Angiokines Associated with Targeted Therapy Outcomes in Patients with Non–Clear Cell Renal Cell Carcinoma

Non–clear cell renal cell carcinoma (NC-RCC) comprises a genetically and histologically diverse set of cancers that arise from the kidney, including type 1 and 2 papillary RCC, chromophobe RCC, translocation carcinoma, as well as many other rare subtypes, some of which remain histologically unclassified (1, 2). NC-RCC accounts for about 25% of all cases of RCC. However, in the metastatic setting, non–clear cell histology is less common at less than 20% of all RCC cases (2, 3), and the subtypes of NC-RCC that are most commonly found in the metastatic setting are collecting duct carcinoma, type 2 papillary and unclassified NC-RCC given their more aggressive disease course.

We and others have recently reported on randomized prospective clinical trials comparing the VEGF inhibitor sunitinib with the mTOR inhibitor everolimus in patients with metastatic NC-RCC (ASPEN and ESPN; refs. 4, 5). The ASPEN trial is the largest global study conducted in this population to date, and was highly challenging to conduct, requiring 18 centers to enroll 108 patients over 3 years, testifying to the relative rarity of this disease in the metastatic setting. In these trials, sunitinib provided superior response rates and more durable control of disease; however, outcomes were heterogeneous based on histologic subtypes. For example, patients with papillary RCC and unclassified RCC, as well as those patients with good/intermediate-risk disease had superior outcomes with sunitinib, while patients with chromophobe RCC and those with poor-risk disease had superior outcomes with everolimus. These data support the concept that these non–clear cell tumors should be regarded as distinct molecular and phenotypic entities with distinct treatment outcomes with molecularly targeted therapies (6, 7).

A major unmet need in the field of RCC therapy is for predictive biomarkers of treatment benefit. While histology (clear cell disease) is predictive of the benefits of high-dose IL2, and serum lactate dehydrogenase may be predictive of the benefits of mTOR inhibition as compared with IFN treatment in poor-risk RCC (8), there are no other clear predictors of improved overall survival (OS) benefits with specific therapies. However, biomarkers of angiogenesis and immune activation or evasion may predict for differential response and PFS with anti-VEGF and immune therapies in RCC, such as with bevacizumab and atezolizumab in clear cell RCC (9). A recent analysis of the RECORD-3 trial comparing sunitinib and everolimus identified several composite prognostic biomarkers for progression-free survival (PFS) with everolimus, but these were unable to predict OS and the analyses were largely restricted to clear cell RCC (10). Given that NC-RCC is a distinct group of cancers, dedicated analyses in this patient population are needed.

To accurately quantify key regulators of tumor angiogenesis, inflammation, and extracellular matrix remodeling, we have developed a protein multiplex array, termed the Angiome. This assay has been reviewed and approved by the NCI Biomarker Review Committee to be used as an integrated biomarker in NCTN trials for the measurement of circulating angiokines. We worked closely with the NCI in developing our validation approach, performing the necessary experiments to ensure analytic data quality. The objectives of this analysis were to identify prognostic and predictive factors of PFS and OS, and to describe changes in patient angiokine levels over time throughout the course of treatment. We sought to characterize circulating angiokines in the plasma of patients with metastatic NC-RCC as part of an international, randomized, prospective clinical trial comparing sunitinib and everolimus (ASPEN). We asked whether angiokines were differentially expressed by histologic subtypes, whether clinical efficacy outcomes differed by angiokine levels at baseline, and explored longitudinal changes in plasma angiokines during treatment response and progression to provide insights into potentially important biology that may be driving adverse outcomes.

The current analysis was a preplanned correlative analysis of a prospective, open-label nonblinded randomized FDA IND-exempt trial conducted across 17 participating global sites, including the United States, Canada, and the United Kingdom. Regulatory oversight in Canada and the United Kingdom was obtained for this trial. After meeting eligibility, randomized subjects were assigned 1:1 to either sunitinib malate or everolimus at approved doses until disease progression.

Patients were eligible if they had histologically confirmed advanced RCC with non–clear cell pathology after local site review by pathology, including unclassified subtypes. Mixtures of these non–clear cell variants were allowed provided they consisted predominantly (≥50%) of papillary, chromophobe or undifferentiated histology. Patients with minor clear cell components (<50%) were permitted provided the dominant histology and presumed primary histology was non-clear cell. Exclusion criteria for the study included active untreated central nervous system metastases, prior systemic therapy for RCC, and collecting duct or medullary histology. Full eligibility details are provided in the primary clinical article (4).

