Predictive biomarkers for 5-fluorouracil and oxaliplatin-based chemotherapy in gastric cancers via profiling of patient-derived xenografts

Abstract

Gastric cancer (GC) is commonly treated by chemotherapy using 5-fluorouracil (5-FU) derivatives and platinum combination, but predictive biomarker remains lacking. We develop patient-derived xenografts (PDXs) from 31 GC patients and treat with a combination of 5-FU and oxaliplatin, to determine biomarkers associated with responsiveness. When the PDXs are defined as either responders or non-responders according to tumor volume change after treatment, the responsiveness of PDXs is significantly consistent with the respective clinical outcomes of the patients. An integrative genomic and transcriptomic analysis of PDXs reveals that pathways associated with cell-to-cell and cell-to-extracellular matrix interactions enriched among the non-responders in both cancer cells and the tumor microenvironment (TME). We develop a 30-gene prediction model to determine the responsiveness to 5-FU and oxaliplatin-based chemotherapy and confirm the significant poor survival outcomes among cases classified as non-responder-like in three independent GC cohorts. Our study may inform clinical decision-making when designing treatment strategies.

Fig. 1. Classification of gastric cancer (GC) PDX models, based on responsiveness to 5-FU-based chemotherapy, and similarities to patients’ clinical responses.

a Graphical overview of the study. This figure was generated using BioRender. WES whole-exome sequencing, WTS whole-transcriptome sequencing. b Percent tumor growth inhibition (TGI (%)) of 32 PDX models. Asterisks indicate significant differences between the responder and non-responder groups (*P < 0.05; **P < 0.0001). Violin plot shows the TGI values of PDX models of the responder (n = 13), non-responder (n = 11), and intermediate-responder (n = 8) groups. In the box and whisker plots, data are presented as median value and standard deviation, and the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. Two-tailed Mann–Whitney U test using normal distribution (P = 0.000038) was performed. c Computed tomographic images of two representative patients with clinical responses that were consistent with their respective PDX responses. Cycles of chemotherapy are indicated in the bracket. The yellow arrow indicates tumor mass. M months. d Clinical outcome of patients treated with 5-FU-based chemotherapy. Of four patients whose PDX models were classified as responders, three patients (patient #3, #5, and #19) showed prolonged survival, without recurrence or progression to XELOX and TS-1 treatment. All of the six patients whose PDX models were classified as non-responders showed tumor progression and poor prognosis. XELOX capecitabine (prodrug of 5-FU) and oxaliplatin, FOLFOX folinic acid, 5-FU, and oxaliplatin, TS-1 a combination of tegafur, gimeracil, and oteracil potassium. TS-1 is converted into 5-FU after absorption. The number of treatment cycles is indicated after the drug names. e Progression-free survival of patients, based on the responses of their respective PDX models. Two-sided log-rank test (P = 0.0212).

Fig. 2. Comprehensive genomic profiles associated with responsiveness to 5-FU and oxaliplatin-based chemotherapy.

a Overall mutation patterns for both the responder (R) and non-responder (NR) groups. Tumor mutation burden (TMB) and clinicopathologic information (top), somatic mutation patterns of GC-associated genes (middle), and exclusively mutated cancer-associated genes (mutation frequency >20% only in the R or NR groups; bottom) were compared between the R and NR groups. b Trinucleotide mutation frequencies of the R and NR groups (top) and mutational signatures of single base substitutions (middle). Each mutational signature was grouped by the proposed etiology. Representative mutational signatures of 13 responders and 11 non-responders are shown with violin plots (bottom). In the box and whisker plots, data are presented as median value and standard deviation (SD), and the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. c Somatic indel frequencies of the R and NR groups (top) and differential mutational signatures of small insertions and deletions (bottom). In the box and whisker plots, indel signature proportions of 13 responders and 11 non-responders are presented as median value and SD, and the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. d Gene-level somatic copy number alterations in GC-associated genes (top) and differentially altered cancer-associated genes between the R and NR groups (bottom; Two-sided Fisher’s exact test P < 0.1. Data are provided in Source data).

Fig. 3. Identification of response-associated gene expression features for 5-FU and oxaliplatin-based chemotherapy.

a Gene set enrichment analysis (GSEA) of the R versus NR PDX models, using RNA sequencing data. GSEA demonstrated that metabolism-related pathways were enriched in the R group and Wnt/β-catenin and angiogenesis signaling pathways were enriched in the NR group. Normalized enrichment score (NES) and nominal P value is described. b Identification of core differentially expressed genes (DEGs) between the R and NR groups in both patient tumors and PDX tumors (Top). A total of 40 upregulated genes were identified in the R group and 83 upregulated genes were identified in the NR group. The bottom panel shows the genetic and transcriptomic alterations in TP53-associated genes in the R and NR groups. c Protein–protein interaction enrichment analysis based on STRING network for the 83 DEGs that are upregulated in the NR group. Enrichment P value are corrected for multiple testing using the method of Benjamini and Hochberg.

Fig. 4. Expression profiles for the tumor microenvironment (TME) that are associated with responsiveness to 5-FU and oxaliplatin-based chemotherapy.

a GSEA of the presumed TME expression pattern, based on mouse sequencing reads from PDX tumors. The responder (R) group shows the enrichment of reactive oxygen species (ROS) and metabolism pathways, and the non-responder (NR) group shows the enrichment of TGF-β, Wnt/β-catenin, and angiogenesis signaling pathways. b Core gene expression from the top gene sets identified in the GSEA results. Log₂FPKM expression levels are shown with box plots (n = 13 for Rs, n = 9 for NRs). Data are presented as median value and standard deviation (SD), and the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. The two-tailed Wald test P values are shown. c Candidate TME genes were selected by comparing patient tumors and PDX TME expression levels between the R and NR groups. d Protein–protein interaction (PPI) enrichment analysis for 23 candidate TME genes in the NR group, showing the protein–protein interactions associated with the regulation of angiogenesis. Enrichment P values are corrected for multiple testing using the method of Benjamini and Hochberg. e Estimated endothelial cell (EC) proportion in both patient and PDX tumors (n = 13 for Rs, n = 9 for NRs). Data are presented as box and whisker plot with median value and SD, and the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. Two-tailed Mann–Whitney test P values are shown. n.s. not significant.

Fig. 5. Development of a response prediction model for 5-FU-based chemotherapy.

a The gene expression of 30 classifier genes from among the core DEGs in PDX tumors. b–d Survival analyses for GC patients in ACRG (b), TCGA (c), and Singapore-Duke (d) studies, according to prediction model. Patients were stratified by our prediction model, and the responder-like (R-like) patient group showed a significantly better prognosis than the non-responder-like (NR-like) group. P values were calculated from the log-rank test. ACRG the Asian Cancer Research Group. e–g Reassignment of GC patients from the ACRG study (e), TCGA (f), and Singapore-Duke (g) studies according to prediction model (Top). Survival analyses for subgroups in ACRG (e), TCGA (f), and Singapore-Duke (g) studies, according to prediction model (bottom). P values were calculated from the two-sided log-rank test.

Fig. 6. Summary of integrative characteristics associated with 5-FU-based therapy responding and non-responding tumors.

a Representative pathways enriched in the R group. b Representative pathways enriched in the NR group. The figures were generated with BioRender.