Single-Cell Gene Expression Reveals a Landscape of Regulatory T Cell Phenotypes Shaped by the TCR

Home , FOXP3, GNLY, GPR83, GSTO1, IL2RB, SAMHD1

Articles https://doi.org/10.1038/s41590-018-0051-0

Single-cell gene expression reveals a landscape of regulatory T cell phenotypes shaped by the TCR

David Zemmour1, Rapolas Zilionis2,3, Evgeny Kiner1, Allon M. Klein2, Diane Mathis1* and Christophe Benoist 1*

+ CD4 T regulatory cells (Treg) are central to immune homeostasis, their phenotypic heterogeneity reflecting the diverse envi- ronments and target cells that they regulate. To understand this heterogeneity, we combined single-cell RNA-seq, activation + – reporter and T cell receptor (TCR) analysis to profile thousands of Treg or conventional CD4 FoxP3 T cells (Tconv) from mouse lymphoid organs and human blood. Treg and Tconv pools showed areas of overlap, as resting ‘furtive’ Tregs with overall similarity to Tconvs or as a convergence of activated states. All Tregs expressed a small core of FoxP3-dependent transcripts, onto which additional programs were added less uniformly. Among suppressive functions, Il2ra and Ctla4 were quasiconstant, inhibitory cytokines being more sparsely distributed. TCR signal intensity did not affect resting/activated Treg proportions but molded activated Treg programs. The main lines of Treg heterogeneity in mice were strikingly conserved in human blood. These results reveal unexpected TCR-shaped states of activation, providing a framework to synthesize previous observations of Treg heterogeneity.

egulatory T cells are dominant-negative regulators of many contexts, in order to promote peaceful coexistence with commensal facets of the immune system, controlling immune responses microbiota17 or fetal antigens. 18 and enforcing peripheral tolerance to self, symbiotic com- The TCR plays a central role in the Treg life story . It is neces- R 1 mensals and fetal antigens . In addition, some Tregs reside in non- sary for Treg differentiation, and the signals that it delivers upon lymphoid tissues, where they help control tissue homeostasis and major histocompatibility complex (MHC)-peptide recognition, sterile inflammation2. conditioned by costimulatory and other modulators, rescues pre- 1, 3 Tregs constitute a diverse constellation of cells . They have mul- cursor cells from clonal deletion. Continued TCR presence and 4 tiple origins : many Tregs differentiate in the thymus, but others engagement by MHC molecules is required for suppressive activ- + 19 arise in the periphery from naive CD4 T cells upon suboptimal ity and differentiation to an activated phenotype . The Treg TCR exposure to antigens, in particular microbial antigens. Their organ- repertoire is skewed toward recognition of self-antigens but is as 18 ismal locations vary: they reside not only in the T cell zones of lym- broad as that of Tconvs . phoid organs but also in B cell areas, where they control antibody Understanding Treg molecular diversity and definition, in rela- maturation and production (T follicular regulators (Tfr cells)) in tion to Tconv cells, is thus complex and confounded by the different autoimmune or tumoral lesions or at body–microbiota interfaces. states that both populations can adopt in response to various stim-

Their effector pathways are heterogeneous: Tregs utilize cell-surface uli. Single-cell transcriptome analysis offers the potential to illumi- inhibitors, such as CTLA4; inhibitory cytokines, such as IL-10, nate these questions, in an unbiased manner that does not rely on IL-35 or TGF-β; cytokine capture via the IL-2 receptor; purine- assumptions of cell-type identities. Although single-cell (sc) RNA- mediated suppression; or direct cytoxicity5. These facets correspond seq remains challenging20,21 because of the limiting sensitivity of 1,3 to diverse Treg subphenotypes . Particular Treg subtypes have been detection and the large dimensionality of the data, the approach has recognized on the basis of expression of chemokine receptors, such been transformative, for example, in identifying novel cell types and + as CXCR3 (CXCR3 Tregs are particularly adept at suppressing Th1 in dissecting transcriptional differences previously masked by the responses6) or CXCR5 (in Tfr cells)7), or activation markers (in ‘averaging’ inherent in profiling RNA from pooled cells22. 8–11 ‘eTregs’ or ‘aTregs’) . These more activated types of Tregs are particu- Here, we applied scRNA-seq to profile thousands of single Treg 2 larly represented among extralymphoid Tregs in inflammatory sites . and Tconv cells, in mice and humans, to reveal the diversity of tran- Tregs and Tconvs have opposite immune functions, but their molec- scriptional phenotypes that can be adopted by Tregs. We concentrated ular distinction can be complicated. Stable expression of FoxP3 is on two driving questions: how are Tregs and Tconvs related, and how do semantically eponymous for Tregs, and FoxP3 controls a substantial TCR-mediated signals affect Treg activation. For focus, we limited 12 fraction of the characteristic transcriptional signature of Treg cells . the present analysis to Treg and Tconv cells from lymphoid organs. The However, FoxP3 is not sufficient, and several other transcription results reveal an unexpected degree of overlap between Tregs and factors not specific to Tregs but also present in Tconvs are required by Tconvs and provide a framework integrating many prior observations 4 Tregs . Further blurring the Treg/Tconv distinction, FoxP3 itself can be on effector-Treg states. 13 14 expressed transiently upon activation in human and mouse Tconvs. Conversely, whereas the Treg phenotype is generally stable, Tregs can Results 15,16 + lose FoxP3 expression under stress, such as IL-2 deprivation . CD4 Treg and Tconv scRNA-seq datasets. In order to maximize the Finally, Tregs can differentiate directly from Tconvs in tolerogenic power to find small (sub)populations of cells and significant gene

1Division of Immunology, Department of Microbiology and Immunobiology, Harvard Medical School, and Evergrande Center for Immunologic Diseases, Harvard Medical School and Brigham and Women’s Hospital, Boston, MA, USA. 2Department of Systems Biology, Harvard Medical School, Boston, MA, USA. 3Institute of Biotechnology, Vilnius University, Vilnius, Lithuania. *e-mail: [email protected]; [email protected]

in T , as expected (Fig. 1c). More generally, the changes in gene a c conv 8 Foxp3 expression between Tregs and Tconvs were conserved, as compared with

Treg standard-population RNA-seq data, and substantially overlapped 12 published Treg- and Tconv-specific signatures (Supplementary Fig. 1d).

+ Tconv Transcriptional relationship between CD4 Treg and Tconv cells. As /FoxP3 4 discussed above, Treg and Tconv cells are ontogenetically related and GFP CD4 can derive from one another in some circumstances. Many profiling + + analyses at the population level have shown that a clear and repro- Gated on TCRβ CD4 ducible gene expression signature distinguishes the two cell types,

b T #1 T #3 with Foxp3 as the most extreme example of a Treg-specific gene. The reg reg Tconv Treg

Treg #2 Tconv averaging generated by population-level profiling can mask com- 15 Il2ra plexities in this relationship, so we explored our scRNA-seq data Normalized counts )

10 specifically in this light. We focus here on splenic populations from 3 unperturbed mice, in which Treg and Tconv populations are generally thought to be stable, as opposed to locations of cell stress, where instability has been observed24. 5 We first used t-distributed Stochastic Neighbor Embedding (tSNE)25 for dimensionality reduction to relate independently Number of expressed genes detected (lo g 2 sorted splenic Treg and Tconv cells (sorted independently according 234 gfp to phenotype in the Foxp3 mice; Fig. 2a,b). Although Treg and Tconv Number of unique T T conv reg cells generally partitioned into two main areas (D and E in Fig. 2a), mRNA molecules (log ) 10 Single cell they were also somewhat imbricated. First, some Treg cells (hereaf- ter referred to as ‘furtive’ Tregs) were found in the main Tconv area,

Fig. 1 | Treg and Tconv scRNA-seq datasets. a, Flow cytometric plot of gated and vice versa (best seen in Fig. 2b, top); second, there were groups + + gfp CD4 TCRβ Treg and Tconv cells from Foxp3 mice, sorted for scRNA-seq of cells in which Tregs and Tconvs comingled apart from the bulk Treg library construction. Representative of 5 experiments. b, Single-cell and Tconv pools (areas A–C, demarcated in Fig. 2a by bootstrap- transcriptome coverage, shown as the number of uniquely identified mRNA optimized partition clustering). Differences in sequencing coverage reads per cell (x axis) versus the number of expressed genes per single cell, were unlikely to explain this clustering, because the distributions of in three Treg biological-replicate datasets (details in Supplementary Table the detected genes and number of reads were similar in the different

1) and the Tconv dataset. Single cells with <500 transcripts (dashed line) areas (Supplementary Fig. 2a,b). In addition, similar patterns were sequenced were excluded from further analysis. c, Scatter plot representing observed in additional Treg and Tconv datasets generated with two the expression (normalized counts) of Foxp3 and Il2ra in the single Treg other scRNA-seq platforms: the 10x Genomics platform or 96-well-

(n = 708) and Tconv cells (n = 1,098). sorted cells profiled by CEL-seq (Supplementary Fig. 2c–e). Furtive Tregs represented 26% of splenic Tregs in these experiments. Foxp3 transcripts were detectable in many of them (Fig. 2b, bottom), correlations, we performed single-cell transcriptomics on thou- at levels comparable to those of other Tregs (Fig. 2c). Cytometric sort- sands of single cells by using InDrop, encapsulating single Treg and ing of Tregs into plates for scRNA-seq allowed us to relate protein 21,23 + + + Tconv cells in microfluidic droplets . We sorted CD4 TCRβ GFP levels to the transcriptome in single cells. The results showed that – gfp Treg and GFP Tconv cells from Foxp3 mice (Fig. 1a) and encapsu- furtive Tregs expressed FoxP3 protein at the same level as did other lated them separately at high efficiency (70–80%) before scRNA-seq Tregs (Supplementary Fig. 2e,f). Furtive Tregs were therefore unlikely library construction23. Over the course of this study, we analyzed to result from misexpression of the reporter or from contaminating

4,237 splenic Tregs (in three independent cell cohorts) and 1,093 cells in the initial sort (moreover, at 26% of the splenic Tregs profiled, splenic Tconvs (one cohort matching Treg #1). In order to minimize furtive Tregs represented far more than the 3.5% contamination rate batch effects and to assess the robustness of the results, we analyzed in the dataset). Canonical Treg-signature genes were also overex- each Treg cohort independently and confirmed the reproducibility pressed in these cells compared with surrounding Tconv cells (Fig. 2d). of the cell or gene clusters in the biological replicates. We extended However, some of the Treg signature transcripts were shifted in those + these data by scRNA-seq of total splenic CD4 T cells (n = 2,508, cells relative to bulk Tregs (Fig. 2e), including Il2ra and Ikzf2 (encodes 26 containing Tregs and Tconvs) through a different microfluidic platform Helios), an important transcription factor in Treg cells . Furtive Tregs (10x Genomics) and library protocol, and through a plate-based also overexpressed Rel, Cd177 and several transcripts of genes nor- approach in which individually sorted Tregs were analyzed with CEL- mally associated with Tconvs (e.g., Itgb4) (Fig. 2f). 20 seq (n = 200) (Supplementary Table 1). In addition to these Tconv-like Tregs, we also observed some Tregs On average, 1,751 unique mRNA molecules representing 787 and Tconvs in discrete clusters at the interface (areas A–C in Fig. 2a), genes were detected per single cell in the primary InDrop datasets, which were separated from the major Treg and Tconv populations altogether surveying 16,720 genes (Fig. 1b). All batches had similar and clustered together instead. In area A (Fig. 2g,h), both Tregs and coverage (Fig. 1b), and we observed good concordance among them Tconvs overexpressed a set of genes associated with residence in B cell for both the mean (Supplementary Fig. 1a) and the variation in gene follicles (i.e., T follicular helper (Tfh) and Tfr cells), including the 27 expression (Supplementary Fig. 1b). One of the advantages of single- characteristic Cxcr5 and Icos transcripts . In area B (Fig. 2i,j), Tconv cell profiling over population profiling is that rare contaminating and Treg cells both upregulated a set of genes belonging to the early cells from the sort can be identified, and we used for identification a response to TCR engagement in both cells (Nr4a1, Egr1, Egr2, Myc naive Bayes algorithm that compares each single-cell transcriptome and Dusp2)28, thus suggesting that TCR signals may drive similar with each of the 249 immune-cell states profiled by the Immgen programs in both Tregs and Tconvs. Consortium (here, 3.5% of the cells were identified as contaminants— Although Treg and Tconv cells thus converged to similar subpheno- mostly B and myeloid cells—and culled; Supplementary Fig. 1c). For types defined by the integration of all transcripts through the tSNE verification, Foxp3 and Il2ra, two important Treg markers, were found algorithm, is there nevertheless an overarching set of transcripts to be expressed throughout the different Tregs and rarely expressed that generally demarcates Treg and Tconv identities, independently

abT T c T (n = 708) conv reg reg * T (n = 1,098) conv NS * All cells E 86% 6 26%

C Tconv Treg 9% 4 tSNE2 5% 3% Foxp3 expression 11% B + (normalized counts) D Foxp3 1% cells 6% 2 A 45% (mRNA) 6%

