Clinical Genomic Sequencing of Pediatric and Adult Osteosarcoma Reveals Distinct Molecular Subsets with Potentially Targetable Alterations

Osteosarcoma, the most common primary malignant bone tumor, accounts for approximately 1% of all cancer cases in the United States (1, 2). The incidence of osteosarcoma shows a bimodal distribution with one peak in childhood/adolescence and the other in adults over 50 years of age (1). The current standard therapies, which include combination chemotherapy and surgical resection, were originally developed in the 1980s and have significantly improved the 5-year disease-free survival of patients with osteosarcoma to approximately 70% (3, 4). Furthermore, the response to preoperative combination chemotherapy is highly prognostic in patients with localized disease (5). However, 20% to 30% of patients remain refractory to conventional treatment, and the survival rate for patients presenting with localized disease has remained essentially unchanged for over 20 years (4, 6). Patients with unresectable primary tumors or metastases have poor clinical outcomes (7, 8). Older studies have reported on kinases or their ligands including VEGF, IGF1, PDGF, HER2, and MET as potential therapeutic targets in osteosarcoma based on their overexpression by IHC analysis (9).

Next-generation sequencing (NGS) technology has made the comprehensive analysis of cancer-related genes more clinically accessible, opening new avenues in treatment modalities for a variety of tumor types (10, 11). The implementation of precision medicine for the treatment of rare tumors such as osteosarcoma has been difficult due to a lack of targetable driver mutations or fusions involving well-established drug targets such as kinases (12). In the present study, we analyzed clinical sequencing data in osteosarcoma using the MSK-IMPACT (Integrated Mutation Profiling of Actionable Cancer Targets) panel assay (11) to identify the proportion of patients with potential somatic actionable alterations as defined by the OncoKB precision oncology knowledge base (13).

This project was approved by the Institutional Review Board of Memorial Sloan-Kettering Cancer Center (MSKCC) and was conducted in accordance with the U.S. Common Rule. A total of 92 formalin-fixed paraffin-embedded osteosarcoma samples from patients treated at MSKCC between 2004 and 2016 were submitted for clinical sequencing using the MSK-IMPACT panel (11). In all cases, the diagnosis of osteosarcoma was confirmed by sarcoma pathologists. The MSK-IMPACT assay generated data for 81 of the 92 osteosarcoma samples (Supplementary Table S1), with the remaining 11 samples (12%) being insufficient or inadequate for NGS. This percentage is in keeping with our general experience with MSK-IMPACT testing, where approximately 9% of samples overall are found to have insufficient tumor or insufficient DNA extracted to proceed with MSK-IMPACT NGS (11). The remaining 80 cases consisted of 71 samples of classic high-grade osteosarcoma (including six samples of postradiation osteosarcoma) that were used for the analyses of genomic and clinicopathologic correlates, and a separate group of nine cases of special osteosarcoma subtypes (extraskeletal osteosarcoma, n = 7; dedifferentiated osteosarcoma, n = 2) that were excluded from further analysis in this study (Supplementary Table S1).

Among the 71 high-grade osteosarcoma samples (from 66 patients), 54 samples (from 49 patients) underwent clinical sequencing in a prospective manner, whereas 17 samples (from 17 patients) were selected and sequenced retrospectively. To confirm and select the tumor and corresponding normal tissue for the retrospective group, slides from all the tissue blocks were reviewed by a sarcoma pathologist (M. Hameed). In the prospective group, matched blood was used as the germline sample after obtaining patient consent. Tumor and germline DNA were sequenced using MSK-IMPACT, an FDA-cleared, hybridization capture-based NGS assay capable of detecting all somatic protein-coding mutations, copy-number alterations (CNA), and select promoter mutations and structural rearrangements in a panel consisting of 341 cancer-related genes (version 1) later expanded to 410 (version 2) and then 468 genes (version 3; ref. 11). Of the genes discussed in this study, only VEGFA was not present in all three versions (versions 2 and 3 only). The sequence read alignment processing, nonsynonymous mutations, and rearrangements were determined as previously described (11).

