The epithelial splicing regulator ESRP2 is epigenetically repressed by DNA hypermethylation in Wilms tumour and acts as a tumour suppressor

Wilms tumour (WT), an embryonal kidney cancer, has been extensively characterised for genetic and epigenetic alterations, but a proportion of WTs still lack identifiable abnormalities. To uncover DNA methylation changes critical for WT pathogenesis, we compared the epigenome of foetal kidney with two WT cell lines, filtering our results to remove common cancer‐associated epigenetic changes and to enrich for genes involved in early kidney development. This identified four hypermethylated genes, of which ESRP2 (epithelial splicing regulatory protein 2) was the most promising for further study. ESRP2 was commonly repressed by DNA methylation in WT, and this occurred early in WT development (in nephrogenic rests). ESRP2 expression was reactivated by DNA methyltransferase inhibition in WT cell lines. When ESRP2 was overexpressed in WT cell lines, it inhibited cellular proliferation in vitro, and in vivo it suppressed tumour growth of orthotopic xenografts in nude mice. RNA‐seq of the ESRP2‐expressing WT cell lines identified several novel splicing targets. We propose a model in which epigenetic inactivation of ESRP2 disrupts the mesenchymal to epithelial transition in early kidney development to generate WT.


Introduction
Wilms tumour (WT; nephroblastoma) is an embryonal kidney cancer [1,2], which originates from foetal kidney (FK), due to the failure of the mesenchymal to epithelial transition (MET) that the metanephric blastema undergoes during early nephrogenesis. Premalignant lesions (nephrogenic rests; NRs) are often found as microscopic lesions in the normal kidney (NK) adjacent to WTs [3]. It is hypothesised that genetic and epigenetic defects occur during renal development that block MET, leading to the formation of NRs, some of which progress to WT [1][2][3].
Epigenetic alterations are also common in WT, especially at the 11p15 locus, where the frequent loss of imprinting of the foetal growth factor gene IGF2 is associated with DNA hypermethylation at H19 [14]. Other epigenetic alterations in WT include loss of imprinting at 11p13 involving imprinted WT1 transcripts [14,15], global hypomethylation [16], DNA hypermethylation at individual tumour suppressor genes such as RASSF1A [16][17][18], and long-range epigenetic silencing of the PCDHG@ gene clusters [19].
Despite the identification of many loci with genetic and/or epigenetic lesions in WT, a proportion of WTs still lack identifiable driver defects, implying that additional novel genes are involved in WT pathogenesis [7]. We previously used genome-wide DNA methylation analysis to identify novel epigenetic lesions in WT [19], and here, we report further studies comparing WT cell lines to foetal kidney. We have identified novel differentially methylated genes, one of which is the alternative splicing regulator ESRP2 (epithelial splicing regulatory protein 2). ESRP2 is known to be important in epithelial to mesenchymal transitions and MET [20], suggesting that epigenetic deregulation of MET may be an important factor in WT development. We show that ESRP2 is frequently silenced by DNA hypermethylation in WT and that it acts as a tumour suppressor gene, regulating alternative splicing in novel genes, some of which affect pathways known to be important in kidney development.

Ethical statement
WT samples were from Bristol Children's hospital (BCH), or from collaborators at the Royal Marsden Hospital (RMH), as part of a UK collaboration. Samples were obtained with informed written consent (from parent and/or legal guardian for children less than 18 years old) and with appropriate ethical approval (E5797, Southwest -Central Bristol Research Ethics Committee (UK)). All methods were performed in accordance with the relevant regulations specified in the UK Human Tissue Act 2004. The study methodologies conformed to the standards set by the Declaration of Helsinki. All animal experiments and procedures were approved by the UK Home Office in accordance with the Animals (Scientific Procedures) Act 1986. Mice were maintained at the Biological Services Unit, University of Exeter, UK. Housing and handling of mice have been done according to the UK Home Office Code of Practice: https://assets.publishing.service.gov.uk/ government/uploads/system/uploads/attachment_data/ file/388895/COPAnimalsFullPrint.pdf.

