Meta-analysis of Neuroblastomas Reveals a Skewed ALK Mutation Spectrum in Tumors with MYCN Amplification

Neuroblastoma is the most common solid extracranial pediatric tumor, with an annual incidence of 1 in 100,000 children below the age of 15 years (1). Despite intensive multimodal treatment, neuroblastoma remains fatal in almost half of the unfavorable patients. Insights into the molecular pathogenesis of this disease are required for the development of less toxic and more effective molecular targeted therapy. Detailed studies of patterns of DNA copy number alterations have been instrumental in our understanding of the clinical and biological heterogeneity of this tumor. Three major genomic subtypes represent >80% of all cases, i.e., hyperploid neuroblastoma with whole chromosome gains and losses, near diploid neuroblastoma with 11q deletions and 17q gain, and MYCN-amplified neuroblastoma with 1p deletions and 17q gain (2–5). The discovery of rare but recurrent high-level amplification of the ALK gene and a genetic study of familial neuroblastomas led to the discovery of activating ALK mutations in neuroblastoma (6–10). The frequency of ALK mutations in primary neuroblastoma varied between 6% and 11% in the different studies (6–10). The relatively low number of mutations described in each of the individual studies has precluded a thorough analysis of the frequency and distribution of recurrent ALK mutations across the different genomic subtypes. Moreover, in the first published series, there was a significant bias towards analysis of high-stage tumors, thus preventing a more general assessment of frequency and distribution of mutations across different stages (6–8). In our current study, we screened an additional 254 neuroblastoma cases including all clinical stages and genomic subtypes. In a meta-analysis, these findings were combined with those from 455 published cases (8–10) for which genomic subtype and clinical information were available. This strategy enabled us to analyze the ALK mutation profile in relation to genomic and clinical data in 709 neuroblastomas, which revealed a distinct mutation spectrum in relation to genomic subtype. Two hotspot mutations, F1174L and R1275Q, were shown to induce ALK autophosphorylation and were able to transform interleukin-3 (IL-3)-dependent Ba/F3 cells into cytokine-independent growth. In addition, we evaluated ALK gene expression levels and showed that high ALK expression is correlated with poor survival.

In total, 254 primary untreated neuroblastoma tumors with a tumor percentage >60% were investigated, including 44 stage 1, 30 stage 2, 34 stage 3, 113 stage 4, and 33 stage 4S tumors [according to the International Neuroblastoma Staging System (INSS); ref. 11]. Patient information and genomic subtypes for 455 published tumors screened for ALK mutations were retrieved from the publications or were made available by the authors (Supplementary Table S1; refs. 8–10). In addition, 39 neuroblastoma cell lines were included. The cell lines were obtained from several sources (see Supplementary Table S2). All of the cell lines were genotyped by DNA fingerprinting (PowerPlex, Promega).

Genomic DNA was isolated using the Qiagen DNA isolation kit (Qiagen) or a standard proteinase K/SDS procedure.

For the first tumor cohort (146 cases) and the 39 cell lines, all 29 ALK coding exons were analyzed, whereas the remaining 108 tumors were screened for only the tyrosine kinase domain. Constitutional DNA from blood samples was available and analyzed for 12 of the 17 patients with a mutation in the primary tumor. Exons were amplified from genomic DNA (primer information in Supplementary Table S3). PCR products were subjected to directional or bidirectional sequencing using BigDye Terminator V1.1/V3.1 Cycle Sequencing chemistry on an ABI3730XL sequencer (Applied Biosystems). Electropherograms were analyzed using Seqscape v2.5 software (Applied Biosystems).

To determine DNA copy number alterations, array comparative genomic hybridization (arrayCGH) was done by using an in-house developed 1 Mb resolution bacterial artificial chromosome array (BAC) array (37 samples) as previously described (3) or by using a custom-designed 44K array enriched for regions with recurrent imbalances in neuroblastoma (1p, 2p, 3p, 11q, 17; 217 samples; Agilent Technologies). For the latter, 150 ng of tumor and reference DNA were labeled with Cy3 and Cy5, respectively (BioPrime ArrayCGH Genomic Labeling System, Invitrogen). Further processing was done according to the manufacturer's guidelines. Features were extracted using the feature extraction v10.1.0.0.0 software program and processed with an in-house developed visualization software arrayCGHbase (http://medgen.ugent.be/arrayCGHbase; ref. 12), including circular binary segmentation for scoring of DNA copy number alterations (13).