Copy-number aberrations were identified using an in-house–developed algorithm by comparing sequence coverage of targeted regions in a tumor sample relative to a standard diploid normal sample (11), as extensively validated for ERBB2 (HER2) amplification (14). Specifically, coverage values were normalized for the overall coverage of the sample, square root transformed, and adjusted for the guanine/cytosine content of each target region using Loess normalization (14). The following criteria were used to determine significance of whole-gene gain or loss events: fold change >2.0 (gain) or <−2.0 (loss), P < 0.05 (FDR-corrected for multiple testing).

Somatic structural rearrangements including putative gene fusions were identified by Delly (v0.6.1; ref. 15) based on supporting read pairs and split reads (16). Candidate rearrangements were flagged for manual review if the tumor harbored ≥3 discordant reads with a mapping quality of ≥5 and the matched normal sample harbored ≤3 discordant reads (sites of known recurrent rearrangements) or if the tumor harbored ≥5 discordant reads with mapping quality of ≥20 and the matched normal sample harbored ≤1 discordant read (novel rearrangement sites). All candidate somatic structural rearrangements were annotated using in-house tools and manually reviewed using the Integrative Genomics Viewer (17).

The somatic genomic alterations in the sequenced osteosarcoma samples were then analyzed using cBioPortal for Cancer Genomics tools (18, 19). Germline alterations in cancer susceptibility genes were not evaluated in this study as consent issues did not allow germline variant calling across this entire set of patients with osteosarcoma. A systematic analysis of germline cancer susceptibility across pediatric solid cancers (including osteosarcoma) in the MSK-IMPACT dataset is in progress and will be published separately.

Potentially actionable genetic events were categorized into one of four levels using MSK-Precision Oncology Knowledge base (OncoKB; www.OncoKB.org; ref. 13). The level of evidence on a specific molecular alteration is based on FDA labeling, National Comprehensive Cancer Network (NCCN) guidelines, disease-focused expert group recommendations, and scientific literature (13). Tumors with two or more level 1–4 oncogenic drivers were grouped with the highest level actionable driver alteration per the following OncoKB criteria. Individual mutational events are annotated by the level of evidence that supports the use of a certain drug in an indication that harbors that mutation. The levels of evidence are tiered as follows:

FDA-recognized biomarkers that are predictive of response to an FDA-approved drug in a specific indication.

Standard care biomarkers that are predictive of response to an FDA-approved drug in a specific indication.

FDA-approved biomarkers predictive of response to an FDA-approved drug detected in an off-label indication.

FDA- or non–FDA-recognized biomarkers that are predictive of response to novel targeted agents that have shown promising results in clinical trials in a specific indication.

FDA- or non–FDA-recognized biomarkers that are predictive of response to novel targeted agents that have shown promising results in clinical trials for another indication.

Non–FDA-recognized biomarkers that are predictive of response to novel targeted agents on the basis of compelling biologic data.

The clinical characteristics of the 67 patients with high-grade osteosarcoma are summarized in Table 1, whereas clinical, pathologic, and predominant molecular characteristics of all osteosarcoma cases with DNA sequencing belonging to multiple cohorts are shown in Supplementary Tables S1 and S7. The cutoff age of disease presentation for pediatric osteosarcoma was defined as up to 18 years. The median age at diagnosis was 14 for the pediatric group (n = 33; age range, 8–18) and 32 for the adult group (n = 34; age range, 19–80). Thirty-eight (56.7%) of the patients were male, and 29 (43.3%) were female. The primary sites included extremities (n = 53, 79.1%), trunk (n = 9, 13.4%), and other (n = 5, 7.5%). The histologic subtypes for high-grade osteosarcoma and all sequenced cohorts are shown in Supplementary Table S1. Thirty-five samples were collected from the primary site, five from local recurrences, and 32 from metastatic lesions. Upon NGS, one sample (No. 40) failed QC metrics for tumor content (flat copy-number profile + no nonsynonymous somatic variants + no silent somatic variants) and therefore the subsequent MSK-IMPACT data analyses were performed on the remaining 71 osteosarcoma samples from 66 patients.

