Maternal SMCHD1 regulates Hox gene expression and patterning in the mouse embryo

Parents transmit genetic and epigenetic information to their offspring. Maternal effect genes regulate the offspring epigenome to ensure normal development. Here we report that the epigenetic regulator SMCHD1 has a maternal effect on Hox gene expression and skeletal patterning. Maternal SMCHD1, present in the oocyte and preimplantation embryo, prevents precocious activation of Hox genes post-implantation. Without maternal SMCHD1, highly penetrant posterior homeotic transformations occur in the embryo. Hox genes are decorated with Polycomb marks H2AK119ub and H3K27me3 from the oocyte throughout early embryonic development; however, loss of maternal SMCHD1 does not deplete these marks. Therefore, we propose maternal SMCHD1 acts downstream of Polycomb marks to establish a chromatin state necessary for persistent epigenetic silencing and appropriate Hox gene expression later in the developing embryo. This is a striking role for maternal SMCHD1 in long-lived epigenetic effects impacting offspring phenotype.


Introduction
It is now clear that epigenetic information can be passed from generation to generation via the germline, changes in which can have long-lasting effects in the offspring. One of the most notable of these effects is transmission of epigenetic information from the oocyte to the zygote. The oocyte supplies the entire cytoplasm containing all expressed mRNA and proteins to the zygote, sustaining it through its initial cell divisions until its own zygotic genome is transcribed, at embryonic day (E) 2.0 in mice 1 . Genes whose expression is required in the oocyte for normal development of the offspring are known as maternal effect genes.
A classic example of the role of maternal effect genes in passing long-lived epigenetic information from parent to offspring is genomic imprinting, where genes are monoallelically expressed in a parent-oforigin-specific manner. Epigenetic imprints are imparted by germ cell-derived DNA methylation or trimethylation of lysine 7 on histone 3 (H3K27me3) 2, 3, 4 . Maternal effect genes important for imprinting generally have a role in establishing and maintaining these germline marks 5,6 .
Structural maintenance of chromosomes hinge domain containing 1 (Smchd1) is a recently defined maternal effect gene that is expressed in the oocyte and is required for genomic imprinting in the mouse placenta 7,8,9 . In its zygotic form, SMCHD1 plays a key role in epigenetic silencing of imprinted loci, along with other clustered gene families and the inactive X chromosome 10,11,12,13,14,15 . Heterozygous variants in SMCHD1 are also associated with the human diseases Facioscapulohumeral muscular dystrophy (FSHD) and Bosma arhinia microphthalmia (BAMS) 16,17,18,19 , demonstrating the important role SMCHD1 plays in normal development.
SMCHD1 is a member of the SMC family of proteins, large chromosomal ATPases important for chromosome structure 20 . SMCHD1 also plays a role in chromatin architecture, mediating long-range interactions at its targets 12,21,22,23 . Recruitment to at least one of its targets, the inactive X chromosome, is dependent on the polycomb repressive complex 1 (PRC1) mark ubiquitination of lysine 119 of histone H2A (H2AK119ub) 22,24 . For imprinted genes we have proposed that SMCHD1 is recruited downstream of PRC2's mark H3K27me3 7 . Precisely how zygotic or maternal SMCHD1 enables gene silencing is not yet clear.
One of the clustered gene families zygotic SMCHD1 binds and silences is the Hox genes 11, 12, 25 , a highly conserved set of transcription factors that are responsible for correct patterning of body segments along the anterior-posterior (A-P) axis during embryonic development 26,27,28,29 . Hox genes are only expressed at specific times and in specific tissues during post-implantation embryonic development 30,31 . At all other times they are silent and marked by H2AK119ub and H3K27me3 32,33 , including in the oocyte and through pre-implantation development 34,35,36 , opening the exciting possibility of maternal effects on Hox gene expression. Based on these data and Smchd1's role as a maternal effect gene, we investigated whether maternal SMCHD1 has long-lasting effects at its targets in the embryo, specifically on the Hox genes. This was made possible because, unlike many maternal effect genes, deletion of maternal Smchd1 does not result in embryonic lethality 7 .
In this study we show that maternal SMCHD1, found in the preimplantation embryo, is required to prevent premature Hox gene activation in the early post-implantation embryo. Interestingly, these changes occurred without disruption of H2AK119ub or H3K27me3 marks over Hox genes in the pluripotent state, suggesting that maternal SMCHD1 acts downstream of Polycomb to regulate Hox gene expression and normal skeletal patterning post-implantation.

