Molecular and Immunologic Signatures are Related to Clinical Benefit from Treatment with Vocimagene Amiretrorepvec (Toca 511) and 5-Fluorocytosine (Toca FC) in Patients with Glioma

High-grade glioma (HGG) comprises World Health Organization grade III and IV brain tumors that are usually anaplastic astrocytoma or glioblastoma (GBM) in histology (1). HGG is associated with dismal prognosis with 5-year survival rates of 6.8% (2), primarily due to cancer invasiveness and resistance to treatment that leads to tumor recurrence in nearly all patients (3). Numerous genetic alterations are believed to play a role in glioma etiology and progression, including mutations and copy-number variation (CNV) in genes from pivotal signaling pathways, such as EGFR, NF1, P53, RB1, RTK, RAS, and PI3K (4). Putative indicators of improved clinical outcomes include younger age and IDH1 mutations (5). IDH1 wild-type tumors are often stratified into three distinct prognostic subtypes using CNV data: a W1 subtype associated with worst survival, a W2 subtype associated with intermediate survival, and a W3 subtype associated with best survival (6). In addition, TERT promoter mutations are present in approximately 55% of GBM tumors, are associated with poorer prognosis, and are rarely observed with IDH1 mutations, chromosome 7 amplifications, or chromosome 10 deletions (4). Certain HLA alleles are associated with more favorable glioma clinical outcomes (7). The HLA gene complex on human chromosome 6 is one of the most polymorphic regions in the human genome and contributes in large part to the diversity of the immune system. Patient HLA class I (HLA-A, -B, and -C) genotype can influence cancer response to immunotherapy, and LOH at HLA class I loci in tumors is linked to decreased survival, especially among patients with low tumor mutational burden (TMB; ref. 8).

The tumor microenvironment is a complex mixture of factors in and around the tumor that can both promote and inhibit growth, including noncancerous cells such as leukocytes, fibroblasts, and endothelial cells (9). Tumor purity is the proportion of cancer cells in a tumor sample, and is hypothesized to be an intrinsic tumor characteristic associated with clinical covariates, as opposed to being an artifact of sample collection methodology (10). Previous work suggests that lower tumor purity is associated with GBM histology, mesenchymal molecular subtype, and shorter survival, whereas higher tumor purity is associated with anaplastic astrocytoma histology, proneural molecular subtype, and prolonged survival (11). Furthermore, immune cells in the tumor microenvironment are widely believed to affect tumor progression and response to therapy. Glioma tumors can evade immune surveillance by inducing an immunosuppressive tumor microenvironment through recruitment and induction of specific leukocyte subsets, such as tumor-associated macrophages (TAMs), microglia, myeloid-derived suppressor cells (MDSCs), and regulatory T cells (Tregs), which leads to an altered immune composition compared with normal brain (12–14). Immune modulatory cytokines, including IL1β, IL6, TNFα, VEGF, and GM-CSF, can induce immunosuppressive leukocytes (15). Cytokine measurements have been linked to glioma risk (16) and can potentially serve as biomarkers of clinical outcomes (17).

Vocimagene amiretrorepvec (Toca 511) is a cancer selective, retroviral replicating vector that encodes cytosine deaminase (18). When administered, extended release 5-fluorocytosine (Toca FC) is converted by cytosine deaminase into the potent, short lived, chemotherapeutic agent, 5-fluorouracil, which diffuses into the tumor microenvironment from Toca 511–infected cells (19, 20). Preclinical models have shown that this therapeutic regimen kills malignant cells as well as nearby immunosuppressive cells, eliminating tumor tissue and promoting an antitumor immune response that includes T-cell activation and trafficking in addition to B-cell activation and expansion (19–23). A functional immune system is required to elicit a durable response to Toca 511 and Toca FC cancer therapy in mice (21–23). A phase I open-label, ascending dose, multicenter clinical trial (NCT01470794) investigating intracranial Toca 511 and oral Toca FC treatment in 56 patients with recurrent HGG (rHGG) included safety, efficacy, and molecular profiling as endpoints. All patients received injections of Toca 511 into the surgical resection cavity wall and then 7-day courses of oral Toca FC every 6 weeks. As described by Cloughesy and colleagues (2018), median survival for all efficacy evaluable patients (n = 53) was 13.6 months [95% confidence interval (CI), 10.8–20.0], six patients exhibited durable complete responses, and median duration of follow-up for responders was 35.7 months (24). As of December 31, 2018 (33.9–52.2 months after Toca 511 administration), all responders were alive and remained in response, suggesting a positive association of durable response with overall survival. Furthermore, a subgroup of 23 patients that appeared to benefit from treatment helped to define entry requirements and dosage for a follow-up phase III clinical trial (NCT02414165). Hogan and colleagues (2018) analyzed patient samples from three phase I clinical trials of Toca 511 and Toca FC (NCT01156584, NCT01470794, and NCT01985256), and found that the vector was tumor selective and persistent, Toca 511 and Toca FC showed excellent tolerability, and there was no evidence for clonal expansion of cells with integrated Toca 511 DNA or preferential retrieval of integration sites near oncogene (25). Here, we describe molecular and immunologic profiles of patients with rHGG from a phase I clinical trial of Toca 511 and Toca FC (NCT01470794).

