PHD2 Constrains Antitumor CD8⁺ T-cell Activity

Since the discovery of tumor-specific and tumor-associated antigens in the nineties (1), numerous studies (2) have demonstrated that the immune system can constrain tumor development in humans and rodents. However, several sequential steps are required for effective tumor control in vivo, and these are subject to negative feedback mechanisms. These steps include priming in the draining lymph nodes, migration to the tumor bed, reactivation in the tumor microenvironment, and tumor lysis. It is clearly established that the tumor microenvironment displays several features that determine the outcome of immune reactivity to tumor cells. Among them, hypoxia (low oxygen tension) is a common characteristic of most solid tumors and may impede antitumor immunity by regulating the function and/or recruitment of various immune cell types.

Cell adaptation to hypoxia is regulated via a set of oxygen sensors, among which members of the oxygen-dependent prolyl hydroxylase family [prolyl hydroxylase domain (PHD) or EGLN], comprising 3 members, play a major role. Hypoxia-inducible factors (HIF), of which there are three (HIF-1α, HIF-2α, and HIF-3α), are a set of evolutionary conserved transcriptional regulators that represent the best-characterized substrates of PHDs (3). These transcription factors are active as heterodimers consisting of a cytoplasmic HIF (mostly HIF-1α and/or HIF-2α) subunit and a nuclear HIF-1β subunit also known as ARNT (aryl hydrocarbon receptor nuclear translocator). Both subunits are constitutively expressed by all mammalian cells studied to date. Under normoxic conditions, HIFα is hydroxylated on two prolines by one of the 3 members of the PHD family, leading to its poly-ubiquitination by the E3 ubiquitin ligase complex VHL (composed of Von Hippel-Lindau protein, elongin B and C, Cullin 2 and Rbx1) and in its degradation by the proteasome (4). Under hypoxic conditions, PHD enzymes are inhibited and unable to hydroxylate the HIFα subunit, which can then bind to the HIFβ subunit and activate and/or inhibit the expression of over 1,000 genes (3).

The impact of hypoxia on immune reactivity in the tumor microenvironment has been studied using direct and indirect (through stabilization of HIF) approaches. A number of reports suggest that hypoxia may sustain tumor progression by (i) triggering the protumoral function of macrophages entrapped in hypoxic niches (5); (ii) recruiting other immunosuppressive cell types, such as myeloid-derived suppressor cells and regulatory T cells (Treg; ref. 6); (iii) downregulating Th1-type cell activation (7–10); and (iv) favoring Th17 differentiation (11, 12). The effect of hypoxia on Treg remains controversial: depending on the study and model considered, HIF-1α, the major oxygen sensor stabilized during hypoxia, has been shown to repress the Treg transcriptional program in vitro by inducing the degradation of Foxp3 (11, 12), or to favor their differentiation and function in a murine model of inflammatory bowel disease (13). Hypoxia may also promote immune evasion by inducing HIF-1α−dependent expression of immune checkpoint proteins (14, 15). In addition, hypoxia displays intrinsic effects on tumor cells, promoting the expression of genes that induce epithelial–mesenchymal transition, affect cell adhesion, and drive cancer development in the case of long-lasting HIF signaling (16–19). By contrast, there is increasing evidence that enhanced HIF activity may potentiate the cytotoxic activity of CD8⁺ T lymphocytes, which are major actors of antitumor immunity. Doedens and colleagues showed that loss of VHL (the main regulator of HIF stability) led to enhanced control of persistent viral infection in vivo and that VHL-deficient OT-I CTLs displayed a superior capacity to control ovalbumin (OVA)-expressing melanoma upon transfer (20). Hypoxic conditions in vitro increase the production of the cytolytic molecule granzyme B in a HIF-1α–dependent manner (21, 22).

These observations highlight the contrasting role of proteins involved in hypoxia sensing in the regulation of immunity/tolerance in the tumor microenvironment and raise the question of its global pro- or antitumoral role. In this study, we used mice genetically deficient for PHD2 (a proline-hydroxylase that mediates control of HIF activity) selectively in T lymphocytes (PHD2^ΔT) to characterize the phenotype and function of tumor-infiltrating CD8⁺ T lymphocytes in mice displaying elevated HIF-1 activity selectively in T lymphocytes.

