More than 50% of mammalian genomes consist of retrotransposon sequences. Silencing of retrotransposons by heterochromatin is essential to ensure genomic stability and transcriptional integrity. Here, we identified a short sequence element in intracisternal A particle (IAP) retrotransposons that is sufficient to trigger heterochromatin formation. We used this sequence in a genome‐wide shRNA screen and identified the chromatin remodeler Atrx as a novel regulator of IAP silencing. Atrx binds to IAP elements and is necessary for efficient heterochromatin formation. In addition, Atrx facilitates a robust and largely inaccessible heterochromatin structure as Atrx knockout cells display increased chromatin accessibility at retrotransposons and non‐repetitive heterochromatic loci. In summary, we demonstrate a direct role of Atrx in the establishment and robust maintenance of heterochromatin.
Atrx contributes to retrotransposon silencing in ES cells by promoting inaccessible heterochromatin formation. The data also suggest that retrotransposon silencing requires the histone chaperone Daxx but is independent of histone H3.3.
We identify a small heterochromatin initiation site in IAP‐Ez retrotransposons.
Atrx assists in Setdb1‐/Trim28‐dependent heterochromatin establishment and maintenance.
A locus‐specific chromatin accessibility assay reveals that Atrx‐deficient heterochromatin is more accessible and vulnerable to challenges.
Silencing of retrotransposons is essential for embryonic development. Firstly, novel retrotransposon insertions can impair genome stability , and secondly, aberrant activation of retrotransposon‐based enhancers can confuse the transcriptome by affecting the expression of nearby genes , . In light of these findings, recent data have implicated retrotransposons in the development of human diseases, such as cancer . Hence, it is important to understand how mammalian retrotransposons are targeted for silencing.
Silencing of retrotransposons involves the establishment of a heterochromatic domain covering the important regulatory regions of these elements. This heterochromatic structure shows a very similar modification signature to pericentric heterochromatin and features high levels of DNA methylation as well as H3K9me3 and H4K20me3 . Chromatin modifications on retrotransposons are established by the same molecular machineries that act on pericentric heterochromatin, except for H3K9me3 which is not predominantly mediated by Suv39h enzymes but rather depends on the histone methyltransferase Setdb1 . Targeting of the repressive machineries to retrotransposons is incompletely understood. In part, DNA methylation imprints of germ cells are used to re‐establish heterochromatin during early embryonic development , . In addition, there is strong evidence for de novo pathways that recognize specific retrotransposon sequences , , . These pathways are active in embryonic stem (ES) cells and can be utilized to study de novo heterochromatin establishment.
To identify novel players in heterochromatin establishment and maintenance on retrotransposons, we chose the abundant class of IAP retrotransposons as model system. In ES cells, IAP silencing depends on the transcriptional repressor Trim28 that interacts with Setdb1 and thus mediates the recruitment of H3K9me3. Trim28 and Setdb1 also regulate the establishment and turnover of DNA methylation at IAP sequences in ES cells . Interestingly, in differentiated cells IAP silencing does not require the Trim28 pathway anymore and largely depends on DNA methylation, suggesting that Trim28 is mainly an important factor in the de novo pathway for heterochromatin establishment , . How IAP sequences specifically recruit Trim28 and Setdb1 in ES cells is largely unclear. Likely candidates for targeting Trim28 to IAPs are sequence‐specific KRAB zinc finger proteins. Recruitment of Trim28 by KRAB zinc finger proteins is well described in vitro , , ,  and on few endogenous targets including MLV proviruses and imprinted regions , , . This hypothesis predicts that retrotransposons contain nucleation sites for silencing, bound by KRAB zinc fingers. In accordance with this model, novel IAP insertions into the genome of mouse ES cells recruit Trim28‐dependent silencing , .
Here, we systematically tested sequence elements of IAP retrotransposons for their ability to induce heterochromatin formation and identified a small region of 160 bp (SHIN) that is sufficient to trigger silencing. Based on this sequence, we developed a shRNA screen and identified the SNF2‐type chromatin remodeler Atrx as strong modifier of IAP silencing. Atrx was initially identified as the gene responsible for the X‐linked alpha thalassemia/mental retardation (ATR‐X) syndrome . ATRX‐deficient cells display numerous phenotypes connected to heterochromatic regions, such as defective sister chromatid cohesion and telomere dysfunction , , . ATRX loss is further connected to alterations in DNA methylation patterns at imprinted regions, rDNA loci, telomeric repeats, and pericentromeric heterochromatin , . Recruitment to heterochromatic domains may be due to direct binding of the Atrx ADD domain to H3K9me3 in the context of unmethylated H3K4 , . Interaction with HP1, which is highly enriched at heterochromatin, has been shown to also contribute to stable recruitment of Atrx , .
