Mastermind-Like 3 Controls Proliferation and Differentiation in Neuroblastoma

This article is featured in Highlights of This Issue, p. 409

Neuroblastoma is the most common extracranial solid pediatric cancer and is most prevalent in early childhood (1). It affects the peripheral nervous system and often develops in the adrenal glands but is also found in other peripheral nerve tissues. It is a heterogeneous disease with diverse clinical manifestations, ranging from spontaneously regressing tumors to very aggressive tumors with poor prognosis. Several genetic aberrations have been linked to neuroblastoma, most notably MYCN amplification in 20% of neuroblastomas (2), ALK mutations (6%–10%; refs. 3, 4), as well as loss of the 1p36 and 11q23 chromosomal regions (5, 6). Treatment options for neuroblastoma patients include surgery, radiation, bone marrow transplantation, chemotherapy, and retinoic acid (RA) for high-risk patients, but options to treat late-stage metastatic disease are very limited and urgently need improvement.

The vitamin A metabolite RA plays an important role during embryonic development, including development of the nervous system (7). The physiologic functions of RA are exerted through the nuclear receptors RA receptor (RAR) and retinoid X receptor (RXR), which form a heterodimer that binds to DNA elements known as RA response elements (RARE; ref. 8). In the absence of RA, RAR/RXR is in complex with corepressors like NCoR, SMRT, and HDACs that repress the transcription of RA target genes (9,10). Binding of RA to RAR results in a conformational change of RAR, which leads to an exchange of repressor proteins for transcriptional coactivator proteins like CBP/p300, PCAF, CTBP2, and p160, enabling the transcription of RA target genes (11–13).

The observation that retinoids drive neuronal differentiation in neuroblastoma cell lines has led to the incorporation of 13-cis-RA in the treatment regimen for high risk neuroblastoma (14). However, treatment of high-risk neuroblastoma patients with 13-cis-RA after intensive chemotherapy showed only a modest increase in event-free survival (15, 16). This is caused in part by the fact that both intrinsic and acquired resistance to retinoid therapy are observed frequently in neuroblastoma patients. Understanding mechanisms of resistance to retinoids is crucial to improve treatment with this class of drugs and to attain a long-lasting effect. Several mechanisms of resistance to RA have been described to date, including mutations in the ligand-binding domain of PML-RARα in APL patients (17) and the epigenetic silencing of the RARβ receptor (18, 19), which leads to the inability of RA to activate the pathway and hence to RA resistance. CTBP2, generally a corepressor protein, was identified as a positive cofactor of RAR-RXR in mouse F9 cells, and its suppression conferred resistance to RA (13). Using loss-of-function genetic screens Huang and colleagues showed that knockdown of the zinc finger protein ZNF423 leads to the loss of activation of RA target genes. ZNF423 binds to the RAR/RXR complex and acts as a transcriptional coactivator (20). In another study, loss of the tumor suppressor NF1 was shown to determine responsiveness to RA. Loss of NF1 activates RAS signaling, which in turn represses ZNF423 expression, leading to RA resistance. Low expression levels of ZNF423 and/or NF1 in neuroblastoma patients predict poor outcome (21).

In this study, we use the validation-based insertional mutagenesis (VBIM) system (22) to perform a gain-of-function genetic screen to identify determinants of RA responsiveness in the SK-N-SH neuroblastoma cell line. We identify here an unexpected link between the coactivator mastermind-like 3 (MAML3) and RA signaling, which had hitherto only been implicated in NOTCH signaling.

All-trans retinoic acid (R2625) was purchased from Sigma. OSI-906 (S1091) was purchased from Selleckchem. Phosphatase inhibitor cocktail 2 (P-5726), and phosphatase inhibitor cocktail 3 (P0044) were purchased from Sigma-Aldrich. Dynabeads Protein A (10002D) was purchased from Life Technologies. MAML1 (A300-673A), MAML2 (A300-682A), and MAML3 (A300-684A) antibodies were purchased from Bethyl Laboratories. AKT (2920), pAKT (4060), IGF1R (3027), and pIGF1R (3024) antibodies were purchased from Cell Signaling Technology. HSP90 (sc-7947), SNAP25 (sc-7539), GAP43 (sc-7457), and normal rabbit IgG (sc-2027) antibodies were purchased from Santa Cruz Biotechnology. V5 (R960-25) antibody was purchased from Invitrogen.

pLX-GFP, pLX-MAML3, and pLX-MAML2 vectors were part of the Broad Institute ORF library (Sigma). pBabe-puro (pBp) and pHAGE-MAML1 vectors were purchased from Addgene.

