MYCN overexpression inhibits RA-induced neuronal differentiation
SY5Y neuroblastoma cells treated with RA undergo neuronal differentiation to become dopaminergic neurons [45, 48–51]. We profiled global transcriptional changes mediated by RA in the MYCN Dox-inducible SY5Y-MYCN cell line, which was previously generated from the parental SY5Y cell line by the Westermann lab [42–44]. To assess the effect of MYCN overexpression on neuronal differentiation we imaged SY5Y-MYCN cells treated with RA while overexpression of the MYCN transgene was either induced or un-induced (Fig. 1a). A differentiation ratio for each treatment group was then calculated by dividing the length of the longest axon of a cell by the cell’s width. Like SY5Y cells, SY5Y-MYCN cells underwent RA-mediated differentiation in the absence of MYCN induction. However, when MYCN expression was induced (reaching 10–15 times higher levels than in un-induced cells; Additional file 1: Figure S1a) the ability of RA to efficiently differentiate these cells strongly and significantly was attenuated (t-test, RA versus RA and Dox p < 0.0001). While endogenous MYCN mRNA (parental SY5Y cells) expression was downregulated by RA treatment, ectopic MYCN in SY5Y-MYCN cell lines was not reduced as it is not under the control of the endogenous MYCN promoter (Additional file 1: Figure S1b; also see Duffy et al. [45]). Confirming that the RA was active, it reduced the expression of endogenous c-MYC mRNA by a similar extent in both SY5Y and un-induced SY5Y-MYCN cell lines (Additional file 1: Figure S1b).
MYCN overexpression antagonises the normal transcriptional response to RA treatment
The mechanisms through which MYCN blocks RA-mediated neuronal differentiation are highly relevant to MYCN-amplified neuroblastoma patients, who generally do not respond well to retinoid treatment [20, 27]. Therefore, to identify these mechanisms we conducted mRNA-seq of SY5Y-MYCN cells under four treatment conditions: (i) 24-h DMSO (control), (ii) 24-h RA, (iii) 24-h RA and 48-h Dox, and (iv) 48-h Dox. Firstly, we confirmed that a number of the genes encoding the RA receptors, which are required to facilitate cellular responsiveness to RA, were expressed in SY5Y-MYCN cells (Additional file 1: Figure S1c). Of these the expression of both RARA and RARB was upregulated upon RA treatment. In total, between 511 and 839 differentially expressed (DE) genes were detected per treatment group, with a high degree of overlap between the co-treatment (RA and Dox) sample and the individual treatments (Fig. 1b; Additional file 2: Table S1; Additional file 3: Table S2; Additional file 4: Table S3; Additional file 5: Table S4). MYCN overexpression predominantly downregulated gene expression as we have previously described [42], while RA treatment produced a very similar number of up- and downregulated DE genes (Fig. 1c). While there was a trend for the greatest fold changes to occur in genes which had lower pre-treatment expression states, genes across the full range of pre-treatment expression levels were differentially expressed (Fig. 1d).
Of the 169 genes regulated in common between MYCN overexpression and RA treatment, 95 were regulated in opposing directions (up- or downregulated) by each treatment (Fig. 2a). These differentially activated genes are likely key to MYCN’s ability to block RA-mediated neuronal differentiation and contain both known and novel components of differentiation signalling (see below). To validate the accuracy of the RNA-seq analysis we analysed the changes in expression (by qPCR) of MYCN (Additional file 1: Figure S1a) and seven selected genes, RET, DKK1, EGR1, FZD7, ASCL1, LMO4 and c-MYC (Fig. 2b), which were identified as being DE in the RNA-seq data. The results confirmed the reliability of the RNA sequencing data (Fig. 2b; Additional file 1: Figure S1a). The qPCR also confirmed the differing direction of regulation for RET, FZD7, EGR1, ASCL1 and LMO4 between the RA treatments and MYCN induction (Fig. 2b). To eliminate any Dox-related, non-MYCN-dependent, role in the expression changes of these genes we treated parental SY5Y cells with Dox. Dox treatment in SY5Y cells did not reproduce the expression changes observed when MYCN was overexpressed in SY5Y-MYCN cells via Dox treatment (Additional file 1: Figure S1d).
