Neuropathic mutations in MORC2 perturb GHKL ATPase dimerization dynamics and epigenetic silencing by multiple structural mechanisms

Missense mutations in MORC2 cause neuropathies including spinal muscular atrophy and Charcot-Marie-Tooth disease. We recently identified MORC2 as an effector of epigenetic silencing by the HUSH complex. Here we report the biochemical and cellular activities of MORC2 variants, alongside crystal structures of wild-type and neuropathic forms of a human MORC2 fragment comprising the GHKL-type ATPase module and CW-type zinc finger. This fragment dimerizes upon binding ATP and contains a hinged, functionally critical coiled coil insertion absent in other GHKL ATPases. We find that dimerization and DNA binding of the MORC2 ATPase module transduce HUSH-dependent silencing. Disease mutations change the dynamics of dimerization by distinct structural mechanisms: destabilizing the ATPase-CW module, trapping the ATP lid or perturbing the dimer interface. These defects lead to modulation of HUSH function, thus providing a molecular basis for understanding MORC2-associated neuropathies.


Introduction
Microrchidia CW-type zinc finger proteins (MORCs) are a family of transcriptional regulators conserved in eukaryotes. More specifically, MORCs regulate the epigenetic control of transposons and newly integrated transgenes at different developmental stages in plants 1,2 , nematodes 1,3 and mammals 4,5 . Four mammalian genes (MORC1-4) have been annotated.
MORC1 is required for spermatogenesis in mice 6 , as an effector of transposon silencing 5 .
We recently showed that human MORC2 is necessary, in conjunction with the human silencing hub (HUSH), for heterochromatin maintenance and the heterochromatinization of transgenes integrated at chromatin loci with histone H3 trimethylated at lysine 9 (H3K9me3), an epigenetic hallmark of heterochromatin 4,7 . We showed that MORC2 is additionally recruited to transcription start sites bearing H3K4me3, an epigenetic mark enriched at active promoters 4 . MORC2 has also been reported to have ATP-dependent chromatin remodelling activity, which contributes to the DNA damage response 8 and to down-regulation of oncogenic carbonic anhydrase IX in a mechanism dependent on histone deacetylation by HDAC4 9 . MORC3 localizes to H3K4me3-marked chromatin, but the biological function of MORC3 remains unknown 10 .
Despite growing evidence of their importance as chromatin regulators, MORCs have been sparsely characterized at the molecular level. Mammalian MORCs are large, multidomain proteins with an N-terminal gyrase, heat shock protein 90, histidine kinase and MutL (GHKL)-type ATPase module, a central CW-type zinc finger domain and a divergent Cterminal region with one or more coiled coils thought to enable constitutive dimerization 11 . SMCHD1 (Structural maintenance of chromosomes flexible hinge domain containing protein 1) shares some of these key features and could therefore be considered as a fifth mammalian MORC, but lacks a CW domain and has a long central linker connecting to an SMC-like hinge domain 12 . As with several other members of the GHKL superfamily, the ATPase module of MORC3 dimerizes in an ATP-dependent manner 10 . The recently reported crystal structure of the ATPase-CW cassette from mouse MORC3 consists of a homodimer, with the non-hydrolyzable ATP analogue AMPPNP and an H3K4me3 peptide fragment bound to each protomer 10 . The trimethyl-lysine of the H3K4me3 peptide binds to an aromatic cage in the CW domains of MORC3 and MORC4 10,13,14 . The MORC3 ATPase domain was also shown to bind DNA, and the CW domain of MORC3 was proposed to autoinhibit DNA binding and ATP hydrolysis by the ATPase module 14 . Based on observed biochemical activities, MORCs have been proposed to function as ATP-dependent molecular clamps around DNA 10 . However, the CW domains of MORC1 and MORC2 lack the aromatic cage and do not bind H3K4me3, suggesting that different MORCs engage with chromatin via different mechanisms 4,13 . Moreover, MORC1 and MORC2 contain additional domains, including a predicted coiled coil insertion within the ATPase module that has not been found in any other GHKL ATPases.
