Deep Sequencing Reveals Occurrence of Subclonal ALK Mutations in Neuroblastoma at Diagnosis

Intratumor heterogeneity has been shown to occur in the majority of human malignancies, including chromosomal alterations or gene mutations, and has been shown to play a role in clonal evolution (1–6).

With fractions of mutated alleles evolving over time, the study of the exact mutated allele fractions, of their subclonal distribution and possible role in clonal evolution gains more importance (7–9).

In neuroblastoma, the most frequent extracranial solid cancer of early childhood, efforts to develop improved therapies must also take into account its molecular heterogeneity. Genetic alterations in neuroblastoma at diagnosis mainly concern copy number alterations, with MYCN amplification in 20% to 25% of cases, and segmental copy number alterations involving more extensive chromosome regions (10–14). Genome sequencing studies of neuroblastoma at diagnosis have revealed a low mutation rate involving only few recurrently altered genes mainly involved in chromatin-remodeling or neuritogenesis (15–17).

Activating anaplastic lymphoma kinase (ALK) mutations have been shown to occur in both familial and sporadic cases of neuroblastoma, with somatically acquired ALK mutations observed in 6% to 12% of sporadic neuroblastomas (15, 16, 18–21). The most frequent substitutions, observed in approximately 85% of mutant ALK in neuroblastoma, are localized within the kinase domain of ALK at the F1174 (mutated to L, S, I, C, or V) and R1275 (mutated to Q or L) hotspots (12, 21). In neuroblastoma, ALK can also be activated by genomic amplification (approximately 1%–2% of neuroblastomas) or more rarely following structural rearrangements (22).

Activating ALK mutations are thought to play a role in neuroblastoma oncogenesis based on in vitro and in vivo observations, in some instances through collaboration with MYCN (23–25). Thus, ALK might represent a bone fide target in neuroblastoma therapy. We have recently used deep sequencing to search for ALK mutations in neuroblastoma relapse samples (26). Furthermore, in a recent study, the feasibility of a new method of ALK mutations detection in circulating tumor DNA (ctDNA) using droplet digital PCR (ddPCR) has been described. This method showed the presence of the F1174L and R1275Q ALK mutations in circulating DNA from 21% of neuroblastoma patients at diagnosis (27). Altogether, these studies highlight the importance of next-generation deep sequencing techniques when determining the genetic status of this predictive biomarker.

In order to detect ALK mutations with a high sensitivity and to determine the frequency of subclonal events in neuroblastoma, we have now analyzed a series of 276 diagnostic neuroblastoma samples using targeted deep sequencing.

Patients with neuroblastoma for whom a tumor sample was addressed to the laboratory for diagnostic molecular analysis were included in this study to constitute a representative neuroblastoma patient cohort. For a total of 276 patients, 125 patients with stage IV disease (46%), 123 with localized disease (44%), 25 with stage 4s disease (9%), and 55 patients with MYCN amplification (20%) were included (Supplementary Table S1). Pathologic examination by hematoxylin and eosin (H&E) staining method documented at least 50% of tumor cells in all samples.

MYCN status and tumor genomic copy number profiles were determined as described previously, using an in-house 4k or a commercial 72k platform (Nimblegen; ref. 6).

Patients were treated in French centers according to the relevant national or international protocols. Written informed consent was obtained from parents according to national law and Ethics approval of protocols was obtained according to national guidelines. This study was authorized by the ethics committees “Comité de Protection des Personnes Sud-Est IV,” references L07–95/L12–171, and “Comité de Protection des Personnes Ile de France,” reference 0811728.

Twenty-four germline genomic DNAs from CEPH (Centre d'Etude du Polymorphisme Humain) from healthy donors served as controls.

For sequencing library construction, 50 ng of genomic DNA from each sample were used to amplify the ALK regions of interest (exon 23:chr2:29443647–29443776; exon 25: chr2:29432603–29432704). The regions containing ALK hotspots F1174 (exon 23) and R1275 (exon 25) were amplified in a two-step PCR procedure using the Phusion Hot Start II high fidelity DNA polymerase (Thermo Scientific). During the first PCR round, region-specific primers with adapter overhang on both forward and reverse primers (Supplementary Table S2) were used for a total of 10 cycles. The second PCR round, for a total of 30 cycles, attached Illumina single direction primers at the ends of amplicons. The adapter sequences and the unique barcodes contained in the Illumina primers were added at the 3′ and 5′ ends, respectively [Access Array Barcode Library for Illumina Sequencers 384 (Single direction) P/N 100-4876]. Following clean-up using the vacuum ultrafiltration kit (Macherey Nagel), control of size range using LabChip devices (Caliper, PerkinElmer) and Qubit quantification, PCR products of all samples were then pooled together in an equimolar solution: the cases to be studied, the germline controls, and negative controls. The final pooled library was then loaded on the HiSeq2500 flow cell.

