DNA breaks and chromatin structural changes enhance the transcription of autoimmune regulator target genes

The autoimmune regulator (AIRE) protein is the key factor in thymic negative selection of autoreactive T cells by promoting the ectopic expression of tissue-specific genes in the thymic medullary epithelium. Mutations in AIRE cause a monogenic autoimmune disease called autoimmune polyendocrinopathy-candidiasis-ectodermal dystrophy. AIRE has been shown to promote DNA breaks via its interaction with topoisomerase 2 (TOP2). In this study, we investigated topoisomerase-induced DNA breaks and chromatin structural alterations in conjunction with AIRE-dependent gene expression. Using RNA sequencing, we found that inhibition of TOP2 religation activity by etoposide in AIRE-expressing cells had a synergistic effect on genes with low expression levels. AIRE-mediated transcription was not only enhanced by TOP2 inhibition but also by the TOP1 inhibitor camptothecin. The transcriptional activation was associated with structural rearrangements in chromatin, notably the accumulation of γH2AX and the exchange of histone H1 with HMGB1 at AIRE target gene promoters. In addition, we found the transcriptional up-regulation to co-occur with the chromatin structural changes within the genomic cluster of carcinoembryonic antigen-like cellular adhesion molecule genes. Overall, our results suggest that the presence of AIRE can trigger molecular events leading to an altered chromatin landscape and the enhanced transcription of low-expressed genes.

Immunological tolerance protects against harmful immune responses to endogenous body constituents (1). The initial stages of T cell tolerance occur in the thymus, where thymic stromal cells present self-peptides to thymocytes (2). The repertoire of self-peptides and the potential to detect and eliminate autoreactive thymocytes is critically dependent on the autoimmune regulator (AIRE) 3 protein. In thymic medullary epithelial cells, AIRE enhances the expression of genes encoding the tissue-specific antigens (TSAs) that are restricted to peripheral tissues (3). In humans, mutations in the AIRE gene cause an autosomal recessive disease known as autoimmune polyendocrinopathy-candidiasis-ectodermal dystrophy (APECED), which is associated with destructive autoimmune reactions against multiple (primarily endocrine) organs (4,5). The monogenic nature of the disease has rendered AIRE deficiency an important model for studying the mechanisms of autoimmunity and central tolerance. The mechanism by which AIRE targets and regulates the transcription of a broad repertoire of TSA genes remains largely unknown. AIRE enhances transcription and interacts with multiple proteins implicated in the modification of chromatin structure, activation of transcription, and mRNA splicing (6,7). At the epigenetic level, it acts via binding to hypomethylated H3K4 residues (8,9), although a later study demonstrated a broader AIRE protein presence on genomic sites, linking RNA polymerase II enrichment at transcriptional start sites with AIRE binding (10).
According to the proteomic screens, a top-ranked AIRE interaction partner is topoisomerase II␣ (TOP2A), which, together with its paralog TOP2B, plays a significant role in relaxing torsional tensions of the DNA double helix (6). Eukaryotic TOP proteins are large homodimeric enzymes that generate transient DNA breaks that result in the removal of positive and negative DNA supercoils (11). The TOP2 proteins can be blocked at different functional stages by various small-molecule inhibitors such as etoposide and merbarone (12). Studies in various model systems have demonstrated a role for TOP2 in regulating transcription (13)(14)(15), and Abramson et al. (6) linked its catalytic activity to the changes in the expression of AIRE target genes.
In this study, we applied the RNA-seq approach to systematically analyze the effect of etoposide, which specifically blocks the religation activity of the TOP2 enzymes, in the doxycycline-inducible AIRE-Tet cell line (16). We found that low-dose treatment of AIRE-expressing cells with etoposide results in a robust and synergistic up-regulation of AIRE target genes. Furthermore, we observed the up-regulation of an additional large set of genes with low expression levels, the characteristic feature of genes activated by AIRE. Etoposide treatment did not enhance AIRE-dependent gene expression after TOP2A depletion by RNAi, nor could it enhance transcription when the AIRE CARD domain contained a patient mutation. Furthermore, the CARD domain mutation disrupted the protein interaction between AIRE and TOP2A, as assayed by co-IP. Additionally, we observed AIRE-related structural rearrangements in the chromatin structure, such as the enrichment of ␥H2AX and HMGB1, and the reduction of histone H1 at AIRE target gene promoters. In light of these findings, we performed FAIRE-seq to assess AIRE-related genome-wide changes of chromatin accessibility and found that a large fraction of differentially enriched FAIRE peaks overlap with CTCF sites. Finally, we tested whether this overlap implies that the expression of AIRE could entail changes in CTCF-mediated chromatin interactions using the 3C method, and we were able to determine altered chromatin looping in the CEACAM gene cluster.