We also confirmed by qPCR and western blotting the strong induction of the BDNF receptor NTRK2 (TrkB) upon RA treatment, which was revealed in the RNA-seq results (Additional file 1: Figure S1e, f). This RA-mediated induction of NTRK2 was sustained, and indeed continued to rise over longer RA treatments (Additional file 1: Figure S1e, f). While high NTRK2 expression combined with amplified MYCN is a marker for high-risk neuroblastoma [52], NTRK2 pathway activation by BDNF ligand treatment is also known to aid RA-mediated differentiation [53]. Our results indicate that RA dramatically upregulates NTRK2 expression, potentially priming the cells to respond to BDNF signalling.
Transcriptome-wide profiling reveals novel regulatory mechanisms of known differentiation-associated genes
Of the genes regulated in opposing directions by RA and MYCN (Fig. 2a), we examined three in more detail: CYP26A1, LMO4 and ASCL1 (Fig. 2c). These genes were selected as they have previously been associated with either neuronal differentiation, MYCN or neuroblastoma, but our analysis reveals their opposing transcriptional regulation by RA and MYCN. Our RNA-seq analysis revealed that the expression of the CYP26A1 gene was massively increased upon RA treatment, jumping from almost undetectable to highly expressed (0.05–36.59 CPMkb; Fig. 2c). This increase was further enhanced by the combination of RA and Dox, despite Dox alone slightly reducing CYP26A1 expression (Fig. 2c). CYP26A1 is a member of the cytochrome P450 family and is involved in a negative feedback loop, where RA activates its expression while the CYP26A1 protein inactivates RA by hydroxylation [54–58]. CYP26A1 also regulates the production of migratory cranial neural crest cells [59]. Our data show a trend for MYCN overexpression to enhance the RA-induced expression of the RA inhibitor CYP26A1 (Fig. 2c).
LMO4 is a transcriptional regulator involved in the epithelial-to-mesenchymal transition of neuroblastoma and neural crest cells [60]. It can also inhibit differentiation of mammary epithelial cells and is overexpressed in breast cancer [61]. Its paralogue, LMO1, is a neuroblastoma oncogene which is duplicated in 12.4% of tumours, and is associated with aggressive disease [62]. LMO4 interferes with neuritogenesis in SY5Y cells [63], has a role in the differentiation of progenitor cells of motor neurons and the cranial neural crest and is highly expressed in proliferating mouse epithelial tissues [64, 65]. Our results reveal that LMO4 mRNA levels are upregulated by MYCN and downregulated by RA, while in the combination treatment MYCN overexpression partially reverses RA’s inhibitory effects on LMO4 expression (Fig. 2c).
The ASCL1 transcription factor stimulates neuronal differentiation, but its pro-differentiation functions are blocked by MYCN at the protein level, where MYCN maintains the phosphorylation of ASCL1 [9]. ASCL and MYCN also share some of the same promoter targets, but direct opposing regulation of these shared targets [66]. In addition to MYCN’s role in regulating phosphorylation of the ASCL1 protein, our data revealed that MYCN overexpression regulates ASCL1 mRNA levels (Fig. 2c). MYCN overexpression increased the level of ASCL1 mRNA, while RA treatment strongly reduced it (Fig. 2c). Combination treatment partially rescued the effect of RA on ASCL1. Therefore, ASCL1 is another gene differentially regulated by RA and MYCN overexpression, which is likely to contribute to MYCN’s ability to block neuronal differentiation.
