All donors donating blood for this study provided written informed consent for their participation. Study procedures were approved by the Ethics Committee of the Charité - Universitätsmedizin Berlin.

Peripheral blood from healthy donors was used for enrichment of natural regulatory T cells (nTreg). Two consecutive CliniMACS runs, compromising a CD8⁺ cell depletion followed by a CD25⁺ cell selection, were performed according to the manufacturer‘s instructions (Miltenyi Biotec). Target cell fraction was cultured in complete nTreg-medium, containing X-Vivo 15 medium (Lonza) supplemented with fetal calf serum (FCS, Hyclone), recombinant human IL-2 (Miltenyi Biotec) and Rapamycin (Pfizer). Depending on cell numbers, expansion process of Tregs was performed in 96-well plates and 24-well plates, respectively, for 23 days, using complete nTreg-medium and repetitive stimulation with anti-CD3/CD28 MACSiBead particles (Treg Activation/Expansion Kit, Miltenyi Biotec). Cells were collected for downstream DNA methylation analysis at several time-points during culture.

Peripheral blood was obtained from healthy donors, and peripheral blood mononuclear cells (PBMCs) were isolated by gradient centrifugation with Biocoll (Biochrom). The CD4⁺ CD25^high CD127^– Treg cells were sorted by a flow cytometry-based sorting system, the MACSQuant Tyto Cell Sorter (Miltenyi Biotec). Staining was performed using fluorophore-conjugated human anti-CD4, anti-CD25 and anti-CD127 antibodies (all Miltenyi Biotec), achieving purities of over 90%. Treg cells were cultured in a 96-well round-bottom plates (Eppendorf) with complete nTreg-medium compromising X-Vivo 15 medium (Lonza) supplemented with FCS (Biochrom), Penicillin/Streptomycin (Biochrom), Rapamycin (Sigma-Aldrich) and recombinant human IL-2 (Novartis). Treg cells were repetitively stimulated with anti-CD3/CD28 MACSiBead particles (Treg Activation/Expansion Kit, Miltenyi Biotec) for a total of 8 times throughout culture for 21 days.

PBMCs were isolated from 50 mL of peripheral blood obtained from healthy donors, by Ficoll-Paque Plus (Thermo Fisher Scientific) density gradient centrifugation followed by erythrocyte lysis using Buffer EL (Qiagen). After CD4 enrichment using magnetic cell separation with the autoMACS instrument (Miltenyi Biotec) using CD4 MicroBeads (Miltenyi Biotec), cells were left overnight at 4°C in basal cell culture medium [RPMI 1640 Medium with GlutaMAX Supplement-10 (Thermo Fisher Scientific), 10% FCS (Corning), 50 nM 2-mercaptoethanol (Thermo Fisher Scientific), 1 mM pyruvate (Biochrom), and 25 mM HEPES (Gibco)]. The following day, CD4⁺ memory T cells (CD4⁺ CD25⁺ CD127⁺ CD45RO⁺) and Tregs (CD4⁺ CD25⁺ CD127^–) were FACS-sorted with a BD FACSAria II SORP (Becton Dickinson). Sorted CD4⁺ memory T cells and Tregs were initially cultured in 96-well round-bottom plates (Greiner Bio-One) with the aforementioned cell culture media supplemented with 500 IU/mL rhIL-2 (R&D Systems) and 10 nM Rapamycin (STEMCELL Technologies) (day 0). The following day, cells were stimulated with anti-CD3/CD28 MACSiBead particles (Treg Activation/Expansion Kit, Miltenyi Biotec) according to manufacturer’s guidelines. Cells were stimulated at the same time points as the 1st and 2nd generation products, and collected for downstream DNA methylation analysis.

