PI3 expression predicts recurrence after chemotherapy with DNA‐damaging drugs in gastric cancer

Abstract

Despite recent advances in gastric cancer therapy, chemotherapy resistance and lack of methods for selecting combination regimens remain major problems. Organoids, which provide a culture system that more closely resembles tumor cell organization than traditional cell lines, can be established from surgical specimens with a high success rate and are widely used for drug sensitivity assays. In this study, we aimed to identify a novel biomarker for predicting multidrug resistance using gastric cancer organoids (GCOs). We evaluated 5‐fluorouracil or oxaliplatin‐resistant GCOs to find novel biomarkers that reflect multidrug resistance in gastric cancer. To examine the resistance mechanisms, RNA‐sequencing analysis and ex vivo drug sensitivity testing were performed. The association of biomarkers with patient prognosis and chemotherapy efficacy was evaluated using three original cohorts with a total of 230 cases. The results were also validated with two independent public cohorts and single‐cell RNA sequence data. Increased expression of peptidase inhibitor 3 (PI3) was detected in all 5‐fluorouracil or oxaliplatin‐resistant GCOs. Our findings suggest a potential association of PI3 expression with ribosome biosynthesis and RNA metabolism under organoid conditions. We also found that PI3 overexpression promoted 5‐fluorouracil/oxaliplatin/cisplatin resistance but not paclitaxel resistance. Immunohistochemical evaluation of PI3 expression revealed that the PI3‐positive gastric cancer group had a poorer outcome, especially in terms of time to recurrence. PI3 positivity was also an independent predictor of relapse after chemotherapy with DNA‐damaging agents. PI3 promotes DNA‐damaging drug resistance through multiple downstream regulations related to RNA and ribosomal metabolism. PI3 may be useful as a biomarker for the therapeutic selection of non‐DNA‐damaging agents. © 2025 The Author(s). The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.

Introduction

Despite recent breakthroughs in cancer treatment, gastric cancer remains a leading cause of cancer‐related death worldwide [1]. Gastric cancer has long been treated with a limited number of cytotoxic chemotherapeutic agents, including 5‐fluorouracil (5‐FU), cisplatin (CDDP), oxaliplatin (L‐OHP), taxane agents, and irinotecan, emphasizing the importance of selecting the optimal treatment regimen among multiple anticancer agent combinations [2]. Despite advancements in genomics that have facilitated the selection of molecularly targeted therapies, progress in chemotherapy has been slow owing to the diversity and complexity of the mechanisms of action of anticancer agents and the development of resistance to these drugs [3, 4]. Understanding the mechanisms underlying chemotherapy and resistance is crucial for developing more effective therapeutic strategies. To date, much effort has been devoted to identifying biomarkers for selecting drugs or predicting the efficacy of chemotherapy [5, 6, 7], but none has achieved clinical translation. With the recent recognition of the potential benefits of predicting the efficacy of combination chemotherapy in patients [8], the importance of such investigations has been increasingly emphasized.

The use of patient‐derived organoids has recently increased in anticancer drug‐related research [9]. Organoids are generated using an ex vivo culture method that employs Matrigel and media containing niche factors, allowing the generation of three‐dimensional (3D) cultures using cells derived from surgical specimens. Compared with traditional two‐dimensional (2D) cell lines, intestinal organoids are thought to more closely mirror in vivo conditions in humans, preserving the genetic and phenotypic characteristics of the original tumors [10]. In addition, organoids have a higher establishment efficiency than mouse xenografts and are widely used for chemotherapy screening and drug sensitivity testing [5, 9]. Cancer organoids are also known to contain abundant cancer stem cells (CSCs), a subpopulation of tumor cells with self‐renewal and differentiation capabilities. CSCs are thought to contribute to tumor initiation, metastasis, and the acquisition of chemotherapy resistance by altering their phenotype during treatment [11, 12, 13]. Existing methods for investigating CSCs include the use of marker genes such as CD44, AQP5, and CD133, as well as the analysis of characteristics such as drug efflux capacity or tumorigenic potential. The effectiveness of organoids in studying CSCs has also been clearly established [14, 15, 16, 17, 18, 19]. However, the mechanisms underlying the chemotherapy resistance of CSCs and stem‐like cells in the stomach remain unclear, especially under multidrug conditions. Therefore, in this study, we used patient‐derived gastric cancer organoids (GCOs) to understand the mechanism of multidrug resistance in cells, including CSCs, and to identify novel biomarkers to predict chemoresistance.

In this study, we first reanalyzed microarray data on multidrug resistance using anticancer drug‐resistant GCOs established in our previous reports [17, 18]. We focused on two agents frequently used in the treatment of gastric cancer in Japan, 5‐FU and L‐OHP. Peptidase inhibitor 3 (PI3, alias ELAFIN), also known as trappin‐2, was identified as a novel biomarker associated with 5‐FU/platinum‐drug‐based chemotherapy resistance in gastric cancer. PI3 is a whey acidic protein family serine protease inhibitor [20]. PI3 exhibits antibacterial, anti‐inflammatory, and wound‐healing effects by inhibiting the protease activities of human neutrophil elastase and proteinase 3 [21, 22]. It has also been shown to play a variety of highly complex roles in malignant tumors [23]. It has been reported that PI3 expression may be regulated by the nuclear factor (NF)‐κB pathway [24] and that PI3 shields cyclin E from proteolysis by cellular elastases regulating the cell cycle [23, 25]. The extracellular secretion of PI3 has also been reported to act on the epidermal growth factor receptor (EGFR) [26]. However, there has been no previous report evaluating the involvement of PI3 in gastric cancer. The association between PI3 and chemotherapy resistance has been investigated in several studies on ovarian cancer, including the higher expression of PI3 in chemotherapy‐resistant ovarian cancer [24, 25, 27]. In particular, Wei et al demonstrated that knockdown and forced expression of PI3 alter the CDDP sensitivity of ovarian tumor cells [27]. However, the mechanisms underlying these findings have not yet been elucidated. Here, we investigated the association between PI3 and drug resistance, as well as PI3 expression in gastric cancer tissues, and its significance as a prognostic and predictive factor for chemotherapy outcomes.