Maternal SMCHD1 is required for normal skeletal patterning
Given that previous work in our lab has shown that Smchd1 mutants exhibit homeotic transformations 12, 25 , we first assessed whether Smchd1 maternal knockout embryos also show abnormal skeletal patterning. We set up F1 crosses between C57BL/6 and Castaneus strain (Cast) parents, using MMTV-Cre or Zp3-Cre to knock out Smchd1 in the oocyte (Fig. 1a-c) as we have previously 7 . We set up three types of F1 crosses. The first was a control cross yielding embryos with wild-type SMCHD1 function (Smchd1 wt ). This established a baseline of skeletal patterning in the F1 embryos (Fig. 1a). Almost all of these embryos had normal skeletal patterning, with 97% and 86% of control mice from the MMTV-Cre and Zp3-Cre colonies respectively having the expected 7 cervical vertebrae, 13 thoracic vertebrae, 6 lumbar vertebrae and 4 sacral vertebrae (Fig. 1d). In the second cross, Smchd1 was deleted in the oocyte with either MMTV-Cre or Zp3-Cre, yielding Smchd1 heterozygous embryos which lacked maternal SMCHD1 (Smchd1 matΔ , Fig. 1b). The third cross, performed with the MMTV-Cre only, was reciprocal to the maternal deletion cross and generated both Smchd1 del/+ (Smchd1 het ) and Smchd1 wt embryos, with the oocytes from which they were generated having wild-type levels of SMCHD1 (Fig. 1c). This latter cross tested whether any phenotype observed in the Smchd1 matΔ embryos was due to haploinsufficiency for SMCHD1 after zygotic genome activation rather than lack of maternal SMCHD1, and controlled for the direction of the interstrain cross.
Smchd1 matΔ embryos exhibited a highly penetrant addition of a rib on the seventh cervical element (C7), suggesting that C7 adopts the identity of T1 (Fig. 1b, e). We observed several morphological variations of this additional rib including a short ectopic rib, a rib which fused with T1 with and without subsequent bifurcation before joining the sternum, and a full rib which joined the sternum independently of T1 (Supplementary Dataset 1). Grouped together, any indication of C7 transformation was observed in the MMTV-Cre and Zp3-Cre models at a penetrance of 97% and 91% respectively (Fig. 1d).
Additional posteriorising transformations were observed in a subset of Smchd1 matΔ embryos when Smchd1 was deleted with MMTV-Cre. These included (i) the loss of ribs on T13 leading to a complete T13-to-L1 transformation or severely hypomorphic rib(s) on T13, and (ii) a L6-to-S1 transformation.
These phenotypes were observed at a lower penetrance than the additional C7 rib; 63% and 52% for the transformation altering T13 and L6 respectively (Fig. 1d). All three of these phenotypes had significantly higher penetrance following maternal deletion of Smchd1 compared to both the control and reciprocal cross (p<0.001, Chi-square test), and there was no sex-specificity in the phenotypes observed (Supplementary Dataset 1). Of note, when a lumbar transformation was observed, it was almost exclusively coincident with transformations at cervico-thoracic and thoraco-lumbar transitions.
This suggests serial homeotic transformation in these embryos, supported further by examples of transformation of vertebra surrounding these transition points where specific morphology can be delineated (e.g. C5→C6 and T1→T2; Supplementary Dataset 1). Taken together, deletion of maternal Smchd1 results in a highly penetrant posterior homeotic transformations that can encompass multiple axial regions, implying a potential global shift in patterning effectors. Given there were no abnormalities observed in the Smchd1 heterozygous skeletons (Fig. 1c, d), these data support the view that maternal SMCHD1 is required for appropriate axial patterning.  