Molecular prognostication of liver cancer: End of the beginning

Abstract

New technological developments have frequently preceded major advances in biomedical research and medicine [[1]Sander C. Genomic medicine and the future of health care.Science. 2000; 287: 1977-1978Crossref PubMed Scopus (140) Google Scholar]. For example, the development of fluorescent DNA sequencing techniques made it possible to establish the large-scale high-throughput technology needed for human genome sequencing. Polymerase chain reaction (PCR), fluorescent DNA sequencing, and other techniques have enabled the discovery of about 1700 mendelian disease genes [[2]Institute of Medical Genetics, Cardiff, UK (http://www.hgmd.cf.ac.uk).Google Scholar]. The advent of the DNA microarray based technologies has now made it possible to measure simultaneously the expression of tens of thousands of genes in different tissues under a variety of conditions. This high-throughput technology has afforded biomedical scientists a unique opportunity to integrate the descriptive characteristics (i.e. ‘phenotype’) of a biological system under study with the genomic readout (i.e. gene expression). The opportunity to contemplate the integrated view of biological systems has provoked a shift in biological sciences away from the classical reductionism to systems biology [1Sander C. Genomic medicine and the future of health care.Science. 2000; 287: 1977-1978Crossref PubMed Scopus (140) Google Scholar, 3Guttmacher A.E. Collins F.S. Realizing the promise of genomics in biomedical research.J Am Med Assoc. 2005; 294: 1399-1402Crossref Scopus (105) Google Scholar, 4Hood L. Heath J.R. Phelps M.E. Lin B. Systems biology and new technologies enable predictive and preventative medicine.Science. 2004; 306: 640-643Crossref PubMed Scopus (883) Google Scholar]. The systems approach to a disease is based on the hypothesis that disease processes perturb a regulatory network of genes and proteins in a way that differs from the respective normal counterpart. Consequently, by using multi-parametric measurements it may be possible to transform current diagnostic and therapeutic approaches and enable a predictive and preventive personalized medicine [[4]Hood L. Heath J.R. Phelps M.E. Lin B. Systems biology and new technologies enable predictive and preventative medicine.Science. 2004; 306: 640-643Crossref PubMed Scopus (883) Google Scholar]. The application of microarray technologies to characterize tumors at the gene expression level has significantly impacted clinical oncology [5Pusztai L. Ayers M. Stec J. Hortobagyi G.N. Clinical application of cDNA microarrays in oncology.Oncologist. 2003; 8: 252-258Crossref PubMed Scopus (57) Google Scholar, 6Wang Y. Gene expression-driven diagnostics and pharmacogenomics in cancer.Curr Opin Mol Ther. 2005; 7: 246-250PubMed Google Scholar]. Global gene expression analysis of various human tumors has resulted in the identification of gene expression patterns or signatures related to tumor classification, disease outcome and response to therapy. The microarray technology has also been used to investigate the mechanism of action of specific cancer therapeutics. In this review, we will briefly discuss the general impact of the global gene expression analysis on cancer research and devote the rest to the microarray based gene expression analysis of human hepatocellular carcinoma (HCC). It is well established that cancer even in the same tissue is a very heterogeneous disease that differs widely in clinical outcome and in response to therapy. It is now clear that this heterogeneity is due to different molecular defects that can induce similar tumor phenotypes. Although, histopathological and biochemical markers constitute important tools for identifying groups of tumors that differ with respect to prognosis and responses to treatments, the genes and molecular pathways associated with these markers have not been comprehensively defined. The application of microarray technologies which rely on thousands of pieces (i.e. genes, proteins) of information would a priori be expected to be proficient in identifying tumor subtypes. Indeed, microarray-based global gene expression analysis of human tumors has already revealed the identification of gene expression patterns or signatures related to tumor classification, prognosis and response to therapy [7Bittner M. Meltzer P. Chen Y. Jiang Y. Seftor E. Hendrix M. et al.Molecular classification of cutaneous malignant melanoma by gene expression profiling.Nature. 2000; 406: 536-540Crossref PubMed Scopus (1696) Google Scholar, 8Alizadeh A.A. Eisen M.B. Davis R.E. Ma C. Lossos I.S. Rosenwald A. et al.Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling.Nature. 