MAML3-ΔExon1 DNA was cloned into pBp vector by PCR-amplifying MAML3 with primers spanning from exon 2 to exon 5 (forward: GCAGGATCCAGATGCTACAAGAGACTGTGAAAAGGAAGTTGG, Reverse: CGTGTCGACTGTTAGGGGTTACCAAACAATTCATCAAGCTCC). Full-length cDNA clone, which was used as a template, was purchased from Source BioScience, clone #IRCBp5005D2310Q. The PCR-amplified MAML3 DNA was cloned into pBp using BamHI and SalI restriction sites. Short hairpins against MAML3 (TRCN0000236445) and IGF2 (TRCN0000062429, TRCN0000062431, and TRCN0000062432) were part of the MISSION TRC short hairpin library (Sigma).

SK-N-SH, SH-SY5Y, and N206 cells were from the laboratory collections of R. Bernards. SK-N-BE cells were kindly provided by Prof. Rogier Versteeg of the Academical Medical Center in Amsterdam. Details of the cell culture and viral transduction methods are given in the Supplementary Document.

Cells were seeded in 6-well plates (20–40 × 10⁴ cells per well for SK-N-SH and SH-SY5Y, 1 × 10⁵ for N206 and SK-N-BE) and cultured in the absence or presence of drugs as indicated. 1 μmol/L RA was used in each proliferation assay, unless otherwise indicated. For each cell line, cells cultured in different conditions were simultaneously fixed in 3.7% paraformaldehyde and subsequently stained with 0.1% crystal violet and subsequently photographed.

Cells were plated in triplicates in 96-well plate at 1,500 cells per well and treated as indicated. The plate was incubated in the IncuCyte (Essen BioScience), and cells were allowed to grow to confluency. The IncuCyte measured and recorded confluency every 4 hours. These data were subsequently converted into growth curves.

Cells were lysed in RIPA buffer containing 150 mmol/L NaCl, 50 mmol/L Tris pH 8.0, 1% NP-40, 0.5% sodium deoxycholate, and 0.1% SDS supplemented with protease inhibitors (cOmplete, Roche) and phosphatase inhibitor cocktails II and III (Sigma). 10× reducing agent and 4× sample preparation buffer (NuPAGE) was added. Subsequently, the samples were boiled for 5 minutes and centrifuged at 14,000 rpm for 5 minutes. Equal amounts of protein were subjected to SDS gel electrophoresis using NuPAGE precast gels and MOPS buffer, followed by Western blotting.

Human Phospho-RTK Array Kit was purchased from R&D Systems (cat no. ARY001B). Lysate preparation and receptor tyrosine kinase (RTK) array development of SK-N-SH parental and SD3.23 cells were performed according to the manufacturer's instructions.

Genomic DNA was prepared from cells using DNAzol. gDNA was used for subsequent nested PCR reaction. Primers for nested PCR reaction are provided in the Supplementary Document.

RNA was isolated from cell lines using Zymo Research Quick-RNA MiniPrep (cat. no. R1055). qRT-PCR assays were performed using 7500 Fast Real-Time PCR System (Applied Biosystems). Relative mRNA levels of genes shown were normalized to the mRNA level of GAPDH (housekeeping gene). The details of the primer sequences are provided in the Supplementary Data.

For endogenous MAML3, RAR, and RXR coimmunoprecipitations, total cell lysates were prepared by lysing the cells in a buffer containing 0.25 mol/L NaCl, 0.1% NP-40, and 50 mmol/L Hepes-KOH (pH 7.3). Protease and phosphatase inhibitors were added just before the lysis step. For each immunoprecipitation, 1 mL of the total cell lysate (1 mg total protein) was used. Two micrograms of each antibody was used to coat the protein A Dynabeads. Antibody-coated protein A Dynabeads were added to the lysates and incubated for 3 hours at 4°C. Antibody-immune complexes were recovered using magnetic separation and washed 4 times using binding buffer. After the final wash, the beads containing immune complexes were suspended in 2× sample buffer and analyzed by Western blotting.

