Multi-scale Pan-cancer Integrative Analyses Identify the STAT3-VSIR Axis as a Key Immunosuppressive Mechanism in Head and Neck Cancer

Immunotherapies targeting immune checkpoint proteins (PD-1, PD-L1, CTLA4) provide remarkably deep and durable responses in a subset of patients across many different tumor types and disease settings, but additional tumor-intrinsic as well as adaptive immunosuppressive mechanisms limit benefit for a substantial majority of patients. VISTA (HGNC symbol: VSIR) is a key immunosuppressive immune checkpoint protein that has been shown to act independently of the PD-L1/PD-1 axis (1, 2). While VSIR expression in myeloid cells and lymphocytes has been widely studied (3–5), its expression directly on tumor cells may also contribute to unique mechanisms of immune escape and tumor progression (4, 6, 7). The significance of VSIR overexpression across cancer types, the impact on the immune microenvironment and the potential signaling mechanisms underlying VSIR overexpression in tumor cells remains largely uncharacterized. We therefore initiated a pan-cancer multi-omic integrative analysis to delineate both leukocytic and non-leukocytic sources of VSIR expression, thus identifying specific cancer types that aberrantly express VSIR on tumor cells with significant impact on the tumor microenvironment and patient outcomes, and the signaling events likely contributing to tumor-cell specific VSIR expression.

Detailed methods are provided as Supplementary Materials and Methods.

Molecular profiling data on >11,000 tumor samples spanning 30 solid cancer types was obtained from the publicly available The Cancer Genome Atlas (TCGA) pan-cancer dataset. Specifically, we obtained pan-cancer genome-scale RNA sequencing (RNA-seq) as well as DNA Methylation profiles from the GDC's pan-cancer atlas repository. Gene expression levels for all unique coding genes in log₂-transformed FPKM units (fragments per kilobase of mRNA per million reads) were estimated using the RNA-seq data as previously described by us (8, 9). The DNA Methylation profiles were used to estimate leukocytic infiltration fraction on a per-sample basis as outlined by the pan-cancer atlas research consortium (10).

Tissue from 274 patients diagnosed with squamous cell carcinoma of the oral cavity at The Ohio State University Wexner Medical Center's James Cancer Hospital and Solove Research Institute were used to construct seven tissue microarrays (TMA; ref. 11). These studies were conducted in accordance with the ethical guidelines outlined in the U.S. Common Rule and were performed after approval by the Ohio State University Institutional Review Board, which waived Health Insurance Portability and Accountability Act research authorization and the need to obtain written consent from patients because this was a retrospective study performed on discarded tissue samples. All patients were treated surgically with or without adjuvant therapy (11). Paraffin-embedded archival tissue blocks and their matching hematoxylin and eosin (H&E)-stained slides were retrieved from the Department of Pathology Tissue Archives. A pathologist marked the areas with cancer and adjacent normal on the H&E slides. When available, representative regions (three cores of tumor tissue and one core of adjacent normal tissue) were sampled using a 1.0 mm punch on master TMA blocks. The TMA's were constructed by the Histology Core in the Department of Pathology at The Ohio State University (Columbus, OH). Unstained sections were cut and used for quantitative immunofluorescence (QIF) assessments using the AQUA platform to quantify tumor and stromal VSIR expression; or the InForm (VECTRA) platform to quantify CD4/CD8 T-cell infiltration per tumor sample.

The relationship between leukocytic infiltration and VSIR expression from TCGA was mathematically modeled using LOcal regrESSion (LOESS), a non-parametric and non-linear weighted regression framework that combines multiple regression models using a nearest-neighbor framework. The LOESS model–based framework identified VSIR High and Low tumors wherein the VSIR expression level was not explainable by their respective leukocytic infiltration levels. CIBERSORT (12), a method for characterizing cell composition of complex tissues from their gene expression profiles, was used to estimate relative infiltration levels of a total of 22 immune cell subsets. Since CIBERSORT estimates are relative to the total leukocytic infiltration per sample, the absolute infiltration estimates of each immune cell subset on a per-sample basis was inferred by multiplying the CIBERSORT relative infiltration estimates with the DNA methylation–based leukocytic infiltration fraction.

