A central role for STAT5 in the transcriptional programing of T helper cell metabolism

Abstract

Activated lymphocytes adapt their metabolism to meet the energetic and biosynthetic demands imposed by rapid growth and proliferation. Common gamma chain (cγ) family cytokines are central to these processes, but the role of downstream signal transducer and activator of transcription 5 (STAT5) signaling, which is engaged by all cγ members, is poorly understood. Using genome-, transcriptome-, and metabolome-wide analyses, we demonstrate that STAT5 is a master regulator of energy and amino acid metabolism in CD4⁺ T helper cells. Mechanistically, STAT5 localizes to an array of enhancers and promoters for genes encoding essential enzymes and transporters, where it facilitates p300 recruitment and epigenetic remodeling. We also find that STAT5 licenses the activity of two other key metabolic regulators, the mTOR signaling pathway and the MYC transcription factor. Building on the latter, we present evidence for transcriptome-wide cooperation between STAT5 and MYC in both normal and transformed T cells. Together, our data provide a molecular framework for transcriptional programing of T cell metabolism downstream of cγ cytokines and highlight the JAK-STAT pathway in mediating cellular growth and proliferation.

Activation answers

T cell activation requires changes in metabolism needed for the energy demands of rapid growth and proliferation. Cytokines that engage common gamma chain (c#) receptors on T cells are critical to promoting metabolic changes needed for activation, and here, Villarino et al. examine the role of STAT5 engagement, which is a signaling pathway shared by all c# cytokines. STAT5 was defined as a master regulator of amino acid metabolism in CD4+ T helper cells through interactions with enhancers and promoters of genes encoding a wide array of enzymes and transporters. STAT5 controlled transcription of members of the mTOR pathway to license T cells for IL-2–mediated mTOR signaling and promoted MYC-driven metabolic changes. Together, these findings provide molecular insights downstream of IL-2 engagement that are critical to T cell activation. – CNF.

INTRODUCTION

CD4⁺ T helper (T_H) cells orchestrate both protective and pathogenic immune responses. Before activation, they persist in a naive state characterized by low nutrient uptake, low biosynthesis, and low energy consumption. Upon engagement of the T cell receptor (TCR) and costimulatory receptors, CD4⁺ T_H cells escalate their metabolism to support clonal expansion and energetically demanding effector functions (1). Nutrient uptake and protein and lipid synthesis are massively induced, and energy production is both augmented and diversified (2, 3). Effector T cells (T_eff) get much of their energy from oxidative phosphorylation (OxPhos), a process aided by increased mitochondrial mass and enhanced mitochondria function after activation (4). T_eff also engage in glycolysis, which is less energetically efficient but generates biochemical byproducts needed for pyruvate metabolism, as well as protein, DNA, and lipid synthesis. The ability of T_eff to engage in glycolysis under oxygen-rich conditions mirrors that of cancer cells in a process termed aerobic glycolysis or Warburg metabolism (5, 6). Therefore, it stands to reason that some, if not many, of the cellular pathways that drive aerobic glycolysis in T_eff also play a role in rapidly dividing cancer cells.

Cytokines operating through the common gamma chain (cγ) receptor profoundly affect T cell metabolism, with interleukin-2 (IL-2) as the representative example (7). IL-2 is closely tied to CD4⁺ T_H cells because they are both its principal source and exquisitely sensitive to its effects as one of few cell types that express the high-affinity IL-2 receptor (IL-2R) (8). It has long been known that IL-2 promotes glycolysis, OxPhos, and amino acid metabolism in T cells and that it can induce key metabolism–related genes, like the glucose importer GLUT1, the amino acid importer SLC7A5, and the transcription factor MYC (7, 9–12). Downstream signaling involves both the Janus kinase–signal transducer and activator of transcription (JAK-STAT) and the mammalian target of rapamycin (mTOR) pathways (7). Whether IL-2 directly activates the phosphatidylinositol 3-kinase (PI3K)–AKT pathway is a matter of debate, as is the relationship between PI3K-AKT and mTOR in T cells. It is known that receptor tyrosine kinases, like insulin receptors, directly engage PI3K-AKT signaling (13, 14). However, it is also evident that in T cells, (i) IL-2R does not mobilize signaling intermediates upstream of PI3K activation, (ii) AKT is not required for downstream mTOR activity, and (iii) AKT is not required for downstream proliferation or anabolic metabolism (7, 15–17). It is not disputed that IL-2R mobilizes mTOR signaling in T cells and that, in turn, mTOR affects multiple aspects of their metabolism, including glycolysis, OxPhos, pyruvate metabolism, amino acid uptake, protein synthesis, and lipid synthesis (17–19). STAT5 signaling is also involved because it is known to induce IL-2R, thus establishing a positive feedback loop (20), and the transcription factor MYC, which has broad impacts on glycolysis, OxPhos, amino acid metabolism, and other metabolism pathways relevant to both normal and transformed lymphocytes (10, 21–23). Beyond these and other indirect effects, the role of STAT5 in T cell metabolism is poorly understood.

Using gain-of-function (GOF) and loss-of-function (LOF) approaches, we demonstrate that STAT5 instructs CD4⁺ T_H cell metabolism in multiple ways. First, it controls transcription of essential, often rate-limiting, enzymes and transporters involved in energy metabolism and amino acid synthesis. Second, it controls transcription of key elements in the mTOR pathway, like SLC7A5 (solute carrier family 7 member 5) and RHEB (Ras homolog enriched in brain).

RESULTS

IL-2 has a pervasive impact on T_H cell metabolism

To study the role of IL-2–STAT5 signaling in T_H cell metabolism, we devised a model system that dissociates IL-2–driven processes from those driven by the TCR or costimulation (CD28). Naive CD4⁺ T cells were cultured with agonist anti-TCR and anti-CD28 antibodies in the presence of blocking anti-mouse IL-2 antibodies, then washed, rested, and pulsed with recombinant human IL-2 (Fig. 1A). First, we confirmed that the system yields robust STAT5 activity as denoted by phosphorylation of tyrosine 694 (p-STAT5) and induction of IL-2Rα, a well-known STAT5 target gene (Fig. 1B). Next, we compared metabolomes before and after IL-2 exposure and found that 36% of the captured biochemicals (metabolites) were significantly changed (Fig. 1B). Nearly all differentially expressed biochemicals (DEB) were positively regulated and could be classified as carbohydrate, amino acid, nucleic acid, or lipid (Fig. 1, B to E; fig. S1A; and table S1). These positively regulated DEB were highly enriched for pathways involved with energy, amino acid, and nucleic acid metabolism (Fig. 1, C and D).

Fig. 1. IL-2 has a pervasive impact on T cell metabolism.

(A) Modeling IL-2 responses in T_eff. Naive CD4⁺ T cells are purified from WT mice (1 and 2); activated with agonist anti-CD3 and anti-CD28 antibodies in the presence of blocking anti-mouse IL-2 antibodies (3); and then washed, rested, and pulsed with human IL-2 (3 to 5). LN, lymph node. (B to E) Metabolomes and transcriptomes were measured pre– and post–IL-2. (B) Flow cytometry dot plot shows p-STAT5 and IL-2Rα at 0 and 18 hours post–IL-2. The bar plot shows total and DEB or DEG. Red and blue fractions denote positively and negatively regulated DEB or DEG, respectively. (C) The left bar plot shows total captured biochemicals and positively or negatively regulated DEB within the indicated classes. The right bar plot shows the top five enriched pathways among positively regulated DEB. Energy and amino acid metabolism pathways are highlighted (D) Scatterplot shows relative abundance of select DEB before and after IL-2. Network diagrams highlight carbohydrate, amino acid, and nucleic acid pathways. Blue and red elements denote positively and negatively regulated biochemicals, respectively. The circumference is proportional to the P value. (E) The left volcano plot shows fold change (FC) and variance for all captured biochemicals. Red and blue points indicate DEB. The right volcano plot details amino acids. (F and G) Transcriptomes were measured and DEG were called as in (B). (F) Positively and negatively regulated DEG were subjected to HGT. Line graph shows enrichment P values and ranks for all KEGG pathways. The top five metabolism pathways are noted, along with P value ranks. (G) Biochemicals were paired to genes involved in their synthesis, degradation, and/or functions. Contour plot shows log₂ fold change values for each pair across metabolome and transcriptome datasets. Red and blue points denote positively or negatively pairs. The scatterplot shows top KEGG metabolism pathways among positively regulated pairs. Point size and color are proportional to DEG hit count (number shown) and q value, respectively. BH, Benjamini-Hochberg.

To establish a link between transcriptional and biochemical events downstream of IL-2, we performed RNA sequencing (RNA-seq) on the same pool of cells used for our metabolome studies. IL-2 mobilized a comparable proportion of the transcriptome (28%) but, unlike DEB, differentially expressed genes (DEG) were both positively and negatively regulated (Fig. 1B and fig. S1B). Metabolism pathways were enriched mostly within the positively regulated fraction, with carbon metabolism, nucleic acid metabolism, amino acid synthesis, and glycolysis prominently represented (Fig. 1F and fig. S1, C to E). Results were similar by hypergeometric testing (HGT) of DEG or gene set enrichment analysis (GSEA) of ranked transcriptomes, with some exceptions. Purine metabolism and glycolysis were registered only by HGT, whereas OxPhos was registered only by GSEA (fig. S1F). Next, we paired biochemicals to genes involved in their synthesis, degradation, and/or function and determined which pathways are most enriched among DEB/DEG pairs (table S2). Glycolysis and amino acid synthesis were again prominently represented (Fig. 1G). These data link transcriptional and biochemical events downstream of IL-2 and are consistent with studies showing that IL-2 induces expression of proteins involved with glycolysis, OxPhos, and amino acid biology in CD8⁺ cytotoxic T cells (9).