2000; 403: 503-511Crossref PubMed Scopus (7896) Google Scholar, 9Beer D.G. Kardia S.L. Huang C.C. Giordano T.J. Levin A.M. Misek D.E. et al.Gene-expression profiles predict survival of patients with lung adenocarcinoma.Nat Med. 2002; 8: 816-824Crossref PubMed Scopus (1638) Google Scholar, 10van de Vijver M.J. He Y.D. van't Veer L.J. Dai H. Hart A.A. Voskuil D.W. et al.A gene-expression signature as a predictor of survival in breast cancer.N Engl J Med. 2002; 19: 1999-2009Crossref Scopus (5210) Google Scholar, 11Valk P.J. Verhaak R.G. Beijen M.A. Erpelinck C.A. Barjesteh van Waalwijk van.Doorn-Khosrovani Boer J.M. et al.Prognostically useful gene-expression profiles in acute myeloid leukemia.N Engl J Med. 2004; 350: 1617-1628Crossref PubMed Scopus (1097) Google Scholar, 12Lee J.S. Chu I.S. Heo J. Calvisi D.F. Sun Z. Roskams T. et al.Classification and prediction of survival in hepatocellular carcinoma by gene expression profiling.Hepatology. 2004; 40: 667-676Crossref PubMed Scopus (697) Google Scholar, 13Roepman P. Wessels L.F. Kettelarij N. Kemmeren P. Miles A.J. Lijnzaad P. et al.An expression profile for diagnosis of lymph node metastases from primary head and neck squamous cell carcinomas.Nat Genet. 2005; 37: 182-186Crossref PubMed Scopus (336) Google Scholar, 14McLean L.A. Gathmann I. Capdeville R. Polymeropoulos M.H. Dressman M. Pharmacogenomic analysis of cytogenetic response in chronic myeloid leukemia patients treated with imatinib.Clin Cancer Res. 2004; 10: 155-165Crossref PubMed Scopus (74) Google Scholar]. One of the early landmark studies on the application of DNA microarray to tumor classification involved diffused large B-cell lymphoma (DLBCL) (for review see Ref. [[15]Staudt L.M. Dave S. The biology of human lymphoid malignancies revealed by gene expression profiling.Adv Immunol. 2005; 87: 163-208Crossref PubMed Scopus (204) Google Scholar]). Global gene expression profiling uncovered the existence of new diagnostic categories that were comprised of multiple molecularly and clinically distinct diseases. For example, DLBCL now consists of three gene expression subgroups, i.e. germinal center B-cell-like (GCB) DLBCL, activated B-cell-like (ABC) DLBCL, and primary mediastinal B-cell (PMBL). These DLBCL subgroups are posited to arise from different stages of normal B-cell differentiation, utilize distinct oncogenic mechanisms, and differ in their curative responses to chemotherapy. It has also been revealed that key regulatory factors and their target genes are differentially expressed among these subgroups, including BCL-6, Blimp-1, and XBP1. Furthermore, two subtypes (ABC DLBCL and PMBL) are dependent on constitutive activation of the NF-κB pathway for their survival. In contrast, GCB DLBCL is not, indicating that this pathway may be a potential therapeutic target for certain DLBCL subgroups. Importantly, gene expression profiling of DLBCL, mantle cell lymphoma, and follicular lymphoma, has also been used to create gene expression-based models of survival. These models have also identified the biological characteristics of the tumors that influence their behavior. For example, in mantle cell lymphoma, the length of survival following diagnosis is primarily influenced by the tumor proliferation rate, that is quantitatively reproduced by a proliferation gene expression ‘signature’. This proliferation ‘signature’ can now be viewed as a fundamental reflection of all the important oncogenic lesions in DLBCL, mantle cell lymphoma, and follicular lymphoma that increase progression from the G1 to the S phase of the cell cycle. Gene expression profiling in DLBCL and follicular lymphoma has also revealed that the molecular characteristics of non-malignant tumor-infiltrating immune cells have a major influence on the length of survival. Numerous global gene expression profiling studies have been performed for the molecular classification and molecular prognosis of breast cancer (for review see Ref. [[16]Brenton J.D. Carey L.A. Ahmed A.A. Caldas C. Molecular classification and molecular forecasting of breast cancer: ready for clinical application?.J Clin Oncol. 2005; 23: 7350-7360Crossref PubMed Scopus (726) Google Scholar]). These studies, in addition to confirming the old ideas that breast cancer is a heterogeneous group of diseases, have been able to show that the clinical heterogeneity of the disease can be explained by differences in the genomic composition of the primary tumor. Thus, application of gene expression profiling has permitted classification of breast cancers into three broad subtypes with different prognosis, i.e. luminal subtypes (A–C), HER2 subtype, and basal-like breast cancer. In fact, a 70 gene prognostic profile proves to be a more powerful predictor of outcome in young breast cancer patients than the standard system based on clinical and histological criteria [[10]van de Vijver M.J. He Y.D. van't Veer L.J. Dai H. Hart A.A. Voskuil D.W. et al.A gene-expression signature as a predictor of survival in breast cancer.N Engl J Med. 2002; 19: 1999-2009Crossref Scopus (5210) Google Scholar]. The general conclusion that can be drawn from these and other studies not cited is that application of gene expression