Proteomic and Genomic Technologies for Biomarker Discovery Where Are the Biomarkers? Vathany Kulasingam and Eleftherios P

02_Diamandis_Editorial_1.qxd 11/21/06 3:06 PM Page 5

Proteomic and Genomic Technologies for Biomarker Discovery Where Are the Biomarkers? Vathany Kulasingam and Eleftherios P. Diamandis*

Laboratory Medicine and Pathobiology, University of Toronto,Toronto, ON, Canada and Department of Pathology and Laboratory Medicine, Mount Sinai Hospital, 600 University Ave.,Toronto, ON M5G 1X5, Canada

Introduction cancer, as it is a model disease for studying such applications. Recently, we have witnessed the emergence of the “-omics” era with proteomics, Current Cancer Biomarkers peptidomics, degradomics, metabolomics, fractionomics, transcriptomics, interactomics, Despite tremendous progress in our under- and so forth. However, the only “-omics” standing of the molecular basis of many dis- other than genomics to become a “buzz- eases, cancer continues to be a major cause of word” is proteomics, a term used for studying morbidity and mortality among men and the proteome of an organism. With the com- women. There is convincing evidence that early pletion of the Human Genome Project, opti- detection of cancer can lead to better patient mistic views were expressed that novel outcomes through early administration of disease-specific biomarkers will be discov- therapy (1). Also, better clinical outcomes can ered through various high-throughput tech- be achieved with individualized treatments, niques, such as microarrays and mass which are more effective in subgroups of spectrometry (MS). Did the current technolo- patients, rather than in the total patient popu- gies deliver the goods? In the following, we lation. One of the best ways to diagnose cancer discuss briefly the current status and future early, or to predict therapeutic responses, is by of proteomics and genomics with respect to using serum or tissue biomarkers.

*Author to whom all correspondence and reprint requests should be addressed: Eleftherios P. Diamandis, Laboratory Medicine and Pathobiology, University of Toronto, Toronto, ON, Canada and Department of Pathology and Laboratory Medicine, Mount Sinai Hospital, 600 University Ave., Toronto, ON M5G 1X5, Canada. E-mail: [email protected]. 5 02_Diamandis_Editorial_1.qxd 11/21/06 3:06 PM Page 6

6 ______Kulasingam and Diamandis

Currently, a number of cancer biomarkers proteomics and genomics community to dis- are used clinically, including prostate-specific cover novel biomarkers. antigen (PSA; for prostate cancer), carcinoem- Some Proteomic and Genomic bryonic antigen (CEA; for colon, lung, and breast cancer), the carbohydrate antigens (CA) Approaches to Biomarker 125 (for ovarian cancer), CA 15.3 (for breast Discovery cancer), CA 19.9 (for gastrointestinal cancer), A number of different proteomic- and α-fetoprotein (AFP; for liver and testicular genomic-based approaches have been utilized cancer), and human choriogonadotropin (hCG; to date to discover disease-specific biomarkers. for testicular and gestational cancers). All of Some of the approaches and strategies are these markers, despite being generally not suit- briefly described here along with an example able for population screening (because of low of how these methodologies were applied to diagnostic sensitivity and specificity), play a aid in biomarker discovery or to develop pat- major role in aiding diagnosis and for moni- terns for early diagnosis. toring patients during therapy. Unfortunately, these cancer biomarkers have proven to be Two-Dimensional Gel Electrophoresis ineffective in detecting the early stages of dis- and Mass Spectrometry ease and many suffer from low positive pre- Two-dimensional polyacrylamide gel elec- dictive values. trophoresis, known traditionally as the gold Renewed Interest in Cancer standard, discovery-based tool for proteomics, aims to resolve and visualize protein Biomarkers spots that are differentially expressed in Historically, every era of cancer biomarker healthy vs diseased samples. The spots can discovery was associated with the emergence be excised from the gel, digested, and sub- of a powerful analytical technology. For sequently identified via MS. Using this example, CEA was discovered in the 1960s methodology, Celis et al. (3) developed a mainly as a result of the introduction of novel comprehensive database for bladder cancer and relatively sensitive immunological tech- profiles of both transitional and squamous niques (such as radial immuno-diffusion), cell carcinomas. Furthermore, Kageyama et al. which allowed for the detection of the anti- (4) used a differential display method of gen in cancer tissues with high specificity bladder cancer vs healthy urothelial tissue and reasonable sensitivity. CA 125 and other and MS to identify proteins, which are CAs were developed, in the late 1970s, increased in cancer tissues. Calreticulin, a because of the establishment of the mono- potential tumor marker, was identified in clonal antibody technology. Most of these this study as it was found in urine of tumor markers were discovered by using cell patients with bladder cancer. The authors lines or tumor extracts as immunogens and confirmed their data with Western blot anal- then selecting specific hybridoma clones that ysis, immunoprecipitation, and immunohis- recognized these tumor antigens (2). With the tochemistry. They reported a sensitivity of completion of the Human Genome Project 73% and a specificity of 86%. However, a and the introduction of technologies that caveat to this technology is that the identi- enabled simultaneous examination of thou- fied proteins are usually of high abundance sands of proteins and genes at any given and the method is labor intensive. Similar time, such as MS and protein and DNA strategies have been used by many other arrays, a renewed interest was taken by the groups.