Exome sequencing data from patients with genetically unsolved neuropathies have recently reported missense mutations in the ATPase module of the MORC2 gene [15][16][17][18][19][20][21][22] . A range of symptoms have been detailed, all subject to autosomal dominant inheritance, with a complex genotype-phenotype correlation. Several reports described Charcot-Marie-Tooth (CMT) disease in families carrying MORC2 mutations including R252W (most commonly) 15,16,19,20 : patients presented in the first or second decade with distal weakness that spread proximally, usually accompanied by signs of CNS involvement. Two other mutations, S87L and T424R, have been reported to cause congenital or infantile onset of neuropathies 15,18,20,21 . Severe spinal muscular atrophy (SMA) with primary involvement of proximal muscles and progressive cerebellar atrophy was detailed in patients with the T424R mutation 18,21 , while diagnosis of patients with the S87L mutation was CMT with SMA-like features 15,20 . We recently showed that the CMT-associated MORC2 mutation R252W hyperactivates HUSHmediated epigenetic silencing in neuronal cells 4 . Disease mutations in MORC2 map to the ATPase module, as in the related SMCHD1 protein, where mutations have recently been associated with Bosma arhinia microphthalmia syndrome (BAMS) 23,24 . However, the lack of biochemical or structural data on MORC2 has precluded efforts to understand its molecular function and rationalise the phenotypes of this rapidly growing list of neuropathic MORC2 variants.
Here we report the biochemical and cellular activities of MORC2 variants, alongside structures of wild-type and neuropathic variant forms of a MORC2 fragment comprising the GHKL-type ATPase module and CW-type zinc finger. Our data reveal several functionally critical features of MORC2. We show how the ATPase activity of MORC2 is tuned, and how ATP-dependent dimerization and DNA binding transduce HUSH-dependent silencing. We propose distinct structural mechanisms that explain how these functions of MORC2 are misregulated in neuropathy-associated variants: destabilization of the ATPase-CW module, trapping the ATP lid or perturbing the dimer interface. Together, our data provide a molecular understanding of the multiple structural mechanisms underlying the neuropathic effects of MORC2 mutations.

Isolation of a stable, functional ATPase fragment of human MORC2
To explore its molecular functions and establish a template to study the effect of neuropathic disease mutations, we set out to purify human MORC2 constructs suitable for biochemical and structural studies. MORC2 is a 1032-amino acid protein predicted to contain several functional domains including an N-terminal GHKL-type ATPase module with a coiled-coil insertion (CC1) (Fig. 1A). MORC3 lacks the CC1 insertion and its ATPase module can be produced in E. coli 10,14 , but we were unable to purify soluble human MORC2 with an intact CC1 from bacteria. We purified MORC2(1-282) (i.e. the GHKL domain, but not including CC1 or the remainder of the ATPase module) and used native differential scanning fluorimetry (DSF) to monitor its interaction with non-hydrolysable ATP analogue AMPPNP.
Native DSF exploits the intrinsic fluorescence properties of proteins to monitor thermal unfolding processes, which are usually accompanied by a redshift in the fluorescence maximum wavelength from 330 nm to 350 nm. By monitoring the ratio of fluorescence at these wavelengths as a function of temperature, a melting temperature (Tm) may be extracted in the absence and presence of stabilizing ligands. We found that although we could detect a small thermal stabilization of MORC2(1-282) by Mg 2+ /AMPPNP, indicating an interaction, no ATPase activity was detected using this construct in an endpoint assay based on the detection of inorganic phosphate ( Fig. 1B and Supplementary Fig. 1a). We could express and purify full-length MORC2 bearing a cleavable tandem StrepII tag in insect cells using a baculovirus-based expression system, but obtained low yields and the protein was highly sensitive to proteolytic cleavage. We therefore used limited proteolysis to identify a protease-resistant N-terminal fragment that ran at approximately 75 kDa by SDS-PAGE and showed evidence of ATP hydrolysis activity (Supplementary Fig. 1a,b). Based on sequencing of tryptic peptides by tandem mass spectrometry, we designed a construct spanning residues 1-603, encompassing the N-terminal ATPase module, the CW-type zinc finger (CW) domain and the second predicted coiled coil (CC2). MORC2(1-603) from insect cells was monomeric in isolation according to size exclusion chromatography ( Supplementary Fig. 1c), and was stabilized by Mg 2+ /AMPPNP, leading to a large increase in the protein Tm from 51.5 °C to 67.3 °C as measured by DSF (Fig. 1B). Notably, the unfolding transition became multiphasic in the presence of the nucleotide (Fig. 1C). This is consistent with a mixture of unbound monomers, singly AMPPNP-bound dimers and doubly bound dimers, as observed in the gas phase by native mass spectrometry of MORC3 10 .
Concentrations up to 2 mM of Mg 2+ /ADP and inorganic phosphate (the products of ATP hydrolysis) did not stabilize the protein (Supplementary Fig. 1d).
