Molecular Screening for Cancer Treatment Optimization (MOSCATO-01) in Pediatric Patients: A Single-Institutional Prospective Molecular Stratification Trial

The technical evolution to perform exhaustive pangenomic molecular characterization of tumors at an unprecedented level has opened the door for alternative, more tumor-biology–based and potentially “individualized” treatment approaches. Pediatric cancers constitute a unique group in oncology due to their cellular source, their pathogenesis and their relative rareness, which makes it difficult to generate relevant case numbers for exhaustive investigation. Today, truly “actionable” molecular alterations with proven therapeutic benefit seem to be highly restricted to selected diagnosis subgroups. Several examples have been reported in pediatric malignancies that are currently explored in early clinical trials and highlighting the importance of detecting oncogenic drivers or specific molecular alterations in a particular patient: the inhibition of the BCR/ABL fusion in leukemias (1), anaplastic lymphoma kinase (ALK) gene alterations in anaplastic large cell lymphoma and inflammatory myofibroblastic tumors (2–5), BRAF in BRAFV600–mutated gliomas and Langerhans cell histiocytosis (6, 7), and very recently tropomyosin receptor kinase (TRK) fusion-positive cancers leading to the accelerated FDA approval of larotrectinib for adults and children (8, 9). However, efforts to systematically profile relapsed or resistant pediatric tumors at a large scale have just begun. With growing numbers of patients analyzed, hopefully this will lead to further characterization of subgroups which might benefit from targeted treatment. Pediatric tumors differ markedly from adult cancers in terms of the tissue they originate from and by far a lower non-synonymous mutation rate (10, 11). Nevertheless, the experience that has been gained by providing innovative, biology-driven treatments to adult patients with cancer is of great interest for the childhood cancer community as well.

The Molecular Screening for Cancer Treatment Optimization (MOSCATO-01; NCT01566019) trial aimed at implementing genomic analysis in the clinical management of patients with cancer with recurrent/resistant tumors by performing intentional tumor biopsies at recurrence or progression (12). The identification of targetable individual molecular traits would theoretically enable treatment by a matching drug, preferably within the frame of early phase clinical trials. MOSCATO-01 was the first prospective clinical trial to perform dedicated biopsies in children with cancer in a large European cancer center.

This report presents the results of the pediatric cohort and shows the feasibility and interest of integrating extensive molecular screening into the clinical care of selected pediatric patients. It provides further evidence for the need for clinical studies to safely access novel agents and to better define the genomic alterations for which treatment with specific targeted agents translates to clinical benefit.

The trial recruited patients with incurable, relapsed or resistant solid malignancy. The study was amended in 2012 to include patients aged above 6 months as well as children with central nervous system (CNS) tumors. Patients were required to have received at least one prior treatment line, to intervention accessible disease, life expectancy of >3 months, performance scale of the Eastern Cooperative Oncology Group (ECOG) 0/1 or Lansky play scale ≥70%, adequate bone marrow and organ function. Informed consent of patients and parents was obtained before the biopsy or surgical intervention.

MOSCATO-01 (NCT01566019) is a single-institution, non-randomized, prospective trial, which aims to show that clinical application of high-throughput genomics leads to improved outcome in patients with metastatic cancers. For the adult group, the primary objective was to evaluate clinical benefit measured from patients presenting a PFS on matched therapy (PFS2) longer than 1.3-fold the PFS on prior therapy (PFS1). For the pediatric cohort, the study design was descriptive due to the expected sample size in a heterogeneous population.

Following informed consent, tumor samples was collected by resection or CT, MRI, or ultrasound-guided intentional tumor biopsy (13). Biopsy intervention in patients harboring CNS tumors and all surgical resections took place in Hôpital Necker-Enfants Malades, Paris. All other procedures and tumor sample processing were done at Gustave Roussy. Scientists, geneticists, pathologists, and clinicians reviewed the output of the molecular analysis on a weekly basis to determine the biological significance of the alteration and to match the patients to the most relevant targeted therapy available. The definition of “actionable” in this publication refers to a detected molecular alteration or affected pathway in the patient's tumor and/or germline analysis which theoretically would be targetable by an approved drug or an investigational agent in any phase of clinical development, either directly or indirectly in the affected pathway. Known pathogenic or oncogenic variants and copy number changes had the strongest impact on target definition. Alterations predicted to alter protein quality or quantity were chosen as targets individually case-by-case after exhaustive literature search and if an appropriate drug was available. Prioritization was applied based on the strength of evidence of the molecular alteration and the accessibility of a targeted treatment.

