Prospective Genotyping of Hepatocellular Carcinoma: Clinical Implications of Next-Generation Sequencing for Matching Patients to Targeted and Immune Therapies

Hepatocellular carcinoma (HCC), a leading worldwide cause of cancer-related morbidity and mortality, frequently presents as incurable liver-limited or widespread metastatic disease (1, 2). Systemic treatment options include several multitargeted tyrosine kinase inhibitors (TKI) with meaningful survival advantages over best supportive care (3–5), as well as immune checkpoint inhibitors with documented durable antitumor activity (6). Nevertheless, the majority of patients will not respond to standard systemic agents or ultimately progress on these therapies. Thus, the development and validation of biomarkers to aid in treatment selection for this heterogenous disease remains of critical importance.

The genetic landscape of hepatocellular carcinoma has been studied extensively (7–16). This body of work has identified recurrent somatic oncogenic drivers or tumor suppressors in several cellular pathways (i.e., telomere maintenance, WNT signaling, cell-cycle control, chromatin remodeling, TP53, PI3K–TOR and MAPK pathways), differential genomics subsets, etiologic-dependent genomic heterogeneity, and several potential therapeutic targets. The vast majority of profiled tumor specimens though were collected from patients with early-stage disease, lack response, and disease-specific outcome data following treatment with contemporary therapies (i.e., TKIs and immune checkpoint inhibitors), or were not analyzed in real-time, and thus could not have been used to inform clinical management.

The purpose of this study was to demonstrate the clinical utility of prospective next-generation sequencing in patients with advanced HCC, to explore potential predictive and prognostic genomic biomarkers of treatment response in this population, and to define the value of sequencing in matching patients to molecularly driven therapeutics.

Patients were identified from August 2014 through August 2017 and were eligible if they had a confirmed histologic diagnosis of conventional HCC. Those with fibrolamellar variant or biphenotypic morphology such as mixed HCC-cholangiocarcinomas were excluded. Written informed consent for tumor profiling was obtained from each patient on a prospective genotyping protocol (clinicaltrials.gov, NCT01775072) that was approved by the Memorial Sloan Kettering Cancer Center (MSKCC) Institutional Review Board. The study was conducted in accordance with the U.S. Common Rule.

The electronic medical record was reviewed to extract information on patient sex, date of birth, race, HCC etiologic factor [hepatitis B virus (HBV), hepatitis C virus (HCV), and nonviral], date of diagnosis, tumor histology, histologic grade, specimen location (liver, local recurrence, or extrahepatic metastasis), extent of disease [Barcelona Clinic Liver Cancer (BCLC) staging and proportion with intrahepatic, extrahepatic, and/or gross vascular invasion], treatment history (transplantation, surgical resection, and/or regional therapy), Child-Pugh (CP) classification, alpha-fetoprotein (AFP, ng/mL), type, number and dates of systemic therapy annotated with radiographic response per the treating physician, the last date of follow-up or date of death, and vital status.

All tumor samples were reviewed and confirmed as HCC by at least two gastrointestinal pathologists and with diagnostically challenging cases review was completed at gastrointestinal pathology consensus conference. After pathologic review, formalin-fixed paraffin-embedded tissue blocks were sectioned (5–20 μm) and macrodissected when appropriate. Tissue was deparaffinized and DNA extracted using DNeasy Blood & Tissue Kit (Qiagen).

Genomic DNA from tumor tissue and patient-matched normal blood was subjected to targeted NGS using MSK-IMPACT, a custom, deep-coverage targeted sequencing assay authorized by the New York State Department of Health and the FDA as a clinical test (17). The standard input of DNA was 250 ng, and a DNA minimum input of 50 ng was used in cases where DNA quantity was limited. Briefly, barcoded DNA libraries from tumor and normal samples were captured using custom oligonucleotide probes, sequenced on an Illumina HiSeq 2500 platform, and subjected to a custom analysis pipeline to identify single-nucleotide variants, small indels (<30 bp), copy number alterations (CNA), and selected structural rearrangements in all exons and select introns of 341 or more known cancer-associated genes, as described previously (17). As the assay was expanded during the study period, 341, 410, or 468 genes were interrogated in 21, 70, and 36 cases, respectively (Supplementary Table S1). All candidate variants were manually reviewed using the Integrative Genomics Viewer (18). Testing was performed in a CLIA-certified clinical laboratory.

