The HUSH complex cooperates with TRIM28 to repress young retrotransposons and new genes

1Although <2% of DNA sequence in the human genome codes for proteins, the vast majority plays an enigmatic role and has thus been referred to as genomic dark matter. However, this extra DNA serves a purpose: First, it contains regulatory elements that control when and where genes are expressed, a role only now being realized. Second, it is involved in building heterochromatin, for example at the nuclear periphery. Little is understood about how heterochromatin is formed, but its content is dominated by retrotransposons, which contribute to its establishment from plants to mammals. Retrotransposons replicate through an RNA intermediate, which has allowed them and their regulatory sequences to accumulate and coevolve with their hosts.

2The human silencing hub (HUSH) complex, composed of FAM208A (also known as TASOR), MPHOSPH8 (also known as mpp8), and PPHLN1 (periphilin 1) is recruited to genomic loci rich in H3K9me3 and interacts with SETDB1 and MORC2. The HUSH complex mediates position-effect variegation at reporter constructs that are integrated into silent chromatin, and depletion of HUSH components reduces H3K9me3 and alters transcription. It is unknown whether the HUSH complex is required for the repression of retrotransposons.

3TRIM28, in contrast, is known to silence retrotransposons early in development and is targeted to DNA through KRAB-zinc finger proteins (KZFPs in mouse or KZNFs in human), most of which are specific for transposon sequences. TRIM28 recruits chromatin writers, readers, and erasers including SETDB1, CBX5 (Hp1alpha), and CHAF1A. The resulting silent H3K9me3 mark at retrotransposons overlaps with H3F3A/B (histone variant 3.3), ATRX, and DAXX and spreads to nearby genes.

4We asked here if the HUSH complex, like TRIM28, is necessary for retrotransposon repression and whether it cooperates with TRIM28 and ATRX at specific genomic loci.

5We used a human retrotransposon reporter (Fig. 1A) that is repressed through ZNF91 binding to an SVA-type D variable number tandem repeat (SVA VNTR) sequence. We found that the SVA reporter was repressed (3.3×) in POU5F1-positive human embryonal NTERA-2 cells, which naturally express ZNF91, but not in 293T cells (Fig. 1B). Reporter repression could be engineered in POU5F1-expressing mouse embryonic stem cells (mESCs) by transfection of the cognate ZNF91, but not of a control KZNF, ZNF93 as expected (Supplemental Fig. S1A; ), and even in 293T cells (Supplemental Fig. S1B). We used 293T cells for subsequent reporter assays because of their amenity to genetic manipulation.

6We validated that reporter repression (6.7×) (Fig. 1C) was TRIM28-dependent by using TRIM28 knockout and complemented 293T cells. Having set up this system, we depleted the HUSH complex using shRNA before introduction of the reporter gene (Fig. 1D). All three HUSH components contributed to SVA repression (Fig. 1E), with the most striking effect observed for MPHOSPH8 depletion (14× derepression). Although these results were obtained with single hairpins, we validated the role of Mphosph8 with an independent hairpin in mESCs (Supplemental Fig. S1C). Considering that the HUSH complex is thought to function similarly to CBX proteins, we depleted all three CBX family members, which revealed that they too all contributed to repression in 293Ts and HeLa cells (Supplemental Fig. S1D). Finally, we identified a role for H3F3A (4× derepression) and its ATRX–DAXX chaperone complex in establishing SVA repression (Fig. 1E; Supplemental Fig. S1C).

7We reasoned that the HUSH complex may be targeted to the repressed SVA reporter through the chromodomain of MPHOSPH8 that interacts with H3K9me3; indeed, we found H3K9me3 to be enriched (2.4×) on the repressed reporter (Supplemental Fig. S1E). Of note, plasmids are chromatinized and subject to H3K9me3. We also detected cognate ZNF-specific de novo DNA methylation (23.5%) (Fig. 1F), as observed before for retroviral reporters. DNA methylation was not necessary for repression but contributed to it, since we detected 4.5× SVA repression in Dnmt knockout mESCs, compared to 9× repression in wild-type mESCs (Supplemental Fig. S1F). This suggested HUSH may be necessary for handover of H3K9me3 to DNA methylation and in support of this, depletion of HUSH components coincided with a decrease (5.8×) in de novo DNA methylation at the SV40 promoter along with a reduction (up to 2.7×) in SVA reporter repression (Fig. 1G; Supplemental Fig. S1G). In sum, these data on the nonintegrated reporter suggest that HUSH may be required for the maintenance of retrotransposon repression.

