Low level DUX4 expression disrupts myogenesis through deregulation of myogenic gene expression

Loss of silencing of the DUX4 gene on chromosome 4 causes facioscapulohumeral muscular dystrophy. While high level DUX4 expression induces apoptosis, the effects of low level DUX4 expression on human myogenic cells are not well understood. Low levels and sporadic expression of DUX4 have been reported in FSHD biopsy samples and myoblast cultures. Here, we show that a large set of human myogenic genes is rapidly deregulated by DUX4, including MYOD1 and MYF5, which are efficiently repressed even by low, non-toxic levels of DUX4. Human myoblasts modified to express low nontoxic levels of DUX4 were significantly impaired from differentiating into myotubes in vitro. Surprisingly, inhibition of differentiation does not require the transcriptional activation domain, thus is likely a feature of all mammalian DUX genes. DUX4 does not bind near the MYF5 gene, but has a prominent ChIP-seq peak within the MYF5 −118 kb enhancer. We find that when DUX4 binds at this site, it directs enhancer activity towards a nearby transcriptional start site for a noncoding nonfunctional RNA we name DIME (DUX4-induced MYF5 enhancer) transcript. These data highlight the anti-myogenic properties of DUX4 in human myogenic progenitor cells, and provide an example of enhancer disruption in the downregulation of MYF5.

is depleted through excessive cycles of damage and regeneration, such as is seen in these other dystrophies, lacks a strong foundation in the histological data. Based on work in murine C2C12 myoblasts, we have previously proposed that regeneration is impaired in FSHD 12 . Very low levels of DUX4 expression, i.e. low enough that mouse myoblasts are not killed, will impair their differentiation into multinucleated myotubes in vitro, a result that has been replicated in primary cells from DUX4-inducible mice in vitro 19 , primary myoblasts transduced with DUX4 20 , and recently, in DUX4-inducible mice in vivo 21 . Although in humans, DUX4 expression levels are higher in differentiated myotubes than in proliferating myoblasts 18 , DUX4 is nevertheless detected in undifferentiated human myoblasts, for example by immunostaining in very rare cells mentioned above, and at levels that can activate expression of a 5xDUX4-binding site GFP reporter gene delivered by lentivector 22 . Since the frequency of GFP+ cells in these lentivector reporter experiments was more than an order of magnitude greater than the frequency of DUX4+ nuclei determined from immunostaining experiments, the DUX4 reporter assay must be more sensitive than DUX4 immunostaining. It is therefore reasonable to consider the possibility that many DUX4+ proliferating myoblasts are expressing low levels of DUX4 protein below the threshold for detection by immunostaining, levels that do not lead to cell death, but that may impair regeneration.
Whether human cells expressing low levels of DUX4 would show impaired differentiation potential like their mouse counterparts is not self-evident. This is because DUX4 induces overlapping but distinct sets of target genes in mouse and human cells 12,13 . The first transcriptional profiling study on FSHD myoblasts did indeed show reduced expression of MYOD1 target genes 23 . Interestingly, a recent meta-analysis of published human biopsy gene expression data has identified a gene set that serves as an FSHD signature of equal or greater strength compared to the DUX4 target gene signature, and this set comprises the inverse of the PAX7 downstream gene expression profile 24 . Because PAX7 is necessary in the skeletal muscle stem cell for maintenance of muscle regeneration, this profile suggests the possibility that, in spite of differences between human and mouse targets, DUX4 might impair the differentiation capability of human myogenic progenitors, as it does those of mouse. In the present study, we directly test this proposition, and find that DUX4 impairs differentiation of human myogenic progenitors; we discover the unusual mechanism by which DUX4 perturbs expression of the myogenic regulator, MYF5; and probe regions of the DUX4 protein necessary for inhibition of myogenesis in the human system.