Tconv Treg Treg tSNE1 Expression (in area E) (in area E) (outside E) Low High

d e f Treg/Tconv signature Treg/Tconv signature Izumo1r Cd83 0 Foxp3 0 Nrp1 Il2ra Srgn Il2ra (in E) Ikzf2 10 Il2rb Bmyc Il2rb Ctla4 con v Ikzf2 Socs2 ) ) Tnfrsf9 Sgk1 Il2rb Tnfrsf4 Lrrc32 2 Prdm15 10 10 Rgs1 Id3 Izumo1r Egr1 Nedd1 Tnfrsf18 4 Dapl1 Il2ra S100a10 S100a6 Mab 1 Tnfrs9 Bcl2a1b 4 Tnfrsf4 Itgb4 Actn1 S100a4 S100a4 Gsto1 Rel P value (–log P value (–log Ift80 Itgb1 Ahnak Helios Sgk1 Cd177 Samhd1 Vipr1 Id3

Igf2r (outside E) versu s T P = 0.0003 6 8 Foxp3 reg Dapl1 Tnfrsf1b P = 0.015 P = 0.02 Nr4a1 0.25 Igfbp4 P = 0.34 FC T 0.2 0.5 1 25 0.25 1 10 0.2 0.5 1 25

FC Treg (in E) versus Tconv (in E) FC Treg (in E) versus Tconv (in E) FC Treg (in E) versus Treg (outside E) g h k A versus main Tfr-signature genes

S100a4 (in E) Cxcr3 Itgb1

conv 10 S100a11 T T Icos conv reg Ccr2 E ABCD E ABCD

1 tSNE2 Foxp3

(in A) versus T Sell Il2ra

conv Il2rb 0.01 Tnfrsf4

FC T 0.01 1 10 tSNE1 Ikzf2 Ctla4 FC T (in A) versus T (in D) Low High reg reg Density of Gpr83 Capg expressing cells i j Izumo1r B versus main TCR-activation-signature genes Chchd10 Tnfrsf18

(in E) Dapl1 Egr1 Nr4a1 Igfbp4 conv 10 Dusp2 Egr2

Myc

1 tSNE2 (in B) versus T

conv Pim1 0.01

FC T 0.01 1 10 tSNE1 Low High FC Treg (in B) versus Treg (in D) Density of expressing cells

Fig. 2 | Some Tregs and Tconvs show closely related transcriptomes. a, Two-dimensional tSNE plot of Treg (red) and Tconv (black) single-cell transcriptomes

(each dot represents one cell) with different areas (A–E) identified by unsupervised clustering. The proportions of Tregs and Tconvs in each area relative to the total number of Tregs and Tconvs is indicated. b, Same tSNE plots as in a, but split to show Treg and Tconv cells separately (top) and cells in which Foxp3 transcripts were detected (bottom; the size of each dot indicates the level of Foxp3 RNA expression). c, Foxp3 expression (normalized counts) in furtive

Tregs in area E in comparison to Tregs outside area E. Bars indicate the means. *P < 0.05; NS, not significant (two-tailed t test, n = 5 Tconvs, 46 Tregs in E and

235 Tregs outside E, all expressing Foxp3 mRNA). d, Furtive Tregs show the usual biased transcriptome for Treg-signature genes. Volcano plot (fold change

(FC) versus P value) comparing the gene expression profiles of furtive Treg versus main Tconv (both from area E of the plot). Up- and downregulated Treg- 12 2 signature genes are highlighted (red and blue, respectively). χ-test P values. e, Treg/Tconv expression ratio for furtive Tregs in area E (x axis) versus Treg/Tconv ratio outside area E (x axis). Up- and downregulated Treg-signature genes are highlighted (red and blue, respectively). Gene lists are the genes gated on the plot. f, Down-tuning of the Treg signature in furtive Tregs. Volcano plot comparing the gene expression profiles of furtive versus other Tregs (in and outside 2 area E, respectively). Up- and downregulated Treg-signature genes are highlighted (red and blue, respectively). χ-test P values. g, Tregs and Tconvs in area A differentially express the same set of genes, as compared with their respective main populations. Ratio of gene expression in Tregs in A versus D (x axis) 7 against the ratio of expression in Tconvs in A versus E (y axis). h, Same tSNE plot as in a, highlighting cells expressing Tfr-signature transcripts . i, Tregs and

Tconvs in area B differentially express the same set of genes, as compared with their respective main populations. Ratio of gene expression in Tregs in B versus

D (x axis) against the ratio of expression in Tconvs in B versus E (y axis). j, Same tSNE plot as in a, highlighting cells expressing the most upregulated genes 28 after TCR activation . k, Presence/absence (black/white) heat map showing the core Treg transcripts expressed in all Tregs regardless of their area (each vertical line represents one cell).

a b c Correlations in Tregs and Tconvs Correlations in Tregs Treg/Tconv signature Core T genes Ctla4 reg Klrg1 Il2ra Itgae Il2rb Ebi3 Tnfrsf4 Foxp3 Tnfrsf1b Tnfrsf18 Ctla4 Ikzf2 Cd83 1.00 Ikzf2 Nrp1 Snx18 Tnfrs4 Gzmb Icam1 Tnfrsf18 Penk Serpina3f Tnfrsf9 Tnfrsf1b

(mean expression) 0.01 re g T

0.01 1 Foxp3 Il2ra Tconv (mean expression) Il2rb <–0.2 >0.2 <–0.2 >0.2 Correlation Correlation

Fig. 3 | Treg-signature expression in single Treg and Tconv cells. a, Mean expression in collapsed Treg and Tconv single-cell datasets (normalized counts) 12 (n = 708 Tregs and 1,098 Tconvs). Canonical up- and downregulated Treg-signature genes are highlighted (red and blue, respectively). b, Gene–gene correlation heat map for the canonical Treg-signature genes, calculated across all Treg and Tconv single-cell datasets. c, Gene–gene correlation as in b, but calculated in the Treg single-cell dataset only.

of subphenotype variation? We systematically compared Tregs with observed: a small but relatively tight cluster that included the TNFR their closest Tconvs (correlation distances) (Supplementary Fig. 3a) family members Tnfrsf4 and Tnfrsf18 (encoding OX40 and GITR, and identified a small gene set (Il2ra, Il2rb, Ikzf2, Ctla4, Capg, respectively); another cluster that grouped Foxp3 and Il2ra; and a Tnfrsf4, Tnfrsf18, Izumo1r, Chchd10, Gpr83 and ex officio Foxp3) more loosely coordinated cluster that included several other mol- that was overexpressed by all Tregs irrespective of their location on ecules that mediate Treg function (Ebi3, Gzmb and Ctla4). Overall, the tSNE plot (Fig. 2k). The lack of expression of these genes in Tconv this paucity of correlation of Treg signature transcripts within Tregs cells falling in areas A, B or C indicated that these were true Tconv indicates that the different subphenotypes result from a diversity of cells, not activated cells transiently expressing FoxP3. These genes regulatory influences rather than from a few dominant programs. are all direct targets of FoxP3: they all contain enhancer elements In conclusion, the Treg signature identified at the population level that bind FoxP3 in chromatin immunoprecipitation experiments29,30 can be deconvoluted into a tight core of co-regulated genes, which (Supplementary Fig. 3b) and are up regulated upon ectopic FoxP3 are likely to be tightly controlled by FoxP3, and otherwise heteroge- expression in Tconvs. neous gene sets diversely expressed by individual Tregs. Thus, these data show that although most Tregs are generally distinct from Tconvs, and a small core of Treg-identifying transcripts can Splenic Treg heterogeneity revolves around several different poles. be defined, specific subsets of the two cell types do have consider- In the analyses presented above, we parsed primarily Treg heterogene- able overlap. ity in relation to Tconv cells and to the classic Treg signature. Yet other transcriptional features are likely to be important in defining the

Treg signature: core transcripts and heterogeneity of expres- functional identity of a given Treg cell—features that are not necessar- sion. Switching the analysis to the gene axis, we revisited at the ily Treg specific. We thus evaluated the distribution of all transcripts single-cell level the classic Treg signature identified by population among splenic Treg cells. We centered further analyses on the 100 12,31 transcriptomics . Collapsing the present single-cell data reca- transcripts with the most variability among Tregs. These transcripts pitulated the signature observed in previous microarray or popula- had variability above that expected from Poisson sampling statistics tion RNA-seq datasets, thus confirming congruence between the (Supplementary Fig. 4a) and were thus the most informative21 (the techniques (Fig. 3a). most variable genes were the same in the three batches of splenic

The natural cell–cell variability in gene expression revealed in Tregs) (Supplementary Table 2). Unsupervised partition clustering, scRNA-seq can be valuable in identifying modules of coregulated using a bootstrap approach to optimize the number of clusters and genes21,32. We thus searched for gene correlations in our data and assess their stability33 (Supplementary Fig. 4b,c), grouped the splenic asked whether the Treg signature behaves as a collection of dis- Tregs into six clusters (Fig. 4a on a subset of transcripts that best char- crete modules, as suggested by our earlier analyses12. Within the acterized the different groups) (Supplementary Tables 3 and 4). entire dataset including both splenic Tregs and Tconvs, there was, as These clusters should probably be thought of as discrete states in expected, a clear partitioning of Treg-up versus Treg-down signature a continuum rather than as strictly independent entities, as shown transcripts (Fig. 3b), and a generally positive but weak correlation below. Importantly, the analysis was performed on one cohort of between Treg-up transcripts, except for a small cluster of highly cor- Tregs, and these distinctions were reproduced in independent data- related genes (Fig. 3b). This cluster encompassed the same core set sets of splenic Treg cells (Supplementary Fig. 4d), even though the of Treg-specific transcripts (including Il2ra, Ctla4 and Foxp3) identi- relative proportions of cells in each cluster varied somewhat. fied in Fig. 2 as being shared by all Treg cells, independently of other Three main poles were observed within this partitioning. First, variations. This result confirms that only this small core of overex- there was a general gradient separating resting and activated Tregs, pressed genes identifies Tregs, which variably express other gene sets as judged by the frequency of transcripts typical of resting Tregs, Sell as a function of their location or functional subphenotypes, some of (encoding CD62L) and Ccr7. Tregs in clusters 4–6 seemed to be in a which can be shared with Tconv counterparts. resting state, while those in clusters 1–3 were in an activated state As expected, this tightly co-regulated core set vanished when we (Fig. 4a and Supplementary Fig. 4d). Second, beyond this separa- tested for correlations within Tregs only, aiming to identify co-regu- tion, cells in cluster 1–3 seemed to adopt two different poles of acti- lated components independently of their covariation relative to Tconvs vated phenotypes. Cells in cluster 3 predominantly expressed a set (Fig. 3c). Little correlation was seen, and only three clusters were of transcripts typical of early cell activation (Nr4a1, Egr1, Egr2, Myc

a c Cluster : 1234 56 0.2 Foxp3 1 (11%) Il2ra 3 (4%) 2 Ccr2 Id2 (4%) S100a4 S100a6 Icos 0 Itgb1 Cxcr3 CAP2 Dusp2 4 Nr4a1 Egr1 (25%) Egr2 −0.2 5 Ikzf2 (10%) Bach2 6 Satb1 −0.3 (46%) Ccr7 Sell −0.1 0 0.1 0High CAP1 Expression b d Cluster : 1234 56 Cluster : 123