Clinical procedures, survival data, and responses for the phase I clinical trial of intracranial Toca 511 and oral Toca FC in patients with rHGG (NCT01470794) have been described previously (24, 26). Briefly, 56 patients with HGG that recurred after initial treatment with subtotal resection, postoperative radiation, and temozolomide were enrolled from February 2011 to October 2015. Planned, surgical resection of at least 80% of the enhancing tumor was carried out, and then ascending doses of Toca 511 were injected into the resection cavity wall. Tumor samples collected at time of surgery were used for molecular and immunologic analyses. Oral Toca FC was administered approximately 6 weeks after Toca 511 for weekly cycles repeated every 6 weeks. The earliest peripheral blood plasma samples suitable for proteomic analysis were collected from patients at time of first Toca FC administration. One cohort also received bevacizumab (10 mg/kg) intravenously every 2 weeks or lomustine (110 mg/m²) orally every 6 weeks beginning with the first cycle of Toca FC. The primary endpoint was to identify dose-limiting toxicities of Toca 511 and Toca FC treatment. Objective response was determined by independent radiology review, accounting for neurologic status and corticosteroid use, using modified Macdonald criteria for all cohorts except one, which was evaluated using modified Response Assessment in Neuro-Oncology criteria. Most patients were using ≤2 mg/day systemic corticosteroids with dexamethasone. Among patients with a response, only 1 was on corticosteroids at varying doses of dexamethasone of 1–4 mg. Evaluable and measurable disease required two dimensions of the lesion be at least 1 cm in length for assessment of partial or complete response; smaller lesions were considered evaluable, nonmeasurable disease that could only be assessed for complete response. Baseline measurement for radiologic assessment of all patients was an MRI scan obtained approximately 6 weeks after surgical resection and Toca 511 injection, just prior to Toca FC administration.

Venous whole blood was collected from each patient at the clinical site into EDTA tubes. Plasma was separated via centrifugation, aliquoted into cryovials, and then stored at −80°C. Cytokines were quantified in patient plasma using a digital ELISA-based Single Molecule Array (Simoa) custom HumanMAP (Myriad RBM). The following proteins were measured in each plasma sample: BDNF, E-Selectin, Eotaxin-1, Factor VII, GM-CSF, ICAM-1, IFNγ, IP-10, IL1α, IL1β, IL1ra, IL2, IL3, IL4, IL5, IL6, IL6r, IL7, IL8, IL10, IL12p40, IL12p70, IL15, IL17, IL18, IL23, MIP-1-alpha, MIP-1-beta, MIP-3-alpha, MMP-3, MMP-9, MCP-1, MCP-2, MCP-4, MIG, MPIF-1, SCF, TNFα, NFβ, and VEGF.