The study was approved by an independent ethics committee, the national medical authorities, and conducted according to the ethical principles of the Declaration of Helsinki.

In all collected specimens, the percentage of tumor cells was determined by an experienced pathologist. The most representative sample with the highest tumor cell content was chosen for the molecular analysis. Tumor cell content was required to be ≥30% for further processing; targeted gene panel sequencing (TGPS) was still performed for samples with more than 10% of tumor cell content (14).

Tumor DNA, RNA, and germline DNA from whole blood samples were extracted using AllPrep DNA/RNA mini kit and DNeasy Blood and Tissue Kit, respectively, according to the manufacturer's instructions (Qiagen). Molecular analysis using TGPS was carried out as previously described (14), using Ion AmpliSeq workflow with a customized 75 gene panel with library preparation by multiplex-PCR and sequenced with Personal Genome Machine (PGM, Ion Torrent) according to the manufacturer's recommendation (Thermo Fischer Scientific, Courtaboeuf, France). Details of the gene panel, bioinformatics analysis and variant interpretation are provided in Supplementary Methods.

CGHa analysis was performed as previously described (15). Briefly, DNA was labeled and hybridized on using SurePrint G3 Human aCGH Microarray 4 × 180 K (Agilent technologies) or Affymetrix CytoScan HD array/Oncoscan (Affymetrix) according to the manufacturer's recommendations. The data from microarray scans were extracted with Feature Extraction using default parameters (v10.5.1.1, Agilent Technologies) and Chromosome Analysis Suite (ChAS; v3.1, Affymetrix), respectively. Then data were analyzed and annotated with our own bioinformatic pipeline described in Supplementary Methods.

Whole exome was captured from 400 ng of tumor and paired constitutional DNA using Agilent SureSelect V5 (50 Mb) or Clinical Research Exome (54 Mb) kit. Sequencing of subsequent libraries was performed using Illumina sequencers (NextSeq 500 or Hiseq 2000/2500/4000) in 75 bp paired-end mode. Bioinformatics processing is detailed in Supplementary Methods. The mutational load was calculated by dividing the number of somatic non-synonymous mutations by the number of bases having a depth greater than or equal to 4 in the tumor BAM file.

Libraries were prepared with TruSeq Stranded mRNA kit following recommendations. The key steps consist of PolyA mRNA capture with oligo dT beads 1 μg total RNA, fragmentation to approximately 400 bp, cDNA double strand synthesis, and ligation of adaptors, library amplification and sequencing. Sequencing was performed using Illumina sequencers (NextSeq 500 or Hiseq 2000/2500/4000) in 75 bp paired-end mode. For the optimized detection of potential fusion transcripts an in-house designed metacaller approach was used (details on sequencing and bioinformatics in Supplementary Methods; refs. 16–20).

All molecular analysis results, from TGPS, CGHa, WES, or RNAseq were reviewed one by one, by both a molecular geneticist responsive of generating molecular report highlighting annotated molecular abnormalities to be discussed in Molecular Tumor Board and a pediatric physician-scientist.

Seventy-five pediatric patients with recurrent or refractory solid tumors were included (Flowchart Fig. 1). One patient withdrew consent, and one had no confirmed tumor lesion. Of the 73 patients who underwent intervention, 33 were male (45.2%), median age was 10.9 years (range, 0.8–24.3); 11 patients (15.1%) were ≥18 years at inclusion. The median performance status was Lansky Play scale 90% and ECOG 0. The disease spectrum comprised 25 different entities, 37.0% of patients with CNS tumors, 63.0% with extra-cranial solid tumors (Table 1). Five patients harbored a suspected or known cancer predisposition [two had neurofibromatosis type 1 (NF1) and Li-Fraumeni syndrome, one Gorlin-Goltz syndrome]. One patient with Li-Fraumeni syndrome also had Williams-Beuren syndrome. Patients had a median of two prior lines of treatment (range, 1–8) and a median time since initial diagnosis of 26.8 months (range, 2.4–153.1).

