H3K27 Methylation Dynamics during CD4 T Cell Activation: Regulation of JAK/STAT and IL12RB2 Expression by JMJD3

CD4 T cells are an integral component of the adaptive immune response, facilitating Ag-specific “memory” via adaptation following exposure to pathogens and other foreign invaders. The molecular mechanisms responsible for the transition to memory are still poorly understood and likely stem from multiple epigenetic processes in the cell following activation (1). Numerous epigenetic marks have been explored in cancer and embryonic stem (ES) cell differentiation, including DNA methylation, histone modifications, and chromatin regulators, all of which appear to play roles in modulating gene expression (2–5). These marks can target multiple regulatory regions, including promoters, enhancers, superenhancers, gene bodies, and intergenic regions, which confer another layer of complexity to this already enigmatic system. Epigenetic studies in T cells have been more limited and in this early phase of investigation are still largely descriptive, although clear correlations between epigenetic modifications and T cell differentiation have been illustrated with genome-wide studies (6–13).

H3K27me3, in particular, is a conventionally “repressive” histone modification (9, 14) that plays a role in CD4 T cell differentiation (15, 16). However, its dynamics during CD4 T cell activation and early differentiation has not been explored, and its role in early differentiation is still poorly characterized. One study performed global mapping of H3K27me3 after in vitro polarization of murine CD4 T cell differentiation to reveal that the presence of H3K27me3 in Th-related genes corresponded to silencing of those genes in their opposing lineages (7). A study of murine CD8 T cell dynamics after viral infection also demonstrated a profound loss of H3K27me3 following activation, supporting the role of repressive H3K27me3 marks in naive CD8 T cells to maintain a state of restraint during rest (8).

Two demethylases, JMJD3 and UTX, are known to catalyze H3K27me3 demethylation. Recently an exploration into the role of Jmjd3 in mice upon the regulation of CD4 T cell differentiation found that a conditional knockout of Jmjd3 resulted in skewing to Th2 and Th17 differentiation (15). Both demethylases are required for in vivo thymocyte differentiation in mice (17). In ES cells, JMJD3 appears to delocalize polycomb repressive complex (PRC) proteins, which is essential for further development. Additionally, UTX is a component of the MLL complex, strongly suggesting that H3K27 demethylation can be coupled with the “activating” methylation of H3K4 by MLL (18). UTX is ubiquitously expressed in tissues and is also important for embryonic cell development (19). In contrast, JMJD3 is commonly induced during inflammation or upon exposure to antigenic or oncogenic stimuli (18, 20, 21). JMJD3 inhibits somatic cell reprogramming in inducible pluripotent stem cells, whereas UTX is essential for it, suggesting contrasting roles for these two enzymes (22, 23). The two enzymes also play contrasting roles in acute lymphoblastic leukemia, with JMJD3 inducing the neoplastic process and UTX acting as a tumor suppressor (24).

In the current study, we have examined the dynamics of promoter-associated H3K27me3 upon activation of human naive and memory CD4 T cells. We find that, in both subsets, profound demethylation of H3K27 is observed by 1 d after activation, which is in contrast to H3K4 methylation, for which changes are not observed until days later (25). Mapping specific states of H3K27me3 to known immune pathways demonstrates that loss of H3K27me3 early in activation corresponds to pathways crucial to T cell function, including T cell activation and the JAK/STAT pathways. Mechanistic studies by perturbation of H3K27 demethylation with a small-molecule inhibitor (GSK-J4) and small interfering RNA (siRNA) knockdown of the two H3K27 demethylases confirms that H3K27 demethylation by JMJD3 is important for key members of early differentiation–related pathways. Altogether, these data confirm that H3K27 is a highly dynamic epigenetic modification in CD4 T cells during early activation, and the nature of these dynamic changes is crucial to CD4 T cell function.

All of the studies were covered by Human Subjects Research Protocols approved by the Institutional Review Board of The Scripps Research Institute. Informed written consent was obtained from all study subjects.

Peripheral blood was collected from healthy donors, and PBMCs were collected by centrifugation through a Histopaque (Sigma) gradient. CD4 T cells were negatively selected from four donors using an EasySep Human Naive CD4+ T or Memory CD4 T Cell Enrichment Kit (STEMCELL Technologies). Cell purity was assessed by flow cytometry with Abs specific for CD4, CD45RA, and CD45RO (SK3, HI100, UCHL1; eBioscience). Data acquisition was conducted on an LSR II (BD Biosciences), and analysis was performed using FlowJo (TreeStar). Live cells were gated based on forward by side scatter area, and doublets were excluded based on forward scatter height by forward scatter width and side scatter height by side scatter width. Live cells were then gated on CD4 staining, and cell purity following isolation was determined by CD45RA versus CD45RO staining. Cell purity for all donors was >94%.

CD4 T cells were cultured in RPMI 1640 (Mediatech) supplemented with 100 U/ml penicillin, 100 μg/ml streptomycin, and 10% FBS at 37°C and 5% CO₂. T cells were activated with Dynabeads Human T-Activator CD3/CD28 (Invitrogen) for 1 and 5 d. Cells maintained in culture out to 2 wk received 30 U/ml human rIL-2 every 2 d (National Institutes of Health repository) beginning at day 5 after activation. Samples for chromatin immunoprecipitation (ChIP)-sequencing (ChIP-seq) and RNA sequencing (RNA-seq) were collected from the four donors at rest at 1 and 5 d and 2 wk after activation. GSK-J4 experiments were conducted with a 24-h incubation in 12.5 μM GSK-J4 in 0.2% DMSO alongside a 0.2% DMSO vehicle control prior to activation. T cells were activated with Dynabeads Human T-Activator CD3/CD28 (Invitrogen) for 8 and 24 h and collected for quantitative RT-PCR, ChIP–quantitative PCR (qPCR), and flow cytometry, as described below. For IL-6 rescue experiments, 20 ng/ml human rIL-6 (Tonbo Biosciences) resuspended in PBS was added to media at the time of activation, whereas an equal volume of PBS (2 μl/ml) was added to the control.