Cell growth assays
For mass culture assays, cells were seeded into 6-well plates (1 9 10 6 cellsÁwell À1 ) and treated with 2 µgÁmL À1 doxycycline or DMSO vehicle control, with medium changes every 3 days. Cells were trypsinised and counted using a Countess Cell Counter and trypan blue stain to exclude dead cells. For colony assays, cells were seeded into 6-well plates (2 9 10 5 cellsÁwell À1 ) and treated with 2 µgÁmL À1 doxycycline or DMSO vehicle control. Medium was changed every 3 days, and then at 14 days, cells were fixed, stained with methylene blue and colonies counted manually.
To monitor proliferation in real time, cells were seeded in 24-well plates (5 9 10 3 cellsÁwell À1 ) and images were taken in four different fields per well every 2 h (IncuCyte ZOOM live cell imaging system; Essen BioScience) and phase confluence was calculated as a surrogate for growth.

Transwell assay
Cells were pretreated for 4 days with 2 µgÁmL À1 doxycycline or control media and then seeded into Transwell inserts (2 9 10 5 cellsÁinsert À1 in FBS-free DMEM; 8 µm pore, PET membrane; Falcon, 353093) in a 6-well plate. Wells were filled with 1.7 mL 10% FBS in DMEM to produce a chemotactic gradient. After 24 h, inserts were washed and cells on underside of membrane were fixed and stained with crystal violet and counted manually using light microscopy.

Cells
were seeded into 24-well plates (7.5 9 10 5 cellsÁwell À1 ) and treated with 2 µgÁmL À1 doxycycline or DMSO vehicle control, prior to a scratch being performed manually in the centre of each well. Wells were washed with PBS to remove dead cells, control/doxycycline media replaced, and wells were analysed at 24 and 48 h via Widefield microscopy, using ImageJ software to determine percentage wound closure.
2.7. Cell Trace Violet (CTV) proliferation assay 1x10 6 cells were incubated for 20 min at 37°C in the dark in 1 mL of diluted CTV stain (Thermo Fisher; C34571; prepared according to manufacturer's instructions), then staining was quenched using 10% FBS in DMEM, and cells were seeded into T12.5 flasks. Controls were made by fixing 3x10 5 stained cells in 1% paraformaldehyde and stored at 4°C in the dark. Seeded cells in T12.5 flasks were treated for 6 days with 2 µgÁmL À1 doxycycline or control media, and intensity of CTV staining was analysed using a Novo-Cyte Flow Cytometer and FLOWJO software.

Xenografting into nude mice
V200 and E200L cells were transduced with lentivirus expressing firefly luciferase (Amsbio LVP326), and transduced cells were selected with blasticidin, according to the manufacturer's protocol. For orthotopic kidney implantation, male nude mice (2 months old; Charles River) were anaesthetised using isoflurane, an incision was performed in the left flank of the mice, the kidney was exteriorised and 3 9 10 6 cells were injected. Mice were imaged twice weekly (Xenogen IVIS), following intraperitoneal injection with luciferin. When a bioluminescent signal above background was detected (demonstrating the establishment of tumour growth), mice were injected intraperitoneally with doxycycline three times/ week (50 mgÁkg À1 in 5% glucose). Mice were culled either when tumours grew to the maximum allowed size (10mm in diameter, according to the animal licence) or after two months of imaging. The sample size was determined by power calculations using existing data from similar experiments performed routinely in Dr Oltean's lab. More specifically, the sample size was obtained to be able to see a significant difference (P > 0.05) for tumour growth with a power value of 0.80 (> 80%). We have used statistical principles and formulas available on the following websites: www.nc3rs.org.uk; http:// www.statisticalsolutions.net/pss_calc.php. We have not done randomisation in the animal experiments, and there was no blinding of the investigator.

RNA extraction, cDNA synthesis and RT-PCR
Total RNA was extracted using TriReagent (Sigma) and DNase treated with TURBO DNA-free (Ambion, Gloucester, UK). Human foetal kidney RNA was obtained from BioChain. cDNA was synthesised using the Superscript IV RT-PCR system (Invitrogen, Gloucester, UK). Gene-specific primers (Table S1) were used for end-point PCR (HotStarTaq Plus DNA Polymerase; Qiagen), to detect inclusion or exclusion of alternative exons, after electrophoresis on agarose gels (1.5%). Quantitative real-time PCR (qPCR) using genespecific primers (Table S1) was performed using Quanti-Nova SYBR Green Mix (Qiagen) on an MX3000P real-time PCR machine (Stratagene), normalising the amount of target gene to the endogenous level of TBP. Human universal RNA (Agilent) was used as a reference to standardise results between qPCR batches.