This study was registered as an International Standard Randomized Controlled Trial with ClinicalTrials.gov number NCT01108445. All patients provided informed consent under an institutional Institutional Review Board–approved consent form. This was an investigator-initiated study conducted in accordance with the U.S. Common Rule, with the Duke Cancer Institute (Durham, NC) as lead coordinating center and biorepository. A contract research organization, inVentiv Health Clinical, oversaw the collection of data and safety monitoring on behalf of Duke globally.

Prospective double-spun, platelet-poor plasma was collected longitudinally on all patients for correlative studies at baseline (pretreatment), on day 1 of cycle 3, and at progression or end of treatment due to toxicity. The majority of angiokines were measured using the SP-X imaging and analysis system from Quanterix according to the manufacturer's instructions. VEGF-C was measured using the Ella System (Protein Simple). Other assays including BMP-9, CD73, soluble HER3 and soluble TGFβ-R3, were developed in our laboratory as described previously (11–13). Patients who progressed at or before cycle 3 had only two collections. Angiokines were measured as previously described and validated (14, 15). All samples underwent one freeze-thaw cycle prior to analysis. Angiokines were analyzed in a single batch at the end of the study and the angiokine laboratory investigator (A.B. Nixon) was blinded to the clinical outcomes of the ASPEN study.

The primary endpoint of this preplanned secondary correlative angiokine analysis was the association of baseline plasma angiokines with OS, defined as the time from randomization until death from any cause, with censoring at the end of the last documented mortality assessment. The primary clinical endpoint and a key secondary outcome was the association of the plasma angiokines with radiographic PFS, defined as the time from randomization until disease progression as defined by RECIST 1.1, a new primary malignancy, or death, whichever occurred first. Other prespecified efficacy secondary endpoints included radiographic response rates per RECIST 1.1, and clinical benefit response, defined as the composite sum of partial response, complete response, and prolonged stable disease for more than 6 months. The final database was locked on January 13, 2015 for clinical outcomes.

The current prospectively defined preplanned correlative analysis was defined prior to data lock and analysis. Data were averaged among the two duplicate measurements and transformed using the log₂ function. Biomarker measurements were dichotomized at the median value for each plasma angiokine and defined as either low (≤ median) or high (> median). The PFS and OS distributions by the dichotomized plasma angiokines were estimated using the Kaplan–Meier method. The proportional hazards model was used to explore the prognostic importance of each of the 23 plasma angiokines in predicting OS and PFS. Univariate analyses were performed as well as multivariable analyses adjusting for treatment and the two stratification variables histology and Memorial Sloan Kettering Cancer Center (MSKCC; New York, NY) prognostic risk groups. In the proportional hazards models, the biomarkers were modeled using the log₂-transformed values. The association of each biomarker was summarized with HRs and 95% confidence intervals (CI). In addition, the proportional hazards model was used to test for the plasma angiokine-treatment interaction in predicting PFS and OS. FDRs are reported in all analyses testing for the prognostic significance of the 23 plasma angiokines. P values were not adjusted for multiple testing in the treatment-biomarker interaction terms. No P values will be reported for the exploratory analysis (subgroup analysis in papillary RCC). We also report the median levels and changes of each angiokine for both treatment groups at baseline, cycle 3 (18 weeks after study start), and end of treatment for the longitudinal angiokine analysis. All analyses were performed using R version 3.5.3.

From September 23, 2010 through October 28, 2013, we accrued 109 subjects across three (3) countries and 17 participating sites. One subject did not receive a study drug and withdrew and was not replaced, leaving 108 evaluable subjects who were then randomized to sunitinib (51 subjects) or everolimus (57 subjects). Of the 108 patients, 9 patients had missing baseline plasma samples, leaving 99 evaluable patients for the plasma angiokine analysis (Supplementary Fig. S1). This analysis provides updated OS and PFS data from the original ASPEN article, with a final data lock of January 13, 2015.

Baseline demographic information for the study population has been previously published and 92% of patients were included in this correlative biomarker analysis. The majority of patients (62/108, 57.4%) had metastatic papillary RCC, largely non-type 1, with only 6 patients with type 1 papillary RCC (Table 1). Patients without plasma specimens had similar characteristics to patients with plasma specimens except for race, histologic subtype and OS duration. Patients who self-identified as Black/African American comprised 22.2% of the group (n = 2) without plasma collections and 10.1% of the group (n = 10) with evaluable plasma, but this number is small. The proportion of patients with papillary RCC (61% vs. 44%) is higher in the evaluable population compared with the nonevaluable population, while fewer included patients had chromophobe or unclassified tumors. Patients with available baseline plasma data have greater OS duration (17 vs. 9.3 months) than those without available plasma data. There were a total of 94 PFS and 64 deaths and the median follow-up in 44 alive patients was 26.9 months (95% CI, 24.2–34.7).