Recently, Atrx was found to interact with the histone H3.3‐specific chaperone Daxx to facilitate H3.3 deposition in telomeric and pericentromeric regions , , . Notably, this H3.3 deposition pathway is frequently mutated in pancreatic neuroendocrine tumors, glioblastoma, and pediatric neuroendocrine tumors , , , , . These mutations are generally associated with an activation of the alternative lengthening of telomeres (ALT) pathway, a telomerase‐independent mechanism to extend telomere length by recombination , . It is currently unclear if H3.3 deposition by Atrx/Daxx is really necessary to establish proper heterochromatin domains across telomeres and pericentromeric heterochromatin. Furthermore, it is not known if Atrx affects heterochromatin in other regions of the genome.
In this study, we show that Atrx is crucial for heterochromatin formation at IAP retrotransposons. Interestingly, heterochromatin establishment does not involve Daxx‐mediated histone H3.3 incorporation. Instead, Atrx is important for generating an inaccessible chromatin structure on retrotransposons and non‐repetitive regions in the genome which is necessary for robust maintenance of heterochromatin. Thus, our data reveal a novel role of Atrx in heterochromatin organization and may help to explain pathological phenotypes associated with ATRX syndrome or ATRX‐deficient tumors.
Identification of a short heterochromatin inducing sequence (SHIN)
To monitor the silencing potential of specific DNA sequence elements of mouse IAP retrotransposons, we developed a lentiviral hEF1α‐EGFP reporter system (Fig 1A). When we combined the reporter with the well‐characterized repressor binding site (RBS) of the MLV retrovirus , , strong silencing of EGFP expression was observed 2 days post‐transduction (Fig 1A). Similar effects were observed with the GAG region of IAP retrotransposons (Fig 1A). In order to more systematically test silencing potential of IAP sequence elements, we analyzed different IAP regions in our reporter assay (Fig 1B). A previously characterized region containing the UTR of IAP retrotransposons  displayed reporter silencing; however, the strongest effect was exerted by the ~2,000‐bp‐long IAP‐GAG region (Fig 1B). As de novo silencing of IAP retrotransposons is restricted to ES cells , we did not detect silencing in mouse embryonic fibroblasts (MEF). To refine potential silencing initiating elements within the GAG region, we further dissected this element into 400‐bp fragments (Supplementary Fig S1A) and could finally determine a region of 160 bp which was sufficient to induce strong reporter silencing (Fig 1C). This sequence could not be further shortened without compromising silencing potential (Fig 1C) and is highly conserved in more than 600 IAP elements, predominantly of the IAP‐Ez subclass (Supplementary Fig S1C, Supplementary Table S1). As will be outlined below, this short sequence element is sufficient to trigger heterochromatin formation; thus, we termed this region SHIN for short heterochromatin inducing sequence. Notably, SHIN silencing is not restricted to the hEF1α promoter (Supplementary Fig S1B) and is orientation independent (Supplementary Fig S1B), suggesting that silencing is initiated by sequence‐specific DNA binding factors.
The SHIN sequence induces heterochromatin
In ES cells, IAP silencing is mediated by heterochromatin that depends on Trim28 which, in turn, recruits the H3K9me3 methyltransferase Setdb1 , , . Similar to pericentric heterochromatin, the H3K9me3‐rich domains across IAP elements coincide with H4K20me3 and DNA methylation. To test if SHIN silencing is dependent on any of these chromatin modifications, we assessed the extent of SHIN repression in ES cell lines that are depleted for the major silencing machineries. Here, we measured the number of EGFP‐positive cells relative to the control reporter that does not contain the SHIN sequence. Interestingly, cells deficient for Suv39h, Suv4‐20h, Dnmt1, or Dnmt3ab can still effectively silence the SHIN reporter (Fig 2A). In contrast, Setdb1 knockout cells are severely compromised in SHIN silencing (Fig 2A). Also, lentiviral knockdown of Trim28, which did not even completely abolish Trim28 expression (Supplementary Fig S1D and E), resulted in strong silencing defects. These data demonstrate that SHIN silencing requires the major players that also regulate silencing of endogenous IAP elements.