IL-2 also influences the metabolism of T regulatory cells (T_reg), a subset that limits T_eff responses (24). To compare effects in T_reg and T_eff, we measured transcriptomes in natural CD4⁺ T_reg before and after IL-2 exposure, then called DEG and cross-referenced with metabolism-related genes whose expression is strictly IL-2–STAT5 dependent in T_eff (defined as in Fig. 3C). Unexpectedly, we found limited overlap, with only 24% of metabolism-related genes being shared (fig. S1G). Discordance was also evident at the pathway level because glycolysis was enriched only in T_eff (fig. S1H). Amino acid synthesis was enriched in both T_eff and T_reg but was driven by different sets of genes within each subset. Thus, IL-2 propagates largely distinct gene expression programs in T_eff and T_reg and, in turn, has largely distinct metabolic consequences.

Fig. 3. Transcriptional programing of T cell metabolism via the IL-2–STAT5 axis.

(A) WT and Stat5^−/− CD4⁺ T cells were cultured and transduced as in fig. S3A, and then transcriptomes were measured. Flow cytometry contour plots show p-STAT5 and IL-2Rα in cells transduced with CA-STAT5. The bar plot shows percentage of positively (red) or negatively (blue) regulated DEG. STAT5 LOF compares WT and Stat5^−/− cells. STAT5 GOF compares Stat5^−/− cells transduced with control or CA-STAT5 RV. (B) Scatterplot compares log₂ fold change values for IL-2–regulated DEG (from Fig. 1B) across the STAT5 LOF and GOF datasets. Venn plot compares DEG across IL-2 GOF, STAT5 LOF, and STAT5 GOF datasets. Union defines IL-2–STAT5–regulated genes. (C) STAT5 ChIP-seq was performed in WT CD4⁺ T cells cultured with IL-2. The scatterplot compares amplitude of gene-associated peaks (x axis) and transcript fold change values (y axis; log₂ transformed) for all IL-2–STAT5–regulated genes [from (B)]. Amplitudes are summed for genes associated with multiple peaks. Venn diagram segregates IL-2–STAT5-regulated genes based on the presence or absence of proximal STAT5 binding sites. Union defines IL-2–STAT5 signature genes. (D) Bar plot enumerates total (white) and IL-2–STAT5 signature genes (orange) within select metabolism pathways. (E) Planetary plot shows enrichment of KEGG metabolism pathways among IL-2–STAT5–regulated genes that are bound (x axis) or not bound (y axis) by STAT5. Element size is proportional to the gene hit count. Adjusted P values are shown on both axes. (F) GSEA plots show enrichment of MSigDB hallmark pathways among genes bound or not bound by STAT5. (G) Positively and negatively regulated IL-2–STAT5 signature genes were subjected to HGT against the KEGG database. The line graph shows enrichment P values and ranks for all pathways. The top four metabolism pathways are noted, along with P value ranks. AA, amino acids. (H) Heatmap shows membership of positively (red) and negatively (blue) regulated IL-2–STAT5 signature genes across selected KEGG pathways. (I) Transcriptomes were measured in WT and Stat5^−/− CD4⁺ T cells cultured under nonpolarizing (T_H0) or subset-polarizing conditions (T_H1, T_H2, T_H17, and iT_reg). DEG were then called across genotypes and subjected to HGT against the KEGG database. Line graphs show enrichment P values and ranks for all pathways. Colored numbers denote P value ranks for glycolysis and amino acid synthesis across culture conditions. (J) GSEA plots show enrichment of positively regulated, negatively regulated, and metabolism-related IL-2–STAT5 signature genes in IL-2–treated human T cells. The left plot compares transcriptomes from pre– and post–IL-2 controls. The right plot compares transcriptomes from normal controls and a STAT5 LOF patient.

The IL-2–STAT5 axis instructs transcription of metabolism-related genes

Building on our IL-2 findings, we next considered the role of downstream STAT5 signaling. First, we compared the broad cellular consequences of IL-2 and STAT5 LOF. For IL-2 LOF, we cultured wild-type (WT) CD4⁺ T cells with or without blocking anti–IL-2/IL-2R antibodies. As expected, we found that cell size and cell division were each depressed, and p-STAT5 and IL-2Rα were each abolished in the absence of IL-2R signaling (Fig. 2, A and B, and fig. S2A). For STAT5 LOF, we compared WT and Stat5^−/− CD4⁺ T cells cultured without blocking anti–IL-2/IL-2R antibodies. We found that STAT5-deficient cells were smaller than WT counterparts, proliferated less, and failed to express IL-2Rα, despite exposure to IL-2 (Fig. 2C and fig. S2B). We also measured the extracellular acidification rate (ECAR), an indicator of glycolysis-driven lactate production, and oxygen consumption rate (OCR), an indicator of OxPhos, and found that both were greatly diminished in STAT5-deficient cells (Fig. 2, D and E). Together, these data implicate STAT5 in IL-2–driven metabolic and biosynthetic changes that accompany T cell activation.

Fig. 2. Altered T cell metabolism in the absence of IL-2–STAT5 signaling.

(A to C) WT CD4⁺ T cells were cultured with or without blocking anti–IL-2 and IL-2R antibodies and then assayed by flow cytometry. (A) Dot plots show p-STAT5 across cellular divisions. (B) Line graphs show changes in cell size, cell division, p-STAT5, and IL-2Rα relative to WT controls. (C) WT and Stat5^−/− CD4⁺ T cells cultured without blocking anti–IL-2 and IL-2R antibodies and assayed by flow cytometry. Line graphs show changes in cell size and IL-2Rα relative to WT controls. (D and E) WT and Stat5^−/− CD4⁺ T cells were cultured as in (C) for 72 hours and then rested and restimulated with IL-2 or anti-CD3/CD28. (D) Line graph shows ECAR measurements from metabolic flux analysis. (E) Scatterplot compares ECAR and OCR. (F) Modeling IL-2 responses with antigen-derived T_eff. Naive CD4⁺ T cells are purified from OT-II TCR transgenic mice and cocultured with OVA-loaded antigen-presenting cells for 5 to 6 days (1 and 2). Viable, dividing cells are then sorted and restimulated with IL-2 before metabolic flux analysis (2 and 3). CFSE, carboxyfluorescein diacetate succinimidyl ester; BM-DCs, bone marrow–derived dendritic cells; mIL-2, mouse IL-2. (G) The scatterplot compares ECAR and OCR between WT and Stat5^−/− cells. *P < 0.05 in paired t test.

We next sought to affirm the physiological relevance of our findings. First, we tested whether the concentration of human IL-2 used for our cytokine pulses approximated native IL-2 responses. We found that cell death, cell division, p-STAT5, and IL-2Rα expression were each comparable in cells exposed to autocrine IL-2 or exogenous human IL-2 (100 U/ml) (fig. S2, A and C), Next, we tested whether the metabolic phenotypes seen with polyclonal, antibody-driven stimulation hold true with monoclonal, antigen-driven stimulation. Naive T cells from WT or Stat5^−/− OT-II TCR transgenic mice were cocultured with ovalbumin (OVA) peptide–loaded antigen-presenting cells (APCs); then, 6 days later, viable, proliferating T cells were sorted, restimulated with IL-2 (in the absence of OVA-APCs), and subjected to bioflux analysis (Fig. 2F). Crucially, we found that ECAR and OCR were sharply diminished in STAT5-deficient cells, thus confirming a key role in both glycolysis and OxPhos (Fig. 2G). The fact that only viable, actively proliferating cells were included in the assay also suggests that the metabolic phenotypes seen under polyclonal stimulation are not simply due to the inability of Stat5^−/− cells to become activated or proliferate (Fig. 2, A and B).

To better understand how STAT5 influences T cell metabolism, we integrated LOF and GOF approaches to define its target gene repertoire downstream of IL-2. For LOF, we compared transcriptomes in WT and Stat5^−/− cells cultured as above (Fig. 2C and fig. S2B). For GOF, we measured transcriptomes in Stat5^−/− T cells that were transduced with constitutively active STAT5A (CA-STAT5) and pulsed with human IL-2 (Fig. 3A and fig. S3, A to C). Consistent with our IL-2 GOF studies (Fig. 1B), we found that STAT5 LOF and STAT5 GOF each affected 20 to 30% of transcriptomes, with comparable proportions of positively and negatively regulated DEG (Fig. 3A), and that metabolism pathways were enriched mainly within positively regulated fractions (fig. S3, D and E). We next cross-referenced the IL-2 GOF, STAT5 LOF, and STAT5 GOF datasets to identify genes subject to both IL-2 and STAT5, hereafter termed IL-2–STAT5–regulated genes. Most DEG fell within the union of the three datasets, and those shared DEG were highly enriched for metabolism pathways (Fig. 3B and fig. S3F). Thus, STAT5 appears to control much of the transcriptional response downstream of IL-2, including genes involved in cellular metabolism.