A total of 79 interventions were performed, in 63.3% on tumor metastases. Fifty-one patients (69.9%) underwent 56 image-guided interventions (ultrasound: 24.1%, CT 26.6%, MRI 21.5%). Twenty patients (27.4%) had 21 surgical resections, two had blood or bone marrow samples. In one patient with embryonal rhabdomyosarcoma (eRMS, patient #26) a relapsed primary and metastatic site were biopsied, five patients had sequential interventions due to low tumor cell content in the first biopsy or new tumor progression (#19, 24, 27, 35, 41).

The complication rate of intervention was 7.6%; four of the six events were following a research biopsy, two grade 1 alveolar hemorrhages, and two grade 3 respiratory events that were related to advanced disease status. Pathological analysis confirmed the initial diagnosis in all but two patients with previous medulloblastoma, whose biopsy revealed a high-grade glioma (#10, 13). Median turnaround time between biopsy/surgery and molecular results (CGHa & TGPS) was 26 days [interquartile range (IQR), 19–41 days].

Histologically selected analyzed specimens had a median tumor cell content of 70% (range, 0%–100%), 60% on the research samples. According to the minimum threshold for undergoing genomic analysis, 69 of the 73 patients had a molecular analysis: 66/73 (90.4%) had CGHa and TGPS, 3 (4.1%) had TGPS only; in four patients (5.5%) neither analysis could be performed. Median sequencing depths was over ×700 for hotspot variants. Fifty patients had additional WES (n = 48) and/or RNAseq (n = 48), 46 patients underwent both. Median million reads obtained for normal/tumor WES samples was 117 and 175 and median coverage ×86 and ×125, respectively. Quality data of the WES and RNAseq analyzes are presented in Supplementary Table S1.

Genomic alterations that are currently considered as clearly defined biomarkers associated with proven clinical benefit of a registered treatment in the disease were not detected in any of our patients. Molecular events that are known oncogenic drivers in the underlying or other diseases and that are currently explored in a clinical setting in childhood cancers and for which early activity signal have been reported have been found in four cases: BRAFv600 mutation in a low-grade glioma (#45), SMARCB1 deletion and INI1 expression loss in an epitheloid sarcoma (#35), ALK p.R1275Q mutation in a neuroblastoma (#68), PTCH1 p.Asp301IIefs*23 mutation in a Gorlin-Goltz associated medulloblastoma (#66). Overall, we defined molecular alterations with potentially actionable implications in 42 of the 69 patients (60.9%) with at least one analysis. All potentially targetable and other reportable alterations are presented in the Oncomap (Fig. 2; Supplementary Table S2). 42.4% of patients harbored a target detected by copy-number analysis, 33.3% had an actionable mutation, 14.5% had both. Twenty-six percent had more than one actionable alteration. Where applicable, all alterations detected by CGHa, and the TGPS were confirmed by WES. In nine patients, WES and RNAseq revealed potential new targets, mainly corresponding to mutations which were not included in the target gene panel, rarely due to insufficient coverage in the TGPS. We detected or confirmed disease-defining gene fusions in 12 patients, one of which resulted in histological review and eventually changed from the presumed low-grade glioma to an ependymoma (#48). One patient with an alveolar soft-part sarcoma (ASPS; #21), characterized by ASPSCR1-TFE3 fusions that have been functionally shown to lead to strong aberrant overactivation of MET (21), had already undergone a therapeutic approach by MET inhibitor crizotinib before inclusion into the study. We did not observe any additional fusion events that would have been therapeutically accessible, for example, by constitutive kinase activation.

