A Phase I Dose-Escalation Study to Evaluate the Safety and Tolerability of Evofosfamide in Combination with Ipilimumab in Advanced Solid Malignancies

This article is featured in Highlights of This Issue, p. 2959

Tumor hypoxia predicts poor outcomes across all cancers, and has long been recognized as a critical driver of resistance to both chemotherapy and radiotherapy (1) and, more recently, of immune evasion (2, 3). Indeed, we recently showed that T cells fail to accumulate in hypoxic zones of prostate cancers, and that hypoxia suppresses key elements of the tumor stroma to establish and maintain an immune “cold” and immunotherapy-resistant state (4). Accordingly, alleviation of hypoxia has been demonstrated to reverse T-cell exclusion and restore therapeutic sensitivity to immunotherapy in preclinical models of prostate cancer, head and neck cancer, melanoma, and lung cancer (3–6).

Immune checkpoint blockade, particularly by antibodies targeting the CTL-associated antigen 4 (CTLA-4) and/or the programmed death 1 (PD-1) T-cell coinhibitory receptors, has revolutionized therapy for numerous cancers. These antibodies contributed significantly to a record drop in U.S. cancer deaths in the last year of reporting, although efficacy remains limited outside of immunologically “hot” or “warm” cancers that are already infiltrated with immune cells and/or that generate more mutational neoantigens (7). In these “best case” settings, response rates to checkpoint blockade range from lows of 15% [in human papillomavirus (HPV)-negative head and neck, bladder, and gastric cancer] to as high as 87% (in Hodgkin lymphoma). Most patients, however, present with “cold” tumors that are almost completely resistant to checkpoint blockade (7, 8), leaving the overall percentage of patients achieving objective clinical response below 15%, even within approved indications (9). Thus, developing therapies that sensitize “cold” cancers to checkpoint blockade has become a preeminent goal in cancer immunotherapy.

While the role of hypoxia in mediating immune privilege has been previously recognized in “cold” tumors (10), recent data also show a pivotal role in driving checkpoint resistance in “hot” tumors like melanoma (11). We previously showed that evofosfamide (TH-302) significantly reduces hypoxia across murine cancer models via a tissue remodeling process that restores normal, healthy vasculature capable of maintaining oxygenated blood supply (4, 5). Evofosfamide is a hypoxia-activated, 2-nitroimidazole prodrug of bromo-isophosphoramide mustard, a cytotoxin that was clinically successful in numerous trials as an adjuvant to chemotherapy until failing a pivotal phase III trial with gemcitabine in pancreatic cancer (12). Follow-up studies showed, however, that this failure was due to poor bioavailability from a new formulation of the drug, resulting in suboptimal dosing (13). Evofosfamide-mediated remodeling of hypoxic zones and the associated immunosuppressive microenvironment may enable ipilimumab, an antibody that blocks CTLA-4, to trigger immune rejection of otherwise resistant solid tumors (14). In addition, evofosfamide-induced immunogenic cell death may enhance presentation of tumor antigens, leading to clonal expansion of tumor-specific T cells, especially in combination with ipilimumab (4, 15). Therefore, we sought to evaluate the safety and tolerability of combining evofosfamide and ipilimumab in patients with advanced solid tumors (NCT03098160).

This phase I study assessed the safety, preliminary efficacy, and pharmacokinetics of evofosfamide in combination with ipilimumab. Adult patients with metastatic or locally advanced pancreatic cancer, castration-resistant prostate cancer, melanoma refractory to immune checkpoint blockade, or HPV-negative (HPV⁻) head and neck squamous cell carcinoma were eligible. Other eligibility criteria included an Eastern Cooperative Oncology Group (ECOG) score of ≤ 2, adequate organ function within 7 days of day 1 of therapy, and measurable disease as defined by the immune-related RECIST (irRECIST) in pancreatic cancer, melanoma, or HPV⁻ head and neck cancer (16, 17). As conventional RECIST 1.1 does not account for the response pattern typically seen with immune checkpoint inhibitors, that is, extended time to response and tumor response after initial disease progression, we chose to use irRECIST, which requires confirmation of disease progression on two occasions to ensure that true disease progression is observed rather than pseudoprogression due to early, transient tumor enlargement following infiltration by immune cells. On the other hand, patients with castration-resistant prostate cancer could have measurable or evaluable disease defined by any of the following: PSA progression, as indicated by a minimum of two rising values (three measurements) obtained a minimum of 7 days apart with the last result being ≥ 1.0 ng/mL; new or increasing nonbone disease (RECIST 1.1); or positive bone scan with two or more new lesions. Major exclusion criteria were history of grade 3 or 4 adverse events from prior ipilimumab or ipilimumab combination therapy, known risk factors for bowel perforation, long-term systemic steroid use (>10 mg daily prednisone equivalent initiated > 2 weeks before start of study treatment), and presence of active or uncontrolled autoimmune disease.

