Deregulation of the histone H3K9 di-methylation landscape suppresses canonical Wnt signaling in embryonal rhabdomyosarcoma

The Wnt signaling pathway is down-regulated in embryonal rhabdomyosarcoma (ERMS) and contributes to the block of myogenic differentiation. Epigenetic mechanisms leading to its suppression are unknown and could pave the way towards novel therapeutic modalities. In this study, we demonstrate that the H3K9 lysine methyltransferase G9a suppresses canonical Wnt signaling by activating expression of the Wnt antagonist DKK1. Inhibition of G9a expression or activity reduced DKK1 expression and elevated canonical Wnt signaling resulting in myogenic differentiation in vitro and in vivo. Mechanistically, G9a impacted Sp1 and p300 enrichment at the DKK1 promoter in a methylation-dependent manner. The reduced tumor growth upon G9a deficiency was reversed by recombinant DKK1 or LGK974, which also inhibits Wnt signaling. Consistently, among thirteen drugs targeting chromatin modifiers, G9a inhibitors were highly effective in reducing ERMS cell viability. Together, our study demonstrates that ERMS cells are vulnerable to G9a inhibitors and suggest that targeting the G9a-DKK1-β-catenin node holds promise for differentiation therapy.


G9a inhibitors reduce ERMS cell viability
We recently reported that G9a is overexpressed in ARMS (20). To examine whether G9a expression is de-regulated in ERMS, and if it is functionally relevant in these tumor subtype, we first examined its expression in 16 ERMS patient tumor sections using an anti-G9a antibody. High nuclear expression relative to normal muscle was apparent ( Figure 1A). In addition, compared to primary human skeletal muscle myoblasts (HSMM), G9a overexpression at both mRNA and protein levels was apparent in three ERMS patient derived cell lines RD18, JR1 and RD ( Figure 1B and Figure 1C). To examine if the G9a pathway is functionally relevant, we treated JR1 and RD cell lines with 13 methyltransferase inhibitors at three different concentrations. Viability was measured 8 days later using MTS assay. Drugs targeting BRD4, PRMT5 and G9a showed a strong effect on viability of both cell lines ( Figure 1D and Figure 1E). Consistently, treatment of JR1 cells with UNC0642, a small molecule inhibitor of G9a led to a striking reduction in colony formation ( Figure 1F). Together, these results indicate that G9a is overexpressed and functionally relevant in ERMS.

G9a inhibition promotes myogenic differentiation and inhibits proliferation in ERMS cell lines
To examine the role of G9a in ERMS, we depleted its endogenous expression in RD18, JR1 and RD cells using small interfering RNA ( To differentiate, myoblasts irreversibly exit the cell cycle (21). Given the enhanced myogenic differentiation upon G9a depletion, we investigated the impact of G9a loss on proliferation by labelling S-phase cells with BrdU. Both siG9a cells and UNC0642 treatment resulted in a significant decrease in BrdU + cells compared to their respective control in RD18 cells ( Figure 2D Figure S2H). Together, these results indicate that G9a inhibition permits cells to exit the cell cycle and undergo myogenic differentiation.