Details of RNA sequencing are provided in the Supplementary Document. Briefly, libraries were prepared as previously described (23) and sequenced on Illumina HiSeq 2000. Reads were aligned to hg19; HTseq-count data were processed using limma package (24) in R.

Chromatin immunoprecipitations (ChIP) were performed as described previously (25). Detailed methods can be found in the Supplementary Document.

To identify novel genes that interfere with RA-mediated differentiation in neuroblastoma, we performed a gain-of-function screen with the VBIM lentiviral insertional mutagenesis system. Random insertion of this vector can place a GFP marker gene and a strong promoter into nearly all genomic loci (22). SK-N-SH neuroblastoma cells were used to perform this screen because of their high sensitivity to RA, leading to proliferation arrest and visible morphologic changes. We infected three pools of each 3 × 10⁶ SK-N-SH cells with one of the three VBIM virus vectors (SD1, SD2, and SD3), enabling insertional activation of cellular genes in all three possible reading frames. After infection, each pool was plated at low density in 1 μmol/L RA–containing medium. As a control, uninfected parental SK-N-SH cells were plated in 1 μmol/L RA–containing medium. The cells were selected for 3 weeks to allow RA-resistant colonies to form (Supplementary Fig. S1A).

Uninfected control cells ceased to proliferate and differentiated into cells with a neuronal morphology. Of the three pools, RA-resistant colonies formed only in the SD3-infected pool. Subsequently, the viral integration sites were retrieved by inverse PCR. One of the RA-resistant clones, SD3.23, had a viral integration site in the MAML3 gene locus between exon 1 and 2 (Supplementary Fig. S1B). This potentially leads to an overexpressed transcript that lacks exon 1 (which encodes the first 156 amino acids of the MAML3 protein). To test this, we performed qPCR-using primer set 1 with sequences in exon 1 and primer set 2 with sequences in exon 2 of MAML3 to compare MAML3 expression between parental SK-N-SH and SD3.23 cells. With primer set 1, SD3.23 had reduced expression of MAML3, consistent with the inactivation of one allele of the gene by viral insertion. However, MAML3 was found to be 9-fold upregulated in SD3.23 cells using primer set 2, consistent with increased expression of a truncated protein lacking exon 1 (Fig. 1A). We also measured MAML3 protein using Western blotting with MAML3 antibodies. In SD3.23, a faster migrating band at 120 kDa was present just below the full-length MAML3. This is in line with our prediction of an overexpressed truncated MAML3 protein in the SD3.23 clone, lacking the first 156 of the 1,133 (full-length MAML3) amino acids (Fig. 1B).

The RA-resistant phenotype of SD3.23 was confirmed by growing the cells in the presence and absence of RA in a long-term colony formation assay. SD3.23 cells conferred potent resistance to RA compared with the parental SK-N-SH cells (Fig. 1C). We also investigated whether MAML3 overexpression confers a more general drug-resistant phenotype. To this end, we treated SK-N-SH parental and SD3.23 cells with cisplatin. Supplementary Figure S1C shows that both cell lines are equally sensitive to this cytotoxic agent, indicating that MAML3 overexpression does not lead to a general drug resistance of SK-N-SH cells. To evaluate the RA-resistant phenotype of SD3.23 cells quantitatively, we measured plate confluency in real time using an IncuCyte device. Parental SK-N-SH cells ceased to proliferate in 1 μmol/L RA, whereas SD3.23 cells proliferated significantly faster than parental cells, which was virtually unaffected by RA treatment (Fig. 1D). The SK-N-SH cell line consists of two morphologically distinct cell types, “N” (neuroblastic) and “S” (substrate adherent) type. An “I” (intermediate) type also exists, which shares characteristics of both N and S groups (26). To evaluate the subtype characteristics of SD3.23, we determined the expression of vimentin and βIII-tubulin in SK-N-SH and SD3.23 cells by immunofluorescence. SD3.23 is I type as they express high levels of vimentin (marker of S type) and tyrosine hydroxylase and dopamine β-hydroxylase, markers of N type (Supplementary Fig. S1D and data not shown).