STATs instruct gene transcription by directly engaging DNA regulatory elements, such as enhancers and promoters, associated with target gene loci (25). To determine whether STAT5 directly engages genes involved with T cell metabolism, we mapped STAT5 distribution by chromatin immunoprecipitation followed by DNA sequencing (ChIP-seq). We found that nearly two-thirds of IL-2–STAT5–regulated genes were proximally bound by STAT5, including 94 involved in metabolism pathways (Fig. 3C). Hereafter, we refer to this group as IL-2–STAT5 signature genes. The remaining one-third, which did not exhibit proximal STAT5 binding, likely results from parallel signaling pathways or downstream transcription factors. Crucially, metabolism pathways were far more enriched within the STAT5 “bound” fraction, specifically among positively regulated genes (Fig. 3, D to G). However, the “unbound” fraction was not inert. HGT showed that unbound, IL-2–STAT5–regulated genes are uniquely enriched for the glycine, serine, and threonine metabolism pathway, and GSEA showed enrichment of both glycolysis and OxPhos, albeit orders of magnitude less than for STAT5-bound genes (Fig. 3, E and F). We also found that IL-2–STAT5 signature genes were similarly affected by deletion of STAT5A or STAT5B, indicating that both paralogs are similarly involved (fig. S3, G and H). Thus, STAT5 directly engages many genes involved in T cell metabolism and is especially prominent within certain pathways.

Focusing on glycolysis, we found that IL-2–STAT5 signature genes pervade the pathway and include key enzymes, like hexokinase 2 (Hk2), aldolase A (Aldo1), and pyruvate kinase (Pkm) (Fig. 3H and fig. S4). Accordingly, all captured biochemicals downstream of HK2 were strongly induced by IL-2, including fructose 1,6-diphosphate and pyruvate (Fig. 1D and fig. S4). We also noticed that IL-2 mobilized biochemicals along the citrate cycle, particularly fumarate and malate (Fig. 1D and table S1). This was notable because oxygen consumption is depressed in STAT5-deficient cells, and, similar to glycolysis, OxPhos was enriched among IL-2–STAT5 signature genes (Figs. 2E and 3F). Thus, the IL-2–STAT5 axis influences T cell metabolism through direct transcriptional programming of at least two key energy production pathways.

Our metabolome studies showed that IL-2 prompts accumulation of all amino acids, with possible exceptions of glycine, alanine, valine, and tryptophan (Fig. 1, D and E). It had comparable effects on essential and nonessential amino acids, but the data point to distinct underlying mechanisms (Fig. 1E). On one hand, we found that STAT5 directly engages the Slc7a5 locus, which encodes the principal transporter for essential amino acids in T cells (table S4). However, it does not appear to target transporters for nonessential amino acids but, instead, targets key elements of their synthesis pathways. One notable example is Got1, which encodes a critical enzyme for conversion of aspartate to oxaloacetate, which is necessary for several nonessential amino acid synthesis pathways (Fig. 3H and fig. S4).

Five IL-2–STAT signature genes are shared between glycolysis and amino acid synthesis pathways, namely, Eno1, Eno1b, Eno2, Aldoa, and Pkm (Fig. 3H). Given that products of glycolysis are used for amino acid synthesis, we can infer that the ability of STAT5 to drive the former affects the latter. Accordingly, IL-2 induces several biochemicals that are shared between glycolysis and amino acid synthesis, including 3-phosphoglycerate, phosphoenolpyruvate, and pyruvate, as well as several amino acids that are closely tied to glycolysis, including serine, tyrosine, and leucine (Fig. 1D and fig. S4). Thus, the IL-2–STAT5 axis promotes accumulation of amino acids by instructing transcription of genes involved in both import and synthesis, the latter involving glycolysis-dependent and independent pathways.

Enzymes involved in amino acid metabolism can also affect DNA synthesis and regulation. One notable example is adenosylhomocysteinase (Ahcy), which plays a central role in cysteine and methionine synthesis and promotes DNA methylation by converting S-adenosylmethionine (SAM) to S-adenosylhomocysteine (SAH). Ahcy registers as an IL-2–STAT signature gene, and IL-2 triggers accumulation of SAH, suggesting that, in turn, STAT5 may also promote DNA methylation (tables S1 and S4). STAT5 also controls genes involved with nucleic acid import and salvage, including the nucleoside importer, SLC29a1, and the salvage enzyme, HPRT, as well as genes involved with polyamine synthesis, a process vital for DNA packaging (table S4). Regarding the latter, spermidine synthase registers as an IL-2–STAT signature gene and is crucial for polyamine synthesis (table S4). Thus, the IL-2–STAT5 axis may leverage shared pathway components to coordinate amino acid synthesis and gene transcription.

CD4⁺ T_eff are categorized on the basis of stereotypical expression of transcription factors and cytokines (26). To determine whether STAT5 controls metabolism across T_eff subsets, we compared transcriptomes in WT and Stat5^−/− T cells cultured under nonpolarizing (T_H0) or T_H1, T_H2, T_H17, and T_reg polarizing conditions. Unexpectedly, we found STAT5 dependency varied greatly (Fig. 3I). Glycolysis was highly STAT5 dependent under T_H0, T_H1, T_H2, and T_reg conditions but not T_H17 conditions, whereas amino acid synthesis was highly STAT5 dependent under T_H1 and T_reg conditions but not T_H0, T_H2, and T_H17 conditions. We also noted that, in contrast to the in vivo “natural” T_reg assayed earlier (fig. S1H), IL-2–STAT5 signaling did mobilize glycolysis in these “induced” T_reg (fig. S1H). Thus, we can conclude that STAT5-driven metabolism is more prominent in some T cell subsets than others and may even vary within subsets, depending on how they are generated and/or sourced.

To further establish physiological relevance, we next sought to confirm that IL-2–STAT5 signature genes are similarly regulated in mouse and human T cells. We mined transcriptome datasets comparing IL-2–treated T cells from “normal” controls with those from an individual with an LOF mutation in STAT5 (27). First, we ranked transcriptomes based on pairwise comparisons and then ran GSEA against positively regulated, negatively regulated, and metabolism-related IL-2–STAT5 signature gene sets (defined in mice as in Fig. 3C). We found that (i) all three gene sets were highly enriched when comparing pre– and post–IL-2 samples (within normal controls), and (ii) positively regulated and metabolism-related IL-2–STAT5 signature genes were enriched in normal controls relative to STAT5 LOF (Fig. 3J). Thus, the IL-2–STAT5 axis regulates similar sets of genes in human and mouse T cells, including genes involved in metabolism pathways.

STAT5 directs p300 activity at metabolism-related gene loci

STATs promote gene expression, in part, by recruiting p300 to gene promoters and enhancers. Given this relationship, we hypothesized that STAT5 may recruit p300 to metabolism-related genes. To explore that idea, we first compared genome-wide distributions for STAT5 and p300 in IL-2–treated T_eff (Fig. 4, A and B, and fig. S5A). We found that nearly all p300-bound regions (87%), termed “peaks,” overlapped with STAT5-bound regions, in line with the idea that p300 physically interacts with STAT family proteins (28). However, the converse was not true because only 34% of STAT5 peaks colocalized with p300 peaks (Fig. 4C). Focusing on gene-associated peaks, we noted that overlapping (STAT5 + p300) and nonoverlapping (STAT5 only) sites often decorate the same loci. This was true for STAT5 signature genes—62% had overlapping peaks, of which 75% also had nonoverlapping peaks—and particularly so for those involved in metabolism pathways (Fig. 4C).

Fig. 4. STAT5 instructs enhancer activity at metabolism gene loci.

(A to J) STAT5, p300, H3K27ac, and H3K4me1 ChIP-seq was performed in IL-2–treated WT CD4⁺ T cells. (A) Genome browser tracks show selected IL-2–STAT5 signature gene loci. (B) Histogram and heatmap show p300 distribution relative to STAT5 peak centers. Only overlapping regions are shown. STAT5 motif enrichment is presented below. (C) Venn plots compare genomic coordinates for all STAT5- and p300-bound regions, those near annotated genes, those near IL-2–STAT5 signature genes, and those near metabolism-related IL-2–STAT5 signature genes. (D) Line graphs show signal-based ranking of p300- and STAT5-bound regions. The number of SE is shown; the top five are listed. (E) Histogram shows distribution of p300, H3K27ac, and H3K4me1 relative to STAT5 peak centers. (F) STAT5, p300, H3K27ac, and H3K4me1 ChIP-seq peaks were categorized on the basis of genomic coordinate overlaps. The line graph shows the total number of peaks per category. (G) The violin plot compares STAT5 peak scores for regions bound by STAT5 alone or STAT5 plus p300. (H) Box plots show transcript fold change values for genes bound by STAT5 alone or STAT5 plus p300 (metab, metabolism related genes). LOF and GOF data are from Fig. 3A. (I) IL-2–STAT5 signature genes (from Fig. 3C) were segregated on the basis of p300 binding and subjected to HGT against the KEGG database. The left line graph shows enrichment P values and ranks for all pathways. The right line graph details metabolism-related pathways. The top five pathways are noted for both plots. (J) Positively and negatively regulated p300-bound IL-2–STAT5 signature genes were subjected to HGT against MSigDB. The line graph shows enrichment of all hallmark pathways. Selected pathways are noted, along with P value ranks. EBV, Epstein-Barr virus.

Super enhancers (SE) are dense enhancer clusters associated with highly inducible, lineage-restricted genes (29). To determine whether IL-2 promotes SE assembly at metabolism-related loci, we mapped SE using a p300 ChIP-seq signal. Unexpectedly, we found that only 6 of 243 SE-bearing genes are associated with metabolism pathways, and these were among the lowest-scoring SE per p300 content (Fig. 4D and fig. S5, B and C). Results were similar when we used the STAT5 signal to call SE. Thus, STAT5 appears to drive p300 recruitment to both standard enhancers and SE, but the latter are not particularly relevant for metabolism-associated genes.

p300 promotes gene expression by mediating acetylation of histone 3 at lysine 27 (H3K27ac). Therefore, to gain further mechanistic insight, we mapped H3K27ac as well as monomethylation of histone 3 lysine 5 (HeK4me1), a marker of active transcription, and then cross-referenced with our STAT5 and p300 ChIP-seq datasets (Fig. 4E and fig. S5A). Regions associated with STAT5 alone were most common, likely because this dataset had the greatest number of peaks. More interesting was the prevalence of regions associated with STAT5, p300, H3K27ac, and HeK4me1 (Fig. 4F). These strongly support a model whereby STAT5 drives enhancer activity via p300-mediated histone acetylation. We also noted a substantial number of regions with STAT5 and H3K27ac but no p300 (Fig. 4F). These imply p300-independent epigenetic activities, although prior p300 occupancy has not been ruled out.