G9a regulates DKK1 and canonical Wnt signaling
In order to identify mechanisms underlying G9a function, we performed RNA-Sequencing (RNA-Seq).
Cluster analysis of differentially expressed genes from control RD and G9a knockdown cells was done in triplicates ( Figure 3A). Volcano plot of differentially expressed genes ( Figure 3B) revealed that 872 genes were significantly up regulated in siG9a cells compared to the control, of which 494 genes had a fold change >1.2. Among the 1098 genes that were significantly down regulated in siG9a cells, 695 genes had a fold change >1.2. Gene Ontology (GO) analysis showed that among the top 20 unique biological processes associated with differentially expressed genes in siG9a cells were cell cycle progression and Wnt signaling ( Figure 3C). Given its relevance in ERMS, we focused on the Wnt pathway. Interestingly, negative regulators of the Wnt pathway such as DKK1, DKK3 and ITGA3 (22,23) were downregulated in siG9a cells, whereas positive regulators such as Wnt3 and FRAT2 were upregulated (Supplementary Figure 3A DKK1 is a member of the Dickkopf family that inhibits canonical Wnt/ß-catenin signaling by binding to and inhibiting the Wnt co-receptor Lrp5/6 (16). Consistent with the transcriptomic data, downregulation of DKK1 mRNA was apparent in siG9a cells compared to control cells by qPCR ( Figure 3D).
Interestingly UNC0642 treatment also resulted in a decrease in DKK1 mRNA expression ( Figure 3E). DKK1 protein levels also decreased in both siG9a and UNC0642 treated cells ( Figure 3F and G).
Moreover, the reduction in DKK1 levels correlated with increased active-ß-catenin in siG9a cells ( Figure   3H), shG9a cells ( Figure 3I) and upon UNC0642 treatment ( Figure 3J) when compared to their respective controls. These results indicate that loss of G9a leads to downregulation DKK1 with concomitant activation of canonical Wnt signaling and myogenic differentiation.

G9a regulates DKK1 through Sp1/p300 occupancy
To investigate mechanisms by which G9a activates DKK1 in a methylation dependent manner, we analysed enrichment of the transcription factor Sp1 and the co-activator p300 that have previously been shown to regulate DKK1 expression (24,25) . Sp1 occupancy was detected at the DKK1 promoter ( Figure 4A). Intriguingly, both Sp1 and p300 enrichment were decreased upon treatment with UNC0642 compared to control cells. Correspondingly a reduction in H3K9ac, a mark of transcriptional activation was apparent ( Figure 4B-D). A decrease in p300 and H3K9ac occupancy was also observed in shG9a cells (Supplemental Figure S4A-B).
Sp1 interacts with p300 through its DNA binding domain (26). We therefore examined if Sp1 and p300 interaction was altered by UNC0642 by proximity ligation assay (PLA). The interaction between Sp1 and p300 decreased significantly upon UNC0642 treatment even though that of G9a-Sp1 and G9a-p300 remained unchanged ( Figure 4E and Supplementary Figure S4C, D). We then examined mechanisms leading to loss of Sp1 at the DKK1 promoter. Cyclin dependent kinase 1 (CDK1) phosphorylates Sp1 in its DNA binding domain which decreases its DNA binding capacity (27-29).
Interestingly, CDK1 was upregulated in siG9a cells in RNA-Seq analysis and qPCR validation confirmed its increase in both siG9a cells and upon UNC0642 treatment at the mRNA and protein level ( Figure   4F-I). Similar increase of CDK1 expression upon G9a inhibition was also observed in RD cells (Supplementary Figure 4E-H). To identify if CDK1 is a direct target, ChIP-seq analysis of G9a occupancy was performed which revealed its binding mostly at gene promoters ( Figure 4J) and its enrichment was apparent at the CDK1 promoter ( Figure 4K). To validate these results, ChIP-PCR was done in RD18 cells and a significant enrichment was seen indicating that G9a directly binds to the CDK1 promoter ( Figure 4L). These results indicate that G9a represses CDK1 expression in a methyltransferase activity-dependent manner. Upregulation of CDK1 expression upon G9a inhibition may result in increased Sp1 phosphorylation and its decreased DNA binding resulting in reduced Sp1-p300 occupancy and consequently downregulation of DKK1 expression.