To further prove that the RA resistance phenotype of SD3.23 is due to MAML3 overexpression, we used an shRNA vector targeting MAML3 to suppress the elevated levels of MAML3 in SD3.23 cells. Knockdown of MAML3 resensitized SD3.23 cells to RA (Fig. 1E and 1F). In addition, SD3.23 cells with MAML3 knockdown displayed a characteristic neurite outgrowth phenotype in the presence of RA, suggesting that these cells underwent differentiation. However, neurite outgrowth was clearly absent in the cells with control knockdown vector (Supplementary Fig. S2A).

RA-induced neuroblastoma cells differentiation is accompanied by the induction of proteins expressed in neuronal cells. To assess if MAML3-mediated RA resistance blocks the induction of such proteins, we evaluated the expression after RA treatment of two bona fide neuronal proteins as markers of neuronal differentiation, SNAP25 and GAP43 (27,28). Basal levels of GAP43 and SNAP25 are higher in SD3.23-MAML3 knockdown cells compared with control SD3.23 cells and increased slightly after RA treatment (Supplementary Fig. S2B). In contrast, the levels of these two proteins remained low in SD3.23-pLKO control cells, suggesting an undifferentiated state. Taken together, these experiments further support that the RA-resistant phenotype of SD3.23 cells is indeed due to overexpressed MAML3.

To validate the effect of MAML3 expression, we PCR amplified a partial MAML3 cDNA that lacked exon 1 and cloned it into the pBp vector (pBp-MAML3-ΔE1). Overexpression of MAML3-ΔE1 after retroviral infection and puromycin selection of SK-N-SH cells conferred resistance to RA similarly to SD3.23 (Fig. 2A). Western blot of the MAML3-ΔE1 lysate confirmed overexpression (Fig. 2D). More importantly, V5 epitope-tagged full-length MAML3 expressed using the lentiviral vector pLX also conferred resistance to RA (Fig. 2B, overexpression shown by Western blot analysis in Fig. 2E). SH-SY5Y cells, derivative of SK-N-SH, also became resistant to RA treatment when infected with pLX-MAML3 (Fig. 2C, overexpression shown by Western blot analysis in Fig. 2F). Overexpression of MAML3 also conferred resistance to RA in two additional neuroblastoma cell lines: N206 and SK-N-BE cells (Fig. 2G and H, quantification of the colony formation in Supplementary Fig. S2G and S2H). It is noteworthy that N206 and SK-N-BE are MYCN-amplified neuroblastoma cell lines (in contrast to SK-N-SH and SH-SY5Y), indicating that overexpression of MAML3 can confer resistance to RA in neuroblastoma cells irrespective of MYCN amplification status.

MAML1 and MAML2 also belong to the mastermind-like family of coactivators. These two proteins are structurally and functionally related to MAML3 (29, 30). However, MAML1 and MAML2 expression in SK-N-SH and SH-SY5Y cells did not mediate resistance to RA (Fig. S2C and Supplementary Fig. S2E, overexpression shown by Western blot in Supplementary Fig. S2D and S2F. In summary, these data indicate that full-length MAML3 is able to confer resistance to RA in neuroblastoma cells by conferring a proliferation advantage to these cells, whereas the related MAML1 and MAML2 proteins do not mediate RA resistance.

We performed cell-cycle analysis in parental SK-N-SH and SD3.23 cells by 5-Ethynyl-2′-deoxyuridine incorporation followed by FACS analysis to investigate the differences in cell-cycle progression. A clear block in G₁ phase occurred in the parental SK-N-SH cells after 72 hours of RA treatment, but not in SD3.23 cells (Fig. 3A and B). Furthermore, we assessed the induction of the neuronal markers GAP43 and SNAP25 after RA treatment. As expected, SK-N-SH parental cells treated with RA showed induction of these proteins. However, in SD3.23 cells, there was negligible induction of these genes. In addition, the basal expression of these markers is lower in MAML3-overexpressing cells (Fig. 3D). Similar results were seen with full-length MAML3 expression (Supplementary Fig. S3). These results imply that MAML3 overexpression leads to an abrogation of RA-induced cell-cycle arrest and also lack of differentiation.