The protocol was a single-center open-label trial approved by the Institutional Review Board at MD Anderson Cancer Center (Houston, TX), and conducted in compliance with the Declaration of Helsinki. Written informed consent was obtained from all patients or their legal representatives.

A standard 3+3 design was utilized to combine 3 mg/kg ipilimumab with escalating doses of evofosfamide. In particular, 3–6 patients were enrolled sequentially at 400, 480, 560, and 640 mg/m² evofosfamide (Supplementary Table S1). Evofosfamide was administered as a 60-minute intravenous infusion on days 1 and 8 of cycles 1 and 2 of four 3-week cycles. Ipilimumab was administered as a 90-minute intravenous infusion on day 8 of each cycle for a maximum of four doses. Patients who achieved complete response, partial response, or stable disease after completion of cycle 4, as assessed by irRECIST, were eligible for retreatment, as were patients who experienced disease progression at any time after start of treatment, if the investigator determined that such retreatment had positive benefit-risk ratio, in consultation with the sponsor's medical monitor. In patients eligible for retreatment, the same dose and schedule for evofosfamide and ipilimumab were used for cycles 5 to 8.

Investigators used irRECIST at baseline and every two cycles to determine tumor response. Safety was evaluated by history at baseline and at regular intervals, along with physical and laboratory assessments. Severity of adverse events was assessed according to the NCI Common Terminology Criteria for Adverse Events version 4.03, beginning after the first dose and up to 30 days after the last dose.

For pharmacokinetic analysis, blood samples were drawn predose and at 0.5, 1, 2, and 4 hours after evofosfamide infusion. C_max, AUC_last, and T_max for evofosfamide and bromo-isophosphoramide mustard (Br-IPM) were derived from plasma concentrations using noncompartmental analysis. For analysis of circulating biomarkers, blood samples were drawn before the first dose and after each 3-week cycle. Following whole-blood collection, peripheral blood mononuclear cells (PBMC) were separated at room temperature along a Ficoll gradient (Histopaque-1077, Sigma), as described previously (18). At the conclusion of the trial, PBMC samples were analyzed by 18-parameter flow cytometry on a BD LSR II instrument. For pharmacodynamic analysis, pimonidazole, a weak base that allows quantification of tumor hypoxia by IHC, was administered with patient consent 5–24 hours prior to tumor biopsy, at an oral dose of 0.5 g/m² (19). Paired tumor biopsies (at baseline and at week 7) were obtained and were analyzed by flow cytometry for quantitative and qualitative changes in hypoxia and in the immune microenvironment as described previously (4). DNA and RNA were also isolated using AllPrep DNA/RNA Mini Kit (Qiagen). RNA sequencing was performed by Avera in 2 × 100 bp paired-end runs.

RNA sequencing data were analyzed in collaboration with Baylor College of Medicine Multi-Omics Data Analysis Core. Briefly, sequencing reads were trimmed using Trim Galore (The Babraham Institute, Babraham, Cambridge, United Kingdom), mapped using HISAT (20) against the human genome build UCSC hg38, and quantified using StringTie (21) against the Gencode model (22). Gene expression [fragments per kilobase of transcript per million mapped reads (FPKM)] was quantile normalized in R, and genes differentially expressed between tumor and normal samples were identified by parametric t test at P < 0.05 and fold change 1.25. Pathway enrichment was evaluated in gene set enrichment analysis (GSEA) (http://software.broadinstitute.org/gsea/index.jsp; ref. 23); significance was achieved for adjusted q values (q < 0.25). Heatmaps were generated using Matplotlib, NumPy, and SciPy libraries in Python. Sequencing data have been deposited in Gene Expression Omnibus under accession number GSE159237.

The safety profile of combining evofosfamide and ipilimumab, as well as the recommended phase II dose of evofosfamide in this context, were the primary endpoints, with objective response rate and pharmacokinetics as secondary endpoints. Baseline and change from baseline in peripheral blood and tumor tissue immune and hypoxia parameters were exploratory endpoints.

Dose-limiting toxicity was defined as any of the following treatment-emergent adverse events: grade 3 or 4 nonhematologic toxicity; grade 3 or 4 nausea, vomiting, or diarrhea that cannot be managed medically; grade 3 or 4 aspartate aminotransferase (AST) or alanine aminotransferase (ALT) elevation lasting > 7 days; grade 4 neutropenia lasting > 5 days; or febrile neutropenia. The surveillance period for dose-limiting toxicity was 21 days (cycle 1).