Clinical Proteomics ______Volume 2, 2006 02_Diamandis_Editorial_1.qxd 11/21/06 3:06 PM Page 7

Editorial ______7

Protein Arrays noncancer. This pattern was used to classify Another area of proteomic research that is 116 blinded serum samples. The results were gaining popularity is the use of protein arrays exceptional with a sensitivity of 100% at 95% to identify novel biomarkers. Chinnaiyan and specificity. These numbers are far superior to colleagues recently published data suggesting the sensitivities and specificities obtained with that autoantibody signatures may improve the current serologic cancer biomarkers. early detection of prostate cancer (5). Using a MS-Imaging (MALDI-MS) combination of phage-display technology and protein microarrays, they identified new Another proteomic pattern analysis approach autoantibody-binding peptides derived from involves the use of matrix-assisted laser des- prostate cancer tissue. Serum samples from 119 orption/ionization (MALDI)-MS on small patients with prostate cancer and 138 control amounts of fresh frozen tissue sections. In a patients were analyzed using this methodol- recent application of this methodology, termed ogy. Their results were impressive. A 22-phage MS imaging, pioneered by Caprioli and col- peptide detector had 88% specificity and 82% leagues (7), the authors profiled the protein sensitivity in discriminating between the two expression of surgically resected lung tumor groups. This panel of peptides seemed to per- tissues (8). To detect proteomic patterns in form better than PSA in distinguishing lung tumors, the protein expression profiles of between prostate cancer and the control group. 50 tissue samples (42 lung tumors and 8 normal lung samples) in a training cohort was Proteomic Pattern Analysis (SELDI-TOF) assessed. Mass spectra were obtained directly The ability to identify diseases through pro- from 1-mm regions of single frozen tissue sec- teomic patterns in biological fluids appeared tions. More than 1600 distinct protein species to be a new paradigm shift in the use of MS- were observed, of which 82 discriminating MS based methodologies. This approach attracted signals were selected. A class-prediction model considerable attention over the past few years. was built and it was able to correctly classify Developed by Petricoin, Liotta, and colleagues, all 42 lung tumors and the 8 normal samples. this potentially revolutionary methodology This model was further validated in an inde- involves the use of a minute amount of unfrac- pendent blinded test cohort yielding a phe- tionated serum sample added to a “protein- nomenal result of perfect classification of the chip,” which is subsequently analyzed by samples. surface-enhanced laser desorption/ionization time-of-flight (SELDI-TOF) spectrometry to Quantitative Proteomics generate a proteomic signature of the serum. Quantitative proteomics aims to measure These patterns reflect part of the blood pro- quantitatively the relative or absolute abun- teome, without knowledge of the actual iden- dance of proteins across samples. The emer- tity of the proteins that make up the proteome. gence of experimental protocols, such as The potential of proteomic pattern analysis isotope-coded affinity tags (ICAT), iTRAQ™, was first demonstrated in the diagnosis of stable isotope labeling by amino acids in cell ovarian cancer (6). In this study, a preliminary culture (and SILAC), now allow for quantita- training set of mass spectra, generated by tive proteomic approaches to biomarker dis- SELDI-TOF, from 50 unaffected women and 50 covery. As a result, quantitative differential ovarian cancer patients was obtained. Search- analysis can be performed on an MS signal— ing algorithms were used to identify a pro- the end result being the identification of the teomic pattern that discriminated cancer from relative abundance of the molecular species