Materials and methods

Ethics approval

This study was approved by the Ethics Committee for Human Genome Research of Hiroshima University (E‐597‐01) and the Institutional Review Board of the National Cancer Center (2005‐043, 2020‐242, 2022‐059) and was conducted in accordance with the Ethical Guidance for Human Genome/Gene Research of the Japanese Government. Written informed consent was obtained from all patients for the establishment of organoids and sample collection from the cohort.

Human tissues

The human gastric cancer and normal gastric tissues used for organoid establishment were obtained from patients who underwent surgery at the Department of Gastroenterological and Transplant Surgery, Hiroshima University Hospital, Hiroshima, Japan; Kure Medical Center, Chugoku Cancer Center, Kure, Japan; and National Cancer Center Hospital East, Kashiwa, Japan. Drug‐resistant GCOs were established in our previous study [17, 18].

Establishment and culture of human GCOs

Human parental/chemoresistant GCOs were established and cultured in organoid media containing niche factors, as described previously [17], and were passaged twice a week with a split ratio of 1:3/1:6. Clinical data of the GCOs used in this study and details of the composition of the GCO medium are summarized in Supplementary materials and methods and supplementary material, Tables S1 and S17, respectively.

Microarray

The microarray of GCOs using the Clariom D array, Human (Thermo Fisher Scientific, Waltham, MA, USA) was analyzed in our previous studies [17, 18].

Immunohistochemistry

We prepared 4‐μm‐thick sections from formalin‐fixed paraffin‐embedded tissue blocks. Immunohistochemistry was performed using a BenchMark ULTRA system (Ventana Medical Systems, Tucson, AZ, USA). The tissue sections were stained for rabbit polyclonal antibodies against human PI3 (HPA017737, 1:500; Atlas Antibodies, Bromma, Sweden). The OptiView DAB IHC Detection Kit and OptiView Amplification Kit (Ventana Medical Systems) were used for color reactions. Two of our researchers evaluated the unequivocally positive area in the cancer region without knowledge of the clinical and pathological parameters or patient outcomes. Hematoxylin and eosin (H&E) staining was also performed on the serial sections.

Statistical analyses

All statistical analyses were performed using Python (https://www.python.org/, version 3.9.7) or R software (https://www.r-project.org/, version 4.2.1). All tests were two‐sided, and statistical significance was set at p < 0.05. Graphs and their error bars are presented as mean ± SD. Categorical factors were compared between groups using the χ ² test. The independent Student's t‐test, ANOVA, Mann–Whitney U‐test, and Kruskal–Wallis test were used to compare normally and nonnormally distributed continuous variables. The Spearman's rank correlation coefficient was used for correlation analyses. The correlation between PI3 expression levels and clinicopathological characteristics was analyzed using Fisher's exact test. Overall survival (OS) and time‐to‐recurrence (TTR) were calculated as the time from surgery or chemotherapy to physical or radiographic evidence of disease recurrence, date of death, or date of last contact if no death or recurrence occurred. For all survival analyses, patients were grouped into high‐ and low‐expression groups using the optimal cutoff values obtained from the time‐dependent receiver operating characteristic (ROC) curve analysis. Cutoffs were calculated using staining percentages of immunostaining for the original cohorts and raw data on RNA expression normalized signal intensity for the Asian Cancer Research Group (ACRG) and Yonsei University Severance Hospital (YUSH) cohorts. Each case was classified into the low group if its immunostaining or RNA expression was lower than the cutoff or into the high group if it was equal to or higher than the cutoff. The specific cutoff values used for each cohort are displayed in the figures. The Kaplan–Meier method was used to estimate survival times, and distributions were compared using the log‐rank test. Multivariate Cox proportional hazards regression analysis was performed for all variables available for use. Two‐tailed 95% confidence intervals (CIs) and p values were calculated.

Additional methods and details are available in Supplementary materials and methods.

Results

Overexpression of PI3 in drug‐resistant GCOs

In our previous studies [17, 18], we established anticancer drug‐resistant GCOs by continuous exposure of parental GCOs to 5‐FU or L‐OHP for 48–160 days (Figure 1A; see supplementary material, Figure S1, Table S1). The half maximal inhibitory concentration (IC₅₀) values of the resistant GCOs were more than two‐fold higher than those of the parental GCOs (Figure 1B). Microarray analysis of these GCOs revealed elevated expression of 357 genes, only four of which were common to both 5‐FU and L‐OHP resistance (Figure 1C; see supplementary material, Table S2). We further focused on these four genes to identify common characteristics in the mechanisms underlying resistance to these two DNA‐damaging drugs. In The Cancer Genome Atlas Stomach Adenocarcinoma (TCGA–STAD) data, when patients were divided into two groups based on the signatures of these four genes, disease‐free survival (DFS) was significantly worse in the group with a higher signature [hazard ratio (HR) = 2.0; log‐rank p = 0.00043; Figure 1D]. Furthermore, analysis of individual gene expressions demonstrated a unique relationship between PI3, also known as PI3 in protein form (HR = 1.9; log‐rank p = 0.0018), and laminin subunit gamma 2 (LAMC2; HR = 1.6; log‐rank p = 0.014) with a shorter DFS (see supplementary material, Figure S2A). The expression of each gene in each resistant GCO was confirmed using RT–qPCR, and PI3 expression was particularly high in all resistant GCOs (Figure 1E). Overall, among the four candidate genes, PI3 was considered the most promising gene associated with drug resistance and gastric cancer prognosis.

