Epigenetic deregulation of GATA3 in neuroblastoma is associated with increased GATA3 protein expression and with poor outcomes

To discover epigenetic changes that may underly neuroblastoma pathogenesis, we identified differentially methylated genes in neuroblastoma cells compared to neural crest cells, the presumptive precursors cells for neuroblastoma, by using genome-wide DNA methylation analysis. We previously described genes that were hypermethylated in neuroblastoma; in this paper we report on 67 hypomethylated genes, which were filtered to select genes that showed transcriptional over-expression and an association with poor prognosis in neuroblastoma, highlighting GATA3 for detailed studies. Specific methylation assays confirmed the hypomethylation of GATA3 in neuroblastoma, which correlated with high expression at both the RNA and protein level. Demethylation with azacytidine in cultured sympathetic ganglia cells led to increased GATA3 expression, suggesting a mechanistic link between GATA3 expression and DNA methylation. Neuroblastomas that had completely absent GATA3 methylation and/or very high levels of protein expression, were associated with poor prognosis. Knock-down of GATA3 in neuroblastoma cells lines inhibited cell proliferation and increased apoptosis but had no effect on cellular differentiation. These results identify GATA3 as an epigenetically regulated component of the neuroblastoma transcriptional control network, that is essential for neuroblastoma proliferation. This suggests that the GATA3 transcriptional network is a promising target for novel neuroblastoma therapies.

In order to identify epigenetic changes associated with the development of neuroblastoma, we previously used genome-wide DNA methylation analysis to compare neuroblastoma cells to neural crest cells, their cellular precursors, and identified MEGF10 as an epigenetically repressed putative tumour suppressor gene 24 . Interestingly, like other recent studies 31,32 , we found a preponderance of hypomethylated genes, suggesting that epigenetic activation of normally silenced genes in neural crest cells is critical in neuroblastoma pathogenesis. In this paper we have therefore investigated the hypomethylated genes identified in our previous work 24 , demonstrating that GATA3, a transcription factor known to be critical in development of the sympathetic nervous system [36][37][38] and a gene often dysregulated in diverse human cancers 39 , is frequently affected by epigenetic deregulation in neuroblastoma.

Results
Hypomethylated genes in neuroblastoma. We have previously described our genome-wide analysis of DNA methylation in neuroblastoma, in which we used methyl CpG immunoprecipitation (MCIP) and promoter microarrays, to compare neuroblastoma cell lines with human neural crest cells (hNCC) 24 . In our previous paper, we described our results on genes that were hypermethylated in neuroblastoma compared to hNCC 24 ; in this paper we discuss hypomethylated genes.
A total of 67 genes were found by MCIP to be hypomethylated in neuroblastoma cell lines compared to hNCC ( Fig. 1 and Supplementary Table S1). Gene ontology analysis showed this group to be enriched in genes regulating biological processes, G-protein coupled receptor signalling and sensory perception (Supplementary Table S2). To filter the hypomethylated genes for those likely to be functionally important in neuroblastoma pathogenesis, we used publicly available databases to search for genes that; (1) were overexpressed in neuroblastoma cell lines and tumour tissue compared to hNCC (reasoning that hypomethylated genes would be expected to have elevated expression) and (2) showed an association of high gene expression in tumours with poorer patient survival (to identify genes affecting neuroblastoma biological properties). Six genes had both the required over-expression and association with poor patient survival, but only one, GATA3, had a high CpG-content CpG island (CGI), suggesting it might be most susceptible to epigenetic deregulation ( Fig. 1 and Supplementary Table S1). In addition, GATA3 was known to be of importance in development of sympathetic nervous tissue, from which neuroblastoma derives 1,2 . We therefore went on to examine the DNA methylation, expression and functional relevance of GATA3 in neuroblastoma.
GATA3 methylation in neuroblastoma. Examination of the MCIP data showed that the major area of hypomethylation in neuroblastoma cell lines compared to hNCC, centred on the start of the GATA3 antisense transcripts in the CGI ( Fig. 2A). We used two commercially-available pyrosequencing assays to examine DNA methylation in this region ( Fig. 2A) and found that neuroblastoma cell lines and tumour tissue were significantly hypomethylated compared to a panel of control tissues, consisting of hNCC, fetal adrenal tissue (FA) and dorsal root ganglia/sympathetic ganglia cell lines (DRG/SG) (Fig. 2B,C).
We extracted GATA3 DNA methylation data from the publicly available dataset GSE54719, which confirmed hypomethylation across the start of the GATA3 antisense transcripts in neuroblastoma tumours, compared to adrenal tissue ( Supplementary Fig. S1).