Volume 2, 2006 ______Clinical Proteomics 02_Diamandis_Editorial_1.qxd 11/21/06 3:06 PM Page 8

8 ______Kulasingam and Diamandis

detected in the sample. In a study by Martin and Drug Administration for approval et al. (9), a cell culture model system was (Table 1). used to identify secreted and cell-surface pro- Did Proteomic and Genomic teins from neoplastic prostate epithelium Methodologies Deliver the Goods? using ICAT reagents and tandem MS. Pro- teomic analysis of the media identified more Rapid advances in proteomic and genomic than 600 proteins, of which 524 were quanti- technologies have created optimistic views fied. This study demonstrated the feasibility that many more cancer biomarkers will be dis- of using high-throughput quantitative pro- covered through various high-throughput tech- teomics to identify proteins in conditioned niques. However, these predictions have yet to media. come true. Very few, if any, new serum cancer biomarkers have been introduced at the clinic Genomic Approaches over the last 15 yr. Organizations like the Amer- Genomic microarrays represent a highly ican Society of Clinical Oncology do not encour- powerful technology for genome-wide stud- age the widespread use of tumor markers, ies on gene expression, single-nucleotide unless they affect patient outcome measures polymorphism identification, mutational (23). On the other hand, there is general agree- analysis, and so forth. Microarray experi- ment that a combination of multiple biomark- ments are usually performed with DNA or ers may increase the diagnostic sensitivity and RNA isolated from tissues. Thus, unlike sero- specificity of individual markers. This is partic- logic markers, this methodology does not ularly important in relation to the recent devel- lend itself to minimally invasive diagnostics. opment of powerful bioinformatic algorithms However, tissue biopsies can be very useful (such as artificial neural networks and logistic for confirming diagnosis, for tumor sub- regression), which can interpret multiple para- classification and for predicting therapeutic meters much more efficiently than the average response. It was recently hypothesized that clinician (24,25). the phenotypic differences of tumors from Most of the serum proteomic pattern analy- different patients may be associated with sis technologies (such as SELDI-TOF), although unique gene expression profiles, which can promising, have not as yet been evaluated at be captured by using DNA microarrays (10). the clinic and a number of important limita- The proof-of-principle for the cancer sub- tions have already been identified (26,27). classification hypothesis has been provided These limitations include the possibility of bias for various malignancies, such as breast stemming from artifacts related to the clinical cancer, leukemias, and many other cancer samples used to generate published data, the types (11–18). Some of the most celebrated inherent qualitative nature of MS and/or the studies on cancer subclassification and sub- bioinformatic analysis applied to the raw data sequent selection of therapeutic intervention (28–31). To illustrate this point, a meta-analysis have created excitement in the media (19,20). of prostate cancer proteomic data from four Other approaches, based on the same articles by three different research groups was hypothesis, used alternative technologies, compiled (27). The discriminatory peaks identi- such as quantitative reverse transcriptase- fied in the four papers using SELDI-TOF were polymerase chain reaction or fluorescence in very different. Interestingly, two papers pub- situ hybridization to derive prognostic and lished among the same group using the same predictive information (21,22). Some of these experimental data but different bioinformatic tests have now been submitted to the Food tools yielded different discriminatory peaks.