To further explore the relationship between STAT5 and p300, we segregated STAT5 peaks on the basis of p300 colocalization and compared genomic characteristics. We found that co-occupied regions have greater amplitude and evolutionary conservation than those occupied by STAT5 alone (Fig. 4G and fig. S5D). We also cross-referenced with our transcriptome datasets and learned that genes with overlapping STAT5 and p300 peaks are more labile than those associated with STAT5 alone, with a pronounced effect among metabolism-related genes (Fig. 4H). We then segregated IL-2–STAT5 signature genes on the basis of p300 occupancy and were surprised to find that only those associated with p300 were enriched for Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways, suggesting that this is the more biologically organized fraction (Fig. 4I). However, none involved metabolism, likely because genes involved with metabolism pathways were split in terms of p300 co-occupancy. As expected, metabolism pathway was far more enriched among positively regulated than negatively regulated p300-bound, IL-2–STAT5 signature genes (Fig. 4J and fig. S5E). Thus, STAT5 appears to recruit p300 to metabolism-related genes and may also have relevant, p300-independent activities.

STAT5 primes cytokine-driven mTOR signaling

Both our STAT5 LOF and GOF studies registered mTOR signaling among the most enriched pathways. mTOR was more affected than IL-2–STAT5, glycolysis, or OxPhos (Fig. 5A). Given these findings and the prominent role of mTOR in T cell metabolism, we sought to determine whether there are direct links between STAT5 and mTOR. Using flow cytometry, we first confirmed that STAT5 deficiency impairs mTOR activity at a biochemical level. We found that phosphorylation of S6 kinase (p-S6), a well-known target of mTOR complex 1 (mTORC1), was markedly reduced in Stat5^−/− cells (Fig. 5, B and C and fig. S6A). This phenotype was not due to the inability of cells to proliferate or differentiate into effectors because it held true even when only actively dividing cells were considered (fig. S6B). We also found that total S6 protein levels were diminished (Fig. 5D), in line with studies showing that IL-2 induces both expression and phosphorylation of S6 (9). Other events associated with mTOR activation were also depressed, including serine phosphorylation of AKT, a known target of the mTORC2, and phosphorylation of mTOR kinase itself (fig. S6C). Diminution of p-Ser AKT was less marked than that of p-S6 but, crucially, was not due to the smaller size of STAT5-deficient T cells (fig. S6D). It was also notable that mTOR activity became evident in WT cells only at 24 hours after IL-2 exposure and never at 1 hour (Fig. 5B). This delay was not seen with p-STAT5, which was readily detected at both time points. Together, these data establish that STAT5 is required for IL-2–driven mTOR signaling and suggest a discrete time window for STAT5-dependent priming to occur.

Fig. 5. STAT5 is required for IL-2–driven mTOR activation.

(A) The scatterplot shows GSEA results for STAT5 LOF and GOF studies (MSigDB hallmark database; transcriptomes are from Fig. 3A). Point size and color are proportional to P values and q values, respectively. NES, normalized enrichment score. (B) WT and Stat5^−/− CD4⁺ T cells were cultured for 72 hours, washed, rested, and recultured with IL-2 for 1 or 24 hours. Line graphs show changes in p-STAT5, p-S6, and cell size relative to untreated controls. (C and D) Contour plots show p-STAT5, p-S6, and total S6 kinase levels at 0 or 24 hours post–IL-2. (E) Modeling IL-2 responses with in vivo–generated T_eff. WT and Stat5^−/− mice were immunized with LCMV GP 61-80 peptide. Seven days later, draining lymph nodes were dissected, and single-cell suspensions were cultured with IL-2. Box plots show flow cytometry mean fluorescence intensity (MFI) values for IL-2Rα and p-S6 in antigen-specific effector and memory CD4⁺ T cells and polyclonal naive CD4⁺ T cells (see fig. S6E for gating strategy). CFA, complete Freund’s adjuvant. (F) Contour plots show p-S6 and IL-2Rα cytometry. (G) Western blots show threonine phosphorylation (p-Thr) of FOXO1 and serine phosphorylation of AKT (p-Ser) in T_eff generated in vitro and restimulated with anti-CD3 and anti-CD28 or IL-2. (H) CD4⁺ T cells were cultured as in (C). Contour plots show p-Thr FOXO1 and p-Ser AKT. (I) WT CD4⁺ T cells were cultured as in Fig. 1A with or without AKT, mTOR, or JAK inhibitor during the final IL-2 pulse. Flow cytometry contour plots show p-STAT5 and p-S6 at 24 hours post–IL-2/drugs. *P < 0.05 in paired t test for the 24-hour time point versus 0 hour. **P < 0.05 in paired t test for both the 1- and 24-hour time points versus 0 hour (within the same genotype).

To determine whether STAT5 is required for mTOR activity in vivo, we studied T_eff from immunized mice (Fig. 5, E and F, and fig. S6E). Specifically, WT and Stat5^−/− mice were immunized with a model antigen, then single-cell suspensions from draining lymph nodes were pulsed with IL-2 and mTOR activity was measured in antigen-specific T_eff. As with our in vitro model, we found that p-S6 and p-Ser AKT were each reduced in T_eff from Stat5^−/− mice compared with their WT counterparts (Fig. 5, E and F). IL-2Rα was also depressed, which confirms the concurrent loss of STAT5 signaling. Together, these studies establish that STAT5 is required for IL-2–driven mTOR signaling in T_eff generated through natural, antigen-driven processes.

Recent studies dispute the widely held notion that IL-2 activates mTOR via the PI3K-AKT pathway (7, 15). To address this issue, we measured phosphorylation of AKT at Thr³⁰⁸, which is mediated by PI3K, and phosphorylation of FOXO1 at Thr²⁴, which is mediated by AKT. We found that both were induced by IL-2 in WT cells and diminished in STAT5-deficient cells (Fig. 5, G and H, and fig. S7, A and B). Next, we tested whether AKT is required for IL-2–driven mTOR activation. Crucially, we found that AKT inhibitors had little impact on p-S6, an mTORC1-driven event, or p-Ser AKT, an mTORC2-driven event (Fig. 5I and fig. S7C). This was in sharp contrast to the mTOR inhibitor, rapamycin, and the JAK-STAT inhibitor, tofacitinib, the latter of which was notably more potent. As expected, only tofacitinib suppressed p-STAT5, and all drugs suppressed Ki67, an indicator of cellular proliferation (fig. S7C). These data suggest that the ability of IL-2 to propagate PI3K-AKT signaling in T_eff is largely unrelated to mTOR signaling.

To further explore the relationship between STAT5 and mTOR, we performed GOF studies. First, we measured mTOR activity in cells transduced with CA-STAT5 or an AKT variant known to trigger mTOR signaling (CA-AKT) (30). We found that only CA-STAT5 could induce p-S6 in STAT5-deficient cells (Fig. 6A). Given that both constructs were active in WT cells, we interpret that STAT5 is required for mTOR activation even when it is uncoupled from IL-2R. We also noted that CA-AKT was able to induce p-Thr FOXO1 and p-Ser AKT in both genotypes, which confirms that signaling downstream of AKT is operational in Stat5^−/− cells and suggests fundamental differences in how STAT5 drives mTORC1 and mTORC2 activation (figs. S7B and S8, A and B).

Fig. 6. STAT5 is required for AKT-driven mTOR activation.

(A to D) Transcriptomes were measured in WT and Stat5^−/− CD4⁺ T cells transduced with control, CA-STAT5, or CA-AKT RV. (A) Flow cytometry contour plots show p-STAT5 and p-S6 in transduced cells. (B) DEG were called relative to genotype-matched controls. The left bar plot enumerates total (white bars) and metabolism-related DEG (black bars). The right bar plot enumerates the total and metabolism KEGG pathways enriched among DEG. (C) Venn plot compares DEG mobilized by CA-STAT5 and CA-AKT in Stat5^−/− CD4⁺ T cells. The bar plot enumerates the total and metabolism KEGG pathways enriched among Venn groups. (D) Scatterplots show enrichment of select MSigDB hallmark and KEGG pathways across Venn groups. Point size and color are proportional to DEG hit count and P value, respectively. (E and F) WT and Stat5^−/− CD4⁺ T cells were transduced with paired combinations of control, CA-STAT5, and CA-AKT RV and then transcriptomes were measured. (E) DEG were called relative to genotype-matched controls. Venn plot segregates DEG based on RV combinations. The scatterplot shows enrichment of select KEGG metabolism pathways within Venn groups. Point size and color are proportional to DEG count and P value, respectively. (F) Transcriptomes were ranked against genotype-matched controls and then subjected to GSEA. Scatterplots show NES (x axis) and false discovery rates (FDR; y axis) against KEGG, MSigDB hallmark, and custom rapamycin and tofacitinib gene sets (detailed in table S6). KO, knockout.