C57BL/6 mice were purchased from Envigo (Horst, the Netherlands). C57BL/6 Phd2^flox/flox mice were generated as previously described (23, 24) and kindly provided by Dr. Bart Lambrecht (Ghent University). C57BL/6 Hif1α^flox/flox (B6.129-Hif1atm3Rsjo/J) mice were kindly provided by Dr. F. Bureau (Université de Liège) and CD4 Cre (B6.Cg-Tg(Cd4-cre)1Cwi/BfluJ) mice by Dr. G. Van Loo (Ghent University). Mice carrying Hif1α and Egln1 (Phd2) loxP-flanked alleles were crossed with CD4 Cre mice to obtain T cell–specific gene deletion. Mice were housed in individual ventilated cages and used at 6 to 9 weeks of age. The experiments were performed in compliance with the relevant laws and institutional guidelines and were approved by the Animal Care and Use Committee of the Institute for Molecular Biology and Medicine (protocol numbers: CEBEA-IBMM-2017–22–01 and 03).

EG7-OVA (source: ATCC CRL-2113) T lymphoblasts are derived from the C57BL/6 (H-2^b) mouse lymphoma cell line EL4, which was transfected by electroporation with the plasmid pAc-neo-OVA which carries a complete copy of chicken OVA mRNA and the neomycin (G418) resistance gene. B16-OVA (clone MO4, kindly provided by Kris Thielemans, VUB, Belgium) is a cell line that was isolated from skin tissue of a mouse with melanoma and was transfected with chicken OVA cDNA. It exhibits a morphology of spindle-shaped and epithelial-like cells. No full authentication was carried out, but the EG7-OVA cell line was characterized by T-cell markers; the B16-OVA cell line displays morphologic characteristics (see above), both cell lines were evaluated for SIINFEKL OVA/K^b expression using 25-D1.16 mAb (eBioscience 17–5743–82).

EG7-OVA and B16-OVA tumor cell lines were cultured in DMEM (Lonza 12–741F) containing 10% heat-inactivated FCS (Gibco 10270–106), sodium pyruvate 1 mmol/L (Lonza 13–115E), 1% nonessential aminoacid mixture (Lonza 13–114E), l-glutamine 2 mmol/L (Lonza BE17–605E), penicillin–streptomycin 100 units/mL (Lonza DE17–6025E), β-mercaptoethanol 50 μmol/L (Sigma-Aldrich M-7522) and G418 400 μg/mL (Sigma-Aldrich A1720). Tumor cell medium was changed every 2 days and a new cell line was regenerated every month. Cell lines were treated every 6 months with Mycoplasma Removal Reagent (Millipore 30–500–44).

10⁶ EG7-OVA tumor cells were injected subcutaneously in the right flank. Mice were monitored every 2 days for tumor growth and survival. The tumor volume (mm³) is expressed as (A × B²)/2, where A and B are tumor length and width, respectively. Some mice were injected intraperitoneally with 1-mg anti-CD8β (53–5.8, BioXCell BE0223) or IgG from rat serum (Sigma-Aldrich I8015) 5 days after tumor inoculation, and further treated with 250 μg antibodies weekly.

5×10⁵ B16-OVA tumor cells were injected subcutaneously in the right flank. Some mice were injected intraperitoneally with 200–280 μg anti–programmed cell death protein 1 (PD-1; RMP1–14, BioXCell BP0146) or isotype control (rat IgG2a, 2A3, BioXCell BE0089) 7, 9, 11, and 13 days after tumor inoculation.

CD8⁺ T lymphocytes were positively selected from spleen suspensions by magnetic-activated cell sorting using CD8 beads (Miltenyi 130–117–044) according to the manufacturer's protocol.

For transcriptomic analysis, enriched CD8⁺ T lymphocytes were stained with anti-TCRβ and anti-CD8 mAbs and further purified using a BD FACSAria III Cell Sorter. Sorted CD8⁺ T lymphocytes were activated with plate-bound anti-CD3 (2 μg/mL; clone: 145–2C11; BioXCell) and soluble anti-CD28 (1 μg/mL, clone: 37.51; BioXCell) in RPMI1640 medium (Gibco 31870–025) containing 10% FCS, sodium pyruvate, nonessential amino acid mixture, l-glutamine, penicillin–streptomycin, and β-mercaptoethanol.

To measure glucose dependency, MACS-enriched CD8⁺ T lymphocytes were cultured with DMEM medium without glucose (Life Technologies 11–9660–25) supplemented with 10 mmol/L glucose (Sigma-Aldrich G8270) or galactose (Sigma-Aldrich G5388) and 10% dialyzed FCS, sodium pyruvate, nonessential amino acid mixture, l-glutamine, penicillin–streptomycin, and β-mercaptoethanol.

