High-Throughput Genomics and Clinical Outcome in Hard-to-Treat Advanced Cancers: Results of the MOSCATO 01 Trial

Oncogenic drivers are genomic alterations that lead to malignant transformation and cancer progression (1, 2). Targeting oncogenic drivers improves outcomes in patients with metastatic cancers. For example, ALK and EGFR inhibitors have been shown to improve outcomes in patients presenting with ALK translocations or EGFR activating mutations (3, 4). Nevertheless, the number of approved targeted therapies that are matched to a genomic alteration remains rare. As of 2016, only nine genomic alterations are assessed in daily practice in patients with metastatic cancers to decide about targeted therapies. In the last 5 years, sequencing efforts have generated considerable advances in the knowledge of cancer biology. First, although only a few genetic somatic abnormalities are tested in oncology, it is believed that >400 genes could drive tumor progression (5). Second, most of the frequent cancers present a large number of very rare targetable genomic alterations (6). Several cases reported in the literature support the concept that targeting a rare genomic event could translate into patient benefit. As an illustration, a patient with an activating mTOR mutation presented an outlier response to everolimus (7). Third, some oncogenic drivers, like BRAF mutations or ERBB2 amplifications, are shared across several diseases (8, 9), and their targeting could improve outcomes (10, 11).

These considerations have generated the hypothesis that testing a large number of genes across all tumor types could improve outcomes in patients with “hard-to-treat” advanced cancers. New advances in biotechnologies allow testing of a large number of gene abnormalities in individuals with good analytic validity (12). Several trials have shown that these technologies can be delivered in the context of clinical initiatives (13, 14). Nevertheless, there is as yet no evidence that this approach improves outcome. In the present single-center single-arm open-label trial, we have evaluated the clinical benefit of testing a large panel of genes in patients with advanced “hard-to-treat” cancers.

The study flow is reported in Fig. 1. Overall, 1,035 adult patients gave signed informed consent between December 2011 and March 2016 for the MOSCATO trial (NCT01566019), a successful tumor biopsy was performed in 948 of them, and a molecular portrait was obtained in 843 (89%). Patient characteristics are reported in Table 1. The most frequent tumor types included digestive cancer (n = 197), lung cancer (n = 170), urological cancer (n = 158), breast cancer (n = 135), and head and neck cancers (n = 111). The main sites of biopsy were liver (n = 353), lung (n = 203), lymph nodes (n = 153), and bone metastases (n = 44). The following complications occurred: pneumothorax (n = 13), hemorrhage (n = 2), and others (n = 13). All patients recovered after medical treatment without any deaths related to biopsy procedure.

The techniques used to obtain molecular portraits for 843 patients at the start of the MOSCATO trial included targeted sequencing (feasible for 837 patients) and array comparative genomic hybridization (aCGH) analysis (feasible for 745 patients). During the course of the trial, RNA sequencing (RNA-seq; 427 patients) and whole-exome sequencing were added (166 patients). Genomic reports were validated by a molecular geneticist and discussed during a weekly multidisciplinary molecular tumor board dedicated to the MOSCATO trial. Actionability of the target was defined by the molecular board. An actionable target was detected in 411 patients (49% of the 843 patients with a molecular portrait). In some patients, actionable targets were identified by multiple tests. Actionable targets were identified by targeted sequencing in 255 patients, by aCGH in 252 patients, by IHC in 17 patients, and by FISH in 16 patients. The median time from signed consent to biopsy and from biopsy to molecular board (encompassing sequencing and aCGH analyses) was 19 days [interquartile range (IQR), 6–33 days] and 21 days (IQR, 14–27 days), respectively. We did not observe any meaningful difference in time from biopsy to molecular tumor board between the 64 patients with clinical deterioration (median of 21 days; IQR, 19–26) and the other patients (median, 21 days; IQR, 14–26 days).

