Comprehensive molecular characterization of gastric cancer patients from phase II second-line ramucirumab plus paclitaxel therapy trial

Abstract

Background: Gastric cancer (GC) is a heterogenous disease consisted of several subtypes with distinct molecular traits. The clinical implication of molecular classification has been limited especially in association with treatment efficacy of ramucirumab or various targeted agents.

Methods: We conducted a prospective non-randomized phase II single-arm trial of ramucirumab plus paclitaxel as second-line chemotherapy in 62 patients with metastatic GC who failed to respond to first-line fluoropyrimidine plus platinum treatment. For integrative molecular characterization, all patients underwent pre-ramucirumab treatment tissue biopsy for whole-exome/whole-transcriptome sequencing to categorize patients based on molecular subtypes. We also systematically performed integrative analysis, combining genomic, transcriptomic, and clinical features, to identify potential molecular predictors of sensitivity and resistance to ramucirumab treatment.

Results: Sixty-two patients were enrolled in this study between May 2016 and October 2017. Survival follow-up in all patients was completed as of the date of cut-off on January 2, 2019. No patient attained complete response (CR), while 22 patients achieved confirmed partial response (PR), resulting in a response rate (RR) of 35.5% (95% CI, 23.6-47.4). According to TCGA molecular classification, there were 30 GS, 18 CIN, 3 EBV, and 0 MSI tumors. The RR was 33% in GS (10/30), 33% in CIN (6/18), and 100% in EBV-positive GC patients with significant statistical difference for EBV(+) against EBV(-) tumors (P = 0.016; chi-squared test). Moreover, responsive patients were marked by activation of angiogenesis, VEGF, and TCR-associated pathways, while non-responder patients demonstrated enrichments of sonic hedgehog signaling pathway and metabolism activity. Integrative multi-layer data analysis further identified molecular determinants, including EBV status, and somatic mutation in GNAQ to ramucirumab activity.

Conclusions: Prospective molecular characterization identified a subset of GC patients with distinct clinical response to ramucirumab therapy, and our results demonstrate the feasibility of personalized therapeutic opportunities in gastric cancer.

Trial registration: The study was registered on ClinicalTrial.gov ( NCT02628951 ) on June 12, 2015.

Keywords: Angiogenesis signature; Gastric cancer; Ramucirumab; TCGA molecular subtype.

Conflict of interest statement

All authors declare no competing interests.

Figures

Fig. 1

Genomic landscape of GC patients in response to ramucirumab. a Genomic landscape of gastric cancer patients with waterfall plot of response to ramucirumab (top panel). The y-axis represents the percentage of maximum tumor reduction assessed according to RECIST 1.1 criteria. The lower dotted line represents tumor reduction of 30%, as per RECIST, representing partial response (PR). The bar graphs are colored based on TCGA molecular classification. The first middle panel represents age distribution, the second middle panel depicts overall mutational burden (non-synonymous mutation), the third middle panel shows histopathological classification, the fourth panel demonstrates HER2 expression level, the fifth panel demonstrates objective response rate to ramucirumab and paclitaxel response, and the last middle panel represents mutational signatures. The bottom panel demonstrates mutational landscape. b Swimmer plot. Each lane represents a single patient’s data. The x-axis represents the duration of ramucirumab therapy for each patient

Fig. 2

Ramucirumab response based on molecular classification. a Percentage of GC patients with clinical response based on TCGA (left panel) and ACRG (right panel) classification. b Box plots demonstrating pathway enrichment scores of each corresponding pathway. Box plots span from the first to third quantiles, and the whiskers represent the 1.5 interquartile range. c Genomic landscape of gastric patients focusing on RTK-RAS, PI3K, and DDR encoding genes. Genomic amplifications are highlighted in red and mutations are highlighted in green. Patients are ordered based on TCGA molecular classification. P values in a were derived from chi-squared tests, and the P values in b were derived from one-way ANOVA

Fig. 3

Transcriptome correlates of clinical response to ramucirumab. a Volcano plot representation of differentially expressed gene analysis between ramucirumab responders (patients who obtained partial response) and non-responders (patients who achieved stable or progressive disease). Genes with > 0.5 log2 fold change and < 0.05 P value are colored in red, and those with < − 0.5 log2 fold change and < 0.05 P value are colored in blue. b Gene Ontology (GO) analysis of differentially expressed genes from a. c Violin plot representations of the pathway enrichment scores via single-sample Gene Set Enrichment Analysis (ssGSEA). Horizontal lines within the violin plots represent 0.25, 0.50, and 0.75 quantiles. P values are calculated using two-sided Wilcoxon’s rank-sum test. d Gene Set Enrichment Analysis (GSEA) between ramucirumab-sensitive and resistance patients

Fig. 4

Identification of molecular determinants to ramucirumab response via integrative analysis. a Elastic-net regression results of transcriptome and genomic features that predict clinical response to objective response rate. The bottom scatter plots represent ramucirumab response based on tumor reduction rate. The upper heatmap shows the top extracted features in the model. The left bar graph shows the averaged weight of each predictive feature. The number of appearance from 100 bootstraps is indicated in parentheses. b Predictive features of ramucirumab response identified by the elastic-net regression model-based analysis are plotted based on their appearance frequency and effect size. Associations are colored in red for overall survival, blue for progression-free survival, and green for tumor reduction rate. The averaged weights for all the features that were extracted against tumor reduction rate were plotted inversely to show consistency in clinical prognosis. Node size is proportional to the single clinical-feature linear correlation