Next, we asked if the SHIN sequence can directly trigger Trim28‐/Setdb1‐dependent heterochromatin formation. To address this question, we used a well‐established recombinase‐mediated cassette exchange (RMCE) cell line to integrate the SHIN sequence into a defined position of the ES cell genome (Supplementary Fig S1F) , . We combined the SHIN sequence with a Tet‐inducible EGFP‐T2A‐zeocin reporter which is not transcribed in the absence of doxycycline (Fig 2B). We also generated a control reporter lacking the SHIN sequence (Fig 2B). RMCE of the two reporter constructs resulted in the HA36::SHIN and HA36::control cell lines (Fig 2B). ChIP‐qPCR analysis revealed high levels of H3K9me3 across the RMCE locus in HA36::SHIN cells, comparable to endogenous IAP elements (Fig 2B). In contrast, only residual occupancy for H3K9me3 was detected in HA36::control cells (Fig 2B). To test if H3K9me3 across the SHIN reporter depends on Setdb1, we performed CrispR/Cas knockout of Setdb1 in HA36::SHIN cells. Prolonged knockout of Setdb1 is lethal for ES cells. Thus, we could not select for Setdb1 knockout cells and used a mixed population of Setdb1‐deficient/Setdb‐proficient cells for ChIP analyses 4 days post‐sgRNA transduction (Fig 2C, Supplementary Fig S1G). H3K9me3 levels at major satellite repeats are dependent on Suv39h enzymes and therefore remained almost unchanged upon Setdb1 depletion (Fig 2C). Levels of H3K9me3 at endogenous IAP sequences were only mildly affected, indicating that this modification may have a low turnover on these sequences. Importantly, H3K9me3 was clearly reduced on a unique Setdb1 target region (pos control) and across the RMCE locus containing the SHIN sequence (Fig 2C). In summary, our data show that the SHIN sequence represents an initiation site for heterochromatin formation from which H3K9me3 can spread over the entire locus.
SHIN silencing requires the Snf2‐type chromatin remodeler Atrx
Heterochromatin formation across SHIN strictly depends on the Trim28/Setdb1 pathway and recapitulates silencing of endogenous IAP retrotransposons. Thus, we utilized our reporter assay to screen for additional factors that regulate retrotransposon silencing using a genome‐wide shRNA screen (Fig 3A). We transduced wild‐type ES cells with a pooled genome‐wide shRNA library, selected shRNA expressing cells with puromycin and then transduced the SHIN reporter. Cells that escaped SHIN silencing were isolated based on their EGFP fluorescence, and shRNA sequences from these escaper cells were cloned into a plasmid library (Fig 3A). We tested 71 individual shRNA plasmids in a secondary screen. One shRNA which most strongly inhibited SHIN repression was found to target the SNF2‐type chromatin remodeler Atrx (Fig 3B, Supplementary Table S2). This shRNA clearly reduces Atrx mRNA and protein levels in mouse ES cells (Supplementary Fig S2A and B). However, to rule out off‐target effects, we generated four independent knockout ES cell lines (Supplementary Fig S2C) with loss of Atrx protein (Fig 3C). Atrx knockout (ko) cells grow normally and show no obvious alterations in their cell cycle profile (Supplementary Fig S2D). Importantly, SHIN silencing was defective in all Atrx knockout ES cell lines (Fig 3D), which confirmed the shRNA screening result.
We then asked if Atrx is important for the Trim28/Setdb1 pathway and performed SHIN silencing assays in wild‐type versus Atrx ko cells, depleted for Setdb1 and Trim28, respectively (Supplementary Fig S2E and F). Although Setdb1 could be depleted by around 70% (Supplementary Fig S2E), SHIN silencing was only mildly impaired (Fig 3E). Setdb1 depletion in Atrx ko cells, in contrast, resulted in strong silencing defects (Fig 3E). This suggests that Atrx enhances the efficiency of Setdb1‐dependent repression, although we cannot exclude the possibility that Atrx may act independently of Setdb1 in repressing the SHIN reporter, leading to synergistic effects when both Setdb1 and Atrx are depleted. Similar effects were observed for shRNA knockdown of Trim28 (Fig 3E).
