Aicardi–Goutières Syndrome associated mutations of RNase H2B impair its interaction with ZMYM3 and the CoREST histone-modifying complex

DNA-RNA hybrids arise in all cell types, and are removed by multiple enzymes, including the trimeric ribonuclease, RNase H2. Mutations in human RNase H2 result in Aicardi–Goutières syndrome (AGS), an inflammatory brain disorder notable for being a Mendelian mimic of congenital viral infection. Previous studies have shown that several AGS-associated mutations of the RNase H2B subunit do not affect trimer stability or catalytic activity and are clustered on the surface of the complex, leading us to speculate that these mutations might impair important interactions of RNase H2 with so far unidentified proteins. In this study, we show that AGS mutations in this cluster impair the interaction of RNase H2 with several members of the CoREST chromatin-silencing complex that include the histone deacetylase HDAC2 and the demethylase KDM1A, the transcriptional regulators RCOR1 and GTFII-I as well as ZMYM3, an MYM-type zinc finger protein. We also show that the interaction is mediated by the zinc finger protein ZMYM3, suggesting that ZMYM3 acts as a novel type of scaffold protein coordinating interactions between deacetylase, demethylase and RNase H type enzymes, raising the question of whether coordination between histone modifications and the degradation of RNA-DNA hybrids may be required to prevent inflammation in humans.


Introduction
Aicardi-Goutières syndrome (AGS) is a rare, largely autosomal-recessive disorder characterised by microcephaly, basal ganglia calcification and elevated levels of lymphocytes and interferon-alpha in the cerebrospinal fluid, with occasional extra-neurological involvement of the liver, spleen and skin [1,2]. AGS usually presents at birth or within the first few months of life, and is strikingly reminiscent of congenital viral infection of the brain, although the failure to find a causative pathogen suggests that AGS may be a disorder of the immune system [2]. The AGS phenotype is thought to result from elevated levels of interferon-alpha, and can be  , ranked by number of peptides identified. Results for the region corresponding to band 4 from two independent pulldowns are listed separately. The most abundant peptides correspond to the RNase H2 subunits and known associated proteins such as PCNA (in black). The common protein identified in band 4 corresponds to the zinc finger MYM-type protein 3 (bold). (D) Endogenous untagged ZMYM3 is associated with endogenous RNase H2B. The schematic representation indicates the protein size in amino acids (aa). Rabbit anti-RNase H2B immunoprecipitaes from HEK293T cell lysates compared control rabbit IgG from serum. The same result was observed in three independent pull-downs. The input corresponds to~6% of the lysates used for IP. Rabbit anti-ZMYM3 was developed forms of human RNase H2B in HEK293T cells, immunoprecipitated these along with any interacting proteins, and then compared all binding partners of wild-type and mutant RNase H2 by SDS-PAGE and silver staining, using an empty vector transfection as control. This strategy reproducibly identified a single band, in the wild-type RNase H2B samples that was lost (S159I and K162T) or partially lost (T163I) in the mutants (Fig 1B and S1 Fig). This band identified a single protein of molecular weight between 160kDa and 260kDa which did not match any of the known binding partners of RNase H2. Bands corresponding to approximately 39kDa, 35kDa and 18kDa ( Fig 1B) were excised and as expected were identified by mass spectrometry as RNase H2B, RNase H2A and RNase H2C, respectively ( Fig 1C). To focus on identifying the large unknown protein present in the RNase H2B sample but lost in the mutant samples (indicated in Fig 1B by an open arrow), we repeated the pull-down and excised the region between 160kDa and 260kDa from Flag-tagged RNase H2B and controls after denaturing SDS-PAGE and SYPRO RUBY-staining and subjected the excised region to mass spectrometry. Multiple peptides corresponding to the uncharacterised zinc finger protein ZMYM3 were reproducibly recovered from the RNase H2B pull down ( Fig 1C) but not in the control sample. The interaction was further recapitulated in HEK293T cells by co-immunoprecipitation of endogenous RNase H2B and ZMYM3 using a previously-validated antibody [18] (Fig 1D). No difference was apparent in the stoichiometry of the A and C subunits co-immunoprecipitated by mutant and wild-type RNase H2B (Fig 1B), confirming previous observations that suggested the AGS associated cluster of mutations in RNase H2B had no significant effect on the stability of the complex [16,17]. To exclude any effect of these mutations on the cellular localisation of RNase H2, wild-type and mutant forms were transiently expressed in HEK293T cells and visualised by confocal microscopy. The nuclei of cells transfected with RNase H2B displayed punctate regions of higher intensity staining (S2 Fig), consistent with previously-described RNase H2-containing replication foci [19], but showed no difference in the localisation of wild-type and mutant RNase H2B (S2 Fig).