Somatic alterations detected by MSK-IMPACT in the 71 high-grade osteosarcoma samples from 66 patients are shown in Fig. 1A and listed in Supplementary Tables S2 and S3. Among the common mutations, TP53 mutations were identified in 22 samples (31%; Fig. 1A; Supplementary Table S2). As MSK-IMPACT is not designed to pick up TP53 intron 1 rearrangements, recently reported in osteosarcoma (20), the prevalence of TP53 mutations may even be higher. We also identified alterations in ATRX (nine mutations in seven samples, 10%), RB1 (seven mutations in seven samples, 10%), and SETD2 (five mutations in five samples, 7%; Supplementary Table S2). Approximately 13% of samples (9/71) did not show alterations in any of the genes in Fig. 1A but did show other somatic mutations and/or CNAs. Tumor adequacy was not deemed to be an issue in these cases because they showed similar tumor mutational burdens (TMB) as the cases with the more common alterations (range, 0.9–16.7 mutations/Mb). The mutations seen in these nine cases are listed in Supplementary Table S8.

With respect to CNAs (Fig. 1A; Supplementary Table S3), amplifications at 6p12-21 harboring VEGFA (n = 17/64 samples; 27%), often also including CCND3, were the most frequent CNAs. Deletions at 9p21 involving CDKN2A (n = 16; 22%) and CDKN2B (n = 16; 22%) were the second most frequent CNAs (Table 2). Amplifications at 12q14 harboring MDM2 (n = 11; 15%) and CDK4 (n = 9; 13%) were frequent (Figs. 1 and 2; Table 2; Supplementary Table S4). As expected, MDM2 and CDK4 amplifications were mutually exclusive with TP53 and CDKN2A alterations, respectively (Supplementary Fig. S1; Supplementary Tables S5 and S6), consistent with previous data in osteosarcoma (21, 22). Furthermore, CDK4 and CDKN2A alterations were mutually exclusive with RB1 alterations, such that, in aggregate, this pathway was altered in about half of osteosarcoma samples. Likewise, the TP53/MDM2 pathway is altered in at least half of cases.

Notably, we also identified a subset of tumors with 4q11-12 amplification, including KIT (n = 11; 15%), KDR (n = 11; 15%), and PDGFRA (n = 13; 18%). Consistent with their chromosomal proximity, amplifications of PDGFRA and KDR frequently cooccurred with KIT amplification (P < 0.001; Fig. 1A and B; Table 2; Supplementary Table S4). Tumors with 4q11-12 amplification were mutually exclusive from those with 6p12-21 amplification with the exception of a single 4q12-amplified case that also showed borderline 6p12 gain (Fig. 1A). In addition, cases with 4q12 gene amplification were mutually exclusive not only with 6p12-21 amplification, but also with 12q14 gene amplification involving MDM2 (Supplementary Tables S5 and S6). Perhaps not unexpectedly, given that cases with 4q12 gain were mutually exclusive with MDM2 amplification, they appeared enriched for TP53 alterations. In addition, four cases with 11q13 gene amplification involving CCND1 and the FGF cluster were nonoverlapping with CCND3 gains at 6p12 and PDGFRA/KIT/KDR gains at 4q12 (Supplementary Table S6). Other less common regions of recurrent amplification are shown in Fig. 1A and Supplementary Table S3.