Tumor mutation burden (TMB) per sample was calculated as the total number of nonsynonymous mutations divided by the actual number of bases analyzed. MSIsensor scores were calculated for all cases and represent the percentage of unstable microsatellites of all tested microsatellites (19). MSIsensor interrogates the aligned sequencing data for available microsatellite regions with sufficient coverage in a tumor/normal pair where it identifies deletion length variation. χ² test is used to identify significantly varied loci, and the percentage of unstable loci, after multiple testing correction on the P values, is reported as a MSIsensor score. Tumors with values ≥10 were defined as microsatellite instability-high (MSI-H) status.

Genomic alterations were filtered for oncogenic variants using OncoKB (20), a precision oncology knowledge-base that tracks the effects of cancer variants and their potential clinical actionability (http://oncokb.org). Genomic alterations were classified as actionable using a level of evidence scale of 1 to 4, where Level 1–2A alterations indicated a standard therapeutic intervention and levels 2B-4 included investigational therapeutic alterations, which may direct a patient toward a clinical trial relevant to that biomarker.

All data are available for visualization and analysis via the cBioPortal for Cancer Genomics (http://cbioportal.org/study?id=hcc_mskimpact_2018; refs. 21, 22), and variant calls can be downloaded from the cBioPortal DataHub (https://github.com/cBioPortal/datahub).

All HCC tumor samples sequenced over the prospective sampling period were included in the analysis. Two-sided Fisher exact test was performed to compare the frequency of oncogenic alterations to clinicopathologic characteristics within the MSK cohort and to those of The Cancer Genome Atlas (TCGA) for HCC (7). Mutual exclusivity was assessed by calculating the OR and FDR-corrected P value for each gene alteration. In a descriptive fashion, TMB for all solid tumors in cBioPortal for Cancer Genomics was determined to illustrate the relative degree of hypermutation for HCC in comparison with other solid tumors.

Radiographic response was adjudicated by the treating physician in the case of patients treated with sorafenib and by RECIST version 1.1 for patients treated with immune checkpoint inhibitors, everolimus, or other investigational agents (23). Patients who received concurrent regional therapy such as embolization or radiotherapy were excluded from clinical correlation due to an inability to attribute objective response to systemic therapy. The objective response rate (ORR) was defined as the proportion of confirmed complete responses (CR) or partial responses (PR) in all evaluable patients for a specific therapy. The disease control rate (DCR) was defined as the proportion of CR, PR, and stable disease (SD) for ≥ 4 months in all evaluable patients for a specific therapy. Fisher exact test was performed to assess the impact of select clinicopathologic features and gene pathway alterations on response. Wilcoxon rank-sum test was used to assess the impact of median TMB on objective response to immune checkpoint inhibitors.

Progression-free survival (PFS) and overall survival (OS) for patients treated with sorafenib and immune checkpoint inhibitors was calculated from the date of start of treatment to the date of radiographic disease progression, death, or last evaluation. PFS and OS were calculated using Kaplan–Meier methodology. Given the sample size of our cohort against the total number of evaluated genes (>341), multiple comparisons for each altered gene would diminish power and interfere with meaningful interpretation. We therefore grouped genes into known oncogenic signaling pathways operant in HCC, specifically the TP53 pathway, WNT signaling, cell-cycle control, chromatin remodeling, PI3K–TOR, and the MAPK pathways (24), and investigated associations between these pathways and PFS and OS by log-rank test. HRs were report with 95% confidence interval (95% CI). P value of ≤0.05 were considered significant. Statistical analysis was completed using Rv3.4.3 (https://www.R-project.org).