G9a inhibits differentiation and promotes proliferation through DKK1-mediated antagonism of Wnt signaling.
Canonical Wnt signaling induces myogenic differentiation and decreases proliferation (30, 31). As DKK1 is a well characterized inhibitor of canonical Wnt signaling, we examined if the effect of G9a is mediated by DKK1. Correlating with high endogenous G9a expression, DKK1 was also overexpressed in all three lines compared to HSMM at both mRNA and protein level ( Figure 5A and B). Moreover, analogous to G9a knockdown, DKK1 knockdown in RD18 cells resulted in a significant decrease in BrdU + cells compared to control cells ( Figure 5C). A corresponding increase in differentiation was also apparent by elevated MHC levels and myogenin expression ( Figure 5D). To determine whether DKK1 mediates the effects of G9a, we performed rescue experiments. Recombinant DKK1 (rDKK1) was added to siG9a cells for 24 hr that resulted in the reduction of active b-catenin seen in siG9a cells ( Figure 4E). Interestingly, in presence of rDKK1, the increase in MHC + cells and myogenin expression in siG9a cells was reversed to control levels ( Figure 5E). Similarly, the decrease in BrdU + cells upon G9a knockdown were restored to levels comparable to control. To further validate that G9a mediates its effects on canonical Wnt signaling, we used another Wnt antagonist, a porcupine inhibitor, LGK974.
Similar to rDKK1, LGK974 reversed the effects of G9a knockdown on proliferation, differentiation and active b-catenin levels ( Figure 5F) indicating that G9a mediates the differentiation block by activating DKK1 expression that in turn suppresses canonical Wnt signaling.
In order to examine the effect of G9a in regulating DKK1 and Wnt signaling in vivo, we injected RD cells in BALB/c nude mice. Once the tumors were palpable, mice were injected intraperitoneally every two days with UNC0642 or with control vehicle. Treatment with UNC0642 resulted in reduced tumor growth compared to the control group without any significant changes in body weight ( Figure 6A). By immunohistochemical analysis ( Figure 6B) we confirmed a decrease in H3K9me2 in tumors from mice treated with UNC0642 indicating efficacy of the drug in vivo. The proliferation marker Ki67 was decreased, whereas MHC + cells were increased in tumors from mice treated with UNC0642. Moreover, DKK1 was decreased upon UNC0642 treatment, and correspondingly active-ß-catenin levels were elevated ( Figure 6B).
To verify that the effects of G9a are mediated via an impact on Wnt signaling in vivo, we next injected control shRNA and shG9a cells. Once the tumors were palpable, mice injected with shG9a cells were treated with LGK974 or treated with vehicle alone every alternate day. The shRNA control group was also injected with the vehicle ( Figure 6C). Tumor volume in mice injected with shG9a cells was reduced compared to the control group. However, mice injected with shG9a cells and treated with LGK974 showed tumor volumes comparable to the control group. Body weight of mice did not show any significant changes over the course of this treatment. We analysed two tumors from each cohort by IHC. As expected, G9a expression was decreased in mice injected with shG9a cells that correlated with decrease in Ki67 + cells and increase in MHC + cells compared to tumors of the control group. This alteration was however not seen in the tumors from mice injected with shG9a cells that underwent LGK974 treatment. Consistently, the reduction of DKK1 and increase in active-ß-catenin in tumors from shG9a injected mice was reversed in upon LGK974 treatment ( Figure 6D). Taken together, these studies indicate that inhibition of G9a expression or activity decreased tumor progression by regulating Wnt signaling.