To test if Atrx is required for Trim28/Setdb1‐dependent repression outside of IAP retrotransposons, we performed silencing assays using a heterologous reporter system. The well‐established repressor binding site of MLV retrotransposons (RBS) is recognized by the KRAB zinc finger protein Zfp809, which in turn recruits Trim28 and Setdb1 to induce silencing in a variety of cell types. In wild‐type cells, the RBS‐EGFP reporter was efficiently silenced (Fig 3F, Supplementary Fig S2G). However, RBS silencing was compromised upon knockdown of Atrx in MEFs and in Atrx ko ES cells (Fig 3F, Supplementary Fig S2G). Taken together, our data demonstrate Atrx as a crucial component of the Setdb1/Trim28 silencing pathway.
SHIN silencing is independent of Daxx‐mediated H3.3 deposition
On telomeres and pericentric heterochromatin, Atrx recruits Daxx to mediate H3.3 deposition , , . In order to test if the Daxx/H3.3 pathway plays a role in SHIN repression, we generated Daxx ko cells (Supplementary Fig S3A). We found that SHIN silencing is defective in Daxx ko cells (Supplementary Fig S3B), suggesting that Daxx is an important component for retrotransposon silencing. To address the question whether H3.3 deposition by Daxx is critical for SHIN silencing, we generated rescue cell lines expressing wild‐type and mutant Daxx proteins (Supplementary Fig S3C). Re‐expression of full‐length Daxx protein in Daxx ko cells leads to a rescue in SHIN silencing (Supplementary Fig S3D). However, the expression of Daxx‐mutant proteins with relaxed binding specificity for histone H3.3 (E231A) or impaired H3.3 interaction (R257A) also resulted in rescued SHIN silencing (Supplementary Fig S3D and E). In contrast, the expression of Daxx with a deletion of the C‐terminal repressor domain  does not rescue SHIN silencing (Supplementary Fig S3D). These data point toward a H3.3‐independent function of Daxx in SHIN repression. To confirm that H3.3 is really dispensable for retrotransposon silencing, we generated H3.3 ko ES cell lines (Supplementary Fig S4A). Histone H3.3 is mainly transcribed from two loci in the mouse genome, H3f3a and H3f3b. Consequently, H3f3a/H3f3b double knockout cells have essentially lost H3.3 protein (Supplementary Fig S4B). Notably, SHIN silencing is completely intact in H3.3 ko cells (Supplementary Fig S4C). Due to H3f3a duplications in the mouse genome that could eventually give rise to functional H3.3 transcripts (see blat search and alignment report in Supplementary Table S3), we performed H3f3a knockdown in H3f3b ko cells. These cells have lost H3.3 (Supplementary Fig S4D) but display robust SHIN silencing (Supplementary Fig S4E). Finally, we re‐analyzed published H3.3 ChIP‐seq datasets in control and Atrx ko ES cells . We found only low enrichment of H3.3 in control ES cells (Supplementary Fig S4F). Surprisingly, upon deletion of Atrx, H3.3 becomes strongly enriched at IAP elements (Supplementary Fig S4F), suggesting that Atrx does not promote H3.3 incorporation at endogenous IAP elements, but rather inhibits excessive H3.3 deposition. Alternatively, higher chromatin turnover at IAP elements in the absence of Atrx may indirectly lead to more H3.3 incorporation. Together with our finding that H3.3 interaction mutants of Daxx can repress retrotransposon sequences and H3.3‐depleted cells are not impaired in SHIN repression, our data demonstrate that SHIN silencing is independent of Daxx‐mediated histone H3.3 incorporation.
Atrx is required for efficient heterochromatin formation
Next, we sought to investigate how Atrx mediates SHIN silencing on the molecular level. One possibility was that Atrx could mediate efficient heterochromatin formation across the SHIN locus. To test this hypothesis, we performed a re‐silencing assay in HA36::SHIN ES cells in which we depleted Atrx by shRNA‐mediated knockdown (Fig 4A). Strong induction of the Tet‐inducible promoter using doxycycline alleviates SHIN silencing in a small population of cells, and constant selection pressure with zeocin allowed us to isolate a homogeneous population of cells with strong EGFP expression. Lack of promoter stimulation by doxycycline removal resulted in the rapid loss of EGFP expression within 4 days in both control (shScr) and Atrx‐depleted (shAtrx) cells (Fig 4B, left panel). Loss of EGFP expression coincided with presence of H3K9me3 across the RMCE locus in shScr cells (Fig 4C, left panel, Supplementary Fig S5A). H3K9me3 could also be established in shAtrx cells; however, abundance of H3K9me3 appeared reduced in the test region 1 most distal to the SHIN sequence (Fig 4C, left panel), indicating reduced heterochromatin spreading. Surprisingly, re‐silencing was very efficient in shScr cells despite continuous doxycycline induction (Fig 4B, right panel, Supplementary Fig S5A). A large proportion of these cells displayed strongly reduced EGFP expression and the establishment of H3K9me3 across the RMCE locus (Fig 4B and C, right panels). In contrast, shAtrx cells are completely unable to induce heterochromatin formation across the RMCE locus in continuous presence of doxycycline (Fig 4B and C, right panels). In summary, these data indicate that Atrx is necessary for efficient establishment of heterochromatin. On transcriptionally inactive loci, heterochromatin can be formed, but spreading appears reduced. In addition, Atrx is strictly required when heterochromatin formation is challenged by transcriptional activity.