We conducted RNA-seq in tailbud tissue from Smchd1 wt and Smchd1 matΔ embryos, with somite-matched replicates (Supplementary Figure 1, Fig. 2a, b). We chose 6-11 somites as we theorised that loss of maternal SMCHD1 may lead to precocious Hox gene activation. There was very limited differential expression genome-wide in these somite-matched samples. Indeed, what limited differential expression was present can be explained by the sex disparity in samples at each somite number (Supplementary Dataset 2). Moreover, there was no difference in somite range between control and Smchd1 maternal null embryos (Supplementary Dataset 2), suggesting there was no striking developmental delay following loss of maternal SMCHD1. Consistent with this, Hox gene expression was approximately normal in control and Smchd1 matΔ tailbud samples (Fig. 2c, d). When we compared these two somite series, we saw a collective, albeit modest, upregulation of anterior Hox genes in somite 8-11 tissue, specifically a trend towards precocious activation of the Hox2 to 7 paralogues in somite 10 and 11 tissue ( Fig. 2e). We also observed a concomitant downregulation of posterior Hox genes particularly at the earlier somite stages (Fig. 2d).
To further explore the precocious anterior Hox activation, we opted for an in vitro approach by differentiating murine embryonic stem cells (mESCs) into NMPs. We derived Smchd1 wt and Smchd1 matΔ mESCs, deleting in the oocyte with Zp3-Cre and performed RNA-seq in the mESCs, where we observed no significantly differentially expressed genes (n=3, Supplementary Dataset 3, Supplementary Figure   2). Next, we differentiated the Smchd1 wt and Smchd1 matΔ mESCs, harvesting RNA from differentiating cells every 12 to 24 hours from their pluripotent to NMP-like state (n=4, Fig. 3a). Just as observed in vivo, there was no differential expression when analysed genome-wide (Supplementary Figure 2). The differentiation progressed as expected with the loss of pluripotency factor expression and increase in differentiation factors (Supplementary Figure 2). 12 hours after Wnt activation (day 2.5), we observed precocious activation of several anterior Hox genes in the Smchd1 matΔ cells (Fig. 3b), which corresponds to approximately E8.5 in vivo 40 . Given the Hox genes are just switching on at this time, this was most noticeable as a larger log fold change between day 2 and day 2.5 of differentiation in the maternal null compared with control cells for anterior Hox genes (Fig. 3c, d, p<0.001). At day 3 (NMPs and MPs) and day 4 (24h after GDF11 addition) corresponding to E9.5 in vivo, we observed a general downregulation of Hox gene expression, consistent with the Hox gene activation we observe at day 2.5 being precocious but not sustained (Fig. 3b).
Taken together, these in vivo and in vitro data suggest that the Smchd1 matΔ skeletal phenotype may in part be explained by premature upregulation of anterior Hox genes. The normal Hox gene silencing in the pluripotent state, and the expected downregulation of Hox genes observed later in differentiation, is consistent with the relatively modest effects on skeletal patterning in the absence of maternal SMCHD1.  n=4 replicates for each genotype of mESC, from two separate mESC lines for each genotype. c.
Heatmap of the average log2 fold change of Hox gene expression between the day 2.5 and day 2 samples, for each genotype. d. The log2 fold change for the Hox1-9 genes between day 2.5 and day 2 of differentiation, for each of the 4 replicates per genotype (Student's t-test, two-tailed, equal variance *** p<0.001).