One hundred ninety-nine patients received a therapy matched to the genomic alteration in the subsequent line of therapy, based on CGH (105 patients), targeted sequencing (85 patients), and MET-positive IHC (9 patients). This represents 19% of the patients who consented (95% CI, 17%–22%) and 24% of the patients for whom a molecular portrait was successfully obtained. Reasons for not providing targeted therapies in the other 212 patients with a targetable alteration are outlined in Fig. 1. Patient characteristics of the matched treatment population are reported in Table 1. Patients received a median of 4 prior lines of treatment for advanced disease (1–14). The median Royal Marsden Hospital (RMH) score was 1 (0–3). The vast majority of the patients received matched therapy in the context of phase I/II trials (n = 149, 75%). The matched therapies used in the trial are listed in Table 2. One hundred twenty-seven patients were treated with single-agent targeted therapy, and 72 patients received a combination (chemotherapy in 42 cases; targeted therapy in 30).

The actionable genomic alterations were located in 53 genes; 98 were amplifications and 23 deletions or loss, 103 mutations, 18 were translocations, and 8 were based on IHC.

The genes that contained molecular alterations that were the actual targets of the chosen matched treatment are reported in Fig. 2 (and according to tumor type in Supplementary Fig. S1). The most frequent targeted alterations were PIK3CA mutations and amplifications (n = 32), ERBB2 mutations and amplifications (n = 24), PTEN mutations and deletions (n = 15), FGFR1 mutations and amplifications (n = 13), EGFR mutations and amplifications (n = 13), and NOTCH1/2/3/4 mutations, amplifications, or translocations (n = 12).

The primary endpoint could be evaluated in 193 patients (PFS1 was not available in 5 patients, and 1 patient had PFS2/PFS1 < 1.3 without progressive disease under matched treatment during follow-up). In total, 33% of the patients (63 of 193) presented a PFS2/PFS1 > 1.3 (two-sided 95% CI, 26%–39%). The null hypothesis that the true probability is lower than or equal to 15% is rejected with a one-sided P value of <0.001. In a preplanned secondary analysis, we directly compared the paired failure times PFS2 and PFS1 through a χ² test (15). The null hypothesis of equality of the paired survival times was rejected in favor of PFS2 (χ² = 8.25 with 1 df, P = 0.004). A Kaplan–Meier plot of the PFS2/PFS1 ratio is reported in Fig. 3A. Individual PFS1 and PFS2 of the patients are reported in Fig. 3B, and an alternative display of these times in Supplementary Fig. S2. We tested in a sequential manner the null hypothesis in the subgroup of 27 patients for whom target treatment was considered level of evidence A: 37% of the patients (10 of 27) presented a PFS2/PFS1 > 1.3 (two-sided 95% CI, 19%–58%). The null hypothesis that the true probability is lower than or equal to 15% is again rejected (one-sided P = 0.004).

Responses by RECIST 1.1 criteria among the 194 patients with non-missing PFS1 included 2 complete responses (1%) and 20 partial responses (10%) for an overall response rate of 11% (95% CI, 7%–17%), 100 stable disease (52%), 33 progressive disease (17%), and 39 not available (20%; Fig. 4). One patient with breast cancer and an FBXW7 mutation has presented a complete response after treatment with a NOTCH inhibitor. Moreover, one patient with a HER-positive biliary cancer was treated with a trastuzumab and chemotherapy combination and has also presented a complete response. Among the 20 patients who presented a partial response, the following treatments were provided: HER inhibitors (n = 8 patients), FGFR inhibitor (n = 4 patients), EGFR inhibitor (n = 3 patients), ALK inhibitor (n = 3 patients), and MAPK or PI3K–AKT–mTOR inhibitors (n = 2 patients). The majority of patients without RECIST data had an early clinical deterioration (28 of 39). The median follow-up for progression-free survival on matched therapy or PFS2 was 20 months. The 6-month estimated PFS2 was 16% (95% CI, 12%–22%), and median PFS2 was estimated at 2.3 months (95% CI, 1.9–2.7 months). The 1-year estimated overall survival (OS) probability was 49% (95% CI, 42%–58%) and the estimated median OS was 11.9 months (95% CI, 9.5–14.3 months).