profiles can: (a) classify tumors into homogeneous subtypes that are characterized by distinct molecular pathways that maintain the malignant phenotype; (b) discover genes associated with cancer prognosis and/or pathological features of the tumors; and (c) provide potentially new therapeutic targets and monitor response to therapy. Hepatocellular carcinoma (HCC) is one of the most common cancers in the world, accounting for an estimated 600,000 deaths annually [[17]Parkin D.M. Bray F. Ferlay J. Pisani P. Global cancer statistics, 2002.CA Cancer J Clin. 2005; 55: 74-108Crossref PubMed Scopus (17205) Google Scholar]. While HCC is common in southeast Asia and sub-Sahara Africa, the incidence rates of HCC have continued to increase in the United States and western Europe over the past 25 years and the incidence and mortality rates of HCC are expected to double over the next 10–20 years [18El Serag H.B. Mason A.C. Rising incidence of hepatocellular carcinoma in the United States.N Engl J Med. 1999; 340: 745-750Crossref PubMed Scopus (2707) Google Scholar, 19El Serag H.B. Davila J.A. Petersen N.J. McGlynn K.A. The continuing increase in the incidence of hepatocellular carcinoma in the United States: an update.Ann Intern Med. 2003; 139: 817-823Crossref PubMed Scopus (835) Google Scholar, 20Davila J.A. Morgan R.O. Shaib Y. McGlynn K.A. El Serag H.B. Hepatitis C infection and the increasing incidence of hepatocellular carcinoma: a population-based study.Gastroenterology. 2004; 127: 1372-1380Abstract Full Text Full Text PDF PubMed Scopus (424) Google Scholar]. Although much is known about both the cellular changes that lead to HCC and the etiological agents (i.e. HBV, HCV infection, and alcohol) responsible for the majority of HCC, the molecular pathogenesis of HCC is not well understood [21Bruix J. Boix L. Sala M. Llovet J.M. Focus on hepatocellular carcinoma.Cancer Cell. 2004; 5: 215-219Abstract Full Text Full Text PDF PubMed Scopus (496) Google Scholar, 22Llovet J.M. Burroughs A. Bruix J. Hepatocellular carcinoma.Lancet. 2003; 362: 1907-1917Abstract Full Text Full Text PDF PubMed Scopus (3751) Google Scholar, 23Thorgeirsson S.S. Grisham J.W. Molecular pathogenesis of human hepatocellular carcinoma.Nat Genet. 2002; 31: 339-346Crossref PubMed Scopus (1260) Google Scholar]. The goal of all staging systems is to separate patients into groups with homogeneous prognosis, which then form the bases for the selection of the most appropriate treatments. Much work has been devoted to establishing prognostic models for HCC by using clinical information and pathological classification in order to provide information at diagnosis on both survival and treatment options [24Bruix J. Llovet J.M. HCC surveillance: who is the target population?.Hepatology. 2003; 37: 507-509Crossref PubMed Scopus (51) Google Scholar, 25Calvet X. Bruix J. Gines P. Bru C. Sole M. Vilana R. et al.Prognostic factors of hepatocellular carcinoma in the west: a multivariate analysis in 206 patients.Hepatology. 1990; 12: 753-760Crossref PubMed Scopus (225) Google Scholar, 26Chevret S. Trinchet J.C. Mathieu D. Rached A.A. Beaugrand M. Chastang C. A new prognostic classification for predicting survival in patients with hepatocellular carcinoma. Groupe d'Etude et de Traitement du Carcinome Hepatocellulaire.J Hepatol. 1999; 31: 133-141Abstract Full Text Full Text PDF PubMed Scopus (430) Google Scholar, 27Okuda K. Ohtsuki T. Obata H. Tomimatsu M. Okazaki N. Hasegawa H. et al.Natural history of hepatocellular carcinoma and prognosis in relation to treatment. Study of 850 patients.Cancer. 1985; 56: 918-928Crossref PubMed Scopus (1771) Google Scholar, 28Pugh R.N. Murray-Lyon I.M. Dawson J.L. Pietroni M.C. Williams R. Transection of the oesophagus for bleeding oesophageal varices.Br J Surg. 1973; 60: 646-649Crossref PubMed Scopus (6641) Google Scholar, 29Tan C.K. Law N.M. Ng H.S. Machin D. Simple clinical prognostic model for hepatocellular carcinoma in developing countries and its validation.J Clin Oncol. 2003; 21: 2294-2298Crossref PubMed Scopus (53) Google Scholar, 30CLIP Investigators. A new prognostic system for hepatocellular carcinoma: a retrospective study of 435 patients: the cancer of the liver italian program (CLIP) investigators. Hepatology 1998;28:751–5.Google Scholar]. Although much progress has been made (reviewed in Ref. [[22]Llovet J.M. Burroughs A. Bruix J. Hepatocellular carcinoma.Lancet. 2003; 362: 1907-1917Abstract Full Text Full Text PDF PubMed Scopus (3751) Google Scholar]), many issues still remain unresolved. For example, a staging system that reliably separates patients with early HCC, as well as intermediate to advanced HCC, into homogeneous groups with respect to prognosis does not exist. This is of particular importance because the natural course of early HCC is unknown and the natural progression of intermediate and advanced HCC are known to be quite variable [[31]Llovet J.M. Fuster J. Bruix J. Prognosis of hepatocellular carcinoma.Hepatogastroenterology. 