EG7-OVA tumors were treated with DNAse I (Roche 10104159001) and Liberase (Roche 05401020001) to generate a single-cell suspension and viable cells were enriched on density gradient Lymphoprep (Stemcell Technologies 07851). Tumor-infiltrating cells were incubated with monoclonal antibodies conjugated to fluorochrome or biotin: TCRβ (H57–597), CD8α (53–6.7), CD44 (IM7), CD62 L (MEL-14), TIGIT (1G9), CD279/PD-1 (J43), TIM-3 (5D12) from BD Biosciences. Streptavidin was purchased from BD Bioscience (557598). Dead cells were excluded by LIVE/DEAD (Invitrogen L34976) staining and nonspecific Fc-mediated interactions were blocked by anti-mouse CD16/CD32 (2.4G2, BioXCell). To evaluate the functional status of tumor-infiltrating T lymphocytes, cells were stimulated in vitro for 3 hours with pharmacologic agents bypassing TCR signaling (50 ng/mL PMA (Sigma P8139) and 1 μg/mL ionomycin (Sigma I 0634) in the presence of a protein trafficking inhibitor allowing the intracellular retention of activation-induced cytokines (brefeldin A from Invitrogen 00–4506–51). Cells were subsequently fixed with BD cytofix/cytoperm kit (51–2090KZ), permeabilized with BD Perm/Wash (51–2091KZ) and stained with conjugated monoclonal antibodies directed at IFNγ (XMG1.2, Biolegend), Granzyme B (NGZB, Thermo Fisher), TNFα (MP6-XT22, Life Technologies), Ki67 (B56, BD Bioscience), T-bet (4B10, Thermo Fisher), Eomes (DAN11MAG, Thermo Fisher), Glut-1 (SPM498, Abcam), TOX (TXRX10, Thermo Fisher), and/or TCF1/TCF7 S33–966, BD Pharmingen). The expression of a transcription factor cascade (Tox, TCF1, T-bet, Eomes) was used to define four exhausted PD-1⁺CD8⁺ T lymphocyte subsets, in particular the terminally exhausted subset TCF1^–T-bet^loTox^hiEomes^hi (25).

Flow cytometry analysis was performed on a Canto II (BD Biosciences) or CytoFLEX (Beckman Coulter) and the data generated were analyzed using the FlowJo software (Tree Star) and the FlowSOM technique. This visualization technique analyzes flow or mass cytometry data using a self-organizing map (26).

C57BL/6 mice were immunized by injection of dendritic cells pulsed with SIINFEKL peptide (synthesized by PEPTIDE.2). Spleen cells were digested with collagenase type 3 (Gestimed LS004183) for 30 minutes at 37°C, further dissociated in Ca²⁺-free medium, and separated into low and high density fraction on a Nycodenz gradient (LUCRON 1002424). CD11c⁺ cells were enriched from low-density spleen cells by magnetic positive selection using anti-CD11c (N418) Microbeads (Miltenyi 130–108–338). Dendritic cells were incubated for 1 hour at 37°C with SIINFEKL peptide, washed and suspended in 200 μL PBS. 5×10⁵ SIINFEKL-pulsed dendritic cells were injected into the hind and fore footpads. Mice were tested 5 days later.

RNA was extracted from CD8⁺ T lymphocytes (purified by cell sorting) using TRIzol method (Sigma-Aldrich T9424) and reverse transcribed with Superscript II reverse transcriptase (Invitrogen 18064–014) according to the manufacturer's instructions. Quantitative real-time RT-PCR was performed using the SYBR Green Master mix kit (Thermo Fisher, K1082) and StepOne Plus system (Applied Biosystems). Amplification reactions were conducted for 40 PCR cycles (each cycle: 95C for 15 seconds, 60C for 1 hour) in duplicates. The following RT PCR primers were used: Ubiq-Fw: cgtctgaggggtggctatta; Ubiq-Rev: taaattggggcaagtggcta; Glut-1-Fw: caaacttcattgtgggcatg; Glut-1-Rev: agcaccgtgaagatgatgaa; Sprouty 2-Fw: cgatcacggagttcagatgt; Sprouty 2-Rev: agctctggcctccatcag. Transcript amounts were calculated by using a standard curve and normalized to Ubiquitin-transcripts, which was used as the housekeeping gene.