Activated vs. resting Blimp+ vs. Blimp– CD44hi vs. CD62L p300 WT vs. KO Bach2 WT vs. KO Foxp1 WT vs. KO Gata3 WT vs. KO Anti-CD3 vs. unstim. 456 Anti-CD3/28 vs. unstim. Signatures TCR WT vs. KO CD62Lhi TCR WT vs. KO CD44hi Tfr vs. Tnaive Bcl6 WT vs. KO

Tissue Treg-up genes Tissue Treg-down genes Minimum Maximum

Relative expression tSNE 2 tSNE 1 e f Effector molecules Gated on CD4+TCRb+FoxP3+ * Il2ra Ctla4 Lrrc32Tgfb1 100 8 ● Flow *** *** ● cytometry

● 4 ● ● 50 ● CXCR3 CXCR3 0 NR4A1-GFP (MFI) CD62L – + NR4A1-GFP CXCR3 CXCR3 0 Gzmb Ebi3 Il10 Areg 6 100 expressing the gen e *** *** *** ***

4 regs 4 scRNA-seq 50 2 (corrected counts) 2 % of T

0 0 0 CXCR3 CXCR3 024 048 Il2ra Lrrc32 CD62L NR4A1 >0.2 Areg Ctla4 Il10 Gzmb Tgfb1 Correlation <0.2 Ebi3

Il2ra Areg Il10 Ebi3 Lrrc32 Ctla4 GzmbTgfb1

Fig. 4 | Resting and activated Treg states identified by scRNA-seq. a, Biclustering heat map of the expression of the most characteristic genes for each

Treg cluster (downsampled to equilibrate cluster size; each vertical line represents one cell). b, Biclustering heat map of expression signatures in Treg cells belonging to different clusters (cell and clusters aligned with a). WT, wild type; KO, knockout; stim, stimulated. c, CAP plot39. Each single cell is projected on a circular plan according to its likelihood of belonging to any of the six clusters defined in a (n = 708). The proportion of cells in each cluster is indicated. d, Same tSNE plots as in Fig. 2a, highlighting the locations of individual Treg cells from each of the Treg clusters. e, Cxcr3, Cd62l and Nr4a1 expression in single

Treg cells, as determined by flow cytometry (top row, representative of three experiments) or by scRNA-seq (corrected counts, bottom row). Top right, – + Nr4a1-GFP expression in CXCR3 and CXCR3 Tregs (as gated), *P < 0.05 (two-tailed paired t test, n = 3 mice). f, Proportion of cells in each Treg cluster (defined in a) that express different effector transcripts. Proportion calculated after correction for dropout events (Methods). ***P < 10−4 (ANOVA). and also Dusp2) and thus evoked TCR-mediated activation, while To integrate these data with the body of existing observations on 6–10 cells in clusters 1 and 2 were characterized by a more diverse set of Treg heterogeneity , we probed a large set of signatures distinguish- transcripts (Ccr2, Itgb1, S100a4, S100a6, Icos or Cxcr3). ing Treg cells, retrieved from databases or curated in house (Fig. 4b

a b Nr4a1 expression (normalized count) Low High 01050 Cell density % positive

Nr4a1hi 16.4 %

Nr4a1med 7.1 % Nr4a1-GFP lo 2.9 % FoxP3 Thy1.1 Nr4a1

+ + Gated on TCRβ CD4 tSNE 2 tSNE 1 c scRNA-seq: Flow cytometry: d e + Activated Treg resting Tregs resting CD62L Tregs Flow cytometry (% Treg cells) (% total Tregs) (% total Tregs) 100 50 0 0 50 100 0 10 20

hi hi Nr4a1 Nr4a1hi Nr4a1

med med Nr4a1 Nr4a1 Nr4a1med *

lo Nr4a1lo Nr4a1 Nr4a1lo CD62L 0 20 40 Cluster 5/6 CXCR3 + Cluster 1 CXCR3 (%) + + + Cluster 2 Gated on TCRβ CD4 FoxP3 Cluster 3

+ + Fig. 5 | Treg states associated with different TCR-signaling strength. a, Foxp3-Thy1.1 versus Nr4a1-GFP expression in gated CD4 TCRβ T cells, determined high hi medium med low lo by flow cytometry showing the gating strategy used to sort Nr4a1-GFP (Nr4a1 ), Nr4a1-GFP (Nr4a1 ) and Nr4a1-GFP (Nr4a1 ) Treg cells before microfluidic encapsulation for scRNA-seq library construction. Right, Nr4a1 expression in the corresponding Nr4a1hi, Nr4a1med and Nr4a1lo scRNA- seq datasets (n = 1,266, 1,265 and 1,567 cells, respectively) (red bar, means and 95% confidence intervals in expressing cells). b, Two-dimensional tSNE hi med lo plot of Nr4a1 , Nr4a1 and Nr4a1 Tregs. The heat colors represent cell density. c, Left, bar plot representing the proportion of Tregs belonging to the hi med lo resting clusters 5 and 6 (defined in Fig. 4) in Nr4a1 , Nr4a1 and Nr4a1 Tregs. Left, in the scRNA-seq datasets (clusters 5 and 6, in green and turquoise, respectively). Right, expression of CD62L (flow cytometry, representative of n = 4 experiments) among the three Nr4a1 cell sets. d, Bar plot representing hi med lo hi the proportion of Treg cells belonging to activated Treg clusters (defined in Fig. 4a) in Nr4a1 , Nr4a1 and Nr4a1 Tregs. e, CXCR3 expression in Nr4a1 , med lo + hi med Nr4a1 and Nr4a1 Tregs, determined by flow cytometry. Left, representative CXCR3 histograms. Right, proportion of CXCR3 cells in Nr4a1 , Nr4a1 and lo + + + Nr4a1 Tregs (as gated in the histograms; dashed lines). Gated on CD4 TCRβ FoxP3 cells. *P < 0.05 (two-tailed t test, n = 4 mice in two experiments). and Supplementary Fig. 5). The first set distinguished activated and We projected the likelihood of each single cell belonging to any of resting Tregs, and effectively confirmed that clusters 5 and 6 corre- these main states, on the basis of the expression of the defining genes 39 spond to the most naive Tregs, with signatures of a/eTregs defined by in Fig. 4a (circular a posteriori projection (CAP) plots) (Fig. 4c). 34 high Prdm1 (encoding BLIMP1) or resulting from homeostati- Most splenic Tregs projected cleanly in one of the main clusters, cally driven expansion19,35. For Ep300 (encoding p300) knockout36, but the best definition was observed for activated clusters 1–3, as the bias toward activated Tregs was consistent with the decreased expected from the gene distinctions of Fig. 4a and Supplementary lo proportion of activated CD62L Tregs in these mice. Interestingly, Fig. 4d. Only cluster 4 remained less sharply defined, appearing the transcriptional signatures of Tregs deficient in several transcrip- almost like a transitional group between the resting clusters 5 and tional cofactors that support Treg differentiation (Foxo1, Bach2 and 6 and the TCR-responsive cluster 3. These assignments were again 37,38 Gata3) were under-represented in cells from the activated clus- robust and applied effectively to independently derived Treg scRNA- ters, a result congruent with the distribution of Bach2 expression seq datasets (Supplementary Fig. 4e). We also mapped these clusters predominant in resting clusters (Fig. 4a). The third set of signatures back onto the Treg/Tconv tSNE plot from Fig. 2 (Fig. 4d). The resting 28 identified TCR-delivered signals, defined by direct engagement clusters mapped to the core Treg area but also included many furtive 19 or by acute gene ablation in vivo . These signatures clearly tagged Tregs. However, the activated clusters mapped to the more peripheral cells of cluster 3 as those with the strongest response to TCR signals, regions of overlap with Tconvs (areas A–C), thus indicating that Treg consistent with their expression of early-response genes. The Bcl6- activation leads to some convergence with Tconvs. dependent Tfr signature7 was most clearly enriched in cluster 1, Because this scRNA-seq analysis identified a main resting– as was a generic signature distinguishing Treg cells that localize to a activated axis as well as clear distinctions among phenotypes of variety of tissues and that we already knew to include a substantial activated Tregs, it was important to independently validate these component of cell activation. This juxtaposition suggests that Tfr observations. We applied statistical methods to correct for the low cells, which reside in the B cell areas of the spleen and lymph nodes, sensitivity of the scRNA-seq by calculating the technical dropout are nonetheless very similar to cells that reside in non-lymphoid tis- probability32,40 (Fig. 4e). When corrected counts were used, the sues and sites of sterile inflammation. scRNA-seq data predicted CXCR3 to be predominantly displayed

a TRA V12-3/J57 CALNQGGSAKLIF TRA V12/J17 CALREINSAGNKLTF TRA V6/J56 CALSERATGGNNKLTF TRA V13D/J40 CAPVNTGNYKYVF TRB V3/J1-2 CASSLVGRDSDYTF TRB V31/J1-6 CAWSLNRGNSPLY TRB V2/J1-2 CASSQDPSSDYTF TRB V19/J2-4 CASSPRLKNTLYF

TRA J57/V21 CGARNWGEQNTLYF TRA V14/J22 CAASKGSWQLIF TRA V6-5/J40 CALSGPLNTGNYKYVF b *** TRB V20/J2-4 CILSHQGGSAKLIF TRB V1/J1-4 CTCSADANSNERLFF TRB V15/J1-1 CASSLNRNTEVFF 1.00

0.50 Cell−cell correlation 0

tSNE 2 Same Different tSNE 1 clonotype clonotype

Fig. 6 | Tregs that express the same TCRα/β sequence have related transcriptional programs. a, Two-dimensional tSNE plot of Tregs from the colon lamina propria, in seven panels with single Treg cells with the same rearranged TCRα/β sequences highlighted in red. V/J alleles and CDR3 sequences are indicated above each panel. Data from two different experiments (top four and bottom three panels). b, Cell–cell correlation between colonic Treg cells that express the same rearranged TCRα/β (n = 14 cell–cell correlations) or unrelated TCRs (n = 1,000 correlations between any Treg in the data, randomly sampled). ***P < 10−4 (two-tailed t test). Box plot: center line, median; box limits, upper and lower quartiles. by CD62Llo activated T cells and predicted CXCR3 and NR4A1 to trajectory from low to high expression (Fig. 5b), reproducing the be largely mutually exclusive. These predictions were confirmed, main axis of Treg heterogeneity observed in Fig. 4. lo with expression of CXCR3 on CD62l Tregs, and the highest levels of The data (1,265 to 1,567 cells per bin) were processed and ana- – CXCR3 on NR4A1 Tregs (Fig. 4e). We observed a higher proportion lyzed in the probabilistic framework described for Fig. 4, and each of NR4A1+CXCR3lo cells than predicted from the scRNA-seq data, cell was assigned according to its probability of belonging to clusters perhaps owing to stability of the CXCR3 protein. 1–6 (Supplementary Fig. 6). Two main observations stemmed from We then interrogated these data to determine the frequency this analysis. First—and perhaps surprisingly, given that high TCR 5 and distribution of expression of Treg effector molecules (Fig. 4f). signals might be expected to promote an activated state—the pro- We estimated their frequencies of expression after dropout cor- portion of cells falling in the resting pool (i.e., mapping to clusters rection as in Fig. 4e, which were consistent with the frequencies 5 and 6) was essentially identical for all three bins of GFP expres- observed by intracellular flow cytometry analysis of splenic Tregs sion (Fig. 5c). These results were confirmed by flow cytometry on (data not shown). Effector molecule expression was not mutu- samples from other mice from this line, which showed an equivalent ally exclusive, and we identified coexpressed effector molecules distribution of CD62L across windows of GFP expression (Fig. 5c). (Fig. 4f). As expected from the above, Il2ra and Ctla4 were by far These data are consistent with prior observations that also found no 10 the most frequent inhibitory molecules. In contrast, there was par- relationship between Nr4a1-GFP activity and eTreg status . tial and reciprocally biased expression of Lrrc32 (encoding GARP, However, mirroring the biased representation of Nr4a1 tran- which presents processed TGF-β) and Tgfb1, on the one hand, ver- scripts among activated Treg clusters (Fig. 4), the distribution of sus Gzmb and Ebi3. Finally, Il10 and Areg were preferentially rep- activated Tregs varied markedly with Nr4a1-GFP intensity. The pro- resented in activated clusters but quite rare, consistent with their portion of cells in cluster 3 increased with GFP, while the proportion being deployed preferentially in tissue Tregs or at inflammatory sites. of cells in clusters 1 and 2 decreased (Fig. 5d and Supplementary Thus, splenic Tregs adopt a continuum of states gravitating around Fig. 6). These results were consistent with the distribution of Nr4a1 one resting and two activated states with distinct phenotypic and transcripts, which were highest in Tregs of cluster 3 in normal mice. In effector functions. addition, and also consistent with the negatively correlated expression of Nr4a1 and Cxcr3 (Fig. 4), we found an under-representation + med hi Role of TCR signaling in influencing Treg states. The variable of CXCR3 Tregs in the Nr4a1 and Nr4a1 pools (Fig. 5e). induction of the Nr4a1 gene module among activated Tregs suggested Thus, in physiological settings and in the absence of inflam- differences in how Tregs integrate TCR-derived signals. Because matory challenge, the level of TCR signals does not influence the signals from the TCR affect multiple levels of Treg differentiation, proportion of activated Tregs but markedly skews their phenotypic activation and suppressive function, we sought to further explore choices. the role of TCR signaling intensity in shaping Treg states. Nr4a1 is a transcription factor whose expression is induced rapidly upon Role of TCR specificity in influencing Treg phenotypes. We then T cell activation, in a manner proportional to TCR signaling inten- used another approach to investigate the relationship between TCR 28 gfp sity . We took advantage of Nr4a1 reporter mice, in which the signals and Treg subphenotypes. If they were connected, Treg cells level of GFP displayed by each cell reflects the intensity of the TCR- expressing an identical TCR clonotype (same V composition and 41 delivered signal , and performed scRNA-seq on Tregs sorted from rearranged CDR3 sequence) would be predicted to share a tran- three different zones of GFP expression (Fig. 5a). As expected, scriptional subphenotype. We adapted the InDrop protocol to obtain the frequency and levels of Nr4a1 conformed to the origin of the both the transcriptome and the TCRα/β variable region sequences Tregs (Fig. 5a). Tregs from different Nr4a1 expression bins formed a from the same single cells (detailed in Methods). We sequenced the