2002; 49: 7-11PubMed Google Scholar]. This is especially troublesome since the accuracy of imaging techniques is rapidly evolving and affording detection of early HCC nodules [32Kim C.K. Lim J.H. Lee W.J. Detection of hepatocellular carcinomas and dysplastic nodules in cirrhotic liver: accuracy of ultrasonography in transplant patients.J Ultrasound Med. 2001; 20: 99-104Crossref PubMed Scopus (123) Google Scholar, 33Lencioni R. Cioni D. Bartolozzi C. Tissue harmonic and contrast-specific imaging: back to gray scale in ultrasound.Eur Radiol. 2002; 12: 151-165Crossref PubMed Scopus (156) Google Scholar]. Although the pathological diagnosis of high-grade dysplastic nodules (DN) and early HCC is at present controversial, it is likely that many HCC evolve from the DN [[34]Kojiro M. Roskams T. Early hepatocellular carcinoma and dysplastic nodules.Semin Liver Dis. 2005; 25: 133-142Crossref PubMed Scopus (228) Google Scholar]. However, prognostic predictions based on morphological characteristics of these early lesions are still tentative. Numerous studies dealing with gene expression profiling of HCC have appeared during the last 5 years (see Table 1). In addition, several review articles addressing the application of the DNA microarray platform in studies on HCC have recently been published [35Lee J.S. Thorgeirsson S.S. Genetic profiling of human hepatocellular carcinoma.Semin Liver Dis. 2005; 25: 125-132Crossref PubMed Scopus (52) Google Scholar, 36Lee J.S. Grisham J.W. Thorgeirsson S.S. Comparative functional genomics for identifying models of human cancer.Carcinogenesis. 2005; 26: 1013-1020Crossref PubMed Scopus (47) Google Scholar, 37Lee J.S. Thorgeirsson S.S. Genome-scale profiling of gene expression in hepatocellular carcinoma: classification, survival prediction, and identification of therapeutic targets.Gastroenterology. 2004; 127: S51-S55Abstract Full Text Full Text PDF PubMed Scopus (144) Google Scholar, 38Zhang L.H. Ji J.F. Molecular profiling of hepatocellular carcinomas by cDNA microarray.World J Gastroenterol. 2005; 11: 463-468PubMed Google Scholar, 39Kim J.W. Wang X.W. Gene expression profiling of preneoplastic liver disease and liver cancer: a new era for improved early detection and treatment of these deadly diseases?.Carcinogenesis. 2003; 24: 363-369Crossref PubMed Scopus (48) Google Scholar]. The molecular profiling of HCC presents challenges that are not commonly seen in other human tumors. This is primarily due to the complex pathogenesis of this cancer [[23]Thorgeirsson S.S. Grisham J.W. Molecular pathogenesis of human hepatocellular carcinoma.Nat Genet. 2002; 31: 339-346Crossref PubMed Scopus (1260) Google Scholar]. HCC arises most commonly in cirrhotic livers following infection with HBV or HCV. However HCC can also occur under a variety of other conditions such as hemochromatosis, excessive alcohol consumption and non-alcoholic steatohepatitis. Each of these conditions represents complex and different constellations of chromosomal aberrations and genetic and epigenetic alterations as well as changed molecular pathways [[23]Thorgeirsson S.S. Grisham J.W. Molecular pathogenesis of human hepatocellular carcinoma.Nat Genet. 2002; 31: 339-346Crossref PubMed Scopus (1260) Google Scholar]. Nevertheless, global gene expression profiling, because of its extraordinary power of resolution, may currently be the most appropriate technology platform to molecularly resolve the complex pathogenesis of HCC. Indeed, application of gene expression profiling of HCC has so far identified subgroups of patients according to etiological factors [40Okabe H. Satoh S. Kato T. Kitahara O. Yanagawa R. Yamaoka Y. et al.Genome-wide analysis of gene expression in human hepatocellular carcinomas using cDNA microarray: identification of genes involved in viral carcinogenesis and tumor progression.Cancer Res. 2001; 61: 2129-2137PubMed Google Scholar, 41Delpuech O. Trabut J.B. Carnot F. Feuillard J. Brechot C. Kremsdorf D. Identification, using cDNA macroarray analysis, of distinct gene expression profiles associated with pathological and virological features of hepatocellular carcinoma.Oncogene. 2002; 21: 2926-2937Crossref PubMed Scopus (96) Google Scholar, 42Iizuka N. Oka M. Yamada-Okabe H. Mori N. Tamesa T. Okada T. et al.Comparison of gene expression profiles between hepatitis B virus- and hepatitis C virus-infected hepatocellular carcinoma by oligonucleotide microarray data on the basis of a supervised learning method.Cancer Res. 2002; 62: 3939-3944PubMed Google Scholar], early pre-neoplastic lesions [[43]Nam S.W. Park J.Y. Ramasamy A. Shevade S. Islam A. Long P.M. et al.Molecular changes from dysplastic nodule to hepatocellular carcinoma through gene expression profiling.Hepatology. 