Reply to: “Network-based discovery of gene signature for vascular invasion prediction in HCC”

Network-based discovery of gene signature for vascular invasion prediction in HCCJournal of HepatologyVol. 56Issue 6PreviewWe read with great interest the paper by Minguez and colleagues [1]. In this study, the authors successfully defined a 35-gene signature of vascular invasion (VI) by gene expression profiling of HCV-related hepatocellular carcinoma (HCC) samples, and validated this gene panel in an independent mixed cohort of patients with various etiologies, including HBV, HCV, and alcohol. It is already known that VI can predict recurrence and survival in HCC patients after tumor resection or liver transplantation [2]. Full-Text PDF Open AccessGene-expression signature of vascular invasion in hepatocellular carcinomaJournal of HepatologyVol. 55Issue 6PreviewVascular invasion is a major predictor of tumor recurrence after surgical treatments for hepatocellular carcinoma (HCC). While macroscopic vascular invasion can be detected by radiological techniques, pre-operative detection of microscopic vascular invasion, which complicates 30–40% of patients with early tumors, remains elusive. Full-Text PDF Liu and colleagues raise some issues regarding our recently published study [[1]Mínguez B. Hoshida Y. Villanueva A. et al.Gene-expression signature of vascular invasion in hepatocellular carcinoma.J Hepatol. 2011; 55: 1325-1331Abstract Full Text Full Text PDF PubMed Scopus (111) Google Scholar] to which we would like to make the following comments. We acknowledge the limitations that a gene-expression-based biomarker could have, and that our signature is not unique. Certainly, previous attempts to find such a signature have been published in the past [[2]Ho M.C. Lin J.J. Chen C.N. Chen C.C. Lee H. Yang C.Y. et al.A gene expression profile for vascular invasion can predict the recurrence after resection of hepatocellular carcinoma: a microarray approach.Ann Surg Oncol. 2006; 13: 1474-1484Crossref PubMed Scopus (51) Google Scholar]. We also know that, as in other gene expression studies, potential bias could occur. In fact, reported prognostic signatures are often not reproducible, in most of the cases due to suboptimal study design, small sample sizes, and also because many of them have been based on retrospectively collected tissue samples [[3]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]. Even after taking into account these sources of bias or inconsistencies, it so happens that only a small minority of the reported signatures truly retain prognostic significance. In fact, our recent outcome analysis including 22 gene signatures with prognostic significance in HCC (18 from the tumor, and four from the non-tumoral adjacent tissue) showed that only two signatures retained independent prognostic value [[4]Villanueva A. Hoshida Y. Battiston C. et al.Combining clinical, pathology, and gene expression data to predict recurrence of hepatocellular carcinoma.Gastroenterology. 2011; 140: 1501-1512Abstract Full Text Full Text PDF PubMed Scopus (316) Google Scholar]. In our work, we select a training cohort based upon a homogeneous etiology to minimize the risk of molecular heterogeneity and to identify a clean and distinct signature. Patients with HCV-related HCC were selected, since this is the most common etiology in the Western countries. Then, we validate the signature in an independent multi-etiologic cohort of patients and the accuracy remained stable when an etiology-dependent subgroup analysis was performed [[1]Mínguez B. Hoshida Y. Villanueva A. et al.Gene-expression signature of vascular invasion in hepatocellular carcinoma.J Hepatol. 2011; 55: 1325-1331Abstract Full Text Full Text PDF PubMed Scopus (111) Google Scholar]. The study was aimed at providing a gene-set to ease the preoperative diagnosis of vascular invasion, but was not designed for defining outcome prediction. Nonetheless, we have data indicating that the presence of a vascular invasion signature correlates with poor outcome, since the signature was found to be associated with early recurrence (p = 0.057), and was enriched in patients sharing signatures of poor prognosis [[4]Villanueva A. Hoshida Y. Battiston C. et al.Combining clinical, pathology, and gene expression data to predict recurrence of hepatocellular carcinoma.Gastroenterology. 2011; 140: 1501-1512Abstract Full Text Full Text PDF PubMed Scopus (316) Google Scholar]. Even considering that the question posed is simple (to identify a gene-signature capturing vascular invasion) the characteristics of patients, sample collection, sampling issues, technical variation, validation of results, and bioinformatics approaches are certainly heterogeneous, and thus the results might vary. In most instances, however, the different signatures seem to be able to capture common oncogenic mechanisms, as reflected by their capacity to adequately allocate patients into a poor or good prognosis group [[5]Kim K. Zakharkin S.O. Allison D.B. Expectations, validity, and reality in gene expression profiling.J Clin Epidemiol. 2010; 63: 950-959Abstract Full Text Full Text PDF PubMed Scopus (39) Google Scholar]. By applying a different methodological approach (weighted gene co-expression network analysis) to our data, Liu et al. provide a 9-gene signature with similar accuracy and no overlap with our 35-gene signature. The method applied is based on systems biology to find clusters of highly correlated genes across microarray samples, identify hubs of each module and correlate them with clinical traits [[6]Langfelder P. Horvath S. WGCNA: an R package for weighted correlation network analysis.BMC Bioinf. 2008; 29: 559Crossref Scopus (9185) Google Scholar]. This analysis is based on the hypothesis that information on signaling pathways is crucial to understand how genes are connected to each other and how they influence cellular functions in both normal and cancer conditions. This result further underlines the need for integrating the vast amount of available data and the development of powerful bioinformatics resources (annotation, methodologies, technical platforms, etc.). A more relevant question is when can our signature-alone or in combination with tumor size- be translated into clinical practice. Strict rules have been proposed recently by Simon and colleagues [[7]Simon R.M. Paik S. Hayes D.F. Use of archived specimens in evaluation of prognostic and predictive biomarkers.J Natl Cancer Inst. 2009; 101: 1446-1452Crossref PubMed Scopus (792) Google Scholar]. Following this proposal, the EASL-EORTC guidelines on management of HCC have outlined a list of requirements in order to adopt molecular signatures in the clinical practice [[8]EASL-EORTC Clinical Practice Guidelines: Management of hepatocellular carcinoma. J Hepatol 2012;56:908–943.Google Scholar], which are as follows:1.First, the signature should be generated in the setting of randomized studies or in case of cohort studies, it should follow the training/validation approach.2.The signature should retain independent prognostic value when tested along known clinico-pathological variables.3.The results should be confirmed by independent investigators in a separate set of samples. Thus, according to these rules, in order to implement our signature in the decision-making process, for instance in the waiting list of liver transplantation, it should be validated by independent investigators in a novel set of samples. Ideally, the signature has to be reproduced in a device, which should give similar results. Only then, data is ready for acceptance in guidelines. It is a long path, but the only one for translation of genomic results into our practice. The authors declared that they do not have anything to disclose regarding funding or conflict of interest with respect to this manuscript.