The primary objectives of this study were to evaluate the effects of metformin on (1) systemic and local immunity and (2) tumor growth inhibition in a mouse model of HPV-induced head and neck cancer. Wild-type C57BL/6 J 6–8 weeks old male and female mice used in the experiments were purchased either from The Center for Comparative medicine, Baylor College of Medicine (BCM) or Jackson Laboratory. These mice were housed at 4–5 per cage under specific pathogen-free conditions in appropriate lighting and temperature conditions with free access to food and water. Tumor inoculation was performed when mice reached 7–9 weeks of age. All animal experiments were conducted in facilities at the BCM, Houston, Texas, USA. The study protocols used in the in vivo studies have been reviewed and approved prior to initiation of the study by the Institutional Animal Care and Use Committee at BCM. The precise number of mice used within each experiment is presented in the corresponding figure legend.

Two mouse cell lines (mEER and mEERL95) stably expressing HPV16 E6-E7 proteins and oncogenic H-ras were used in the current studies. The mEER cell line established by in vitro transformation of male murine pharyngeal epithelial cells with HPV16 E6–E7 proteins and oncogenic H‐ras was obtained from Chad Spanos at the Sanford Research center/University of South Dakota and maintained in E-media as described by Spanos et al.28 The mEERL95 cell line derived from a tumor explant of the original mEERL cell line implanted in a C57BL/6J female mouse, is suitable for preclinical and mechanistic related studies on HPV-related head and neck cancer in immunocompetent mouse models using both female and male mice as tumor cell recipients. This was obtained from University Hospital of Lausanne, Department of Otolaryngology, Head and Neck surgery and maintained in Dulbecco’s modified Eagle medium/nutrient mixture F‐12 culture medium with supplements as described previously.29 In the current study, mEER was used in male mice and mEERL95 was used in female mice. Tumor cells in culture were harvested at 80% confluency, washed, counted and resuspended in sterile phosphate buffered-saline (PBS) before injection into mice. C57BL/6 J male and female mice were injected subcutaneously with 10⁶ mEER and mEERL95 cells respectively in the left flank. The mice were monitored two times per week and the tumor growth was measured using calipers. Tumor area (mm²) was calculated as A=L x W where L is length and W is width. The mice were sacrificed when either their tumors reached 225 mm², or if the tumors became severely ulcerated or necrotic and started to interfere with their well-being.

Tumors were allowed to develop for 18–21 days to an average size of 65 mm² prior to the treatment (tumors were randomized to treatment). Order of treatment and measurements were varied as to minimize potential confounder effects. Metformin hydrochloride (M2009-100G; Spectrum Laboratory Products) was administered either intraperitoneally (i.p solubilized in PBS) at a concentration of 40 mg/kg of mice weight once daily for varying timelines in a total of 100 µL injection volume or in drinking water at a concentration of 5 mg/mL. ICI, InVivoMAb anti-mouse programmed cell death protein 1 (PD-1) (clone RMP1–14; BioXCell; 250 µg per dose) and/or InVivoMAb anti-mouse cytotoxic T-lymphocyte associated protein 4 (CTLA-4) (clone 9H10; BioXCell; 100 µg per dose), were administered using i.p. injections as previously described by our collaborators.24 25

To assess the tumor immune cell infiltration, the tumors were harvested, processed, and stained as described previously.24 25 Briefly, tumors were digested at 37°C for an hour in a digestion cocktail buffer prepared in Roswell Park Memorial Institute (RPMI) 1640 medium (Sigma-Aldrich) containing DNase I (20 U/mL; Sigma-Aldrich), Collagenase I (1 mg/mL; EMD Millipore) and Collagenase IV (250 U/mL; Worthington Biochemical Corporation), followed by mechanical disruption to form single cell suspensions. Single cell suspensions were also prepared from blood, spleen, and inguinal lymph nodes with spleen and blood needing an additional lysis of the red blood cells (RBC) using Invitrogen’s RBC lysis buffer. Cells were blocked with the anti-mouse CD16/CD32 Fc block (BD Biosciences) and then separately stained for the extracellular T-cell and the myeloid cell antibodies panel with details of all the antibodies included in online supplemental table S1. For intracellular antibody staining cells were first fixed and permeabilized (eBioscience Intracellular Fixation and Permeabilization Buffer Set; Invitrogen). Data for both antibody panels were acquired on LSRFortessa (BD Biosciences) flow cytometer and analyzed using FlowJo V.10 software (FlowJo). Gating strategies are outlined in online supplemental figure S1.

