Rapid chromatin repression by Aire provides precise control of immune tolerance

Abstract

Aire mediates the expression of tissue-specific antigens in thymic epithelial cells to remove dangerous self-reactive T lymphocytes. However, the mechanism that allows expression of tissue-specific genes at levels that prevent harm is unknown. Here we show that Brg1 generates accessibility at tissue-specific loci to impose central tolerance. We found that Aire harbors an intrinsic repressive function that restricts chromatin accessibility and opposes Brg1 across the genome. Aire exerted this repressive influence within minutes upon recruitment to chromatin and restrained the amplitude of active transcription. Autoimmune mutations that impair Aire-induced activation also impair the its repression function, indicating dual roles for Aire. Together, Brg1 and Aire fine-tune the expression of tissue-specific genes at levels that prevent toxicity, yet promote immune tolerance.

The functional competence of the immune system requires development of T cells that promote defense against pathogens and tumors, while silencing or removing T cells that react against self-constituents. The resultant T cell repertoire is limited by the range of self-antigens presented in the thymus¹. The ectopic expression of thousands of tissue-specific self-antigens (TSAs) encompassing all parenchymal organs leads to immune tolerance to these self-antigens^2,3. This ectopic transcription is driven largely by Aire, which operates in medullary thymic epithelial cells (mTECs) and mutations in this transcription factor cause autoimmune polyendocrine syndrome type-1 (APS-1)^4,5.

Tissue-specific expression programs are generally defined by the developmentally coordinated actions of positive and negative regulators that influence the recruitment and release of RNA polymerase II (Pol II) at enhancers and promoters of lineage-specific genes⁶. Aire releases paused Pol II for productive elongation^7–10 but the mechanisms that lead to accessibility and Pol II binding are unclear. The determinants of this poising mechanism in mTECs are distinct from those in peripheral tissues, as the lineage-specific transcription factors essential in the latter contexts are unnecessary for their thymic expression¹¹.

While Aire has a unique ability to activate a large spectrum of tissue-specific genes, mTECs must also limit the expression of these genes because many of them encode proteins that would perturb physiological processes if expressed at levels comparable to their later expression. Indeed, the expression of Aire-induced PTGs in mTECs is orders of magnitude lower than in their respective peripheral tissues^10,12,13. The expression of PTG subsets within single mTECs is transient and shuttles between distinct gene subsets, reducing the number of mTECs necessary to present the entire PTG repertoire¹⁴. Elucidating how the duration and amplitude of tissue-specific expression is controlled is essential to understanding how Aire induces precisely quantified transcription.

Here, we report that Aire restricts chromatin accessibility through its multimerizing and histone-binding activities to restrain the amplitude of PTG transcription. We also found that Brg1 promotes accessibility at loci encoding tissue-specific antigens, thereby allowing Aire to function.

RESULTS

mTEC^hi differentiation promotes chromatin accessibility

To elucidate the Aire-independent mechanisms that poise tissue-specific genes in mTECs, we determined when tissue-specific loci become accessible by examining the developmental control of chromatin accessibility during mTEC differentiation. Employing ATAC-seq¹⁵ and gene expression profiling, we defined accessibility landscapes and transcriptomes for progenitor mTECs (mTEC^lo) and Aire⁺ differentiated mTECs (mTEC^hi) distinguished by their surface expression of major histocompatibility class II (MHCII) (Supplementary Fig. 1a–e). We focused our ATAC-seq analyses on the main axis of variance from principal component analysis (PCA) that separates mTEC^hi and mTEC^lo samples (Supplementary Fig. 1f) to reduce the contribution of batch effects and other technical noise¹⁶. Using MAVRIC (Moskowitz and Greenleaf, manuscript submitted), a bioinformatic framework that attributes axes of variance to underlying sources (ATAC-seq peaks in this case), we identified 59,818 differentially accessible peaks that correlated with mTEC^hi maturation, accounting for ~ one-fourth of the total accessible genome detected (Fig. 1a). Nearly 70% of these peaks exhibited increases in accessibility during mTEC^hi differentiation and were enriched for their nearest gene being tissue-specific (Fig. 1a–d and Supplementary Fig. 1g). The mTEC^lo state exhibited low levels of accessibility at these regions, comparable to those in embryonic stem (ES) cells and naïve CD8⁺ T cells (Fig. 1d). Polycomb repressive complexes were enriched in ES cells (Supplementary Fig. 1h), suggesting that these tissue-specific sites were likely within facultative heterochromatin, consistent with the reported enrichment of lysine 27 trimethylation of histone H3 (H3K27me3) and depletion of Pol II binding and active histone marks at Aire-regulated tissue-specific genes¹⁷. Accessibility changes at the mTEC^hi-induced but not mTEC^hi-repressed peaks robustly correlated with the neighboring PTGs’ transcriptional changes (Fig. 1c). 90% of these peaks were > 1 kb away from the nearest transcriptional start site (TSS) (Fig. 1d,e). These results indicate a profound shift in the chromatin landscape during mTEC^hi differentiation that is driven by accessibility changes at distal cis-regulatory elements controlling transcription of tissue specific genes.

Figure 1.

mTEC^hi differentiation promotes chromatin accessibility at tissue-specific loci. (a) Heatmap of normalized ATAC-seq fragment density at differential peaks (rows). n = 4 independent experiments. (b) Tissue restrictions of indicated peak sets (filtered for peaks with accessibility fold-change > 2) assessed by tissue expression profile of the nearest gene. (c) Cumulative distribution function (CDF) plot of the transcriptional fold-change between mTEC^hi and mTEC^lo at tissue-specific peaks classified by differential accessibility changes (key). Representative of 4 independent experiments. Mann-Whitney U tests (two-tailed) performed for comparisons between classes of peaks. Confidence intervals (95%) are specified for each comparison. (d) Density plots of ATAC-seq fragment dyads from mTEC^hi, mTEC^lo, embryonic stem cells (ESC) and CD8+ T cells at induced peaks near tissue-specific genes upregulated in mTEC^hi (left) or near housekeeping genes (right). Representative of 4 independent experiments. (e) Clustered correlation of chromatin accessibility at ATAC-seq peaks distal (left) and proximal (right) to tissue-specifc genes with induced transcription in mTEC^hi vs. mTEC^lo. n = 4 independent experiments.