A total of 3 × 10⁷ cells was fixed for 10 min in cell culture medium with 1% formaldehyde at room temperature. Fixation was quenched with 10% glycine for 5 min at room temperature. Cell pellets were flash frozen in liquid nitrogen and stored at −80°C until chromatin isolation. Chromatin was isolated from cell pellets using the ChIP-It Express Enzymatic Shearing Kit (Active Motif), per the manufacturer’s instructions. For immunoprecipitation, 7.5 μg of chromatin was diluted in a total of 1 ml of low-salt wash buffer (0.1% SDS, 1.0% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl [pH 8.1], 500 mM NaCl) with cOmplete Protease Inhibitor Cocktail (Millipore Sigma) and precleared with 30 μl of Invitrogen Dynal Protein G magnetic beads for 2 h at room temperature with rotation. Twenty microliters of precleared chromatin was saved for input analysis and stored at −80°C. Remaining chromatin was incubated with 2 μl of anti-H3K27me3 Ab (catalog number 17-622; Millipore) overnight at 4°C with rotation. Thirty microliters of Invitrogen Dynal Protein G beads was added to each ChIP assay and incubated at 4°C with rotation for 2 h. Beads were washed three times in low-salt wash buffer and then two times with high-salt wash buffer (0.1% SDS, 1.0% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl [pH 8.1], 500 mM NaCl) with 5 min of rotation at 4°C for each wash. Beads were resuspended in 150 μl of elution buffer (1% SDS, 0.1 M NaHCO₃) and incubated in a ThermoMixer at 65°C for 30 min at 1200 rpm to reverse cross-linking. Two microliters of Proteinase K (Invitrogen) was added to each sample, and 6 μl of 5 M NaCl was added for a total concentration of 200 mM. Samples were then incubated in a ThermoMixer at 65°C overnight at 1200 rpm. Eluted samples were removed from the beads and purified using the QIAGEN QIAquick PCR purification kit, per the manufacturer’s instructions.

RNA for RNA-seq was isolated from purified cells using an AllPrep kit (QIAGEN), following the manufacturer’s instructions, and purified total RNA was converted to cDNA using Ovation RNA-seq (NuGEN), followed by S1 endonuclease digestion (Promega), as previously described (26). Digested cDNA libraries were then end-repaired and A-tailed. Indexed adapters were ligated, and ligation product was purified on Agencourt AMPure XP Beads (Beckman Coulter Genomics), followed by size selection from 2% agarose. Purified product was amplified with 15 cycles of PCR, followed by size selection from 2% agarose. Libraries were assessed on an Agilent Bioanalyzer using a DNA chip and quantitated using the Quant-iT dsDNA Assay Kit, broad range (Invitrogen) and a Qubit Fluorometer (Invitrogen). Cluster generation and sequencing of 100-bp single-end reads on an Illumina HiSEquation 2000 system was conducted with purified libraries, following the manufacturer’s instructions. Depth of sequencing per sample was 15 million aligned reads.

For ChIP-seq, 10 ng of purified DNA from individual ChIP assays were end-repaired and A-tailed. Indexed adapters were ligated, and ligation product was purified on Agencourt AMPure XP Beads and processed using Illumina DNA TruSeq protocols.

For JMJD3 ChIP-seq, 3 million naive CD4 T cells were fixed by resuspending in 1 ml of culture medium and cross-linking with 100 μl of fixation buffer (High Sensitivity ChIP kit; Active Motif) at room temperature for 15 min. Fifty-five microliters of stop solution was added to quench unreacted formaldehyde and incubated at room temperature for 10 min. Fixed cells were centrifuged at 700 × g at 4°C for 10 min and washed with 5 ml of cold PBS. After aspirating the supernatants, fixed pellets were stored at −80°C. Prior to chromatin preparation, pellets were thawed on ice and resuspended in 1 ml of ice-cold Lysis Buffer supplemented with 5 μl of cOmplete Protease Inhibitor Cocktail + 5 μl of PMSF (ChIP-IT Express Enzymatic Shearing Kit; Active Motif). Pellets were incubated on ice for 30 min and then centrifuged at 5000 rpm for 10 min at 4°C. Pelleted nuclei were resuspended in 130 μl of ice-cold shearing buffer (10 mM Tris [pH 8.1] + 0.1% SDS + 1 mM EDTA) and transferred into a Covaris microTUBE. Chromatin was sheared into 200 bp (5% duty, Intensity: 4, 200 cycles per burst, 30 s per cycle for 30 min, 4°C). Sheared chromatin was centrifuged at maximum speed for 10 min at 4°C. Newly prepared chromatin was transferred to a new tube and resuspended in 250 μl of PBS with 1 μl of cOmplete Protease Inhibitor Cocktail. Chromatin was precleared with 10 μl of protein G beads and rotated for 2 h at 4°C. Precleared chromatin samples were incubated with 2 μg of anti-JMJD3 Ab (catalog number ab38113; Abcam) with rotation overnight at 4°C, and immunoprecipitation was performed as for H3K27me3 ChIP.

Human naive CD4 T cells were transfected by electroporation with 2 μM JMJD3-specific siRNA (ON-TARGETplus siRNA SMARTpool; GE Dharmacon) or scrambled siRNA controls. Cells were then activated with anti-CD3/CD28 beads at 10 μl per million cells 24 h after transfection, and cells were collected 24 h after activation. Total RNA from 5 million JMJD3 small interfering RNA–treated cells or scramble control cells of four donors was isolated using an RNeasy Plus Mini Kit (catalog number 74136; QIAGEN) and quantified by Qubit. Twenty-five nanograms of total RNA of each sample was used as input for library preparation with an Ion AmpliSeq Transcriptome Human Gene Expression Kit (catalog number A26325; Life Technologies). Barcoded libraries were multiplexed and templated on an Ion Chef System and then sequenced on an Ion Proton using an Ion AmpliSeq Transcriptome Human Gene Expression Panel.

Reads were aligned to hg19 using Bowtie 2 (27). Sample normalization factors adjusting for sequencing depth and compositional bias for each histone mark were determined by unweighted trimmed mean of M values on 10-kb bin read counts (28). These normalization factors were used in all differential binding analyses described below. For each condition, peaks were called independently for each of the four donors using the MACS peak caller on the aligned reads. Then, biologically reproducible consensus peaks were determined using the Irreproducible Discovery Rate (IDR) framework with a threshold of 0.01. The same process was repeated with all aligned reads from all conditions to obtain a single set of condition-independent consensus peaks for cross-condition comparison. This procedure was repeated for H3K27me3 and H3K4me3 samples to obtain condition-specific and condition-independent peak sets for both histone marks. For each histone mark, reads whose 5′ mapping location overlapped each condition-independent consensus peak were counted for each sample. Counts were analyzed for differential binding between conditions using edgeR’s quasi-likelihood F-test (29), with a model including the condition as the main effect and donor as a batch effect. The p values were adjusted for multiple testing using the Benjamini–Hochberg procedure (30). Sliding window–based differential binding analysis was performed (28) using the same model as for the peak analysis. In each comparison, windows were only included in the analysis for that comparison if they had a mean log counts per million (CPM) of −1 or more across all samples in that comparison.