Over the study period, a total of 134 HCC samples were analyzed from 127 unique HCC patients. In comparison with the TCGA dataset of surgically resected, liver-limited tumors, the MSK cohort was enriched with specimens from patients who were not eligible for curative resection or transplantation (61%), and included, a subset of locally recurrent/extrahepatic (16%) tumors (Fig. 1A; Table 1; ref. 7). Similar to the TCGA, the majority of the MSK cohort were male (79%) of non-Asian descent (76%) with equal proportions of virally mediated (51%) and nonvirally associated (49%) HCC.

Eight-four (63%) of 134 specimens were obtained from core biopsies, while the remaining samples were obtained from resection or hepatic explants. Median estimated tumor purity for sequenced samples was 60% (range, 10%–90%) and a mean sequencing coverage of 690× was achieved (minimum 120×). We identified an average of 4.7 nonsynonymous mutations per tumor sample (range, 1–14).

Genomic alterations were observed in 127 of 134 unique samples (95%). The most frequently mutated genes occurring in ≥10% of all samples included TERT (55.6%), CTNNB1 (35.7%), TP53 (32.5%), ARID1A (12.7%), and TSC2 (10.6%; Fig. 1B). In addition to mutations in CTNNB1, alterations in canonical members of the WNT pathway such as AXIN1 (6.3%), AXIN2 (0.05%), and APC (5.2%) were also identified. CNA included amplifications in CCND1/FGF19 (7%), MYC (6.1%), MET (1.5%) and VEGFA (<1%), and loss in HLA-B (2.7%). Structural alterations, specifically fusion events, were uncommon. Median TMB in HCC was 4.08. When compared with other solid histologies, the variance in TMB was low (SD = 2.71), second only to germ cell tumors (SD = 1.75), and MSI-H and hypermutation were not observed in this tumor type in the MSK cohort (Supplementary Fig. S1). Despite differences in the baseline clinical characteristics between the MSK and TCGA cohorts, a comparison of mutational frequencies did not reveal a significant difference between the relative proportion of altered genes (Supplementary Table S2).

Consistent with prior tumor genomic profiling efforts, TP53 mutations were rarely seen in conjunction with WNT pathway alterations (P = 0.003), effectively delineating at least two major genomic subsets (Fig. 1C). Clinicopathologic features, including sex, ethnicity, specimen location (liver limited, local recurrence, and extrahepatic metastasis), and histologic grade did not associate with genomic alterations. Virally mediated HCC, when compared with nonvirally mediated HCC, was enriched for TP53 mutations (45% vs. 21%, P = 0.001) and was numerically less commonly associated with WNT pathway alterations (41% vs. 60%, P = 0.7).

Of the 127 unique HCC patients, 87 (68.5%) received systemic therapy while 38 (29.9%) were not candidates for systemic treatment as these patients were free of disease, undergoing serial embolization, or were unfit for systemic treatment. Two patients (1.6%) had missing data regarding systemic therapy. Eighty-one patients were treated with sorafenib, a multitargeted TKI, as a standard of care or in the context of a clinical trial (Table 2). Two patients were excluded from analysis as they were lost to follow-up. The median sorafenib PFS for the cohort overall was 4.8 months, which is similar to contemporary large first-line pivotal studies of sorafenib (3.6 months; ref. 5), and the median OS was a favorable 16.4 months (12.2 months; ref. 5).

To identify potential genomic biomarkers of response to sorafenib, we correlated genomic pathway alterations with treatment response and outcomes. We observed that patients whose tumors harbored activating mutations in the PI3K–mTOR pathway exhibited lower rates of clinical benefit (DCR 8.3% vs. 40.2%, P = 0.05; Fig. 2A) from sorafenib treatment and had a shorter median PFS [1.9 vs. 5.3 months, HR 3.8 (95% CI, 2.0–7.5) P < 0.0001; Fig. 2B] and median OS [10.4 vs. 17.9 months, HR 2.5 (95% CI 1.21–5.31), P = 0.01; Fig. 2C] when compared with patients without such mutations. Mutations predicted to activate the WNT or MAPK pathway (including alterations in receptor tyrosine kinases that are potential targets of sorafenib) as well as the TP53 pathway, cell-cycle control, and chromatin remodeling had no effect on outcomes (Supplementary Table S3). Finally, VEGFA amplification, a proposed biomarker for extreme sorafenib responders (25), was rarely observed (3.9%) and at least anecdotally did not appear to associate with clinical benefit (Fig. 2A).