The TOP2 inhibitor etoposide enhances the expression of AIRE target genes
In an earlier study, etoposide, a drug that specifically stabilizes the TOP2-DNA complex and promotes DNA breaks, was demonstrated to induce the expression of AIRE target genes in HEK293 cells. However, the study did not find an additive or synergistic effect of AIRE and etoposide (6). In contrast to the previous study, we detected a strong synergistic increase in target gene S100A8 and IVL transcription with low-dose etoposide (2 M) treatment of doxycycline-induced, AIRE-positive AIRE-Tet cells after 2 h of treatment (supplemental Fig. 1A). After 24 h of 2 M etoposide treatment, this up-regulation reached an even higher increase in expression: over 8000-fold and 150-fold for S100A8 and IVL genes, respectively (Fig. 1A). Treatment with etoposide did not cause apoptosis in AIRE-Tet cells (supplemental Fig. 1B), and the low concentration of etoposide (2 M) induced the largest change in expression, as we could not establish further increase with higher etoposide concentrations (10 and 50 M) (supplemental Fig. 1A). We noticed, in agreement with Abramson et al. (6), that etoposide alone (Etop) was able to increase, albeit to a lesser extent (4-, 7-, and 3-fold, respectively), the expression of AIRE-dependent S100A8 and IVL and AIRE-independent PSMD4 genes (Fig. 1A). These results suggest a strong synergistic effect of AIRE and low-dose etoposide on target gene expression, which is much higher than etoposide alone. We performed subsequent experiments by treating cells with 2 M etoposide for 24 h, i.e. under conditions where we observed the strongest synergistic effect.

Genes that are synergistically activated by AIRE and the TOP2 inhibitor have low expression levels
The observation that AIRE induction and etoposide strongly enhanced the transcription of select genes posed the question of whether immobilizing TOP2 enzymes in the DNA-cleavable complex could induce AIRE-dependent gene transcription on a genome-wide scale. To assess the impact of etoposide treatment on the global level of gene expression, we performed RNA-seq with AIRE-Tet cells. Consistent with earlier results (16,17), doxycycline (Dox)-induced AIRE was responsible for the up-regulation of otherwise low or moderately expressed genes in AIRE-Tet cells (Fig. 1B, top panel). AIRE up-regulated 691 of 692 differentially expressed (DE) genes and thus acted almost exclusively as an activator. Strikingly, AIRE activation together with 2 M etoposide led to substantial genome-wide transcriptional activation, as altogether 6118 genes were activated in the DoxϩEtop-treated cells (Fig. 1, B, center panel, and C). AIRE together with etoposide still predominantly acted as an activator, as 73% of DE genes were up-regulated and 27% of genes (2215) were down-regulated in DoxϩEtop cells. The finding that AIRE had a mostly up-regulating effect is in agreement with previous RNA-seq with sorted primary thymic epithelial cells from Aire-deficient and wild-type mice (18,19). We found that a large proportion of the genes activated by AIRE and etoposide were also normally expressed at low levels (81% of the 4506 genes in the DoxϩEtop only group, Fig. 1C). In contrast, etoposide treatment alone influenced the expression of large number of genes with various levels of expression, and the numbers of up-and down-regulated genes were approximately similar (56% and 44% of the 4413 genes, respectively) (Fig. 1B, bottom panel).
As expected, AIRE primarily activated protein-coding genes (93%), in agreement with its role to activate self-antigen proteins to be presented by thymic epithelial cells, whereas etoposide treatment increased the fraction of pseudogenes and RNA genes (supplemental Fig. 2A). We did not identify non-canonical transcription start sites among AIRE target genes (data not shown), in contrast to an earlier report (20). In agreement with our initial qPCR results, the RNA-seq data analysis confirmed that the S100A8 and IVL genes were up-regulated under Dox and DoxϩEtop conditions (Fig. 1D). In subsequent experiments, we included an additional subset of AIRE-activated genes (HBG2, PDYN, DMBT1, KRT73, and CEACAM5) for analysis. The AIRE-mediated expression of these genes was enhanced by etoposide, as demonstrated by the RNA-seq experiment (Fig. 1D) and validated by qPCR (data not shown).

AIRE with etoposide enhanced the activation of tissue-specific genes and gene clusters
In further analyses of AIRE-regulated genes, we found that 99.4% of genes up-regulated by Dox were also up-regulated in DoxϩEtop samples (687 of 691) ( Fig. 2A). This almost complete inclusion of Dox up-regulated genes within the DoxϩEtop group was remarkable, in contrast to the overlap between the Dox and Etop gene lists that shared 123 genes ( Fig. 2A). Thus, although AIRE and etoposide independently regulate distinct subsets of genes, the addition of etoposide to AIRE-expressing cells strongly enhances AIRE transcriptional activity. The genes up-regulated by AIRE alone showed a further, ϳ4-fold increase in the presence of AIRE and etoposide (Fig. 2B, DoxϩEtop). In contrast, overall there was no increase in the expression of AIRE target genes after treatment with etoposide only (Fig. 2B), although few AIRE target genes (including S100A8 and IVL) showed an elevation in expression, as seen in Fig. 1A.
Consistent with previous studies, most of the AIRE-induced genes were associated with tissue-specific expression patterns (79%), and this phenomenon was largely preserved after etoposide co-treatment (75%, Fig. 2C). However, only 32% of etoposide alone-activated genes were associated with a tissue-specific expression pattern (Fig. 2C), which is comparable with the overall proportion of tissue-specific genes (34%) determined by the Human Protein Atlas Project. A similarly low percentage of tissue-specific genes was observed in the down-regulated gene groups (22% and 37% in DoxϩEtop and Etop, respectively) (Fig. 2C).
The genes activated by AIRE in Dox and DoxϩEtop cells were enriched for known APECED autoantigens (6 of 692 and 25 of 8333 genes, respectively, of 31 autoantigens reported in Ref. 4) compared with four autoantigens among the 4413 DE genes under Etop conditions (p ϭ 4 ϫ 10 Ϫ5 and 0.03, respectively). We observed highly significant chromosomal clustering among 21% of AIRE-regulated genes, with an average of 4.5 genes/cluster. In comparison, 27% of DE genes that were only in DoxϩEtop cells were in clusters (on average 4.2 genes/cluster). Only 14.2% of DE genes in Etop cells were in clusters, which was significantly less than in Dox or DoxϩEtop (p ϭ 5.3 ϫ 10 Ϫ6 and p Ͻ 2.2 ϫ 10 Ϫ16 , respectively), and these clusters tended to be smaller in gene numbers (on average 3.5 genes/cluster).

