IL-10 Produced by Induced Regulatory T Cells (iTregs) Controls Colitis and Pathogenic Ex-iTregs during Immunotherapy

CD4⁺ CD25⁺ Foxp3⁺ regulatory T cells (Tregs) are essential to the balance between pro-inflammatory and anti-inflammatory responses at mucosal surfaces (1). There are two subsets of Tregs: natural regulatory T cells (nTregs), which develop in the thymus, and induced regulatory T cells (iTregs), which arise from conventional T (Tconv) cells in the periphery (2–6). Immunologic tolerance requires both Treg subsets, which act synergistically (6, 7). Additionally, iTregs can be generated in vitro by T cell activation in the presence of TGF-β1 and IL-2, which makes them an attractive alternative for the treatment of human autoimmune disorders unresponsive to current approaches (8–11). In vitro–derived iTregs also suppress inflammation in animal models of inflammatory bowel disease, diabetes, and autoimmune gastritis (12–14). Importantly, in vitro–derived iTregs contribute to tolerance in disease models where in vivo–derived iTregs are not present (6, 7). The pathways used by iTregs to support disease resolution in these models remain unknown, and it is unclear to what extent the method of iTreg derivation influences the acquisition of the full complement of Treg suppressive mechanisms.

Although much work has been done to uncover the molecular mechanisms of Treg suppressive activity, the “division of labor” between nTregs and iTregs remains largely unresolved (2). Molecules such as CTLA-4, granzyme B, IL-10, and TGF-β have been proposed as mechanisms of nTreg-mediated suppression (15–18). Of these mechanisms, the overall importance of IL-10 to immune homeostasis is exemplified by the chronic colitis that develops with complete or APC-specific IL-10 deficiency (19, 20). IL-10 is particularly important for Tregs at environmental interfaces, as a Treg-specific inactivation of IL-10 results in spontaneous colitis (16). In the CD45RB^hi transfer model of colitis, IL-10 present in the CD45RB^low population, which contains Tregs, was found to be necessary for disease prevention and treatment (21, 22). Furthermore, Treg-derived IL-10 was shown to control Th17 and Th1+Th17 cells (23). To date, studies have been limited to the role of IL-10 in nTreg function, leaving unresolved the role of IL-10 as an iTreg suppressive mechanism.

In this study, we used cells from mice that harbor a cassette encoding a mutant Foxp3-enhanced green fluorescent fusion protein within the Foxp3 locus (Foxp3^ΔEGFP) to create colitis in an environment free of iTregs (6, 24). After immunotherapy with nTregs and iTregs that were selectively deficient in IL-10, we investigated how the source of IL-10 impacted disease progression. We also examined the clonal and transcriptional relationships between those iTregs that maintained Foxp3 expression or lost Foxp3 expression (ex-iTregs). Our results identify in vivo selection of a potent iTreg pool that is clonally distinct from pathogenic ex-iTregs and IL-10–dependent reciprocal compensation between Treg subsets as critical for the maintenance of mucosal tolerance.

Foxp3^EGFP and Foxp3^ΔEGFP mice on the BALB/c background were generated and screened as previously described (24). Thy1.2⁺ Foxp3^ΔEGFP newborn mice were rescued by i.p. transfer of 60 × 10⁶ unfractionated Thy1.1⁺ BALB/c splenocytes to generate naive Thy1.2⁺ CD4⁺ CD45RB^hi T cells with the nonfunctional Foxp3^ΔEGFP allele. Rag1^−/− and IL10^−/− mice were obtained from The Jackson Laboratory. The Animal Resource Committee at the Medical College of Wisconsin approved all animal experiments.

Pooled splenocytes and lymph node cells (axillary, brachial, inguinal, and mesenteric) were stained with either anti-CD4–allophycocyanin (RM4-5; BD Biosciences) or anti-CD4–Pacific blue (RM4-5; Invitrogen), plus anti-CD90.1–PerCP (OX-7; BD Biosciences) and anti-CD45RB–allophycocyanin (C363.16A; eBioscience) as appropriate and sorted on the basis of Ab and enhanced GFP (EGFP) fluorescence. All sorting was done on a FACSAria (BD Biosciences). The average purity and viability of the sorted CD4⁺ populations was 98.96 ± 0.14 and 84.31 ± 0.68 (n = 120), respectively.

Colitis was induced in 6- to 8-wk-old Rag1^−/− BALB/c mice by i.p. injection of 4 × 10⁵ CD4⁺ CD90.1⁻ EGFP⁻ CD45RB^hi cells. Mice were weighed twice weekly. In some experiments, when mice lost 2.5% (±5.7%) of their initial body weight or began to exhibit symptoms of colitis (diarrhea, hunched posture), they were treated by i.p. injection of nTregs (Thy1.2⁺) plus iTregs (Thy1.1⁺) (5 × 10⁵ each) purified by cell sorting. The nTregs were isolated on the basis of EGFP expression from the spleen and lymph nodes of Foxp3^EGFP BALB/c mice or IL10^−/− Foxp3^EGFP BALB/c mice. The iTregs were generated in culture (as described below) and isolated on the basis of EGFP expression.