Atrx is important for heterochromatin maintenance
Thus far, out data showed that Atrx is necessary for efficient de novo establishment of heterochromatin. However, in embryonic stem cells endogenous IAP elements are already covered by a heterochromatic structure which needs to be maintained. We thus wondered if Atrx is important for heterochromatin maintenance. When we analyzed published ChIP‐seq datasets for Atrx, Setdb1, Trim28, and H3K9me3 in ES cells , , , we found strong enrichment of Atrx across IAP retrotransposons with highest occupancy over the SHIN region (Fig 5A). These data suggest that Atrx has a permanent role on these elements. ChIP‐qPCR analyses for Atrx could confirm the enrichment on IAP elements and a non‐repetitive control region (Fig 5B, top panel). However, heterochromatin marks, such as H3K9me3 and DNA methylation, were not altered in Atrx ko cells (Fig 5B, Supplementary Fig S5B and C).
The strong enrichment of Atrx, H3K9me3, Trim28, and Setdb1 on IAP sequences close to the SHIN region (Fig 5A) suggests that Atrx is recruited by H3K9me3 and/or Trim28/Setdb1. Thus, we tested if Atrx is de novo recruited to a Setdb1‐dependent heterochromatin locus by analyzing HA36::SHIN cells. Enrichment of Atrx at the SHIN region within the reporter locus demonstrates that the recruitment of Atrx occurs in the context of Setdb1 and Trim28 (Fig 5C). Thus, we investigated whether Atrx generally associates with Setdb1/Trim28 target region in repetitive and non‐repetitive regions of the genome. The overlap between Atrx, Trim28, and Setdb1 in non‐repetitive regions is relatively small (Supplementary Fig S7A). Prominent enrichment was mainly observed on imprinted genes and few intragenic loci (Supplementary Fig S7B–D, Supplementary Tables S5 and S6). In contrast, our analyses revealed that Atrx co‐occupies many ERV classes which are also enriched for Setdb1, Trim28, and H3K9me3, for example, IAP‐Ez, MusD/ETn, and MMERVK10 elements (Fig 5D, Supplementary Fig S6, Supplementary Table S4). Thus, we sought to investigate if some of these ERV sequences become de‐repressed in Atrx ko cells. Out of the three tested ERV classes with the enrichment of Setdb1/Trim28/Atrx, we detected the de‐repression of MusD/ETn retrotransposons (Fig 5E), suggesting that Atrx function is rate limiting for the silencing of distinct ERV classes.
Challenges like DNA replication or transcription through intronic IAP repeats require heterochromatin to be constantly re‐established. Based on our finding that Atrx is crucial for efficient establishment of heterochromatin, we wondered if heterochromatin maintenance is perturbed in Atrx ko cells when the Setdb1/Trim28 pathway has reduced activity. Moderate knockdown of Trim28 in wild‐type cells resulted in slight up‐regulation of endogenous IAP elements, demonstrating that heterochromatin maintenance is not fully ensured. However, when we performed Trim28 knockdown in Atrx ko cells, we observed severe derepression of IAP retrotransposons (Fig 5F). These findings indicate that Atrx is required and rate limiting for the repression of endogenous IAP elements when heterochromatin maintenance is challenged.
Another way of challenging heterochromatin is to provoke transcription across a heterochromatinized locus. In HA36::SHIN cells, the EGFP reporter gene is not transcribed and covered with H3K9me3 (Fig 2C). However, transcription can be induced by doxycycline that allows the recruitment of the reverse tetracycline transactivator (rtTA) to the Tet‐responsive promoter of the reporter gene. Due to the heterochromatic state of the reporter gene in control cells (shScr + dox), doxycycline induction leads to a very low percentage of EGFP‐positive cells (Fig 5G, Supplementary Fig S5D). In contrast, doxycycline induction upon Atrx knockdown (shAtrx + dox) results in a much higher percentage of EGFP‐expressing cells (Fig 5G, Supplementary Fig S5D). These data demonstrate that heterochromatin is more vulnerable in Atrx ko cells.