MORC2 was previously reported to have ATPase activity in an assay using cellular extracts and in which the D68A point mutant was used as a negative control 8 . We were unable to purify MORC2 constructs bearing the D68A mutation from either bacterial or eukaryotic cells, suggesting that it may cause misfolding of the ATPase module. Since GHKL-type ATPases are usually inefficient enzymes, a robust negative control is essential to rule out background activity from more efficient contaminating ATPases. Hence, we performed an ATPase assay with purified components, using the classical NADH-coupled system that has been used for DNA gyrase and Hsp90 in order to measure enzyme kinetics in continuous mode 25,26 . For the negative control, we mutated the highly conserved active site residue Asn39; the N39A mutation did not compromise the folding of MORC2 but abrogated binding of Mg 2+ /AMPPNP according to DSF data (Fig. 1B). Purified wild-type MORC2(1-603) was found to have low ATPase activity, while the N39A mutant was inactive (Fig. 1D). The kinetics we measured were typical of GHKL ATPases, with a fitted kcat of 0.10 min -1 and Km (ATP) of 378 ± 53 µM ( Fig. 1E). Together these data indicate that the wild-type MORC2 N-terminal ATPase module dimerizes upon ATP binding and that dimers dissociate upon ATP hydrolysis, like other GHKL-type ATPases.

Structure of a homodimeric N-terminal fragment of human MORC2 bound to AMPPNP
Having isolated a MORC2 construct competent for nucleotide binding and hydrolysis, we sought to generate mechanistic insights into the biochemical properties of MORC2 and the molecular basis of MORC2-associated neuropathies via structural analysis. We obtained crystals of human MORC2  in the presence of a molar excess of AMPPNP. The structure was determined by molecular replacement, using the murine MORC3 ATPase module structure 10 as a search model. The asymmetric unit contained two MORC2 molecules and the structure was refined to 1.8 Å resolution (Supplementary Table S1).
The overall architecture of the crystallized MORC2 fragment bound to AMPPNP is an almost symmetric, parallel homodimer resembling the letter M ( Fig. 2A). A 2,778 Å 2 surface from each monomer is buried at the dimer interface. Structural alignment of the ATPase modules of MORC2 and MORC3 showed an rmsd of 1.29 Å for 2,200 backbone atoms, with 36% sequence identity. The MORC2 ATPase module consists of two α-β-α sandwich domains which we have distinguished as the GHKL domain (residues 1-265) and the transducer-like domain (residues 266-494, previously annotated as the S5 domain) due to its resemblance to the transducer domain of gyrase 27,28 . Notably, the β-sheet in the transducer-like domain contains an 80-amino acid antiparallel coiled coil insertion, CC1 (residues 282-361), which forms a 6-nm projection emerging from the ATPase module. A similar insertion is predicted in MORC1 but is absent in other GHKL superfamily members. The transducer-like domain is capped by a helix-loop-helix motif that links to the CW domain (residues 495-545). This motif is disordered in the MORC3 structure and moreover, the CW domain of MORC2 is in a completely different position and orientation relative to the ATPase module. Our MORC2 structures span residues 1-551, including all reported sites of neuropathy-causing mutations ( Supplementary Fig. 2a,b). We did not observe electron density for the second predicted coiled coil, CC2 (residues 551-603). A tetrahedrally coordinated zinc atom carried over from the purification was observed bound to the CW domain. The presence of zinc in the MORC2 crystals was confirmed by X-ray fluorescence spectroscopy (Supplementary Fig. 2c).
MORC2 has a prototypical GHKL ATPase active site. One AMPPNP molecule, stabilized by an octahedrally coordinated Mg 2+ ion, is bound in the active site of both protomers. All critical residues involved in ATP binding and hydrolysis from the four signature motifs in the Nterminal GHKL ATP binding domain 29

Nucleotide binding of MORC2 ATPase stabilizes the dimer interface
GHKL ATPases usually dimerize on binding ATP but the composition and dynamics of the ATP lid that can close over the active site vary across the GHKL superfamily 29 . In the wildtype MORC2 structure, the ATP lid (residues 82-103) is in the closed conformation in both protomers, leaving only a narrow channel between the bound AMPPNP and the solvent.
Aside from residues in the four motifs detailed above, protein-nucleotide interactions made by the sidechains of Ser87 (a neuropathy mutation site) and Lys89 with the β-phosphate, and by the backbone atoms of Gln99 and Tyr100 with the γ-phosphate, stabilize the lid conformation (Fig. 2B). Residues in the lid form a significant part of the dimer interface, with Ile82, Phe84, Arg90, Tyr100 and Asn102 forming hydrogen bonds and hydrophobic contacts with residues 12-24 of the other protomer. A loop in the transducer-like domain (residues 422-437), also contributes to the dimer interface. This loop coordinates the γ-phosphate of AMPPNP through Lys427 (and includes another neuropathy mutation site, Thr424) (Fig. 2C).

