Purpose: Sorafenib is the current standard therapy for advanced hepatocellular carcinoma, but validated biomarkers predicting clinical outcomes are lacking. This study aimed to identify biomarkers predicting prognosis and/or response to sorafenib, with or without erlotinib, in hepatocellular carcinoma patients from the phase III SEARCH trial.
Experimental Design: A total of 720 patients were randomized to receive oral sorafenib 400 mg twice daily plus erlotinib 150 mg once daily or placebo. Fifteen growth factors relevant to the treatment regimen and/or to hepatocellular carcinoma were measured in baseline plasma samples.
Results: Baseline plasma biomarkers were measured in 494 (69%) patients (sorafenib plus erlotinib, n = 243; sorafenib plus placebo, n = 251). Treatment arm–independent analyses showed that elevated hepatocyte growth factor [HGF; HR, 1.687 (high vs. low expression); endpoint multiplicity adjusted (e-adj) P = 0.0001] and elevated plasma VEGFA (HR, 1.386; e-adj P = 0.0377) were significantly associated with poor overall survival (OS) in multivariate analyses, and low plasma KIT [HR, 0.75 (high vs. low); P = 0.0233; e-adj P = 0.2793] tended to correlate with poorer OS. High plasma VEGFC independently correlated with longer TTP (HR, 0.633; e-adj P = 0.0010) and trended toward associating with improved disease control rate (univariate: OR, 2.047; P = 0.030; e-adj P = 0.420). In 67% of evaluable patients (339/494), a multimarker signature of HGF, VEGFA, KIT, EPGN, and VEGFC correlated with improved median OS in multivariate analysis (HR, 0.150; P < 0.00001). No biomarker predicted efficacy from erlotinib.
Conclusions: Baseline plasma HGF, VEGFA, KIT, and VEGFC correlated with clinical outcomes in hepatocellular carcinoma patients treated with sorafenib with or without erlotinib. These biomarkers plus EPGN constituted a multimarker signature for improved OS. Clin Cancer Res; 22(19); 4870–9. ©2016 AACR.
Validated biomarkers of prognosis and response to sorafenib have not yet been identified in patients with advanced hepatocellular carcinoma. We assessed whether baseline concentrations of 15 plasma biomarkers and various combinations of these biomarkers, as well as mutations in 19 oncogenes, could predict prognosis or treatment response in patients with advanced hepatocellular carcinoma enrolled in SEARCH, a phase III trial of sorafenib with or without erlotinib. We found that high baseline plasma HGF and VEGFA correlated significantly with shorter OS, and high KIT with longer OS. In addition, high VEGFC correlated significantly with better TTP and DCR. Two multimarker signatures, one consisting of 2 markers (HGF and VEGFA) and the other of 5 markers (HGF, VEGFA, KIT, VEGFC, and EPGN), showed significant correlations with OS. These findings, if confirmed, could potentially guide clinicians on how to predict patient prognosis or response to treatment.
Hepatocellular carcinoma is the second most frequent cause of cancer-related deaths worldwide (1). Most patients are diagnosed with advanced stage disease, when curative treatments, including resection, liver transplantation, and ablation, are no longer an option (2, 3). The oral multikinase inhibitor sorafenib, which targets Raf-1, VEGFR1-3, PDGFR, KIT, RET, and other tyrosine kinases, has both antiproliferative and antiangiogenic effects (4). Sorafenib has demonstrated survival benefits in patients with advanced unresectable hepatocellular carcinoma and remains the standard of care for this disease based on two phase III trials (5, 6).
Erlotinib is a potent, orally active inhibitor of EGFR tyrosine kinase approved to treat patients with advanced non–small cell lung and pancreatic cancers (7–9). In two single-arm phase II trials, erlotinib showed modest prolonged progression-free survival and promising disease control in patients with unresectable hepatocellular carcinoma (10, 11). Moreover, a phase II trial of the combination of erlotinib and the antiangiogenic mAb bevacizumab, which binds to and inhibits VEGF, found that this combination led to clinically meaningful progression-free survival and response rates in patients with advanced hepatocellular carcinoma (12).
