Hypoxia is a common characteristic of many tumor microenvironments, and it has been shown to promote suppression of antitumor immunity. Despite strong biological rationale, longitudinal correlation of hypoxia and response to immunotherapy has not been investigated.
In this study, we probed the tumor and its surrounding microenvironment with 18F-FMISO PET imaging to noninvasively quantify tumor hypoxia in vivo prior to and during PD-1 and CTLA-4 checkpoint blockade in preclinical models of breast and colon cancer.
Longitudinal imaging identified hypoxia as an early predictive biomarker of therapeutic response (prior to anatomic changes in tumor volume) with a decreasing standard uptake value (SUV) ratio in tumors that effectively respond to therapy. PET signal correlated with ex vivo markers of tumor immune response including cytokines (IFNγ, GZMB, and TNF), damage-associated molecular pattern receptors (TLR2/4), and immune cell populations (macrophages, dendritic cells, and cytotoxic T cells). Responding tumors were marked by increased inflammation that were spatially distinct from hypoxic regions, providing a mechanistic understanding of the immune signaling pathways activated. To exploit image-guided combination therapy, hypoxia signal from PET imaging was used to guide the addition of a hypoxia targeted treatment to nonresponsive tumors, which ultimately provided therapeutic synergy and rescued response as determined by longitudinal changes in tumor volume.
The results generated from this work provide an immediately translatable paradigm for measuring and targeting hypoxia to increase response to immune checkpoint therapy and using hypoxia imaging to guide combinatory therapies.
Hypoxia is a well-known component of many tumors and acts in an immunosuppressive manner through multiple distinct pathways. 18F-FMISO PET imaging has been demonstrated to accurately quantify hypoxia in a clinical setting, although it has not been investigated for immunotherapy. Our work demonstrates that effective immunotherapy prevents tumor hypoxia and that uptake of 18F-FMISO was predictive of subsequent changes in anatomical tumor size. We further demonstrate that PET imaging of hypoxia is correlated with phenotypic characteristics of inflammation through transcriptomics, immune cell spatial correlation, and direct measurement of secreted inflammatory proteins. The upregulation of damage associated molecular pattern signaling in responding tumors warranted investigation of the addition of the hypoxia targeted prodrug to nonresponding tumors to enhance checkpoint blockade efficacy. The addition of evofosfamide improved tumor oxygenation and response, providing rationale for immediate clinical investigation of both 18F-FMISO PET imaging and combination evofosfamide therapy for improved cancer immunotherapy.
The use of immunotherapy has substantially increased in use in both clinical trials and as a standard-of-care treatment for malignant tumors in both the primary and metastatic setting (1, 2). Unfortunately, it remains effective in only a minority of patients due to a combination of intrinsic tumor and microenvironment factors that result in a coordinated resistance to treatment (3). Currently, biomarker studies focus on individual proteins that correlate with response; however, the physical properties of the tumor may have an equal or more profound impact (4). Genetic and histologic techniques that are commonly used in precision medicine are limited in accurately defining physical properties of a tumor, as they rely on correlation to surrogate proteins or genes and provide data from only a snapshot in time. Imaging techniques that permit non-invasive in vivo measurement to longitudinally quantify aspects of the tumor microenvironment and correlate them to immune response have the potential to unlock personalized approaches to cancer immunotherapy.
One property of the tumor microenvironment that has been shown to have a negative impact on immune response is hypoxia. It is well known that hypoxia increases as tumors progress (5–7). Tumors benefit from hypoxia, as it dampens the response to radiation and chemotherapy and hinders antitumor immunity (6–9). In preclinical studies, hypoxia has also been shown to decrease the effectiveness of checkpoint blockade therapies, increasing PD-L1 expression and secretion of CTLA-4 by tumor stroma (10). Because of the correlation between hypoxia and treatment futility, previous studies have attempted to monitor hypoxia using invasive techniques such as biopsy or intratumoral polarographic oxygenation (9). However, given the limitations of invasive hypoxia measurements, the ability to quantitatively measure hypoxia noninvasively in tumors is needed to improve immunotherapy treatment outcomes.
Molecular imaging techniques such as positron emission tomography (PET) provide the ability to noninvasively visualize and quantify biomarkers associated with the interactions that occur between tumor cells and the immune system. Numerous PET agents that target either immune cell lineage markers or functional proteins associated with immune function have been explored (11). These approaches provide non-invasive quantification of the presence and phenotype of specific immune cell populations but lack the ability to directly measure immunosuppressive physical characteristics of the tumor. Conversely, hypoxia imaging is a well-established technique that has been previously investigated clinically with a number of treatment modalities, but it has not been investigated in the setting of immunotherapy. The PET imaging agent, fluoromisonidazole (18F-FMISO), quantitatively measures hypoxia levels intratumorally, as it is retained by irreversible binding to the thiol-rich metabolic proteins at rates that are inversely proportional to oxygen concentration (12). 18F-FMISO has been extensively used in clinical trials in glioblastoma, but its widespread incorporation has been hindered by a lack of sensitivity to measure less hypoxic regions of tumors (11). However, this characteristic may be beneficial for immunotherapy, where severe hypoxia is more likely to be associated with a suppressive immune microenvironment.