Having confirmed the interaction between endogenous RNase H2B and endogenous ZMYM3, we next confirmed the loss of interaction in the RNase H2B AGS-associated mutants by western blot (Fig 2A). ZMYM3 is a known interactor of several chromatin-modifying and transcription factors [20], including HDAC2, KDM1A (LDS1), GTFII-I and RCOR1 (CoR-EST), therefore we also monitored the presence of these proteins in the complex associated with ZMYM3/RNase H2B by western blot. As can be seen in Fig 2A, this revealed each of these proteins to be associated with RNase H2B, with each of these interactions disrupted by each AGS-associated mutation of the S159I-K162T-T163I cluster, and to a lesser degree by the V185G mutation. By contrast, none of the mutations disrupted the well-characterised association of RNase H2B with PCNA [19] (Fig 2A), suggesting that the mutations specifically impair the interaction of RNase H2 with the ZMYM3 protein and the KDM1A/HDAC/CoREST complex.

RNase H2 binds to a C-terminal region of ZMYM3 containing a 'PXP' motif
To establish the region(s) of ZMYM3 responsible for mediating its interaction with RNase H2B, we systematically probed the binding of RNase H2B to C-terminal truncations of with goat anti-Rabbit Peroxidase, and the rabbit anti-RNase H2B with TrueBlot anti-rabbit IgG Peroxidase. (  and RNase H2B are associated as part of the larger CoREST complex. Transfected FLAG-RNase H2B pull downs in HEK293T cell extracts brings down ZMYM3 and CoREST, but both are lost or partially lost in mutants S159I, K162T and T163. On the left, individual developed films are shown as an overlap to allow comparison of the different protein sizes. The input on the right (~3% of the lysates) is shown for the endogenous RNase H and members of the CoREST complex as well as the Flag-RNase H2B transfected protein. Antibodies used for western blotting were Rabbit anti-ZMYM3, Mouse anti-FLAG M2 Peroxidase, Rabbit anti-RNase H2C, Rabbit anti-GFII-I, Rabbit anti-CoREST, Rabbit anti-HDAC2, Goat anti-Rabbit IgG Peroxidase and Goat anti-Mouse IgG Peroxidase. The � indicates an isoform of GTFII-I. The same result was observed in four independent pull-downs. (B) HA-tagged ZMYM3 protein fragments in HEK293T cell extracts co-immunoprecipitated with FLAG-tagged RNase H2B. The schematic shows the domain architecture of ZMYM3 and the C-terminally truncated fragments, including the nuclear localisation region from the DUF3504 domain (black). The first and last amino acids of full and truncation proteins are indicated according to fulllength human ZMYM3 (UniProtKB: Q14202.2). PPD: Proline-rich region. NLS: Nuclear localisation signal. Each pull ZMYM3 designed on the basis of predicted secondary structure features and domains ( Fig  2B). HA-tagged ZMYM3 truncation mutants were co-expressed with FLAG-tagged RNase H2B in HEK293T cells, co-immunoprecipitated with anti-FLAG beads and their binding documented by SDS-PAGE and western blot (Fig 2B and 2C). We identified a region encompassing aa 863 to 942 of ZMYM3 that was required for the co-precipitation of FLAG-RNase H2B and ZMYM3 fragments, and which contained a repeating proline-valine/isoleucine-proline, or PXP motif (see S3 Fig). This was confirmed by internal deletion of aa 863-942 which completely abrogated the interaction, whereas deletion of the neighbouring polyproline domain aa 813-827 had no effect ( Fig 2C). Loss of the interaction with the truncation mutants cannot be attributed to a reduction in their abundance ( Fig 2C and 2D), nor to a change in their cellular localisation, which was unaffected when transfected in HEK293T and monitored by confocal microscopy (S4 Fig).