2005; 42: 809-818Crossref PubMed Scopus (146) Google Scholar], stages of the disease [44Smith M.W. Yue Z.N. Geiss G.K. Sadovnikova N.Y. Carter V.S. Boix L. et al.Identification of novel tumor markers in hepatitis C virus-associated hepatocellular carcinoma.Cancer Res. 2003; 63: 859-864PubMed Google Scholar, 45Ye Q.H. Qin L.X. Forgues M. He P. Kim J.W. Peng A.C. et al.Predicting hepatitis B virus-positive metastatic hepatocellular carcinomas using gene expression profiling and supervised machine learning.Nat Med. 2003; 9: 416-423Crossref PubMed Scopus (711) Google Scholar], rate of recurrence [46Iizuka N. Oka M. Yamada-Okabe H. Nishida M. Maeda Y. Mori N. et al.Oligonucleotide microarray for prediction of early intrahepatic recurrence of hepatocellular carcinoma after curative resection.Lancet. 2003; 361: 923-929Abstract Full Text Full Text PDF PubMed Scopus (438) Google Scholar, 47Kurokawa Y. Matoba R. Takemasa I. Nagano H. Dono K. Nakamori S. et al.Molecular-based prediction of early recurrence in hepatocellular carcinoma.J Hepatol. 2004; 41: 284-291Abstract Full Text Full Text PDF PubMed Scopus (103) Google Scholar], and survival [[12]Lee J.S. Chu I.S. Heo J. Calvisi D.F. Sun Z. Roskams T. et al.Classification and prediction of survival in hepatocellular carcinoma by gene expression profiling.Hepatology. 2004; 40: 667-676Crossref PubMed Scopus (697) Google Scholar]. These datasets represent an impressive progress in the use of gene expression profiling in elucidating the molecular pathogenesis of HCC and hold the promise of improving the diagnostic and prognostic prediction for HCC patients. The dataset is also large enough to warrant a critical examination of reproducibility and validation of the molecular classification of HCC and the predictive expression ‘signatures’ (markers) generated by the microarray based global gene expression profiling of HCC. The molecular classification of HCC has to be evaluated in the context of the currently used clinco-pathological classifications of HCC. Although this is not the subject of this review, it appears evident that a dynamic integration of the molecular dataset with the clinical and pathologic features of the tumors would provide the most useful clinical information. Here, we will focus on the predictive gene expression ‘signatures’ for HCC.Table 1Summary of gene expression profile studies of liver diseaseYearDiseaseProbesPlatformNo. of patientNo. of probes (or gene features)No. of significant genesRefs.2000HCCcDNAMembrane10432516Oncol Res. 2000;12(2):59–692001HepatitiscDNASlide321080137Gastroenterology 2001 Mar;120(4):955–662001HCCcDNASlide10108010Hepatology 2001 Apr;33(4):832–402001HCCcDNAMembrane914,000Not describedCancer Res. 2001 Apr 1;61(7):3176–812001HCCcDNASlide2023,040335Cancer Res. 2001 Mar 1;61(5):2129–372002HCCcDNAMembrane15109844Oncogene 2002 Apr 25;21(18):2926–372001HCCOligoAffymetrix1142,0001235Cancer 2001 Jul 15;92(2):395–4052001HCCcDNAMembrane2912,3932253Proc. Natl. Acad. Sci. USA 2001 Dec 18;98(26):15089–942002HCCcDNAMembrane35885Mol. Carcinog. 2002 Feb;33(2):113–242002HCCOligoAffymetrix45712983Cancer Res. 2002 Jul 15;62(14):3939–442002HCCcDNASlide1212,8001820J. Cancer Res. Clin. Oncol. 2002 Jul;128(7):369–792002HCCcDNASlide8223,0753180Mol. Biol. Cell. 2002 Jun;13(6):1929–392002HCCcDNASlide8123,0757830Cancer Res. 2002 Aug 15;62(16):4711–212003HCCOligoAffymetrix712,60092Hepatology 2003 Jan;37(1):198–2072003HCCOligoAffymetrix60600012Lancet 2003 Mar 15;361(9361):923–92003HCCPCR203072220J. Hepatol. 2003 Dec;39(6):1004–122003HCCcDNASlide409180153Nat. Med. 2003 Apr;9(4):416–232003HCCOligoAffymetrix45600089Oncogene 2003 May 15;22(19):3007–142003HCC and cirrhosiscDNASlide3213,597132Hepatology 2003 Dec;38(6):1458–672003HCCcDNASlide2513,5972302Cancer Res. 2003 Feb 15;63(4):859–642004HCCPCR203072117J. Exp. Clin. Cancer Res. 2004 Mar;23(1):135–412004HCCcDNASlide739180489Hepatology 2004 Feb;39(2):518–272004HCCOligoSlide9021,329406Hepatology 2004 Sep;40(3):667–762004HCCcDNASlide228464668World J. Gastroenterol. 2004 Dec 15;10(24):3569–732004HCCcDNASlide68740044Biochim. Biophys. Acta 2004 Dec 24;1739(1):50–612004HCCcDNASlide379000218Hepatology 2004 Apr;39(4):944–532004HCCOligoAffymetrix456000176Int. J. Oncol. 2004 Mar;24(3):565–742004HCCPCR100307292J. Hepatol. 2004 Aug;41(2):284–912005HCCOligoSlide4218,664120Hepatology 2005 Oct;42(4):809–182005HCC and HCAOligoAffymetrix1454,00063Am. J. Surg. Pathol. 2005 Dec;29(12):1600–82005HCC and cirrhosiscDNASlide36107240Biochem. Biophys. Res. Commun. 2005 Aug 26;334(2):681–8 Open table in a new tab It has long been recognized that survival prediction of HCC patients is more challenging than with most other cancers. This is, in the cases of HBV and HCV, the consequence of the underlying viral driven non-neoplastic disease, i.e. chronic hepatitis and cirrhosis that can and does inflict functional impairment on the liver that may affect the outcome of the HCC patients. However, in a recent study on survival of HCC patients, it was demonstrated that the HCC was the prime cause of death in patients with compensated cirrhosis [[48]Sangiovanni A. Del Ninno E. Fasani P. De Fazio C. Ronchi G. Romeo R. et al.Increased survival of cirrhotic patients with a hepatocellular carcinoma detected during surveillance.Gastroenterology. 