As emerging data indicate that a subset of patients with liver cancer have durable responses to immune checkpoint inhibitors, we sought to explore genomic biomarkers of responsiveness or resistance in the 31 patients within the 127 patient MSK cohort treated with immune checkpoint inhibitors (Table 3). Patients received immunotherapy either as standard of care, on a compassionate use program, or in the context of a clinical trial. Of the 31 patients, 1 patient received anti–CTLA-4 monotherapy, 25 patients received monotherapy with a mAb targeting PD-1 or PD-L1, and 5 patients received an anti–PD-1 antibody in combination with antibodies targeting CTLA-4, KIR, or LAG-3. Twenty-seven patients were evaluable for radiographic response; 1 was excluded due to concomitant treatment with radioembolization, and 3 were not evaluable due to incomplete imaging or nonmeasurable disease. In total for evaluable patients, the ORR was 11.1%. We observed 1 (3.7%) CR, 2 (7.4%) PR, 10 (37%) SD, and 14 (51.9%) progression of disease as best response. The median PFS and OS from the start of immune checkpoint inhibitor therapy was of 5.4 and 12.9 months, respectively.

As WNT pathway activation results in T-cell exclusion and innate resistance to immune checkpoint blockade in vivo (26, 27), we correlated radiographic response to immunotherapy with the presence of activating mutations in the WNT pathway. Of 10 patients with WNT pathway alterations, none achieved clinical benefit (i.e., all patients had progression of disease at first interval scan), whereas 9 (53%) of 17 non-WNT pathway–altered patients had durable stable disease (≥ 4 months) or better as best response (P = 0.009; Fig. 3A). To further explore this finding, we correlated patient outcomes with genomic pathway alterations. When comparing WNT-activated tumors to nonaltered tumors, we found a shorter median PFS (2.0 vs. 7.4 months, HR 9.2; 95% CI: 2.9–28.8, P < 0.0001; Fig. 3B) and a numerically shorter median OS (9.1 vs. 15.2 months, HR 2.6; 95% CI 0.76–8.7, P = 0.11). As noted above, activating WNT pathway alterations had no effect on sorafenib median PFS (4.5 for WNT activating vs. 5.1 for non-WNT patients; Fig. 3C). No other pathways correlated with responsiveness or resistance to checkpoint inhibitor treatment.

Mutational burden has been shown to correlate with clinical benefit from treatment with immune checkpoint inhibitors in patients with melanoma, lung, and other cancer types. For the 27 evaluable patients treated with immune checkpoint inhibitors, the median TMB was 3.8—only 1 patient in the cohort had a TMB of >10. The median TMB was 3.8 and 3.9 (P = 0.98) for responders and nonresponders, respectively. The median PFS stratified by median TMB was similar (TMB <3.8: 3.1 months; TMB ≥3.8: 3.0 months, P = 0.21).

In total, 24% of patients harbored at least one actionable mutation as annotated by OncoKB (Fig. 4A; ref. 20). Consistent with the lack of validated predictive biomarkers of drug response in this disease type, no patient had a level 1 or 2A alteration. Potentially actionable mutations included truncating and inactivating mutations in the mTOR modulators, TSC1 (<1%) and TSC2 (7%), truncating or homozygous deletions in PTEN (3.9%), FGF19 (6.3%), and MET amplifications (1.5%), and oncogenic missense mutations in HRAS, NRAS, and PI3KCA (all <1%).