For RNA-sequencing, tumor pieces were separated from bulk tumor, total RNA was isolated using Maxwell 16 Total RNA Purification Kit (Promega, AS1050), and libraries were prepared using TruSeq Stranded Total RNA LT (Illumina) according to the manufacturer's recommendations. For whole-exome sequencing (WES), about 1 μg of genomic DNA isolated from tumor or blood was fragmented and prepared for sequencing using SureSelect V5 (Agilent) according to the manufacturer's instructions. All exome and RNA-sequencing were performed on the Illumina HiSeq 2000/2500 Platforms by Macrogen Corp. and Siemens Healthineers. Upstream data processing and sequence quality control were done on the DNAnexus platform. For RNA-sequencing data, quantification of RNA expression levels was performed with Salmon v 0.7.2, and reads were also aligned to the human ENSEMBL reference RNA database and then to the human genome with STAR v2.5.3a. Exome sequencing data were aligned to the human genome reference using BWA-MEM v0.7.12. We considered the intersection of high-confidence variants called using Sentieon v201611 and MuSE v1.0rc. Variants were annotated using SNPeff v4.3. Four‐digit–typed HLA alleles of each patient were identified from exome sequencing data using Omixon HLA Explore v1.7.0 (http://www.omixon.com). NetMHCpan v4.0 and NetMHCIIpan v3.2 were used to predict binding of mutant peptides to the patients' MHC class I and class II molecules, respectively. Exome data were analyzed for copy-number variants using CNVkit version 0.9.1. Copy-number subtypes were defined by four genetic loci: gain of whole chromosome 1 (gChr1), gain of whole chromosome 19 (gChr19), and coamplification of CDK4/MDM2 (caCDK4/MDM2); subtypes were called according to the algorithm described in Cimino and colleagues 2018 (6), as follows: W1 = [No gChr1 + No gChr19 + caCDK4/MDM2] or [gChr1 + caCDK4/MDM2]; W2 = [No gChr1 + No gChr19 + No caCDK4/MDM2]; and W3 = [No gChr1 + gChr19] or [gChr1 + No caCDK4/MDM2].

Biostatistical analyses and graphical visualizations were performed using R version 3.4.1 or later (R Core Team, 2017), unless otherwise specified. Data were converted from wide format to long format using the reshape and reshape2 packages to facilitate plotting using ggplot2, ggfortify, and ggforce packages. Cox proportional hazard survival regressions were carried out using the survival package and Kaplan–Meier plotting was carried out using ggplot2 and survminer packages. Tumor-infiltrating leukocytes were enumerated from bulk tumor RNA-sequencing using iSort (CiberMed, Inc.), a gene expression deconvolution framework based on CIBERSORT technology that currently supports the detection of 23 functionally defined human immune subsets in bulk tumor expression data in a manner that is robust to changes in the immune cell phenotypes in the tumor milieu (27, 28). This allows for the “relative” quantification of leukocyte composition in relation to total immune content and “absolute” quantification of leukocytes in relation to total tumor cellularity. Two-sided Wilcoxon rank-sum tests were used to determine significant differences in tumor-infiltrating leukocyte estimates between responders and nonresponders, because tumor-infiltrating leukocyte values were not normally distributed and the directionality of differences was not assumed. The described tumor immune signature was calculated using the relative frequencies of four tumor-infiltrating leukocyte subsets. The positive integer 1 was included in the denominator to provide equal weighting to immune subsets that were positively and negatively associated with clinical benefit, as well as to prevent division by zero in the absence of natural killer (NK) cell and M0 macrophage infiltrates. Patients in the top quartile of tumor immune signature values were designated as the “high” group, while all other patients were designated as the “low” group for survival analysis. Similarly, the described cytokine signature, based on relative rankings within all cytokines measured, used the maximum concentration of three cytokines detected by multiple digital ELISA in peripheral blood plasma samples. The positive integer 1 was included in the denominator to provide equal weighting to cytokines that were positively and negatively associated with clinical benefit, as well as to prevent division by zero in the absence of IL6. Patients above the median cytokine signature value were designated as the “high” group, while all other patients were designated as the “low” group for survival analysis.

The described phase I clinical trial of intracranial Toca 511 and oral Toca FC in rHGG (NCT01470794) was approved by the institutional review board at each site and was in compliance with the Declaration of Helsinki, Good Clinical Practice guidelines, local laws, and international ethical guidelines for biomedical research involving humans. All participants provided written informed consent prior to inclusion in the study.

We sought to generate patient profiles that include genetic, molecular, and immunologic features to elucidate therapeutic mechanisms of action and identify potential indicators of clinical benefit in a phase I clinical trial of intracranial Toca 511 and oral Toca FC (NCT01470794). Because of limited sample and data availability, this work includes three subpopulations of patients whose demographic characteristics and clinical outcomes, including objective response and survival from time of surgery, are shown in Table 1. Surgery coincides with start of treatment, because Toca 511 virus was injected into the resection cavity wall, and we analyzed rHGG tumor tissue that was collected at this time. Plasma samples were derived from blood samples that were collected at least 6 weeks after surgery, when Toca FC prodrug was first administered. Nonresponders refer to patients with stable or progressive disease, while responders refer to patients who experienced partial or complete objective response as determined by independent radiology review.