Total RNA was extracted from the flash frozen metformin treated and untreated tumor tissues harvested on day 36 from the chronic metformin exposure experimental group using the Qiagen’s RNeasy Mini Kit as per the manufacturer’s instructions. RNA sample quality was assessed on the Bioanalyzer prior to gene expression profiling (GEP). Assays were performed with approximately 100 ng total RNA using the nCounter PanCancer Immune Profiling Panel for mouse (NanoString Technologies) containing 770 genes involved in cancer immune response. GEP was performed using the NanoString nCounter Gene Expression system. The process in brief includes the hybridization of total RNA with the NanoString assay panel containing 770 unique pairs of reporter probes, biotin-labeled capture probes and internal reference controls. Excess probes were then removed with magnetic bead purification on the nCounter Prep Station (software V.4.0.11.2) on the High Sensitivity assay. The samples were then quantified by identifying the fluorescent spots using the nCounter Digital Analyzer (software V.3.0.1.4). Raw data were analyzed using NanoString’s nSolver Analysis Software. Raw read counts were normalized, background subtracted, and assessed for cell type score and differential gene expression using NanoString nSolver (V.4.0) following the manufacturer’s instructions under the R computation environment (R V.3.4.0). Genes with a fold-change larger than 1.5 and an adjusted P value (Benjamini-Yekutieli procedure) of less than 0.05 between the control and treated groups were analyzed further. Using differential gene expression analysis, the metformin treated group demonstrated a statistically significant expression change with a Benjamini-Yekutieli false discovery rate corrected p<0.01 in 130/710 genes (online supplemental table S2). All the genes were categorized into gene sets based on their immune response category. The results from the differential expression testing for all the genes were then summarized at the gene set level, where most differentially expressed genes from each gene sets were identified to generate a directed ‘global significance score’, calculated as a square root of the mean signed squared t-statistic for the genes in a gene set, with t-statistics coming from the linear regression underlying the differential expression analysis. All the gene sets were found to be upregulated in the metformin treated group (details on the ‘directed global significance score’ for each gene set in online supplemental table S3) and these results were confirmed in the volcano plot displaying each gene -log10 (p value) and log2 fold change in the metformin treated versus untreated groups. Immune infiltrate cell type profiling was done to estimate the relative abundance of each cell type and a tumor infiltrating lymphocyte (TIL) score was calculated from the abundance of all the considered cell populations together. Relative cell type measurement evaluated more subtle patterns in the TIL landscape considering the variable amounts of total immune infiltrates between samples; relative versus total TIL values in the samples were compared against the treatment covariant.

Subsequent to receiving approval from BCM’s and the Michael E. DeBakey Veterans Affairs Medical Center’s Institutional Review Boards, we reviewed the records of Veterans with a diagnosis of oropharyngeal squamous cell carcinoma between January 1, 2000 and January 1, 2015. All collection and analysis of the current data was performed in a manner consistent with existing standards for clinical research (Declaration of Helsinki, US Federal Policy for the Protection of Human Subjects). Analysis of TILs was performed on all pretreatment tumor specimens as previously described using conventional immunohistochemistry coupled to quantitative imaging.30

Analysis was performed using GraphPad Prism V.8 software and the statistical significance was determined using either unpaired t test for two-tailed data or analysis of variance test followed by selected comparison using Tukey’s multiple comparison tests with multiple comparison correction (*p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001; ns, non-significant). All error bars indicate SE of measurement unless indicated otherwise.