To determine the effective promoter radius for each histone mark, the distance from each unique transcription start site (TSS) annotated in the UCSC gene database to the distance to the nearest peak was determined for each condition’s consensus peak set. The distribution of these distances was plotted. It was assumed that these histone marks would be uniformly distributed throughout most of the genome but enriched in promoters. For all histone marks and conditions, visual inspection revealed a peak at small distances and flat background level at larger distances, consistent with the assumption of enrichment of peaks near promoters. For H3K27me3, the distribution flattens out to the background level near 2.5 kb. These distances were used as the effective promoter radius when defining the promoter region associated with each TSS. The promoter region of each annotated transcript was defined by extending a region upstream and downstream from the TSS by the determined radius. Overlapping promoters from the same gene were merged into one. The number of reads overlapping each promoter in each sample was counted. Promoter counts were analyzed using the same model as for the peak analysis. As a negative control, the ChIP-seq input samples were also analyzed in the same way to verify that the differential binding test did not give false positive results.

Because dispersions were found to vary with time point, each test for differential binding between conditions was conducted using dispersions estimated from only the samples from time points associated with the conditions being tested. For example, for comparing memory at rest versus memory 5 d after activation, all samples from rest and 5 d were used for estimating dispersions. Results for differential binding were filtered with a false discovery rate (FDR) cut-off of 0.1 and a fold-enrichment cut-off of log₂ fold change >1 or less than −1.

As described for the ChIP-seq analysis, differential binding was calculated when comparing samples across time points, allowing for evaluation of changes in overall enrichment. This type of analysis is a powerful adjunct to the analysis of peaks because it allows for examination of histone modification changes outside of called peaks within a larger sequence frame and has been used in other studies (31–33). Within the 5-kb radius around the promoter that we examined (i.e., 2.5 kb upstream and downstream), peaks might be called in two or more conditions, but calculations for differential binding still suggest that there is differential enrichment. Therefore, we have used both types of analyses in this study when comparing different conditions: one comparison in which a quantitative difference in enrichment is calculated and another in which qualitative comparisons are made for the presence or absence of peaks within the promoter radius.

RNA-seq reads were aligned to the UCSC hg19 transcriptome and genome using TopHat 2 (34). The number of reads aligning unambiguously to each gene in each sample was computed. Genes without at least five reads assigned in at least one sample were considered not detected and were discarded. Normalization factors were computed using trimmed mean of M values, and these normalization factors were used for differential expression analysis and quantification. Gene counts were analyzed for differential expression using the same model as for the peak analysis. Gene expression levels for each sample were quantified as fragments per kilobase per million fragments sequenced (FPKM) using the length of the longest transcript isoform for each gene. Batch-corrected average gene expression levels for each condition were quantified by back-transforming the fitted model coefficients for each condition onto a raw count scale and then normalizing to FPKM as for the sample counts. Cut-offs imposed for differential expression analysis included an FDR < 0.05 and log₂ fold change of ±1.

Tests were performed for correlation between the presence of a ChIP-seq peak at a given experimental condition and either RNA-seq expression level (FPKM values) at a given experimental condition or expression log₂ fold change between two conditions, for all genes in the genome. First, genes were partitioned by promoter peak presence or absence, a Kolmogorov–Smirnoff test (a nonparametric test for distributional differences) was performed to test for significant differences in the RNA-seq statistic of interest between the partitions, and 95% confidence intervals for the difference in means were constructed (based on an assumption of a normal distribution).

Functional pathway mapping was conducted using Ingenuity Pathway Analysis (IPA; QIAGEN, www.qiagen.com/ingenuity) and included p values and log fold changes for differential enrichment or differential expression where appropriate. Pathway mapping in parallel using ImmuneMap, Panther, and Gene Ontology supplemented the final pathway results.

Cells harvested for RNA (5 × 10⁶ cells) were washed three times in 1 ml of PBS without calcium and magnesium (Corning), flash frozen in liquid nitrogen, and stored at −80°C until RNA purification. Total RNA was isolated using an RNeasy Mini Kit (QIAGEN), and RNA concentration was determined by NanoDrop. First-strand cDNA was synthesized from 500 ng of RNA in a 20-μl reaction volume using a Bio-Rad iScript system (Bio-Rad). The complete reaction mix was incubated for 5 min at 25°C, 30 min at 42°C and 5 min at 85°C. The synthesized cDNA was stored at −20°C for further qPCR analysis.

For quantitative real time PCR (qRT-PCR), synthesized cDNAwas mixed with PerfeCTa SYBR Green SuperMix (Quantabio), and one of the primer sets. Reactions were performed on a HT7900 Fast-real-time PCR System (Applied Biosystems) under optimized cycling conditions consisting of a 10 min initial denaturing step at 95°C, followed by up to 40 cycles of amplification (denaturation at 95°C for 10 s, annealing and extension at 60°C for 30 s). Then a melting curve was measured from 65 to 98°C. The specificity of each of the qRT-PCR products was confirmed when a narrow peak appeared in the melting curve above 72°C. The expression level of each gene was evaluated relative to β-2 microglobulin (B2M) using the 2 −ΔΔthreshold cycle (CT) method (35). For ChIP-qPCR, 2 μl of purified DNA from H3K27me3 ChIP or 2% input samples from each condition were subjected to qPCR using the same protocol as for cDNA, and the percentage input was calculated using the equation 100 × 2^{CT(input)−CT(immunoprecipitation)}. Primer sequences included the following: JAK2 forward, 5′-TCTGGGGAGTATGTTGCAGAA-3′ and reverse, 5′-AGACATGGTTGGGTGGATACC-3′; JMJD3 forward, 5′-GGAGACCTTTATCGCCTCTG-3′ and reverse, 5′-TCCCTTTCACCTTGGCATT-3′; UTX forward, 5′-GGACATGCTGTGTCACATCCT-3′ and reverse, 5′-CTCCTGTTGGTCTCATTTGGTG-3′; STAT3 forward, 5′-CAGCAGCTTGACACACGGTA-3′ and reverse, 5′-AAACACCAAAGTGGCATGTGA-3′; IL12RB2 forward, 5′-ATCATGGTGGGCATTTTCTCA-3′ and reverse, 5′-GCTACACCACTGAGGTCTGAG-3′; JAK2 promoter forward, 5′-GCTCTCCCCCAGCCTCTAT-3′ and reverse, 5′-TAGCTTCGAACTCAGCCTCC-3′; and STAT3 promoter forward, 5′-CTGGCTGAACCAAGTCATAACAC-3′ and reverse, 5′-TGGCTGGCTGTGCTGATAAAGC-3′.