It is well established that genetic alterations are linked to glioma etiology and progression (29). Therefore, we used DNA sequencing to characterize the genomes of tumor and blood samples collected from patients with rHGG who received intracranial Toca 511 and Toca FC. WES of DNA isolated from 32 patients achieved 175-fold mean target coverage for tumor and 98-fold mean coverage for blood, with 91% of bases covered at least 50-fold for tumor and 88% of bases covered at least 30-fold for blood. This allowed us to assess genetic alterations that are potentially associated with clinical outcomes, including CNV, TMB, and chromosomal aberrations (Fig. 1). Patients with longer survival were enriched for tumors with IDH1 mutations. Four patients had mutations in IDH1, three with IDH1-R132H and one with IDH1-R132S. All four patients also had mutations in TP53. As expected, the four patients with IDH1 mutations did not have TERT promoter mutations, although at least 41% of patients had TERT promoter mutations, including C228T and C250T. In addition, 69% of tumors had copy-number changes or mutations in the p53 pathway (MDM2, MDM4, TP53, and CDKN2A), 72% had alterations in the Rb pathway (CDK4, CDK6, CCND2, CDKN2A/B, and RB1), 25% had alterations in the PI3K pathway (PIK3CA, PIK3R1, PIK3C2G, PIK3CG, PIK3CB, PIK3C2B, PIK3C2A, and PIK3R2), 69% had alterations in either the PI3K pathway or PTEN, and 69% had alterations in the RTK pathway (EGFR, PDGFRA, MET, and FGFR1/2/3). RTK dysregulation is largely accounted by EGFR mutations or CNVs (66% of tumors). These include 28% of tumors with mutations in EGFR and 63% with whole chromosome 7 amplifications. In tumor samples from three patients, no driver mutation or copy-number variant known to be associated with GBM was called, even when expanding the list of putative GBM-associated gene targets. There were no commonalities between these samples regarding molecular subtype, previously determined from RNA-sequencing (26), or estimated tumor purity. All three patients had low tumor mutational load and few copy-number alterations called. Two of the patients had overlapping losses at 16p11.2. All three patients had CNVs at 8p11.22 specifically affecting ADAM3A, including one gain and two heterozygous losses (Supplementary Fig. S1). Small CNVs altering ADAM3A were common; among the 45 patients assessed, nearly 69% gained and 31% lost copies. The only patient without a CNV affecting ADAM3A had an IDH1 wild-type tumor, and exhibited a complete response to Toca 511 and Toca FC therapy. We compared our CNV results with two publicly available GBM patient datasets: The Cancer Genome Atlas (TCGA, n = 256) and German Glioma Network (GGN, n = 243). As expected, the moderate survival W2 copy-number subtype was the most commonly observed in our patient cohort, and the distribution of the three subtypes skewed toward the W2 subtype compared with that of control population cohorts (Supplementary Fig. S2). In this study, 14% of IDH1 wild-type patients fit the W3 copy-number subtype, compared with 34% in TCGA and 32% in GGN datasets.

If tumor mutations give rise to neoantigens that a patient's immune system can recognize via MHC binding, they are more likely to mount antitumor immune responses when cancer cells are lysed by Toca 511 and Toca FC treatment. To determine whether immunogenic neoantigens are present, we combined HLA genotyping of blood with mutational analysis of tumor, and considered whether any observations were associated with clinical outcomes. Five patients were homozygous for one HLA class I locus (HLA-A, -B, or -C), 23 patients were heterozygous for all three loci, and four patients had one locus whose zygosity could not be determined. Patients who survived for more than 2 years were either heterozygous or indeterminate for all three HLA class I loci, and these patients were enriched for the B*07-Cw*07 haplotype. Responders were enriched for the HLA-B44 supertype, but the results were not statistically significant. Our results do not support other previously published correlations between GBM occurrence or survival and Cw*01, A*32 (specifically A*3201), B*55, the HLA-B62 supertype (including HLA-B*15:01), B*13, or B*07 (7, 8, 30). Following analysis of somatic mutations and patient-specific HLA typing, an in silico workflow (31) identified candidate tumor neoantigens for each patient, excluding hypermutations. Mutations in PIK3CA gave rise to MHC class I neoantigens in three patients and MHC class II neoantigens in four patients. In addition, mutations in EGFR, SYNE1, CAD, ARSI, and IDH1 each generated MHC class II neoantigens in two or more patients. Total neoantigen load was not correlated with survival, although patients who survived for more than 2 years were enriched for the aforementioned neoantigens. Among patients with nonhypermutated tumors, IDH1, PIK3CA, EGFR, or SYNE1 neoantigens were detected in 86% of patients who survived for more than 2 years, but in only 26% of patients who survived less than 2 years.