Immunofluorescence
Cells were grown on sterile glass slides, fixed for 30 min at room temperature in 1% paraformaldehyde in PBS, permeabilised for 10 min in 0.5% Triton X-100 in PBS and finally rinsed in 50mM glycine in PBS. Fixed cells were stained using a primary antibody against FLAG (mouse, Sigma F3165) and secondary antibody against mouse IgG (Alexa Fluor 488-labelled; Invitrogen) to detect transfected ESRP2, together with Alexa Fluor 594-labelled phalloidin (Invitrogen) to detect actin. Antibodies were diluted in PBS + 1% bovine serum albumin, containing 0.1 µgÁmL À1 DAPI to image nuclei. Slides were mounted in Fluoroshield (Sigma) and examined with a confocal microscope, acquiring eight images at 1 µm spacing/field. Maximum intensity projections were merged using ImageJ software (http:// imagej.nih.gov/ij/).

RNA sequencing (RNA-seq)
RNA was extracted from E200L cells 96 h after treatment with 2 µgÁmL À1 doxycycline, or control solvent (DMSO), using an RNAeasy kit (Qiagen), then DNase treated, and quality confirmed using an Agilent ScreenTape RNA assay. Two biological replicates were used for RNA-seq (i.e. four samples total). Sequencing libraries were prepared from total RNA (500 ng) using the TruSeq Stranded mRNA Library Preparation Kit (Illumina, Inc., Cambridge, UK) and uniquely barcoded adapters (RNA LT adapters, Illumina, Inc). Libraries were pooled equimolarly for sequencing, which was carried out on the NextSeq500 instrument (Illumina, Inc.) using the NextSeq High Output v2 150-cycle kit (Illumina, Inc.). Approximately 300 million paired reads (passing filter, PF) were obtained, divided between the four experimental samples. Next-Seq Control Software version 2.0.0 and RTA v2.4.6 were used for instrument control and primary analysis, respectively. Reads from the four samples were mapped to the human genome (hg19) using the new Tuxedo Suite of programs (HISAT2, StringTie, Ballgown; https://www.ncbi.nlm.nih.gov/pubmed/?term= 27560171). To identify RNA splicing alterations, the four BAM files generated by HISAT2 were used as input for rMATS ( [26] http://rnaseq-mats.sourceforge. net/user_guide.htm). Bam files were viewed in the Integrative Genomics Viewer (http://software. broadinstitute.org/software/igv/) to produce Sashimi plots of alternative splicing. RNA-seq data are accessible through GEO Series accession number GSE154496: https://www.ncbi.nlm.nih.gov/geo/query/ acc.cgi?acc=GSE154496.

Statistical analysis
Comparisons of two datasets were performed using Student's t-test or a Mann-Whitney U-test, depending on whether the data met the normal distribution. A comparison of three or more groups was performed using one-way analysis of variance (ANOVA) with Dunnett's post-test or using Tukey's pairwise test. The Chipmonk software used for MCIP analysis and the rMATS software used for RNA-seq analysis use Benjamini and Hochberg FDR correction for multiple testing. For smaller numbers of samples, Bonferroni correction was used for multiple testing. Numbers of samples quoted in figure legends (n) refer to biological replicates. P < 0.05 was considered to indicate a statistically significant difference.

Genome-wide DNA methylation analysis
We used MCIP to identify 225 genes that were hypermethylated in two WT cell lines compared with foetal kidney ( Fig. 1A and Table S2). Gene ontology analysis showed that these genes were particularly involved in chromatin organisation, developmental processes and transcriptional regulation (Fig. 1B  and  Table S4). To distinguish genes that were methylated specifically in WT, two filters were applied: (a) genes were removed that are polycomb repressive complex marked in embryonic stem cells, since such genes are predisposed to DNA methylation in a wide range of human cancers [27][28][29] and therefore might not be WT-specific; (b) positive selection was applied for genes that are upregulated in early nephrogenesis, since their inactivation could induce the MET block that is critical for WT development [2]. Using these criteria, four candidate genes were pinpointed: CHST2, KIT, PTTG1IP and ESRP2 (also known as RBM35B) (Fig. 1A and Table S2), of which ESRP2 was the most consistently methylated in WT (Fig. 1C).
ESRP2 was particularly attractive for further study, because of its known involvement in epithelial to mesenchymal transitions in cancer [30]. Support for a role in WT came from examination of the ESRP2 target ENAH. ESRP2 induces inclusion of the epithelialspecific exon 11a in ENAH RNA transcripts [24]. Using RT-PCR, less exon 11A was found expressed in WTs compared with normal kidney (NK) and foetal kidney (FK), consistent with downregulation of ESRP2 in WT (Fig. 1D). We therefore went onto examine DNA methylation and expression of ESRP2 in two large cohorts of WTs using pyrosequencing (Figs. S3 and S4).