GMP-compliant Treg products and PBMCs were stained using fluorescently conjugated monoclonal antibodies for CD3 (BV650, clone OKT3), CD4 (PerCP-Cy5.5, clone SK3), CD8 (BV510, clone RPA-T8), CD127 (APC-A700, clone A019D5), and CD25 (APC, clone M-A251) at 4°C for 30 min. Backbone-stained samples were subsequently split to be stained for further 30 min with PE-conjugated antibodies for different human surface antigens: CD5 (clone UCHT2), CD80 (clone 2D10), CD86 (clone IT2.2), CD59 [clone p282 (H19)], CD160 (clone BY55), CD279 (clone EH12.2H7), CD366 (clone F38-2E2), and TIGIT (clone A15153G). All antibodies were purchased from BioLegend. To exclude dead cells, LIVE/DEAD Fixable Blue Dead Cell Stain dye (Thermo Fisher Scientific) was added. Lymphocytes were gated on the basis of the forward scatter (FSC) vs. side scatter (SSC) profile after exclusion of doublets via FSC-Height vs. FSC-Area. Tregs were defined as CD25^high CD127^low CD4⁺ CD8^– CD3⁺ and conventional T cells (Tcon) as CD25^– CD127⁺CD4⁺ CD8^– CD3⁺. Cells were analyzed on a Cytoflex LX (Beckmann Coulter) flow cytometer. Cytometric raw data were analyzed using Kaluza^TM software (Beckman Coulter), mean expression values were visualized using the nautilus R package (available from¹).

Genomic DNA of cell pellets was isolated using AllPrep DNA/RNA Mini Kit (Qiagen) or Zymo’s Quick-DNA MicroPREP Kit (Zymo Research), following manufacturer’s instructions. DNA concentration was assessed with Qubit dsDNA HS Assay Kit and the Qubit Fluorometer (Molecular Probes/Life Technologies). 200–250 ng of genomic DNA was used as input for bisulfite conversion, which was performed with Zymo’s EZ DNA Methylation-Gold Kit (Zymo Research) or Zymo’s EZ DNA Methylation Kit (Zymo Research). Methylation analysis was performed with the Infinium MethylationEPIC Kit (Illumina EPIC-8 BeadChip) following the manufacturer’s instructions. Bead Chips were imaged on Illumina’s Microarray Scanner iScan.

All analyses were conducted in the R statistical software (Version > 4.0).

Treg products from three healthy donors were generated twice by two indepenent manufacturing runs. These first generation (1st gen.) Treg products are generated from peripheral blood-derived starting material which is depleted of CD8⁺ cells and enriched for CD25-expressing cells using the CliniMACS system, yielding a starting populations enriched for CD8^– CD4⁺ CD25⁺ Tregs. For generation of the final product, the cells undergo expansion culture for 23 days with repetitive TCR stimulation (using antiCD3/CD28 microbeads, 8 stimulations in total) in the presence of IL2 and Rapamycin. Supplementation of Rapamycin is meant to inhibit the outgrowth of contaminating conventional proinflammatory CD4⁺ T cells (Battaglia et al., 2005). We harvested samples at the starting (day 0), final (day 23), and relevant additional time points during cell expansion (Figure 1A). The samples were subjected to genome-wide methylation analysis using the Illumina Methylation EPIC microarray, which covers a selection of 850,000 CpG methylation sites genome-wide with a special emphasis on regulatory elements such as promoters and enhancers.

Principal component analysis (PCA) of the EPIC array methylation profiles demonstrated that the number of days in culture had a substantial impact on the DNA methylome, as this feature accounted for PC2 (11% of the total variation in the data set) (Figure 1B). Inter-donor variances could also be observed and were reflected on PC3 (Figure 1B). Importantly, within the same donor, the samples did not separate based on production runs. These observations indicate that there is a progressive change in DNA methylation with prolonged culture, which is reproducible between donors and production runs. We also explored methylation changes at partially methylated domains (PMDs) of the genome, which are large transcriptionally silenced regions characterized by a highly disorganized methylation pattern, a generally reduced DNA methylation level and poor gene content that undergo methylation loss as a result of strong episodes of proliferation (Salhab et al., 2018; Decato et al., 2020). We indeed observed a loss of PMD methylation with increased expansion rate (Figure 1C), which is consistent with previous findings that demonstrated that accumulation of cell divisions drive PMD hypomethylation in cancer and aging (Zhou et al., 2018), and during T cell differentiation (Durek et al., 2016).