On the basis of these findings, the SEARCH trial was designed to compare the efficacy and safety of sorafenib plus erlotinib with sorafenib plus placebo as first-line treatment in patients with advanced/unresectable hepatocellular carcinoma (13). In this trial, 720 patients were randomized to sorafenib plus erlotinib (n = 362) or sorafenib plus placebo (n = 358), with a primary endpoint of overall survival (OS; ref. 13). The trial did not meet the primary endpoint; median OS was similar in the sorafenib plus erlotinib and sorafenib plus placebo groups (HR, 0.929; P = 0.204; 9.5 vs. 8.5 months, respectively), as was median TTP (HR, 1.135; P = 0.091; 3.2 vs. 4.0 months; ref. 13). Of note, the overall response rate (ORR) with sorafenib plus erlotinib was almost twice as high as the ORR with sorafenib plus placebo (7% vs. 4%; ref. 13). In contrast, the disease control rate (DCR) was significantly lower (43.9% vs. 52.5%, P = 0.010) in the sorafenib plus erlotinib group compared with sorafenib plus placebo, perhaps due to the shorter treatment duration in the sorafenib plus erlotinib group (13).
Despite this trial not demonstrating a benefit for the addition of erlotinib to sorafenib, much can be learned by assessing biomarkers that may predict prognosis and/or benefit from treatment. A previous prospective study demonstrated that lower baseline plasma levels of insulin-like growth factor-1 and higher plasma VEGF levels correlated with advanced clinicopathologic parameters and poor OS in patients with hepatocellular carcinoma (14). In addition, an analysis of the 602 patients in the phase III Sorafenib Hepatocellular Carcinoma Assessment Randomized Protocol (SHARP) trial randomized to receive oral sorafenib or placebo found that baseline plasma concentrations of angiopoietin 2 and VEGFA were independent prognostic predictors of patient survival in the entire patient population and the placebo cohort (15). Also in SHARP, trends toward enhanced survival benefit from sorafenib treatment were seen in patients with high plasma KIT and/or low hepatocyte growth factor (HGF) concentrations at baseline (Pinteraction = 0.081 and 0.073, respectively). The goal of the current exploratory biomarker analysis of patients in SEARCH was to identify biomarkers that may correlate with outcome in patients treated with sorafenib, or predict treatment with erlotinib versus placebo in combination with sorafenib, in patients with advanced hepatocellular carcinoma. In this biomarker study, 15 candidate mechanistic plasma biomarkers, proteins with known or hypothesized relevance to sorafenib's and/or erlotinib's mechanism of action or to hepatocellular carcinoma outcome, were examined. Analytes selected for plasma biomarker analysis are either molecular targets of sorafenib or ligands of those targeted receptors (VEGFA, VEGFC, soluble KIT, and PDGF-BB), or are ligands of the EGFR targeted by erlotinib [amphiregulin, betacellulin, EGF, EPGN, epiregulin, heregulin, heparin binding EGF (hbEGF), and TGFα], or have been implicated in the pathogenesis of hepatocellular carcinoma and were found to correlate with measures of outcome in the SHARP trial (BFGF, IGF2, and HGF; ref. 15). In addition, the mutational status of 19 genes was analyzed in available tumor samples; the hotspot mutation panel utilized includes genes targeted by sorafenib and erlotinib as well as genes reported to be mutated in hepatocellular carcinoma (16), although most at low prevalence.
