Targeted Dynamic Phospho-Proteogenomic Analysis of Gastric Cancer Cells Suggests Host Immunity Provides Survival Benefit

Abstract

Despite of massive emergence of molecular targeting drugs, the mainstay of advanced gastric cancer (GC) therapy is DNA-damaging drugs. Using a reverse-phase protein array-based proteogenomic analysis of a panel of 8 GC cell lines, we identified genetic alterations and signaling pathways, potentially associated with resistance to DNA-damaging drugs, including 5-fluorouracil (5FU), cisplatin, and etoposide. Resistance to cisplatin and etoposide, but not 5FU, was negatively associated with global copy number loss, vimentin expression, and caspase activity, which are considered hallmarks of previously established EMT subtype. The segregation of 19,392 protein expression time courses by sensitive and resistant cell lines for the drugs tested revealed that 5FU-resistant cell lines had lower changes in global protein dynamics, suggesting their robust protein level regulation, than their sensitive counterparts, whereas the cell lines that are resistant to other drugs showed increased protein dynamics in response to each drug. Despite faint global protein dynamics, 5FU-resistant cell lines showed increased signal transducer and activator of transcription 1 phosphorylation and PD-L1 expression in response to 5FU. In publicly available cohort data, expression of signal transducer and activator of transcription 1 and NFκB target genes induced by proinflammatory cytokines was associated with prolonged survival in GC. In our validation cohort, total lymphocyte count, rather than PD-L1 positivity, predicted a better relapse-free survival rate in GC patients with 5FU-based adjuvant chemotherapy than those with surgery alone. Moreover, total lymphocyte count⁺ patients who had no survival benefit from adjuvant chemotherapy were discriminated by expression of IκBα, a potent negative regulator of NFκB. Collectively, our results suggest that 5FU resistance observed in cell lines may be overcome by host immunity or by combination therapy with immune checkpoint blockade.

Keywords: 5-fluorouracil; NFκB; PD-L1; STAT1; cisplatin; copy number variation; gastric cancer; proteogenomics.

Graphical abstract

Fig. 1

Integrative multiplatform analyses of GC cell lines.A, experimental outline. A panel of 8 GC cell lines were analyzed using multiple platforms including: (i) water-soluble tetrazolium salt-based cell viability assay; (ii) a large panel-based targeted gene sequencing; (iii) RNA sequencing (RNA-seq); and (iv) reverse-phase protein array (RPPA). Sensitivity to each drug was evaluated on the basis of 50% growth inhibitory concentration (GI₅₀). Pathway activity was determined based on the fold changes in mRNA and protein levels from baseline to 24 h after drug treatment. 5FU, 5-fluorouracil; CIS, cisplatin; ETP, etoposide; DTX, Docetaxel; PGx, pharmacogenomics. B–F, GI₅₀ for each drug tested (B), cell line summary (C), copy number variation (CNV) landscape (D), and mRNA and protein expression at baseline (E and F) are depicted. Cell lines are ordered by 5FU sensitivity. DTX was used as a non-DNA-damaging drug. Relevant genes for cell line summary and expression profiles were selected based on previous studies of GC subtypes (17, 18). EBV gene expression was detected by quantitative RT-PCR (Supplemental Fig. S1C).