In addition to utilizing hypoxia as a predictive immunotherapy biomarker, several hypoxia treatments have been previously explored, including hypoxia-activated prodrugs (13). Of particular interest, evofosfamide is a chemotherapy that is reduced at the nitroimidazole site of the prodrug by intracellular reductases when exposed to hypoxic conditions, ultimately inducing hypoxic cells to undergo apoptosis (13, 14). Evofosfamide treatment alone or in combination with additional chemotherapies in phase III clinical trials failed to show a statistically significant improvement in overall survival in solid tumors (14). However, a recent phase I clinical trial combining evofosfamide and immunotherapy (ipilimumab) achieved partial response or stable disease in 15 of 22 patients, but drug-related toxicities were common (15). These data support an imaging-guided paradigm for precisely selecting patients most likely to benefit from combination therapy, to reduce unnecessary toxicity, improve efficacy, and provide quantitative biomarkers to guide the schedule of treatment.
The overall goal of this study was to noninvasively stratify tumors based on a clinically relevant hypoxia imaging approach to monitor immunotherapy. Here, we have shown in preclinical models that hypoxia and immunotherapeutic efficacy are inversely correlated, and imaging-based stratification can provide a valuable technique to enable a personalized approach to treatment. This provides a potential avenue for identifying patients who will benefit from immunotherapy prior to or shortly after treatment initiation using tumor microenvironment targeted imaging agents.
Materials and Methods
MC38 murine colorectal cancer cells were obtained from Kerafast (September 2019, catalog number: ENH204-FP, RRID:CVCL_B288) and E0771 murine breast cancer cells were obtained from CH3 BioSystems (March 2020, catalog number: 94A001, RRID:CVCL_GR23). MC38 is an immunogenic, grade III adenocarcinoma of colorectal cancer, that is characterized by microsatellite instability (16). E0771 tumors are derived from spontaneous breast cancer, express known immunomodulatory molecules, including PD-L1, and are moderately responsive to immunotherapy (17). Cells were cultured in Dulbecco's Modified Eagle Medium (DMEM) with 10% fetal bovine serum (FBS), 2 mmol/L L-glutamine, and 1 mmol/L sodium pyruvate in a humidified incubator with 5% CO2 at 37°C. Cells were maintained at passage numbers less than 20. All experiments were done with cells acquired and frozen within 1 month to maintain the phenotype of the cell line. MC38 and E0771 cells tested negative for pathogens, including Mycoplasma, via real-time polymerase chain reaction (real-time PCR) by Charles River Research Animal Diagnostic Services (CR RADS; MC38 – August 2020, E0771–May 2020) and both cell lines were received directly from the authenticated source (MC38 – Kerafast, E0771 – CH3 BioSystems). All cell lines were used within 6 months of testing.
All animal procedures and housing were maintained in accordance with the guidelines provided by Institutional Animal Care and Use Committee of The University of Alabama at Birmingham. For both cell lines, approximately 5 × 105 cells were diluted in 40% Matrigel and 60% serum free DMEM and subcutaneously injected into the upper right shoulder of 6- to 12-week-old C57BL/6 mice (Charles River Laboratories, catalog number: 027, RRID:IMSR_CRL:027). Tumors were grown until they reached a volume of a 100 mm3 (7–10 days postinoculation). All tumors were measured individually on day 0 and allocated to divide tumor volumes equally across groups. The mean standard tumor volume on day 0 is 179 ± 57.1 (N = 13) for normoxic tumors and 256 ± 124 with (N = 8) for hypoxic tumors for these groups.
Mice were treated via intraperitoneal injection on days 0 (7–10 days postinoculation), 3, and 6 with saline or 200 μg anti–PD-1 + 100 μg anti–CTLA-4 combination therapy. Anti-mouse PD-1 (clone RPM1–14, Bio X Cell Cat. No. BE0146, RRID:AB_10949053) and anti-mouse CTLA-4 (clone 9H10, Bio X Cell Cat. No. BE0131, RRID:AB_10950184) were obtained from Bio X Cell. Mice were treated via intraperitoneal injection on days 6 to 10 following tumor inoculation with saline or 50 mg/kg evofosfamide (TH-302 Selleckchem Catalog No. S2757). Immunotherapy and Evofosfamide doses were chosen based on concentrations that reflected human clinical studies and tumor volume was assessed every other day for changes in longitudinal tumor viability.
18F-FMISO was prepared by The University of Alabama at Birmingham's Cyclotron Facility on a GE FastLab2 or a Synthra RNplus synthesizer according to current literature (18–20). Chemical and radiochemical purity of the final product is confirmed using HPLC and GC and radionuclidic purity is confirmed using an HPGe detector. Overall, nondecay corrected yields are approximately 26%. 18F-FMISO was obtained with an average radiochemical purity of >95% and a specific activity of >2,000 Ci/mmol.