For serial adoptive transfer experiments, 1 × 10⁵ CD4⁺ Thy1.1⁺ EGFP^– ex-iTregs were isolated by sorting cells from the mesenteric lymph node (MLN) and spleen of mice that were successfully treated with Thy1.1⁺ wild-type (WT) iTregs plus Thy1.2⁺ WT nTregs and injected i.p. into Rag1^−/− hosts. In some experiments, the mice were treated with 0.5 × 10⁶ Thy1.2⁺ nTregs plus 0.5 × 10⁶ Thy1.2⁺ iTregs and were coinjected with 15,000 Thy1.1⁺ EGFP^– T cells (based on a 3% sort impurity, three times the calculated impurity). The Thy1.1⁺ EGFP^– T cells were isolated from iTregs that had lost Foxp3 expression in culture.

Sorted CD4⁺ EGFP⁻ cells from Foxp3^EGFP or IL10^−/− Foxp3^EGFP mice (1 × 10⁶/ml) were cultured with anti-CD3 mAb (clone 14-2C11 at 2.5 μg/ml) coated dishes in the presence of soluble anti-CD28 mAb (1 μg/ml; clone 37.51), TGF-β1 (5 ng/ml; R&D Systems), and 100 U/ml IL-2. After 72 h, cells were resorted based on EGFP fluorescence and used for adoptive transfer or maintained in culture with IL-2.

The entire colon and the distal 15 cm of the small intestine were used as the source of intraepithelial lymphocytes (IELs) and lamina propria lymphocytes (25). IELs were removed by gentle shaking of 0.5-cm intestinal sections for 30 min in buffer containing 5% (v/v) FCS, 1 mM DTT (Sigma-Aldrich), and 5 mM EDTA. IELs were washed and isolated on a discontinuous Percoll gradient (67%, 44%). Washed intestinal sections were digested with collagenase D (1 mg/ml; Roche) in the presence of DNase I (Invitrogen). A discontinuous Percoll gradient (67%, 44%) was used to isolate washed lamina propria lymphocytes.

Cells were collected from the spleen, MLN, colon and small intestine and stained as indicated. The anti-mouse Abs used were Pacific blue–conjugated anti-CD4 (RM4-5; Invitrogen), PerCP-conjugated anti-CD90.1 (OX-7; BD Biosciences), Alexa Fluor 700–conjugated anti-CD44 (IM7; BioLegend), PE-conjugated anti-CD62L (MEL-14; BD Biosciences), Alexa Fluor 647–conjugated anti-CD103 (2E7; BioLegend), PE–Texas red–conjugated anti-CD25 (PC61 5.3; Invitrogen), PE–Cy7–conjugated Klrg1 (2F1; eBioscience), allophycocyanin eFluor 780–conjugated anti-TCRβ (H57-597; eBioscience), and allophycocyanin-conjugated anti-GITR (DTA-1; eBioscience). A four-laser custom LSRII was used to collect the data, and FlowJo software was used for analysis.

Intracellular cytokine staining was performed after a 5-h restimulation with PMA (5 ng/ml; Sigma-Aldrich) and ionomycin (0.5 μM; Sigma-Aldrich) in the presence of brefeldin A (1 μl/ml; BD Biosciences). Surface staining of cells was performed using a modified FACS buffer containing 10 μg/ml brefeldin A. Cells were stained on ice for 30 min with the primary anti-mouse Abs PE–Cy7–conjugated anti-CD4 (RM4-5; BD Biosciences), PerCP-conjugated anti-CD90.1 (OX-7; BD Biosciences), and allophycocyanin eFluor 780–conjugated anti-TCRβ (H57-597; eBioscience) then washed with the modified FACS buffer and fixed in 1% paraformaldehyde overnight at 4°C. After this incubation, cells were washed with 1 ml PBS and then permeabilized with 1 ml 0.1% Triton-X. Intracellular staining was performed for 30 min at room temperature with allophycocyanin-conjugated anti–IFN-γ (XMG1.2; BD Biosciences), Alexa Fluor 700–conjugated anti–TNF-α (MP6-XT22; BD Biosciences), Pacific blue–conjugated anti–IL-17A (TC11-18H10.1; BioLegend) or with Pacific blue–conjugated anti-Helios (22F6; BioLegend), and allophycocyanin-conjugated anti–CTLA-4 (UC10-4B9; BioLegend). A four-laser custom LSRII was used to collect the data, and FlowJo software was used for analysis. Serum cytokines were measured using the eBioscience FlowCytomix kit following the manufacturer’s recommendations.

Complete colons were fixed in formalin, processed, and stained with H&E using a histology core facility. Blinded sections from the entire colon were examined by a pathologist (N.H.S.), and large intestine colitis scores were determined for inflammatory changes on a 4-point semiquantitative scale with 0 representing no change (26). The following features were considered: severity, depth and chronic nature of the inflammatory infiltrate, crypt abscess formation, granulomatous inflammation, epithelial cell hyperplasia, mucin depletion, ulceration, and crypt loss.