Fig. 2

CNV and mesenchymal gene expression in CIS-sensitive GC cells.A, GI₅₀ profiles of GC cell lines to define sensitivity and resistance to CIS and ETP. B, scatter plot showing correlation between area under the dose-response curve (AUC) values for CIS and ETP across gastric and colon cancer cell lines. Each dot indicates an individual cell line (n = 17) shared between the AUC data sets. C, heatmaps represent pairwise Pearson correlations between AUC values of CIS and ETP assessed in the PRISM repurposing secondary screen (30). D, number of CNV loci detected. Each dot indicates an individual cell line. The two-tailed p values were obtained with a nonparametric Mann-Whitney U test. E, binary matrix representing genes that commonly lost its copy number in cell lines with dual CIS/ETP sensitivity. The two-tailed p values were obtained with Fisher’s exact test. F, scatter plot showing correlation between AUC values of CIS or ETP and KIT copy number across gastric and colon cancer cell lines. Each dot indicates an individual cell line. G, hallmark GSEA signatures from RNA-seq data ranked by Normalized Enrichment Score (NES) for CIS/ETP-sensitive versus CIS/ETP-resistant cell lines (left). GSEA plots showing positive and negative enrichment of “EMT” and “DNA repair” gene sets in CIS/ETP-sensitive cell lines (right). FDRq, false discovery rate (q value). H, time course RPPA data showing changes in cleaved PARP levels after CIS treatment. Error bars represent s.e.m. I, caspase-focused RPPA analysis of dual CIS/ETP-sensitive (mean, n = 5) and -resistant (mean, n = 3) cell lines (left) and a schematic of CIS-activated signaling outcomes in dual CIS/ETP-sensitive cell lines (right). c-C3−9, cleaved caspase-3−9; c-PARP, cleaved PARP.

Fig. 3

5FU-induced MAPK signaling pathways.A, GI₅₀ profiles of GC cell lines to define 5FU sensitivity and resistance. Two pairs of 5FU resistant and parental cell lines are highlighted. B, pharmacogenomics panel shows drug metabolizing genes harboring single-nucleotide variants (SNVs) in each cell line. DPYD is highlighted as a canonical 5FU metabolism pathway gene. C, comprehensive cancer panel. TP53, TET2, FGFR2, and NF1 are highlighted for common, MKN74-unique, and 45FU-unique mutations. D, CNV analysis showing no shared copy number changes between 74FU and 45FU. Copy number changes were calculated by subtracting the copy number of each chromosomal position in matched parental cell lines from that in resistant cell lines (ΔCNV). Chromosomal positions with maximum and minimum ΔCNVs are highlighted. E, Pearson correlations between GI₅₀ 5FU and protein expression at baseline (top panel) and 24 h after 5FU treatment (bottom panel). F, temporal proteomic changes in 8 cell lines within 24 h of 5FU, CIS, ETP, or DTX treatment. Seven clusters were determined by K-means clustering and further grouped for early, intermediate, late, and no response based on their kinetics. G, proportions of protein expression time courses segregated into three doses for each drug. Two-tailed p values were obtained with Fisher’s exact test. H, proportions of protein expression time courses from high-dose conditions segregated into six signaling pathways including DNA damage response (DDR), MAPK, PI3K, STAT, NFκB, and WNT pathways. I, MAPK pathway-focused RPPA analysis of 45FU and parental MKN45 cells. J, temporal changes in MEK1-pS²⁹⁸ levels after 5FU treatment. K, MAPK pathway-focused RPPA analysis of 74FU and parental MKN74 cells. L, temporal changes in MEK1-pS²⁹⁸ levels after 5FU treatment. M, 5FU-activated MAPK signaling pathways in 45FU cells. N, 5FU-activated MAPK signaling pathways in 74FU cells. Error bars represent s.d. (A) or s.e.m. (J and L).

Fig. 4

5FU elicits PD-L1 expression in 5FU-resistant GC cells.A, hallmark GSEA signatures from RNA-seq data ranked by NES for 45FU versus parental MKN45 cells. B, GSEA plots showing positive and negative enrichment of “G2/M checkpoint” and “IFNα response” gene sets in 45FU cells. C, hallmark GSEA signatures from RNA-seq data ranked by NES for 74FU versus parental MKN74 cells. D, GSEA plots showing positive and negative enrichment of “IFNα response” and “G2/M checkpoint” gene sets in 74FU cells. E, DNA damage response (DDR) pathway-focused RPPA analysis of 45FU and parental MKN45 cells (left) and a schematic summarizing signaling outcomes in 45FU cells (right). F, DDR pathway-focused RPPA analysis of 74FU and parental MKN74 cells (left) and a schematic summarizing signaling outcomes in 74FU cells (right). c-PARP, cleaved PARP; c-C3−9, cleaved caspase-3−9. G, STAT and NFκB (proinflammatory) pathway-focused RPPA analysis of 45FU and parental MKN45 cells (left) and a schematic summarizing signaling outcomes in 45FU cells (right). H, proinflammatory pathway-focused RPPA analysis of 74FU and parental MKN74 cells (left) and a schematic summarizing signaling outcomes in 74FU cells (right). I, Pearson correlations between STAT1-pY⁷⁰¹ expression and expression of other proteins (left) and a representative scatter plot showing correlation between STAT1-pY⁷⁰¹ and PD-L1 (right). J, Pearson correlations between IκBα-pS^32/36 expression and expression of other proteins (left) and a representative scatter plot showing correlation between IκBα-pS^32/36 and PD-L1 (right). FDRq, false discovery rate (q value) (B and D).