Maternal SMCHD1 acts downstream of Polycomb-mediated H3K27me3 and H2AK119ub in mESCs
To investigate the mechanism by which loss of maternal SMCHD1 caused upregulation of anterior Hox genes, we assessed whether the histone marks H2AK119ub and H3K27me3 were perturbed in

Discussion
In this study we have shown that maternal SMCHD1 is required for appropriate patterning of the axial skeleton, linked to its role in silencing Hox genes. Deletion of Smchd1 in the oocyte results in the highly penetrant posteriorising homeotic transformations of C7-to-T1, T13-to-L1 and L6-to-S1. This phenotype is similar to what is observed in zygotic PRC1 subunit knockouts and is attributable to Hox gene overexpression 43,44 . We too observed a modest but consistent upregulation of anterior Hox genes both in vivo in the developing tailbud of ~E8.5 embryos and in vitro soon after induction of Wnt signalling in mESC differentiating into neuromesodermal progenitors. Hox gene silencing was restored later in differentiation, consistent with the relatively subtle axial patterning defects observed in the Smchd1 maternal null embryos. Interestingly, maternal SMCHD1 was not required to maintain appropriate Hox gene silencing earlier in development, either in the pluripotent state or in the morula.
These data suggest that maternal SMCHD1, which controls the embryo in the preimplantation period 7 , is required to ensure Hox genes are not prematurely activated in the post-implantation period. This is a long-lived effect of maternal SMCHD1 at the Hox clusters from around E2.75 when zygotic SMCHD1 is activated to around E8.5 when precocious Hox gene activation is observed.
Given the important role of PRC1 and PRC2 in silencing the Hox genes, their role in long-lived mitotic epigenetic memory 45 (and the fact that they mark the Hox genes with H2AK119ub and H3K27me3 from the oocyte stage onwards 4, 34, 35, 36, 41 , we asked whether maternal SMCHD1 may function together with the PRCs to silence Hox genes, using our mESC model. In undifferentiated mESCs we observed no change in H3K27me3 or H2AK119ub marks genome-wide in Smchd1 maternally deleted cells compared to control. These data suggest that maternal SMCHD1 acts downstream of polycomb marks in this system, consistent with previous work showing that zygotic SMCHD1 acts downstream of H2AK119ub on the inactive X chromosome 24 , that H3K27me3 is unchanged over SMCHD1 targets in zygotic Smchd1 null NSCs 11 , and the role for maternal SMCHD1 at non-canonical imprinted genes controlled by H3K27me3 and H2AK119ub 7 . Potentially, the retention of H2AK119ub and H3K27me3 in the absence of maternal SMCHD1 explains why Hox genes are not aberrantly expressed prior to E8.5. H2AK119ub coverage was also reduced, but only at the 1-cell stage. Seeing as this is the window of time when exclusively maternal SMCHD1 protein is present in the embryo 7 , maternal SMCHD1 may affect anterior Hox genes because lower levels of H3K27me3 and H2AK119ub create a higher dependence on maternal SMCHD1 for an appropriate chromatin state at the Hox genes. These studies also show that neither the dynamic changes in Polycomb marks as the early embryo develops, nor the allele-specificity of them, is captured in mESC, as we also find in our CUT&RUN data from mESCs.
Hence, although our Smchd1 maternal null mESCs appear to retain the maternal effect of SMCHD1 as they exhibit anterior Hox upregulation, in the future it will be important to study the effects of maternal SMCHD1 in the preimplantation period to fully elucidate the role of maternal SMCHD1 at the Hox clusters.
Maternal SMCHD1 acting downstream of Polycomb does not fully answer the question of how deletion of Smchd1 in the oocyte has such a long-lasting effect on the embryo, days after activation of zygotic SMCHD1. Given that zygotic SMCHD1 has a role in maintaining chromatin architecture, specifically long-range chromatin interactions 12,21,23 , including at Hox clusters 12 , it is possible that this long-lasting epigenetic memory exists in the form of a particular chromatin conformation at Hox clusters that is put in place early in development by maternal SMCHD1. Without maternal SMCHD1, we propose that the chromatin state of the Hox clusters is destabilised leaving Hox genes prone to inappropriate activation over time (Fig. 4k). While the Polycomb marks remain, zygotic SMCHD1 activated at the late morula stage appears to be insufficient to ensure appropriate Hox gene silencing later in development.
Potentially this is because zygotic SMCHD1 cannot restore the chromatin architecture required for Hox silencing at the late morula stage or afterwards, as the establishment of such a chromatin state needs to occur within the context of the dynamic epigenetic reprogramming that happens earlier in preimplantation development. If the Polycomb marks are sufficient for silencing in the short term, why would a maternal SMCHD1-mediated chromatin state be required to prevent premature Hox gene activation post-implantation? The early post-implantation period is another time of wholesale epigenome remodelling as the embryo undergoes germ-layer specification and gastrulation. Potentially a destabilised chromatin state created by the absence of maternal SMCHD1 is liable to disruption in the context of such genome-wide remodelling.
Although further work is required to elucidate how maternal SMCHD1 has a long-lasting epigenetic memory in the developing embryo, this study shows that maternal SMCHD1 is required for appropriate Hox gene expression and, consequently, is also required for normal skeletal patterning in the mouse embryo. This work is relevant to our understanding of how maternal proteins influence offspring phenotypes, and may be relevant to humans considering the pathogenic variants in SMCHD1 observed in several human diseases 16,17,18,19 .