a b c T T Treg (n = 741) conv reg

Tconv (n = 1,008)

All cells

Treg

A 88% T T 55% conv reg T C conv 5% CD25 3% + CD127 Foxp3 Gated on CD4+ (mRNA) 7% cells 42% B tSNE2 tSNE2 tSNE1 tSNE1 Expression Low High d ef T * Treg signature Tconv signature reg

0 P < 10–6 P < 10–5 6 IL2RB 2 BACH2 FOXP3

) IL2RA BACH2 10

4 CTLA4

FUT7 PRDM1 P value (–log 6 BLIMP1 IL2RG SATB1 NEAT1 SATB1 IL10RA CCR7

Foxp3 expression (normalized counts ) 8 HLA-DRA

0 GBP5 tSNE 2 T T T T conv conv reg reg 0.5 1 23 tSNE1 (A) (outside A) (A) (outside A) Expression FC Treg (A) versus Treg (outside A) Low High

gh1 2 3 4 5 6 i 33% 31% 2% 9% 3% 21% Cluster : GNLY GZMK FOXP3 1

(A ) IL2RA 10 CCL5 GZMA

conv CCR2 ID2 3 2 CCR2 S100A4 S100A6 LGALS1 ICOS

1 ITGB1 CAP2

(C) versus T CXCR3 IL6ST DUSP2 conv NR4A1 SELL CCR7 EGR1 4 FC T EGR2 5 0.1 6 IKZF2 0.1 110 BACH2 CAP1 SATB1 FC Treg (C) versus Treg (A) CCR7 SELL 0 High Expression

GSE76598: Anti CD3/28 vs. unstimulated Tregs GSE15659: Activated CD45RA– (Fr.II) vs. other Foxp3+ CD4+ T cell fractions GSE76598: Memory (CD45RA–) vs. naive (CD45RA+) CD25high CD4+ T cells GSE15659: Resting (Fr.I) vs. other Foxp3+ CD4+ T cell fractions Minimum Maximum Relative expression

+ Fig. 7 | Similitudes in human and mouse Treg heterogeneity. a, Flow cytometric plot of gated CD4 Treg and Tconv cells in human blood, as defined by CD127 and CD25 expression, showing the sorting gates used before microfluidic encapsulation and scRNA-seq library construction. Representative of cells isolated from two different donors. b, Two-dimensional tSNE plot from transcriptomes of single human Treg (red, n = 741) and Tconv (black, n = 1,008) cells, with different areas (A–C) identified by unsupervised clustering. Each dot represents an individual cell. The proportions of Tregs and Tconvs in each area relative to the total numbers of Tregs and Tconvs are indicated. c, Same tSNE plots of human Treg and Tconv cells as in b, but split to show Treg and Tconv cells separately (top) and cells in which FOXP3 transcripts were detected (bottom; the size of each dot indicates the level of FOXP3 RNA expression. d, FOXP3 expression (normalized counts) in Tregs in area A (furtive Tregs) in comparison to Tregs outside area A (mean and 95% confidence interval). *P < 0.05 (two-tailed t test) (n = 54 Tregs in A and 52 Tregs outside A, all expressing FOXP3 mRNA). e, Volcano plot comparing the gene expression profiles of human furtive Tregs (in area A) versus all other Tregs (outside area A). Human up- and downregulated Treg-signature genes are highlighted (red and blue, 2 respectively). χ test P values. f, tSNE plots as in b, highlighting BACH2- and SATB1-expressing Treg cells (red). Dot size, level of expression in each cell. g, Fold change/fold change plot comparing the gene expression profiles of Tregs in C and A areas (x axis) with the gene expression profiles of Tconvs in C and A areas (y axis). Highlighted genes are those specifically expressed by both Tregs and Tconvs in area C. h, Biclustering expression heat map of human blood Tregs, grouped into the equivalent of the six mouse clusters (defined in Fig. 4) by using a likelihood model (Methods) based on expression of human orthologs of the mouse genes (top). Biclustering heat map of published activated/resting signatures8,42 (bottom). Each vertical line represents one cell.

The proportions of cells in each cluster are indicated. i, Circular projection plot of the likelihood of each single human Treg cell belonging to any of the six clusters defined by homology to the mouse Treg clusters defined in Fig. 4.

+ TCRs of FoxP3-GFP Treg cells from the spleen and colon, the latter difference was not due to the tissue of origin (mouse blood Tregs have because microbiota-driven differentiation and/or expansion of par- the same Nr4a1-GFP profile as splenic Tregs; Supplementary Fig. 7f) 17 ticular clones leads to a higher frequency of repeated clonotypes . but may be due to different environmental influences on Treg activa- Of the 32 and 138 Tregs from the spleen and colon for which both tion in humans and mice. With this exception, however, Treg hetero- TCRα and TCRβ chains could be determined unambiguously, none geneity appears to follow very similar lines in mouse and human was repeated in splenic Tregs, as expected, because the repertoire of immune systems. splenic Treg is very diverse. However, seven clonotypes (defined by identical TCRα and TCRβ nucleotide sequences) were found in Discussion two or more of the colonic Treg cells. With two exceptions, these These results provide an unprecedented perspective on Treg diver- clonotype pairs belonged to cells that mapped close to each other sity, and on what exactly defines a Treg cell, with two overarching on the tSNE plot (Fig. 6a). In order to evaluate the importance of considerations. First, Tregs are generally distinguishable from Tconvs this apparent proximity, we calculated pairwise cell distances (as within the CD4+ T cell space, but the two populations are multiply Pearson correlation) between the transcriptomes of these shared- imbricated, with an important and previously unrecognized degree clonotype cells or between randomly sampled pairs of Tregs. Indeed, of overlap, whether among naive cells or their more differentiated the transcriptomes of Tregs that shared their TCRs were more closely progeny. Second, signals from the TCR play unanticipated roles in −4 correlated than by chance (P < 10 ), thereby confirming the signifi- shaping the different facets that activated Tregs can adopt. This study cance of the observation (Fig. 6b). Thus, Treg cells that display the provides a unifying framework for diverse observations made on same TCR clonotype tend to share transcriptional identity, in keep- Tregs in the context of immune homeostasis. ing with the conclusion that signals from the TCR can mold the Although scRNA-seq lacks the sensitivity to measure the whole subphenotype adopted by activated Tregs. transcriptome of a single cell, we leveraged the power of performing scRNA-seq on thousands of single cells by using droplet microflu-

Human and mouse Tregs show similar patterns of heterogeneity. idics and applied statistical methods to take into account the tech- Finally, we investigated whether the conclusions reached about nical dropout probability32,40. Thus, conclusions were drawn not mouse splenic Tregs might also apply in human cells. Blood Tregs from a single cell but from several single cells with closely related + high low + – and Tconvs were sorted (CD4 CD25 CD127 and CD4 CD25 programs, residing in the same area of the multidimensional space. CD127high, respectively; Fig. 7a) from two healthy donors, and their With small aggregates of single cells, both technical noise and gene transcriptomes were examined by scRNA-seq (Supplementary expression stochasticity are mitigated. This approach identifies dif-

Fig. 7a–d). Aspects of the heterogeneity and inter-relationships ferent cell clusters and estimates the proportion of Tregs using differ- between Treg and Tconv cells, and of the genes underlying them, proved ent suppression strategies. strikingly similar to those in mice in several important respects. Treg and Tconv cells have diametrically opposite functions, thus First, the relative disposition of human CD4+ T cells from a tSNE suggesting sharp boundaries to their transcriptional programs, representation (Fig. 7b,c) was similar to that in the mouse popula- a notion consistent with the hundreds of transcripts previously 12,31 tions. Most Tregs and Tconvs clustered apart, but again furtive Tregs tres- found to be differently expressed by population transcriptomics . passed into the Tconv zone of the plot (Fig. 7b), actually in even higher However, revisiting this notion at the single-cell level showed that proportions (55%) than observed for mouse splenocytes (26%). many of these transcripts were not ubiquitously expressed in Tregs. These furtive Tregs again expressed FOXP3 (Fig. 7c), albeit at lower In fact, only a small core set of transcripts was uniformly expressed levels than were observed in other Tregs (Fig. 7d), and Treg signature throughout all Tregs, and consistently missing or under-represented transcripts were shifted in furtive Tregs compared with other Tregs in Tconvs, and as such may truly define Treg identity. Some of these (Fig. 7e). Particularly striking was the expression of the two Treg reg- transcripts are involved in essential Treg pathways, such as Il2ra, ulators BACH2 and SATB1, which were mostly expressed in furtive which anchors the IL-2 paracrine loop (made by Tconvs and used in 43 Tregs (Fig. 7f). This over-representation of BACH2 and SATB1 was Tregs ); other transcripts encode determining transcription factors also observed in mouse cells, albeit less strikingly. Another region (Foxp3 and Ikzf2) or several members of the TNF receptor family 44 of the tSNE spread exhibited tight comingling of Tregs and Tconvs (area (Tnfrsf4 and Tnfrsf18) . However, these core Treg transcripts also 45 C in Fig. 7b), reminiscent of areas A–C in mice. Treg and Tconv cells include some whose function remains conjectural (Gpr83) or have in these areas still maintained their core transcriptional identity yet to be explored (Izumo1r, Capg or Wdr92). Conversely, Tregs con- (FOXP3, IL2RA, IKZF2 and CTLA4 for Tregs, and THEMIS, ID2 sistently lacked expression of Igfbp4 and Dapl1. and IL7R for Tconvs) but shared common transcriptional characteris- Beyond the core transcripts, Tregs and Tconvs appeared much more tics, uniquely expressing GZMA, GZMK, GNLY and CCR2 (Fig. 7g imbricated than would be suggested by the comfortable dichot- and Supplementary Fig. 7e). omy of GFP profiles in Foxp3gfp mice. The overall transcriptomes