We next performed differential methylation analysis to identify genetic loci undergoing significant methylation changes during Treg product generation. Increased DNA methylation in promoters, first introns, and enhancers generally correlate with transcriptional downregulation (Varley et al., 2013; Anastasiadi et al., 2018), while hypomethylation correlates with transcriptional activation. In contrast, hypermethylation at gene bodies (mainly on exons) has been reported to correlate with increased transcription (Ball et al., 2009) and increased methylation at alternatively spliced exons has been implicated in differential splicing by mediating exon retention during transcription (Sun et al., 2020).

Although we focus much of our later analyses on differentially methylated regions (DMRs) containing at least 3 CpGs, we also characterized differentially methylated positions (DMPs) of individual CpGs in case biologically important loci were not sufficiently covered by multiple adjacent CpG probes due to microarray design limitations. We detected approximately 40,700 and 30,000 CpGs that were differentially methylated in production Runs 1 and 2, respectively, between the final (day 23) and starting (day 0) products (Figure 2A, Supplementary Table 1). For both runs, majority of DMPs mapped to introns (38%), followed by distal intergenic regions (29%) and promoters (27%) (Figure 2B). Other DMPs mapping to the 5′ or 3′ UTRs, exons, and genomic regions immediately downstream of the gene body (<300 bp) comprised only 5% of all significant sites. Furthermore, promoters and introns consistently had a higher number of hypermethylated DMPs, whereas hypomethylated DMPs were more frequent in intergenic regions. Interestingly, a few CpGs were consistently among the top 5 most significant DMPs (stratified by genic annotation) in both production runs, such as a hypermethylated DMP mapping to the promoter of PDCD4-AS1, a long non-coding RNA that has been shown to positively regulate PDCD4, a tumor suppressor gene that has also been implicated in autoimmune disorders (Hilliard et al., 2006; Jadaliha et al., 2018; Figure 2C).

We then performed DMR analysis to identify genomic regions containing adjacent differentially methylated DMPs, and identified 1,753 DMRs associated with 1,515 unique genes in Run 1, and 1,246 DMRs associated with 1,103 genes in Run 2; the remaining 10,852 genes from Run 1 and 9,597 genes from Run 2 did not contain a DMR despite having at least 1 significantly DMP. However, these non-DMR DMPs may still function as important regulators of Treg or T cell biological processes: When we performed gene set enrichment analysis of Run 1 promoter DMP genes that were not represented by any DMRs, we found significant enrichment for genes important for lymphocyte activation (Figure 2D).

We next focused on significant DMRs that were present in both production runs as they would reflect reproducible methylation changes during Treg product generation. We defined “shared DMRs” as production run 1 DMRs overlapping with run 2 DMRs by 50% or more. Furthermore, we were also interested in delineating early epigenetic changes in response to initial activation versus changes occurring only later, the latter resulting from progressive changes due to prolonged culture. We therefore labeled shared DMRs as “Early” (first identified in day 10 vs. day 0 comparison and still significant in day 23 versus day 0) or “Late” (first identified in day 23 vs. day 0 in both production runs). As expected, Early DMRs displayed stronger initial methylation change within the first 10 days of culture (indicated by the bend in the methylation time series traces) (Figure 3A).

In total, there were 838 shared DMRs (Supplementary Table 2), with 62% hypermethylated by the final time point (Figure 3B). Although the directionality (hyper- vs. hypomethylated) of the DMRs was consistent between the two production runs, the methylation kinetics (“Early” or “Late”) differed slightly (Figure 3B): 16% of shared DMRs had different kinetics between the two runs. This may be a result of variability in expansion rates between the two runs (Figure 1D), differences in the proportion of contaminating cell types in the starting day 0 cultures, or simply lack of statistical power. Gene promoters accounted for the majority of all shared DMRs (63%), followed by introns and distal intergenic elements, both of which frequently contain enhancers.

