Formate Supplementation Enhances Antitumor CD8⁺ T-cell Fitness and Efficacy of PD-1 Blockade

The immune system provides a critical defense against tumor progression (1). CD8⁺ T cells are key immune effector cells in antitumor immunity. Thus, cancer immunotherapy strategies have aimed to increase CD8⁺ T-cell function. One promising strategy is immune-checkpoint blockade (ICB), which disrupts pathways inhibiting antitumor immunity (2). ICB has revolutionized cancer treatment by providing durable responses, even for patients with advanced disease (3–7). Although ICB therapy is efficacious in multiple cancer types, many patients either fail to respond, relapse after treatment, or have tumors inherently resistant to immune therapy (8). To extend the therapeutic potential of ICB, a better understanding of mechanisms of T-cell dysfunction is essential to identify novel targets to combine with ICB.

Metabolic reprogramming is a hallmark of cancer, allowing rapidly dividing tumor cells to generate macromolecules necessary for proliferation and survival (9, 10). Recent studies have shown metabolites are also crucial for T-cell activation, differentiation, and function (11, 12). To support their rapid proliferation, activated T cells increase the uptake of fuels, such as glucose and amino acids, to generate precursors required for macromolecular synthesis and energy production. Metabolic impairments in the tumor microenvironment (TME) are emerging as important resistance mechanisms exploited by cancer, as tumor-infiltrating lymphocytes (TIL) acquire metabolic insufficiencies restricting their function (13). T cells and cancer cells are dependent on similar nutrients to support cellular proliferation and function. However, tumor cell nutrient depletion and accumulation of waste products can impair T-cell function (12). Understanding the connections between metabolic dependencies and compromised T-cell function in the TME provides a vulnerability to exploit to improve antitumor immunity.

One-carbon (1C) metabolism is the most rapidly induced metabolic pathway during early T-cell activation (14–16). 1C metabolism uses serine to generate single carbon units for purine and thymidine biosynthesis and promotes redox homeostasis by generating glycine for glutathione synthesis and NADPH to reduce oxidized thioredoxin and glutathione (17, 18). Serine starvation or defects in 1C metabolism result in depressed nucleotide synthesis and reduced proliferation, but supplementation with formate as a 1C donor can partially restore these deficits (19–21). 1C metabolism also contributes to mitochondrial protein synthesis, supporting oxidative phosphorylation in tumors (22). T cells with depressed 1C metabolism exhibit decreased proliferation and increased DNA damage, resulting in cell death (15). 1C metabolism consists of both a cytosolic and a mitochondrial compartment. In both compartments, serine is converted to glycine by the enzyme serine hydroxy-methyl transferase (SHMT1-cytoplasm and SHMT2-mitochondria) (23). This reaction allows for transfer of a methyl group from serine to tetrahydrofolate (THF), which is converted to formate by methylenetetrahydrofolate dehydrogenase (MTHFD1-cytoplasm, MTHFD2-mitochondria; ref. 24). Notably, exposure of tumor cells to formate, directly or from decreased expression of aldehyde dehydrogenase 1 family member 2 (ALDH1L2), enhances formyl-peptide receptor-dependent cancer cell migration (25). Critically, cancer cells are dependent on serine for de novo purine synthesis to support their cellular proliferation. Activity of this pathway is elevated across human cancers. Thus, inhibiting folate metabolism with drugs such as methotrexate has been a foundation of cancer therapy for the past 70 years. Whether 1C metabolism limits antitumor T-cell function has not been investigated.

In this study, we demonstrate that induction of 1C metabolism is linked to T-cell receptor (TCR)–dependent activation of CD8⁺ T cells, and the TME contains low serine, which may contribute to dysfunctional antitumor CD8⁺ T cells. Remarkably, exhausted CD8⁺ T-cell function can be enhanced by supplementation of 1C units with formate. Therapeutic supplementation of 1C metabolism with formate caused global changes in the transcriptional activity of metabolic pathways, effector function, and cell cycle in CD8⁺ T cells. Supplementation with formate can synergize with anti–PD-1 treatment to enhance antitumor CD8⁺ T-cell proliferation and effector function and consequently tumor clearance. Thus, our data identify formate supplementation as an unexpected strategy to overcome a metabolic vulnerability in the TME and improve the efficacy of ICB.

Using a novel coculture system (26), we investigated TCR-dependent metabolic changes in CD8⁺ T cells upon their interaction with tumor cells. TCR-transgenic CD8⁺ T cells specific for chicken ovalbumin (OVA) epitope H2-K^b:OVA_257–264 (OT-1) were activated in vitro using anti-CD3/anti-CD28, effector cell polarized using IL2/IL12, and then cocultured with tumor cells expressing cognate antigen (B16-OVA) or not (B16; Fig. 1A). Using a rapid (few seconds) filtration method based on the size differential between T cells and tumor cells (26), we assessed metabolic dependencies in CD8⁺ T cells after coculture. We performed ¹³C-glucose tracing to identify the metabolic fate of glucose following tumor cell interactions with T cells (Fig. 1B). Serine biosynthesis from glucose was uniquely increased in OT-1 cells following interactions with B16-OVA cells, demonstrated by an increased proportion of the T-cell serine pool labeled with ¹³C following coculture (Fig. 1C). Serine biosynthesis in T cells was dependent on TCR interactions with tumor cells (B16-OVA), as glucose-derived serine was not increased following coculture with tumor cells not expressing OVA (B16; Fig. 1C). Serine is the rate-limiting metabolite for 1C metabolism and is directly linked to glucose utilization via this pathway. Carbon from ¹³C-glucose-derived serine was metabolized into ¹³C-glycine, demonstrating that newly synthesized serine contributes to 1C metabolism in OT-1 cells (Fig. 1D). By contrast, the glucose contribution to other metabolites downstream of glycolysis, lactate and pyruvate, was increased in OT-1 cells cocultured with either B16-OVA or B16 cells, demonstrating glucose contributions to lactate and pyruvate were independent of TCR interactions with tumor cells (Fig. 1E and F).

The first step in 1C metabolism is serine conversion to glycine. To determine whether there were TCR-dependent alterations in serine–glycine conversion, parallel experiments were performed in serine/glycine-deficient media using ¹³C-serine or ¹³C-glycine (Fig. 1G). Analysis of serine uptake demonstrated OT-1 cells in coculture reached a steady state in the intracellular serine pool (Fig. 1H). However, the conversion of glycine to serine was abrogated in OT-1 cells in coculture, driven instead by serine synthesis from glucose compared with activated T cells cultured alone (Fig. 1I).

We probed whether serine biosynthesis from extracellular glucose could compensate for low serine environments during OT-1 cell activation. Naive OT-1 cells were stimulated in vitro in media lacking serine and glycine (Fig. 1J). OT-1 cells stimulated in serine/glycine-deficient media proliferated and upregulated activation markers (CD44) similarly to T cells from control media; thus, serine biosynthesis is sufficient to support these features of early activation (Fig. 1K–N). However, OT-1 cells activated in the absence of serine and glycine had diminished cytotoxic potential with decreased tumor cell killing in vitro (Fig. 1O). These defects in cytolysis were rescued by the addition of 1C donor formate to T cells activated in the absence of serine and glycine (Fig. 1O). Together, these results demonstrate that TCR-dependent CD8⁺ T-cell interactions with tumor cells activate serine biosynthesis from glucose. When activated in the absence of serine and glycine, serine biosynthesis was sufficient to support proliferation but not cytolytic function.