Acknowledgements
We thank Jessica Martin (WEHI) for her valuable assistance with mouse lines. We thank Dr Andrew

Author contributions
NB acquired, analysed and interpreted the data and drafted the paper. QG, ATdF and AK analysed and interpreted data. TB and KB acquired data. EM contributed to design of the work, data acquisition, analysis and interpretation. MB conceived and designed the work, analysed and interpreted the data.
All authors edited the manuscript.

Competing interests
All authors declare no competing financial or other interests.

Mouse strains and genotyping
Mice were bred, housed and maintained in accordance with standard animal husbandry procedures and allele-specific genomic analysis due to differential SNPs between the C57BL/6 and Cast genomes.
Genotyping was carried out as previously described for Smchd1 and the X and Y chromosomes 12 ; and for the Cre transgene 48 .

Skeletal preparations
Whole-mount skeletal staining was performed on E17.5 embryos as previously described 25 (Rigueur and Lyons, 2014. Skin and organs were removed, embryos dehydrated and remaining tissue dissolved in acetone. After staining, skeletons were cleared in KOH, washed through a glycerol/water series and imaged in 100% glycerol. Images were acquired with a Vision Dynamic BK Lab System at the Monash University Paleontology Lab. Images were taken with a Canon 5d MkII with a 100mm Macro lens (focus stop 1:3/1:1). Multiple images were taken to extend the focal depth, and stacked in ZereneStacker using the PMax algorithm. Two people independently scored vertebral formulae of each skeleton, blind to genotype and sex.

Tailbud dissection
Tailbud dissection and somite counting was performed as previously described 49 . In brief, embryos were dissected in ice-cold DEPC-treated PBS. Tailbud tissue was horizontally dissected at a distance of 1.5 somites below the last segmented somite to ensure no contaminating somite tissue was included.
Tailbud tissue was snap frozen on dry ice and stored at -80°C for later RNA extraction. The yolk sac was used for genotyping. Somites were counted before fixing each embryo in 4% DEPC-treated then 2i+LIF media. Outgrowths were mechanically disrupted by pipetting in the 2i+ LIF media and then were transferred into a 24-well to be cultured as mESC lines. Cell lines were genotyped to check Smchd1 knockout and only male lines were selected and these lines were grown in non-tissue culture treated plates in suspension in 2i + LIF medium at 37°C with 5% (v/v) carbon dioxide and 5% (v/v) oxygen, and were passaged using Accutase (Sigma-Aldrich) every other day.