AIRE regulates alternative splicing in a different subset of genes
AIRE interacts with the proteins involved in splicing (6,7), and recent studies demonstrated AIRE-dependent differential inclusion of exons in transcripts whose overall levels were only slightly affected or unaffected by AIRE (18,21). In our RNA-seq data, we found that the number of genes featuring differential exon usage (DEU) following AIRE induction was in approximately the same range (568 genes) with DE genes. Etoposide treatment increased the number of DEU genes both in Doxϩ Etop and Etop cells, but the increase was very similar (1989 and 1850 DEU genes, respectively) ( Fig. 2D). Furthermore, a large fraction of the DEU genes found in Dox cells overlapped with DEU genes both in DoxϩEtop and Etop cells (428 of 568 and 406 of 568, respectively) ( Fig. 2D), which suggests that the DEU genes are not synergistically affected by AIRE and etoposide. Additionally, there was very little overlap between the DE and DEU genes in Dox samples (Fig. 2E). The overlap was larger in DoxϩEtop cells (Fig. 2E), but this was mostly due to a greater number of DEU genes the DoxϩEtop cells share with Etop cells. In general, the alternatively spliced genes tended to be highly expressed (Fig. 2F) and were less likely to represent tissue-specific genes compared with AIRE target genes (supplemental Fig.  2B). However, it should be noted that the initial low expression of AIRE target genes might hinder the analysis of alternative splicing.

Gene activation requires the interaction of TOP2A with intact AIRE
To confirm that TOP2A is directly involved in enhancement of AIRE target gene activation, we used shRNA to knock down TOP2A expression in AIRE-Tet cells (Fig. 3A). The effect of gene transcription was significantly reduced in cells transfected with a TOP2A-specific shRNA (shTOP2A4) compared with the scrambled control shRNA (Scrambled) that retained the ability to enhance AIRE target gene expression (Fig. 3B).
We also wanted to demonstrate that functional AIRE is required for the enhancement of gene activation by etoposide. For this, we investigated the effect of etoposide on the transcription-enhancing activity of AIRE carrying the L28P APECED mutation in the CARD domain and compared it with the intact AIRE protein. We used stably transfected HEK293 cells constitutively expressing wild-type AIRE, the AIRE-L28P mutant protein, or YFP as a negative control. In a previous study, we demonstrated that, although AIRE-L28P localizes to the nucleus, the mutated protein lacks the speckled pattern observed with wild-type AIRE and is unable to activate AIRE target genes (22). Following etoposide treatment of AIRE-L28P cells, we did not observe changes in expression of most of the genes studied, whereas cells expressing wild-type AIRE reacted to etoposide treatment by enhancing the transcription of AIRE target genes similarly to the doxycycline-induced AIRE in AIRE-Tet cells (Fig. 3C). Although the intact AIRE protein in doxycycline-induced AIRE-Tet or in stably transfected HEK293 cells was able to interact with TOP2A with or without etoposide addition (Fig. 3, D, second and third lanes, and E, second and fifth lanes), the mutant AIRE-L28P was unable to interact with TOP2A (Fig. 3E, third and sixth lanes).

AIRE-mediated gene activation utilizes the DNA cleavage activity of both type 1 and 2 topoisomerases
Our results prompted us to confirm that 2 M etoposide is sufficient to significantly affect the chromosomal DNA integrity as a determinant of AIRE-mediated transcriptional activity. For this, we treated AIRE-Tet cells with doxycycline and etoposide and analyzed the Br-dUTP-labeled DNA breaks by flow cytometry. Additionally, we treated the cells with 1 M camptothecin, a topoisomerase I (TOP1)-specific inhibitor that, similarly to etoposide, freezes the TOP-DNA cleavable complex; 1 M merbarone, a TOP2 inhibitor that blocks enzymatic activity before the DNA cleavage reaction; and 1 M ␤-lapachone, a TOP1 inhibitor that blocks enzymatic activity before the DNA cleavage reaction (Fig. 4A). In this way, two of the inhibitors studied, etoposide and camptothecin, cause TOP2-and TOP1related DNA breaks, respectively, and two other inhibitors, merbarone and ␤-lapachone, inhibit TOP2 and TOP1, respectively, without causing DNA breaks. We observed an increase, although not significant, in the percentage of cells containing DNA breaks in doxycycline-induced cells after treatment with etoposide and camptothecin (Fig. 4B). As expected, and in agreement with the inhibitory mechanism of topoisomerases before DNA cleavage, neither merbarone nor ␤-lapachone had an effect on the occurrence of DNA lesions (Fig. 4B). Based on these findings, we tested the effect of TOP1 inhibitors on AIREdependent gene expression and found that camptothecin treatment essentially mimicked that of etoposide by strongly and . Bottom panel, a representative Western blot of the TOP2A knockdown experiment using anti-TOPa and anti-␤-actin antibodies as described under "Experimental Procedures." B, expression analysis of AIRE-dependent genes (S100A8, IVL, HBG2, KRT73, and DMBT1) and an AIRE-independent PSMD4 gene in AIRE-Tet cells using qPCR after shRNA-mediated knockdown of TOP2A. Log 10 -transformed data points together with their mean Ϯ S.D. are from three to six independent experiments. Statistical significance was assessed by two-sample t test (*, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001). C, expression analysis of AIRE-dependent genes (S100A8, IVL, HBG2, KRT73, and DMBT1) and an AIRE-independent gene (PSMD4) in HEK293 cells stably expressing YFP, WT AIRE, or mutant AIRE (L28P). Log 10 -transformed data points together with their mean Ϯ S.D. are from three to six independent experiments. Statistical significance was assessed by two-sample t test (*, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001). D, analysis of the AIRE and TOP2A interaction in Ctrl, Dox, and DoxϩEtop AIRE-Tet cells by co-IP. E, co-IP analysis of the AIRE and TOP2A interaction in HEK293 cells stably expressing YFP, WT AIRE, or mutant AIRE (L28P) with or without etoposide treatment. In both co-IP experiments, anti-TOP2A antibody was used for immunoprecipitation, and anti-TOP2A and anti-AIRE antibodies were subsequently used in the Western blot. Input was used as a protein loading control, and immunoprecipitation with anti-IgG antibody was used to determine the background binding of AIRE.
significantly up-regulating AIRE target genes (Fig. 4C), suggesting that both types of topoisomerases can influence AIRE-related transcriptional processes via DNA breaks. Although the role of TOP1 with respect to the function of AIRE remains to be elucidated, TOP1 is known to support the expression of genes with low transcription levels (23).