Among the 66 patients with MSK-IMPACT data, 14 (21%) had at least one potentially actionable alteration (level 2 or 3) as defined by the OncoKB classification (www.OncoKB.org; ref. 13; Table 3). Overall, 32 of 66 cases (48%) were annotated as levels 2 to 4 by OncoKB. None of the alterations were level 1, reflecting the lack of biomarker-driven FDA approvals in this disease.

Nine patients (14%) with CDK4 amplification were classified as level 2B potentially actionable somatic alterations by OncoKB. CDK4, an intracellular kinase, is altered by amplification in a diverse range of cancers, including liposarcoma, and CDK4 inhibitors, including abemaciclib (NCT02846987) and palbociclib (23, 24) are treatment options for patients with well-differentiated and dedifferentiated liposarcomas in the NCCN compendium. A somatic BRCA2-truncating mutation and two cases with BRCA2 deletions were annotated as a level 2B alteration. BRCA2 is a tumor-suppressor gene involved in DNA damage repair by homologous recombination (25, 26). PARP inhibitors olaparib (25) and rucaparib (26) are currently approved by the FDA for use in the treatment of BRCA2-mutant ovarian cancer. Interestingly, a recent analysis identified a genomic signature of homologous recombination deficiency in approximately 27% of osteosarcoma samples (27).

MDM2 amplifications, detected in nine patients (14%), are classified as a level 3B alteration. MDM2, an ubiquitin ligase that negative regulates p53, is amplified in a diverse range of cancers, including well-differentiated and dedifferentiated liposarcomas (28, 29). There are promising clinical data supporting the use of MDM2-inhibitors such as RG7112 (28) and DS-3032b (29) in patients with MDM2-amplified liposarcoma. A GULP1-PTCH1 fusion, likely inactivating, was detected in one case and was classified as a level 3B potentially actionable alteration by OncoKB. PTCH1, a tumor-suppressor gene and inhibitor of the hedgehog pathway, is recurrently mutated in basal cell carcinoma (30, 31). Currently, there are promising clinical data to support the use of hedgehog pathway inhibitors such as sonidegib (30) and vismodegib (31) in patients with basal cell carcinoma harboring truncating PTCH1 mutations.

PTEN deletion and truncating mutation were identified in two of 66 patients (3%). PTEN, a tumor-suppressor gene and phosphatase, is one of the most frequently altered genes in cancer. Although there are no FDA-approved or NCCN-compendium listed treatments specifically for patients with PTEN-deleted bone cancer, functional studies and clinical trials using ARQ 751, AZD5363+olaparib, AZD8186, GSK2636771, and palbociclib + gedatolisib are in progress for various malignancies (32–41). CDKN2A alterations were identified in 18 cases (27%), and an NF1 deletion was identified in a single case.

A previously underappreciated prevalence of 4q12 amplification, including KIT, KDR, and PDGFRA, was noted in this series, being identified in 13 of 66 patients (20%; Figs. 1A and B and 2; Tables 2 and 4). Of the 13 patients with 4q12 amplifications, IHC was performed for PDGFRA [Clone: 1C10; Novus (NBP2-46357); 1:600 (1.7 μg/mL)] on nine patients with available material: tumors from eight of nine patients showed strong cytoplasmic expression (2+ to 3+ intensity; Fig. 2), whereas one showed weak expression (1+). IHC was also performed for KDR [VEGF Receptor 2; Clone: 55B11; Cell Signaling Technology (2479); 1:250 (0.1 μg/mL)] on five patients with available material and two of these showed focal cytoplasmic expression (Supplementary Fig. S1). IHC for KIT [Clone: YR145; Cellmarque (117R); 1:300 (0.1 μg/mL)] was negative in this subset of cases.

These findings may provide a rationale for closer evaluation of multikinase inhibitors targeting these kinases. For example, pazopanib and regorafenib both target VEGFR, PDGFR, and KIT (42–44). Interestingly, both agents have been recently shown to produce objective responses in a subset of patients with osteosarcoma. Furthermore, olaratumab, an mAb to PDGFRA (45), could be evaluated in patients in this 4q12-amplified subset of osteosarcoma.