To characterize the activity of candidate transcription factors in ovarian serous adenocarcinoma (ovarian SA) and head and neck squamous cell carcinoma (HNSCC) tumors within TCGA pan-cancer dataset, we first employed InFlo (13), an integrative systems biology framework for characterizing activities of complex signaling networks in individual patient samples. InFlo infers deregulation of pathway subnetworks in individual biological tumor samples compared with normal samples by integrating RNA-seq profiles with detailed regulatory information derived from pathway network annotations. In addition, we employed the VIPER statistical methodology (14) to analyze RNA-seq profiles of individual tumor samples to infer the activity levels of transcription factors by integrating the expression levels of their transcriptional targets.

Distinct ovarian cancer (OVCAR3 and OVCAR8) and HNSCC cell lines (BICR18, BICR22, and CAL27) were seeded in 6-well tissue culture plates. Next day, cells were treated with small-molecule inhibitors S3I-201 (Selleckchem, catalog no. S1155) at 50 or 75 μmol/L or Tofacitinib (Selleckchem, catalog no. S2789) at 10 and 20 μmol/L for 48 hours, using DMSO (Sigma-Aldrich, catalog no. D2650) as a vehicle control. Total lysates were consolidated and protein extracted from the treated cell lines, followed by Western Blotting analysis to test for the expression of pSTAT3 705, pSTAT3 727, tSTAT3, and VSIR.

To identify cancers in which VSIR may be playing a significant immunosuppressive role, we first explored VSIR mRNA expression in 9,735 tumor samples spanning 30 cancer types using RNA-seq data obtained from the pan-cancer TCGA cohort (10). Given that VSIR has previously been shown to be expressed in immune cells (3–5), we evaluated whether the wide variation in VSIR expression in the pan-cancer cohort was explainable by the extent of leukocytic infiltration in the tumors. Accordingly, we estimated the extent of leukocytic infiltration in individual tumor samples using a DNA methylation–based quantification of leukocytic cellular fraction within each tumor sample (See Materials and Methods). Integration of the DNA methylation–based leukocytic infiltration estimates with RNA-seq profiles in the pan-cancer cohort indeed revealed higher VSIR expression in tumor samples with high leukocytic infiltration (Supplementary Fig. S1; Supplementary Table S1). However, we observed substantial variability in VSIR expression levels (>500-fold) in the context of tumors with low leukocytic infiltration (Supplementary Fig. S1; Supplementary Table S1). We therefore hypothesized that VSIR expression in a subset of tumor samples may be derived from cellular sources other than tumor-infiltrating leukocytes. Therefore, we developed an integrative non-parametric nonlinear Local regrESSion (LOESS) framework to model the relationship between non-leukocytic cellular fractions and VSIR expression (see Materials and Methods), resulting in the identification of tumor samples whose VSIR expression level was significantly (P < 10⁻⁶) higher than what might be expected based on their leukocytic infiltration levels (Fig. 1A). Accordingly, tumor samples that were identified by the LOESS model as exhibiting substantial non-leukocytic VSIR expression were classified as VSIR High and tumors whose observed VSIR expression was lower than what was expected by the leukocytic fraction were classified as VSIR Low (Fig. 1A). Furthermore, given that VSIR expression varies across immune cell subpopulations, and immune cells belonging to the myeloid lineage tend to exhibit high VSIR expression (15), we additionally performed RNA-seq–based deconvolution of immune cell infiltrations (see Materials and Methods) in the pan-cancer cohort, to test whether VSIR High tumors exhibited higher myeloid cell infiltration as compared with VSIR Low tumors. Interestingly, LOESS-derived VSIR High tumors exhibited, on average, lower myeloid cell infiltration as compared with VSIR Low tumors (Supplementary Fig. S2), thus further supporting the specificity of the LOESS-model as identifying aberrant VSIR expression in the non-leukocytic compartment of tumors.