As MAML3 is a known transcriptional regulator, we asked whether MAML3 binds the RAR/RXR complex. Coimmunoprecipitation experiments in SK-N-SH cells suggested that MAML3 binds to RXR, but not to RARα (Supplementary Fig. S4 and data not shown). This interaction indicates that MAML3 could play a role in the regulation of the RA signaling pathway. Next, we performed ChIP followed by sequencing (ChIP-seq) using MAML3 and RARα antibodies in SK-N-SH cells to determine chromatin-binding regions of these proteins in a genome-wide context, both in absence and presence of RA. We identified 6,902 and 8,816 binding regions of MAML3 in the absence and presence of RA, respectively. We also identified 1,346 and 1,537 RARα-bound regions in the absence and presence of RA, respectively. MAML3 and RARα–binding sites strongly overlapped, irrespective of RA treatment (Fig. 4A). MAML3 has more binding sites throughout the genome as compared with RARα, indicating additional roles for MAML3 apart from RA signaling. For both MAML3 and RARα, the number of binding regions increases slightly following RA treatment. Most MAML3 binding sites are found either in gene introns or in the distal intergenic regions, suggesting that MAML3 may be enhancer-bound in these cells. To investigate whether MAML3-binding sites overlap with known marks of enhancers, we used publicly available ChIP-seq data of H3K4me1 and p300 in SK-N-SH cells and H3K27ac ChIP-seq data from the closely related SH-SY5Y cell line (Supplementary Table S5). MAML3-binding regions are enriched for active enhancer marks such as H3K27ac and H3K4me1. Chromatin-binding patterns of p300, another mark of active enhancers, also showed strong overlap with MAML3 binding, suggesting that MAML3 could be part of an activation complex assembled on enhancers (Fig. 4B).

To gain a better understanding of the transcriptional changes that take place upon MAML3 overexpression, we performed RNA sequencing analysis of SK-N-SH parental cells versus SD3.23 cells, in the absence and presence of RA. A heatmap was generated, representing hierarchical clustering of the most differentially expressed genes (Fig. 4C). In the absence of RA, MAML3 overexpression led to increased expression of 502 genes (log₂FC > 1.5 and adjusted P < 0.05), whereas 326 genes were downregulated (log₂FC < −1.5 and adjusted P < 0.05). Using the same criteria, MAML3 overexpression led to increased expression of 582 genes and downregulation of 369 genes in the presence of RA (Supplementary Table S1).

One way cells can become resistant to RA-induced differentiation is by loss of RA target gene expression (20, 21). To determine if ectopic MAML3 expression leads to the loss of RA-responsive gene expression, we integrated our ChIP-seq and transcriptome data. First, we defined RA target genes using two criteria: (i) having a RARα binding site within 20 kb upstream of the transcription start site or within the gene body (N = 819) and (ii) log fold change in expression above 1.5 (with an adjusted P < 0.05) after RA treatment (N = 216). Applying these criteria yielded a list of 53 genes (P < 10⁻¹⁸), among which are a number of well-established RA target genes, including RARβ, CYP26A1, RBP1, and CRABP2 (Supplementary Table S2). Of these 53 genes, 24 had a loss of RA responsiveness in the cells overexpressing MAML3 (log₂FC below 1; Fig. 4D; Supplementary Table S3). Strikingly, 21 of 24 of these genes also had a proximal MAML3-binding site, found within 20 kb of the transcription start site or in the gene body (Supplementary Table S3). This suggests that MAML3 overexpression interferes with activation of a subset of RA target genes upon RA treatment, disrupting RA signaling and resulting in RA resistance.

Further analysis of the ChIP-seq and transcriptome data indicated that 311 genes had proximal MAML3-binding sites and a log fold change of at least 1.5 in expression upon MAML3 overexpression (Fig. 4E). Because of the proximal MAML3 binding sites, these genes are potentially under direct control of MAML3. Of these genes, 80 were RA responsive (i.e., log₂FC > 1.5 upon RA treatment) and 231 were RA unresponsive. This latter finding implies that MAML3 controls additional processes apart from RA signaling, in line with the previously mentioned MAML3 unique binding sites that are devoid of RARα (Fig. 4A).