The library was sequenced using the paired-end procedure on the Illumina Hiseq2500 according to the manufacturer. Sample barcoding enabled 384 individual samples to be sequenced on each lane, aiming for a depth of coverage of at least 16,000X for the amplicons in every sample (Fig. 1).

Once paired-end reads merged and adaptors trimmed by SeqPrep with default parameters, merged reads were aligned via BWA allowing up to one difference in the 22-base-long seeds and reporting only unique alignments (26).

Variant calling software was not used, since we aimed to predict also variations at low frequencies. Such variants require a custom approach using DepthOfCoverage functions of the Genome Analysis Toolkit (GATK) v2.13.2 and additional statistical analysis (28).

In order to focus on high-quality data, only reads with a mapping quality of >20 and a base quality of >10 were considered for the determination of the depth of coverage. First, in order to analyze the noise and to determine the background level of variability over the studied regions, the entire amplicons chr2:29443647–29443776 and chr2:29432603–29432704 (Human Genome Browser, http://genome.ucsc.edu/; hg19) were analyzed in 24 germline control samples. Next, in order to highlight variants in the neuroblastoma samples, for each sample the frequencies of each base at each position were compared with those observed in the germline controls. Statistical analyses were performed with the R statistical software (http://www.R-project.org). Fisher exact two-side tests with a Bonferroni correction were performed to compare percentages of base frequencies (allele fractions) between the datasets, that is, for a given base between a case and the controls. It takes into account for each coordinate and each base the impact of its environment since controls were sequenced together with cases for comparison. In complement, in order to limit false-positive predictions of variants, we kept as P value threshold the lowest one within intronic regions of the amplicons once polymorphisms filtered, since no causal variants were expected in those regions (exon 23: 1.1E−154 and exon 25: 6.1E−193, respectively). Finally, significant variations were filtered-in once (i) a significant increase in the percentage of variant base and (ii) a significant decrease in the percentage of its reference base following our P values criteria were observed.

The experiments were performed twice in an independent manner with the same approach. Only results predicted twice were considered.

Correlation analyses using χ² test and survival analysis using multivariate logistic regression analysis were done with MedCalc (Medical Calculator) software (version 13.3.0.0). Progression-free survival (PFS) and overall survival (OS) were estimated using the Kaplan–Meier method, and comparisons were made using log-rank tests.

In order to determine the frequency of ALK mutations with a higher sensitivity than conventional Sanger sequencing, we sequenced the ALK F1174 and R1275 hotspots in a series of 276 neuroblastoma samples in a two-step PCR procedure using HiSeq technology with a very high depth of coverage (Supplementary Table S1).

In a first step, the sensitivity of the technique across a series of 24 germline controls was determined. The mean overall depth of coverage for the germline controls was 33,000X (range, 25069–56620) for the exon23 and 41,000X (range, 25062–56639) for the exon25 amplified region, respectively (Fig. 1). At this deep coverage, within targeted regions, the mean overall background variability, or noise, was 0.014% ± 0.031% in the germline controls (Supplementary Fig. S1).

To determine the expected sensitivity, we then calculated for each position the number of variation-supporting reads that would result in a statistically significant difference from the background noise. Bonferroni correction was applied, as multiple tests were performed for each base at each position. Considering a mean coverage of 33,000X, a variation supported by 688 reads or more, or observed with a frequency higher than 0.21%, would result in a statistically significant difference from the controls (two-sided Fisher exact test; details in Materials and Methods).

In the studied neuroblastoma samples, the depth of coverage and the overall background variability was not different from that observed in the germline controls. Furthermore, except a variant reported in dbSNP and in the 1000Genomes project (PMID:11125122) as a polymorphism (rs3738868) observed in more than 1% of the population (chr2:29432625, G>T), no variants outside the studied hotspots were observed in the targeted regions.

