Analysis of Circulating Tumor DNA and Clinical Correlates in Patients with Esophageal, Gastroesophageal Junction, and Gastric Adenocarcinoma

Globally, gastric and esophageal malignancies (herein gastroesophageal cancers) are one of the most frequent types of cancers with approximately 1.5 million patients diagnosed each year. They are associated with high mortality and 1.1 million patients are estimated to die each year (1). At the time of diagnosis, about 50%–60% of patients have regional lymph node involvement or distant metastatic disease (2). Combinations of systemic chemotherapies are generally used for the management of metastatic gastroesophageal cancers. Systemic therapies with antimetabolite- (5-fluorouracil or capecitabine) and platinum- (oxaliplatin or cisplatin) based therapies, as well as taxanes are widely used; however, median survival remains poor (9–11 months; refs. 3, 4). Thus, there is an urgent need to better understand the molecular biology of these neoplasms.

Along with the rapid advances in next-generation sequencing (NGS) technology, the molecular nosology of gastroesophageal cancers is now better understood. The Cancer Genome Atlas Research Network categorized patients with gastric cancer into four different subtypes: (i) Epstein–Barr virus-related group, which are associated with PIK3CA mutations, PD-L1/2 overexpression, and CDKN2A silencing; (ii) microsatellite unstable group, associated with hypermutation and MLH1 silencing; (iii) chromosomal unstable group associated with receptor tyrosine kinase and RAS activation; and (iv) genomically stable subtypes (5).

Clinically, several genomic and/or protein markers are now being used to guide treatment decisions for patients with gastroesophageal cancers. Examples include targeting of HER2 overexpressed/amplified gastric and gastroesophageal junction cancers with trastuzumab (anti-HER2 antibody; ref. 6) and PD-L1–overexpressed cases with pembrolizumab (anti-PD-1 antibody; refs. 7, 8), which are now FDA approved. Although some achievements were seen with the aforementioned targeted therapy approaches, clinical benefit has been modest (trastuzumab: 2.7 months of survival gain when added to chemotherapy; ref. 6; pembrolizumab response rate of ∼20%; ref. 7). Moreover, targeting other markers including FGFR amplification or MET overexpression, has not been able to demonstrate clinical benefit, at least as monotherapies, in the setting of gastroesophageal cancers (9, 10).

One of the major challenges to addressing genomic alterations in the clinic is tumor heterogeneity. Pectasides and colleagues compared genomic alterations between primary and metastatic lesions in 26 patients with gastroesophageal adenocarcinomas and revealed extensive differences (11). Kim and colleagues reported that more than half of the ERBB2-amplified gastroesophageal adenocarcinomas had additional oncogenic alterations, with each patient harboring unique molecular patterns that could explain the reason for the modest effects achieved with ERBB2-targeting agents (12). Moreover, dynamic change in underlying genomic alterations can evolve along with tumor progression and therapeutic pressure, which further confounds targeted therapy approaches (13).

NGS of plasma-derived circulating tumor DNA (ctDNA; also known as cell-free tumor DNA) has recently been evaluated in several tumor types (14–17). Detection of tumor-specific mutations that are shed into the blood from cancer cells can be performed on a small vial of blood, enabling the characterization of genomic alterations of tumors in a timely fashion. Although analysis of ctDNA among gastroesophageal cancers has been investigated (18), clinical application of the results is not well described. Here we investigated clinical characteristics and therapeutic outcomes among patients with gastroesophageal adenocarcinomas whose ctDNA was interrogated by clinical-grade NGS and compared the results to tissue NGS results in those patients in whom both tests were performed.

We investigated the genomic alteration status and clinical outcomes of 55 patients with gastroesophageal adenocarcinomas seen at the UC San Diego Moores Cancer Center (January 2014 to July 2017; San Diego, CA). Blood samples were evaluated at the clinical laboratory improvement amendments-licensed and College of American Pathologist-accredited clinical laboratories–Guardant Health, Inc. (http://www.guardanthealth.com) or Foundation Medicine (https://www.foundationmedicine.com) for ctDNA analysis. Tissue samples were sent to Foundation Medicine for tissue DNA testing using NGS. All investigations followed the guidelines of the UCSD Internal Review Board for data collection (NCT02478931) and for any experimental therapeutic trials for which consents were obtained.