DNA methylation of ESRP2 in Wilms tumour
The first cohort of WTs from Bristol Children's Hospital (BCH) consisted of tumour samples of all stages, obtained at surgical resection, prechemotherapy. 72% of these WTs were hypermethylated at the ESRP2 (DNA methylation > 25%) compared with normal tissue (NT) ( Fig. 2A and Fig. S4A). The second cohort from the Royal Marsden Hospital (RMH) were from stages 1 to 3, taken at surgical resection, postchemotherapy. In this different cohort, 78% of WTs were hypermethylated ( Fig. 2B and Fig. S4C). ESRP1 DNA methylation was also tested in the RMH cohort and found to be very low (< 2%) in NT and WT, and not significantly different (Fig. 2B). Additional independent DNA methylation data were extracted for the ESRP1 and ESRP2 promoters from the data set GSE59157, which showed hypermethylation of ESRP2 in WTs, with much lower methylation of ESRP1, that was only marginally different between NT and WT (Fig. 2C). Thus, ESRP2 DNA was hypermethylated in three independent cohorts of WTs, but the ESRP2 paralog ESRP1 was not hypermethylated.
ESRP2 is located on chromosome 16q22, a chromosomal region showing frequent loss of heterozygosity (LOH) in WT [31]. No difference was observed in the ESRP2 methylation in WTs with or without 16q LOH (Fig. S5C).
Most WTs are thought to develop via premalignant lesions (NRs) [3]. To characterise the phase of WT development at which ESRP2 DNA methylation occurs, it was assayed in two sets of matched NK, NR and WT. ESRP2 was found to be at a similar level of hypermethylation in NRs and matched WTs compared with NKs (Fig. 2D). In contrast, RASSF1A, a tumour suppressor gene frequently hypermethylated in WT [18], was not hypermethylated in NRs (Fig. 2D), as previously reported [16]. Methylation values were also extracted for NRs from data set GSE59157, and similarly, ESRP2 was significantly more methylated in both NR and WT compared with NK (Fig. 2E), but RASSF1A was only hypermethylated in WTs and not NRs compared with NK (Fig. 2E).
To investigate whether epigenetic changes, including ESRP2 hypermethylation, are associated with other clinical and molecular features, the BCH cohort of WTs were grouped by hierarchical clustering of DNA methylation values at four loci: ESRP2, the WT1 antisense regulatory region [15], H19 [14] and RASSF1A [18] (Fig. S8 and Table S5). Interestingly, of the 22 WTs studied for WT1 mutations, all six WT1-mutant WTs were in the same cluster (group 3), whereas ten of the WT1 wild-type WTs were in group 1 or 2 and six in group 3 (P = 0.015, Fisher exact test). This difference in epigenetic profiles between WT1-mutant and wild-type WTs is supported by similar findings in a recent comprehensive characterisation of molecular defects in WT [7].
ESRP2 DNA hypermethylation was also observed in 10 of 16 (63%) non-WT childhood renal tumours (Fig. S9A), especially in clear cell sarcomas of the kidney and in rhabdoid tumours. In data sets GSE73187 and GSE4487, ESRP2 was also found to be hypermethylated in clear cell sarcomas (Figs S9B, C) and rhabdoid tumours (Fig. S9C)  TCGA data showed DNA methylation changes in ESRP2 in several adult cancers, including hypermethylation in two adult kidney cancers (renal clear cell carcinoma and renal papillary cell carcinoma; Fig. S9D). This suggests that epigenetic inactivation of ESRP2 may be involved in the pathogenesis of several types of renal tumours in adults and children, not just in WT.