Gene expression data were available for 440 tumors, comprising a published dataset of 251 tumors profiled on custom Agilent 44k arrays (ref. 14; downloaded from the EBI ArrayExpress database E-TABM-38), an unpublished dataset of 101 tumors profiled on the Human Exon 1.0 ST Affymetrix arrays (normalized and summarized at the transcript level using RMAsketch), and an unpublished dataset of 88 tumors profiled on the Affymetrix HG-U133plus2.0 platform (normalized using MAS5 and downloaded from the R2 database).¹⁵

For the establishment of a tissue microarray, three representative areas from each tumor were selected on H&E-stained slides from 70 formalin-fixed and paraffin-embedded primary untreated neuroblastoma tumors (Supplementary Table S4). Neuroblastoma tumors were classified according to the International Neuroblastoma Pathology Classification (INPC) scoring system (15), which divides tumors into undifferentiated, poorly differentiated, and differentiating. Of each tumor, three cores were punched into the recipient block. Immunohistochemistry was done using a monoclonal mouse anti-human CD246 ALK antibody (clone ALK1, Dako), and slides were scored for immunoreactive neuroblastoma cells where 0 meant no or weak staining intensity (<10% of the cells), 1 meant weak staining intensity (10-50% of the cells), 2 meant medium staining intensity (50-80% of the cells), and 3 meant high staining intensity (>80% of the cells). The medium score of the three punches was calculated for Kaplan-Meier and log-rank analysis.

Lysis of cells was done when the neuroblastoma cell lines reached 70% confluency. Total cell lysates (50 μg of protein) were analyzed by standard procedures (16) using anti-phospho-ALK (Tyr1604), anti-ALK (Cell Signaling), and anti-ERK2 (Santa Cruz) antibodies. The Aida Image Analyzer v.4.22 was used for quantification of Western blots.

Ba/F3 cells were cultured in RPMI-1640 supplemented with 10% fetal bovine serum and 1 ng/mL murine IL-3 (Peprotech). ALK F1174L and ALK R1275Q constructs were generated by PCR, cloned into the retroviral vector pMSCV-neo (Clontech), and transduced in Ba/F3 cells. Experiments were done in triplicate. Transduced Ba/F3 cells were selected with G418 (500 μg/mL medium). The amount of viable cells was assayed on regular intervals. For proliferation curves, Ba/F3 cells were washed with PBS, and 5 × 10⁵ cells were seeded in 5 mL medium without IL-3. Viable cells were counted using a Vi-CELL Cell Viability Analyzer (Beckman Coulter) on days 3, 5, 7, and 9.

Fisher's exact tests, Mann-Whitney tests, and correlation analyses were done using R (version 2.8.1). The R Survival package was used to generate Kaplan-Meier plots and to carry out log-rank analyses. Multivariate logistic regression analysis was done using the glm function (R-base package).

In a series of 254 sporadic, nonfamilial primary neuroblastoma tumors, a total of 17 ALK mutations (6.7%) and 2 ALK amplifications (0.8%) were identified (Supplementary Fig. S1). In a first series of 146 cases, all 29 ALK coding exons were analyzed but no mutations were found outside the tyrosine kinase domain. Therefore, only the tyrosine kinase domain was analyzed for the remaining 108 tumors. The most frequent mutations were located at residues R1275 and F1174, and were detected in 3.9% (10 of 254) and 2.0% (5 of 254) of the cases, respectively. A third recurrent but less frequent mutation affecting residue F1245 (6–9) was not detected in our series. One of the mutations previously reported in only one tumor was also detected as a single case in our series (Y1278S; ref. 10), thus providing further evidence for its contribution to neuroblastoma pathogenesis. In addition to the previously reported mutations we observed one new missense mutation, R1231Q. Sequence analysis of the constitutional DNA of 12 of 17 patients with an ALK mutation showed that these mutations were somatically acquired.

Mutation analysis was also done in 39 neuroblastoma cell lines. All mutations, except for one (D1091N in LAN-6), were exclusively found within the tyrosine kinase domain. Recurrent mutations were found in two cell lines that were previously not analyzed, STA-NB-8 and NB-14. All results of previously reported cell lines were in concordance with published results except for one (see Supplementary Table S1).