2004; 126: 1005-1014Abstract Full Text Full Text PDF PubMed Scopus (503) Google Scholar]. In a recent global gene expression analysis of human HCC, Lee et al. [[12]Lee J.S. Chu I.S. Heo J. Calvisi D.F. Sun Z. Roskams T. et al.Classification and prediction of survival in hepatocellular carcinoma by gene expression profiling.Hepatology. 2004; 40: 667-676Crossref PubMed Scopus (697) Google Scholar] identified two distinctive subclasses that are highly associated with the survival of the patients. A limited number of genes were identified that both predicted the length of survival of the HCC patients and provided new molecular insights into the pathogenesis of HCC (Fig. 1). Application of a knowledge-based annotation of the 406 ‘survival’ genes revealed molecular pathways responsible for the biological differences observed in the two subclasses of HCC. As expected, measurement of cell proliferation and apoptosis provided the best quantitative separation of the two survival subclasses. However, there were other notable differences. For example, the low survival subclass displayed higher expression of genes involved in ubiquitination and histone modification, suggesting an etiological involvement of these processes in accelerating the progression of HCC. Indeed, it is well established that the ubiquitin system is frequently deregulated in cancers [[49]Pagano M. Benmaamar R. When protein destruction runs amok, malignancy is on the loose.Cancer Cell. 2003; 4: 251-256Abstract Full Text Full Text PDF PubMed Scopus (74) Google Scholar], and has been proposed as a possible predictive marker for recurrence of human HCC [[50]Shirahashi H. Sakaida I. Terai S. Hironaka K. Kusano N. Okita K. Ubiquitin is a possible new predictive marker for the recurrence of human hepatocellular carcinoma.Liver. 2002; 22: 413-418Crossref PubMed Scopus (20) Google Scholar]. The predictive power of the survival gene expression ‘signature’ has now been validated in an independent HCC dataset (unpublished data). Nevertheless, it needs to be emphasized that a considerable molecular heterogeneity still exists within each of the HCC survival subclasses. It is therefore likely that more subclasses of HCC that differ in the length of survival will emerge as we learn more about the molecular pathogenesis of HCC and additional data become available. HCC recurrence is a serious complication following resection of the primary tumor and happens in 50% of cases 3 years after the operation [[22]Llovet J.M. Burroughs A. Bruix J. Hepatocellular carcinoma.Lancet. 2003; 362: 1907-1917Abstract Full Text Full Text PDF PubMed Scopus (3751) Google Scholar]. In 75% of the cases this is due to intrahepatic metastasis whereas the remaining 25% are due to de novo HCC [[51]Cheung S.T. Chen X. Guan X.Y. Wong S.Y. Tai L.S. Ng I.O. et al.Identify metastasis-associated genes in hepatocellular carcinoma through clonality delineation for multinodular tumor.Cancer Res. 2002; 62: 4711-4721PubMed Google Scholar]. The major histopathological features that predict HCC recurrence are vascular invasion, degree of differentiation of the tumor, and multi-nodularity [[22]Llovet J.M. Burroughs A. Bruix J. Hepatocellular carcinoma.Lancet. 2003; 362: 1907-1917Abstract Full Text Full Text PDF PubMed Scopus (3751) Google Scholar]. Several recent studies have employed the gene expression profiling to address the issue of HCC recurrence following resection and intrahepatic metastasis. We selected three studies to illustrate both the promise and challenge of selecting and using ‘recurrence specific gene expression signature’. Iizuka et al. [[46]Iizuka N. Oka M. Yamada-Okabe H. Nishida M. Maeda Y. Mori N. et al.Oligonucleotide microarray for prediction of early intrahepatic recurrence of hepatocellular carcinoma after curative resection.Lancet. 2003; 361: 923-929Abstract Full Text Full Text PDF PubMed Scopus (438) Google Scholar] investigated mRNA expression profiles in tissue specimens from a training set, comprising 33 patients with hepatocellular carcinoma, with high-density oligonucleotide microarrays representing about 6000 genes. The training set was used in a supervised learning manner to construct a predictive system, consisting of 12 genes. The predictive performance of the system was then compared on a blinded set of samples from 27 newly enrolled patients. Early intrahepatic recurrence within 1 year after curative surgery occurred in 12 (36%) and 8 (30%) patients in the training and blinded sets, respectively. The system correctly predicted early intrahepatic recurrence or non-recurrence in 25 (93%) of 27 samples in the blinded set and had a positive predictive value of 88% and a negative predictive value of 95%. In the second study, Kurokawa et al. [[47]Kurokawa Y. Matoba R. Takemasa I. Nagano H. Dono K. Nakamori S. et al.Molecular-based prediction of early recurrence in hepatocellular carcinoma.J Hepatol. 