Germline sequencing of 48 patients confirmed the presence of a cancer-predisposing germline alteration in four of the five patients who were suspected or proven before with a cancer-predisposition syndrome; one patient with NF1 did not undergo WES, but the targeted sequencing was highly suggestive of the underlying germline mutation in NF1 (#14). This accounts for a rate of 10.2% in this cohort. No other incidental findings according to American College of Medical Genetics (ACMG) Recommendations were detected in the remaining patients (22). This is in keeping with prevalences reported in the literature of 8.5% in a series of 1,120 pediatric patients with cancer (23) and in two pediatric clinical sequencing trials (refs. 24, 25; 10 and 12%, respectively), thereby emphasizing the importance of incidental causes in childhood cancer.

The median rate of non-synonymous mutations per Mb sequenced for the entire cohort that underwent WES was 1.76/Mb (range, 0.61–22.39/Mb). Six tumors carried >10 somatic mutations/Mb: ASPS (#21), primitive neuroectodermal tumor (PNET; #24), Ewing sarcoma (#22), nephroblastoma (#18), spindle epithelial tumor with thymus-like differentiation (SETTLE; #39) and high-grade glioma (#42) all without clinical context of a predisposition syndrome.

Among the patients with two biopsies, patient #24 with PNET was biopsied a second time after a 4-week interval due to presumed material insufficiency but both biopsies yielded valid results. A NRAS p.Q61H mutation together with a very focal FGF3/4 amplification were detected in one sample whereas absent in the other sample. Patient #19 with a Sertoli-Leydig granulosa cell tumor had two liver biopsies at an interval of 1.5 years which revealed the identical MAP2K1 p.G128V mutation in a similar proportion as well as superposable CGHa profiles. Patient #26 with eRMS had a biopsy of both the recurrent primary tumor and a distant lymph node metastasis. Both samples presented largely distinct copy-number alterations while no mutation was detected.

Although dealing with a heterogeneous cohort, two-thirds of all alterations identified as being potentially “actionable” fell among four key signaling systems, namely the receptor tyrosine kinases and their ligands and adapters, the cell cycle, the RAS/RAF/MEK/ERK and the PI3K–AKT–mTOR signaling pathways (Fig. 3). The most frequent, yet nontargetable genomic hit affected TP53 in 18.8% of patients, followed by CDKN2A/B deletions in 10 patients, among them all patients with Ewing sarcoma who underwent copy-number analysis. In seven patients, this alteration was evaluated “actionable” due to focality, homozygosity or the underlying disease. Five patients had alterations in codons 13 and 61 of H-RAS/N-RAS, three of them with eRMS. Five patients, including three with a high-grade glioma, had oncogenic PI3KCA mutations in the helical or kinase domain (codons 545, 546, and 1047).

Of the 42 patients presenting genomic alterations that a “matching” agent theoretically might exert antitumoral effects on, 14 patients (33.3%) received a respective agent, representing 19.2% of the total study population (Table 2). Three patients received twice a targeted agent. Eight treatments were administered within clinical trials, six of which were single-agent phase I and II were phase I combination trials; nine treatments used registered agents, mostly combined with chemotherapy. Five of these patients experienced objective tumor response (during 4 months each in a cervix clear cell carcinoma, #9, and an alveolar rhabdomyosarcoma, #50) or prolonged stable disease (during 4 months in a PNET, #24, 6 months in a diffuse infiltrative pontine glioma, #43, and more than 12 months in a low grade glioma, #45).

The majority of the patients (n = 28) did not receive a treatment based on the individual tumor profile. In 13 patients, this was related to drug access (trial not open yet, trial open uniquely for adults, no respective age slot available, inclusion criteria not fulfilled). In 11 patients, an alternative regimen was chosen or an on-going therapy continued, preserving the option for a targeted therapy at a later time point. Three patients experienced disease progression, which was too rapid to initiate targeted treatment. A further two patients had previously received the respective drug (#66 and #21).

In the following section, we present four illustrative examples that highlight the challenges of translating molecular alterations into clinical treatments of children.

A 4.7-year-old girl underwent biopsy of the first recurrence of an alveolar rhabdomyosarcoma. Treatment was initiated with vincristine-irinotecan-temozolomide chemotherapy resulting in stable disease (−7.5% tumor regression). Following the detection of a PIK3CA p.H1047R mutation, the mTOR inhibitor temsirolimus was added to the chemotherapy. The girl experienced partial tumor response (−57%) after four cycles and underwent subsequent surgical complete resection.

