PVRIG Expression Is an Independent Prognostic Factor and a New Potential Target for Immunotherapy in Hepatocellular Carcinoma

Simple Summary Hepatocellular carcinoma (HCC) represents the third cause of cancer-related death in the world, and identification of new prognostic factors and/or therapeutic targets remains a major issue. Recently, immune checkpoint inhibitors (ICIs) have opened promising therapeutic options for this disease. Using a large series of transcriptomic data of clinically annotated HCC samples, we provide evidence that the TIGIT/DNAM-1 axis, and notably the PVRIG molecule, might be an interesting therapeutic candidate for HCC tumors. Abstract Hepatocellular carcinoma (HCC) is a frequent and deadly cancer in need of new treatments. Immunotherapy has shown promising results in several solid tumors. The TIGIT/DNAM-1 axis gathers targets for new immune checkpoint inhibitors (ICIs). Here, we aimed at highlighting the potential of this axis as a new therapeutic option for HCC. For this, we built a large transcriptomic database of 683 HCC samples, clinically annotated, and 319 normal liver tissues. We interrogated this database for the transcriptomic expression of each member of the TIGIT/DNAM-1 axis and tested their prognostic value for survival. We then focused on the most discriminant one for these criteria, i.e., PVRIG, and analyzed the clinical characteristics, the disease-free and overall survivals, and biological pathways associated with PVRIG High tumors. Among all members of the TIGIT/DNAM-1 axis, PVRIG expression was higher in tumors than in normal liver, was heterogeneous across tumors, and was the only member with independent prognostic value for better survival. PVRIG High tumors were characterized by a higher lymphocytic infiltrate and enriched for signatures associated with tertiary lymphoid structures and better anti-tumor immune response. These results suggest that patients with PVRIG High tumors might be good candidates for immune therapy involving ICIs, notably ICIs targeting the TIGIT/DNAM-1 axis. Further functional and clinical validation is urgently required.


Introduction
Hepatocellular carcinoma (HCC) represents the third cause of cancer-related death in the world [1][2][3]. Its incidence is increasing [4,5], with one million of new cases expected worldwide in the upcoming years. HCC represents a heterogeneous disease with multiple therapeutic opportunity for the treatment of several tumor types, potentially including HCC [24].
Based on these observations, we analyzed the transcriptomic expression of the members of the TIGIT/DNAM-1 axis to search for correlations between mRNA expression and clinicopathological data, including survival. The overall objective was to assess their relevance as potential targets for new ICIs in HCC.

Gene Expression Datasets
We gathered clinicopathological and gene expression data of HCC clinical samples from public repositories: Gene Expression Omnibus (GSE14520 and GSE54236) and TCGA database (Table S1). We collected 683 HCC samples and 319 healthy liver tissues (1002 samples in total) that we pooled in our database. Patients were from both Western and Eastern countries. The analysis's workflow is detailed in Figure 1. Among the 683 HCC samples, the follow-up and survival data were available in two datasets (GSE14520 and TCGA), representing 602 HCC samples ( Figure 1).
PVRIG and TIGIT enhanced trastuzumab-mediated NK cell response against breast cancers [26]. There is currently one anti-PVRIG antibody, which is currently tested in an ongoing clinical trial, which is either administered alone or in combination with nivolumab or anti-TIGIT ICI in several solid tumors (NCT03667716). Altogether, these clinical trials suggest that the TIGIT/DNAM-1 axis is indeed a target for the upcoming generation of ICIs, and a new therapeutic opportunity for the treatment of several tumor types, potentially including HCC [24].
Based on these observations, we analyzed the transcriptomic expression of the members of the TIGIT/DNAM-1 axis to search for correlations between mRNA expression and clinicopathological data, including survival. The overall objective was to assess their relevance as potential targets for new ICIs in HCC.

Gene Expression Datasets
We gathered clinicopathological and gene expression data of HCC clinical samples from public repositories: Gene Expression Omnibus (GSE14520 and GSE54236) and TCGA database (Table S1). We collected 683 HCC samples and 319 healthy liver tissues (1002 samples in total) that we pooled in our database. Patients were from both Western and Eastern countries. The analysis's workflow is detailed in Figure 1. Among the 683 HCC samples, the follow-up and survival data were available in two datasets (GSE14520 and TCGA), representing 602 HCC samples ( Figure 1).