Like many cancers, glioma can induce an immunosuppressive microenvironment to evade immune destruction, and clinical outcomes can be related to the quantities of different white blood cell subsets in and around the tumor (12–14). In addition, Toca 511 and Toca FC cancer treatment has a putative mechanism of action that includes T-cell–mediated antitumor immune activity (21). Therefore, we sought to determine the immune composition of the tumor microenvironment in patients with rHGG treated with intracranial Toca FC and oral Toca 511 to identify potential predictive indicators of clinical benefit from therapy. A deconvolution platform built upon CIBERSORT technology, iSort (27, 28), estimated 23 different tumor-infiltrating leukocyte subsets using RNA-sequencing data for 109 tumor samples derived from 45 patients. These estimates can be expressed as either the absolute frequency of all cells in the sample, or as the relative frequency of immune cells in the sample. Absolute and relative leukocyte subset frequencies for all tumor samples from each patient were averaged (Fig. 2) and then used to investigate potential associations with clinical outcomes, including objective response and survival. At time of surgery, complete responders exhibited higher tumor-infiltrating M1 macrophages, activated memory CD4 T cells, gamma delta T cells, monocytes, and neutrophils, but lower tumor-infiltrating M0 macrophages and resting NK cells, compared with nonresponders (Fig. 3; Supplementary Fig. S3). On the basis of these results, we devised a tumor immune signature using the following equation:

Higher values of this signature indicate that more activated memory CD4 T cells, more M1 macrophages, fewer resting NK cells, and fewer M0 macrophages were detected in a patient's tumor. As expected, this signature was higher in responders than nonresponders (Wilcoxon rank-sum test, P < 0.001). Patients in the top quartile, or highest 25%, of the tumor immune signature exhibited longer overall survival times than patients with lower values (Fig. 2C). More than half of these patients were still alive at last contact, and Cox proportional hazards regression accounting for patient age and gender confirmed that these patients survived significantly longer than other patients in the study (HR, 0.17; 95% CI, 0.063–0.45).

Data from TCGA and Immune Prediction of Clinical Outcomes from Genomic Profiles indicate that higher M0 macrophage infiltration may be prognostic of poorer rHGG outcomes regardless of treatment, but resting NK cells, activated CD4 memory T cells, and M1 macrophage tumor infiltrates are not prognostic to the disease in other settings (Supplementary Fig. S5). Consistent with TCGA data (32), IDH1-mutant tumors exhibited less M0 macrophage infiltration than IDH1 wild-type tumors (Fig. 3A). In addition, more total leukocyte infiltration was detected in patient tumors with greater than 300 mutations (Fig. 3B). Tumors collected from patients with multiple recurrences exhibited higher levels of tumor-infiltrating resting NK cells and resting memory CD4 T cells, as well as lower levels of monocytes and neutrophils than patients at first recurrence (Supplementary Fig. S4), but immune cell infiltrates did not show any associations with tumor diameter or histology (data not shown). In addition, normal brain tissue from the Genotype-Tissue Expression (GTEx) project (33) had a different distribution of leukocyte subsets and less absolute immune content than the glioma tumors in this study (Fig. 3C and D), suggesting that our results were unlikely to represent normal brain tissue in the samples.

We also used DNA sequencing data to estimate tumor purity for 32 patients to investigate whether it was potentially linked to molecular and immunologic features, or to clinical benefit from therapy. Consistent with previously published work, neural subtype tumors were less pure than other molecular subtypes (Supplementary Fig. S6A). Tumor purity was positively correlated with tumor-infiltrating M1 macrophages, follicular Th cells, resting memory CD4⁺ T cells, and Tregs, and inversely correlated with memory B cells and naïve CD4⁺ T cells (Supplementary Fig. S6B–S6D). However, tumor purity did not appear to be associated with molecular features or with clinical outcomes (Fig. 1). Tumor purity estimates of multiple samples from the same patient were correlated (R² = 0.44; P < 0.0001; Supplementary Fig. S6E). In low-purity samples, low-frequency mutations fell below the limit of detection or were attributed to sequencing error. Higher tumor purity was associated with more called mutations and CNVs (P = 0.005; Supplementary Fig. S6F).