Spleen and MLN cells of mice that were successfully treated with 0.5 × 10⁶ Thy1.1⁺ WT iTregs plus 0.5 × 10⁶ Thy1.2⁺ WT nTregs were stained with Pacific blue–conjugated anti-CD4 (RM4-5; Invitrogen) and PerCP-conjugated anti-CD90.1 (OX-7; BD Biosciences) and were sorted on the basis of Ab and EGFP fluorescence. Ex-iTregs (CD4⁺ Thy1.1⁺ EGFP^–), iTregs (CD4⁺ Thy1.1⁺ EGFP⁺), and nTregs (CD4⁺ Thy1.1^– EGFP⁺) were isolated by flow cytometry sorting on a FACSAria. Total RNA was extracted with the RNeasy Micro Kit for <100,000 cells or the RNeasy Mini Kit for >100,000 cells (Qiagen) according to the manufacturer’s protocol. cDNA was synthesized with the Superscript III First Strand Synthesis System and oligonucleotide (dT) primers (Invitrogen) according to the manufacturer’s protocol. Isolated cDNA was used for quantitative PCR, spectratype analysis, or CDR3 sequencing.

Quantitative PCR was performed in a StepOnePlus PCR System (Applied Biosystems) using the TaqMan Fast Universal PCR Master Mix (Applied Biosystems) and predeveloped specific TaqMan primers for Il10 (Mm00439616_m1). GAPDH was used (Mm99999915_g1) as a housekeeping gene (Applied Biosystems). The standard curve for Il10 was developed by isolating bone marrow cells from C57BL/6J and culturing the cells at 4 × 10⁴ cells/ml in DMEM/F12-10 medium (DMEM/F12 [Life Technologies], 10% [v/v] FBS [Atlantic Biologics], 10 mM l-glutamine [Life Technologies], 100 IU penicillin/ml [Life Technologies], 100 μg/ml streptomycin [Life Technologies]) at 4 × 10⁶ cells/ml in the presence of 10 U/ml MCSF (14-8983-80; eBioscience). The cells were incubated at 37°C for 7 d (medium was changed on day 3). Macrophages were isolated using Cellstripper (Cellgro), washed, and resuspended in DMEM/F12-10 at 2 × 10⁵ cells/ml. The cells were plated for 8 h at 37°C in a 24-well plate containing 50 ng/ml LPS (Sigma), 1 μg/ml OVA (323–339 peptide), and 15 μl/ml anti-OVA (0220-1682; AbD Serotec). cDNA was isolated as described, and the Il10 gene from the exon 2/3 boundary (forward primer: 5′-AATGCAGGACTTTAAGGGTTACTTGGG-3′) to the exon 4/5 boundary (reverse primer: 5′-CTTGTAGACACCTTGGTCTTGGAG-3′) was cloned into the pCR4-TOPO vector (Invitrogen) and transformed into DH5α cells. The plasmid was isolated with a mini prep (Promega) and then purified by agarose gel extraction. The concentration was determined using log-dilutions and measured with a NanoDrop (Invitrogen). The standard curve was created with 10-fold dilutions based on copy number.

Total RNA for the iTreg, nTreg, and ex-iTreg sets was isolated with TRIzol (Invitrogen), according to the manufacturer’s protocol, from sorted cells pooled from 14 mice treated with 0.5 × 10⁶ WT nTregs plus 0.5 × 10⁶ WT iTregs. Labeled target was prepared and hybridized to Affymetrix 430 2.0 GeneChips in accordance with the manufacturer’s protocol. Two technical replicates were performed, and the results were averaged. Probe sets that revealed a 2-fold difference (|log₂ ratio| >1.0) relative to Tconv cells were identified and used in subsequent analyses. The data were normalized with the robust multi-array analysis algorithm derived by the Bioconductor group (http://www.bioconductor.org) (27). The mean fold change was calculated from two independent arrays for each cell type and was scored p < 0.05 with a false discovery rate <10% by the non-parametric rank product test (28). A false negative result is recovered for Foxp3⁺ Tregs because the gene array probes for Foxp3 lie distal to the poly-A initiation site in the Foxp3^EGFP allele. The microarray data are available in the Gene Expression Omnibus database (http://www.ncbi.nlm.nih.gov/gds) under the accession number GSE35543. Tconv cell data were taken from Gene Expression Omnibus microarray data set GSE6875.