ATP-dependent dimerization is required for MORC2 effector function in HUSH silencing
The MORC2(1-603) N39A mutant was a monomer in solution and did not bind or hydrolyse ATP (Figs. 1B, 1D and Supplementary Fig. 1c). Since ATP binding by the MORC2 ATPase module is coupled to dimerization, we conclude that the catalytically inactive N39A mutant does not form dimers via the ATPase module. We previously established a genetic complementation assay to assess the capacity of different disease-associated variants of MORC2 to rescue HUSH-dependent transgene silencing in MORC2 knockout (KO) cells.
Briefly, we isolated clonal HeLa reporter cell lines bearing a HUSH-repressed GFP reporter.
CRISPR-mediated KO of MORC2 in these clones led to cells becoming GFP bright, allowing complementation with exogenous MORC2 variants, which can be monitored as GFP rerepression using FACS 4 . The lentiviral vector used expresses mCherry from an internal ribosome entry site (IRES), enabling us to control for multiplicity of infection (MOI) by monitoring mCherry. Using this assay, we previously found that the N39A mutant failed to rescue HUSH-dependent silencing 4 . Together with our biochemical data ( Fig. 1), this shows that ATP binding or dimerization of MORC2 (or both) is required for HUSH function.
To decouple the functional roles of ATP binding and dimerization, we used our MORC2 structure to design a mutation aimed at weakening the dimer interface without interfering with the ATP binding site. The sidechain of Tyr18 makes extensive dimer contacts at the twofold symmetry axis, but is not located in the ATP binding pocket (Fig. 2C). Using the genetic complementation assay described above, we found that whereas addition of exogenous V5-tagged wild-type MORC2 rescued HUSH silencing in MORC2-KO cells, the Y18A MORC2 mutant failed to do so (Fig. 2D). The inactive MORC2 Y18A variant was expressed at a higher level than wild-type despite the same MOI being used (Fig. 2E).
We then purified MORC2(1-603) Y18A and analysed its stability and biochemical activities using DSF and the ATPase assay. The mutant was properly folded based on DSF data on the unliganded protein, and was stabilized by AMPPNP suggesting that it was capable of nucleotide binding (Fig. 2F). However, consistent with our design, the high-Tm species corresponding to MORC2(1-603) dimers were absent. Despite its inability to form dimers, MORC2(1-603) Y18A was an active ATPase with slightly increased activity over the wildtype construct (Fig. 2G). This suggests that dimerization of the MORC2 N-terminus is not required for ATP hydrolysis. Taken together, we conclude that ATP-dependent dimerization of the MORC2 ATPase module transduces HUSH silencing, and that ATP binding and hydrolysis are not sufficient.

CC1 has rotational flexibility that may be coupled to nucleotide binding and dimerization
The most striking feature of the MORC2 structure is the projection made by CC1 (residues 282-361) that emerges from the core ATPase module. The only other GHKL ATPase with a similar coiled coil insertion predicted from its amino acid sequence is MORC1, for which no structure is available. Elevated B-factors in CC1 suggest local flexibility and the projections emerge at different angles in each protomer in the crystal structure. The orientation of CC1 relative to the ATPase module also varies from crystal to crystal, leading to a variation of up to 19 Å in the position of the distal end of CC1 (Fig. 3A). Although the orientation of CC1 may be influenced by crystal contacts, a detailed examination of the structural variation reveals a cluster of hydrophobic residues (Phe284, Leu366, Phe368, Val416, Pro417, Leu419, Val420, Leu421, Leu439) that may function as a 'greasy hinge' to enable rotational motion of CC1. Notably, this cluster is proximal to the dimer interface. Furthermore, Arg283 and Arg287, which flank the hydrophobic cluster at the base of CC1, form salt bridges across the dimer interface with Asp208 from the other protomer, and further along CC1, Lys356 interacts with Glu93 in the ATP lid (Fig. 3B). Based on these observations we hypothesize that dimerization, and therefore ATP binding, may be coupled to rotation of CC1, with the hydrophobic cluster at its base serving as a hinge.
Positively charged residues on the distal end of CC1 contribute to DNA binding and are required for HUSH-dependent silencing CC1 has a predominantly basic electrostatic surface, with 24 positively charged residues distributed across the surface of the coiled coil (Fig. 3C). MORC3 was shown to bind double-stranded DNA (dsDNA) through its ATPase module, but this interaction was autoinhibited by the CW domain 14 . We therefore sought to determine whether the MORC2 ATPase-CW cassette binds DNA, and whether the charged surface of CC1 contributes to DNA binding. We first performed electrophoretic mobility shift assays (EMSAs) with nucleosome core particles (NCPs) and observed that wild-type MORC2(1-603) bound to both free DNA and nucleosomal DNA present in the NCP sample, with an apparent preference for free DNA (Fig. 3D).