Using survival data available for our patient cohort, we demonstrated that complete absence of DNA methylation within the hypomethylated region was associated with worse relapse-free survival and worse overall survival in neuroblastoma patients (Fig. 2D,E).
These results suggested that epigenetic deregulation of GATA3 by altered DNA methylation might be functionally important in neuroblastoma. We therefore investigated whether DNA methylation affected GATA3 expression in neuroblastoma.

Relationship between GATA3 methylation and expression.
In almost all neuroblastoma cell lines, there was little or no GATA3 DNA methylation, and much higher levels of expression of both the sense and antisense GATA3 RNAs, compared to normal tissues (FA, hNCC and DRG/SG), which had 7-22% DNA methylation, and very low RNA expression (Fig. 3A). The exceptions were SH-EP and GI-ME-N, S-type neuroblastoma cell lines (Supplementary Table S3), which had some GATA3 methylation and low RNA expression (Fig. 3A). In general, GATA3 sense and antisense RNA levels correlated positively, with N-type neuroblastoma cells generally having the highest antisense expression ( Fig. 3A and Supplementary Fig. S2C). There was an inverse correlation between GATA3 DNA methylation and RNA expression in normal and neuroblastoma cell lines ( Supplementary  Fig. S2A,B).
GATA3 protein was overexpressed in nearly all neuroblastoma cell lines compared to normal tissue (Fig. 3B). The S-type cell line GI-ME-N had undetectable GATA3 protein and SH-EP had one of the lowest levels of expression (Fig. 3B), which agrees with the low levels of GATA3 RNA expression detected in these two S-type cell lines (Fig. 3A). Overall, there was a good correlation between GATA3 RNA and protein expression in the neuroblastoma cell lines ( Supplementary Fig. S2D).
These findings suggested a possible relationship between GATA3 DNA methylation and expression, which we tested by treating DRG/SG cells with the DNA methylation inhibitor 5-aza-2′-deoxycytidine (AZA). AZA treatment resulted in a five-fold increase in GATA3 RNA expression, as well as increased expression of GATA3 protein (Fig. 3C,D), implying that DNA methylation plays a mechanistic role in regulating GATA3 expression in  Table S1. NB data: "Survival" shows genes whose increased expression is significantly associated with reduced relapse-free survival in neuroblastoma (p < 0.05, log rank test); data generated in R2 using GSE16476. "Expression" shows genes whose RNA expression is increased in both neuroblastoma cell lines (GSE28019) and neuroblastoma tumours (GSE16476) compared to neural crest cells (GSE14340); comparison was made using the "Megasampler" function in R2 Genomics Analysis and Visualization Platform (http://r2.amc.nl). See Table S1 for full results. www.nature.com/scientificreports www.nature.com/scientificreports/ sympathetic nervous system-derived cells. We then proceeded to investigate GATA3 RNA and protein expression in neuroblastoma tumour tissues from different stages, to search for any relationship between in vivo biological properties of tumours and GATA3 expression.
GATA3 RNA and protein expression in neuroblastoma. GATA3 sense RNA expression was significantly elevated in neuroblastoma cell lines and tumours compared to normal tissues (FA, hNCC and DRG/SG; Fig. 4A,B). These results were replicated in publicly available gene expression data ( Supplementary Fig. S3A), which also showed that high GATA3 RNA expression was associated with poor overall survival in neuroblastoma ( Supplementary Fig. S3B).
Almost all neuroblastoma tumours expressed detectable GATA3 protein, of which 61% expressed higher levels than FA (Fig. 4C,D). In those tumours for which survival data were available, relapse-free survival and overall survival were both worse in patients whose tumours expressed more GATA3 protein than FA (Fig. 4E,F).
These results suggested that expression levels of GATA3 were correlated with the clinical properties of neuroblastoma tumours, so we examined GATA3 effects on cell proliferation and death, which could provide an explanation for these correlations.