Gene Expression Data Processing
Data analysis required pre-analytic processing, as previously described [27]. Briefly, each DNA-microarray dataset was normalized separately, by using quantile normalization for the available processed Agilent data and Robust Multichip Average (RMA) with the non-parametric quantile algorithm for the raw Affymetrix dataset. Normalization was

Gene Expression Data Processing
Data analysis required pre-analytic processing, as previously described [27]. Briefly, each DNA-microarray dataset was normalized separately, by using quantile normalization for the available processed Agilent data and Robust Multichip Average (RMA) with the non-parametric quantile algorithm for the raw Affymetrix dataset. Normalization was performed in R using Bioconductor and associated packages (version 3.5.1, R Foundation for Statistical Computing, Vienna, Austria). We log 2 -transformed the available TCGA RNA-seq data that were already normalized. Then, we mapped hybridization probes across the different technological platforms. We used EntrezGene (Homo sapiens gene information db, release date: 9 December 2008, ftp://ftp.ncbi.nlm.nih.gov/gene/) to retrieve and update the non-Affymetrix gene chips annotations, and NetAffx Annotation files (www.affymetrix.com; release date: 1 December 2008) for Affymetrix annotations. The probes were then mapped according to their EntrezGeneID. When multiple probes represented the same GeneID, we retained the one with the highest variance in a particular dataset. Next, we corrected the three studies for batch effects using z-score normalization. Briefly, for each gene expression value in each study separately, all values were transformed by subtracting the median of the gene in that dataset divided by its standard deviation, median, and standard deviation being measured on HCC samples only. The comparison of gene expression levels (continuous values) in HCC versus normal tissues (NT) was performed using the Student t-test. PVRIG expression in tumors was measured as a discrete value (High versus Low) by using the median expression level across the whole series as cut-off.
To explore the biological pathways associated with PVRIG expression in HCC, we performed a supervised analysis with the GSE14520 dataset [39] as learning set, which included 121 PVRIG High and 121 PVRIG Low samples. We used the TCGA dataset [40] as an independent validation set, which included 180 PVRIG High and 180 PVRIG Low samples. In the learning set, we compared the whole-transcriptome expression profiles between PVRIG High and PVRIG Low samples using a Student t-test and applied a false discovery rate (FDR) correction. Significant genes were defined by the following thresholds: adjusted p-value < 0.05 and fold change (FC) superior to |1.5x|. Ontology analysis of the resulting gene list was established with the GO biological processes of the Bioconductor package clusterProfiler. We tested the robustness of the resulting gene list in the validation set by computing the metagene score for PVRIG High samples and PVRIG Low samples, respectively.

Statistical Analysis
Continuous variables were displayed as median (range) values and categorical variables with counts (n) and proportion (%). Correlations between the PVRIG expression-based tumor groups (PVRIG High and Low) and clinicopathological and molecular variables were analyzed using the Student t-test or the Fisher's exact test when appropriate. The disease-free survival (DFS) was calculated from the date of diagnosis until the date of relapse (local, regional, or distant) or death from any cause. OS was calculated from the date of diagnosis to the date of death from any cause. Follow-up was measured from the date of diagnosis to the date of last news for event-free patients. We compared DFS and OS of the tumor groups by using Kaplan-Meier curves and log-rank tests. Uni-and multivariate prognostic analyses for DFS and OS were based on the Cox regression analysis (Wald test) and searched for associations between the gene expression level (continuous value) of members of the TIGIT/DNAM-1 and PD-1/PD-L1 axes and DFS and OS, then between the PVRIG expression-based tumor groups and the clinicopathological variables and DFS and OS. To measure the association between the PVRIG expression-based tumor groups and immune-and fibroblast-related classifiers/scores, we used the glm R function based on a binomial model with logit link. The multivariate analyses were run using the variables significant in univariate analysis (p < 0.05). All statistical tests were two-sided at the 5% level of significance. Statistical analysis was performed using the survival package
As a pre-screen strategy to identify which of these members was important to predict HCC outcome, we searched for a correlation between the gene expression of each member of the TIGIT/DNAM-1 axis and DFS and OS data. Gene expression was analyzed as a continuous variable. As shown in Figure 3, PVRIG and NECTIN-1 were the only members of the TIGIT/DNAM-1 axis that were significantly associated with OS in uni-and multivariate analyses (in univariate analysis: p = 3.2 × 10 −6 and p = 1.8 × 10 −2 , respectively; in multivariate analysis: p = 4.6 × 10 −4 and p = 5.5 × 10 −3 , respectively; Wald test p = 2.7 × 10 −5 ). PVRIG being also significantly associated with DFS, contrary to NECTIN-1 (Table S2), we decided to further focus our analysis on PVRIG. As a pre-screen strategy to identify which of these members was important to predict HCC outcome, we searched for a correlation between the gene expression of each member of the TIGIT/DNAM-1 axis and DFS and OS data. Gene expression was analyzed as a continuous variable. As shown in Figure 3, PVRIG and NECTIN-1 were the only members of the TIGIT/DNAM-1 axis that were significantly associated with OS in uni-and multivariate analyses (in univariate analysis: p = 3.2 × 10 −6 and p = 1.8 × 10 −2 , respectively; in multivariate analysis: p = 4.6 × 10 −4 and p = 5.5 × 10 −3 , respectively; Wald test p = 2.7 × 10 −5 ). PVRIG being also significantly associated with DFS, contrary to NECTIN-1 (Table S2), we decided to further focus our analysis on PVRIG.

