Landscape of splicing factors in early-onset gastric cancer reveals SRSF1 as a key driver of oxaliplatin resistance

Abstract

The incidence of early-onset gastric cancer (EOGC) is increasing. While RNA alternative splicing critically regulates cancer progression, and abnormal changes in splicing factors (SFs) can affect alternative splicing regulation, their roles in EOGC remain unclear. Using multi-omics approaches, we explored the expression and regulatory patterns of 75 SFs in EOGC and further analyzed the differences associated with different regulatory patterns. We investigated the role of serine/arginine-rich splicing factor 1 (SRSF1) in regulating oxaliplatin (OXA) resistance and malignant phenotypes in EOGC. The results showed that the expression levels of most SFs in the EOGC samples were significantly upregulated, while the somatic mutation rate of SFs was low. Based on the expression of SFs, the EOGC population can be stably divided into three splicing regulatory patterns, which differ in immune function, tumor mutational burden, and the anticipated response to chemotherapy drugs. Overexpressing SRSF1 confers OXA resistance to EOGC cells, promotes colony formation, and inhibits apoptosis, and it could promote exon skipping in downstream genes, thereby altering tumor-related functions. This study reveals the expression landscape of SFs in EOGC and highlights the disparities in biological functions across various splicing regulatory patterns. SRSF1 could be a potential therapeutic target and biomarker for overcoming OXA resistance in EOGC.

In the past few years, the worldwide prevalence of gastric cancer (GC) (1). However, it is a notable concern that the prevalence of early-onset gastric cancer (EOGC, GC with an onset age of ≤ 50 years) is continuously rising (2). The latest epidemiological data show that in 2019, EOGC constituted more than 30% of newly diagnosed GC cases in the United States (3), while in China, the proportion was approximately 11% during the same period (4). Compared to the late-onset gastric cancer (LOGC, GC with an onset age of > 50 years), EOGC presents more unfavorable clinicopathological features of EOGC. For example, it is predominantly characterized by poorly differentiated or undifferentiated adenocarcinoma and a diffuse Lauren classification. These features lead to a poorer prognosis and often result in chemotherapy resistance, significantly complicating the treatment process (5). Thus, comprehensive investigation into the basic biological attributes of EOGC and the identification of treatment targets are critically essential.

Accumulating evidence has shown that abnormal gene splicing creates a microenvironment conducive to tumor initiation, progression, and treatment resistance (6, 7). For instance, NOTCH2 and FLT3 genes' splice variants make regulatory proteins that mediate resistance to targeted inhibitors by modulating key downstream signaling pathways (8). The splicing factor SNRPA mediates alternative splicing of ERCC1, leading to cisplatin resistance and enhanced DNA damage repair in lung adenocarcinoma cells (9). In GC, aberrant splicing can regulate oncogenic transcription factors and apoptotic pathways, contributing to the development of a chemotherapy-resistant phenotype (10, 11). Given the unfavorable clinicopathological features of EOGC, it is reasonable to speculate that the biological alterations mediated by splicing regulation may be more extensive.

Constitutive splicing (CS) and alternative splicing (AS) are the two main types into which splicing may be generally divided. In the process of CS, introns were removed from pre-mRNA and exons were precisely ligated to produce mature mRNA, which mainly encodes proteins essential for basic cellular functions. In contrast, AS generates multiple mature mRNAs from a single pre-mRNA through various splicing mechanisms. These mRNAs are then translated into proteins with diverse structures and functions, playing a crucial role in organism development and environmental adaptation (12). Splicing factors (SFs) are essential in the tumor AS process. They influence the splicing of downstream genes and subsequently regulating gene product networks and multiple tumor-related functions (13). Particularly in solid tumors such as liver cancer, pancreatic cancer, and GC, alterations in the expression levels and copy numbers of SFs are particularly prominent. In recent years, smallmolecule inhibitors targeting splicing factors or their co-proteins have emerged as a new area of focus in cancer treatment research (14, 15).

Thus, this study aims to comprehensively investigate the regulatory patterns of SFs in EOGC and their association with tumor biological characteristics. The objectives are to elucidate the role of SFs in the dysregulation of AS events in EOGC, provide therapeutic approaches that target SFs with a theoretical foundation, and identify new research directions to overcome chemotherapy resistance in EOGC.

Results

The genetic variation landscape of SFs in EOGC

Through an analysis of the expression profile of 75 SFs in normal tissues, EOGC and LOGC samples, we found that the majority of SFs were significantly overexpressed in EOGC compared to the normal group. Notably, genes such as YBX1, SRSF1, HNRNPA2B1, HNRNPK, HNRNPA1, HNRNPU, HNRNPUL1, SFPQ, PTBP1, and SRSF3 showed the most significant upregulation (p < 0.05). Compared to LOGC, a higher proportion of SFs maintained elevated expression levels in EOGC, including differential genes such as HNRNPH2, SRSF4, HNRNPL, HNRNPC, SRSF9, HNRNPK, HNRNPUL1, HNRNPF, HNRNPUL2, and HNRNPM (Fig. 1, A and B). To strengthen the analysis, we assessed additional gene sets using the same dataset, including cancer driver gene sets, the mTOR signaling pathway gene set, and ferroptosis gene sets, to confirm whether the observed upregulation of SFs is specific The results indicated that the expression of these gene sets in EOGC and LOGC tissues failed to exhibit statistically significant differences as pronounced as those observed for SFs; see Fig. S1.

Figure 1.

The expression landscape of 75 SFs in EOGC.A, the expression of SFs differs in EOGC samples compared to normal samples, as shown by the heatmap. B, the heatmap displays the different SFs expression levels in EOGC and late-onset gastric cancer data. C, somatic mutation features of SFs in 27 TCGA-EOGC patient samples, with mutation types displayed in each column of the waterfall plot. D, somatic mutation types in EOGC samples. E, types of base substitutions in EOGC samples. F, top mutated genes in EOGC samples. G, the frequency of copy number variation in SFs, (∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001). SF, splicing factor; EOGC, early-onset gastric cancer.