Taken together, we show that MAML3 and RARα largely overlap in their genomic-binding regions and that MAML3 interacts with RXR. MAML3 overexpression leads to transcriptional changes and loss of activation of a subset of RA target genes. This is accompanied with a block in the induction of neuronal differentiation marker proteins upon RA treatment.

In addition to resistance to RA, we noticed that cells that overexpress MAML3 also had an increased proliferation rate in the untreated condition (Fig. 1D). To identify signaling pathways that could contribute to this hyperproliferative phenotype, we made use of RTK antibody arrays, which allows a global survey of activated RTKs in a cell lysate. We subjected the total cell lysates of SK-N-SH parental and SD3.23 cells treated with and without RA to RTK array analysis. Interestingly, we found that SD3.23 cells had a marked increase in phosphorylated (activated) insulin-like growth factor 1 receptor (IGF1R) and to a much lesser extent of insulin receptor (INSR; Fig. 5A). To validate this, the lysates used on the RTK arrays were also subjected to Western blot analysis to measure IGF1R phosphorylation. We found that both IGF1R and its downstream kinase AKT were significantly activated as judged by their phosphorylation on the activating sites (Fig. 5B). Strikingly, one of the genes that showed the highest differential expression in MAML3 overexpressing cells versus parental cells was the IGF1R ligand IGF2 (Fig. 4C). This was independently validated by qRT-PCR (Fig. 5C). IGF2 was also highly upregulated in SK-N-SH and SH-SY5Y cells overexpressing full-length MAML3 (Supplementary Fig. S5A), which was accompanied by phosphorylation of IGF1R and pAKT in these cells (Supplementary Fig. S5B).

If activation of IGF1R signaling by IGF2 is causally involved in the hyperproliferative phenotype, knockdown of IGF2 should impact proliferation of MAML3-overexpressing cells. To test this, we infected SD3.23 cells with three lentiviral shRNA vectors targeting the IGF2 gene. Two of three hairpins (sh1 and sh4) gave good knockdown of IGF2 expression, whereas one hairpin (sh3) did not give any knockdown (Fig. 5E). Indeed, IGF2 knockdown in SD3.23 cells reduced proliferation approximately to the rate of SK-N-SH parental cells infected with empty vector. IGF2 shRNA3, which did not knockdown IGF2, also did not affect proliferation (Fig. 5D). Total cell lysates from empty vector and IGF2 knockdown cells were subjected to biochemical analysis. Phosphorylation of IGF1R and AKT dramatically decreased upon knockdown of IGF2, confirming that IGF2 is the factor that activates IGF1R signaling in MAML3-overexpressing cells (Fig. 5F).

We asked if IGF2 alone could mediate resistance to RA. The addition of up to 400 ng/mL recombinant IGF2 to the growth medium of parental SK-N-SH cells did not confer resistance to RA, even though cells clearly have a proliferation advantage (Fig. 5G). Western blot analysis of the lysates showed increase in both pIGF1R and pAKT (Fig. 5H). Of note, RA by itself activates AKT signaling. Qiao and colleagues found that RA induces PI3K/AKT signaling, leading to cellular differentiation (31). However, we show that in the absence of RA, AKT activation by IGF2 leads to increased proliferation. We hypothesize that upon MAML3 overexpression, the upregulation of AKT signaling by IGF2 also gives these cells a proliferation advantage in the presence of RA due to the differentiation block in these cells. A similar observation was made in SH-SY5Y cells (Supplementary Fig. S5C and S5D). These results show that IGF2 activates IGF1R and AKT signaling in MAML3 overexpressing cells, which results in hyperproliferation. However, IGF2 alone cannot mediate resistance to RA.