Figure 1A depicts the log₂-transformed distributions of each angiokine at baseline. Osteopontin (OPN) has the highest overall median levels whereas PIGF was the least abundant angiokine. It is important to note that abundance in the plasma is not directly related to biological relevance and potency, but merely reflects the plasma concentration. We also present the plasma angiokines by histology (Fig. 1B). Patients with papillary RCC in ASPEN did not have higher median hepatocyte growth factor (HGF) levels at baseline compared with other subtypes but did have higher median baseline OPN levels (Fig. 1B). Patients with chromophobe RCC had higher PDGF-AA and stromal derived factor 1 (SDF-1) levels but overall angiokine levels were similar across histologies. Supplementary Table S2A describes the absolute levels of each angiokine overall and by histology and clinical risk groups. Figure 1C shows the difference in angiokine levels across the MSKCC prognostic risk groups. Poor-risk patients with NC-RCC typically had higher levels of multiple angiokines as compared with good-risk patients—particularly Angiopoeitin-2 (Ang-2), thrombospondin-2 (TSP-2), IL6, TIMP-1, CD73, ICAM-1, HGF, and VEGF levels, with clear increases from good to intermediate to poor-risk groups for many of these angiokines suggesting their importance to underlying biology or disease burden (Supplementary Table S2B). Good/intermediate-risk patients and patients with chromophobe metastatic NC-RCC had greater SDF-1 levels than other risk groups. SDF-1 was the only angiokine reduced with poor-risk disease. Unclassified tumors frequently had poor-risk angiokine profiles such as high IL6, HGF, and OPN levels.

The primary clinical endpoint for the ASPEN trial was PFS and with the exception of HER-3, SDF-1, TGFb-R3, VEGF-R1, and VEGF-R2, higher angiokine levels were associated with increased risk of progression or death (HR > 1; Table 2). In univariate analysis for PFS, the HR for OPN was 1.2 (95% CI, 1.1–1.4). However, OPN was not statistically significant in the multivariable analysis adjusting for treatment, histology, and MSKCC risk, and when adjusting for multiplicity (Table 2). In addition, we explored the association of plasma angiokines in predicting PFS among the subset of 58 patients with papillary RCC with similar results obtained, suggesting OPN has univariate prognostic importance (HR, 1.3; 95% CI, 1.1–1.5; Supplementary Table S3). The median PFS (by dichotomized angiokines) showing PFS differences between low and high levels and are presented for most of the angiokines that had the largest difference (Fig. 2).

The primary endpoint for this correlative angiokine analysis was OS, and for most angiokines, the HRs from the univariate analysis were associated with worse OS (HR > 1) except for HER-3, BMP-9, SDF-1, and TGFβ-R3 (Table 3). In multivariable analyses of OS after adjusting for treatment and stratification factors (histology and MSKCC risk), HGF, OPN, TIMP-1, TSP-2, and VCAM-1 were independently associated with OS (Table 3) with HRs of 1.5 (95% CI, 1.2–1.8), 1.3 (95% CI, 1.1–1.6), 1.7 (95% CI, 1.1–2.5), 1.8 (95% CI, 1.2–2.7), and 2.2 (95% CI, 1.2–4.1), respectively. The median OS by dichotomized angiokines showed OS differences between low and high levels and are presented for the angiokines that had the largest difference (Fig. 3). In addition, we explored the association of plasma angiokines in predicting OS in patients with papillary RCC as this was the largest histologic subgroup and found similar results as compared to the overall analysis (Supplementary Table S4).

Table 4 summarizes the treatment-biomarker interactions for PFS and OS. None of the angiokines are statistically significant even when ignoring multiple testing adjustments, demonstrating the lack of predictive discrimination for any biomarkers for the selective benefit for patients receiving either sunitinib or everolimus. The Kaplan–Meier plots for PFS and OS by the biomarker and treatment groups are presented in Supplementary Fig. S4A and S4B and outcomes are further described by treatment group in Supplementary Table S5. Finally, we did not identify a pretreatment angiokine associated with differential objective response rates by RECIST 1.1 (Supplementary Table S6).