PVRIG Gene Expression and Correlation with Clinico-Pathological Features and Molecular Subtypes Classification
PVRIG gene expression was heterogeneous across samples, with a range of intensities covering six intervals in log2 scale ( Figure S2). We thus divided the population into two groups, based on PVRIG's median expression level in the whole series, respectively, defining a PVRIG Low group and a PVRIG High group. We first searched for correlations between the clinicopathological variables and molecular subtypes classification and the

PVRIG Gene Expression and Correlation with Clinico-Pathological Features and Molecular Subtypes Classification
PVRIG gene expression was heterogeneous across samples, with a range of intensities covering six intervals in log 2 scale ( Figure S2). We thus divided the population into two groups, based on PVRIG's median expression level in the whole series, respectively, defining a PVRIG Low group and a PVRIG High group. We first searched for correlations between the clinicopathological variables and molecular subtypes classification and the PVRIG Low versus High gene expression status. As shown in Table 2, no difference was found with patients' age, TNM stage, pathological grade, AFP level, HBV and HCV infection status, alcohol consumption, and hemochromatosis. By contrast, differences existed with a higher proportion of females (p = 2.9 × 10 −2 ), with more cases positive for NAFLD (p = 3.6 × 10 −2 ) and more Hoshida's S1 subtypes (p = 1.9 × 10 −7 ) in PVRIG High tumors (Table 2). Overall, there was no major clinical feature associated with PVRIG High tumors, except that these tumors were significantly enriched for the Hoshida's S1 subtype, which is characterized by a fibrotic and immune-rich environment.

PVRIG Expression and Correlation with Biological Processes
To investigate the biological functions associated with PVRIG in HCC pathogenesis, we applied a supervised analysis to search for the genes differentially expressed between the PVRIG Low and PVRIG High groups. We first worked with a learning set (GSE14520, n = 242 divided in PVRIG High (n = 121) and Low (n = 121) groups). The supervised analysis identified 181 differentially expressed genes, including 15 genes downregulated and 166 genes upregulated in the PVRIG High group ( Figure 5A and Table S3). The robustness of this gene signature was confirmed in the independent validation set (TCGA dataset), including a total of 360 HCC samples, by comparing the metagene score of PVRIG Low samples and PVRIG High samples using a Student t-test. As expected, the metagene's difference was not only significant in the learning set but, more importantly, also in the 360 samples from the independent validation set ( Figure 5B), showing its robustness.