Expression of ESRP2 in Wilms tumour
In the BCH cohort, expression of ESRP2 in WT was very low compared with NT ( Fig. 3A and Fig. S4B) and hypermethylation was associated with reduced expression of ESRP2 (Fig. 3B). In the RMH cohort, the expression of ESRP2 was also reduced in WT compared with NT ( Fig. 3C and Fig. S4D), but ESRP1 expression was not significantly different (Fig. 3C). In data set GSE2712, ESRP2 expression was lower in WT compared with NT, but ESRP1 expression did not differ significantly (Fig. 3D). Like methylation results, there was no relationship between ESRP2 expression and tumour histology (Fig. S7B, D,  E). These results showed that ESRP2 but not ESRP1 expression was reduced in WTs compared with NT and that reduced expression of ESRP2 was associated with hypermethylation. When the two WT cell lines were treated with the DNA methylation inhibitor 5aza-2'-deoxycytidine (Aza), there was a 5-to 10-fold increase in ESRP2 RNA expression (Fig. 3E), suggesting a mechanistic link between ESRP2 methylation and gene expression.

Biological function of ESRP2 in vitro
The results described above suggested that Wit49. Overexpression of ESRP2, but not of ESRP1, produced strong growth inhibition in both cell lines (Fig. S10). Due to the strong growth inhibition by ESRP2, we were unable to establish stable cell lines using these constitutively active expression vectors. We therefore transfected the WT cell lines with an inducible ESRP2 expression vector (Fig. S2). Unfortunately, we were unable to establish a stable cell line from 17.94, but the WT cell line Wit49 was successfully transfected, producing the E200L cell line (V200 was the control cell line transfected with empty vector). E200L showed strong doxycycline-induced expression of ESRP2 RNA (Fig. 4A) and protein (Fig. 4B), with the expected nuclear localisation of ESRP2 protein (Fig. 4D). Induction of ESRP2 drove the splicing of the known target gene ENAH [24] towards its epithelial splice form (+exon 11a; Fig. 4C), demonstrating that the construct produced biologically active ESRP2 in a WT cell line. There was slight leakiness from the expression vector, with more ESRP2 RNA detected in uninduced E200L cells compared with V200 cells (Fig. 4A), which probably explains the increased level of ENAH exon 11a in uninduced E200L cells compared with V200 cells (Fig. 4C).
Overexpression of ESRP2 was associated with an apparent redistribution of actin filaments towards the cell periphery (Fig. 4D), compared to a more cytoplasmic distribution of actin stress fibres in uninduced cells (Fig. 4D), as reported in other systems [32,33].
ESRP2 overexpression caused decreased colonyforming efficiency (Fig. 4E), as well as reduced growth rate in mass cultures (Fig. S11). Real-time analysis of cell density showed a slower cell proliferation rate in the doxycycline-induced E200L cells (Fig. 4F), associated with a small but significant decrease in the rate of cell division (Fig. 4G, H). Cell invasion (Fig. S12A) and cell motility (Fig. S12B) showed no changes upon induction of ESRP2 expression.

Xenograft assays of ESRP2 function in vivo
We used orthotopic xenografts of the Wit49-derived cell lines, under the kidney capsule of nude mice [34] (Fig. 5), to examine the effect of ESRP2 expression in vivo. After treatment with doxycycline, tumours produced by V200 cells continued to proliferate, whilst tumours produced by E200L cells stopped growing, or regressed ( Fig. 5A and Fig. S13A, B). V200 cells produced large tumours in four of five mice, but only one mouse out of five injected with E200L cells (1E-L) produced a large tumour (Fig. 5B  and Fig. S13C). Western blotting of excised tumours demonstrated that doxycycline treatment had induced high-level ESRP2 expression in all E200L tumours, with the notable exception of 1E-L (where the tumour grew larger) and V200-induced tumours (Fig. 5C). This therefore demonstrated a strong correlation between ESRP2 expression and suppression of tumour growth.