Second, orthologs of the genes that clustered mouse Tregs (Fig. 4c) of furtive Tregs were very similar to those of the main Tconv popula- separated human Tregs, and we were able to project human Tregs to the tion, while still maintaining their core identity (Foxp3 expression as mouse Treg clusters (Fig. 7h,i). The proportions of human blood Tregs RNA and protein). These cells formed a substantial portion of Tregs in each cluster differed from those in the mouse splenic Tregs, with (>20%) in both mouse spleen and human blood. They seemed to be a higher presentation of the activated clusters. CCR7, SELL, SATB1 ‘weaker’ Tregs, because they showed lower expression of Foxp3- and and BACH2 again characterized cells belonging to the resting Treg Treg-signature genes, and strongly expressed Satb1 and Bach2, two clusters (4–6), whereas IKZF2 was high in cluster 3, and S100A4, repressors of effector functions37,46. They probably correspond to 8 S100A6 and ITGB1 were frequent in activated clusters 1 and 2. the poorly suppressive resting Tregs previously described . Because of lo hi Integrating these data with the existing observations on human Treg their proportions and CD44 CD62L naive phenotype, it is unlikely heterogeneity, we analyzed the expression of published signature that furtive Tregs represent activated Tconvs that transiently express sets8,42 in our single cells (similarly to analysis in Fig. 4b). The sets FoxP3 upon activation13,14 or, at the opposite end of the spectrum, distinguishing activated/memory and resting/naive Tregs were differ- Tregs in the process of losing FoxP3, and become activated effector ently expressed in clusters 1 and 2 and 4–6, thus highlighting the T cells47. One limitation to these speculations is that there is no rig- conservation of activation signals between human and mouse Tregs. orous way to infer precursor–product relationships from these data. In a striking departure from our observations for mouse splenic Also blurring the demarcation was the observation that some Treg Tregs, however, very few human blood Tregs expressed the set of tran- and Tconv cells converged to similar transcriptome locations. These scripts dependent on TCR signaling (NR4A1, EGR1 or EGR2). This cells were activated (by CD44/CD62L criteria), and they expressed

specific gene sets overriding the core Treg/Tconv transcriptional differ- Tregs that share the same clonotype may result from similar qual- ence. This confluence is reminiscent of findings in earlier reports ity and intensity of TCR engagement by MHC-peptide ligands. that transcription factors characteristic of terminally differentiated Alternatively, the shared antigenic specificity may drive these Tregs to Tconv states also have a critical role in Treg suppression (Tbx21, Rorc the same anatomical location, where they integrate the same envi- 3,48 and Bcl6) . In one of those convergences, Tregs and Tconvs expressed ronmental cues (such as MHC-peptide, cytokines and metabolic transcripts known to characterize follicular T cells (in Tfh or Tfr constraints). Lastly, this phenomenon may reflect influences that 7 cells) . It is possible that these Tregs and Tconvs converge transcription- are received early during Treg differentiation or peripheral activa- ally as a consequence of residing in the same anatomical areas of the tion and imprint specific programs that persist after emigration to spleen, where they receive the same environmental cues7,45. peripheral organs.

These results are therefore consistent with a model defining In comparing human and mouse Tregs at the single-cell level, we 12 Treg identity as modular . All Tregs express a very small core of Treg- observed many similarities. In both species, Treg and Tconv cells were specific transcripts, onto which additional programs are added, as distinguishable by the ubiquitous expression of a small core set of a function of differentiation and/or location. More generally, our transcripts (FOXP3, IKZF2, TNFRSF1B, IL2RA and IL2RB). Furtive unbiased analysis of Treg heterogeneity revealed that Tregs form a con- Tregs were found in both species, more frequently in humans, thus – lo tinuum gravitating around three major poles. One axis of variation providing a likely explanation for why CD45RA FOXP3 Tregs in encompasses the previously known resting/activated difference, humans have very low suppressive activity8. In humans, as in mice, 8–11 reflected by CD62L or CCR7 (refs ). Two transcription factors, some activated Tregs and Tconvs co-expressed similar transcriptional 30 Bach2 and Satb1, which modulate Treg differentiation , were com- programs that overrode the canonical Treg/Tconv difference. Activated paratively overexpressed in resting Tregs, and may help maintain Tregs adopted diverse phenotypes in both species, albeit with the this state; interestingly, they seemed to be reciprocally expressed in striking difference that the population with evidence of high TCR resting clusters 5 and 6, thus potentially suggesting redundant func- signals was essentially missing in human blood. tions. However, most of the phenotypic divergence was revealed In conclusion, this study provides a general model for lymphoid within the activated Treg compartment, with two orthogonal states Treg identity and heterogeneity that is largely conserved between of activation associated with differences in TCR signaling strength humans and mice. It also sheds light on a new and unexpected and suppressor-molecule expression. mechanism for the TCR in setting Treg fates. Analysis of complex The single-cell data provided us with the opportunity to corre- T cell populations through scRNA-seq provides unique opportuni- late TCR sequence and signaling intensity with the transcriptional ties to identify TCR sequences and other mechanisms modulating T output of each Treg. We expected that the activated Treg phenotype cell fate in health and disease, and manipulating these states should would be driven by the intensity of TCR signaling. Unexpectedly, ultimately open avenues to restore homeostasis. the TCR signal intensity, as indicated by expression of Nr4a1, and confirmed across windows of activity of the Nr4a1gfp reporter, did Methods not correlate with the naive/activated Treg ratio (reflected by the fre- Methods, including statements of data availability and any asso- quency of cluster 5 and 6 Tregs, or of Satb1 and Bach2 expression). ciated accession codes and references, are available at https://doi. This observation is consistent with the even distribution of CCR7+ org/10.1038/s41590-018-0051-0. gfp 10 Tregs across Nr4a1 that was previously reported , but it appears at odds with findings from studies showing that activated Tregs elec- Received: 23 June 2017; Accepted: 17 January 2018; 19 tively disappear upon TCR ablation . This discrepancy may merely Published online: 12 February 2018 reflect the difference between complete ablation and quantitative variation in transmitted signals. The puzzling disconnect between References TCR-derived signals and activated status may also reflect different 1. Josefowicz, S. Z., Lu, L. F. & Rudensky, A. Y. Regulatory T cells: mechanisms time scales: Nr4a1 is an early-response gene and the reporter prob- of diferentiation and function. Annu. Rev. Immunol. 30, 531–564 (2012). ably responds to short-term stimuli, whereas the aTreg phenotype 2. Panduro, M., Benoist, C. & Mathis, D. Tissue Tregs. Annu. Rev. Immunol. 34, may integrate cues over a longer timeframe. 609–633 (2016). However, TCR signaling intensity seemed to have a strong influ- 3. Campbell, D. J. & Koch, M. A. Phenotypical and functional specialization of FOXP3+ regulatory T cells. Nat. Rev. Immunol. 11, 119–130 (2011). ence shaping the different forms of activated Treg differentiation, and 4. Luo, C. T. & Li, M. O. Transcriptional control of regulatory T cell hence Treg effector functions. Tregs in cluster 3, with the signs of the development and function. Trends. Immunol. 34, 531–539 (2013). strongest TCR signals, preferentially expressed Cd24a and Tgfb1. 5. Vignali, D. A., Collison, L. W. & Workman, C. J. How regulatory T cells Perhaps paradoxically, the cluster with the most strongly activated work. Nat. Rev. Immunol. 8, 523–532 (2008). phenotype (cluster 1), which most actively expressed migratory 6. Koch, M. A. et al. Te transcription factor T-bet controls regulatory T cell homeostasis and function during type 1 infammation. Nat. Immunol. 10, molecules (Itgb1, Itgae and Ccr2), had the lowest expression of the 595–602 (2009). TCR-response (Nr4a1, Egr1 and Egr2) cluster. It is possible that those 7. Linterman, M. A. et al. Foxp3+ follicular regulatory T cells control the Tregs circulate to (or from) tissues where they encounter cognate anti- germinal center response. Nat. Med. 17, 975–982 (2011). 8. Miyara, M. et al. Functional delineation and diferentiation dynamics of gen. Alternatively, it is known that TCR engagement triggers several + 49 human CD4 T cells expressing the FoxP3 transcription factor. Immunity 30, signaling cascades and that Nr4a1 induction is not equally depen- 899–911 (2009). dent on all such signaling cascades (for example, independent of 9. Kleinewietfeld, M. et al. CCR6 expression defnes regulatory efector/ NF-AT)50. Thus, low Nr4a1, Egr1 and Egr2 induction may reflect a memory-like cells within the CD25+CD4+ T-cell subset. Blood 105, particular balance of TCR signaling paths, in cluster 1 and 2 Tregs. In 2877–2886 (2005). addition, we cannot rule out that the different T clusters indicate 10. Smigiel, K. S. et al. CCR7 provides localized access to IL-2 and defnes reg homeostatically distinct regulatory T cell subsets. J. Exp. Med. 211, different time periods of activated Treg differentiation, and that TCR 121–136 (2014). signals are dampened by negative feedback in Tregs of clusters 1 and 2. 11. Cretney, E., Kallies, A. & Nutt, S. L. Diferentiation and function of Foxp3+ In keeping with the notion that the TCR is a critical component efector regulatory T cells. Trends. Immunol. 34, 74–80 (2013). 12. Hill, J. A. et al. Foxp3 transcription-factor-dependent and -independent shaping Treg fate, Tregs that share the same TCR were more transcrip- tionally similar than T as a whole. These similarities did not apply regulation of the regulatory T cell transcriptional signature. Immunity 27, regs 786–800 (2007). to all clonotypes, in keeping with the observation that Tregs in TCR 13. Gavin, M. A. et al Single-cell analysis of normal and FOXP3-mutant human transgenic mice can occupy different organismal and phenotypic T cells: FOXP3 expression without regulatory T cell development. Proc. Natl niches (ref. 17 and data not shown). These similar fine programs in Acad. Sci. USA 103, 6659–6664 (2006).

