Prognostic significance of hypoxic and metabolic gene profiling in hepatocellular carcinoma

Hepatocellular carcinoma (HCC) is characterized by high clinical and biological heterogeneity, depending on the extremely variable combinations of pathways, linked with immune mechanisms, neo‐angiogenesis, ECM remodeling, metabolism and/or hypoxia. We recently identified a 5‐genes neo‐angiogenic transcriptomic signature (TS), able to discriminate between “aggressive” HCCs (TS‐positive) from “bland” HCCs (TS negative), the former having extremely poor survival. The aim of this study was to compare gene expression of our HCC cohort with gene expression of well‐characterized, published signatures, which have been related with several different functions potentially relevant in carcinogenesis (ie immune control, hypoxia, metabolism, vascular invasion). We also aimed to ascertain the prognostic power for survival.


| INTRODUC TI ON
In the last years, a huge mass of biologic data has unveiled the high clinical heterogeneity of hepatocellular carcinoma (HCC). 1 It is now quite clear that HCCs are characterized by the up-regulation or down-regulation of extremely different combinations of pathways, which eventually influence the course of the disease and the response to therapy.
Despite the vast number of molecular signatures identified so far, 2 none has entered clinical practice. Possible reasons include their complexity, 3 and the applicability to only a fraction of HCC, since the HCC source is mostly represented by surgical samples, 3 given the scarce propensity to perform biopsy in HCC. 4 In a prospective study, we recently identified a neoangiogenic signature composed of 5 genes, which accurately assess HCC prognosis at first presentation. 5 "Aggressive" HCCs defined by the presence of this signature were also shown to express unique molecular features, such as marked local up-regulation of both PD-1 and PD-L1 and concurrent FoxP3-positive lymphocytic infiltrate, a loss of E-cadherin, gain of epithelial-to-mesenchymal transition (EMT) phenotype and extreme poor differentiation at histology. 6 From a histological point of view, the aggressive HCCs have often Edmondson-Steiner (E-D) grade 2 or 3, but the relationship between E-D score and the neoangiogenic signature is partial, as patients with aggressive HCCs harboring a high signature score can have low E-D score. Despite the neoangiogenic composition of the signature, at presentation typical wash-in/wash-out radiological features do not characterize aggressive HCCs, suggesting relevant influence of hypoxia on the HCC course. Furthermore, in a small percentage of these aggressive cases, features reminiscent of intrahepatic cholangiocarcinoma (iCCA) can be highlighted.
Starting from these considerations, the aim of this study was to investigate the impact of microenvironment-related genes on the clinical outcome of patients with HCC.

| Patients
We prospectively collected a cohort of patients with Child-Pugh class A liver cirrhosis of any etiology, undergoing ultrasound (US) surveillance at 6-months interval in our Unit. Clinical features of the cohort were previously reported in detail. 5 Briefly, patients with a new CT-confirmed HCC diagnosis underwent US-guided liver biopsy both inside the lesion and in the surrounding tissue. Tumor (T) and non-tumor (NT) liver samples were collected in cold RNAlater (Qiagen) and immediately processed for gene expression analysis.
For each biopsy, a portion was also fixed in 10% formaldehyde, paraffin-embedded, and stained with H&E. The diagnosis of HCC was based on established histological criteria. 7 The Ethics Committee of Azienda Ospedaliero-Universitaria, Modena approved the study protocol (IRB10/08_CE_UniRer; ClinicalTrials ID: NCT01657695). These findings can also offer an interpretation for the somewhat disappointing results of immune-based therapeutic strategies and lead the way to new pre-clinical studies on mechanisms really involved in the progression of HCC.