Naive CD4⁺ T cells were resuspended in 100 μl of Nucleofector solution (82 μl of P3 Primary cell solution plus 18 μl of Supplement 1; Lonza) and electroporated with 2 μM human KDM6B, human KDM6A, or human JAK2 ON-TARGETplus SMARTpool siRNA (Dharmacon) using a Primer T cell unstimulated HE program (Lonza). Scrambled siRNA (2 μM; Invitrogen) was used for normalization. Then cells were resuspended in prewarmed RPMI 1640 medium with 10% FCS and 1% penicillin streptomycin. Cells were stimulated with anti-CD3/CD28 beads 24 h later and harvested for further analysis at the indicated time points after activation.

Cells from each condition studied were harvested and washed in 2 ml of PBS without calcium and magnesium and stained for 15 min at room temperature with 1 μl of Zombie Violet viability stain (BioLegend) in 100 μl of PBS. Cells were then washed with 2 ml of FACS wash buffer (PBS with 1% FCS and 0.1% sodium azide) prior to each Ab staining protocol.

For IL12RB2, 1 × 10⁵ cells were surface stained on ice for 15 min in 100 μl of FACS wash buffer with 10 μl of anti–IL-12R β2 (clone REA333; Miltenyi Biotec). Cells were then washed in 2 ml of FACS wash buffer and stored in FACS fixation buffer (1× PBS with 4% paraformaldehyde) at 4°C protected from light until collection.

For STAT3, 1 × 10⁶ cells were fixed with 1 ml of FoxP3 Fixation/Permeabilization working solution (eBioscience) for 1 h at 4°C, washed with 2 ml of 1× Permeabilization Buffer (eBioscience), and incubated at room temperature for 1 h in 100 μl of 1× Permeabilization Buffer with 20 μl of anti-STAT3 (clone M59-50; BD Biosciences) protected from light. Cells were then washed in 2 ml of FACS wash buffer and stored in FACS fixation buffer at 4°C protected from light.

For phospho-flow, 1 × 10⁶ cells were incubated at room temperature following viability staining for 1 h at room temperature in 100 μl of 1× Permeabilization Buffer protected from light. Cells were then resuspended in 1 ml of cold 100% methanol and stored at −20°C until staining. Methanol-fixed cells were washed with 2 ml of FACS wash buffer and stained with 5 μl of anti–p-STAT Abs (p-STAT3, LUVNKLA, and p-STAT4, 4LURPIE; eBioscience) in 100 μl of FACS wash buffer at room temperature for 1 h protected from light. Cells were then washed with 2 ml of FACS wash buffer and resuspended in FACS fix at 4°C protected from light until collection.

Data acquisition was conducted on an LSR II (BD Biosciences), and analysis was performed using FlowJo (TreeStar). Live cells were gated based on forward by side scatter area, and doublets were excluded based on forward scatter and side scatter height and width. Cell populations were then gated on Zombie Violet⁻ cells for viability and assessed for their respective Ab stains.

Cells harvested for protein (5 × 10⁶ cells) were washed three times in 1 ml of PBS without calcium and magnesium, flash frozen in liquid nitrogen, and stored at −80°C until protein isolation. Cell pellets were resuspended in 100 μl of RIPA buffer (50 ml Tris-HCl, 150 mM NaCl, 0.1% SDS, 0.5% sodium deoxycholate, 1% Triton X-100) containing complete EDTA-free cOmplete Protease Inhibitor Cocktail and rotated at 4°C for 1 h. Lysed cells were centrifuged at maximum speed at 4°C for 10 min, and supernatants were transferred to a fresh tube and quantitated using a BCA assay (Pierce).

For Western blots, 10 μg of protein was run on a 4–12% Bis-Tris gel in MOPS buffer (Novex) and transferred to a low-fluorescence polyvinylidene difluoride membrane (Immobilon-FL; Millipore). Membranes were blocked with Odyssey Blocking Buffer (LI-COR Biosciences) with 3% goat serum at room temperature for 2 h and then incubated with anti-JAK2 Ab 1:1000 (catalog number ab108596; Abcam) in Odyssey Blocking Buffer overnight at 4°C. Membranes were washed three times for 5 min with PBS containing 0.1% Tween-20 and then incubated in IRDye 680RD Goat anti-Rabbit (LI-COR Biosciences) 1:10,000 in Odyssey Blocking Buffer with 0.001% SDS and 0.02% Tween-20 for 45 min protected from light. Membranes were washed again in PBS/0.1% Tween-20, rinsed in deionized distilled water, and scanned with a LI-COR Odyssey CLx Imaging System (LI-COR Biosciences). Blots were stripped with Restore Western Blot Stripping Buffer (Thermo Fisher) for 20 min at room temperature and then blocked for 1 h at room temperature in Odyssey Blocking Buffer with 3% goat serum. Blots were incubated with anti–α-tubulin Ab at 1:1,000 (catalog number 3873P; Cell Signaling Technology) in Odyssey Blocking Buffer at 4°C for 2 h, washed, incubated in secondary Ab at 1:10,000 (IRDye 800RD Goat anti-Rabbit; LI-COR Biosciences), washed, and scanned as previously described for anti-JAK2. Densitometry with ImageJ was used to quantitate protein band density in scanned blot images.

Hi-C library generation in this study was based on a previously described in situ Hi-C protocol (36) and was optimized for naive CD4 T cells. Briefly, naive CD4 T cells were fixed in culture medium containing 1% formaldehyde for 5 min at room temperature, and 7.5 M Tris-HCl (pH 7.5) was added to a final concentration of 2.5 M to quench the formaldehyde. After washing with PBS twice, cells were resuspended in cell lysis buffer (the same recipe as for JMJD3 and H3K27me3 ChIP) and incubated at 4°C for 30 min. Nuclei were pelleted and subjected to SDS permeabilization, followed by Triton X-100 to quench SDS. Nuclei pellets were subjected to HindIII digestion and religation, at which point biotin-dCTP was induced into the de novo ligation sites. Newly religated genomic DNA was purified and sheared into 200-bp fragments, and fragments containing biotin-labeled de novo ligation sites were purified by M280 streptavidin beads (Invitrogen) and subjected to library preparation. Hi-C libraries were sequenced on the Illumina HiSEquation 2500, targeting 100 million paired reads for each library. Hi-C reads were trimmed at the de novo religation sites (NheI sites generated from religating HindIII digestion) and then mapped to the hg38 reference genome by Bowtie 2. Then, unique mapped reads were subject to filtering to remove self-ligation, dangling ends, and random breaking. Filtered reads were normalized by the iterative correction and eigenvector decomposition method (37), and the Pearson correlation coefficient was calculated at 1 Mb resolution. Because biological replicates in our data set showed high reproducibility (Pearson correlation coefficient > +0.9), all three biological replicates were combined, and chromatin interactions at 25-kb resolution were calculated by HOMER.