Most ctDNA analyses were done through Guardant Health, Inc (N = 49/55). ctDNA was extracted from whole blood collected in 10 mL Streck tubes, and 5–30 ng of ctDNA was prepared for sequencing as described previously (19). All ctDNA was sequenced, including the somatic ctDNA and the germline ctDNA that is derived from natural leukocyte lysis. Germline alterations were filtered out and not reported. The fractional concentration or variant allele fraction for a given somatic mutation is calculated as the fraction of ctDNA harboring that mutation in a background of wild-type ctDNA fragments at the same nucleotide position. The analytic sensitivity reaches detection of 1–2 single mutant fragments from a 10 mL blood sample (0.1% limit of detection) and analytic specificity is greater than 99.9999%. Throughout the timeframe of this study, the ctDNA assay performed by Guardant Health, Inc. expanded from 54 to 68 to 70 to 73 genes (N = 49; Table 1; Supplementary Table S1). Degree of copy number alterations were reported as follows: 1+, 2.13–2.40, which is the 10th to 50th percentile; 2+, 2.41–4.00, which is >50th to 90th percentile; and 3+, greater than 4.0 copy numbers, which is >90th percentile. The ctDNA assay performed by Foundation Medicine interrogated 67 genes (N = 6; Table 1; Supplementary Table S2) which was described previously (20). Only nonsynonymous alterations were analyzed in this study. The variant allele fraction or fractional concentration for a given somatic mutation is derived from the fraction of ctDNA harboring that mutation in a background of wild-type ctDNA fragments at the same nucleotide position (19). When patients had multiple ctDNA evaluated at different time points, the results of ctDNA at the earlier time point was used for the analysis.

Thirty-one patients had both ctDNA and tissue DNA analysis. All tissue DNA analyses were performed by Foundation Medicine as described previously (21, 22). The assay for tissue DNA was designed to include all genes known to be somatically altered in human solid tumors that are validated targets for therapy and interrogated 236 genes, as well as 47 introns of 19 genes commonly rearranged in cancer (n = 2) and 315 genes, as well as introns of 28 genes commonly rearranged in cancer (N = 29).

Descriptive statistics were used to summarize the genomic alterations identified in this study. The Mann–Whitney U test was used for continuous data and Fisher exact test was used for categorical data. Concordances of ctDNA and tissue DNA were quantified by concordance percentage and Kappa value with the SE. Kappa values are interpreted by commonly used agreement categories: κ = 1 (perfect agreement) to κ = 0 (no agreement other than would be expected by chance). Overall survival (OS) was calculated from the time of ctDNA analysis to last follow-up. Survival analyses were assessed by Kaplan–Meier analysis and Cox proportional hazard model was used to estimate HRs with 95% confidence intervals (CI). For multivariate analysis, variables with P < 0.20 in univariate analysis were included in the multivariate regression model.

A total of 55 patients with gastroesophageal adenocarcinomas were evaluated for ctDNA. Tumors located in esophagus, gastroesophageal junction, and stomach were represented as follows: 20.0% (N = 11), 30.9% (N = 17), and 49.1% (n = 27), respectively (Table 1; Fig. 1). Among all patients with gastroesophageal adenocarcinomas (N = 55), the median age at the time of diagnosis was 62.6 years old (range, 23.5–91.5), and 63.6% (N = 35) were men. ctDNA tests were performed prior to surgical resection in 9 patients (16.4%). Among patients who had ctDNA analysis, tissue NGS was performed in 31 patients (56.4%; Table 1).

Among these 55 patients evaluated for ctDNA, 76.4% (N = 42) had at least one genomic alteration [includes characterized alterations and variants of unknown significance (VUS)] and 69.1% (N = 38) had at least one characterized alteration. The median number of alterations per patient was 2 (range: 0–15); the median number of characterized alterations per patient was 1 (range: 0–7). The most common characterized alteration was TP53 (50.9%, N = 28) followed by PIK3CA (16.4%, N = 9), ERBB2 (14.5%, N = 8), and KRAS (14.5%, N = 8; Fig. 2; Supplementary Table S3). Thirteen patients (23.6%) had ≥5% of allele frequency and the median percentage of highest allele frequency was 0.8% (range: 0%–50.7%; Table 1).