To determine how T-cell stimulation regulates serine metabolism, naive CD8⁺ T cells were activated in vitro and expression kinetics of the rate-limiting enzymes of serine biosynthesis (PHGDH) and 1C metabolism (SHMT1, SHMT2) were evaluated over the next three days (Fig. 2A and B). Although minimally expressed in naive T cells, expression of these enzymes increased during this time (Fig. 2C). These changes in enzyme expression directly correlated with serine biosynthesis, as steady-state levels of glucose-derived serine increased over the three days after CD8⁺ T-cell activation (Fig. 2D). To determine the impact of TCR signal strength on 1C pathway activity, CD8⁺ T cells were activated across a range of concentrations of SIINFEKYL peptide-pulsed splenocytes (for OT-1 CD8⁺ T cells) or anti-CD3 (for bulk CD8⁺ cells) for three days (Fig. 2E; Supplementary Fig. S1A–S1C; Supplementary Fig. S2A). For the peptide studies, we developed a congenic marker-based (CD45.1 vs. CD45.2) rapid magnetic separation model for metabolomic analysis. Splenocytes (CD45.2) were pulsed with varying concentrations of SIINFEKYL peptide and then cultured with congenically distinct OT-1 cells (CD45.1) for three days. We added CD45.2-PE antibody to the cell culture prior to magnetic bead separation using anti-PE microbeads. We used these rapidly isolated OT-1 cells to perform Western blot analysis and labeled metabolite tracing. Increased expression of 1C (SHMT1/SHMT2) and serine biosynthesis (PHGDH) enzymes was directly correlated with increasing concentrations of SIINFEKYL peptide or anti-CD3 (Fig. 2F; Supplementary Fig. S2B). Increased expression of these enzymes was also associated with increased contributions of glucose-derived serine to the OT-1 and CD8⁺ T-cell intracellular serine pool (Fig. 2G; Supplementary Fig. S2C). These data demonstrate that increased serine biosynthesis is linked to TCR signal strength and is not exclusive to OT-1 cells but rather a feature of all CD8⁺ T cells.

To probe the importance of 1C metabolism in T cells, naive CD8⁺ T cells were activated in vitro in the presence of a competitive inhibitor of SHMT1 and SHMT2 (SHMTi; ref. 27; Fig. 2H). Consistent with the role of 1C units in generating purines for cell proliferation, CD8⁺ T cells did not divide in the presence of SHMTi (Fig. 2I). Notably, formate supplementation was sufficient to overcome impaired cellular division in SHMTi-treated CD8⁺ T cells (Fig. 2I). In contrast, activation (denoted by CD44 expression) of CD8⁺ T cells was not impaired by SHMT inhibition (Fig. 2J). To test whether SHMTi would affect the cytolytic function of differentiated effector CD8⁺ T cells, SHMTi was added at the time of coculture of CD8⁺ T cells (previously activated in the absence of SHMTi) with tumor cells to determine whether blocking 1C metabolism affected cytotoxicity. Notably, SHMT inhibition did not alter CD8⁺ T-cell–specific killing (Supplementary Fig. S3). Collectively, these findings demonstrate that 1C metabolism is dependent on T-cell activation status and strength of the TCR signal. Inhibition of 1C metabolism during activation of naive CD8⁺ T cells impairs their proliferation but not CD44 upregulation. Inhibition of 1C metabolism also does not impair the cytolytic function of effector CD8⁺ T cells.

To determine how tumor cells shape the metabolic TME and affect antitumor CD8⁺ T-cell–mediated immunity, we first evaluated the impact of tumor cells on metabolite concentrations in vitro. We analyzed metabolite levels by mass spectroscopy in tumor-conditioned media (CM) obtained from B16-OVA cells cultured in standard RPMI media for 24 hours. Among all measured metabolites, serine was the most depleted relative to the control media (Fig. 3A). Using ¹³C-glucose tracing, we determined serine biosynthesis was minimal in B16-OVA and B16 tumor cells cultured in control (RPMI) or tumor CM alone or with OT-1 CD8⁺ T cells (Fig. 3B). Unlike T cells, which derived ∼15%–20% of serine from glucose (Fig. 1C), B16-OVA tumor cells derived only ∼2% of serine from glucose, indicating these tumor cells rely on environmental serine. These results highlight a potential T-cell–specific vulnerability for intracellular serine biosynthesis. However, other tumor cell lines have the ability to synthesize serine from glucose, highlighting that these metabolic features differ between tumor cell types (28).

To test how tumor cells affect metabolites in the TME in vivo, we performed a metabolomic analysis of the tumor interstitial fluid (TIF) and plasma of mice with B16-OVA tumors (Fig. 3C; Supplementary Fig. S4A). Similar to the CM, serine and glucose were decreased relative to plasma (Fig. 3D and E). In contrast, other metabolites such as methionine, glycine, lactate, and pyruvate were increased in the TIF relative to the plasma (Supplementary Fig. S4A–S4E). These findings demonstrate that in B16-OVA tumors, the TME is relatively depleted of serine in vivo, suggesting that serine biosynthesis may be particularly important in CD8⁺ T cells within the TME.

To determine the functional significance of CD8⁺ T-cell serine biosynthesis in the TME, we utilized the CRISPR-based Chimeric Immune Editing (CHIME; ref. 29) system to delete Phgdh (Supplementary Fig. S5A), the rate-limiting step of serine biosynthesis, in OT-1 cells. OT-1 cells lacking Phgdh (Phgdh-KO) showed defects in serine biosynthesis (Supplementary Fig. S5B–S5D). Cotransfer of control and Phgdh-KO OT-1 cells into B16-OVA tumor-bearing mice revealed a decreased abundance of Phgdh-KO OT-1 cells, relative to control cells, in B16-OVA tumors (Fig. 3F and G). However, Phgdh-KO T cells did not differ in expression of markers of proliferation (Ki-67), function (Granzyme B), or T-cell exhaustion (PD-1, TCF-1, and TIM-3) relative to control cells (Supplementary Fig. S6A–S6E). Together, these findings demonstrate the importance of cell-intrinsic CD8⁺ T-cell serine biosynthesis in the TME.

We investigated whether 1C supplementation with formate would improve the efficacy of anti–PD-1. Anti–PD-1 treatment drives the proliferation and effector functions of tumor-specific CD8⁺ T cells (30). Previous studies demonstrated that formate supplementation could partially rescue the expansion of T cells during serine starvation or from defects in 1C metabolism (15, 17). Thus, we provided formate to mice in their drinking water beginning on day −1 before tumor cell implantation and administered anti–PD-1 after B16-OVA tumors became palpable (Fig. 4A). Formate treatment alone did not affect tumor growth or mouse survival in the conditions tested (Fig. 4B–D; Supplementary Fig. S7A). However, mice receiving formate combined with anti–PD-1 demonstrated profound improvement in tumor growth control and a marked increase in tumor clearance compared with anti–PD-1 treatment alone (Fig. 4C and D; Supplementary Fig. S7B). To test whether CD8⁺ T cells were required for the increased tumor clearance observed in mice receiving formate and anti–PD-1, parallel experiments were performed following CD8β⁺ cell depletion (Supplementary Fig. S8A). CD8β⁺ cell depletion abrogated the beneficial effects of combination therapy (Fig. 4E–G; Supplementary Fig. S8B and S8C). In contrast, formate combined with anti–PD-1 did not increase the clearance of EO771 or 4T1 tumors (Supplementary Fig. S9A–S9H). Collectively, these findings demonstrate that metabolic supplementation with formate during anti–PD-1 treatment can enhance tumor clearance in a tumor model–selective manner, and CD8⁺ T cells are essential for these therapeutic effects.