Figure 1.

Overexpression of PI3 in drug‐resistant GCOs. (A) Schematic diagram of establishment of anticancer drug‐resistant GCOs from our previous research [17, 18]. Each concentration of the drug was continuously added to the medium for the indicated number of days for establishment. (B) IC₅₀ values calculated from drug sensitivity tests of parental GCOs and resistant GCOs [17, 18]. (C) Venn diagram of genes upregulated in resistant GCOs identified by microarray analysis [17, 18]. Genes upregulated in both resistant GCOs are shown in center. (D) Survival curves for disease‐free survival of TCGA–STAD data. The dashed line represents the 95% CI. p = 0.00043 from log‐rank test. (E) Heatmap showing fold‐change of expression in resistant GCOs compared to parental GCOs by RT‐qPCR for each gene expression (n = 3). NA indicates that the expression is below the detection limit. (F) Extracellular PI3 in parental and resistant GCOs examined by ELISA, representing the fold‐change of PI3 secretion in resistant GCOs compared to parental GCOs (n = 3). N.S., not significant from Student's t‐test. (G) PI3 expression fold‐change of resistant to parental cell lines measured by RT‐qPCR (n = 3). N.S., not significant from Student's t‐test. (H) Results of drug sensitivity tests of eight different GCOs. Each dot represents the average IC₅₀ value of a single GCO (n = 3). GCOs are divided into two groups (high/low) based on the PI3 expression level of each GCO measured by RT‐qPCR. (I) Heatmap of PI3 expression and IC₅₀ values for eight GCOs. All values are min–max normalized by column and shown between zero and one. Results of hierarchical clustering using IC₅₀ values for each drug are also shown. p = 0.02 from Student's t‐test between PI3 high/low for the three drugs: CDDP, L‐OHP, and 5‐FU.

Although previous reports have indicated that PI3 is secreted extracellularly [26], no significant change was observed in the extracellular secretion of PI3 in our resistant GCOs (Figure 1F). We also found that the expression of PI3 was not upregulated in resistant cancer cell lines cultured under 2D conditions (Figure 1G). These findings suggest that the expression of PI3 in gastric cancer cells cultured under in vivo‐like conditions is associated with the acquisition of resistance. Next, we comprehensively examined PI3 gene expression levels and the IC₅₀ values of 5‐FU, L‐OHP, CDDP, and paclitaxel (PTX) using eight GCOs. When the GCOs were divided into two groups based on PI3 expression, the average IC₅₀ values of 5‐FU, L‐OHP, and CDDP were higher in the PI3 high‐expressing group, and the difference was reversed for PTX, although the differences were not significant owing to the limited number of cases (Figure 1H; see supplementary material, Figure S2B). Hierarchical clustering after normalization of the IC₅₀ values showed that the patterns for L‐OHP and CDDP were similar, whereas PTX was placed in a separate cluster from the other anticancer agents (Figure 1I). Moreover, the normalized IC₅₀ values of 5‐FU, L‐OHP, and CDDP were significantly higher in the group with high PI3 expression (p = 0.02; Figure 1I). These results indicate that PI3 expression may be involved in resistance to DNA‐damaging agents in GCOs.

PI3 expression regulates ribosome biogenesis and RNA metabolism in gastric cancer

We performed PI3 overexpression (PI3OE) and control vector transfection (Ctrl) in three gastric cancer cell lines and three GCOs to investigate the effect of PI3 expression (supplementary material, Figure S3). Gene expression was comprehensively analyzed using RNA sequencing (RNA‐seq). The principal component analysis plots of the cell lines and GCOs were highly distinct (Figure 2A). Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis comparing the Ctrl‐to‐Ctrl expression patterns of cell lines and GCOs yielded 39 enriched pathways in GCOs (p < 0.01; supplementary material, Table S3). Of these, oxidative phosphorylation, peroxisome proliferator‐activated receptor (PPAR) signaling pathway, and glutathione metabolism have been reported to be associated with CSCs [28, 29]. Next, we performed Gene Ontology (GO) analysis to infer the function of PI3 and found that the number of GOs altered by PI3 forced expression (p < 0.01) was greater in GCOs than it was in the cell lines, and only two GOs were commonly altered (Figure 2B; supplementary material, Tables S4 and S5). Importantly, multiple nucleic acid/ribosome‐associated GOs were identified among the enriched GOs only in PI3OE GCOs (Figure 2C; supplementary material, Figure S4A). We subsequently analyzed the differentially expressed genes (DEGs; p < 0.05) using PI3OE in GCOs or cell lines and found that no genes were shared between the GCOs and cell lines (supplementary material, Tables S6–S9). Next, we performed a prognostic analysis of TCGA–STAD data using the DEGs of the PI3OE GCOs. Among the 36 DEGs whose expression were downregulated in the PI3OE GCOs and whose expression data were also included in TCGA–STAD, we found no correlation between the signature and DFS (HR = 1.2; log‐rank p = 0.36; see supplementary material, Figure S4B). In contrast, a significant correlation was observed between the high signature of upregulated genes and poor DFS (HR = 1.6; log‐rank p = 0.022; Figure 2D), which is consistent with our analysis of PI3 expression alone (supplementary material, Figure S2A).