Results
DUX4 perturbs myogenic gene expression. We previously engineered the immortalized human myoblast cell line LHCN-M2 25 with a doxycycline-regulated DUX4 transgene to obtain the derivative cell line, LHCN-M2-iDUX4 14 , in which DUX4 can be induced to variable levels by treating cells with doxycycline at different concentrations. To identify the earliest gene expression changes induced by DUX4, we evaluated RNA-seq data from LHCN-M2-iDUX4 cells treated with doxycycline for 6 hours. In addition to a large number of upregulated genes, DUX4 provoked downregulation of a large number of genes, which is remarkable because both rapid cessation of transcription followed by turnover of previously expressed transcript would be necessary in order for downregulated targets to be identified within this short 6 hour time frame. Notable among these downregulated targets were the key myogenic regulatory factors, MYF5 and MYOD1. This prompted us to evaluate the global effects of DUX4 on the general myogenic program. Within the set of genes altered within the first 6 hours of DUX4 expression, we found that many important myogenic regulatory genes were strongly altered, in both directions -some being overexpressed, but the majority being repressed (Fig. 1). Because DUX4 is a transcriptional activator, the rapid downregulation of so many genes was unexpected. We therefore evaluated this list of perturbed myogenic genes for nearby peaks of DUX4 binding in LHCN-M2-iDUX4 ChIP-seq data. This revealed that the majority of upregulated targets had a DUX4 binding site within 10 kb of their transcription start site (TSS), whereas none of the downregulated targets had such a nearby DUX4 binding site (Fig. 1). This suggested that the downregulation of these many myogenic genes is not likely associated with DUX4-mediated recruitment of silencing factors, i.e. that DUX4 does not act as a direct repressor of transcription.

Low level DUX4 expression in human myoblasts impairs MYOD1 and MYF5 expression and myotube differentiation.
To determine the effect of DUX4 expression on differentiation of human myoblasts, we exposed LHCN-M2-iDUX4 cells to a low dose of doxycycline, 3.1 and 12.5 ng/mL, to induce low levels of DUX4, which do not lead to cell death within 48 hours. While in the absence of doxycycline, LHCN-M2-iDUX4 cells could be differentiated into large, multinucleated, myosin heavy chain + myotubes, this differentiation was strongly impaired in the presence of 12.5 ng/mL doxycycline ( Fig. 2A). This corresponded with reduced mRNA levels for MYOD1, MYOG, DESMIN and MYH3 (Fig. 2B). It was previously shown that in mouse myoblasts, DUX4 rapidly downregulates both the RNA and protein levels of Myod1 12 . The gene expression changes shown in the heat map in Fig. 1 are at a dose of doxycycline that gives close to maximum output of the Tet-on system (250 ng/mL). To carefully investigate the activity of DUX4 against MYOD1 mRNA and protein in the human system, we exposed LHCN-M2-iDUX4 cells to a doxycycline 4x dilution series, from 200 -0.8 ng/mL doxycycline. DUX4 was detectable by western blot from 12.5 ng/mL, a dose at which MYOD1 was measurably reduced, both at the protein level and at the RNA level (Fig. 2C,D, Supplementary Fig. 1). At higher levels of induction, MYOD1 was undetectable by western blot, and reduced to near zero by RTqPCR. Because MYOD1 is a relatively short-lived protein, the time frame of protein loss was rapid, being virtually complete within 14 hours (Fig. 2C). The myogenic regulatory factors MYOD1 and MYF5 are differentially expressed in the myogenic hierarchy, but they show strong phenotypic compensation 26 . As MYF5 was also downregulated by high level DUX4 expression, we performed a similar dose-response experiment to investigate regulation of MYF5. MYF5 was measurably reduced at 50 ng/mL doxycycline and above (Fig. 2D). Inhibition of differentiation does not require the C-terminal 98 amino acid activation domain of DUX4. The C-terminal 98 amino acid transcriptional activation domain of DUX4 is essential for its cytotoxicity. To determine whether it is also necessary for its inhibition of differentiation, we generated a number of mutant versions of DUX4, in which different lengths of C-terminus were lacking. We also tested DUX4C, a protein expressed from a satellite D4Z4 repeat that has a frameshift mutation