Patients and Methods
Patients and samples
The SEARCH trial has been described in detail previously (13). Eligible patients with histologically or radiologically confirmed advanced/metastatic hepatocellular carcinoma not amenable to local therapies, an Eastern Cooperative Oncology Group (ECOG) status of 0 or 1, and Child-Pugh class A (determined during screening; n = 720) were randomized to treatment with sorafenib 400 mg twice daily plus either erlotinib 150 mg once daily (n = 362) or matching placebo (n = 358), stratified by ECOG performance score (0 vs. 1), macrovascular invasion and/or extrahepatic spread (yes vs. no), smoking status (current vs. former vs. never), and geographic region (North America/South America vs. Europe/South Africa vs. Asia-Pacific). The primary endpoint was OS. Secondary endpoints included TTP by independent radiologic review, DCR, ORR, and safety.
Blood samples were collected before treatment (during screening or predose on day 1 of cycle 1) into tubes containing ethylenediamine tetraacetic acid as anticoagulant. Plasma was prepared by centrifugation and removed to a separate tube. The tubes were stored at −70°C (although storage at −20°C was acceptable if no −70°C freezer was available) and shipped on dry ice every 6 weeks (4–5 weeks if stored at −20°C) to the central laboratory.
Formalin-fixed paraffin-embedded archival tumor biopsy samples or unstained slides were collected at screening.
Submission of biomarker samples and consent for tumor genetics were optional per the SEARCH protocol; all plasma samples received were assayed for plasma biomarkers, and all usable tumor samples received from patients who gave genetic consent were assayed for tumor mutations.
Candidate plasma biomarkers were chosen on the basis of their known or hypothesized relevance to the mechanisms of action of sorafenib and/or erlotinib and/or their relevance to hepatocellular carcinoma, and the planned analyses were prespecified at study design (15). All plasma biomarkers were measured by AssayGate Laboratories. Plasma concentrations of VEGFC, heregulin, soluble KIT (R&D Systems), EPGN (USCN Life Science Inc), and IGF2 (Mediagnost GmbH) were measured by commercially available ELISA kits; and plasma concentrations of VEGFA, HGF, amphiregulin, betacellulin, EGF, epiregulin, hbEGF, TGFα, BFGF, and PDGF-BB were measured using multiplex Luminex Bead Assays (Thermo Fisher Scientific), according to the manufacturer's instructions. Mutations in 19 oncogenes (ABL1, AKT1, AKT2, BRAF, CDK4, EGFR, ERBB2, FGFR1, FGFR3, FLT3, HRAS, JAK2, KIT, KRAS, MET, NRAS, PDGFRA, PIK3CA, RET) were analyzed in tumor DNA by Quintiles Laboratories using the Sequenom OncoCarta 1.0 multiplex assay system, according to the manufacturer's instructions (the list of mutations assayed is shown in Supplementary Fig. S2). All personnel associated with the laboratories performing the assays were blinded to treatment group assignments and all clinical data, including outcome.
Statistical methods and analyses
Statistical analyses were performed using SAS and R software version 9.2 (SAS Institute Inc). Clinical outcome measures included in the biomarker analyses were OS, TTP, and DCR. Each biomarker was analyzed as a continuous variable and a dichotomized variable (dichotomized using the median and an optimized cut-off point determined using the maximum χ2 method, which tests all possible cut-off points between the 25th and 75th percentiles and selects the optimal cut-off value), as well as being included in multimarker models.
For multimarker models, feature selection methods were used to identify a subset of biomarkers to be used in developing multimarker signatures associated with clinical outcomes (17). The 14 plasma biomarkers tested on a continuous scale (excluding Heregulin, for which a large proportion of values were BLQ) were included in feature selection and were analyzed as log2-transformed continuous variables, utilizing a model adjusting for treatment. Penalized Cox regression using a bootstrap elastic net (BELNET) was used to investigate the stability of feature selection. This procedure repeatedly performs feature selection on bootstrap samples (n = 50) drawn from the observed data to calculate a selection probability for each feature (i.e., the proportion of times each of the 14 biomarkers was selected across all bootstrap samples). Selection probability thresholds of 0.9 and 0.8 were considered to identify sets of features of interest. A composite score was generated using the selected biomarkers from each analysis, and optimal cut-offs were identified using the maximum χ2 method.
