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.

Introduction

Vigil (gemogenovatucel-T) is a novel autologous tumor cell immunotherapy, which is constructed from harvested malignant tissue[1-3]. 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 high-grade 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 immuno-therapeutics[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[11-16] 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[11-15,18].

Nevertheless, we have studied patient subpopulations most sensitive to Vigil therapy based on molecular profile using NanoString assessment, and demonstrated that TIShigh 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.

Table 1

Demographics summary of all patients by ENTPD1/CD39 status.

	ENTPD1/CD39 Status, No. (%)
Characteristic	ENTPD1/CD39 Low	ENTPD1/CD39 High
No. of patients	45	46
Frontline chemotherapy
Neoadjuvant	6 (13.3%)	9 (19.6%)
Adjuvant	39 (86.7%)	37 (80.4%)
Stage
III	38 (84.4%)	39 (84.8%)
IV	7 (15.6%)	7 (15.2%)
Age (years)
Median (IQR)	62.0 (56–70)	63.5 (55–68)
Range	38–79	42–84
<65	27 (60%)	26 (56.5%)
>= 65	18 (40%)	20 (43.5%)
ECOG
0	31 (68.9)	30 (65.2)
1	14 (31.1)	16 (34.8)
Residual disease post-surgery
Macroscopic	13 (28.9%)	14 (30.4%)
Microscopic/NED	32 (71.1%)	32 (69.6%)

Fig. 1

CONSORT Diagram.

Flow of patients through the VITAL trial.

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 360TM 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.

Fig. 2

NanoString Statistical Algorithm.

Flow chart of all patients’ analysis. Analyzed both with genes as raw continuous data and with genes dichotomized. Genes were selected if the interaction term was significant in both analyses. 5% alpha was used unless noted.

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[5-7,19,20]. 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.

Table 2

Two-sided p values of the interaction term in the Cox model.

	Interaction term (continuous)* OS	Interaction term (continuous)* RFS	Interaction term (binary)** OS	Interaction term (binary)** RFS
ENTPD1/CD39	0.00751	0.00375	0.0158	0.00014
CCL13	0.0190	0.00271	0.044	0.00998
CD79B	0.00426	0.00280	0.0303	0.0152
MRC1	0.01040	0.0169	0.0173	0.000822

*Analyzed with genes as raw continuous data.

**Analyzed with genes dichotomized.

Table 3

One-sided p values of log-rank test comparing two KMs and hazard ratios and 90% CI from the univariate Cox proportional hazards model based on four predicted genes from multivariate analysis.

	Vigil ≥ median vs. Vigil < median OS		Vigil ≥ median vs. Vigil < median RFS		Vigil ≥ median vs. placebo ≥ median OS		Vigil ≥ median vs. placebo ≥ median RFS
	P value	HR	P value	HR	P value	HR	P value	HR
ENTPD1/CD39	0.002	0.177 [0.059, 0.524]	0.0003	0.238 [0.114, 0.498]	0.013	0.257 [0.087, 0.761]	0.00007	0.200 [0.094, 0.427]
CCL13	0.0005	0.119 [0.034, 0.423]	0.0003	0.236 [0.113, 0.493]	0.019	0.228 [0.063, 0.824]	0.006	0.338 [0.161, 0.709]
CD79B	0.006	0.248 [0.092, 0.670]	0.014	0.423 [0.219, 0.817]	0.027	0.324 [0.118, 0.892]	0.010	0.421 [0.224, 0.793]
MRC1	0.0001	0.058 [0.010, 0.325]	0.0004	0.229 [0.105, 0.502]	0.005	0.109 [0.019, 0.613]	0.001	0.245 [0.112, 0.535]