RAS–MAPK Pathway-Driven Tumor Progression Is Associated with Loss of CIC and Other Genomic Aberrations in Neuroblastoma

Neuroblastoma is a pediatric tumor of the peripheral sympathetic nervous system, which has a highly variable clinical outcome (1). Whole genome sequencing (WGS) studies have identified recurrent aberrations primarily in MYCN, ALK, ATRX, and TERT (2–5). We recently identified that mutations that activate the RAS–MAPK signaling pathway occur frequently in neuroblastoma relapse tumors and that these mutations confer sensitivity to MEK inhibitors (6). As in other tumor types the RAS pathway is frequently affected in neuroblastoma through mutations in the RAS genes themselves, but also via mutations in regulatory proteins and downstream signaling components, like NF1 and PTPN11, or constitutive activation of receptor kinases that trigger this pathway, like ALK (7). These results have led to the inclusion of neuroblastoma relapse patients in clinical trials with MEK inhibitors. However, the role of MEK inhibitors in an induction treatment protocol is under debate, since the role of active RAS–MAPK signaling in primary neuroblastoma tumors was not obvious.

Gene expression signatures of oncogenic pathways have been proven to be a powerful tool in cancer research. These signatures are able to identify patterns of pathway activation in cell lines and tumors and can also predict response to targeted therapeutics (8, 9). Several gene signatures for the RAS–MAPK pathway have been published and these show a wide range of genes being expressed downstream of RAS (8–10). Generally, expression of RAS–MAPK negative feedback regulators is associated with an active pathway, which suggests that it is very tightly regulated (9, 11). Moreover, the RAS–MAPK pathway affects expression of downstream transcription factors like the ETS family of transcription factors, c-FOS and MYC (12–14). One problem is that the majority of published signatures show little overlap (9), suggesting that the effects of RAS activation could be largely background dependent.

We developed a neuroblastoma RAS–MAPK gene signature to determine RAS–MAPK pathway activation in neuroblastoma primary tumors. Because WGS data of a lot of these tumors was available, the signature also enabled us to verify known mutational biomarkers for pathway activation and identify new biomarkers. We show that RAS–MAPK pathway activation is associated with poor prognosis in primary neuroblastoma tumors. Mutations in genes that are known to cause activation, such as ALK, SOS1, and PTPN11, are associated with high pathway activity. In other activated tumors, we detect recurrent aberrations in CIC, DMD, and PHOX2B. We show that these mutations affect this pathway to varying degrees both in vitro and in vivo. Especially deletion of CIC causes a significant increase in signature gene expression and growth speed in vivo. This increase is independent of phosphorylated ERK, indicating that CIC functions downstream of the RAS–MAPK pathway.

All cell lines were cultured in DMEM supplemented with 10% FBS, 20 mmol/L l-glutamine, 10 U/mL penicillin, and 10 μg/mL streptomycin and maintained at 37°C under 5% CO₂. Cell line identities are confirmed by short tandem repeat (STR) profiling using the PowerPlex16 system and GeneMapper software (Promega) upon the start of culture. Cell lines are screened monthly for the presence of mycoplasma.

All analyses were performed using the R2 bio informatics platform (http://r2.amc.nl). Expression data of NRASQ61V overexpression, NF1 CRISPR lines, Trametinib and appropriate controls were profiled in triplicate on Affymetrix u133p2 arrays and are available under GSE115406. For our cohort of primary neuroblastoma tumors two expression series (GSE16476 and GSE73537) were combined, and batch correction with the combat algorithm in R was applied to control for the different labeling kits used. Accession numbers of other used datasets can be found in Supplementary Table S1.

Heatmaps represent z-score normalized data and is represented in the order of clustering. Colored boxes (tracks) above or below expression data represent sample characteristics or clustering outcome. Signature scores were calculated as the average z-score of the signature genes.

For mutational analyses WGS data of 88 primary neuroblastoma tumors (EGA, EGAS00001000222) combined with 21 additional tumors (EGA, EGAS00001001479) was analyzed as was published before.

Lentivirus was packaged by cotransfection of constructs with the third-generation packaging plasmids pMD2.G, pRRE, and pRSV/REV with Fugene HD (Roche) into HEK293T cells. Cell lines were lentivirally infected and subsequently selected with blasticidin or puromycin.