For each biomarker associated with prognosis (including both single biomarkers and multimarker signatures), multivariable analyses were performed that included clinical variables identified as associated with prognosis. Cox regression models were used to identify the clinical variables prognostic for OS, TTP, and DCR in hepatocellular carcinoma. The clinical variables tested for prognostic significance were as follows: age, ECOG PS, gender, geographic region, race, stage at randomization, ascites, macroscopic vascular invasion, extrahepatic spread, cirrhosis, smoking status, Child–Pugh score, BCLC score, hepatitis B, and hepatitis C. Multivariable models were then run, which included the biomarkers of interest and the identified prognostic clinical covariates.
Populations of patients evaluated for biomarkers
A total of 720 patients were randomized in the SEARCH trial, 362 to sorafenib plus erlotinib and 358 to sorafenib plus placebo (13). Plasma samples were obtained from 494 patients (68.6%) at baseline, 243 (67.1%) in the sorafenib plus erlotinib group and 251 (70.1%) in the sorafenib plus placebo group. Baseline demographic and disease characteristics of patients in the biomarker subpopulations were similar to those in the overall SEARCH population (Table 1). Clinical outcomes were also similar in the SEARCH biomarker and SEARCH overall populations. In the biomarker population, OS in the sorafenib plus erlotinib and in the sorafenib plus placebo groups was 9.7 and 8.9 months, respectively (HR, 0.922; 95% CI, 0.749–1.133), and TTP was 3.2 and 3.9 months (HR, 1.166; 95% CI, 0.937–1.450). In the overall SEARCH population, OS of the sorafenib plus erlotinib and sorafenib plus placebo groups was 9.5 and 8.5 months (HR, 0.929; 95% CI, 0.781–1.106), respectively, and TTP was 3.2 and 4.0 months (HR, 1.135; 95% CI, 0.944–1.366; Table 2).
Plasma biomarkers correlating with clinical outcomes in the full biomarker population
Median, mean, range, and 25th/75th percentiles are shown for key biomarkers in Supplementary Table S3. In the first set of analyses, the treatment arms were combined into one group because all patients had been treated with sorafenib, making this a treatment arm–independent analysis (Table 3). Plasma biomarkers were first assayed for their ability to predict OS. When dichotomized using the maximum χ2 method, high baseline HGF correlated significantly with shorter OS [HR, 1.672; 95% CI, 1.352–2.074; max χ2 P = 0.00005; endpoint multiplicity adjusted (e-adj) P = 0.0007; Fig. 1A]. High baseline HGF also correlated significantly with shorter OS when dichotomized at the median (HR, 1.595; 95% CI, 1.294–1.967; max χ2 P = 0.00001; e-adj P = 0.0002), when analyzed as a continuous variable (HR, 1.148; 95% CI, 1.070–1.233; max χ2 P = 0.0001; e-adj P = 0.0015; Table 3), as well as when analyzed among only those patients with stage IV disease (HR, 1.708; 95% CI, 1.295–2.254). High baseline VEGFA showed a trend toward correlation with shorter OS when dichotomized at the optimized cut-off (HR, 1.385; 95% CI, 1.124–1.704; max χ2 P = 0.03; e-adj P = 0.39; Fig. 1B), when dichotomized at the median, or analyzed as a continuous variable (Table 3), as well as when analyzed among only those patients with stage IV disease (HR, 1.480; 95% CI, 1.126–1.946). High baseline KIT showed a trend toward correlation with longer OS (analysis using optimized cut-off: HR, 0.713; 95% CI, 0.562–0.897; max χ2 P = 0.05; e-adj P = 0.60; Fig. 1C; analysis as a continuous variable; Table 3). In multivariate analyses including known prognostic clinical variables, HGF was independently prognostic for OS whether analyzed as a dichotomized (e-adj P = 0.0001) or continuous (e-adj P = 0.0108) variable (Table 3). VEGFA was also indpendently prognostic for OS (dichotomized, e-adj P = 0.0377; continuous, e-adj P = 0.0457), and both HGF and VEGFA remained independently prognostic when included in the same multivariable model together (Table 3). While the P values for the association between KIT and OS in multivariate analyses were 0.0233 and 0.0323 for dichotomized and continuous analyses, respectively, the adjusted P values did not reach <0.05, and thus this association is not statistically significant.