Unsupervised hierarchical clustering based on shared DMR mean methylation values at days 0, 10, and 23 depicted samples of the same time points clustering together regardless of their corresponding production run (Figure 3C). Genes implicated in T cell activation and differentiation were present in both hypo- and hypermethylated DMRs. We performed Gene Ontology (GO) enrichment analysis on the shared DMRs that were stratified by directionality and methylation kinetics to evaluate biological processes that may be regulated by epigenetic mechanisms. Genes that were significantly hypomethylated early during manufacturing are implicated in T cell activation, differentiation, adhesion, and other related pathways (Figure 3D). Enrichment of these GO terms was maintained even in the late hypomethylated DMRs. Examples of genes undergoing early strong promoter hypomethylation (>15%) include genes important for cell signaling, intracellular signal transduction, and proliferation such as GPR87 and MAPKAPK3 (Figure 3E; Zhang et al., 2010; Cargnello and Roux, 2011). Genes containing early hypermethylated DMRs did not enrich for any GO term, but still contained interesting candidate genes. We observed strong hypermethylation at the promoter FOXP1 (Figure 3E), perhaps indicating a path toward terminal differentiation of the Treg products as DNA methylation-controlled transcriptional downregulation of FOXP1 has been shown to correlate with end-differentiated T cells (Goronzy and Weyand, 2012; Durek et al., 2016).

We also observed early DNA hypomethylation of genes reported to negatively modulate TCR-mediated signaling, such as PDCD1, CD59 (Sivasankar et al., 2009), and IL1RN (McIntyre et al., 1991; Arend et al., 1998; Figure 3F). This may indicate the beginning of a negative feedback loop that is normally part of a physiological response to restrict T cell effector function in order to prevent overshooting of the immune response during antigen encounter. Furthermore, many DMRs mapping to genes important for T cell activation were significantly hypermethylated later in culture (Figure 3D). For example, a progressive increase in promoter methylation was detected at genes which positively regulate TCR signaling and T cell differentiation, such as RHOH (Wang et al., 2011), HLX (Allen et al., 1995), and RARA (Hall et al., 2011), suggesting transcriptional inactivation that may occur through repetitive stimulation-induced epigenetic reprogramming (Figure 3G). To demonstrate that hypermethylation of these TCR signaling loci occurs independently of starting material purity, we also profiled these promoters’ methylation status in FACS sorted primary conventional CD4⁺ memory T cells (CD4mem) and Tregs which have also been stimulated repeatedly with antiCD3/CD28 beads in similar culture conditions. Indeed, both Tregs and CD4mem at day 23 displayed significantly increased promoter methylation compared to the starting populations (day 0) (Figure 3H), indicating that frequent TCR stimulation for 23 days is capable of epigenetically reprogramming and potentially stably downregulating these critical genes.

We also identified promoter-associated Late DMRs that were significantly and strongly hypomethylated (methylation loss > 15%) at HAVCR2 (encoding for TIM-3), TIGIT, and CD160 (Figure 4A), which are co-inhibitory surface molecules that are upregulated in T cells and Tregs during prolonged antigen encounter. Although upregulation of these co-inhibitory surface molecules is often associated with exhaustion, a hypo-responsive and dysfunctional state in conventional T cells (Tcon) caused by persistent TCR signaling (reviewed in Blank et al., 2019), they are also linked to enhanced immunosuppressive capacity in Tregs (Sakuishi et al., 2013; Fuhrman et al., 2015; Liu et al., 2018).