PVRIG Gene Expression Status and Correlation with Immune and Stromal Features
To explore further the quality of the immune response observed in PVRIG High tumors, we determined if there was a difference in the quantity of immune or stromal cells infiltrating PVRIG Low and PVRIG High tumors. For this we used the TILs score tool, which provides not only scores for tumor cells' purity but also for stromal, lymphoid, and myeloid cells' infiltration within the analyzed tumor. PVRIG High samples showed the highest score for lymphoid cells, followed by myeloid cells and stromal infiltrates (respectively, p = 1.1 × 10 −31 , p = 1.0 × 10 −19 , and p = 1.5 × 10 −6 ). No difference was observed between PVRIG Low and PVRIG High tumors regarding tumor cells. This identified PVRIG High tumors as being more infiltrated by immune cells, notably lymphocytes, than PVRIG Low tumors ( Figure 6A, and Table S5). Cancers 2023, 15, x FOR PEER REVIEW 13 of 23 The ontology analysis of the 15 genes upregulated in PVRIG Low tumors highlighted oxidation-reduction processes (p = 1.8 × 10 −2 ), a bile acid biosynthetic process (p = 3.6 × 10 −2 ), and a tricarboxylic acid cycle (p = 4.9 × 10 −2 ). This number of genes was however very small, and these data should be interpreted with caution. Inversely, the functional enrichment analysis of the 166 genes upregulated in PVRIG High tumors revealed pathways related to an immune response (p = 1.6 × 10 −35 ), a T cell co-stimulation (p = 1.1 × 10 −13 ), an We next looked at the detailed composition and functional orientation of these immune infiltrates. As expected from the ontology, the analysis of the 24 Bindea's immune cell types, defined as the immunome [30], revealed that PVRIG High tumors differed from PVRIG Low tumors by a higher infiltration with immune cell populations (p < 0.05).

PVRIG Gene Expression Status and Correlation with Immune and Stromal Features
To explore further the quality of the immune response observed in PVRIG High tumors, we determined if there was a difference in the quantity of immune or stromal cells infiltrating PVRIG Low and PVRIG High tumors. For this we used the TILs score tool, which provides not only scores for tumor cells' purity but also for stromal, lymphoid, and myeloid cells' infiltration within the analyzed tumor. PVRIG High samples showed the highest score for lymphoid cells, followed by myeloid cells and stromal infiltrates (respectively, p = 1.1 × 10 −31 , p = 1.0 × 10 −19 , and p = 1.5 × 10 −6 ). No difference was observed between PVRIG Low and PVRIG High tumors regarding tumor cells. This identified PVRIG High tumors as being more infiltrated by immune cells, notably lymphocytes, than PVRIG Low tumors ( Figure 6A, and Table S5).  We next looked at the detailed composition and functional orientation of these immune infiltrates. As expected from the ontology, the analysis of the 24 Bindea's immune cell types, defined as the immunome [30], revealed that PVRIG High tumors differed from PVRIG Low tumors by a higher infiltration with immune cell populations (p < 0.05). The top fifteen subsets infiltrating PVRIG High tumors were the lymphocytes (T cells: p = 6.  Finally, we looked at the composition of stromal cells, and more especially of CAF. We used the Givel et al., Kieffer et al., and Biffi et al. signatures to search for enrichment in specific CAF subsets. PVRIG High tumors were enriched with both CAF-S1 and CAF-S4 subsets (p = 1.5 × 10 −12 , p = 1.5 × 10 −9 , respectively) [47]. Similarly, both inflammatory CAF (iCAF) and myofibroblastic CAF (myCAF) subsets were detected in PVRIG High tumors. The iCAF subset was slightly predominant, especially the IFNy-iCAF (p = 6.3 × 10 −21 ), the IL-iCAF (p = 6.3 × 10 −15 ), and the detox-iCAF (p = 5.8 × 10 −11 ) ( Figure 6D and Table S5). These subsets secrete immunomodulatory molecules that can elicit or maintain an anti-tumor immune response. The pro-tumoral myCAF subsets (wound-, ecm-, and TGFb-myCAFs) were also well represented (wound-myCAF: p = 1.4 × 10 −9, ecm-myCAF: p = 1.7 × 10 −8 , TGFβ-myCAF: p = 3 × 10 −8 ). These subsets are pro-tumoral cells ( Figure 6D).
Altogether, these results suggest that cytotoxic immune cells are infiltrating PVRIG High tumors, concomitant with a mixed anti-tumoral iCAF-like and pro-tumoral myCAFlike stroma.