18F-FMISO PET imaging
Animals (MC38 N = 31; E0771 N = 21) were imaged with 18F-FMISO-PET on days 0 and 5. At each imaging time point, mice were injected with approximately 150 μCi of 18F-FMISO (143.5 μCi ± 6.7, mean ± SD) via retro-orbital injection and then 80 minutes later were transferred under anesthesia for preclinical small animal PET/CT imaging (Sofie Biosciences) to quantify uptake of 18F-FMISO (21). Anesthesia was maintained at a rate of 2% isoflurane in air. Animal body temperature was maintained at 37°C. Static PET images were acquired for 20 minutes, and CT images were acquired for 5 minutes. Uptake of 18F-FMISO in the tumor and muscle were quantified by drawing regions of interest (ROI) SUV using the CT anatomic guidance. The radiotracer dose was decay-corrected to the time of injection and mean standardized uptake values were calculated as activity concentration/(injected dose/body weight) and tumor to background (TBR) was determined as tumor SUV/muscle SUV, as has been described previously (22). Mice were monitored for an additional 15 days after final antibody treatment for long term changes in tumor viability.
MC38 (N = 16, N = 8 treated, and N = 8 control) and E0771 (N = 15, N = 7 treated, and N = 8 control) tumors were extracted from mice on days 0 and 5 and flash frozen in OCT in liquid nitrogen and stored in −80°C. Frozen tumors were sliced and adhered to microscopy slides by UAB Pathology Core and stored at −80°C. Tumors were allowed to thaw and soaked in PBS for 10 minutes and washed twice in 0.05% Tween20 PBS for 5 minutes each. Normal goat serum (Abcam Cat. No. ab7481, RRID:AB_2716553) in TBS with BSA was used to block tissues. Primary antibodies were prepared according to the manufacturer's recommendations in PBS with BSA and were applied and left in 4°C overnight. Primary antibodies used include CD8 (Thermo Fisher Scientific Cat. No. MA5–18153, RRID:AB_2539527), GZMB (Biorbyt Cat. No. orb10738-CF647, RRID:AB_2893361), Texas Red Tomato Lectin preinjected 1 minute before tumor extraction (Vector Laboratories Cat. No. TL-1176, RRID:AB_2336563), pimonidazole preinjected 1 hour prior to tumor extraction (Hypoxyprobe Cat. No. HP1–1000, RRID:AB_2811309), and secondary Hypoxyprobe Red PE (Hypoxyprobe HP11–100kit, RRID:AB_2893362), TLR4 (Novus Biologicals NB100–56723AF647, RRID:AB_2893363), PD-1 (Thermo Fisher Proteintech Cat. No. CL647–66220, RRID:AB_2883732), F4/80 (Abcam Cat. No. ab204467, RRID:AB_2810932), TNFa (Abcam ab237353, RRID:AB_2893364), Arginase (Thermo Fisher Proteintech Cat. No. CL488–66129, RRID:AB_2883268), and CD11c (Thermo Fisher Proteintech Cat. No. CL488–60258, RRID:AB_2883128). DAPI was received from BioLegend (4′,6-Diamidino-2-Phenylindole, Dilactate, Cat. No. 422801) and ProLong Gold Antifade Mountant (Thermo Fisher P36930, RRID:SCR_015961) was used. Tumor sections were imaged on a Leica Stellaris 5/DMi8 platform with LAS X software using a Leica PlanApo CS2 20x/0.75 NA oil-immersion objective. Laser lines 405, 488, 514, and 647 were detected using four Leica Power HyD S tunable hybrid detectors. Further negligible image adjustments were made to minimize background and LUT assignments were completed in NIH ImageJ (RRID:SCR_003070; ref. 23).
Cytokine analysis − ELISA
ELISA assays were used to analyze MC38 colorectal tumors extracted on day 0 or day 5 after PET imaging and completed according to their coordinating protocol: Mouse HMGB1 (high mobility group protein B1; Novus Biologicals Cat. No. NBP2–62767, RRID:AB_2893367), BioLegend ELISA Max Standard Set Mouse TNFa (BioLegend Cat. No. 430901, RRID:AB_2883995), BioLegend ELISA Max Standard Set Mouse IFNg (BioLegend Cat. No. 430801, RRID:AB_2893366), and R&D DuoSet ELISA Development System Human Granzyme B (GZMB; crossreacts with mouse and human; R&D Systems Cat. No. DY2906–05, RRID:AB_2893368). Biotek Synergy Mx Plate Reader and Gen5 Software were used to image the ELISA assays (Gen5, RRID:SCR_017317).
RNA analysis – NanoString and nSolver
MC38 (N = 16 total, N = 8 treated, and N = 8 control) and E0771 (N = 16 total, N = 8 treated, and N = 8 control) tumor-bearing mice were analyzed using 18F-FMISO-PET imaging prior to and after initiation of dual checkpoint blockade on days 0 and 5, respectively. Tumors were collected on day 0 prior to treatment as a baseline and on day 5 after treatment initiation. Tumors were excised and stored in RNAlater (Thermo Fisher AM7021). The mRNA was extracted according to the manufacturer-recommended protocol using TRIzol Plus RNA Purification Kit (Invitrogen 12183555) and quantified using the NanoString nSolver Mouse PanCancer Immune Profiling Panel and nSolver 4.0 with Advanced Analysis 2.0 software (nSolver Analysis Software, RRID:SCR_003420).