To determine the best region in which to evaluate H3K27me3 enrichment, we calculated the average distance from TSSs to H3K27me3 peaks. We (38) and other investigators (6) have recently shown that H3K4 methylation peaks are generally within 1 kb of the TSS. In contrast, our analysis showed that H3K27me3 peaks are primarily clustered within a 2.5-kb radius of the TSS (Supplemental Fig. 1A, 1B). Across all time points, naive cells contained H3K27me3 peaks near a total of 5123 gene promoters and memory cells contained called peaks near a total of 5490 gene promoters within this radius. Therefore, we examined H3K27me3 enrichment 2.5 kb upstream and downstream of the TSS. ANOVA was performed to evaluate donor variability in H3K27me3 enrichment versus variability between conditions. H3K27me3 varies significantly based upon activation, time, and cell type rather than by individual donor (see p value plots, Supplemental Fig. 1C).

To determine whether there was significant variation in signal/background ratios across samples, we plotted the distribution of coverage depths across the genome for each sample (Supplemental Fig. 1D), the majority of which have consistent signal/background ratios. Principal coordinate analysis (39) revealed that the biological effects of interest were visible in the first four principal coordinates, demonstrating that technical confounding factors were not an issue (Supplemental Fig. 1E, 1F).

To ensure that called peaks were robust in the face of technical and biological variation, we used the IDR framework developed for the ENCODE project. Although ENCODE typically uses two biological replicates to assess biological reproducibility, we have four replicates for each group, with the exception of memory at 2 wk, which only includes three donors. This results in six pairwise comparisons for each group, with the exception of memory at 2 wk (three). For each of the pairs, the IDR method produces a count of peaks that can be confidently called at our chosen threshold of reproducibility: IDR = 2%. For each group, we selected the maximum of these six counts, as recommended by the ENCODE documentation for datasets of more than two samples. Then we selected that number of peaks from the top of the list of peaks called on the combined dataset of all replicates. The number of peaks selected for each group is shown in Supplemental Fig. 1G. Because the combined peak calls are based on four times the sequencing depth, the peaks are called more confidently than in the individual replicates; therefore, the 2% threshold represents an upper bound on the fraction of irreproducible peaks. The IDR thresholds were generally consistent across all replicates, and variations in the number of peaks called by IDR do not associate with low signal/background ratios in the coverage depth distributions. Thus, the IDR threshold of 2% is well above the technical noise floor imposed by any issues with ChIP efficiency.

It is also possible that differences in ChIP efficiency could confound estimates of fold changes between sample groups. To assess this possible technical limitation, we looked at the MA-plots of each sample versus all of the other samples of the same type after normalization (Supplemental Fig. 1H, 1I). Each MA-plot generally has a low-abundance “noise mode” [log₂(CPM) < 2], representing the promoters with only background noise, and a high-abundance “signal mode” [log₂(CPM) ∼ 4.5], representing the promoters containing ChIP signal. In the analyzed samples, the signal mode is centered vertically within 0.5 log₂ fold difference of 0, indicating that the edgeR normalization used is effectively controlling for global differences in ChIP efficiency and other technical artifacts. The few samples for which the signal mode is far away from 0 are samples with lower sequencing depth, which means that these samples receive less weight in the model-fitting stage. Given these observations, variations in ChIP efficiency are not responsible for any perturbation larger than ∼0.5 in reported log₂ fold changes.

To examine the dynamics of histone modifications over time, naive (CD45RA⁺CD45RO⁻) and memory (CD45RA⁻CD45RO⁺) CD4 T cells were isolated from the peripheral blood of four healthy human donors and activated with anti-CD3/anti-CD28 beads and cultured in rIL2-supplemented media for 1, 5 d, and 2 wk. Purity of each subset was >94% for all donors (Supplemental Fig. 2A). ChIP-seq for H3K27me3 was performed on cells from all conditions and time points alongside RNA-seq for the same conditions. The raw data for ChIP-seq and RNA-seq from each condition are available at the National Institute of Health Gene Expression Omnibus under accession number GSE73214 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE73214).

Based on our previous data, histone modifications are dynamic during CD4 T cell activation in naive and memory CD4 T cells (25); to our knowledge, this is the first kinetic study of postactivation H3K27me3 marks comparing naive and memory CD4 T cells. We previously showed enrichment for H3K4me2 and H3K4me3 changes in both directions but only relatively late (i.e., 5 d postactivation) in naive and memory cells. In contrast, promoter H3K27me3 methylation changes are extensive by 1 d in both subsets (3801 genes in naive and 1351 in memory) (Fig. 1A, 1B). Of those gene promoters with differential H3K27me3 enrichment, 67% (2981/4473) were demethylated in naive cells, whereas 98% (1548/1575) were demethylated in memory cells (Supplemental Fig. 2B).

H3K27me3 is typically considered a repressive mark associated with lower gene expression. We examined the presence of H3K27me3 peaks at each time point in relation to RNA expression and verified that low expression correlated with the presence of H3K27me3 promoter peaks during activation of naive and memory CD4 T cells (Fig. 1C, 1D). We validated these results by an orthologous analysis for low versus high gene expression to compare the percentage of peaks at all time points in each subset (Fig. 1E, 1F). Surprisingly, promoter H3K27me3 enrichment changes after activation of naive cells appeared only transient, whereas changes in promoter H3K27me3 peak status after activation tended to correspond to more permanent chromatin remodeling, particularly for promoters gaining H3K27me3 (Supplemental Fig. 2C). Interestingly, naive cells showed a dramatic increase in the percentage of promoters containing H3K27me3 peaks 1 d after activation (from 9% at rest to 23%) and ultimately maintained these marks out to 2 wk, particularly in low-expression genes. In contrast, memory cells showed very different dynamics, with an increase in H3K27me3 promoter peaks at 5 d (not at 1 d). However, this change in state was only transient and was lost by the 2 wk time point (Fig. 1F). The genes associated with these promoter peaks in memory mapped to ES cell pluripotency pathways, which heavily overlap with WNT signaling (Supplemental Fig. 2D), and half (20 of 39 genes) overlapped with peaks gained in naive cells at 1 d.