Five (5.7%) of 87 patients who received systemic therapy were matched to treatments targeting mTOR or MET based on the clinical genomic report. Four (40%) of 10 patients who harbored alterations in TSC1/2 received everolimus, an mTOR inhibitor, off-label or on an investigational study (Fig. 4B). Of the 2 patients with MET amplifications, one patient with a 3.7-fold VEGFA and 2.1-fold MET-amplified tumor received mAbs to VEGFR2 and MET in the context of a clinical trial. For 4 of these 5 patients, first-line sorafenib treatment was ineffective with radiographic progression of disease within 1.5 to 2.5 months. Second-line targeted therapy resulted in stabilization of disease at first interval scan in all 4 patients—3 of 4 patients had clinical benefit for 5 months including 1 patient who rapidly progressed on prior sorafenib who had clinical improvement and disease stabilization with some areas of tumoral shrinkage (Fig. 4C). The fifth patient received everolimus as a first-line treatment for 5.0 months until progression of disease.

In this study, we demonstrate the utility of prospective sequencing of advanced liver-limited or metastatic HCC in clinical practice. We confirm genomic findings derived from retrospectively analyzed cohorts of surgically-resected, early-stage, liver-limited HCCs and extend these observations to an advanced patient population more representative of those treated as part of routine clinical practice. As our sampling and analysis strategy occurred in tandem with clinical care, we were able to explore potential predictive and prognostic genetic factors to contemporary therapies, such as sorafenib and immune checkpoint inhibitors. We found that mutations predicted to activate the PI3K–mTOR pathway were associated with poor outcomes in sorafenib-treated patients and that mutations predicted to activate the WNT pathway were associated with innate resistance to immune checkpoint blockade. Furthermore, potentially actionable mutations were identified in a small subset of patients and NGS permitted real-time interpretation and utilization of results in clinical practice.

Our data are consistent with prior reports regarding the frequency and overall distribution of oncogenic alterations in HCC. Specifically, we confirm that mutations predicted to result in WNT/β-catenin activation are present in nearly half of all HCCs, and represent a unique biological subset (28). While other series have identified a MSI-H/mismatch repair–deficient phenotype and/or a high-to-moderate TMB in a small minority of patients, we did not observe any such cases in this prospective series. This may have been due to demographic differences in the populations studied or to the advanced nature of our patient cohort, as MSI-H has been shown to be a favorable prognostic indicator in other cancer types. Finally, except for slight enrichment of TP53 mutations in virally mediated HCCs, genomic alterations did not correlate with specific HCC etiologic factors (7–16).

Multitargeted TKIs have altered the natural history of HCC, and sorafenib is an FDA-approved first-line standard systemic therapy (3, 5, 29). Aside from VEGFA amplifications, which are rare, few genomic biomarkers of drug response have been identified in retrospective studies of multitargeted TKIs. As a signaling pathway downstream from the putative targets of sorafenib, PI3K–mTOR activation, either through primary genomic alterations or negative feedback loops, could abrogate the dependence of HCCs on upstream kinases, and thus may mediate resistance to multitargeted TKIs (30–33). Indeed, preclinical models indicate that sorafenib-resistant HCCs utilize PI3K–mTOR as one potential bypass pathway, and PI3K pathway inhibition via RNA interference or small-molecule inhibitors, restores sorafenib responsiveness (34–36). In clinical samples, mTOR activation has been associated with resistance to sorafenib and our data corroborate these findings (30). mTOR activation also portends worse clinical outcomes postresection in patients with localized HCC (37) and in the metastatic setting (33, 38). Thus, pharmacologic interference with mTOR signaling (37–42) may have clinical import in this HCC molecular subset (38), and several platform/umbrella studies, such as NCI-MATCH (NCT02465060) and ASCO TAPUR (NCT02693535), are evaluating the clinical utility of PI3K–mTOR blockade across cancer histologies.