These findings suggest that alternative splicing regulation may have a more pronounced biological impact on EOGC. Subsequently, by comparing EOGC with normal samples and applying the screening conditions (FDR < 0.05, |log₂(fold change)| > 2), we ultimately identified 35 SFs with specific high expression in EOGC, including PTBP1, HNRNPDL, HNRNPA3, FUS, SFPQ, YBX1, HNRNPD, SRSF2, HNRNPA0, HNRNPAB, HNRNPF, RBM25, SRSF1, SRSF3, SYNCRIP, TRA2B, HNRNPK, PCBP2, HNRNPA1, SF3B1, HNRNPC, HNRNPR, SF1, HNRNPA2B1, HNRNPH1, SRRM1, MBNL1, KHDRBS1, DAZAP1, KHSRP, HNRNPM, RBMX, HNRNPL, UPF1, and HNRNPU (Marked with a red "star" in Fig. 1A).

Somatic mutation analysis revealed that SFs in EOGC had a relatively lower mutational burden, predominantly missense mutations. The base changes primarily involved ‘C’ to ‘T’ transitions (Fig. 1, C–E). The three genes with the highest mutation frequencies were TTN (26%), SPTA1 (22%), and TP53 (37%) (Fig. 1F). Copy number variation analysis showed an increased frequency of copy number gains in genes such as SRSF1, HNRNPK, HNRNPU, PCBP2, SFPQ, SRSF2, TRA2B, and YBX1, which could be connected to the relevant mRNA levels being upregulated (Fig. 1G). Creating this expression profile provides valuable insights into the regulatory network of SFs during the initiation and progression of EOGC. To visualize the genomic landscape of these splicing factors, we mapped their chromosomal localizations using a circus plot (Fig. S2). The results demonstrate a widespread distribution of key SFs across the human genome, with several critical regulatory factors—such as SRSF1 and SRSF2 on chromosome 17, and the HNRNP family members across multiple chromosomes—localized to specific genomic hotspots, suggesting their broad potential for systemic transcriptional impact.

We further analyzed the relationship between 35 SFs and chemotherapy response. We compiled the clinicopathological data of 30 EOGC patients; including information on 5 complete response (CR) and 3 progressive disease (PD) cases (Table S1). We performed a heatmap analysis to visualize the SF expression profiles in these 8 patients. Notably, we observed that the expression levels of several SFs were markedly elevated in patients with PD compared to those with CR. Similarly, SRSF1 is one of the relatively significant SFs (Fig. S3).

Splicing patterns regulated by SFs in EOGC

Next, based on the expression of the 35 SFs, we proceeded with an unsupervised clustering analysis. The Figure 2A represents the delta area plot from the Consensus-Cluster-Plus analysis, which depicts the relative change in the area under the cumulative distribution function curve for different cluster numbers (k). The peak of the Delta Area curve at k = 3 indicates that the clustering stability reached its optimal point at this value, and further increasing k resulted in a sharp decline in the relative gain of consensus. This justifies the selection of three clusters for the subsequent analysis. Cluster 2 exhibited the lowest levels of SF expression, while Cluster 3 exhibited the highest levels. Cluster 1's overall expression levels were midway between that of Clusters 2 and 3 (Fig. 2C). Cluster 1 patients exhibited the greatest frequency of CDH1 mutations, as established by somatic mutation analysis (Fig. 2D); CDH1 is the gene encoding epithelial-cadherin protein, which is widely recognized as a primary genetic marker for hereditary GC. Germline mutations in the CDH1 gene can be detected in the peripheral blood of about 30%–50% of families with diffuse GC (16). Patients in Cluster 2 exhibited higher mutation frequencies of MAP2K7 and TGFBR1 (Fig. 2E). Previous studies have shown that inactivating mutations in MAP2K7 inhibit JNK-mediated pro-apoptotic functions, while mutations in TGFBR1 lead to the transformation of the TGF-β signaling pathway from tumor suppression to metastasis promotion. Both mutations synergistically promote epithelial-mesenchymal transition and immune evasion (17, 18). Many patients in Cluster 3 had multiple mutated genes, potentially related to the abnormal activation and expression of SFs, with the highest mutation rate observed in the classical TP53 gene (Fig. 2F). TP53 mutations can significantly reduce the efficacy of various drugs by disrupting the DNA damage response, enhancing pro-survival signaling, and causing genomic instability (19). To further analyze the molecular signatures of the identified clusters, we compared the expression levels of candidate SFs across the three clusters. As shown in Figure 2G, Cluster 3 exhibited a pervasive and significant upregulation of nearly all analyzed splicing factors compared to Clusters 1 and 2 (p < 0.001). This distinct 'SF-high' profile suggests that the dysregulation of the splicing machinery is a predominant feature of the Cluster 3 subtype, and further analysis of its specific functional pathway characteristics is needed.

Figure 2.

The regulatory patterns of SFs in EOGC.A, patients with EOGC were stratified into three distinct clusters utilizing a consensus clustering algorithm based on the expression of the 35 SFs. B, validation of cluster segregation efficacy was performed through t-Stochastic Neighbour Embedding analysis. C, differential expression patterns of SFs across the identified clusters. D, somatic mutations in Cluster 1. E, somatic mutations in Cluster 2. F, somatic mutations in Cluster 3. G, expression differences of SFs in the three clusters, (∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001). SF, splicing factor; EOGC, early-onset gastric cancer.

Enrichment signatures across splicing regulatory patterns

To more thoroughly examine the biological distinctions between a variety of splicing regulatory patterns, we used the Gene set variation analysis (GSVA) algorithm to assess the enrichment differences in cancer-related hallmark gene sets and Kyoto Encyclopedia of Genes and Genome (KEGG) gene sets associated with oncogenesis (Fig. 3, A and B). In the analysis of cancer-related hallmark gene sets, we found that proliferation-related gene sets in Cluster 3 were significantly more upregulated than those in Clusters 1 and 2. For example, DNA repair, E2F targets, G2/M checkpoint, and MYC targets showed increased activity, suggesting that individuals in Cluster 3 may have stronger drug resistance (Fig. 3A).

Figure 3.