| Discriminatory gene analysis (DGA)
To determine if there were genes differentially expressed across all tumor samples considered, we performed a discriminatory gene analysis (DGA), as previously described. 5 Briefly, each T sample was compared individually with the combined group of NT samples. We identified 243 genes able to discriminate two types of tumor with different aggressiveness, herein indicated as 'fast-growing' and 'slow-growing' tumor samples. Details of the statistical analysis were previously described in reference (5), and are reported in brief in Supplemental Methods. We included the most discriminatory genes, ANGPT2, ESM1, NETO2, NR4A1 and DLL4, known to be involved in angiogenesis, in a novel neoangiogenic transcriptomic signature (TS), able to accurately identify fast growing tumors with gloomy prognosis.
In the current study we have tested different existing signatures described by literature in HCC and in other solid tumors, involving immune, hypoxic, metabolic and vascular pathways, against the prospective cohort cited above. We have indeed investigated if these genes were differently expressed in our cohort of patients characterized by presence/absence of neoangiogenic TS and which genes were related with survival and outcome of patients.

| Statistical analysis
Patients in the original study 5 were censored at the time of liver transplant (LT), death, or last available follow-up. We used the nonparametric Mann-Whitney U test to evaluate differences in expression levels of the genes of the various signatures examined.
The Kaplan-Meier method was used to estimate the cumulative probability of overall survival; the factors evaluated to estimate survival were the median gene expression levels of the different signatures. Differences in observed probability were assessed using the log-rank test.

| RE SULTS
A total of 78 patients were enrolled in the Microarray study. Clinical characteristics are detailed in Table S1. Data were censored in December 2012 (mean follow-up 28.1 ± 13.2 months).

| Expression profile of HCC tissues
Expression data of T and NT tissue samples for genes relevant to immunity, extracellular matrix (ECM), profibrotic growth factors, proliferation and hypoxia, were analyzed according to different histopathologic (severity of fibrosis, intensity of inflammation, grading according to Edmondson-Steiner score) and molecular (presence of the neoangiogenic transcriptomic signature) features.

| Severity of fibrosis and inflammation
Several genes were altered in both T and NT tissues (Table 1). Among them, KRT19 was the only down regulated gene, while all the others were up regulated.
When analyzing gene expression in relation with intensity of inflammation, only few genes were found altered, all but Notch3 linked with collagen and ECM expression (Table 1).

| Edmondson-Steiner score (E-S score)
Few genes (JAG1, TGFb2) were found altered in tumor tissue when analyzed in relation with HCC grading according to E-S score. None of the genes were found significantly altered in non-tumor tissue ( Table 1).

| Neoangiogenic transcriptomic signature
When analyzing data according to transcriptomic signature (TS), only COL4A2 gene was found hyper-expressed in tumor tissue of aggressive HCC while COL22A1 was down regulated in non-tumor tissue.
Interestingly, BGN gene was down regulated in both tumor and nontumor tissue (Table 1).

| Relationship between transcriptomic signature of aggressiveness and hyper-expressed genes in different prognostic signatures
Genes included in several different well-characterized signatures, related to functions potentially relevant in carcinogenesis (ie immune control, 23 hypoxia, 24 proliferation/fibrosis 25 ), were analyzed with respect to HCC aggressiveness, defined by the presence/absence of the neoangiogenic TS as TS-positive and TS-negative HCC. 5
Ten out of 26 genes from the hypoxic signature described by Eustace et al 15 were either up regulated (7 genes) or down regulated (3 genes) in aggressive HCC (Table 2). They were mostly involved in metabolism/glycolysis, and in vascular and ECM remodeling. Chang et al 16 described two different hypoxic signatures, common to several solid cancers, one associated with poor prognosis and one with good prognosis. In the 'good prognosis' signature, one gene (P4HTM) was up regulated in neoangiogenic TS-positive HCC, while TA B L E 1 List of differentially expressed genes according to higher intensity of fibrosis, higher severity of inflammation, higher Edmondson-Steiner score or aggressive neoangiogenic signature. Gene expression was analyzed by Mann-Whitney test. Red: up regulated genes; Blue: down regulated genes (fold change; P value) two (KDM6B, ALKBH4) were down regulated. In the 'poor prognosis' signature, only PLOD2 was up regulated in T (P = .034) and had borderline significance in NT (P = .063) tissue.