In order to explore which patients could derive more benefit from this approach, we retrospectively assessed the PFS2/PFS1 ratio according to post hoc subgroups. The benefit of sequencing was not significantly different according to the RMH score (exact test for difference in proportions, P = 0.67), drug family (P = 0.72), year of inclusion (P = 0.95), disease (P = 0.66), and level of evidence for the target gene molecular abnormality (P = 0.15; Table 2).

There are controversies about whether the use of high-throughput genomics could improve outcome in patients with hard-to-treat cancers. In the present study, we have shown that tumor sequencing improves outcome in 33% of patients with advanced cancers. Previous trials had evaluated the efficacy of multigene molecular abnormality screening approaches to personalized therapy. In the SAFIR01 trial (14), only 30% of the treated patients presented an objective response or stable disease. In the SHIVA randomized trial (13), no significant improvement in progression-free survival could be observed in the precision medicine arm as compared with the standard-of-care arm. There are several reasons that could explain why the MOSCATO trial is positive, whereas previous ones failed. First, 75% of the patients received last-generation targeted therapies in the context of phase I/II trials. It is well established that only therapies that hit the target with high bioactivity can improve outcome. As an illustration, vemurafenib dramatically improves outcome of BRAF^V600-mutant melanoma, whereas sorafenib, a weak BRAF inhibitor, does not (16). Second, it is very likely that the MOSCATO trial team benefited from the experience acquired in previous precision medicine trials (13, 17, 18–23).

In the present nonrandomized study, the ratio PFS2/PFS1 was chosen as a primary endpoint. Von Hoff first proposed this endpoint and concluded that a ratio >1.3 would indicate a treatment benefit. This is supported by the observation that PFS decreases over the lines of therapy in the natural course of the disease (24). In that regard, in a study that included 934 patients with metastatic breast cancer, median PFS times were 9.3, 5.5, 4.9, 4.1, and 0.2 months in the first, second, third, fourth, and fifth lines of therapy, respectively (25). This endpoint has been successfully used to develop oxaliplatinum and to find optimal dosing of imatinib, but has been criticized to be of little value when the within-patient correlation between consecutive PFS times in the natural course of the disease is low (26). Although this study reports, for the first time, evidence that sequencing a large panel of genes improves outcome of a subset of patients, it does not therefore provide level I evidence that this approach improves outcome in the general population of patients with metastatic cancers. Indeed, the quantification of the impact of such an approach in the overall population requires randomized controlled trials. SAFIR02_Breast and SAFIR02_lung trials (NCT02299999 and NCT02117167) are currently addressing such questions. The other limitation of the current trial is the lack of comparison with patients, treated by the same targeted therapies, whose tumors do not have matched genomic abnormalities. Nevertheless, it is unlikely that this would explain the efficacy observed in the current trial, because efficacy of unmatched targeted therapies has so far been observed only with CDK4 and mTOR inhibitors.

There are several ways to further improve the efficacy of precision medicine. We first need better tools to identify the relevant genomic alterations. In the present study, although RNA-seq and whole-exome sequencing were performed in 427 and 166 patients, respectively, they were not useful to drive patients to therapy. Further trials that include immunotherapeutics and therapies targeting translocations will better assess the utility of these technologies. Also, RNA-seq could be useful to define pathway activation and target expression, but this was not assessed in the current study. Second, we need to better define which patients derive more benefit from this approach. Several studies have suggested that patients with high genomic instability (27) and/or subclonal alterations are less likely to derive benefit from targeted therapies. Finally, there is a need to develop combinations, especially in patients who present multiple driver alterations (28).