Wild type C56BL/6J mice without any manipulation (ie, tumor inoculation) were treated with metformin or saline (PBS) control administered i.p for five consecutive days; inguinal nodes and spleen were harvested from all animals and were analyzed for the T cell and myeloid subsets by flow cytometry. A substantial decrease in the total immune cell population was observed in the spleens of metformin-treated mice. This change was consistent across all the subpopulations of CD11b+ cells, dendritic cells and CD3+ cells. The inguinal lymph nodes also showed a similar pattern with an overall decrease in all the T-cell and myeloid populations in the metformin treated group with the exception of dendritic cells (online supplemental figure 2). We next evaluated the effect of acute metformin exposure on systemic immunity in tumor-bearing mice. In mice carrying established mEER tumors, metformin treatment triggered an increase in the number of CD4 +T cells and a decrease in CD4 +FOXP3+T-regulatory cells and LY6G+ granulocytic myeloid derived suppressor cells (G-MDSCs) in tumor-draining lymph nodes, whereas all the other T-cell and myeloid populations showed no change between the treated and control groups. Metformin treatment significantly decreased the G-MDSCs in the spleens of treated mice. No significant differences in any other cell populations were observed (figure 1).

We previously showed limited ICI activity in this HNSCC model, even when combined with conventional chemoradiation regimens.25 Based on the significant effects of metformin identified above, we sought to determine whether metformin exposure would shift TIME in the context of ICI treatment. As shown in figure 2, metformin treatment generated a shift towards substantial protective immune modulatory effects in both tumor and spleen tissue with a decrease in regulatory T-cells (Treg) and G-MDSCs and an increase in CD8+, CD4+, monocytes and dendritic cells. Ki67 expression is upregulated in proliferating T cells and an upregulation of specifically Ki67 +PD1+CD8+ and Ki67 +PD1+CD4+cell responses are elicited in the group treated by metformin and checkpoint inhibitors. Despite the favorable immune shifts noted, no changes were seen in tumor growth delay as a function of metformin treatment over this short time course even when combined with ICI treatment although this experiment was geared primarily to measurements of shifts in TIME rather than tumor growth delay effects (figure 2).

Given the substantial immunomodulatory effects noted for acute metformin exposure, we sought to clarify whether these shifts are sufficient to generate antitumor activity in a dedicated experiment. As shown in figure 3, metformin treatment delivered either i.p or in drinking water failed to arrest tumor growth to any significant degree. Systemic immune profiling by flow cytometry analysis did not show any differences in the CD8+ and the CD4 +T cell types between the treated and the untreated mice in this experiment. A statistically significant decrease in the Treg population was observed in the metformin i.p treated group, and this was reflected in an increased CD8 +T cell/Treg ratio as well in this group, but no change was seen in the metformin in drinking water group (figure 3). In the myeloid cell subtypes, only an increase in the inflammatory monocytes was seen in the group that received metformin in drinking water. Even though there were beneficial immunomodulatory effects observed (ie, increase in the CD8 +T cells/Treg ratio, increase in inflammatory monocytes), these acute responses were apparently insufficient to mount a substantial antitumor response.

In contrast to most preclinical model and clinical trial regimens, including that utilized in figure 3, metformin utilization in retrospective cohorts consists of longitudinal exposure lasting months to years, which predates the initial cancer diagnosis.4–6 In order to partially mimic this regimen, we generated a chronic exposure treatment as outlined in figure 4. Pretreatment with metformin prior to tumor inoculation resulted in a significant and substantial (>50%) reduction in tumor growth velocity. Flow cytometry analysis of the systemic immune response to chronic metformin exposure showed a statistically significant increase in the CD8+ cells in the group treated with metformin in drinking water, but nodifference in the i.p treated metformin group. No significant differences were found in the CD4+ and the Treg types. Similarly, a significant increase in the ratio of CD8+/Tregs was also observed in the metformin in drinking water group but not in the metformin i.p treated group (figure 4). Conversely, analysis of myeloid cell subpopulations showed a beneficial myeloid response in the metformin i.p treated group but not any differences were seen in the group that received metformin in drinking water. An approximate twofold increase in the immune-stimulating myeloid subsets like the dendritic cells and the inflammatory monocytes which have been reported to support antitumor immune responses was seen in the metformin i.p treated group.31 On the other hand, a significant increase in circulating G-MDSCs which are immunosuppressive in nature was observed in the metformin i.p treated group, but this does not correlate with the overall antitumor response seen in this group. This might indicate that the presence of tumor infiltrating MDSCs and not the peripheral MDSCs are responsible for the differences in response to therapy and chronic metformin exposure might serve as an effective approach in combination with other conventional treatment modalities to improve treatment efficacy.