| Metabolic signatures
Zhu et al 17 described a metabolic ten-gene signature in HCC with excellent ability for predicting survival prognosis. Among these genes, 7 were up regulated in aggressive HCC (Table 2). Among  Table 2).
In the 4-genes signature described by Liu et al, 20 only UCK2 (P < .000) and GOT2 (P = .023) were differentially expressed in aggressive HCC, while significance of ACAT1 was borderline (P = .064). Liu et al 26 also evaluated a series of metabolic genes derived from the Cancer Genome Atlas-Liver Hepatocellular Carcinoma data set (TCGA-LIHC). We evaluated the 172 most differentially expressed of this series. We identified 92 genes, out of 172 tested that were up regulated (33 genes) or down regulated (59 genes) in aggressive HCC (Table S2).  Table 2).

| Vascular invasion signature
In the 35-gene signature predictive of vascular invasion described by Minguez et al, 22 13 genes were up regulated in bland tumors and 9 in aggressive tumors ( Table 2). Two genes (GLYAT and ADH4) were significantly down regulated in aggressive tumors. TA B L E 2 Genes included in the hypoxic, immune, metabolic, and vascular invasion signatures found altered in the prospective cohort of 78 hepatocellular carcinoma (HCC) patients at first HCC diagnosis in relation with presence of the aggressive phenotype according to the neoangiogenic transcriptomic signature. 5 Different levels of gene expression were analyzed by Mann-Whitney test (P value ***P < .001, **P < .01, *P < .05, # borderline)

| Relationship between hyper-expressed genes of different prognostic signatures with survival and relationship with the neoangiogenic transcriptomic signature
Relationship between level of gene expression and survival was evaluated by Kaplan-Meier analysis.

| Immune signatures
Only few genes obtained from the different immune signatures evaluated (Jiang et al, 8 Okrah et al, 9 Sia et al, 10 Lal et al, 11 Wang et al 12 and Xu et al 13 ) were significantly related to survival ( Table 3).
None of the genes identified by Jiang et al 8

but PLCG1
(P = .005) was significantly related with survival, the relationship, TA B L E 3 Genes included in the hypoxic, immune, metabolic, and vascular invasion signatures, which were found altered when tested in the gene expression database of the prospective cohort of 78 hepatocellular carcinoma (HCC) patients at first HCC diagnosis described in reference (5). The relationship with worst survival of up regulated or down regulated genes was evaluated by Kaplan-Meier analysis (Level of significance: ***P < .001, **P < .01, *P < .05, # borderline significance) after stratification according to TS, being with aggressive tumors only.
In the T cell signatures described by Okrah et al, 9 3 genes were up regulated; for all of them, increased expression was related with reduced survival. Only ADAMTS2 was down regulated and significantly associated with lower survival (Table 3). Evaluating their expression in HCC according to neoangiogenic TS, only JAG1 was up-regulated in aggressive HCCs (P = .042; Table 4). The opposite was found for IGFBP5 (bland: P = .010; aggressive: NS; Table 4). Of note, 3 genes, whose expression was not different in the unstratified cohort, were up regulated specifically in aggressive HCCs (KRT19, P = .032; EMILIN1 (P = .038), GLI2 (P = .032; Table 4).  (Table 4).
Among the 8 genes identified by Xu et al, 13 only CKLF_C (P = .031) was significantly related with survival, the relationship holding true only for aggressive HCC after stratification according to TS (Tables 3   and 4). Wang et al 12 developed a 9-gene prognostic model for survival.
Among them, only 2 genes were up-regulated (ANGPT1 P = .053 and NDRG1 P < .0001) and one was down-regulated (OSGIN1, P = .040; Table 3). When evaluating them in accordance to the neoangiogenic TS, NDRG1 was significant in bland HCC (P = .002) and borderline in TA B L E 4 Genes included in the hypoxic, immune, metabolic, and vascular invasion signatures, which were found altered when tested in the gene expression database of the prospective cohort of 78 hepatocellular carcinoma (HCC) patients at first HCC diagnosis described in reference (5). The relationship with worst survival of up-or down regulated genes according to the aggressive phenotype, defined by the neoangiogenic transcriptomic signature, was evaluated by Kaplan-Meier analysis (Level of significance: ***P < .001, **P < .01, *P < .05, # borderline significance) aggressive HCC (P = .063; Table 4). OSGIN1 was significantly down regulated in bland HCCs only.