Aire and Brg1 have opposing influences on chromatin state in mTEC^hi

Subunits of the mSWI/SNF or BAF chromatin remodeling complex appeared in a hypothetical network for functional partners for Aire⁸ suggesting that BAF complexes specifically contribute to the poising of PTGs in mTECs. Thus, we profiled the accessibility landscapes in mTEC^hi and mTEC^lo from Foxn1^ex9cre;Brg1^F/F mice, which delete Smarca4, the catalytic subunit of Brg1 in TECs (Brg1-cKO hereafter). The frequencies of mTEC compartments and the punctate localization of Aire were largely normal in Brg1-cKO mice (Supplementary Fig. 2a–c). To compare the influence of Brg1 on the mTEC^hi chromatin state with that of Aire, we profiled the accessibility landscapes in mTECs from Aire-deficient (Aire-KO hereafter) mice. Hierarchical clustering based on the correlations of accessibility at mTEC^hi-induced peaks near PTGs separated the Brg1-cKO and Aire-KO mTEC^hi samples (Fig. 2a), indicating distinct influences of Brg1 and Aire on the mTEC^hi chromatin landscape. Aire-KO and Brg1-cKO mTEC^hi samples distributed at opposite poles of the principal component of variability attributable to the genotypes of mTEC^hi samples (PC2) (Fig. 2b), suggesting divergent influences of Aire and Brg1 on the mTEC^hi-specific accessibility state.

Figure 2.

Aire and Brg1 are determinants of an mTEC^hi-specific chromatin state with opposing influences on accessibility. (a) Correlations between indicated ATAC-seq samples at tissue-specific regions exhibiting induced accessibility during mTEC^hi differentiation. n = 4 independent experiments. (b) Principal component analysis of ATAC-seq data. (c) Comparison of the influences of Aire and Brg1 on mTEC^hi-specific accessibility state (ATAC-seq peaks correlated to PC2). Peaks whose nearest genes are upregulated in mTEC^hi vs. mTEC^lo are highlighted in orange and enumerated in each cartesian quadrant with total peaks (blue). P values indicate one-sided Fisher’s exact test with 95% confidence intervals (lower bound). (d) Aire (top) and Pol II (bottom) ChIP-seq fragment dyad density at Aire-induced or Aire-repressed ATAC-seq peaks. Representative of 2 independent experiments. (e) Heatmap of Aire (top) or Pol II (bottom) ChIP-seq fragment dyad density at Aire-repressed ATAC-seq peaks. Representative of 2 independent experiments. (f) Aire (top) and Pol II (bottom) ChIP-seq fragment dyad density at Brg1-induced or Brg1-repressed ATAC-seq peaks. Representative of 2 independent experiments. (g) Genomic signal tracks of ATAC-seq fragments at two loci from indicated mTEC^hi samples (top). ChIP-seq signal tracks (representative of 2 independent experiments) from WT mTEC^hi samples. Red arrowheads indicate differentially accessible regions.

Using MAVRIC, we identified peaks correlated to PC2 and found a negative correlation (r = −0.43) when comparing accessibility promoted by Aire versus that promoted by Brg1 (Fig. 2c). Using arbitrary fold-change cutoffs to classify induced and repressed peaks (FC >2, <0.5), cumulative distribution function (CDF) plots show that Brg1-induced peaks were predictive for Aire repression and Aire-induced peaks were predictive for Brg1 repression (Supplementary Fig. 2d), indicating that Aire and Brg1 exert opposing forces at the same sites. To assess the functional relevance of these regions for mTEC^hi differentiation, we assigned the nearest TSS to each ATAC-seq peak. mTEC^hi-activated genes were far more frequently near peaks whose accessibility was repressed by Aire and induced by Brg1 (Fig. 2c) than to peaks induced by Aire and repressed by Brg1 or those repressed by both Aire and Brg1 (Fig. 2c). To validate the negative influence of Aire on chromatin accessibility, we analyzed published ATAC-seq samples¹⁸ and found increases in fragment density in Aire-KO compared to wild-type mTEC^hi at peaks correlated to the primary axis of variance separating the genotypes (Supplementary Fig. 2h,i). Taken together, these data unveil that Aire has a repressive function at BAF-promoted accessible regions near genes upregulated during mTEC^hi differentiation.

To assess whether Aire is directly mediating these accessibility changes, we used published mTEC^hi Aire ChIP-seq data¹⁸ to assay Aire occupancy at differentially accessible peaks. Aire localization was robustly enriched at regions where chromatin accessibility was repressed by Aire compared to unaffected regions (Fig. 2d–g and Supplementary Fig. 2g,j). Furthermore, Aire bound to regions where accessibility was induced by Brg1 (Fig. 2f,g and Supplementary Fig. 2g), suggesting a direct role for Aire in repressing accessibility at Brg1-promoted sites. These Aire-repressed, Brg1-promoted regions were mostly gene-distal with only ~7% of peaks within promoters (Supplementary Fig. 2e,f). These regions were also enriched for Pol II, acetylation at lysine 27 of histone H3 (H3K27ac), topoisomerases I and IIa (Top1 and Top2a) and γ-H2AX deposition (Fig. 2d–g and Supplementary Fig. 2g,3), further supporting their enhancer potential. In contrast, regions where accessibility was increased by Aire were depleted of Aire, Pol II, H3K27ac, Top1, Top2a and γ-H2AX localization (Fig. 2d–g and Supplementary Fig. 3), suggesting an indirect influence of Aire on accessibility at these loci. These data suggest that the repressive function of Aire is executed after BAF-promotes accessibility.

Accessibility at PTGs is promoted by Brg1 and repressed by Aire

To gain insight into the mechanism of the ectopic expression of tissue-specific genes in mTECs, we focused our analyses on the regions whose accessibilities could serve to directly promote the expression of PTGs during mTEC^hi differentiation. We defined these regions as ATAC-seq peaks whose accessibility is specifically induced in mTEC^hi compared to mTEC^lo, is bound by Aire, and whose nearest gene is tissue-restricted. 2,255 peaks qualified for these criteria (Fig. 3a), which, considering the heterogeneity of individual mTECs expressing distinct subsets of the collective PTG repertoire^{10,11,13,14,19}, is likely only a fraction of the total. These qualifying regions presumably reflect PTGs with higher probability of expression (up to 48% detected) in individual mTECs¹¹. The accessibility changes at these peaks exhibited a strong correlation with the transcriptional changes during mTEC^hi maturation (Fig. 3b), indicating these peaks may serve as cis-regulatory elements that control PTG expression.

Figure 3.

Accessibility at tissue-specific loci are promoted by Brg1 and repressed by Aire. (a) MA plot of differential ATAC-seq peaks between mTEC^hi and mTEC^lo samples (blue), highlighted by those whose nearest gene is tissue-specific (light blue) and those that have Aire occupancy and neighbor tissue-specific genes (red). (b) CDF plot of the transcriptional fold-change between mTEC^hi and mTEC^lo at indicated loci classified by differential accessibility changes. P value from two-tailed Mann-Whitney U test with 95% confidence intervals. Representative of two independent experiments. (c) Density plots of ATAC-seq fragment dyads from mTEC^hi samples of indicated genotypes at Aire-bound, ATAC-seq peaks near tissue-specfic genes (left) or housekeeping genes (right). Representative of 2 independent experiments. (d) Heatmap of normalized ATAC-seq fragment dyad density between samples of indicated genotypes at mTEC^hi-induced, Aire-bound, tissue-specific loci. n = 2 independent independent experiments. (e) Heatmap of Aire ChIP-seq fragment dyad density at same regions as (d). Representative of 2 independent experiments.