The methylation status of the Treg-specific demethylation region (TSDR) of Foxp3 in donor male nTregs, iTregs, and ex-iTregs purified from mice treated with WT nTregs plus WT iTregs by cell sorting was assessed by bisulfite sequence analysis (29). Briefly, genomic DNA was treated by bisulfite, to convert unmethylated cytosines into uracil leaving methylated cytosines unchanged. Bisulfite conversion of DNA was performed using EZ DNA Methylation-Direct Kit (Zymo Research) according to the manufacturer’s instructions. The TSDR of converted DNA was amplified by methylation-specific primer sequences 5′-TATTTTTTTGGGTTTTGGGATATTA-3′ (forward) and 5′-AACCAACCAACTTCCTACACTATCTAT-3′ (reverse). The PCR product was purified and inserted into a TOPO TA cloning vector (Invitrogen). The ligation product was used to transform competent bacteria (10-β competent Escherichia coli, New England Biolabs), and clones were selected on kanamycin. Plasmid DNA was extracted by Qiagen miniprep kit, and clones that present a 300-bp fragment after EcoRI digestion were selected. Sequencing was done with M13R primer. Blast analysis was done by comparing the M13R sequence and converted Foxp3 gene sequence.

cDNA was amplified by PCR with a Cβ primer (5′-CTCAAACAAGGAGACCTTGGGTGG-3′) and a Vβ primer from one of 22 Vβs (30). An ABI 3100 Genetic Analyzer was used to analyze the length distribution of amplified cDNA products, and Xplorer v2.4.2 (http://www.dnatools.com/download.html) was used to create histograms.

cDNA was amplified by PCR with the Cβ primer and a Vβ8.2 primer (5′-GCTACCCCCTCTCAGACATCAGTG-3′). PCR products were purified using the QIAquick PCR purification kit (Qiagen) according to the manufacturer’s protocol and concentrated using ethanol precipitation. The DNA sample was purified using Agencourt AMpure beads. The purified samples were used to generate libraries for Next Generation Sequencing using the Ion Torrent Personal Genome Machine (Life Technologies, Carlsbad, CA) following the Ion Fragment Library Kit protocol. During preparation of the libraries, samples were size selected using the Sage Science Pippen Prep instrument and 2% agarose cassettes. Completed libraries were analyzed and quantified on the Agilent 2100 BioAnalyzer. The obtained libraries were processed for sequencing by dilution following the recommendations for the Ion Xpress Template Kit protocol that used an emulsion PCR, breaking, and enrichment of each sample. Sequencing was performed using the Ion Torrent 314 or 316 chip and respective reagents. Sequence analysis and base calling was performed using the built-in sequence software (v.1.9).

The purified PCR products were TA cloned into the pCR4-TOPO vector (Invitrogen) to create cDNA libraries. Individual colonies were subcloned, and plasmids containing inserts were grown for 16 h in Luria–Bertani medium, frozen at −80°C in 50% glycerol, and sent to Beckman Coulter for sequencing. The CDR3 regions were identified as the sequence between the second conserved cysteine encoded by the 5′ Vβ gene segment and the conserved phenylalanine encoded by the 3′ Jβ segment (IMGT).

The comparisons between groups for overall survival functions were done using the log-rank test. The random coefficient model was used to generate a quadratic fit of the weight change over time. For the colitis scores, cell frequencies and numbers, and serum cytokine levels, a non-parametric Kruskal–Wallis test was used to compare the measurements between the groups. For the pairwise comparisons, a Mann–Whitney U test was performed. The TCR repertoire data were analyzed using the Morisita–Horn Index, which was calculated using EstimateS software.

To establish the relevance of iTreg-produced IL-10 in a setting containing the nTreg subset, we induced colitis in BALB/c Rag1^−/− mice by transferring 4 × 10⁵ Thy 1.2⁺ EGFP^– CD45RB^hi cells isolated by cell sorting from BALB/c Thy1.2⁺ Foxp3^ΔEGFP mice. The transferred CD45RB^hi cells have a nonfunctional Foxp3 allele, and colitis develops with accelerated kinetics in the absence of in vivo–derived iTregs (6). When mice lost ∼2.5% of their initial body weight, we treated them with different combinations of 5 × 10⁵ iTregs plus 5 × 10⁵ nTregs, where one or both of the Treg subsets lacked the capacity to produce IL-10 (Fig. 1). For all experiments, the iTregs were derived in vitro from CD4⁺ EGFP^– cells isolated from the spleens and lymph nodes of Foxp3^EGFP or Il10^−/− Foxp3^EGFP mice and cultured with TGF-β1 and TCR cross-linking to upregulate Foxp3 expression. The nTregs were isolated by cell sorting from the spleen and lymph nodes of these same mice.

In the absence of any therapeutic intervention, mice rapidly lost weight, became moribund, and were sacrificed, with a mean survival of 45 d (Fig. 1A, 1B). In treated mice, where the transferred iTreg compartment supplied the only source of Treg-derived IL-10, weight gain and survival were similar to those of control mice that were treated with WT iTregs plus WT nTregs (Fig. 1A). Reversing the experiment, where the nTreg compartment supplied the IL-10, showed similar patterns of weight gain and survival. In contrast, complete absence of all IL-10 produced by Tregs, while better than no therapy, still resulted in weight loss, and only 25% of the mice survived to 125 d (Fig. 1A, 1B). When Treg-derived IL-10 was limited to iTregs, there was also less inflammatory infiltrate in the colons of treated mice (Fig. 1C, 1D). Serum levels of IFN-γ were reduced when IL-10 was produced by at least one Treg compartment, and TNF-α was reduced in all treatment groups irrespective of the capacity of Tregs to produce IL-10 (Fig. 1E). Serum levels of IL-17A were not reduced with Treg treatment.