Cells were transduced with virus produced from pLentiCRISPRv2 (15) with the appropriate gRNA sequences (Supplementary Table S2) or pInducer (6) with open reading frames for the mentioned protein and selected with puromycin. Genomic DNA was isolated and CRISPR efficiency verified using PCR and subsequent Sanger sequencing. pLentiCRISPR v2 was a gift from Feng Zhang (Addgene plasmid #52961).

Samples of inducible overexpression lines were harvested after 72 hours exposure to doxycycline. Samples were lysed in Laemli buffer and subjected to Western blotting with antibodies against RAS (Cell Signaling Technology; #3965), total ERK (Cell Signaling Technology; #9102), pMEK (Cell Signaling Technology; #9154), pERK (Cell Signaling Technology; #9101), total MEK (Cell Signaling Technology; #8727), NF1 (Santa Cruz Biotechnology sc-67), and PHOX2B (Santa Cruz Biotechnology sc-376997).

RNA was isolated using Trizol and purified using RNAeasy columns (Qiagen). cDNA was generated from 2.5 μg of RNA using the IScript cDNA Synthesis Kit (Bio-Rad) according to the manufacturers protocol. Gene expression was determined using SYBR green (Bio-Rad) on a C1000 thermal cycler (Bio-Rad). HPRT1, SDHA, and UBC were used as reference genes to normalize gene expression. Primer sequences can be found in Supplementary Table S3.

All experiments were conducted after obtaining ethical approval from the DEC (animal experiments committee) of the AMC under number DAG103101. NMB cell lines were generated as described above and 2 × 10⁶ cells were xenografted into female NMRI nu-/nu- mice at 5 to 7 weeks of age (n = 6 per group). Tumor size was monitored two times a week. Tumor burden was determined according to the formula (π/6)d³, where d represents the mean tumor diameter obtained by caliper measurement. Mice were sacrificed when tumors volumes exceeded 1,000 mm³. Statistical analysis was performed using a Gehan Breslow Wilcoxon test to determine significant differences between survival curves.

To identify transcriptional changes associated with activation of the RAS–MAPK signaling pathway in neuroblastoma, we generated several cell line systems in which this pathway is activated. Activating mutations in NRAS and inactivation of NF1 are both associated with increased RAS–MAPK signaling and occur at low frequency in neuroblastoma tumors (7, 16). An NRASQ61V mutated protein was overexpressed and the NF1 gene was inactivated using CRISPR, both in the cell line NMB. This cell line did not contain any known mutations affecting the RAS–MAPK pathway on forehand as determined by WGS and did not show activation of RAS–MAPK signaling pathway (Supplementary Fig. S1). Both overexpression of mutated NRAS and inactivation of NF1 caused activation of the RAS–MAPK signaling pathway as measured by Western blot analysis (Fig. 1A). Gene expression profiling was performed for all cell lines and appropriate controls.

Mutations in NRAS and NF1 are also known to activate other signaling pathways. To further restrict the gene set to only genes directly associated with RAS–MAPK signaling, we performed gene expression profiling of the cell line SJNB8 treated with the MEK inhibitor Trametinib for 6 hours. SJNB8 has a KRAS G12D mutation and concordantly shows high levels of phosphorylated ERK on Western blot analysis, which is completely abrogated by treatment with 1 μmol/L Trametinib for 6 hours (Fig. 1B).

Genes that were regulated by expressing the different mutations and were conversely regulated by MEK inhibitor treatment by more than 2.5-fold were included in the mRNA signature (Supplementary Table S4; Supplementary Fig. S2). Gene set enrichment analysis of the differentially expressed genes shows KRAS signaling as the top enriched category for each treatment (Supplementary Table S5), suggesting that changes in RAS–MAPK pathway were indeed the most relevant change in our models. Analysis of the overlapping genes yields a set of six genes that are consistently upregulated upon activation and downregulated upon inhibition of the RAS–MAPK pathway (Fig. 1C). This set contains several genes that are known to be associated with RAS–MAPK signaling in other systems and are part of established RAS–MAPK signatures, such as DUSP4, DUSP6, and ETV4 (9, 17). Analysis of expression data from keratinocytes transfected with different oncogenes shows a strong upregulation of these six genes in RAS transfected cells in contrast to cells transfected with GFP or other oncogenes, showing that upregulation of these genes is specifically associated with increased RAS signaling in multiple cell systems (Fig. 1D; ref. 8).