Preclinical evidence suggests that durable responses to Toca 511 and Toca FC treatment, which we observed in the clinical trial under investigation, should be accompanied by antitumor immune activity. As a result, immune signaling proteins indicating an ongoing response to treatment may be detectable in patients who were experiencing clinical benefit from therapy. Therefore, we used a digital multiplex ELISA platform to quantify 40 different cytokines in 87 peripheral blood plasma samples that were collected from 24 patients with glioma at multiple timepoints after surgery and subsequent Toca 511 and Toca FC administration. These measurements enabled investigations of temporal immunologic changes associated with clinical outcomes. Ten cytokines were below the limit of detection and/or were invariant between samples, but the remaining 30 cytokines exhibited intra- and interpatient variation (Supplementary Fig. S7). Quantities of different cytokines were highly correlated (Supplementary Fig. S8A), therefore, we performed principal components analysis to help determine whether any cytokine variability may be related to patient outcomes. The first seven cytokine principal components accounted for more than 60% of variance observed in the dataset (Supplementary Fig. S8B). The eighth and 12th cytokine principal components showed associations with complete response to therapy and overall survival (Supplementary Fig. S9). The top eigenvectors in these components included the cytokines E-selectin, MIP-1-beta, and IL6, each of which individually showed associations with patient outcomes. Responders showed elevated plasma concentrations of E-selectin and/or MIP-1-beta within 5 months of receiving Toca FC for the first time, whereas nonresponders showed elevated IL6 in the same timeframe (Fig. 5A). Moreover, all three of these cytokines showed univariate associations with patient survival and had higher relative rankings (Supplementary Table S1). Thus, we devised a cytokine signature that incorporates the maximum value of these three cytokines across all posttreatment samples from each patient using the following equation:

Increasing values of this cytokine signature indicate higher peak E-selectin, higher peak MIP-1-beta, and lower peak IL6 in the patient's peripheral blood during and after Toca 511 and Toca FC treatment. Patients above the median value for this posttreatment cytokine signature exhibited significantly longer survival, even when adjusting for age and gender (HR, 0.13; 95% CI, 0.038–0.45), and more than half of this patient group was still alive at last contact (Fig. 5B). A multivariate model including concentrations of these three cytokine across all tumor samples, as well as patient age and gender, further indicated that improved survival was associated with higher posttreatment E-selectin (HR, 0.47; 95% CI, 0.22–0.98) and MIP-1-beta (HR, 0.38; 95% CI, 0.22–0.63), but worse survival was associated with higher posttreatment IL6 (HR, 2.10; 95% CI, 1.22–3.63).

This work investigated potential relationships between Toca 511 and Toca FC administration clinical outcomes and glioma patient profiles composed of molecular, genomic, and immunologic features. In smaller study cohorts, such as this phase I trial under investigation, risk enrichment for DNA copy-number subtypes associate with better survival due to the selection of patients who do well enough to enter a clinical trial or who are deemed well enough to undergo resection at recurrence (6). Compared with publicly available datasets, CNV for IDH1 wild-type patients in this study skewed toward the intermediate prognostic W2 subtype, and we observed a smaller proportion of patients in the best prognostic W3 subtype, suggesting there was no bias toward longer surviving patients. This is consistent with our previous observation that patients with durable responses were not predisposed toward tumor mRNA profiles associated with improved survival (26, 34). The two overlapping deletions observed at 16p11.2 were unlikely to be drivers of tumor progression, because CNV in this genomic region is frequently observed in the general population (35). CNV results also corroborated previous research suggesting that alterations to ADAM3A and other ADAM family genes are frequently found in cancer, including glioma, and may be associated with poor prognosis (36–38). HLA genotyping results, on the other hand, did not support previously published associations between patient outcomes and the following haplotypes: Cw*01, A*32 (specifically A*3201), B*55, the HLA-B62 supertype (including HLA-B*15:01), the HLA-B44 supertype, B*13, B*07, or B*07-Cw*07 (7, 8, 30, 39). However, patients who survived longer than 24 months were enriched for the B*07-Cw*07 haplotype, which is associated with GBM risk in Caucasian populations (7). In addition, patients with improved survival had tumor neoantigens arising from IDH1, PIK3CA, EGFR, or SYNE1, genes where mutations have previously established association with various cancers, including glioma (40, 41). The strong correlations observed between tumor purity estimates from samples collected from the same patient in this study are consistent with the hypothesis that purity is an intrinsic characteristic of the tumor developing in a suitable microenvironment. Similar to previous studies, there were differences in tumor purity between molecular subtypes (11). Neural subtype tumors generally had the lowest purity estimates, followed by mesenchymal, while proneural subtype tumors had the highest purity estimates. Tumor purity was positively correlated with the number of mutations that were called. We previously reported that samples with the neuronal phenotype were potentially from the nonenhancing region of the tumor, which has a higher fraction of nonneoplastic cells (26). Consistent with this, samples with neuronal phenotype consistently had low tumor purity (Supplementary Fig. S6). The most abundant cell types in the brain, neurons, microglia, and astrocytes, were not measured by the iSort deconvolution method applied herein.