Induction of AIRE causes structural changes in chromatin at AIRE target gene promoters
Consistent with the increased proportion of cells with DNA breaks, we observed a significant enrichment of ␥H2AX, a known marker of DNA double-stranded breaks (24), at the promoter regions of the AIRE-dependent S100A8 and IVL genes in DoxϩEtop cells compared with Ctrl and Etop cells (Fig. 5A). The ␥H2AX signal was weaker in control regions located ϳ10 kb away from IVL and S100A8 promoters (Fig. 5A).
The TOP2A-induced DNA breaks generally attract and activate the poly(ADP-ribose) polymerase 1 (PARP1) enzyme (13). Activated PARP1 can use histone H1 as a substrate for poly(ADP-ribosyl)ation, thereby excluding it from a subset of PARP1-containing promoters (25). Histone H1 is widely viewed as a transcriptional repressor (26), and removal of histone H1 or exchange of H1 with the high-mobility group protein B1 (HMGB1) facilitates gene expression (27,28). We investigated whether the transcriptional activation observed with AIRE, especially in the presence of etoposide, is accompanied by changes in the abundance of histone H1 and HMGB1 by performing ChIP analyses using histone H1 and HMGB1-specific antibodies at S100A8 and IVL promoters. Our results indicated that, in the presence of AIRE, the linker histone H1 signal was significantly reduced at AIRE target gene promoters compared with Ctrl, and it further declined after treatment with etoposide (DoxϩEtop) (Fig. 5B). Additionally, we found that more HMGB1 was recruited to AIRE target gene promoters in Dox and DoxϩEtop cells (Fig. 5B). In contrast, the HMGB1 signal at the promoters of AIRE-independent PSMD4 and GAPDH genes was consistently higher than histone H1 under all experimental conditions, in agreement with their high expression in AIRE-Tet cells (Fig. 5B).

AIRE induces changes in CTCF-mediated chromatin looping in the CEACAM gene cluster
Topoisomerase and RNA polymerase activities remodel DNA supercoiling, influencing the folding of large-scale chromatin structures and creating domains that are flanked by CTCF insulator protein binding sites (28). To address the genome-wide changes in chromatin structure induced by AIRE and etoposide, we adopted the FAIRE-seq approach, which allowed us to analyze chromatin accessibility in AIRE-Tet cells and compare this in AIRE-induced and etoposide-treated cells. We found changes in ϳ1000 and 1900 genomic regions in Dox and DoxϩEtop cells, respectively, whereas, in Etop cells, we could not detect any FAIRE peaks that were differentially accessible compared with Ctrl. Interestingly, Ͼ90% of the FAIRE peaks in Dox-and DoxϩEtop-treated cells became less accessible, and 42-70% of the detected changes occurred in gene promoters, as assessed by the overlapping of FAIRE peaks with genic and intergenic features (Fig. 6A). However, very few of the differentially accessible regions in Dox and DoxϩEtop cells were in the vicinity of AIRE-activated genes (Fig. 6B). A significant fraction of the changed regions in Dox and DoxϩEtop cells overlapped with HEK293-specific chromatin architectural protein CTCF binding sites (Fig. 6C). As the induction of AIRE changed chromatin accessibility in CTCF binding sites, we next focused on a 310-kb genomic region on chr19 containing CEACAM genes to investigate the potential impact of AIRE on CTCF binding and CTCFmediated chromatin interactions (Fig. 6D). The genomic region contains CEACAM3, CEACAM4, CEACAM5, CEACAM6, CEACAM7, and CEACAM21 genes, which are co-expressed with AIRE in human medullary thymic epithelial cells (29,30). The cluster structure together with RNA-seq, FAIRE-seq, and CTCF binding data is depicted in Fig. 6D. Our RNA-seq data showed CEACAM genes as AIRE-activated genes in AIRE-Tet cells, prompting the analysis of genomic changes in 24 CTCF binding sites within this locus by chromosome conformation capture (3C) assay (31). The 3C assay combines protein crosslinking and proximity ligation of DNA to detect long-range chromatin interactions between genomic loci. We set a FAIRE peak (site 10 between CEACAM5 and CEACAM7 genes, Fig.  6E) that overlapped with the CTCF binding site and became less accessible after AIRE induction as the anchor region for the 3C experiments. The chromatin interactions between the anchor region and the remaining 23 CTCF sites were analyzed by qPCR using primers that corresponded to Sau3AI-specific restriction fragments that either overlapped or flanked the CTCF sites. We compared the 3C signal from Dox cells with Ctrl AIRE-Tet cells and found that, in AIRE-expressing cells, the anchor CTCF site formed stronger chromatin contacts with sites 1 and 4 flanking CEACAM21 and with site 12 near CEACAM5. In addition, contacts with sites 3 and 19 close to the CEACAM21 and CEACAM3 genes, respectively, became significantly weaker in AIRE-expressing cells. This suggests that AIRE affects long-range restructuring of chromatin and determines the partitioning of the CEACAM cluster into chromatin loops.