In order to determine if the results obtained from the cell line were relevant to neuroblastoma tumour biology, we examined the effect of these genes on neuroblastoma patient survival in three large neuroblastoma tumour datasets (Versteeg [67], SEQC [68] and Kocak [69], with 88, 498 and 649 tumours, respectively), using the R2: Genomics Analysis and Visualization Platform (http://r2.amc.nl). CYP26A1, LMO4 and ASCL1 mRNA expression levels were each prognostic of patient outcome (Fig. 2d; Additional file 1: Figure S1g), consistent across the three datasets. Furthermore, the elevated expression of ASCL1 and LMO4 seen in the MYCN-overexpressing SY5Y-MYCN cells matched the high expression of these genes in the poor outcome tumours. Conversely, tumours with low ASCL1 and LMO4 expression had better prognosis, matching the cell line results in which these genes were downregulated by RA. The correlation between the RNA-seq and the tumour data was not as straightforward for CYP26A1. Expression of CYP26A1 was induced by RA and augmented further by MYCN induction, but not by MYCN induction alone (Fig. 2c). High levels of CYP26A1 were also indicative of poor outcome. Thus, while not activated by MYCN alone, CYP26A1 expression is induced by RA treatment even more strongly in the presence of elevated MYCN, and subsequently inactivates RA, resulting in retinoid resistance.
To move beyond the single-gene level and identify additional mechanisms through which MYCN overexpression can interfere with RA signalling we performed global pathway and network-based analysis of the RNA-seq data.
Global analysis of mRNA-seq results reveals MYCN and RA differentially activate a range of transcriptional regulators
We analysed the RNA-seq data using the GO disease and function terms tool of the IPA programme. Using existing knowledge, GO term analysis identifies patterns of gene regulation in the transcriptomic data which match patterns related to biological events such as apoptosis, ribosome biogenesis, proliferation and DNA replication. GO term analysis confirmed the phenotypic observations, showing that RA-activated genes are involved in neuronal differentiation processes and RA-inhibited genes are involved in cell movement (Fig. 3a). Conversely, MYCN overexpression repressed differentiation-associated processes, while combination treatment tended to fall between the two extremes but still with a bias towards the repression of neuronal differentiation (Fig. 3a). Disease and function GO analysis of the top 15 GO terms per condition revealed that RA inhibited proliferation and cancer-associated processes (Additional file 1: Figure S2a).
IPA analysis also showed that these phenotypic changes were achieved by RA differentially regulating the components of a number of signalling pathways associated with neuronal differentiation, including RAR and VDR/RXR, which were in the top 15 signalling pathways altered during differentiation (Fig. 3b). In particular, the expression of components of the RAR pathway itself were regulated by RA at all levels when projected onto a RAR pathway map (Fig. 3c). Aside from known RA-associated pathways, our analysis highlighted that a large array of signalling pathways participate in the differentiation of neurons, including axonal guidance, protein kinase A, eNOS and G-protein coupled receptor signalling (Fig. 3b).
The IPA suite was next used to identify the ITRs of DE genes of each treatment condition. Given the wealth of transcriptomic experiments publically available, a vast database exists regarding how genes are transcriptionally regulated in response to a wide array of regulators (genes, proteins or chemical compounds). ITR analysis harnesses this prior knowledge to identify patterns of transcriptional regulation in our datasets which match the patterns produced by known regulators. This comparison of known patterns versus patterns observed in the data enables the inference of which regulators are likely responsible for the differential gene expression seen in our transcriptomics data. Given this prior knowledge, ITR analysis can not only infer the regulators likely altering transcriptional regulation but also predict their activation status, i.e. whether these regulators were activated or inhibited in the treatment groups compared to the control cells [42, 70]. For more information on the statistical algorithms employed to match the detected changes in gene expression to known gene regulatory modules from the curated IPA knowledge database see Krämer et al. [70]. RA itself was a top ITR in each of the treatment groups (Fig. 4a, b). The analysis correctly and independently predicted it to be activated in both of the conditions in which RA treatment was performed (24-h RA and 24-h RA and 48-h Dox), providing a positive validation of the ITR analysis. As previously shown [42], RA as an ITR was inhibited by 48-h Dox treatment (Fig. 4a, b), revealing that MYCN DE regulates known RA target genes. Indeed, consistent with MYCN repressing RA’s effects on its target genes, the activation z score of RA for the combination treatment (24-h RA and 48-h Dox) was lower than for the 24-h RA single treatment, despite both conditions receiving the same dose and duration of RA treatment (Fig. 4a, b).