Descriptive statistics were used to summarize safety, tumor response, and biomarker data. Safety data are reported using preferred terms defined in the Medical Dictionary for Regulatory Activities. Enrollment of up to 30 patients was anticipated to determine recommended phase II dose. Safety was assessed in all patients who received one or more doses of evofosfamide or ipilimumab. A safety review committee reviewed safety and pharmacokinetic data prior to each dose escalation. Tumor response was assessed in all patients, and objective response rate was calculated as the proportion of patients with measurable disease at baseline who subsequently experienced complete or partial response. For patients with prostate cancer with evaluable biomarkers, a decline in PSA of > 50% from baseline, measured twice at least 4 weeks apart, was considered a response. For subjects with an observed decline in PSA while on study, subsequent PSA progression was defined as ≥ 25% increase and ≥ 2 ng/mL above nadir, confirmed by a second measurement at least 3 weeks later. For patients with no observed decline in PSA while on study, progression was defined as ≥ 25% increase and ≥ 2 ng/mL above baseline after at least 12 weeks on study. Similarly, a decline in CA 19-9 of > 50% from baseline was considered a response; such response was considered confirmed if the decline was observed twice at least 4 weeks apart.

Twenty-two patients were enrolled at MD Anderson's Department of Investigational Cancer Therapeutics between June 2017 and March 2019. Three patients were enrolled at 400 mg/m² evofosfamide, 6 at 480 mg/m², 6 at 560 mg/m², and 6 at 640 mg/m². All patients received 3 mg/kg ipilimumab as described in Patients and Methods. One patient enrolled at 560 mg/m² withdrew consent before initiating treatment. Thus, 21 patients were included in safety analysis.

Baseline patient characteristics are shown in Table 1. Median age was 67 years (range, 47–78), with patients being 81.0% male (17/21) and 19.0% (4/21) female. ECOG performance status was 1 in 81.0% (17/21) and 0 in 19.0% (4/21). Eleven patients had metastatic castration-resistant prostate cancer, 7 had pancreatic cancer, 2 had checkpoint blockade refractory melanoma, and 1 had checkpoint blockade refractory HPV⁻ head and neck cancer. Fifteen patients (71.4%) had received ≥ 3 prior lines of systemic therapy, and 4 (19.0%) had received prior immune checkpoint inhibitor therapy.

Treatment was discontinued for all 3 patients enrolled at 400 mg/m² evofosfamide due to disease progression. Similarly, treatment was discontinued for 4 patients at 480 mg/m² due to disease progression (66.7%), while the other 2 withdrew from the study (33.3%). At 560 mg/m² (n = 6), treatment was discontinued for 5 patients due to disease progression (83.3%), while ipilimumab was discontinued for 1 patient due to toxicity (16.7%). At 640 mg/m² (n = 6), treatment was discontinued for 4 patients (66.7%) due to disease progression and ipilimumab was discontinued for 1 patient due to toxicity (16.7%). One patient remained on treatment at 640 mg/m² (16.7%) at the time of analysis. As no dose-limiting toxicity was observed, 640 mg/m² was considered the MTD, although 560 mg/m² is the recommended phase II dose based on response as detailed below.

A total of 193 treatment-related adverse events occurred in 21 patients, of which 162 (83.9%) were grade 1 or 2 events and 31 (16.1%) were grade ≥ 3 events (Table 2). There were 29 serious adverse events. The most common any grade treatment-related adverse events were hematologic (anemia 100%, thrombocytopenia 71.4%, leukopenia 61.9%, lymphopenia 57.1%, neutropenia 33.3%), rash (81.0%), elevated liver enzymes (ALT 90.5%, AST 71.4%, alkaline phosphatase 23.8%), fever (47.6%), nausea (23.8%), vomiting (38.1%), diarrhea (19.0%), constipation (19.0%), and colitis (14.3%). There were six treatment-related grade ≥ 3 adverse events at 400 mg/m², eight at 480 mg/m², 10 at 560 mg/m², and seven at 640 mg/m². The most frequent treatment-related grade 3 adverse events included elevated ALT (33.3%), elevated AST (19.0%), anemia, thrombocytopenia, lymphopenia (14.3% each), rash, hypotension, and neutropenia (9.5% each). Three patients required transfusions of packed red blood cells, but none required platelet transfusions. One patient at 640 mg/m² had delayed grade 3 thrombocytopenia following five cycles of treatment, an episode that lasted more than 3 weeks. One patient had grade 3 ipilimumab-related lymphocytic myocarditis and grade 3 ipilimumab-related transaminitis. The former was treated with intravenous immunoglobulin and the latter required treatment with mycophenolate mofetil, resulting in complete resolution of both. One patient experienced grade 3 ipilimumab-related transaminitis and grade 3 ipilimumab- or evofosfamide-related pneumonitis, both of which resolved with steroids. One patient had grade 3 ipilimumab-related colitis that resolved with infliximab. No grade 5 adverse events were reported. Overall, 4 patients (19.0%; 1 at 480 mg/m², 2 at 560 mg/m², and 1 at 640 mg/m²) required evofosfamide deescalation due to toxicity. One patient at 640 mg/m² discontinued ipilimumab due to toxicity. None of the adverse events met the criteria for dose-limiting toxicity.