A 17-year-old patient with a SMARCB1-heterozygously deleted, INI-negative epithelioid sarcoma experiencing rapid tumor progression of pleural metastases, was treated with a combination of erlotinib and rapamycin based on a suspected EGFR amplification (poor CGHa quality) and TSC2 variant of unknown significance, based on preclinical data (26). She experienced stable disease with 20% tumor regression during 4 months. Subsequent confirmatory FISH and WES did not confirm EGFR amplification, and the TSC2 variant was shown to be germline. WES further confirmed the SMARCB1 alteration and the patient was included in the phase I/II study of EZH2 (enhancer of zeste homologue 2)-inhibitor tazemetostat (NCT01897571). EZH2, the enzymatic subunit of Polycomb repressor complex 2 that methylates repressive histone mark H3K27, is a potent cancer and progression driver in tumors such as rhabdoid tumors which harbor inactivating events in subunits of the SWItch/Sucrose Non-Fermentable (SWI/SNF) chromatin remodeling complex, leading to a loss of its' antagonizing function (27). The patient progressed but experienced complete response to pazopanib; none of the known pazopanib targets had been evoked by the analysis.

A 6-year-old patient with Gorlin-Goltz syndrome who had developed a desmoplastic medulloblastoma at the age of 4.5 years underwent at relapse treatment with the smoothened inhibitor Sonidegib on a compassionate use basis at 450 mg for 4.5 months. After the second cycle, a partial response was observed, but the patient progressed after four cycles. The tumor was biopsied 3 weeks later, genomic analysis confirmed a germline PTCH1 p.Q371fs mutation and revealed a somatic PTCH1 p.D301Ifs*23 mutation as well as a SMO p.V321M variant. In basal cell carcinoma, a malignancy characterized by alterations primarily affecting the PTCH1 gene, as well as medulloblastoma, SMO mutations could be observed in tumors that had developed resistance to vismodegib (28), including the p.V321M, which presumably confers drug resistance by its location in the transmembrane region and proximity to the drug-binding pocket (29). It is therefore highly probable in this patient that the alteration was associated with the acquired resistance to Sonidegib.

A 20-year-old patient with ASPS underwent treatment with MET/ALK-inhibitor crizotinib within the CREATE study (NCT01524926). ASPS is defined by the presence of fusion gene ASPSCR1-TFE3 t(X;17)(p11.2;q25; 30). The rationale for this treatment was based on MET upregulation by direct transcriptional activation of the resulting oncoprotein (21). A phase II monotherapy study with MET inhibitor tivantinib achieved disease-control rates of 78% in ASPS (31). Our patient experienced fatigue and mucositis and withdrew after three cycles with stable disease. Biopsy for the purpose of MOSCATO-01 was performed 1 month following the end of treatment with crizotinib, the genomic profile revealed the pathognomonic fusion but no other druggable targets.

To our knowledge, this is the first study, performing intentional sample acquisition at relapse in pediatric patients with cancer to define the current state of a patient's tumor for molecular analysis. The feasibility of this approach is underlined in the low complication rate for tissue collection procedures in pediatric patients with advanced cancers, the potentially actionable alterations identified in 60.9% of patients and the change of the presumed diagnosis in three patients, including two secondary cancers.