A clear trend emerged from the top ITRs of each condition: RA primarily activated transcriptional regulators (14/15) while MYCN primarily repressed them (12/15) (Fig. 4b). This trend was also clear across the top 100 ITRs (Additional file 1: Figure S3a; Additional file 6: Table S5). In line with this trend and the mutual antagonism of RA and MYCN, the combination treatment showed an almost equal number of ITRs to be activated and repressed, (seven and eight, respectively). The top protein ITRs of RA treatment formed a highly interconnected network, revealing the complexity of the molecular mechanisms involved in RA-mediated neuronal differentiation (Fig. 4c). Interestingly, almost half of the 24-h RA ITRs were chemical compounds, suggesting that additional drugs, if co-administered, may be able to improve the differentiation efficiency of clinical RA treatments.
Antagonistic regulation of transcriptional regulators by RA and MYCN
To identify the transcriptional regulators through which MYCN exerts its inhibition of RA treatment, we next examined transcriptional regulators which were differentially activated between the treatment groups. A number of the top 15 ITRs were differentially regulated between the RA and MYCN overexpressing conditions, such as TGFB1, HIF1A, APP and FGF2 (Fig. 4b). To identify all ITRs which were differentially activated between RA and Dox treatments we overlapped the ITRs and their activation/inhibition status (Additional file 1: Figure S4a). Then, we generated protein interaction maps to reveal the transcriptional regulator networks which are likely to mediate MYCN inhibition of neuronal differentiation (Additional file 1: Figure S4a). The RA-inhibited and MYCN-activated protein ITRs were enriched for β-catenin binding genes (molecular function GO analysis, p = 8.52E-07) and Wnt signalling-related genes (KEGG pathway enrichment analysis, p = 4.84E-02), with all of the Wnt-related proteins present being antagonists of the pathway. Conversely, the protein–protein interaction network for the ITRs activated by RA and inhibited by MYCN overexpression were enriched for MAPK pathway-related proteins (KEGG pathway enrichment analysis, p = 6.829E-17). This network also included the WNT1 ITR, which combined with the results from the first network suggests that RA activates WNT1 signalling and represses Wnt antagonists. MYCN has the inverse effect, inhibiting WNT1 and activating Wnt antagonists. We have recently independently shown that Wnt and MAPK signalling are involved in regulating differentiation in MYCN-amplified neuroblastoma cells [42, 71]. In SY5Y cells, which are MYCN single-copy, the Wnt-RA cross-talk has been shown to be mediated by PSEN1 (Presenilin 1) [72]. We previously discovered novel cross-talk between the MYCN oncogene and GSK3 (one of the Wnt-related ITRs) [45], β-catenin [71] and MAPK [42]. Therefore, the protein–protein interaction network of regulators differentially activated by RA and MYCN overexpression identified here confirm our previous findings, support the validity of the current analysis and reveal that MYCN’s ability to inhibit RA-mediated differentiation involves the regulation of Wnt and MAPK signalling components.
To identify novel ITRs which may enhance the clinical response to RA when given as combination therapies, we next collated ITRs which were strongly differentially regulated between the three conditions (Fig. 4d). One of these regulators was MYCN itself. The effects of MYCN overexpression on known MYCN target genes predominated over RA effects, with MYCN overexpression strongly activating MYCN target gene expression, an effect which was only mildly attenuated by RA co-treatment (Fig. 4d). It should be noted, however, that this was in a MYCN-inducible system where MYCN expression was not under the control of its endogenous promoter and enhancers. Although artificial, this scenario mirrors highly amplified MYCN neuroblastoma where over 70 additional copies of the MYCN gene can be inserted in the tumour’s genome, often losing their endogenous promoters and enhancers. Interestingly, the strongly differentially regulated ITRs were significantly enriched for the Neurotrophin signalling pathway (BDNF, NGF and Trk receptors etc., see 'MYCN overexpression antagonises the normal transcriptional response to RA treatment' section above; KEGG p = 5.739E-10; Additional file 1: Figure S4b), which is strongly associated with neuronal differentiation and neuroblastoma outcome. These data suggest a convergence of the molecular mode of action of RA and neurotrophin (NGF/BDNF) mediated differentiation. The most highly connected nodes in the network included histone deacetylases (HDACs), which have recently been shown to synergise with RA treatment [8, 73–75], and TGFB1. TGFB1 is a ligand of the transforming growth factor beta (TGF-β) signalling pathway, with known roles in modulating differentiation [76–79]. The TGFB1 ITR was strongly differentially activated between RA and MYN overexpression conditions; RA activated TGFB1 while MYCN overexpression strongly inhibited it (Fig. 4d). The effect of MYCN was more dominant and TGFB1 activity was also inhibited in the co-treatment (Fig. 4d). We therefore further assessed the possibilities for TGFB1-informed therapy to enhance the response of MYCN-amplified cells to retinoid therapy.