Discussion
In this study, we uncovered a G9a-dependent epigenetic node that results in repression of Wnt signaling in ERMS. We propose that by activating DKK1 expression, G9a maintains Wnt signaling in a repressed state, and thus prevents the transition of myoblasts to a differentiated state. These findings underscore specific epigenetic mechanisms to reactivate Wnt signaling and induce differentiation in ERMS.
There has been a resurgence of interest in differentiation therapy as a viable treatment option for solid tumors (32). The Wnt, Notch and Hedgehog pathways play a pivotal role major role in balancing proliferation, self-renewal and differentiation during embryonic myogenesis. Not surprisingly, deregulation of these developmental pathways has been reported in ERMS (12,33,34) and consequently, drugs targeting each of these pathways are being tested. However, current inhibitors do not selectively target specific pathways and have either unacceptable toxicity, or do not show marked clinical improvement. Several studies have demonstrated suppression of Wnt signaling in RMS cells, which mostly do not stain positively for nuclear b-catenin. Also, no mutations in the b catenin gene have been reported (35,36). The suppression of Wnt signaling is a critical contributor to the differentiation block in ERMS, and induction of the pathway leads to cell cycle exit and differentiation (12,14). GSK3ß inhibitors are the most prominent Wnt signaling activators and showed promising effects in inducing differentiation in a zebrafish model of ERMS (14). GSK3ß is a constitutive serine/threonine protein kinase that inhibits canonical Wnt signaling by phosphorylating ß-catenin and triggering its degradation in the cytoplasm (37). However, GSK3ß is involved in many pathways with more predicted substrates than any known kinase. Thus, its inhibition is not a specific strategy to induce Wnt signaling. This is also emphasized by slow progress of existing GSK3ß inhibitors toward clinical translation (38). Tideglusib, an irreversible GSK3ß inhibitor, was recently tested against rhabdomyosarcoma PDX models, where the highest safe dosage failed to both induce myogenic differentiation as well as affect cancer progression in PDX models (39). Consequently, development of alternative molecularly targeted therapies that induce Wnt signaling is critical goal in this disease.
Our data demonstrates an epigenetic mechanism to activate Wnt signaling and overcome the differentiation block in ERMS. Interestingly G9a activates DKK1 in a methylation-dependent manner resulting in the suppression of Wnt signaling in ERMS. While methylation-dependent silencing or repression by G9a has been described, only a few studies have demonstrated methylation-dependent activation of gene expression. Consistent with previous studies, we found Sp1 binding at the DKK1 promoter and inhibition of G9a led to its decreased occupancy. Sp1 transcriptional activity is regulated through a variety of post-translational modifications (27). CDK1/cyclin B1 phosphorylates Sp1 at Threonine (Thr) 739 at the Sp1 DNA binding site (29) and decreases its DNA binding. Since CDK1 is repressed by G9a in a methyltransferase-dependent manner, its elevated levels upon UNC0642 treatment may result in reduced Sp1 binding. Phosphorylation of Sp1 at the DNA binding domain might also result in the dissociation of p300 from Sp1, as Sp1 interacts with p300 through its DNA binding domain (26). Consequently, the activation mark H3K9ac also decreased at the DKK1 promoter. Thus, our data establish a novel mechanism in which G9a methyltransferase activity is required for Sp1-p300 association for activation of DKK1 expression. However, there may be additional mechanisms by which G9a regulates DKK1 expression.
Pan-HDAC and pan-DNMT inhibitors have been explored in ERMS (9,40). HDAC inhibitors often exert effects independent of their epigenetic roles (9) and application of DNMT inhibitors (41) are restricted due to their toxicity in healthy cells. EZH2 inhibitors are the only other specific epigenetic inhibitors that demonstrate a strong phenotype in ERMS (8). In this study, a drug screen of 15 methyltransferase inhibitors in two different ERMS cell lines showed that small molecule inhibitors targeting G9a activity are very effective. Thus, our data support targeting G9a as a therapeutic approach in ERMS, particularly since its deletion does not impact development of muscle (42). We have recently shown that G9a is deregulated in ARMS as well (20). Together with the herein described role of G9a in ERMS, these observations clearly suggest the importance of differentiation in these therapies and imply common features between these two subtypes of RMS.

Materials and Methods
Cell culture and drug sensitivity assays
Appropriate secondary antibodies (IgG-Fc Specific-Peroxidase) of mouse or rabbit origin (Sigma Aldrich) were used.

Proximity Ligation Assay
Proximity ligation assay (PLA) was performed using the Duolink in situ-fluorescence (Sigma DUO92101). For G9a and Sp1 interaction, anti-G9a (Cell Signaling, 1:50) and anti-Sp1(1:50 Santa Cruz) antibodies were used. For Sp1 and p300 interaction studies, Sp1 (Millipore, 1:100) and p300 (Abcam, 1:1000) antibodies were used. Images were captured under FluoView FV1000 confocal fluorescence microscope (Olympus) at 60X (oil). For quantifying PLA signals, particle analysis was performed using Fiji/ImageJ software, and pixel area size of 2-50 was assigned for calculating the total number of PLA signals per field. PLA signals as dots per nuclei were calculated for at least 3 microscopic fields.