Distinct splicing regulatory patterns and their signaling pathways.A, comparative analysis of enrichment scores for cancer hallmark gene sets associated with tumor progression across Cluster 1, 2, and 3. B, assessment of enrichment score variations in Kyoto Encyclopedia of Genes and Genome pathways linked to cancer development among Cluster 1, 2, and 3. C, the Upset plot of AS events in EOGC (AS events include Mutually exclusive exons, Retained intron,Alternate Donor site, Alternate acceptor site, Alternate terminator, Alternate promoter, Exon skipping. D, the Gene Ontology functional enrichment analysis of differential AS events in EOGC. E, the Kyoto Encyclopedia of Genes and Genome enrichment analysis of differential AS events in EOGC. AS, alternative splicing.

Further KEGG gene sets related to cancer development enrichment analysis revealed that genes in Cluster 3 were primarily enriched in drug metabolism pathways (such as drug metabolism enzymes and drug metabolism—cytochrome P450) and proliferation-related pathways (such as DNA repair and the cell cycle). In contrast, genes in Cluster 1 were mainly concentrated in immune response pathways, such as antigen processing and presentation, and cytokine receptor interaction (Fig. 3B).

Additionally, we investigated differential AS events in EOGC samples and found that these events predominantly involved exon skip (ES)and AT (Alternate terminator) (Fig. 3C). The results of the analysis of differential AS events under different splicing regulatory patterns are shown in Table S2 Subsequently, using selection criteria of log₂(fold change) > 2 and p < 0.05, GO and KEGG functional enrichment analyses showed that the differential AS events (EOGC vs. adjacent cancer tissue) were associated with cell proliferation, cell cycle, and apoptotic activities (Fig. 3, D and E). These results further indicate that EOGC incidence and development are strongly linked to AS.

Immune trait variations across splicing regulation patterns

Different splicing regulatory patterns showed variances in immune-related pathways in the preceding GSVA analysis. Hence, we delved deeper into the splicing regulatory patterns in relation to immune-related checkpoints, immunological functional activity, and immune-cell infiltration. Except for Cluster 3, which contained a greater percentage of CD4 memory T cells, no significant variations were observed in the quantity of most immune cells across the three groups (Fig. 4A). Regarding immune function activity (Fig. 4B), Cluster 2 exhibited high activity in aDCs, DCs, type I IFN response, and type II IFN response. Cluster 1 showed higher expression of mast cells. Immune function activity in Cluster 3 was generally weaker than in the other clusters. Analysis of differential immune checkpoint differential expression revealed that CD276, CD160, TNFRSF14, and TNFRSF25 were significantly more expressed in Clusters 1 and 3 compared to Cluster 2 (Fig. 4C). Conversely, the expression of the immunological checkpoint PD1 was heightened in Cluster 2.

Figure 4.

Variations in immune-related characteristics and forecasting treatment responsiveness across distinct splicing regulatory patterns.A, variations in 22 immune cell infiltrations between the three groups. B, immune function activity levels varied amongst the three groupings. C, variations in immunological checkpoint expression across the three clusters. D, prediction of response to anti-CTLA4 in three clusters. E, anti-PD1 response prediction in three clusters. F, cisplatin's IC50 prediction in the three clusters. G, fluorouracil's IC50 prediction in the three clusters, (∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001). EOGC, early-onset gastric cancer.

Splicing regulatory landscape predicts drug sensitivity

Next, we compared the immune treatment response predictions for the three splicing regulatory patterns with classic immune checkpoint targets CTLA4 and PD1. The computational analysis indicated that the predicted response for CTLA4 inhibitors was not significant across the three clusters (Fig. 4D). However, predictions suggested that anti-PD-1 therapy might have potential therapeutic value for Cluster 2 patients (Fig. 4E). By examining global gene expression under various splicing regulatory patterns, we were able to predict the treatment sensitivity of two commonly used GC chemotherapy medications: the platinum-based drug cisplatin and the fluorouracil-based drug 5-fluorouracil. The chemotherapy regimen combining platinum and fluorouracil (such as SOX, FLOT, XELOX, etc.) is the first-line chemotherapy regimen for GC. The results indicated that Cluster 3 exhibited the maximum predicted IC50 of both drugs (Fig. 4, F and G). Among them, the predicted IC50 values for patients in Cluster 3 were the highest, which is consistent with the results of our previous enrichment analysis.

Regulation of SRSF1 expression affects EOGC progression

Previous results indicated that SFs in Cluster 3 may be closely related to drug resistance and disease progression in EOGC patients. We observed significant upregulation of SRSF1, PTBP1, and SRSF3 in Cluster 3, ranking among the top three. We thus carried out further studies to examine the relationship between SRSF1, PTBP1, and SRSF3 expression and the prognosis of EOGC and LOGC patients. The results indicated that only high levels of SRSF1 were significantly associated with a bad outcome in EOGC patients (GSE63354, Log-rank p = 0.025; GSE84437, Log-rank p = 0.018), while no statistical difference was found in LOGC patients (Figs. 5, A and B and S4). Furthermore, SRSF1 displayed a significant frequency of copy number gains in copy number variations. Based on these results, we conducted an in-depth study of SRSF1 to explore its role in EOGC drug resistance and disease progression.

Figure 5.

SRSF1's prognostic characteristics in EOGC and its correlation with the malignant phenotype of EOGC cells.A, the relationship between SRSF1 expression and prognosis in the EOGC population from GSE62254. B, the relationship between SRSF1 expression and prognosis in the EOGC population from GSE84437. C, efficiency of transfection targeting SRSF1 mRNA levels in MKN7 cells. D, efficiency of transfection targeting SRSF1 protein levels in MKN7 cells. E, colony formation assay of SRSF1-KO MKN7 cells. F, colony formation assay of SRSF1-overexpression MKN7 cells. G, FACS analysis of SRSF1-KO MKN7 cells, after staining with PI and anti-Annexin V. H, FACS analysis of SRSF1-overexpression MKN7 cells, after staining with PI and anti-Annexin V. I, efficiency of transfection targeting SRSF1 protein levels in MKN74 cells. J, efficiency of transfection targeting SRSF1 mRNA levels in MKN74 cells. K, CCK8 assay of SRSF1-overexpression MKN74 cells. L, CCK8 assay of SRSF1-KO MKN74 cells. M, colony formation assay of SRSF1-KO and overexpression MKN74 cells, (∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001). Data is presented as mean ± SD from at least three independent biological replicates. Individual data points as independent biological replicates. EOGC, early-onset gastric cancer.