MYCN regulates TGFB1 and its target genes
To compare the effect of MYCN overexpression and RA on TGFB1 with the effects of amplified MYCN, we next examined sequencing datasets (RNA-seq and MYCN ChIP-seq) which we had previously generated (ArrayExpress (http://www.ebi.ac.uk/arrayexpress/) accession numbers E-MTAB-2690, E-MTAB-2691, E-MTAB-1684, E-MTAB-4100 and E-MTAB-2689) [42, 45, 71, 80]). These datasets revealed that TGFB1 was also a top regulator of the differences in the MYCN regulator and effector networks between single copy MYCN and MYCN-amplified neuroblastoma cell lines [42] (Fig. 5a). Mirroring the effect of MYCN overexpression in SY5Y-MYCN cells, the TGFB1 ITR was strongly inhibited in all MYCN-amplified cell lines compared with MYCN single copy cells. Our publicly available datasets also revealed that, similar to RA, induction of apoptosis by LiCl treatment (GSK3 inhibition) activated the TGFB1 ITR (Fig. 5a). Therefore, both cellular phenotypes associated with good outcome, i.e. differentiation and apoptosis, activated TGFB1 signalling.
We next examined MYCN ChIP-seq data [42] to determine if the inhibitory effect of MYCN on TGFB1 functioning (IPA ITR analysis) was achieved by MYCN binding to TGFB1 targets or binding to components of the TGF-β signalling pathway. Both overexpressed and amplified MYCN bound to the genes of a large number of TGFB1 targets (IPA ITR analysis), with the number of genes bound increasing with higher MYCN expression levels (Additional file 1: Figure S5a). Similarly, a proportion of the genes bound by MYCN have known SMAD regulatory elements (Additional file 1: Figure S5b), as revealed by DiRE analysis [81]. SMADs are the direct and prime transcriptional effectors of TGF-β signalling [82, 83]. In addition, genes bound by MYCN in KCNR cells were enriched for the TGF-β signalling pathway, as revealed by IPA (pathway analysis, p-value of overlap = 1.56E-02, ratio of overlap = 0.218), with 19 genes that encode components of the pathway being genomically bound by MYCN (Additional file 1: Figure S5c). TGF-β signalling pathway components were also bound by MYCN in KCN (17 genes) and 48-h Dox-induced SY5Y-MYCN (15 genes).
These results identified TGFB1 as a key regulator of RA-mediated differentiation, which is differentially activated in a MYCN context-dependent manner, being suppressed in MYCN-overexpressing and amplified cell lines. Therefore, we next examined if RA treatment and MYCN overexpression could modulate TGFB1 gene expression. MYCN overexpression modestly but significantly reduced TGFB1 mRNA levels compared to RA-only treated cells (t-tests, MYCN versus RA p value = 0.0076, RA versus co-treatment [MYCN and RA] p value = 0.0010; Fig. 5b). In line with the MYCN overexpression results, three of the four MYCN-amplified cell lines had lower TGFB1 mRNA expression than SY5Y cells which are MYCN single-copy, with expression of TGFB1 in KCNR cells being almost absent (Fig. 5c). The inhibitory effect of MYCN upon TGFB1 mRNA expression was not rescued by RA treatment (Fig. 5b), suggesting a novel mechanism through which MYCN can inhibit RA-mediated neuronal differentiation.