All plasma biomarkers were also assessed for correlation with other efficacy outcomes (TTP and DCR). High baseline VEGFC correlated with longer TTP when dichotomized using the max χ2 method (HR, 0.615; 95% CI, 0.493–0.767; max χ2 P = 0.0003; e-adj P = 0.0042; Fig. 1D), when dichotomized at the median (HR, 0.679; 95% CI, 0.544–0.846; max χ2 P = 0.0006; e-adj P = 0.0078), and when analyzed as a continuous variable (HR, 0.877; 95% CI, 0.806–0.957; max χ2 P = 0.0032; e-adj P = 0.0486). High baseline VEGFC also correlated with higher DCR when dichotomized at the median (OR, 1.819; 95% CI, 1.225–2.714; max χ2 P = 0.003; e-adj P = 0.0421), and showed similar trends when dichotomized using an optimized cut-off or when analyzed as a continuous variable. VEGFC remained independently prognostic for TTP in multivariable models when analyzed as either a dichotomized (e-adj P = 0.0010) or as a continuous (e-adj P = 0.0195) variable (Table 3). None of the other plasma biomarkers assayed was associated with any efficacy outcome.
Analysis of plasma biomarkers as predictors of treatment benefit
In the second, “predictive” set of analyses, differences in clinical outcome between treatment arms were analyzed in biomarker subgroups. Because one arm received sorafenib plus erlotinib and the other received sorafenib plus placebo, the biomarker data could be analyzed for correlations between biomarkers and erlotinib treatment effect, thus attempting to identify biomarkers predicting benefit from one treatment regimen over the other. For example, neither patients with low (i.e., <195.365 pg/mL) nor high (i.e., ≥195.365 pg/mL) baseline betacellulin showed significant survival benefit from the addition of erlotinib to sorafenib treatment compared with sorafenib plus placebo (unadjusted interaction P = 0.357), though a trend toward benefit from the addition of erlotinib was seen in the high betacellulin group (HR, 0.725; 95% CI, 0.522–1.004; P = 0.0531; Fig. 2). Likewise, none of the other plasma biomarkers showed a significant relationship with treatment effect, suggesting that none of the candidate biomarkers significantly predict benefit from one treatment arm over the other (Supplementary Tables S1 and S2; Supplementary Fig. S1).
In the third set of analyses, feature-selection methods were used to identify multimarker signatures associated with clinical outcomes including generation of a composite score. The multimarker signature analysis was performed for both treatment arm–independent and predictive analyses. Two biomarkers, HGF and VEGFA, met the stringent BELNET threshold of 0.9 in the treatment arm–independent analysis. When patients were divided into those with (n = 270) and without (n = 224) this signature based on a composite score, median OS was significantly longer in the former group (11.7 vs. 6.8 months; HR, 0.573; 95% CI, 0.465–0.705; P < 0.00001; Fig. 3A). Five markers (KIT, EPGN, HGF, VEGFA, and VEGFC) met the relaxed BELNET threshold of 0.8, with 339 patients with and 155 without the signature. A multimarker composite score defined by these 5 markers also showed a statistically significant association with OS in treatment arm–independent analysis (11.5 vs. 6.0 months; HR, 0.505; 95% CI, 0.407–0.627; P < 0.0001; Fig. 3B). Both the 2 marker (HR, 0.050; 95% CI, 0.016–0.151; P < 0.00001) and the 5 marker (HR, 0.150; 95% CI, 0.078–0.287; P < 0.00001) sets remained independently prognostic of OS in multivariable models including the clinical covariates identified as prognostic for OS (see footnote of Table 3). No predictive multimarker signature correlating with treatment effect (i.e., in a predictive analysis as described above) could be identified (data not shown).