We were interested in profiling methylation changes at additional genes which have been reported to be transcriptionally upregulated during prolonged antigen exposure. We curated a list of genes by combining T cell dysfunction/exhaustion gene modules from Singer et al. (2016) with genes identified in CD4⁺ exhausted T cells from Crawford et al. (2014), and checked whether these genes were present in our shared DMR dataset. Indeed, we found several genes known to be upregulated in exhausted T cells which were consistently differentially methylated during Treg product generation, with most of them experiencing strong hypomethylation at their promoters (Figure 4B). We confirmed that strong promoter hypomethylation corresponded to upregulation of these chronic activation-induced genes (i.e., HAVCR2/TIM-3, TIGIT, CD160, CD59, CD80, CD86, and CD5), as the final product had higher surface protein expression than either of the ex vivo Treg and Tcon populations present in the day 0 starting material (Supplementary Figure 1).

Epigenetic changes have also been correlated to the establishment of functional exhaustion, most well studied in the context of tumors and chronic infections (Khan et al., 2019; Scott et al., 2019). In a recent study conducted by Abdelsamed et al. (2020), the authors generated whole genome bisulfite sequencing (WGBS) data from CD8⁺ effector memory T cells (TEM) derived from healthy donors and patient-derived exhausted PD-1⁺ HIV-reactive CD8⁺ T cells (Texh). We took the opportunity to reanalyze this same WGBS dataset to define DNA methylation features distinguishing chronically activated CD8⁺ T cells (Texh) from TEM, and determined whether these epigenetic differences were also shared with our first generation products which have undergone repetitive TCR stimulation. Our analysis identified approximately 52,000 genomic regions that were differentially methylated between TEM and exhausted T cells, which we refer to as exhDMRs.

We rationalized that the consistently Late shared DMRs identified in our Treg products would likely contain exhaustion-associated changes because these late changes were likely induced by long-term TCR stimulation. We found that approximately 20% of all Late shared DMRs overlapped with the exhDMRs (116 overlapping DMRs of 585 Late shared DMRs, Supplementary Table 3), which is likely an underestimation due to the limited number of CpGs captured by the EPIC microarray. Of the overlapping DMRs, 91 were significantly hypermethylated in exhausted T cells compared to TEM, of which 74% were consistently hypermethylated in the final Treg products (day 23) compared to day 0 (Figure 4C). A vast majority (85%) of these hypermethylated DMRs also mapped to promoters of genes (highlighted red in Figure 4C), suggesting that a set of hypermethylated exhaustion-associated genes are similarly transcriptionally silenced by epigenetic mechanisms during in vitro manufacturing of Treg products.

Preservation of Treg immunosuppressive function and identity is a primary concern for the development of adoptive Treg therapies, as persistent TCR-mediated activation has been shown to correlate with downregulation of the lineage-defining transcription factor FOXP3, although perhaps only selectively in memory (CD45RA^–) Tregs (Hoffmann et al., 2009; Arroyo Hornero et al., 2017; Guo et al., 2019). The downregulation of FOXP3 is linked to the hypermethylation of an intronic enhancer in the FOXP3 gene, the so-called FOXP3 Treg-Specific Demethylated Region (FOXP3-TSDR), which is used as an epigenetic marker to distinguish de facto Tregs from closely related conventional CD4⁺ memory T cells (CD4mem) due to its highly selective demethylated status only in Tregs (Huehn et al., 2009). The consequential gain of FOXP3-TSDR methylation in response to persistent TCR stimulation demonstrates the need to profile additional culture-induced epigenetic changes at other Treg-specific loci, as these changes may be predictive of impaired Treg immunosuppressive function.

Recent WGBS analysis comparing primary conventional CD4⁺ T cell and Treg subsets identified an additional 33 Treg-selective demethylated regions (Treg-DRs; Ohkura et al., 2020). Unsurprisingly, several of these Treg-DRs map to transcription factors important for Treg stability, mediators of IL-2 signaling, and immunosuppressive surface molecules, among other genes. We asked whether these Treg-DRs gained methylation in response to repetitive CD3/CD28 stimulation during the manufacturing of the Treg products, which may indicate TCR activation-induced destabilization of the Treg identity. As additional positive and negative controls, we included EPIC array data from FACS-sorted and expanded primary Tregs and CD4mem T cells (n = 3). In addition, we had the chance to analyze the final time point of second-generation Treg products, which were generated through GMP-compliant FACS and expanded with the same protocol as the first generation products (n = 4; see Methods).