While RA did not rescue the effect of MYCN overexpression on TGFB1 mRNA levels, it partially rescued TGFB1 signalling, as revealed by the ITR analysis (Fig. 5a). Additionally, the inhibitory effects of MYCN on TGFB1 functioning in the other cell lines (Fig. 5a) did not always correlate directly to TGFB1 mRNA expression levels (Fig. 5c), suggesting further levels of cross-talk. Therefore, to further probe the functional relationship between TFGB1 and MYCN we next examined MYCN’s protein–protein interactions.
MYCN protein interacts with TGF-β signalling-associated proteins
While RA slightly increased TGFB1 mRNA levels and MYCN overexpression reduced them, these changes do not fully account for the differences in TGFB1 activation revealed by the ITR analysis. Therefore, to investigate additional potential cross-talk between MYCN and TGF-β signalling we performed MYCN interaction proteomics using the same experimental conditions as for the RNA-seq experiments: (i) 24-h DMSO (control); (ii) 24-h RA; (iii) 24-h RA and 48-h Dox; and (iv) 48-h Dox. We performed IPA ITR analysis on the proteins bound by MYCN in all conditions, which revealed that 32 TGFB1-regulated proteins had protein–protein interactions with MYCN (Fig. 5d). Of these, only MYCN’s interaction with HDAC2 was previously known (present in the String database). Fourteen of the 32 proteins were differentially bound to MYCN when RA-only treatment was compared with the RA and Dox co-treatment, with the majority of them binding less strongly to MYCN in the MYCN overexpressing sample (Fig. 5e). For 13 of the 14 differentially bound proteins, the change in MYCN binding was not as a result of altered transcriptional regulation (Additional file 1: Figure S5d); rather, altered binding was likely mediated by post-translational events. Interestingly, the one protein in which altered transcriptional regulation may have contributed to its differential binding to MYCN protein, at least partially, was the neuroblast differentiation-associated protein AHNAK. AHNAK was differentially expressed at the mRNA level, with opposing regulation by MYCN overexpression and RA treatment (Fig. 2a; Additional file 1: Figure S5d). AHNAK was bound to MYCN in all conditions, and has been described as a tumour suppressor that can stimulate the growth suppressing functions of the TGF-β pathway [84].
Taken together, our findings reveal cross-talk between MYCN and the TGF-β pathway at several levels, including the regulation of TGFB1 mRNA expression, the regulation of TGFB1 target genes and pathway components, and protein–protein interactions between MYCN- and TGFB1-regulated gene products. They also show that MYCN and RA drive opposing functional regulation of TGFB1.
The ability of MYCN overexpression to inhibit the normal RA-mediated activation of TGFB1 signalling, revealed here, further underscores the importance of TGF-β suppression in contributing to MYCN’s oncogenic potential. We therefore further investigated associations of TGF-β signalling genes and outcome in neuroblastoma tumour data and whether pharmaceutical modulation of the TGF-β pathway can enhance the effectiveness of RA treatment in MYCN-amplified cell lines.