replacing the C-terminal 98 amino acid activation domain with a nonsense C-terminus after amino acid 326 (Fig. 3A). It was shown in the mouse system that DUX4C also represses MYF5 27 although the opposite result was shown in the human system 28 . These constructs were transduced into LHCN-M2 cells using the same dox-inducible lentivectors used to express full length DUX4. Expression of the constructs were confirmed by western blot and, and nuclear localization by immunofluorescence ( Fig. 3B,C, Supplementary Fig. 2). We first tested cytotoxicity, and determined that all C-terminal deletions showed no or greatly reduced (DUX4 ) cytotoxicity within 48 hours of induction (Fig. 3D). We tested induction of apoptosis directly by Annexin V staining, and found that no C-terminal deletion mutants were able to induce apoptosis, with the exception of DUX4 , which showed a marginal but statistically significant increase in Annexin V staining (Fig. 3E,F). We also tested each construct for effects on transcription (Fig. 4A)    We then induced this deletion panel of cell lines to differentiate (Fig. 4B). This showed that the C-terminal activation domain was not necessary for the inhibition of differentiation phenotype: both DUX4[1-326] and DUX4C potently inhibited the formation of multinucleated myotubes. However, the shorter constructs, having only the N-terminal half of the protein, had no discernable effect on differentiation. Evaluation of a number of genes associated with terminal differentiation further confirmed the phenotype (Fig. 4C). Coimmunostaining with MHC and DUX4C showed expression of DUX4C in cells that are not differentiating (Fig. 4D). DIME: a DUX4-Induced MYF5 Enhancer transcript. As mentioned above, neither MYOD1 nor MYF5 have nearby DUX4 binding sites. However, we discovered a DUX4 ChIP-seq peak within the −118 kb MYF5 enhancer (Fig. 5A). The −118 kb enhancer corresponds to the −111 kb enhancer identified by transgenic studies in mice 29,30 . MYF5 has a complex set of regulatory elements, most of which are distal to the neighboring gene, MYF6, actually within introns of the next gene along, PTPRQ. The DUX4 ChIP-seq peaks within the −118 kb MYF5 enhancer colocalize with five putative DUX4 binding motifs. Interestingly, three of these cluster within a DNAse hypersensitive site in this cell line that is actually outside of the most highly conserved elements of this enhancer, while two sites cluster within the zone of greatest conservation ( Supplementary Fig. 3). Looking more closely, we found that the DUX4 peak is associated with new RNA-seq reads on the opposite strand (Fig. 5A) suggesting that DUX4 is converting the −118 kb element from a distal enhancer for MYF5 to a proximal promoter for a new transcript. We name this non-coding transcript the DUX4-Induced MYF5 Enhancer (DIME) transcript.
The DIME transcript appears to be non-coding, lacking any ORFs, and thus probably non-functional. To test this, we cloned DIME from DUX4-induced cells, over expressed it in LHCN-M2 cells and found that it did not have any effect on MYOD1 and MYF5, or on myogenic differentiation ( Supplementary Figs 4 and 5).
To investigate the dose sensitivity of induction of the DIME transcript, we designed and employed an intron-spanning qPCR assay, which showed that DIME begins to show weak expression at 12.5 ng/mL dox, but shows quite strong expression at 50 ng/mL dox and above (Fig. 5B), the same point on the curve that MYF5 expression is perturbed. Although it is strongly expressed, it is less sensitive than the most responsive DUX4 target genes, e.g. ZSCAN4 and MBD3L2, which show a transcriptional response at 3.1 ng/mL dox (Fig. 5C). Repurposing of the MYF5-118 enhancer to direct DIME expression rather than MYF5 expression would be expected to cause MYF5 expression levels to decline. To independently confirm DUX4-specific binding to the MYF5-118 enhancer and to demonstrate functional activity, we designed luciferase reporter constructs that carry one of two known distal MYF5 enhancer elements: −118 kb and −136 kb. Interesting, only the −118 kb element showed functional activity in response to DUX4, as seen by elevated luciferase signal after 24 hours of DUX4 induction in iC2C12-DUX4 and 293T-iDUX4 cell lines (Fig. 5D). These data demonstrate that DUX4 can bind to, and activate transcription from, the −118 kb MYF5 enhancer.