Transcriptome analysis and Quantitative real-time polymerase chain reaction (qPCR)
For RNA sequencing analysis, RNA was isolated from control and siG9a cells in triplicate using Trizol.
RNA was sequenced using Illumina high-throughput sequencing platform. CASAVA base recognition was used to convert raw data file to Sequence Reads and stored in FASTQ(fq) format. Raw reads were then further filtered in order to achieve clean reads using the following filtering conditions: reads without adaptors, reads containing number of base that cannot be determined below 10% and at least 50% bases of the reads having Qscore denoting Quality value <=5. For mapping of the reads STAR software was used to align the reads against hg19 Homo sapiens reference genome. 1M base was used as the sliding window for distribution of the mapped reads. For analysis of the differentially expressed genes, Gene Ontology analysis was done using cluster Profiler (43) software for GO terms with corrected p value less than 0.05.
For qPCR analysis, total RNA was extracted using Trizol (Thermo Fisher Scientific) and quantified using binding. Peaks were annotated using ChIPseeker (44) and ChIPpeakAnno (45) against genes model of UCSC, hg19, knownGene (TxDb.Hsapiens.UCSC.hg19.knownGene). Differential Expression count matrix were analysed using R. Genes with an adjusted p-value less than 0.05 were labelled as significantly differentially expressed. From that list, genes with an absolute fold change >= 1.2 were selected for further analysis. GO analysis of gene subsets were performed using Metascape (46). The ChIP-Seq data are compliant with MIAME guidelines and have been deposited in the NCBI GEO database.

Mouse xenograft experiments
6-week-old C.Cg/AnNTac-Foxn1 nu NE9 female BALB/c nude mice (InVivos, Singapore) were injected subcutaneously in the right flank with control RD cells (10 X 10 6 ). Once tumors were palpable, one group (n=10/group) was injected intraperitoneally with vehicle [5% DMSO in PBS], and the other with UNC0642 (5mg/kg body weight in 5% DMSO) every alternate day. Tumor diameter and volume was calculated as described (20). Resected tumors were fixed and paraffin sections were immunostained with various antibodies. To determine the role of Wnt signaling, one group of mice were injected with RD shcontrol cells (n=10/group) and two groups with RD shG9a cells. Once tumors were palpable, the control group and one shG9a group were injected intraperitoneally with control vehicle [2% DMSO in corn oil], and one group of shG9a mice was injected with LGK974 (5mg/kg body weight in 2% DMSO).
Tumour growth and body weight were recorded as described (20). All animal procedures were approved by the Institutional Animal Care and Use Committee.

Immunohistochemistry (IHC)
Paraffin sections from 16 primary ERMS archival tumour specimens and three normal muscle from National University Hospital (NUH) and KK Women's and Children Hospital in Singapore were analysed by IHC using anti-G9a antibody (1:50 dilution, Cell Signaling) as described (20). Negative controls were performed using secondary antibody only. Images were captured with Olympus BX43 microscope (Inashi, Nagano, Japan). Approval was obtained from the ethics committee (

Statistical analysis
For statistical analysis, two-tailed non-parametric unpaired t test was used to evaluate significance with the use of GraphPad prism 9.0 software. Each experiment had three biological replicates. Standard error of mean (SEM) was calculated for all data sets and a p value less than 0.05 was considered statistically significant. *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001. For in vivo experiments repeat-measure two-way ANOVA was used to calculate the statistical significance between different groups.

Data availability
ChIP-Seq data has been deposited in GEO under the accession number GSE125960. Private token for referees: qbgtkgganjmlrwd. RNA-Seq data been deposited in GEO under the accession number GSE142975. Private token for referees: unylciiwfvorjgd.      (D) Tumors from two different mice in each group were analyzed by IHC for Ki67, H3K9me2, DKK1, MHC and active-ß-catenin staining as described above. Scale bar: 50µm. Inset shows zoomed in images.