8We reasoned that the HUSH complex may exert its greatest impact in naïve pluripotent cells, in which chromatin awaits stable epigenetic programming. We depleted epigenetic factors in mESCs (Fig. 2A) and compared mixed population to naïve cultures, the latter in which we verified enhanced NANOG expression (Fig. 2B). Depletion of Mphosph8, as well as Atrx and Trim28, was sufficient to reactivate retrotransposons in two different strains of serum-cultured mESCs (Supplemental Fig. S2A). However, although Trim28 or Atrx depletion mainly affected intracisternal A-particle (IAP) elements, Mphosph8 depletion mainly affected L1s. Parallel culture of J1 ESCs in serum versus 2i conditions consistently led to more pronounced reactivation in the naïve cells (Fig. 2C; Supplemental Fig. S2B) with HUSH components Mphosph8 and Fam208a both affecting L1 elements (up to 13× reactivation in naïve cells) (Fig. 2C). Depletion of Trim28, Setdb1, Atrx, or Fam208a resulted in IAP GAG protein accumulation, whereas depletion of Fam208a and Mphosph8 led to increased production of L1 ORF1 protein (Fig. 2D). Finally, TRIM28, SETDB1, ATRX, MPHOSPH8, and FAM208A protein expression levels were elevated in naïve cells (Fig. 2E), in support of their critical role early in development.

9We sought to identify genomic sites where TRIM28 collaborates with FAM208A, SETDB1, and ATRX and therefore performed mRNA-sequencing (Fig. 3A) and focused on up-regulated genes. Samples within treatment groups clustered together (Supplemental Fig. S3A) and a large proportion of TRIM28-repressed genes were co-repressed by FAM208A (94 genes), SETDB1 (183 genes), or ATRX (89 genes) (Fig. 3B; Supplemental Fig. S3B; Supplemental Table S4). These three groups of TRIM28-repressed genes are likely direct rather than indirect targets, because up to 81% overlapped TRIM28 peaks (TRIM28-FAM208A and TRIM28-SETDB1 gene sets) and up to 77% overlapped H3K9me3 (all three gene sets), compared to randomly-selected gene groups (“Random”) (Fig. 3C). All three groups were enriched in protein-coding genes (Fig. 3D), and follow-up of the TRIM28-FAM208A genes revealed them to be functionally related and involved in developmental pathways (Fig. 3E), unlike random genes. Of note, we verified that up-regulation was detectable across the length of the transcripts within this group (Supplemental Fig. S3C).

10TRIM28 regulates genes through binding repeats. We therefore asked if these sets of genes were enriched in repeats and divided repeats by size (Fig. 3F; Supplemental Fig. S3D). TRIM28-SETDB1 genes were enriched for all repeats here (9× for LINEs >5 kb), whereas TRIM28-ATRX genes were enriched for full-length LTRs (4×); most interestingly, TRIM28-FAM208A genes were enriched for LINE elements of any size but particularly those >5 kb (7×) (Fig. 3F; Supplemental Fig. S3D) and for satellites (2×). In fact, 81.72% of TRIM28-FAM208A genes contained an L1 or satellite or both within 20 kb (only TRIM28-regulated L1s were included), and L1s were often present in arrays (Fig. 3G). We verified that TRIM28-FAM208A genes contained significantly more TRIM28-L1s than random genes (e.g., 2.3× for L1Md_T) (Supplemental Fig. S3E). Because TRIM28-FAM208A genes were most enriched in LINEs >5 kb (Fig. 3F; Supplemental Fig. S3D), we selected all of these (within 20 kb), ordered them by their evolutionary age, and found 98% to be <5 million years old and absent from rat genomes. We also ordered them by mean divergence (Supplemental Fig. S3F) and found them all to be <10% diverged from consensus sequences and the L1Md_A integrants to be the youngest by this method (mean divergence 1.53%).

11In a complimentary approach, we mapped mRNA-sequencing reads to Repbase, which showed repeats to be overexpressed in all four treatment groups (Fig. 3I; Supplemental Table S5). Scoring the top five repeats derepressed in each treatment group showed that IAP elements are co-repressed by TRIM28, SETDB1, and ATRX, whereas FAM208A mainly represses young L1s (from the TF and GF families) (Supplemental Fig. S3G). In total, 25 families of repeats were co-repressed by TRIM28 and FAM208A (Fig. 3J). This included 13 L1 families that we classified by age and found 77% to be <3 million years old (Fig. 3K). The rest of the TRIM28-FAM208A co-repressed repeats fell into the ERV or satellite classes with satellites most highly derepressed (14×) (Fig. 3L,M). Overall, these data suggest that TRIM28 and FAM208A co-repress young L1s. Of note, we did not assess polymorphic or de novo L1 insertions.