We next tested whether formate could synergize with anti–PD-1 in mice with existing tumors. Mice with established B16-OVA tumors received formate in their drinking water (day 9 after tumor implantation) one day before anti–PD-1 treatment (day 10 after tumor implantation; Fig. 4H). Formate treatment alone did not significantly alter tumor control (Fig. 4I; Supplementary Fig. S10A). Treatment with anti–PD-1 alone attenuated tumor growth compared with isotype control. However, formate combined with anti–PD-1 had greater efficacy, resulting in clearance of B16-OVA tumors in ∼60% of mice compared with ∼12% in mice receiving anti–PD-1 alone (Fig. 4J and K; Supplementary Fig. S10B). Together, these data demonstrate formate supplementation can improve responses of B16-OVA tumors to anti–PD-1.

We utilized single-cell RNA-sequencing (scRNA-seq) to understand the effects of formate supplementation on the global CD8⁺ T-cell transcriptional landscape in B16-OVA tumors in an unbiased fashion. We administered formate (day 9 after tumor implantation) and anti–PD-1 (day 10 after tumor implantation), either alone or in combination, to mice with established B16-OVA tumors and purified tumor-infiltrating CD8⁺ T cells for scRNA-seq analysis on day 12 post-implantation (Fig. 5A). Unsupervised cluster analysis was performed on the integrated datasets from each of the four treatment groups: control isotype, formate isotype, control anti–PD-1 (αPD-1), and formate anti–PD-1 (αPD-1). This analysis identified eight distinct tumor-infiltrating CD8⁺ T-cell populations across treatment groups by gene-expression signatures (Fig. 5B; Supplementary Fig. S11A–S11G). The individual cell clusters were annotated by expression patterns of genetic markers or gene-expression programs associated with functional properties (Supplementary Fig. S11A and S11B), including clusters with features of T-cell exhaustion.

Chronic antigen stimulation of CD8⁺ T cells in tumors (or during infections) drives progressive loss of effector functions toward a hypofunctional state known as exhaustion. Exhausted CD8⁺ T cells within the TME are heterogenous. A “stem-like” progenitor-exhausted cell population retains proliferative potential and undergoes self-renewal and/or differentiation into populations that retain some effector functions (30, 31). PD-1 pathway blockade increases progenitor cell proliferation and differentiation into highly functional transitory exhausted cells (32). This transitory population retains proliferative and effector functions but eventually differentiates into highly dysfunctional terminally exhausted cells.

We observed that metabolite supplementation with formate promoted the increase of transitory exhausted cells with enhanced proliferative and effector potential. Although each CD8⁺ T-cell cluster was represented across treatment conditions (Supplementary Fig. S11C–S11F), formate supplementation altered the frequencies of these cell populations in mice receiving either isotype control or anti–PD-1 antibodies (Fig. 5B and C). Formate treatment resulted in decreased frequencies of the CD8⁺PD-1⁺TCF-1⁺ progenitor exhausted cells, which retain proliferative potential and undergo self-renewal and/or differentiation into other exhausted populations (Fig. 5C). Therapeutic formate supplementation resulted in increased gene signature scores for cellular activation and cell cycle in mice receiving either isotype or anti–PD-1 relative to controls (Supplementary Fig. S11G–S11K). These data suggest formate treatment may function by improving the proliferative potential of tumor-infiltrating CD8⁺ T cells responding to anti–PD-1. Consistent with this hypothesis, formate combined with anti–PD-1 resulted in an increased frequency predominantly in the cell cluster containing genes associated with entering or exiting cell cycle (Fig. 5C). Formate combined with anti–PD-1 resulted in decreased frequencies of exhausted cell populations with gene signatures of interferon (IFN) signaling or dysfunctional cytotoxic activity (CD8⁺ T cells coexpressing Lag3 and Tnfrf9; Fig. 5C; ref. 33). Additionally, formate combined with anti–PD-1 was associated with reduced gene signatures for terminally exhausted cells or high IFN signaling (Supplementary Fig. S11L–S11O; ref. 30).

To identify whether changes in gene signature scores could be mapped to metabolic alterations associated with formate supplementation in combination with anti–PD-1, we projected a curated set of 61 Kyoto Encyclopedia of Genes and Genomes (KEGG) metabolic gene signatures onto all cells (34). Forty-six of the 61 metabolic pathways were differentially expressed in CD8⁺ T cells from tumors of anti–PD-1-treated mice compared with tumors of mice receiving combination formate and anti–PD-1 (Fig. 5D). These surprising findings indicated metabolic supplementation with formate combined with anti–PD-1 drives global metabolic pathway transcriptional rewiring of tumor-infiltrating CD8⁺ T-cell populations within the TME.

To determine the effects of formate within individual cell populations, we calculated the KEGG metabolic signature for each cell cluster and compared average scores in anti–PD-1-treated mice from control or formate treatment groups (Fig. 5E). Top differentially expressed KEGG signatures from the anti–PD-1-treated groups were analyzed from control (glycosphingolipid 3) and formate (glycolysis/gluconeogenesis)-treated groups. In anti–PD-1-treated mice, the increased expression of the glycosphingolipid 3 pathway was restricted to only half the cell clusters (Fig. 5F), but in the formate plus anti–PD-1-treated mice, the glycolysis/gluconeogenesis score was increased across most cell clusters (Fig. 5G). Interestingly, formate treatment alone drove many metabolic alterations without anti–PD-1 treatment (Supplementary Fig. S12A–S12F). Collectively, these findings demonstrate metabolic rewiring occurring after formate treatment can be restricted to individual cell clusters or differ globally across CD8⁺ T cells and is distinct between individual metabolic pathways.

To understand whether formate supplementation altered expression levels of genes involved in 1C metabolism, the expression of individual genes within this pathway was analyzed. Remarkably, although formate supplementation alone drove modest increases in gene expression of 1C pathway enzymes, formate combined with anti–PD-1 resulted in even greater expression of genes involved in serine biosynthesis, folate cycle, and purine synthesis (Fig. 5H). When the average expression of 1C pathway genes (Fig. 5H) was assessed within individual CD8⁺ T-cell populations, the increased expression driven by formate in anti–PD-1-treated mice was restricted to the proliferating and exhausted (IFN stimulated and dysfunctional) clusters (Fig. 5I). The expression levels of the individual genes for the rate-limiting enzymes for serine biosynthesis (Phgdh) and the folate cycle (Shmt2) were associated with the same groups of cell clusters (Fig. 5J and K). Together, these findings indicate formate supplementation during anti–PD-1 treatment can selectively modulate specific CD8⁺ T-cell subpopulations. Despite global metabolic pathway gene-expression changes, formate treatment in the absence of anti–PD-1 treatment had minimal effects on expression of genes in the folate cycle and serine biosynthesis pathways (Supplementary Fig. S12G–S12I), instead promoting the methionine cycle and transsulfuration (Supplementary Fig. S12J and S12K). Thus, our findings show that formate supplementation alters the transcription of 1C enzymes in a manner dependent on T-cell activation status, as well as the response of specific cell populations to anti–PD-1 treatment.

