Glutathione peroxidase 2 is a metabolic driver of the tumor immune microenvironment and immune checkpoint inhibitor response

Abstract

Background: The existence of immunologically 'cold tumors' frequently found across a wide spectrum of tumor types represents a significant challenge for cancer immunotherapy. Cold tumors have poor baseline pan-leukocyte infiltration, including a low prevalence of cytotoxic lymphocytes, and not surprisingly respond unfavorably to immune checkpoint (IC) inhibitors. We hypothesized that cold tumors harbor a mechanism of immune escape upstream and independent of ICs that may be driven by tumor biology rather than differences in mutational neoantigen burden.

Methods: Using a bioinformatic approach to analyze TCGA (The Cancer Genome Atlas) RNA sequencing data we identified genes upregulated in cold versus hot tumors across four different smoking-related cancers, including squamous carcinomas from the oral cavity (OCSCC) and lung (LUSC), and adenocarcinomas of the bladder (BLCA) and lung (LUAD). Biological significance of the gene most robustly associated with a cold tumor phenotype across all four tumor types, glutathione peroxidase 2 (GPX2), was further evaluated using a combination of in silico analyses and functional genomic experiments performed both in vitro and in in vivo with preclinical models of oral cancer.

Results: Elevated RNA expression of five metabolic enzymes including GPX2, aldo-keto reductase family 1 members AKR1C1, AKR1C3, and cytochrome monoxygenases (CP4F11 and CYP4F3) co-occurred in cold tumors across all four smoking-related cancers. These genes have all been linked to negative regulation of arachidonic acid metabolism-a well-established inflammatory pathway-and are also known downstream targets of the redox sensitive Nrf2 transcription factor pathway. In OCSCC, LUSC, and LUAD, GPX2 expression was highly correlated with Nrf2 activation signatures, also elevated in cold tumors. In BLCA, however, GPX2 correlated more strongly than Nrf2 signatures with decreased infiltration of multiple leukocyte subtypes. GPX2 inversely correlated with expression of multiple pro- inflammatory cytokines/chemokines and NF-kB activation in cell lines and knockdown of GPX2 led to increased secretion of prostaglandin E2 (PGE2) and interleukin-6. Conversely, GPX2 overexpression led to reduced PGE2 production in a murine OCSCC model (MOC1). GPX2 overexpressing MOC1 tumors had a more suppressive tumor immune microenvironment and responded less favorably to anti-cytotoxic T-lymphocytes-associated protein 4 IC therapy in mice.

Conclusion: GPX2 overexpression represents a novel potentially targetable effector of immune escape in cold tumors.

Keywords: head and neck neoplasms; immune evation; immune tolerance; immunotherapy; tumor escape.

Figure 1

Identifying correlates of tumor immunity in tobacco-related malignancies. (A) Clustering of OCSCC, LUSC, LUAD and BLCA based on single sample gene set enrichment analysis score for 13 leukocyte subtypes (and log2 FOXP3 RNA for regulatory T-cells) identifies subsets of immunologically hot and cold tumors. OCSCC samples included tumors from the oral cavity, oral tongue (excluding base of tongue), floor of mouth, buccal mucosa, alveolar ridge, hard palate, and lip. (B) Differential analysis of genes upregulated in hot and cold tumors across all four tumor histologies. AKR1, aldo-keto reductase family 1; BLCA, bladder cancer; CYP, cytochrome P; GCLC, glutamine-cysteine ligase catalytic subunit; GPX2, glutathione peroxidase 2; LUAD, lung adenocarcinomas; LUSC, lung squamous cell carcinomas; OCSCC, squamous cell carcinomas of oral cavity; NQO1, NAD(P)H quinone dehydrogenase 1.

Figure 2

Nrf2-dependent and Nrf2-indepdent GPX2 regulation in HNSCC tumor lines. (A) Correlation between GPX2 RNA expression and Nrf2 activation (ie, Nrf2 single sample gene set enrichment analysis ssGSEA score) in a large panel of HNSCC cell lines annotated by KEAP1/NRF2 mutational status or relative amount of Nrf2 activation (U.Q., upper quartile). Remaining WT samples with Nrf2 scores below the UQ are represented by black circles. (B) Correlation between GPX2 RNA and TP63 expression in the same panel of HNSCC cell lines. (C) Clustering HNSCC cell lines by Nrf2 activation and expression of GPX2, TP63, or a second downstream Nrf2 target (GCLC) revealed five clusters or patterns of expression. GCLC, glutamine-cysteine ligase catalytic subunit; GPX2, glutathione peroxidase 2; HNSCC, human head and neck squamous cell; ssGSEA, single sample gene set enrichment analysis.