Next, to assess the importance of CC1 in HUSH-dependent silencing, we examined the effect of a panel of charge reversal mutations in CC1 in the cell-based HUSH complementation assay. The charge reversal point mutations R319E, R344E, R351E and R358E all rescued HUSH function in MORC2-KO cells, but R326E, R329E and R333E (or combinations thereof) failed to do so ( Fig. 3E and Supplementary Fig. 3a). Again, inactive variants were expressed at higher levels than active ones (Supplementary Fig. 3b).
Residues 326, 329 and 333 form a positively charged patch near the distal end of the second α-helix of CC1. We therefore made a MORC2(1-603) triple mutant, R326E/R329E/R333E, and compared its dsDNA binding to that of the WT construct. We confirmed that WT MORC2(1-603) bound to the canonical Widom 601 nucleosome positioning sequence with high apparent affinity, and observed a 'laddering' effect on the DNA at the lowest (250-750 nM) protein concentrations. This is consistent with multiple DNA binding surfaces on the protein and/or multiple proteins binding to a single piece of DNA. The triple CC1 charge reversal mutant still bound dsDNA, but with weaker apparent affinity, and no laddering of the DNA bands was observed (Fig. 3F). The WT MORC2 GHKL domain alone (residues 1-282) also bound dsDNA, albeit with a much lower affinity and with no laddering, while the CW domain in isolation did not bind DNA in the EMSA (Supplementary Figs. 3b,c). Together, these data suggest that MORC2 binds dsDNA through multiple sites including a positively charged surface near the distal end of the CC1 arm, and that the latter is required for transduction of HUSH-dependent silencing.

The CW domain has a regulatory role in the HUSH effector activity of MORC2
Several recent studies have shown that the CW domain of MORC3 binds H3K4me3 peptides selectively over histone 3 peptides with other epigenetic marks 10,13,14 . By contrast the MORC2 CW domain does not bind to the H3K4me3 mark due to a missing tryptophan at the 'floor' of the CW aromatic cage (Thr496 in MORC2, Fig. 4A) 4,13 . Indeed, the MORC2 CW domain was found not to interact with any of a wide variety of histone H3 and histone H4 peptides 13 . We confirmed that the lack of interaction with DNA and/or histones is not due to a folding defect or a reliance on the ATPase module for folding, since isolated 15 N-labelled MORC2 CW domain showed well-dispersed peaks in a 1 H, 15 N-heteronuclear single quantum coherence (HSQC) experiment (Supplementary Fig. 4a).
The orientation of the CW domain relative to the ATPase module differs by approximately 180˚ in the MORC2 and MORC3 structures, with the degenerate histone binding site of the MORC2 CW domain facing towards the ATPase module rather than towards solvent ( Supplementary Fig. 4b). The CW domain binds an array of arginine residues in the transducer-like domain: conserved residue Trp505, providing the 'right wall' of the methyllysine-coordinating aromatic cage, forms a cation-π interaction with the sidechain of Arg266.
Thr496 (the degenerated 'floor' residue) makes a water-mediated hydrogen bond with the backbone amide of Arg266. Asp500 forms a salt bridge with Arg254. Gln498 forms a hydrogen bond with the backbone carbonyl oxygen of Arg252. Glu540 forms a salt bridge with the Arg252 sidechain, which also forms a hydrogen bond with the backbone oxygen atom of Leu503 (Fig. 4B). The latter interactions are especially notable since a number of recent studies have shown that the R252W mutation causes Charcot-Marie-Tooth (CMT) disease 15,16,19,20 . We recently demonstrated that this mutation causes hyperactivation of HUSH-dependent epigenetic silencing 4 , leading to enhanced and accelerated re-repression of the GFP reporter in our functional assay. The R252W mutation, by removing the salt bridge to Glu540, may destabilize the ATPase-CW interface, which could account for the misregulation of MORC2 function in HUSH-dependent silencing. To test this hypothesis, we designed a mutation aimed at causing a similar structural defect, R266A, which disrupts the cation-π interaction with Trp505 described above. We performed a timecourse experiment, monitoring GFP reporter fluorescence in MORC2-KO cells after addition of the exogenous MORC2 variant. The R266A mutation recapitulated the hyper-repressive phenotype of R252W in the reporter clone tested (Fig. 4C-E), supporting the notion that the ATPase-CW interaction in MORC2 has a regulatory function in HUSH transgene silencing. Since the CW domain is directly linked to the second coiled coil (CC2), which is the putative HUSH binding module 4 , destabilization of this intramolecular association could have direct consequences on HUSH assembly or activity.