12LTRs provide genetic material (promoters, enhancers, and first exons) to create new genes or new expression patterns, and active L1s can create new genes through retroposing cellular mRNAs. We asked if TRIM28-FAM208A genes (Supplemental Table S6) were enriched in new genes, (which we define here as mouse-specific) and tissue-specific genes. We first found 41% of genes were not conserved across placental mammals and were mouse-specific, compared to 14% of random genes (Fig. 4A, left); focusing only on the protein-coding genes and their last common ancestor as a measure of their evolutionary age also revealed TRIM28-FAM208A genes to be enriched in mouse-specific genes (11.8% compared to 0.9% of all genes in the mouse genome) (Fig. 4A, right). Forty percent of TRIM28-FAM208A genes exhibited tissue-specific expression patterns (Fig. 4B), whereas 44% had an unknown expression pattern and this group was enriched for new genes (Fig. 4B). We verified that example loci of these new or tissue-specific genes were associated with arrays of young L1s or ERVs and epigenetic regulation (Fig. 4C; Supplemental Fig. S4). Finally, we observed that LTRs 3 kb upstream of TRIM28-FAM208A genes were biased to reside in a sense orientation (69%) suggesting they may function as promoters (Fig. 4D).

13We asked if FAM208A mainly regulates young L1s because it binds selectively to young L1s. We addressed this by performing ChIP-seq using an antibody recognizing mouse FAM208A and mapped reads to Repbase. FAM208A binds a range of retrotransposons, primarily ERVs and L1s (Fig. 5A; Supplemental Table S7), coating their 3′ halves (Supplemental Fig. S5A). Repbase L1s bound by FAM208A (22 families with an enrichment of more than 4×) were inactive families lacking full-length copies mostly older than 13 million years, because they were present before the mouse–rat split (Fig. 5B). We therefore also mapped reads to the genome to see if we could detect FAM208A binding to young L1s. We found 1045 FAM208A peaks (Fig. 5C; Supplemental Table S8), which clustered together (75% of ChIP-seq peaks were within 50 kb of another FAM208A peak) (Fig. 5D), suggesting FAM208A binding spreads with 34% of FAM208A peaks overlapping H3K9me3 (Fig. 5E; ). We could detect FAM208A binding to young L1s: Of 194 peaks targeting L1s, 6% of them were young L1s, whereas the rest were inactive L1s (Fig. 5F; for an example of a bound young L1, see Supplemental Fig. S5B). FAM208A may spread to young L1s mainly from its tethering to inactive L1s and ERVs. In support of this, we found that 61% and 87% of TRIM28-FAM208A genes (that are enriched for young L1s) (Fig. 3H) contained a FAM208A-bound L1 or an ERVK within 20 kb, respectively (Fig. 5G).

14Mechanistic studies showed that like TRIM28 and SETDB1, FAM208A contributes to H3K9me3 maintenance at the locus Zfp180, which was bound by TRIM28 and FAM208A and at global IAP and L1 elements (Fig. 6A,B; Supplemental Fig. S6A). This decrease in H3K9me3 was sufficient for up-regulation of Zfp180 (Supplemental Fig. S6B), an increase in H3K27ac (Fig. 6C), and retrotransposon reactivation (Fig. 2). The greatest shift in H3K27ac was apparent at L1s (4.1× for FAM208A) (Fig. 6C), at which we observed most derepression (13×) (Fig. 2). Of note, levels of preexisting DNA methylation were not affected in knockdown samples (Supplemental Fig. S6C). At baseline, L1s exhibited lower levels of H3K9me3 (2.4× less), DNA methylation (3× less), and TRIM28 binding (2.5× less), suggesting that they recruit “leaky” heterochromatin and as such are readily reactivated (Fig. 6B,D,E). This fits with the enrichment we observed of L1s within active TRIM28-FAM208A genes (Fig. 3F; Supplemental Fig. S3D), which recruited less H3K9me3 than TRIM28-SETDB1 genes (63% versus 77%) (Fig. 3C). Finally, we found that the HUSH complex is required to repress L1 elements and the TRIM28-repressed locus ZNF274 in human embryonic cells, again contributing to H3K9me3 maintenance (Fig. 6F–H). L1 elements displayed lower H3K9me3 (1.4× less), DNA methylation (3.7× less), and TRIM28 binding (3.8× less) compared to SVA elements (Fig. 6H–J), perhaps explaining their adept derepression (4.2× upon Fam208a depletion) (Fig. 6G). Furthermore, we could detect FAM208A binding to L1 elements in human cells (Fig. 6J).