To determine whether BAF promotes the poising of these tissue-specific regions by facilitating their accessibility, we quantified the ATAC-seq fragment density at these sites and found a substantial reduction of fragment density in Brg1-cKO mTEC^hi compared to wild-type (Fig. 3c,d). In contrast, the ATAC-seq fragment density at these sites in Aire-KO mTEC^hi increased (Fig. 3c–e), indicating that Aire repressed accessibility at these regions near transcriptionally upregulated PTGs. Furthermore, the increase in chromatin accessibility at these tissue-specific loci in mTEC^hi compared to mTEC^lo was also present in Aire-KO mTEC^hi cells (Supplementary Fig. 4a), indicating accessibility at these loci was induced in Aire-deficient mTECs. In contrast, the transcriptional activation of Aire-dependent tissue-specific genes was compromised in Aire-KO mTEC^hi compared to wild-type (Supplementary Fig. 4b). Similarly, the the transcriptional activation of PTGs during the mTEC^lo to mTEC^hi transition in wild-type mice was largely Aire-dependent (Supplementary Fig. 4d). In contrast, Aire was broadly dispensable for the accessibility induction at PTGs during normal mTEC^hi differentiation (Supplementary Fig. 4c). This is consistent with reports that the amount of DNA methylation and promoter-proximal transcription of Aire-induced PTGs does not change in Aire-KO mTECs compared to wild-type mTECs^7,10. Taken together, these results suggested that Brg1 was essential for promoting accessibility near PTGs in mTEC^hi that Aire later targeted for transcriptional activation. Moreover, the negative influence of Aire on accessibility may restrain the transcriptional amplitude at PTGs.

NF-κB target sites are hypersensitive to regulation by Aire and Brg1

Because the 8 DNA-binding domains of the BAF subunits have only limited sequence specificity²⁰ and Aire does not exhibit sequence-specific DNA-binding^21,22, we sought to identify transcription factors potentially working in tandem with either complex that promote the mTEC^hi-specific chromatin state. Steric hindrance between transcription factors and the Tn5 transposase can be inferred from ATAC-seq data, allowing enrichment analyses for transcription factor footprinting¹⁵. Using an established framework to quantify the changes in accessibility across ATAC-seq peaks sharing known transcription factor motifs²³, we identified a battery of transcription factors that were differentially enriched in wild-type mTEC^hi compared to mTEC^lo. Peaks with motifs for the transcription factors cRel, RelA and NF-κB1 exhibited the most prominent increases in accessibility (Fig. 4a and Supplementary Fig. 5a), consistent with the essential roles of NF-κB signaling in mTEC^hi maturation^24,25. We also inferred the differential increase in binding of NF-κB complexes in mTEC^hi vs. mTEC^lo using transcription factor footprinting (Supplementary Fig. 5e). We also noticed significant decreases in accessibility at sites containing STAT, P53 and P63 motifs in mTEC^hi compared to mTEC^lo (Fig. 4a and Supplementary Fig. 5a), along with the putative loss in binding of these factors (Supplementary Fig. 5f,k,l,n–p), consistent with the reported role of Stat3 and p63 in the expansion and survival of the progenitor mTEC^lo compartment^26–28.

Figure 4.

Regions containing NF-kB motifs are highly sensitive to opposing regulation by Aire and Brg1. (a) Heatmap of deviations from expected ATAC-seq signal from mTEC^hi or mTEC^lo samples at ATAC-seq peaks containing known trans-factor motifs. n = 4 independent experiments (b) Change in deviation from expected accessibility signal at ATAC-seq peaks containing known trans-factor motifs (key) between indicated samples. Mean (diamond) and replicates (circles) from n = 2 independent experiments. (c) Accessibility footprints at NF-kB motifs within ATAC-seq peaks from indicated mTEC^hi samples. (d) Top known motif enrichment within Aire ChIP-seq peaks. (e) Heatmap of Aire ChIP-seq fragment dyad density at ATAC-seq peaks with NF-kB motifs. Representative of 2 independent experiments.

Because NF-κB depends on pre-existing accessibility for binding²⁹, the substantial increase in accessibility at sites with NF-κB motifs during mTEC^hi maturation suggested NF-κB-binding requires chromatin remodeling to impose the mTEC^hi state. Indeed, Brg1 promoted chromatin accessibility most robustly at peaks containing NF-κB motifs (Fig. 4b and Supplementary Fig. 5b,d). Furthermore, the NF-κB footprinting along with the flanking accessibility was substantially reduced in Brg1-cKO mTEC^hi at these sites compared to wild-type (Fig. 4c). These data indicate that BAF prepared the chromatin landscape to allow NF-κB-binding at target loci and drive mTEC^hi differentiation.

Activation of NF-κB signaling in mTECs precedes the expression of Aire²⁵. To determine the influence of Aire on the NF-κB and Brg1-induced chromatin accessibility, we quantified the changes in accessibility between Aire-KO mTEC^hi and wild-type mTEC^hi across peaks containing all known motifs. The largest changes were observed at peaks with NF-κB motifs, whose accessibilities were repressed by Aire (Fig. 4b and Supplementary Fig. 5c,d). Moreover, the flanking accessibility and magnitude of the NF-κB footprinting was increased in Aire-KO mTEC^hi vs. wild-type controls (Fig. 4c). Robust Aire occupancy at NF-κB motif-containing ATAC-seq peaks indicated direct repression by Aire at Brg1-promoted sites (Fig. 4d,e). Considering the central role of NF-κB activation in mTEC^hi differentiation and the negative regulation of mTEC^hi cellularity by Aire^5,13,30,31, these results suggest Aire directly curtails fundamental components of the mTEC^hi differentiation program by limiting chromatin accessibility. Indeed, Aire bound to and diminished chromatin accessibility near genes encoding factors essential for mTEC^hi differentiation, e.g. Cd80, H2-Eb2 and Fam18a (Fig. 2g and Supplementary Fig. 2g).