Gene expression for promoters gaining H3K27me3 peaks 1 d after activation of naive cells decreased from 0 h to 2 wk (Fig. 2A) and mapped strongly to WNT signaling (Table I). We next compared the H3K27me3 peaks found in freshly isolated (“at rest”) naive cells with those in freshly isolated resting memory cells. We discovered that memory cells had considerably more H3K27me3 promoter peaks than naive cells (1891 versus 1065, respectively, with 672 shared, Fig. 2B). These results suggest that predetermined H3K27-mediated suppression of specific genes and pathways is one feature of memory cell differentiation and identity. Pathway mapping demonstrated once again that genes marked uniquely in memory cells mapped to pluripotency and WNT signaling (Supplemental Fig. 2E). We mapped the subset of genes whose H3K27me3 marks increased 1 d after naive cell activation, which also mapped heavily to the pluripotency/WNT pathways (Table II), including WNT4 (Fig. 2C) and WNT5B (Fig. 2D). Additionally, this mapping identified bone morphogenetic protein and PDGF pathways, both of which have been linked to pluripotency and self-renewal (40, 41). These data strongly suggest that pathways reinforcing pluripotentiality (42) require suppression during T cell activation and differentiation.

We analyzed RNA expression to compare genes with H3K4me3 peaks from our previous data set (25) versus H3K27me3 and confirmed that genes from all conditions containing H3K27me3 promoter peaks had significantly lower RNA expression (Supplemental Fig. 2F, Supplemental Table I). Therefore, as previously described (6), genes with H3K27me3 peaks are associated with suppression of RNA expression, and this association is maintained throughout T cell activation in naive and memory CD4 subsets.

Why are H3K27me3 changes a primary epigenetic event in the first day after activation? Resting RNA expression is high in genes that subsequently lose H3K27me3 early after activation in naive and memory cells (Fig. 3A, 3B; Supplemental Fig. 2G, 2H). The metric used was the percentage of genes showing an increase or decrease in promoter H3K27me3 during activation compared with their RNA expression at rest. Conversely, genes with low RNA expression at rest gain H3K27me3 with activation (Fig. 3A, 3C), but increased H3K27me3 enrichment occurs primarily in naive, but not memory, cells shortly after activation (Supplemental Fig. 2B). Assuming that promoter H3K27me3 exerts the canonical repressive effects described in the literature, we hypothesize that activation-induced changes to promoter H3K27me3 enrichment early in activation function to maintain gene expression established at rest for selected genes as a function of cell type and differentiation states.

These results raise the question of how changes in H3K27me3 enrichment regulate RNA expression changes with activation. Either decreasing H3K27me3 is operating actively to increase RNA expression after activation, or this mark is passively maintaining the higher levels of expression originally determined for a selected set of genes in the resting or quiescent state. The results demonstrate that in naive (Fig. 3D) and memory (Fig. 3E) CD4 cells, decreasing H3K27me3 marks in the first day correlate with higher levels of gene expression at all time points. In memory cells, there is a transient, but significant, increase in gene expression between rest and 1 d for promoters decreasing in H3K27me3. However, in naive cells, overall levels of gene expression do not increase in the first 5 d after activation, as would be predicted if the impact of decreased H3K27me3 was actively increasing RNA expression rather than simply reinforcing it passively.

Presumably, a naive T cell awaiting activation would have this reinforcement mechanism in place for selected genes that are critical to the process of activation. Indeed, we were able to map decreases in H3K27me3 at 1 d to numerous important canonical functions (Table III), including TCR signaling (Fig. 4A), apoptosis signaling, and CTLA4 signaling, all networks that are integrally related to T cell activation. Although peaks were not always apparent in these data, a measurable decrease in enrichment occurred in many of these sites, as exemplified by the CD45 promoter in Fig. 4B. In Fig. 4C, we show a heat map based on hierarchical clustering of 30 selected immune activation genes in which H3K27me3 enrichment changed in either direction after activation as a function of time (resting and 1, 5, and 14 d). One gene that dramatically loses the H3K27me3 mark during activation and differentiation is DUSP4, which plays a key role in T cell proliferation (43). This gene is heavily enriched for H3K27me3 at rest in naive and memory cells, progressing to low H3K27me3 enrichment after activation (Fig. 4C, 4D) for both cell types, with concurrent and dramatic upregulation of RNA expression by 1 d in naive cells (Fig. 4E).

JAK/STAT pathways were among the top canonical pathways affected by H3K27 demethylation during the early activation of naive CD4 T cells (Fig. 5A). JAK2 and STAT3 are two prominent pathway members, and their promoters are significantly demethylated 1 d after activation (Fig. 5B, 5C). JAK2 phosphorylates several STATs in multiple pathways, including the IL-12R pathway mediated by STAT3 and STAT4, which is necessary for Th1 differentiation (44). Treatment of naive CD4 T cells with the selective H3K27 demethylase inhibitor GSK-J4 (45) increased H3K27me3 at the JAK2 and STAT3 promoters, with a concomitant decrease in RNA expression of JAK2 and STAT3 (Fig. 5D). Because GSK-J4 inhibits both known H3K27 demethylases, JMJD3 and UTX (45), knockdown of each was performed. In JMJD3 knockdowns, JAK2 expression was decreased 8 h after activation, whereas JAK2 expression in UTX knockdowns was not significantly different from scrambled controls. In contrast, STAT3 expression was significantly decreased 8 h after activation in JMJD3 and UTX knockdowns (Fig. 5E, Supplemental Fig. 2I, 2J). Protein expression of JAK2 and STAT3 was reduced in GSK-J4–treated cells compared with vehicle controls (Fig. 5F, 5G).

Phosphorylation of both JAK2 targets, STAT3 and STAT4, was impaired in GSK-J4–treated cells (Fig. 6A). Knockdown of JAK2 confirmed that impaired phosphorylation of STAT3 and STAT4 results from the loss of JAK2 expression (Fig. 6B, Supplemental Fig. 2K). Interestingly, addition of IL-6, a known proinflammatory cytokine whose receptor signals through JAK1 (46), resulted in a rescue of STAT3 phosphorylation in GSK-J4–treated cells (Fig. 6C). RNA-seq of JMJD3 knockdowns 1 d after activation revealed that IL12RB2 and JMJD3 (KDM6B) were the top three and four most differentially expressed and downregulated genes (FDR < 10%) compared with scrambled siRNA controls (Fig. 6D, Supplemental Table II). Additionally, GSK-J4–treated cells showed a decrease in IL12RB2 surface expression (Fig. 6E). Because p-STAT4 drives transcription of IL12RB2, we hypothesized that decreased JAK2 expression was responsible for this effect. Indeed, JAK2 knockdown in naive CD4 T cells resulted in a comparable decrease in IL12RB2 mRNA to JMJD3 knockdown (Fig. 6F, Supplemental Fig. 2L). We then performed ChIP-seq for JMJD3 and UTX. Analysis of the IL12RB2 gene revealed peaks of JMJD3, but not UTX, binding at 8 h postactivation (Fig. 6G). Of three JMJD3 binding sites, all located in the introns of IL12RB2, two sites matched perfectly with the loss of H3K27me3 peaks at 24 h.