CD8⁺ T lymphocytes, purified as indicated in the “cell culture” section of the methods, were stimulated with anti-CD3 and anti-CD28 antibodies for 15 hours. RNA extraction was performed using a RNeasy kit (Qiagen, 74004) and sample quality was tested on a 2100 Bioanalyzer (Agilent). Total RNA underwent paired-end sequencing by using a TruSeq Stranded mRNA kit (25×106 paired-end reads/sample, Novaseq 6000 platform) performed by BRIGHTcore ULB-VUB, Belgium (http://www.brightcore.be). Adapters were removed with Trimmomatic-0.36 (with the following parameters: Truseq3-PE.fa:2:30:10 LEADING:3 TRAILING:3 SLIDINGWINDOW:4:15 MINLEN:36 HEADCROP:4). Reads were then mapped to the reference genome mm10 by using STAR_2.5.3 a software with default parameters. Reads were then sorted from the alignment according to chromosome positions and index the resulting BAM-files. Read counts in the alignment BAM-files that overlapped with the gene features were obtained using HTSeq-0.9.1 with “–nonunique all” option (if the read pair aligns to more than one location in the reference genome, it was counted in all features to which it was assigned and scored multiple times). Genes with no raw read count greater or equal to 20 in at least 1 sample were filtered out with an R script, raw read counts were normalized and a differential expression analysis was performed with DESeq2 by applying an adjusted P value < 0.05 and an absolute log2-ratio larger than 0.5.

All statistical analyses were conducted using GraphPad Prism (GraphPad Software, version 6). Statistical difference between two groups was determined by unpaired, two-tailed Student t tests. A one-way or two-way ANOVA was used for multigroup comparisons together with Tukey multiple comparisons post hoc tests.

Survival significance in tumor bearing mice was determined by a Log-rank Mantel–Cox test. Data are judged to be statistically significant when P value < 0.05. In figures, asterisks denote statistical significance (*, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P <0.0001).

The RNA sequencing (RNA-seq) data reported in this study have been deposited in the Gene Expression Omnibus repository with the accession code no. GSE216536. All other data are available in the main text or the Supplementary Materials or are available from the corresponding author on reasonable request.

To explore the role of the HIF pathway in the adaptive response to tumor in vivo, we deleted the gene encoding PHD2 selectively in T cells and monitored the growth of EG7-OVA lymphoma cells. Mice with loxP-flanked Egln1 (Phd2) alleles [generated as previously described (23, 24)] were crossed onto CD4 Cre mice, giving rise to the Egln1^fl/fl CD4^CRE (named PHD2^ΔT) mouse strain harboring a selective deletion of Egln1 (Phd2) in T lymphocytes. Expression of HIF-1α was increased in CD8⁺ T cells upon 24h in vitro stimulation with anti-CD3/CD28 (Supplementary Fig. S1A). We subcutaneously injected EG7-OVA lymphoma cells into PHD2^ΔT mice and compared tumor growth with that seen in wild-type (WT) mice and control littermates (CD4 Cre⁻Phd2^fl/fl). The data in Fig. 1A indicate that a significant proportion of PHD2^ΔT mice (mean = 67% in all experiments performed) rejected the tumor and survived at endpoint, as compared with 2.6% in WT mice. The tumor growth in Phd2^fl/fl was slower than in WT mice, presumably due to altered, hypomorphic expression of the floxed Phd2 allele (Fig. 1S, panel B). Of note, deletion of HIF-1α in otherwise PHD2-deficient mice (HIF-1α PHD2^ΔT) reverted the phenotype, as assessed by the control-like survival rate of these mice upon tumor inoculation (Fig. 1A and B). The control of tumor growth was dependent on CD8⁺ T lymphocytes, as the injection of depleting anti-CD8 increased mortality in both WT and PHD2^ΔT mice (Fig. 1C and D). Administration of anti-CD4 resulted in reduced tumor growth and death, presumably because of Treg depletion (Supplementary Fig. S2). Similar results, i.e., enhanced tumor regression, were observed in PHD2^ΔT mice injected with B16-OVA melanoma (Fig. 2, see isotype-treated groups). Thus, stabilization of HIF-1α resulted in increased control of tumor growth in two murine models.