RNA-seq analysis of alternative splicing in Wilms tumour cell lines
In biological duplicates, we carried out RNA-seq on E200L cells that were doxycycline-induced (ESRP2expressing) or uninduced (non-expressing), obtaining between 70 and 80 million paired-end reads per sample. These reads were mapped onto the human genome, examined for differential gene expression and used in rMATS software [26] to identify alternative splicing events.
Very few transcripts, apart from ESRP2, showed significant changes in RNA expression (P < 0.05, fold change > 2) when ESRP2 expression was induced (Fig. 6A, B). Interestingly, one induced gene was GRHL1, and grainyhead-like transcription factors are important in both kidney development and MET [35], making them good candidates for an involvement in WT. However, we found no difference in expression of GRHL1 between NT and WT (Fig. S14), which does not support a role for altered GRHL1 expression in WT pathogenesis.
In contrast to the lack of altered gene transcription, ESRP2 induction was associated with over 900 splicing events involving over 700 genes, with significant changes (false discovery rate, FDR < 0.05) in skipped exons, mutually exclusive exons and retained introns (Fig. 6C, Table S6, S7, and S8). The genes involved were particularly enriched for biological processes concerned with vesicular and intracellular transport (Table S9). Although we found many ESRP target genes in common with other reports [36,37], we also identified over 600 novel target genes (Fig. 6D). Comparison with a recent study of MET-associated alternative splicing changes during kidney development [38] also revealed overlap with some of our target genes (Fig. 6E). Interestingly, the two lists of genes identified as overlapping our ESRP2 targets, included five genes (33-36%) in common (CTNND1, CTTN, FLNB, MAP3K7 and MPRIP; shown in bold in Fig. 6D, E), emphasising the importance of ESRP-regulated alternative splicing in kidney development.
We validated a selection of putative targets by specific RT-PCR assays, to examine exon inclusion upon ESRP2 induction. We successfully validated several previously identified targets; CD44, ENAH, FGFR2, SCRIB and SLK (Fig. S15), as well as the novel targets LEF1, NPHP1 and RAC1 (Fig. 6F, G, H). However, some putative target genes showed no altered splicing after ESRP2 induction (Fig. S16 and Table S10).
To investigate the possible role of ESRP2 target genes in WT pathogenesis, we examined alternative splicing of 12 genes (seven novel and five previously described) in FK, NK and WT ( Fig. 7 and Fig. S17). Five genes (42%) showed significant changes in the degree of alternative splicing between normal tissues and WT (Fig. 7), and seven (58%) did not (Fig. S17, Table S10).