Effects of GATA3 silencing in neuroblastoma cell lines. IMR32, Kelly and NGP neuroblastoma cell lines, all of which expressed high levels of GATA3 sense RNA and GATA3 protein (Fig. 3A,B), were transfected with siRNAs to knock-down GATA3 expression, and then cell proliferation and cell death were assessed by a variety of assays (Fig. 5, Supplementary Fig. S4). In all cell lines, the GATA3 siRNAs were effective in down regulating both GATA3 RNA expression and GATA3 protein expression (Fig. 5A,B, Supplementary Fig. S4A,B,F,G). GATA3 knock-down inhibited cell proliferation (Fig. 5C, Supplementary Fig. S4C,H) and increased the number of dead cells in the cultures (Fig. 5D, Supplementary Fig. S4D,I). Cleaved PARP (c-PARP) protein levels increased in GATA3 siRNA-treated cells (Fig. 5A, Supplementary Fig. S4A,F), and this increase was abrogated by treating cells with the caspase inhibitor quinolyl-valyl-O-methylaspartyl-[2,6-difluorophenoxy]-methyl ketone (QVD) (Fig. 5E, Supplementary Fig. S4E,J). QVD treatment also abrogated the increased dead cell counts after GATA3 knock-down (Fig. 5D, Supplementary Fig. S4D,I), which together with PARP cleavage, showed that the cell death was caused by caspase-mediated apoptosis.
To further investigate the possible pathways by which GATA3 influences neuroblastoma proliferation, we examined cyclin D1 (CCND1), one of the known targets of GATA3 that is involved in cell cycle regulation 40 . Flow cytometry demonstrated that in almost all GATA3 siRNA knock-downs, there was a reduced proportion of cells in G2-M, consistent with disrupted cell cycle control ( Supplementary Fig. S5). Knock-down of GATA3 in neuroblastoma cell lines reduced CCND1 protein and RNA expression substantially (Fig. 5A,B, Supplementary  Fig. S4A,B,F,G). In neuroblastoma tumours, most showed increased expression of CCND1 compared to FA, where it was undetectable ( Supplementary Fig. S6A) and there was a good correlation between GATA3 protein expression and CCND1 protein expression ( Supplementary Fig. S6B). This suggested that one of major mechanisms by which GATA3 affects neuroblastoma cell proliferation may be via modulating CCND1 levels.
To assess whether GATA3 knock-down might be affecting cellular differentiation, we investigated the expression of a range of differentiation markers in the siRNA-transfected neuroblastoma cell lines ( Supplementary  Fig. S7). We assayed markers of both mesenchymal differentiation (VIM, PRRX1 and SNAI1) and adrenergic differentiation (NCAM, NF68), but observed no reproducible changes in gene expression, suggesting that GATA3 knock-down only affects proliferation and death, but not differentiation, in neuroblastoma cells.

Discussion
In this paper we have examined genes that were hypomethylated in neuroblastoma, as identified by our previous genome-wide analysis of DNA methylation, comparing hNCC to neuroblastoma cell lines 24 . We have now shown that GATA3, an important regulator of sympathetic nervous system development [36][37][38] , is hypomethylated in neuroblastoma cell lines and tumour tissues (Figs. 2B,C, 3A and Supplementary Fig. S1). Hypomethylation of GATA3 correlated with upregulated expression of both GATA3 RNA and GATA3 protein (Figs. 3A,B, 4A-D and Supplementary Fig. S3).
These results suggest that the development of neuroblastoma from neural crest derivatives involves a change in the epigenetic state of GATA3, that is controlled at least partially by DNA methylation, which is known to be an important factor in neural crest development 41 . Interestingly, we found a maximum of 22% GATA3 DNA methylation in the normal tissues that we examined (Figs. 2C and 3A), suggesting that only a subpopulation of these cells have DNA methylated at GATA3, which could reflect an intrinsic heterogeneity in the neural crest population 42 .
The inverse correlation between GATA3 DNA methylation and RNA expression ( Fig. 3A and Supplementary  Fig. S2A,B), implied a functional role for DNA methylation in the repression of GATA3 transcription, which was supported by the increase in GATA3 expression following AZA treatment of DRG/SG cells (Fig. 3C,D). The five-fold increase in GATA3 RNA expression was perhaps surprising, considering the relatively low levels of GATA3 DNA methylation in DRG/SG cells (Fig. 2C). This could suggest that there is a small subpopulation of methylated cells, which show greatly elevated GATA3 expression when they become demethylated, or bars). The genomic positions of assays 01 and 02 are shown in part A. (D,E) Kaplan-Meier survival curves (D, relapse-free survival; E, overall survival) taken from the dataset of NB patients in B and C for whom survival data were available. Me-, tumours with no GATA3 DNA methylation; Me + tumours with DNA methylation (using the average of pyrosequencing assays 01 and 02). p values from log-rank test. (2019) 9:18934 | https://doi.org/10.1038/s41598-019-55382-6 www.nature.com/scientificreports www.nature.com/scientificreports/ alternatively that the transcriptional regulation of GATA3 may be via another epigenetically regulated factor, rather than by a direct epigenetic effect on GATA3 itself.