Next, we measured transcriptomes and called DEG relative to genotype-matched controls. Few DEG were evident in WT cells whether transduced with CA-STAT5 or CA-AKT, likely because both the JAK-STAT and AKT pathways are already highly active (Fig. 6B). On the other hand, CA-STAT5 mobilized thousands of DEG in STAT5-deficient cells and CA-AKT mobilized several hundred. However, unlike those mobilized by CA-STAT5, DEG mobilized by CA-AKT had little to do with metabolism, because only 37 could be assigned to KEGG metabolism pathways versus 330 for CA-STAT5 (Fig. 6B). That disparity held true at the pathway level—only one metabolism pathway was enriched among CA-AKT–responsive DEG—but did not reflect a general lack of organization because 18 nonmetabolism pathways were enriched (Fig. 6B and fig. S8C). Rather, it may reflect a genuine lack of metabolism-related DEG downstream of AKT, in line with the idea that it acts mainly on posttranscriptional events or a requirement for STAT5 in downstream transcriptional responses. Consistent with the latter point, most of the DEG mobilized by CA-AKT were also mobilized by CA-STAT5 (Fig. 6C). However, note that these shared DEG are not enriched for metabolism pathways or downstream targets of mTOR, which, instead, are enriched only among DEG mobilized by CA-STAT5 alone (Fig. 6D). Thus, we can conclude that STAT5 is the more relevant factor with respect to transcriptional programing of T cell metabolism.

To address cooperation between STAT5 and mTOR, we compared transcriptomes from Stat5^−/− cells transduced with CA-STAT5, CA-AKT, or both constructs. First, we called DEG relative to control retrovirus (RV). By far, the greatest number of DEG was found in cells transduced with CA-STAT5, whether alone or in combination with CA-AKT (group 3). There were also many DEG that became apparent in cells transduced with either CA-STAT5 or CA-AKT (group 4), CA-STAT5 plus CA-AKT (group 2), CA-STAT5 alone (group 1), or CA-AKT alone (Fig. 6E). To further explore these DEG groups, we performed HGT against the KEGG database. We found that glycolysis and amino acid synthesis were enriched only in group 3, genes mobilized by CA-STAT5 with or without CA-AKT (Fig. 6E). GSEA of ranked transcriptomes confirmed that CA-STAT5 was sufficient to mobilize both glycolysis and amino acid synthesis and further indicated that cotransduction with CA-AKT provides no additional bonus (Fig. 6F). GSEA also showed that CA-STAT5 is far better at mobilizing genes involved with mTORC1 and IL-2–STAT5 signaling, whether testing against Molecular Signatures Database (MSigDB) gene sets or custom gene sets of rapamycin (mTOR)– and tofacitinib (JAK-STAT)–sensitive genes (Fig. 6F and table S6). Principal components analysis (PCA) and hierarchical clustering further illustrated the disparity between CA-STAT5– and CA-AKT–driven transcriptional responses (fig. S8, E to G). Together, these studies affirm that STAT5 is necessary and sufficient for transcriptional programing of T cell metabolism.

Given that CA-AKT bypasses IL-2R yet still cannot activate mTOR in STAT5-deficient cells, we reasoned that STAT5 must control elements of the mTOR pathway that are independent or downstream of IL-2R signaling. To explore this idea, we asked whether and where IL-2–STAT5 signature genes appear along the core mTOR pathway. First, we noted a prominent cluster involved with import and sensing of large neutral amino acids (Fig. 7A). Among these was the aforementioned Slc7a5, a critical transporter for amino acid–driven mTOR activation (11). Thus, we confirmed that IL-2 mobilizes all known SLC7A5 substrates in T cells (Fig. 7B) and that expression of CD98, a heterodimer of SLC7A5 and SLC3A2, is depressed in Stat5^−/− cells and restored by CA-STAT5 (Fig. 7C). We also noted that IL-2–STAT5 signature genes occupy other key stations along the mTOR pathway, including Rheb, a central bottleneck for mTORC1 activation, and Pik3r3, a critical regulator of upstream PI3K-AKT activity (Fig. 7A). These genes and all mTOR-related, IL-2–STAT5 signature genes were marked with STAT5, p300, HeK27ac, and H3K4me1 at promoters and proximal enhancers (Fig. 7D). Thus, the IL-2–STAT5 axis enables mTOR signaling by instructing transcription of key pathway elements, particularly within the amino acid import and sensing machinery.

Fig. 7. Transcriptional mechanisms for STAT5-depent mTOR activation.

(A) The KEGG pathway map highlights IL-2–STAT5 signature genes within the core mTOR pathway. IL-2–STAT5 signature genes are defined as in Fig. 3C. Tripartite segments correspond to DEG sets (IL-2 GOF, STAT5 LOF, or STAT5 GOF), and color denotes transcriptional effect. AMPK, AMP-activated protein kinase. (B) Metabolomes were measured as in Fig. 1B. Volcano plot shows fold change and variance for amino acids known to be transported via SLC7A5. (C) Flow cytometry contour plots show surface CD98 and IL-2Rα in WT and Stat5^−/− CD4⁺ T cells transduced with control or CA-STAT5 RV. (D) ChIP-seq was performed as in Fig. 4A. Genome browser tracks show STAT5, p300, H3K27ac, and H3K4me1 at mTOR-associated gene loci. (E) Flow cytometry contour plots show IL-2Rα and IL-2Rβ in WT and Stat5^−/− CD4⁺ T cells transduced with control or CA-STAT5 RV. (F) Flow cytometry histograms show IL-2Rα and p-S6 in CD4⁺ T cells transduced with control, IL-2Rα, or CA-STAT5 RV (MFI values noted). (G) Stat5^−/− CD4⁺ T cells were transduced with control or IL-2Rα RV, and then transcriptomes were measured. The volcano plot shows fold change and variance for all expressed transcripts. The red and blue points denote positive and negatively regulated DEG.

We next considered whether loss of STAT5-driven IL-2R expression contributes to diminished mTOR signaling in Stat5^−/− cells. First, we measured IL-2Rα and IL-2Rβ and found that both were strongly induced by IL-2, sharply reduced in the absence of STAT5, and restored by CA-STAT5 (Fig. 7E and fig. S9A). We also noted that mTOR activity was most apparent in cells with high IL-2Rα levels and increased progressively across null, low, medium, and high expression bins (fig. S9B). IL-2Rα levels appeared to have greater influence on mTOR activity than a cytokine dose because p-S6 remained stable within expression bins across a hundred-fold range of IL-2 concentrations (fig. S9C). However, when we transduced Stat5^−/− cells with IL-2Rα, we found that a threefold increase in surface IL-2Rα had little effect on p-S6 and mobilized only eight DEG (Fig. 7, F and G). This was in stark contrast to CA-STAT5, which yielded a >20-fold increase in IL-2Rα, robust S6 phosphorylation, and thousands of DEG. Predictably, IL-2Rα transduction also did little in WT cells because they already express high levels (Fig. 7F). Together, these data imply that IL-2–driven mTOR signaling occurs mainly in T cells that reach a high receptor threshold, which is typically achieved through STAT5-mediated positive feedback.

STAT5 cooperates with MYC to program T cell metabolism

STAT5 is a potent inducer of MYC (10, 31). Accordingly, we observed that (i) MYC is an IL-2–STAT5 signature gene; (ii) the Myc locus is decorated with co-incident STAT5, p300, H3K27ac, and H3K4me1; and (iii) MYC targets are highly enriched among IL-2–STAT5 signature genes (Figs. 4J, 6D, and 8A and fig. S3E). Given these findings, we next addressed whether MYC deprivation contributes to metabolic defects seen in Stat5^−/− cells. First, we confirmed that IL-2 induces MYC protein expression via the JAK-STAT pathway (Fig. 8B). Then, we restored MYC expression in Stat5^−/− T cells and measured downstream transcriptional consequences. Unexpectedly, we found that MYC mobilized only 171 DEG compared with 1493 DEG mobilized by CA-STAT5 (Fig. 8C). MYC transduction did make cells larger and affected several metabolism pathways but, unlike CA-STAT5, did not restore the transcriptional repertoire to WT levels (Fig. 8, C and D, and figs. S8C and S10A). This may explain why we found many more enriched pathways by GSEA than HGT, because GSEA is better at detecting coordinated, small-scale changes in gene expression (Fig. 8E and fig. S10A). Building on the latter idea, we compared dynamic ranges for STAT5 and MYC as per Tong et al. (32). We ranked DEG on the basis of fold change and found that all MYC-regulated genes exhibited 2- to 10-fold changes, whereas STAT5-regulated genes had a much broader range of outcomes, with many exhibiting >100-fold changes (Fig. 8F). Thus, MYC and STAT5 both affect transcription of metabolism-related genes but do so in different ways.

Fig. 8. Cooperation between STAT5 and MYC in programming T cell metabolism.