Immunologic assessments of the patients with rHGG in this phase I clinical trial support a mechanism of action for Toca 511 and Toca FC, previously determined in preclinical studies, that includes (i) alleviating immune suppression by killing MDSCs and TAMs in the tumor microenvironment and (ii) promoting a T-cell–mediated antitumor immune response by releasing tumor-associated antigens (TAAs), pathogen-associated molecular patterns, and damage-associated molecular patterns (19, 21–23, 42). In this study, RNA-based tumor-infiltrating leukocyte estimates suggest that lower NK cells, lower M0 macrophages, higher M1 macrophages, and higher activated CD4⁺ memory T cells in the tumor microenvironment at time of surgery can potentially predict therapeutic benefit. Elevated MIP-1-beta and E-selectin in peripheral blood plasma appears to reflect an ongoing response to therapy, whereas elevated IL6 indicates no ongoing response. Tumor-infiltrating NK cells potentially reduce efficacy by inhibiting Toca 511 infection and spread via antiretroviral activity (43). Tumor-infiltrating M0 macrophages are immunosuppressive and provide a reservoir for MDSCs and TAMs, creating a barrier to overcoming tumor tolerance and establishing a durable antitumor immune response (32, 44). Lower M0 macrophage infiltration in IDH1-mutant tumors may contribute to the previously reported improved clinical outcomes for those patients (26). Unlike other macrophage subsets, M1 macrophages in the tumor microenvironment are believed to be associated with better prognosis in multiple cancers, including glioma (45). Activated CD4⁺ memory T-cell infiltrates suggest a halted antitumor immune response that can resume when immunogenic TAAs are released and immune suppression is reduced (13, 46). This could explain the increased T-cell activation and proliferation previously reported in these patients (26). A robust antitumor immune response could involve leukocyte retention in the tumor microenvironment through endothelial expression of the cell adhesion molecule, E-selection (47), as well as activation of M1 macrophages in the tumor microenvironment to secrete T-cell chemoattractants like MIP-1-beta (48, 49).

Multiple patients in this phase I clinical trial experienced a durable response and survival greater than 1 year, which is unusual for patients with rHGG (24). Previous preclinical work has demonstrated that Toca 511 and Toca FC mechanism of action operates as an immuno-gene therapy (19–23). Moreover, flow cytometric immunophenotyping of PBMCs in some patients with rHGG from this study suggested that complete responders exhibit increased activation and proliferation of T cells during therapy (24). This led us to expect that pretreatment immunologic status and temporal immune modulations during treatment are key factors for patient outcomes and could potentially help predict and monitor responses to therapy. The patient profiles described here support such a hypothesis. Pretreatment tumor immune infiltrates and neoantigens, as well as posttreatment cytokine concentrations showed associations with objective response and survival, suggesting that immune measurements could help select patients and monitor therapeutic benefit to improve clinical decision-making. A pretreatment tumor immune signature and posttreatment cytokine signature were both associated with improved survival times when accounting for patient age and gender in multivariate statistical models. However, due to sample size and trial design limitations these results need to be tested in additional studies. The larger sample size and standard-of-care control arm in a phase III clinical trial of patients with rHGG (NCT02414165) should help elucidate potential indicators of clinical benefit.