An increase in iTreg frequency and/or number was seen in the MLNs, colon, and spleen of mice treated with WT iTregs plus Il10^−/− nTregs compared with control mice treated with WT iTregs plus WT nTregs (Fig. 2A, 2B, Supplemental Fig. 1A, 1B, and data not shown). However, the number of nTregs recovered from these tissues was similar between all treatment groups, demonstrating that there was no effect of Treg-derived IL-10 on nTreg recovery (Fig. 2B, Supplemental Fig. 1B). These data indicate that a compensatory expansion of WT iTregs occurs when nTregs lack the capacity to produce IL-10 and that differences in clinical outcome between the treatment groups are not due to an overall reduction in Treg numbers.

Next, we examined the impact of iTreg-produced IL-10 on effector T cells in the target tissues. Overall, the number of CD4⁺ T cells was reduced in the MLN and spleen of treated mice compared with untreated controls, irrespective of the IL-10 status of the Tregs used to treat the mice (Fig. 2C and data not shown). CD4⁺ T cells were further reduced in the MLN of mice treated with WT iTregs plus Il10^−/− nTregs, suggesting a modest effect of iTreg-produced IL-10 in this location. Consistent with the reduction in CD4⁺ T cells, the number of IFN-γ⁺, IL-17A⁺, or IFN-γ⁺ IL-17A⁺ effector CD4⁺ T cells in the MLN of treated mice was also reduced when the iTreg compartment selectively produced IL-10 (Fig. 2D, 2E) and closely matched that seen in mice treated with Il10^−/− iTregs plus WT nTregs. In the colon, the frequency of IL-17A⁺ and IFN-γ⁺ IL-17A⁺ T cells was reduced when Treg IL-10 production was limited to iTregs (Supplemental Fig. 1C). These data demonstrate that iTregs are a potent source of IL-10 and that iTreg suppression of lymphoproliferation and Th1 and Th17 cell differentiation is closely associated with IL-10 production. The data also confirm a similar outcome with a reverse in experimental design. Additional mechanisms of Treg-mediated suppression may be operative, as a modest Treg-associated effect on weight change and survival was observed in the complete absence of Treg-derived IL-10.

In all treatment groups, many Thy1.1⁺ Foxp3⁺ iTregs lost Foxp3 expression (ex-iTregs) and were maintained in the MLN, colon, spleen, and small intestine of treated mice (Fig. 2A, Supplemental Fig. 1A, and data not shown). Contaminating EGFP^– cells from the sorted iTreg cultures contributed little (12%) to the ex-iTreg population (Supplemental Fig. 1D, 1E). When comparing groups, the frequency of ex-iTregs was highest in the MLN and spleen of mice treated with IL-10–sufficient iTregs (Fig. 3A, 3B). This pattern was reversed in the colon and small intestine (Supplemental Fig. 1F). The ex-iTregs isolated from mice treated with WT nTregs plus WT iTregs did not continue to produce IL-10 as demonstrated by RT-PCR of sorted cells (Supplemental Fig. 1G). However, in vitro stimulation of ex-iTregs isolated from multiple tissues resulted in the production of IFN-γ and IL-17A (Fig. 3C and Supplemental Fig. 1H). Importantly, the frequency of ex-iTregs in the MLN that were IL-17A⁺ or IFN-γ⁺ IL-17A⁺ was reduced when iTregs produced IL-10 (Fig. 3C). The frequency of ex-iTregs producing IL-17A was increased (18–27%) in the small intestines of mice from all treatment groups relative to the other tissues (data not shown). These data demonstrate that ex-iTregs can produce proinflammatory cytokines and that the capacity of ex-iTregs to produce IL-17A and the localization of ex-iTregs may be influenced by iTreg-produced IL-10.

We investigated the phenotypic and molecular relationship between nTregs, iTregs, and ex-iTregs by examining their cell surface phenotype, gene expression profile, and their methylation status at the TSDR of conserved noncoding sequence 2 (CNS2) within the Foxp3 promoter. Flow cytometric analysis of cells from the MLN of mice treated with WT iTregs plus WT nTregs revealed that iTregs and nTregs had similar expression of several cell surface markers including CD25, CD62L, CD44, CD103, KLRG1, GITR, and CTLA-4 (Fig. 4A). Helios was expressed in 20.6% (±5.8%) of the iTregs in the spleen, but was generally not found in iTregs in the MLN or in ex-iTregs (Fig. 4A, 4B, and data not shown). Several differences were noted between ex-iTregs and iTregs found in the MLN. Compared to iTregs, the ex-iTregs had lower levels of CD25, CD103, KLRG1, and CTLA-4, although they had similar expression of CD62L, CD44, and GITR (Fig. 4B).