To assess the use of this gene signature in neuroblastoma, we used our panel containing 17 neuroblastoma cell lines. Twelve of these cell lines contain mutations affecting RAS–MAPK signaling (ALK, NF1, RAS, and BRAF) and it was previously shown that mutation status is associated with sensitivity to MEK inhibition in these cell lines (6). We confirmed that mutation status also correlates with phosphorylated MEK and ERK on Western blot analysis (Supplementary Fig. S1).

Expression data from this cell line panel was subjected to k-means clustering in two groups using the signature genes. The cell lines with mutations affecting the RAS–MAPK pathway cluster together on one side and consistently show high expression of the signature genes (Fig. 2A). Furthermore, when the average expression of these genes is calculated as one signature score per cell line, these signature scores show significant differences between mutated and nonmutated cell lines (Supplementary Fig. S3). Signature scores also correlate with MEK and ERK phosphorylation and sensitivity to three different MEK inhibitors (Fig. 2B and C).

The same analyses were performed in an independent dataset containing 23 neuroblastoma cell lines for which mutation status and sensitivity to a variety of targeted inhibitors is available (18). Again, all mutated cell lines show high signature scores. One cell line for which no mutations were reported also clustered together with the positive lines, suggesting that this line may contain undetected mutations that activate this pathway (Supplementary Fig. S4A). Also in this dataset signature scores correlate with sensitivity to three different MEK inhibitors, which are distinct from the ones in the previous analysis (Supplementary Fig. S4B). These results suggest that the RAS–MAPK gene signature can accurately determine the activation of this pathway and sensitivity to targeted treatment in neuroblastoma cell lines.

To determine if this signature also performs well in vivo, analyses were performed in mRNA profiles of transgenic neuroblastoma mouse models. In transgenic mice, MYCN amplification alone is enough to induce tumor formation; however, tumors that also contain ALK mutations display increased RAS–MAPK signaling and have a shorter latency and are much more aggressive (19, 20). The tumors that have both ALK mutations and MYCN amplification have significantly higher signature scores than the tumors with only MYCN amplification in both datasets (Fig. 2D), consistent with the observation that ALK mutations activate RAS–MAPK signaling in these models. This shows that expression of our signature also correlates with activation of this pathway in vivo in neuroblastoma mouse models.

The RAS–MAPK gene signature was used in expression data of our set of 122 primary neuroblastoma tumors to determine the effect of pathway activation on prognosis. The data were subjected to k-means clustering in two groups, which divides the tumors into a group of 26 tumors with high expression of the signature genes and a group of 96 tumors with low expression (Fig. 3A; Supplementary Fig. S5A). The patient group represented in the cluster with high expression showed a significantly shorter overall survival than the patients in the cluster with low expression (Fig. 3B; Supplementary Fig. S5B). The same analysis was performed in an independent larger dataset of primary neuroblastoma containing 498 tumors (21). Also here, a smaller group (126/498) of tumors clustered together and showed high expression of the gene signature and this was associated with significantly shorter overall survival (Fig. 3C and D). In both datasets, pathway activation was also associated with shorter progression-free survival (Supplementary Fig. S6A and S6B). These results suggest that the RAS–MAPK pathway is more active in a subset of neuroblastoma tumors and that activation of this pathway is associated with poor prognosis.

Neuroblastoma is classified in five stages (1–4S) with stage 4 being the most aggressive. To determine if the prognostic effect of activation of RAS–MAPK signaling is caused by association with a higher stage we looked at the effect of RAS–MAPK activation on survival within each stage. Multivariate cox proportional hazard analysis shows that RAS–MAPK activation is still a significant predictor for survival when we take risk group into account (Supplementary Table S6). Accordingly, pathway activation is associated with worse prognosis in stage 3 and 4 tumors (Supplementary Fig. S7A and S7B), showing that the effect of RAS–MAPK pathway activation on survival is not caused solely by association with a specific stage. In stage 1 and 2 and 4S, there is no clear correlation between RAS–MAPK pathway activation and survival (Supplementary Fig. S7C–S7E). These results suggest that RAS–MAPK activation does not determine tumor stage but does cause a more aggressive tumor when it occurs in high stages.