Interestingly, when we evaluated the C-terminal deletion constructs, we found that both DUX4[1-326] and DUX4C showed a measurable increase in DIME expression (Fig. 5E). Although this was less than WT DUX4, these constructs were much more potent at activating DIME than ZSCAN4 or MBD3L2 (Fig. 4A), WT DUX4 being about 25x more potent than DUX4  at activating DIME, while being >1,000x more potent at activating ZSCAN4, and even more so for MBD3L2. How can a version of DUX4 lacking its activation domain nevertheless activate expression of DIME? Given that DIME is adjacent to the MYF5 −118 enhancer, it seems likely that DUX4 simply binding to this enhancer perturbs its normal regulation, allowing its activity to be inappropriately directed towards DIME transcription. DIME is induced strongly by DUX4 in myogenic cells. In order to understand whether DIME transcription in response to DUX4 is unique to the LHCN-M2 cell line, and whether it is more effective in cell types in which the MYF5 −118 enhancer is active, i.e. myogenic cell types, we tested DIME induction by DUX4 in four additional cell lines. These included two rhabdomyosarcoma cell lines (RH30-iDUX4 and RD-iDUX4) and two non-myogenic cell lines (A204-iDUX4 and 293T-iDUX4). A204 is a human cell line established from sarcoma (originally thought to be a myosarcoma, but later revised), and 293 T is a human kidney cell line. We modified each cell line with the same dox-inducible DUX4 lentivectors used to generate LHCN-M2-iDUX4. We observed that the rhabdomyosarcoma cell lines expressed their signature markers MYOD1 as well as downstream myogenic genes MYOGENIN and DESMIN (Fig. 6). In all cell lines, DUX4 robustly upregulated its canonical targets, ZSCAN4, MBD3L2, and TRIM43 (Fig. 6). Similar to its activity in immortalized human myoblasts, DUX4 was able to downregulate MYOD1 expression in the rhabdomyosarcoma cell lines. Remarkably, the DIME transcript was only transcribed at highly elevated levels in myogenic cell lines; in nonmyogenic cells, only very low levels of DIME transcript were induced by DUX4 (Fig. 6). Thus, we infer that DIME activation involves more than simply the DUX4 binding site, i.e. it seems likely that the −118 enhancer itself contributes to DIME expression through cooperative activity of enhancer-bound factors with DUX4.

Discussion
These data demonstrate that DUX4 impairs myogenic differentiation in the human system. Similar results were observed when DUX4 was expressed in mouse myoblasts 12 , however the significance of that result was questioned because of the differences between the target genes of DUX4 in the mouse and human systems. Therefore, directly testing whether low non-toxic levels of DUX4 will affect human myogenic gene expression and impair human myogenesis is critical. Two additional findings are interesting in this regard: it was previously reported that FSHD myoblasts formed measurably thinner myotubes when subjected to differentiation 31 ; and it was recently reported that the strongest gene expression profile detected in FSHD biopsies was not that of DUX4 up-regulated targets, but rather that of inverse-regulated PAX7 targets 24 . Both of these studies involving clinical material from FSHD patients lend support to the notion that myogenesis may be impaired in FSHD. This impairment of myogenic differentiation by DUX4 is most likely due to the misregulation of a large number of myogenic genes, not to a specific single target. Because it is rapid, i.e. myogenic gene expression disruption is detected within 6 hours of DUX4 expression, the impaired myogenesis is not a secondary non-specific toxic effect of DUX4. In addition, certain non-toxic mutants of DUX4, for example deletion of the C-terminal activation domain, also impair myogenesis, providing strong support to the notion that the impaired myogenesis is not a secondary effect of DUX4 cytotoxicity. This correlates with recent findings that overexpression of the non-toxic protein DUX4C, which carries an alternative C-terminus due to a frameshift mutation, inhibits differentiation of mouse myoblasts 27 and provokes cytoskeletal defects, and disorganization in human myotubes 32 . It is also notable that translocations of DUX4 into the IgH locus have been implicated in B cell leukemia, and that all leukemia-associated translocations lack the 3′ region of the DUX4 ORF, leading to overexpression of a version of DUX4 lacking the C-terminus 33,34 . This raises the possibility that inhibition of differentiation, probably through downregulation rather than upregulation of target genes, is a feature of DUX4 in multiple cell types.
Interestingly, there are three lineages of DUX genes in the mammalian genome, DUXA, DUXB, and DUXC 35 . DUX4 is the human representative of the DUXC family 36 . Only the DUXC family has the transcriptional activation domain associated with cytotoxicity. Because the transcriptional activation domain and cytotoxicity in general are not necessary for impairment of differentiation, the DUXA and DUXB genes would also be predicted to inhibit differentiation.
Whether the effect of inhibition of differentiation by DUX4C could be responsible for some symptoms of FSHD is an open question. Although it has been reported to be expressed in FSHD 28 , it seems unlikely to be regulated by the D4Z4-distal polyadenylation signal thought to be causative of FSHD 11 . One possibility that bears investigation is whether there are sequence variants allowing DUX4C expression in linkage disequilibrium with 4qA alleles, as DUX4C is very near to the 161 SSLP associated with FSHD, which being proximal to D4Z4 is in fact actually much more closely linked to DUX4C than to the poly A signal sequence.