Given the changes in gene signatures associated with activation and cell cycle, we hypothesized formate was promoting proliferation and effector function of tumor-specific CD8⁺ T-cell populations. We examined CD8⁺ T-cell proliferation and effector potential in tumors 14 days after B16-OVA cell implantation in mice given formate relative to control mice beginning on day 9 and anti–PD-1 or isotype control on days 10 and 13 after tumor implantation (Fig. 6A). Tumor-specific CD8⁺ T cells were identified using fluorophore-conjugated H2-K^b:OVA_257–264 tetramers. Formate treatment alone did not lead to appreciable differences in tumor-specific T-cell numbers within the tumors (Fig. 6B). In contrast, formate combined with anti–PD-1 resulted in a marked increase in tumor-specific CD8⁺ T cells per mg of tumor (Fig. 6C). To validate whether the observed increase in gene signature scores for cell cycle (Supplementary Fig. S11G/J–K) was driven by increased cell proliferation, formate- and control-treated mice (with or without anti–PD-1 treatment) were parenterally administered the synthetic pyrimidine bromo-deoxy-uridine (BrdUrd) 12 hours prior to TIL isolation. Although formate therapy alone did not increase recently proliferating TILs, formate combined with PD-1 resulted in a significant increase in recently proliferating TILs (Fig. 6D and E).

The transcriptional profile of cells from mice receiving formate and anti–PD-1 suggested this combination may overcome exhaustion to improve antitumor immunity. To determine whether observed decreases in transcriptional features of exhaustion (Supplementary Fig. S11G/L–O) in mice treated with formate and anti–PD-1 promote effector function, we evaluated granzyme B expression in tumor-specific CD8⁺ T cells. CD8⁺ T cells increased expression of granzyme B once entering the tumor bed, as minimal levels were detected in CD44⁺CD8⁺ T cells from the draining lymph node (dLN in black fill; Fig. 6F and G). Combined formate and anti–PD-1 treatment led to increased numbers of granzyme B⁺ CD8⁺ T cells in tumors relative to controls (Fig. 6F and G). We also examined CD8⁺PD-1⁺TCF-1⁻TIM-3⁺CD101⁻ transitory exhausted T cells which have an effector-like transcriptional signature and retain proliferative and effector functions, including expression of the chemokine receptor CX3CR1, proinflammatory cytokines, and granzyme B. Recently proliferating transitory exhausted CD8⁺ T cells with effector properties can control chronic viral infections and cancer. These transitory exhausted T cells are the progeny of stem-like exhausted CD8⁺ T cells and can be identified by expression of the chemokine receptor CX3CR1 (35, 36). We compared the numbers of the CX3CR1-expressing tumor-specific CD8⁺ T cells in all treatment groups. Mice receiving combined formate and anti–PD-1 had an increased abundance of CX3CR1⁺ CD8⁺ T cells compared with mice receiving anti–PD-1 alone (Fig. 6H and I). This enhanced functional capacity was limited to tumor-specific CD8⁺ T cells, as no differences were observed between groups among tetramer-negative “bystander” cells (Supplementary Fig. S13A–S13H). TCF-1⁺ stem-like exhausted CD8⁺ T cells give rise to the proliferative burst after PD-1 blockade. Thus, we tested whether the number and frequency of TCF-1⁺TIM-3⁻ CD8⁺ T cells were altered following treatment with formate either alone or in combination with anti–PD-1. We did not find significant changes among conditions (Supplementary Fig. S14A–S14H; Supplementary Fig. S15A–S15H). These findings suggest formate and anti–PD-1 combination therapy improves CD8⁺ T-cell proliferative and cytolytic capacity by increasing the number of CX3CR1⁺ CD8⁺ T cells.

In this study, we identify 1C metabolism as a pathway that restricts antitumor CD8⁺ T-cell responses to anti–PD-1 therapy. Our findings demonstrate that impaired antitumor immunity can be overcome via supplementation with the 1C donor formate when combined with anti–PD-1. The combined therapeutic benefits of formate supplementation and anti–PD-1 therapy improve CD8⁺ T-cell–mediated B16-OVA tumor clearance. These findings reconfigure the conventional view whereby antagonizing folate metabolism reduces cancer cell proliferation (37). Importantly, we show that formate supplementation does not reduce the growth of B16-OVA tumors following CD8⁺ T-cell depletion. Our findings demonstrate that formate supplementation can enhance metabolic function in CD8⁺ T cells reinvigorated by anti–PD-1 therapy. Formate administration during PD-1 blockade results in enhanced B16-OVA tumor clearance.

T-cell metabolism is essential for T-cell activation and function by providing metabolic intermediates responsible for fueling T-cell proliferation and influencing the CD8⁺ T-cell epigenome (11, 12). However, characterizing metabolic fitness, especially in complex multicellular systems, has proven difficult (10). Using our unique coculture platform that allows rapid separation of tumor and T-cell populations, we analyzed cell type–specific metabolic events that occur upon CD8⁺ T-cell–tumor cell interactions and identified a previously unrecognized TCR-dependent role for serine biosynthesis following T-cell stimulation by tumor cells. These results are in line with previous reports demonstrating the link between TCR signaling and metabolic rewiring. First, GLUT1-dependent glucose utilization is linked to TCR signaling and enhanced by CD28 activation in both an AKT-dependent and -independent manner (38, 39). Second, mTOR activation through TCR signaling modulates purine synthesis and 1C metabolism via ATF4 expression (40). Thus, TCR signaling drives glucose utilization to support serine biosynthesis and promotes 1C metabolism via the mTOR pathway (41). Directly linking these signaling pathways with metabolic rewiring in T cells during early phases of activation and antitumor immunity are important areas of future investigation.

Our studies demonstrate a reduction of serine in the TME of B16-OVA tumors, which has the potential to restrict CD8⁺ T-cell antitumor immunity. Our results suggest that within B16-OVA tumors, where serine and glucose levels are lowered, therapeutic supplementation of formate can enhance the efficacy of anti–PD-1. As serine metabolism varies among different cancer subtypes, the resulting restriction of serine within the TME may be dependent on tumor type (28). Further studies are required to understand how serine meta­bolism in different types of cancer cells can shape metabolite concentrations in the TME and whether these metabolites in the TME affect interactions with immune cells. In our studies, the effect of formate supplementation combined with anti–PD-1 varied across tumor models. One possible explanation is that differences in serine metabolism in tumor types may drive differential outcomes of combined formate and anti–PD-1 treatment in tumor models such as EO771 and 4T1. In addition, multiple factors (i.e., other immunogenic, metabolic substrate, or host factors) may contribute to variability in the observed therapeutic effect of formate and anti–PD-1 (32). Future studies addressing the role of each of these important factors are necessary to extend the potential benefits of formate supplementation in combination with anti–PD-1 to other tumor types.

Transcriptional analysis of individual tumor-infiltrating CD8⁺ T cells revealed formate supplementation combined with anti–PD-1 therapy drives the expression of genes for multiple metabolic pathways. Although the transcription of some metabolic pathway genes increased across all CD8⁺ T-cell populations, other changes were limited to specific CD8⁺ T-cell populations. Changes in metabolic pathway transcriptional activities were associated with changes in gene-expression profiles related to cellular activation and proliferation. Notably, formate administration in combination with anti–PD-1 upregulated genes involved in the 1C metabolic pathways of the folate cycle and purine synthesis. Interestingly, our finding that 1C metabolites increase gene expression of metabolic enzymes also has been seen in Caenorhabditis elegans and Sacchromyces cerevisiae, suggesting an evolutionarily conserved mechanism of metabolic pathway autoregulation (42–44).