Clinical Proteomics ______Volume 2, 2006 02_Diamandis_Editorial_1.qxd 11/21/06 3:06 PM Page 9

Editorial ______9

Table 1 New and Emerging “-omic” Cancer Tests Company Product Approval status Characteristics Agendia MammaPrint Pending at FDA ➣ A 70-gene expression signature measured on an Agilent DNA microarray ➣ Predicts prognosis for certain breast cancer patients Correlogic OvaCheck Pending at FDA ➣ Protein expression pattern systems analysis ➣ More than 150,000 starting data points are reduced to 5–10 for final evaluation Exagen Breast cancer Submission for FDA ➣ Traditional FISH test, derived prognosis test for PMA anticipated from gene expression as well as by end of 2005 other data Genomic Oncotype Dx Not an FDA-regulated ➣ A 21-gene expression assay health product. Launched done using RT-PCR for as a laboratory- prognosis in breast cancer developed test in California, January 2004 Roche Leukemia array Launch anticipated ➣ A 300-gene expression 2006–2007 microarray test Modified from ref. 35. FDA, Food and Drug Administration; FISH, fluorescent in situ hybridization; PMA, pre-market approval; RT-PCR, reverse transcriptase-polymerase chain reaction.

This raises the problem of data overfitting and because their concentrations in serum and/or the lack of reproducibility with some of these biological fluids are too low and therefore technologies. cannot be measured or the proteins cannot In a study by Gygi and coworkers, six com- be purified, unless specific immunological panies were asked to analyze the same protein reagents and highly sensitive enzyme-linked extract from 10,000 human cells by liquid immunosorbent assay methods are available. chromatography-tandem MS after trypsin Thus, in the initial discovery phase for novel digestion (32). Only 52 proteins, representing cancer biomarkers, a less complex sample 3% of all the data, were identified by all six (elimination of high-abundant proteins) is participants. Furthermore, about two-thirds of essential. Therefore, proteomic technologies the proteins were identified by only one par- have yet to deliver the “goods” because they ticipant. This exemplifies a very important are more difficult to find than originally point—different MS platforms identify differ- anticipated. ent proteins in the same mixture. In addition, On the genomics side, Michiels et al. (33) it should be noted that serum is a complex recently performed a meta-analysis of seven of mixture and it is possible that new candidate the most prominent studies on cancer progno- biomarkers have not yet been identified sis, by using microarray-based expression

Volume 2, 2006 ______Clinical Proteomics 02_Diamandis_Editorial_1.qxd 11/21/06 3:06 PM Page 10