All 21 patients were evaluable at the time of analysis, although only 18 had measurable disease at baseline. Two of these 18 patients achieved confirmed partial response (11.1%), one of whom had HPV⁻ head and neck cancer and the other had metastatic castration-resistant prostate cancer (Fig. 1A). One additional patient with metastatic castration-resistant prostate cancer had an unconfirmed partial response (5.6%); thus, the objective response rate was 16.7%. Twelve patients (66.7%) had stable disease (5 with metastatic castration-resistant prostate cancer, 5 with pancreatic cancer, and 2 with checkpoint blockade refractory melanoma). Overall disease control rate (complete response + partial response + stable disease) was 83.3% (15/18). Five patients (2 with pancreatic cancer, 2 with metastatic castration-resistant prostate cancer, and 1 with HPV⁻ head and neck cancer) were retreated. Two patients who achieved partial response in the first course of treatment, had a repeat partial response on retreatment (1 with metastatic castration-resistant prostate cancer and 1 with HPV⁻ head and neck cancer), while all others continued to present stable disease (1 with metastatic castration-resistant prostate cancer and 2 with pancreatic cancer). The median duration of response for patients with partial response was 85 days (range, 78–183 days), while the median duration of disease control (complete response + partial response + stable disease) was 64 days (range, 2–244 days; Fig. 1B and C).

Serial tumor markers were also assessed for 18 of 21 evaluable patients (11 with metastatic castration-resistant prostate cancer and 7 with pancreatic cancer). One patient with pancreatic cancer had a confirmed CA 19-9 response (62% decrease from baseline) as best response. One patient with metastatic castration-resistant prostate cancer had a confirmed PSA response (100% decrease from baseline) as best response. The 3 patients with non-measurable disease at baseline had metastatic castration-resistant prostate cancer, of whom 1 had confirmed PSA progression (48% increase from baseline) as best response, and 2 had stable PSA as best response (Fig. 1D).

At 400 mg/m² (n = 3), all 3 patients achieved stable disease as best response. At 480 mg/m² (n = 6), 1 patient with prostate cancer achieved an unconfirmed partial response (16.7%) as best response, 4 achieved stable disease (66.7%), and 1 progressed (16.7%). Best outcomes were achieved at 560 mg/m² (n = 6), with 2 patients (1 with prostate cancer and 1 with HPV⁻ head and neck cancer) achieving confirmed partial response (33.3%), 2 achieving stable disease (33.3%), and 2 experiencing progressive disease (33.3%) as best response. At 640 mg/m² (n = 6), 3 patients achieved stable disease (50.0%), 2 achieved stable PSA (33.3%), and 1 experienced PSA progression (16.7%) as best response.

Pharmacokinetic studies were performed at 560 and 640 mg/m² evofosfamide. No differences in mean plasma concentration, T_max, C_max, and AUC_last were observed for evofosfamide and its active metabolite Br-IPM (Supplementary Fig. S1). These studies demonstrated a plateau of active evofosfamide exposure at 560 mg/m² (C_max = 0.18 μg/mL) that was higher than the active range defined in the successful phase II MAESTRO study (C_max = 0.13 μg/mL), and which was never achieved in non-Japanese patients at phase III (C_max = 0.08 μg/mL; ref. 13).

For all pharmacodynamic analyses, patients were classified into four groups (partial response, stable disease ≥ 60 days, stable disease < 60 days, and progressive disease), of which the first two groups were considered “responders” and the latter two “nonresponders” based on molecular clustering. In addition, we note that stable disease lasting less than 60 days reflects a lack of durable response to therapy and can include patients who were slowly but steadily progressing. In contrast, stable disease lasting more than 60 days indicates continuing response to therapy beyond the treatment period. These response classifications were all based on irRECIST excepting 2 patients with prostate cancer whose progressive disease was diagnosed on the basis of PSA progression. Both patients are included in the analysis of PBMCs and RNA, and 1 is included in the flow cytometry analysis of paired biopsies.

