Candidate biomarkers identified by trans-omics analysis

In our previous proteomic dataset utilizing aptamers, we identified 236 differentially expressed serum proteins specific to GC (Supplementary Table 1). Their corresponding genes were found in the UniProt database. Analysis of the TCGA-STAD transcriptomic dataset uncovered 10,637 differentially expressed genes (Supplementary Fig. 1). There were 119 overlapping DEGs between the two datasets (Supplementary Table 2, Supplementary Fig. 2).

To identify key DEGs, we employed LASSO regression and random forest analyses on the overlapping DEG set. The LASSO regression with λ = 0.002 selected 22 key DEGs (Fig. 1A, B). The random forest analysis pinpointed the 24th decision tree with the lowest OOB error rate of 0.0172 for GC vs. normal sample classification, in which six genes had importance scores > 2, as evaluated by Gini importance (Fig. 1C, D). The LASSO regression and random forest selection identified four DEGs that overlapped, namely interleukin enhancer-binding factor 2 (ILF2), immunoglobulin J chain (JCHAIN), chromodomain helicase DNA-binding protein 7 (CHD7), and phosphoglucomutase-2-like 1 (PGM2L1) (Fig. 1E), which were designated as candidate biomarkers for GC.

Fig. 1

Identification of candidate biomarkers for gastric cancer. A, B: The selection of key overlapping differentially expressed genes (DEGs) by least absolute shrinkage and selection operator (LASSO) regression analysis. C, D: The selection of key overlapping DEGs by random forest (RF) analysis. E: Venn diagram of key overlapping DEGs selected by LASSO regression and RF analysis

Validation of candidate biomarker expression

In the TCGA-STAD dataset used for internal validation, ILF2 (Interleukin enhancer-binding factor 2), CHD7 (Chromodomain-helicase-DNA-binding protein 7), and PGM2L1 (Glucose 1,6-bisphosphate synthase) demonstrated significantly higher mRNA expression levels in GC than in control tissues, whereas JCHAIN (Immunoglobulin J chain) showed significantly lower mRNA expression levels (Fig. 2A). In the external validation using the merged GSE dataset, these candidate biomarkers showed the same trends in mRNA expression levels and significant differences between GC tissues and adjacent normal tissues (Fig. 2B). Before analysis, the external validation datasets (GSE27342, GSE54129, and GSE66229) retrieved from the GEO database underwent standardization, resulting in consistent gene expression distributions (Supplementary Fig. 3A). Following batch-effect correction, the differences among these datasets in the principal component space were significantly reduced (Supplementary Fig. 3B, C), indicating successful merging of the three GES datasets.

Fig. 2

Expression and diagnosis value of the four candidate biomarkers in gastric cancer and normal control tissues. * P < 0.05, ** P < 0.01, *** P < 0.001. A: STAD dataset from The Cancer Genome Atlas database; B: Merged datasets of GSE27342, GSE54129 and GSE66229 datasets from the Gene Expression Omnibus (GEO) database. C: STAD dataset from The Cancer Genome Atlas database; D: Merged dataset of GSE27342, GSE54129 and GSE66229 datasets from the Gene Expression Omnibus (GEO) database; AUC: area under the curve

Diagnostic value of candidate biomarkers for GC

We assessed the diagnostic significance of mRNA levels for the four potential biomarkers in GC using the receiver operating characteristic curve (ROC). In the TCGA-STAD dataset, the AUROCs for diagnosing GC with PGM2L1, ILF2, CHD7, and JCHAIN were 0.950, 0.920, 0.919, and 0.629, respectively (Fig. 2C). In the external validation dataset (the merged GSE dataset), the AUROCs for PGM2L1, ILF2, CHD7, and JCHAIN in diagnosing GC were 0.820, 0.784, 0.745, and 0.736, respectively (Fig. 2D). These results showed that PGM2L1, ILF2, and CHD7 (but not JCHAIN) all had AUROCs greater than 0.7 in both the datasets, highlighting their diagnostic potential for GC and were therefore selected as valuable biomarkers for subsequent studies.

Associations of the valuable candidate biomarkers with tumor immunity

Using CIBERSORT analysis, we compared the immune cell infiltration status in GC tissues between high and low expression levels of each biomarker. The results showed that ILF2, CHD7, and PGM2L1 were significantly associated with 14, 11, and 2 of 22 types of infiltrating immune cells, respectively (Fig. 3A-C). From the transcriptome data in the TCGA-STAD dataset, we extracted expression levels of 8 immune checkpoint genes (TIGIT, HAVCR2, CD274, SIGLEC15, LAG3, PDCD1, PDCD1LG2, and CTLA4). Comparing differences in their expression between high and low levels of each biomarker, we found that ILF2 expression was significantly associated with CD274, CTLA4, and SIGLEC15; CHD7 was significantly associated with CD274; and PGM2L1 was significantly associated with SIGLEC15 (Fig. 4A-C). These findings suggest that, among the three biomarkers, ILF2 exhibits the strongest association with tumor immunity in GC, involving multiple immune checkpoint genes. Therefore, ILF2 was selected as a potential biomarker for subsequent studies.