We also observe that pathway activation as determined by clustering is associated with MYCN amplification in the larger dataset (Fisher exact P value = 6.03e−16), suggesting that the two events may cooperate. However, in a set of only non-MYCN amplified tumors the patients with tumors that cluster in the group with high expression still have a significantly poorer prognosis (Supplementary Fig. S8A and S8B; ref. 22). Moreover, Cox proportional hazards analysis show a significant association with survival also if we take MYCN amplification into account in the larger dataset, whereas this is borderline significant in our dataset (Supplementary Table S6). These data suggest that the effect of RAS–MAPK pathway activation on prognosis is not dependent on MYCN amplification. Activation of telomere maintenance through mutations in TERT and ATRX was also recently shown to be an important risk factor for poor prognosis, however, these groups were too small to test the association between an activated RAS–MAPK pathway and the presence of these mutations (Supplementary Fig. S8C).

To determine if activation of this pathway is associated with the presence of activating mutations, we analyzed WGS data (available for 67 of 122 tumors) and targeted sequencing data for PHOX2B and ALK of these tumors. We detect mutations that have been described to activate the RAS–MAPK pathway in 10 tumors (Supplementary Table S7), and eight of these fall into the MEK active cluster (Fisher exact P-value = 3.4E−07). These are mostly mutations in ALK, but also mutations in other genes associated with RAS–MAPK signaling such as PTPN11, SOS1, and NF1. Tumors with these mutations have significantly higher signature scores than tumors without (Supplementary Fig. S9).

To identify potential other genetic aberrations that lead to activation of this signaling pathway in neuroblastoma, genes were selected that were mutated two times or more in the RAS–MAPK active cluster (Fig. 4A). We previously identified several tumors with frameshift or nonsense mutations in PHOX2B (23), of which five were also present in this dataset (Fig. 4B). Three of these tumors fall into the RAS-MAPK active group as determined by k-means clustering, whereas the other two also show relatively high signature scores. Furthermore, two tumors in the RAS-MAPK active cluster show deletions in the DMD gene (Fig. 4C) and two other tumors from this cluster show homozygous inactivation of the CIC locus (Fig. 4D). In our series of 108 tumors that were subjected to WGS a third tumor with homozygous inactivation of CIC was detected (Supplementary Fig. S10). Unfortunately, no mRNA expression data was available for this tumor.

To determine if these aberrations play a role in the activation of the RAS–MAPK pathway in neuroblastoma, we generated overexpression or knockout cell lines. Because PHOX2B mutations are thought to exert a dominant negative effect (24), a mutant PHOX2B protein derived from the SKNSH cell line was expressed in the NMB and SJNB1 cell lines (Fig. 4B). This indeed leads to activation of this signaling pathway as identified by Western blot analysis, whereas this activation was not observed when wild-type PHOX2B was expressed (Fig. 5A).

Inactivation of CIC in similar locations as the breakpoints detected in the tumors using CRISPR-CAS9 did not lead to an increase in MEK and ERK phosphorylation (data not shown). It did, however, lead to upregulation of the signature genes compared with the unmodified cell line (Fig. 5B). CIC is known to function downstream of ERK and is involved in the activation of ERK target genes (25). Indeed, treating the CIC mutant NMB line with Trametinib does not cause a shift of the CIC mutated line from the active to the inactive cluster, indicating that expression of the signature genes is not severely affected. In contrast, this shift is induced by treating NF1 deleted lines with Trametinib (Fig. 5B). These suggest that loss of CIC induces expression of RAS–MAPK target genes independent of activated ERK. In lines where DMD was knocked out or where mutant PHOX2B was overexpressed, no increased expression of the signature genes was observed (Supplementary Fig. S11).