Mutations in tumor samples
Only 33 tumor samples were evaluable for oncogene mutations. Of these, 30 were negative for mutations in the 19 oncogenes assessed. Tumors of 3 patients were positive for gene mutations. One patient, with both HRAS and PDGFRA mutations, was in the sorafenib plus erlotinib treatment group and had a best response of stable disease. The second patient, with an EGFR T790M mutation, was treated with sorafenib plus erlotinib, and had a best response of progressive disease. The third patient had a mutation in MET, although this T992I mutation is at best a weak activating mutation and its presence in healthy individuals suggests it may be a nononcogenic SNP (18); this patient was in the sorafenib plus placebo treatment group and had a best response of progressive disease.
This study, conducted in the setting of the phase III SEARCH trial (13), is one of the largest studies to date to attempt to identify biomarkers predictive of prognosis and/or treatment benefit of erlotinib over placebo in addition to sorafenib in patients with advanced hepatocellular carcinoma. Of the 720 patients in the overall SEARCH population, 494 (68.6%) were included in the SEARCH biomarker population. Demographic characteristics and clinical outcomes (OS and TTP) were similar in the biomarker and overall populations, indicating that the biomarker population was representative of the full study population.
In treatment arm–independent analyses of the SEARCH biomarker population, high baseline levels of plasma HGF showed a significant correlation with poorer OS. HGF is the ligand for the receptor tyrosine kinase c-MET. The HGF-MET cascade is associated with hepatocarcinogenesis (19). Elevated HGF levels have been associated with poor prognosis in patients with hepatocellular carcinoma (15), and high expression of c-MET has been associated with poor outcomes in patients with hepatocellular carcinoma treated with sorafenib (20). The current clinical results, showing that elevated HGF levels at baseline were associated with significantly shorter OS in patients with advanced hepatocellular carcinoma, are consistent with these earlier findings. The current study also indicated that high KIT levels tended to correlate with better OS.
Because of the study design, in which both arms were treated with sorafenib, it could not be determined whether correlation with outcome in these treatment arm–independent analyses was due to a biomarker's indication of prognosis or due to a biomarker's correlation with sorafenib benefit (or both); such a clear distinction would only be possible to achieve with the inclusion of an additional arm without sorafenib treatment. In the phase III SHARP trial, high HGF (in univariate analysis) and high VEGFA (in both univariate and multivariate analyses) correlated with poor prognosis, whereas KIT was not prognostic (13). These findings suggest that the correlations of HGF and VEGFA with OS observed in the SEARCH trial were due, at least in part, to the prognostic effects of these biomarkers in hepatocellular carcinoma. In addition, in SHARP, high KIT and low HGF showed a trend toward predicting greater benefit from sorafenib treatment, whereas VEGFA was clearly not predictive. These results suggest that in SEARCH, at least part of the correlation observed between KIT or HGF and OS in the treatment arm–independent analyses may be due to a role in predicting benefit from sorafenib treatment.