To examine whether the Phd2 deletion affected the differentiation of CD8⁺ T lymphocytes into effector and/or memory cells, we used flow cytometry to monitor the capacity (upon PMA/ionomycin restimulation) of tumor-infiltrating CD8⁺ T lymphocytes to express IFNγ, TNFα, and Granzyme B as well as the glucose transporter GLUT-1, a direct target of HIF-1α. We used FlowSOM (self-organizing map) as a starting point for an analysis of a pool of cells from 4 experimental groups. Six major clusters were identified, from which two populations (identified as POP 1 and 2) were strongly enriched selectively in PHD2^ΔT mice (Supplementary Fig. S3). We split the data set to visualize the proportion of the populations in the four experimental groups (Fig. 3A and B). The expression of IFNγ, TNFα,  and GLUT-1 was strongly increased in tumor-infiltrating CD8⁺ T cells from PHD2^ΔT mice, reaching about 40% of CD8⁺ T cells (POP 1 and 2), as compared with 10% to 22% in other groups, whereas Granzyme B was overexpressed in POP 2 only, representing 15% of CD8⁺ T lymphocytes (Fig. 3B and C). Concomitantly, the proportion of CD8⁺ T lymphocytes expressing lower levels of four genes (POP 6) was strongly decreased. The absolute number of polyfunctional CD8⁺ T lymphocytes per gram of tumor reached 41,000 in PHD2^ΔT mice, a 2-fold increase, as compared with 19,000 and 21,000 cells in WT and CD4 Cre⁻ PHD2^fl/fl mice, respectively (Fig. 3D).

A summary of all experiments performed (Fig. 4A) revealed that 36% of tumor-infiltrating CD8⁺ T lymphocytes had the capacity to express IFNγ, TNFα, and Granzyme B in PHD2^ΔT mice, a more than 2-fold increase as compared with other groups. The calculation of the absolute numbers of cells showed a 6-fold selective increase in the number of CD8⁺ T lymphocytes expressing these three cytotoxic-associated factors in the same mice, reaching a mean of 65×10³ per gram of tumor (Fig. 4B). The proportion of some cells expressing a single functional marker (TNFα) as well as dual expressors (IFNγ⁺TNFα⁺ and IFNγ⁺Granzyme B⁺) was also significantly increased in PHD2-deficient cells (Supplementary Fig. S4).

The glucose transporter GLUT-1 was expressed by 65% (mean of all experiments) of tumor-infiltrating CD8⁺ T cells in PHD2^ΔT mice, as compared with 17% to 32% in other groups (Fig. 4C) and the absolute number of polyfunctional CD8⁺ T lymphocytes expressing GLUT-1 was significantly increased in PHD2^ΔT mice, suggesting a potential role for glucose-dependent metabolism (Fig. 4D). In addition, the proportion of polyfunctional tumor-infiltrating CD8⁺ T lymphocytes was higher among GLUT-1⁺ than among GLUT-1^– cells in PHD2^ΔΤ mice (Fig. 4E).

FACS analyses of tumor-infiltrating CD8⁺ lymphocytes revealed a slight increase in KI67 expression in PHD2^ΔΤ mice, as compared with WT (Supplementary Fig. S5A) despite a similar effector memory phenotype, as assessed by CD44 and CD62 L expression (Supplementary Fig. S5B). However, expression of the transcription factor Eomes (as assessed by its relative frequency and mean fluorescence signal) was downregulated in PHD2-deficient cells (Supplementary Fig. S6A), an observation in line with a potential role for this transcription factor in cell exhaustion. RNA-seq data confirmed a 1.3-fold decrease in Eomes mRNA levels (see later discussion). The expression of Tbx21 was not significantly different between groups.

We next used flow cytometry to examine the differentiation program of tumor-infiltrating CD8⁺ T cells in PHD2^ΔT mice and CD4 Cre⁻Phd2^fl/fl mice. The expression of several markers of differentiation/exhaustion revealed a change in the exhaustion program in PHD2^ΔT mice, i.e., an increased proportion of effector cells relative to “terminally exhausted” CD8⁺ lymphocytes (Fig. 5A; ref. 25). Indeed, the major populations infiltrating the tumor in both strains of mice differed by their expression of PD-1, Tox, and Eomes, with a more than 2-fold decrease in the proportion of PD-1⁺ cells among CD8⁺ TILs (Fig. 5B). Among these PD-1⁺ cells, we observed a decreased expression of the exhaustion-associated transcription factor TOX and a nonsignificant trend for increased proportion of TCF1⁺TIM3^lo progenitor cells in PHD2^ΔT mice (Fig. 5C).

Finally, we evaluated the capacity of lymphocytes from PHD2^ΔT mice to acquire cytolytic activity independently of tumor growth. WT and PHD2^ΔT mice were primed by injection of OVA peptide–pulsed splenic dendritic cells and tested for cytotoxic activity in vivo 5 days later. The data in Supplementary Fig. S6B show a significant increase in the lysis of OVA-pulsed targets in PHD2^ΔT mice, as compared with WT mice, suggesting that HIF-1α stabilization resulted in increased lytic activity of CD8⁺ T cells even upon transient antigenic stimulation.