Figure 4 illustrates the changes in plasma angiokine levels over time by treatment to illustrate the pharmacodynamic impact of VEGF tyrosine kinase inhibitor (TKI) or mTOR inhibitor therapy in patients with metastatic NC-RCC during treatment response (cycle 3) and at progression or end of treatment. We observed that multiple angiokine levels increased in patient plasma throughout the study at disease progression as compared with baseline or on-treatment cycle 3 assessments. For example, OPN levels declined with sunitinib initially, but rose at progression with both therapies. HGF levels consistently rose over time with both therapies. VEGF-R2 and VEGF-R3 decreased with both treatments, while VEGF-R1 levels rose with sunitinib treatment and fell with everolimus treatment. With sunitinib, increases at progression were observed for most angiokines, particularly CD-73, ICAM-1, IL6, OPN, PIGF, SDF-1, TIMP-1, TGF-b2, TGFb-R3, TSP-2, VCAM-1, VEGF, and VEGF-D. With everolimus, increases in Ang-2, CD-73, HER-3, HGF, IL6, OPN, PDGF-AA PDGF-BB, SDF-1, TGF-b1–3, TIMP-1, TSP-2, VEGF-D, and VEGFR-1 were particularly notable.

Here we report a comprehensive, prospective circulating angiokine analysis of plasma samples from the international randomized ASPEN trial in patients with metastatic papillary and other NC-RCC histologies treated with sunitinib or everolimus. While ASPEN established sunitinib as a more effective agent broadly in this patient population based on improved PFS and OS outcomes, significant heterogeneity by clinical risk and histologic subtypes were observed, with everolimus having similar if not improved outcomes in patients with poor-risk RCC or metastatic chromophobe histologies, and sunitinib having improved outcomes in good/intermediate risk and papillary or unclassified tumors (4). We speculated that circulating angiokines and immunokines may mediate some of this heterogeneity and thus could be suitable targets for future clinical trial investigation if proven to be prognostic or predictive, or should certain biomarkers emerge at disease progression.

In this analysis, we found that a range of angiokines are associated with a poor prognosis in patients with metastatic NC-RCC including papillary RCC. Several angiokines were consistently associated with poor-risk criteria by MSKCC and poor OS and these are nominated to be of particular relevance for further investigation, preclinical modeling, and validation for therapeutic targeting. These include OPN, TSP-2, ICAM-1, Ang-2, and HGF, while only one angiokine, SDF-1 was associated with improved outcomes. The increases of these angiokines at disease progression also suggests potential biologic relevance or associations with disease burden, and only interventional or mechanistic studies will be able to discern whether these are actionable targets in NC-RCC variants.

OPN, encoded by the SPP1 gene, is implicated in immune modulation and Th17 polarization of T cells, IL17 production, and is known to facilitate tumor-associated myeloid cell activation, invasion/metastasis, and bone mineralization biology (16). High OPN levels are associated with a poor prognosis in patients with clear cell RCC treated with anti-VEGF strategies (17), are prognostic in other solid tumors, and may be predictive for the benefits of anti-VEGF therapy in certain contexts (14, 18). The HGF–c-MET axis has been clearly associated with papillary RCC through c-MET amplifications or activating mutations, or pathway activation, and is targetable by selective or nonselective c-MET inhibitors such as cabozantinib. Cabozantinib has activity in patients with papillary RCC (19) and is presently in randomized testing in the NCI Cooperative Group PAPMET trial (NCT02761057) against sunitinib, and our data support the relevance of this axis to poor outcomes during sunitinib or everolimus therapy. TSP-2 is a secreted glycoprotein with antiangiogenic properties implicated in invasion/metastasis (20). IL6 is an acute phase reactant implicated in inflammation, TGFβ signaling, and immune modulating, angiogenesis, and tumor metastasis, and high IL6 levels have been linked previously to differential poor outcomes in patients with clear cell RCC treated with anti-VEGF targeted therapies (17).

We did not identify specific angiokines predictive of the therapeutic benefits of everolimus nor sunitinib in the ASPEN trial. For example, high OPN or HGF levels were prognostic for worse OS irrespective of treatment group, and sunitinib treatment led to improved survival over everolimus irrespective of OPN or HGF levels. In nearly all cases, sunitinib outperformed everolimus, and thus we were unable to identify an mTOR-sensitive angiokine predictor. Because of the limitations of small sample sizes of the many subgroups, we had limited power to detect treatment-biomarker interactions beyond the established differences observed by clinical risk and histology previously reported. Future studies and meta-analyses in specific groups defined by histology and MSK risk may help to identify such interactions.