Despite the clear efficacy of immune checkpoint inhibitors in HCC, few patients respond to this treatment. Our data suggest that TMB (43–45) and mismatch repair deficiency/microsatellite instability (46, 47), are unlikely to be major determinants of response to immunotherapy in this disease. On the other hand, the prevalence of WNT/β-catenin alterations in HCC may have important implications for immune-based treatment. Rendering mutant tumors immunologically cold, oncogenic WNT signaling leads to T-lymphocyte exclusion (26, 27) and insensitivity to combination anti–PD-L1 and anti–CTLA-4 mAb therapy in vivo (26). Herein, we present clinical data indicating that patients with HCC with mutations predicted to result in WNT pathway activation are innately resistant to immune checkpoint blockade. In our series, all patients with tumors harboring activating CTNNB1 or inactivating AXIN1 mutations had progressive disease as best response whereas patients with non-WNT pathway–altered tumors were significantly more likely to respond or derive clinical benefit from immunotherapy. As WNT pathway alterations did not correlate with sorafenib PFS and OS, WNT activation may indeed be a predictive biomarker of resistance to immune checkpoint blockade in HCC. Acknowledging these data are preliminary and may be confounded by both known and unknown clinicopathologic factors not controlled for due to the small sample size of the cohort, we believe our clinical findings will prompt further preclinical investigations that will further elucidate the biologic underpinnings of these intriguing initial observations. Moreover, larger retrospective datasets and prospective evaluation of this preliminary observation are imperative, and if validated, this finding would be highly impactful for both patient care and clinical trial design. We also note that other factors, such as HLA-I heterogeneity (48), may also contribute to innate resistance, and such factors also deserve further exploration.

Although the direct clinical impact of prospective NGS for HCC clearly has less utility than in other solid tumors such as lung cancer and melanoma (49), 24% of tumors had potentially actionable alterations that could be targets for currently available FDA-approved drugs, or agents in active clinical development. As an example, our data and other series specifically support continued investigation of mTOR inhibitors in tumors with TSC1/2 alterations (50). Albeit, it is important to recognize that at least in this small case series, mTOR inhibition appeared to have modest cytostatic benefit at best. Emerging data also indicate that an amplicon that includes the FGF19 gene may contribute to HCC oncogenesis, and that inhibition of this pathway with FGF19 mAbs or with highly selective inhibitors of FGFR4, the receptor for FGF19, could be clinically effective in patients with HCC (51). Indeed, several compounds are now in clinical development, with promising preliminary data. Although rare, IDH altered HCCs (<1%) also likely represent a unique biologic subset that exhibit clinical and genetic features similar to cholangiocarcinomas (7), and such tumors may be susceptible to selective small-molecule inhibitors of IDH1 (52). Although initial studies targeting MET (1.5%) failed to demonstrate meaningful clinical activity (53), more potent and specific MET inhibitors are being evaluated, and such efforts have been bolstered by recent positive data with cabozantinib.

Our study has several notable limitations. First, the targeted capture approach used in this study could not by definition detect alterations in genes not included in the assay design, epigenetic mechanisms of gene suppression, or viral HBV DNA sequences. Second, tumor heterogeneity is well documented in HCC and sampling a single site of disease can never fully assess clonal complexity in patients with multisite metastatic disease. To address the issue of tumor heterogeneity, future studies in patients with HCC could seek to include an assessment of genomic alterations in circulating cell-free DNA or in single cells. Third, given the small sample size of the current study, potential confounding variables, such as HCC etiology, AFP levels, extrahepatic spread/vascular invasion, should be evaluated in future studies to assure such factors are not the predominant determinants of therapeutic outcome. To address this limitation, all clinical and genomic data from this study have been made available through the cBioPortal for Cancer Genomics to facilitate future meta-analyses. Finally, the favorable OS observed in the sorafenib cohort compared with the published literature, suggest a selection bias but may also be due in part to access to novel therapies and immunotherapies at our center as well as the high proportion of patients with intact hepatic function in the cohort. Despite these limitations, our data represent the first attempt to link real-time NGS to clinical practice and provide needed insight into feasibility and the potential impact of this investigational modality to direct patient care.

