2.1 Data Collection

RNA-seq expression data of stomach adenocarcinoma (STAD) and patients' clinical information were downloaded from the TCGA database (https://portal.gdc.cancer.gov/), and we obtained 375 cancer samples and 32 normal samples; the collection of anoikis-associated genes was obtained from the GeneCards database (https://www.genecards.org/).

2.2 Screening for Anoikis-Related Differential Genes

The “Limma” package in R language was used to normalize the data and screen for DEGs. The Wilcoxon rank sum test was used to perform non-parametric tests on the data. |Fold Change| > 1.5 and False Discovery Rate(FDR) < 0.05 were used to screen up and down regulated ARGs in gastric cancer samples. The “Pheatmap” package and “ggplot” package in R language were used to plot heatmaps and volcano maps of DEGs. Pearson’ correlation was applied to calculate differential gene expression correlations and miRNA-mRNA regulatory relationships were downloaded in Starbase (http://starbase.sysu.edu.cn/).

2.3 Single Sample Gene Set Enrichment Analysis

Single sample gene set enrichment analysis (ssGSEA) is an extension of the gene set enrichment analysis (GSEA) method that calculates enrichment scores for each sample and gene set pair. Each ssGSEA enrichment score represents the extent to which a specific gene set member is up or down regulated in the sample. Consensus clustering of cancer samples was performed by calculating the ssGSEA score of differentially expressed ARGs for each gastric cancer sample.

2.4 Consensus Clustering Analysis

Consensus clustering analysis was performed on the cancer samples using the “ConsensusClusterPlus” package in R language to further investigate the role and prognostic importance of ARGs in STAD, and the Calinski-Harabaz index was used as an evaluating index for the clustering of data to determine the optimal k-value of the clusters. It was obtained from the ratio of separation to tightness, with separation representing the degree of dispersion between subgroups and tightness representing the degree of aggregation within categories, and the larger the ratio indicates the better the clustering effect. The “Survival” package in R language was used for survival analysis, and p < 0.05 was considered to be statistically significant.

2.5 Weighted Gene Correlation Network Analysis

The “WGCNA” package in R language was used to construct co-expression networks for each subtype. Firstly, the samples of each subtype were hierarchically clustered to detect and eliminate outliers; secondly, the pickSoft Threshold function was used to build a scale-free topology network for each subtype; thirdly, the adjacency matrix was constructed and transformed into a TOM matrix, the co-expression network was constructed based on the optimal soft threshold, and the gene clustering tree was drawn after dividing the genes into different modules; finally, a correlation heatmap between genes and a correlation heatmap between modules and clinical features, p < 0.05 was considered statistically significant.

2.6 Enrichment Analysis

The “Metascape” (https://metascape.org) was used for the enrichment analysis and visualization of Gene Ontology (GO) and Kyoto Encyclopedia of Genes and genomes (KEGG) pathways for each subtype. Reactome Database (http://reactom.org/) was used for the analysis of reactome pathways for each subtype. GO is a common method for gene function enrichment, which covers three main categories: molecular function (MF), cellular component (CC), and biological process (BP). KEGG is a database that integrates genomic, chemical, and systemic functional information and is mainly used for analyses of biometabolic pathways. The Reactome database is similar to the KEGG database in that it brings together the reactions and biological pathways of human biological processes.

2.7 Immune Cell Infiltration Analysis

The “Cibersort” package in R language was used to calculate the immune cell infiltration level of each subtype after consensus clustering. The “Pheatmap” package in R language was used to draw heat maps. In addition, the degree of enrichment of ARGs within each sample and the enrichment score within each immune cell were calculated by ssGSEA, respectively. Spearman's correlation was used to calculate the correlation between the pooled scoring of differentially expressed anoikis genes and the immune cell infiltration level.

2.8 Screening for Hub-Genes

Using ANOVA to identify genes that are differentially expressed in different subtypes, FDR < 0.05 was considered statistically significant; enrichment analysis of differentially expressed genes was performed based on “Metascape”. The “limma” package in R language was applied to screen the differentially expressed genes of each subtype. The criteria of |Fold Change| > 1.5 and FDR < 0.05 were followed. “Pheatmap” package and “ggplot” package were used to draw heatmaps and volcano maps of differentially expressed genes.

2.9 Prognostic Analysis and Construction of Nomogram

Independent prognostic genes were screened by “Survival” package by performing univariate and multivariate COX regression survival analyses for genes differentially expressed in at least two subtypes, and p < 0.05 was considered statistically significant. The “rms” package in R language was applied to integrate the independent prognostic genes, and a nomogram was constructed to predict the 3–5-year survival of the patients, and the calibration graph was used to verify the degree of agreement between the model and the actual situation.

