In order to identify the glycolytic enzymes that play the largest role in Treg differentiation, we analyzed multiple publicly available transcriptomic and proteomic datasets. We first investigated an RNA sequencing (RNA-seq) dataset that compared human naïve CD4 cells cultured under either Th0 or Treg polarizing conditions ³¹. After library normalization, we examined genes included in the “Glycolysis and Gluconeogenesis” gene ontology (GO) term and found that PGAM1 was the most differentially expressed glycolytic gene, with upregulation in in vitro-induced Tregs (iTregs) compared to conventional Th0 cells (Figure 1A). To determine whether PGAM expression is similarly upregulated in ex vivo-derived Tregs, we examined a single cell RNA-seq (scRNA-seq) dataset of peripheral blood mononuclear cells (PBMCs) from human healthy controls ³² and compared glycolytic gene expression between Tregs and total CD4 cells. After removing glyceraldehyde-3-phosphate dehydrogenase (GAPDH) for scaling purposes, we found that PGAM1 was similarly upregulated in Tregs (Figure 1B). A proteomics dataset from human healthy donor PBMCs ³³ demonstrated significantly upregulated protein levels of PGAM1 and PGAM2 in Tregs compared to total CD4 cells (Figure 1C), confirming findings from the mRNA level.

The differential upregulation of PGAM in Tregs suggested that it might play a physiologic regulatory role in these cells. We therefore wondered whether PGAM inhibition impacts Treg differentiation. To answer this question, we used epigallocatechin gallate (EGCG), a known inhibitor of PGAM ³⁴. We cultured naïve murine CD4 cells under Treg polarizing conditions with either vehicle (solvent alone) or EGCG and assessed Treg differentiation by quantifying the percentage of CD25⁺FOXP3⁺ cells by flow cytometry after 96 hours. Treatment with EGCG significantly reduced Treg differentiation (Figure 1D) without affecting proliferation or cell viability (Figure S1). To validate the effects of pharmacologic inhibition with genetic knockdown, we utilized antisense oligonucleotides (ASOs) targeting PGAM to reduce gene expression. Experiments conducted with fluorescently-tagged ASOs demonstrated near-complete ASO uptake by cultured CD4 cells after 24 hours (Figure S2). Compared to scrambled control ASOs, we found that ASOs targeting PGAM reduced PGAM1 protein levels (Figure S3A) and significantly reduced Treg differentiation assessed by flow cytometry (Figure 1E). Because glycolysis is crucial for cell health, we wanted to know if PGAM knockdown was affecting survival and/or proliferation. We found that PGAM ASOs did not affect cell survival and led to a minor increase in cell proliferation (Figures S3B and S3C).

To gain further insights into the effects of PGAM knockdown on Treg differentiation, we performed RNA-seq transcriptional profiling from unsorted naïve murine CD4 cells cultured under Treg polarizing conditions with either scrambled or PGAM-specific ASOs. PCA analysis showed that PGAM ASOs led to major transcriptional changes (Figure 1F), with major effects on MYC targets and mTORC1 signaling (Figure 1G), among other changes. We then performed differential expression analysis on a subset of genes important in the Treg-Th17 axis and found that cells treated with PGAM ASOs had significantly higher expression of pro-inflammatory genes such as Il17, Il22, and Il9, while significantly lower expression of genes involved in Treg differentiation such as Foxp3 and Izkf3 (Figure 1H). A complete list of differentially expressed genes can be found in Table S1.

Given the observed changes in transcriptional profile and differentiation markers, we next asked whether PGAM knockdown leads to functional changes in Treg suppressive activity. Compared to treatment with scrambled ASOs, PGAM ASO-treated Tregs were less effective at suppressing effector T (Teff) cell proliferation, indicating diminished suppressive function (Figure 1I). To determine whether PGAM expression is associated with Treg suppressive activity in human disease, we examined a published scRNAseq dataset of tumor-infiltrating Tregs (TIL-Tregs) isolated from surgically resected non-small cell lung cancer ³⁵. Among the 10 distinct subsets of TIL-Tregs identified in this study, one subset, the OX40^hiG-ITR^hi cluster [termed “Activated (1)” by the authors] had markedly increased suppressive activity and was enriched in patients who were resistant to immune checkpoint blockade. Strikingly, PGAM1 was significantly overexpressed in the highly suppressive OX40^hiGITR^hi cluster compared to the 9 other TIL-Treg subsets (Figure 1J). Moreover, PGAM1 expression was selectively increased in this subset relative to proximal, rate-limiting glycolytic enzymes, suggesting a unique association between PGAM expression and Treg suppressive activity.