To test the hypothesis that activation of RAS–MAPK signaling leads to increased tumorigenicity in neuroblastoma, the NMB lines modeling the various mutations were used in xenograft experiments. Doxycycline induced expression of NRAS Q61V in xenograft tumors causes activation of the RAS–MAPK pathway as measured by increased expression of RAS–MAPK signature genes, although the difference is not as clear as in vitro (Fig. 5C). This was associated with increased tumor growth and a significantly shorter survival time than mice engrafted with the same cells that did not receive doxycycline (Fig. 5D). We did not observe effects on tumor growth induced by expression of mutant PHOX2B or knockout of DMD using CRISPR. Concordantly, no increased expression of the RAS–MAPK signature genes was observed (Supplementary Fig. S12A and S12B).

Inactivation of CIC in our xenograft tumors does cause a significant increase in the expression of RAS–MAPK signature genes (Fig. 5C) and a significantly shorter survival (Fig. 5D). We observed in vitro that deletion of CIC caused increased expression of the signature genes without inducing ERK phosphorylation. The observation that CIC deletions cause increased tumor growth suggests that expression of the RAS–MAPK target genes induces the functional effects of an activated RAS–MAPK pathway, even in the absence of active ERK. This also suggests that tumors with CIC inactivation will not respond to MEK inhibition, because the pathway is activated downstream of the target of the inhibitor. In line with this hypothesis, the presence of inactivating mutations in CIC were identified as the strongest predictor of resistance to Trametinib in a pan-cancer panel of 930 cell lines (Supplementary Fig. S13; ref. 26; https://www.cancerrxgene.org).

In this paper we show through a mRNA signature that increased RAS–MAPK pathway activity in primary neuroblastoma tumors is associated with a poor prognosis. In several tumors with an active signature, previously described mutations are present that can account for activation of this pathway. Furthermore, we show that activation of this pathway in vivo promotes tumor growth. Finally, we identify CIC as a tumor suppressor in neuroblastoma that induces the downstream effects of RAS–MAPK pathway activation, whereas not directly affecting activation of the pathway itself (Fig. 6).

The mRNA signature was generated through analysis of gene expression data of activation and inhibition of the RAS–MAPK pathway in neuroblastoma cell lines. The stringent analysis yields a list of only six genes, but the expression of these genes is highly correlated with MEK and ERK phosphorylation and sensitivity to MEK inhibition in cell lines, suggesting it has functional relevance. Furthermore, the observation that several of these genes, like DUSP4 and DUSP6, are bona fide RAS–MAPK target genes strengthens this hypothesis (9, 17).

When used in tumor series, the gene signature identifies a group of tumors that have high activation of the pathway and a significantly worse prognosis. However, activation of the pathway is only weakly associated with higher stage and happens across all stages, reminiscent of ALK mutations, which show a similar distribution and relationship to prognosis (27). This suggests that RAS–MAPK activation can occur in all tumor stages, and is not necessarily associated with a high-risk phenotype. When it occurs in high-risk tumors, however, it is associated with a significantly poorer prognosis. Whether activation of this pathway is also associated with poor prognosis in low risk tumors, remains to be determined, because it is possible that our study was underpowered to detect this effect.

We also observe an association between pathway activation and MYCN amplification, suggesting that these events may cooperate. It is known that activated RAS can stabilize the MYCN protein in neuroblastoma (28), which could explain our observations. However, high expression of the signature is still associated with poor prognosis in non-MYCN amplified tumors, showing that its effect is not only dependent on MYCN amplification.

Gene signatures provide valuable insight into the biology behind cancer. However, classification based on these signatures is highly dependent on the context of the patient cohort used and can, therefore, be difficult to reproduce, so their use as biomarkers for treatment is limited (29). Mutations are generally easier to detect and may provide a more robust biomarker of pathway activation and the application of targeted treatment. Therefore, we combined the mRNA signatures with mutational data and identify that mutations in certain genes are associated with activation of the RAS–MAPK pathway in neuroblastoma.

Mutations in genes known to activate the pathway were identified in 8/26 tumors, which were classified as RAS–MAPK active by clustering. This means that 31% of active tumors show possible causative mutations for RAS–MAPK pathway activity. Not all tumors were profiled using WGS, so it is possible that this frequency is even higher. These suggest that it is feasible to detect RAS–MAPK active tumors, which may be amenable to MEK inhibitor treatment, using mutations as a biomarker. Similar analyses in larger datasets of matched expression and mutation data will hopefully yield more biomarkers to also detect the remainder of active tumors. This would obviate the need for clinical use of our gene signature and the associated downsides.

