Summary: Khan and colleagues demonstrate how serial blood-based liquid biopsies integrated with imaging and mathematical modeling can accurately “forecast” the time to treatment failure in patients with metastatic colorectal cancer treated with EGFR blockade, by early detection of molecular alterations associated with drug resistance in cell-free DNA. Cancer Discov; 8(10); 1213–5. ©2018 AACR.
See related article by Khan et al., p. 1270.
Almost 70 years after the presence of fragmented cell-free DNA (cfDNA) in the blood was discovered (1), the ability to detect and analyze cfDNA released by tumor cells has advanced rapidly in recent years. Often referred to as “liquid biopsy,” analysis of tumor-derived cfDNA can identify tumor-specific alterations—including somatic point mutations, loss of heterozygosity, gene fusions, gene copy-number variations, and DNA methylation changes—all from a routine peripheral blood specimen (2). By providing a noninvasive means to determine the molecular profile of a cancer directly from the circulation, liquid biopsy offers an attractive alternative when tumor biopsies are difficult or unsafe to perform, or when the tissue obtained is insufficient for analysis.
However, the potential clinical applications of cfDNA analysis are not limited to clinical scenarios where tumor tissue is unavailable. Because it requires only a routine peripheral blood draw, cfDNA analysis can be performed serially throughout treatment, providing a means to study clonal evolution and to identify mechanisms of acquired resistance without the need for repeat invasive tumor biopsies. Liquid biopsy has also offered researchers insight into the complexity of tumor heterogeneity in the setting of acquired resistance. Because cfDNA harbors fragments of DNA shed by tumor cells throughout the body, it is possible to detect distinct resistance mechanisms emerging concurrently in different metastatic lesions that would not be feasible or safe to assess through multiple invasive tumor biopsies (2–4).
Our understanding of resistance to anti-EGFR antibodies in metastatic colorectal cancer (mCRC) exemplifies the power of cfDNA analysis in this setting. cfDNA analysis facilitated the discovery of the first acquired resistance mechanisms to anti-EGFR therapy in mCRC—mutations in KRAS and NRAS, which bypass the dependency on EGFR signaling—and also helped identify other resistance mechanisms, including genomic alterations in other downstream effectors of the EGFR pathway such as BRAF and MAP2K1 and in parallel pathways such as MET and ERBB2, as well as EGFR itself (5, 6). Moreover, liquid biopsy studies have demonstrated how many of these resistance mechanisms may coexist in distinct tumor subclones in an individual patient, leading to mixed treatment responses (6, 7). Still other potential clinical applications of cfDNA include the possibility to noninvasively monitor patients undergoing treatment to obtain real-time assessments of response or resistance.
In this issue of Cancer Discovery, Khan and colleagues present data from a prospective phase II clinical trial (PROSPECT-C) of single-agent anti-EGFR antibody (cetuximab or panitumumab) in patients with RAS wild-type mCRC (8). This trial integrated genomic profiling of serial cfDNA and matched sequential tissue biopsies. Their work highlights several important applications of liquid biopsy that have the potential to guide clinical decision-making.
First, the authors compared the tumor molecular profiles determined from cfDNA and tumor biopsies obtained pretreatment. Interestingly, the authors found that ∼50% of patients harbored preexisting alterations in the RAS pathway detectable in the pretreatment cfDNA, and that these patients demonstrated significantly inferior progression-free survival and overall survival as well as a trend toward lack of response compared with patients without baseline RAS pathway alterations in cfDNA. Importantly, they found that even though enrollment in this trial was restricted to those patients with wild-type KRAS and NRAS based on clinical testing of tumor tissue, many of these patients actually harbored low-frequency KRAS or NRAS mutations detectable in cfDNA. The remainder of the patients with baseline RAS pathway alterations in cfDNA harbored alterations in other RAS pathway components, such as mutations in BRAF or amplification of receptor tyrosine kinases, such as ERBB2 (HER2). The discrepancies between baseline cfDNA and tumor tissue could be due to several factors: For alterations in RAS pathway components other than KRAS or NRAS, many of the differences could be because these alterations were not covered by the clinical assay used to determine study eligibility. However, for patients with KRAS or NRAS mutations detected in cfDNA despite being originally classified as KRAS and NRAS wild-type by clinical testing of tumor tissue, the differences could be due to the presence of these RAS mutations in a rare subclonal population of tumor cells, below the limit of detection of the clinical assay, or due to sampling bias and spatial heterogeneity of the tumor. Nevertheless, these results illustrate how baseline cfDNA analysis has the potential to identify patients classified as KRAS or NRAS wild-type by standard clinical testing who are unlikely to benefit from administration of anti-EGFR antibodies.