VEGFA at 6p12 was amplified in 14 of 59 patients (24%), pointing to angiogenesis pathways as potential targets in this subset of patients with osteosarcoma (Fig. 1A and C). Several antiangiogenic agents have shown in vitro and in vivo antitumor activity in osteosarcoma in association with amplification of VEGF (46–51). Clinical studies have reported activity of antiangiogenic therapies such as antibodies and small-molecule inhibitors which target the VEGF–VEGFR axis in some patients with osteosarcoma (52–54), a subset that we now speculate may represent VEGFA/6p12-amplified cases. Sorafenib has also been shown to produce long-lasting partial responses in a small subset of osteosarcoma (55), and intriguingly, it has also been shown to be effective in VEGFA-amplified hepatocellular carcinoma (56).

No significant differences were found between pediatric and adult osteosarcoma groups in the frequency of potentially actionable alterations, commonly altered genes, or distinct molecular subsets. Furthermore, we did not identify any molecular alterations that were unique to pediatric or adult osteosarcoma cases. However, we did find differences in overall TMB (see below).

The samples obtained from primary site included samples from pretreatment biopsies (24 samples) as well as posttreatment resections (11 samples; Table 4). The frequency of the most common CNAs was then calculated for each of the specimen types. Amplification of 6p12-21 including VEGFA was identified in 14 of 34 metastatic/recurrent samples (41.2%) as compared with three of 31 primary samples (9.7%; Fig. 1A; Table 4). This difference was found to be statistically significant (P < 0.01, χ² test). Overall, the 37 metastatic/recurrent samples in the cohort were enriched for amplification of 12q14 including MDM2 (10 samples, 27%), but the differences did not reach statistical significance (Fig. 1A; Table 4). When cases were divided into two prognostic groups based on the development of recurrence and/or metastasis within 5 years of diagnosis, cases with 6p12-21 gain showed a trend toward faster disease progression (recurrence and/or metastasis within 5 years) when compared with the rest of the cohort (32.1% vs. 12.8%, P = 0.05, χ² test). No differences were observed in overall or disease-free survival between groups with different genomic alterations (data not shown).

Four cases had two or more samples tested (highlighted samples in Supplementary Table S7). All cases with multiple samples were posttreatment metastatic specimens that lacked matched primary tumor data. In three of four cases, the alterations found were concordant across samples, with some alterations identified at subthreshold levels that did not meet criteria for clinical reporting (Supplementary Table S7). In one patient, where both samples were posttreatment lung metastases resected one and 1.5 years after initial presentation, only one of the two samples showed an MDM2 amplification (samples 34 and 35, Supplementary Table S7).

The range of TMB scores, based on the ratio of nonsynonymous somatic mutations to sequencing territory (adjusted for MSK-IMPACT version), spanned 0.9 to 16.7 mutations/Mb (Fig. 1A). The average TMB for patients with an age of diagnosis up to 18 years was lower (1.9 mutations/Mb) than patients aged 19 years or older at disease presentation (2.9 mutations/Mb; t test, P < 0.05).

Knowledge of a tumor's genetic profile has proved to be useful in diagnosis, prognosis, and targeted therapy selection for a variety of common and rare cancers including sarcomas (11, 57–61). High-grade osteosarcomas are genetically unstable tumors with generally complex, chaotic karyotypes (62). Their genomic instability is highlighted by high levels of somatic structural variations and many CNAs (63–67). Whole-genome sequencing studies have shown recurrent TP53, RB1, and ATRX somatic mutations (64, 68–70). TP53, RB1, CDKN2A/B, CDKN2AP14ARF, and CDKN2AP16INK4A have been previously shown to be frequently affected by deletions and/or LOH, whereas MDM2 and VEGFA have been the most frequent amplified genes previously reported (64, 68–74).