Distinct zinc fingers of ZMYM3 interact with HDAC2, KDM1A, CoREST and GTFII-I
The multi-domain nature of ZMYM3 raises the possibility that it could function as a scaffold protein, acting as a platform to bind together RNase H2 and HDAC2, KDM1A, CoREST and GTFII-I. Having established the ZMYM3 binding site of RNase H2, we therefore sought to establish the ZMYM3 binding sites of HDAC2, KDM1A (LSD1), RCOR1 (CoREST) and GTFII-I. As the eighth and ninth zinc fingers of the closely-related ZMYM2 protein have been shown to bind to the complex comprising HDAC1, KDM1A and RCOR1 [21], we sought to establish whether the highly-conserved eighth and ninth zinc fingers of ZMYM3 would play a similar role by constructing a set of N-terminally FLAG-tagged zinc finger arrays (see Fig 3A  schematic), all of which were C-terminally appended with the ZMYM3 bipartite NLS contained within the C terminal domain of unknown function (DUF3504) to ensure nuclear localisation. Expression and immunoprecipitation of these zinc finger arrays in HEK293T cells revealed that, as expected, deletion of the eighth and ninth zinc fingers (fragment 304-757 [Δ620-719] ) significantly diminished (but did not abolish) interaction with endogenous RCOR1 and KDM1A, and that either one of these zinc fingers alone (fragments 620-678 or 661-719) was sufficient to bind to RCOR1 and KDM1A ( Fig 3A). Using the same assay with HA-tagged zinc finger arrays, we showed that the first zinc finger of ZMYM3 (fragment 304-352) is both necessary and sufficient for the interaction with the transcription factor GTFII-I endogenous protein (Fig 3B).

RNase H2 can also interact with the ZMYM3 homologues ZMYM2 and ZMYM4
As its name suggests, ZMYM3 is a member of a family of proteins, the six members of which each possess arrays of MYM-type zinc finger motifs [22]. As shown in Fig 2B and 2C, the Cterminal region (aa 862 to 943 including the PXP motif) of ZMYM3 is required for interaction with RNase H2. Presumably ZMYM proteins containing homologous regions may interact down is shown next to lysates transfected with HA-tagged ZMYM3 fragments and Flag-vector controls. The input equals 3% of the total lysate. Antibodies used to visualise the proteins were Mouse anti-FLAG M2 Peroxidase and Rat anti-HA Peroxidase. (C) Small internal deletions of the conserved proline rich region of motif ZMYM3 prevent its interaction with FLAG-tagged RNase H2B. As B, amino acids shown in the schematic on either side of the deleted segment are according to full-length human ZMYM3 (UniProtKB: Q14202.2). Each pull down is shown next to lysates transfected with HA-tagged ZMYM3 fragments and Flag-vector controls. The input equals 3% of the total lysate. Antibodies used to visualise the proteins were Mouse anti-FLAG M2 Peroxidase and Rat anti-HA Peroxidase. The same results were observed for B and C in three independent pull-downs each.
https://doi.org/10.1371/journal.pone.0213553.g002  protein and ZMYM1 and ZMYM6 contain a different C-terminal region with structural homology to the RNAse-H fold. Using overexpressed HA-tagged proteins we showed that both ZMYM2 and 4 clearly co-immunoprecipitate with FLAG-RNase H2B, whereas ZMYM1 and ZMYM6 do not (Fig 4B). ZMYM5 was very highly expressed and bound non-specifically to beads used for Co-immunoprecipitation of RNase H2B and ZMYM family members, making it impossible to study its interaction (or lack thereof) with RNase H2B. As before, confocal microscopy of transfected ZMYM proteins suggested that the differences in the binding to RNase H2 are not due to lack of subcellular co-localisation, with the possible exception of ZMYM1, which showed a peri-nuclear distribution (S5 Fig).

ZMYM3 is not required for the stability or genome protection function of RNase H2B in murine ES cells
The results presented above suggested to us that ZMYM3 may act as a scaffold protein to bring together RNase H2 and members of a histone-modifying complex, for the purpose of transcriptional regulation, DNA repair, or other processes. To determine whether ZMYM3 is required for any of these functions, we investigated the in vitro phenotype of targeted While expression of the ZMYM3 protein was completely lost in the targeted ES cells, we observed no change in the levels of γ-H2AX or RNase H2B, suggesting that the interaction of ZMYM3 with RNase H2 is not required for the role of the latter in maintaining genomic stability in the absence of external damage or cellular stress in ES cells (Fig 4C and 4D).