Moving beyond analysis of baseline cfDNA, the authors also investigated how serial cfDNA analysis during treatment could provide insight into clonal evolution and the emergence of therapeutic resistance (Fig. 1). Specifically, the authors integrated serial cfDNA analysis with imaging and mathematical modeling to determine whether cfDNA analysis could “forecast” the time to treatment failure. Consistent with prior studies, the authors found that the majority of patients (86% in this series) harbored RAS pathway alterations at the time of disease progression. However, in the majority of cases, these resistance mechanisms could be detected in the blood several months before radiographic progression was observed (5, 7, 9, 10). Based on these observations, Khan and colleagues developed a mathematical model for the emergence of resistance based on the assumption that the tumor is composed of a population of treatment-sensitive cells and a population of treatment-resistant cells, which have determined growth and death rates. Using this model, the authors were able to leverage serial cfDNA assessments to predict with reasonable accuracy the time until radiographic relapse was observed by RECIST v1.1 measurements. Time to treatment failure was influenced by the initial frequency of the resistant population prior to treatment initiation, with responders harboring a rare or a slow-growing resistant subclone, whereas most nonresponders harbored an already prevalent resistant population at baseline. In some instances, this model could predict which resistant subclone would emerge first to drive relapse. The authors also found that the predictive power of the model could be enhanced by more frequent cfDNA sampling, with monthly or even biweekly blood draws providing the greatest accuracy.
Although mathematical modeling of serial cfDNA evaluations to predict time to progression will require prospective validation in future studies with larger numbers of patients, the potential ability to predict the time to relapse through serial cfDNA analysis would have important implications for clinical practice. The ability to detect impending treatment failure sooner and to detect the specific resistance alterations driving resistance in the blood would enhance our ability to more optimally tailor therapeutic decisions for individual patients.
As technologies continue to improve and the cost of cfDNA analysis declines, routine integration of liquid biopsies into clinical practice in combination with radiologic imaging as a means of treatment monitoring may become feasible. Serial blood-based cfDNA analysis could be used much like standard serum tumor markers as a routine means of monitoring patient response during treatment, while also providing real-time insight into the genomic mechanisms driving resistance, ultimately guiding clinical decisions with greater precision.
Disclosure of Potential Conflicts of Interest
R.B. Corcoran reports receiving commercial research grants from AstraZeneca and Sanofi and is a consultant/advisory board member for Amgen, Avidity, Spectrum Pharmaceuticals, Symphogen, Taiho, Warp Drive Bio, Astex, BMS, Genentech, FOG Pharma, LOXO Oncology, N-of-One, Roche, Roivant, Spectrum Pharmaceuticals, Symphogen, Taiho, and Warp Drive Bio. No potential conflicts of interest were disclosed by the other author.
G. Siravegna is supported by a 3-year AIRC fellowship and “Roche per la ricerca” grant 2017. R.B. Corcoran is supported by NIH/NCI Gastrointestinal Cancer SPOREP50 CA127003, R01CA208437, U54CA224068, and a Stand Up To Cancer (SU2C) Colorectal Dream Team Translational Research Grant (grant number SU2C-AACR-DT22-17). Research grants are administered by the American Association for Cancer Research, the scientific partner of SU2C.