Our pan-cancer LOESS model suggests that some tumor cells across cancer types have co-opted VSIR expression as a means of immune evasion. As such, we next evaluated whether VSIR High tumors are associated with poorer clinical outcomes across tumor types. Indeed, in the pan-cancer setting, VSIR High tumors exhibited poorer overall survival compared with their VSIR Low counterparts (P = 0.0029; Supplementary Fig. S3A). However, the observed difference in survival did not remain significant after adjusting for tumor type (Supplementary Fig. S3B and Supplementary Table S2), a finding supported by VSIR High tumors being enriched for cancer types with relatively poorer outcomes (Supplementary Table S3). Given that HNSCC was one of the cancer types enriched within the VSIR High tumors, with oral cavity squamous cell carcinoma (SCC) subgroup of HNSCC also being enriched (Supplementary Table S3), we tested whether human papillomavirus (HPV) positivity, a major contributor of SCC development, was associated with VSIR High HNSCCs or oral cavity SCCs, but found no such association (Supplementary Fig. S4). We next tested whether non-leukocytic VSIR expression results in poorer clinical outcomes within each of the cancer types that were enriched in the VSIR High tumors (Supplementary Table S3). Notably, of all the cancer types enriched in VSIR High tumors, only ovarian SA exhibited a significant association between non-leukocytic VSIR expression and poorer overall survival (Fig. 1B; Supplementary Fig. S5). We therefore performed RNA-seq–based deconvolution of immune cell infiltrations in Sarcomas, followed by assessment of whether VSIR High sarcomas exhibited higher lymphocytic infiltration (see Materials and Methods). Indeed, we found that VSIR High sarcomas also exhibited significantly higher CD8 tumor-infiltrating lymphocytes (Supplementary Fig. S6), suggesting that VSIR expression in Sarcomas may merely be a surrogate for higher lymphocytic infiltration, consistent with the improved prognosis in this disease setting (Supplementary Fig. S5).

Nonethless, given the likelihood that high lymphocytic infiltration could be a key modulator of patient outcomes, we assessed for the association of tumor-cell specific VSIR expression with outcomes within each of the cancers enriched for VSIR High tumors (Supplementary Table S3), by first segregating tumors based their lymphocytic infiltration levels. We categorized tumors within each cancer type (Supplementary Table S3) into either harboring an immunogenic profile (high lymphocytic infiltration) as compared to a non-immunogenic profile (low lymphocytic infiltration) using model-based assessments of total lymphocytic infiltration (see Materials and Methods; Supplementary Figs. S7, S8A, and S8B). Interestingly, we found patients harboring VSIR High tumors with an immunogenic profile to be associated with significantly poorer overall survival, independent of stage or age, in both ovarian SA (log-rank P = 0.034; multivariable Cox regression HR = 2.177, P = 0.022, Fig. 1C; Supplementary Fig. S8B; Supplementary Table S4) and HPV-negative oral cavity squamous cell carcinomas (log-rank P = 0.032; multivariable Cox regression, HR = 2.11, P = 0.053, Fig. 1D; Supplementary Fig. S8A; Supplementary Table S4), but not any of the other cancers enriched within VSIR High tumors (Supplementary Fig. S9), thus suggesting that oral cavity and ovarian cancer tumor cells may have co-opted VSIR expression as a means of immune evasion in the context of high lymphocytic infiltration.

Given that VSIR High tumors were associated with poorer survival in ovarian and oral cavity cancers harboring an immunogenic profile, we next explored whether the association of non-leukocytic VSIR expression and poorer clinical outcomes in ovarian SA and HPV-negative oral cavity SCC tumors could be explained by the modulation of the activity or infiltration of specific immune cell subsets using predefined gene-expression signatures and model-based assessments (See Materials and Methods). Importantly, the poorer survival outcomes in patients harboring VSIR High ovarian SA and HPV-negative oral cavity SCC tumors was not explainable by significantly higher PD-L1 expression or myeloid cell infiltration independent of overall lymphocytic infiltration levels (Fig. 2A and B; Supplementary Fig. S10).