Among 55 cases with gastroesophageal cancers, a total of 159 alterations were identified (including VUSs). Among those alterations, 68.6% (109/159) were characterized alterations, including substitutions [51.6% (82/159)] and amplifications [17.0% (27/159)]. Among the 109 characterized alterations, 89.9% (98/109) were potentially targetable with FDA-approved agents as on- or off-label use, and an additional 8.2% (8/109) were theoretically targetable with therapies that are currently in clinical trials. Altogether, among all characterized alterations, 97.2% (106/109) were theoretically actionable either with agents that are approved by the FDA (including off-label) or with agents that are in clinical trials. Among all 55 patients with gastroesophageal cancers, 69.1% (N = 38; all patients with ≥1 characterized alteration) had ≥1 theoretically actionable alterations by an FDA-approved agent (on- or off-label). Patients who did not have targetable alterations were those who only had VUSs (7.3%, N = 4) or patients without detectable alteration (23.6%, N = 13).

Among the 38 patients who had ≥1 characterized alteration, no 2 patients had identical molecular portfolios (e.g., TP53 R273H and TP53 R175H considered as molecularly distinct), while 8 patients had identical genomic portfolios (e.g., TP53 R273H and TP53 R175H considered as genomically identical; ID#15, 27, 29, 31, 40, 48, 51; Supplementary Table S4).

Of 31 patients who had both ctDNA and tissue DNA analyses, the median time interval between tissue and ctDNA collection was 1.3 months (range 0–47.1 months). The overall concordance rate was 61.3% for TP53, 83.9% for ERBB2, 74.2% for PIK3CA, and 87.1% for KRAS alterations (statistically significant concordance only seen with ERBB2 alterations (P = 0.048; Supplementary Table S5A). When overall concordance was compared between ctDNA and tissue NGS, 1 patient [3.2% (1/31)] had complete concordance (TP53 R342*, PTEN H75fs*2, and KRAS amplification were found in both ctDNA and tissue; Supplementary Table S4; ID: 43). Partial concordance (e.g., ERBB2 amplification found in both ctDNA and tissue NGS, but TP53 alteration only found in tissue NGS) was seen in 54.8% (17/31) of cases. There was no concordance seen in 41.9% (13/31) of cases (including 10 patients without detectable characterized alterations from ctDNA; Supplementary Table S4). Similar concordance was observed when the interval between blood draw and tissue biopsy was ≤6 months (N = 19 patients) versus >6 months (N = 12 patients), but the small number of individuals in each group precludes definitive conclusions (Supplementary Table S5B). When concordance was evaluated depending on the site of biopsies [primary site (N = 23) vs. metastatic site (N = 8)], higher concordance was observed between ctDNA and metastatic sites for TP53 and ERBB2 alterations (concordance rate between ctDNA and primary site vs. ctDNA and metastatic site was as follows: TP53: 52.2% vs. 87.5%, ERBB2: 78.3% vs. 100%); however, these differences were not statistically significant (Supplementary Table S5C). Because treatment can affect ctDNA levels, we ran the concordance analysis of the 20 patients who had not received systemic therapy or were off treatment for at least 4 weeks. We found similar concordance rates to those in the analysis of all patients (Supplementary Table S5D).

