ENTPD1/CD39 as a predictive marker of treatment response to gemogenovatucel-T as maintenance therapy in newly diagnosed ovarian cancer

Background Broadened use of predictive molecular and phenotypic profiling amongst oncologists has facilitated optimal integration of targeted- and immuno-therapeutics into clinical care. However, the use of predictive immunomarkers in ovarian cancer (OC) has not consistently translated into clinical benefit. Vigil (gemogenovatucel-T) is a novel plasmid engineered autologous tumor cell immunotherapy designed to knock down the tumor suppressor cytokines, TGFβ1 and TGFβ2, augment local immune function via increased GMCSF expression and enhance presentation of clonal neoantigen epitopes. Methods All patients enrolled in the VITAL trial (NCT02346747) of maintenance Vigil vs. placebo as front-line therapy with homologous recombination proficient (HRP) stage IIIB-IV newly diagnosed ovarian cancer underwent NanoString gene expression analysis. Tissue was obtained from surgically resected ovarian tumor tissue following surgical debulking. A statistical algorithm was used to analyze the NanoString gene expression data. Results Using the NanoString Statistical Algorithm (NSA), we identify high expression of ENTPD1/CD39 (which functions as the rate-limiting step in the production of the immune suppressor adenosine from ATP to ADP) as a presumptive predictor of response to Vigil versus placebo regardless of HRP status on the basis of relapse free survival (median not achieved vs 8.1 months, p = 0.00007) and overall survival (median not achieved vs 41.4 months, p = 0.013) extension. Conclusion NSA should be considered for application to investigational targeted therapies in order to identify populations most likely to benefit from treatment, in preparation for efficacy conclusive trials.


Plain Language Summary
Treatment options are limited in patients with advanced stage ovarian cancer. Treatments that stimulate the immune system to target the cancer are sometimes effective, however determining which patients will have benefit has been difficult. It is therefore important to develop new markers to predict which patients will respond to therapy. In this study, we looked at the levels of a large number of genes in tumors from patients treated with Vigil (gemogenovatucel-T), a treatment that modifies patient's own tumor cells to activate the immune system. We demonstrate that high expression of a gene named ENTPD1/CD39 predicts a positive response to Vigil therapy. This finding could help clinicians to determine which patients might benefit from Vigil treatment and therefore might guide decisions on who should receive this treatment.
V igil (gemogenovatucel-T) is a novel autologous tumor cell immunotherapy, which is constructed from harvested malignant tissue [1][2][3] . It incorporates a multigenic plasmid encoding the human immune-stimulatory GMCSF gene and a bifunctional short-hairpin RNA construct, which specifically knocks down the proprotein convertase furin and its downstream targets TGFβ1 and TGFβ2 1,3,4 . It is also designed to facilitate both cancer-associated antigen and neoantigen expression, upregulate MHC-II and enhance bone-marrow derived dendritic cell maturation, thereby augmenting the afferent immune response and generating a systemic antitumor effect. The VITAL study (NCT02346747) was a Phase IIb double-blind, placebo-controlled trial involving women 18 years and older with Stage III/IV highgrade serous, endometroid or clear cell ovarian cancer (OC) in clinical complete response (CCR) following carboplatin and paclitaxel induction chemotherapy 5,6 . Results recently published in a subset of 67 patients with BRCA1/2-wildtype (wt) OC showed improved relapse free survival (RFS; HR = 0.51, p = 0.02) and overall survival (OS; HR = 0.49, p = 0.049) compared to placebo 5 . Moreover, ad hoc analysis of a subset of 45 patients with homologous repair proficient (HRP) tumors by Myriad MyChoice CDx (Myriad Genetics, Salt Lake City, UT) also showed improvement in RFS and OS (HR = 0.39, p = 0.007 and HR = 0.34, p = 0.019, respectively) 6 . Long term follow-up confirmed a durable survival effect 7 . Three-year survival proportion from time of procurement was 83% for Vigil and 40% for placebo (p = 0.0006) 7 . A correlation of systemic immune response to Vigil clinical benefit was noted using ELISPOT assay 3,8 .