14. Miyao, T. et al. Plasticity of Foxp3+ T cells refects promiscuous Foxp3 39. Jaitin, D. A. et al. Massively parallel single-cell RNA-seq for marker-free expression in conventional T cells but not reprogramming of regulatory T decomposition of tissues into cell types. Science 343, 776–779 (2014). cells. Immunity 36, 262–275 (2012). 40. Shalek, A. K. et al. Single-cell RNA-seq reveals dynamic paracrine control of 15. Rubtsov, Y. P. et al. Stability of the regulatory T cell lineage in vivo. Science cellular variation. Nature 510, 363–369 (2014). 329, 1667–1671 (2010). 41. Moran, A. E. et al. T cell receptor signal strength in Treg and iNKT cell 16. Zemmour, D., Pratama, A., Loughhead, S. M., Mathis, D. & Benoist, C. Flicr, development demonstrated by a novel fuorescent reporter mouse. J. Exp. a long noncoding RNA, modulates Foxp3 expression and autoimmunity. Proc. Med. 208, 1279–1289 (2011). Natl Acad. Sci. USA 114, E3472–E3480 (2017). 42. Pesenacker, A. M. et al. A regulatory T-cell gene signature is a specifc and 17. Lathrop, S. K. et al. Peripheral education of the immune system by colonic sensitive biomarker to identify children with new-onset type 1 diabetes. commensal microbiota. Nature 478, 250–254 (2011). Diabetes 65, 1031–1039 (2016). 18. Li, M. O. & Rudensky, A. Y. T cell receptor signalling in the control of 43. Fontenot, J. D., Rasmussen, J. P., Gavin, M. A. & Rudensky, A. Y. A function regulatory T cell diferentiation and function. Nat. Rev. Immunol. 16, for interleukin 2 in Foxp3-expressing regulatory T cells. Nat. Immunol. 6, 220–233 (2016). 1142–1151 (2005). 19. Levine, A. G., Arvey, A., Jin, W. & Rudensky, A. Y. Continuous 44. Mahmud, S. A. et al. Costimulation via the tumor-necrosis factor receptor requirement for the TCR in regulatory T cell function. Nat. Immunol. 15, superfamily couples TCR signal strength to the thymic diferentiation of 1070–1078 (2014). regulatory T cells. Nat. Immunol. 15, 473–481 (2014). 20. Hashimshony, T., Wagner, F., Sher, N. & Yanai, I. CEL-Seq: single-cell 45. Liu, Z. et al. Immune homeostasis enforced by co-localized efector and RNA-Seq by multiplexed linear amplifcation. Cell Rep. 2, 666–673 (2012). regulatory T cells. Nature 528, 225–230 (2015). 21. Klein, A. M. et al. Droplet barcoding for single-cell transcriptomics applied to 46. Beyer, M. et al. Repression of the genome organizer SATB1 in regulatory embryonic stem cells. Cell 161, 1187–1201 (2015). T cells is required for suppressive function and inhibition of efector 22. Wagner, A., Regev, A. & Yosef, N. Revealing the vectors of cellular identity diferentiation. Nat. Immunol. 12, 898–907 (2011). with single-cell genomics. Nat. Biotechnol. 34, 1145–1160 (2016). 47. Zhou, X. et al. Instability of the transcription factor Foxp3 leads to the 23. Zilionis, R. et al. Single-cell barcoding and sequencing using droplet generation of pathogenic memory T cells in vivo. Nat. Immunol. 10, microfuidics. Nat. Protoc. 12, 44–73 (2017). 1000–1007 (2009). 24. Sawant, D. V. & Vignali, D. A. Once a Treg, always a Treg? Immunol. Rev. 48. Sefk, E. et al. Individual intestinal symbionts induce a distinct population of 259, 173–191 (2014). RORγ+ regulatory T cells. Science 349, 993–997 (2015). 25. van der Maaten, L. & Hinton, G. Visualizing high-dimensional data using 49. Malissen, B. & Bongrand, P. Early T cell activation: integrating t-SNE. J. Mach. Learn. Res. 9, 2579–2605 (2008). biochemical, structural, and biophysical cues. Annu. Rev. Immunol. 33, 26. Kim, H. J. et al. Stable inhibitory activity of regulatory T cells requires the 539–561 (2015). transcription factor Helios. Science 350, 334–339 (2015). 50. Ashouri, J. F. & Weiss, A. Endogenous Nur77 is a specifc indicator of antigen 27. Sage, P. T. & Sharpe, A. H. T follicular regulatory cells. Immunol. Rev. 271, receptor signaling in human T and B cells. J. Immunol. 198, 657–668 (2017). 246–259 (2016). 53. Ferraro, A. et al Interindividual variation in human T regulatory cells. Proc. 28. Wakamatsu, E., Mathis, D. & Benoist, C. Convergent and divergent efects of Natl Acad. Sci. USA 111, E1111–E1120 (2014). costimulatory molecules in conventional and regulatory CD4+ T cells. Proc. Natl. Acad. Sci. USA 110, 1023–1028 (2013). Acknowledgements 29. Samstein, R. M. et al. Foxp3 exploits a pre-existent enhancer landscape for We thank E. Sefik for help with colon profiling and K. Hattori, C. Araneo K. Waraska, regulatory T cell lineage specifcation. Cell 151, 153–166 (2012). M. Thorsen and R. Steen for help with mice, profiling, sorting and sequencing. This work 30. Kitagawa, Y. et al. Guidance of regulatory T cell development by Satb1- was supported by NIH grants AI051530, AI116834 and AI125603 to C.B./D.M. D.Z. and dependent super-enhancer establishment. Nat. Immunol. 18, 173–183 (2017). E.K. were supported by a PhD fellowship from the Boehringer Ingelheim Fonds. 31. Gavin, M. A., Clarke, S. R., Negrou, E., Gallegos, A. & Rudensky, A. Homeostasis and anergy of CD4+CD25+ suppressor T cells in vivo. Nat. Immunol. 3, 33–41 (2002). Author contributions 32. Meredith, M., Zemmour, D., Mathis, D. & Benoist, C. Aire controls gene D.Z., E.K. and R.Z. performed the experiments; D.Z., A.M.K., C.B. and D.M. expression in the thymic epithelium with ordered stochasticity. Nat. Immunol. designed the study, and analyzed and interpreted the data; D.Z., D.M. and C.B. 16, 942–949 (2015). wrote the manuscript. 33. Hennig, C. Cluster-wise assessment of cluster stability. Comput. Stat. Data. Anal. 52, 258–271 (2007). Competing interests 34. Cretney, E. et al. Te transcription factors Blimp-1 and IRF4 jointly control The authors declare no competing interests. the diferentiation and function of efector regulatory T cells. Nat. Immunol. 12, 304–311 (2011). 35. Feng, Y. et al. A mechanism for expansion of regulatory T-cell repertoire and Additional information its role in self-tolerance. Nature 528, 132–136 (2015). Supplementary information is available for this paper at https://doi.org/10.1038/ 36. Liu, Y. et al. Inhibition of p300 impairs Foxp3+ T regulatory cell function and s41590-018-0051-0. promotes antitumor immunity. Nat. Med. 19, 1173–1177 (2013). Reprints and permissions information is available at www.nature.com/reprints. 37. Roychoudhuri, R. et al. BACH2 represses efector programs to stabilize T(reg)-mediated immune homeostasis. Nature 498, 506–510 (2013). Correspondence and requests for materials should be addressed to D.M. or C.B. 38. Ouyang, W. et al. Novel Foxo1-dependent transcriptional programs control Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in T(reg) cell function. Nature 491, 554–559 (2012). published maps and institutional affiliations.

Methods After purification of the DNA/RNA duplex with 1.2× AMPure beads (Beckman), Mice. C57BL/6 J mice were obtained from Te Jackson Laboratory. Foxp3ires-gfp/B6 second-strand cDNA synthesis was performed (NEB) for 2.5 h at 16 °C. The library (ref. 51) mice (Foxp3gfp in the main text) were maintained in our colony. All was then amplified through T7 in vitro transcription for 15 h at 37 °C (HiScribe T7 experimentation was performed following the animal protocol guidelines of High Yield RNA Synthesis Kit, NEB). After purification, half of the amplified RNA Harvard Medical School (reviewed and approved HMS IACUC protocol 02954). was used for further processing. First, the aRNA was fragmented for 3 min at 70 °C Foxp3ires-thy1.1 (ref. 52) mice (Foxp3thy1.1 in the main text) were crossed with the (magnesium RNA fragmentation kit, Ambion) and purified using AMPure Beads Nur77gfp reporter mice41. (1.2×). Then reverse transcription with random hexamers was done for 1 h at 42 °C (PrimeScript Reverse Transcriptase, Takara Clontech). A final PCR was done to Human blood. Blood was obtained in the context of the PhenoGenetic amplify the library and to add the P5–P7 and Illumina index primers (Kapa 2× study, reviewed and approved by the Institutional Review Board at Harvard HiFi HotStart PCR mix, Kapa Biosystems). The number of cycles was chosen Medical School (protocol IRB15-0504). The two healthy donors, one male and one by first running a qPCR on one-twentieth of the RT reaction. The optimum was female) gave their written informed consent. between 11 and 14 cycles. Library size was analyzed with a Tapestation (High Sensitivity D1000 Flow cytometry. Blood was collected by cardiac puncture from euthanized ScreenTape, Agilent Technologies), quantified by qPCR and sequenced using mice, and coagulation was prevented by the addition of heparin (20–50 U per ml NextSeq 500 and custom primers (read 1, 40 bp; index, 7 bp; read 2, 51 bp). blood) (heparin sodium salt, Sigma-Aldrich H3393). A single-cell suspension Data processing and normalization (InDrop). InDrop sequencing data were of mouse splenocytes was obtained in parallel after physical dissociation with a processed as described by Zilionis et al.23. Read 1 contained transcript information. 40-μm mesh (Falcon). Lysis of red blood cells was performed with 500 μl of ACK Read 2 contained the Unique Molecular Identifier (UMI) (6 bp), single-cell (Lonza) for 2 min on ice. Antibody staining was done in ice-cold buffer (DMEM barcode (16–19 bp) and a conserved sequenced named W1. Fastq reads were first without phenol red, 2% FBS) for 5 min at a dilution of 1/100 with antibodies to filtered for quality (>80% with Sanger Q >20) and on the expected structure of CD4 (GK1.5; BioLegend), CD25 (PC61; BioLegend), TCRβ (H57-597; BioLegend), the reads: paired reads were kept only if read 2 contained the W1 and poly(T) CXCR3 (clone SA011F11, BioLegend) and CD62L (MEL-14; BioLegend). Tregs were sequence and if read 1 did not contain them. The UMI sequence was added to isolated as CD4+TCRβ+ Foxp3-GFP+, CD4+TCRβ+ Foxp3-Thy1.1+ (for the Nur77 the read ID of read1 at the fourth position (positions separated by ':'). Single-cell + + – experiments). Tconvs were isolated as CD4 TCRβ Foxp3-GFP . demultiplexing was performed against the possible barcode space, and only reads 48 + + + The protocol of Sefik et al. was used to isolate CD4 TCRβ FoxP3-GFP Tregs mapping unambiguously and with fewer than two mismatches were kept. For from the colonic lamina propria. Colonic tissue was dissociated using the following each single-cell library, the reads were then mapped against the mouse mm10 steps. Epithelial cells were removed by treating the tissue with RPMI containing transcriptome (complemented with the sequence of the Foxp3-cd90.1 or Foxp3-gfp 1 mM DTT, 20 mM EDTA and 2% FBS at 37 °C for 15 min, then mincing and transgenes) using Tophat2 v2.0.10 (--library-type fr-firststrand)54. Reads mapping dissociating in a collagenase solution (1.5 mg/ml collagenase II (Gibco), 0.5 mg/ml to multiple regions or having a low alignment score (MAPQ <10) were filtered dispase and 1% FBS in RPMI) with constant stirring at 37 °C for 45 min. Single-cell out (flag 256). Duplicated reads were filtered as follows, using UMIs and custom suspensions were then filtered and washed with 4% RPMI solution. CD45+ cells scripts (Zemmour_Code/CorrectAndCountUMIsEditDist): for each set of reads were enriched before sorting using microbeads (Miltenyi Biotec). mapping to one gene, we kept all reads with UMIs with a pairwise string distance Human Tregs and Tconvs were isolated from peripheral blood mononuclear cells ≥2 (Levenshtein distance for taking into account indels and substitution). A final as described by Ferraro et al.53. An equal volume of room-temperature PBS/2 mM gene table with genes in rows and cells in columns was then saved. A CellDataset EDTA was mixed into 15 ml of blood and carefully layered over 14 ml Ficoll- object from the Monocle package was used to store the data (Zemmour_Code/ Hypaque solution (GE Healthcare). After centrifugation for 30 min at 900 g (with Zemmour_Code.Rmd: **Make a CellDataset object (from Monocle)**). no break), at room temperature, the mononuclear cell layer was washed three times A first quality control (QC) was performed after demultiplexing by calculating with excess HBSS (Gibco) (10 min at 400 g) and resuspended in 2 ml of HBSS. An the proportion of total reads that mapped to the most abundant single-cell FBS gradient was then used to remove platelets and debris by layering the 2-ml cell barcodes (Zemmour_Code/Zemmour_Code.Rmd: **QC Plots**). Good libraries suspension over 8 ml FBS and centrifuging for 10 min at 300 g at room temperature. had >70% of reads mapping to a small set of single-cell barcodes (500–3,000), The pellet was resuspended in FACS buffer (phenol red–free DMEM, 2% FBS, as expected from successful encapsulation. A second quality control involved 0.1% azide and 10 mM HEPES, pH 7.9) for antibody staining in ice for 10 min the distribution of mapping rates and sequencing coverage. Good libraries were with the following antibodies: anti-CD4 (clone RPA-T4; BioLegend), anti-CD25 normally distributed with an average 80% mapping rate (5% s.d.) and >90% UMI (clone BC96; Biolegend), anti-CD127 (clone A019D5; BioLegend) and FcBlock duplicated reads (indicating sequencing saturation). – + + low (homemade). Tregs and Tconvs were sorted as DAPI CD4 CD25 CD127 and Single cells with more than 500 genes were kept for the analysis. Total UMI – + – high DAPI CD4 CD25 CD127 , respectively. count normalization was performed and scaled to the median UMI count (Zemmour_Code/Zemmour_Code.Rmd: **Normalization and filtering of genes Single-cell RNA-seq (InDrop). Library construction. scRNA-seq was done and cells**). by following the InDrop protocol described at length by Zilionis et al.23. Tis technique sequences poly(A) mRNAs from thousands of single cells in a timely Single-cell RNA-seq (CEL-seq). scRNA-seq using the CEL-seq protocol was done (<30 min) and highly efcient manner (>70–80% of the input cells can be 32 + + sequenced). Unique molecular identifers (UMIs) are used in order to minimize following the protocol described at length in ref. . Briefly, Splenic Treg (CD4 TCRβ FoxP3-GFP+) and T (CD4+TCR + FoxP3-GFP–) cells were index-sorted with a the noise introduced during amplifcation. Only 3′ ends are sequenced, keeping conv β strand information. BD FACSAria II instrument in 96-well hard-shell PCR plates filled with 4.4 μl of lysis buffer (HSP963, Bio-Rad) containing 0.125 µl of RNAseOut (40 U per 40,000 Tregs or Tconvs were first sorted by flow cytometry in complete medium (RPMI with 10% FBS). Just before encapsulation, cells were spun for 5 min at 4 °C 1 µl stock; 10777-019, Invitrogen), 0.25 µl of RT primer (25 ng per 1 µl stock) and and 500 g and resuspended in PBS with 15% OptiPrep Density Gradient Medium 4 µl of RNase-free water (AM9932; Ambion). Index-sorting allowed us to record, (OptiPrep; Sigma-Aldrich) at a concentration of 80,000 cells/ml. This density for each sorted single cell, the fluorescence intensity of each flow cytometry marker gradient prevents sedimentation during encapsulation and allows a uniform used. After cell sorting, the plates were quickly covered with an aluminum seal flow of cells to enter the microfluidic device. Encapsulation is done by flowing (AlumaSeal96; F96100; Excel Scientific), then vortexed for 10 s, centrifuged for four different inputs in the device: cells, reverse transcriptase buffer (RT buffer), 1 min at maximum speed (>2,250 g at 4 °C), frozen on dry ice and kept at −80 °C primer hydrogels and oil (water in oil droplets). Primer hydrogels contain reverse for up to 3 weeks. RNA was then denatured for 3 min at 70 °C and reverse transcription (RT) primers with a unique single-cell barcode. From 5′ to 3′, the transcribed using ArrayScript Reverse Transcriptase (AM2048 Ambion) and an primer contains a T7 promoter for amplification (in vitro transcription), the PE1 mRNA Second Strand Synthesis Module (E6111L; NEBNext). Single-cell cDNA sequencing primer, a unique single-cell barcode (16–19 bp) (147,456 possible libraries containing different barcodes were pooled into one tube, and in vitro barcodes), a UMI (6 bp) and a poly(T) sequence. transcription was done using a MEGAshortscript T7 transcription kit (AM1354; Around 5,000 single cells were then encapsulated in droplets of 3–4 nl Ambion) for 14 h. RNA was then fragmented, ligated to the 3′ Illumina adaptor containing a primer hydrogel bead and the RT buffer (SuperScriptIII, Invitrogen). and reverse transcribed. Libraries were amplified by PCR and sequenced on a Encapsulation was performed in <30 min in order to maintain cell viability, and HiSeq 2500 instrument. 32 the emulsion was collected in a 500-μl tube. Data processing was as described in length in ref. . It followed similar steps to RT was performed immediately after encapsulation. First, the primers those in InDrop, filtering reads for quality, mapping to the mm10 transcriptome were released from the gels by exposing the emulsion to UV for 7 min. RT was and filtering duplicated reads out using UMIs. done at 50 °C for 2 h and was followed by enzyme inactivation for 15 min at 70 °C. Emulsions were then split into small aliquots to ensure no more than Single-cell RNA-seq (10x Genomics). 7,000 splenic CD4+ cells from C57/Bl6 mice 3,000 cells per tube, and were broken down by addition of 20% 1H,1H,2H,2H- were sorted by flow cytometry into a well of a chilled 96-well plate filled with 14 µl perfluorooctanol (PFO) in HFE-7500 oil (3 M Novec 7500 Engineered Fluid, PBS with 0.06% BSA. Cells were then encapsulated in one lane of a 10x Chromium Novec) and kept at –80 °C. instrument, and libraries were constructed with a Single Cell 3′ Reagent Kit (V2 Samples were thawed on ice and spun at 4 °C for 5 min at 19,000 g to pellet cell chemistry) (https://support.10xgenomics.com/single-cell-gene-expression/library- debris. Excess primers and hydrogel beads were removed by filtration through a prep/). Libraries were sequenced on the NextSeq 500 platform (26/8/0/98, nucleic acid purification column and enzyme digestion (ExoI and HinfI). Read1/i7/i5/Read2). Data were analyzed using the Cell Ranger 1.0 Pipeline