We carried out a meta-analysis on the new cohort of 254 cases together with 455 previously analyzed and published cases to relate the ALK mutations to clinical and genomic data (8–10; Table 1 and Supplementary Table S1). The 709 samples had in total 49 ALK mutations (6.9%). The two most frequently occurring mutations were the F1174 (34.7%) and R1275 mutations (49%; Supplementary Fig. S1). When the entire group of different ALK mutations was taken into account in the evaluation of the overall frequency of ALK mutations, no significant difference was observed for the frequency of mutations in favorable (INSS 1, 2, and 4S; 14 of 245, 5.7%) versus unfavorable (INSS 3 and 4; 33 of 440, 7.5%) neuroblastomas (Fisher's exact test P = 0.087). Mutation frequencies were also compared within different stages, MYCN status, or age, but no significant differences were found (see Table 1 for P values). However, when looking at the different types of mutations in relation to the clinical and genomic parameters, we noticed a skewed distribution for MYCN status in the F1174 mutated cases versus the tumors with wild-type ALK (P = 0.001; Fig. 1A and Table 1). ALK F1174 mutations were found in 1.3% of the MYCN-single copy tumors, compared with 6.1% of MYCN-amplified tumors. Of the 17 tumors with the F1174 mutation, 58.8% had MYCN amplification, compared with a frequency of MYCN amplification of 21.6% in the tumors with wild-type ALK (Fig. 1A). In contrast, the frequency of MYCN amplification was similar for R1275 mutated versus wild-type cases. As most cell lines were MYCN amplified, we sequenced 39 neuroblastoma cell lines to verify whether the association of F1174 with MYCN amplification could also be detected. Indeed, F1174 mutations were present in 5 of 27 MYCN-amplified cell lines whereas only one R1275 mutation was present in this cohort (Supplementary Table S2).

To further explore the relationship between ALK mutation status and genomic alterations, we classified the tumors into genomic subclasses based on arrayCGH data (3). For 659 tumors, we could establish the genomic subtype: subtype 1 with numerical imbalances only (n = 218), subtype 2A with 11q deletion and without MYCN amplification (n = 126), subtype 2B with MYCN amplification (n = 158), and subtype 3 without any detectable DNA copy number alterations (n = 78). This classification covered most of the cases (88%), with 79 remaining unclassified (Supplementary Fig. S2).

A comparison of the ALK mutation frequency in relation to genomic subtype revealed that ALK mutations were most frequently observed in MYCN-amplified tumors (subtype 2B; 8.9% mutated), followed by subtype 1 tumors (7.3% mutated), subtype 2A tumors (4.0%), and subtype 3 tumors (1.3%). Interestingly, all infrequent mutations (6 of 49) were present in subtype 1 tumors (Fig. 1B).

No significant survival differences were found in tumors with or without ALK mutations (or amplifications; log-rank P = 0.317; Fig. 2A). However, when the survival of patients with R1275 mutation or wild-type patients was compared with patients with the F1174 mutation type, Kaplan-Meier analysis showed significant survival differences (P = 0.027 and P = 0.002; Fig. 2B and C). This might largely be explained by the high frequency of MYCN amplification within the F1174 mutated tumors compared with the R1275 mutated tumors.

Interestingly, although not statistically significant, we noticed that 9 of 10 patients with a MYCN amplification and F1174 mutation died of disease, suggesting that within the MYCN-amplified subgroup, patients with the F1174 mutation may have a particularly poor survival as compared with a 32% 5-year overall survival rate for MYCN-amplified cases without the F1174 mutation.

Overall, ALK amplifications could be detected in only 12 of 709 tumors (1.7%), and none of these carried activating ALK mutations. All ALK-amplified tumors, except for one, were also found to be MYCN amplified (P < 0.001), which accounts for 6.7% of the total of MYCN-amplified tumors. Like ALK mutation, amplification is not a statistically significant independent marker for survival when analyzed in a model with MYCN, stage, and age in a logistic regression analysis (data not shown).

Low copy number gain of chromosome 2p material or whole chromosome 2 gain was detected in 19.3% (49 of 254) and 17.7% (45 of 254) of the cases, respectively, in keeping with the high occurrence reported in previous studies (6, 8–10, 17). No focal low copy number gains were detected for ALK. In cases with partial 2p gain, the extra chromosomal segments varied in size from 15 to 87 Mb and included the MYCN gene except for one case. In the latter, MYCN amplification was present together with a more distal 15 Mb gain with a telomeric breakpoint immediately proximal to the ALK locus (Supplementary Fig. S3). The ALK gene was included in all except four cases of partial 2p gain (91.8%). An interesting observation, particularly in view of the high frequency of ALK copy number gain, was that 2p gains encompassing the ALK locus were present in only 2 of 17 mutated tumors (11.8%), indicating that 2p gain is not a common mechanism for increasing mutated ALK copy number.