Immunofluorescence
Immunofluorescence was performed on differentiating mESCs to NMPs as previously described 12 . In brief, cells grown on coverslips were 13 mm circular glass coverslips (Hecht, cat no. 6.071 724) were washed 3 times for 5 minutes each in PBS before fixation in 4% paraformaldehyde for exactly 10 minutes. Fixed cells were then stored in 0.02% sodium azide in PBS at 4 °C for up to a week until all samples from differentiation were ready for processing. Cells were then washed 3 times for 5 minutes each again in PBS before permeabilisation in 0.5% TritonX in PBS for exactly 5 minutes on ice. Cells were washed again 3 times for 5 minutes each in PBS then non-specific binding sites were blocked in 1% bovine serum albumin (Sigma-Aldrich, A9418) in PBS for approximately 1 hour at room temperature. Primary antibodies against T/Brachyury (Abcam, #ab209665) and Sox2 (ThermoFisher Scientific, cat #14-9811-82) were then added in 1% BSA solution at a dilution of 1:100 and incubated with cells at 4 °C overnight in a humidified chamber. Cells were then washed 3 times for 5 minutes each in PBS before incubation with secondary goat anti-rabbit 647 (ThermoFisher Scientific, cat #A21244) and goat anti-rat 568 (Invitrogen, cat #A11077) antibodies in a dark humidified chamber for one hour at room temperature, before washing 3 times for 5 minutes each again in PBS and counterstaining with DAPI 1:10,000 in PBS for 1 minute at room temperature. Cells were washed 3 times for 5 minutes each again in PBS before mounting on Polysine microscope slides (LabServ, cat #LBSP4981) with Vectashield Vibrance Antifade mounting medium (Vector Laboratories, cat #H-1700). Cells were imaged on an LSM 880 (Zeiss) confocal microscope at 40X magnification and zstacks were merged and composite images generated using the ImageJ distribution package FIJI 51 .
CUT&RUN CUT&RUN was performed as previously described 52 . mESCs grown in 2i media were taken out of culture and counted using a haemocytometer before washing by centrifuging at 600 g for 5 minutes and  between the histone mark ChIP and input libraries. The three biological replicates each for Smchd1 wt and Smchd1 matΔ were merged, and peaks were called against the MACS peaks from the publicly available data using the feature probe generator function in SeqMonk (±2.5kb). Probes were quantitated using the read based quantitation option, normalising to library size and log2 transforming the read count. Read counts were exported from SeqMonk and correlation scatterplots were made using GraphPad Prism v9.0.0. CUT&RUN browser tracks were made by quantifying probes over 1000 bp windows (normalising to library size), sliding by 500 bp, smoothing peaks over 20 adjacent probes.

RNA-sequencing
RNA-sequencing was carried out on tailbud tissue, 2i mESCs and differentiating cells by first extracting RNA using a Quick-RNA Miniprep Kit (Zymo) with DNase I treatment according to the manufacturer's instructions. 100 ng total RNA (or less if <100 ng was yielded from a PSM) was used to prepare libraries using either a TruSeq RNA Library Prep Kit v2 (Illumina) or a TruSeq Stranded mRNA kit (Illumina) according to the manufacturer's instructions. Libraries were size-selected for 200-600 bp and primer dimers cleaned up using Ampure XP beads (Beckman Coulter Life Sciences) and were quantified using a D1000 tape on a 4200 Tapestation (Agilent Technologies). Libraries were then pooled and sequenced on the Illumina NextSeq platform, with 75 bp single-end reads, except for 2i mESC RNA-seq libraries which had paired-end sequencing.

RNA-sequencing analysis
For tailbud, mESC and NMP differentiation RNA-seq, adapter trimming of Fastq files was performed using TimGalore! v0.4.4 with Cutadapt v1.15 53 then QC was carried out using FastQC v0.11.8. Reads were mapped to the GRCm38.p6 version of the mouse reference genome using hisat2 v2.0.5 58  All genomic data is available on the Gene Expression Omnibus under number GSE183740.