Contemporary clinical management of oncology patients is increasingly being guided by predictive molecular and phenotypic profiling in order to optimize the use of targeted-and immunotherapeutics 9 , e.g., tumor mutational burden (TMB), MMR, PD-1, and PD-L1 10 . However, the use of predictive biomarkers for immunotherapy in OC has not consistently translated into clinical benefit [11][12][13][14][15][16] despite documented responses in some patients 17 . Although genomically unstable, OC is not mutationally driven, thus the clinical efficacy of immunotherapy in this disease has been dismal (<10% which generally correlates with high TMB, a presumptive marker of neoantigen content), represented by several failed phase III clinical trials [11][12][13][14][15]18 .
Nevertheless, we have studied patient subpopulations most sensitive to Vigil therapy based on molecular profile using NanoString assessment, and demonstrated that TIS high score (tumor inflammation score) and MHC-II expression correlated with ELISPOT reactivity and clinically to OS and RFS 19 . Likewise, using NanoString technology to assess OS and RFS in patients enrolled in the VITAL study 20 , we showed marked benefit in patients with BRCA1/2-wt and HRP profiles and improved outcomes in patients whose tumors had mutant TP53 (p = 0.0013). The current study explores the relationship of mRNA expression via NanoString analysis in harvested baseline tumor to RFS and OS in Vigil treated patients from the VITAL study. ENTPD1/ CD39 demonstrated clinical significance as a presumptive predictor of Vigil response versus placebo regardless of HRP status.

Methods
Study design and Vigil construction. All patients provided written informed consent prior to study enrollment in the VITAL study. Briefly, the VITAL study (NCT02346747) was a phase 2b randomized, double-blind, placebo controlled trial involving women 18 years and older with stage III or IV high-grade serous, endometroid or clear cell ovarian cancer in clinical complete response. As specified in the approved clinical protocol (Mary Crowley IRB), patients provided consent for excess tissue to be used for additional immunotherapy research. Specimens were obtained from excess tissue harvested at the time of procurement for vaccine manufacture. Tissue is dissociated into cell suspension and cells are frozen at a concentration of 1.33 million cells/ml in freeze media (10% DMSO v/v in 1% HSA/plasma-Lyte A solution and stored long term in vapor phase nitrogen. Homologous recombination status [homologous recombination deficient (HRD) or HRP] was determined for all patients using the Myriad MyChoice CDx assay as previously described 6,7 . Patient demographics and CONSORT diagram are presented in Table 1 and Fig. 1, respectively.
Vigil plasmid construction and cGMP manufacturing have been previously described 5,6 . Following VITAL study protocol guidelines, ovarian tumor tissue was excised at the time of initial tumor cytoreduction surgery and shipped to Gradalis, Inc. (Dallas, TX) for tissue processing, transfection, and vaccine manufacture.
RNA isolation and gene expression analysis. RNA expression was determined from total RNA isolated using RNeasy Mini Kit (Qiagen, Venlo, The Netherlands). NanoString PanCancer Immuno-Oncology 360 TM CodeSet using the nCounter SPRINT platform (NanoString Technologies, Seattle, WA, USA), which includes 750 cancer expression pathway genes, was used to analyze gene expression per manufacturer protocol.
Statistics. For all 750 genes a NanoString statistical algorithm (NSA) was defined prior to gene analysis (Fig. 2) to assess the correlation of NanoString gene expression results with clinical benefit as measured by both RFS and OS advantage effect with Vigil to specific mRNA expression. First, a univariate Cox model was used with gene Z-scores as a continuous variable and run for both OS and RFS in Vigil treated patients. From this data, the two-sided p-value, HR and corresponding 95% confidence interval (CI) were extracted. Genes that were significant for both OS and RFS advantage at the 1% significance level were identified. The more stringent variable selection criterion of 1% significance level was used due to the relatively small number of OS/RFS events compared to the large number of genes assessed. Next, Cox proportional hazards model with interaction term for each gene identified in the univariate Cox model was used to identify genes that were predictive of response to Vigil by analyzing data of both Vigil and placebo patients. A Cox proportional hazards model was used to determine if the interaction term between gene and treatment group was significant. The Cox model included the treatment group, gene and treatment-by-gene interaction term. The gene was considered predictive if the interaction term was significant (p ≤ 0.05). The model was run using the gene as a continuous variable or using binary high or low gene assignment. When using binary gene assignment, the median gene value for all 91 patients was calculated for each of the 750 cancer expression pathway genes. Patients were dichotomized into high or low gene expression groups if their value was either above or below the median.