(A) ChIP-seq was performed as in Fig. 4A. Genome browser tracks show the Myc locus. (B) WT T cells were cultured as in Fig. 1A with rapamycin or tofacitinib. Flow cytometry contour plots show IL-2Rα and MYC at 24 hours post–IL-2 ± drugs. (C to F) Transcriptomes were measured in WT and Stat5^−/− T cells transduced with control, CA-STAT5, or MYC RV. (C) DEG were called relative to genotype-matched controls. Bar plots enumerate the total (white bars) and metabolism-related (black bars) DEG or enriched KEGG pathways. (D) Box plots show TPM values for DEG that are positively regulated by STAT5 (top) or MYC (bottom) in Stat5^−/− cells. (E) Planetary plot shows HGT- and GSEA-derived P values for Stat5^−/− T cells transduced with MYC. Only KEGG metabolism pathways are included. Element size is proportional to DEG hit count. (F) STAT5- and MYC-induced DEG were ranked on the basis of fold change. The line graph shows percentage exhibiting 3 to 10, 10 to 100, or >100-fold induction. (G) Dot plots show absolute transcript abundance (TPC) in transduced WT or Stat5^−/− T cells. The black line denotes 1:1 correlation (slope = 1). The red line denotes the observed trend. (H to N) CD4⁺ T cells were transduced with paired combinations of control, CA-STAT5, and/or MYC RV, and then transcriptomes were measured. (H) The scatterplot shows PCA for unabridged transcriptomes. (I) Transcriptomes were ranked against genotype-matched controls and then subjected to GSEA. The scatter plots show NES (x axis) and FDR (y axis) for MSigDB hallmark pathways. (J) DEG were called relative to genotype-matched controls. Venn plot segregates DEG by RV combinations. The scatterplot shows enrichment of KEGG metabolism pathways within Venn groups. Point size and color are proportional to DEG hit count and P value, respectively. (K) DEG were called as in Figs. 6E and 8J. Box plots shows mean fold change values for DEG within the following Venn groups: (1) DEG mobilized by STAT5 and inhibited by AKT/MYC, (2) DEG mobilized by the combination of STAT5 and AKT/MYC, (3) DEG mobilized by STAT5 with or without AKT/MYC, and (4) DEG mobilized by STAT5 or AKT/MYC. (L) The scatterplot shows mean TPM values for select STAT5- and MYC-regulated genes. (M and N) MYC and STAT5 ChIP-seq data were cross-referenced with IL-2 transcriptome data (from Fig. 1B). (M) The pie chart shows a proportion of MYC-bound regions co-occupied by STAT5. Genome browser tracks show colocalization at the Ldha locus. (N) Venn diagram segregates IL-2–regulated DEGs based on whether they are bound by STAT5 alone (4), MYC alone (3), or both at overlapping (1) or nonoverlapping sites (2). The scatterplot shows enrichment of select KEGG pathways across Venn groups. Point size and color are proportional to DEG count and P value, respectively.

We next compared transcripts per cell (TPC), a quantitative measure of mRNA abundance. We found that, similar to MYC deficiency (33), STAT5 deficiency results in global transcriptome diminution and that, crucially, only CA-STAT5 could rescue the deficit. MYC alone was not sufficient (Fig. 8G and fig. S10B). Thus, STAT5 is required not only for MYC expression but also for MYC-driven transcriptome amplification. This latter finding implied cooperation, so we next compared transcriptomes in cells transduced with CA-STAT5, MYC, or both. Hierarchical clustering and PCA each illustrated the marked effect of CA-STAT5 and relatively muted effect of MYC (Fig. 8H and fig. S10C). PCA also hinted at cooperation; MYC cotransduction pushed STAT5-deficient cells further toward WT counterparts than CA-STAT5 alone (Fig. 8H). Similar trends were also observed at the pathway level. GSEA showed that MYC is sufficient to drive enrichment of MYC targets, glycolysis, OxPhos, and several other metabolism-related pathways, but, in each case, CA-STAT5 was more potent, and its effects were boosted by MYC cotransduction (Fig. 8I and fig. S10D).

To further explore the relationship between STAT5 and MYC, we called DEG in cotransduced cells. Given the relative paucity of MYC-regulated DEG (171 per Fig. 8C), we were surprised to find that >1000 DEG were called only in cells transduced with both CA-STAT5 and MYC (group 2) (Fig. 8J). It was also notable that the metabolism pathways enriched among these group 2 DEG differed from those among group 3 DEG, which were called in cells transduced with CA-STAT5 with or without MYC. The former were enriched for the (TCA) tricarboxylic acid cycle and OxPhos, whereas the latter were enriched for glycolysis and amino acid synthesis (Fig. 8J and fig. S10E). Hierarchical clustering further showed that glycolysis and amino acid synthesis were enriched mainly within a cluster of DEG that were strongly induced by CA-STAT5, largely insensitive to MYC alone, and boosted by MYC cotransduction (fig. S10F). This “boosting” effect was evident across entire DEG groups, except group 1, and for individual metabolism-related genes, like Hk2 and Slc7a5 (Fig. 8, J to L). No such boosting effect was observed in cells cotransduced with CA-STAT5 and CA-AKT (Fig. 8K). We interpret that STAT5 and MYC cooperate genome wide and that certain metabolism pathways, like OxPhos, are particularly dependent on that cooperation.

Considering that STAT5 and MYC are each DNA binding transcription factors, we next asked whether they colocalize at metabolism-related gene loci. First, we cross-referenced a MYC ChIP-seq dataset captured in effector CD4⁺ T cells (33) with our STAT5 ChIP-seq dataset and found that 49% of MYC-bound regions overlapped with STAT5-bound regions (Fig. 8M). Then, we assigned peaks to genes, cross-referenced with our IL-2 GOF dataset, and tested for pathway enrichments. We found that amino acid synthesis was highly enriched among IL-2–regulated genes with overlapping MYC and STAT5 peaks (group 1) but not those with only nonoverlapping peaks (group 2) or those lacking MYC altogether (group 4; STAT5 alone) (Fig. 8N and fig S10G). Glycolysis was not enriched in any group, although STAT5 and MYC did overlap at key glycolysis genes (Fig. 8M and fig. S10H). This latter finding is consistent with studies showing that MYC directly engages lactate dehydrogenase A (LDHA) and PKM and suggests that cooperation is relevant for this pathway (34, 35). Thus, cooperation between STAT5 and MYC is evident broadly across amino acid synthesis pathways and more electively within the glycolysis pathway.

Evidence for cooperation between STAT5 and MYC in lymphoid malignancies

Given similarities between activated and transformed lymphocytes, we were intrigued that CA-STAT5 mobilized cancer-related pathways in Stat5^−/− T cells (fig. S11A). Even more intriguing was the fact that MYC did not, despite strong engagement of canonical MYC target genes (fig. S11A). To extend these findings, we performed HGT against human disease ontology databases and, again, found that cancer-related pathways, particularly leukemia- and lymphoma-related pathways, were highly enriched among STAT5-regulated genes but not MYC-regulated genes (fig. S11A). Next, we revisited our cotransduction studies and found that leukemia- and lymphoma-related pathways were enriched only in group 3, DEG mobilized by CA-STAT5 with or without MYC (fig. S11, B and C). Focusing on T cell malignancy, we then performed GSEA on ranked transcriptomes from peripheral T cell lymphomas (PTCL) stratified by patient outcome (36). This analysis showed that group 3 was highly enriched in samples from patients with poor prognoses, as was group 2, DEG called only in cells transduced with both CA-STAT5 and MYC (fig. S11D). We also noted that, in both cases, enrichment was greater when comparing poor versus good prognosis than poor prognosis versus normal controls. These data imply that cooperation between STAT5 and MYC may be particularly relevant for PTCL.

We next explored whether colocalization of STAT5 and p300 is relevant in the context of lymphoid cancers. First, we noted that IL-2–STAT5 signature genes with coincident p300 are far more enriched for leukemia- and lymphoma-related pathways than those without (fig. S11E). We also found that IL-2–STAT5 signature genes with coincident p300 are strongly enriched in PTCL patients with poor prognoses but in not diffuse large B cell lymphoma (DLBCL) patients with poor prognoses, with one notable exception (fig. S11, F and G) (37). BCL2/MYC “double hit” DLBCL was different and more similar to PTCL in two important ways: (i) IL-2–STAT5 signature genes were highly enriched, and (ii) STAT5-bound genes were more enriched than unbound genes (fig. S11G). Thus, we conclude that STAT5 drives expression of genes associated with lymphoid cancers through canonical transcriptional mechanisms involving p300 and noncanonical mechanisms that may not involve p300.

DISCUSSION

Many studies have addressed the role of mTOR in T_eff metabolism, but few have addressed STAT5, a cardinal feature of cγ cytokine signaling (1, 3, 7). Here, we report that the IL-2–STAT5 axis instructs transcription of key genes involved with anabolic and catabolic metabolism, genes involved with mTOR activation, and the transcription factor MYC. It does so by localizing to relevant promoters and enhancers, where it mediates both p300-dependent and independent transcriptional effects. We also report that, with few exceptions, IL-2–STAT5 signaling does not induce SE at metabolism-related gene loci. This finding is consistent with previous studies (38) and suggests that genes involved with metabolism pathways are more similar to “housekeeping” genes than lineage-defining genes (29). Notwithstanding, we propose that STAT5 is a master regulator of T_eff metabolism, specifically of glycolysis and amino acid synthesis, in most (if not all) settings where T_eff are generated, whether protective or pathogenic (fig. S12). We also hypothesize that STAT5 plays a similar role in other cell types, downstream of other stimuli. Supporting this latter point, STAT5 and mTOR are each activated by insulin, growth hormone, and other growth factors with broad cellular ranges.

To explore the role of STAT5, we first asked which metabolites are most affected by IL-2, the principal STAT5 stimulus in T_eff. As expected, we found that IL-2 rapidly induced multiple glycolysis intermediates and later determined that downstream STAT5 signaling instructs transcription of key enzymes along the core glycolysis pathway, including Hk2, Eno1, and Pkm. We were also impressed that it rapidly induced nearly all amino acids and later determined that downstream STAT5 signaling instructs transcription of key genes involved with amino acid import and synthesis, including SLC7A5, the principal transporter for essential amino acids in T cells (11), and GOT1, which catalyzes the conversion of aspartate to oxaloacetate and thereby provides a source of nitrogen for synthesis of nonessential amino acids. The latter is particularly relevant here because unlike quiescent cells, which metabolize glutamate through glutamine dehydrogenase 1, rapidly proliferating T cells do so mainly via glutamic-oxaloacetic transaminase (GOT) enzymes (39). We also noted that IL-2 promotes accumulation of biochemicals necessary for DNA replication and packaging, specifically nucleotides and polyamines, and, again, the data point to STAT5-driven transcription of genes involved with their synthesis and/or import.