Figure 2.

PI3 expression is associated with ribosome biogenesis and RNA metabolism in gastric cancer. (A) Principal component analysis of Ctrl, PI3OE cell lines (MKN‐1, MKN‐45, and HSC44PE), and GCOs (GCO1, GCO2, and GCO3). Each point represents an individual sample. Solid lines represent a cluster of organoids, and dotted lines represent a cluster of cell lines. (B) Number of GOs altered by PI3OE [nominal (NOM) p < 0.01]. Upregulated GOs are shown in red, downregulated GOs in blue. (C) Normalized enrichment score (NES) of representative enhanced GOs in PI3OE GCOs among GOs with NOM p < 0.01. (D) Disease‐free survival analysis of TCGA–STAD based on the signature of differentially expressed genes upregulated in PI3OE GCOs. p = 0.022 from log‐rank test. (E) Uniform Manifold Approximation and Projection of public gastric cancer scRNA‐seq data showing heterogeneity of PI3 expression levels in gastric cancer cells. (F) Venn diagram of enhanced GOs in both PI3OE GCOs and PI3‐positive cell‐rich cluster from public gastric cancer scRNA‐seq data. The numbers in the diagram indicate the number of all GOs with p < 0.01. Three GOs shown in the NES graph were common to both data.

To validate the RNA‐seq results and link them to clinical data, we analyzed publicly available single‐cell RNA‐seq (scRNA‐seq) data from resected or biopsied gastric cancer samples. After a sufficient quality check and filtering out nontumor cells, the gastric cancer cells showed a heterogeneous PI3 expression pattern, revealing a cluster of PI3‐high‐expressing gastric cancer cells (Figure 2E; supplementary material, Figure S5A–E). CD44 expression was also high in this cluster (supplementary material, Figure S5F,G). GO analysis of this cluster revealed 196 GO terms that were significantly upregulated compared with the other tumor cells (p < 0.01; supplementary material, Table S10). Three GOs overlapped with our original RNA‐seq data and the public scRNA‐seq data, all of which were related to ribosome biosynthesis and RNA metabolism (Figure 2F; supplementary material, Figure S5H). Taken together, these results indicate that there is an association between PI3, DFS, and resistance to DNA‐damaging agents potentially involving the regulation of nucleic acid and ribosomal metabolism in GCOs.

PI3 promotes DNA‐damaging agent resistance only in cells under organoid culture medium conditions

Functional analyses using both cell lines and GCOs were conducted to elucidate the differences between cell lines and GCOs. Drug sensitivity tests were initially performed using the gastric cancer cell line MKN‐1 cultured in 2D; however, no significant differences were observed in PI3 expression (Figure 3A). Similar results were observed in the MKN‐45 and HSC44PE cell lines (see supplementary material, Figure S6A). We also conducted 2‐week exposure experiments to examine the effect of PI3 on the proliferation potential during long‐term exposure, but no significant differences were observed (see supplementary material, Figure S6B). We then assessed drug sensitivity in GCOs and discovered that PI3OE GCOs were significantly resistant to 5‐FU and L‐OHP (p < 0.01; Figure 3B; see supplementary material, Figure S7A). Notably, we observed a similar pattern of changes with CDDP (p < 0.01); however, we found no significant difference with PTX (Figure 3C; see supplementary material Figure S7B). Taken together, these results indicate that PI3 may be responsible for the acquisition of resistance to DNA‐damaging agents.

Figure 3.

PI3 promotes DNA‐damaging agent resistance only in the cells under organoid culture medium conditions. (A) Sensitivity test of 5‐FU and L‐OHP using gastric cancer cell line MKN‐1. N.S., not significant from Student's t‐test. (B) Sensitivity test of 5‐FU and L‐OHP using GCO 2. **p < 0.01 from Student's t‐test. (C) Sensitivity testing of CDDP and PTX using GCO2. **p < 0.01, N.S., not significant from Student's t‐test. (D) Schematic of experimental design for three dimensioning cell lines with Matrigel. Cell lines were embedded in Matrigel, and subsequent experiments were performed after more than 3 weeks. (E) Schematic of experimental system for culturing cell lines in organoid medium. The medium was changed to organoid medium instead of normal medium, and subsequent experiments were performed after more than 3 weeks. (F) Sensitivity test of the 3D gastric cancer cell line MKN‐1 to 5‐FU and L‐OHP. N.S., not significant from Student's t‐test. (G) Sensitivity test of medium‐modified MKN‐1 to 5‐FU and L‐OHP. **p < 0.01, *p < 0.05, from Student's t‐test. All data represent n = 3 results, with each plot representing the respective measured value together with the calculated IC₅₀.

To examine the differences between the cell lines and organoids, we focused on the culture conditions. Matrigel was used in one experiment to replicate the culture dimensions of the organoids (Figure 3D), whereas another experiment involved switching the culture medium to an organoid medium (Figure 3E). The results of drug sensitivity tests under all culture conditions showed no significant differences when the dimensions were modified (Figure 3F; supplementary material, Figure S8A); however, differences only appeared when the culture medium was changed (p < 0.05; Figure 3G; supplementary material, Figure S8B). These results indicated that the effects of PI3 on drug resistance may be specific to cells cultured in an organoid medium or in vivo‐like medium conditions.