10 ______Kulasingam and Diamandis

profiling. Surprisingly, in five of the seven epithelial ovarian cancer. N. Engl. J. Med. 309, studies on cancer prognosis, the original data 883–887. could not be reproduced (34). The other two 3. Celis, A., Rasmussen, H. H., Celis, P., et al. (1999) Short-term culturing of low-grade studies showed much weaker prognostic superficial bladder transitional cell carcinomas information than the original data. The meta- leads to changes in the expression levels of analysis also indicated that the list of genes several proteins involved in key cellular activ- identified as predictors of prognosis was ities. Electrophoresis 20, 355–361. highly unstable and that the molecular signa- 4. Kageyama, S., Isono, T., Iwaki, H., et al. tures were strongly dependent on the selection (2004) Identification by proteomic analysis of calreticulin as a marker for bladder cancer of patients in the training sets. This article sug- and evaluation of the diagnostic accuracy of gested that the results of the aforementioned its detection in urine. Clin. Chem. 50, studies, as published, are highly over opti- 857–866. mistic and that they need careful validation 5. Wang, X., Yu, J., Sreekumar, A., et al. (2005) before conclusions can be drawn. Autoantibody signatures in prostate cancer. N. Engl. J. Med. 353, 1224–1235. Where Do We Stand Today? 6. Petricoin, E. F., Ardekani, A. M., Hitt, B. A., et al. (2002) Use of proteomic patterns in serum to The emergence of the “-omics” era is an identify ovarian cancer. Lancet 359, 572–577. exciting time for the scientific community. The 7. Caprioli, R. M. (2005) Deciphering protein technologies and tools needed to decipher the molecular signatures in cancer tissues to aid proteome of organisms and to dig deeper into in diagnosis, prognosis, and therapy. Cancer Res. 65, 10,642–10,645. the biomarker mine are available. There is 8. Yanagisawa, K., Shyr, Y., Xu, B. J., et al. (2003) no doubt that, if these new technological Proteomic patterns of tumor subsets in non- advances prove to be successful in identifying small-cell lung cancer. Lancet 362, 433–439. cancer biomarkers for early cancer detection, 9. Martin, D. B., Gifford, D. R., Wright, M. E., et al. the clinical benefits are likely to be enormous. (2004) Quantitative proteomic analysis of pro- What is not clear is which technologies will be teins released by neoplastic prostate epithelium. Cancer Res. 64, 347–355. the most successful as it is too early to say if 10. Perou, C. M., Sorlie, T., Eisen, M. B., et al. we have indeed succeeded in delivering the (2000) Molecular portraits of human breast “goods.” We should also not forget that other tumors. Nature 406, 747–752. diagnostic modalities, such as imaging, may 11. Alizadeh, A. A., Ross, D. T., Perous, C. M., and eventually become so sensitive and specific van de Rijn, M. (2001) Towards a novel classi- that the use of biomarkers for patient diagno- fication of human malignancies based on gene expression patterns. J. Pathol. 195, 41–52. sis and management may become redundant. 12. Weigelt, B., Hu, Z., He, X., et al. (2005) Molec- Have we come upon a new paradigm shift in ular portraits and 70-gene prognosis signature the use of proteomic and genomic technolo- are preserved throughout the metastatic pro- gies for biomarker discovery? Only meticu- cess of breast cancer. Cancer Res. 65, 9144–9158. lous validation steps and time will tell. 13. Alizadeh, A. A., Eisen, M. B., Davis, R. E., et al. (2000) Distinct types of diffuse large B-cell lym- References phoma identified by gene expression profiling. Nature 403, 503–511. 1. Etzioni, R., Urban, N., Ramsey, S., et al. (2003) 14. Rosenwald, A., Wright, G., Chan, W. C., et al. The case for early detection. Nat. Rev. Cancer (2002) Lymphoma/leukemia molecular profil- 3, 243–252. ing project: The use of molecular profiling to 2. Bast, R. C., Jr., Klug, T. L., St. John, E., et al. predict survival after chemotherapy for dif- (1983) A radioimmunoassay using a mono- fuse large-B-cell lymphoma. N. Engl. J. Med. clonal antibody to monitor the course of 346, 1937–1947.

Clinical Proteomics ______Volume 2, 2006 02_Diamandis_Editorial_1.qxd 11/21/06 3:06 PM Page 11

