Linking Genotype to Phenotype: Bench to Bedside

In this issue of Clinical Cancer Research, Nixon and colleagues (1, 2) used biospecimens collected within cooperative group trials to look for features of a tumor or host biologic response that indicate the level of tumor aggressiveness or risk to the patient's health (i.e., prognostic biomarkers) and uncover elements that identify which patients are most likely to benefit from a therapy (i.e., predictive biomarkers). In both studies, baseline circulating levels of proteins involved in angiogenesis, inflammation, and immune response were measured and correlated with progression-free survival (PFS) and overall survival (OS) following treatment with the antiangiogenic agent, bevacizumab.

Human populations are remarkably heterogeneous, and the cancers that plague us are just as diverse. The vast complexity of cancer phenotypes, genotypes, and environmental influences makes finding the right treatment for the right patient an incredible challenge. Scientists have uncovered many features of cancer that illustrate how interplay between tumor, environment, and host allows a malignancy to grow and spread (Table 1). Experimental techniques to characterize cancer generate vast amounts of data that are interrogated using powerful new computational tools. We can examine the genome of a single cell, characterize the microbiome of an individual, analyze blood specimens to identify nucleic acids, contents of exosomes, and obtain gene expression data from tissue and blood. Clinical genotyping of tumors has become almost routine, and germline data are increasingly available. Researchers have achieved notable successes when, from among all the inevitable noise, tumor and host response signals that guide effective treatment were identified.

To uncover features that contribute to our understanding of the complex biology of cancer and also can be exploited to achieve clinical benefit, biological assays must be associated with carefully characterized clinical variables. However, too much noise in routine clinical data can lead to misleading results, and clinical data are notoriously difficult to standardize. Because of this, the optimal environment for assessing clinical behavior is a prospective clinical trial, where clinical variables are rigorously defined.

CALGB 90206 was a phase III trial comparing bevacizumab plus IFNα versus IFNα alone as first-line treatment for patients with advanced renal cell carcinoma (RCC; ref. 1). The study examined a panel of plasma proteins representing factors involved in angiogenesis, inflammation, and immune response and twelve biomarkers were identified as prognostic of OS including IL6, Ang-2, OPN, hepatocyte growth factor (HGF), TSP-2, PIGF, IL8, CRP, IGFBP-1, IGFBP-2, VEGFR1, and VEGF. This study also found that IL6 and HGF were potential predictive markers of bevacizumab benefit, with improved OS observed in a subpopulation of patients with higher than median IL6 and lower than median HGF levels. In a second study, CALGB/SWOG 80405, patients with previously untreated KRAS wild-type (WT) metastatic colorectal cancer were randomized to receive either bevacizumab or the anti-EGFR antibody, cetuximab, in combination with chemotherapy (FOLFOX6 or FOLFIRI, physician choice). Plasma markers prognostic of both OS and PFS included IL6, Ang-2, OPN, HGF, TSP-2, CD73, ICAM-1, TIMP-1, VCAM-1, and VEGFR-3. High levels of the VEGF family member, PlGF, predicted lack of PFS benefit from bevacizumab, and low VEGF-D levels predicted PFS benefit from bevacizumab for a subset of patients receiving FOLFOX chemotherapy (2). These two studies examined different histologies and comparator treatments, and used different analytic approaches, yet significant overlap in the markers identified as associated with clinical benefit from bevacizumab was observed.

Taken alone, the contribution from these two studies is modest, and neither changes clinical practice. However, these two studies do not exist in isolation, but are part of much larger datasets. Resources from these trials have contributed results that, taken together, add much to our understanding of the highly complex nature of metastatic renal cell and colorectal cancer. CALGB/SWOG 80405 randomized 1,137 patients with KRAS WT metastatic colorectal cancer between September 2005 and March 2012, and primary study results were published in 2017, showing no difference in OS with treatment using cetuximab versus bevacizumab added to the mFOLFOX6 or FOLFIRI chemotherapeutic regimens as first line treatments (4). This trial also collected health-related quality of life and pharmacoeconomic data, diet and lifestyle assessments, and patient-reported outcomes [PRO–Common Terminology Criteria for Adverse Events (CTCAE)]. Available biospecimens include tumors and serial blood and urine samples. As of this time, a total of 20 additional manuscripts reported correlative science results from CALGB/SWOG 80405. These include analysis of germline DNA revealing variants associated with survival (5), and with bevacizumab-induced hypertension (6). Tumor genotyping showed that microsatellite instability, tumor mutational burden, consensus molecular subtyping, and loss of neurofibromin-1 (NF1; ref. 7) all characterized disease subsets with distinct behavior (8, 9). Diet and metabolism were identified as important features of clinical outcomes through studies examining diet quality, body mass index (BMI) and weight loss, diabetes, physical activity and circulating levels of insulin-like growth factor (IGF)-binding proteins, adiponectin and vitamin D. Age was investigated, characterizing differences in outcomes for younger patients and older patients with attention to comorbidities associated with aging (10, 11). Finally, a study examining the influence of race found no difference in survival or response to therapy between Black patients and White patients in an equal treatment setting (12). CALGB 90206 randomized a total of 732 patients with metastatic RCC between October 2003 and July 2005. Study results published in 2008 showed that first line treatment with bevacizumab plus IFNα produced superior PFS and objective response rate compared with IFNα monotherapy (13). Final study results published in 2010 showed that OS favored the bevacizumab plus IFNα arm but did not meet the predefined criteria for significance, and that hypertension may be a biomarker of outcome with bevacizumab plus IFNα (14). Tissue from pretreatment nephrectomy specimens was used to determine expression levels of 424 candidate genes, identifying a prognostic signature containing eight genes (15). Of note, one of these genes, HGF, encodes a circulating protein associated with poor prognosis in both CALGB/SWOG 80405 and CALGB 90206, and predictive of improved response to bevacizumab in CALGB/SWOG 80405 (1, 2).