Atrx regulates heterochromatin accessibility
Heterochromatin is generally characterized by a high compaction grade which makes it largely inaccessible to transcriptional activators and refractory to challenges. Our finding that transcription of a heterochromatic locus can be more easily stimulated in the absence of Atrx suggests that structural properties of heterochromatin like local nucleosome density or higher order folding are compromised when Atrx is lost. To test this hypothesis, we performed MNase accessibility assays in wild‐type versus Atrx ko ES cells. Atrx is only binding to a small fraction of the mouse genome, and thus, we did not detect global alterations in MNase accessibility  (Fig 6A and B). To investigate Atrx target regions, we employed a qPCR strategy to measure the MNase digestion degree on specific genomic loci with increasing MNase concentrations (Fig 6C). Curve fitting through these data points allows statistical assessment of the digestion degree between wild‐type and Atrx‐mutant cells (Supplementary Fig S7E). Further, we derive a chromatin accessibility score which represents the digestion rate at which 50% of the target locus is digested (Supplementary Fig S7E). MNase accessibility at endogenous IAP retrotransposons is significantly increased, whereas control regions, such as promoters of active and inactive genes, show no alterations (Fig 6D and E, Supplementary Fig S7F). Interestingly, chromatin accessibility was also increased on non‐repetitive Setdb1/Trim28 targets (Fig 6E, Supplementary Fig S7E). These data demonstrate that Atrx renders Setdb1‐dependent heterochromatin more inaccessible.
In the mammalian genome, heterochromatin is very abundant across repetitive elements. Pericentric heterochromatin is mainly organized by the Suv39h–HP1–Suv4‐20h pathway , , , while repetitive elements outside the pericentric compartment are controlled by the Trim28/Setdb1 pathway, with support by Suv39h enzymes , , , . A major characteristic of the heterochromatic state is its modification pattern of H3K9me3/H4K20me3/DNA methylation and the buildup of an inaccessible chromatin structure which does not allow strong transcriptional activity. Heterochromatin formation is thought to initiate on specific nucleation sites from which it can spread over large distances . Initiation sites in pericentric heterochromatin may be binding sites for specific transcription factors  or involve other recruitment mechanisms for Suv39h/HP1 complexes . In the context of retrotransposons, recruitment mechanisms may involve KRAB zinc finger proteins which could serve as targeting platform for Trim28/Setdb1 . Our work identified a novel nucleation site in IAP retrotransposons (SHIN) that is sufficient to induce heterochromatin and transcriptional repression of strong promoters (Fig 7). SHIN silencing requires the Trim28/Setdb1 pathway for heterochromatin formation across an integrated reporter gene (Fig 2C). Cumulative coverage maps across IAP elements revealed strong enrichment of Setdb1 and Trim28 across the SHIN sequence; however, additional coverage was detected at the LTR/UTR region and in the beginning of the POL coding sequence (Fig 5A). These findings suggest that endogenous IAP elements may feature several independent nucleation sites for heterochromatin formation. In support of this hypothesis, previous analyses of the LTR/UTR sequence revealed significant silencing potential , although to a lower extent as compared to the SHIN sequence (Fig 1). Which proteins recognize the SHIN sequence is currently unclear. As silencing can only be initiated in ES cells, KRAB zinc finger proteins with selective expression in ES cells are likely candidates. As we could not further shorten the SHIN sequence to less than 160 bp without compromising silencing potential, it is quite likely that multiple DNA binding proteins are involved in Trim28/Setdb1 recruitment.
Identification of the SHIN sequence as a strong heterochromatin nucleation site allowed a pilot screen for regulators of heterochromatin establishment through which we identified the chromatin remodeler Atrx. Atrx co‐localizes with Trim28/Setdb1 and H3K9me3 on IAP elements, other ERV classes, and several single copy loci (Fig 5A, Supplementary Fig S7A–D). We found that Atrx is de novo recruited to Setdb1‐dependent heterochromatin (Fig 5C), suggesting that Atrx could be directly recruited by Setdb1 or Trim28. However, as Atrx features domains which bind H3K9me2/3 and HP1 , , a complex combinatorial nature of interactions leading to Atrx recruitment is likely. This may be further complicated by potential interactions with non‐coding RNAs  that may affect heterochromatin formation on IAP elements .