Discussion
Based on the gene expression analysis of the TIGIT/DNAM-1 axis in a large cohort of HCC clinical samples, we showed that PVRIG is an independent biomarker predictive for favorable clinical outcome. PVRIG expression was lower in HCC samples than in normal hepatic tissues. Its expression was heterogeneous across tumors and allowed us to search for correlations between PVRIG expression levels and clinicopathological variables. Using PVRIG median as a cutoff enabled us to interrogate our database, avoiding time-consuming immunohistochemistry development and analyses, while providing opportunities to work on a relatively large series of samples. We were then able to search for association with the expression of other genes and multigene signatures to investigate pathways associated with PVRIG gene expression levels more in depth.
To our knowledge, this is the third study highlighting PVRIG expression as a new potential target for immune therapy in HCC. PVRIG High expression has indeed been associated with improved DFS and OS in HCC patients [48], and an eight-gene prognostic signature was constructed to anticipate the HCC outcome, in which PVRIG was the most important variable [49].
PVRIG is an immune checkpoint receptor with inhibitory function for cytotoxic CD8 + T cell and NK cells [50,51]. In murine models with solid tumors, PVRIG interaction with its ligand NECTIN-2 on tumor or dendritic cells surface suppresses cytotoxic signals, cytokines production, and triggers exhaustion [51,52].
To our knowledge, the correlation between PVRIG expression and immune variables has never been explored further, and this is the first report characterizing the tumor immune microenvironment and biological pathways in PVRIG High tumors. The HBV and HCV infection status being equally distributed in PVRIG High or Low tumors, the background immune response due to the viral infection can be considered as equivalent. Compared to PVRIG Low tumors, several immune variables were enriched in PVRIG High tumors and suggested: (i) higher infiltration of immune cells; (ii) higher infiltration of both adaptive and innate immune cells, notably B cells, γδ T cells, T cells with a Th 1 profile, cytotoxic CD8 + T cells, and activated NK CD56dim cells; and (iii) higher cytotoxic activity. Altogether, these results suggested that PVRIG High tumors have a higher potential for anti-tumor immune response than PVRIG Low tumors. Nonetheless, despite this apparent anti-tumoral environment, the local immune cells did not prevent HCC occurrence and progression. The high expression of NECTIN-2, the ligand of PVRIG, which has also been reported by others in HCC tumors [53], might bridle anti-tumor response efficiency. Indeed, we did not find major evidence for exhaustion, as often observed in highly immunogenic tumors [27,54,55]. The higher infiltration with both immunosuppressive and pro-tumoral CAF subsets in PVRIG High tumors was also in line with the hypothesis of a poor anti-tumoral response.
An apparent contradiction in our results lies in the association of PVRIG High expression, which can trigger a strong immunosuppressive signal, with better disease-free and overall survivals. We and others have previously reported similar results with expression of PD-L1 [55][56][57] and IDO1 inhibitory molecules [58]. In these cases, our explanation is that a very strong anti-tumoral immune response, associated with better clinical outcome, also activates an inhibitory feedback loop, in the prospect of a return to homeostasis with minimum tissue damages. IFN-γ was in part responsible for the overexpression of PD-L1 and the initiation of the negative feedback loop in these tumors [53]. Here, we did not report major signs of exhaustion, except for IDO1, and we did not detect strong IFN-γ production either, but there was evidence for general IFN and cytotoxic immune responses. One hypothesis might be that the PVRIG/NECTIN-2 axis is a new negative feedback loop, either proper to the tumor liver microenvironment or associated with a specific context of immune response. Indeed, the TIGIT-DNAM-1 axis, with all its receptors and ligands expressed in normal tissues, is much more complex than the PD-1/PD-L1 axis.
The role of PVRIG as a potential regulator of HCC outcome however still remains to be proven in functional and animal model studies. Whereupon this validation, ICIs targeting PVRIG-NECTIN-2 interactions might be envisaged in HCC. For now, ICIs targeting the CTLA-4 and/or PD-1/PD-L1 axes have shown promising results in HCC [59,60]. Our observations suggest that the TIGIT/DNAM-1 axis, and notably the PVRIG/NECTIN-2 interaction, might be another interesting target for ICIs. This has never been considered for HCC yet. However, clinical trials evaluating the safety, tolerability, and clinical activity of COM701, an anti-PVRIG antibody, in monotherapy or in combination with a PD-1 inhibitor (NCT03667716) or an anti-TIGIT antibody (NCT04570839, BMS-986207), have been initiated in subjects with advanced solid tumors. Overall, these trials suggest that blocking the PVRIG/NECTIN-2 interaction to enhance anti-tumor cytotoxic function is a new therapeutic opportunity for the treatment of solid tumors [51]. Based on our work, our hypothesis is that immune cells, including cytotoxic cells, are recruited at the tumor site in PVRIG High tumors. They receive an activation signal and become activated, but they also receive negative signals from PVRIG, which is strongly engaged with NECTIN-2, which might limit anti-tumor activation. In these tumors, precluding the PVRIG/NECTIN-2 interaction might unleash a cytotoxic anti-tumor immune response ( Figure 7) and improve the clinical outcome.  PVRIG High HCC are infiltrated with cytotoxic cells, whose activity is not fully unleashed. Our hypothesis is that anti-PVRIG treatment might alleviate this inhibitory signal (one among others) in innate and adaptive cytotoxic cells expressing PVRIG. This might favor anti-tumor activation and lead to better tumor regression and better clinical outcome.