Fig. 3

Heatmaps of immune cell scores and comparisons between high and low biomarker expression groups. *P < 0.05, **P < 0.01, ***P < 0.001. A: ILF2; B: CHD7; C: PGM2L1

Fig. 4

Bioinformatics analyses of ILF2 in gastric cancer. *P < 0.05, **P < 0.01, ***P < 0.001. A-C: Comparison of immune checkpoint-related gene expression levels between the high and low expression groups of biomarkers. D-G: Comparison of sensitivity to anticancer drugs between the high and low ILF2 expression groups. H-K: Comparison of response to immunotherapy between the high and low ILF2 expression groups; IPS, the Immunophenoscore; POS, positive; NEG, negative

Associations of ILF2 with immunotherapy and chemotherapy responses

We utilized the GDSC database to analyze the variations in response to standard chemotherapy drugs (5-fluorouracil, paclitaxel, docetaxel, and cisplatin) [26] among gastric cancer patients with different levels of ILF2 expression. The findings indicate that the IC50s were lower in the high-level ILF2 group compared with the low-level ILF2 group. Hence, GC with increased ILF2 levels exhibits enhanced response to these drugs (Fig. 4D-G). Additionally, using the TCIA database, we calculated CTLA4- and PD-1-based immunophenotype scores (IPS) for GC patients and compared the response to anti-CTLA4 and anti-PD-1 immunotherapy between patients with high and low ILF2 expression levels. The results indicated that individuals with elevated ILF2 levels exhibited significantly attenuated responses to CTLA4 and PD-1 inhibitors (Fig. 4H-K).

Clinical validation of the diagnostic significance of serum ILF2 levels in gastric cancer

Utilizing ELISA (its standard curve shown in Figure 5A), serum levels of ILF2 protein were measured and notably elevated in the GC group when compared to the control group (443.23 ± 303.29 ng/mL vs. 72.31 ± 45.18 ng/mL, P < 0.0001) (Fig. 5, B). ILF2 levels were valuable for diagnosing GC with an AUROC of 0.944 (Fig. 5C). The baseline information of the patients is shown in Supplementary Table S3.

Fig. 5

Clinical and functional validation of ILF2. A: The standard curve for the detection of ILF2 by enzyme-linked immunosorbent assay. B: Serum ILF2 levels and comparison between gastric cancer and control group. C: The receiver operating characteristic curve of serum ILF2 levels for the diagnosis of gastric cancer. D: ILF2 levels in cell culture supernatants. E: Relative mRNA expression levels of ILF2 in different gastric cancer cell lines; F, G: The effect of siRNA-mediated ILF2 knockdown on mRNA expression in HGC-27 and AGS cells; H-J: Comparisons of HGC-27 cell proliferation between siRNA-mediated ILF2 knockdown and control groups. K: The effect of siRNA-mediated ILF2 knockdown on clone formation in AGS cells. AUC, Area Under Curve; CI, Confidence Interval; OD, Optical Density

Detection of ILF2 expression in the supernatant of gastric cancer cell culture

The presence of ILF2 protein was observed in the culture supernatants of BGC823 GC cells, and its concentration increased as the culture time progressed. However, it was not detected in the supernatants of GES-1 gastric epithelial cells (Fig. 5D), indicating that gastric cancer cells can secret ILF2.

Expression levels of ILF2 in GC cells

The ILF2 mRNA expression was detected by RT-qPCR in four GC cell lines (AGS, MKN45, HGC27, and BGC823) and one control cell line (GES-1). The primer sequences used for the PCR amplification of ILF2 were 5’- GGGGAACAAAGTCGTGGAAAG-3’ (forward) and 5’- CCAGTTTCGTTGGTCAGCA-3’ (reverse). The results indicated a significant upregulation of ILF2 mRNA expression in all four GC cell lines compared to the control (Fig. 5E).

Effect of ILF2 knockdown on GC cell growth

We knocked down ILF2 transcription using siRNAs in HGC27 and AGS cells. Three ILF2 mRNA-specific siRNA sequences (siILF2-1, siILF2-2, and siILF2-3) and one control sequence (siILF2-CONT) were employed in the knockdown experiments. Their nucleic acid sequences (sense) were as follows:

siILF2-1: 5’-GAUAGUAACACCUUCAGAATT-3’.

siILF2-2: 5’-CUUUGUACCACAUAUCCCATT-3’.

siILF2-3: 5’-GCUACAGUGAAGAUUCUCATT-3’.

siILF2-CONT: 5’-UUUCUCCGAACGUGUCACGUTT-3’.

The findings indicated that the siRNA depletion notably reduced the ILF2 mRNA levels (Fig. 5F, G). Among the three siRNAs, siILF2-2 exhibited the most effective knockdown. In CCK-8 experiments, there was a notable reduction in cell proliferation at 48, 72, and 96 h in HGC27 cells and 72 and 96 h in AGS cells transfected with siILF2-2 and siILF2-3 compared to the control (Fig. 5H-J). In clone formation experiments (Fig. 5K), the number of AGS cell clones with knocked-down ILF2 decreased significantly.

Effect of ILF2 overexpression on the growth of gastric cancer cells

The ILF2 overexpression plasmid was designed by Shanghai Genechem Co., Ltd (China), and vector details are presented in Fig. 6A. The results indicated that ILF2 overexpression significantly increased the level of ILF2 mRNA, as demonstrated in Fig. 6B, C. In the CCK-8 assay, BGC823 and MKN45 cells with overexpression of ILF2 exhibited significantly enhanced proliferative capacity compared to the control group (Fig. 6D-E). Additionally, In the clone formation experiment (Fig. 6F), the number of MKN45 cell clones overexpressing ILF2 increased significantly.

Fig. 6

The effect of overexpressing ILF2 on the function of gastric cancer cells. A The vector of ILF2. B, C The effect of ILF2 overexpression on mRNA expression in BGC823 and MKN45 cells. D, E: Comparisons of cell proliferation between ILF2 over-expressed groups and control groups in BGC823 and MKN45 cells. F. The effect of ILF2 over-expressed on clone formation in MKN45 cells. OE, over expression