Notably, while we observed no change in CD8 T cell–associated cytolytic activity (see Materials and Methods) between VSIR High and VSIR Low tumors, we found VSIR High ovarian and HPV-negative oral cavity tumors exhibiting significantly lower levels of CD4 follicular helper T-cell infiltration as compared with VSIR Low tumors (Fig. 2A and B). Strikingly, this reduction in CD4 follicular helper T-cell infiltration in VSIR High tumors was also evident in the subpopulation of ovarian and oral cavity cancers harboring immunogenic profiles (Supplementary Fig. S10). Furthermore, while we found no significant reduction in CD4⁺ Th1 or CD4⁺ Th17 cells in both ovarian and oral cavity cancers, we observed a significant reduction in CD4⁺ regulatory t cells (Treg) and CD4⁺ Th2 cells uniquely in oral cavity cancers (Supplementary Fig. S11). Furthermore, our model-based assessments of immune cell infiltrations also revealed significant reduction in total CD4 T-cell infiltration independent of immunogenic profiles, in oral cavity cancers (Supplementary Figs. S12–S15). Notably, we found no difference in total CD8 T-cell infiltration in oral cavity cancers (Supplementary Figs. S12–S14), consistent with the CD8 cytolytic activity signature assessments (Fig. 2B; Supplementary Fig. S10). These findings, taken together with the association of VSIR High cancers with poorer overall survival (Fig. 1), suggest that non-leukocytic VSIR expression is likely contributing to a pro-tumorigenic immune microenvironment in specific subsets of cancers.

Given the likely immunomodulatory effects of non-leukocytic VSIR expression in both ovarian and oral cavity cancers, we next assessed VSIR expression in clonal cell line models represented in the Cancer Cell Line Encyclopedia (CCLE; ref. 16), identifying subsets of HNSCC and ovarian cancer cell lines as exhibiting high expression levels of VSIR. This finding, when coupled with the recent evidence supporting the use of immunotherapy targeting the PD-L1/PD-1 axis in recurrent/metastastic oral cavity SCC (17), led us to prioritize validation of our findings in oral cavity SCC using orthogonal immunofluoresence-based measurement modalities in a large independent tissue microarray (TMA) of oral cavity tumors (Supplementary Table S5) and normal squamous tissues accrued from patients treated at Ohio State University (OSU TMA; ref. 11) and for whom we had close to 20 years of clinical follow-up (see Materials and Methods).

As such, we performed QIF measurements (18, 19), to assess for non-leukocytic VSIR protein expression and lymphocytic infiltration in oral cavity tumors and normal squamous tissue samples represented in the OSU TMA (see Materials and Methods; Supplementary Fig. S8C). Interestingly, we observed substantial concordance in VSIR protein expression across independent TMA spots for each tumor sample within the cytokeratin-negative stromal (Supplementary Fig. S16A) and cytokeratin-positive tumor compartments (Supplementary Fig. S16B), in addition to a strong correlation between VSIR expression across the tumor and stromal compartments (Supplementary Fig. S17) of the tumors in the OSU TMA. In addition, we detected significantly higher levels of total VSIR protein expression in oral cavity tumors as compared with normal tissues in the OSU cohort (P = 2.22 × 10⁻¹⁶; Fig. 3A). On the basis of this comparison, we defined a VSIR^Low category of tumor samples, corresponding to the lowest quartile tumor samples whose expression falls within one SD of the mean VSIR expression in the normal samples, with the upper three quartiles forming a VSIR^High category of tumors (Fig. 3A).