All statistical analyses were performed using GraphPad Prism Version 9 software (GraphPad Prism, RRID:SCR_002798). For correlations between imaging and tissue-based analyses, multiple unpaired t tests, Pearson correlation, Spearman correlation, Kendall Tau correlation, repeated measure analysis, ANOVA, or Kaplan–Meier survival curves with Gehan–Breslow–Wilcoxon statistical tests were completed. Specific comparisons were made based on the choice of immunotherapy providing an underlying hypothesis, thus no adjustment for multiple comparisons were made. Normality was checked prior to any analysis using Kolmogorov–Smirnov and Anderson–Darling tests with univariate analysis procedure using Statistical Analysis Software (SAS) when normal assumption applied; otherwise, nonparametric methods were used or data were transformed (e.g., logarithm) when appropriate (Statistical Analysis System, RRID:SCR_008567). Correlations were considered significant if the P value for rejecting the null hypothesis of a zero slope was less than 0.05.
RNA data have been deposited into the Gene Expression Omnibus (GEO No. GSE184863) database repository (RRID:SCR_005012). Additional data generated in this study are available upon request from the corresponding author.
18F-FMISO PET imaging is predictive of response to checkpoint blockade
To first assess the correlation between hypoxia and longitudinal response to checkpoint inhibitor therapy, 18F-FMISO PET imaging was performed prior to and early during immunotherapy treatment in MC38 colon cancer and E0771 triple negative breast cancer (TNBC) tumors. Representative qualitative images (Fig. 1A) revealed visual demarcation of hypoxic (left) and normoxic tumors (right) following initiation of therapy. Quantification of 18F-FMISO accumulation in tumors stratified by downstream therapeutic response (tumor volume < 125 mm3 7 days posttreatment) demonstrated relatively stable oxygenation of responding tumors, whereas hypoxia increased significantly in nonresponding tumors. Interestingly, there was differentiation in initial hypoxia levels between tumor types. MC38 responding tumors were significantly more hypoxic at baseline prior to therapy, as measured by 18F-FMISO tumor to background (muscle; TBR) ratio, prior to therapy initiation compared with nonresponders (responder TBR = 1.80 ± 0.37 vs. nonresponder TBR = 1.20 ± 0.53; P < 0.05), whereas E0771 responders were significantly less hypoxic (TBR = 1.41 ± 0.33 vs. 1.90 ± 0.43; P < 0.05). Despite initial conditions, in both tumor models, nonresponding tumors became significantly more hypoxic after the initiation of therapy (P < 0.0001), whereas responding tumor oxygenation remained relatively stable through time (P = 0.71). On day 5 posttherapy initiation, the TBR of responding E0771 TNBC tumors was significantly less compared with nonresponding tumors (1.45 ± 0.35 vs. 1.96 ± 0.43, respectively; P = 0.0087). Trends were consistent with the MC38 colorectal tumors, with the TBR for responders at 1.67 ± 0.27 and nonresponders at 2.40 ± 0.37 (P = 0.0019; Fig. 1B and C). The increase in 18F-FMISO in nonresponding groups was not dependent on changes in tumor volume, as the slope of the nonparametric linear correlations between TBR and tumor volume was determined not statistically significantly different than non-zero for MC38 and E0771 tumors, respectively (P = 0.35 day 0, P = 0.11 day 5; Figs. 1D and E).
Next, we sought to determine the utility of PET imaging to stratify tumors based on hypoxic tumor microenvironment. To facilitate this, 18F-FMISO PET imaging on day 5 was used to quantify hypoxia and stratify tumors into imaging metric designated responders and nonresponders and subsequent tumor progression was measured by longitudinal measurements of tumor volume. A receiver-operating characteristic analysis was performed to determine a hypoxia threshold for tumor response. For MC38 tumors, the threshold used was 1.86 with a sensitivity of 100% and specificity of 83.3%, and for E0771 it was 1.77, with a sensitivity of 100% and a specificity of 87.5% (Supplementary Fig. S1). For imaging-predicted groups from day 5, significant differences in tumor volume between hypoxic and normoxic tumors could be measured by day seven posttreatment in MC38 tumors (normoxic volume 196 ± 170 mm3, hypoxic volume 478 ± 366 mm3; P = 0.006). For the E0771 cohort, imaging-predicted groups differences were seen as early as two days following treatment (175 ± 55 mm3 vs. 385 ± 209 mm3; P = 0.003; Figs. 1F and G). Across both tumor lines, divergence in hypoxia preceded tumor volume changes, with more rapid regression of tumors occurring in the less hypoxic E0771 tumor model. In accordance with tumor volume measurements, overall survival of normoxic colorectal tumors (median survival = undefined vs. 13 days; P = 0.01) and normoxic TNBC (median survival = undefined vs. 9 days; P = 0.003) were significantly extended in comparison with hypoxic tumors (Fig. 1H and I). Taken together, these data supported the role of 18F-FMISO PET imaging to predict and stratify response to immune checkpoint treated tumors based on hypoxia.