Genes mediating TGF-β signalling activation in response to the small molecule kartogenin are prognostic of neuroblastoma patient outcome
Having identified a role for TGF-β signalling in MYCN’s blocking of RA-mediated neuronal differentiation, we next assessed whether this shared node of MYCN and RA signalling was amenable to therapeutic intervention. Kartogenin (KGN) is a recently developed small molecule which enhances the activation of TGF-β signalling through indirectly regulating the activity of the SMAD transcriptional effectors [85, 86]. KGN has been shown to promote chondrocyte differentiation in vitro and in vivo [85–87]. KGN also strongly upregulates the expression of TGFB1 itself [87]. KGN competitively binds filamin A (FLNA), thus inhibiting it from interacting with core-binding factor β subunit (CBFB) [85]. CBFB that is not bound by FLNA is then free to translocate from the cytoplasm to the nucleus, where it complexes with the RUNX-SMAD transcriptional machinery and regulates target gene expression (Fig. 6a). Interestingly, we showed that SMADs 1, 2 and 6 and RUNX2 were genomically bound by amplified MYCN (Additional file 1: Figure S5c) and there was a protein–protein interaction between MYCN and FLNA (Fig. 5d, e), revealing further levels of cross-talk between TGF-β signalling, KGN’s mode of action and oncogenic MYCN. The RUNX developmental regulators have previously been implicated in a variety of cancers [88]. Therefore, to examine whether the FLNA-CBFB-RUNX2-SMADs module is associated with neuroblastoma tumour biology and patient outcome, we examined the R2 neuroblastoma datasets (used above; http://r2.amc.nl) for these genes. Low CBFB and high FLNA (inhibitor of CBFB) mRNA expression was associated with poor patient outcomes (Fig. 6b; Additional file 1: Figure S6a, b). In line with the differential activation of the TGFB1 ITR by RA and MYCN, the SMAD transcriptional effectors of the pathway were also largely differentially activated between the two conditions, with the canonical SMADs being inhibited by MYCN overexpression (Fig. 6c). Interestingly, SMAD7, an antagonist of TGF-β signalling, was activated by MYCN and inhibited by RA, potentially revealing another mechanism through which MYCN inhibits RA-mediated activation of TGFB signalling. Accordingly, high SMAD7 mRNA expression was predictive of poor patient outcomes (Fig. 6d). Additionally, a number of the other SMAD genes were predictive of patient outcome, with high SMAD1 and SMAD2 expression associated with good outcomes (Additional file 1: Figure S6c), while RUNX3 showed a better survival prediction than RUNX2 (Fig. 6b; Additional file 1: Figure S6a–c).
KGN shows differentiating potential for neuroblastoma but does not overcome the amplified-MYCN-mediated differentiation block
Having demonstrated cross-talk between MYCN and the TGF-β signalling pathway, and having shown that pathway components targeted by KGN are associated with neuroblastoma patient outcome, we next treated MYCN-amplified IMR32 cells to determine their response to KGN individually and in combination with RA. IMR32 cells are less responsive to RA-mediated differentiation than MYCN single-copy cells. Low dose KGN (0.1–5 μM) was almost as effective at differentiating IMR32 cells as RA (Fig. 7a, b), with all treatment groups being significantly different to untreated controls (p < 0.0001 for all groups versus controls). It should be noted that MYCN-amplified IMR32 cells are far more resistant to differentiation than MYCN single-copy neuroblastoma cells (see Fig. 1a for comparison). Interestingly, when given in combination with RA, none of the low dose KGN (0.1–5 μM) treatments enhanced the differentiating potential over that seen in RA-only treatments (t-test, p = 0.226–0.982). High dose KGN (10 μM) had a significantly different differentiation ratio to untreated cells (t-test, p < 0.0001), but the extent of differentiation was less than that seen in lower dose KGN treatments (0.1–5 μM KGN). It was also able to partially inhibit the differentiating effects of RA in the combination treatment compared with RA-only treatment (t-test, p < 0.0001).
While both RA and KGN increased the differentiation ratio of IMR32 cells, the combination of these compounds did not act synergistically to overcome the differentiation block imposed by amplified MYCN (Fig. 7a, b). However, combination treatment of RA with a high dose of KGN (10 μM) resulted in a large proportion of cells with an apoptotic-like rounded appearance (Fig. 7a; Additional file 1: Figure S7a). These apoptotic-like cells were not included in the calculation of the differentiation ratios due to their complete lack of axons; only surviving cells were measured. In order to further assess the effect of KGN on neuroblastoma cells, we next assessed cell viability.