Others have reported DNA methylation at the GATA3 CGI in breast cancer 43,44 bladder cancer 45 and some leukaemias 46 , with direct evidence that DNA methylation regulates GATA3 expression 46,47 . In our results, GATA3 methylation was confined to the start of the GATA3 antisense transcript ( Fig. 2A), although AZA treatment upregulated GATA3 sense RNA and GATA3 protein in methylated DRG/SG cells (Fig. 3C,D), suggesting a possible role for the antisense RNA in regulating GATA3 sense expression. We found a good correlation between GATA3 sense and antisense expression ( Supplementary Fig. S2C), in agreement with a previous report 48 . That report suggested that the proximity of the transcriptional start sites of the GATA3 sense and antisense RNAs could imply the presence of a bidirectional promoter 48 . However, it has recently been shown that the GATA3 antisense RNA can bring MLL to the GATA3 promoter, to increase permissive chromatin marks and thus directly upregulate GATA3 expression 49 . Thus, it is likely that demethylation of GATA3 in neuroblastoma has an indirect effect, by increasing GATA3 antisense expression, which in turn attracts positive epigenetic regulators to the GATA3 promoter region.
Our siRNA experiments demonstrated that knock-down of GATA3 expression reduced cell proliferation and increased apoptosis in neuroblastoma cell lines (Fig. 5, Supplementary Fig. S4). These results agree with previous studies that have shown an essential role for GATA3 in the proliferation of neuroblastoma cells in vitro and for tumorigenicity in vivo 50,51 . We found that knock-down of GATA3 decreased CCND1 expression in neuroblastoma cell lines (Fig. 5, Supplementary Fig. S4) and that in tumour samples, there was high expression of CCND1 compared to FA and a good correlation between GATA3 and CCND1 expression ( Supplementary Fig. S6), as found by others 40,52,53 . Thus, one of the likeliest mechanisms by which GATA3 affects cell proliferation in neuroblastoma, is via regulation of its transcriptional target cyclin CCND1 40,54 , which when knocked-down, shows similar effects on cell proliferation to GATA3 53 . Given that we observed no effects on cellular differentiation www.nature.com/scientificreports www.nature.com/scientificreports/ when GATA3 was knocked-down ( Supplementary Fig. S7), it seems likely that the major mechanism by which GATA3 affects neuroblastoma development may be via its effects on cellular proliferation and death ( Fig. 5 and Supplementary Fig. S4), rather than by affecting cellular differentiation.
Two recent papers have delineated super-enhancer-associated transcriptional circuits that divide neuroblastoma cells into two distinct differentiation states 55,56 ; corresponding to undifferentiated mesenchymal cells (MES), as found in hNCC, and a sympathetic adrenergic identity (ADRN), in which GATA3 plays a critical role as a master transcription factor 55,56 . Neuroblastomas contain mixtures of these two cell types, but with the majority being highly enriched in adrenergic cells, leading to high GATA3 expression, as we have observed (Fig. 4).
Our genome-wide DNA methylation study compared hNCC (mesenchymal differentiation) to neuroblastoma cell lines, which mostly have a predominately adrenergic state 55,56 . Hierarchical clustering of this data demonstrates the ability of DNA methylation to discriminate between cell lines representing these different lineages ( Supplementary Fig. S8A). Interestingly, this analysis showed that SK-N-AS clusters separately from the three other neuroblastoma cell lines, which agrees with the super-enhancer profiling, which places SK-N-AS as being intermediate between the ADRN and mesenchymal MES signatures 55,56 .
We also showed that adrenergic (ADRN) signature genes were more highly methylated in hNCC than were mesenchymal (MES) signature genes 56 (Supplementary Fig. S8B). Thus, distinct patterns of DNA methylation are associated with alternate differentiation states, suggesting that epigenetic modifications may play an important role in defining or stabilising cell identity in neuroblastoma.
Among neuroblastoma cell lines, the S-type cells GI-ME-N and SH-EP are in the mesenchymal differentiation pathway, like hNCC cell lines 55,56 . Interestingly, our results identified GI-ME-N and SH-EP as the only two neuroblastoma cell lines that had GATA3 DNA methylation, like hNCC, with corresponding low levels of expression of GATA3 protein (Fig. 3). Therefore, our results identify DNA methylation of GATA3 as a characteristic of the mesenchymal hNCC-like differentiation pathway, with loss of GATA3 methylation in neuroblastoma cell lines and tumours, which contain mainly adrenergic cells. However, we found that knock-down of GATA3 did not alter the differentiation status of the three neuroblastoma cell lines tested (Supplementary Fig. S7). This suggests that www.nature.com/scientificreports www.nature.com/scientificreports/ inactivation of just one of the transcription factors defining the adrenergic phenotype, is insufficient to drive neuroblastoma cells towards the mesenchymal phenotype, presumably due to the functional redundancy within the complex regulatory network. Our results therefore show that GATA3 is essential for the sustained proliferation of adrenergic neuroblastoma cells, but not solely responsible for defining their lineage.