| Hypoxic signatures
Expression in the tumor tissue of seven genes present in the signature by Eustace et al 15 was significantly related with decreased survival (six of them being up regulated, one down regulated; Table 3).
Among 13 genes in the signature by Hu et al, 14 only 5 were related to survival (Table 3). Only one, SLC1A7 (P = .036), was found significant down regulated in the neoangiogenic TS− (Table 4).

Of the genes reported by Chang et al, 16 down regulated KDM8
and ALKBH7 and up regulated P4HTM (indicated in the Chang's study as being related with better prognosis) were instead related with decreased survival as well as up regulated PLOD2 when tested in our series (Table 3). Following stratification by neoangiogenic TS, we found altered P4HTM in TS− but not in TS+ HCC, while PLOD2 was found up regulated in the TS+ HCC only (Table 4). 26 KDM8 was down regulated in TS− only and ALKBH7 down regulated in TS+ only (Table 4).
Four genes from the 9-gene Zhang's signature (3 up regulated, one down regulated) 18 were significantly related with survival (Table 3).
Among those not significantly related with survival in the unstratified cohort, AGRN and RBCK1 were up regulated in the 'aggressive' tumors with significantly lower survival (P = .015; Table 4).   (Tables 3 and 4).

| Cox regression analysis
The different signatures (neoangiogenic, immune, hypoxic, metabolic, vascular invasion) reported in HCC at univariate analysis were then tested to identify in each category, which was more powerful in predicting survival (Table 5). Signatures were tested in different combinations at multivariate analysis ( Table 5). The two best fitting models are reported in Table 5. In one, the neoangiogenic transcriptomic signature, the vascular signature described by Yi et al 21 and independently related with survival. In none of the models, the immune signature was independently related with survival. Several signatures recently described in HCC patients, focused on gene expressions related to immune system 9,10 for its potential druggability in HCC derived from the availability of new immunotherapeutic strategies. 27 Alterations of the immune system (ie number of immune cells or cytokine levels) were shown to contribute to progression of HCC by regulating tumor tolerance and tumor surveillance. 28 In these studies, the immunophenotype of HCC was related to patient survival, suggesting that a strong immune activation within the tumor could be able to hamper HCC progression with a beneficial effect on survival. 29 Starting from these observations, HCC features, namely vascular invasion, early recurrence, intrahepatic metastasis, and significantly shorter overall survival. 33 The other gene found overexpressed in aggressive TS+ HCC was KRT19