(https://support.10xgenomics.com/single-cell-gene-expression/software/overview/ These probabilities were visualized as described by Jaitin et al.39, using circular welcome/). a posteriori (CAP) plots. The probabilities were projected on a two-dimensional sphere, using radial positions reflecting the relationship between clusters (a = (pi/2, Algorithm to find contaminant cells. Contamination cells/doublets were pi/4, 3 × pi/4, –3 × pi/4, –pi/4, –pi/2) for clusters 1 to 6). For each single cell c and identified in the gene table, using the following criteria. First, we compared each cluster k, the x and y coordinates were then calculated as single-cell transcriptome to the Immgen dataset (http://www.immgen.org/) and calculated the likelihood of each single cell being any of the Immgen cell types xp=×cos(a ) c ∑ ck k (Zemmour_Code/Zemmour_Code.Rmd: **Immgen cell type prediction**). k The Immgen matrix of gene expression was used to provide prior probabilities (probability to express gene i in cell type j = pij), and we calculated for each single cell c the likelihood of being of cell type j (Lcj) (multinomial model). Log posterior yp=×sin(a ) c ∑ ck k probabilities were derived by normalizing so that they summed to 1 for each k single cell.

Lc=×logp() cj ∑ i ij Treg signatures and single-cell score. Gene signatures were curated from published i datasets (references in Supplementary Fig. 4) and computed by comparison between two conditions (for example, WT versus KO). Data were downloaded Contaminant cells were flagged when a T cell type was not amid the top five from the Gene Expression Omnnibus (GEO) (https://www.ncbi.nlm.nih.gov/ most likely cell types. Second, we performed a PCA on the dataset, using the most geo/). Only datasets containing replicates were used. To reduce noise, genes with variable genes. Contaminant cells were flagged as they clustered in the first PCs. a coefficient of variation between biological replicates >0.6 in either comparison Cells usually matching a combination of these two criteria were removed from the groups were selected. Up- and downregulated transcripts were defined as having analysis. Third, we analyzed the distribution of UMIs and genes in each single cell. a fold change in gene expression >1.5 or <2/3 and a t-test P value <0.05. Other Single cells with more than 6,000 UMIs were also filtered out (possible doublets) signatures were obtained from Godec et al.59. (Zemmour_Code/Zemmour_Code.Rmd: **Removing Contaminant cells**). A signature score for each single cell was computed by summing the counts for the upregulated genes and subtracting the counts for the downregulated Gene expression comparison between groups of single cells. Differential gene genes. Z scores were plotted in the heat map (Fig. 4b and Supplementary Fig. 4) 55 expression was performed using the DESeq statistical model (Zemmour_Code/ (Zemmour_Code/Zemmour_Code.Rmd: **Treg signatures and single cell score**). Zemmour_Code.Rmd: **Differential gene expression function**). The distribution of counts Cij for each gene i in group j was modeled by a negative binomial of Correcting technical dropouts in scRNA-seq to estimate frequency of parameter μij (mean) and dispersion θi expression. Frequency of expression is underestimated in single cell RNA-seq studies because of the low sensitivity of the technique. However, several statistical CNB(,) 32,40 ij =μij θi methods have been developed to estimate technical dropouts (Zemmour_Code/ Zemmour_Code.Rmd: **Correcting technical dropouts in scRNA-seq to estimate First, the dispersion was estimated using the background model according to frequency of expression**). Genes were first distributed in 25 bins of expression 56 2 ref. , in which a linear model was fitted between the log ()μ and log ()σi for each according to the population average. Then, for each single cell c, we calculated 2 2 i 2 μi 2 the fraction of genes that were expressed in each bin (count ≥1) and fit a logistic gene i. For each gene i, the dispersion θi =−62 μ , was estimated,sμ max(μ + 10− ,)s i i i μi regression to predict the dropout probability for each gene in the single cell. being the fitted variance of gene i. A generalized linear model was then fit for each Goodness of fit was assessed by calculating the area under the ROC curve (AUC). gene i in each group j [glm(formula = gene ~groups, data = tmp, family = negative. 99% of the cells had an AUC between 0.88 and 0.9. Dropout-corrected gene binomial(θi))], and we used an ANOVA to compare the coefficients in each model expression for each single cell c and gene i (CorrectedCounti,c) was calculated, using and obtain a P value. FDR was calculated to correct for multiple testing. these dropout probabilities (dropouti,c) as weight and the average expression of the To identify the core Treg transcripts independently of other variables such as ten closest genes (by correlation distance) (µi) as expected counts. activation, we compared matched Tregs and Tconvs (Zemmour_Code/Zemmour_ CorrectedCountCount (1 dropoutd) ropout Code.Rmd: **Core Treg genes**). For each single Treg cell (n = 708), we compared its ic,,=×ic −+ic,,μ×i ic 50 closest Tregs and 50 closest Tconvs (correlation distance) and deemed genes with a P value <0.05 to be significant. Core Treg genes scored as significantly different in Frequency of expression was calculated using the dropout-corrected gene >80% of the comparisons. expression (CorrectedCount >0). tSNE representation and clustering. Data were visualized using the t-distributed Paired single-cell TCRα/β sequencing. Single-cell TCRα/β sequencing was done 25 Stochastic Neighbor Embedding (t-SNE) dimensionality reduction algorithm from one batch of splenic Tregs and two batches of colonic Tregs isolated from SPF (Zemmour_Code/Zemmour_Code.Rmd: **t-SNE**). A principal component mice. Single-cell TCRα/β sequencing was done on the same material as for the analysis (PCA) was first performed on the top 100 most variable genes that were whole-transcriptome library construction, by taking a small amount of material expressed in more than 1% of the cells. The Fano factor (variance/mean) was after the RNA amplification and before the fragmentation step. The protocol used as a metric for variability because of its independence from the mean in started by taking an aliquot of the aRNA library before fragmentation, because it Poisson distributed data. The number of significant principal components (PCs) retained the full transcript sequence, and performing a reverse transcription using were determined by comparison to PCA over a randomized matrix, as described TRAV and TRBV external primers (Supplementary Table 5). The sequences of the by Klein et al.21. Two-dimensional tSNE was run using the Rtsne function56 on external and internal primers were obtained from ref. 60 for TRBV and ref. 61 for the significant PCs, setting the seed for reproducibility and using the following TRAV. The Illumina PE1 sequencing primer was added to the 5′ end of the internal parameters (perplexity = 50; max_iter = 1,000). primers (Supplementary Table 5). TCRα and TCRβ RT reactions were performed Clustering was done independently of the tSNE projections based on k-means separately by following the protocol for SuperScriptIII (Invitrogen): 1 μl of the and the method described by Krijthe et al.57, using the clusterboot function in aRNA was mixed with 1 μl of RT primers (external TRAV or TRBV primers) R and kmeansCBI58 on the significant PCs (Zemmour_Code/Zemmour_Code. (10 μM), 1 μl of dNTP and 10 μl of RNase-free water, and incubated at 65 °C Rmd: **Clustering**). Cluster stability was assessed by bootstrapping and analyzed for 5 min before addition of 4 μl of SuperScriptIII (Invitrogen) 5× buffer, with the Jaccard index (average similarity between the closest clusters during 1 μl of DTT and 1 μl of RNAseOut (Invitrogen) and incubation for 50 min at 50 °C, bootstrapping) and silhouette plots (difference between the average distance for then for 70 °C for 15 min. The cDNA library was purified and size-selected using each single cell to cells in the same cluster versus cells in the closest cluster). The 0.5× AMPure SPRI beads. The first PCR used internal primers and added PE1 number of clusters (k) was optimized by running the algorithm with different k and sequencing primers. A qPCR was performed beforehand to optimize the number choosing the largest k with the most stable clusters (all clusters with a Jaccard index of cycles needed for PCR1 by using one-tenth of the previous purified cDNA >0.6). Cluster coherence was also assessed by visualizing them on the tSNE plots. (1 μl), 5 μl water, 10 µl KAPA ReadyMix 2× (Kapa Biosystems), 2 µl SYBRgreen (Invitrogen), 1 µl of primers (RTP_TRAV or RTP_TRBV) (10 µM) and 1 µl of Circular a posteriori projection plots. In order to compare the different single- P7_Rd2_idxN (10 μM). The cycle parameters were 95 °C for 3 min; 40 cycles of cell-dataset replicates, we calculated the likelihood of each single cell belonging to 98 °C for 20 s, 60 °C for 30 s and 72 °C for 1 min. Fluorescence was then read. The the spleen Treg clusters defined in Fig. 4 (Zemmour_Code/Zemmour_Code.Rmd: number (X) of optimized cycles corresponded to the middle of the exponential **Circular A Posteriori Projection plots**). The average expression within each phase (to which log2(10) = 3.3 cycles were subtracted, because the PCR used nine cluster in Fig. 4 was used as prior probability, and we used a multinomial mixture times more cDNA than the qPCR). PCR1 was then performed as follows: cDNA model, as described by Jaitin et al.39 (and above for the contamination algorithm) (10 µl), 11 µl water, 25 µl KAPA ReadyMix 2× (Kapa Biosystems), 2 µl of primers to calculate the log posterior probability of each single cell belonging to any of (RTP_TRAV or RTP_TRBV) (10 µM), 2 µl of P7_Rd2_primer_idxN_R (10 μM). the spleen clusters. For the human data, we used the ortholog genes (http://www. The cycle parameters were 95 °C for 3 min; X cycles of 98 °C for 20 s, 60 °C for 30 s informatics.jax.org/downloads/reports/index.html#homology/). and 72 °C for 1 min. The PCR product was then purified and size selected using