Editorial ______11

15. Bhattacharjee, A., Richards, W. G., Staunton, J., 25. Stephan, C., Cammann, H., Semjonow, A., et al. et al. (2001) Classification of human lung (2002) Multicenter evaluation of an artificial carcinomas by mRNA expression profiling neural network to increase the prostate cancer reveals distinct adenocarcinoma subclasses. detection rate and reduce unnecessary biopsies. Proc. Natl. Acad. Sci. USA 98, 13,790–13,795. Clin. Chem. 48, 1279–1287. 16. Ramaswamy, S., Ross, K. N., Lander, E. S., and 26. Diamandis, E. P. (2003) Point: proteomic pat- Golub, T. R. (2003) A molecular signature of terns in biological fluids: do they represent the metastasis in primary solid tumors. Nat. Genet. future of cancer diagnostics? Clin. Chem. 49, 33, 49–54. 1272–1275. 17. Pemeroy, S. L., Tamayo, P., Gaasenbeek, M., 27. Diamandis, E. P. (2003) Analysis of serum pro- et al. (2002) Prediction of central nervous teomic patterns for early cancer diagnosis: system embryonal tumor outcome based on drawing attention to potential problems. gene expression. Nature 415, 436–442. J. Natl. Cancer Inst. 96, 353–356. 18. Iizuka, N., Hammoto, Y., and Oka, M. (2004) 28. Karsan, A., Eigl, B. J., Fibotte, S., et al. (2005) Predicting individual outcomes in hepatocel- Analytical and preanalytical biases in serum lular carcinoma. Lancet 364, 1837–1839. proteomic pattern analysis for breast cancer 19. van de Vijver, M. J., He, Y. D., van’t Veer, I. J., diagnosis. Clin. Chem. 51, 1525–1528. et al. (2002) A gene-expression signature as a 29. Banks, R. E., Stanley, A. J., Cairns, D. A., et al. predictor of survival in breast cancer. N. Engl. (2005) Influences of blood sample processing on J. Med. 347, 1999–2009. low-molecular-weight-proteome identified by 20. van’t Veer, I. J. and Weigelt, B. (2003) Road surface-enhanced laser desorption/ionization map to metastasis. Nat. Med. 9, 999–1000. mass spectrometry. Clin. Chem. 51, 1637–1649. 21. Lossos, I. S., Czerwinski, D. K., Alizadeh, 30. Baggerly, K. A., Morris, J. S., Edmonson, S. R., A. A., et al. (2004) Prediction of survival in dif- and Coombes, K. R. (2005) Signal in noise: fuse large-B-cell lymphoma based on the evaluating reported reproducibility of serum expression of six genes. N. Engl. J. Med. 350, proteomic tests for ovarian cancer. J. Natl. 1828–1837. Cancer Inst. 97, 307–309. 22. Cronin, M., Pho, M., Dutta, D., et al. (2004) 31. Ransohoff, D. F. (2005) Lessons from contro- Measurement of gene expression in archival versy: ovarian cancer screening and serum paraffin-embedded tissues: development and proteomics. J. Natl. Cancer Inst. 97, 315–319. performance of a 92-gene reverse transcriptase- 32. Elias, J. E., Haas, W., Faherty, B. K., and Gygi, polymerase chain reaction assay. Am. J. Pathol. S. P. (2005) Comparative evaluation of mass 164, 35–42. spectrometry platforms used in large-scale 23. Bast, R. C., Jr., Ravdin, P., Hayes, D. F., et al. proteomics investigations. Nat. Methods 2, (2000) Update of recommendations for the use 667–675. of tumor markers in breast and colorectal 33. Michiels, S., Koscielny, S., and Hill, C. (2005) cancer: clinical practice guidelines of the Prediction of cancer outcome with microar- American Society of Clinical Oncology. J. Clin. rays: a multiple random validation strategy. Oncol. 19, 1865–1878. Lancet 365, 488–492. 24. Finne, P., Finne, R., Auvinen, A., et al. (2000) 34. Ioannidis, J. P. (2005) Microarrays and molecular Predicting the outcome of prostate biopsy in research: noise discovery? Lancet 365, 454–455. screen-positive men by a multilayer percep- 35. Branca, M. A. (2005) Omic diagnostics trip up tron network. Urology 56, 418–422. on way to clinic. Nat Biotechnol. 23, 769.

Volume 2, 2006______Clinical Proteomics