J.J. Harding reports receiving commercial research grants from Bristol-Myers Squibb, and is a consultant/advisory board member for Bristol-Myers Squibb, CytomX, Eisai, and Eli Lilly. J.F. Hechtman reports receiving speakers bureau honoraria from Medscape. M.F. Berger is a consultant/advisory board member for Roche. D.M. Hyman reports receiving commercial research grants from AstraZeneca, Loxo Oncology, and Puma Biotechnology, and is a consultant/advisory board member for AstraZeneca, Atara Biotherapeutics, Bayer, Boehringer Ingelheim, Chugai Pharma, CytomX Therapeutics, Genentech, and Pfizer. D.S. Klimstra holds ownership interest (including patents) in Paige.AI, is a consultant/advisory board member for Merck and Paige.AI, and received royalties from American Registry of Pathology and UpToDate. Y.Y. Janjigian reports receiving commercial research grants from Bayer, Boehringer Ingelheim, Bristol-Myers Squibb, Eli Lilly, Genentech/Roche, and Merck, and is a consultant/advisory board member for Bayer, Bristol-Myers Squibb, Eli Lilly, Imugene, Merck, Merck Serono, and Pfizer. D.B. Solit is a consultant/advisory board member for Illunina, Loxo Oncology, and Pfizer. G.K. Abou-Alfa is a consultant/advisory board member for 3DMedcare, Agios, Alignmed, Amgen, Antengene, Aptus, Aslan, Astellas, AstraZeneca, Bayer, Beigene, Bioline, Boston Scientific, BridgeBio, Bristol-Myers Squibb, Carsgen, Celgene, Casi, Cipla, CytomX, Daiichi, Debio, Delcath, Eisai, Exelixis, Genoscience, Gilead, Halozyme, Hengrui, Incyte, Inovio, Ipsen, Jansen, Jazz, Kyowa, Kirin, Lilly, Loxo, Merck, Mina, Newlink Genetics, Novella, Onxeo, PCI Biotech, Pfizer, Pharmacyte, Pharmacyclics, Pieris, QED, Redhill, Sanofi, Servier, Silenseed, Sillajen, Sobi, Targovax, Tekmira, Twoxar, Vicus, Yakult, and Yiviva. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J.J. Harding, J. Armenia, M.I. D'Angelica, D.S. Klimstra, Y.Y. Janjigian, D.B. Solit, N. Schultz, G.K. Abou-Alfa

Development of methodology: J.J. Harding, J. Shia, J.F. Hechtman, Y.Y. Janjigian, D.B. Solit, N. Schultz, G.K. Abou-Alfa

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.J. Harding, M. Ly, J.F. Hechtman, R.K. Do, T.P. Kingham, M.I. D'Angelica, D.M. Hyman, W.R. Jarnagin, Y.Y. Janjigian, D.B. Solit, G.K. Abou-Alfa

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.J. Harding, S. Nandakumar, J. Armenia, J. Shia, M.F. Berger, D.B. Solit, N. Schultz, G.K. Abou-Alfa

Writing, review, and/or revision of the manuscript: J.J. Harding, S. Nandakumar, J. Armenia, D.N. Khalil, M. Albano, J. Shia, J.F. Hechtman, R. Kundra, I. El Dika, R.K. Do, T.P. Kingham, M.I. D'Angelica, D.M. Hyman, W.R. Jarnagin, D.S. Klimstra, Y.Y. Janjigian, D.B. Solit, N. Schultz, G.K. Abou-Alfa

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.J. Harding, M. Ly, R. Kundra, Y. Sun, D.M. Hyman, Y.Y. Janjigian, D.B. Solit, N. Schultz

Study supervision: J.J. Harding, D.B. Solit, N. Schultz, G.K. Abou-Alfa

This work was funded in part by a Cancer Center Support Grant (P30 CA008748), the Marie-Josée and Henry R. Kravis Center for Molecular Oncology, Cycle for Survival, Robertson Foundation (to N. Schultz), and the NCI (U24 CA220457).

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.