To further confirm the biological role of SRSF1 in EOGC progression, we constructed transfection plasmids for overexpressing and knocking down SRSF1, which were used to infect the EOGC cell line MKN7. After screening, we successfully established cell lines with overexpression of SRSF1 (oeSRSF1) and knockdown of SRSF1 (shSRSF1). As shown in Figure 5, C and D, we verified the transfection efficiency by Western blot and PCR, and selected the two knockdown sites, sh1 and sh2, with higher knockdown efficiency from the three knockdown sites for subsequent studies.

Next, we assessed the impact of SRSF1 knockdown and overexpression on the colony formation of the EOGC cell line MKN7 (Fig. 5, E and F). The results showed that the proliferation rates of the SRSF1 knockdown groups (SRSF1-sh1 and SRSF1-sh2) were 15.24 ± 0.46% and 11.57 ± 0.9%, respectively, significantly lower than the control group (Control-shSRSF1, with a proliferation rate of 36.86 ± 0.92%). In the SRSF1 overexpression group (SRSF1-oe), the proliferation rate was 39.49 ± 0.98%, substantially raised compared to the control group (Control-oeSRSF1, with a proliferation rate of 22.92 ± 1.70%). In order to investigate the regulatory role of SRSF1 in the apoptosis of MKN7 cells, we used flow cytometry to analyze the effects of SRSF1 knockdown (SRSF1-sh1/sh2) and overexpression (SRSF1-oe) on cell apoptosis. The results demonstrated that SRSF1 knockdown significantly increased apoptosis rates: the control group (Control-shSRSF1) showed an apoptosis rate of 9.10 ± 0.50%, while the sh1 and sh2 groups exhibited increased rates of 20.70 ± 0.73% and 25.46 ± 0.51%, respectively (p < 0.001). In the SRSF1 overexpression group, the apoptosis rate decreased from 4.95 ± 0.30% in the control group to 4.23 ± 0.06% (Fig. 5, G and H).

To enhance the reliability of the experiment, we performed validation studies in an additional EOGC cell line, MKN74. SRSF1 knockdown (using the most effective shRNA construct sh2 identified in prior cell lines) and overexpression were performed in MKN74 cells; the corresponding transfection efficiencies are depicted in Figure 5, I and J. CCK-8 and colony formation assays revealed that SRSF1 markedly modulates the proliferation of EOGC cells (Fig. 5, K–M).

SRSF1 can regulate OXA resistance in EOGC

The findings demonstrated that the SRSF1 knockdown groups' IC50 values (SRSF1-sh1 and SRSF1-sh2) were 50.52 ± 0.97 μM and 30.36 ± 1.40 μM, respectively, which were significantly lower than that of the control group (Control-shSRSF1, IC50 = 102.65 ± 1.81 μM), indicating that SRSF1 knockdown significantly increased EOGC cell line MKN7 sensitivity to OXA. Conversely, the IC50 value of the SRSF1 overexpression group (SRSF1-oe) was 209.31 ± 1.21 μM, considerably higher than that of the control group (Control-oeSRSF1, IC50 = 110.12 ± 1.73 μM), see Figure 6, A and B. We observed the same trend in another EOGC MKN74 cell line. The SRSF1-sh2 group exhibited an IC50 value of 24.95 ± 0.96 μM, significantly lower than the control group (Control-shSRSF1, IC50 = 67.80 ± 0.67 μM), demonstrating that SRSF1 knockdown substantially enhanced OXA sensitivity in EOGC MKN74 cells. Conversely, the SRSF1-overexpression group (SRSF1-oe) displayed an IC50 of 169.09 ± 0.64 μM, substantially higher than its Control- (oeSRSF1, IC50 = 70.45 ± 0.81 μM), as shown in Figure 6, C and D. Finally, colony formation assays revalidated that SRSF1 overexpression conferred enhanced resistance to OXA in EOGC MKN7 and MKN74 cells, whereas SRSF1 knockdown heightened their sensitivity, see Figure 6, E and F.

Figure 6.

SRSF1 expression modulates the sensitivity of EOGC cells to OXA.A, determination of oxaliplatin IC50 in SRSF1-overexpression MKN7 cells. B, determination of oxaliplatin IC50 in SRSF1-KO MKN7 cells. C, determination of oxaliplatin IC50 in SRSF1- overexpression MKN74 cells. D, determination of oxaliplatin IC50 in SRSF1-KO MKN74 cells. E, colony formation assay of SRSF1-overexpression and KO MKN7 cells. F, colony formation assay of SRSF1-overexpression and KO MKN74 cells, (∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001). Data is presented as mean ± SD from at least three independent biological replicates. Individual data points as independent biological replicates. EOGC, early-onset gastric cancer; OXA, oxaliplatin.