Next, we sorted nTregs, iTregs, and ex-iTregs from the spleens and MLNs of mice treated with WT iTregs plus WT nTregs and examined their gene expression profiles. From the three groups, we identified a total of 3250 probe sets that were differentially regulated relative to naive CD4⁺ T cells (Supplemental Table I). This analysis captured 319 of the 603 probe sets found in the canonical Treg transcriptional signature (31). Notably, 1437 (44%) probe sets were common to all three groups and included ∼40% (132 of 319) of the Treg probe sets that we recovered (Fig. 4C). These included directionally concordant expression of the Treg signature genes Itgae, Lef1, Dusp4, Ctla4, Ahr, Gzmb, Ccr6, and Rora. In most cases, however, the ex-iTregs had the smallest fold changes in level of expression (Fig. 4D). Most of the remaining Treg signature genes were suppressed or not differentially regulated in the ex-iTreg expression profile, including Ikzf4, Gpr83, Nrp1, Ikzf2, Pde3b, Il2ra, Ebi3, and Klrg1. The genetic signatures of in vitro–derived iTregs that were maintained in vivo and nTregs were remarkably similar (64% overlap, Fig. 4C), consistent with our previous data examining the relationship between in vivo–derived iTregs and nTregs (7).

Ex-iTregs also differentially regulated a number of genes not contained within the canonical Treg signature (Fig. 4E). Ex-iTregs showed induction of Il17a (30-fold versus Tconv cells) and Ifng (11-fold versus Tconv cells), consistent with the intracellular cytokine staining. In contrast, Il10 expression was largely confined to nTregs and iTregs (52-fold and 27-fold increases versus Tconv cells, respectively). The ex-iTregs expressed Il2, Il22, Ccl5, and Gzmk. Several other genes not associated with the Treg signature were expressed in ex-iTregs as well as both Treg subsets, such as Lyz1, Gp49a, S100a8, S100a9, and Klf4. In all three cell types, Ikzf1 (Ikaros) and Sox4 were repressed.

To investigate further the molecular relationship between the nTreg, iTreg, and ex-iTreg subsets, we analyzed the methylation status of the TSDR within CNS2 of the Foxp3 promoter of mice rescued with WT nTregs plus WT iTregs. As expected, the CpG islands in the TSDR of WT nTregs were demethylated (Fig. 4F, top panel) (7, 29). Consistent with their lack of Foxp3 expression, the CpG islands in ex-iTregs showed extensive methylation at the examined CpG motifs. In aggregate, CpG islands in the TSDR of stable in vitro–derived iTregs isolated from successfully treated mice showed partial demethylation, although there was considerable variability in the methylation patterns found in individual mice. Notably, iTregs isolated from at least two mice were largely demethylated (Fig. 4F, bottom panel). The extensive iTreg demethylation seen in these two mice did not correlate with Treg or ex-iTreg recovery, rate of weight loss after colitis induction, or the weight gained after treatment (data not shown).

Taken together, these gene and protein expression data indicate that ex-iTregs retain a component of the Treg canonical signature. However, loss of Foxp3 expression is associated with changes in the expression of several genes important for Treg suppressive function and in the acquisition of genes related to the inflammatory response and T cell effector capability (32–36). The general failure of most in vitro–derived iTregs to demethylate CpG islands in the TSDR likely contributed to the large number of ex-iTregs present in the treated mice, as demethylation of these residues has been shown to be important to iTreg stability (37). Fully demethylated iTreg TSDRs were occasionally observed, suggesting that iTregs per se are not excluded from stable Foxp3 expression.

The ex-iTregs expressed genes associated with regulation but also expressed genes associated with inflammation. Because of the dual nature exhibited by ex-iTregs, we tested their pathogenic potential by the adoptive transfer of 1 × 10⁵ sorted ex-iTregs into Rag1^−/− hosts. Recipients rapidly lost weight, became moribund, and were sacrificed after ∼40 d (Fig. 5A, 5B). Histological analysis of the colon revealed lymphocytic infiltration and severe colitis (Fig. 5C). Notably, an average of 2% of the transferred cells found in the MLN regained Foxp3 expression in vivo. This frequency of iTregs matched that seen after the transfer of 1 × 10⁵ CD4⁺ EGFP^– CD45RB^hi cells into Rag1^−/− mice (Fig. 5D). Thus, in the absence of Tregs, ex-iTregs cause severe disease. The capacity of ex-iTregs to upregulate Foxp3 is retained and is similar to that of naive CD4⁺ T cells.