Brg1 imposes immunological tolerance

Because Aire acts on a preexisting chromatin landscape, we asked whether the induced accessibility at tissue-specific loci was important for ectopic transcription of PTGs and central tolerance induction. Accessibility changes at tissue-specific loci during the mTEC^lo to mTEC^hi transition were robustly predictive of the accessibility changes between Brg1-cKO and wild-type mTEC^hi (Fig. 5a). Gene-expression profiling of Brg1-cKO mTEC^hi showed that the activation of hundreds of PTGs was diminished in Brg1-cKO mTEC^hi compared to wild-type, especially at Aire-regulated genes (Fig. 5b). In addition, the top decile of Brg1-upregulated genes was highly enriched for PTGs expressed in a single peripheral tissue (Fig. 5c). Taken together, these results suggested that BAF bolstered PTG expression by promoting accessibility at distal regulatory elements during mTEC differentiation. To determine how the compromised activation of PTGs in Brg1-cKO mice affected the self-reactivity of the T cell repertoire, we first assayed the frequency of activated T cells in the periphery. While the frequency of thymocyte compartments were largely normal (Supplementary Fig. 6a,b), the Brg1-cKO mice exhibited a doubling of the frequency of CD4⁺ CD44^hi CD62L^lo T effector cells compared to wild-type, with nearly half of the splenic T cell compartment consisting of activated CD44^hi CD62L^lo T effector cells at 6 months of age (Fig. 5d–f). However, the number of total splenic T cells in Brg1-cKO mice was approximately half of that in wild-type mice (Supplementary Fig. 6e), likely due to the compromised function of Brg1-deficient cTECs that promote early thymocyte differentiation (Supplementary Fig. 6c,d). We next asked whether the activated effector T cells in Brg1-cKO mice could provoke autoreactive tissue damage. We found substantial lymphocytic infiltration in the kidney, liver, lung, lacrimal and salivary glands at 3 – 6 months old Brg1-cKO mice (Fig. 5g,h). Immunohistochemistry showed strong CD3 staining within these tissue infiltrates (Supplementary Fig. 6h).

Figure 5.

Brg1 imposes immunological tolerance. (a) CDF plot of accessibility fold-changes at tissue-specific peaks classified as differentially accessible in WT mTEC^hi vs. mTEC^lo. P value (two-tailed Mann-Whitney U-test with 95% confidence intervals). Representative of 2 independent experiments. (b) Role of Brg1 on the ectopic induction of tissue-specific genes during mTEC^hi differentiation. Dotted lines indicate 1.5- and 0.66 fold-change for each comparison. Corner numbers indicate total (black) and Aire-upregulated (orange) genes in each quadrant. P values correspond to genes in upper right quadrant (one-tailed Fisher’s exact test with 95% confidence intervals). (c) Fraction of genes from top decile of Brg1-activated genes in mTEC^hi in each indicated tissue-restriction group (number of peripheral tissues expressing gene) compared to random group of genes. (d) Frequencies of activated splenic CD4+ T cells in 4 wk-old mice (n = 8 independent experiments). Mean +/− s.e.m. **P = 0.002 (two-tailed t-test). (e) Comparison of the frequencies of activated splenic T cells as a function of age from indicated genotypes. (f) Representative cytometry plot from 4 independent experiments of the frequency of activated CD4+ T cells in spleen from 6 month-old mice of indicated genotypes. (g) Histological analyses of indicated tissues (left) from 3–6 month-old WT or Brg1cKO mice via H&E staining for infiltrating lymphocytes. Each octagon represents individual mouse. n = 6 independent experiments. (h) Representative H&E stainings (n = 6 independent experiments) of the histopathology in diseased tissues of Brg1cKO mice. Scale bars = 200um. (i) Histological analyses of tissues via H&E staining for infiltrating lymphocytes from nude mice, 14 wk post-thymus transplants from indicated donors (right). Representative H&E stainings (left) from 2 independent experiments. Scale bars = 100um.

We next addressed whether the autoimmunity in Brg1-cKO mice was due to failure in the negative selection of self-reactive T cells and/or the positive selection of Foxp3⁺CD4⁺25⁺ regulatory T cells (T_reg cells) or to the diminished expression of tissue-specific self-antigens (TSAs) in mTECs. Brg1-cKO mice had normal frequency, number and function of splenic Foxp3⁺ T_reg cells compared to those in wild-type mice (Supplementary Fig. 6f,g), suggesting a functional T_reg cell compartment. To test whether the autoimmunity in Brg1-cKO mice was driven by the reduced thymic output, which might extend the neonatal window of physiologic lymphopenia, allowing aberrant homeostatic proliferation of self-reactive T cells³², or by deficiencies in the selection or function of the T_reg cell compartment, we grafted thymic stroma depleted of hematopoietic cells by treatment with 2-deoxyguanosine from wild-type mice, Brg1-cKO mice or both under the kidney capsules of wild-type athymic nude mice. This experimental setting allowed us to test whether the normalized thymic output and the provision of T_reg cells from the wild-type co-transplanted thymus could rescue the autoimmune pathology driven by the self-reactive T cells escaping negative selection in the Brg1-cKO thymus. Mice transplanted with both wild-type and Brg1-cKO thymic stroma exhibited lymphocytic infiltration in multiple organs (Fig. 5i), indicating the self-reactive T cells from the Brg1-cKO thymus had a dominant activity. However, the range of organs affected in these co-transplanted chimeric hosts was reduced to the liver and salivary gland compared to those transplanted with only Brg1-cKO thymic stroma that exhibited infiltrations in the kidney and lacrimal glands (Fig. 5i), suggesting a potential role for defective positive selection of T_reg cells and/or lymphopenia-induced proliferation to the autoimmunity in Brg1-cKO mice. Taken together, these data underscore the importance of Brg1 in promoting central tolerance induction by facilitating the accessibility at tissue-specific loci, an essential component for TSA expression and selection of a functional T cell repertoire.