Conclusions
We showed that PVRIG expression is heterogeneous in HCC clinical samples and higher expression is associated with better DFS and OS and with strong, although nonoptimal, local immune response. Our study displays several strengths: (i) its originality, (ii) a relatively large size of series, and (iii) the biological and clinical relevance of PVRIG expression. It also includes a few limitations: (i) its retrospective nature and associated biases, and (ii) the mRNA assessment, rather than protein, analysis on bulk tissue samples. Of course, analyses in larger series are warranted to confirm our observation, as well as functional analyses on HCC preclinical models to validate PVRIG's role in HCC progression. However, and considering the inhibitory role of PVRIG, our results highlight the potential value of anti-PVRIG immunotherapies to enhance the anti-tumoral immune local response in HCC. Our data suggest that the microenvironment of PVRIG High samples (strong local cytotoxic immune response) might be more favorable for ICIs efficiency than that of PVRIG Low samples. However, the link between the target expression level and tumor response to ICIs remains unclear in many cancers. For example, in the CheckMate- Figure 7. Hypothetical evolution of HCC after treatment with an anti-PVRIG in PVRIG High tumors: PVRIG High HCC are infiltrated with cytotoxic cells, whose activity is not fully unleashed. Our hypothesis is that anti-PVRIG treatment might alleviate this inhibitory signal (one among others) in innate and adaptive cytotoxic cells expressing PVRIG. This might favor anti-tumor activation and lead to better tumor regression and better clinical outcome.

Conclusions
We showed that PVRIG expression is heterogeneous in HCC clinical samples and higher expression is associated with better DFS and OS and with strong, although nonoptimal, local immune response. Our study displays several strengths: (i) its originality, (ii) a relatively large size of series, and (iii) the biological and clinical relevance of PVRIG expression. It also includes a few limitations: (i) its retrospective nature and associated biases, and (ii) the mRNA assessment, rather than protein, analysis on bulk tissue samples. Of course, analyses in larger series are warranted to confirm our observation, as well as functional analyses on HCC preclinical models to validate PVRIG's role in HCC progression. However, and considering the inhibitory role of PVRIG, our results highlight the potential value of anti-PVRIG immunotherapies to enhance the anti-tumoral immune local response in HCC. Our data suggest that the microenvironment of PVRIG High samples (strong local cytotoxic immune response) might be more favorable for ICIs efficiency than that of PVRIG Low samples. However, the link between the target expression level and tumor response to ICIs remains unclear in many cancers. For example, in the CheckMate-040 trial dedicated to patients with HCC [18,19], the responses to nivolumab occurred irrespectively of the PD-L1 staining status. Clearly, the testing of anti-PVRIG ICIs in HCC is warranted at both pre-clinical and clinical levels. In this setting, analysis of pre-treated tumor samples will allow to assess PVRIG expression and to test if it can predict the therapeutic response. Overall, our study also highlights the complexity behind immune checkpoints function, revealing important roles for other potential candidates, such as here, with the members of the TIGIT/DNAM-1 axis.
Supplementary Materials: The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/cancers15020447/s1, Figure S1: Survival in patients with HCC; Figure S2: Distribution of PVRIG expression in HCC samples. Table S1: Datasets included in the study; Table S2: Association between expression of members of TIGIT/DNAM-1 axis and PD-1/PD-L1 axis related RNA expression and survival in HCC; Table S3: List of the 181 genes differentially expressed between PVRIG High and Low groups in the learning set (GSE14520, n = 242); Table S4: Ontology analysis of the 181 genes differentially expressed between PVRIG High and Low groups in the learning set (GSE14520, n = 242); Table S5: Logistic regression results measuring association between PVRIG groups and functional gene sets.  Institutional Review Board Statement: The study was approved by our institutional review board.
Informed Consent Statement: Our in silico study is based on public data from published studies in which the informed patient's consent to participate and the ethics and institutional review board were already obtained by authors.
Data Availability Statement: All datasets are publicly available, and references are described in Table S1.

Conflicts of Interest:
The authors declare no conflict of interest.