Discussion
This is the first demonstration of ESRP2 repression caused by DNA hypermethylation in WT, which implicates RNA splicing alterations as an important pathogenic factor in WT development. Investigation of matched sets of NK, NR and WT (Fig. 2D, E) suggested that inactivation of ESRP2 by DNA methylation occurs at an early stage in kidney development, prior to NR formation. We propose that ESRP2 is essential for the differentiation of the metanephric blastema into nephrons (Fig. 8B) and that loss of ESRP2 expression causes a differentiation block, initiating NRs, that can undergo further genetic and     [36,37]. (E) Venn diagram comparing genes identified in this study (SE + MXE + RI) with an RNA-seq analysis of MET-associated splicing changes in the developing kidney [38]. (F, G and H) Alternative splicing of novel targets LEF1 (F), NPHP1 (G) and RAC1 (H). Left-hand panels: Sashimi plots of RNA-seq data from E500L cells uninduced (-Dox) or induced to express ESRP2 (+Dox). Right-hand panels: Agarose gels of RT-PCRs of amplicons spanning alternatively spliced exons (Table S1 for primers), in V200 and E200L cells, either uninduced, or doxycycline-induced to produce high-level ESRP2 expression in E200L cells. epigenetic defects to produce WT (Fig. 8C). Support for this model comes from studies showing that the Esrp paralogs are expressed in the developing kidney [39], with increased expression of Esrp 1 and 2 when renal precursors undergo epithelial differentiation [38], and that knockout of Esrp genes in mice decreases kidney volume, due to a lack of nephrons [40].
Inactivation of ESRP2 as an early premalignant event in WT development probably explains why we found no association with clinical features (Figs S5 and S6). It also explains why we found no association between ESRP2 methylation and LOH at 16q (Fig. S5C), where the ESRP2 gene is located, because we have previously demonstrated that 16q LOH occurs after NR formation [14], that is after ESPR2 hypermethylation.
We have shown that ESRP1, though an ESRP2 paralogue, is not repressed by hypermethylation in WTs (Fig. 2B, C and Fig. 3C, D). This implies that ESRP1 and ESRP2 may have different biological functions and are regulated differently in some instances, as recently reported in prostate cancer, where ESRP2 but not ESRP1 is regulated by androgens [41].
Splicing alterations are frequent in human cancers [42], including ESRP-induced changes in breast cancer [43,33], prostate cancer [44,41], renal cell carcinoma [45] and colorectal cancer [46]. Most studies have reported expression changes without finding underlying genetic or epigenetic defects in the ESRP genes themselves [46,43,45,41,33]. However, there are reports of genetic defects in ESRP genes in human cancers, specifically, microsatellite indels [47] or duplications [44] of ESRP1. In addition, there are reports of DNA methylation changes in ESRP1 in prostate cancer [48] and of ESRP2 in breast cancer [49], and our examination of TCGA data (Fig. S9D) demonstrated ESRP2 methylation changes in several other adult cancer types. Thus, our results add to a growing body of evidence that ESRP genes can be either genetically or epigenetically deregulated in a wide range of human cancers.
Our functional studies suggested that the main biological effect of ESRP2 is to regulate cell proliferation by slowing cell division (Fig. 4E-H and Fig. S11). Whilst we observed some actin cytoskeleton rearrangement (Fig. 4D), we did not observe significant expression changes in classical epithelial marker genes (Fig. 6A, B), nor any changes in cell motility or invasion (Fig. S12), unlike what occurs when ESRP expression is modulated in adult human cancer cell lines [32,33,30]. Coupled with our xenograft experiments that identify ESRP2 as a bona fide tumour suppressor gene (Fig. 5), these results suggest that the tumour suppressor activity of ESRP2 in WT cell lines occurs mainly by altering cell growth properties, rather than by affecting cellular differentiation.
Mechanistically, our RNA-seq results demonstrated that ESRP2 modulated the splicing of a diverse range of genes, including both well-established and novel targets ( Fig. 6 and Tables S6 to S8). A subset of these genes showed reduced expression of their epithelial splice forms in WT (Fig. 7) (Table S1 for  hypermethylation-induced downregulation of ESRP2 in WT (Fig 2 and Fig. 3). Interestingly, of the 728 genes that we identified as having their splicing modulated by ESRP2 (Fig. 6), only 62 (9%) are WT1 DNA-binding targets [50], whereas 244 (34%) are WT1 RNA-binding targets [51] (Fig. 8A). The WT1 RNA-binding targets include all five of the ESRP2regulated genes that we found in common between our results and two other RNA-seq studies (Fig. 6D, E). This suggests that WT1 and ESRP2 are involved in the post-transcriptional regulation of a similar set of genes during renal development. Since ESRP2 hypermethylation is an early event, like WT1 mutation [52,53], this suggests that ESRP2 hypermethylation may be another important early event in WT development, which contributes to WT pathogenesis by inhibiting MET (Fig. 8C). These results, together with genetic evidence showing defects in miRNAprocessing genes in WT [8][9][10][11][12], reinforce the critical role that post-transcriptional gene regulation plays in WT pathogenesis.

Conclusions
Our genome-wide DNA methylation analysis of WT has identified ESRP2 as a novel differentially methylated gene.  Fig. 6) with 1663 WT1 DNA-binding targets identified by chromatin immunoprecipitation in developing kidney [50] and 4503 WT1 RNA-binding targets (protein-coding genes) identified by RNA immunoprecipitation in M15 mesonephric cells [51]. (B) ESRP2 may be required for epithelial differentiation, to form nephrons during kidney development. (C) Loss of ESRP2 function by hypermethylation may inhibit normal differentiation and therefore promote persistence of undifferentiated blastema, leading to nephrogenic rest formation and eventual progression to Wilms tumour. B and C adapted from Fig. 2 in reference [1].
suppressed tumour growth of orthotopic xenografts in nude mice, demonstrating that ESRP2 acts as a tumour suppressor gene in WT. Using RNA-seq of the ESRP2-expressing WT cell lines, we have identified several novel splicing targets, some of which affect pathways known to be important in kidney development. We propose that epigenetic inactivation of ESRP2 disrupts the regulation of alternative splicing during the mesenchymal to epithelial transition in early kidney development, to generate WT.