Although no JMJD3 or UTX peaks were found in the JAK2 promoter, two JMJD3 peaks surrounding the JAK2 locus were identified (Fig. 7A). One peak (chromosome [Chr] 9–7) was located 10 kb upstream of the JAK2 TSS and was associated with a reduction in H3K27me3 at that site after activation (Fig. 7B, 7D), whereas the other (Chr 9–10) was located 1.66 Mbp downstream of the JAK2 TSS and did not exhibit a reduction in H3K27me3 (Fig. 7C, 7E). The two JMJD3 peaks overlapped with H3K4me1 peaks and DNase hypersensitivity sites found in ENCODE for human CD4⁺ T cells. Additionally, no H3K27ac peaks were observed at these sites, consistent with previously described silencing enhancer sites (47, 48). Hi-C revealed that the distal peak interacts with the JAK2 promoter (Fig. 7A), suggesting a mechanism for the significantly decreased H3K27me3 surrounding the TSS and further supporting JMJD3’s involvement in the regulation of this gene. No JMJD3 or UTX peak was identified surrounding the STAT3 locus, and Hi-C did not reveal any distal JMJD3 sites interacting with the promoter, calling into question how the observed loss of H3K27me3 around the STAT3 promoter is occurring.

Taken together, these data suggest that H3K27 demethylation during naive T cell activation is critical for regulating at least two well-established pathways important for T cell function and differentiation (44, 49). However, 34 other significantly differentially expressed genes were seen in the JMJD3 knockdowns (Supplemental Table II), the majority of which are also established in T cell activation and differentiation pathways as worthy candidates for future studies to build on our current work.

Based on our results, loss of H3K27me3 during CD4 T cell activation participates in the regulation of several genes that are key to CD4 T cell function, including JAK2, STAT3, and IL12RB2. Conversely, the addition of H3K27me3 during naive CD4 T cell activation targets pluripotency-related genes, and these marks are perpetuated in memory cells. Our results demonstrate that promoter H3K27me3 fits the paradigm of a repressive mark in CD4 T cells, associating with low-expression genes throughout activation in naive and memory CD4 T cells. The mechanisms behind H3K27me3-mediated silencing are not well characterized, and little is known about how these mechanisms are regulated during immune activation of CD4 naive versus memory cells. PRC2 is responsible for H3K27 methylation, and conventional thinking is that silencing mediated by H3K27me3 is due to binding of the modified histone by PRC1 (50, 51). H3K27me3 has been proposed to be a marker deposited in association with gene silencing and not a direct cause of it (51, 52). In fact, contrary to the paradigm that H3K27me3 is a repressive mark, promoter H3K27me3 in ES cells has been shown to associate with actively transcribed promoters (32). In contrast, our data, as well as previous data from CD4 T cells (6), suggest a strong correlation between the presence of promoter H3K27me3 and gene silencing, with the vast majority of genes containing H3K27me3 in their promoters demonstrating a low level of expression.

Loss of H3K27me3 enrichment was a prominent finding early after T cell activation (i.e., day 1) in both subsets. However, naive CD4 T cells demonstrated a simultaneous gain of H3K27me3 enrichment in 1492 gene promoters and that gain was also associated with low gene expression at rest. This low expression state remains consistent for these same genes at all time points studied after activation, and many maintained additional H3K27me3 peaks in resting memory cells. Thus, for this subset of genes, increasing H3K27me3 at day 1 after CD4 T cell activation appears to reinforce low gene expression. Based on current literature, we speculate that the mechanism is binding of PRC to the histones with this modification. Mapping of this subset of 1492 genes revealed connections to WNT signaling and pluripotency pathways. Importantly, in contrast with naive CD4, relatively few genes show increased promoter enrichment of H3K27me3 in activated memory cells, suggesting that one feature of naive T cell activation is a need to maintain a subset of genes in a repressed state. In contrast, memory cells demonstrate far more H3K27me3 promoter peaks at rest, many of which are concomitant with increased promoter enrichment and H3K27me3 peak gain in naive cells after activation. These genes also map to the same WNT-related pluripotency pathways. Therefore, we conclude that, during CD4 memory cell commitment, this subset of critical genes has already been marked for repression. WNT signaling was previously shown to maintain a pool of undifferentiated cells for further selection, differentiation, and effector function (42), suggesting that its repression during activation might promote T cell differentiation.

It is clear that many changes to promoter enrichment for H3K27me3 after activation correlate with basal RNA expression at rest for naive and memory cells (Fig. 3A–C, Supplemental Fig. 2F, 2G). Thus, genes showing an activation-induced increase in promoter H3K27me3 enrichment demonstrate low gene expression prior to activation, whereas genes showing a decrease in promoter H3K27me3 enrichment have higher levels of gene expression at rest. Loss of overall H3K27me3 enrichment is a profound change after activation in naive and memory CD4 T cells and appears to be mostly transient. Our data show that this loss is associated with consistently higher gene expression for affected genes at all time points studied (Fig. 3D, 3E). Mechanistically, this might occur via the recognition by H3K27 demethylases of areas featuring partially open chromatin during the resting state and subsequent removal of H3K27 methylation early in activation to open these regions completely and reinforce baseline gene expression. Not surprisingly, genes affected by this epigenetic regulatory mechanism map to pathways highly important for T cell activation, such as TCR and IL signaling (Table III). Recent work has demonstrated that the H3K27me3 demethylase, JMJD3, is necessary for Th1 polarization during CD4 T cell differentiation, and Jmjd3 deficiency in mice resulted in a loss of plasticity in CD4 T cells, supporting our results that H3K27me3 demethylation is integral for CD4 T cell function (15). The other H3K27 demethylase, UTX, has not been evaluated in this context, but our data with UTX knockdown do not suggest it is critical for early activation events in CD4 cells. Indeed, JMJD3 is upregulated in response to numerous stimuli, whereas UTX is described as more of a maintenance demethylase with ubiquitous expression (53, 54). How regulation and specific gene targeting by either enzyme are accomplished to support CD4 T cell agendas are clearly going to be important to study.