PBMCs collected pretreatment, on-treatment [generally after four doses of evofosfamide (range, 2–6) and two doses of ipilimumab (range, 1–3)], and post-treatment were available for 5 nonresponding patients (3 progressive disease and 2 stable disease < 60 days) and 7 responding patients (3 partial response and 4 stable disease ≥ 60 days). Statistically significant accumulation of proliferative CD8⁺ and CD4⁺ effector T cells, as measured by Ki67 expression, was observed in responders relative to nonresponders (Fig. 2A; Supplementary Fig. S2). Nonresponders also tended to accumulate exhausted peripheral CD8⁺ T cells (PD-1⁺ LAG3⁺; Fig. 2B). These peripheral T-cell responses are plotted across therapy for individual patients in Supplementary Fig. S3. Levels of the immunosuppressive enzyme arginase also trended higher in circulating myeloid-derived suppressor cells at every timepoint examined (Fig. 2C). Similarly, we found significantly higher pretreatment levels of arginase in circulating dendritic cells (CD11c⁺b⁻) in nonresponders compared with responders (Fig. 2D). A similar trend was observed in monocytic dendritic cells (CD11c⁺b⁺). Mirroring arginase, we found a higher fraction of PD-L1 expression pretreatment in dendritic cells from nonresponders versus responders (Fig. 2E). Taken together, the data suggest the potential of a suppressive phenotype in the circulating myeloid repertoire as a potential pretreatment biomarker, and of peripheral effector T-cell proliferation as a pharmacodynamic biomarker of response to cotreatment with evofosfamide and ipilimumab.

We obtained pretreatment tumor biopsies from 16 patients, 9 of whom also consented to on-treatment biopsies at week 7. These biopsies were analyzed by 20-parameter flow cytometry to evaluate the effect of evofosfamide and ipilimumab cotreatment on hypoxia and on the immune composition of the tumor microenvironment (Supplementary Fig. S4). Hypoxia was assessed using Hypoxyprobe, an antibody to the pimonidazole adduct and conjugated to the fluorescent dye ATTO 594. Consistent with restoration of antitumor T-cell immunity, CD8⁺ cytotoxic T cells and CD4⁺ effector T cells tended to accumulate in responders (Fig. 3A). Greater densities of T cells and antigen-presenting dendritic cells in hypoxic regions also correlated with response; statistically significant differences in CD4⁺ effector T cells were observed even with the small number of samples available for analysis (Fig. 3B). In addition, responders tended to have reduced proliferation of immunosuppressive CD33⁺CD14⁻CD15⁻ tumor-associated macrophages (Fig. 3C). Responses for all of these populations are plotted for individual patients in Supplementary Fig. S5A. Both CD4 and CD8 hypoxia-exposed T cells tended to exhibit higher PD-1 expression compared with their nonhypoxia resident counterparts; however, this difference only reached statistical significance for pretreatment CD8 T cells (Fig. 3D; Supplementary Fig. S5B). We also observed a nonsignificant trend toward lower PD-1 expression by hypoxia-exposed CD8 T cells (PD-1 mean MFI of 6,565 vs. 3,139) following treatment with evofosfamide and ipilimumab. We hope to investigate this phenomenon further in a larger study. Increased PD-L1 expression correlated with better response to therapy, consistent with the generation of an inflammatory tumor microenvironment, as shown for a representative patient who achieved partial response (Fig. 3E).

RNA sequencing data showed that patients who eventually achieved partial response have distinct tumor gene expression profiles pretreatment from patients who eventually experienced progressive disease and stable disease < 60 days (Fig. 4A). These differences were also apparent in comparing responders overall with nonresponders (Fig. 4B; Supplementary Table S2). Significant differences were also observed between responders and nonresponders on treatment (Fig. 4C; Supplementary Table S2), although gene expression was altered in response to therapy in both responders and nonresponders (Supplementary Fig. S6). GSEA against the Hallmark gene set collection revealed that patients who eventually responded to treatment show evidence of preexisting adaptive and innate antitumor immunity (Fig. 4D). Hence, restoring T cell access to and survival in formerly hypoxic areas of the tumor may be sufficient to sensitize cancers in this setting to T-cell checkpoint blockade. On the other hand, patients who progress on treatment show a hypermetabolic phenotype at baseline. In this environment, relieving hypoxia still leaves T cells at a tremendous metabolic disadvantage, likely explaining the lack of response to CTLA-4 blockade. Finally, enrichment analysis against the transcription factor targets collection showed that targets of HIF1α, the cardinal hypoxia-responsive transcription factor, accumulated in nonresponders on treatment, but diminished in responders (Fig. 4E).