Figure 3

GPX2 suppression increases inflammatory mediator production by HNSCC cell lines. FaDu and UM22A cells were infected with lentiviral constructs containing either empty vector (EV) or shRNA targeting GPX2. (A) Western blot validation of GPX2 protein knockdown 72 hours post infection. Insert demonstrates suppression of GPX2 protein levels at 48 hours post infection. Conditioned media was analyzed at 48 and 72 hours post-infection using ELISA to quantitate levels of secreted IL-6 (B) or PGE2 (C). Soluble mediator levels are normalized to total number of cells. Data are presented as means, with error bars denoting SD. * denotes p<0.05; ** denotes p<0.01; *** denotes p<0.001, **** denotes p<0.0001. IL, interleukin; GPX2, glutathione peroxidase 2; HNSCC, human head and neck squamous cell; PGE2, prostaglandin E2.

Figure 4

GPX2 overexpression suppresses PGE2 production in murine HNSCC cell lines. (A) Western blot validation of murine GPX2 OE in MOC1 cells. Clones 1 and 2 (C1 and C2) were derived from polyclonal stable cell lines selected with puromycin following infection with either an empty vector or one with the mouse GPX2 complementary DNA insert. (B) GPX2 OE doubled the intracellular GPX2 enzymatic activity measured in MOC1 cell lines. (C) PGE2 secretion was reduced by twofold in MOC1 clones with GPX2 OE. Stable overexpression of either empty vector or GPX2 containing constructs were accomplished in the MOC1 murine cell line (insert denotes Western blot). Individual clones (C) were established for both the empty vector control (EV) and for the GPX2 overexpressing constructs. and reduced PGE2 secretion. Data are presented as means, with error bars denoting SD. *** denotes p<0.001; **** denotes p<0.0001. GPX2, glutathione peroxidase 2; HNSCC, human head and neck squamous cell; PGE2, MOC, murine oral cancer; OE, overexpression; PGE2, prostaglandin E2.

Figure 5

GPX2 drives tumor immune microenvironment shifts. MOC1 tumors (N=6) generated from parental (wild-type (WT)), empty vector (EV) or GPX2 containing constructs were harvested at the same size and analyzed via flow cytometry. All data are presented as means, with error bars denoting SD and individual tumor volumes showed using individual circles. * denotes p<0.05; ** denotes p<0.01; *** denotes p<0.001, **** denotes p<0.0001, after performing a Tukey multiple comparison test and further adjusting p values with a Benjamini-Hochberg correction (FDR<0.1) to control the family wise error within a cell line group due to multiple testing across different leukocyte subpopulations. The arcsine transformation was applied to percentages prior to statistical analysis. EV, empty vector; G-MDSC, granulocyte-myeloid derived suppressor cells; GPX2, glutathione peroxidase 2; MOC, murine oral cancer.

Figure 6

GPX2 reduces ICI effectiveness. Pooled analysis of two independent experiments in which MOC1 tumors with EV (A, B) or GPX2 OE (C, D) were established and allowed to grow in the flank model. Established tumors underwent treatment with three total injections of either isotype (iso) control (A,C) or anti-CTLA4 antibody (B, D). Tumor measurements are presented as individual symbols (square=experiment 1; circle=experiment 2) and each tumor is represented as an individual curve for the growth panels (red=EV; blue=OE). Each treatment group had 20 tumors (exp I=12; exp II=8) (E) Survival is denoted using Kaplan-Meier curves as an aggregate of the treatment groups. (F) Table of responders and non-responders following anti-CTLA4 antibody treatment. CTCLA-4, cytotoxic T-lymphocytes-associated protein 4; EV, empty vector; GPX2, glutathione peroxidase 2; ICI, immune checkpoint inhibitor; MOC, murine oral cancer; OE, overexpression.

Figure 7

Both GPX2 and Nrf2 activation correlate with poor survival in OSCC. Recursive partitioning was used to dichotomize patient tumors as either high or low for GPX2 RNA expression (left) or Nrf2 activation (right). Kaplan-Meier curves of overall survival were analyzed for statistical significance with a log-rank test. HRs for tumors expressing high GPX2 and Nrf2 activation are 1.593 and 1.452, respectively. GPX2, glutathione peroxidase 2; OSCC, squamous cell carcinomas of oral cavity.