The JAK/STAT pathways are heavily represented in our pathway mapping of H3K27 demethylation (Fig. 5A). Our explorations of the role of H3K27 demethylation in JAK2 and STAT3 gene expression demonstrate that this mechanism is important for maintaining their high expression during activation. Thus, inhibition of H3K27 demethylation during activation results in increased promoter H3K27me3 and decreased expression of these genes (Fig. 5C–G). In our studies, JMJD3 appears to be solely responsible for the demethylation of the JAK2 promoter, whereas H3K27 demethylation of the STAT3 promoter is affected by siRNA-driven loss of JMJD3 or UTX (Fig. 5D). Interestingly, previous reports demonstrated that Stat3, in addition to Stat1, is responsible for transcription of Jmjd3 in rat microglial cells (55), suggesting that there might be a feedback loop between JMJD3 and STAT3. However, using our data, we were never able to prove a direct interaction of JMJD3 with the STAT3 promoter or with any distal elements interacting with the STAT3 promoter, rendering us unable to support this conclusion in the context of CD4 T cells. Knockout of Jak2 in mice results in embryonic lethality from severe defects in hematopoiesis (56, 57). Conditional knockouts also suffer defects in hematopoiesis, with the erythroid and thrombocytic stem cell compartments being primarily affected (58–60). Naturally occurring and activating somatic mutations of JAK2 in humans have also been documented to result in myeloproliferative syndromes (56–58).

The knowledge surrounding the pairing of JAKs with specific receptors has been complicated by the fact that they frequently can substitute for each other (61). Nonetheless, these same studies demonstrate that JAK2 is a necessary molecule for cellular function. Indeed, in our studies, loss of JAK2 resulted in reduced phosphorylation of STAT3 and STAT4 (Fig. 6B), as well as a subsequent reduction in IL12RB2 RNA expression (Fig. 6F), all of which are integral for Th1 differentiation (44). Thus, loss of JMJD3 resulting in a reduction in JAK2 is likely to be responsible, at least in part, for the skewing away from Th1 differentiation observed in Jmjd3-knockout mice (15). Additionally, we used ChIP-seq to demonstrate the direct binding of JMJD3 to the IL12RB2 gene outside of the promoter (but within the 2.5-kb radius of the TSS), which corresponds to the loss of its H3K27me3 peaks downstream of the promoter and the TSS (Fig. 6G) 1 d after activation. Strikingly, our data demonstrate that the loss of these peaks is perpetuated long term, because resting memory cells do not contain them. It has previously been shown that JMJD3 is a requirement for RNA polymerase II to progress through gene bodies enriched with H3K27me3 (62). We reason that the site-specific loss of the repression-associated H3K27me3 peak also contributes directly to the dramatic decrease in Il12rb2 gene expression seen in Jmjd3-knockout mice (Fig. 6D).

H3K27me3 has been characterized as an important epigenetic mark for CD4 T cell differentiation. In contrast to the Th2 polarization that occurs in the absence of JMJD3, CD4 T cells lacking EZH2, the central component of PRC2, will overproduce IFN-γ and skew toward a Th1 phenotype (63). Our data did not demonstrate any clear differential enrichment for H3K27me3 at the IFN-γ promoter or its interacting distal elements; however, we cultured our cells under nonpolarizing conditions that result in a mixture of Th subtypes after activation. Therefore, Th subset–specific peaks will not necessarily be prominent in our data. Considering the strong impact of H3K27me3-mediated mechanisms upon the expression of IL12RB2, which regulates IFN-γ production upstream, we propose that the effects upon IFN-γ are indirect in human CD4 T cells and are a direct result of defective JAK2 signaling and decreased IL12RB2 expression. Indeed, we were able to show clearly that loss of JAK2 in GSK-J4–treated cells (a small-molecule inhibitor of JMJD3 and UTX) resulted in reduced phosphorylation of STAT3 and STAT4. Additionally, this was not an off-target effect of the drug, because STAT3 phosphorylation could be rescued by the addition of IL-6 (whose receptor signals through JAK1), and knockdown of JMJD3 demonstrated the same effect.

One limitation of the current study is that we restricted most of our analysis to regions within 2.5 kb of the TSS. As described in our results, this distance was calculated based upon the average distance of H3K27me3 peak summits from the TSS, the majority of which fell within a 2.5-kb radius (Supplemental Fig. 1A). This result suggested that H3K27me3 peaks generally fell within 2.5 kb of the TSS. As previously reported, H3K27me3 peaks primarily affect gene promoters containing CpG islands (CGIs) (64). In our data, we found that ∼85% of the gene promoters with called H3K27me3 peaks near the TSS contained CGIs, which supports the conclusion that the skewing of H3K27me3 peaks toward the TSS is being driven by CGI promoters. Additionally, it became clear during our mechanistic studies evaluating the role of JMJD3 in JAK2 expression that the loss of H3K27me3 in regions outside of the specified radius was potentially an important mechanism, despite the fact that the majority of H3K27me3 peaks were not called in these genomic territories. Therefore, we are currently performing further analyses to assess the role of JMJD3 and H3K27me3 loss in these regions.

Another limitation of the current study is the potential for some of the effects observed on gene expression to be perpetuated from altered expression of upstream regulators. Technology does not currently allow manipulation of H3K27 demethylase activity targeted to specific loci, which makes the study of direct H3K27me3 effects challenging. Because of the broad activity encompassed by GSK-J4, as well as siRNAs targeting JMJD3 and UTX, we attempted to examine the H3K27me3 status of the pathway members that we evaluated in detail to compensate for this difficulty. However, we acknowledge that upstream regulators that are currently not mapped to the JAK/STAT and IL12RB2 pathways could also be playing a role in the altered expression of some of these genes.

Ultimately, this study demonstrates that the states of H3K27me3 in promoters are highly dynamic during CD4 T cell activation and that these dynamics differ markedly between naive and memory cells. Both T cell subsets rely upon H3K27me3 demethylation by JMJD3 during activation to facilitate and maintain high expression of genes that are crucial to T cell function and differentiation. Conversely, naive cells are unique in increasing H3K27me3 enrichment in genes pertinent to differentiation-related pathways, and these immune activation–induced marks are maintained in cells that enter the memory effector pool. Further studies will explore how H3K27me3 dynamics in other genomic regions affect CD4 T cell function to broaden our understanding of this important epigenetic mark.

We thank Dwight Kono, Michael McHeyzer-Williams, and Peiqing Sun for input into this work.

The authors have no financial conflicts of interest.