18F-FMISO PET imaging tumor stratification is correlated with an inflamed phenotype
We next sought to understand whether the differences in checkpoint inhibition response were driven by divergent immunologic phenotypes. PET imaging was acquired on either day 0 or 5, and 18F-FMISO TBR was used to characterize tumors as normoxic or hypoxic based on the threshold TBR that was previously defined. Given the similarities in PET imaging, tumor growth kinetics, and overall response across TNBC and colon cancer, both tumor types were combined for phenotypic analysis (Supplementary Fig. S2). Excised tumors were analyzed by NanoString mRNA profiling with the Mouse PanCancer Immune Profiling Panel. Phylogenetic clustering demonstrated that the majority of normoxic tumors had an overall inflamed phenotype, whereas the majority of hypoxic tumors were categorized as noninflamed (Fig. 2A). To determine the specific biologic mechanisms that differed between normoxic and hypoxic tumors, we compared mRNA expression across a number of immune signaling pathways. Normoxic tumors, in comparison to hypoxic tumors, had significantly increased mRNA expression of adaptive immune signaling proteins (3.34 ± 4.57 vs. −4.29 ± 3.89; P = 0.0033), chemokines (3.10 ± 4.13 vs. −3.987 ± 3.094; P = 0.002), cytokines (4.41 ± 5.83 vs. −5.674 ± 4.566; P = 0.0021), and interleukins (3.533 ± 4.776 vs. −4.543 ± 3.968; P = 0.0028; Fig. 2B). The increased levels of immune signaling were also reflected in transcriptional analysis of cellular phenotypes. Interestingly, in comparison with hypoxic tumors, normoxic tumors had lower levels of cytotoxic T-cell mRNA relative to total immune cells (0.235 ± 0.389 vs. 0.732 ± 0.438; P = 0.03). However, functionally, normoxic tumors contained a significantly lower relative number of exhausted Cd8 cells compared with hypoxic tumors (−0.135 ± 0.736 vs. −1.027 ± 0.507; P = 0.012). Dendritic cells relative to the total immune population were also enriched in normoxic tumors compared with hypoxic tumors (normoxic −0.255±1.243 and hypoxic −1.393±0.475; P = 0.039; Fig. 2C). A comparison of the absolute number of cells identified normoxic tumors as containing a greater number of tumor-infiltrating lymphocytes (7.446 ± 0.886 vs. 5.957 ± 0.827; P = 0.004), macrophages (9.438 ± 1.301 vs. 7.774 ± 0.938; P = 0.0129), and dendritic cells (7.191 ± 1.588 vs. 4.564 ± 1.250 N = 7 P = 0.003) compared with hypoxic tumors, thus providing a direct correlation between immune cells present and oxygenation (Fig. 2D).
Spatial organization of the tumor microenvironment differs between 18F-FMISO-stratified tumors
The presence and spatial localization of cell types identified by mRNA profiling were next explored using immunofluorescence. As demonstrated in Fig. 3, normoxic tumors had increased TNF, Arginase, and F4/80 expression, with a majority of TNF and arginase spatially colocalized in the center core (Fig. 3A and C), Pimonidazole staining was enhanced in hypoxic tumors, which also contained fewer macrophage markers compared to normoxic tumors (Fig. 3B and D). This indicated an increase in both classically activated (M1) and alternatively (M2) activated macrophages in responding tumors, which is consistent with a TH1 response. In addition, normoxic tumors demonstrated higher expression of CD11c localized in the center core (Fig. 3E) when compared with hypoxic tumors which had much lower levels of CD11c that was expressed only outside of hypoxic regions (Fig. 3F), indicating increased presence and localization of dendritic cells in responding tumors. Furthermore, normoxic tumors show increased expression localized in the inner core of CD8 cells and an outer core of granzyme B (Fig. 3G), when compared with hypoxic tumors (Fig. 3H). Normoxic tumors were also marked by an increased expression of TLR4 (Fig. 3I) when compared with hypoxic tumors (Fig. 3J). Taken together, these data indicated oxygenation (normoxia) contributes to a spatial stratification of antigen presenting cells such as macrophages and dendritic cells in an organized structure when compared with areas of hypoxia, which appear less structurally organized. Furthermore, hypoxic tumors reveal decreases in immune cell populations and localization pattern determined via decreased amounts of TNF, arginase, F4/80, CD8, granzyme B, and TLR4, supporting that normoxic tumors respond due to differences in cellular presence and spatial organization.