In the present trial, we have shown that the molecular screening strategy led to a matched targeted treatment for 19% of successfully screened patients, and 33% of those matched-treated patients or 7% of those successfully screened obtained a PFS ratio above the predefined threshold. Next steps include the validation of clinical utility in randomized trials, better defining which patients could derive benefit from this approach, and the further development of new methods to detect cancer drivers in individuals.

The Molecular Screening for Cancer Treatment Optimization trial (MOSCATO; NCT01566019) is a monocentric, prospective clinical trial. The trial accrued patients between December 2011 and March 2016. This study aimed to show that high-throughput genomics improve a progression-free survival ratio in patients with advanced cancers. Patients were eligible if they presented a locally advanced, unresectable, or metastatic cancer that had progressed during at least one line of prior therapy. Only patients who were considered noncurable by a multidisciplinary board were eligible for the trial. Patients who presented with a genomic alteration that rendered them eligible for a targeted therapy approved in the indication were not eligible for this trial. Children were eligible for the trial, but were not included in the population of patients where the primary endpoint was evaluated as prespecified in the protocol. This study is registered with ClinicalTrials.gov, MOSCATO 01 (NCT01566019). All patients gave signed informed consent for the trial and genomic analyses. This trial protocol was approved by an institutional review committee and done in accordance with the Declaration of Helsinki.

After having given signed informed consent, the patient underwent either a biopsy or a tumor resection. The tumor cellularity was assessed by a senior pathologist on a hematoxylin and eosin slide from the same biopsy core than the one used for nucleic acid extraction and molecular analysis. In case of tumor cellularity >30%, the association of aCGH and targeted sequencing was performed, completed by RNA-seq analysis for cases collected after 2014 and whole-exome sequencing for selected cases, when the quality and the quantity of nucleic acid were sufficient. In cases with tumor cellularity with 10% to 30%, only targeted sequencing was performed. Tumor DNA, RNA, and germline DNA from whole-blood samples [used for whole-exome sequencing (WES)] were extracted using, respectively, AllPrep DNA/RNA Mini Kit and DNeasy Blood and Tissue Kit according to the manufacturer's instructions. The molecular analysis using targeted sequencing and aCGH was carried out as described previously (29, 30). Briefly, targeted sequencing was performed using Personal Genome Machine (Ion Torrent PGM, ThermoFisher Scientific) with Ion AmpliSeq multiple genes panels, based on multiplex-PCR. For biopsies collected from May to November 2012, the targeted gene panel consisted of the Ion AmpliSeq Cancer Panel (CP1) covering 190 amplicons in 40 cancer genes (ThermoFisher Scientific). For biopsies collected from December 2012 to September 2013, the targeted gene panel consisted of the Ion AmpliSeq Cancer Hotspot Panel v2 covering 207 amplicons in 50 cancer genes (ThermoFisher Scientific). Finally, for biopsies collected after September 2013, the targeted gene panel (MOSC3) covered 75 critical oncogenes or tumor suppressor genes using Ion AmpliSeq custom design (details available in Supplementary Methods), combining 1,218 amplicons with the CHP2 panel. The library preparation as well as the sequencing analysis were performed according to the manufacturer's recommendation for Ion AmpliSeq Workflow. The bioinformatic analysis was performed with Torrent Suite software, variantCaller (ThermoFisher Scientific) completed by our own filtering and annotation pipeline. Each retained variant was reviewed by a molecular geneticist and classified as pathogenic variant, unknown pathogenicity variant, or probably nonpathogenic variant (details available in Supplementary Methods). The median coverage depth on retained variants is over 700 reads, offering a sensitivity down to 5% of allelic frequency. The whole-genome aCGH, SurePrint G3 Human aCGH Microarray 4 × 180K, Agilent Technologies) was performed as described previously (30). The microarray scanning was performed using default parameters, and quantification of Cy5 and Cy3 signals was extracted with Feature Extraction v10.5.1.1 (Agilent Technologies), then analyzed and annotated with our own bioinformatic pipeline, described in Supplementary Methods. The aCGH analysis was mainly used to detect tumor gene amplifications and/or deletions (an amplifications >×0.7 log₂ ratio and deletions < 0.5 log₂ ratio) and was discussed during the tumor board. The alterations with a length less than 10 Mb were considered of interest. When needed, CGH results were confirmed by FISH analysis. Between 2014 and March 2016, CGH arrays were substituted by Cytoscan technology to quantify copy-number alterations. Whole-exome analysis and RNA-seq analysis were performed on a limited number of cases by IntegraGen pattern platform (IntegraGen), starting from the same nucleic acid prepared for targeted gene sequencing (TGS) or aCGH. Analysis protocols used for WES and RNA-seq are reported in Supplementary Methods.