Patients with locally advanced, metastatic or recurrent gastroesophageal cancers (N = 46) were included in the survival analysis. Characteristics with at least 5 patients affected were included. When OS was evaluated from the time of ctDNA analysis to last follow up, in the univariate analysis, gender, PIK3CA, and KRAS alterations, as well as lines of systemic therapies (≥2 lines) before the ctDNA analysis were not statistically associated with the OS outcome. On the other hand, patients with TP53 alterations, highest %ctDNA ≥ 1.65% and cumulative total %ctDNA ≥ 2.3% had a trend towards worse OS (HR: 2.28; 95% CI: 0.72–7.29, P = 0.154; HR: 2.49, 95% CI: 0.78–8.00, P = 0.115; and HR: 2.50, 95% CI: 0.75–8.28, P = 0.122, respectively; cutoff of %ctDNA ≥ 1.65% and ≥2.3% was chosen because it was the median %ctDNA). Patients whose ctDNA was obtained ≥6.8 months from the time of metastatic/recurrent disease had a trend toward better OS (HR, 0.48; 95% CI, 0.15–1.46; P = 0.183; cutoff of 6.8 months was chosen because it was the median). ERBB2 alterations were significantly associated with worse OS (HR, 8.02; 95% CI, 2.41–26.69; P < 0.001) in univariate analysis (Table 2; Fig. 3). After the multivariate analysis, ERBB2 alterations continued to be an independent factor associated with poor OS (HR, 14.06; 95% CI, 2.44–81.03; P = 0.003). Patients whose ctDNA were obtained ≥6.8 months from the time of metastatic/recurrent disease was associated with better OS (HR, 0.18; 95% CI, 0.04–0.82; P = 0.026; Table 2).

Among 9 patients whose ctDNA was examined prior to resection, 3 (33.3%) had detectable characterized alterations, while 76.1% (35/46) of patients with metastatic, recurrent, or locally advanced diseases had detectable alterations (P = 0.019). Among patients who underwent surgery (N = 9), the median number of alteration (0 vs. 3, P = 0.009) and median value of highest allele frequency of %ctDNA (0 vs. 1.7, P = 0.001) were significantly lower than that of patients with metastatic, recurrent, or locally advanced diseases (Supplementary Table S6).

Among 55 patients evaluated for ctDNA, 24 patients received systemic therapy after the ctDNA analysis. Among those 24 patients, 70.8% (17/24) had at least one actionable genomic alteration (the remaining 7 patients had no characterized alterations). However only 1 of 24 patients (4.2%) was treated with matched targeted therapy approach (Fig. 1; Supplementary Table S7).

The summary of the patient who received matched targeted therapy is given in Supplementary Table S7.

This is a 68-year-old woman with gastroesophageal junction adenocarcinoma who was found to have multiple liver metastases and lymphadenopathy at diagnosis. Both ctDNA and tissue NGS revealed EGFR amplification. The patient was started on dual anti-EGFR therapy (ref. 23; cetuximab and erlotinib) and achieved partial response [67% decrease by RECIST 1.1, progression-free survival (PFS) of 18 months; ref. 17]. Along with radiographic response (Fig. 4A), serial ctDNA analysis 4 and 12 months after the first ctDNA analysis showed normalization of EGFR copy number in plasma (pretreatment EGFR copy number = 143, down to reference range; Fig. 4B).

Patients with advanced gastroesophageal cancers have a poor prognosis with a median survival of 9–11 months (3, 4). Although molecular characteristics of gastroesophageal cancers have been investigated (5), to date, understanding clinical correlates of genomic data and capitalizing on this information in the patient care setting has been limited (6, 7, 9, 10). Herein, we report the biologic and clinical correlates of genomic alterations among 55 patients with mostly advanced-stage gastroesophageal cancers using blood-derived ctDNA interrogated by clinical-grade NGS.

Altogether, 76.4% (42/55) of patients had at least one nonsynonymous alteration detected from ctDNA and 69.1% (38/55) had at least one characterized alteration (VUSs excluded). The most frequent characterized alterations were in the TP53 gene (50.9%, 28/55) followed by the PIK3CA (16.4%, 9/55), ERBB2 (14.5%, 8/55), and KRAS genes (14.5%, 8/55; Fig. 2; Supplementary Table S3), which is consistent with a previous report (18). Among frequently altered genes, concordance rate between tissue DNA and ctDNA varied from 61%–87% depending on the alterations (Supplementary Table S5A). This observation is consistent with a previous report by Pectasides and colleagues, which showed discordance between tumor DNA and ctDNA in gastroesophageal cancers (11). This may not be surprising because genomic alterations from the areas of cancer that were not biopsied (e.g., distant metastases) or intratumoral heterogeneity are likely being uncovered by ctDNA analysis. Indeed, Pectasides and colleagues noted discordance in findings between primary tumors and metastases in gastroesophageal cancers (11). In our study, comparison between the blood draw and tissue biopsy ≤ 6 months apart (N = 19) versus >6 months apart (N = 12) did not reveal a difference in concordance rate among frequently altered genes (Supplementary Table S5B). These observations differ from previous reports in nongastric cancers that showed high concordance between tissue DNA and blood-derived ctDNA when the interval between two tests was short (24, 25). The relatively small number of patients in our study and in previous work (11) may have confounded the results and/or the concordance in gastric cancers may be lower than in other types of malignancies. Technical differences between the tissue and blood ctDNA assays cannot be ruled out.