Downstream AS events regulated by SRSF1 and correlation analysis

Since SRSF1 is an essential splicing factor in EOGC, we performed RNA sequencing using SRSF1-overexpressing and WT MKN7 cells to identify SRSF1-regulated AS types (Table S3). We found that SRSF1 overexpression led to changes in 1203 splicing events, including 942 ES events, 105 alternative 5′ splice site (A5SS), 98 alternative 3’ (acceptor) splice site (A3SS), 88 mutually exclusive exon (ME), and 70 RI (Fig. 7A). Furthermore, based on the flowchart shown in Figure 7B (The relevant data are in Table S3), we narrowed down the range of differential AS genes that are likely to interact with and be regulated by SRSF1 (including CLSTN1, PDGFA, ENAH, RAI14, ADAM15, CD47, PFKM, ABLIM1, ANO1, AFMID, PRMT7, PHF8, GKAP1, ACP5, STK39, MAZ, FBLN2, TBRG1). To further verify the regulatory relationship between SRSF1 and its potential downstream AS targets, we performed Real-time polymerase chain reaction (RT-PCR) to detect the relationship between SRSF1 expression and different AS events in 10 EOGC tissues and their corresponding adjacent tissues. Our analysis revealed that the percent splice-in index [Splice-In Reads/(Splice-In Reads + Splice-Out Reads)] of RAI14 (p = 0.0172, r = −0.5137) and PDGFA (p = 0.0069, r = −0.5706) exhibited the most robust inverse correlations with SRSF1 mRNA levels (Fig. 7, C and E). This significant negative correlation indicates that higher levels of SRSF1 are associated with increased ES of RAI14 and PDGFA. However, no significant correlations were detected between SRSF1 expression and the splicing patterns of ENAH, ADAM15, CLSTN1, PFKM, CD47, or TBRG1 (p > 0.05; Fig. 7, D, and F–J). These observations imply that the splicing of these candidates may be subject to multi-factorial regulation by additional splicing factors or modulated by the tumor microenvironment. Finally, validation of four splicing events by agarose gel electrophoresis showed that SRSF1 overexpression promotes exon 11 skipping in RAI14 (Fig. 7, K–N). Additionally, we investigated whether SRSF1 regulates the alternative splicing of ERCC1, given that ERCC1 has been previously validated to be associated with platinum resistance (9). However, our results indicate that SRSF1 likely does not mediate the splicing of ERCC1 isoforms (Fig. S5). We further analyzed the splicing of RAI14 and PDGFA using five pairs of EOGC tissues and matched normal adjacent tissues (NAT). The results showed that RAI14-S was significantly higher in EOGC tissues than in NAT (p < 0.05). However, the PDGFA-S showed no significant difference between EOGC and NAT (Fig. 7, O and P).

Figure 7.

Differential AS events regulated by SRSF1 and correlation analysis.A, RNA-seq analysis to investigate the AS events affected by SRSF1. B, flowchart outlines the strategy for selecting candidate target AS genes of SRSF1. C–J, correlation analysis of SRSF1 with RAI14-PSI, ENAH-PSI, CLSTN1-PSI, PFKM-PSI, PDGFA-PSI, ADAM14-PSI, CD47-PSI, and TBRG1-PSI. K–N, representative SRSF1-affected exon inclusion events with RNA-seq reads coverage, RT-PCR results and quantification of their RNA products measured as percent splicing index. Note that alternative exons for SRSF1-mediated exon skips are marked in blue. O, comparison of RAI14 exon 11 skipping between EOGC and normal adjacent tissue. P, comparison of PDGFA exon 6 skipping between EOGC and normal adjacent tissue, (∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001). Data is presented as mean ± SD from at least three independent biological replicates. Individual data points as independent biological replicates. AS, alternative splicing; EOGC, early-onset gastric cancer.

SRSF1-related splicing variants RAI14-S promotes tumorigenic potential of EOGC cells

We next investigated the functional impact of overexpressing distinct splicing variants of the SRSF1-associated genes RAI14 and PDGFA. The transfection efficiencies of the RAI14-L/S and PDGFA-L/S constructs of MKN74 were validated by the results presented in Figure 8, A and B. The CCK-8 and colony formation assays demonstrated that RAI14-S significantly promoted EOGC cell proliferation, whereas PDGFA-S had no significant effect (Fig. 8, C–F). IC50 assays revealed that the IC50 value was significantly higher in the oe-RAI14-S group (97.21 ± 0.86 μM) compared to the oe-NC and oe-RAI14-L groups. A more modest increase was observed in the oe-PDGFA-S group (87.32 ± 1.21 μM) (Fig. 8, G and H). Colony formation assays confirmed that RAI14-S enhanced resistance to OXA in MKN74 cells. In contrast, no statistically significant differences were observed among the PDGFA-S, PDGFA-L and control groups (Fig. 8, I and J).

Figure 8.

Functional phenotype of SRSF1 related splicing variants RAI14 and PDGFA in EOGC.A, efficiency of transfection targeting variants RAI14 RNA levels in MKN74 cells. B, efficiency of transfection targeting variants PDGFA RNA levels in MKN74 cells. C, CCK8 assay of variants RAI14-oe MKN74 cells. D, CCK8 assay of variants PDGFA-oe MKN74 cells. E, colony formation assay of variants RAI14-oe MKN74 cells. F, colony formation assay of variants PDGFA-oe MKN74 cells. G, determination of oxaliplatin IC50 in variants RAI14-oe MKN74 cells. H, determination of oxaliplatin IC50 in variants PDGFA-oe MKN74 cells. I, colony formation assay of control and variants RAI14-L/S+OXA J, colony formation assay of control and variants PDGFA-L/S+OXA, (∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001). Data is presented as mean ± SD from at least three independent biological replicates. Individual data points as independent biological replicates. EOGC, early-onset gastric cancer; OXA, oxaliplatin.

Discussion

The trend of younger-onset cancer is becoming more prominent. Globally, new cancer cases in individuals under 50 have increased by 79% over the past 30 years, and cancer-related deaths have risen by 27.7% (20). GC is no exception, with the number of EOGC cases rising annually (2). The reasons behind this trend remain unclear. In recent years, the role of gene aberrant splicing in tumor onset and development has drawn the attention of scientists. Research indicates that, compared to adjacent normal tissue, most tumors display extensive splicing abnormalities, which may result in the inactivation of tumor suppressor genes or increased expression of oncogenes (21, 22). During these aberrant splicing events, mutations or expression alterations in SFs play a critical role (23, 24). However, there is still a lack of understanding regarding the overall expression of SFs in EOGC and their roles in genomic mutations, immune suppression, metabolic reprogramming, and other aspects.

Thus, we investigated the expression characteristics and regulatory patterns of SFs retrievable from two major RNA-binding protein databases in EOGC. The somatic mutation rate of SFs in EOGC patients was determined to be low, which is consistent with previous research (25). However, the majority of SFs were upregulated in the EOGC samples, which showed notable changes in the SFs transcriptome when compared to the normal and LOGC samples. This suggests that SFs promote the development of EOGC by regulating the splicing process. Subsequently, we categorized the EOGC population according to the expression of SFs utilizing unsupervised clustering technique. Tumor categorization is a technique employed to differentiate the aggressiveness and heterogeneity of tumors. Currently, the molecular subtypes commonly used in GC are the TCGA classification (26) and the Asian Cancer Research Group (ACRG) classification (27). However, there is limited research on EOGC, and whether these two subtypes apply to EOGC is still unknown. In this study, we identified three splicing regulatory patterns in different EOGC patients, which displayed significant differences in their biological processes.