Discussion
The experiments presented here identify RNase H2 as a novel component of the co-repressor of transcription KDM1A/RCOR1/HDAC2 LCH complex, via its interaction with ZMYM proteins ZMYM3 and ZMYM2. ZMYM2 and 3 have been shown to associate with the KDM1A/ RCOR1/HDAC2 complex [20], although their precise functions within the complex are not known. CoREST was originally identified as a repressor enforcing the transcriptional silencing of neuronal genes in non-neuronal tissues and neuronal stem cells [24] but it also acts on different lineage-specific genes, for example repressing non-erythroid gene expression in haematopoietic stem cells and pro-inflammatory responses in astrocytes [25] or acting to regulate enhancer activity and transcriptional activation of androgen receptor regulated genes in human prostate cell lines [26]. The multiplicity of functions associated with the CoREST complex are a result of the diverse composition of these multiprotein complexes in different cell types and at different developmental stages, but are generally associated with transcriptional silencing and depend on the core KDM1A/RCOR1/HDAC (LCH) complex and its histone H3 lysine4 and lysine9 demethylase activity and histone deacetylase activities [27].
Our results identify several novel interactions of RNase H2, including the uncharacterised zinc finger protein ZMYM3, the transcription factor TFII-I (GTFII-I), the transcription repressor RCOR1, the histone deacetylase HDAC2 and the histone demethylase LSD1 (KDM1A), all of which bind to ZMYM3 [20]. We further identify the domains within ZMYM3 required for the interactions with the components of the CoREST complex. Our data suggest that the zinc finger domain 1 (aa 304-352) mediates the interaction with GTFII-I, quite separately from the zinc finger domains 8 and 9 (aa 620-719) which are required for binding to the main KDM1A/RCOR1/HDAC2/complex, whereas the highly conserved proline rich region in the C terminal domain of ZMYM3, the PXP repeat-containing region (aa 862-949), is sufficient to confer binding to RNase H2.
Although our analyses cannot formally exclude mislocalisation or misfolding of the fragmented proteins used to map the interactions, at least in the case of the interaction with RNase H2B, removal of the PXP repeat-containing region (aa 863-942) either directly (ZMYM3 [Δ863-942] ) or by C-terminal truncation of the protein (ZMYM3 [Δ863-1370] , ZMYM3 [Δ944-1370] , ZMYM3 [Δ1124-1370] , ZMYM3 [Δ1183-1370] ) did not lead to major changes in the nuclear distribution (S4 Fig). In the case of the zinc finger domain1-GTFII-I and the zinc finger domain 8/ 9-LCH interactions, the interactions are lost by truncation of discrete zinc fingers, which is less likely to result in misfolding due to the ability of zinc fingers of the MYM-type to fold autonomously through their zinc-binding residues [21]. In addition, the fact that isolated zinc finger domain 1 or zinc finger domain 8/9 alone are sufficient to maintain the interaction with GTFII-I and LCH respectively, supports the notion that they are responsible for mediating these interactions in full-length ZMYM3.
While this work clearly implicates the zinc finger domain 8 and 9 of ZMYM3 in binding to KDM1A/RCOR1/HDAC2 LCH complex, the precise architecture of these interactions remains unclear. Previous studies with recombinant ZMYM2, KDM1A/LSD1, CoREST and HDAC1 revealed the interaction of CoREST with ZMYM2 to be dependent on the presence of both KDM1A/LSD1 and HDAC1, and likewise, the interaction of HDAC1 with ZMYM2 to be dependent on the presence of an KDM1A/LSD1-CoREST fusion protein, altogether suggesting that an intact LCH ternary sub complex is required to allow interaction with ZMYM2 [21]. Given the high degree of conservation between ZMYM3 and ZMYM2, and in particular of their LCH-interacting zinc fingers, a similar arrangement seems likely to exist in the case of ZMYM3. However, which of the three LCH members bind directly to the zinc finger domains 8/9 remains to be elucidated for either ZMYM2 or ZMYM3.