It is widely believed that IL-2 activates mTOR signaling via the PI3K-AKT pathway. We find that IL-2 does engage PI3K-AKT in CD4⁺ T_eff but also that PI3K-AKT is not required for downstream mTOR signaling, in line with previous studies (7, 15). Instead, we demonstrate that STAT5 is required and offer two nonexclusive mechanisms for how it promotes mTOR activity. First, it instructs transcription of the α and β components for IL-2R, which sets the stage for both JAK-STAT and mTOR signaling. Second, it instructs transcription of key genes along the core mTOR pathway, including key elements of the amino acid sensory system. Building on the latter, we demonstrate that defects in IL-2R expression do not wholly account for the lack of mTOR activity seen in STAT5-deficient T cells, which affirms that STAT5 plays a key role downstream of IL-2R. Consistent with previous reports (21), we also show that STAT5 instructs transcription of MYC, a transcription factor with broad effects on T cell metabolism, and that ectopic MYC does not correct the metabolic defects seen in STAT5-deficient T cells, at least not at the transcriptional level. This later finding was unexpected given that many of the metabolism pathways affected by STAT5 deficiency are known to be regulated by MYC (22, 23, 40). We do not infer that MYC is metabolically inert without STAT5—it still made STAT5-deficient cells larger and mobilized certain metabolism pathways—but propose that they cooperate to control expression of metabolism-related genes. We also stress that they operate in different ways. STAT5 prompts digital, high-amplitude changes in gene transcription, whereas MYC prompts analog, low-amplitude changes. This implies that STAT5 first turns genes on and then MYC amplifies them, which is consistent with current models of how each affects gene expression.

Beyond T_eff metabolism, we demonstrate that STAT5 is required for MYC-driven transcriptome amplification, a capacity that likely contributes to its oncogenic properties (22, 33). We also present evidence that cooperation between STAT5 and MYC is relevant for lymphoid cancers and, on the basis of post hoc analysis of patient transcriptomes, further propose that high STAT5 activity pushes toward poor prognosis in PTCL and DLBCL with dual MYC and BCL2 mutations (i.e., double hit lymphomas), in line with previous studies linking STAT5 and pathogenesis in B cell leukemias (41, 42). Given that IL-2 is not a major STAT5 stimulus for B cells, our findings also speak to its general oncogenic character and suggest that relevant target genes may be agnostic to cellular lineage or upstream stimulus. Thus, despite limitations inherent to our reliance on mouse models, in vitro systems, and bioinformatics, the data presented here coalesce to establish that STAT5 is central to T cell metabolism and further suggest that this property is relevant for lymphoid malignancies.

MATERIALS AND METHODS

Study design

The objective of this study was to define the role of IL-2–STAT5 signaling in T_H cell metabolism. In vivo studies were conducted in mice. In vitro studies involved primary mouse cells. Multiple experimental techniques were used, including flow cytometry, mass spectrometry, RNA-seq, ChIP-seq, and bioflux analysis. The sample size and the number of independent biological replicates for all experiments are detailed in table S9.

Animals

Stat5a/b ^flox/flox Cd4-Cre^+/− and Cd4-Cre^−/− littermate controls were generated as described (43). These were crossed with OT-II TCR and Thy1.1 transgenic mice to generate the Stat5a/b ^flox/flox Cd4-Cre^+/− OT-II^+/− Thy1.1^+/+ mice used for natural antigen studies (the Jackson Laboratory). Stat5a^−/− Stat5b^+/− (labeled as Stat5a^−/− in fig. S3) and Stat5a^+/− Stat5b^−/− (labeled as Stat5b^−/− in fig. S3) were generated as described (44). WT C57BL/6J mice were purchased from the Jackson Laboratory (strain: 000664). Both male and female mice were used and sex-matched within experiments. Animals were housed and handled in accordance with National Institutes of Health (NIH) guidelines, and all experiments were approved by the National Institute of Arthritis and Musculoskeletal and Skin Diseases (NIAMS) Animal Care and Use Committee.

Immunizations

Stat5a/b^flox/flox Cd4-Cre^+/− and Cd4-Cre^−/− mice were immunized with 10 μg of lymphocytic choriomeningitis virus (LCMV) GP 61-80 peptide (GLKGPDIYKGVYQFKSVEFD; GenScript) emulsified with Complete Freund’s adjuvant (Sigma-Aldrich). Nine days later, draining lymph nodes were dissected, and single-cell suspensions were cultured overnight with or without human IL-2 (100 U/ml; NIH/National Cancer Institute Biological Resources Branch Preclinical Repository) (fig. S6E). Four or five mice were immunized per genotype over two independent trials. APC-labeled, GP 61-80–loaded major histocompatibility II (MHC II) tetramer was used to track antigen-specific T cells (NIH MHC tetramer core).

Cell cultures

Unless otherwise noted, in vitro cultures were seeded with naive CD4⁺ T cells sorted from pooled lymph nodes and spleens of WT or STAT5-deficient mice (CD4⁺ CD44^low CD25⁻, >95% purity) (fig. S3B). These were cultured at 1 × 10⁶ cells/ml and stimulated with plate-bound anti-CD3ε (10 μg/ml; clone 17A2, Bio X Cell, USA) and soluble anti-CD28 (1 μg/ml; clone 37.51, Bio X Cell) under conditions detailed in table S7. For studies involving natural T_reg, CD4⁺ FoxP3-GFP^high cells were sorted from pooled lymph nodes and spleens of WT FoxP3-GFP “reporter” mice (fig. S1G). These were immediately processed for RNA-seq (“no cytokine group”) or cultured for 16 hours with human IL-2 (10 U/ml) (no further sorting before sample collection). For differentiation studies, cultures were seeded with naive CD4⁺ T cells and stimulated with the following cytokine/antibody combinations: T_H0 = anti–IL-4 + anti–interferon-γ (IFN-γ), T_H1 = IL-12 + anti–IL-4, T_H1 = IL-4 + anti–IFN-γ, T_H17 = IL-6 + transforming growth factor-β (TGF-β) + anti–IL-4 + anti–IFN-γ, and induced T_reg (iT_reg) = TGF-β + anti–IL-4 + anti–IFN-γ (Fig. 3I). For drug cytometry studies, cells were treated with hIL-2 and ipatasertib, MK-2206, rapamycin, or tofacitinib for 24 hours (Selleck Chemicals, USA). For drug RNA-seq studies, naive CD4⁺ T cells were stimulated with plate-bound anti-CD3ε and anti-CD28 (10 μg/ml each) and 0.5 μM rapamycin or 0.3 μM tofacitinib for 72 hours. Mouse IL-4, IL-6, IL-12, and TGF-β were purchased from R&D Systems (USA) and used at 10 ng/ml [except TGF-β (1 ng/ml) for T_H17 conditions]. Anti-mouse IL-2 (clones S4B6 and JES6-5H4), anti-mouse IL-2Rα (clone PC61), and anti-mouse IL-2Rβ (clone TM-b) were purchased from Thermo Fisher Scientific, USA and used at 10 μg/ml. All cultures were in RPMI 1640 medium supplemented with 10% fetal calf serum, 1% sodium pyruvate, 1% nonessential amino acids, 10 mM Hepes, 0.1% β-mercaptoethanol, penicillin (100 U/ml), and streptomycin (100 mg/ml). Unless otherwise noted, cultures were supplemented with anti-mouse IL-4 and anti-mouse IFN-γ (10 μg/ml each; clones 11B11 and XMG1.2, Bio X Cell).

Flow cytometry

For surface antigens, cells were stained in phosphate-buffered saline supplemented with 0.5% bovine serum albumin and 0.1% sodium azide. For intracellular antigens, cells were fixed with Cytofix/Cytoperm (BD Biosciences) and permeabilized with 100% methanol. Fluorochrome-labeled anti-mouse AKT1 p-S473, CD3ε, CD4, CD8α, CD19, CD25 (IL-2Rα), CD44, CD45, CD62L, CD98, CD122 (IL-2Rβ), CD127 (IL-7Rα), FOXP3, human nerve growth factor receptor (hNGFR), Ki67, mTOR p-S2448, MYC, STAT5 p-Y694, S6 p-S235/6, and total S6 kinase antibodies were purchased from Thermo Fisher Scientific, BD Biosciences, or BioLegend (USA). Unlabeled rabbit polyclonal anti-FoxO1 p-Ser256 was used in conjunction with phycoerythrin (PE)–labeled anti-rabbit antibody (Cell Signaling Technologies). Dead cells were excluded with Live/Dead Aqua (Invitrogen, USA). Data were collected on FACSVerse (BD Biosciences) or Aurora (Cytek Biosystems, USA) cytometers and analyzed with FlowJo software (FlowJo LLC, USA). Boxplots and line graphs are compiled from >3 biological replicates and normalized to genotype-matched controls. Horizontal lines indicate means, whiskers indicate minimum or maximum values, and dotted red lines indicate twofold thresholds.

Retroviral gene transduction

Retroviral vectors containing mutant mouse STAT5A (CA-STAT5 = S711F + H299R), myristoylated mouse AKT1 (CA-AKT), and mouse c-MYC were generated as described (10, 45, 46). For IL-2Rα vectors, mouse Il2ra cDNA was synthesized on the basis of reference sequence ENSMUST00000028111.6 and then subcloned into an MoMLV-based plasmid vector immediately before an internal ribosome entry sequence and green fluorescent protein (GFP) marker. Vectors and pCL-Eco “helper” plasmid were transfected into 293T cells (American Type Culture Collection, USA) using Lipofectamine (Invitrogen), and then virus-containing supernatants were collected 48 hours later. For transductions, naive CD4⁺ T cells were stimulated with plate-bound anti-CD3ε and anti-CD28 (10 μg/ml each) in the presence of anti-mouse IL-2 antibodies for 48 hours and then exposed to viral supernatant for 1 hour (at 2200 rpm, 18°C), rested for 18 hours (with anti-mouse IL-2), and pulsed with human IL-2 for 24 hours (as per fig. S3A).