To evaluate the possible impact of ALK copy number gain on ALK expression, we compared Affymetrix exon array expression data and arrayCGH data for 101 neuroblastomas. This showed a strong correlation between copy numbers of the ALK gene and expression levels (Spearman correlation coefficient, 0.308; P = 0.002; one ALK-amplified sample was omitted from this analysis; Supplementary Fig. S4). This was confirmed by Mann-Whitney analysis comparing the expression in tumors with normal ALK copy number versus tumors with ALK gain (CGH result >0.3; P = 0.001).

Next, we evaluated the relation between ALK gene expression and survival. We analyzed expression levels in three independent datasets including a total of 440 tumor samples (14).¹⁶

In addition to the relation between ALK transcript expression and survival, we also investigated the ALK protein expression status and patient survival using a tissue microarray containing 70 primary tumors. Kaplan-Meier and log-rank analysis show a significant correlation between ALK protein expression and overall survival (OS; P = 0.014) and progression-free survival (PFS; P = 0.002; Fig. 4). The three cases with ALK mutation (R1275Q) present on the tissue microarray had a median expression value (score 2) whereas in the remaining ALK wild-type tumors, ALK reactivity varied from low (score 0) to high (score 3).

For the evaluation of ALK activity, the level of activated/phosphorylated ALK (p-ALK) was investigated. However, as the specificity of p-ALK antibodies remains to be determined, we carried out Western blot analysis rather than immunohistochemistry experiments on a panel of 22 neuroblastoma cell lines. This allowed for comparison of ALK mRNA expression levels (P = 0.011) and native ALK protein levels (P = 1.55E-07) versus p-ALK protein level. We clearly show a significant correlation between both ALK mRNA expression levels and native ALK protein levels with p-ALK protein levels (Supplementary Fig. S5). Interestingly, we could also show that cell lines harboring the F1174 mutation (except for the SK-N-SH cell line with very low ALK expression levels) or ALK amplification have relatively more phosphorylated ALK than do cell lines with the other hotspot mutation. Moreover, cell lines with wild-type ALK have very low levels of phosphorylated ALK (Supplementary Figs. S6 and S7).

Given the observed concordance between mutation type and ALK activity, we also compared the transforming capacity of both ALK hotspot mutations in IL-3–dependent Ba/F3 cell lines.

Although both mutants were able to transform Ba/F3 cells to IL-3–independent growth, cells expressing ALK F1174L transformed the cells significantly faster than did Ba/F3 cells expressing ALK R1275Q (Fig. 5).

The present meta-analysis of ALK mutations of 709 neuroblastomas in relation to genomic profiles and clinical parameters resulted in a number of new important observations. First, substitutions at residue F1174, one of the two hotspot mutations, were significantly overrepresented in MYCN-amplified tumors. Moreover, patients with the F1174 mutation present with a particularly poor outcome. Second, our results indicate that, although both hotspot mutations have constitutive ALK phosphorylation, the F1174L mutation has stronger autophosphorylation and transformation capacity of Ba/F3 cells than does the R1275Q mutation. Third, in contrast to some previous reports, we show that ALK mutations also occur in a significant proportion of tumors with favorable stages 1, 2, and 4S (5.7% versus 7.5% in stage 3 and 4 tumors). Fourth, we show that copy number gain of the chromosome 2 region encompassing ALK is associated with an increased ALK expression. Fifth, increased ALK expression is associated with a worse outcome in the global population. Finally, we show that chromosome 2p is not frequently gained in tumors with ALK mutations, indicating that mutated ALK alleles are not selected for high expression by copy number gain.

The functional relevance of the high proportion of F1174 mutations in the subset of MYCN-amplified ALK-mutated neuroblastomas remains undetermined, but their co-occurrence may suggest a cooperative effect between both aberrations in these tumors. The F1174 mutation might contribute to an additional growth and survival benefit in MYCN-amplified neuroblastoma cells, which may explain the particularly poor survival of these patients. The fact that neuroblastoma cell lines with MYCN amplifications have a relatively high frequency of F1174 mutations might also point at a particular growth advantage that may have facilitated in vitro growth of these cells. Of further interest in this context is the observation that F1174 mutations in the germline have not been reported up to now, which could suggest embryonic lethality. Of particular interest was the observation that only 1 of the 10 patients with MYCN amplification together with a F1174 mutation survived, in contrast to a 32% 5-year overall survival rate in patients with MYCN-amplified tumors without the F1174 mutation.