In addition, to assess for the extent of tumor-infiltrating lymphocytes (TIL) per oral cavity tumor sample in the OSU TMA, we performed VECTRA-based assessments of CD4 as well as CD8 T-cell densities (see Materials and Methods) per spot in the OSU TMA. We then defined total TILs as the sum of CD4 and CD8 T-cell densities in the stromal (cytokeratin-negative) compartment of tumors (Supplementary Fig. S18). Our assessments revealed significantly higher levels of CD4⁺ CD8 T-Cell infiltration in the stromal compartment of tumors as compared with normal samples (Fig. 3B) and enabled further categorization of the tumors into TILs-High and TILs-Low oral cavity cancers (see Materials and Methods, Supplementary Fig. S19).

We next integrated VSIR expression and lymphocytic infiltration assessments and performed correlative studies with patient outcomes, thus identifying VSIR overexpression as being significantly associated with poorer overall survival within oral cavity tumors exhibiting high CD4⁺CD8 T-cell infiltration (Fig. 3C). Despite the relatively small number of events in the TILs-High and VSIR^Low category (Fig. 3C), these findings when taken together with similar trends in TCGA Oral Cavity SCC cohort (Fig. 1D), support an immunomodulatory role for VSIR in this disease. In addition, give findings of non-leukocytic VSIR expression as being associated with significantly lower CD4 T-cell infiltration, but not CD8 T-cell infiltration in the oral cavity cancers harboring an immunogenic profile within TCGA pan-cancer dataset (Supplementary Figs. S13 and S14), we evaluated whether VSIR overexpression was associated with differences in CD4 or CD8 T-cell infiltration in the OSU oral cavity cohort. Interestingly, we observed significantly lower CD4 T-cell infiltration in VSIR overexpressing immunogenic oral cavity tumors, albeit without any changes in CD8 T-cell infiltration (Fig. 3D). Importantly, these findings, when taken together with the poorer survival outcomes observed in the oral cavity SCC cancers in both the OSU and TCGA cohorts (Figs. 1D and 3C), suggests a potential role of VSIR in suppressing CD4 T cell–mediated antitumor activity in the tumor microenvironment.

Given the likely immunomodulatory role played by VSIR in both ovarian and oral cavity cancers, we next assessed for potential upstream regulators of tumor-cell specific VSIR expression in this disease setting. We first assessed for transcription factor binding sites on the VSIR promoter using the transcription factor binding site (TFBS) database (20), identifying multiple binding sites for STAT3, SMAD3, and ZBTB7A. We then employed our InFlo systems biology framework that is designed to model activity levels of unmeasured signaling network components in individual tumor samples by integrating RNA-seq measurements with genome-scale signaling network definitions (13). Interestingly, InFlo-based assessments (see Materials and Methods) in TCGA ovarian SA and HNSCC cohorts revealed significantly higher STAT3 signaling network activity in VSIR High as compared with VSIR Low tumors for both ovarian (P < 0.05) and HNSCC (P < 0.0005) cancers (Fig. 4A; Supplementary Fig. S20), both of which were also orthogonally confirmed using VIPER-based assessments of STAT3 activity (Fig. 4A). However, neither SMAD3 nor ZBTB7A transcription factor activities, as assessed by InFlo or VIPER, were consistently associated with VSIR overexpression in TCGA cohorts (Supplementary Fig. S21 and S22). We therefore prioritized STAT3 signaling for pharmacologic studies designed to test whether this signaling axis indeed modulates VSIR expression in ovarian and oral cavity SCC cell line models (Fig. 4B). Indeed, pharmacologic inhibition of STAT3 activity using a STAT3 inhibitor (S3I201) revealed a dose-dependent reduction in VSIR protein expression in two distinct VSIR-High oral cavity SCC (BICR22 and CAL27) as well as ovarian cancer (OVCAR3 and OVCAR8) cell lines, concomitant with reduction in the phosphorylation of STAT3 at the Tyrosine 705 residue (Fig. 4C). Furthermore, treating the VSIR-High cell lines using an FDA-approved JAK-STAT inhibitor, tofacitinib, also resulted in a dose-dependent reduction in VSIR expression in both ovarian and HNSCC cell lines (Fig. 4D). Notably, STAT3 regulation of VSIR protein expression was orthogonally validated using siRNA-based knockdown of STAT3 in both HNSCC and ovarian cancer cell lines (Supplementary Fig. S23). Taken together with our findings in the primary tumor cohorts, these findings support a STAT3 signaling as a likely mechanism regulating tumor-cell specific VSIR expression in ovarian and oral cavity cancers.