Kaplan-Meier (KM) curves were generated for genes identified as predictive for both OS and RFS. Since the identified predictive genes may not be independent, further model selection was performed using a multivariate Cox model in Vigil treated patients to further refine identification of relevant genes. We used the my.stepwise.coxph function in R (open source, R Core Team), which employs both forward selection and backward elimination methodology to further select genes that were significantly associated with the time-to-event data (OS or RFS) in Vigil treated patients 21 . The significance level for variable entry and for stay in the model was set at 0.01 and variable stay we set at 0.01 to account for potential multiplicity in the model selection process.   Biomarker vigil benefit over placebo. Previous analyses of Vigil relationship to BRCA1/2-wt, HRP and TP53 mutation (p53-mu) subpopulations revealed correlation to clinical benefit [5][6][7]19,20 . These subpopulations were explored via KM analysis to assess the effect of combination biomarkers BRCA1/2-wt, HRP, p53-mu and genes identified as significant following multivariate analysis in this study on Vigil and placebo treatment effects as measured by OS and RFS.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Results
Univariate analysis vigil patients only. First, a univariate Cox model was performed with the gene Z-score as a continuous variable to obtain the two-sided p-value, HR and 95% CI in Vigil treated patients only (n = 47). This analysis identified 13 genes that were statistically significant at the 1% significance level for both OS and RFS (Supplementary Table 1). All of these genes are associated with critical immunologic modulation function as per NanoString Pan Cancer Immuno-Oncology 360™ Code set (NanoString Technologies, Seattle, WA, USA).
Predictive genes using all patients data. While the previous analysis was able to identify genes of interest, they were not able to specify if genes were predictive. To determine genes predictive of Vigil treatment efficacy, Cox proportional hazards model with interaction term was used to analyze data from both Vigil and placebo patients (n = 91). The Cox model included the treatment group, gene and treatment-by gene interaction term. Demographics between Vigil and placebo were previously shown to not impact clinical benefit results 5,6 . Four genes were identified as predictive in both Cox models using continuous and binary data for both OS and RFS (CD79B, CCL13, ENTPD1/CD39 and MRC1). Four separate KM curves were generated for each gene in: (1) Vigil patients with gene expression < median and ≥ median; (2) placebo patients with gene expression < median and ≥ median; (3) Vigil patients with gene expression < median and placebo patients < median; and (4) Vigil patients with gene expression ≥ median and placebo patients ≥ median. KM curves for OS (Fig. 3) and RFS (Fig. 4) for placebo vs. Vigil with < or ≥ median expression from these 4 genes demonstrate benefit  Tables 2 and 3.
Multivariate Analysis Vigil Patients Only. To further select significant gene associations with OS or RFS in Vigil treated patients, the my.stepwise.coxph function in R was used as the stepwise variable selection procedure (with iterations between the 'forward' and 'backward' steps) including the 4 genes showing RFS and OS advantage to Vigil treatment over placebo. Two common strategies for adding or removing variables in a multiple regression model are backward elimination and forward selection. Backward elimination begins with all genes included in the model and eliminates variables one-by-one until the model cannot be improved per the model fitting criterion. Forward selection starts with no variables included in the model, then adds variables according to importance (e.g., based on p values) until no other significant variables are found. The significance level for variable entry in the model was set at 0.01 and for variable stay was set at 0.01 to account for potential multiplicity in the model selection process. ENTPD1/CD39 was the only gene identified through this stepwise model selection process for both OS and RFS (p value < 0.001).