In the current study, in which all patients were treated with sorafenib, high baseline VEGFA, which promotes vascular angiogenesis through activation of endothelial cell associated VEGFR-1 and VEGFR-2 (21), tended to correlate with poorer OS in univariate analysis and correlated significantly in multivariate analysis; as the placebo-controlled SHARP trial showed that elevated VEGFA was prognostic of poor outcome but not predictive of sorafenib benefit, the observation in the current study that elevated VEGFA correlates with shorter survival in the full study population (all treated with sorafenib) is consistent with VEGFA having a prognostic role in hepatocellular carcinoma. In contrast to the VEGFA results, those with high baseline VEGFC, a ligand for VEGFR-2 and VEGFR-3 that promotes angiogenesis and lymphangiogenesis (21), had longer TTP in the current study, although no similar relationship was observed with OS. Although the absence of a non-sorafenib arm in the current study precludes decisive determination of whether the VEGFC result is due to a prognostic effect or is predictive of treatment benefit from sorafenib, it has been shown in previous studies that elevated tumor levels of VEGFC or peritumoral levels of VEGFC in combination with VEGFR-1 and VEGFR-3 correlate with poor prognosis in hepatocellular carcinoma, including shorter disease-free survival, time to recurrence, and OS (22–24). While these studies examined VEGFC protein levels in tissue and not in circulation, these published findings suggest that VEGFC, like VEGFA, is an indicator of poor prognosis in hepatocellular carcinoma. In contrast, a phase II study of advanced hepatocellular carcinoma patients treated with the VEGFR inhibitor sunitinib showed that elevated plasma VEGFC concentrations correlated with improved outcomes, including longer TTP and OS and increased DCR (25). This finding in combination with the VEGFC result from the current study suggests that elevated circulating VEGFC levels may enhance the antitumor activity of therapies targeting the VEGFR pathway in hepatocellular carcinoma, although the role of circulating VEGFC in hepatocellular carcinoma as compared with tumor or peritumor VEGFC is not well studied. In addition, the current study demonstrated that two multimarker signatures correlated with OS in treatment arm–independent analyses. One signature included two biomarkers, HGF and VEGFA, and the other included five markers, HGF, VEGFA, KIT, VEGFC, and EPGN.
These biomarkers were also tested in predictive analyses to determine whether their baseline concentrations correlated with treatment benefit in one treatment arm versus the other. However, none of these biomarkers, either individually or in multimarker analyses, significantly predicted differences in benefit from sorafenib plus erlotinib versus sorafenib plus placebo.
The biomarker analyses in SEARCH were exploratory and hypothesis generating. Although several potentially prognostic and predictive biomarkers were identified in the advanced hepatocellular carcinoma setting, further investigations are needed to confirm and validate their predictive and/or prognostic value.
Disclosure of Potential Conflicts of Interest
T.R.J. Evans is a consultant/advisory board member for Bayer. P. Ross reports receiving speakers bureau honoraria from Bayer and Sirtex; and is a consultant/advisory board member for Bayer, Bristol-Myers Squibb, and Sirtex. A. Vogel reports receiving speakers bureau honoraria from and is a consultant/advisory board member for Bayer and Lilly. No potential conflicts of interest were disclosed by the other authors.
Conception and design: A.X. Zhu, Y.-K. Kang, A. Santoro, M. Jeffers, G. Meinhardt, C.E.A. Peña
Development of methodology: A.X. Zhu, G. Meinhardt, C.E.A. Peña
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.X. Zhu, Y.-K. Kang, O. Rosmorduc, T.R.J. Evans, A. Santoro, P. Ross, E. Gane, A. Vogel, M. Jeffers
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.X. Zhu, A. Santoro, P. Ross, A. Vogel, G. Meinhardt, C.E.A. Peña
Writing, review, and/or revision of the manuscript: A.X. Zhu, Y.-K. Kang, O. Rosmorduc, T.R.J. Evans, A. Santoro, P. Ross, E. Gane, A. Vogel, M. Jeffers, G. Meinhardt, C.E.A. Peña
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.X. Zhu
Study supervision: A.X. Zhu, E. Gane, G. Meinhardt, C.E.A. Peña
The study was supported by Bayer HealthCare Pharmaceuticals and Onyx Pharmaceuticals, an Amgen subsidiary.
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.
The authors thank Scott M. Wilhelm and Chetan Lathia of Bayer HealthCare Pharmaceuticals for scientific discussion and BelMed Professional Resources and C4 MedSolutions, LLC, a CHC Group company, for editorial support.
Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/).
- Received November 25, 2015.
- Revision received April 6, 2016.
- Accepted May 10, 2016.
- ©2016 American Association for Cancer Research.