In a next step, the mutational status of the ALK gene within exons 23 and 25 at the hotspots F1174 and R1275 was analyzed. At positions chr2:29443695–29443697 (F1174 hotspot), ALK mutations were observed in 15 cases: 13 cases harbored a mutation leading to the amino acid change F1174L; F1174C and F1174V were each detected in one case, with a range of fractions of the mutated allele of 0.562% to 40.409% (Table 1, panel A). At position chr2:29432664 (R1275 hotspot), ALK mutations were detected in 12 cases: 11 samples showed the R1275Q mutation and one case showed the R1275L mutation, with a range of mutated allele fractions from 0.811% to 73.001% (Table 1, panel B).

Altogether, deep sequencing revealed the presence of ALK mutations at the two hotspots F1174 and R1275 in 27/276 (9.8%) samples (Fig. 2; Table 2), with a wide range of mutated allele fractions (range: 0.562%–73.001%), even when corrected for tumor cell content and chr2p copy number status (Table 2 and Supplementary Table S1).

Indeed, in this study, contamination with normal cells should be considered. For this study, contamination of tumor samples by normal cells of up to 50% was tolerated, and thus it is expected that in these samples heterozygous mutations would be present in 25% of all analyzed DNA fragments in a diploid context when occurring in all tumor cells. Thus, we use the term “clonal” for a mutated allele fraction > 20%, which would most likely concern a majority of tumor cells, and the term “subclonal” for a mutated allele fraction < 20%, which would most likely concern a smaller tumor cell population. On the basis of this definition and on the deep-sequencing results, 12 ALK mutations observed with an allele fraction higher than 20% might be considered as clonal, whereas 15 ALK mutations detected with an allele fraction of lower than 20% might be considered as subclonal events.

All deep-sequencing results were validated in a second independent experiment. Concordance between the first and second experiments was documented by (i) concordance of mutated allele fractions in the neuroblastoma samples between the two separate experiments and by (ii) concordance of frequencies of the allele of the rs3738868 polymorphism (Supplementary Fig. S2). Only in two of 276 cases was a discordance observed between the first and the second experiment and these cases were excluded from the analysis, both detected at very low mutated allele fractions in one experiment but not the other. These results indicate a very low false discovery rate of 2/276 (0.7%).

In addition, in two of 257 (0.8%) samples with available genomic copy number profiles, an ALK gene amplification was detected. One of these samples (NB1418T) also showed a R1275Q mutation (mutated allele fraction 73%; Table 2).

All ALK mutated samples were also tested by Sanger sequencing which confirmed all ALK mutations occurring at a clonal level with a mutated allele fraction >20%. Because of the limits of detection and due to the background noise generally present in the Sanger elecropherograms, ALK mutations would not have been retained by Sanger sequencing in subclonal cases.

A correlation between the presence of an ALK aberration (mutations and/or amplification) and MYCN amplification was observed (Table 3, panel A), with an enrichment of F1174 ALK mutations in tumors harboring MYCN amplification (P < 0.0001). There were no other statistically significant correlations with clinical parameters (Table 3, panel A).

Furthermore, no statistically significant correlations between the ALK status and main clinical parameters were observed among patients whose tumors harbored ALK mutations either at a clonal or subclonal level, respectively.

FISH analysis of MYCN status correlated with the tumor cell content determined by H&E staining, and in no instance was heterogeneous MYCN amplification observed. In particular, in cases with ALK mutations detected at a subclonal level, MYCN status determined by FISH was homogeneous throughout all tumor cells, and in amplified cases concerned all tumor cells (>50% of cells in the sample).

The overall survival of patients with ALK wild-type versus ALK-mutated and/or ALK-amplified tumors was compared, and Kaplan–Meier analysis showed a statistically significant poorer OS in patients whose tumors harbored an ALK aberration (P < 0.02; Fig. 3A). However, no statistically significant differences were found in PFS between these two patient groups. A statistically significant difference in OS was observed between patients with ALK F1174-mutated tumors versus all other patients (P = 0.0014; Fig. 3B). The comparison of survival of patients with ALK wild-type or ALK aberrations, with or without MYCN amplification showed a poorer OS in patients whose tumors harbor MYCN amplification, with or without ALK aberration (Fig. 3C).

Multivariate analysis of OS using a Cox's regression model further confirmed that in this series, independent predictors of poor outcome are INSS stage IV disease and MYCN amplification, and that ALK mutational status does not add prognostic information to the clinical parameters “INSS stage” and “MYCN status” (Table 3, panel B). No statistically significant differences of OS and PFS in patients with ALK mutations at a clonal versus subclonal level were observed (log-rank test: P = 0.2 and P = 0.98, respectively).