In MORC3, the CW domain prevents binding of the ATPase module to DNA in the absence of H3K4me3 14 . In MORC2, however, the CW domain does not inhibit DNA binding since MORC2(1-603) bound tightly to DNA despite the presence of an unliganded CW domain Figs. 3D,F). We note that many of the sidechains forming key contacts in the ATPase-CW domain interfaces of MORC2 and MORC3 are not conserved in the two proteins. These nonconserved residues are Arg254, Arg266 and Thr496 in MORC2; and Glu184, Arg195, Lys216, Tyr217, Arg405, Arg444, and Asp454 in MORC3. Hence, it appears unlikely that the CW domain can bind to the MORC2 ATPase module in the same configuration as in MORC3, and vice versa. Together, our data show that the CW domain of MORC2 has a degenerate aromatic cage that explains its lack of binding to epigenetic marks on histone tails, and suggest that the association of the CW domain to the ATPase module antagonizes HUSH-dependent epigenetic silencing. Moreover, we conclude that MORC2 and MORC3 have evolved CW domains with distinct regulatory mechanisms.

Disease mutations modulate the in vitro and in vivo activities of MORC2
We next tested whether MORC2 mutations reported to cause neuropathies affected the ATPase activity of MORC2. We purified MORC2(1-603) variants containing the R252W, T424R and S87L point mutations. All of the variants were well folded and were stabilized by addition of 2 mM Mg 2+ /AMPPNP (Supplementary Fig. 5a,b). We found a wide range of effects on ATPase activity (Fig. 5A). MORC2(1-603) bearing CMT mutation R252W 15,16,19,20 showed a small decrease in the rate of ATP hydrolysis. In contrast, SMA mutation T424R 18,21 increased ATPase activity approximately threefold. The S87L variant (for which the clinical diagnosis was CMT with SMA-like features 15,20 ) eluted from a size-exclusion column as two species: a major dimeric species with elevated 260 nm absorbance, indicating the presence of bound nucleotide, and a minor monomeric species (Supplementary Fig. 5a); this variant displayed very low ATPase activity, near the detection threshold.
The R252W MORC2 variant hyperactivates HUSH-mediated transgene silencing 4 but has reduced ATPase activity in vitro. We used the timecourse HUSH functional assay in two distinct MORC2-KO GFP reporter clones (i.e. two different HUSH-repressed loci) to investigate further the correlation of these activities. S87L (which forms stabilized dimers and also has reduced ATPase activity in vitro) also matched or outperformed wild-type MORC2 at each time point measured. Conversely, T424R (which has increased ATPase activity in vitro) was significantly less efficient at GFP reporter repression than wild-type at both loci ( Fig. 5B and Supplementary Figs. 5c,d). Together, these data indicate that unlike the point mutants incompetent for ATP binding (N39A) or dimerization (Y18A), which altogether fail to transduce HUSH silencing, the disease-associated variants are all capable of ATP binding, dimerization and hydrolysis. Further, we find that the efficiency of HUSH-dependent epigenetic silencing decreases as the rate of ATP hydrolysis increases. A summary of the properties of neuropathic and engineered MORC2 variants is shown in Table 1.

Neuropathic mutations S87L and T424R perturb MORC2 dimer interface
Two MORC2 mutations, S87L and T424R, have been reported to cause congenital or infantile onset of neuropathies, distinct from the later onset that was reported for patients bearing the R252W (or other) mutations. The consequences of S87L and T424R mutations on the biochemical activities of MORC2 are drastic. The locations of these mutation sites -Ser87 in the ATP lid and Thr424 at the dimer interfaceare also at functionally important regions in the structure and we therefore determined the crystal structures of these variants to understand better the observed activities (Supplementary Table S1). T424R MORC2 was co-crystallized with AMPPNP using the same protocol as for wild-type MORC2, but since S87L was dimeric and bound to ATP upon purification from insect cells, we determined its structure bound to ATP. The overall homodimeric structure of the two MORC2 disease variants was very similar to that of wild-type (Supplementary Fig. 6). The orientation of CC1 relative to the ATPase module varied in each protomer within the same range as in wild-type.
The ATP molecules bound to S87L MORC2 are in a nearly identical conformation to AMPPNP in the wild-type and T424R structures, confirming that AMPPNP is a reasonable mimic of the natural nucleotide substrate in this case.