IGF2 and concomitant IGF1R signaling is important for cancer growth and has been implicated in a variety of cancers, including neuroblastoma (32, 33). To translate our findings of IGF2-induced IGF1R signaling in MAML3-overexpressing cells to a clinically useful strategy to inhibit proliferation of these cells, we made use of a small-molecule inhibitor of IGF1R/INSR, OSI-906. First, we determined IC₅₀ values for RA and OSI-906 in both SK-N-SH parental and SD3.23 cells (Fig. S6A and S6B). It is noteworthy that the IC₅₀ for RA in MAML3-overexpressing cells was 10-fold higher than the parental SK-N-SH cells (∼1μmol/L in SK-N-SH cells and >10 μmol/L in SD3.23 cells). On the other hand, the IC₅₀ for OSI-906 in MAML3-overexpressing cells was 4-fold lower than SK-N-SH cells (∼0.4μmol/L in SK-N-SH cells and ∼0.1μmol/L in SD3.23 cells).

Subsequently, we evaluated the effects of IGF1R/INSR inhibition in colony formation assays. Single-drug treatment of OSI-906 had modest effect in SK-N-SH parental cells but had strong effect on proliferation of MAML3-overexpressing cells. However, single-drug treatment of RA had strong effect on SK-N-SH cells and no effect on MAML3-overexpressing cells. Dual inhibition by RA and OSI-906 was synergistic in both SK-N-SH and SD3.23 cells (Fig. 6A). To determine the level of synergy, we calculated the combination indices (CI). Combined treatment of OSI-906 and RA resulted in a CI of 0.52 (synergistic) for MAML3-overexpressing SD3.23 cells and 0.19 (highly synergistic) for SK-N-SH parental cells (Supplementary Fig. S6C and S6D). Both these indices represent synergy between the two drugs (34). Interestingly, the parental cells showed stronger synergy for these two drugs. One reason for this is that in SK-N-SH cells, RA can induce pAKT just prior to differentiation, most likely through IGF1R activation. OSI-906 can reverse this activation of pAKT by RA (Fig. 6B).

Western blotting with lysates of OSI-906 and RA-treated cells revealed that both phosphorylation of IGF1R and AKT were nearly completely inhibited (Fig. 6B), providing a biochemical explanation for the phenotype. However, combination treatment did not lead to a differentiated phenotype as judged by the induction of the neuronal markers SNAP25 and GAP43, indicating that the effect of OSI-906 was mainly on proliferation and not on differentiation (Fig. 6C).

Using a gain-of-function genetic screen, we identify here MAML3 as a novel gene whose ectopic expression confers resistance to RA-induced differentiation in neuroblastoma. MAML3 is a known transcriptional coactivator of Notch signaling along with its related family members, MAML1 and MAML2 (29,30). MAML proteins act as scaffolding proteins and assemble the coactivation complex on NOTCH target genes. It is therefore not surprising that this attribute of MAML proteins can be turned into potent fusion oncoproteins in cancer. MAML3 forms a fusion protein with PAX3 in biphenotypic sinonasal sarcoma. This PAX3–MAML3 fusion protein is a potent transcriptional activator of PAX3 response elements (35). MAML2 is well documented to form a fusion protein with CRTC1 following a t(11;19) translocation, aberrantly activating cAMP/CREB signaling (36–38). This translocation has been recurrently found in salivary gland tumors, mucoepidermoid carcinoma, Warthin tumor, and clear cell hidradenoma of the skin. In AML and MDS patients, MLL–MAML2 fusions have been described (39). This fusion disrupts NOTCH signaling and that possibly contributes to carcinogenesis in these malignancies. MAML1 has been found to be a coactivator for p53-mediated transcription (40), as well as for MEF2C-mediated transcription in muscle differentiation (41). In addition, MAML1 is known to modulate NF-κB signaling (42) and was shown to have a role in bone development by enhancing RUNX2-mediated transcription (43).

In this study, we provide evidence that MAML3 mediates resistance to RA and promotes hyperproliferation in neuroblastoma. We demonstrate that MAML3 expression, but not MAML1 or MAML2, can block the response to RA in neuroblastoma, indicating that MAML3 has a unique role in this disease. MAML3 overexpression interfered with two important consequences of RA treatment, namely cell-cycle arrest and differentiation. RA restricts the cells from entering the S-phase, which was not seen in MAML3-expressing cells (Fig. 3A and 3B). In addition, MAML3 expression resulted in a failure to induce the neuronal markers GAP43 and SNAP25 after RA treatment (Fig. 3D).