In order to determine the frequency of ALK mutations in diagnostic neuroblastoma samples with a higher sensitivity than conventional sequencing methods, we have used ultra-deep sequencing to analyze a large series of representative neuroblastoma samples (Supplementary Table S1).

Traditional Sanger sequencing has been widely used in clinical laboratories for mutation testing, but the sensitivity is limited to the detection of 20% to 30% of mutated alleles in a wild-type background (25). A higher sensitivity has been shown for droplet digital PCR (ddPCR) and NGS techniques (25, 27). Indeed, a limit of detection of 2% of mutated allele fractions has been evidenced for BRAF mutations using NGS methods (29). We have reported on a sensitivity limit of 0.17% using PGM technology for detection of ALK mutations in a series of diagnostic and relapse neuroblastoma samples (30).

Different algorithms, such as ABSOLUTE, have been described to detect mutations with low mutated allele fractions (31). However, these algorithms have not been developed specifically for ultra-deep sequencing data. Thus, in this study, a custom approach has been used to define mutations with very low mutated allele fractions in deep-sequencing data.

In this study, the targeted sequencing method achieved a very high depth of coverage over the region of interest (mean coverage 33,000X). On the basis of this coverage, a mutated allele fraction of 0.21% or more was considered significantly different from the germline controls, which is well below the lower limit defined in the technical specifications of Hiseq Illumina. This calculated sensitivity is considered relevant only if (i) analyzed data correspond to similar depth of coverage, (ii) sequencing is performed with the same technology, (iii) controls show similar low variability, and (iv) the targeted region has a base environment similar to the regions analyzed in this experiment. This detection limit also corresponds to the theoretical limit of detection based on the DNA input for one experiment (50 ng of DNA, equivalent to approximately 5,000 diploid genomes).

Between the different mutation types, a higher mean mutated allele fraction was observed in the R1275 cases (mean of mutated allele fractions: 11.6 vs. 29.5 in F1174 vs. R1275 cases, respectively; P = 0.02, t test). When considering chr2p status, in cases with three copies of chr2, the mutated allele most likely concerns the non-duplicated chr2p for F1174, and the duplicated chr2p for R1275 (Table 2). This might further support the hypothesis that R1275 mutations are less aggressive than F1174 mutations, and that more copies of the R1275 mutated allele are necessary to transfer selective advantage to the tumor cells, as previously suggested (32, 33).

Previous studies describing ALK mutations in neuroblastoma samples, frequently analyzing the whole tyrosine kinase domain, have used conventional sequencing methods, such as Sanger sequencing, or whole-genome/whole-exome sequencing with standard resolution. The incidence of ALK mutations observed in these studies varies from 6% to 12% of all neuroblastoma cases (15, 16, 18–21). In our cohort, studying only exons 23 and 25 of the ALK gene, 27 of 276 neuroblastoma (9.8%) presented an ALK mutation. An incidence of only 4.3% would have been observed by Sanger sequencing, with the incidence doubling when moving to ultra-sensitive deep sequencing.

To date, in neuroblastoma, the co-occurrence of an ALK hotspot mutation and amplification has been reported only for a cell line: CLB-GE harbors both the F1174V mutation and an ALK high level amplification (15). In this series, one sample had both ALK amplification and a R1275 mutation, indicating that, although rare, these alterations are not mutually exclusive (Table 2). A mutated allele fraction of 73% in the context of a genomic amplification of >10 copies suggested that the ALK gene underwent amplification before an R1275 mutational event with subsequent further amplification rounds.

ALK F1174L mutations have been reported in a higher frequency in MYCN-amplified tumors (19). The co-occurrence of ALK F1174L and MYCN amplification probably represents a cooperative effect between both alterations in neuroblastoma, which might explain the particularly poor survival of these patients (33, 34).

In vitro and in vivo studies have indicated the efficacy of ALK inhibitors on cells harboring ALK mutations at a clonal level (21, 28). It now remains to be determined whether tumor populations composed of subclones harboring ALK R1275 or F1174 mutations have different growth properties and different responses to ALK inhibitors in in vitro and in vivo experimental systems.