Resistance to anti-VEGF TKI therapy generally developed at around 8 months, with only 17% and 9% of patient remaining progression free at 12 and 24 months, respectively, illustrating the need for better therapies at progression. Presently, no standard of care exists in this second-line setting in papillary RCC, and many potentially active agents exist. On the basis of our longitudinal analysis of plasma angiokines during sunitinib treatment and disease resistance, we found a host of angiokines change over time and which may be associated with driver biology or disease burden. On the basis of the consistent upregulation at progression irrespective of therapy, our lead angiokine candidates for further study are OPN (16, 21), CD-73, and adenosine signaling (22), HER-3, HGF, PlGF, IL6, SDF-1, TGFβ signaling, VCAM-1, TIMP-1, and TSP-2. CD73, IL6, TGFβ, and HGF axis targeting are presently under study in RCC, and evaluation in patients with advanced NC-RCC is warranted.

The strengths of our study include inclusion of the largest global therapeutic study of metastatic NC-RCC to date, the careful clinical annotation of outcomes by RECIST with long-term survival follow-up, the stratification by histology and risk categories, and the blinded nature of the angiokine analysis with the clinical outcomes. No prior work in this setting has been developed and thus, our findings warrant further validation and mechanistic studies in deciding which of the many circulating biomarkers of inflammation, angiogenesis, and immune evasion warrant targeting, and which may be simply associated with disease burden and poor survival. Further limitations beyond the small sample sizes of subsets include the lack of available genomic information from patients, the lack of tissue-based validation and mechanistic studies for each angiokine, and the heterogeneity of our patient population. Even within papillary RCC, subsets of patients defined by genetic, metabolic, and epigenetic alterations exist with differing prognosis, and future preclinical and clinical studies should further evaluate targeting and/or measuring these angiokines in order to determine their mechanistic importance to patient outcomes (23, 24).

In conclusion, we have comprehensively characterized the circulating plasma angiokines in patients with metastatic NC-RCC, and identified key potential prognostic candidates associated with poor outcomes during treatment with everolimus or sunitinib and which may emerge over time during disease progression. Several consistent candidates include OPN, TSP-2, IL6, HGF, ICAM-1, and Ang-2, and offer a range of possibilities for planning preclinical mechanistic and efficacy trials as well as prospective clinical therapeutic studies.