We observed that MAML3 overexpression results in loss of activation of a subset of RA target genes, including a number of bona fide RA target genes, such as TGM2, HOXD4, and NTRK2 (Fig. 4D). Other studies have shown that loss of activation of RA target genes blocks RA-induced differentiation (20,21). It is therefore likely that the loss of activation of a subset of RA target genes contributes to RA resistance in MAML3-overexpressing cells. ChIP-seq analyses of MAML3 and RARα revealed that more than 80% of endogenous RARα-binding regions overlap with endogenous MAML3-binding regions, suggesting the involvement of MAML3 in coregulating these genes (Fig. 4A). We also demonstrate that MAML3 physically binds RXR, further reinforcing the notion of a role for MAML3 in RA signaling (Supplementary Fig. S4A). Although MAML3 is known as a transcriptional coactivator, we hypothesize that overexpression of MAML3 interferes with proper transcriptional activation of a subset of RA target genes upon RA treatment, leading to loss of activation of these genes.

In addition, we found that MAML3 overexpression upregulates IGF2 transcription, leading to the activation of IGF1R and downstream AKT signaling (Fig. 5A and B). The IGF1R pathway drives increased proliferation of MAML3-overexpressing cells. However, activation of IGF1R signaling alone does not lead to resistance to RA (Fig. 5G and Supplementary Fig. S5C), indicating that MAML3 controls response to RA and proliferation via distinct mechanisms and pointing at the importance of MAML3 in neuroblastoma oncogenesis. We show that knockdown of IGF2, as well as administration of the small-molecule inhibitor OSI-906, which targets IGF1R and INSR, inhibits proliferation of MAML3-expressing neuroblastoma cells (Figs. 5D and 6A). When OSI-906 is combined with RA, we observed a modest synergy in the inhibition of cell proliferation (Supplementary Fig. S6D). It is noteworthy that the parental cells showed stronger synergy for these two drugs. We consistently observed an increase in pAKT soon after RA treatment. OSI-906 reversed pAKT activation, suggesting that this signal could be in fact coming from IGF1R-INSR receptors and providing a potential explanation for the synergy between the two drugs. This combination treatment could therefore be a potential clinical strategy in MAML3-mutated as well as MAML3 wild-type neuroblastomas.

Interestingly, MAML3 mutations have been found in neuroblastoma patients. Molenaar and colleagues performed whole-genome sequencing on 87 neuroblastoma patient samples and found MAML3 gene mutations in two samples (44). Although we yet do not know the impact on neuroblastoma cells of these MAML3 mutations, they indicate that the MAML3 gene is affected in neuroblastoma, and this merits future study. In the online The Cancer Genome Atlas collection of tumor data, MAML3 is also mutated and/or amplified at a lower frequency in a range of other cancers. The relevance of these aberrations of MAML3 in these cancers remains to be elucidated, but it could point towards a role of MAML3 in malignancies beyond neuroblastoma, which would be in line with the marked increase in proliferation rate we observe here for cells that overexpress MAML3. In summary, our work links MAML3 functionally to differentiation and proliferation, which may explain why this gene is mutated in cancer.

No potential conflicts of interest were disclosed.

Conception and design: G.J.J.E. Heynen, W. Zwart, R. Bernards, P.K. Bajpe

Development of methodology: G.J.J.E. Heynen, P.K. Bajpe

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G.J.J.E. Heynen, S. Palit, N.J. Basheer

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): G.J.J.E. Heynen, E. Nevedomskaya, S. Palit, N.J. Basheer, C. Lieftink, A. Schlicker

Writing, review, and/or revision of the manuscript: G.J.J.E. Heynen, A. Schlicker, W. Zwart, P.K. Bajpe

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): G.J.J.E. Heynen, P.K. Bajpe

Study supervision: W. Zwart, R. Bernards, P.K. Bajpe

The authors thank the George Stark laboratory for providing them with the plasmids and protocols of the VBIM system, NKI Genomics Core Facility for support, and Rogier Versteeg for useful discussions and for providing them with the SK-N-BE cell line.

This work was supported by a grant from the Dutch Cancer Society (KWF) and the Cancer Genomics Netherlands Consortium (CGC.nl). The data discussed in this publication have been deposited in NCBI's Gene Expression Omnibus under accession numbers GSE69121, GSE69119, GSE69183.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.