The observation of ALK-mutated subclones at diagnosis suggests the coexistence of ALK nonmutated and ALK-mutated cells, which might coexist in an advantageous equilibrium and might crucially affect the dynamics of cancer progression at a later stage. Cooperation of different tumor cell subsets has been reported to contribute to the malignant cancer phenotype (30, 32). Single cell experiments and in situ approaches will enable to elucidate how ALK-mutated cells are distributed throughout a neuroblastoma tumor and how these cells might potentially collaborate with cells with different genetic alterations.

A higher frequency of ALK mutations at relapse of neuroblastoma has recently been reported, with the possibility of expansion of a minor ALK-mutated subclone at diagnosis to a dominant ALK-mutated clone at relapse (26). Further detailed sequencing studies in larger series of diagnosis-relapse samples will be required to elucidate the role of subclonal ALK mutations in clonal evolution and progression of neuroblastoma.

In conclusion, given the oncogenic role of ALK, the possibility of evolution of ALK mutated clones from diagnosis to relapse and the possibility to target ALK mutations with ALK inhibitors such as crizotinib (PF-02341066) or other second-generation ALK inhibitors such as LDK378 (Novartis Pharmaceuticals), the determination of the exact genetic ALK status in all neuroblastoma is crucial (21, 29). We clearly demonstrate the advantages of deep-sequencing NGS technology for the detection of mutational events in a common pediatric cancer. Our approach, relative to conventional methods, has shown several advantages, including (i) in vitro library construction of a large number of samples, (ii) high resolution, (iii) short time analysis, and (iv) high accuracy. Owing to the higher sensitivity of deep sequencing, it is very likely that, in the near future and with growing experience, these techniques will replace conventional methods for mutation detection.

The high-resolution technique identifies clonal and subclonal mutational events, thus doubling the number of identified ALK mutation events compared with those detected with conventional sequencing techniques. To date, no formal recommendation for targeted therapy in a context of low subclonal events can be given. Indeed, in other pathologies, targeted therapies such as dabrafenib for V600E/K mutations in melanoma are most frequently proposed in a context of mutation detection using standard resolution techniques which do not reveal lower mutated allele fractions below 5% to 10% (35). Our findings now highlight the importance of further functional studies analyzing the interactions between ALK-mutated and nonmutated subclones on the one hand, and the possibilities of clonal evolution on the other hand. Such studies should especially focus on the analysis of readily accessible surrogate samples, such as serial blood samples for the study of ctDNA.

No potential conflicts of interest were disclosed.

Conception and design: D. Valteau-Couanet, O. Delattre, G. Schleiermacher

Development of methodology: A. Bellini, Q. Leroy, T. Rio Frio, G. Schleiermacher

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Q. Leroy, G. Pierron, V. Combaret, E. Lapouble, H. Rubie, E. Thebaud, C. Bergeron, N. Buchbinder, S. Taque, A. Auvrignon, J. Michon, I. Janoueix-Lerosey, G. Schleiermacher

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Bellini, V. Bernard, C. Bergeron, I. Janoueix-Lerosey, G. Schleiermacher

Writing, review, and/or revision of the manuscript: A. Bellini, V. Bernard, G. Pierron, V. Combaret, P. Chastagner, A.S. Defachelles, C. Bergeron, D. Valteau-Couanet, O. Delattre, G. Schleiermacher

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Bellini, G. Pierron, E. Lapouble, N. Clement, O. Delattre, G. Schleiermacher

Study supervision: G. Schleiermacher

The authors thank the following colleagues for their contribution to this study: Catherine Devoldere, Paul Fréneaux, Vannina Giacobbi-Milet, Hayet Hanbli, Philippe Le Moine, Odile Minckes, Isabelle Pellier, Michel Peuchmaur, Dominique Plantaz, Emmanuel Plouvier, and Nicolas Sirvent.

This work was supported by the Annenberg foundation and the Ligue Nationale Contre le Cancer (équipe labellisée). Funding was also obtained from SiRIC/INCa (Grant INCa-DGOS-4654) and from the CEST of Institute Curie and by the Associations Enfants et Santé, Association Hubert Gouin Enfance et Cancer, Les Bagouz à Manon, Les amis de Claire. Next generation experiments (NGS) were conducted on the Institute Curie's ICGex NGS platform funded by the EQUIPEX “investissements d'avenir” program (ANR-10-EQPX-03) and ANR10-IneuroblastomaS-09-08 from the Agence Nationale de le Recherche, and by the Canceropôle Ile de-France.

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.