Tumors from the chronic metformin exposure and the untreated control groups were comprehensively profiled for immune-related gene expression using Nanostring’s mouse PanCancer Immune Profiling Panel (figure 5). The housekeeping genes that are used for the normalization of the data showed stable expression level across all the samples. Metformin treatment was the covariate used in the analysis. A global distribution of the p values for the experiment generated by testing each gene’s univariate association with the covariate showed a significant mass of data with low p value strong evidence for an association between the covariate and the gene expression data (online supplemental figure 3B). Normalized data were used to generate a heatmap via unsupervised hierarchical clustering, wherein the metformin treated samples clustered together characterized by an overall upregulation of the immune-related genes (online supplemental figure 3A). Many genes that are associated with T-cell inflamed status (CD8, Stat1, Ccl5, Pdcd1, Tnf, Cd86) (online supplemental table S2) were upregulated in the metformin treated group characterized by their involvement in antigen presenting, MHC and T-cell functions (figure 5). The top gene sets that were upregulated are involved in antigen processing, dendritic cell functions, inflammation and macrophage function (online supplemental table S3). Metformin treated tumors demonstrated high TIL scores with highest infiltration of CD8 +T cells, T cells, natural killer (NK) cells and cytotoxic T cells and a decrease in B-cells, Th1 cells, mast cells when measured against TILs. Plotting of differential expression analysis to known protein-based Kyoto Encyclopedia of Genes and Genomes pathways revealed highest enrichment for the ‘cytokine-cytokine receptor interaction’ (figure 6A) and ‘cell adhesion molecules (CAMs) pathway (figure 6B) along with other pathways such as the JAK-STAT signaling, chemokine signaling and those involved in viral infections. An upregulation of several CC and CXC subfamily chemokines, class I cytokines, TNF family, CAMs involved in the immune system along with and some of their corresponding receptors was seen in the metformin treated group. All observed gene expression changes in tumors treated chronically with metformin represent a shift toward an immune favorable gene expression which has a potential for generating a beneficial antitumor response. An additional analysis was performed focusing on genes included in the previously described T-cell inflamed GEP (T-GEP) which has been correlated with response to blockade of the PD1 therapeutic axis (NKG7 is not available in this panel precluding calculation of an actual T-GEP score).32 33 As illustrated in figure 6C, there was a substantial increase in expression of T-GEP genes in metformin treated tumors.

To date, there have been fairly limited immunologic datasets from HNSCC patients taking metformin. We previously reported on TILs as a predictor of oncologic outcomes in Veterans with HNSCC limited to the oropharynx site.30 In the current study, we performed a secondary analysis of this dataset using clinically annotated data regarding the presence of diabetes mellitus (DM) at the time of cancer diagnosis and metformin utilization at the time of diagnosis and during treatment delivery. Of 143 patients, 21 patients carried a diagnosis of DM; among these, 11 patients were taking metformin at time of diagnosis and during treatment and 10 were utilizing other means of glycemic control. Quantitative IHC of CD3+ TILs and CD8+ TILs was performed as previously described.30 DM patients taking metformin had higher fractions of CD3+ and CD8+ TILs compared with DM patients not taking metformin (CD3 +28.3% vs 18.2%, p value 0.04; CD8 +16.2% vs 7.2%, p value 0.02).