Molecular classification of hepatocellular carcinoma: The view from metabolic zonation

Potential conflict of interest: Nothing to report. See Article on Page 1502 Hepatocellular carcinomas (HCCs) arise from diverse etiological backgrounds and display remarkable biological and clinical heterogeneity. Unsupervised analyses of gene expression profiles variably subdivide HCCs into between two and six subclasses with distinct transcriptomic profiles and clinicopathologic features.1 Nevertheless, HCC molecular subclassification has not been implemented in the clinic, hampered, in part, by a lack of consensus between the various classifications and their prognostic and, in particular, predictive implications. Sorafenib, the only approved systemic treatment, on average prolongs patient survival by 2.8 months.6 However, robust predictive biomarkers for sorafenib response have not been defined. Indeed, prognostication and treatment decisions for HCC patients are currently made primarily based on the Barcelona Clinic Liver Cancer (BCLC) staging system. Nonetheless, several recurrent themes have emerged from the molecular analyses. HCCs can be broadly classified as proliferative and nonproliferative in roughly equal proportions. Compared to nonproliferative HCCs, proliferative HCCs are typically more aggressive and less differentiated3 and are frequently associated with high serum α‐fetoprotein (AFP) levels, TP53 mutations, and poor outcome.1 Furthermore, proliferative HCCs may display transforming growth factor beta (TGF‐β), MET, AKT, and/or insulin‐like growth factor 2 pathway activation1 and/or the progenitor cell phenotype.3 By contrast, the nonproliferative subgroup appears to be heterogeneous with less‐certain biological significance. Whereas a subset (∼30%) of the nonproliferative HCCs harbor mutations in β‐catenin (encoded by catenin beta‐1 [CTNNB1]) and show β‐catenin pathway activation,1 there is no consensus on the distinguishing features of the remaining nonproliferative HCCs. In the current issue of Hepatology, Désert et al.7 further our understanding of nonproliferative HCCs through the lens of the metabolic zonation program of the liver. Liver parenchymal cells display a gradient of metabolic processes along the porto‐central axis relative to the vascular structure of the liver (Fig. 1). For instance, gluconeogenesis and urea synthesis are primarily performed by hepatocytes near the portal vein (“periportal”), whereas lipogenesis and glycolysis are increased on the central end (“perivenous” or “pericentral”). β‐catenin‐mediated Wnt signaling and hepatocyte nuclear factor 4 alpha (HNF4A)‐regulated gene networks play important roles in governing metabolic zonation,8 where in the periportal hepatocytes, T‐cell factor 4 (TCF4) induces the transcription of HNF4A‐regulated genes in the absence of β‐catenin, and in the periportal hepatocytes, β‐catenin allows TCF4 to bind to Wnt‐response elements, thus inducing the transcription of β‐catenin‐induced genes.9 Through a meta‐analysis of 1,113 HCCs previously profiled using gene expression microarrays, Désert et al.7 report that nonproliferative HCCs are divided into two distinct subclasses, each preserving the periportal or perivenous phenotypes of the metabolic zonation that is critical for normal liver functioning.Figure 1: Nonproliferative HCCs recapitulate metabolic zonation. The metabolic zonation program of the liver stipulates that the metabolic functions of hepatocytes form a spatial gradient in relation to their proximity to either the portal vein (“periportal”) or the central vein (“perivenous”). Periportal hepatocytes are involved in functions such as gluconeogenesis and amino acid catabolism, whereas perivenous hepatocytes perform functions such as glycolysis and lipogenesis. Nonproliferative HCCs exhibit molecular features that are reminiscent of either the periportal (“periportal‐type HCC”) or the perivenous (“perivenous‐type HCC”) hepatocytes. Periportal‐type HCCs are associated with favorable outcome, well‐differentiated HCC, and HNF4A‐associated gene networks. Perivenous‐type HCCs are associated with less‐favorable outcome, increased frequency of CTNNB1 mutations and β‐catenin activation, and increased expression of metastasis gene signature.To enable the prediction of CTNNB1 mutation status in the microarray metadata set, the investigators first defined a robust five‐gene signature using independent training and validation data sets with known CTNNB1 mutation status. The signature consists of three (glutamate‐ammonia ligase [GLUL], leucine‐rich repeat‐containing G‐protein coupled receptor 5 [LRG5], and odontogenic ameloblast‐associated protein [ODAM]) and two (histidine ammonia lyase [HAL] and vanin 1 [VNN1]) genes whose expression positively and negatively, respectively, correlates with CTNNB1 mutation status. Whereas GLUL and LRG5 are well‐known Wnt target genes, the remaining three constitute novel HCC