Fig. 5

Survival curves for patients with stage II/III GC stratified by potential confounding factors.A, Kaplan-Meier curves for overall survival (OS) in PD-L1^– (left) or PD-L1⁺ (right) patients treated with or without S-1. B, Kaplan-Meier curves for OS in patients with low total lymphocyte count (TLC^–) or high total lymphocyte count (TLC⁺) who did or did not receive S-1 treatment. C, subgroup analysis of OS in PD-L1^– (n = 188) or PD-L1⁺ (n = 78) patients (left) and TLC^– (n = 128) or TLC⁺ (n = 138) patients (right) evaluated for surgery alone and surgery followed by S-1 adjuvant chemotherapy. D, positive correlation between inflammation score and H. pylori positivity (left). Two-tailed p values were obtained with the nonparametric Mann-Whitney U test. Subgroup analysis of OS in H. pylori^– (n = 131) or H. pylori⁺ (n = 66) patients evaluated for surgery alone and surgery followed by S-1 adjuvant chemotherapy (right). E, Kaplan-Meier curves for OS in H. pylori^– (left) or H. pylori + (right) patients who did or did not receive S-1. F, hazards for OS were evaluated for the expression of genes regulated by GC cell-intrinsic signaling in response to proinflammatory cytokines. G, Kaplan-Meier curves for OS in ACRG GC patients (n = 283) stratified based on tumor IRF1 (left) and NFKBIA (right) expression levels. Cox proportional hazards model was used to determine the hazard ratio of each mRNA expression level (C, D, and F). Error bars represent 95% confidence intervals. The p values were obtained with log-rank test (A, B, E, and G). Stratification strategies are shown in Supplemental Fig. S6A. ACRG, Asian Cancer Research Group (18).

Fig. 6

Survival curves for patients with stage II/III GC stratified by IκBα level and potential confounding factors.A, representative immunohistochemical staining in a IκBα^– tumor region. B, Kaplan-Meier curves for OS in IκBα^– patients stratified by treatment (i.e., S-1 and Surgery) divided into TLC^– and TLC⁺ groups. C, subgroup analysis stratified by interaction of TLC^– (n = 53) or TLC⁺ (n = 64) based on the hazard for OS was evaluated by treatment. D, Kaplan-Meier curves for RFS in IκBα^– patients stratified by treatment divided into TLC^– and TLC⁺ groups. E, subgroup analysis stratified by interaction of TLC^– (n = 53) or TLC⁺ (n = 64) based on the hazard for OS was evaluated by treatment. F, representative immunohistochemical staining in a IκBα⁺ tumor region. G, Kaplan-Meier curves for OS in IκBα⁺ patients stratified by the treatment divided into TLC^– and TLC⁺ groups. H, subgroup analysis stratified by interaction of TLC^– (n = 58) or TLC⁺ (n = 49) based on the hazard for OS was evaluated by treatment. I, Kaplan-Meier curves for RFS in IκBα⁺ patients stratified by treatment divided into TLC^– and TLC⁺ groups. J, subgroup analysis stratified by interaction of TLC^– (n = 58) or TLC⁺ (n = 49) based on the hazard for RFS was evaluated by treatment. Scale bar, 50 μm (A and F). Risk for survival was evaluated using Cox proportional hazards models (C, E, H and J). Error bars represent 95% confidence intervals. The p values were obtained with a log-rank test (B, D, G and I).