Clinical significance of PI3 expression in patients with gastric cancer

To evaluate the clinical significance of PI3, immunostaining for PI3 was performed on samples from Original Cohort 1, which consisted of 198 patients with gastric cancer who underwent surgical resection. PI3 immunostaining was minimal in the nontumor area, but those cases with high PI3 immunostaining intensity in the nontumor area frequently exhibited severe inflammation (Figure 4A,B; supplementary material, Figure S9A,B). In contrast, tumor cells were stained at a significantly higher rate than nontumor tissues (Figure 4C). PI3 immunostaining was observed within the cytoplasm, as well as various staining patterns, including the staining of single or small clusters of cells (Figure 4A,B; supplementary material, Figure S9C,D).

Figure 4.

Immunohistochemistry of PI3 in patients with gastric cancer. (A and B) Representative H&E‐stained image and PI3‐immunostained image of the same region. The scale bar is 500 μm, and a red frame enlarges the region where the single cancer cell is stained. In this case, the scale bar is 50 μm. (C) PI3‐positive rates in tumor and nontumor regions for all patients in Original Cohort 1. Each dot represents an individual case. **p < 0.01 from ANOVA. (D) Tumor and normal PI3 expression in TCGA–STAD and GTEx. Each dot represents an individual case. **p < 0.01 from ANOVA. (E) Overall survival curve of Original Cohort 1. The black line represents the PI3‐negative group (n = 109), the red line represents the PI3‐positive group (n = 89). The cutoff value was calculated using staining percentages of immunostaining, displayed in the figure. p = 0.0119 from the log‐rank test. (F) Multivariate Cox proportional hazards regression analysis of OS in Cohort 1. Rhombus represents the HR, and the line represents the 95% CI. P values in red represent a statistically significant difference.

PI3 expression was higher in tumors than in normal tissues in TCGA–STAD and Genotype–Tissue Expression (GTEx) normal stomach tissues (Figure 4D). Cases with a PI3‐positive area ratio greater than 1% were defined as PI3‐positive, and their association with clinicopathological factors were investigated. We found that PI3 positivity was correlated with differentiation, N grade, lymphatic invasion, and vascular invasion (p < 0.05; Table 1; supplementary material, Figure S10A). In a subsequent analysis based on OS, the 5‐year survival rate was significantly worse in the PI3‐positive group (negative = 0.922%, 95% CI = 0.851–0.960; positive = 0.767%, 95% CI = 0.663–0.843; p = 0.0119; Figure 4E). Multivariate Cox proportional hazard regression analysis of OS also revealed that PI3 positivity was an independent poor prognostic factor (p = 0.02; Figure 4F).

Table 1.

Clinicopathological features of Original Cohorts 1, 2, and 3.

		Original Cohort 1 (n = 198)			Original Cohort 2 (n = 93)			Original Cohort 3 (n = 26)
		All cases in a year			Adjuvant chemotherapy			Neoadjuvant chemotherapy
		PI3 Negative	PI3 Positive	p value	PI3 Negative	PI3 Positive	p value	PI3 Negative	PI3 Positive	p value
Age (years)	<65	41	29	0.550	21	16	0.210	7	5	0.695
	65 < =	68	60		24	32		6	8
Sex	Female	40	24	0.170	16	12	0.366	6	3	0.411
	Male	69	65		29	36		7	10
Differentiation	Diff	44	51	0.022	11	22	0.050	5	6	1.000
	Undiff	65	38		34	26		8	7
T grade	pT1	68	45	0.200	7	5	0.877	2	0	0.122
	pT2	10	10		7	8		2	3
	pT3	17	13		14	14		1	5
	pT4	14	21		17	21		8	5
N grade	pN0	80	50	0.002	16	10	0.018	4	3	1.000
	pN1	13	11		13	10		4	4
	pN2	12	10		12	11		3	3
	pN3	4	18		4	17		2	3
M grade	M0	107	89	0.503	41	47	0.194	9	11	0.645
	M1	2	0		4	1		4	2
Stage	pStage I	74	47	0.066	8	5	0.288	0	1	0.465
	pStage II	17	17		16	17		4	3
	pStage III	17	25		17	25		4	7
	pStage IV	1	0		4	1		5	2
Lymphatic invasion	Negative	80	40	<0.001	16	11	0.253	9	4	0.115
	Positive	29	49		29	37		4	9
Vascular invasion	Negative	76	44	0.005	20	11	0.047	6	5	1.000
	Positive	33	45		25	37		7	8

Note: Bold: p < 0.05, PI3 status is based on immunohistochemistry (protein).

PI3 expression as a predictive factor for relapse after DNA‐damaging drug treatment

To evaluate the significance of PI3 in chemotherapy performance, we established Original Cohort 2, which included 93 surgical specimens from patients who received postoperative chemotherapy. In this cohort, PI3 positivity was linked to N grade and vascular invasion as a clinicopathological variable (p < 0.05; Table 1; supplementary material, Figure S10B). Analysis of the TTR revealed a significantly poorer 5‐year recurrence‐free rate in the PI3‐positive group (negative = 0.704, 95% CI = 0.545–0.817; positive = 0.491, 95% CI = 0.342–0.624; p = 0.0436; Figure 5A). Furthermore, multivariate analysis using all available factors revealed that PI3 expression was an independent predictor of the response (p = 0.047; Figure 5B; supplementary material, Table S11).

Figure 5.