Cotreatment with evofosfamide and ipilimumab appears to be both safe and well tolerated, with most drug-related adverse events being grade 1–2. Common grade ≥ 3 adverse events included hematologic toxicities that were not dose limiting, that were comparable across evofosfamide doses, and that were similar to those observed with single-agent evofosfamide (24). Grade 3 ALT/AST elevation and grade ≥ 3 rash were observed more frequently than previously reported with either agent alone, suggesting a possible additive effect (18, 24). Other immune-related grade 3 adverse events were comparable with that previously reported with single-agent ipilimumab (18). On the other hand, mucositis was significantly less frequent (9.5%) compared with single-agent evofosfamide and was exclusively all grade 1–2. Of note, adverse events were well managed with supportive measures and dose interruption, except in 1 patient with immune hepatitis who required immunosuppression with mycophenolate mofetil. Collectively, safety data indicate that toxicities from combining evofosfamide and ipilimumab were comparable with that of either agent as monotherapy with the exception of milder mucositis, elevated but not dose-limiting hepatotoxicity, more frequent skin rashes, and, in one instance, delayed/prolonged thrombocytopenia. The safety profile is also comparable with that of the combination of evofosfamide with gemcitabine, as assessed in a randomized phase III study (MAESTRO) of 693 patients with advanced pancreatic cancer (12).

We observed multiple partial responses in patients with hormone- and chemotherapy-refractory prostate cancer and HPV⁻ head and neck cancer, as well as stable disease in pancreatic cancer and checkpoint blockade refractory melanoma, for whom treatment options are typically scarce. While ipilimumab was previously demonstrated to be of limited efficacy against metastatic castration-resistant prostate cancer (25, 26), it is important to note that patients in these prior studies were chemotherapy-naïve and, in most cases, were treated with 10 mg/kg, rather than 3 mg/kg, ipilimumab. Given these differences, the lack of similarity between the transcriptomic response we observed and the classic transcriptomic response to ipilimumab in melanoma (27), and the transcriptomic separation between responders and nonresponders and between pre- and on-treatment biopsies, we assert that these responses are the result of therapeutic cooperativity between evofosfamide and ipilimumab. Of note, we determined the recommended phase II evofosfamide dose when combined with 3 mg/kg ipilimumab to be 560 mg/m², at which two confirmed partial responses were observed, even though the MTD appeared to be 640 mg/m², at which neither response nor dose-limiting toxicity was observed.

Furthermore, the observation of excellent disease control among patients who were retreated indicates that treatment with a limited number of cycles, followed by retreatment if necessary, may be feasible with this combination, thereby avoiding prolonged and undue toxicities. However, the duration of response and duration of disease control were relatively short for many patients, highlighting the need for correlative studies and better combinatorial strategies, including, potentially, regimens that block PD-1, as evofosfamide synergized with anti-CTLA-4 and anti-PD-1 antibodies in multiple preclinical models of murine prostate cancer (4), and as PD-L1 accumulated in responders. Indeed, a follow-up study to combine evofosfamide with anti-CTLA-4 and anti-PD-1 is planned.

Although prior trials demonstrated that levels of circulating myeloid-derived suppressor cells can be predictive of subsequent responses to ipilimumab, the phenotypic traits of these cells (e.g., arginase expression) were not examined (28). We have now observed that CD8⁺ and CD4⁺ effector T cells significantly accumulate in responders versus nonresponders while on treatment, even in a small cohort. This not only suggests the value of monitoring peripheral T-cell phenotypes as a biomarker of response to the combination of evofosfamide and ipilimumab, but also shows that this regimen elicits a peripherally measurable mobilization of circulating effector T cells in responding patients.

Comprehensive flow cytometry of pre- and on-treatment tumor biopsies also revealed a T cell–inflamed milieu in responders. T-cell proliferation, activation, and survival under hypoxia also improved or trended toward improvement, accompanied by concomitant reduction in myeloid-derived suppressor cells and tumor-associated macrophages. These observations suggest that evofosfamide cooperates with ipilimumab to remodel the tumor microenvironment of traditionally immune-refractory cancers to one that is conductive to T-cell persistence and function. The data also reinforce the notion that specific biomarkers can be used to assess the efficacy of the combination.

Gene expression analysis indicated that a hypermetabolic phenotype at baseline often precludes response. Similarly, we found that a similar hypermetabolic phenotype featuring elevated oxidative phosphorylation and heightened glycolysis was highly enriched when a checkpoint blockade–sensitive melanoma cell line was multiply passaged in vivo in mice receiving immune checkpoint inhibitors until complete immunotherapy resistance was acquired (29). Thus, we speculate that hypoxia reduction alone may be insufficient to relieve T-cell inhibition due metabolic starvation, as happens in the context of a hypermetabolic tumor.