Taken together, these findings demonstrate a critical role for PGAM in promoting Treg differentiation and suppressive function, in vitro as well as in healthy and diseased human states. The positive regulatory role of PGAM in Tregs was intriguing, since glycolysis has generally been associated with decreased Treg function. Indeed, whereas inhibition or knockdown of PGAM led to decreased Treg differentiation, our own prior work found that similar inhibition of GAPDH, which catalyzes a reaction only two steps upstream from PGAM in glycolysis, had the opposite effect²⁸ – a finding we replicated here with the GAPDH inhibitor koningic acid (Figure S4). Consistent with this observation, the ratio of PGAM1 to GAPDH mRNA levels increased in human iTregs vs. Th0 cells (Figure S5A), and PGAM2 protein levels increased out of proportion to GAPDH (and other glycolytic enzymes) in ex vivo Tregs (Figure 1C). Similarly, a review of published data from the ImmPres immunological proteome resource (http://immpres.co.uk/)³⁶ shows increased PGAM1/GAPDH protein copy number ratio in murine iTregs versus Th17 cells (Figure S5B), further implicating a specific role for PGAM in the Treg-Th17 axis. The discordant roles of GAPDH and PGAM in Tregs led us to focus on the glycolytic metabolites downstream of GAPDH and upstream of PGAM, as the concentrations of these metabolites would be differentially impacted by the two enzymes (Figure 2A).

PGAM metabolizes the glycolytic intermediate 3PG into 2PG ^37,38. In addition to its role in glycolysis, 3PG is the starting substrate for the de novo serine synthesis pathway ^39,40. Given previous work implicating various roles for serine in T cells ^41–43, we hypothesized that PGAM enzymatic activity regulates Treg differentiation and function by controlling levels of 3PG with downstream effects on serine synthesis (Figure 2A). To test this hypothesis, we first cultured naïve murine CD4 cells under Treg-polarizing conditions with either vehicle or the PGAM inhibitor EGCG and measured intracellular serine levels from unsorted cells by mass spectrometry. As anticipated, PGAM inhibition with EGCG increased intracellular levels of serine in Tregs (Figure 2B). We next asked whether glucose-derived de novo serine synthesis is physiologically regulated during Treg differentiation, focusing on the reciprocal Treg-Th17 axis. Following incubation with uniformly 13-carbon labeled glucose (U¹³C-glucose), serine that is newly synthesized from 3PG has a mass-shift of 3 (M+3) when analyzed by mass spectrometry, whereas unlabeled, non-glucose-derived serine has no mass-shift (M+0). We differentiated naïve murine CD4 cells into either Tregs or Th17 cells for 72 hours. After 6-hour glucose starvation, the cells were incubated with U¹³C-glucose for an additional 6 hours. Using mass spectrometry, we found that ¹³C-labeling of serine was significantly higher in Th17 cells compared to Tregs (Figure 2C), indicating that flux through de novo serine synthesis is physiologically downregulated in Tregs compared to Th17 cells. Because serine can be recycled via one-carbon metabolism, we observed an increase in one carbon-labeled (M+1) serine as well as M+3 serine in Th17 cells relative to Treg cells.

If the role of PGAM in Treg differentiation depends on de novo serine synthesis, then the effects of PGAM inhibition should be rescued by simultaneously blocking the serine synthesis pathway. To this end, we cultured naïve murine CD4 cells under Treg-polarizing conditions and treated them with the PGAM inhibitor EGCG ± NCT-503, an inhibitor of phosphoglycerate dehydrogenase (PHGDH, the rate-limiting enzyme of serine synthesis from 3PG) ^44,45. As before, PGAM inhibition with EGCG inhibited Treg differentiation as assessed by FOXP3 expression, but this was reversed by PHGDH inhibition with NCT-503 (Figure 2D). Similarly, if PGAM regulates Tregs by controlling concentrations of 3PG (the substrate for serine synthesis), then direct supplementation of 3PG should replicate the effects of PGAM inhibition. To test this, 3PG was delivered to naïve murine CD4 cells via electroporation followed by Treg polarization. As expected, 3PG supplementation decreased FOXP3 expression, which was rescued by PHGDH inhibition with NCT-503 (Figure 2E). To further validate the role of de novo serine synthesis in Tregs, we used PHGDH-targeted ASOs and found that genetic knockdown of PHGDH led to augmented Treg differentiation as assessed by FOXP3 expression (Figure 2F).