For all our precision medicine approaches, the ultimate aim is to detect and define molecular markers that translate into clinical benefit using adapted treatments. In childhood solid cancers, despite 2 decades of use of targeted agents, only limited clearly defined biomarkers are reported, such as TSC1/2 mutations in subependymal giant cell astrocytoma associated with tuberous sclerosis complex, which led to the marketing authorization of everolimus (32). Several molecular targets are currently under clinical evaluation, some with preliminary activity signals. For additional alterations preclinical data suggest a potential role in childhood oncogenesis. Considering the low occurrence of confirmed biomarkers, we applied an approach that allowed the description of well-defined targets for which clinical trials were ongoing. In addition, we wished to include the reporting of genetic alterations that could be of potential interest in pediatric cancers based on preclinical data. None of the patients in our study population exhibited a genetic alteration that allowed the treatment of an approved drug in the disease. Four patients were found with molecular events that are known oncogenic drivers in the underlying or other diseases and that are currently explored in a pediatric clinical trial and for which early activity signal have been reported (5, 6, 33, NCT01897571). What is far less clear is the significance of activating hot spot mutations in signaling pathways, like PIK3CA and NRAS or HRAS, or copy-number alterations in cell-cycle genes like CDKN2A/B deletion or CDK4/6 amplification which occurred in several of our patients. In a context of therapeutic innovation, it is challenging to define “the” cut-off of the term “actionable” alterations. For the above-mentioned reasons and with the aim to develop further therapeutic proof-of-concept trials, we have chosen for our study to report molecular alterations or affected pathways which theoretically would be targetable with an agent. At a first glance, our rate of reported alterations seems to be higher than recently published, ranging from 43% to 51% (24, 25, 34, 35). Different factors account for this difference as the trials varied in the selection of included entities, the analytic methods applied or simply the denominator for statistical calculations. Discrepancies in the definition of the term “actionable alteration” as outlined above and restrictions in treatment recommendations due to the nonavailability of potential agents clearly affect the formal number of patients identified to exhibit specific, treatable molecular traits. Another relevant aspect is that all specimen analyzed in MOSCATO-01 were derived of patients exposed to chemo and/or radiotherapy, reflecting the contemporary state of the tumor in pretreated patients. It has been shown in a broad range of pediatric tumors that this leads to a higher mutational charge as compared to first diagnosis (25, 36–39). Furthermore, the understanding of initially present, yet undetected subclones and/or the de novo introduction of progression-driving alterations enhance the probability of identifying additional targets, as shown in neuroblastoma or glioma (40, 41). In addition to temporal changes, the potential degree of intra- and intertumoral heterogeneity might be counteracted by longitudinal and multisite biopsies to personalize cancer therapy. This will be difficult to achieve from an ethical perspective especially in patients with solid tumors. Liquid biopsy strategies may help to overcome this issue.

Our initial analytic backbone comprised aCGH for copy-number estimation and targeted gene panel sequencing for a panel of 75 oncogenes and tumor suppressors, intentionally designed for the adult population of the MOSCATO-01 trial (12). WES and RNAseq were included early during the study to narrow gaps in target detection, for example, gene fusions or mutations in genes not included in the TGPS (e.g., PTCH1, Histones) and to reliably identify constitutional pathologies the importance of which has been documented in the pediatric cancer population (23, 42). WES, which was preferred to WGS to balance costs, substantiated results obtained by aCGH and TGPS and added targetable alterations in nine patients, mainly mutations in genes not captured by the panel and in two cases due to insufficient coverage in the TGPS. RNAseq allowed the detection of genomic rearrangements. The combination of aCGH and TGPS met well with clinically relevant expectations in terms of: cost-effectiveness, time, reliability of results and allowed analyzing samples with lower tumor cell content than needed for WES (14). To decipher the underlying mechanisms leading to treatment failure it is indispensable to extensively investigate pediatric cancer tissues at relapse or resistance to generate significant, disease-specific cohorts that allow systematic analyses. The decline in sequencing costs will further enhance applicability of large-scale, comprehensive genomic profiling. Nevertheless, it must be taken into account that WES and WGS generate huge amounts of data that require a robust bioinformatics pipeline for data processing, sufficient storage capacities and are more time-consuming. Comparison of different alignment and variant-calling pipelines revealed a rather low single-nucleotide variant (SNV) overlap (43). For the purpose of purely clinical sequencing in pediatric oncology, based on the growing body of data available, we think that a gene panel specifically tailored to pediatric cancers, gene copy-number analysis and gene fusion detection will be sufficient for target screening of innovative compounds currently available for children and young adults.