2004; 41: 284-291Abstract Full Text Full Text PDF PubMed Scopus (103) Google Scholar] addressed the issue of intrahepatic recurrence by analyzing gene expression using a PCR-based array platform of 3072 genes in 100 HCC patients. The authors selected 92 genes that demonstrated distinct expression patterns differing significantly between recurrence and recurrence-free cases. Using the 20 top-ranked genes (from the 92 selected), the authors were able to correctly predict the early intrahepatic recurrence for 29 of 40 cases within the validation group, with the odds ratio of 6.8 (95%CI 1.7–27.5, P=0.010). The 2-year recurrence rates in the patients with the good signature and those with the poor signature were 29.4 and 73.9%, respectively. The authors further showed (using multivariate Cox analysis) that the 20 gene molecular-signature was an independent indicator for recurrence (hazard ratio 3.82, 95%CI 1.44–10.10, P=0.007). In the third study, Ye et al. [[45]Ye Q.H. Qin L.X. Forgues M. He P. Kim J.W. Peng A.C. et al.Predicting hepatitis B virus-positive metastatic hepatocellular carcinomas using gene expression profiling and supervised machine learning.Nat Med. 2003; 9: 416-423Crossref PubMed Scopus (711) Google Scholar] analyzed the expression profiles of 67 primary and metastatic HCC samples from 40 patients. Using a supervised machine-learning algorithm, the authors generated a 153 gene molecular signature that permitted classification of metastatic HCC patients and identified genes that were relevant to metastasis and patient survival. The authors further showed that the gene expression signature of primary HCC with accompanying metastasis was very similar to that of their corresponding metastases, implying that genes favoring metastasis progression were initiated in the primary tumors. Furthermore, osteopontin, which was identified as a lead gene in the signature, was over-expressed in metastatic HCC and an osteopontin-specific antibody effectively blocked HCC cell invasion in vitro and inhibited pulmonary metastasis of HCC cells in nude mice. Unlike traditional experiments that usually examine single or few genes at a time, the outcome of microarray experiments is inherently under the huge influence of variation because a large number of genes is analyzed in parallel. In general, the increased variation in high throughput studies escalates the level of noise. Moreover, many confounding factors are embedded in the gene expression profile data from human cancer tissues. These factors include ages, hospital care, different protocols of treatments, non-parallel progression of cancer, and unspecified environmental factors which are irrelevant to HCC development. These contribute to the limited success in generating robust gene expression signatures for predicting prognosis. The standard strategy for estimating the accuracy of classification methods is to apply a training-validation approach in which the training set is utilized to identify the molecular signature and the validation set is used to estimate the degree of reliability. In a recent paper, Michiels and colleagues assessed the robustness of the gene expression signatures to predict prognosis of cancer [[52]Michiels S. Koscielny S. Hill C. Prediction of cancer outcome with microarrays: a multiple random validation strategy.Lancet. 2005; 365: 488-492Abstract Full Text Full Text PDF PubMed Scopus (796) Google Scholar]. The aim of the study was to evaluate the extent to which the molecular signature depends on the constitution of the training set and to study the distribution of misclassification rates across validation sets by applying a multiple random training-validation strategy. The authors re-analyzed data from seven large published studies that attempted to predict prognosis of cancer patients (including the study by Iizuka et al. [[46]Iizuka N. Oka M. Yamada-Okabe H. Nishida M. Maeda Y. Mori N. et al.Oligonucleotide microarray for prediction of early intrahepatic recurrence of hepatocellular carcinoma after curative resection.Lancet. 2003; 361: 923-929Abstract Full Text Full Text PDF PubMed Scopus (438) Google Scholar] on prediction of early intrahepatic recurrence of hepatocellular carcinoma after curative resection). In general they discovered that the list of genes identified as predictors of prognosis was highly unstable, and the selected molecular signatures strongly depended on the selection of patients in the training sets. Furthermore, in all but one study, the proportion that was misclassified decreased as the number of patients in the training set increased. Moreover, the results revealed that five of the seven studies did not classify patients better than chance. Their re-analysis of the microarray data from Iizuka and colleagues [[46]Iizuka N. Oka M. Yamada-Okabe H. Nishida M. Maeda Y. Mori N. et al.Oligonucleotide microarray for prediction of early intrahepatic recurrence of hepatocellular carcinoma after curative resection.Lancet. 