Aire rapidly represses accessibility upon recruitment to chromatin

To elucidate the mechanism of the repressive function of Aire, we used the chromatin in vivo assay (CiA) system^33–35, which allows precise temporal control of the recruitment of chromatin regulators to specific genetic loci via chemically-induced proximity within living cells. The minute-by-minute assessment of effects of recruitment of a chromatin regulator allows one to distinguish between rapid direct biochemical actions of Aire from slow secondary and indirect effects, e.g. influences on cell cycle or the actions of factors encoded by Aire-target genes. Mouse embryonic fibroblasts (MEFs) from CiA mice have two arrays of binding sites for transcription factors and GFP inserted into one allele of the Oct4 gene (CiA:Oct4)³³, which is an Aire-regulated, tissue-specific locus in mTEC^{hi 31}(Supplementary Fig. 7a). This locus is faithfully integrated into the endogenous chromatin context with appropriate long-range, topological conformations³⁴. Furthermore, this locus is transcriptionally silent and allows assessment of effects independent of transcription. We generated stable cell lines from these mice expressing the ZFHD1 DNA-binding domain fused to FKBP12 (component of mTOR signaling pathway) and Aire fused to the FRB domain of mTOR and V5 epitope serving as the anchor and recruitment partner, respectively (Fig. 6a). Addition of rapamycin elicited the rapid association of FKBP12 and FRB causing the robust recruitment of Aire (> 200-fold measured by V5 ChIP) to the CiA:Oct4 locus within minutes (Fig. 6b). We assessed the impact of Aire recruitment to the CiA:Oct4 locus and found an extensive loss of chromatin accessibility at the recruitment site (measured by DNase I sensitivity) within 15 minutes of rapamycin addition compared to control ethanol treatment (Fig. 6b). The magnitude of the accessibility loss at the modified Oct4 locus appeared comparable to that seen after 1 hour or after 5 days of Aire recruitment (Supplementary Fig. 7b), suggesting Aire induces its maximum repressive effect within minutes of recruitment. These observations indicated that Aire has a direct repressive activity. This rapid repression was in contrast to the quick gain in accessibility upon BAF complex recruitment to the CiA:Oct4 locus (Supplementary Fig. 7c,d)^34,35. Furthermore, the transcriptionally silent state of the CiA:Oct4 locus indicated that the repressive activity of Aire is uncoupled to transcription (Supplementary Fig. 7e). In a distinct approach, we employed an orthogonal recruitment strategy using dCas9 and MS2-binding aptamers³⁶ in the human cortical thymic epithelium cell line 4D6³⁷ (Supplementary Fig. 7f,g). We chose an open tissue-specific locus upstream of PSMB11 (encoding the β5t subunit of the thymoproteosome³⁸) to test the influence of Aire recruitment on accessibility. Aire recruitment upstream of PSMB11 induced loss of accessibility compared to the recruitment of dCas9 alone (Supplementary Fig. 7h), confirming the results at the CiA:Oct4 locus in MEFs. Taken together, Aire exerted its repressive influence within minutes of recruitment to an Aire-regulated locus independent of transcription, indicating a direct biochemical function.

Figure 6.

Aire rapidly represses accessibility upon recruitment to chromatin. (a) Schematic of inducible CiA system: Recruitment to modified Oct4 locus via rapamycin (Rap)-induced dimerizing of Aire-Frb and Fkbp-Zfhd1 fusion proteins. (b) Changes in DNase I hypersensitivity (left, n = 4 independent experiments - circles) at CiA:Oct4 locus upon Rap-induced (15 min) Aire recruitment measured by V5 ChIP (right, n = 3 independent experiments). Mean +/− s.e.m. (c) Depiction of APS-1 patient mutations in CARD, SAND and PHD1 domains. (d) Western blot of Aire-Frb (and mutants) transgenic expression in CiA system. Representative of two independent experiments. (e) Changes in DNase I hypersensitivity (left, n = 4 independent experiments) at CiA:Oct4 locus upon Rap-induced (15 min) WT Aire or variant (key) recruitment measured by V5 ChIP (right, n = 3 independent experiments). Mean +/− s.e.m. Statistical significance for each mutant relative to WT by two-tailed t-test. P values < 0.05 (*), < 0.01 (**), < 0.001 (***).

Aire repressive function requires the PHD1 and multimerization domains

Aire directly interacts with chromatin through its PHD1, a histone-binding module whose recognition of unmodified H3-tail is necessary for tolerance induction^22,39,40. Because the CiA system uses a ZFHD1-FKBP anchor, the recruitment of Aire to the CiA:Oct4 locus does not depend on its own domains, and as such can be used to uncouple the targeting and repressive functions of Aire. The Aire C311Y mutation in humans and C313Y in mice (Fig. 6c) disrupt the zinc coordination of PHD1⁴¹, preventing Aire interaction with chromatin²². This mutation gives rise to the autoimmune polyendocrine syndrome type-1 (APS-1)⁴. When we recruited a Aire-C313Y variant to the CiA:Oct4 locus via rapamycin, the repressive activity of Aire was abolished compared to the recruitment of wild-type Aire, despite the comparable protein expression and locus recruitment between cells expressing Aire-C313Y and Aire-WT (Fig. 6d,e). Because the unfolding of the PHD1 finger in the Aire-C313Y mutant might affect other domains, we next recruited an Aire variant with a designed mutation (Aire-D299A) that mitigates the histone-binding activity of Aire, but maintains the integrity of the PHD finger^22,41. Aire-D299A was unable to repress accessibility at the CiA:Oct4 locus compared to that of wild-type Aire (Fig. 6e), indicating the importance of the histone-binding module of Aire for both its repressive and activating functions. The G228W APS-1 mutation in the SAND domain of Aire (G229W in mice) exhibits dominant negative activity to cripple Aire localization⁴². In contrast to the PHD1 mutants Aire-C313Y and Aire-D299A, recruitment of Aire-G229W to the CiA:Oct4 locus decreased chromatin accessibility similar to that of Aire-WT (Fig. 6e), indicating that the targeting defects of Aire-G229W were rescued by the ZFHD1-FKBP anchor.

Aire multimerizes through its caspase recruitment domain (CARD) to form a multi-protein complex of ~750 kDa^18,43. The L28P APS-1 mutation (L29P in mice) disrupts the ability of Aire to multimerize, causing defects in subnuclear localization and activation of target genes⁴³. The recruitment of Aire-L28P to the CiA:Oct4 locus was ~50-fold less than the recruitment of Aire-WT, despite comparable protein expression in the respective CiA MEFs (Fig. 6d,e), and the repressive activity of Aire-L28P was largely abolished compared to Aire-WT (Fig. 6e), suggesting that the multivalent conformation of Aire within an oligomerized complex is essential to its repressive function. Taken together, the repressive function of Aire is keyed by the same multimerization and histone-binding activities that are necessary for promoting ectopic transcription, indicating a unique duality in the functional domains of Aire.

Aire recruitment to an active locus restrains transcriptional amplitude

If the repressive influence of Aire on chromatin accessibility serves to constrain the transcription of tissue-specific genes, directed Aire recruitment to an active locus should reduce gene expression. To directly test the impact of Aire on active transcription, we constitutively recruited the transcriptional activator VP16 to the CiA:Oct4 locus and sorted for cells that activated the GFP reporter encoded in the CiA:Oct4 locus before a 3 day treatment with rapamycin to co-recruit Aire (Fig. 7a). Both the intensity of GFP expression and the frequency of GFP⁺ cells were reduced upon Aire and VP16 locus co-recruitment compared to cells not given rapamycin (Fig. 7a–c). In contrast, co-recruitment of VP16 and the BAF complex to the CiA:Oct4 locus increased the frequency of GFP⁺ cells (Fig. 7d,e). The repressive influence of Aire was dependent in part on the histone-binding module and multimerization potential of Aire (Fig. 7f). These observations suggest that multimerized Aire and the recognition of unmodified H3-tails through the PHD1 fingers contribute to fortifying a barrier for active transcription. Taken together, these results indicate that during mTEC development, the initial poising by Brg1 and the subsequent transcriptional triggering by Aire was restrained by the repressive influence of Aire on chromatin accessibility, akin to a rheostat in an electrical circuit (Supplementary Fig. 7i).

Figure 7.