Ser87 is in the lid that covers bound ATP. Its sidechain hydroxyl forms a hydrogen bond with the β-phosphate of AMPPNP in the wild-type structure. In the S87L mutant, we found that the lid is partially missing in one protomer and has a completely different conformation in the other. In the latter protomer, the lid forms additional contacts across the dimer interface in the S87L mutant (Fig. 5C). Leu87 itself forms apolar contacts with Asp141 from the other protomer, but more importantly, Arg90 forms a tight salt bridge with Glu17 across protomers.
In the wild-type structure the Arg90 and Glu17 sidechains are 4-5 Å apart but do not form a salt bridge. Instead, Lys86 can form a salt bridge with Asp141 from the other protomer in wild-type. The increased number of dimer contacts in the S87L mutant is reflected in an increased buried surface area at the dimer interface (3,016 Å 2 buried per protomer versus 2,778 Å 2 in wild-type). These observations provide a plausible structural basis for the observation that S87L forms more stable ATP-bound dimers than wild-type, which in turn affects its cellular function.
The effect of T424R on the crystal structure of MORC2 is more subtle. The backbone structures of wild-type and T424R are essentially identical, including in the loop that contains the mutation (Fig. 5D). The arginine sidechain in the mutant does make an additional salt bridge across the dimer interface, with Glu27 from the other protomer. This additional contact may contribute to the dimer interface, but we did not observe any dimerization of T424R MORC2 during purification, suggesting that the mechanism of misregulating MORC2 is distinct from S87L. Moreover, the buried surface area at the dimer interface is actually decreased upon the T424R mutation (2,527 Å 2 buried per protomer versus 2,778 Å 2 in wildtype). We have described how ATP binding/hydrolysis is structurally coupled to dimerization/dissociation. The contribution of the mutant Arg424 sidechain to the dimer interface, and its position just three residues away from a key active site residue Lys427, can be expected to alter the dimerization dynamics and thus the ATPase activity of MORC2.
Indeed, we found that the T424R mutation increases the rate of ATP hydrolysis, indicating that T424R dimers form and dissociate more rapidly than in the wild-type. It should be noted, however, that MORC2-associated neuropathies are subject to autosomal dominant inheritance. Our structures therefore represent the physiologically less common species in which not one but both protomers bear the mutation. It may be that the effect on molecular function is subtly different in heterozygous MORC2 dimers. Together, these data show that S87L causes kinetic stabilization of MORC2 dimers, whereas T424R increases the rate of dimer assembly and disassembly. These two disease mechanisms are distinct from that of R252W, which we propose above to weaken the regulatory ATPase-CW interaction.

Discussion
Genetic studies have established that MORC family proteins have fundamentally important functions in epigenetic silencing across eukaryotic species 1,4,5 . We recently identified MORC2 as an indispensable effector of the HUSH complex and showed that MORC2 contributes to chromatin compaction across HUSH target loci. The activity of MORC2 was dependent on ATP binding by its GHKL-type ATPase module 4 32 . The Km of MORC3 has not been reported but its activity at 3 mM ATP was 0.4-0.5 min -1 . 14 Hence, MORC2 and MORC3 resemble prototypical GHKL ATPases in that they bind ATP with relatively low affinity and hydrolyse ATP relatively slowly. Due to their low enzymatic turnover, GHKL ATPases are not known to function as motors or deliver a power stroke. Instead, ATP binding and hydrolysis function as conformational switches triggering dimer formation and dissociation, respectively 33 . Since MORC2 has similar enzymatic properties to other GHKL ATPases, we propose that its HUSH effector function arises from dimerization of its ATPase module triggered by ATP binding. Consistent with this model, a mutation outside the active site designed to inhibit dimerization of the ATPase module without affecting ATP binding/hydrolysis (Y18A) causes a loss of HUSH silencing (Fig. 2D-G). Since ATP hydrolysis leads to dissociation of the dimer, one corollary is that increasing the rate of hydrolysis should cause less efficient HUSH function, and decreasing the rate of hydrolysis should hyperactivate HUSH function (so long as ATP can still bind and the protein can dimerize). Indeed this is what we found through analysis of three neuropathy-associated MORC2 mutants: T424R, which hydrolyses ATP more rapidly than wild-type, leads to less efficient HUSH silencing, whereas S87L and R252W, which hydrolyse ATP more slowly than wild-type, match or hyperactivate wild-type HUSH silencing (Fig. 5). We conclude that the lifetime of wild-type MORC2 ATPase dimers is tuned to enable cellular function. Indeed, the properties of engineered and naturally occurring MORC2 variants studied so far (summarized in Table 1) suggest that slight changes in the rate of ATP hydrolysis or dimer stability misregulate MORC2 cellular function by perturbing its N-terminal dimerization dynamics.