We found that in the absence of Atrx reporter silencing is delayed and spreading of heterochromatin is reduced (Fig 4), which could be due to reduced activity of Setdb1 in establishing H3K9me3. Being a putative chromatin remodeler, we hypothesize that Atrx assists in generating a proper nucleosomal array which can then serve as efficient substrate for Setdb1. Consistent with this hypothesis, we found that reduced Setdb1 activity leads to defects in the establishment of silencing when Atrx is not present (Fig 3E). Secondly, heterochromatin spreading and stability rely on the formation of properly spaced nucleosomal arrays by HP1 molecules binding to H3K9me3 . Atrx activity may be required to establish such arrays, and, as a consequence, loss of Atrx would lead to compromised spreading. Alternatively, Atrx may also promote heterochromatin spreading by supporting Suv39h activity at retrotransposons . A third function of Atrx in heterochromatin establishment may be the recruitment of additional factors to assist in this process. On telomeres, Atrx interacts with Daxx to facilitate histone H3.3 deposition , , . We found that H3.3 is dispensable for silencing retrotransposon sequences in reporter assays, suggesting that incorporation of this histone variant is not important for Trim28/Setdb1‐dependent heterochromatin formation. Instead, we find that Daxx is important for this process (Supplementary Fig S3). Recently, Daxx was found to be involved in silencing of exogenous ASV retroviruses by H3K9me3 and DNA methylation . Although it is unclear how silencing is triggered in this context and if Atrx, Trim28, or Setdb1 are involved in this process, these data suggest Daxx as a global regulator of retrovirus repression. Interestingly, we find that Daxx requires a functional SUMO interaction motif to contribute to Setdb1‐/Trim28‐dependent repression (Supplementary Fig S3), suggesting that the interaction with sumoylated proteins is important in this context . This is consistent with the repressive activity of Daxx when recruited by sumoylated transcription factors . As Setdb1 recruitment to Trim28 has been shown to be SUMO dependent, and because several heterochromatin proteins, like HP1, are sumoylated , , it is likely that Daxx recruitment might rely on the sumoylation of Trim28 or other factors. More thorough genome‐wide screens for retrotransposon silencing are necessary to identify such interaction partners and additional components of this silencing pathway.
Heterochromatin formation is the net result of a dynamic balance between factors that build up heterochromatic structures and processes which act antagonistically . Thus, the efficiency of the heterochromatin buildup machinery is particularly critical in genomic regions with prominent activity of antagonistic processes, for example, transcription. Examples for such regions are retrotransposons which reside in intronic regions of highly expressed genes. Unfortunately, due to the lack of polymorphisms, it is not possible to directly examine heterochromatin formation on these elements. However, we could study heterochromatin establishment in competition with transcriptional activity on the RMCE reporter locus. In wild‐type cells, this reporter locus could be efficiently silenced despite strong promoter activity. Atrx knockdown leads to impaired silencing. These data clearly demonstrate that in the absence of Atrx, the activity of the Trim28/Setdb1 pathway is strongly reduced. Importantly, the de‐repression of MusD/ETn elements in Atrx ko cells represents a physiological situation in which the full activity of the Setdb1/Trim28 pathway is necessary for silencing (Fig 5E). The establishment of new heterochromatin domains, for example, in the context of differentiation, may also require Atrx for the full activity of the Setdb1/Trim28 pathway. The strong developmental phenotype of Atrx ko embryos  supports this hypothesis, although more analyses are necessary to identify Atrx target loci during developmental transitions.