The role of the ZMYM3 zinc finger domains 8/9 in binding to KDM1A/RCOR1/HDAC2 could have been anticipated from similar studies of ZMYM2, however, the finding that the zinc finger domain 1 alone is necessary and sufficient for binding to GTFII-I was unexpected but consistent with the structural properties of similar zinc-coordinated domains. As initially described for the zinc-finger LIM domain of zyxin, a single domain is sufficient for its interaction with cysteine-rich protein [28]. Recent data has identified an overlap between the region of ZMYM3 that directly binds dsDNA (aa 300-330) [29] and the region that we identified as binding GTFII-I (aa 304-352), raising the question of whether the DNA binding activity of ZMYM3 is direct. Nonetheless, the fact that GTFII-I and LCH bind to distinct zinc finger domains of ZMYM3 sheds new light on the MYM-type zinc finger domain and the ZMYM3 protein, by showing that it can bring together different proteins in a modular fashion, both within the zinc finger domain region (where individual zinc fingers can act as discrete protein binding modules), and across the protein as whole, with non-zinc finger regions of ZMYM3 (e.g. the PXP-containing region) binding to additional proteins such as RNase H2 (S6 Fig). Although this arrangement suggests a possible scaffolding role for ZMYM3, confirmation of this idea will require further structural studies of the protein and its partners.
By diminishing the interactions of RNase H2B with ZMYM3, it is conceivable that the AGS-mutations investigated here may reduce the stability of the RNase H2 complex, as has been observed for several other AGS-associated RNase H2 mutations that compromise the interaction between the C and A or C and B subunits [13,30]. However, this seems unlikely, as interaction with the A and C subunits is not affected by the mutants in this study, and such a loss of stability would also be expected to affect the abundance of RNase H2B, which is not seen with these mutations (Figs 1 and 2). While the data presented here suggest that loss of interactions between RNase H2 and ZMYM3/GTFII-I/CoREST/HDAC2/KDM1A may result in AGS pathology, and that these interactions may be co-ordinated by ZMYM3, the identification of these interactions raises the more fundamental question: why do they exist at all? This, to our knowledge, is the first time that an RNase H-type enzyme has been shown to interact with a histone-modifying complex, and raises the intriguing possibility that the function(s) of this complex may be dependent on the function(s) of RNase H2, or vice versa. While there is some evidence to suggest that RNase H2 is required for the resolution of R-Loops [10], its best-established role is in the excision of misincorporated ribonucleotides from DNA, failure of which results in catastrophic genomic instability [11,31]. Furthermore, loss of RNase H2 results in accumulation of cytoplasmic aggregates of DNA [32]. The identification of these interactions therefore raises the possibility that the efficient repair of ribonucleotides in vivo may require associated changes in chromatin structure (namely, demethylation by KDM1A/ LSD1 and deacetylation by HDAC2), or vice versa, and that failure to co-ordinate these processes in humans may contribute to the inflammatory brain disorder AGS. The well-established but functionally poorly understood interaction of lncRNAs or small RNAs with several chromatin remodelling and silencing complexes like CoREST or the NuRD and polycomb complexes [33] might explain the advantages of functional association of such complexes with an RNase H2 enzyme, that can efficiently remove RNA in the context of chromatin.
Failure to coordinate the activity of RNase H2 and the CoREST complex could also lead to an alternative source of nucleic acid accumulation and inflammation, by impairing the ability of the CoREST complex to suppress the expression of endogenous transcripts [25] or by permitting spurious transcription [34,35]. While RNase H2 appears to facilitate, rather than inhibit, LINE-1 retrotransposition [32,36], cell lines with RNase H2 mutations also have significantly reduced overall levels of genomic DNA methylation, compared to cell lines derived from TREX1 or SAMHD1 AGS patients [37], suggesting that silencing of genomic regions (usually correlated with stable DNA methylation) is impaired in cells lacking fully functional RNase H2. Given that mutations of ZMYM3 have been implicated in the pathogenesis of medulloblastoma [38], and intellectual disability syndromes [39,40], it will be interesting to establish whether the loss of this putative scaffolding ability plays a role in either of these pathologies.
Regarding the RNase H2/ZMYM3 interaction, our data suggest that other members of the ZMYM family might act redundantly, since we did not observe mayor abnormalities in mouse ES cells devoid of ZMYM3, unlike the differentiation block associated with KDM1A/LSD1 deficiency [41]. Recent data have shown an alternative function for ZMYM3 in facilitating the repair of DNA damage by homologous recombination by acting as a bridge in the recruitment of BRCA1 to chromatin. Our own data would support the idea of ZMYM3 acting as a bridge to coordinate chromatin remodelling and silencing in the context of ribonucleotide excision repair [29]. It will therefore be important to investigate the association of the RNase H2 with the ZMYM3/CoREST complex in tissues where both CoREST and ZMYM3 are likely to have specific nonredundant functions, such as the brain.