The metabolic changes driven by formate treatment combined with anti–PD-1 in mice with existing tumors also altered exhausted CD8⁺ T-cell populations. Combined formate supplementation and anti–PD-1 drove an increase in transitory exhausted intratumoral (CX3CR1⁺) CD8⁺ T cells that retained proliferative capacity and cytotoxic potential compared with other exhausted T-cell subsets. These CX3CR1⁺ CD8⁺ T cells are critical in mediating host protection from viruses and tumors. Gene-expression profiles of exhausted cell populations and gene signatures of cell cycle and immune cell activation were associated with the metabolic changes driven by formate therapy. Recent studies have demonstrated that CD8⁺ T cells differentiate and acquire their effector function as they invade the tumor bed (45). Our findings suggest formate supplementation may support CD8⁺ T cells during their transition from activation through effector differentiation. Together, these findings indicate the enhanced tumor clearance afforded by combined formate and anti–PD-1 treatment is due to the selective support of a specific exhausted CD8⁺ T-cell population. Investigations into the augmentation of other metabolic pathways within this CD8⁺ T-cell subset may provide new therapeutic avenues to further extend the benefits of ICB in cancer therapy. Although metabolic supplementation with formate during anti–PD-1 treatment markedly enhanced tumor clearance in a CD8⁺ T-cell–dependent manner, formate supplementation may also have effects on other immune cells (non-CD8⁺ T-cell), the microbiome, and/or whole animal physiology in antitumor immunity. In addition, features of the TME, such as extracellular acidosis, can also impact 1C metabolism activity leading to augmented T-cell function (46). Further studies are needed to examine the effects of formate on anti–PD-1 therapy in each of these contexts.

In summary, we demonstrate that formate supplementation can improve the efficacy of anti–PD-1 therapy. Combination therapy with formate and anti–PD-1 promoted CD8⁺ T-cell–mediated tumor control and animal survival in the B16-OVA tumor model. Our results in this model, show that formate has the potential to enhance the reinvigoration of exhausted CD8⁺ T cells following anti–PD-1 therapy and promote tumor growth control. Collectively, our studies identify a novel strategy of combining metabolic support in the form of formate with anti–PD-1 to extend the benefits of this therapy. These studies provide a foundation for further investigations into how therapeutic approaches aimed at supporting the metabolic fitness of CD8⁺ T cells can synergize with immune-based therapies.

B16 (RRID:CVCL_F936) and B16 cells expressing chicken ovalbumin (B16-OVA; ref. 47), EO771 cells (CH3 Biosystems; cat. 94A001; RRID:CVCL_GR23), or 4T1 cells (American Type Culture Collection; cat. CRL-2539; RRID:CVCL_0125) were grown in DMEM (Thermo Fisher Scientific; cat.11995065) supplemented with 10% FBS, 100 units/mL penicillin, and 100 mg/mL streptomycin (Thermo Fisher Scientific; cat.5140-122). For in vivo experiments, tumor cells were thawed and implanted into mice within one to two in vitro passages. For coculture experiments, B16-OVA cells were cultured with OT-1 cells in RPMI-1640 (Thermo Fisher Scientific; cat.11875-119) supplemented with 10% dialyzed FBS (Thermo Fisher Scientific; cat.1743489), 100 units/mL penicillin and 100 mg/mL streptomycin (Thermo Fisher Scientific; cat.5140-122), 10 mmol/L HEPES (Thermo Fisher Scientific; cat.15630-130), 1 mmol/L sodium pyruvate (Thermo Fisher Scientific; cat.11360-070), and 55 μmol/L 2-mercaptoethanol (Thermo Fisher Scientific; cat.21985–023; Complete RPMI). All cells were confirmed Mycoplasma negative using PCR-based assays.

C57BL/6J (cat.000664; RRID:IMSR_JAX000664) and BALB/c (cat. 000651; RRID:IMSR_JAX000651) mice were purchased from The Jackson Laboratory. Cas9-expressing mice (29) were bred within our animal facility. OT-1 (C57BL/6-Tg(TcraTcrb)1100Mjb/J; RRID:IMSR_JAX:003831) were crossed with Cas9-expressing and CD45.1 congenic (C57BL/6.SJL-Ptprc^aPepc^b/BoyJ; RRID:IMSR_JAX:002014) to generate OT-1/Cas9/CD45.1 mice. Animals were maintained in a pathogen-free facility and used according to institutional and NIH guidelines. Harvard Medical School is accredited by the American Association of Accreditation of Laboratory Animal Care. All studies were approved by the Harvard Medical School Institutional Animal Care and Use Committee.

CD8⁺ T cells were purified from spleens of OT-1 mice using a naive CD8⁺ cell isolation kit (Miltenyi Biotec; cat.130-096-543). Isolated naive CD8⁺ T cells were activated for 72 hours at 37°C in plates coated with 5 μg/mL anti-CD3 (clone 145-2C11; Bio X Cell; cat.BE0001-1; RRID:AB_2714218) and 5 μg/mL anti-CD28 (clone 37.51; Bio X Cell; cat. BE0015-1; RRID:AB_1107624) in the presence of 100 units/mL IL2 (R&D Systems; cat. 202-IL-050) and 10 ng/mL IL12 (PeproTech; cat. 210-12-50 mg; unless otherwise stated). CD8⁺ T cells were cultured either in complete RPMI or CM. CM was obtained from tumor cells cultured for 24 hours at ∼80% confluency and filtered using 0.2-μm filters.

For in vitro killing assays, 15,000 tumor cells were plated in 96-well plates and incubated at 37°C until cells adhered. 24 hours later, a test plate was counted and preactivated (with plate-bound anti-CD3/ anti-CD28 plus IL2 and IL12 as indicated above) OT-1 CD8⁺ T cells were plated on top using a 1:5 ratio (CD8⁺:tumor cells). Following coculture for 24 hours, plates were trypsinized, tumor cells and CD8⁺ T cells were resuspended in PBS (Invitrogen; cat.14190-250) and analyzed by flow cytometry to determine the numbers of cancer cells and T cells. Percent killing was calculated as previously described (47).

For studies using the SHMT inhibitor (ref. 14; KGD-112) or using RPMI serine and glycine-deficient media (Gibco, custom media), CD8⁺ T cells were activated with anti-CD3/anti-CD28 in the presence of IL2 and IL12 for 72 hours. In all in vitro studies with formate supplementation, sodium formate (Sigma-Aldrich; cat. 456020) was added at a concentration of 1 mmol/L.

Eight-week-old C57BL/6J female mice were anesthetized with 2.5% 2,2,2-tribromoethanol (MilliporeSigma; cat. T48402-25 g), and 250,000 B16-OVA tumor cells were injected in the flank subcutaneously. For orthotopic breast cancer models: 8-week-old C57BL/6J female mice received 1,000,000 EO771 tumor cells and 8-week-old BALB/c female mice received 100,000 4T1 cells (BALB/c) implanted into the mammary fat pad following anesthesia as indicated above. In mice receiving formate supplementation, sodium formate (Sigma; cat. 456020-25G) was added to the drinking water at a concentration of 5 mg/mL. Treatment was started on either day −1 or day 8 after tumor implantation and continued until tumors reached 2,000 mm³. Depletion of CD8⁺ T cells was achieved by administering anti-mouse CD8β antibodies (clone 53–5.8; Bio X Cell; cat. BE0223; RRID:AB_2687706) at 10 mg/kg/dose on days 0, 2, 4, 6, and 8. anti–PD-1 (CD279; clone 29F.1A12; RRID:AB_2687796; ref. 48) or isotype control rat IgG2a (clone 2A3, Bio X Cell; cat. BP0089; RRID:AB_1107769) was administered at days 10 and 13 at 5 mg/kg/dose. Tumor volume was measured with a caliper every 2 to 3 days and calculated using the formula ½ D × d², where D is longer and d is shorter diameter. Mice with tumors exceeding 2,000 mm³ or severely ulcerated, and mice with body condition score ≥2 were sacrificed. In figures, average growth curves are displayed until one animal exhibited a tumor volume of 2,000 mm³. The metabolic composition of TIF and plasma was analyzed on day 14 or 21 as previously described (26). Tumor dissociation and TIL isolation were performed as previously described (47).