Our results demonstrate the need for improved access to a broader range of antineoplastic agents, thereby highlighting molecular pathways in which alterations occur in the pediatric population. The quantitative distribution of our “actionable” findings is determined by agents that exist but also pinpoint molecular traits in recurrent pediatric cancers that warrant further drug development. Two-thirds of patients providing a treatment rationale by displaying a potentially targetable genomic alteration could not be treated. For a significant proportion of these patients an existing matching agent was not accessible. In our cohort, mTOR inhibitors were the most commonly used targeted agents, followed by multityrosine kinase inhibitors within an ongoing phase I pediatric clinical trial. However, in some patients a more specific treatment would have been desirable. This, together with a paucity of clearly defined biomarkers, not sufficiently stringent definition of “actionable” and the occurrence of multiple molecular alterations in one quarter of patients, may contribute to the rather low frequency of responses observed. We have therefore developed a multiarm, proof-of-concept trial (AcSé-ESMART, NCT02813135), which currently covers the most frequently detected altered pathways. This trial aims to determine through an enrichment strategy if theoretically “actionable” alterations in the targeted pathways may bring a clinical benefit to the patient (44). However, translating biological findings into clinical benefit should be the most challenging part in the current precision medicine approach. We have chosen to report in detail on four of our patients who were treated with the proposed targeted therapy. These selected cases reflect the diversity and complexity of this approach. The first illustrates that agents that suboptimally target the PIK3CA mutation through inhibition of mTOR may revert tumor resistance to chemotherapy and result in objective response. The second highlights that a treatment based on a strong genomic rationale may fail, as this is currently thought for EZH2 inhibitors in INI1-deficient tumors, whereas there is no molecular rationale reported for the response observed to the FGFR/VEGFR inhibitor. The third case shows that a research biopsy at progression may allow to detect mechanisms that are involved in resistance to targeted treatments. Finally, the fourth case shows that the targeting of pathways that are transcriptionally activated by fusion oncoproteins is less than straight forward.

It is imperative to pursue the path of pediatric precision medicine trials to further enrich the genomic landscape of pediatric relapsed/resistant tumors, to improve our understanding of tumor escape or resistance mechanisms as previously shown for neuroblastoma or Ewing sarcoma (37, 45). RAS alterations predominated in patients with eRMS, focal CDKN2A/B deletions occurred in all four patients with Ewing sarcoma who had sufficient analysis. A fourth patient with eRMS had a BRAF p.G596R mutation, further supporting a role of the MAPK pathway in this high-risk disease. Despite the lack of statistical power due to small sample size, our results suggest that certain alterations are enriched in the relapse setting when compared to their occurrence in cohorts of predominantly diagnostic material (22% and 28%; refs. 39, 46). This is also in keeping with the observation that the frequency of CDKN2A/B alterations exceeded 50% in Ewing sarcoma cell lines as representatives of rather high-risk phenotypes (39, 47) but further research is needed to define the extent of enrichment at relapse. Consequently, pangenomically well-defined subgroups would also likely benefit from the introduction of targeted therapies earlier in the tumor management. Furthermore, the efficacy of these agents might be enhanced by combination with other targeted or conventional anticancer drugs and/or radiotherapy. The clinical benefit of for example, CDK4/6 inhibitors alone targeting the cell-cycle machinery in pediatric patients with rhabdoid tumors, characterized by SMARCB1 inactivation leading to Cyclin D1 transcriptional activation, seemed to be limited to single patients (48), whereas encouraging results were obtained in patients with mantle cell lymphoma and liposarcoma (49, 50). To develop meaningful effective combinational strategies, it is therefore indispensable to elucidate molecular mechanisms that are the real actors in tumor (re-)growth. A major limitation of the current profiling efforts is that it can only add limited insight in mechanisms of cancer hallmarks. We are convinced that an extended consideration of the cancer complexity is needed to improve outcome, including epigenetic phenomena, noncoding DNA, immune contexture and microenvironment in pediatric tumors that are currently not sufficiently addressed in our studies.

Discussion with patients and parents before inclusion in this profiling trial showed that there was a high expectation of the analysis and demand to find a treatment option but also to improve knowledge on the disease. In our experience, one of the major challenges during the study was the aggressive behavior of pediatric cancers which often did not allow a waiting time of 3 to 4 weeks for the results. We therefore suggest the intervention at an earlier time point when another “salvage” option is still available. The second most challenging point was to temper the expectations of the families of the precision medicine paradigm. This can be addressed in the long term by improving our knowledge and understanding of the oncogenic elements in pediatric cancers and the relationship between these elements and response to specific targeted agents. In the meantime, it is important to convey to families the uncertainty surrounding the definition of "actionable" genomic alterations.