2003; 361: 923-929Abstract Full Text Full Text PDF PubMed Scopus (438) Google Scholar] and a training set of 34 patients (using 500 signatures regenerated from re-analyses) revealed that only four of 12 published signature genes were seen in more than 250 of their signatures, whereas nine not present in the published signature were also selected in more than 250 estimated signatures. In this context it is of interest, and based on the data generated by Michiels and colleagues perhaps expected, that no overlap of the 20 genes that constituted the recurrence gene expression signature in the study of Kurokawa et al. [[47]Kurokawa Y. Matoba R. Takemasa I. Nagano H. Dono K. Nakamori S. et al.Molecular-based prediction of early recurrence in hepatocellular carcinoma.J Hepatol. 2004; 41: 284-291Abstract Full Text Full Text PDF PubMed Scopus (103) Google Scholar] is seen with recurrence of the 12 gene expression signature reported by Iizuka et al. [[46]Iizuka N. Oka M. Yamada-Okabe H. Nishida M. Maeda Y. Mori N. et al.Oligonucleotide microarray for prediction of early intrahepatic recurrence of hepatocellular carcinoma after curative resection.Lancet. 2003; 361: 923-929Abstract Full Text Full Text PDF PubMed Scopus (438) Google Scholar]. Comparison of the survival data from Lee et al. [[12]Lee J.S. Chu I.S. Heo J. Calvisi D.F. Sun Z. Roskams T. et al.Classification and prediction of survival in hepatocellular carcinoma by gene expression profiling.Hepatology. 2004; 40: 667-676Crossref PubMed Scopus (697) Google Scholar] with the data reported by Ye et al. [[45]Ye Q.H. Qin L.X. Forgues M. He P. Kim J.W. Peng A.C. et al.Predicting hepatitis B virus-positive metastatic hepatocellular carcinomas using gene expression profiling and supervised machine learning.Nat Med. 2003; 9: 416-423Crossref PubMed Scopus (711) Google Scholar] and Iizuka et al. [[46]Iizuka N. Oka M. Yamada-Okabe H. Nishida M. Maeda Y. Mori N. et al.Oligonucleotide microarray for prediction of early intrahepatic recurrence of hepatocellular carcinoma after curative resection.Lancet. 2003; 361: 923-929Abstract Full Text Full Text PDF PubMed Scopus (438) Google Scholar] showed that genes associated with early recurrence and intrahepatic metastasis of HCC did not discriminate between the subclass A and B, suggesting that the information (at least from a gene expression standpoint) embedded in these important processes is not sufficient to predict survival. It is therefore likely that the additional information provided by the survival genes (only two of the genes associated with intrahepatic metastasis were among these) is needed for effectively predicting survival. However, considerable molecular heterogeneity still exists within each HCC subclass, as evidenced by quantitative differences in survival gene expression [[12]Lee J.S. Chu I.S. Heo J. Calvisi D.F. Sun Z. Roskams T. et al.Classification and prediction of survival in hepatocellular carcinoma by gene expression profiling.Hepatology. 2004; 40: 667-676Crossref PubMed Scopus (697) Google Scholar] and the small fraction of patients that are frequently misclassified in the prediction models. It is therefore probable that more subclasses of HCC might emerge when gene expression data from more HCC patients become available. The conclusion reached by Michiels and colleagues [[52]Michiels S. Koscielny S. Hill C. Prediction of cancer outcome with microarrays: a multiple random validation strategy.Lancet. 2005; 365: 488-492Abstract Full Text Full Text PDF PubMed Scopus (796) Google Scholar] was that, because of inadequate validation of the results, these studies were overoptimistic and they recommended the use of validation by repeated random sampling. They also emphasized that studies with large sample size are needed before expression profiling can be utilized in the clinic. It is clear from this analysis that inadequate validation of molecular signatures used for prognostic classification of HCC (and other cancers) may be one of the major sources of failure in applying ‘molecular signatures’ in the clinic. Profiling liver cancer and indeed cancer in general with gene expression arrays has become common. The results from early studies, particularly those unraveling novel cancer classifications and identifying novel markers for prediction of clinical outcome, kindled the hope that this technology would provide understanding of the molecular differences between clinical cases and allow individualization of care. There can be no doubt that the DNA microarray technology has provided an extraordinary opportunity to perform integrative analyses of the cancer transcriptome. Furthermore, the results of array based gene expression profiles have started to impact both clinical decision-making in oncology and advanced our understanding of cancer biology, as well as facilitated the development of more effective therapies.