In the present study, the findings of recurrent gene amplifications of CDK4, MDM2, KIT, PDGFRA, KDR, and VEGFA raise the possibility of an umbrella protocol using targeted therapeutics in distinct subsets of patients with osteosarcoma (Fig. 3). Approximately 20% of tumors in this study harbored a chromosome 4q12 amplification, encompassing the genes encoding the targetable receptor tyrosine kinases PDGFRA, KDR, and KIT. KIT has been previously proposed as a target in osteosarcoma (75). IHC analysis of this cohort confirmed strong expression of PDGFRA, moderate expression of KDR, and only weak expression of KIT, suggesting a rationale for combined PDGFRA/KDR inhibition. Recent reports have described patients with osteosarcoma with clinical responses to single-agent multikinase inhibitors with activity against PDGFRA and KDR (42, 76, 77). Although correlative genomic data for these responders were not reported, these findings are compelling for a formal trial of combined PDGFRA/KDR inhibition in 4q12-amplified osteosarcoma. If possible, it would be informative to correlate responses in trials of regorafenib (77, 78) and pazopanib (NCT01759303) for patients with recurrent osteosarcoma with the genomic amplification profiles of the tumor specimens. In a recent study by Holme and colleagues, 18 osteosarcoma cell lines were tested for chemosensitivity to 79 small-molecule inhibitors, and MG-63, an osteosarcoma cell line with PDGFRA amplification, showed sensitivity to imatinib and sunitinib (79).

Approximately 24% of patients in our cohort harbored a 6p12 amplification, involving VEGFA and CCND3. Moreover, our study identified this group of tumors as almost entirely mutually exclusive from tumors harboring 4q12 gene amplifications. Similar to PDGFRA and KDR in 4q12-amplified tumors, VEGFA is a candidate driver that is potentially targetable through kinase inhibition. In IHC studies, the expression of VEGF has been detected in 63% to 74% of osteosarcoma samples and has been associated with pulmonary metastasis, decreased disease-free survival, and overall survival (46, 80). Our study shows a significantly higher proportion of metastatic/recurrent samples harboring VEGFA (14/34 samples, 41.2%) as compared with samples procured from primary sites (3/31 samples, 9.7%; P < 0.01). Furthermore, VEGF signaling inhibition has been reported to suppress cell growth and enhance apoptosis in osteosarcoma cell lines (81, 82). In another study, 32 of 50 osteosarcoma showed VEGFA amplification (46) which was associated with decreased tumor-free survival and increased microvascular density (46, 83). Several antiangiogenic agents have been shown to have antitumor activity against osteosarcoma in vitro and in vivo (44–47, 49). In particular, pazopanib, which targets VEGF, has shown activity in preclinical mouse models with high expression of VEGF (84). As mentioned above, recent reports of clinical responses to pazopanib in small patient cohorts have been published (42). Sorafenib, another multikinase inhibitor with activity against VEGF, demonstrated significant clinical activity in a very small subset of patients with recurrent osteosarcoma (55). In hepatocellular carcinoma, tumors with VEGFA amplifications are distinctly sensitive to sorafenib (56). In a recent study by Sayles and colleagues, whole-genome sequencing performed on tumor specimens from 23 patients with osteosarcoma showed VEGFA amplification in 23% (85). In the same study, patient-derived tumor xenografts with VEGFA amplification showed significant decrease in tumor volume on treatment with sorafenib (85). Together, these findings suggest that osteosarcoma with 6p12 amplifications may be good candidates for VEGF inhibition (42, 76).