Subgroup Vigil/Placebo: HRP, ENTPD1/CD39. Twenty of the 91 patients (22%) enrolled into the VITAL trial (11 Vigil, 9 placebo) had HRP molecular profile 5-7 and ENTPD1/CD39 "high" expression. Note HRP status and TP53 mutations have been identified in previous analyses as predictive of Vigil response 6,7,20 . OS advantage was demonstrated (Fig. 5A) in the Vigil treated HRP/high ENTPD1/CD39 patients relative to placebo (not achieved vs 27 months, HR = 0.23, p = 0.025). In the same subset, the median RFS for Vigil was 21.1 months and 5.6 months for placebo (HR = 0.18, p = 0.004) (Fig. 5B). Despite small sample size, these subgroup results support additional survival benefit in patients whose tumors demonstrate ENTPD1/ CD39 high expression in the HRP subgroup. In order to assess the impact of HRP and ENTPD1/CD39 on outcomes in patients treated with Vigil, multivariate analyses including HRP status and  Subgroup Vigil/Placebo: HRP, p53, ENTPD1/CD39. Evidence of survival advantage was further suggested in patients with tumors demonstrating high ENTPD1/CD39 expression and of HRP/p53-mu profile. Despite the small sample size (n = 13), a trend toward OS benefit with Vigil therapy (median not reached vs 27 months, HR = 0.34, p = 0.099) and robust RFS benefit (21.1 vs 5.6 months, HR = 0.09, p = 0.004) was suggested (Fig. 5C, D).

Discussion
Using NanoString PanCancer Immuno-Oncology 360 TM molecular profiles derived from patient tumor samples in conjunction with NSA, we identified that high expression of ENTPD1/CD39 was associated with a significant and independent improvement in OS and RFS with Vigil maintenance therapy in the VITAL study. ENTPD1/CD39 is highly expressed in OC cell-lines 22 , and functions as a master regulator to maintain the balance between proinflammatory and immunosuppressive regulatory function 23 . The latter largely due the role of ENTPD1/CD39 as the rate limiting step in the conversion of ATP to ADP in the adenosine pathway. Adenosine inhibits both T-cell and NK-cell anti-tumor function. Although adenosine can be exported from the tumor into the extracellular space by nucleoside transport proteins, it is primarily formed via the action of membrane ectoenzymes by phosphohydrolysis from dead cells 24 . In addition, ENTPD1/ CD39 is present on cancer extracellular vesicles (ECVs) 24 . ENTPD1/CD39 is ubiquitously expressed in the vasculature, B cells, NK cells, dendritic cells, monocytes, macrophages, regulatory T cells and monocyte derived suppressor cells in the TME 25,26 . CD8 + T cells demonstrate T cell exhaustion signatures with malignant upregulation of CD39 in the tumor microenvironment [27][28][29] . Moreover, T regulatory (Treg) cell upregulation of ENTPD1/CD39 within the tumor microenvironment generates immunosuppressive activity thereby facilitating malignant growth and survival 30,31 . Inhibition of ENTPD/1CD39 in murine cancer models induces anticancer activity and ENTPD1/CD39 deficient mice demonstrated a reduction in tumor growth [32][33][34][35] . Furthermore, anti-ENTPD1/ CD39 increased cytotoxicity of alloreactive primed T-cell towards fresh OvCA cells 22 .
In the current study, Vigil treated patients with baseline elevated tumor expression of ENTPD1/CD39 was associated with a significantly improved response compared to those patients with tumors with low expression and to those with high tumor expression treated with placebo. The primary VITAL study results suggest that Vigil induction of GMCSF, knock down of TGFβ1 and TGFβ2 and induced CD8 + T cell activity targeted to tumor-specific cancer neoantigens provide anticancer activity beneficially impacts OS and RFS in newly diagnosed Stage III/IV OC patients receiving Vigil as maintenance therapy. This activity appears to be correlated to high ENTPD1/CD39 expression-a presumptive predictive marker. Interestingly, in a murine model, high levels of TGFβ were associated with immunosuppressive CD39 + myeloid derived suppressor cells (MDSC) 36 . Notably, placebo treated patients from the VITAL study with high ENTPD1/CD39 expression tended to show poorer survival compared to patients with lower expression, presumably reflecting the immunosuppressive role of ENTPD1/CD39 in these patients. It is also of interest that ENTPD1/CD39 promotes tumor cell survival in hypoxic regions characterized by increased levels of ATP and   1,3 . Vigil also demonstrated in a small number of patients increase in the number of circulating CD3 + / CD8 + T cells following treatment 37 . In the VITAL trial we demonstrated RFS and OS benefit in patients with HRP molecular profile 6,7 . We also suggested that the presence of mutant  20 . Results of mRNA expression via NanoString signature also indicate enhanced OS and/or RFS endpoint benefits of Vigil maintenance in both these groups. These results support the need for further verification of ENTPD1/CD39 as a biomarker of sensitivity to Vigil treatment in OC and possibly other solid tumors with high ENTPD1/CD39 expression.