A.J. Armstrong reports grants and personal fees from Pfizer and grants from Novartis during the conduct of the study; grants and personal fees from Merck, BMS, Dendreon, Bayer, and AstraZeneca; grants from Genentech and Astellas, and personal fees from Clovis outside the submitted work. A.B. Nixon reports grants from Genentech, MedImmune/AstraZeneca, Medpacto, Seattle Genetics, HTG Molecular Diagnostics, Tracon Pharmaceuticals, Eureka Therapeutics, Leadiant Biosciences, OncoMed Pharmaceuticals, and Promega Corporation; personal fees from Eli Lilly, GlaxoSmithKline, Kanghong Pharma, Promega Corporation, and Leap Therapeutics outside the submitted work. A. Carmack reports master's thesis project. T. Eisen reports grants and non-financial support from Cambridge Biomedical Research Centre during the conduct of the study; grants, nonfinancial support, and other from AstraZeneca and Roche; and grants from Pfizer and Bayer outside the submitted work; and is trustee with Macmillan Cancer Support and Kidney Cancer UK. W.M. Stadler reports nonfinancial support from Pfizer during the conduct of the study; personal fees from AstraZeneca, Bayer, Esai, Merck, Pfizer, Sotio, Treadwell Therapeutics, Caremark/CVS, and Roche/Genentech outside the submitted work. R.J. Jones reports personal fees from Pfizer outside the submitted work. J.A. Garcia reports grants from Pfizer during the conduct of the study; personal fees from Merck, Pfizer, Eisai, Astellas, Janssen, Sanofi-Aventis, Bayer, and MJH outside the submitted work. U.N. Vaishampayan reports personal fees from Bayer, Pfizer, Aveo, and Helsinn; grants and personal fees from BMS, Exelixis, Merck, Alkermes, and AAA outside the submitted work. J. Picus reports grants from Inventiv Health Clinical, LLC during the conduct of the study. R.E. Hawkins reports personal fees from Pfizer, Novartis, and BMS outside the submitted work. J.D. Hainsworth reports grants from Novartis/Pfizer during the conduct of the study. C.K. Kollmannsberger reports personal fees from Pfizer, Merck, Ipsen, Eisai, EMD Serono, Astellas, Bayer, and Janssen during the conduct of the study. T.F. Logan reports grants from Novartis and Pfizer during the conduct of the study; grants from Abbott, Abraxis, Acceleron, Amgen, AstraZeneca, Biovex, Cerulean, Eisai, Eli Lilly, Hoffman LaRoche, Immatics, Merck, Roche, Synta, Threshold, Tracon, EMD Serono, Millenium, Schering Plough, Macrogenics, Peloton, Iovance, MedImmune, Dynavax, and Clinigen; and grants and personal fees from Argos, Aveo, Bristol Myers Squibb, Gelgene, GlaxoSmithKline, Genentech, Novartis, Pfizer, Prometheus, Wyeth outside the submitted work. I. Puzanov reports personal fees from Merck, Amgen, Iovance, and Nouscom outside the submitted work. L.M. Pickering reports personal fees from Pfizer and Novartis outside the submitted work. C.W. Ryan reports grants from Novartis and Pfizer during the conduct of the study; personal fees from Esai, Janssen, AstraZeneca, Bristol-Myers Squibb, Diciphera, and SynOx Therapeutics; grants from Xynomic, Nektar, GlaxoSmithKline/Novartis, Daiichi Sankyo, Merck, Argos Therapeutics, CytRx Corporation, Exelixis, Bristol-Myers Squibb, Genentech, Karyopharm Therapeutics, and Pfizer outside the submitted work. D.J. George reports grants and personal fees from Astellas, AstraZeneca, BMS, Exelixis, Janssen, Pfizer, and Sanofi; personal fees from Bayer, Constellation Pharma, EMD Serono, Flatiron, Ipsen, Merck, Michael J Hennessey, Millennium Med Pub, Modra Pharma, Myovant Sciences, NCI Genitourinary, Nektar Therapeutics, Physician Education Resource, Propella Therapeutics, UroGPO, UroToday, and Vizuri Health Sciences; and grants from Calithera and Novartis outside the submitted work. S. Halabi reports grants and other from Pfizer and Novartis during the conduct of the study. No disclosures were reported by the other authors.

A.J. Armstrong: Conceptualization, resources, data curation, supervision, funding acquisition, validation, visualization, writing–original draft, project administration, writing–review and editing. A.B. Nixon: Data curation, formal analysis, investigation, methodology, writing–review and editing. A. Carmack: Formal analysis, visualization, methodology, writing–review and editing. Q. Yang: Formal analysis, visualization, methodology, writing–review and editing. T. Eisen: Resources, supervision, investigation, writing–review and editing. W.M. Stadler: Resources, investigation, writing–review and editing. R.J. Jones: Resources, investigation, writing–review and editing. J.A. Garcia: Resources, investigation, writing–review and editing. U.N. Vaishampayan: Resources, investigation, writing–review and editing. J. Picus: Resources, investigation, writing–review and editing. R.E. Hawkins: Resources, investigation, writing–review and editing. J.D. Hainsworth: Resources, investigation, writing–review and editing. C.K. Kollmannsberger: Resources, investigation, writing–review and editing. T.F. Logan: Resources, investigation, writing–review and editing. I. Puzanov: Resources, investigation, writing–review and editing. L.M. Pickering: Resources, investigation, writing–review and editing. C.W. Ryan: Resources, investigation, writing–review and editing. A. Protheroe: Resources, investigation, writing–review and editing. D.J. George: Conceptualization, resources, funding acquisition, investigation, writing–review and editing. S. Halabi: Conceptualization, resources, data curation, software, formal analysis, supervision, validation, investigation, methodology, writing–original draft, writing–review and editing.

We wish to acknowledge the trial coordinators and their staff for their support in the conduct of this trial, and the patients and their families for their dedication to this research. We wish to thank Omar Din at Cancer Clinical Trial Centre, Weston Park Hospital, Sheffield, UK and Mary MacKenzie at London Health Sciences Center, London Ontario for their support of this study at their centers. We wish to thank the Duke Center for Human Genetics biorepository and David Layfield for their support. We thank the Cancer Research UK Glasgow Clinical Trials Unit, who were instrumental in delivering the UK sites for this study. We thank both Novartis and Pfizer for their financial support of this investigator initiated trial to both inVentiv Health Clinical and Ergomed for their monitoring and data collection support across all centers.

Investigator initiated/sponsored trial with funding from Novartis and Pfizer.

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.