Because of the large number of DUX4-disrupted myogenic genes and the fact that they are not hierarchically related, impairment of myogenesis is likely multifactorial, i.e. not reversible by simple overexpression of any one of the downregulated factors. Because the majority of DUX4-disrupted myogenic genes are downregulated, and the transcriptional activation domain of DUX4 is not necessary for myogenic inhibition, the data implies that DUX4 is capable of regulating gene expression via multiple mechanisms, at least one of which does not require its C-terminal 98 amino acid activation domain. Furthermore, because the downregulated genes do not have nearby DUX4 binding sites, the mechanisms of downregulation are likely to be indirect. In this study, we probe one specific gene, MYF5, and discover an unexpected and quite unique mechanism of indirect downregulation: interference with activity of a far upstream enhancer by binding and inducing spurious transcription within the enhancer itself. We find that DUX4 binds within the −118 kb enhancer of MYF5, and induces a transcript which we name the DIME (DUX4-Induced MYF5 Enhancer) transcript. The DIME transcript has no obvious open reading frame and is likely to be nonfunctional. We speculate that its main effect is simply the disruption of the −118 doxycycline). Data are presented as mean ± SEM; *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001, by one-way ANOVA with Tukey's post hoc test (RT-qPCR) and t-test (luciferase assay kb enhancer, resulting in disengagement from, and therefore downregulation of, MYF5. The co-involvement of the −118 kb enhancer in regulation of DIME is supported by the fact that strong DIME induction by DUX4 is specific only to myogenic cells (where the −118 kb enhancer is active). Additionally, we found that a mutant version of DUX4 lacking its activation domain was able to induce DIME transcription. Although full-length DUX4 is about 25-fold more potent at DIME activation than DUX4  (DUX4ΔC), this difference is much greater at other DUX4 target genes. For example, full-length DUX4 is approximately 1,000-fold more potent at activation of the classical DUX4 target gene, ZSCAN4, compared to DUX4 . The significant activity of DUX4ΔC on DIME transcription suggests that mere inappropriate binding to the enhancer is capable of appropriating its activity towards alternative nearby targets, such as DIME. In further support, it has been demonstrated that a single DUX4 binding site is insufficient to induce target gene expression at low concentrations (i.e. at single copy gene concentrations of reporter) but rather that DUX4 transcriptional activation is highly cooperative 37 . DUX4 target genes typically have multiple clustered DUX4 binding sites 13,38 . This suggests the likelihood that DUX4 bound to the −118 kb enhancer cooperates with other −118 kb enhancer-bound factors to induce the DIME transcript, i.e. that enhancer activity is appropriated towards DIME. Taken together, these results demonstrate that DUX4 at low, nontoxic, levels interferes with human myogenesis by perturbing myogenic gene expression. We show that effects on myogenesis are primarily due simply to the DNA binding activity of DUX4, facilitated by functional homeodomains, and do not require the C terminal activation domain. Finally, we discover an unexpected interaction mechanism between DUX4 and MYF5, by which a MYF5 distal enhancer is disrupted by DUX4 binding. This is the first evidence of direct interaction of DUX4 with a key myogenic regulator.

Identification of gene expression changes.
We previously generated RNA-seq and ChIP-seq data on LHCN-M2-iDUX4 cells exposed to doxycycline for 6 hours and controls not exposed, which are available at the Gene Expression Omnibus repository under accession code GSE78158 14 . Expression FPKM values were extracted from among the differentially expressed genes (Benjamini-Hochberg adjusted p-value < 0.05), log2 transformed and plotted using the R package Heatmaply. DNase hypersensitive site data for LHCN-M2 cells was obtained from the ENCODE project accession ENCFF001BVR.
Antibodies, immunostaining and western blots. Immunostaining was performed on 4% paraformaldehyde fixed cells, treated with 0.3% Triton X and blocked with 3% BSA (all from Sigma). Primary and secondary fluorochrome conjugated antibodies were diluted in 3% BSA and incubated overnight at 4 °C or for 1 hour at room temperature.
Western blots were performed on proteins separated on 8% SDS-PAGE gels and transferred to PVDF membrane. Primary and secondary HRP conjugated antibodies were diluted in 5% skim milk in TBST and incubated overnight at 4 °C or 1 hour at room temperature. HRP signal was visualized using Pierce ECL western blotting substrate (Thermo Scientific).