This study discovered that the aberrant expression of SFs across distinct clusters correlated with somatic mutation profiles, suggesting that SFs may contribute to the accumulation of somatic mutations by modulating genome stability and post-transcriptional modifications. In Cluster 1, the high-frequency mutation of CDH1 was potentially linked to SF-mediated RNA splicing dysregulation: it is established that the splicing factor HNRNPA1 regulates alternative splicing of CDH1 pre-mRNA, ultimately reducing E-cadherin protein stability and promoting tumor invasion (28). In Cluster 2, low expression of SFs could induce aberrant splicing of MAP2K7 pre-mRNA, yielding truncated proteins that inactivate JNK signaling and block the apoptotic pathway (29). Regarding the prevalent TP53 mutations in Cluster 3, dysregulated SF expression (e.g., RBM10 inactivation) could directly impair p53-dependent DNA damage repair mechanisms: SFs such as RBM10 participate in the negative feedback regulation of the p53 signaling pathway by modulating TP53 transcript stability and the splicing of upstream regulators like MDM2 (30), and loss of their function would exacerbate genetic instability, leading to chemotherapy resistance and enrichment of TP53 mutations.

In the expression pattern characterized by generally low levels of SFs, multiple oncogenic signal transduction pathways were activated, including the KRAS and WNT/β-catenin pathways. In Cluster 3, where SFs are generally highly expressed, we noticed dramatically increased cell proliferation signals. Meanwhile, both pathways connected to the cell cycle and drug metabolism were highly active, which also may indicate that disease progression and chemotherapy resistance are more pronounced in this population. It is widely recognized that SFs govern transcription, and their higher expression and proliferation-inducing capabilities are mutually reinforcing. Due to their strong cell viability, DNA damage repair-related signaling pathways are fully activated by this splicing regulation mechanism, guaranteeing that genetic information is expressed normally (31).

Furthermore, we found that when SFs expression levels were between the two patterns, there was a considerable increase in the infiltration of inflammatory immune cells such as B cells, M2 macrophages, and mast cells, along with a significant increase in immune function scores. This indicates chronic inflammatory development and immunosuppression in the tumor microenvironment, pertaining to many pro-inflammatory and immune-associated signaling pathways, such as B cell receptor, chemokine, and Toll-like receptor signaling. Establishing splicing regulatory patterns can also more clearly define the clinical condition and features of the tumor microenvironment in EOGC patients, thereby guiding the formulation and evaluation of personalized treatment plans for EOGC patients. Additionally, we predicted the treatment sensitivity under different splicing regulatory patterns. Chemotherapy resistance to platinum and fluoropyrimidine medications is more likely to develop in patients with splicing regulation patterns that are typified by usually high expression of SFs. Conversely, patients with low SF expression splicing regulatory patterns exhibit a greater response to anti-PD-1 therapy.

In the splicing regulation pattern analysis, we found that Cluster 3 was unusually active in drug metabolism and cell proliferation-related pathways, and is likely more susceptible to resistance to platinum and fluorouracil-based chemotherapy. Further analysis of the significantly expressed SFs in this cluster revealed that only high SRSF1 expression was linked to a poor outcome in EOGC patients, while its expression had no relation to the prognosis of LOGC. Based on these findings, we conducted further in vitro experiments, which demonstrated that either knockdown or overexpression of SRSF1 could influence the sensitivity of EOGC cells to OXA and significantly alter the proliferation and apoptosis phenotypes of EOGC cells.

Cancer-causing SRSF1 can manage the splicing of important apoptosis pathway proteins such RIPK1, BAX, and Bcl-2, producing isoforms with no pro-apoptotic properties (32). Furthermore, SRSF1 overexpression can prevent breast cancer cells from undergoing apoptosis and cause tumor development in epithelial cells (33). Research indicates that SRSF1 expression correlates with mTORC1 activation (34) and participates in clonal selection signaling pathways during the progression of acute myeloid leukemia and the relapse of minimal residual disease (35). However, the role of SRSF1 in mediating biological functions and OXA resistance in EOGC remains undefined. Our phenotypic assays demonstrate that SRSF1 overexpression or knockdown significantly enhanced or suppressed proliferation and apoptosis in EOGC cell lines (MKN7 and MKN74). Crucially, modulation of SRSF1 expression altered OXA sensitivity to EOGC cells, which contributes to filling this existing research gap. As the role of ERCC1 alternative splicing in cisplatin resistance has been proven, we also wanted to explore whether there is a connection between SRSF1 and ERCC1 alternative splicing. However, our results indicate that SRSF1-mediated regulatory networks likely do not encompass ERCC1 isoform switching in this context. This suggests that SRSF1 may participate in the alternative splicing of other downstream genes involved in DNA repair or programmed cell death (Fig. S5).

This study has certain limitations. Firstly, as this study primarily relies on public databases for analysis, the number of samples available for EOGC is relatively small. Therefore, it is necessary to utilize a larger sample size and clinical cohorts to conduct a reliable prognostic analysis and to further validate the distinctive features of splicing regulatory patterns. Secondly, we have only investigated the association between SRSF1 and EOGC cells in OXA resistance and their malignant phenotypes. In the future, we will further expand the study to examine the characteristics of more SFs and investigate SFs and their downstream splicing regulatory mechanisms in more depth. Furthermore, this study has not fully covered all the potential roles and impacts of SFs in EOGC, especially in immune evasion, tumor metastasis, and other aspects. Nevertheless, using multi-omics data, this study analyzed the expression patterns of common SFs in EOGC. It revealed significant differences in splicing regulatory patterns based on SF expression across various biological processes and immunological characteristics. Recognizing these patterns offers a prospective reference point for assessing the tumor microenvironment and therapeutic pharmacotherapy in EOGC patients. Meanwhile, SRSF1 may potentially become a therapeutic target to overcome OXA resistance. This discovery provides a new perspective on the treatment of EOGC. However, further confirmation of the validity and accuracy of these findings still requires larger clinical data and experiments in subsequent studies.