Discussion
Topological and structural changes of chromatin play a critical role in the regulation of gene expression and in promoting the accessibility of genomic DNA. DNA topoisomerases, the key enzymes that affect chromatin architecture and influence transcription, control the topological state of genomic DNA (11). Among the various topoisomerases, TOP2A has been shown to interact with AIRE (6). Here we report an analysis of chromatin-related molecular events in the context of AIREdependent gene activation that involve the activities of the topoisomerases.
In this study, we describe the genome-wide effect of etoposide treatment in the doxycycline-inducible AIRE-Tet cell line. In contrast to a previous study (6), we found a strong synergistic effect on transcription when AIRE-expressing cells were treated with low-dose etoposide. We observed a remarkable increase in the number of genes expressed and in the fold difference of individual genes. The genes that were synergistically activated displayed features of AIRE target genes, including low expression, tissue-specific expression under physiological conditions, and chromosomal clustering (16,18,(32)(33)(34). Subsequent analyses of the AIRE-expressing AIRE-Tet cells revealed another distinct set of genes that was prone to alternative splicing in AIRE-induced cells. Earlier studies with primary thymic cells have established that alternative exon usage affects central tolerance (35) and that AIRE knock-out mouse thymic epithelial cells have a different gene splicing pattern than wildtype cells (18,21). However, although treatment of AIRE-expressing cells with etoposide increased the number of alternatively spliced genes, these genes did not display features of typical AIRE-regulated genes, and they largely overlapped with genes that became alternatively spliced in the presence of etoposide alone.
Transcription is associated with severe topological perturbations of DNA, and locally targeted TOP2-associated DNA breaks generate a permissive chromatin setting that facilitates A, assessment of DNA breaks as measured by the enrichment of the Ser 139 -phosphorylated histone H2AX (␥H2AX) in AIREdependent S100A8 and IVL gene promoters and distal control regions in Ctrl, Dox, DoxϩEtop, and Etop AIRE-Tet cells using qPCR. The box and whiskers plots show the median and interquartile range of log2-transformed data from three independent experiments. Whiskers cover data points within a 1.5 ϫ interquartile range. Statistical significance was assessed by two-sample t test (*, p Ͻ 0.05; **, p Ͻ 0.01). B, qPCR analysis of H1 and HMGB1 enrichment at AIRE target gene (S100A8, IVL, and DMBT1) and AIRE-independent gene (PSMD4 and GAPDH) promoters in Ctrl, Dox, DoxϩEtop, and Etop AIRE-Tet cells. IgG was used as a negative control for ChIP experiments. The box and whiskers plots show the median and interquartile range of log2-transformed data from five independent experiments. Whiskers cover data points within a 1.5 ϫ interquartile range. Statistical significance was assessed by two-sample t test (*, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001).
gene transcription (11). The interaction between AIRE and TOP2A, which has been determined by co-immunoprecipitation in Ref. 6 and in our study, appears to be one key event in AIRE target gene expression, as TOP2A knockdown significantly reduced the expression of AIRE target genes. Furthermore, the APECED L28P mutation in the AIRE CARD domain abolished the multi-fold induction of AIRE target genes, which might be partially explained by the disruption of the AIRE-TOP2A interaction, although the direct proof for this remains to be shown. These results suggest that DNA breaks induced in the regions of AIRE target genes are needed to activate the transcription of these genes. As etoposide enhances the AIRE and TOP2A interaction and stabilizes TOP2A-DNA complexes, it is conceivable that an increased presence of DNA breaks in AIRE gene promoters would facilitate positive transcription elongation factor b (P-TEFb) recruitment to these loci and activate their transcription, as reported earlier (36,37). In this respect, the recruitment of the bromodomain protein BRD4 to target genes was shown to connect AIRE and P-TEFb to release paused promoter-proximal RNA polymerase II and to induce AIRE-dependent transcripts (38). Interestingly, AIRE-dependent transcription appears to be equally well stimulated by low concentrations of camptothecin, which, analogously to etoposide, freezes the TOP1-DNA cleavable complex covalently to the chromatin. How TOP1 enhances AIRE-mediated gene regulation will have to be elucidated in subsequent experiments. Nevertheless, the idea that single-stranded DNA breaks may play a role at least in the early steps of AIRE-depen- Figure 6. FAIRE-seq and 3C reveal AIRE-dependent changes in chromatin structure. A, percentages of FAIRE peaks that overlap with genic and intergenic regions in Dox and DoxϩEtop AIRE-Tet cells relative to Ctrl. The x axis labels refer to the FAIRE peaks that show either loss or gain of chromatin accessibility (Closed and Open, respectively). The numbers of the peaks are shown in brackets. The overlaps between the peaks and promoters up to 3 kb from the transcription start site were assessed with the relative distance test implemented in the GenometriCorr R package. B, distance between DE genes and the closest FAIRE peaks in Dox and DoxϩEtop AIRE-Tet cells. The statistical analysis did not determine a significant overlap between the genomic regions. C, differentially enriched FAIRE peaks show a strong enrichment of CTCF sites in AIRE-Tet cells both with Dox and DoxϩEtop treatment. The numbers of FAIRE peaks in the overlap are shown on the columns. The statistical significance of the overlaps between the FAIRE-seq peaks and HEK293-specific CTCF sites was assessed with the relative distance test implemented in the GenometriCorr R package. D, annotated overview of the genomic region that comprises the CEACAM cluster at