Heterochromatin is generally characterized by low transcriptional activity. A likely explanation for this property is that binding of transcription factors and RNA polymerase is restricted by the largely inaccessible chromatin structure. We tested chromatin accessibility of different genomic regions in mouse ES cells and found that pericentric heterochromatin is highly inaccessible to MNase. A similarly low accessibility was only observed for a promoter region which is under control of the polycomb system (Six3, Fig 6E). IAP retrotransposons display higher chromatin accessibility as compared to pericentric heterochromatin, which may be due to a different organization of these regions. Pericentric heterochromatin is composed of large domains of major satellite repeats which cluster into higher order structures to form so‐called chromocenters. As retrotransposons form much smaller domains of only several kilobases, higher order folding of the chromatin structure is probably limited. However, chromatin accessibility of IAP elements is still much lower as compared to active promoters and may protect these regions from various challenges. Thus, it is currently unclear if the observed differences in chromatin accessibility result from changes in large‐scale chromatin compaction or from an altered local nucleosomal organization. In this study, we found that chromatin accessibility in Atrx ko cells is significantly increased on IAP elements, demonstrating that Atrx is important for proper heterochromatin organization. Interestingly, this more accessible heterochromatin architecture is indeed vulnerable to challenges. For example, strong transcriptional activators can more easily access their binding sites and lead to stronger transcriptional activation in Atrx ko cells. If Atrx‐deficient heterochromatin is vulnerable to other challenges, for example, DNA damage, remains to be tested. Interestingly, chromatin architecture is not only altered on IAP retrotransposons. We also tested non‐repetitive genomic regions with the enrichment of Atrx and H3K9me3 and consistently detected increased chromatin accessibility. Thus, our data provide strong evidence for a general role of Atrx in the establishment and robust maintenance of heterochromatin domains.
Materials and Methods
Reporter gene silencing assay
Reporter constructs were stably integrated into cells by lentiviral transduction, and the percentage of EGFP+ cells was measured by FACS after 2–4 days. Lentiviral particles were generated using standard protocols, and virus titers were determined by titration in HeLa cells. Mouse ES cells were transduced on gelatinized multi‐well dishes using spinoculation at a low multiplicity of infection to ensure a linear relationship between virus titer and transduction rate. The ratio of the percentage of EGFP+ cells generated by the reporter relative to the percentage of EGFP+ cells generated by a control EGFP vector of the same virus titer was used to quantify reporter silencing (relative % EGFP+ cells). In Fig 1 and Supplementary Fig S1, the reciprocal ratio (fold repression) is plotted.
Feeder‐independent ES cells were transduced with a genome‐wide pooled shRNA library (Lentiplex, Sigma‐Aldrich) at a low MOI, selected for shRNA expression using 0.5 μg/ml puromycin, and then transduced with the SHIN reporter. Cells that escaped SHIN silencing were isolated by FACS sorting based on their high EGFP fluorescence. shRNA sequences of the EGFP+ cells were amplified by PCR and cloned into the pLKO1 backbone. For validation of hits, 71 random shRNA sequences of the primary screen were separately transduced into ES cells and the SHIN reporter assay was performed as indicated above. The number of EGFP‐positive cells was compared to a non‐silencing scrambled control. Details are given in the Supplementary Materials and Methods.
T86 cells (HA36::SHIN cells harboring a reverse Tet transactivator) were treated with 2 μg/ml doxycycline for 2 days, and the re‐expression of the silenced reporter locus was selected with 50 μg/ml zeocin and 2 μg/ml doxycycline for at least 2 weeks. Reactivated reporter cells were then transduced with a lentiviral shRNA targeting either Atrx or a scrambled control sequence, and knockdown cells were selected by the addition of 1 μg/ml puromycin 2 days after transduction. After additional 3 days, zeocin selection was released and resilencing of the EGFP reporter was monitored daily in the presence or absence of doxycyline by FACS. The median EGFP fluorescence was calculated using FlowJo™ (TreeStar). For ChIP experiments, cells were harvested 4 days after zeocin release. Details are given in the Supplementary Materials and Methods.
GS and DS designed the experimental approach; DS, KS and SG carried out the bench experiments; JE performed the FACS sorting; IK, CF and FJT developed and carried out the statistical analyses; and GS and DS wrote the manuscript.
Conflict of interest
The authors declare that they have no conflict of interest.
Supplementary Materials and Methods, Supplementary Figures, Supplementary Legends
Supplementary Table S1
Supplementary Table S2
Supplementary Table S3
Supplementary Table S4
Supplementary Table S5
Supplementary Table S6
Source Data for Supplementary Figures S1–S7
Work in the lab of G.S. was funded by grants from the Deutsche Forschungsgemeinschaft (SFB 1064, SPP1356). F.J.T. and C.F. were supported by ERC grant “LatentCauses” and SPP1356. We thank Sandra Hake for providing antibodies. We are grateful to Tobias Straub and Felix Müller‐Planitz for critical discussions. We thank Dirk Schübeler for generously providing HA36 ES cells. RMCE plasmids were provided by Matthew Lorincz and Dirk Schübeler.
FundingDeutsche ForschungsgemeinschaftSFB 1064SPP1356
- © 2015 The Authors