Collectively, these observations suggest that the stabilization of HIF-1α in T lymphocytes inhibits their differentiation from effector into terminally exhausted TOX^hiPD-1^hi cells, thereby favoring the recruitment, survival, and/or differentiation of polyfunctional CD8⁺ T lymphocytes into the tumor bed.

To gain insight into the mechanism underlying increased CD8⁺ T-cell polyfunctionality in the context of PHD2 deficiency, we performed RNA-seq analysis on CD8⁺ T lymphocytes from WT and PHD2^ΔT mice 16 hours after anti-CD3 and anti-CD28 activation in vitro. We validated the PHD2 inactivation, as shown by the reduced frequency of reads coverage around exon 2 of the Egln1 gene (Fig. 6A). Analysis of the dataset identified 19 downregulated genes and 132 upregulated upon PHD2 deletion (Fig. 6B). Among upregulated genes, were Slc2a1 and Slc2a3, which encode for proteins (GLUT1 and GLUT3 respectively) that facilitate the transport of glucose. The representation analysis for pathways enriched from genes that were upregulated in activated CD8⁺ T lymphocytes from PHD2^∆T mice showed a specific enrichment of terms related to metabolism and energy (Fig. 6C and D). Several genes were involved in carbohydrate catabolic processes that generate pyruvate important for glycogenesis such as Ldha, Aldoa, Pdk1, and Pgk1. In addition, cytokine activity and regulation of immune effector process were strongly enriched implicating genes such as Tbx21, Ifng, Irf7, and Bcl3. Finally, genes such as Vegfa, Rora, and Pdk1 enriched pathways related to hypoxia. These observations suggest a link between increased ability of tumor rejection and enhanced glycogenesis.

To extend our understanding of PHD2 deletion on glycogenesis, we performed a motif enrichment analysis for transcription factors on the promoter of genes upregulated in CD8⁺ T lymphocytes from PHD2^∆T mice and associated with the glycogenesis process. We noticed a strong enrichment for HIF-1α motifs suggesting the involvement of this protein in the increased glycogenesis activity in these cells (Fig. 6E).

On the basis of our observations and a previous report showing that aerobic glycolysis was required for cytokine production (but not activation and proliferation) by CD4⁺ T lymphocytes (27), we compared the expression of cytokines in CD8⁺ T lymphocytes cultured in either glucose (to engage aerobic glycolysis) or galactose (to enforce respiration). The data in Fig. 7 indicate that the proportion of CD8⁺ T lymphocytes expressing IFNγ and Granzyme B was higher in PHD2^ΔT mice than in WT or CD4 Cre⁻ PHD2^ΔT counterparts, when activated in classical media (including glucose). Of note, culture in galactose-supplemented media led to a strong reduction in cytokine production and abrogated the difference among groups of mice. These data support the hypothesis that the increased function of CD8⁺ T lymphocytes from PHD2^ΔT mice upon in vitro stimulation was dependent on glucose metabolism.

There is increasing evidence that cells expressing intermediate levels of PD-1 preferentially expand upon PD-1 blockade (28). We therefore explored the role of PHD2 in T lymphocytes in response to immunotherapy in the B16-OVA melanoma tumor model. We subcutaneously injected WT, control littermates (CD4 Cre⁻Phd2^fl/fl) and PHD2^ΔT mice with B16-OVA melanoma and treated them or not with anti–PD-1 at days 7, 9, 11, and 13. In WT mice, treatment with anti–PD-1 had a minor effect on tumor growth and survival (Fig. 2). In sharp contrast, tumors were completely rejected in PHD2^ΔT and CD4 Cre⁻Phd2^fl/fl mice treated with anti–PD-1, indicating a strong synergism between PHD2 genetic deletion or hypomorphism and checkpoint blockade.

In this study, we show that the stabilization of HIF-1α in T lymphocytes (through genetic ablation of PHD2) enhanced the rejection of OVA-expressing tumors and altered the differentiation of CD8⁺ T cells, preventing their progressive dysfunction. In particular, a larger number of effector/memory T lymphocytes, characterized by an increased capacity to express IFNγ, TNFα, and Granzyme B, infiltrated the regressing tumors in PHD2^ΔT mice. Of note, the stabilization of HIF-1α, even partial as in CD4 Cre⁻ PHD2^fl/fl mice, had a synergistic effect with anti–PD-1 therapy to control tumor growth and survival.