chr19:42068524-42378450. The tracks below the gene annotations show the AIRE-Tet-specific gene expression (RNA-seq), chromatin accessibility (FAIRE-seq), differentially enriched FAIRE peaks, HEK293-specific DNase sequencing, and CTCF ChIP sequencing peaks (ENCODE) and computationally predicted CTCF loops (32). E, the chromatin interactions of the 24 CTCF sites in the CEACAM gene cluster depicted in D were analyzed by 3C. The interaction frequencies were quantified by qPCR and are shown as log 2 -fold changes between the signals from Dox and Ctrl AIRE-Tet cells. CTCF site 10 was used as an anchor region in the qPCR assays. The data show the mean Ϯ S.D. from five independent experiments. Statistical significance was assessed by one-sample t test comparing the log2 -fold changes with the reference value of 0 (Ⅲ, p Ͻ 0.1; *, p Ͻ 0.05). dent gene expression in etoposide-treated cells would be feasible, given that the low concentration of etoposide predominantly results in single-stranded DNA lesions (39). Importantly, our results highlight the functional difference of the inhibitors (regardless of their targeting of TOP1 or TOP2) between those that generate DNA breaks and bind the protein covalently to DNA (etoposide and camptothecin) and compounds that inhibit TOP catalytic activity but do not generate the breaks (merbarone and ␤-lapachone). It should be noted that, in addition to TOP2, the TOP1 activity in association with transcription is well known (11), and a recent study demonstrated that the initiation of transcription requires RNA polymerase II-mediated activation of the TOP1 activity at transcription start sites by BRD4 (40). Thus, it is very likely that both TOP2 and TOP1 inhibitors that induce DNA breaks enhance AIRE target gene expression by similar mechanisms and transcriptional complexes.
Our ChIP experiments demonstrated that ␥H2AX, a marker of double-stranded DNA breaks (24,41), binds the promoter regions of AIRE-activated genes. AIRE co-immunoprecipitates with ␥H2AX and DNA-PK, and DNA-PK was found to be essential for the recruitment of AIRE to gene promoters (42,43). We detected the exchange of histone H1 with the HMGB1 protein at target gene promoters. The enrichment of HMGB1 is an indicator of a more relaxed chromatin environment and associated with actively transcribed genes, whereas histone H1 is crucial for the proper compaction of the chromatin structure (13,28). This exchange requires the enzymatic activity of PARP1, which is part of the DNA repair complex, including the aforementioned DNA-PK, and is stimulated by the TOP2-generated DNA breaks. These interrelated associations were further strengthened by the observation that etoposide treatment of AIRE-expressing cells caused almost complete replacement of histone H1 with HMGB1. Altogether, these results suggest that the effect of etoposide is associated with DNA breaks at AIRE target genes. Although we were able to show the presence of relatively high levels of DNA breaks and break-associated proteins, it has been difficult to demonstrate a substantial increase of AIRE at the target genes. Then how do we explain the relatively low binding of AIRE at target genes? One explanation would be that AIRE interaction with chromatin could occur only at early stages of transcriptional activation or via other proteins such as histone H3, as shown earlier, making the detection of AIRE difficult by the ChIP method. Alternatively, the modest increase in AIRE at target genes versus the higher abundance of DNA break-associated proteins might be explained by the recent findings that TOP-induced breaks can be long-lasting, in contrast to AIRE interaction with chromatin. Earlier studies have identified genomic regions that are targeted by using TOP poisons such as etoposide and camptothecin, although their chromatin features are still not understood (12). In addition, the enrichment of ␥H2AX at activated genes after etoposide treatment in AIRE-expressing cells could be attributed to the action of the passing RNA polymerase II that denatures the topoisomerase-DNA cleavable complexes (44). For more in vivo settings, ChIP sequencing studies in cultured neuronal cells found increased ␥H2AX signals largely confined to actively transcribed genes but not to Polycomb-repressed regions or heterochromatin (45). Consistent with this hypothesis, more DNA breaks were present in MHCII hi -expressing medullary thymic epithelial cells (mTECs) from AIRE-positive wild-type mice compared with AIRE knockout littermates (6). It remains to be studied whether DNA breaks are critical for the expression of a distinct set of thymic TSAs regulated by the transcription factor Fezf2 (46).
We tested the putative chromatin relaxation implied by the ChIP experiments and expression analyses by genome-wide FAIRE-seq approach to investigate chromatin accessibility. To our surprise, we detected relatively few changes within or near the genes whose expression was enhanced by AIRE. Instead, we found a large proportion of AIRE-related changes in chromatin structure occurring in intergenic regions at CTCF sites that are known as chromatin insulator regions, which set boundaries between nearby genomic regions. These intergenic locus-specific structural changes at CTCF sites are involved in the arrangement of chromosomal architecture and can affect interactions between promoter and enhancer sequences (47). We therefore analyzed the changes in CTCF-mediated chromatin interactions that could explain the enhanced transcription of AIRE-dependent genes. We found that, in AIRE-expressing cells, the select anchor region in the AIRE-regulated CEACAM gene cluster forms several chromatin loops over a large distance that are distinct from those in uninduced AIRE-Tet cells and that partially overlap with the CTCF loops that were predicted earlier by computational analysis (48). Thus, the long-distance interactions between CTCF sites within the gene clusters containing AIRE target genes may modulate their transcriptional permissiveness.