Assembly of RNase H2 and ZMYM constructs
N-terminally FLAG-tagged RNase H2B and mutant versions were assembled by PCR using a cDNA template isolated from a human spleen cDNA library (Life Technologies) and using primers introducing an Nhe I restriction site, KOZAK consensus sequence GCCACC, glycinealanine linker and FLAG (DYKDDDDK)-encoding sequence 5' of the protein-coding region (primer 2B N-FLAG Fwd), and a Not I restriction site 3' of the protein-coding region (primer 2B N-FLAG Rev). Nhe I and Not I PCR products and the vector pIRESpuro3 (Clontech) were digested with NheI and NotI, gel purified, ligated to one another to produce pIRESpuro3-2B (or equivalent mutant). pIRESpuro3-2B (or equivalent mutant) and pCDNA4/TO-IRES-GFP, a mammalian expression vector which expresses the gene of interest and EGFP under the control of a CMV promoter (Addgene), were digested with NdeI and NotI, gel purified (the NdeI/ NotI fragment containing the RNase H2B gene and the NdeI/NotI fragment containing the backbone of pCDNA4/TO-IRES-GFP) and ligated to produce pCDNA4/TO-2B-IRES-GFP (or mutant equivalent).
ZMYM3 truncation mutants were generated by PCR from pCMV-SPORT6-ZMYM3 as before. All C-terminal truncations (unless otherwise indicated) contained two nuclear localisation signals from the DUF3504 domain of ZMYM3 immediately 3' of the truncation site while the zinc finger truncation mutants all included the KOZAK/GA/HA module as well as the two predicted nuclear localisation signals from the DUF3504 domain of ZMYM3 and a stop codon. FLAG-tagged ZMYM3 zinc constructs were assembled by PCR as before into pCDNA4/TO-IRES-tdTomato. All primers are listed in S1 Table. HEK293T cell culture, transfection and immunoprecipitation HEK293T cells in Dulbecco's Modified Eagle's Medium (DMEM) supplemented with 100 μg ml -1 penicillin and streptomycin (Sigma) and 10% foetal bovine serum (FBS) (Life Technologies) were transfected with 1 μg of plasmid DNA using GeneJuice Transfection Reagent (Novagen) as per the manufacturer's instructions. Expression of the fluorescent reporter genes was used to monitor transfection efficiency and~4x10 6 cells were washed in PBS, harvested and lysed in 20 mM HEPES, 150 mM NaCl, 0.2% Triton X-100 (Sigma), 1X Complete Protease Inhibitor (Roche), 1 mM Dithiothreitol, 2 mM MgCl 2 and 3 μl/ml Benzonase (Novagen) at 4˚C for 45 mins under gentle rotation. Cell lysates were precleared by centrifugation 19,000 RCF for 20 mins and 460 μl added to pre-washed 15 μl of anti-FLAG M2 affinity gel (Sigma) with30 μl of each lysate set aside to monitor protein input and incubated at 4˚C under gentle agitation. Samples were washed 5x in 500 μl PBS supplemented with 0.1% Triton X-100 (Sigma), centrifuged at 100 RCF for 2 mins in Proteus Clarification Mini Spin Columns (Generon) and the bound proteins eluted 2x in 215 μl 3X FLAG peptide at 10 ng/μl (Sigma) and resuspended in 30 μl NuPAGE LDS Sample Buffer (SB) (Life Technologies) supplemented with 50mM DTT.
For the immunoprecipitation of endogenous RNase H2B and ZMYM3, 10 μl of Dynabeads Protein A beads (Life Technologies) were twice washed in 1ml of RIPA buffer, and resuspended in 100 μl of RIPA, (with 10 μl of 10 μg/μl purified BSA (NEB)) at 4˚C for approximately 45mins, washed three times in 1ml of RIPA, and resuspended in 100 μl of RIPA and then added to 460 μl cell lysates in the presence of 1 μg of rabbit anti-RNase H2B or reagent grade IgG from rabbit serum (Sigma) approximately 0˚C for two to three hours. Beads were washed 3x in 1ml RIPA buffer (in 1% Triton X-100) and the pellet resuspended in 30 μl NuPAGE LDS Sample Buffer (SB) (Life Technologies) supplemented with 50mM DTT.