Clinical significance of PI3 in patients with gastric cancer administered chemotherapy. (A) TTR curve of Original Cohort 2. The black line represents the PI3‐negative group (n = 45), the red line the PI3‐positive group (n = 48). The cutoff value was calculated using staining percentages of immunostaining, displayed in the figure. p = 0.0436 from the log‐rank test. (B) Multivariate Cox proportional hazards regression analysis of TTR in Original Cohort 2. The rhombus represents the HR, and the line represents the 95% CI. Red asterisks represent p < 0.05. (C) TTR curve of YUSH cohort. The black line represents the PI3‐negative group (n = 18), the red line the PI3‐positive group (n = 31). The cutoff value was calculated using RNA expression normalized signal intensity, displayed in the figure. p = 0.0151 from the log‐rank test. (D) Multivariate Cox proportional hazards regression analysis of TTR in YUSH cohort. Red asterisks represent p < 0.05. (E) TTR curve of ACRG cohort. The black line represents the PI3‐negative group (n = 38), the red line the PI3‐positive group (n = 44). The cutoff value was calculated using RNA expression normalized signal intensity, displayed in the figure. p = 0.0567 from the log‐rank test. (F) Multivariate Cox proportional hazards regression analysis of TTR in ACRG cohort. Red asterisks represent p < 0.05. (G) Relationship between absence or presence of neoadjuvant chemotherapy and PI3‐positive frequency for all patients in Original Cohorts 1–3. N.S., not significant from χ ² test. (H) Correlation between duration of neoadjuvant chemotherapy and percentage of PI3‐positive area. p = 0.041 from Spearman rank correlation analysis. (I) TTR curve of Original Cohort 3. The black line represents the PI3‐negative group (n = 13), the red line the PI3‐positive group (n = 13). The cutoff value was calculated using staining percentages of immunostaining, displayed in the figure. p = 0.0408 from the log‐rank test.

We further validated these results using the YUSH and ACRG cohorts. Among the 65 patients who underwent microarray analysis in the YUSH cohort, 49 who received adjuvant chemotherapy were included in the analysis. The correlation between PI3 expression and clinicopathological factors is presented in Supplementary materials and methods, supplementary material, Table S12. In the analysis of TTR, PI3‐positive patients showed a significantly poorer prognosis (5‐year recurrence‐free rate: negative = 0.718, 95% CI = 0.449–0.872; positive = 0.419, 95% CI = 0.247–0.583; p = 0.0151; Figure 5C). Subsequent multivariate analysis of TTR also showed that PI3 was an independent predictor of response (p = 0.01; Figure 5D; supplementary material, Table S13). Notably, chemotherapy in this cohort involved single‐agent 5‐FU therapy or a combination of 5‐FU with CDDP/L‐OHP, doxorubicin, or PTX, but it was not possible to determine the exact regimens administered to each patient [30]. The ACRG cohort analysis included only 82 patients who underwent postoperative chemotherapy with 5‐FU or platinum‐based agents or both and excluded those who did not receive adjuvant chemotherapy and those who received radiation therapy [31]. Correlations between PI3 expression and clinicopathological factors are shown in Supplementary materials and methods, supplementary material, Table S14. TTR analysis revealed a trend toward poor TTR in the PI3‐positive group, although the difference was not significant (5‐year recurrence‐free rate: negative = 0.660, 95% CI = 0.492–0.784; positive = 0.416, 95% CI = 0.248–0.576; p = 0.0567; Figure 5E). In the multivariate analysis of TTR, only PI3 expression was an independent predictor of response (p = 0.029; Figure 5F; supplementary material, Table S15). Although tumor stage was also found to be an independent predictor of response in several cohorts, it was associated with TTR in patients who did not receive adjuvant chemotherapy, suggesting that stage is a poor prognostic factor (supplementary material, Figure S11A–C). In contrast, PI3 was not associated with the prognosis in patients who did not receive chemotherapy, suggesting that PI3 is a better predictor of response than a prognostic factor (supplementary material, Figure S11D–F). The results of these analyses using multiple original and validation cohorts indicate that PI3 expression is a suitable predictor of outcome in response to adjuvant chemotherapy with DNA‐damaging agents rather than a prognostic factor.

Lastly, we established Original Cohort 3, which included 26 patients who received 5‐FU/platinum‐based neoadjuvant chemotherapy, in order to investigate the relationship between treatment and PI3 expression (Table 1; supplementary material, Figure S10C). We first compared the PI3‐positivity rate in Original Cohort 3 with that of the 204 patients in Original Cohorts 1 and 2 who did not receive neoadjuvant chemotherapy (Figure 5G). Although the presence of neoadjuvant chemotherapy did not affect the positive rate, we found a positive correlation between the number of days from neoadjuvant chemotherapy to surgery and the PI3‐positive area rate (p = 0.041; Figure 5H). Moreover, analysis of TTR revealed a significantly worse prognosis in the PI3‐positive group (5‐year recurrence‐free rate: negative = 0.508, 95% CI = 0.214–0.742; positive = 0.154, 95% CI = 0.025–0.388; p = 0.0408; Figure 5I) than that of the PI3‐negative group. Our results indicate that PI3 could also be a useful predictor of the response to neoadjuvant chemotherapy.

Discussion

In this study, we identified PI3, which is commonly upregulated in 5‐FU‐resistant and L‐OHP‐resistant GCOs, as a candidate marker for multidrug resistance. In addition, forced expression of PI3 was found to associate with nucleic acid and ribosome biosynthesis in GCOs. PI3 was linked to the acquisition of resistance to 5‐FU, L‐OHP, and CDDP but did not affect resistance to PTX. This resistance development occurred exclusively in organoids and cell lines cultured in organoid medium. Using our original cohort, we further demonstrated that PI3 expression is an independent prognostic factor and useful predictor of preoperative/postoperative treatment response, which was verified using in silico analyses using publicly available datasets. Our finding that PI3OE is associated with resistance to DNA‐damaging agents aligns with a study in ovarian cancer, where PI3 expression correlated with a poor response specifically to genotoxic drugs [27]. Here, we also provide new insights into the mechanisms of resistance.