15A key question has been whether the HUSH complex participates in retrotransposon repression and whether it collaborates with TRIM28 and its cofactors. Here, we show that the HUSH complex and TRIM28 exert nonredundant roles at evolutionarily young L1s of <5 million years old in naïve pluripotent cells. These young elements are likely prone to derepression because they bind TRIM28 and FAM208A only weakly. In the case of TRIM28, weak binding to young retrotransposons is known to result from them escaping KZNF recognition due to sequence divergence. In contrast, TRIM28 and FAM208A are strongly associated with ERVs, which recruit dense heterochromatin and at which only TRIM28 but not HUSH is required for repression. We found FAM208A binding spreads through chromatin, consistent with the ability of HUSH to mediate position-effect variegation. Our data reveal that TRIM28 and FAM208A coregulate new genes and tissue-specific genes, which are associated with young L1s. This suggests that regions of the genome that are enriched in young L1s and leaky heterochromatin/part euchromatin may be hot spots for the evolution of new genes and new regulation of existing genes. Such genes may hijack incomplete epigenetic repression to gain tissue-specific expression (for a summary model, see Fig. 7). TRIM28 and FAM208A cooption into these gene regulatory networks, therefore, appears to be a byproduct of their leaky regulation of young retrotransposons.

16In summary, this work illustrates the complexity of epigenetic repression pathways that have evolved to regulate cellular genes through retrotransposon sequences. Future work focused on specific retrotransposon integrants will enable us to understand how these elements have been coerced to roles in gene regulation and chromatin organization in mammalian cells.

17Human embryonal NTERA-2 cells (NT2/D1, a kind gift from Peter Andrews, University of Sheffield) were grown in Dulbecco's Modified Eagle's Medium (DMEM, Gibco) high glucose with 2 mM L-Glutamine, 10% fetal calf serum (FCS), and 1% Penicillin/Streptomycin (P/S). They were split in half by cell scraping. 293T cells were grown in standard DMEM + 10% FCS and P/S. J1 ESCs (129S4/SvJae) or their derived triple knockout (TKO) cells, which are knockout for Dnmt3a, Dnmt3b, and Dnmt1 (from M. Okano) were used where stated, or as an independent mESC line, ES3 ESCs. All mESCs were cultured on gelatin-coated plates (0.2%), either in standard media as previously described, or in media containing two small-molecule inhibitors of MEK and GSK3 plus LIF (2i + LIF), as described. Cells were split every other day using accutase (Gibco, A11105-01) and cultured in 2i media for 7 d before transduction.

18Cells (1 × 106 per condition) were fixed and permeabilized using intracellular staining buffers (eBioscience, 00-5523) and stained with POU5F1-PE (eBioscience, 12-5841) or isotype control (eBioscience, 12-4321) antibodies, washed, and analyzed by flow cytometry.

19Dual luciferase assays were performed at ratios detailed elsewhere. 293T cells were plated at 5 × 104 cells per well in a 24-well plate and transfected with 200 ng of KZNF plasmid, 20 ng of luciferase reporter plasmid, and 2 ng of pRT-TK_Renilla luciferase-encoding control plasmid using 1.5 µL of FuGENE 6 (Promega) per well in triplicate wells (or for ESCs, we used 500 ng KZNF, 50 ng luciferase reporter, and 5 ng Renilla luciferase plasmid in 12-well plates). Forty-eight hours post-transfection, cells were lysed, and luciferase was measured using the Dual Luciferase assay kit (Promega, E1910), an opti-plate, and a GloMax 96 microplate Luminometer (Promega) using the Dual Glow program. Raw luciferase values were normalized to Renilla luciferase values to control transfection efficiency.

20Total RNA was extracted using an RNeasy mini kit (Qiagen), treated with DNase (Ambion AM1907), and 500 ng was reverse transcribed using SuperScript II (Thermo Fisher Scientific) and random primers following the manufacturer's instructions. Samples were run on an ABI 7500 Real Time PCR System (Applied Biosystems) using SYBR green Fast PCR mastermix (Life Technologies). CT values for test genes were normalized to B2M and GAPDH for human genes or Cox6a1 and Gapdh for mouse genes using the −ΔΔCt method to calculate fold change. See Supplemental Table S1 for primers.