Aire recruitment to an active locus restrains transcriptional amplitude. (a) Top: Sequential recruitment of Aire (Rap) to CiA:Oct4 locus for 3 days post VP16 recruitment, frequency of GFP+ cells over ethanol control (EtOH) assessedvia flow cytometry. Representative cytometry plots of 6 independent experiments. Bottom: GFP+ cells post-VP16 recruitment were sorted, then treated +/− Rap for 3 days. Representative cytometry of 5 independent experiments. (b) GFP+ frequency quantification via VP16 +/− Aire co-recruitment (n = 6 independent experiments). Mean +/− s.e.m. ***P <0.001 (two-tailed t-test). (c) Representative histogram of Rap- vs. EtOH-treated cells described in (a-bottom) for GFP expression. Representative of 5 independent experiments. (d,e) Sequential recruitment of BAF via Ss18 (Rap) for 3 days post VP16 recruitment, frequency of GFP+ cells quantified (n = 7 independent experiments), Mean +/− s.e.m. ***P <0.001 (two-tailed t-test). (f) Quantification of fold-change in frequency (left) and mean flourescence intensity (right) of GFP+ cells between EtOH- and Rap-treated conditions after sorting as in (a). n = 3 independent experiments. Mean +/− s.e.m. Two-tailed t-tests between each mutant and WT Aire recruitment: P < 0.05 (*), < 0.01 (**), < 0.001 (***), < 0.0001 (****).

DISCUSSION

Here we provide new insight on the underlying mechanisms that promote activation of thousands of tissue-specific genes in the thymic epithelium, at levels that are sufficient for immune tolerance, but below those that could be harmful. We find that Aire, the transcriptional regulator necessary for much of the mTEC expression of tissue-specific genes, has an intrinsic repressive function that restricts chromatin accessibility. Our analyses indicate that Aire bound Brg1-induced, mTEC^hi-specific regulatory sites and repressed their accessibility to curtail mTEC^hi differentiation. Taken together with the repressive actions of Aire at tissue-specific loci, it appears that Aire limits the duration and amplitude of transcription of tissue-specific genes, as well as the number of cells that could become competent for this ectopic expression. The physiological necessity for this negative regulation can be explained by the negative consequences of overexpressing tissue-specific factors such as insulin, calcitonin, and blood coagulation factors in mTECs^44–46.

Our recruitment studies using chemically-induced proximity at an Aire-regulated locus indicates that Aire exerts its repressive influence within minutes of recruitment to chromatin, indicating a direct biochemical mechanism. Because this locus is transcriptionally silent, the repressive influence of Aire on accessibility precludes any dependence on active transcription. This repressive function required key residues that facilitate the multimerization and histone-binding activities of Aire. Importantly, the same domains and residues are essential for expression of tissue-specific genes and prevention of autoimmunity^45,40. The histone-binding activity of Aire stabilizes its interaction with target loci, contributing to its targeting function^39,40. However, the CARD-mediated multivalent state allows the simultaneous recognition of multiple histone tails that likely favor nucleosomal occupancy and perhaps repression of transcription initiation. To our knowledge, this intrinsic functional duality has not been described for a single transcription regulatory factor.

We find that the Brg1 catalytic subunit of the mSWI/SNF remodeling complex is essential for promoting the mTEC^hi-specific chromatin state by facilitating accessibility at tissue-specific loci prior to their expression. The mSWI/SNF or BAF complex represent one of about 30 non-redundant ATP dependent chromatin remodeling complexes with often instructive roles in developmental lineage specification and tumor suppression. BAF antagonizes Polycomb occupancy at target sites via direct physical interaction and ATP-dependent eviction^34,47. Considering the enrichment of Polycomb subunits and H3K27me3 at Aire-regulated loci in peripheral tissues and ES cells¹⁷, BAF likely employs similar mechanisms in mTECs to facilitate the poising of tissue-specific genes. In addition, BAF complexes associate with topoisomerase II (Top2) for the resolution of facultative heterochromatin^35,48,49. These observations are consistent with our studies, as the mTEC^hi-specific regulatory regions whose accessibilities are promoted by Brg1 showed enrichment of Top2a. Importantly, Aire inhibits topoisomerase II^18,50 suggesting a mechanism by which Aire could reduce accessibility and provide a repressive function.

Our studies indicate that Aire is unique in that it has, in the same protein, mediated by the same domains, both activating and repressive functions. These bifunctional domains may have evolved to cooperate with both repressors and activators to allow ectopic expression of tissue-specific antigens at levels generally several orders of magnitude lower than their expression in their peripheral tissues. Although thousands of tissue-specific genes become accessible to the transcription factors present in mTECs, these potential regulatory regions appear to be held in check by Aire’s ability to oppose accessibility. This opposition potentially guards tissue-specific genes from the hundreds of transcription factors (like NF-κB) expressed in mTECs to yield safe but immunologically effective levels of tissue-specific antigens in the thymus.

ONLINE METHODS

Mice

B6. Aire^+/− mice⁵ obtained from Jackson Laboratory were bred to generate Aire^−/− and Aire^+/+ littermates. Brg1^F/F and Foxn1^ex9cre mice^51,52 on C57BL/6 background were bred to generate Brg1^F/F and Foxn1e^x9cre;Brg1^F/F littermates. All experiments used female and male littermate controls. All mice were maintained in accordance with Stanford University’s Animal Care and Use Committee guidelines. Animal protocol is annually approved by the Administrative Panel on Laboratory Animal Care of the Research Compliance Office of Stanford University.

Cell Lines

Mouse embryonic fibroblasts (MEFs) were acquired from mice with the CiA:Oct4 allele at embryonic day 14.5 as previously described³³. MEFs were transformed with simian virus 40 large T antigen and single cell sorted after transfection with LGmCreER (self-deleting) plasmid⁵³ (Addgene #33340) to enrich for cells with excised neo cassette. Clones were screened for growth rate, VP16-mediated eGFP activation and DNase accessibility at the CiA:Oct4 locus. A single clone was used for all CiA recruitment experiments. MEFs were grown in high-glucose DMEM (Life Technologies, 11960) supplemented with 10% FBS (Omega Scientific, FB-11), 10 mM HEPES pH 7.5, Minimal AA, glutaMAX, Na Pyruvate, Pen/Strep, 2-Mercaptoethanol at 37° C., 5% CO₂. The 4D6 human thymic epithelial cell line³⁷ was a gift from Maria Toribio and Diane Mathis. The 4D6 cells were grown in RPMI 1640 (Life Technologies, 21870092) supplemented with 10% FBS (Omega Scientific, FB-11) and Pen/Strep at 37° C., 5% CO₂.