The different distribution across the genomic subtypes and the adverse impact of the F1174 mutations on survival raise the question of whether the F1174 and R1275 mutations may execute distinct effects on tumor biology. George and colleagues have previously also shown that both F1174L and R1275Q mutants could transform Ba/F3 cells, but their analysis did not reveal major differences in oncogenic potential between these two mutants, in part because the proliferation data were not reported in detail (7). In our hands, the F1174L mutant transformed the Ba/F3 cells more efficiently than did the cells expressing the R1275Q mutant. This correlated with higher autophosphorylation levels of ALK F1174L, which was not observed for the R1275Q mutant.

The occurrence of mutations of particular genes in relation to genomic subgroups has been reported in certain tumor entities, such as PIK3CA mutations in head and neck squamous carcinomas without EGFR amplification and β-catenin mutations in medulloblastomas with loss of chromosome 6 (18–20). However, to the best of our knowledge, a different distribution for mutations within the same functional domain of one specific gene, as observed here for the F1174 mutation in ALK, has not been reported.

Our study also showed, in contrast to some of the initial studies, that ALK mutations occur in fairly equal frequency in both low- and high-stage tumors. Therefore, mutation analysis should also be done in patients with low-stage tumors, and the clinical characteristics and behavior of such tumors should be carefully monitored in further studies.

In addition to mutations, ALK activation can also result from high level gene amplification as shown by previous studies (8–10). Meta-analysis showed that this is a recurrent but rare mechanism, detected in only 1.7% of the cases. In keeping with previous studies, amplification of ALK almost exclusively occurs in MYCN-amplified tumors. Apart from such rare ALK high-level amplification, high-stage neuroblastoma tumors often exhibit gain of a large part of 2p that mostly encompasses the ALK locus. In our cohort, mutation analysis showed that only a minority of tumors with 2p gain carried ALK mutations, although 2p gains are present in as much as 19.3% of all neuroblastomas with segmental imbalances. This observation indicates that 2p gain does not act as a mechanism for increased copy number of mutated ALK, in contrast to what has been described for other oncogenes in other tumor entities (21, 22). In view of the variability of centromeric break points for 2p gains but almost consistent presence of ALK in these segments, one could assume that low copy number gain of ALK could also imply a growth or survival advantage for neuroblastoma cells. To test this hypothesis we analyzed the relation between ALK copy number and expression and the impact of increased ALK gene expression on survival. ALK gene expression was indeed shown to be copy number sensitive, and increased ALK expression correlated with poor survival. Using immunostaining, Passoni et al. (2009) recently showed that ALK overexpression correlated with patient survival although no correlation was found between mRNA and protein expression in their tumor cohort (23), which is in contrast with our observations in cell lines. Our findings, together with those of Passoni et al. (2009), indicate that increased ALK expression might be functionally relevant. Therefore, patients with increased ALK expression might benefit from future clinical trials with ALK inhibitors.

In conclusion, this meta-analysis shows for the first time that the recurrent F1174 mutation predominantly occurs in MYCN-amplified tumors, and clearly shows differences in the frequency and distribution of ALK mutations across the different genomic subtypes in neuroblastoma. The F1174 mutation might suggest a poor prognosis in patients with MYCN amplification, but further studies are needed to substantiate this hypothesis. Furthermore we could show that the F1174L mutant displayed a higher degree of ALK phosphorylation and tumorigenicity than did the R1275Q mutant. No significant difference was observed in the frequency of ALK mutations between low- and high-stage tumors.

No potential conflicts of interest were disclosed.

We thank Justine Nuytens for excellent technical support.

Grant Support: EU under FP6 (EET-pipeline, nr.037260 and the Kids Cancer Kinome program, nr.6037390), FWO (G.0198.08), IWT (SBO60848), GOA (01G01910). This article presents research results of the Belgian program of Interuniversity Poles of Attraction, initiated by the Belgian State, Prime Minister's Office, Science Policy Programming, Methusalem-program [BOF08/01M01108], FOD (NKP_29_014), Stichting Kindergeneeskundig Kankeronderzoek, the KIKA Foundation. K. De Preter is postdoctoral researcher with the FWO. R. Noguera is supported by ICIII RD06/0020/0102. C. Kumps and M. Porcu are supported by the IWT, J. Cools and P. Zabrocki are supported by Foundation against Cancer (SCIE2006-34, J.C.) and J. Hoebeeck is a postdoctoral research supported by a grant of the Ghent University (BOF01P07406) and by the fund for Scientific Research Flanders (KAN1.5.207.08).

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.