In summary, the combination of evofosfamide and ipilimumab was well tolerated, with no new or unexpected safety signals. Antitumor activity was observed not only in heavily pretreated patients, but also in checkpoint blockade refractory tumors, although the response duration was limited. Nevertheless, biomarker-driven patient selection and/or addition of PD-1 blockade may further augment and extend the clinical response.

S. Piha-Paul reports other support from AbbVie, Inc., ABM Therapeutics, Inc., Acepodia, Inc., Alkermes, Aminex Therapeutics, Amphivena Therapeutics, Inc., BioMarin Pharmaceutical, Inc., Boehringer Ingelheim, Bristol Myers Squibb, Cerulean Pharma, Inc., Chugai Pharmaceutical Co., Ltd., Curis, Inc., Daiichi Sankyo, Inc., Eli lilly, ENB Therapeutics, Five Prime Therapeutics, Gene Quantum, Genmab A/S, GlaxoSmithKline, Helix BioPharma Corp., Incyte Corp., Jacobio Pharmaceuticals Co., Ltd., Medimmune, LLC., Medivation, Inc., Merck Sharp & Dohme Corp., Novartis Pharmaceuticals, Pieris Pharmaceuticals, Inc., Pfizer, Principia Biopharma, Inc., Puma Biotechnology, Inc., Rapt Therapeutics, Inc., Seattle Genetics, Silverback Therapeutics, Taiho Oncology, Tesaro, Inc., TransThera Bio, and NCI/NIH P30CA016672 outside the submitted work. J. Rodon reports grants and personal fees from Novartis, Spectrum Pharmaceuticals, Inc., and Pfizer; personal fees from Eli Lilly, Orion Pharmaceuticals, Servier Pharmaceuticals, Peptomyc, Roche Pharmaceuticals, Ellipses Pharma, Certera, NovellusDX, and IONCTURA SA; personal fees and other support from Merck Sharp & Dohme, Kelun Pharmaceuticals/Klus Pharma, and Molecular Partners; grants, personal fees, and other support from Bayer; grants from Blueprint Pharmaceuticals, Tocagen, Symphogen, BioAlta, GenMab, CytomX, Kelun-Biotech, Ipsen, and Takeda-Millenium; grants and other support from GlaxoSmithKline; and other support from ESMO, Department of Defense, Louisiana State University, Huntsman Cancer Institute, Cancer Core Europe, Karolinska Cancer Institute, King Abdullah International Medical Research Center, WIN Consortium, Janssen, European Journal of Cancer, VHIO/Ministero De Empleo Y Seguridad Social, Chinese University of Hong Kong, SOLTI, and Elsevier outside the submitted work. T.A. Yap reports grants from Artios, Constellation, Cyteir, Eli Lilly, Forbius, GlaxoSmithKline, Genentech, ImmuneSensor, Ipsen, Jounce, Karyopharm, Kyowa, Novartis, Ribon Therapeutics, Regeneron, Sanofi, Scholar Rock, Tesaro, and Vertex Pharmaceuticals; grants and personal fees from AstraZeneca, Bayer, Clovis, EMD Serono, F-Star, Merck, Pfizer, Repare, and Seattle Genetics; and personal fees from Almac, Aduro, Atrin, Axiom, Bristol Myers Squibb, Calithera, Cybrexa, Guidepoint, Ignyta, I-Mab, Janssen, Roche, Rubius, Schrodinger, Varian, and Zai Labs outside the submitted work. V. Subbiah reports grants from LOXO Oncology/Eli Lilly during the conduct of the study. V. Subbiah also reports consulting or advisory role with MedImmune, Helsinn Therapeutics, Loxo, R-Pharma-US, and QED Pharma; research funding from Novartis (Inst), GlaxoSmithKline (Inst), NanoCarrier (Inst), Northwest Biotherapeutics (Inst), Roche (Inst), Berg Pharma (Inst), Bayer (Inst), Incyte (Inst), Fujifilm (Inst), PharmaMar (Inst), D3 Oncology Solutions (Inst), Pfizer (Inst), Amgen (Inst), AbbVie (Inst), Multivir (Inst), Blueprint Medicines (Inst), Loxo (Inst), Vegenics (Inst), Takeda (Inst), Alfasigma (Inst), Agensys (Inst), Idera (Inst), Boston Biomedical (Inst), Inhibrx (Inst), Exelixis (Inst), Amgen (Inst), and Turningpoint Therapeutics (Inst); travel, accommodations, and expenses support from PharmaMar, Bayer, Novartis, and Helsinn Therapeutics; and other relationship support from Medscape. J.P. Higgins reports personal fees from Molecular Templates during the conduct of the study, as well as personal fees from Molecular Templates outside the submitted work. E.T. Williams reports other support from Molecular Templates outside the submitted work, as well as employment with Molecular Templates. F. Meric-Bernstam reports personal fees from Aduro BioTech, Alkermes, AstraZeneca, DebioPharm, eFFECTOR Therapeutics, F. Hoffman-La Roche, Ltd., Genentech, IBM Watson, Jackson Laboratory, Kolon Life Sciences, OrgiMed, PACT Pharma, Parexl International, Pfizer Inc., Samsung Bioepis, Seattle Genetics, Tyra Biosciences, Xencor, Zymeworks, Immunomedics, Inflection Biosciences, Mersana Therapeutics, Puma Biotechnology, Seattle Genetics, Silverback Therapeutics, Spectrum Pharmaceuticals, Zentalis, Chugai Biopharmaceuticals, Mayo Clinic, and Rutgers Cancer Institute of New Jersey; grants from Aileron Therapeutics, AstraZeneca, Bayer Healthcare Pharmaceutical, Calithera Biosciences, Curis Inc., CytomX Therapeutics, Daiichi Sanko Co. Ltd., Debiopharm International, eFFECTOR Therapeutics, Genentech, Guardant, Millennium Pharmaceuticals, Novartis, Puma Biotechnology, and Taiho Pharmaceutical; and nonfinancial support from Beth Israel Deaconess Medical Center outside the submitted work. M.A. Curran reports personal fees from ImmunoGenesis, Inc. and Agenus, Inc. during the conduct of the study. M.A. Curran also reports personal fees from ImmunoGenesis, Inc., Agenus, Inc., Alligator Bioscience, Inc., ImmunOs, Inc., Oncoresponse, Inc., Pieris, Inc., Nurix, Inc., Aptevo, Inc., Servier, Inc., Kineta, Inc., Salarius, Inc., Kineta, Inc., Xencor, Inc., and Mereo, Inc., as well as grants and personal fees from ImmunoMet, Inc. outside the submitted work; in addition, M.A. Curran has a patent for Methods and Composition for Localized Secretion of Anti-CTLA-4 Antibodies issued to multiple licensees. D.S. Hong reports grants from NIH and Molecular Templates during the conduct of the study. D.S. Hong also reports grants from Abbvie, Adlai Nortye, AstraZeneca, Bristol Myers Squibb, Daichii Sankyo, Eisai, Fate Therapeutics, Ignyta, Infinity, Kite, Kyowa, Lilly, Merck, Medimmune, Mirati Therapeutics, Mologen, Novartis, and Turning Point Therapeutics; grants and personal fees from Adaptimmune, Amgen, Genentech, Pfizer, Seattle Genetics, and Takeda; grants, personal fees and nonfinancial support from Bayer; grants and nonfinancial support from Genmab, Loxo Oncology, and miRNA; personal fees and nonfinancial support from AACR, ASCO, and SITC; and personal fees from Alpha Insights, Acuta, Axiom, Baxter, GLG, Group H, Guidepoint, Infinity, Janssen, Merrimack, Medscape, Prime Oncology, Trieza Therapeutics, WebMD, COG, and EcoR1 outside the submitted work. D.S. Hong also reports Molecular Match (advisor), OncoResponse (founder), and Presagia Inc. (founder and advisor). No disclosures were reported by the other authors.