This study elucidates the expression profiles of various splicing factors in EOGC and investigates the disparities in biological processes among different splicing regulation patterns. These findings offer valuable insights into developing personalized treatment strategies tailored to different EOGC patients. Additionally, regulation of SRSF1 expression can either enhance or diminish the sensitivity of EOGC cells to OXA and influence EOGC malignant phenotypes. SRSF1 could therefore be a promising biomarker and therapeutic target for overcoming oxaliplatin resistance in EOGC.

Experimental procedures

Data collection, processing and tumor samples

This research methodically explored the molecular features of the EOGC utilizing multi-omics data. A total of 30 EOGC samples were obtained from the The Cancer Genome Atlas (TCGA, https://portal.gdc.cancer.gov/). These samples were selected from an initial cohort of 443 TCGA-STAD patients based on the following inclusion criteria: (1) histological diagnosis of adenocarcinoma, and (2) age ≥18 and < 50 years 59 healthy control samples from the Genotype-Tissue Expression (https://www.gtexportal.org/GTEx) database were randomly selected using a reproducible R sampling algorithm after setting a predefined random seed (value: 123,456). Alternative splicing data for TCGA-EOGC were obtained from the TCGA SpliceSeq database (https://bioinformatics.mdanderson.org/TCGASpliceSeq/), which included the percentage of seven types of splicing events [ES, ME, intron retained alternative promoter (AP), AT, alternative donor site (AD)/A5SS, and alternative acceptor site/A3SS], ranging from 0 to 100%. Additionally, the data on copy number variations and somatic mutations for EOGC patients were also obtained from the TCGA database. After being standardised using transcripts per million, the RNA-seq data underwent log₂ transformation. Gene Expression Omnibus (GEO) data (https://www.ncbi.nlm.nih.gov/geo/) was used to obtain microarray expression profiles and related clinical prediction information for the GSE84437 and GSE62254 datasets. The study examined a set of 75 SFs obtained from the SpliceAid 2 and RBPmap databases (36, 37).

Unsupervised clustering of splicing factors

To gain more insight into the SFs' splicing regulatory processes in EOGC, we utilized a consensus clustering approach for unsupervised clustering built around the SFs' expression. The unsupervised consensus clustering was performed using the Consensus-Cluster-Plus R package (v1.68.0). We utilized the k-means algorithm with 1000 iterations and a subsampling ratio of 0.8 to ensure clustering stability. The t-Stochastic Neighbour Embedding method was employed to further validate the dependability of the recognized splicing regulation patterns according to SFs.

Gene set variation analysis (GSVA)

The 'GSVA' package was utilized to perform GSVA enrichment calculations to investigate the variations in biological processes across different splicing regulation patterns. To estimate the activity levels of pathways and biological processes associated with gene sets in the samples, this unsupervised, non-parametric method computes enrichment scores. The gene set “h.all.v2023.2.Hs.symbols” was obtained from the MSigDB database (https://www.gsea-msigdb.org/gsea/msigdb/). By using the KEGG analysis and the 'Cluster Profiler' package, we were able to determine the signaling pathways that are associated with differential genes. Gene Ontology was employed to annotate the biological functions of AS events in EOGC, which were visualized using the 'upset' program. The Gene Ontology enrichment analysis was conducted using the ClusterProfiler R package (v4.4.0). The gene functional annotations were based on the org.Hs.e.g.., db human database. To control the false discovery rate, p-values were adjusted using the Benjamini-Hochberg (BH) method, and terms with an adjusted p < 0.05 were considered significantly enriched.

Prediction and evaluation of drug response

For each cluster, the 'pRRophetic' package predicted the half-maximal inhibitory concentration (IC50) of commonly used chemotherapy drugs. This package employs a Ridge Regression algorithm, establishing a gene expression-drug response model through standardized training sets (72-h drug exposure with CCK8 assay). IC50, defined as the drug concentration required inhibiting 50% of the target's activity, is an important criterion for early-stage drug efficacy evaluation. To evaluate the response of various splicing regulatory patterns to immune checkpoint inhibitors that target anti-PD-1 and anti-CTLA4, we applied SubMap (https://cloud.genepattern.org/gp/pages/index.jsf).

Analysis of immune cell infiltration inside the tumor micro-environment

The CIBERSORT algorithm can be used to quantitatively analyze the makeup of immune cells using transcriptome data, which is a tool for cell type identification. We used the 'CIBERSORT' package to analyze immune cell infiltration under different splicing regulatory patterns (including naive B cells, memory B cells, plasma cells, CD8+ T cells, naive CD4+ T cells, resting CD4+ memory T cells, activated CD4+ memory T cells, and 22 other types of immune cells), immune functional scores, and the expression characteristics of immune checkpoints (including CD40, CD44, PD1, CTLA4, and 45 other types of immune checkpoints). The CIBERSORT algorithm using the LM22 signature matrix. The deconvolution was based on nu-Support Vector Regression (nu-SVR) using the e1071 R package. We performed quantile normalization (via the preprocessCore R package) and ensured the input mixture matrix was in non-log space (with automatic anti-log transformation applied if max values were < 50).

Cell culture

The EOGC cell line MKN7 and MKN74 was acquired from the Cell Bank of the Chinese Academy of Sciences. EOGC cell lines MKN7 and MKN74 were authenticated by Short Tandem Repeat profiling at BIOWING. The MKN7 and MKN74 cell line were grown in RPMI1640 media (Servicebio) placed in a controlled environment with 5% CO2 and 10% fetal bovine serum, along with 1% penicillin-streptomycin, in an incubator set at 37 °C. Plasmid constructs encoding SRSF1-specific siRNA and overexpression were obtained from MIAOLING PLASMID, and transiently transfected into MKN7 cells using Lipofectamine 3000 reagent. RT-PCR was performed utilizing the ABI 7500 instrument and the Thermo Fisher Scientific-SuperScript IV UniPrime reagent (The sequences of plasmids and primers are specified in Table S4).