Our observations point to an elevated glycolytic metabolism induced by sustained HIF-1α signaling as a molecular mechanism by which HIF-1α yields more protective effector/memory CD8⁺ T lymphocytes. Indeed, our data show: (i) an increased expression of the glucose transporter GLUT-1 by tumor-infiltrating CD8⁺ T lymphocytes in PHD2^ΔT mice; (ii) a potent induction of genes related to glycolytic metabolism (glucose transporters GLUT-1/3 and rate limiting glycolytic enzymes) in PHD2-deficient CD8⁺ T lymphocytes activated in vitro; (iii) the glucose dependency of the polyfunctional capacity (IFNγ, TNFα, and granzyme B production) of anti-CD3/CD28-stimulated PHD2-deficient CD8⁺ T lymphocytes. This hypothesis is in line with data in the literature showing that oxidative phosphorylation (OXPHOS) and aerobic glycolysis interchangeably support T-cell proliferation/survival but that aerobic glycolysis is required for optimal IFNγ production in mice (27, 29) and humans (30). Accordingly, glucose deprivation was shown to limit the cytolytic function of effector CTL in vitro (31–33). Two in vivo studies have shown that constitutive HIF-dependent glycolytic metabolism (through conditional deletion of Vhl in T lymphocytes) enhances the effector responses of CD8⁺ T lymphocytes and prevents exhaustion during chronic viral infection (20, 34). T-cell intrinsic deletion of PHD proteins was found to limit tumor colonization in the lung and improve adoptive cell transfer immunotherapy (35). Our data confirm these findings and further show that HIF-1α stabilization enhances the differentiation/recruitment of polyfunctional CD8⁺ T lymphocytes in vivo, resulting in increased tumor rejection and cytotoxicity. In contrast, Mamlouk and colleagues (36) reported that loss of PHD2 in both T lymphocytes and myeloid cells was required for tumor regression, possibly due to macrophage-mediated immune subversion in their tumor model.

Our study may also explain the apparently paradoxical observation of a better survival rate in groups of patients with glioma and acute myeloid leukemia (AML) bearing tumors expressing a mutated allele of the gene encoding isocitrate dehydrogenase 1 [referred to as IDHmut tumors, see (37)]. Differential impact of IDH1/2 mutational subclasses on outcome in adult AML results from a large multicenter study (38). Although mutations in IDH1 are often perceived as an immune evasion mechanism, it is noteworthy that gain-of-function mutations in this enzyme can lead to the accumulation of 2-hydroxyglutarate (2HG), a metabolite that can be produced as an L or a D enantiomer (39). Although both enantiomers can be considered as oncometabolites, their effect on immune cells are clearly distinguishable. In addition to promoting tumor development and fitness (39), D-2HG exerts potent immunosuppressive properties, notably by lowering the NAD/NADH ratio of tumor infiltrating CD8⁺ T cells (40). Oncometabolite D-2HG alters T-cell metabolism to impair CD8⁺ T-cell function (40). In marked contrast, L-2HG is known to inhibit PHD proteins leading to the stabilization of HIF-1α (41). S-2-hydroxyglutarate regulates CD8⁺ T-lymphocyte fate (40, 41), leading to enhanced tumor infiltration and effector function, further confirming, in a clinical setting, the beneficial effect of HIF-1 stabilization during an antitumor response. In conclusion, whereas the hypoxia-derived L-2HG has been shown to affect PHD2 enzymatic activity, oncometabolites produced by IDHmut tumors appear to dysregulate the epigenetic program of both tumor and tumor-infiltrating cells independently of the PHD2–HIF-1α signaling pathway.

It is likely that the increased glucose transport in tumor-infiltrating CD8⁺ T lymphocytes is of critical importance because of the metabolic competition in the tumor microenvironment (42, 43). Chang and colleagues have reported (44) that tumors may directly dampen the effector function of T lymphocytes by consuming glucose. They further show that programmed death-ligand 1 not only inhibits T lymphocytes via PD-1, but may also enhance tumor-cell glycolysis, thereby limiting glucose availability. Checkpoint blockade may therefore directly and indirectly potentiate T-cell function by increasing the level of available glucose in the tumor milieu. Of note, a recent report (45) confirmed that prolonged glucose restriction contributed to hyporesponsiveness of tumor-infiltrating CD8⁺ T lymphocytes and further showed that acetate may restore IFNγ production in glucose-deprived T lymphocytes in vitro and in vivo.