Our data suggest that the biological function of MORC2 arises from cycling between monomeric and dimeric conformational states at a finely tuned rate defined by ATP binding MORC2 is the only gene in the MORC family in which mutations have been reported to cause neuropathies in humans. Here, we have examined three disease mutations (S87L, R252W and T424R) that cause neuropathies diagnosed as CMT and/or SMA. Our work provides insights on the basis of MORC2 misregulation in affected patients and identifies the link between the biochemical activities of MORC2 and its cellular function as an effector of the HUSH complex. Modulation of HUSH-independent activities of MORC2 may also be important in determining disease outcome. For example, MORC2 has been associated with the activity of HDAC4 9 , a protein known to be important in synaptic plasticity and as a transcriptional regulator in the central nervous system 37 , and which was overexpressed in SMA model mice and muscles of SMA patients 38 . More work will be needed understand how MORC2 mutations cause the range of clinical symptoms described, but it is interesting to note that the mutations causing pronounced changes in biochemical properties (S87L, which forms constitutive dimers, and T424R, which increases ATPase activity three-fold) are associated with congenital or infantile onset, unlike the R252W mutation where the biochemical effect is more subtle and affected patients presented later.
Based on our observed relationships between its in vitro and in vivo activities, we conclude that MORC2 is part of a homeostatic system tuned such that a reduction in biochemical activity can cause a gain-of-function cellular phenotype, and vice versa. Similarly, mutations in SMCHD1 cause Bosma arhinia microphthalmia syndrome (BAMS), while others cause facioscapulohumeral muscular dystrophy type 2 (FSHD2) 23,24 , with varied effects on in vitro ATPase activity 23 . These studies provide an interesting example in another GHKL ATPase of (i) pleiotropic disease outcomes resulting from mutations in the same gene, and (ii) gain of function cellular phenotypes resulting from decreased biochemical activity and vice versa.
Since both gain and loss of MORC2 molecular function are linked to neuropathies, there are in theory a large number of MORC2 mutations that could cause disease including those that compromise the stability of the ATPase module. We would therefore predict that there are other MORC2 mutations associated with undiagnosed neuropathies. Here we have       Error bars represent standard deviation between measurements. The WT data are shown for reference and are the same as in Fig. 1. R252W and S87L variants have reduced rates of ATP hydrolysis, whereas T424R has threefold higher activity than WT.  propanediol, 2-propanol, 1,4-butanediol and 1,3-propanediol. Pyramidal crystals appeared in 1-2 days and were frozen in liquid N2 using 35% glycerol as a cryoprotectant. The procedure was the same for the S87L and T424R variants except that the precipitant concentrations for the best crystals were higher (10-11% PEG4000 and 20-22% glycerol). In the case of the S87L variant, the protein was at 4.5 mg/ml and was concentrated from the earlier gel filtration peak (i.e. the nucleotide-bound dimeric fraction), which was assigned as ATPbound; therefore, ATP was supplemented in place of AMPPNP. X-ray diffraction data were collected at 100 K at the European Synchrotron Radiation Facility (beamlines ID29 and ID30b) and processed using the autoPROC package 40   Antibodies. The following antibodies were used: mouse α-V5 (Abcam, ab27671), mouse αβ-actin (Sigma-Aldrich, A5316), donkey α-mouse HRP-conjugated antibody (Jackson ImmunoResearch, 715-035-150).
Immunoblotting. Cells were lysed in 1% SDS plus 1:100 (v/v) benzonase solution (Sigma-Aldrich) for 15 min at room temperature, and then heated to 65 °C in SDS sample loading buffer for 5 min. Following separation by SDS-PAGE, proteins were transferred to a PVDF membrane (Millipore), which was then blocked in 5% milk in PBS + 0.2% Tween-20.
Membranes were probed overnight with the indicated primary antibodies, washed four times in PBS + 0.2% Tween-20, then incubated with HRP-conjugated secondary antibodies for 1 h at room temperature. Reactive bands were visualized using SuperSignal West Pico (Thermo Fisher Scientific). lines indicate absorbance at 280 nm and 260 nm, respectively. The black arrow indicates a larger molecular weight (earlier elution) peak for S87L with elevated 260 nm absorbance, which we have assigned as a nucleotide-bound dimer. Chromatography was performed with a S200 Increase (10/300) column (GE Healthcare). The greyed out data for WT and N39A are for reference and are also shown in Supp.