Normoxic tumors identified by 18F-FMISO PET have increased TH1 signaling
After identifying the immune cells present in normoxic and hypoxic tumors as defined by PET imaging, we next explored specific signaling pathways that would support the differences in immune cell phenotypes through transcriptomics and direct protein measurement. Because the differences in hypoxia measured through 18F-FMISO-PET imaging were only predictive following the initiation of immunotherapy, we hypothesized that normoxic tumors would be more efficient in propagating damage-associated molecular pattern (DAMP) signaling pathways. Thus, genes and proteins specifically associated with a DAMP-driven, TH1 response were assayed. Although there was not a significant difference in the level of Tnf gene transcription either before or after starting immunotherapy, there was enrichment in normoxic tumors, compared with hypoxic, for the gene encoding Ifng (5.23 ± 0.92 vs. 3.75 ± 1.22; P = 0.019) after initiating therapy (Fig. 4A). We also explored the levels of genes encoding for receptors that contribute to initiating (Tlr2 and Tlr4), amplifying (Tnfr2) and recruiting additional immune cells (Ccr2) to propagate the initial DAMP-driven response. Although no differences were detected on pretreatment tumors at baseline, by day 5, normoxic tumors were found to have a significant increase in genes encoding for the Tnf receptor (5.42 ± 1.25 vs. 3.86 ± 0.65; P = 0.015), Tlr2 (7.36 ± 0.66 vs. 6.55 ± 0.39; P = 0.019), Tlr4 (7.16 ± 1.15 vs. 5.45 ± 1.08; P = 0.013), and Ccr2 (9.25 ± 1.03 vs. 7.47 ± 0.92; P = 0.005). Together, these data supported our hypothesis that the differences in immune cell populations between normoxic and hypoxic tumors were driven by immune signaling.
To explore expression of signaling molecules controlled through spatial segregation rather than transcriptional regulation, 18F-FMISO PET stratified tumors (Fig. 4B) were analyzed for high mobility group box-1 (Hmgb1) protein, a nuclear protein that is released from cells following cell damage that activates Tlr2/4 signaling pathways (24–26). Consistent with immune-mediated killing, normoxic tumors had significantly elevated levels of HMGB1 in comparison with hypoxic tumors (2,794 ± 1,040 mg/mL vs. 677 ± 788 mg/mL; P = 0.0003; Fig. 4E). Given the difference in the levels of Tlr-activating Hmgb1, we compared downstream signaling proteins IFNγ and TNF. IFNγ cytokine levels mirrored transcript levels, with significantly higher release in normoxic tumors compared to hypoxic tumors (5,510 ± 5,222 μg/mL vs. 417 ± 385 μg/mL; P = 0.011; Fig. 4C). Interestingly, TNF cytokine expression was significantly different between normoxic and hypoxic tumors (267.2 ± 126.0 μg/mL vs. 77.12 ± 66.41 μg/mL; P = 0.0015); however, the Tnf transcript was not different between the groups (P > 0.05; Fig. 4F). Finally, granzyme B, a downstream measure of successful T-cell activation, was also significantly increased in normoxic tumors compared with hypoxic tumors (4.420 ± 1.935 μg/mL vs. 1.287 ± 1.057 μg/mL; P = 0.0009; Fig. 4D). Together, these data support a biological mechanism in which an increase in HMGB1/DAMPS creates an increase in TLR signaling, leading to enhanced TH1-cell recruitment and activation resulting in response in normoxic tumors, thus providing a direct correlation between immune response and tumor oxygenation (Fig. 4G).
18F-FMISO imaging guides the delivery of hypoxia targeted therapy to rescue immunotherapy nonresponders and elicit an effective response to immunotherapy
Given the correlation between hypoxia and immune suppression, we utilized an image-guided approach to add a hypoxia targeted prodrug in combination with continued immunotherapy to hypoxic tumors that were imaging metric designated non-responders. The goal of this approach was to induce an antitumor immune response through release of DAMPs (Fig. 5A). The hypoxia targeted prodrug, evofosfamide, was given following 18F-FMISO PET stratification of normoxic or hypoxic tumors in immunotherapy treated or control treated tumors. 18F-FMISO PET imaging was acquired again following treatment with evofosfamide to determine the effect of treatment on tumor hypoxia. While vehicle-treated tumors significantly increased in hypoxia (day 5 TBR = 1.86 ± 0.73 day 5, day 10 = 2.62 ± 0.77; P = 0.012), evofosfamide plus immunotherapy significantly reduced hypoxia in hypoxic tumors (day 5 = 2.55 ± 0.40, day 10 = 1.76 ± 0.59; P = 0.005) and reduced hypoxia to the lowest measured level in evofosfamide plus immunotherapy tumors (1.22 ± 0.44; Fig. 5B). Evofosfamide given without immunotherapy did not have a distinguishable difference on either tumor volume (vehicle 2,302 ± 1,374 mm3 and evofosfamide only 2,322 ± 1,072 mm3 on day 26) or overall survival (vehicle median survival of 19 days vs. 15 days for evofosfamide alone) when compared with vehicle controls (Fig. 5C and D). The addition of immunotherapy to normoxic tumors resulted in significant and sustained tumor volume control by day 14 (P = 0.03) and extended overall survival (MS = not reached; P = 0.009), whereas the addition of immunotherapy to hypoxic tumors did not affect tumor growth or median survival compared with controls (16 days; P > 0.05). Despite the ineffectiveness of immunotherapy alone in controlling hypoxic tumors, the addition of evofosfamide to hypoxic tumors undergoing immunotherapy significantly reduced tumor volume by day 20 compared with controls (P = 0.03) and extended overall survival (median survival not determined; P = 0.0005). Surprisingly, the addition of evofosfamide to normoxic tumors also improved both tumor growth control by day 17 in comparison with immunotherapy alone (P = 0.02) and resulted in 100% survival. Taken together, these data support synergy between hypoxia imaging guided timing of evofosfamide and immunotherapy as defined by the Bliss test of independence. The proportion of reduction in 18F-FMISO signal was proportional to the increase as response rate, further supporting the connection between hypoxia prevention and checkpoint blockade response (Fig. 5E).