In reply to these challenges, we have been pursuing since January 2016 with the international, multicenter, prospective genomic sequencing trial MAPPYACTS (MoleculAr Profiling for Pediatric and Young Adult Cancer Treatment Stratification; NCT02613962), run in conjunction with the Institute Curie. MAPPYACTS feeds into the therapeutic proof-of-concept trial AcSé-ESMART, with the aim to maximize our efforts on tailoring treatment in pediatric patients with so far incurable tumors.

P. Varlet is a consultant/advisory board member for Boehringer, Hoffman-Roche, and Novartis. No potential conflicts of interest were disclosed by the other authors.

Conception and design: A.C. Harttrampf, L. Lacroix, S. Michiels, G. Vassal, J.-C. Soria, B. Geoerger

Development of methodology: A.C. Harttrampf, L. Lacroix, S. Michiels, G. Vassal, J.-C. Soria, B. Geoerger

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.C. Harttrampf, L. Lacroix, F. Deschamps, S. Puget, N. Auger, P. Vielh, P. Varlet, Z. Balogh, S. Abbou, D. Valteau-Couanet, S. Sarnacki, L. Gamiche-Rolland, V. Minard-Colin, J. Grill, C. Dufour, N. Gaspar, B. Geoerger

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.C. Harttrampf, L. Lacroix, M. Deloger, N. Auger, A. Allorant, G. Meurice, J. Grill, S. Michiels, J.-C. Soria, B. Geoerger

Writing, review, and/or revision of the manuscript: A.C. Harttrampf, L. Lacroix, N. Auger, P. Vielh, S. Abbou, D. Valteau-Couanet, V. Minard-Colin, L. Brugieres, N. Gaspar, S. Michiels, J.-C. Soria, B. Geoerger

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Lacroix, P. Varlet, Z. Balogh, G. Meurice, S. Michiels, B. Geoerger

Study supervision: L. Lacroix, G. Vassal, B. Geoerger

We are grateful to all patients and their parents for participating in the trial. The authors acknowledge the clinical teams for its implication in the enrollment of patients in MOSCATO-01 trial, the Clinical Research Direction of Gustave Roussy who made this academic-sponsored trial possible (Aurélie Abou-Lovergne, Lisa Lambert, Thibaud Motreff, Delphine Vuillier), Maud Ngo-Camus, Leslie Tsambou, Sandrine Tchao, Aljosa Celebic and Katty Malekzadeh for Clinical Data Management/Data Management. The authors acknowledge laboratory and bioinformatic teams, including Amélie Boichard, Mélanie Laporte, Isabelle Miran, Nelly Motté, Ludovic Bigot, Stéphanie Coulon, Marie Breckler, Catherine Richon, Aurélie Honoré, Magali Kernaleguen, Glawdys Faucher, Felipe Andreiuolo, Jonathan Sabio, Lionel Fougeat, Marie Xiberras, Leslie Girard, Lucie Herard, Catherine Lapage, Romy Chen-Min-Tao, Celine Lefebvre, Guillaume Robert, Marion Pedrero for their support in the preanalytic processing of samples, storage and for tumor sample analysis (Wet lab and bioinformatics) of the patients included in MOSCATO-01 trial. We thank the Tumorothèque of Necker Enfants Malades, Paris. We are grateful to Brenda Mallon for critical reading of the manuscript.

This work was presented in part at the American Society of Clinical Oncology meeting (ASCO) 2014 in Chicago, USA (Proceedings of ASCO, #10050, 2014), the European Association of Cancer Research EACR-OECI Conference “Precision Medicine for Cancer” 2015 in Neumuenster Abbey, Luxembourg (#26), and the 48th Congress of the International Society of Pediatric Oncology 2016 in Dublin, Ireland (#SIOP6-0926). Supported by the Fondation Gustave Roussy and a grant from Fondation Philanthropia (to A.C. Harttrampf).

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.