Immune checkpoint inhibitors have shown dramatic clinical activity and are increasingly being employed across cancer types but a substantial proportion of patients do not benefit from this therapeutic intervention due to both tumor-intrinsic and extrinsic factors (21, 22). In addition to the known contributions of immunosuppressive tumor microenvironmental components, there is a growing recognition of the role of additional immune checkpoint proteins that exert immunosuppressive pressure independent of the PD-L1/PD-1 axis (21, 22). Here we employed multi-omic integrative analyses of pan-cancer cohorts, identifying subsets of cancers that aberrantly express the recently recognized immune checkpoint protein, VSIR, on tumor cells with significant clinical and immunomodulatory impact.

Notably, while prior studies have largely focused on the role of VSIR expression in the tumor immune microenvironment, our study revealed a subset of tumor samples spanning multiple cancer-types wherein the observed VSIR gene expression levels were not explainable by the extent of leukocytic infiltration (Fig. 1; Supplementary Table S1). In addition, our findings revealed that VSIR expression in the tumor-compartment as being associated with poorer outcomes in a subset of cancer types, such as ovarian SA and oral cavity SCCs, especially in the context of high levels of TILs (Fig. 1C and D), thus suggesting that this immune checkpoint axis is potentially clinically significant in a subset of cancer types. Indeed, our findings of non-leukocytic VSIR expression as being associated with poorer survival in immunogenic oral cavity SCCs was identified using both model-based inference of the tumor-cell specific expression of VSIR mRNA in TCGA pan-cancer dataset (Fig. 1D), as well as immunofluorescence-based assessments of VSIR protein expression in the independent OSU oral cavity cohort (Fig. 3; Supplementary Fig. S19). The association of VSIR over-expression with poorer survival, particularly in oral cavity cancers harboring an immunogenic profile is particularly significant given the recent clinical adoption of immune checkpoint blockade in this disease context (17, 23–26).

Interestingly, the association of non-leukocytic VSIR expression and poorer survival outcomes in both ovarian and oral cavity cancers harboring immunogenic profiles (Fig. 1C and D) did not appear to be trivially explained by increased myeloid cell or FoxP3⁺ CD4⁺ Treg infiltration (Fig. 2; Supplementary Fig. S10), since these immunosuppressive subpopulations of the tumor immune microenvironment have also been known to express VSIR (3, 4, 15, 27). Furthermore, our finding that VSIR High tumors also exhibit significantly impaired CD4 follicular helper T-cell levels but not decreased CD8 T-cell cytolytic activity in both ovarian and oral cavity cancer contexts (Fig. 2; Supplementary Fig. S10) suggests a role for VSIR as a suppressor of CD4 T cell–mediated antitumor immune response within the tumor microenvironment. Our findings are also consistent with recent evidence that VSIR induces quiescence and tolerance in naïve T cells, thus acting as an early checkpoint regulator of peripheral naïve T-cell fate (28). These findings, therefore, taken together with the growing recognition of CD4⁺ helper T cells in mediating immunotherapy response (29), and a recent report highlighting the role of CD4⁺ follicular helper T cells in augmenting and maintaining the cytotoxic activity of CD8⁺ T cells (30), suggests the possibility that tumor cells overexpress VSIR as a means to enhance CD4 T-cell tolerance by orchestrating the CD4⁺ T-cell compartment to one that is less active and more quiescent. These findings, when considered in conjunction with prior reports highlighting VSIR as an immune checkpoint protein independent of the PD-L1/PD-1 axis (2) as well as the inability of PD-L1 expression alone to reliably predict response to anti-PD-1 immunotherapy (17, 31), suggest the need for future studies to explore whether VSIR overexpession may be a tumor-intrinsic resistance mechanism to immune checkpoint blockade, specifically in the context of head and neck cancer.