For tumor coculture experiments, previously activated CD8⁺ T cells were plated at a 1:5 ratio with tumor cells with 11 mmol/L of ¹³C-glucose (Cambridge Isotope Laboratories, cat. CLM-1396) for six hours prior to cell separation from coculture plates as previously described (26). For experiments using SIINFEKYL peptide (InvivoGen, cat. vac-sin)-pulsed splenocytes, splenocytes were first cultured for 1 hour at 37°C with the indicated peptide concentrations and then incubated with naive CD45.1^+/+ OT-1 cells for 3 days (with ¹³C-glucose for metabolomics experiments). Anti-CD45.2-PE (BioLegend, cat. 109808, clone 104; RRID:AB_313445) was added to cell culture 30 minutes prior to selection with anti-PE microbeads (Miltenyi Biotec, cat. 103-105-639) according to the manufacturer's protocol. Cell pellets containing enriched OT-1 populations were subsequently used for either western blot or metabolomics experiments. Quenching, metabolite extraction, and LC-MS analysis were performed on isolated cells from either coculture or magnetic separation as previously described (26).

Metabolites were extracted from cell pellets by adding 800 μL 60% methanol containing 6.67 μg/mL glutaric acid (Sigma-Aldrich, G3407) in LC/MS grade water (Fisher Scientific, W6500). After the addition of 500 μL chloroform (Sigma-Aldrich, C2432), samples were vortexed for 10 minutes at 4°C and then spun at 17,000 × g for 10 minutes at 4°C. The upper phase was transferred into a fresh tube and dried down in a Vacufuge plus speed-vac at 4°C. The metabolite extract was separated using an iHILIC-(P) Classic column (2.1 μm, 150 mm × 2.0 mm I.D., The Nest Group) coupled to a Thermo Scientific SII UPLC system. The autosampler and column oven were held at 4°C and 25°C, respectively. The iHILIC-(P) Classic column was used with buffer A (0.1% ammonium hydroxide, 20 mmol/L ammonium carbonate) and buffer B (100% acetonitrile). The chromatographic gradient was run at a flow rate of 0.150 mL/minute as follows: 0–20 minutes: linear gradient from 80% to 20% B; 20–20.5 minutes: linear gradient from 20% to 80% B; 20.5–28 minutes: hold at 80% B. The mass spectrometer was operated in a full-scan, negative-ion model. Mass spectrometry detection was carried out on a Q Extractive HF-X orbitrap mass spectrometer with a HESI source. For metabolite quantification, TraceFinder software (Thermo Fisher) was used. The lower phase after chloroform extraction was also dried down in a Vacufuge plus speed-vac at 4°C and proteins extracted from the pellets in 100 μL RIPA buffer (Pierce, 89901) by vortexing for 10 minutes at 4°C. Samples were then spun at 17,000 × g for 10 minutes at 4°C, the supernatant was transferred to a fresh tube, and the protein concentration was determined using the DC Protein Assay (Bio-Rad, 500-0112). The metabolite levels were normalized by the total protein amount per sample in μg. GraphPad Prism 9 software was used for statistical analysis.

Single-cell suspensions of TIL were prepared by mechanical dissociation using gentleMACS system (Miltenyi Biotec) with 2 mg/mL Collagenase I (Worthington Biochemical; cat. LS004194) and resuspended in staining buffer (PBS containing 1% FBS and 2 mmol/L EDTA) and stained with the indicated antibodies. From BD Biosciences BUV395 anti-mouse CD3ε (cat. 563565; clone 145-2C11; RRID:AB_2738278); BUV496 anti-mouse CD4 (cat. 612952; clone GK1.5; RRID:AB_2813886); BUV805 anti-mouse CD8α (cat. 564920; clone 53-6.7; RRID:AB_2716856); AF488 anti-mouse TCF-7/TCF-1 (cat. 567018; clone S33-966; RRID:AB_2916388) From BioLegend, BV421 anti-mouse CX3CR1 (cat. 149023; clone SA011F11; RRID:AB_2565706), BV510 anti-mouse PD-1 (cat. 135241; clone 29F.1A12; RRID:AB_2715761), BV605 anti-mouse CD62 L (cat. 104437; clone MEL-14; RRID:AB_11125577), BV711 anti-mouse TIM-3 (cat. 119727; clone RMT3-23; RRID:AB_2716208), BV785 anti-mouse CD44 (cat.103059; clone IM7; RRID:AB_2571953), FITC anti-mouse CD8α (cat. 100706; clone 53-6.7; RRID:AB_312745), PerCP-Cy5.5 anti-mouse Ki-67 (cat. 652424; clone 16A8), PE anti-mouse CD45.2 (cat. 109808; clone 104), PE anti-mouse CD44 (cat. 103024; clone IM7; RRID:AB_493687), APC anti-mouse CD45.1 (cat. 110714; clone A20; RRID:AB_313503), and AF700 anti-mouse granzyme B (cat. 372222; clone QA16A02; RRID:AB_2728389). The APC/Cy7 Live/Dead fixable near-IR dead cell stain kit (Thermo Fisher Scientific; cat. L34976) was used to determine cell viability. Cell division was determined using the CellTrace Violet Cell Proliferation kit (Thermo Fisher; cat. C34557). Flow cytometry was performed on BD LSR II and BD FACS Symphony machines. FACS sorting was conducted on a BD Aria II or MoFlo Astrios EQ. Flow cytometry analyses were performed using FlowJo 10.6.1 software (TreeStar; RRID:SCR_008520). All antibodies were used at either a 1:100 or 1:200 dilution. UltraComp beads (Thermo Fisher Scientific; cat.01-2222-42) were used for compensation.

Targeted deletion of the Phgdh gene was accomplished using CHIME (29). gRNA sequences were generated using the Genetic Perturbation Platform (Broad Institute, Cambridge, MA; https://portals.broadinstitute.org/gpp/public) CRISPick design tools. The top-scoring gRNA targeting Phgdh (sequence: GCCACACTGAGAGCCTACCTG) was used in the CHIME system. Hematopoietic stem cells were sort purified by the absence of lineage markers (Lin⁻), c-Kit⁺, and Sca-1⁺ (LSK) from Cas9-expressing OT-1 mice and transduced with lentiviral vectors encoding guide RNAs (gRNA) targeting Phgdh or no genome target (control) and violet-excited GFP (Vex) as a marker of transduction efficiency. Bone marrow chimeras were made in lethally irradiated (1,200 rads) recipient mice using transduced LSK cells. Following 8 weeks the reconstituted hematopoietic compartment contained naive TCR-transgenic T cells in which gRNA expression and gene deletion occurred. Transduced naive OT-1 cells were subsequently purified using naive CD8⁺ negative selection (Miltenyi Biotec; cat.130-096-543) and sorted (Vex⁺CD44^lo) for adoptive transfer (in vivo) or coculture (in vitro) experiments.