biomarkers. Using HCC cell lines, the investigators demonstrate that activating β‐catenin signaling using glycogen synthase kinase 3 beta (GSK3β) inhibitor 6‐bromoindirubin‐3′‐oxime up‐regulates ODAM and attenuates HAL and VNN1 expression. Indeed, a network analysis of mouse liver periportal and perivenous signatures showed that, although ODAM, HAL, and VNN1 were not present in the signatures, they were highly connected to the genes in the signatures. The five‐gene signature was found to have accuracies of 87% and 93% in the training and validation sets and was used to predict CTNNB1 mutation status for the meta‐analysis. The investigators report that hierarchical clustering of the 1,113 HCCs revealed four subclasses with different prognoses. “ECM‐type” (extracellular matrix) and “STEM‐type” HCCs display signatures of high tumor cell proliferation and are associated with the S1/S2 (Wnt/TGF‐β, poor prognosis) subclasses.3 Additionally, the ECM‐type subclass is characterized by signatures of ECM modeling, integrin signaling, and epithelial‐to‐mesenchymal transition, whereas the STEM‐type subclass is enriched for cancer stem cell, metastasis, cell‐cycle progression, and p53 mutation signatures. The remaining two subclasses display signatures of low proliferation and favorable prognosis and both show little intracluster variability. What is interesting, however, is that, gene set enrichment analyses revealed that these two classes reflect the tightly regulated metabolic zonation program of the liver (Fig. 1). These two subclasses were named “perivenous‐type” (PV) and “periportal‐type” (PP) HCCs to reflect their resemblance to the phenotypes at the two ends of the metabolic zonation spectrum. The PV subclass is highly enriched for predicted CTNNB1 mutations and Wnt activation, the G6 subclass,1 and perivenous hepatocyte gene signatures, including lipid and bile salt metabolism signatures. By contrast, the PP subclass is enriched for periportal hepatocyte gene signatures, such as those of gluconeogenesis and amino acid catabolism, and was enriched for the S3 subclass.3 Finally, given the role of HNF4A, the investigators demonstrated that the expression profiles of PP‐type HCCs show a strong enrichment of HNF4A‐regulated genes. Clinically, PP, PV, ECM, and STEM types form a continuum of increasingly aggressive clinical behavior in terms of tumor aggressiveness, tumor node metastasis (TNM) staging, BCLC stages, vascular invasion, serum AFP concentrations, overall survival, and disease‐free survival. The current study by Désert et al.7 highlights the importance of interpreting HCC molecular classification in the context of basic liver biology and anatomy. Previous studies invariably identified a transcriptomically heterogeneous subset of well‐differentiated HCCs that display features consistent with or suggestive of hepatocytes. Whereas some of these nonproliferative HCCs harbor CTNNB1 mutations and show β‐catenin pathway activation,1 there is no agreement on the biological significance of the CTNNB1‐wild‐type subset. When viewed from the perspective of metabolic zonation, Désert et al.7 demonstrate that nonproliferative HCCs recapitulate the phenotypes at the two distinct ends of the zonation program. In this context, the less‐aggressive clinical behavior of nonproliferative HCCs can be interpreted as their higher degree of preservation of normal liver phenotype. In particular, the lack of β‐catenin activation in the PP‐type HCCs would also be consistent with their least aggressive clinical and biological behavior. Notably, on the periportal versus perivenous spectrum, the only subclass that leans toward a perivenous phenotype is the PV‐like, underscoring its unique biology compared to the other subclasses. Finally, the current study also provides a plausible explanation for the lack of prognostic difference between patients with CTNNB1‐mutant or ‐wild‐type HCCs,10 given that PV‐type HCCs driven by β‐catenin activation sit between proliferative HCCs and PP‐type HCCs in terms of prognosis and highlights the molecular heterogeneity within the CTNNB1‐wild‐type subset. A more speculative interpretation of the current results raises the intriguing question of the cell of origin of HCCs. It remains to be seen whether the metabolic signatures observed in the various HCC subclasses merely reflect the metabolic programming along the porto‐central axis or are indicative of their spatial origin. A contextualized view of HCC molecular subclassifcation is crucial toward understanding HCC biology and identifying drug targets that consider the unique biology of liver cancer. The study by Désert et al.7 represents a significant step toward a unifying molecular classification of HCCs. Going forward, it will be of interest to validate the results from the current study on biopsy and resected HCC samples, aiming not only to obtain prognostic indicators, but also to discover predictive biomarkers.