All molecular analysis results from targeted sequencing, aCGH, WES, or RNA-seq were reviewed one by one by a molecular geneticist responsible for generating a molecular report highlighting annotated molecular abnormalities to be discussed during the molecular tumor board. The genomic data from this trial are available in the GENIE database.

In addition, we performed IHC for phospho-MET in order to drive patients to MET inhibitors. The IHC analysis targeting MET and phospho-MET was performed by the clinical pathology laboratory as described previously (31) using antihuman c-MET, clone SP44 (reference M3444, Spring Bioscience) or clone CVD13 (reference 18-2257, Invitrogen) and anti-phospho-MET (Tyr1234/1235) clone D26 (reference 3077, Cell Signaling Technologies).

The genomic analysis report (TGS, aCGH, RNA-seq, and WES) provided by the molecular geneticist was discussed during a weekly multidisciplinary molecular tumor board dedicated to the MOSCATO trial, and a matched therapy was decided accordingly. Actionability of the target was defined by the molecular board. Actionability was defined by the availability of drugs that hit either the target or the pathway activated by the target. Patients who were offered a matched targeted therapy during the multidisciplinary molecular board and who received it constituted the population where primary endpoint would be evaluated. In some patients presenting an actionable genomic alteration, no therapy was proposed at the time of the molecular tumor board, but they could have been further treated with a matched therapy outside MOSCATO trial.

The primary endpoint was defined as the PFS2/PFS1 ratio or growth modulation index (15), in which the progression-free survival on matched therapy (PFS2) was compared with the progression-free survival for the most recent therapy on which the patient had just experienced progression (PFS1). Progression-free survival on matched treatment (PFS2) was defined as the time from start of treatment to progression, as defined by RECIST 1.1, clinical progression, or death from any cause. Progression-free survival on prior therapy (PFS1) was defined as the time from start of the last prior treatment to progression as defined by RECIST 1.1 or clinical progression. The null hypothesis was that the PFS ratio >1.3 for 15% of patients or fewer (17). In order to reject this null hypothesis with 90% power at a one-sided significance level of 0.05, and assuming the true proportion of patients with a PFS ratio >1.3 was equal to 24%, a total of 165 evaluable patients was required. The primary endpoint was tested by a one-sided exact test. We also tested sequentially the PFS ratio in the subgroup of patients for which the target treatment was considered level of evidence A according to recommendation (32), which was scored by an expert blinded from clinical outcome. It was also prespecified as secondary analysis in the protocol to directly compare the paired failure times PFS2 and PFS1 through a χ² test with 1 degreeof freedom in a loglinear model for correlated failure times (15). Secondary endpoints were the number of patients who received a matched treatment to describe feasibility, PFS2, OS, and best overall response (RECIST 1.1). OS was defined as the time from inclusion to death, irrespective of the cause. Patients still alive at the last visit were censored at the date of last follow-up. The Kaplan–Meier method was used to construct survival curves. The median follow-up was calculated by the reverted Kaplan–Meier method. Exploratory post hoc subgroup analyses were performed for the primary outcome using exact tests for differences in proportions. Statistical analyses were performed in SAS 9.4 and R 3.2.2.