Enriched CD8⁺ T cells were lysed in Pierce RIPA Buffer supplemented with Halt Protease and Phosphatase Inhibitor Cocktail (100×) for 15 minutes on ice. Whole-cell lysates were spun down at 20,817 × g at 4°C for 15 minutes. Supernatants were collected and transferred to new Eppendorf tubes, and 10 μL of each lysate was taken for protein estimation using the Pierce BCA Protein Assay Kit to normalize for protein loading. Cleared lysates were denatured with 4× Laemmli Sample Buffer (Bio-Rad) or 4× NuPAGE LDS (Invitrogen) containing 2-mercaptoethanol (BME) and boiled for 5 minutes at 95°C. Fifteen to 40 μg of protein per lysate was loaded and run on a NuPAGE 4%–12% Bis-Tris protein gel and then transferred onto a nitrocellulose membrane. Ponceau staining was performed to check transfer efficiency and protein loading. Membranes were then blocked for 1 hour in Tris Buffered Saline (TBS) supplemented with 1% Tween (TBS-T) and 5% milk or in LI-COR Intercept Blocking Buffer supplemented with 0.2% Tween at room temperature. Membranes were then incubated with the following primary antibodies: SHMT1 (Novus Biologicals; 1:1,000; cat. NBP2-32173), SHMT2 (Novus Biologicals, 1:1,000, cat. NBP1-80754; RRID:AB_11031453), PHGDH (Cell Signaling Technologies; 1:1,000; cat. 13428; RRID:AB_2750870), Histone H3 (Cell Signaling Technologies; 1:1,000; cat.9715; RRID:AB_331563), and β-Actin (Cell Signaling Technologies; 1:1,000; cat. 3700; RRID:AB_2242334) overnight. Membranes were rocked overnight in blocking buffer at 4°C, then washed 3× in TBS-T buffer and incubated with HRP-linked secondary antibodies, HRP-linked anti-rabbit IgG (Cell Signaling Technologies; 1:2,000; cat. 7074; RRID:AB_2099233), HRP-linked anti-mouse IgG (Cell Signaling Technologies; 1:2,000; cat.7076; RRID:AB_330924), and HRP-linked anti-rat IgG (Cell Signaling Technologies; 1:2,000; cat. 7077; RRID:AB_10694715) in blocking buffer for 1 hour at room temperature. Membranes were then treated with SuperSignal West Pico PLUS Chemiluminescent Substrate (Thermo Fisher; cat. 34580) and imaged using the Amersham Imager 600. Membranes incubated with fluorescent secondary antibodies (donkey anti-rabbit IgG; LI-COR; cat. 926-32213 or goat anti-mouse; LI-COR; cat. 926-68070) were directly imaged on LI-COR Odyssey CLx following TBS-T washes. When necessary, membranes were stripped with Restore PLUS Western Blot Stripping Buffer (Thermo Scientific; cat. 46430) for 15 minutes at room temperature, blocked, and reprobed with primary and secondary antibodies as described above.

Congenically marked OT-1 CD45.1⁺ or CD45.2⁺ CD8⁺ T cells generated using CHIME were isolated from mouse spleens. The use of congenic markers allowed us to distinguish adoptively transferred CD8⁺ T cells from endogenous CD8⁺ T cells by flow cytometry, and distinguish between Control-gRNA and Phgdh-gRNA transduced cells. To evaluate whether specific gene deletion affected the competitive advantage of CD8⁺ T cells in the TME, 10,000 OT-1 Control-gRNA and 10,000 OT-1 Phgdh-gRNA cells were cotransferred intravenously to 8-week-old Cas-9 expressing recipient mice prior to tumor implantation. The following day, 250,000 B16-OVA tumor cells were injected in the flank subcutaneously. Fourteen days after tumor implantation, tumor dissociation and TIL isolation were performed as previously described (49). Data were collected using either BD LSR II and BD FACS Symphony machines and analyzed using FlowJo 10.6.1 software (TreeStar; RRID:SCR_008520).

CD8⁺ T cells were sort purified from dissociated B16-OVA tumors on day 12 after implantation for single-cell analysis. B16-OVA tumors were manually digested in Collagenase I (Worthington Biochemical; cat. LS004194) and incubated for 20 minutes with gentle rocking at 37°C. Cells were stained with Live/Dead near-IR dead cell stain (Thermo Fisher Scientific; cat. L34976), PE anti-mouse CD45.2 (BioLegend; cat. 109808; clone 104; RRID:AB_313445), and FITC anti-mouse CD8α (BioLegend; cat.100706; clone 53-6.7; RRID:AB_312745) for sort purification of viable CD45⁺CD8α⁺ cells. Isolated cells from 3 mice were pooled from each group and ∼12,000 cells were loaded on the Chromium Controller (10X Genomics) for a target cell recovery of ∼8,000 cells and processed according to the manufacturer's instructions. One sample (control/anti–PD-1) failed this step. All remaining samples were sequenced on an Illumina NovaSeq 500 sequencer.

Statistical analyses were performed in GraphPad Prism 8.3.1 (RRID:SCR_002798). A two-tailed unpaired Student t test was used for comparison of two unpaired groups. A two-way ANOVA with Sidak's multiple comparison test was used for the comparison of multiple groups. Error bars display standard error measurement unless otherwise noted. P values are noted as *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

Sample demultiplexing, barcode processing, alignment, filtering, and UMI counting were performed using the Cell Ranger pipeline (v5.0.0; RRID:SCR_017344). Genome alignment, QC metrics, and library normalization of the scRNA-seq data were performed as previously reported (34). Data were analyzed by clustering cells to characterize subpopulations and determine gene-expression patterns changing with formate and anti–PD-1 treatment (50). Data preprocessing and filtering were performed using Python (v3.8.5; RRID:SCR_008394) and Scanpy (v1.5.1; RRID:SCR_018139). After merging all samples into a Seurat object, single-cell transcriptomes were filtered using four quality control metrics. Cells in which fewer than 500 unique genes were detected or where mitochondrial encoded transcripts represented greater than 5% of the total library were removed from the analysis. Genes detected in fewer than 1% of all cells across the dataset were removed. Mitochondrial genes were included if they began with the letters “mt-.” Finally, cells were excluded based on expression housekeeping genes (list available on the Satija laboratory; https://satijalab.org/seurat), with cells passing criteria with expression >0 for more than half of genes.

Normalization, integration, and clustering were performed using the R (v3.6.2; RRID:SCR_001905) package Seurat (v3.2.2; RRID:SCR_016341). Data normalization was performed using the SCTransform function on samples separately. Variable features were chosen using the SelectIntegrationFeatures function in Seurat, with nfeatures set to “Inf,” followed by the PrepSCTIntegration function to ensure Pearson residuals required for downstream analysis were calculated. Anchors between datasets were identified by the FindIntegrationAnchors function and passed to the IntegrateData function using normalization.method = “SCT” to produce an integrated dataset. Dimensional reduction was accomplished by performing the principal component analysis on the integrated dataset and then using the first 30 principal components for Uniform Manifold Approximation and Projection (UMAP) using default parameters associated with the RunUMAP function. Unsupervised clustering was done by constructing a shared nearest-neighbor graph using the FindNeighbors function and then performing graph-based clustering using the default algorithm with resolution = 0.3 by the FindClusters function. Differential expression analysis between clusters and treatment conditions was performed using a Wilcoxon rank sum test using the sc.tl.rank_genes_groups in Scanpy. Each cluster was assessed for the presence of genes associated with CD8⁺ T cells. One cluster was determined to be a contaminating cluster of dendritic cells and excluded.