| D ISCUSS I ON
(also known as Cyfra21-1). High expression levels of KRT19 were related to high levels of ERK activation and to a gloomy prognosis. 34 Several other findings indicate a direct promoting effect of KRT19 on cancer cell survival, invasion, and angiogenesis. 35 Not surprisingly, in this same cohort, we previously reported that significantly higher levels of circulating Cyfra21-1 were a feature of aggressive HCCs. 6 Overall, these genes and pathways found overexpressed in HCC, despite being involved in some of the immune-related signatures described in HCC, are more representative of other signaling pathways regulating mechanisms underpinning fibrogenesis, angiogenesis and metastasis.
We found other immune-related genes deregulated in TSnegative HCC only. ADAMTS2 down regulation in TS-negative HCC is consistent with previous reports demonstrating its inhibitory action on angiogenesis. 36  Hypoxia is associated with activation of HIF-1α that plays important roles in many critical aspects of HCC tumorigenesis, progression and metastasis, and is also an indicator of poor outcome. 38 The induction of angiogenesis in hypoxic conditions is relevant for tumor growth by stimulating expression of angiogenic factors. 39 In our study, several genes associated with hypoxia were found deregulated (described in reference [15,16]), especially in aggressive TS+ HCCs, and also associated with decreased survival.
The most striking feature of these tumors, however, was the abnormal metabolism. The liver plays a central role in metabolism owed to the hepatocyte capacity to maintain energy production and metabolic homeostasis. In HCC, altered metabolism also affects the tumor microenvironment in order to sustain cellular proliferation and/or escape from apoptosis, in particular lipid metabolism, with consequent protective effects on tumor growth, proliferation and survival. 40 For these reasons, metabolic changes can provide yet neglected but hopefully promising therapeutic targets in HCC treatment. 41 By the analysis of metabolic signatures, we have found a major correspondence of genes differentially expressed in our cohort of patients (Tables 3 and 4). Among up-regulated genes, SLC16A3, alias MCT4, was up regulated in both 'bland' TS-negative and 'aggressive' TS-positive HCCs. SLC16A3 plays a role in the glycolytic process and it is able to induce HIF1α expression in the microenvironment of large tumors. 42 In HCC, its expression was found at higher levels in tumor than in non-tumor tissue, and was correlated with tumor size and poor prognosis. 43 SLC22A1 is also known as OCT1, one of the organic cation transporters that were constitutively expressed in liver. It plays metabolic functions of uptake, intracellular inactivation, and biliary or urinary excretion of a broad spectrum of endogenous and exogenous compounds, including anticancer drugs. 44 It is also commonly regarded as a marker of 'cancer stemness'. In HCC and CCA, a strong downregulation of SLC22A1 mRNA expression has been described. This finding was related to advanced tumor stages, tumor progression and to a significantly reduced overall survival. 45 In our cases, SLC22A1 down-regulation was a specific feature of aggressive tumors, where it was significantly associated with worst survival.
AGRN and PGK1 were two other up-regulated genes in TS+ patients. AGRN encodes for a proteoglycan (Agrin) that represent an important component of remodeled ECM. It is important for neoangiogenesis in HCC tissues, and it is incorporated into newly formed vasculature. 46 Chakraborty et al, 47 have revealed that Agrin was an important factor activating and coordinating cellular adhesion, migration and invasiveness of HCC cancer cells. In particular, Agrin behaves as mechano-activator of yes-associated protein (YAP), and cooperation between Agrin and YAP leads to liver cancer development, HCC in particular. In HCCs, up-regulation of Agrin was related to decreased survival time and presence of tumor metastasis, likely indicating a prognostic role. 48 A similar value as potential biomarker of enhanced invasiveness was reported for PGK1, known to be overexpressed in several carcinomas, even including pancreatic and gastric carcinoma beside liver. 49

| CON CLUS IONS
Altogether, our data demonstrate that in HCC, and especially in aggressive TS-positive HCC, signaling pathways related with hypoxic and metabolic/glycolytic signatures are more relevant in determining a poorer outcome of HCC than immune-related pathways.
This has profound implications for therapeutic choice and also can offer an interpretation for the somewhat disappointing results of immune-based therapeutic intervention. 26 Indeed, therapeutic protocols involving checkpoints inhibitors have reached the endpoints only when coupled with inhibitors of angiogenesis. 50 On the other hand, the striking relevance of hypoxic and metabolic/glycolytic signatures, especially in 'aggressive' HCCs, gives an account of the grim natural history of these cancers. Unfortunately, there are no effective tools so far to efficiently counteract the activation of these signaling pathways. This means that it is still difficult to interfere with the end product of their activation, ie neoangiogenesis. As the activation of many key pathways seems to be related with processes (like fibrosis establishment and progression) that are unleashed by chronic liver injury, these observations lend support to the notion that most efforts should be put in preventing and/or curing chronic liver disease before events eventually leading to carcinogenesis are kindled.

CO N FLI C T S O F I NTE R E S T
The authors declare no conflict of interest.

AUTH O R CO NTR I B UTI O N S
Conception and design of the study, EV and LF; acquisition of data,