There are challenges to overcome to make the best use of clinical trials resources to answer new research questions. Clinical trials routinely take years to complete, and the pace of technology is such that the best approaches to specimen collection and/or analysis are often not known when the study is launched. Today, for example, we would like to use banked blood to examine circulating nucleic acids, but pre-2015 blood collection methods were not optimal for this use. In addition, the older the trial, the less likely it will be that the regimens tested represent standard of care. This means that the results are less relevant for use as predictive markers, although they still contribute to our understanding of the molecular pathways influenced by the treatments. Finally, data sharing can be difficult. Until recently, clinical trials datasets existed in multiple formats, and their use by outside researchers depended upon collaboration with statisticians and data managers from the original trial, who often don't have sufficient time to devote to new requests. Secondary use of large complex data sets from -omic analyses presents an even greater standardization challenge.

Because the two trials highlighted here were funded by the NCI, the clinical datasets and biospecimens created by them are a shared resource, to be used as widely as possible to advance patient benefit. To facilitate this goal, the NCI now maintains a centralized, controlled-access database, the NCTN/NCORP Data Archive (https://nctn-data-archive.nci.nih.gov/) that includes patient-level deidentified datasets representing variables used in publications of primary outcomes from all phase III trials as of January 1, 2015. The Archive also includes selected nonprimary publication datasets. Each trial submission includes a clinical dataset, a data dictionary, and key metadata fields. To make these datasets available in a timely manner, data providers must submit data within 6 months of publication. Whenever possible, NCTN/NCORP datasets are also made available through an open-access data sharing platform, Project DataSphere (https://data.projectdatasphere.org/projectdatasphere/html/home). CALGB/SWOG 80405, published in 2017, is included in both the NCTN/NCORP Data Archive and Project DataSphere. Biospecimens remaining from NCI-funded trials are accessible via the NCI Navigator https://navigator.ctsu.org/. Finally, NCI's Genomic Data Commons (https://gdc.cancer.gov/ provides a platform for storage and sharing of an ever-increasing number of large datasets resulting from clinical trials -omics studies.

We can't change the fact that human populations are difficult to study, and that clinical trials take years to produce results. We can, however, try to make better use of both new trials and current clinical trial resources. To do this, we need to ensure that clinical trials researchers are integrated with the basic science community in ways that allow new trials to be designed with the goal of producing the specimens and clinical data needed for therapeutic discovery and development. We also need to use what we already have, as fully as possible, recognizing that we owe this to our patients, especially to those who participated on clinical trials.

S. George reports personal fees and other support from Blueprint Medicines and Deciphera Pharmaceuticals; other support from Daiichi Sankyo, Tracon, Merck, Theseus, Eisai, Springworks, BioAtla, Abbott Laboratories, and Wolter Kluwers; and personal fees from Immunicum, CStone, Kayothera, and WCG outside the submitted work; in addition, S. George is the Vice-Chair of the Alliance of Clinical Trials in Oncology and Vice-President of the Alliance Foundation. M.M. Bertagnolli reports research funding for the Alliance for Clinical Trials in Oncology from AbbVie, Agenus, Astellas, AstraZeneca, Baxalta, Bayer HealthCare, Breast Cancer Research Foundation, Bristol-Myers Squibb, Celgene, Complion, Czarnowski, Derse, Inc., Eisai, Exelixis, Flame Biosciences, Genentech, GHI, Gilead Sciences, GSK Total, Incyte Corporation, Janssen, Jazz Pharma, Leap Therapeutics, Leidos, Lexicon Pharma, Lilly, Maltrex, Merck, Millennium, MITRE Corporation, Novartis, Pfizer, Pharmacyclics, Robert Wood Johnson FDN, Roche/Genentech, Sagerock Advisors, Sanofi, STO, Taiho Oncology, Takeda, Tesaro, and Teva; in addition, M.M. Bertagnolli is the Group Chair, Alliance for Clinical Trials in Oncology and serves on the boards of American Cancer Society, Prevent Cancer Foundation, Natera, Inc., Leap Therapeutics, and Oncoclinicas do Brasil Servicos Medico, and is a consultant for MITRE Corp.