Interestingly, patients with early-stage/surgically resectable disease had a significantly lower number of alterations detected from ctDNA, as well as lower variant allele fraction of ctDNA when compared with patients with metastatic/advanced disease (Supplementary Table S6). This observation is consistent with previous reports that showed that the level of ctDNA was associated with underlying tumor burden and can demonstrate dynamic changes along with the therapeutic course (24, 26). Moreover, among patients with early-stage colon cancer, postsurgical detection of ctDNA was strongly associated with tumor recurrence suggesting that ctDNA may be used as a surrogate for minimal residual disease after surgery (27). Further investigation in this respect is necessary in patients with gastroesophageal cancers. In regard to outcome, prior reports have shown an association between high %ctDNA and survival (24, 25). In this study, we found that high %ctDNA (evaluated via the ctDNA alteration presenting the highest variant allele fraction, as well as by calculating the cumulative percentages of all ctDNAs in each patient) showed a trend toward correlation with OS in univariate, but not in multivariate, analysis (Table 2).

Notably, among several alterations that were identified, presence of an ERBB2 alteration was significantly associated with poor OS (HR, 14.06; 95% CI, 2.44–81.03; P = 0.003 by multivariate analysis; ERBB2 altered vs. not; Table 2; Fig. 3). Trastuzumab is approved for gastroesophageal cancers with HER2 overexpression or amplification; addition of trastuzumab to chemotherapy demonstrated clinical benefit (trastuzumab plus chemotherapy vs. chemotherapy alone: median OS: 13.8 vs. 11.1 months; ToGA trial; P = 0.0046; ref. 6). The survival in our patients with ERBB2-altered malignancies was considerably shorter than in the ToGA trial, perhaps because the ToGA trial included only patients who had not had prior therapy in the metastatic setting. Also none of our patients were treated with ErbB2-targeting agents after the ctDNA analysis. In addition, detection of ERBB2 overexpression or amplification can differ depending on the various methodologies used (Supplementary Table S8), which can confound the clinical outcome. Previous reports indicate that gastroesophageal cancers with ERBB2 alterations commonly harbor genomic coalterations that can potentially drive therapeutic resistance to anti-HER2–directed therapies (12). In this study, we have also observed that all patients with ERBB2 alterations (N = 8) harbored at least one coalteration, including anomalies in FGFR2, RAF1, PIK3CA, and KRAS that can be associated with resistance to anti-HER2 regimens (Supplementary Table S4; patient ID: 1, 4, 17, 30, 32, 33, 47, and 53). These observations suggest that cotargeting of resistance signals may enhance the efficacy of Erbb2-antagonist agents. Furthermore, among patients whose tumors harbored ≥1 characterized alteration, no 2 patients had identical molecular portfolios (Supplementary Table S4). Studies with customized combination regimens have been initiated (28).

Although there is growing evidence that matched targeted therapy is potentially promising as a therapeutic approach (29–32), one of the realistic challenges to this strategy has been the low rate of patients receiving matched treatments (∼5%–20% in many studies; refs. 29, 33–35). Low target–drug matching rate is often due to a number of reasons, including lack of drug accessibility, deterioration of patient's condition at the time of matching, or lack of actionable targets (36). In this study, we have also observed that molecularly matched therapies based upon profiling with ctDNA was underutilized in the clinic, as only 1.8% (1/55) of patients were treated in this manner (Fig. 1; Supplementary Table S7) despite the fact 69.1% (38/55) of patients had theoretically tractable targets (Supplementary Table S4). Although there was only 1 patient who received matched cognate agents, this individual showed clinical benefit, with a partial response (67% regression) and PFS of 18 months in a patient with EGFR amplification who was given dual anti-EGFR therapies (cetuximab and erlotinib; Fig. 4; Supplementary Table S7). Further clinical investigation is warranted.