Population sizes among treatment conditions were performed using a two-sided exact binomial test. Metabolic and signaling pathway gene signatures were curated from the KEGG subset of canonical pathways from the C2 collection within MSigDB. Analyses of metabolic pathway transcriptional activity utilized a computational approach developed by our labs to layer KEGG metabolic gene signatures over scRNA-seq data to define single-cell metabolic state and correlate with CD8⁺ T-cell differentiation and functionality by group and within cell clusters (34). Immune signatures were curated from the C7 collection of immunologic signatures within MSigDB. Single-cell signature scores were calculated using the sc.tl.score_genes function from Scanpy.

All code used to process the scRNA-seq data can be found on GitHub at www.github.com/oshahid/Formate_PD1Blockade. Raw and processed scRNA-seq data files were uploaded to the Gene Expression Omnibus repository within the GSE240231 reference series.

J.S. Park reports grants from the NIH/NCI during the conduct of the study. J.D. Rabinowitz reports grants, personal fees, and nonfinancial support from Rafael Holdings, the Barer Institute, and Farber Partners and grants from the NIH during the conduct of the study; personal fees from LEAF Pharmaceuticals, Empress Therapeutics, and Raze Therapeutics, personal fees and nonfinancial support from Bantam Pharmaceuticals, Third Rock Ventures, and Faeth Therapeutics, and grants and nonfinancial support from Princeton University–PKU Collaboration outside the submitted work; and a patent for compositions and methods for enhancing immunotherapy pending, licensed, and with royalties paid from Farber Partners; Farber Partners is a subsidiary of the Barer Institute and Rafael Holdings, and these entities have provided support and financial compensation to J.D. Rabinowitz. G.J. Freeman reports grants from the NIH during the conduct of the study; personal fees from Roche, Bristol Myers Squibb, Origimed, Triursus, iTeos, NextPoint, IgM, Jubilant, Trillium, GV20, IOME, and Geode outside the submitted work; a patent for PD-L1/PD-1 cancer immunotherapy issued, licensed, and with royalties paid from Roche, a patent for PD-L1/PD-1 cancer immunotherapy issued, licensed, and with royalties paid from Merck/MSD, a patent for PD-L1/PD-1 cancer immunotherapy issued, licensed, and with royalties paid from Bristol Myers Squibb, a patent for PD-L1/PD-1 cancer immunotherapy issued, licensed, and with royalties paid from Merck KGA, a patent for PD-L1/PD-1 cancer immunotherapy issued, licensed, and with royalties paid from Boehringer Ingelheim, a patent for PD-L1/PD-1 cancer immunotherapy issued, licensed, and with royalties paid from AstraZeneca, a patent for PD-L1/PD-1 cancer immunotherapy issued, licensed, and with royalties paid from Dako, a patent for PD-L1/PD-1 cancer immunotherapy issued, licensed, and with royalties paid from Leica, a patent for PD-L1/PD-1 cancer immunotherapy issued, licensed, and with royalties paid from the Mayo Clinic, a patent for PD-L1/PD-1 cancer immunotherapy issued, licensed, and with royalties paid from Eli Lilly, and a patent for PD-L1/PD-1 cancer immunotherapy issued, licensed, and with royalties paid from Novartis; and equity in companies pursuing immunotherapies: NextPoint, Triursus, Xios, iTeos, IgM, Trillium, Invaria, GV20, and Geode. M.C. Haigis reports grants from Roche during the conduct of the study, as well as a patent for the topic of 1C metabolism issued. A.H. Sharpe reports grants from the NIH (P50 CA101942, U54 CA224088, R01 CA276866, and P01 AI56299), Roche, the Ludwig Center at Harvard, the Glenn Foundation for Medical Research, the Cancer Research Institute (CRI4006 and CRI 3352), and the European Molecular Biology Organization (ALTF-1078-2017) during the conduct of the study; personal fees from Surface Oncology, Sqz Biotech, Selecta, Monopteros, Elpiscience, Bicara, Fibrogen, Alixia, IOME, GSK, Janssen, and Amgen, grants from Merck, Roche, Ipsen, Novartis, Quark Ventures, AbbVie, Moderna, Erasca, and Vertex, and other support from Corner Therapeutics outside the submitted work; patent 7,432,059 with royalties paid from Roche, Merck, Bristol Myers Squibb, EMD Serono, Boehringer Ingelheim, AstraZeneca, Leica, the Mayo Clinic, Dako, and Novartis, patent 7,722,868 with royalties paid from Roche, Merck, Bristol Myers Squibb, EMD Serono, Boehringer Ingelheim, AstraZeneca, Leica, the Mayo Clinic, Dako, and Novartis, patent 8,652,465 licensed to Roche, patent 9,457,080 licensed to Roche, patent 9,683,048 licensed to Roche, patent 9,815,898 licensed to Novartis, patent 9,845,356 licensed to Novartis, patent 10,202,454 licensed to Novartis, patent 10,457,733 licensed to Novartis, patent 9,580,684 issued, patent 9,988,452 issued, patent 10,370,446 issued, patent 10,457,733 issued, patent 10,752,687 issued, patent 10,851,165 issued, and patent 10,934,353 issued; is on the scientific advisory boards for the Massachusetts General Cancer Center, the Program in Cellular and Molecular Medicine at Boston Children's Hospital, the Human Oncology and Pathogenesis Program at Memorial Sloan Kettering Cancer Center, and the Bloomberg-Kimmel Institute for Cancer Immunotherapy; and is an academic editor for the Journal of Experimental Medicine. No disclosures were reported by the other authors.

J.H. Rowe: Conceptualization, formal analysis, methodology, writing–original draft, writing–review and editing. I. Elia: Concep­tualization, formal analysis, methodology, writing–review and editing. O. Shahid: Formal analysis, investigation, writing–review and editing. E.F. Gaudiano: Formal analysis, investigation, writing–review and editing. N.E. Sifnugel: Formal analysis, investigation, writing–review and editing. S. Johnson: Formal analysis, investigation, writing–review and editing. A.G. Reynolds: Investigation. M.E. Fung: Formal analysis, investigation, writing–review and editing. S. Joshi: Data curation, formal analysis, investigation, writing–review and editing. M.W. LaFleur: Investigation, metho­dology, writing–review and editing. J.S. Park: Investigation. K.E. Pauken: Data curation, visualization, writing–review and editing. J.D. Rabinowitz: Resources, investigation, methodology, writing–review and editing. G.J. Freeman: Resources, investigation, writing–review and editing. M.C. Haigis: Conceptualization, resources, data curation, supervision, funding acquisition, methodology, writing–original draft, writing–review and editing. A.H. Sharpe: Conceptualization, resources, formal analysis, funding acquisition, investigation, writing–review and editing.

We thank all members of the Sharpe and Haigis Laboratories for providing invaluable feedback regarding the project. We acknowledge the Harvard Medical School Biopolymers Facility for assistance with sequencing experiments, and the Harvard Department of Immunology Flow Cytometry Core for assistance with flow cytometry and cell sorting.

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.