We first selected all CpGs from our microarray data which overlapped with the 33 Treg-DRs identified from Ohkura et al. (Supplementary Table 4). As validation, we compared DNA methylation between primary Tregs and CD4mem at day 0, and confirmed that nearly all selected CpGs were significantly hypomethylated in Tregs (Figure 5A), with CpGs mapping to intronic parts of the FOXP3 gene body displaying the greatest absolute difference. Additionally, these Treg-DR CpGs map to strong, active regulatory regions (e.g., active transcription start sites, genic enhancers that correlate most strongly with gene expression, and additional active enhancers) identified in Tregs (Figure 5B; Roadmap Epigenomics et al., 2015). Strikingly, at the identical genomic location in CD4mem cells, most of these regions were identified to be transcriptionally repressed (e.g., weak enhancer, weak transcription, repressed polycomb region, and quiescent region), further emphasizing that these Treg-DRs are important regulatory features distinguishing Treg identity from that of a CD4mem. We observed that the methylation values of the starting material (day 0) of the first generation products were consistently in between sorted primary CD4mem and Tregs (Figure 5C), confirming that MACS-mediated CD8 depletion followed by CD25 enrichment is not sufficient to obtain a pure Treg starting population.

We were next interested in assessing culture-induced changes in Treg identity and performed a PCA with these selected CpGs to determine how all time points of expanded primary CD4mem T cells, Treg, and first- or second-generation products clustered relative to each other. We found strong separation between CD4mem and Tregs along PC1 (76% of total variation) at any time point, with final first generation products (indicated by highest cumulative expansion rates) clustering more closely with CD4mem cells than final second-generation products (Figure 5D). Indeed, unsupervised hierarchical clustering confirmed that highly expanded final first generation products clustered with the CD4mem cells (Figure 5E), with the selected CpGs showing hypermethylated values relative to primary Tregs. However, primary Tregs also exhibited significantly lower expansion rates, which may partially explain the discrepancy in methylation status of these CpGs.

While the observed similarities between the final first generation Treg products and expanded CD4mem T cells might partially stem from the relative impurity of the MACS-enriched starting material, we found indications that the culture process alone can destabilize the epigenetic identity in primary FACS-sorted Tregs: Although expanded primary Tregs had lower Treg-DR methylation relative to CD4mem T cells at day 23, we were keenly aware that a subset of these selected methylation sites gained methylation with days in culture, approaching similar methylation levels as CD4mem T cells. For example, in both the first generation Treg products and primary Tregs, the selected CpGs within the IKZF2 (HELIOS) locus showed progressive gain in methylation throughout culture (Figure 5F).

We therefore assessed whether in vitro expansion resulted in an epigenetic switch of Treg-DRs in the primary FACS-sorted Tregs. We performed a linear regression on all 31 Treg-DR CpGs that were significantly demethylated in Tregs relative to CD4mem at day 0, and used log10 expansion rates as the dependent variable instead of days in culture to adjust for possible donor-specific differences in day-to-day proliferation behavior. Strikingly, several CpGs mapping to FOXP3 and TNFRSF1B (TNFR2) showed very little change in methylation with increased expansion rate (Figure 5G) indicating a stable demethylated behavior even under strong proliferative conditions. Many of the remaining CpGs, however, displayed significantly higher R² values (> 0.4) which indicates strong correlation between hypermethylation of indicated Treg-DR CpGs and cumulative expansion rate of the population. A similar correlation for many of the same sites could not be observed in the CD4mem controls. This expansion-induced increase in methylation on genes essential for Treg function (e.g., IKZF2, IKZF4, IL2RA), are likely to impact the functional behavior of the Tregs and are of general concern for any Treg product, irrespective of the purity of the starting material.