The above findings demonstrate that PGAM regulates Treg differentiation by controlling flux through the serine synthesis pathway. Because serine can be derived from both de novo intracellular synthesis and extracellular sources, we next asked whether extracellular serine also affects Treg differentiation. We cultured naïve murine CD4 cells under Treg-polarizing conditions in either serine/glycine-free media or serine/glycine-free media supplemented with 4 mM serine. Deprivation of both serine and glycine is required to evaluate the role of extracellular serine, as glycine can be converted to serine via the enzyme serine hydroxymethyltransferase (SHMT)⁴⁰. The presence of extracellular serine diminished Treg differentiation (Figure 2G), indicating that both synthesized and exogenous serine impact Treg generation.

After finding that serine derived from both extracellular sources and de novo synthesis inhibits Treg polarization in vitro, we investigated the effect on Tregs of altering serine availability in vivo. We first examined the impact of dietary serine restriction by maintaining mice immediately post-weaning on either a serine/glycine-free diet or isocaloric control diet. After 8 weeks, we measured Treg abundance in peripheral blood by flow cytometry. Mice fed a serine/glycine-free diet had significantly higher proportions of Tregs in peripheral blood (Figure 3A). To determine whether this effect on Tregs translates to protection in a murine model of autoimmunity, we used the myelin oligodendrocyte glycoprotein (MOG)_35-55 experimental autoimmune encephalomyelitis (EAE) model of multiple sclerosis (MS). Mice were transitioned to a serine/glycine-free or isocaloric control diet beginning one week prior to immunization with MOG_35-55, and disease severity was monitored by a blinded scorer. A serine/glycine-free diet significantly attenuated disease severity, delaying the onset of neurologic deficits and reducing peak clinical scores (Figures 3B-D).

We next asked whether de novo serine synthesis is altered in human Tregs in disease. Using publicly available RNA-seq and scRNA-seq data, we estimated the rate of de novo serine synthesis using single cell flux estimation analysis (SCFEA)⁴⁶, which constructs a global set of metabolic reactions as a factor map and uses multi-layer neural networks to model reaction rates from mRNA levels of metabolic enzymes. We applied SCFEA to human Th0 and iTreg cells differentiated from naïve CD4 cells in vitro³¹ and found that serine synthesis was downregulated in iTregs (Figure 3E). Autoimmunity occurs when Tregs fail to adequately regulate the immune response, so we wondered whether the predicted rate of de novo serine synthesis would be altered in Tregs derived from patients with autoimmune disease. Using scRNA-seq data derived from Tregs isolated from the cerebrospinal fluid (CSF) of patients with treatment-naïve relapsing MS and controls ³², we found the predicted rate of serine synthesis was significantly higher in Tregs from MS patients (Figure 3F). Conversely, while autoimmunity results from inadequate Treg suppressive function, Tregs in the tumor microenvironment have pathologically high suppressive function. We examined scRNA-seq data from Tregs derived from healthy prostate and prostate cancer ⁴⁷ from the IMMUcan database ⁴⁸ and found that prostate tumor-infiltrating Tregs had significantly lower predicted serine synthesis than Tregs from healthy prostate (Figure 3G). Consistent with our earlier findings that serine synthesis inhibits Treg function, these analyses suggest that serine synthesis within Tregs is higher in autoimmunity and lower in the tumor microenvironment, thereby negatively correlating with expected Treg suppressive function in human disease.