E. Deutsch reports receiving commercial research grants from Roche Genentech, AZD, Lilly, and Servier, and is a consultant/advisory board member for Merck Pfizer and EISAI. A.M. Eggermont is a consultant/advisory board member for Actelion, Bayer, BMS, GSK, Haliodx, MSD, Nektar, Novartis, Pfizer, and Sanofi. J.-C. Soria is a consultant/advisory board member for AstraZeneca, GSK, Lilly, MSD, Pfizer, Pharmamar, Pierre Fabre, Roche, Sanofi, and Servier. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S. Michiels, M.-C. Le Deley, L. Lacroix, A. Hollebecque, G. Vassal, F. André, J.-C. Soria

Development of methodology: S. Michiels, C. Ferté, M.-C. Le Deley, L. Lacroix, R. Bahleda, T. De Baere, J.-Y. Scoazec, V. Lazar, J.-C. Soria

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): C. Massard, C. Ferté, L. Lacroix, A. Hollebecque, L. Verlingue, E. Ileana, S. Ammari, M. Ngo-Camus, R. Bahleda, A. Gazzah, A. Varga, S. Postel-Vinay, Y. Loriot, C. Even, I. Breuskin, N. Auger, T. De Baere, F. Deschamps, P. Vielh, J.-Y. Scoazec, C. Richon, V. Ribrag, E. Deutsch, E. Angevin, A. Eggermont, J.-C. Soria

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): C. Massard, S. Michiels, C. Ferté, M.-C. Le Deley, L. Lacroix, A. Hollebecque, L. Verlingue, S. Rosellini, S. Ammari, R. Bahleda, Y. Loriot, N. Auger, B. Job, F. Deschamps, P. Vielh, A. Eggermont, F. André, J.-C. Soria

Writing, review, and/or revision of the manuscript: C. Massard, S. Michiels, C. Ferté, M.-C. Le Deley, L. Lacroix, A. Hollebecque, A. Gazzah, A. Varga, S. Postel-Vinay, Y. Loriot, C. Even, P. Vielh, J.-Y. Scoazec, V. Lazar, V. Ribrag, E. Angevin, A. Eggermont, F. André, J.-C. Soria

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): C. Massard, C. Ferté, M.-C. Le Deley, L. Lacroix, E. Ileana, E. Angevin, G. Vassal, J.-C. Soria

Study supervision: C. Ferté, M.-C. Le Deley, L. Lacroix, G. Vassal, J.-C. Soria

The authors acknowledge the global clinical team for its involvement in the enrollment of patients in MOSCATO trial, as well as all relevant personnel from Gustave Roussy Direction de la Recherche who made possible this academic-sponsored trial (Aurélie Abou-Lovergne, Lisa Lambert, Thibaud Motreff, and Delphine Vuillier). Maud Ngo-Camus, Aljosa Celebic, and Katty Malekzadeh are thanked for Clinical Data Management/Data Management. Dorota Gajda, Dienabou Sylla, and Adrien Allorant are thanked for their statistical assistance. 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, Zsofia Balogh, Jonathan Sabio, Lionel Fougeat, Marie Xiberras, Leslie Girard, Lucie Herard, Catherine Lapage, Guillaume Meurice, Romy Chen-Min-Tao, Yannick Boursin, Marc Deloger, Celine Lefebvre, and Marion Pedrero, for their support in the preanalytic processing of samples, storage, and for tumor sample analysis (wet lab and bioinformatic) of the patients included in the MOSCATO 01 trial. The authors especially acknowledge Yuki Takahashi for her editing support and Nelly Hainault for final submission.

This trial was supported by Fondation Gustave Roussy (Revolution Cancer initiative), INCa-DGOS-INSERM 6043 (SIRIC SOCRATE), and ANR-10-IBHU-0001 (MMO), and received an unrestricted grant from Genentech and Sanofi.

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.