We found a correlation between the superior tumor control and increased numbers of tumor-infiltrating CD8⁺ T lymphocytes endowed with the capacity to produce IFNγ, TNFα, and granzyme B in PHD2^ΔT mice. These observations are consistent with several reports showing that polyfunctional CD8⁺ T lymphocytes display a superior ability to reject tumor cells or control viral infections (46–51), a property lost by “exhausted” T cells. Among the mechanisms known to induce/maintain T-cell exhaustion during chronic infection, Eomes, a paralog of T-bet, and Sprouty 2 (SPRY2), a negative regulator of the MAPK/ERK pathway, may be of interest in our setting as they were decreased in PHD2-deficient CD8⁺ T lymphocytes. The role of Eomes is complex, as it regulates both T-cell full differentiation and exhaustion, possibly depending on its level of expression (52). Li and colleagues reported that Eomes expression by tumor-infiltrating lymphocytes increased when tumor progressed and that Eomes was required for the development of antitumor effector cells. However, high levels of Eomes drove T lymphocyte unresponsiveness by inducing expression of T-cell exhaustion genes, such as Havcr2 (which encodes TIM-3; ref. 52). Another study emphasized the complementary role of progenitor (T-bet⁺) and terminal (Eomes⁺) subsets of CD8⁺ T lymphocytes to control viral infection (53). Accordingly, tumor-infiltrating CD8⁺ T lymphocytes in PHD2^ΔT and control littermates differed by their expression of PD-1 and TOX. Furthermore, the decrease in SPRY2 expression in PHD2-deficient CD8⁺ T lymphocytes (activated in vitro) may be relevant to the increased tumor control, as the lack of SPRY2 has been shown to enhance survival and T-cell polyfunctionality of effector CD8⁺ T lymphocytes in mice (54, 55) and humans (56). This hypothesis is further supported by a report showing that HIF-1α and HIF-2α increased the methylation of the SPRY2 promoter, leading to decreased SPRY2 mRNA levels in a hepatoma cell line (57). Finally, our data show that neuropilin 1, a potential biomarker for dysfunctional T lymphocytes, was downregulated in PHD2^ΔT mice, suggesting a potential role as immune memory checkpoint (58, 59). Additional experiments will be required to clarify the role of these genes in the antitumor response of PHD2^ΔT mice.

Broadly, we conclude that the stabilization of HIF-1α in T lymphocytes regulates the metabolic balance within the tumor niche, favoring the differentiation of polyfunctional CTLs. In addition to a direct effect on aerobic glycolysis, the HIF complex has been shown to control expression of CTL effector molecules, trafficking (HIF-null cells retain the expression of lymph-node-homing receptors; ref. 32) and regulation of TCR signaling, suggesting that several mechanisms may contribute to the control of tumor growth in PHD2-deficient mice. There is now increasing evidence that hypoxia may have a beneficial effect on the antitumor function of CD8⁺ T lymphocytes, thereby providing novel potential targets for cancer immunotherapy, eventually in association with currently available immune checkpoint blockers.

S. Goriely reports grants from Walloon Region, Fondation contre le cancer (Belgium); and grants from FNRS and Télévie during the conduct of the study. No disclosures were reported by the other authors.

C. Bisilliat Donnet: Conceptualization, resources, investigation, methodology. V. Acolty: Resources, investigation, methodology. A. Azouz: Validation, investigation. A. Taquin: Investigation, methodology. C. Henin: Investigation, methodology. S. Trusso Cafarello: Investigation. S. Denanglaire: Investigation, methodology. M. Mazzone: Conceptualization, supervision, writing–review and editing. G. Oldenhove: Formal analysis, supervision. O. Leo: Conceptualization, supervision, validation, writing–review and editing. S. Goriely: Supervision, writing–review and editing. M. Moser: Conceptualization, resources, writing–original draft, writing–review and editing.

We are grateful to Peter Carmeliet (KU Leuven, Leuven, Belgium) for providing mice with loxP-flanked Phd2 alleles and to Benoit Van den Eynde and Pierre Coulie (UCLouvain, Belgium) for interesting discussion and helpful suggestions. We thank Fabienne Andris and Kevin Englebert for valuable help, and Caroline Abdelaziz and Véronique Dissy for animal care.

The work was supported by the National Fund for Scientific Research (FNRS), the Fondation contre le Cancer, the Walloon Region (IMMUCAN), the European Regional Development Fund (ERDF-Wallonia BioMed 2014–2020-LIV 45–20), and grants from the Fonds Jean Brachet and the Fonds Hoguet.

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.