Inducible and stable AIRE cell lines
Doxycycline-inducible AIRE expression in the HEK293 cell line (AIRE-Tet) has been described previously (16). Cells were cultured in DMEM supplemented with 10% tetracycline-negative fetal calf serum, 100 units/ml penicillin/streptomycin, and 0.15 mg/ml G418 for at least 24 h before any additional treatments. Untreated cells served as a negative control material, and AIRE expression was induced with 1.5 M doxycycline (631311, Clontech). After 24 h, the control and induced cells were either mock-treated with DMSO or with 2 M etoposide (E1383, Sigma) for another 24 h. In some experiments, the cells were additionally incubated with 1 M merbarone (M2070, Sigma), 1 M camptothecin (C9911, Sigma), and 1 M ␤-lapachone (L2307, Sigma) for 24 h. Stable HEK293 cells expressing WT AIRE, mutant AIRE (L28P), or YFP were cultured as described previously in Ref. 49.

RNA isolation and qPCR
Total RNA was isolated from AIRE-Tet cells at different time points during etoposide using TRIzol reagent according to the protocol of the manufacturer (Life Technologies). The yield and purity of RNA were determined by NanoDrop 1000. Subsequently, 5 g of total RNA from each sample was used as a template for cDNA synthesis using the SuperScript TM III First-Strand Synthesis kit according to the protocol of the manufacturer (Life Technologies). The gene expression levels of AIRE target genes were detected by quantitative real-time PCR (qPCR) using Maxima SYBR Green/ROX qPCR Master Mix (Thermo Scientific) and the ViiA 7 real-time PCR system (Life Technologies). The expression of AIRE target genes was normalized to the housekeeping gene HPRT1 and analyzed using the comparative Ct method (50). For statistical analysis, the -fold change values were log-transformed, standardized, and scaled according to the procedure described in Ref. 51. Experimental groups were compared with a two-sample t test implemented in R. The primers are listed in supplemental Table 1.

RNA-seq sample preparation and data analysis
RNA was isolated with the miRNAeasy Mini Kit (Qiagen) using on-column DNase digestion from 10 6 AIRE-Tet cells induced with 1.5 M doxycycline for 48 h, 2 M etoposide for 24 h, or a combination of both. Uninduced DMSO-treated cells were used as a negative control. Two independent biological replicates from each treatment were prepared from 500 ng of RNA with RNA integrity number (RIN) Ͼ 9. Sequencing libraries were generated with the TruSeq Stranded Total RNA LT (with Ribo Zero Gold) Sample Prep Kit (Illumina) according to the instructions of the manufacturer, with the exception that 12 PCR cycles were used for the DNA enrichment step. Paired-end sequencing (2 ϫ 100 bp) was performed with Illumina HiSeq 2000, which generated 80 -100 ϫ 10 6 sequenced fragments/ sample. Prior to alignment, the reads were trimmed to remove adapter sequences and bases with a Phred score lower than 30 using Trim Galore!. Reads were mapped to the human reference genome GRCh37 (Ensembl release 75) using STAR aligner version 2.4.2a with the two-pass mode (52). Read count tables were generated with the STAR aligner option -quantMode geneCounts. Differential gene expression between the experimental samples was analyzed with DESeq2 (53). Read count tables and group comparisons for differential exon usage were performed with DEXSeq (54). Determination of tissue specificity was based on data from the Human Protein Atlas (55). Chromosomal clusters of genes were detected with the command line tool CROC (56).

Co-IP
Uninduced and doxycycline-induced AIRE-Tet cells, either untreated or treated with etoposide, were cultured on 10-cm plates. Cells were lysed with hypotonic lysis buffer (0.05% Nonidet P-40, 10 mM HEPES (pH 7.4), 1.5 mM MgCl 2 , 10 mM KCl, 5 mM EDTA, and 1:100 EDTA-free complete protease inhibitor mixture, Thermo Scientific) at a concentration of 30 ϫ 10 6 cells/ml and incubated on ice for 15 min. Cell lysates were centrifuged for 10 min at 1000 ϫ g at 4°C. The supernatant was discarded, and the pelleted nuclei were resuspended in nuclear extraction buffer (20 mM HEPES (pH 7.4), 300 mM NaCl, 20 mM KCl, 4 mM CaCl 2 , 1:100 EDTA-free complete protease inhibitor mixture, and 1 unit MNase) at a concentration of 30 ϫ 10 6 cells/ml. The lysates were incubated on ice for 1 h and centrifuged for 10 min at 13,000 ϫ g at 4°C. Lysates (1 mg) were kept overnight on a rotary shaker at 4°C in Pierce spin columns with 10 g of rabbit anti-TOP2A (Abcam, ab2987) or IgG control (Diagenode, AIP-103-110) antibodies. The following morning, 150 l of protein G-Sepharose 4B beads (Invitrogen, 50% sus-pension in IP lysis buffer) was added, and the lysates were further incubated for 5 h on a rotary shaker at 4°C. Beads were then washed five times with IP lysis buffer, and the immunoprecipitates were eluted in Laemmli buffer at 99°C for 10 min.