We revealed that forced expression of PI3 was associated with several biological functions, such as nucleobase biosynthesis, ribosome biogenesis, RNA metabolism, and ribosomal RNA (rRNA) metabolism in organoids. By analyzing publicly available gastric cancer scRNA‐seq data, we confirmed the upregulation of these GOs in the PI3‐high cluster. The relationship between ribosomes, various RNA metabolic pathways, and anticancer drugs is widely known. Treatment with 5‐FU has been reported to induce ribosomal stress, and 5‐FU is incorporated into RNA‐influencing pathways related to its modification and processing [32, 33]. CDDP is also known to inhibit rRNA synthesis [34]. Furthermore, L‐OHP‐induced ribosome biogenesis stress is a key cause of cell death, and the relationship between its GO and resistance has been investigated [35]. Although CDDP and L‐OHP do not always show cross‐resistance [36], PI3 is linked to both the rRNA and ribosomal systems, which could explain the cross‐resistance observed in this study. Moreover, 5‐FU, L‐OHP, and CDDP have been shown to induce ribosome biogenesis stress at sufficiently high doses, whereas PTX does not, even at high concentrations [37]. On the basis of these reports, the various GOs found downstream of PI3 in the present study could conceivably contribute to the process of resistance acquisition. Since PI3 has diverse functions [20, 21, 22, 23, 24, 25, 26, 27], it may be involved in the activity of Akt, ECT2, MYC, or other molecules that regulate the expression and activity of ribosomal proteins and RNA metabolism. In terms of the regulatory mechanism of PI3 expression, the promoter region of PI3 contains binding sites for the transcription factors SP1, CCAAT/EBP, and NF‐kB [24]. Importantly, since NF‐kB is activated by genotoxic stress, it may serve as a consistent regulator of PI3 expression in response to DNA damage induced by anticancer drugs [38]. This interaction indicates the complex network of transcriptional regulation that may underlie the response to DNA‐damaging therapeutic agents.

Functional analysis of PI3 using cell lines and organoids revealed that different culture systems, especially media conditions, affect the function of PI3 in cells. This suggests a close relationship between the tumor chemotherapy response and tumor microenvironment (TME). Several reports have described the relationship between the TME and sensitivity to chemotherapy. Significant changes in drug sensitivity have been reported using organoid models, depending on the type of culture medium used [39]. Other studies have reported that classification based on the expression of HGF, FGF7, and p‐SMAD2 in cancer‐associated fibroblasts (CAFs) is useful in the selection of treatment [40]. The possibility that extracellular molecules released from various cells in tumor tissues can influence drug sensitivity has also been demonstrated [41]. As discussed, relationships among in vivo‐like environments, CSCs, and tumor cell features have been gradually elucidated. Uniquely, this study demonstrated that pathways or cell states related to the TME, or medium condition, have a significant impact on the behavior of PI3 in drug resistance. Details of the composition of the organoid medium and the 2D cell line medium are provided to enhance the reproducibility and understanding of our findings (supplementary material, Table S17). To further understand chemotherapy resistance, future studies are required to analyze the generation and alteration of the TME by cancer cell progression and treatment and investigate how these changes affect tumor cells. The TME is directly involved in tumor histology, with direct evidence of an association between R‐spondin‐expressing fibroblasts and gastric cancer histology [42]. However, the gastric cancer microenvironment consists of CAFs as well as epithelial cells, macrophages, T cells, natural killer cells, mast cells, and a variety of other cells. These cells can be further classified into multiple types based on their expression profiles [43]. As the TME is composed of a diverse range of cells and variables, a better understanding of the cellular communities that constitute tumor tissues is essential to examine the interaction of tumor cells with their environment. Additionally, a deeper understanding of the TME and its impact on PI3 function could provide valuable insights into the development of resistance and guide the design of precision medicine.

We constructed three cohorts and performed PI3 immunostaining on 230 patients from our institution. To the best of our knowledge, this is the first study to examine PI3 in the stomach. We found that PI3 expression in the epithelial cells of normal mucosa was limited to areas of intense inflammation. In contrast to normal tissues, approximately 45% of the patients showed >1% PI3 expression in the tumor region. Given that PI3 has an anti‐inflammatory effect that suppresses immune cells, such as neutrophils and dendritic cells [44], it may have a similar effect in the stomach. However, the regulatory mechanism associated with PI3 expression is not yet fully understood, and further studies are needed to clarify its relationship with inflammation, which may underlie the development of gastric cancer.

This study showed that PI3‐positive gastric cancer had worse OS and TTRs in patients who received 5‐FU/platinum‐based chemotherapy. Furthermore, multivariate analysis showed that PI3 expression was a useful independent predictor of response. Taken together with the ex vivo results, these results suggest that PTX regimens, rather than 5‐FU/platinum regimens, may be selected for patients with PI3‐positive tumors. This is consistent with the findings showing the efficacy of PTX as second‐line therapy after first‐line treatment with 5‐FU or platinum [2, 45]. Furthermore, although PTX/capecitabine did not outperform CDDP/capecitabine in advanced gastric cancer, the former was associated with fewer toxicities and treatment‐related adverse events [46]. In patients with severe peritoneal metastases of gastric cancer, 5‐FU/leucovorin/PTX shows a longer progression‐free survival and acceptable toxicity than 5‐FU/leucovorin [47]. Moreover, XELOX + PTX is effective in patients with gastric cancer with peritoneal metastases [48]. Optimization of treatment regimens and enhancement of response rates may be possible by stratifying patients on the basis of PI3 expression before chemotherapy. Prospective clinical trials are warranted to validate the predictive value of PI3 expression in gastric cancer.