Hypoxia PET imaging with 18F-FMISO represents a valuable tool for measuring a correlative physical property of the tumor microenvironment that elicits global effects on immune cells during immunotherapy. Herein, we demonstrate that hypoxia PET imaging provides robust early predictions of therapeutic response to checkpoint blockade in mouse models of breast and colorectal cancer. 18F-FMISO imaging effectively differentiates normoxic and hypoxic tumors with measurements strongly correlating with underlying tumor immune phenotypes. Imaging-delineated normoxic tumors have increased inflammation and upregulated type I immune response, whereas hypoxic tumors have a suppressed immune response and increased dysfunctional effector cells. Furthermore, we demonstrate that imaging can identify tumors that are susceptible to tumor microenvironment-altering agents to rescue the immune response and sensitize predicted unresponsive tumors to immunotherapy. Therapy targeting hypoxic cells can rescue response to immunotherapy and synergistically increase effectiveness in tumors with severe hypoxia, providing a rational framework for implementing personalized combination immunotherapy.
Quantification of hypoxia using PET imaging is well established, as previous studies have demonstrated that PET imaging using 18F-FMISO correlated with gold standard invasive hypoxia measurements. Furthermore, 18F-FMISO PET imaging has been used previously to show hypoxia negatively correlates with outcome in other treatment modalities, including surgical resection, radiotherapy, and chemotherapy (27, 28). However, despite strong biological rationale, PET imaging of hypoxia has not been investigated in the setting of immunotherapy. Here, we report the first correlative and longitudinal PET imaging of hypoxia during immunotherapy. Our results demonstrate that increased tumor hypoxia precedes anatomical growth of tumors, providing a potential biomarker for predicting checkpoint blockade resistance. Interestingly, in a model of breast cancer, pretreatment hypoxia was predictive of nonresponse to immunotherapy, whereas in a colorectal cancer model responding tumors were significantly more hypoxic pretreatment. This likely demonstrates differences in hypoxia pre-treatment among different tumor models is less influential than immunotherapy treatment, especially in syngeneic subcutaneous models. In addition, it is possible that moderate levels of hypoxia in a short-term may even augment immune function, especially in an acute setting by inducing temporarily increased inflammation (29). We demonstrated that within five days of starting immunotherapy, regardless of tumor type, nonresponding tumors significantly increased in hypoxia, whereas responding tumors remained unchanged, revealing that the mechanism in which the hypoxic tumor microenvironment initially responds to the introduction of immunotherapy plays a major role in eventual therapeutic response. In addition, significant increases in hypoxia after starting immunotherapy were correlated with decreased overall survival. Each of these measures were independent of tumor volume either pretreatment or on day 5 posttherapy, indicating that hypoxia, rather than tumor size or aggressiveness, drove response, and thereby provides an early predictive measure of therapeutic response.
Given the correlation between hypoxia, as quantified by 18F-FMISO PET imaging, and eventual response to checkpoint blockade, we sought to understand whether PET imaging could identify distinct tumor immune microenvironments. Previous studies using invasive or in vitro techniques have demonstrated hypoxia correlates with immune suppression through decreased intratumoral dendritic cells, polarization to an M2 phenotype, and increased regulatory T cells (13, 30, 31). Stratifying tumors based on hypoxia imaging metrics provided the ability to compare differential tumor immune microenvironments using a transcriptomic approach. Phylogenetic mapping of the tumors demonstrated significant clustering of normoxic tumors into a more inflamed microenvironment, whereas hypoxic tumors had a suppressive phenotype that was consistent with previous analyses of nonresponsive tumor immune microenvironments (32, 33). The inflammation in normoxic tumors was broad in nature, including significant enrichment in adaptive, chemokine, cytokine, and interleukin pathways. Upregulation of these signaling pathways in turn led to a greater number of functional proinflammatory immune cells, which supported the observation that normoxic tumors were responsive to checkpoint blockade. Transcriptomic profiling was confirmed by immunofluorescent chemistry and cytokine ELISA, further supporting 18F-FMISO PET imaging as a surrogate biomarker for antitumor immune microenvironments.