The capacity of ex-iTregs to reacquire Foxp3 expression after transfer into Rag1^−/− hosts (Fig. 5D) suggested that the ex-iTregs and iTregs could be interconverting and thus clonally related. To test this hypothesis, we isolated iTregs and ex-iTregs from six mice treated with WT iTregs plus WT nTregs and characterized the TCR repertoires. Notably, mouse 1 and mouse 2 received equivalent aliquots drawn from the same population of pooled Tregs. In a separate experiment, we also treated mouse 4 and mouse 6 by splitting a pooled Treg sample. Mouse 3 and mouse 5 each received an unrelated population of cells. This experimental design allowed us to compare the iTreg and ex-iTreg populations that were maintained within a single individual and between two individuals that received equal fractions of the same inoculum. Fragment analysis of 12 Vb CDR3 regions from six treated mice showed both skewed and Gaussian distributions with limited similarity between the iTreg and ex-iTreg populations, both within individuals and between those individuals that received equivalent Treg populations (Supplemental Fig. 2A). For example, only the Vb1 and Vb2 analyses showed a dominant peak at the same CDR3 length distribution in more than half of the samples. Spectratypes dominated by a single CDR3 length were occasionally observed and in general did not correlate between cell populations and mice.

We next examined the clonal relationship between iTregs and ex-iTregs by comparing TCR sequences from mouse 4 and mouse 6. Vb8.2 cDNA fragments were selected for TCR CDR3 sequencing based on the broad distribution of CDR3 lengths with both Gaussian and skewed distributions seen in all analyzed mice (Fig. 6A). CDR3 regions from each population were amplified by PCR, linked to beads, and examined using non-optical integrated semiconductor sequencing. We obtained 416,207 in-frame reads encoding 5,162 iTreg and 3,738 ex-iTreg CDR3 polypeptides (Supplemental Table II). Repertoire comparisons based on amino acid sequences revealed minimal overlap (∼2%) between unique iTreg and ex-iTreg CDR3 regions (Fig. 6B). Consistent with this observation, the Morisita–Horn Index (MHI) was 0.11 and 0.12 for the two comparisons, indicating little similarity. Ln-rank frequency versus ln-rank plots revealed a two-component distribution consisting of clones that follow a power law relationship and multiple high-rank clonotypes. To determine if the same high-rank clonotypes expanded after transfer, we compared the two iTreg and the two ex-iTreg repertoires that developed in treated mice that received aliquots from a common Treg pool (Fig. 6C). There was little overlap between the iTreg populations (∼1%, MHI = 0.092). The ex-iTreg populations also showed a small 1.4% overlap, although the MHI for this comparison was higher (MHI = 0.488). The higher MHI here indicates that a few of the high-rank clonotypes were recovered from both mice. For control experiments, we compared the iTreg clonotypes recovered from the same samples run on two different semiconductor chips or by direct cloning and sequencing of all four populations (Fig. 6D, Supplemental Fig. 2B, Supplemental Table II). As expected, the TCR repertoires recovered by the different methods were highly similar (average MHI ∼0.8) and in some instances nearly identical (MHI = 0.971). These data validate non-optical integrated semiconductor sequencing in this application and demonstrate that we can find similarity between repertoires where it exists. Collectively, these data indicate that each mouse expanded unique populations of iTregs and ex-iTregs, and the ex-iTreg populations also shared a few high-rank clonotypes between the two mice. Within an individual mouse, the iTreg and ex-iTreg populations were clonally unrelated.

In this study, we modified a T cell transfer model of colitis to create chimeric mice where nTregs and iTregs were selectively deficient in IL-10. Although iTregs composed a fraction (∼20%) of all Tregs recovered from treated mice, the IL-10 supplied by iTregs could replace nTreg-derived IL-10 in the cure of disease. In the reverse experiment, nTreg-derived IL-10 was equally effective. These results demonstrate the principle of reciprocal compensation between Treg subsets, which is an essential tolerogenic mechanism. Importantly, we identified iTregs as a particularly potent source of IL-10 in the gastrointestinal tract that can be used to augment regulatory function in situations where nTregs are defective or depleted. Reductions in Treg numbers and/or function have been implicated in many human autoimmune diseases (38). This therapeutic approach may therefore have broad application, as long-term stable tolerance can be achieved through iTreg transfers when nTregs are also present.

Prior work has shown that the iTreg transcriptional signature 72 h after in vitro induction of Foxp3 with TGF-β was largely independent of Foxp3 expression and distinct from that of nTregs (6, 39). In this study, we extended these studies by determining the transcriptional profile of in vitro–derived iTregs that were maintained in treated mice for approximately 3 mo. We now demonstrate that nTregs and in vitro–derived iTregs that are stable in vivo share similar transcriptional profiles, including the expression of many genes associated with Treg suppressive function in addition to Il10. These data provide further support for the hypothesis that the two Treg subsets use similar suppressive mechanisms. This result also illustrates that the in vitro–derived iTreg and nTreg transcriptional signatures converge as the iTregs are selected and maintained in vivo. One caveat to these studies is that we have compared in vivo–derived nTregs to in vitro–derived iTregs and not to in vivo–derived iTregs. In vitro–derived iTregs that persist in treated mice may not be equivalent to in vivo–derived iTregs under the same conditions. Further studies are needed to compare the function and stability of iTregs and nTregs derived in vivo. This caveat notwithstanding, we have identified important characteristics of stable in vitro–derived iTregs relevant to their use in immunotherapy.