21293T cells were washed with PBS and lysed in 2X Sample Buffer (Invitrogen) supplemented with 5% β-mercaptoethanol (Sigma). ESCs were washed 2× in ice-cold PBS and resuspended in radioimmunoprecipitation buffer with protease inhibitors added (Roche 11836170001) and quantified using the BCA Assay (Pierce). Ten micrograms was loaded per well on mini Tris/glycine gels, run in Tris/glycine/SDS buffer and mini-PROTEAN tanks (Bio-Rad) or on precast Bis-Tris gels (4%–12% or 8% gels), run in MOPs SDS buffer, followed by wet transfers to PVDF membranes. Primary antibodies are in Supplemental Table S2. Secondary antibodies were horseradish peroxidase-conjugated (GE healthcare), and membranes were developed using ECL kits (Amersham).

22The luciferase reporter plasmids named Empty (pGL4cp-OCT4Enh-SV40), SVA_VNTR (pGL4cp-VNTR OCT4Enh E2), and L1PA4 (pGL4cp-L1PA4 OCT4Enh E2), and the human KZNF expression plasmids ZNF91 (pCAG ZNF91 HA) and ZNF93 (pCAG ZNF93) were a kind gift from David Haussler. Dual promoter lentiviral vectors were used for RNAi, encoding both hairpin and puromycin resistance gene (either a HIV SIREN backbone was used for human cells, from Greg Towers, or pLKO.1 for mouse cells from Dharmacon or Sigma-Aldrich). Hairpin sequences were designed (http://bioinfo.clontech.com/rnaidesigner/sirnaSequenceDesignInit.do) and annealed and cloned into BamHI-EcoRI sites. The shRNA pLKO.1 plasmid for SETDB1 was from Miguel Branco. See Supplemental Table S3 for shRNA sequences. VSV-G-pseudotyped lentiviral vectors were produced by FuGENE 6 (Promega) cotransfection of 293T cells in 10-cm plates with 1.5 µg shRNA-encoding plasmid, 1 µg p8.91, and 1 µg pMDG2 encoding VSV-G. The harvested supernatant was used unconcentrated for cell lines or concentrated by ultracentrifugation (20,000g for 2 h at 4°C) for primary cells.

23Mouse 2i + LIF cultured J1 ESCs treated with different shRNA vectors were used for mRNA-sequencing. Cells were cultured for 5 d following puromycin selection before RNA extraction, and samples from three independent experiments were used. See Supplemental Methods for further details.

24Reads were mapped to rodent Repbase (https://www.girinst.org/downloads/) and the latest release downloaded (Repbase 20.06 used here). The SAMtools v.1.19 idxstats utility was used to extract the number of mapped reads per repeat, that were inputed into the R package DESeq2 (https://bioconductor.org/packages/3.2/bioc/html/DESeq2.html) to identify differentially expressed repeats between samples depleted of epigenetic modifiers and controls, as previously described. P-values were adjusted for multiple testing with the Benjamini-Hochberg false discovery rate (FDR) procedure.

25293T cells were harvested using trypsin, whereas NTERA-2 cells and 2i + LIF grown mESCs were harvested using accutase, and chromatin was cross-linked, quenched, and prepared as described, except that sonication was performed on a Bioruptor (Diagenode). See Supplemental Methods for further details.

26DNA was purified with a DNeasy Blood and Tissue Kit (Qiagen), and 1 µg DNA was used for bisulfite conversion using an EpiTect Bisulfite Kit (Qiagen). Four microliters converted DNA was amplified by PCR using primers in Supplemental Table S1, which are from or were designed using http://urogene.org/methprimer/, and PCR products were cloned using the TOPO TA-Cloning Kit (Thermo Fisher Scientific), and at least 10 colonies were sent for sequencing using the T7P primer. Results were analyzed using the QUMA online tool (http://quma.cdb.riken.jp) from the Riken Institute.

27All data in the figures are presented as the standard deviation (where there are three or more samples) or by standard error of the mean (SEM) and assessed by unpaired or paired two-tailed Student t-tests (see figure legends for details). A P-value of <0.05 was considered statistically significant (***P < 0.001; **P < 0.01; *P < 0.05).

28mRNA-sequencing and ChIP-sequencing data and processed files from this study have been submitted to the NCBI Gene Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/geo/) under accession number GSE107840, and processed files are also included in Supplemental Material.