Short-reads RNA-seq

Total RNA was extracted from cells using TRIzol reagent (Thermo Fisher Scientific), and the quality and integrity were validated using an Agilent 2100 Bioanalyzer (Agilent Technologies). Only samples with RNA integrity number > 7.0 were used for library construction. Poly(A) mRNA was isolated from total RNA using VAHTS mRNA Capture Beads (Vazyme). Sequencing libraries were then constructed using the VAHTS Universal Plus DNA Library Prep Kit for Illumina (Vazyme) according to the manufacturer’s instructions. Briefly, the isolated mRNA was fragmented and reverse-transcribed into cDNA. After end-repair and adapter ligation, the libraries were amplified by PCR and sequenced on an Illumina HiSeq 4000 platform with a 150-bp paired-end strategy. To identify and quantify SRSF1-regulated AS events, the rMATS (v4.1.0) algorithm was employed. Five types of AS events, including ES, A5SS, A3SS, ME, and retained intron, were analyzed.

Proliferation of cells, apoptosis, and estimation of the IC50 value

Cell viability was assessed using a Cell Counting Kit-8 (Servicebio) according to the manufacturer's protocol. Briefly, MKN7 and MKN74 cells were seeded into 96-well plates at a density of 500 cells/well in 100 μl of complete growth medium and allowed to adhere for 36 h at 37 °C under 5% CO₂. On the second day, oxaliplatin was prepared at 10 mM by dissolving it in 5% glucose and further diluted in a complete medium to generate a concentration gradient (0.1–100 μM, 2-fold serial dilutions). The diluted oxaliplatin was then added to the wells separately. After 48 h of incubation, 10 μl of CCK-8 reagent was added directly to each well, and the plates were incubated for an additional 2 h. A microplate reader was used to measure the absorbance at 450 nm, with a reference wavelength of 650 nm. The IC50 calculation and curve plotting were carried out using GraphPad Prism 9 (https://www.graphpad.com/scientific-software/prism/). For apoptosis assays, cells were stained using the Annexin V-APC/7-AAD apoptosis kit (MULTI SCIENCES). Each dye was incubated in the dark at room temperature for 15 min. The cells were then analyzed using a flow cytometer, with appropriate compensation and voltage adjustments. For the colony formation assay, single-cell suspensions were seeded at a density of 100 cells/well in 6-well plates. The medium was refreshed every 3 days during the 12-days culture period. After that, after being fixed in methanol for half an hour, the cell colonies were dyed with crystal violet for an hour and photographed. Three independent biological replicates were conducted for each experiment.

Western blot and RT-PCR

Total protein was extracted using RIPA lysis buffer (Servicebio) supplemented with 1% protease inhibitor cocktail (Servicebio). Protein concentrations were quantified using a BCA Protein Assay Kit (Vazyme). Equal amounts of protein were separated by 10% SDS-PAGE at 100 V for 60 min and subsequently transferred onto PVDF membranes (Millipore) at 200 mA for 120 min. After blocking with 5% non-fat milk in TBST for 1 h at room temperature, the membranes were incubated with primary antibodies overnight at 4 °C. Following three washes with TBST, the membranes were incubated with HRP-conjugated secondary antibodies for 1 h at room temperature. Finally, protein bands were visualized using an enhanced chemiluminescence detection system. Antibodies used for Western blotting were as follows: The primary antibodies, anti-SRSF1 (Proteintech, Cat No. 12929-2-AP) and anti-GAPDH (Proteintech, Cat No. 10494-1-AP), were both rabbit monoclonal and utilized at a dilution of 1:1000. The secondary antibody was HRP-conjugated Goat Anti-Rabbit IgG (Servicebio, Cat No. GB23303), applied at a dilution of 1:5000.

Total RNA was isolated from cells using the Super FastPure Cell RNA Isolation Kit (Vazyme) according to the manufacturer's instructions. The obtained RNA was then reverse transcribed into complementary DNA (cDNA) using the HiScript III RT SuperMix for PCR (Vazyme). Briefly, genomic DNA (gDNA) was removed by incubating the RNA samples with 4× gDNA Wiper Mix at 42° for 2 min. Subsequently, the reverse transcription reaction was performed with 5×HiScript III RT SuperMix at 37° for 5 min, followed by inactivation at 85° for 5 s to terminate the reaction. The resulting cDNA was used as a template for polymerase chain reaction (PCR). The PCR amplification was conducted in a 50ul reaction system containing 1uL of cDNA template, 1uL of each forward and reverse primer, 22uL of PrimerSTAR DNA Polymerase, and 25uL of double-distilled water. The thermal cycling conditions were as follows: an initial denaturation at 98° for 3 min, followed by 40 cycles of denaturation at 96° for 30 s, annealing at 55° for 30 s, and extension at 72° for 1 min. A final extension step was performed at 72° for 5 min.

The quantification of RNA bands in agarose gels was performed using ImageJ. The gel images were captured under non-saturating conditions and imported into the software. If necessary, the 'Invert' function was applied to ensure the bands appeared as dark signals on a light background for proper densitometry. Using the 'Gels' analyzer tool, lanes were defined and the peak area for each specific band was calculated. PSI was calculated using the formula: long transcript peak/(long transcript peak + short transcript peak) × 100%.

Statistical analysis

For continuous variables, comparisons between two groups were performed using the independent samples t test. One-way analysis of variance (ANOVA) was used for comparisons among three groups. If the overall test showed a statistical difference (p < 0.05), pairwise comparisons were conducted using the Tukey method. Survival curves were generated using the "survival" package. The significance of all statistical tests was determined at p < 0.05, and they were administered as two-sided tests. Version 4.4.0 of the R software (R Foundation for Statistical Computing) was employed to conduct all analyses.

Data availability

The datasets generated and/or analyzed during the current study available from the corresponding author on reasonable request.

Supporting information

This article contains supporting information.

Conflict of interest

The authors declare that they have no conflicts of interest with the contents of this article.

Reviewed by members of the JBC Editorial Board. Edited by Paul Shapiro

Supporting information

Supplementary Figure 2.

Supplementary Figure 3.

Supplementary Figure 4.

Supplementary Figure 5.