Protein identification
Proteins (typically 20 μl per sample) were separated by SDS-PAGE NuPAGE 4-12% Bis-Tris Gels in 1X NuPAGE MOPS SDS (Life Technologies) and transferred into Immobilin P PVDF membranes (Millipore) in a tris-glycine buffer containing 10% methanol. Membranes were washed 3x5mins in PBS supplemented with 0.05% TWEEN20 (NBS Biologicals) (PBST) under gentle agitation, blocked in PBST supplemented with 5% w/v dried skimmed milk (Marvel) for 45mins and washed 3x5mins in PBST prior to incubation with antibody in PBST supplemented with 2.5% (w/v) milk or PBS at 4˚C. Dilutions and antibodies used are specified in S2 Table. Proteins were visualised by ECL Western Blotting Detection Reagents (GE Healthcare), as per the manufacturer's instructions. Molecular size was estimated from the mobility of Novex Sharp Pre-stained Protein Standards.
For protein identification 35 μl of per sample were separated, gels were fixed and stained using the SilverQuest Staining Kit (Life Technologies), or SYPRO Ruby Protein Gel Stain (Life Technologies), according to the manufacturer's instructions. Bands of interest were excised and identified by tandem LC-MS/MS mass spectrometry, performed by the mass spectrometry facility at the MRC Laboratory of Molecular Biology, UK.

Immunofluorescence
Transfected HEK293T cells were grown on poly-l-lysine (Sigma) treated coverslips (Thermo Scientific) for 36 hours and fixed in 1 ml of 4% paraformaldehyde for 20mins at room temperature, washed 3x in 1ml of PBS, permeabilized in 1ml of 0.4% Triton X-100 for 5mins, then washed again in 1ml of PBS 3x 5mins. After blocking in 250 μl of PBS supplemented with 10% normal goat serum (Life Technologies), 0.1% Triton X-100 and 2.5% FBS, cells were stained in 250 μl of blocking buffer for 1hr at room temperature, then washed with 1ml PBS 3x5mins, before being stained with the appropriate dilution (see S2 Table) of secondary antibody in 250 μl blocking buffer for 1hr at room temperature. Coverslips were washed again in 1ml PBS 3x5mins, mounted in 10 μl of Vectashield with DAPI (Vector Laboratories). Confocal microscopy was performed with Zeiss LSM 710 microscope.
To induce in vitro differentiation, cells were treated according to the 4-/4+ method [23]. Briefly, ES cells were grown on feeders in a 24 well plate, as described until they reached 80% confluence. Cells were washed with 1ml PBS, trypsinized and resuspended in 3 ml of media in the absence of LIF. Cells were grown in No LIF media for four days ('4-' phase) at 37˚C and 10% CO2, with the media changed daily and for a further 4 days in No LIF media supplemented with 0.5 μM retinoic acid (Sigma) and for a further 8 days in No LIF media without supplementation (day16). The cells were then transferred to serum free N2 media (Life technologies) and cultured for a further 14 days, with the media changed every other day.
Supporting information S1  Fig. Replicate of co-immunoprecipitation screen for differential interactions of wildtype and AGS-mutated RNase H2B. Protein extracts from control and HEK293T cells transfected with Flag-tagged RNase H2B were incubated with anti-FLAG beads, and the associated proteins were separated by SDS-PAGE and silver stained to reveal any differences in binding between the wild type protein and mutants corresponding to the single amino acid substitutions indicated. An empty vector (control) transfection was also performed to distinguish specific from non-specific binding. The region indicated by the open arrow indicates the region corresponding to the RNase H2B, specific binding identified by mass spectrometry as ZMYM3.  [43]. Predicted MYM zinc fingers domains (cyan), nuclear localisation signals NLS (black), and the domain of unknown function DUF3504 (blue) are shown. The region responsible for the interaction with RNase H2 is enlarged above, and the degree of conservation between ZMYM2, ZMYM3 and ZMYM4 is indicated by the coloured residues (identical residues in dark red, least conserved in blue). Conservation between ZMYMs 2, 3 and 4 identify a repetitive motif not previously described in ZMYM proteins, the PXP motif (black), comprised of repeats of two prolines interrupted by either an isoleucine or valine residue (i.e. X = I or V). Alignments were performed using online protein sequence aligner PRALINE [44].