TOP2A knockdown
Uninduced and doxycycline-induced AIRE-Tet cells, either untreated or treated with etoposide, were grown on 6-well plates and transfected with four SureSilencing shRNA vectors along with one scrambled negative control shRNA vector (KH01520N, SABiosciences) using Turbofect reagent (Thermo Scientific). The efficiency of knockdown was checked by Western blotting with anti-TOPA antibody 48 h after transfection. Anti-␤-actin antibody (A5441, Sigma) was used to determine protein loading.

TUNEL assay
Uninduced and doxycycline-induced AIRE-Tet cells, either untreated or treated with etoposide, merbarone, camptothecin, or ␤-lapachone, were fixed with 4% formaldehyde. The occurrence of DNA breaks was quantitatively assessed with the APO-BrdU TUNEL assay (Pharmingen). Briefly, the cells were labeled with Br-dUTP using TdT enzyme for 60 min at 37°C and then incubated with FITC-labeled anti-BrdU antibody for 30 min at room temperature. After washing with PBS, the cells were analyzed by LSR Fortessa using FACSDiva software (both from BD Biosciences). The efficacy of the assay was determined with the positive and negative control samples included in the kit.

Apoptosis assay
Uninduced and doxycycline-induced AIRE-Tet cells, either untreated or treated with etoposide, were labeled with the Annexin V-FITC kit (130-092-052, Miltenyi Biotec) according to the protocol of the manufacturer, but propidium iodide was replaced with 7-aminoactinomycin D. The percentages of dead, viable, and early apoptotic cells were determined by FACS with LSR Fortessa using FACSDiva software (both from BD Biosciences).

ChIP
ChIP experiments were performed using 10 7 cells/experiment according to a protocol described previously with slight modifications (57). Briefly, uninduced and doxycycline-induced AIRE-Tet cells, either untreated or treated with etoposide, were cross-linked with 1% formaldehyde for 10 min at room temperature, and the formaldehyde was quenched by addition of glycine to a final concentration of 0.125 M. Cells were lysed in 200 l of lysis buffer (50 mM Tris-HCl (pH 8), 10 mM EDTA, 1% SDS, and 1ϫ Halt protease inhibitor mixture; Thermo Scientific). The samples were diluted 10ϫ in radioimmune precipitation assay buffer containing 1ϫ Halt protease inhibitor mixture. Sonication was performed with a Diagenode Bioruptor for 15 min using 30-s on/off cycles at a high voltage setting. After chromatin preparation, ChIP was performed with IP-Star (Diagenode) using the Chip Direct method (Diagenode), in which the antibodies are first coated onto the surface of magnetic beads, and then the bound antibodies are added to the sheared chromatin. ChIP-grade antibodies against HMGB1 (ab18256) and ␥H2AX (ab2893) were obtained from Abcam and against CTCF from Millipore (07-729). The antibody against histone H1 (sc-34464) was obtained from Santa Cruz Biotechnology. The ChIP signals were measured by qPCR, and the values were log-transformed, standardized, and scaled according to the procedure described in Ref. 51. Experimental groups were compared with a two-sample t test implemented in R. Primers for the ChIP DNA analysis are listed in supplemental Table 1.

Formaldehyde-assisted isolation of regulatory elements coupled with high-throughput sequencing (FAIRE-seq)
Uninduced and doxycycline-induced AIRE-Tet cells, either untreated or treated with etoposide, were cultured and processed as in the ChIP experiments up to the preparation of sonicated lysates. Lysates were then spun at 20,000 ϫ g for 10 min at 4°C to remove cellular debris, and 10% of the sample volume was collected for input sample preparation. The input samples were treated with 10 g of DNase-free RNase A for 30 min at 37°C, followed by 10 g of proteinase K incubation overnight at 65°C. The DNA from the remaining lysates and the proteinase K-treated input samples was prepared with standard phenol-chloroform extraction and additionally purified with a Qiagen PCR purification kit. The FAIRE-seq libraries were prepared following the Illumina Truseq ChIP sample preparation guide and subjected to 100-bp paired-end sequencing on an Illumina HiSeq 2000. The quality of the raw reads was checked with FastQC software. Adapter sequences and lowquality base calls (Phred score Ͻ 30) were removed with Trim Galore!. The remaining paired-end reads were mapped to the human genome (hg19) with bowtie2 in "local" mode (58). Properly paired reads with a mapping quality of Ͼ30 were used for detecting differentially accessible chromatin regions with the csaw R package (59). The differentially accessible regions were annotated with the ChIPseeker R package (60). Statistical sig-nificance of overlap between different sets of genomic intervals was assessed with the GenometriCorr R package (61).

3C
Uninducedanddoxycycline-inducedAIRE-Tetcellswereprocessed according to the detailed protocol in Ref. 31, but the Sau3AI (Thermo) restriction enzyme was used in the DNA digestion step. The chromatin interactions were calculated as the ratio of the qPCR signal in the doxycycline-induced sample relative to the signal in the uninduced control sample. The anchor region was selected based on the FAIRE-seq results. The 3C signals were measured by qPCR, and the values were logtransformed, standardized, and scaled according to the procedure described in Ref. 51. Statistical significance of the log2 -fold change was assessed with a one-sample t test implemented in R. Primers for the 3C qPCR analysis were designed according to the guidelines in Ref. 31 and are listed in supplemental Table 1.