Flow cytometry

mTEC subsets were isolated as previously described^54,55. Thymi were dissected, capsules incised and triturated with glass pipettes to release thymocytes from stromal fragments. Stroma were digested with Liberase TM (Roche) and DNase I (Roche), and TECs were enriched using α-CD45 MACS microbeads (Miltenyi) or centrifugation on a Percoll PLUS gradient (GE Healthcare). Enriched TECs were stained with flurochrome-conjugated antibodies (BioLegend) against CD45 (30-F11), Ly-51 (6C3), MHC-II I-A/I-E (M5/114.15.2) and EpCAM (G8.8) along with fluorescein labeled UEA-I (Vector Labs) and DAPI (Life Tech). Thymocytes were stained with antibodies against TCRβ (H57–597), CD4 (RM4–5), CD8 (53-6.7), CD25 (PC61) and CD69 (H1.2F3). Spleen and lymph nodes were isolated, minced and stained with antibodies against TCRβ, CD4, CD8, CD25, CD44 (IM7) and CD62L (MEL-14). Intracellular staining for Aire (5H12) and Foxp3 (FJK-16s) were performed using the eBiosciences Foxp3 staining kit. All antibody stainings were preceded by FcγR block (2.4G2). Cell sorting was performed on FACS Aria II (BD), data collected using LSR II flow cytometer (BD) and analyzed using FACS Diva (BD) and FlowJo (Tree Star).

Gene Expression Profiling

mTECs were sorted into Trizol LS (Life Technologies), RNA was extracted according to manufacturer’s instructions, then purified using the RNeasy Micro Kit (Qiagen). Purified RNA was processed, amplified, labeled and hybridized to Affymetrix GeneChip MoGene 2.0 ST microarrays as previously described⁵⁶. Expression signals were normalized using the Signal Space Transformation-Robust Multi-Chip Average (SST-RMA) algorithm on the Affymetrix Expression Console software. Normalized signals were analyzed using the Transcriptome Analysis Console software (Affymetrix). Tissue restriction for each gene was determined using assignments from the previously reported dynamic step method¹⁷. Briefly, expression values for 64 non-thymic physiological samples from the GNF Mouse GeneAtlas V3^57,58 were calculated and hierarchically clustered into 35 groups. Guided by thresholds typically chosen for microarray data analysis, we define tissue restricted probes as those with a minimum normalized expression value of 50 that showed a moderated exponential step-up in expression, such that expression was substantially higher in 1–5 tissue groups than in the 6th highest tissue group. Only genes with unanimously tissue-restricted probe sets were designated as tissue-restricted.

ATAC-seq sample preparation

Sixty thousand cells were sorted into a V-bottom 2ml tube (E&K Scientific) and 1 ml of RSB buffer without detergent (10mM Tris 7.4, 10mM NaCl, 3mM MgCl₂) was added prior to pelleting by a swinging bucket centrifuge. Cells were resuspended in 200ul RSB buffer with 0.1% Tween-20 (Sigma-Aldrich) and incubated on ice for 5 min. Cells were pelleted, resuspended in 50ul TD buffer with 2.5ul Tn5 transposase (Illumina) and incubated at 37° C. for 30 min. Transposition reactions were cleaned up with MinElute columns (Qiagen) and libraries were constructed as previously described⁵⁹. Libraries were sequenced using paired-end, dual-index sequencing on an Illumina HiSeq2000 instrument with 51 × 8 × 8 × 51 cycle reads.

ATAC-seq data analysis

ATAC-seq data were processed as previously described^23,60. Reads were trimmed with a custom script^23,60, aligned using Bowtie2 and filtered for unique reads with alignment quality > q30. Reads mapping to mitochondrial, ChrY and unmapped contigs were removed. Peaks were called using MACS2 and filtered with a custom blacklist²³. Peak summits were extended +/− 250-bp and filtered for non-overlapping, maximally significant 500-bp peaks (n = 250,313). ATAC-seq fragment counts for each sample were calculated across all 250,313 peaks, quantile normalized and GC-content normalized as previously described^23,60. The nearest gene to each peak was annotated using HOMER. PCA was performed on normalized data using the MAVRIC R package (Moskowitz and Greenleaf, 2017, manuscript submitted) for the top 100,000 peaks with highest variance (ntop = 100000, alpha =0.01, effsize = 0.05, corMethod = “pearson”). Peaks with variance significantly correlated to a given PC (α < 0.05) were identified using MAVRIC [dimdesc($pcaobj)]. Hierarchical clustering was performed using Pearson correlation as the distance metric on normalized data. Normalized bedGraph files were created for each sample using bedtools genomecov to visualize insertion tracks. The sum of reads within peaks associated with “housekeeping” genes (median microarray signal > 75, coefficient of variation < 0.15, fold-change WT mTEChi vs. mTEClo > 0.9 and < 1.1) was used as the scaling factor for each sample as previously described⁶¹. Transcription factor (TF) deviation scores were calculated as described previously^23,60. TF footprinting was assessed by plotting the normalized distribution of the 5’ ends of fragments (replicates pooled) spanning a 300bp window relative to the motif center. Similarly, the normalized distribution of fragment dyads (midpoint) were plotted spanning a 4 kb window relative to indicated peak center or transcription start-site (TSS).

ChIP-seq data analysis

Fastq files for Aire, Pol II, Top1, Top2a, γ-H2AX and H3K27ac ChIP-seq¹⁸ from mTEC^hi; Lsd1, CoREST, Hdac2⁶², Suz12⁶³, Ring1b⁶⁴, LaminB⁶⁵ in ES cells were aligned to the mouse genome (mm9) with Bowtie using default settings except filtering reads with multiple alignments (command line parameter –m 1). Bam files were processed by removing duplicates, unaligned reads and those aligned to ChrM. Fragment sizes were estimated and reads were extended accordingly for each replicate. MACS2 was used to call peaks using -callpeak function with IgG ChIP as controls. Signal tracks were generated using bedtools genomecov and normalized by sequence depth. These signal track were cross-validated for significance using MACS2 with commands bdgcmp –m logLR –p 0.00001 to generate log10-likelihood bedGraph files. Enriched TF motifs for peak sets were identified using HOMER (findMotifsGenome.pl –size given –mask).

Chromatin in vivo Assay (CiA)

Mouse embryonic fibroblasts (MEFs) were acquired from mice with the CiA:Oct4 allele at embryonic day 14.5 as previously described³³. MEFs were transformed with simian virus 40 large T antigen and single cell sorted after transfection with LGmCreER (self-deleting) plasmid⁵³ (Addgene #33340) to enrich for cells with excised neo cassette. Clones were screened for growth rate, VP16-mediated eGFP activation and DNase accessibility at the CiA:Oct4 locus. A single clone was used for all CiA recruitment experiments. Embryonic stem cells (ESCs) from mice with CiA:Oct4 allele were established as previously described³³. Murine Aire with C-terminal Frb tandem repeat (Aire-Frb2x- V5) or Gal4-binding domain (Aire-Gal4BD-V5) were cloned into a previously described lentiviral backbone³³ for rapamycin-induced- or constitutive recruitment, respectively. BAF was recruited via the Ss18 subunit, using previously described constructs (Frb2x-V5-Ss18, ZFHD1-Ss18)^34,35. CiA anchor (ZFHD1-FKBP12) and VP16 (Gal4BD-VP16) constructs were described previously³³. Lentivirus was generated as previously described⁶⁶. MEFs and ESCs were infected and selected with puromycin (2ug/ml), blasticidin (10ug/ml) and/or hygromycin (200ug/ml). For experiments using chemical-induced proximity (CIP), rapamycin (Selleckchem) was added at 3nM. Media containing rapamycin (Rap) was changed daily for experiments spanning longer than 24h. For early time points (< 20 min), Rap at 12nM was added to media as well as to the fresh trypsin used to dissociate cells.