We observed high signature scores in tumors with mutations in PHOX2B and DMD. However, mimicking these mutations did not cause activation of the signature genes nor an increase in tumor growth, although some activation of the pathway could be shown on Western blot analysis for PHOX2B. However, overexpression of NRAS Q61V, a strong oncogenic driver of RAS–MAPK signaling, only caused a modest increase in tumor growth. Therefore, it is possible that our study was underpowered to detect the effect of other mutations. Moreover, the unmodified NMB xenograft line already showed very rapid growth. This could indicate that it has acquired other aberrations that have a similar functional effect as RAS–MAPK pathway activation, complicating the detection of effects induced by our treatment.

Mutations in PHOX2B were previously linked to retinoic acid resistance in neuroblastoma (30), which was also reported for RAS–MAPK activation through NF1 deletion (16). Losses in DMD were recently reported as oncogenic drivers in tumors from the myogenic lineage (31), whereas mouse models where DMD is deleted show a high frequency of embryonal rhabdomyosarcoma (32), a myogenic tumor that is frequently driven by RAS hyper-activation. These indirect links between these mutations and RAS–MAPK activation merit further study of these genes in neuroblastoma.

We did observe strong activation of our signature genes by deletion of CIC and a significant increase in tumor growth. Inactivation of the CIC gene was recently described to drive metastasis in various cancer types (33). The Capicua tumor suppressor normally represses transcriptional changes induced by activated ERK by binding to promoter region of these genes (34). Active ERK causes detachment of CIC from the DNA, thereby enabling transcription of these genes. Loss of CIC therefore induces downstream gene expression independent of ERK. In concordance with this, we see induction of signature gene expression in CIC knockout lines, which cannot be inhibited by MEK inhibitor treatment. This means that MEK inhibitor treatment is not expected to be effective in CIC mutated tumors, which is concordant with CIC mutations being the strongest predictor for resistance to the MEK inhibitor Trametinib (https://www.cancerrxgene.org; ref. 26). Homozygous inactivation of CIC was detected in 3 of 108 of tumors in our series (5, 35). Another study detected focal deletions in CIC in 3 of 30 high-stage neuroblastoma tumors; ref. 36). These suggest that loss of CIC occurs relatively frequently in neuroblastoma and that further research is necessary to identify appropriate therapeutic targets for tumors with these aberrations.

Recent targeted and WGS studies have shown that mutations in known RAS–MAPK genes, such as NF1, NRAS, and PTPN11, occur in primary neuroblastoma tumors (7, 16). Although the relative frequency of each mutation is low, the cumulative incidence of these known mutations combined with the putative new aberrations we identify represents a significant part of the aggressive primary neuroblastoma tumors. For tumors with ALK mutations, an ALK inhibitor may be a more suitable treatment, but for tumors with other mutations MEK inhibitors may be a viable therapeutic strategy. Further research should therefore indicate whether the presence of these mutations in neuroblastoma tumors can indeed be used as a biomarker for treatment with MEK inhibitors.

No potential conflicts of interest were disclosed.

Conception and design: T.F. Eleveld, J.J. Molenaar

Development of methodology: T.F. Eleveld, L. Schild, J.J. Molenaar

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T.F. Eleveld, L. Schild, G.A. Tijtgat, R. Versteeg

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T.F. Eleveld, J. Koster, D.A. Zwijnenburg, R. Volckmann, J.J. Molenaar

Writing, review, and/or revision of the manuscript: T.F. Eleveld, L. Schild, L.K. Alles, G.A. Tijtgat, J.J. Molenaar

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Schild, J. Koster, D.A. Zwijnenburg, L.K. Alles, M.E. Ebus, R. Volckmann, P. van Sluis, R. Versteeg,

Study supervision: T.F. Eleveld, J.J. Molenaar

The research in this paper was supported by the Kinder Kankervrij Foundation (KiKa) under grants nos. 132 and 189 (primary recipient: Jan J. Molenaar).

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.