An outstanding question that remained was how checkpoint blockade prevented the induction of hypoxia within tumors. Canonical markers of angiogenesis such as Vegf and Vegfr1/2 were not differentially expressed between hypoxic and normoxic tumors in either model. In addition to immune cells, Toll-like receptors such as Tlr2 and Tlr4 are expressed on some endothelial cells and have been shown to play an important role in sprouting angiogenesis, independent of Vegf signaling (34, 35). In this study, Tlr2 and Tlr4 were found to have higher expression following immunotherapy treatment in normoxic tumors, as compared with hypoxic tumors. Furthermore, a ligand of both Tlr2 and Tlr4, Hmgb1, was found at a significantly higher level in the extracellular matrix of normoxic tumors. These data support the hypothesis that effective tumor killing through immunotherapy induces the release of Hmgb1, which in turn activates Tlr2 and Tlr4 signaling not only in immune cells, but also the surrounding tumor endothelial cells. This creates a feedforward loop in which angiogenesis reduces hypoxia and improves immune cell recruitment and function, in turn causing further target cell killing ultimately feeding back into the Tlr pathway. Interestingly, Hmgb1 can alternatively act in an immunosuppressive manner when oxidized, which may suggest that its effectiveness is limited to hypoxic tumor microenvironments (36). These data are also consistent with previous work that describes changes in tumor vascularization in response to effective checkpoint blockade (37).
Given that 18F-FMISO PET imaging could identify checkpoint blockade responsive tumors prior to anatomic changes, and that hypoxia posttherapy correlated with decreased immunogenic cell death, we sought to target hypoxic cells and increase immunogenic cell death through a single mechanism. A number of hypoxia-activated treatments, such as evofosfamide, have been trialed with limited success to date. Clinical trials of evofosfamide alone or in additional to other chemotherapy treatment have shown no significant improvement in progression-free survival despite restoring T-cell infiltration and sensitizing tumors to immunotherapy in prostate cancer and pancreatic adenocarcinoma (30, 33). According to its mechanism of action, we hypothesized that by administering evofosfamide on day five, when hypoxia differentiation occurs, treatment would result in improved response to immunotherapy. Our results demonstrated significant improvement in both tumor volume control and overall survival in tumors that were given combination evofosfamide and checkpoint blockade in comparison with their matched immunotherapy-only control. Consistent with a DAMP-associated immune mechanism of hypoxia control, the addition of evofosfamide in combination with immunotherapy also significantly reduced hypoxia from day 5 to day 10. Taken together, our work demonstrates that noninvasive hypoxia imaging allowed for a rational combination of immunotherapy and hypoxia targeted therapy that resulted in better response than either treatment alone.
In summary, we have demonstrated that noninvasive hypoxia PET imaging can be utilized to predict and characterize the tumor microenvironment to guide effective immunotherapy. 18F-FMISO PET imaging was an early imaging biomarker, predictive of subsequent therapy response (anatomical changes) across multiple tumor types, and PET signal correlated with an inflamed tumor microenvironment. Furthermore, information gleaned from longitudinal imaging allowed for the appropriate timing of evofosfamide to supplement an immune driven feedforward loop that enhanced response through the immune activation and the prevention of hypoxia. Although these studies are limited to syngeneic subcutaneous tumors, they provide a framework for expansion into a clinical setting. These data support initial investigations of 18F-FMISO accumulation prior to and following administration of checkpoint blockade. If the mechanism of hypoxia is conserved across species, then 18F-FMISO can serve to guide the timing of evofosfamide in personalized and image-guided adaptive immunotherapy trials. Both 18F-FMISO and evofosfamide have been approved as investigational new drugs by the FDA; therefore, rapid clinical exploration is possible to test the translation of our described paradigm. If successful, this strategy may act to rescue otherwise nonresponsive tumors and enhance the number of patients who benefit from immunotherapy.
E.S. Yang reports grants and personal fees from Bayer, grants from Eli Lilly, personal fees from AstraZeneca, and grants and personal fees from Clovis outside the submitted work. A.G. Sorace reports grants from NIH NCI and American Cancer Society during the conduct of the study. B.M. Larimer reports personal fees and other support from Cytosite Biopharma outside the submitted work as well as a patent for Hypoxia PET Imaging to Guide Immunotherapy pending. No disclosures were reported by the other authors.
K.M. Reeves: Conceptualization, resources, data curation, software, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. P.N. Song: Conceptualization, resources, data curation, software, formal analysis, investigation, methodology, writing–original draft, writing–review and editing. A. Angermeier: Resources, data curation, software, formal analysis, writing–original draft, writing–review and editing. D. Della Manna: Resources, data curation, software, formal analysis, writing–original draft, writing–review and editing. Y. Li: Data curation, software, formal analysis, validation, writing–review and editing. J. Wang: Resources, data curation, software, formal analysis, writing–original draft, writing–review and editing. E.S. Yang: Resources, data curation, software, formal analysis, writing–original draft, writing–review and editing. A.G. Sorace: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. B.M. Larimer: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.
The authors thank the UAB Cyclotron Facility, UAB NanoString Core, UAB Pathology Core for technical assistance. Funding was provided by the UAB O’Neal Preclinical Imaging Shared Facility Grant (National Cancer Institute P30CA013148), a NIH Directors’ New Innovator Award (NCI DP2CA261453) to B. Larimer, a NIH Pathway to Independence award (NCI R00CA215604) to B. Larimer, NIH research funding (NCI R01CA240589) to A. Sorace, an American Cancer Society Research Scholar Grant (RSG-18–006–01-CCE) to A. Sorace, and an O’Neal Comprehensive Cancer Center Mary Ann Harvard Award to B. Larimer.
The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.
Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/).