Our study had some important limitations. First, the use of organoids and cell lines as models may not fully capture the complexities of drug resistance observed in vivo. Further studies using clinical samples, animal models, or both are required to confirm our results. Unfortunately, the PI3‐overexpressing organoids did not engraft as anticipated, limiting our ability to perform animal experiments. We are actively exploring alternative in vivo models and methods to overcome these challenges in future studies. Second, considering the limited number of DNA‐damaging agents used in this study, our findings may not be generalized to other DNA‐damaging agents. Future investigations should explore the impact of PI3 in the context of additional DNA‐damaging agents. Third, although our results were validated in a public cohort to confirm consistency, our cohort was a single‐center, retrospective study with a limited number of patients. We could not perform a multivariate analysis of the association between neoadjuvant chemotherapy and PI3 because only 26 patients were included in our cohort. Multicenter studies and prospective clinical trials with larger numbers of patients are necessary to investigate the ability of PI3 expression to predict the efficacy of adjuvant chemotherapy in gastric cancer.

In conclusion, we demonstrated that PI3 contributes to the acquisition of resistance to 5‐FU and platinum drugs in in vivo‐like environments and identified GOs related to the ribosome and RNA metabolism as novel PI3 downstream targets. Ex vivo studies also revealed that PI3 expression did not contribute to changes in PTX sensitivity. Furthermore, PI3 expression has the potential to predict the efficacy of adjuvant chemotherapy using 5‐FU/platinum‐based agents. Our results provide new insights into multidrug resistance mechanisms and highlight PI3 as a novel tool for the precise stratification of patients with gastric cancer.

Author contributions statement

KH and NS conceptualized the study. KH, YY and SU analyzed the data. KH, TK, YN, RK, HH, SU and RMa conducted the investigation. KH, TK, YN, RMo, YY, SS and MK curated the data. KH, NS, GI and WY acquired funding. KT, HO, KS, TK, GI and WY provided resources. NS, GI, AO and SI supervised the study. KH wrote the original manuscript. NS, TK, YY, SU, RMa, KS, TK, GI and SI reviewed and edited the manuscript. All authors read and approved the final manuscript.

Supporting information

Supplementary materials and methods

Figure S1. Representative image of morphology examined by phase‐contrast microscopy

Figure S2. Data related to Figure 1

Figure S3. Size‐based Simple Western system analysis detecting PI3

Figure S4. Data related to Figure 2

Figure S5. Single‐cell RNA‐seq analysis of gastric cancer cells

Figure S6. Drug sensitivity tests using gastric cancer cell lines

Figure S7. Drug sensitivity tests using GCOs

Figure S8. Drug sensitivity tests using modified gastric cancer cell lines

Figure S9. Representative H&E staining and immunohistochemistry for PI3

Figure S10. Distribution of PI3‐positive area rate by TNM classification and stages

Figure S11. TTR analysis of cases without adjuvant chemotherapy

Table S1. Clinical information of GCOs

Table S2. List of protein‐coding genes upregulated in 5‐FU‐resistant and L‐OHP‐resistant GCOs

Table S3. List of enriched KEGG pathways in Ctrl organoids compared to Ctrl cell lines (p < 0.01)

Table S4. List of GO terms upregulated and downregulated in PI3OE GCOs compared to Ctrl GCOs (p < 0.01)

Table S5. List of GO terms upregulated and downregulated in PI3OE cell lines compared to Ctrl cell lines (p < 0.01)

Table S6. Upregulated genes in PI3OE GCOs from RNA‐seq (p < 0.05)

Table S7. Downregulated genes in PI3OE GCOs from RNA‐seq (p < 0.05)

Table S8. Upregulated genes in PI3OE cell lines from RNA‐seq (p < 0.05)

Table S9. Downregulated genes in PI3OE cell lines from RNA‐seq (p < 0.05)

Table S10. GO terms enriched in PI3‐positive clusters revealed by single‐cell RNA‐seq analysis (p < 0.01)

Table S11. Multivariate Cox regression analyses of factors influencing time to recurrence in Original Cohort‐2

Table S12. Clinicopathological features of YUSH cohort

Table S13. Multivariate Cox regression analyses of factors influencing time to recurrence in YUSH cohort

Table S14. Clinicopathological features of ACRG cohort

Table S15. Multivariate Cox regression analyses of factors influencing time to recurrence in ACRG cohort

Table S16. List of primers

Table S17. Composition of medium

Data availability statement

Primary microarray data are accessible in NCBI's Gene Expression Omnibus (GEO; https://www.ncbi.nlm.nih.gov/geo/) under the accession number GSE280740. Additionally, RNA‐seq data for gastric cancer cell lines and GCOs can be found under the accession number GSE280755. Our previous microarray data are available at GEO under the accession number GSE154127 [17]. The GSE183904 dataset was used for single‐cell RNA sequencing (scRNA‐seq) data [43]. The GSE13861 dataset from the YUSH cohort [30, 49] and the GSE62254 dataset from the ACRG cohort [31], which contain all patient information, gene expression data, and chemotherapy history for gastric cancer, were also used in this study. Gene expression data and clinical information were downloaded from the GEO and the corresponding articles. The raw data generated or analyzed in this study are available upon reasonable request from the corresponding author.