0.5× AMPure SPRI beads. A second PCR incorporating the P5 Illumina sequence Life Sciences Reporting Summary. Further information on experimental design is contained the following: purified PCR1 product (5 µl), 16 µl water, 25 µl KAPA available in the Life Sciences Reporting Summary. ReadyMix 2× (Kapa Biosystems), 2 µl of primers P5_Rd1 (10 μM) and 2 µl of P7_ Rd2_idxN (10 μM). The cycle parameters were 95 °C for 3 min; X cycles of Data availability. The data reported in this paper have been deposited in the Gene 98 °C for 20 s, 60 °C for 30 s and 72 °C for 1 min. The number (X) of cycles for PCR2 Expression Omnibus (GEO) database under accession no. GSE109742). was optimized using a qPCR as described above (fewer than ten cycles). The PCR3 product was then purified and size selected twice using 0.5× AMPure SPRI beads. Library size was assessed using a High sensitivity D5000 ScreenTape (Agilent). The References TCRα and TCRβ libraries had an expected size of 1,470 and 1,230 bp, respectively. 51. Bettelli, E. et al. Reciprocal developmental pathways for the generation of Library concentration was determined by qPCR. Sequencing was done using pathogenic efector TH17 and regulatory T cells. Nature 441, 235–238 (2006). paired-end NanoMiseq (read 1, 265 bp; index, 7 bp; read 2, 51 bp). 52. Liston, A. et al Diferentiation of regulatory Foxp3+ T cells in the thymic Read 1 contained the TCR transcript sequence. Read 2 contained the cortex. Proc. Natl. Acad. Sci. USA 105, 11903–11908 (2008). UMI (6 bp), single-cell barcode (16–19 bp) and a conserved sequenced named 54. Kim, D. et al. TopHat2: accurate alignment of transcriptomes in the presence W1 (similarly to the whole-transcriptome analysis). As for the transcriptome of insertions, deletions and gene fusions. Genome. Biol. 14, analysis, Fastq reads were first filtered for quality (>80% with Sanger Q >20) R36 (2013). and on the expected structure of the reads: paired reads were kept only if read 55. Love, M. I., Huber, W. & Anders, S. Moderated estimation of fold change and 2 contained the W1 and poly(T) sequence and if read 1 did not contain them. dispersion for RNA-seq data with DESeq2. Genome. Biol. 15, Single-cell demultiplexing was performed against the possible barcode space, and 550 (2014). only reads mapping unambiguously and with fewer than two mismatches were 56. Grün, D. et al. Single-cell messenger RNA sequencing reveals rare intestinal kept. TCRα and TCRβ alignment (V, D, J alignment and CDR3 identification) cell types. Nature 525, 251–255 (2015). was done individually for each single cell against the mouse IMGT database 57. Krijthe, J. H. Rtsne: T-Distributed Stochastic Neighbor Embedding (http://www.imgt.org/) using the Mixcr 1.8.1 program62: mixcr.jar using a Barnes–Hut Implementation. GitHub https://github.com/jkrijthe/ align –f --library local -nw -OminSumScore = 0 --loci $loci -OjParameters.parameters. Rtsne/ (2015). floatingRightBound = False -OdParameters.absoluteMinScore = 0 --species 58. Hennig, C. fpc: Flexible Procedures for Clustering. R package version mmu --report $line.$loci.aln.log $line.fq $line.$loci.aln.vdjca; mixcr.jar 2.1-10. Comprehensive R Archive Network https://CRAN.R-project.org/ exportAlignmentsPretty --skip 0 --limit 10 $line.$loci.aln.vdjca $line.$loci.aln. package=fpc (2015). pretty.txt; mixcr.jar exportAlignments -f -p full --no-spaces $line.$loci.aln.vdjca 59. Godec, J. et al. Compendium of immune signatures identifes conserved and $line.$loci.aln.txt; mixcr.jar assemble -f $line.$loci.aln.vdjca --report $line.$loci. species-specifc biology in response to infammation. Immunity 44, asmbl.log $line.$loci.asmbl.clns; mixcr.jar exportClones -f -p full --no-spaces 194–206 (2016). $line.$loci.asmbl.clns $line.$loci.asmbl.txt. Only sequences with a score >100 were 60. Baker, F. J., Lee, M., Chien, Y. H. & Davis, M. M. Restricted islet-cell reactive kept for further analysis. TCRα and TCRβ sequences, and whole-transcriptome T cell repertoire of early pancreatic islet infltrates in NOD mice. Proc. Natl. data were matched to the same single-cell barcode unambiguously with fewer than Acad. Sci. USA 99, 9374–9379 (2002). two mismatches. Only single-cell barcodes with both TCRα and TCRβ sequences 61. Burzyn, D. et al. A special population of regulatory T cells potentiates muscle and for which the transcriptome was available were kept for further analysis. repair. Cell 155, 1282–1295 (2013). Single cells were grouped in clonotypes when sharing the same TCRα and TCRβ 62. Bolotin, D. A. et al. MiXCR: sofware for comprehensive adaptive immunity sequences (V, D, J and CDR3 sequence). profling. Nat. Methods 12, 380–381 (2015).

Corresponding author(s): Christophe Benoist Initial submission Revised version Final submission Life Sciences Reporting Summary Nature Research wishes to improve the reproducibility of the work that we publish. This form is intended for publication with all accepted life science papers and provides structure for consistency and transparency in reporting. Every life science submission will use this form; some list items might not apply to an individual manuscript, but all fields must be completed for clarity. For further information on the points included in this form, see Reporting Life Sciences Research. For further information on Nature Research policies, including our data availability policy, see Authors & Referees and the Editorial Policy Checklist.

Experimental design 1. Sample size Describe how sample size was determined. Sample size was chosen according to standard practices in the field. 2. Data exclusions Describe any data exclusions. In accordance with current practice in the field, single cell barcodes with too few reads (<500) were considered technical failures and excluded. 3. Replication Describe whether the experimental findings were All of data were successfully reproduced by each attempt. reliably reproduced. 4. Randomization Describe how samples/organisms/participants were NA allocated into experimental groups. 5. Blinding Describe whether the investigators were blinded to No blinding test in this study. group allocation during data collection and/or analysis. Note: all studies involving animals and/or human research participants must disclose whether blinding and randomization were used. 6. Statistical parameters For all figures and tables that use statistical methods, confirm that the following items are present in relevant figure legends (or in the Methods section if additional space is needed). n/a Confirmed

The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement (animals, litters, cultures, etc.) A description of how samples were collected, noting whether measurements were taken from distinct samples or whether the same sample was measured repeatedly A statement indicating how many times each experiment was replicated The statistical test(s) used and whether they are one- or two-sided (note: only common tests should be described solely by name; more complex techniques should be described in the Methods section) A description of any assumptions or corrections, such as an adjustment for multiple comparisons The test results (e.g. P values) given as exact values whenever possible and with confidence intervals noted A clear description of statistics including central tendency (e.g. median, mean) and variation (e.g. standard deviation, interquartile range)

Clearly defined error bars June 2017

See the web collection on statistics for biologists for further resources and guidance.

1 Software nature research | life sciences reporting summary Policy information about availability of computer code 7. Software Describe the software used to analyze the data in this Flowjo, R study.

For manuscripts utilizing custom algorithms or software that are central to the paper but not yet described in the published literature, software must be made available to editors and reviewers upon request. We strongly encourage code deposition in a community repository (e.g. GitHub). Nature Methods guidance for providing algorithms and software for publication provides further information on this topic.

Materials and reagents Policy information about availability of materials 8. Materials availability Indicate whether there are restrictions on availability of No unique material was used in this study. unique materials or if these materials are only available for distribution by a for-profit company. 9. Antibodies Describe the antibodies used and how they were validated All antibodies in this study are described in the Method section and their validation for use in the system under study (i.e. assay and species). data are available on the manufacturer's website. 10. Eukaryotic cell lines a. State the source of each eukaryotic cell line used. NA

b. Describe the method of cell line authentication used. NA

c. Report whether the cell lines were tested for NA mycoplasma contamination.

d. If any of the cell lines used are listed in the database NA of commonly misidentified cell lines maintained by ICLAC, provide a scientific rationale for their use.

Animals and human research participants Policy information about studies involving animals; when reporting animal research, follow the ARRIVE guidelines 11. Description of research animals Provide details on animals and/or animal-derived Primary Tconv and Treg cells were isolated from 6-8 week-old male mice using a materials used in the study. reporter (Foxp3-IRES-GFP/B6, Foxp3-IRES-Thy1.1/B6 X Nur77-GFP/B6)

Policy information about studies involving human research participants 12. Description of human research participants Describe the covariate-relevant population 2 healthy male and female donors characteristics of the human research participants. June 2017

2 nature research | flow cytometry reporting summary June 2017 1 s of Final submission Revised version Initial submission Corresponding author(s):Corresponding Benoist Christophe healthy, live and singlet population. CD4+ cells were further gated as CD4+ healthy, live and singlet population. CD4+ FoxP3GFP+ or Foxp3/Thy1.1+, and and TCRb+. Tregs were further gated as or CD25+ (colon) Tconv as FoxP3/GFP- or Foxp3/Thy1.1-, gates and then DAPI to select Human: Starting cells were gated by FSC/SSC cells were further gated in order healthy, live and singlet population. CD4+ to sort Tregs and Tconvs as CD4+ CD25+ CD127low and CD4+ CD25- CD127high, respectively. the spleen, and CD4+ TCRb+ CD25high in the colon. the spleen, and CD4+ TCRb+ CD25high from the blood as CD4+ CD25+ Humans: Tregs and Tconvs were isolated respectively. CD127low and CD4+ CD25- CD127high, Flowjo sorting Around 96% by single sorting, 99% by double

identical markers).

populations within post-sort fractions. the flow cytometry data. Methodological details Data presentation 4. A numerical value for number of cells or percentage (with statistics) is provided. for number of cells or percentage (with 4. A numerical value 2. The axis scales are clearly visible. Include numbers along axes only for bottom left plot of group (a 'group' is an analysi axes only for bottom left plot of group clearly visible. Include numbers along 2. The axis scales are plots. plots with outliers or pseudocolor 3. All plots are contour 1. The axis labels state the marker and fluorochrome used (e.g. CD4-FITC). the marker and fluorochrome used 1. The axis labels state

Tick this box to confirm that a figure exemplifying the gating strategy is provided in the Supplementary Information. Tick this box to confirm that a figure exemplifying the gating strategy is provided

9. Describe the gating strategy used. by FSC/SSC gates and then DAPI to select Mice: Starting cells were gated 8. Describe the abundance of the relevant cell 8. Describe the abundance of the relevant

7. Describe the software used to collect and analyze 7. Describe the software used to collect 6. Identify the instrument used for data collection.6. Identify the instrument used for data LSRII

5. Describe the sample preparation. in Mice: Tregs were identified as CD4+ FoxpP3/GFP+ or FoxpP3/Thy1.1+ For all flow cytometry data, confirm that: For all flow cytometry

Form fields will expand as needed. Please do not leave fields blank. Please do not will expand as needed. Form fields Flow Cytometry Reporting Summary Reporting Cytometry Flow