We next examined the mechanistic links between serine metabolism and Treg function. Serine can serve as a one-carbon donor by entering the one-carbon metabolism cycle, which depends on conversion into glycine by the enzyme SHMT1/2 ⁴⁹ (Figure 4A). To determine whether the effects of serine on Treg differentiation depend on contributions to one-carbon metabolism, we cultured naïve murine CD4 cells under Treg polarizing conditions with either vehicle or the SHMT1/2 inhibitor SHIN1^50,51. Inhibiting SHMT1/2 increased Treg polarization (Figure 4B) and mitigated the inhibitory effect of extracellular serine on Tregs (Figure 4C). Because methionine can compensate as a one-carbon donor, other groups have found that the contribution of serine to one-carbon metabolism is greater in the absence of methionine⁵². To further the idea that serine inhibits Treg polarization by contributing to one-carbon metabolism, we cultured naïve murine CD4 cells under Treg polarizing conditions in methionine-free media supplemented with either methionine or excess serine. Although methionine supplementation had little effect in the presence of physiologic extracellular serine (0.4 mM), excess extracellular serine (4 mM) was highly potent in inhibiting Treg polarization in the absence of methionine (Figure 4D). These studies suggest that the regulatory role of serine in Tregs is mediated at least in part through one-carbon metabolism.

A crucial role of one-carbon metabolism is to provide methyl groups for methylation reactions, and DNA methylation represents a key epigenetic mechanism used to silence gene transcription⁵³. We therefore asked whether serine entry into one-carbon metabolism contributes to DNA methylation in Tregs, so we conducted a carbon tracing experiment in which Tregs were cultured in media containing uniformly 13-carbon labeled serine (U¹³C-serine) for 12 hours before DNA was isolated from the unsorted cells and analyzed by mass spectrometry. We found that serine directly contributed to cytosine methylation, as measured by carbon labeling of methylcytosine (Figure 4E). Demethylation of the Treg-specific demethylated region (TSDR) in Tregs is a critical step for generation of stable, bona fide Tregs, and methylation in this region is associated with transient or unstable expression of FOXP3 and suppressive activity ⁵⁴. We therefore investigated the effects of extracellular serine and PGAM knockdown on methylation of the TSDR via pyrosequencing. In a first set of experiments, naïve murine CD4 cells were cultured under Treg polarizing conditions in either serine/glycine-free media or media containing serine. The presence of extracellular serine increased TSDR methylation (Figure 4F). Genetic knockdown of PGAM using ASOs similarly increased TSDR methylation (Figure 4G). Taken together, these findings demonstrate that both extracellular serine and PGAM-regulated serine synthesis increase the methylation, and therefore silencing, of Treg-associated genes.

Given the promising results seen with pyrosequencing, we next used a next generation sequencing panel to identify CpG sites in Treg-specific genes whose methylation is regulated by serine (Figure 4H). Methylation of Ikzf2 (Helios) was significantly increased with the addition of extracellular serine, but this was reversed by blocking serine entry into one-carbon metabolism with the SHMT inhibitor SHIN1. In the 5’ upstream region of the Foxp3 gene, one CpG site showed significantly higher methylation in the presence of serine, and this effect was reversed by SHIN1. A complete list of CpG methylation levels can be found in Table S2.

To definitively link methylation to Treg differentiation and suppressive function, we blocked methylation in Tregs with 3-Deazaadenosine (3DZA), an inhibitor of S-adenosylhomocysteine (SAH) hydrolase. SAH hydrolase degrades SAH and prevents recycling of the methyl donor S-adenosylmethionine (SAM) ⁵⁵. In naïve murine CD4 cells cultured under Treg polarizing conditions, 3DZA increased Treg differentiation (Figure 4I) without affecting cell viability (Figure S6). Treatment with 3DZA similarly augmented suppressive function in cultured murine Tregs (Figure 4J). To determine the in vivo relevance of these findings, we used an adoptive transfer RAG colitis mouse model. Conventional murine CD4 cells were activated ex vivo and injected into RAG^-/- mice to induce colitis, along with simultaneous injection of Tregs that had been treated during differentiation with either vehicle or 3DZA. Mice injected with Tregs polarized in the presence of 3DZA had less severe disease as measured by weight loss and colonic damage (Figures 4K and S7), indicating greater Treg suppressive function. Altogether, these findings demonstrate that one-carbon metabolism and methylation impair Treg differentiation and suppressive function both in vitro and in vivo.