There were several limitations to this study. First, the study was performed retrospectively with a small sample size at a single institution. Second, not all patients had tissue NGS as a comparator. Third, it is conceivable that using different ctDNA assays may provide different results, even if the assays are clinical grade. Fourth, the number of genes evaluated changed over time (from 54 to 68 to 70 to 73 gene panels for the Guardant assay), which makes direct comparison challenging; however, it should be noted that only the genes found in both the tissue and the ctDNA panel were compared for concordance. Fifth, some of the concordance was driven by the wild-type cases (where both ctDNA and tissue would be negative for an alteration); hence, the P values might be misleading for positive concordance. Sixth, different systemic therapies at the time of ctDNA analysis can affect the level of variant allele fractions, which may have confounded the survival analysis. Finally, results of germline alterations were filtered out and not reported in this study. To comprehensively understand the biology of cancer in each patient, evaluation of both germline, as well as somatic alterations may be required. Thus, the current findings require further validation with larger numbers of patients in the setting of prospective studies using newer methodologies, as they become available, because the capability to detect alterations is improving with time.

In conclusion, we have evaluated 55 patients with gastroesophageal cancers who had blood-derived ctDNA analysis by clinical-grade NGS. The most frequent alterations were in TP53 (50.9%) followed by PIK3CA (16.4%), ERBB2 (14.5%), and KRAS genes (14.5%). At least one alteration was identified in 76.4% of patients. Concordance between ctDNA and tissue DNA among commonly altered genes ranged from 61.3% to 87.1%. Discordant results may be due to the dynamic changes in ctDNA after treatment or tumor progression, tumor heterogeneity, or from disparities in sensitivity between tissue and ctDNA analysis. Technical issues cannot be ruled out. ERBB2 alterations were associated with significantly worse OS (median OS = 1.7 vs. 20.2 months). Importantly, among patients who had at least one characterized alteration, no 2 patients had identical molecular portfolios, suggesting that customized therapy may be necessary. Although the number of patient who received therapy that matched ctDNA analysis was small, the patient showed benefit. Further investigations of the clinical utility of blood-derived ctDNA among patients with gastroesophageal cancers are needed.

J.K. Sicklick is a consultant/advisory board member for CarsGen, Grand Rounds, Loxo Oncology, and reports receiving commercial research support from Foundation Medicine and Novartis Pharmaceuticals. R. Kurzrock has ownership interests (including patents) at IDbyDNA anD CureMatch Inc., is a consultant/advisory board member for Actuate Therapeutics, LOXO, NeoMed, Roche, and X-Biotech, and reports receiving commercial research grants from Foundation Medicine, Genentech, Guardant Health, Incyte, Konica, Merck Serono, Pfizer, and Sequenom. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S. Kato, R. Okamura, P.T. Fanta, R. Kurzrock

Development of methodology: S. Kato, R. Okamura, R. Kurzrock

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Kato, R. Okamura, J.M. Baumgartner, H. Patel, K. Kelly, P.T. Fanta

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Kato, R. Okamura, H. Patel, P.T. Fanta, S.M. Lippman, R. Kurzrock

Writing, review, and/or revision of the manuscript: S. Kato, R. Okamura, J.M. Baumgartner, H. Patel, L. Leichman, K. Kelly, J.K. Sicklick, S.M. Lippman, R. Kurzrock

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Kato, R. Okamura, S.M. Lippman

Study supervision: R. Okamura, S.M. Lippman, R. Kurzrock

Funded in part by the Joan and Irwin Jacobs Fund (to R. Kurzrock) and the Jon Schneider Memorial Cancer Research Fund (to J.K. Sicklick, P.T. Fanta), as well as by National Cancer Institute grants P30 CA023100 (to R. Kurzrock), K08 CA168999, and R21 CA192072 (all to J.K. Sicklick).

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.