Combination treatment of KGN and RA specifically reduces viability of MYCN-amplified cells, but not MYCN single-copy cells
The viability of IMR32 cells was reduced in a dose-dependent manner by KGN (Fig. 7c), though the effect was relatively mild. Our analysis suggested that combination treatment with these two compounds should provide an additive or synergistic effect. Despite RA having no effect on cell viability, RA treatment sensitised IMR32 cells to KGN, with co-treated cells responding more strongly than to each compound individually (Fig. 7a, c). Combination treatment of 20 μM KGN and 1 μM RA strongly reduced cell viability (t-test, p = 1.0E-04; Fig. 7c). Combination treatment significantly reduced cell viability compared with 20 μM KGN-only treatment (t-test, p = 8.4E-03). MYCN-amplified KCNR cells also showed a strong reduction in cell viability upon combination treatment (Additional file 1: Figure S7b; t-test, p = 4.0E-03). Importantly, in line with the omics analysis, KGN-RA combination treatments only reduced the viability of MYCN-amplified cells, without significant effects on the viability of the MYCN single-copy cell line SY5Y (untreated versus 20 μM KGN and 1 μM RA; t-test, p = 0.4715; Fig. 7d). Combination treatment also significantly reduced the cell viability of NBL-S cells (Additional file 1: Figure S7c; untreated versus 20 μM KGN and 1 μM RA; t-test, p = 1.0E-04), which although being MYCN single copy have elevated levels of MYCN protein due to an increased MYCN protein half-life [89].
While KGN and RA did not cooperate to further differentiate IMR32 cells, they did cooperate to enhance the apoptotic response to treatment. As part of the RA neuronal differentiation programme, RA is known to block proliferation and promote apoptosis of normal neuronal precursors and low-risk neuroblastoma cells, in addition to inducing the differentiating of surviving cells [90–94]. Similarly, ITR analysis revealed that TGFB1 was strongly activated when IMR32 cells were induced to undergo apoptosis by LiCl treatment (Fig. 5a). To further functionally confirm that TGFB1 is involved in directing neuroblastoma cell fate in a MYCN-dependant manner, we mined an RNAi knockdown screen targeting the druggable genome in SY5Y-MYCN cells [42]. TGFB1 was a top ITR (ranked third) of the 674 genes that strongly reduced SY5Y-MYCN cell viability when knocked down (Additional file 1: Figure S7d), confirming the functional role of TGFB1 in neuroblastoma cell fate and supporting its likely therapeutic potential. RA was also a high ranking (12th) ITR of these viability-reducing genes (Additional file 1: Figure S7d).
While elevated MYCN levels can block the pro-apoptotic effects of RA (Additional file 1: Figure S7e), our results reveal that KGN can be used as a combination therapy to promote MYCN-amplified neuroblastoma cell death. Thus, RA and KGN combination treatments represent a novel therapeutic option with potential for targeting high-risk MYCN-amplified tumours.
Pharmacological inhibition of TGF-β signalling strongly attenuates RA-mediated neuronal differentiation
To confirm the functional role of the MYCN-induced inhibition of TGF-β signalling in promoting retinoid resistance we next investigated whether pharmacological inhibition of TGF-β signalling could attenuate RA-mediated differentiation in the absence of MYCN overexpression. We treated SY5Y-MYCN cells with RepSox, a potent and selective inhibitor of TGF-β receptor 1 (TGFBR1) [95]. RepSox can successfully replace Sox2 in reprogramming cells and its use alongside other reprogramming factors is efficient at generating induced pluripotent stem cells [96]. RepSox did not reduce the viability of SY5Y-MYCN cells (Fig. 7e; DMSO-control versus 100 nM RepSox; t-test, p value = 0.2244). When un-induced SY5Y-MYCN cells were treated with RA and RepSox, RepSox blocked the differentiating effect of RA so strongly (1 μM RA-only versus 25 nM RepSox and 1 μM RA; t-test, p value <0.0001.0E-04) as to maintain the differentiation ratio near the same level as that seen in control cells (Fig. 7f). Taken together, our results reveal that TGF-β signalling inhibition, whether achieved by MYCN overexpression or pharmacological treatment, strongly contributes to resistance to retinoid-mediated differentiation in neuroblastoma cells and that pharmacological activation of TGF-β signalling represents a promising strategy to sensitising MYCN-amplified cells to retinoid-mediated apoptosis.