Among other potentially targetable alterations, we identified MDM2 amplification in 9 of 66 (14%) patients, including 6 cases (9%) with coamplification of CDK4 and MDM2. Earlier studies using a variety of methods have reported MDM2 amplification in 6.6% to 14.3% of osteosarcoma (21, 86, 87), and recently whole-genome sequencing studies identified MDM2 amplification in 3.1% to 5.1% of osteosarcoma (70). In clinical trials, MDM2 inhibitors have shown significant antitumor activity in patients with liposarcoma (23, 24). Some MDM2 inhibitors also display significant activity in MDM2-amplified osteosarcoma cell lines (e.g., SJSA) in comparison with non–MDM2-amplified cell lines (88, 89). CDK4 overexpression has been reported in about 10% of osteosarcoma (22, 87, 90). However, to the best of our knowledge, there have been no studies examining the association between CDK4 amplification and the activity of CDK4 inhibitors in osteosarcoma. In well-differentiated and dedifferentiated liposarcomas, several clinical trials have shown that treatment with a CDK4 inhibitor was associated with favorable progression-free survival in patients with CDK4 amplification (23, 24). Based on these findings, targeting of MDM2 and CDK4 appears to be a potential therapeutic option for the 12q13-amplified subset of patients with osteosarcoma.

Mutually exclusive genetic alterations often point to important alternative oncogenic pathways. There were several notable relationships of this type in our dataset. The 17 samples with VEFGA/CCND3 amplification at 6p12-21 were mutually exclusive with the 13 samples with amplification of PDGFRA, KIT, and KDR, at 4q12, with one exception (Log OR, −1.87; Supplementary Table S5). In the single case with gains at both loci, the 4q12 amplification was higher, whereas the 6p12 gain was borderline (results not shown). Amplification of 12q14 (MDM2 and CDK4) was found in 20% (14/71) of the samples and was mutually exclusive with 4q12 amplification (Log OR←10; Supplementary Table S5). These mutually exclusive and targetable oncogenic pathways may represent distinct biological subsets of osteosarcoma with important therapeutic implications. It should be noted that the major copy-number gains highlighted in Fig. 3 could also be detected by methods other than the one used in the present study, such as FISH or array-based copy-number profiling, which might be more widely available. In summary, we were able to identify potentially actionable (OncoKb levels 1–3) somatic alterations in approximately 21% of patients with osteosarcoma (14/66). In addition, distinct osteosarcoma subsets defined by amplification of PDGFRA and KDR at 4q12 or VEGFA at 6p12-21 may offer new therapeutic opportunities.

G. Jour is a consultant/advisory board member for Bristol-Myers Squibb. E. Slotkin reports receiving other commercial research support from Eli Lilly. P. Myers has immediate family members who have received speakers bureau honoraria from Genentech; holds ownership interest (including patents) in Amgen; and is a consultant/advisory board member for Eli Lilly, Astellas, Takeda, and Boehringer. M. Ladanyi is a consultant/advisory board member for Bayer. No potential conflicts of interest were disclosed by the other authors.

Conception and design: Y. Suehara, G. Jour, P. Meyers, J.H. Healey, M. Hameed, M. Ladanyi

Development of methodology: S. Middha, A. Zehir

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Suehara, D. Alex, S. Middha, A. Zehir, L. Wang, G. Jour, T. Hayashi, A.A. Jungbluth, E. Slotkin, J.H. Healey

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D. Alex, A. Bowman, S. Middha, A. Zehir, D. Chakravarty, L. Wang, G. Jour, K. Nafa, T. Hayashi, N. Shukla, P. Meyers, J.H. Healey

Writing, review, and/or revision of the manuscript: Y. Suehara, D. Alex, A. Bowman, S. Middha, A. Zehir, D. Chakravarty, K. Nafa, T. Hayashi, E. Slotkin, N. Shukla, P. Meyers, J.H. Healey, M. Hameed, M. Ladanyi

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.A. Jungbluth, D. Frosina

Study supervision: M. Hameed, M. Ladanyi

This research was supported in part by the NCI of the NIH (P30 CA008748). Y. Suehara was supported by a Grant-in-Aid from the Japan Society for the Promotion of Science (JSPS) KAKENHI (grant number 15KK0353).

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.