A. Hegde: Conceptualization, methodology, writing–original draft, writing–review and editing. P. Jayaprakash: Investigation, methodology, writing–review and editing. C.A. Couillault: Investigation, methodology. S. Piha-Paul: Investigation, writing–original draft. D. Karp: Investigation. J. Rodon: Investigation. S. Pant: Investigation. S. Fu: Investigation. E.E. Dumbrava: Investigation. T.A. Yap: Investigation. V. Subbiah: Investigation. P. Bhosale: Investigation. C. Coarfa: Software, investigation. J.P. Higgins: Investigation. E.T. Williams: Investigation, writing–original draft. T.F. Wilson: Investigation, project administration. J. Lim: Investigation. F. Meric-Bernstam: Investigation, writing–original draft. E. Sumner: Investigation. H. Zain: Investigation. D. Nguyen: Data curation. L.M. Nguyen: Supervision, investigation, project administration. K. Rajapakshe: Software, visualization, methodology, writing–original draft. M.A. Curran: Conceptualization, supervision, methodology, writing–original draft, writing–review and editing. D.S. Hong: Conceptualization, supervision, funding acquisition, investigation, writing–original draft, writing–review and editing.

We thank Molecular Templates, Inc. and Threshold, Inc., particularly Charles Hart, Tillman Pearce, and Barry Selick, for supporting this trial. We thank Amadeo Biter for editorial support.

This trial and associated correlative studies were supported by Molecular Templates, Inc.

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.