Notably, ∼85% of the surviving Thy1.1⁺ population no longer expressed Foxp3 three months after transfer. Although these ex-iTregs retained some elements of the canonical Treg transcriptional signature, the TCRβ CDR3 repertoires of iTregs and ex-iTregs were essentially non-overlapping. Thus, the generation of a large population of ex-iTregs in vivo may be the unintended consequence of in vitro conversion based on nonspecific TCR cross-linking. In the absence of Treg control, the prior history of Foxp3 expression was not sufficient to control the pathogenic potential of ex-iTregs upon retransfer into Rag1^−/− hosts. However, iTreg-derived IL-10 limited the frequency of ex-iTregs adopting a Th1, Th17, or Th1+Th17 cell phenotype. This is consistent with previous studies showing that IL-10 provided exogenously or by CD4⁺ Foxp3⁺ Tregs constrains Th17 and Th1+Th17 cell frequencies (23). Additionally, IL-10 signaling in Tregs and Th17 cells limits Th17 cell-mediated inflammation (23, 40). Collectively, these data support a specific role for iTreg-produced IL-10 in the control of Th17 responses.

After retransfer into Rag1^−/− hosts, ∼2% of the ex-iTregs reacquired Foxp3 expression. Thus ex-iTregs retain the potential to re-express Foxp3 and irrespective of their pathogenic potential may serve as an iTreg reservoir in instances when iTreg responses are exhausted or diminished. Unstable Foxp3 expression in both nTregs and iTregs has been described, although others have concluded that Foxp3 expression in nTregs is heritable and stable (41–48). Our results clearly demonstrate that many in vitro–derived iTregs have unstable Foxp3 expression in vivo. It has been shown that iTregs derived in vitro are unstable due to a failure to fully demethylate the TSDR (29, 37). This incomplete demethylation could contribute to the high number of ex-iTregs observed in treated mice. It is interesting, however, that even mice with iTregs that had an extensively demethylated TSDR still had a sizeable fraction of in vitro–derived iTregs that lost Foxp3 expression and became ex-iTregs. Furthermore, stable iTregs that never fully acquired the nTreg TSDR signature were recovered from successfully treated mice, both in this experimental colitis model and in a model of Foxp3 deficiency, suggesting that pathways other than promoter demethylation may impact the maintenance of iTregs in vivo (7). The amount of intermouse variability in the extent of iTreg TSDR demethylation further supports the notion that factors such as subclinical inflammation and disease status may contribute to iTreg stability. Because iTregs can be specific for gut bacterial Ags, strategies to derive polyclonal populations of Ag-specific iTregs prior to transfer may enhance the stability of Foxp3 expression (49). Treatment with inhibitors of DNA methyltransferases and histone deacetylases have been shown to enhance the stability of Foxp3 expression and could be incorporated into in vitro induction protocols (37, 50, 51).

When a pooled population of nTregs and iTregs was used to treat two mice, iTreg populations emerged that composed a similar fraction of the Treg compartment but contained distinct TCR repertoires. This result supports two conclusions. First, a limited and fixed pool of in vitro–derived iTregs contains a surprisingly large number of clones with TCRs that can be maintained within the iTreg niche. In fact, it has been shown that high TCR diversity is important for the optimal function of Tregs in a model of experimental acute graft versus host disease (52). Second, the iTreg niche is restricted, as selection of iTregs is likely to be stochastic, and there was minimal overlap in the iTreg repertoires. This constraint is probably not due to the number of available Ags, given the complexity of the microbiome. Thus, other factors such as the number of tolerogenic APCs, local concentrations of TGF-β1 and IL-2, and signaling via the programmed death 1–programmed death ligand pathway may determine the size of the iTreg population (53–57). In the absence of nTregs, the iTreg compartment expands ∼5-fold indicating that nTregs, either directly or indirectly, also influence the size of the iTreg niche (6). Manipulation of these factors may provide mechanisms to expand the iTreg compartment thereby decreasing the propensity for the formation of pathogenic ex-iTregs while improving therapeutic outcome.

In conclusion, recent work has established that iTregs and nTregs are functionally nonredundant, based in part on differences in TCR repertoire and presumably TCR specificity (7, 49, 58). We now identify the capacity for reciprocal compensation between the two Treg subsets that relies on a mutual suppressive mechanism. Thus, the answer to the recently posited question “Natural and adaptive Foxp3⁺ regulatory T cells: more of the same or a division of labor?” (2) is that both shared methods and unique characteristics are needed to enforce tolerance.

We thank James Verbsky for kindly providing the Foxp3^EGFP IL10^−/− mice. We thank Brandon Edwards and Kyle Upchurch for assistance with cell sorting. We also thank Mary Williams for administrative assistance and William Drobyski and Jack Gorski for critical review of the manuscript.

The authors have no financial conflicts of interest.