The presence of ENTPD1/CD39 in multiple cell types other than certain cancers (e.g., CD4 + / Treg, CD8 + and MDSC) supports the consideration of therapeutic assessment of combined ENTPD1/CD39 inhibition and Vigil in patients with ENTPD1/ CD39 high tumor expression. ENTPD1/CD39 monoclonal antibodies have demonstrated anticancer activity in murine models as single agents and in combination with checkpoint inhibitors and autologous EBV-specific human T cells 38 . Currently, there are a number of different CD39 targeting agents in early Phase I clinical trials under evaluation [39][40][41] .
It is also possible that in a larger patient population receiving Vigil, supportive evidence demonstrated with the other immune modulatory signals identified by NanoString analysis (i.e., CXCL13 42,43 , CD79B 44 , MRC1 45 ) will also be found to have further impact on OS and RFS. All three of these genes also perform important immunologic functions. MRC1 is expressed on tumor associated macrophages (TAMs) with M2 phenotype. Once activated MRC1 directs TAM's to M1 phenotype thereby activating the innate response 46 . Recent work has shown that high CXCL13 expression in high-grade serous ovarian cancer correlates with increased survival by maintaining CXCR5 + / CD8 + T cells with in tertiary lymphoid structures 47 . CD79b expression is limited to B cells. B cells play an important role in anti-tumor immunity through secretion of cytokines and antigen presentation 48 . Such results may further direct research towards a multiplex of biomarker sensitivity and may even direct novel combination therapeutic approaches with Vigil, including combination treatment regimens based on various molecular signal expression patterns and immune related signal pathways that are relevant to Vigil related benefit.
Molecular biomarker assessment to optimize the proportion of responsive patient populations to Vigil therapy will involve more comprehensive analyses including p53mu, BRCA1/2-wt, ENTPD1/CD39 and HRP molecular profiles. Evidence provided in this report suggests these gene expression signals can act independently in defining sensitive subpopulations and also appears to support the possibility that the combined use of predictive biomarkers can suggest if not identify additive and possibly, synergistic therapeutic activity combinations. Clearly, statistical analyses such as those applied to the VITAL study here will likely continue to help identify optimal subpopulations with potential to benefit via treatment with Vigil as well as suggest the direction for continued Vigil combination studies. Results also justify trial consideration of Vigil in other solid tumor patients with HRP profile, p53mu and those with ENTPD1/CD39 high expression by NanoString PanCancer Immuno-Oncology 360 TM CodeSet analysis.
In conclusion, we identified gene signatures indicative of response to Vigil maintenance therapy as part of frontline treatment in newly diagnosed OC patients. Interestingly, NSA identified ENTPD1/CD39, a gene signal associated with an immunosuppressive tumor microenvironment, was most highly predictive of Vigil responsiveness. Previous work has indicated TGFβ may upregulate ENTPD1/CD39 in immunosuppressive myeloid cells, such that Vigil's effect in downregulating TGFβ may counter this effect and account for its activity in patients with high tumor expression of ENTPD1/CD39. Combining previously identified biomarkers of Vigil response, such as HRP and mutant p53, with ENTPD1/CD39 expression may allow for refined identification of Vigil responsive populations-ultimately allowing a transition from predictive analysis to prescriptive analytics. Such an approach can be more broadly applied to assess for correlations between gene expression signals and survival benefits as well as widen the therapeutic index by optimizing patient selection and treatment allocation with other targeted therapies.

Data availability
Supplementary Data 1 contains source data for the main figures (Figs. [3][4][5] in this manuscript. Additional data that support the findings of this study, and that do not involve proprietary or similar commercial information, will be made available to appropriate parties following an approved data sharing request sent to Laura Nejedlik (lnejedlik@gradalisinc.com) at Gradalis, Inc. Requests that may involve a conflict of interest or a competitive risk maybe declined by Gradalis, Inc. in its sole discretion.