A.A. Alizadeh reports other from CiberMed (equity) and Karyopharm (consulting) and personal fees from Roche during the conduct of the study, as well as has a patent for CIBERSORT issued and licensed to CiberMed and a patent for CAPP-seq issued and licensed to Roche, both assigned to Stanford. M. Diehn reports other from CiberMed (founder) during the conduct of the study and Foresight Diagnostics (founder) outside the submitted work, personal fees from Roche, AstraZeneca, and RefleXion, grants and personal fees from Illumina, and grants from Varian Medical Systems outside the submitted work, as well as has a patent for cancer biomarkers and a patent for circulating tumor DNA pending, one (CTC) licensed to Roche, and both owned by Stanford University. H.E. Gruber reports grants from FDA and other from Apollo Bio (licensing partner) during the conduct of the study and outside the submitted work, as well as has patents for the drugs studied in the article, owned by De Novo Biopharma and licensed to Apollo Bio for Greater China. D.J. Jolly reports personal fees from Tocagen Inc (salary and stock) during the conduct of the study, personal fees from Tocagen Inc. (salary and stock) outside the submitted work, as well as has a patent for US8722867, on a modified cytosine deaminase gene and its expression in viral vectors as an anticancer agent in conjunction with the prodrug 5-fluorocytosine, assigned as part of his employment agreement with Tocagen Inc., and recently owned by De Novo Biopharma. D. Ostertag reports other from Tocagen Inc. (employee) during the conduct of the study, and reports a patent titled cancer treatment with recombinant vector for US20130130986A1 issued, a patent titled cancer combination therapy and recombinant vectors for US20140178340A1 issued, and a patent titled retroviral vector having immune-stimulating activity for US20170175137A1 issued, all owned by De Novo Biopharma. No potential conflicts of interest were disclosed by the other authors.

W.P. Accomando: Conceptualization, data curation, formal analysis, supervision, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. A.R. Rao: Conceptualization, data curation, formal analysis, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. D.J. Hogan: Conceptualization, data curation, formal analysis, supervision, validation, investigation, visualization, methodology, writing-review and editing. A.M. Newman: Resources, supervision, investigation, methodology, writing-review and editing. A. Nakao: Data curation, formal analysis, investigation, visualization, methodology, writing-review and editing. A.A. Alizadeh: Resources, supervision, project administration. M. Diehn: Resources, supervision, project administration. O.R. Diago: Investigation, methodology. D. Gammon: Supervision, investigation, project administration. A. Haghighi: Formal analysis, investigation, methodology. H.E. Gruber: Conceptualization, resources, supervision, funding acquisition, investigation, project administration, writing-review and editing. D.J. Jolly: Conceptualization, resources, supervision, funding acquisition, investigation, project administration, writing-review and editing. D. Ostertag: Conceptualization, resources, data curation, formal analysis, supervision, investigation, methodology, writing-original draft, writing-review and editing.

This work was funded by Tocagen, Inc. The authors would like to thank the patients and their families for participating in Tocagen's clinical trials. We would also like to acknowledge contributions made by all of the clinicians, hospitals, and caregivers involved in the phase I open-label, ascending dose clinical trial (NCT01470794). In particular, we would like to thank Timothy F. Cloughesy (University of California, Los Angeles, CA), Steven N. Kalkanis (Henry Ford Hospital, Detroit, MI), Tobias Walbert (Henry Ford Hospital, Detroit, MI), Tom Mikkelsen (Henry Ford Hospital and Ontario Brain Institute, Detroit, MI), Joseph Landolfi (JFK Medical Center, Edison, NJ), Bob Carter (Massachusetts General Hospital, Boston, MA), Clark C. Chen (University of Minnesota, Minneapolis, MN), Michael A. Vogelbaum, (Moffitt Cancer Center, Tampa, FL), James B Elder, (Ohio State University, Columbus, OH), and David Piccioni (University of California, San Diego, CA). The authors also acknowledge the ABC2 Foundation (Washington, DC), the National Brain Tumor Society (Watertown, MA), the American Brain Tumor Association (Chicago, IL), the Musella Foundation (Hewlett, NY), and Voices Against Brain Cancer (New York, NY) for their support and collaborations. The results published here are in whole or part based upon data generated by TCGA Research Network: https://www.cancer.gov/tcga.

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.