Western blots

Naive WT CD4⁺ cells were stimulated with plate-bound anti-CD3ε and anti-CD28 (10 μg/ml each) in the presence of anti-mouse IL-2 for 72 hours and then recultured with human IL-2 for 24 hours (100 U/ml), rested for 12 hours, and stimulated with either anti-CD3ε and anti-CD28 or IL-2 for the indicated tie points. Cells (10⁷) were then lysed in buffer containing 50 mM Hepes (pH 7.4), 150 mM NaCl, 20 mM NaF, 4 mM EDTA, 50 μM sodium orthovanadate, small peptide inhibitors, 0.5 mM phenylmethylsulfonyl fluoride, and 0.5% (v/v) Triton X-100 (Roche, CH). Protein was precipitated with cold acetone, resuspended in 2× Laemmli reducing sample buffer, separated on a 4 to 12% SDS–polyacrylamide gel electrophoresis gel, and Western blotted onto nitrocellulose (Invitrogen or Bio-Rad, USA). Proteins were identified using anti–p-STAT5 (clone A-9, Santa-Cruz Biotechnologies, USA), p-Thr³⁰⁸ PKB/Akt (catalog 9275, Cell Signaling Technologies, USA), p-Thr²⁴–Foxo1 (catalog 2599, Cell Signaling Technologies), or β-actin (catalog 3700T, Cell Signaling Technologies).

Metabolomes

Naive WT CD4⁺ T cells were cultured as in Fig. 1A. Viable cells (1.5 × 10⁶) were sorted per replicate before and after IL-2 pulses. These were washed in phosphate-buffered saline, snap-frozen in liquid nitrogen, and shipped for analysis by liquid chromatography–tandem mass spectroscopy (Metabolon Inc., USA). Four hundred thirty-four known biochemicals were detected across all groups/replicates. Welch’s two-sample t test was used to call DEB after normalization on the basis of total DNA or protein concentrations, log transformation, and imputation of minimum observable values for each compound. Five replicates were analyzed per group. BioCyc Database Collection analysis tools were used for DEB pathway analysis (https://biocyc.org).

Metabolic flux analysis

For polyclonal studies, naive CD4⁺ cells were stimulated with plate-bound anti-CD3ε and anti-CD28 (10 μg/ml each) in the presence of anti-mouse IL-2 for 72 hours and then washed and cultured with human IL-2 for 24 hours (100 U/ml). These were then transferred to Seahorse glucose-free, unbuffered RPMI media with 100 mM sodium pyruvate, 2 mM glutamine, and 1% fetal calf serum and cultured with hIL-2 (100 U/ml) for an additional 4 hours (Agilent, USA). Next, they were plated onto a XF96 PET cell culture plate coated with Cell-Tak (Corning, USA) at 250,000 cells per well and incubated for 1 hour at 37°C in the absence of CO₂ before measurement of metabolic flux. ECAR and OCR were assessed at basal conditions and after addition of glucose (10 mM), oligomycin (1 μM), and 2-deoxyglucose (20 mM) for ECAR or glucose (10 mM), oligomycin (1 μM), carbonyl cyanide p-trifluoromethoxyphenylhydrazone (1 μM), and rotenone (1 μM) for OCR (all from purchased from Agilent). Three measurements were recorded for basal metabolic rates and after each injection using a Seahorse Extracellular XF96 analyzer.

For natural antigen studies, naive OT-II TCR transgenic CD4⁺ cells were labeled with carboxyfluorescein diacetate succinimidyl ester (Sigma-Aldrich) and cultured with bone marrow–derived dendritic cells pulsed with OVA 323 to 339 peptide (100 ng/ml) (AnaSpec, USA) (2:1 T cell:dendritic cell ratio) (Fig. 2F). Five to six days later, viable cells that had undergone at least two divisions were sorted, cultured overnight with or without hIL-2, and assayed by Seahorse Extracellular XF96 analyzer as above. For these experiments, 250,000 cells per well were preincubated in XF Dulbecco’s modified Eagle’s medium supplemented with glutamine, glucose, and pyruvate (200 μM each in 20 ml of medium) before bioflux analysis.

RNA sequencing

Viable, CD4⁺ T cells were sorted before and after in vitro culture (>95% purity; fig. S3, B and C). For retroviral experiments, cells were further selected on the basis of transduction markers GFP or hNGFR (Fig. S3C). Two to four biological replicates were collected for each experimental group with similar numbers of cells per replicate (10 to 100 × 10³, depending on experimental condition; detailed in table S9). Sorted cells were immediately lysed in TRIzol reagent, and the total RNA was purified by phenol-chloroform extraction with GlycoBlue as co-precipitant (7 to 15 μg per sample; Life Technologies, USA). Poly(A)-tailed mRNA was enriched by oligothymidilate-based magnetic separation, and single-end read libraries were prepared with the NEBNext Ultra RNA Library Prep Kit (New England Bioxlabs, USA). Sequencing was performed with HiSeq 2500 (Illumina, USA), 50–base pair reads (20 to 50 × 10⁶ per sample) were aligned onto mouse genome–build mm9 with tophat2 and assembled with cufflinks, and gene-level counts were compiled with htseq-count. To minimize normalization artifacts, we purged genes failing to reach an empirically defined count threshold using htsfilter. A total of 12 × 10³ to 14 × 10³ genes were typically recovered after filtering, regardless of genotype or experimental condition. Read counts were normalized and DEG called by quasi-likelihood F testing using edgeR. DEG call denotes a >2-fold pairwise change and a Benjamini-Hochberg–adjusted P < 0.05. Transcripts per million (TPM) were compiled with edgeR. An offset value of 1 was added to all TPM, and those failing to reach a value of >2 TPM in any genotype/condition were purged, as were microRNAs, small nucleolar RNAs (sno-RNAs), and small Cajal body–specific RNAs (sca-RNAs). TPM + 1 was the input for Euclidian clustering (hclust) and PCA (prcomp). External RNA Controls Consortium (ERCC) RNA spike-in controls were used to calculate TPC as described (47).

clusterprofiler was used for HGT and GSEA against the KEGG, Gene Ontology (GO), Reactome, MSigDB, Disease Ontology (DO), and DisGeNET databases or custom gene sets (48, 49). All custom gene sets are cataloged in tables S4 and S5. For GSEA, transcriptomes were ranked by a composite metric (Log₂FC × −log₁₀ adjusted P value). GSEA plots were rendered with enrichplot, heatmaps with pheatmap, KEGG pathway maps with pathview, and all other plots with ggplot2 or Datagraph (Visual Data Tools Inc., USA).

Lymphoma transcriptomes

Peripheral T-cell lymphoma not other specified (PTCL-NOS), transcriptomes, and clinical data were downloaded from the National Center for Biotechnology Information Gene Expression Omnibus (NCBI GEO) (GSE 58445) (36). DLBCL, transcriptomes, and clinical data were sourced from (37). All data were processed using AltAnalyze, and patients were classified as having a good or poor prognosis based on the 5-year survival relative to healthy controls. Transcriptomes were ranked on the basis of pairwise fold change values relative to healthy controls and pathway enrichments measured by GSEA.

Chromatin immunoprecipitation sequencing

Naive WT CD4⁺ cells were stimulated with plate-bound anti-CD3ε and anti-CD28 (10 μg/ml each) in the presence of anti-mouse IL-2 for 48 hours, washed, cultured with human IL-2 for 48 hours (100 U/ml; no anti-CD3/CD18), and pulsed again with human IL-2 (1 hour). Viable cells (3 × 10⁷) were then fixed with 1% formaldehyde, lysed, and sonicated. DNA was then split and immunoprecipitated as follows: 50% with rabbit anti-mouse STAT5A and STAT5B (10 μg, PA-ST5A and PA-ST5A; R&D Systems), 25% with rabbit anti-mouse p300 (10 μg; sc-585, Santa Cruz Biotechnology), 12.5% with rabbit anti-mouse H3K27ac (5 μg; ab4729, Abcam, USA), and 12.5% with rabbit anti-mouse H3K4me1 (5 μg; ab8895, Abcam, USA). Recovered protein-DNA fragments, along with unprecipitated input controls, were decross-linked and blunt-end ligated to adaptors, and single-end libraries were constructed using the NEBNext ChIP-Seq Library Prep for Illumina kit. Fifty–base pair reads were aligned using bowtie, and nonredundant reads were mapped to mouse genome mm9 using macs2 with input controls as reference for peak calling. Homer was used to annotate peaks, call SE, and test for transcription factor motif enrichments. Gene proximal peaks were defined as occurring within introns, exons, untranslated regions, promote transcription start sites, transcriptional termination sites (TTS), or <20 kb of transcriptional start sites. Genome browser files were rendered with Integrated Genome Viewer (IGV). One of two individual experiments was used for all analyses. The MYC ChIP-seq dataset is published (33).

Statistics

Statistical variances and distributions were measured by paired t test, Kolmogorov-Smirnov test, or chi-square test as per table S8. Bonferroni correction was used to account for multiple testing in RNA-seq and ChIP-seq datasets. Biological replicates for each experiment are detailed in table S9. When present, error bars denote SD across >2 biological replicates.