The cellular composition of neuroblastomas does not appear to correspond to clinical outcome directly, although epigenetic plasticity may allow interconversion between the cell types, driving drug resistance 55,56 . Others have previously shown a high level of GATA3 expression in neuroblastoma 50,57 , and suggested that GATA3 is a useful distinguishing diagnostic marker for neuroblastoma 58 . Our results showed that a complete absence of GATA3 DNA methylation (Fig. 2D,E) and a high level of GATA3 protein expression (Fig. 4E,F), were both indicators of poor prognosis in neuroblastoma. Thus, although the adrenergic composition of neuroblastomas may not in itself be sufficient to determine the clinical phenotype of neuroblastomas, we suggest that epigenetic defects in GATA3 may deregulate the complex transcriptional networks controlling neuroblastoma and thus influence clinical outcomes. Therefore, the GATA3 transcriptional network is a promising target for novel neuroblastoma therapies.

Materials and Methods
Cell culture and neuroblastoma tumour samples. Cell lines (Supplementary Table S3) were obtained from ECACC, apart from BCH-N-DW, which is a novel neuroblastoma cell line derived from a bone marrow biopsy (kind gift from Dr C. McConville, unpublished data). Cell lines were cultured in DMEM/F12-HAM medium (Sigma) supplemented with 10% fetal bovine serum, 100U/ml penicillin, 0.1 mg/ml streptomycin, 2mM L-glutamine and 1% non-essential amino acids (Sigma) at 37 °C in a 5% CO 2 incubator.
Human neural crest cell lines (hNCC) were cultured as described previously 59,60 . Dorsal root ganglia/sympathetic ganglia cell lines (DRG/SG) were derived from cells growing away from dorsal root and sympathetic ganglia with some proximal nerve fascicles still attached, explanted from normal 8.5 to 9.5 gestational week human embryos (French Agence de la Biomédecine authorization PSF14-011). Within 18 hours of being explanted, the remaining pieces of ganglionic tissue were removed manually, and the cells passed into a new culture dish and cultured thereafter in the same medium as used for human neural crest cells (hNCC) 60 (Etchevers, unpublished). DRG/SG lines expressed sympathetic adrenergic markers DBH and MYCN to a higher level than lines derived from dorsal root ganglia alone (DRG; Supplementary Fig. S9).
Neuroblastoma tumour samples (Supplementary Table S4) were obtained from Bristol Children's hospital with informed consent (from parent and/or legal guardian for children less than 18 years of age) and with appropriate ethical approval (E5797; South West -Central Bristol Research Ethics Committee (UK)). All methods were performed in accordance with the relevant guidelines and regulations, including those specified in the UK Human Tissue Act 2004.
Cell counts. For cell viability and determining concentration, Trypan Blue stain (0.4%; Sigma) was mixed with cells in a ratio 1/1 before being counted by a Countess cell counter (Invitrogen). Dead cells were assayed by counting the number floating (unattached) cells that were trypan blue permeable.
Flow cytometry and cell cycle analysis. Cells were removed from flasks with trypsin, pelleted by centrifugation and fixed in 70% ethanol at −20 °C for 24 hours. They were then treated with RNAase (Qiagen) and stained with 50 μg/ml Propidium iodide (Sigma) and incubated at 37 °C for 15 min in the dark. Stained cells were analysed using LSR Fortessa X20 (BD Biosciences) using FACS DIVA8 software (Becton-Dickinson Immunocytometry Systems) and at least 10,000 events were collected. Cells cycle analysis was performed using Flow Jo software V10.  Table S5; all synthesised by Sigma), using Lipofectamine 2000 transfection reagent (Invitrogen) and harvested after 72 hrs.
RNA extraction, cDNA synthesis and RT-PCR. Total RNA was extracted with a RNeasy kit (Qiagen), DNase treated with TURBO DNA-free (Ambion) and cDNA synthesized using the Thermoscript RT-PCR system (Invitrogen). gene-specific primers (Supplementary Table S6) were used for QPCR (QuantiTect SYBR Green;