer D. Proteomic profiling of cold thyroid against bound peptides. J Immunol 1999;162:6426–32. 10. Wessel D, Flugge UI. A method for the quantitative nodules. Endocrinology 2007;148:1754–63. 28. Obeid M, Tesniere A, Ghiringhelli F, et al. Calreticulin recovery of protein in dilute solution in the presence of 20. Zafon C, Obiols G, Castellvi J, et al. nm23-H1 exposure dictates the immunogenicity of cancer cell detergents and lipids. Anal Biochem 1984;138:141–3. immunoreactivity as a prognostic factor in differentiat- death. Nat Med 2007;13:54–61. 11. Bradford MM. A rapid and sensitive method for the ed thyroid carcinoma. J Clin Endocrinol Metab 2001;86: 29. Paron I, D’Ambrosio C, Scaloni A, et al. A differential quantitation of microgram quantities of protein utiliz- 3975–80. proteomic approach to identify proteins associated with ing the principle of protein-dye binding. Anal Biochem 21. Luo W, Matsuo K, Nagayama Y, et al. Immunohisto- thyroid cell transformation. J Mol Endocrinol 2005;34: 1976;72:248–54. chemical analysis of expression of nm23-H1/nucleoside 199–207.

Cancer Res 2008; 68: (5). March 1, 2008 1580 www.aacrjournals.org

Downloaded from cancerres.aacrjournals.org on October 1, 2021. © 2008 American Association for Cancer Research. Discovery and Validation of Protein Abundance Differences between Follicular Thyroid Neoplasms

Romana T. Netea-Maier, Stephen W. Hunsucker, Brigiet M. Hoevenaars, et al.

Cancer Res 2008;68:1572-1580.

Updated version Access the most recent version of this article at: http://cancerres.aacrjournals.org/content/68/5/1572

Supplementary Access the most recent supplemental material at: Material http://cancerres.aacrjournals.org/content/suppl/2008/02/28/68.5.1572.DC1

Cited articles This article cites 29 articles, 6 of which you can access for free at: http://cancerres.aacrjournals.org/content/68/5/1572.full#ref-list-1

Citing articles This article has been cited by 4 HighWire-hosted articles. Access the articles at: http://cancerres.aacrjournals.org/content/68/5/1572.full#related-urls

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://cancerres.aacrjournals.org/content/68/5/1572. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.