CRISPR-Cas9 guided recruitment

Catalytic-dead Cas9 (dCas9) mediated recruitment was performed as previously described³⁶. Three small guide RNAs (sgRNAs) targeting a DNase hypersensitivity site (DHS) 200 bp upstream of the PSMB11 TSS were selected based on scores for off-target matches as previously described⁶⁷. These sgRNAs were cloned into the sgRNA(MS2)_zeo lentiviral backbone (Addgene #61427) containing MS2-specific hairpin aptamers. Murine Aire with C-terminal MS2 (Aire-MS2-V5) was cloned into a previously described lentiviral backbone³³. 4D6 human thymic epithelial cell line was sequentially transduced with lentiviruses encoding the sgRNAs and dCas9 or the sgRNAs, dCas9 and Aire-MS2.

Chromatin Immunoprecipitation (ChIP)

ChIPs were essentially performed as previously described^34,35,47. Cells were dissociated, washed and 30 million cells were formaldehyde crosslinked (1%) at 37° C. for 12 min. For experiments using CIP (< 20 min), Rap at 3nM was added to media quenching trypsin, PBS wash and fix buffer. Nuclei were sonicated using Covaris E220 ultrasonicator at 5% duty cycle, 4 intensity, 140W PIP and 200 cycles per burst for 13 min. Insoluble chromatin was pelleted, supernatant diluted 1:1 in 2× ChIP buffer (100 mM HEPES pH 7.5, 600mM NaCl, 2 mM EDTA pH 8.0, 2% Triton X-100, 0.2% sodium deoxycholate, 0.1% SDS) and divided into four ChIP reactions (4–5 ug antibody/reaction, α-V5, clone R960-25, Life Technologies) which incubated overnight at 4° C. 25 ul of Protein G Dynabeads (Thermo Fisher) slurry washed in ChIP buffer were added to ChIP reactions and rotated at 4° C. for 1h. Beads were washed 3× in ChIP buffer, 1× in DOC buffer and 1× in TE before consecutive elutions (2×) in 150 ul of 0.1M NaHCO₃, 1% SDS. Eluates were subjected to RNase A and proteinase K before reverse cross-linking at 65° C. overnight. ChIP DNA was precipitated using phenol/chloroform extraction and reconstituted in TE buffer for qPCR reactions. Amplicon detection for each target region was normalized to that at respective control regions for all samples. Enrichment was calculated to be the fold-change between the normalized bound/input values of Rap- treated vs. ethanol treated samples. Primers used for ChIP studies at CiA:Oct4 locus are summarized in Supplementary Table 1^33,34. V5 ChIP at CiA:Oct4 locus was normalized to that at the housekeeping Rps29 promoter (c).

DNase Accessibility Assay

DNase I sensitivity assays were performed as previously described⁶⁸. Cells were lysed in hypotonic Buffer A, washed in Buffer A and 5 million nuclei were pelleted for each DNase I reaction. For experiments using CIP (< 20 min), Rap at 3nM was added to media quenching trypsin, PBS wash and Buffer A. Nuclei were subjected to varying concentrations of DNase I (Sigma) for 3 min at 37° C. Digestions were terminated with Stop buffer and subjected to proteinase K for 1h at 55° C. Samples were treated with RNase A and DNA was purified over MinElute columns (Qiagen) for qPCR reactions. Amplicon detection at CiA:Oct4 locus (Supplementary Table 1) for all DNase I conditions was normalized to that at a DNase I-insensitive region at the Rho locus (Supplementary Table 1). Amplicon detection at promoter of PSMB11 locus for all DNase I conditions was normalized to that at DNase I-insensitive region at the RHO locus (Supplementary Table 1).

Histopathology and Immunohistochemistry

Histopathology was performed as previously described⁴⁰. Briefly, tissues were fixed in buffered 10% formalin and paraffin-embedded. Hematoxylin and eosin stainings were performed using standard methods. Immunohistochemistry was performed on 4-um sections with the ABC Vectastain kits (Vector Laboratories) using α-CD3 antibody (DakoCytomation, cat#A045229), and developed with DAB.

Thymus Transplants

Kidney capsule thymus transplants were performed as previously described⁶⁹. Briefly, thymi from newborn Brg1cKO and WT littermates were cultured in 1.35 mM 2-deoxyguanosine for 7 days to deplete hematopoietic compartments. Thymic stroma were washed and transplanted under the kidney capsule of 6–8 week-old female nude mice. Thymopoiesis was monitored via cytofluorimetric analysis of blood at 5 and 10 weeks post-transplantation. Animals were examined 13 – 15 weeks after transplantation for T cell reconstitution and peripheral organs were collected for histopathology.

Statistics

Fisher’s exact tests for over-representation in fold-change vs. fold-change plots were one-sided tests with 95% confidence intervals for the lower bound of odd-ratios (Fig. 2c, 5b). For cumulative distribution function plots, Mann-Whitney-U tests were used for significance between comparisons with 95% confidence intervals at medians of pairwise comparisons (Fig. 1c,3b,5a and Supplementary Fig. 2d,4). Correlations between different peak sets were performed using Pearson correlation test (Fig. 1e, 2a). Student t-tests were performed as two-tailed tests with n specified for independent experiments in individual figure legends (Fig. 5,6,7, and Supplementary 6).

Data and Code Availability

ATAC-seq and microarray datasets reported in this manuscript, including raw reads and fully processed count matrices can be accessed in GEO with accession codes GSE102526 and GSE102525, respectively. Publicly available ChIP-seq and ATAC-seq datasets referenced in this study are as follows: GSE92597 (ChIP-seq) for Aire, Pol II, Top1, Top2a, γ-H2AX, H3K27ac in mTECs; GSE39513 (ChIP-seq) for Suz12 in ES cells; GSE42466 (ChIP-seq) for Ring1b in ES cells; GSE28247 (ChIP-seq) for LaminB in ES cells; GSE27841 (ChIP-seq) for Lsd1, CoREST, Hdac2 in ES cells; GSE94041 (ATAC-seq) for ES cells. Any custom code will be made available upon request.