There has been a growing effort over the last few years to develop and clinically evaluate VSIR antagonists (32, 33) to augment the efficacy of current immune checkpoint blockade strategies in cancer. Our identification of STAT3 signaling as an upstream regulator of VSIR expression in ovarian as well as head and neck cancers (Fig. 4; Supplementary Figs. S20 and S23) has particular translational significance in this context, because a substantial number of STAT3 antagonists are already being evaluated in the context of cancer immunotherapy (34). However, we note here that while we found STAT3 signaling to be significantly activated in VSIR High tumor subsets within sarcomas and kidney renal clear cell carcinomas, in addition to ovarian and oral cavity cancers, this association was not observable in glioblastomas, low grade gliomas, or thyroid cancers (Supplementary Fig. S24), thus suggesting the possibility of additional cis- and/or trans-regulatory mechanisms that might regulate VSIR expression in cancer cells (6). Therefore, additional in vivo studies would be required to explore the potential utility of STAT3 inhibition in modulating response to immune checkpoint blockade in the context of VSIR overexpressing tumors.

There are some limitations to our study despite using orthogonal measurement technologies in two independent cohorts of oral cavity SCCs. Our findings of VSIR overexpression being associated with poorer overall survival in the context of higher lymphocytic infiltration needs to be validated in a larger primary tumor cohort of oral cavity cancers. Of note, it is intriguing that in the context of ovarian SA, the impact of tumor-cell specific VSIR overexpression on patient outcomes does not become apparent until after about 30 months from the time of diagnosis, thus deserving further validation in an expanded cohort. Furthermore, our observation of VSIR overexpression being associated with lower CD4 T-cell infiltration is correlative, and our study did not specifically interrogate the role of VSIR overexpression using in vivo model systems or in the context of immune checkpoint blockade. As such, further studies are warranted to tease out the specific functional consequences of VSIR overexpression in tumor cells using preclinical model systems as well as in patients being treated with PD1/PD-L1 checkpoint inhibitors to decipher whether VSIR is an independent immunosuppressive axis in oral cavity SCCs.

Taken together, our findings suggest an intriguing possibility that the STAT3-VSIR axis is an immunomodulatory mechanism of clinical importance in a subset of oral cavity and ovarian cancers, thus warranting further studies to explore whether targeting this axis could potentially synergize with approved anti-PD-L1 immunotherapies.

V. Varadan reports grants from Curis, Inc. during the conduct of the study and other support from Merck Research Laboratories outside the submitted work. No disclosures were reported by the other authors.

R. Katabathula: Formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. P. Joseph: Formal analysis, investigation, methodology, writing–original draft, writing–review and editing. S. Singh: Formal analysis, investigation, visualization, writing–review and editing. S. Zhao: Formal analysis, investigation, methodology, writing–review and editing. B. Kumar: Data curation, formal analysis, investigation, writing–review and editing. P. Gaule: Formal analysis, investigation, methodology, writing–review and editing. Q. Pan: Investigation, methodology, writing–review and editing. M. Old: Resources, formal analysis, investigation, writing–review and editing. D.P. Tuck: Conceptualization, supervision, writing–original draft, writing–review and editing. V. Varadan: Conceptualization, resources, formal analysis, supervision, funding acquisition, investigation, writing–original draft, writing–review and editing.

This research was supported by PHS awards: K25 DK115904 (V. Varadan), P30 CA043703 (V. Varadan), P20 CA233216 (V. Varadan), 2P50 CA150964 (V. Varadan), 2U54 CA163060-06 (V. Varadan).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.