2.10 Clinical Samples

Three pairs of fresh gastric cancer tissues, adjacent non-tumor tissues, and liver metastatic tissues were used for qRT-PCR and partially Western Blotting. All samples were collected from January 2022 to September 2022 at the Fourth Hospital of Hebei Medical University, with the three pairs of sample providers being two males and one female, all aged between 60 and 65 (Table S1). The diagnosis of each gastric cancer patient was confirmed by two independent pathologists, and none of the patients underwent radiotherapy, chemotherapy, or immunotherapy before the biopsy. This study was approved by the Ethics Committee of the Fourth Hospital of Hebei Medical University (Project No: 2019ME0039), and written informed consent was obtained from each patient before sample collection. Our research was conducted in accordance with the Declaration of Helsinki.

2.11 Cell Culture

Normal gastric mucosal epithelial cells GES-1 and gastric cancer cell lines (MKN45, AGS, HGC-27) were purchased from the National Infrastructure of Cell Line Resource (NICR). The cells were cultured in DMEM medium (Gibco, USA) containing 10% fetal bovine serum, 100 U/mL penicillin, and 100 μg/mL streptomycin at a temperature of 37°C with 5% CO₂ and humidity.

2.12 Construction of Anoikis Model

After resuspending the adherent cells with trypsin, seed the cells at a density of 1 × 10⁶ cells per well in a 6-well plate with ultra-low attachment. Then, place them in a 37°C, 5% CO₂ incubator for culture. After 24 h of incubation, proceed with the subsequent experiments.

2.13 qRT-PCR

According to the kit instructions, total RNA was extracted from gastric cancer cells and gastric cancer tissues using TRIzol reagent (Invitrogen, USA). Total RNA was reverse transcribed using the PrimeScriptTM RT Reagent Kit (TaKaRa, Japan). qRT-PCR was performed using the Quantstudio DX system (Applied Biosystems, Singapore). β-actin was used as the internal reference gene, and the relative expression levels were calculated using the $$$ {2}^{-\Delta \Delta {C}_{\mathrm{T}}} $$$ method. Primer sequences are listed in Table S4.

2.14 Western Blot

GC cells and tissues were homogenized by adding lysate and Protease Inhibitor Cocktail and centrifuged at 12,000 rpm at 4°C for 5 min, and the supernatant was collected. The protein concentration in the supernatant was detected by the BCA kit. Proteins were separated by SDS-polyacrylamide gel electrophoresis and transferred to a PVDF membrane. This was followed by incubation with a primary antibody overnight at 4°C, and the membrane was washed with TBST and incubated with a labeled secondary antibody at room temperature for 2 h. The following primary antibodies against CYP1B1 (Abcam, ab185954), N-Cadherin (Abcam, ab76011), E-Cadherin (Abcam, ab314063), Vimentin (Abcam, ab92547), Snail (Abcam, ab216347), ITGAV (Abcam, ab179475), CL-caspase-7 (Abcam, ab256469), CL-caspase-3 (Abcam, ab32042), Bcl-2 (Abcam, ab182858), Bax (Abcam, ab32503), β-actin (Abcam, ab8227) were used. The signal was visualized using enhanced chemiluminescence (ECL) (Thermo Fisher, USA). The ratio of the target protein to the internal reference protein was calculated to derive the relative expression of the proteins.

2.15 Transwell Assay and Wound Healing Assay

The cell invasion ability was measured using 24-well Transwell chambers with 8.0-μm pore polycarbonate filters (CoStar, USA). 1 × 10⁵ cells were seeded into Matrigel-coated chambers. The upper chambers contained 200 μL of serum-free medium, and the lower chambers contained 500 μL of complete medium. After 24 h, the cells across the membrane were fixed with 4% paraformaldehyde and stained with 0.2% crystal violet for 20 min to identify the invaded cells. The number of stained cells was counted in three random fields. For the wound healing assay, cells were seeded in a 6-well plate at a density of 2 × 10⁴ per well, cultured until 80%–90% confluence, and scratched with a 1000 μL pipette tip. Cells were rinsed with PBS, and the medium was replaced with serum-free medium at 37°C for 24 h. The artificial wound was assessed at 0 h and 24 h post-scratching. The distance for the migration was measured by Image J.

2.16 Cell Transfection

To construct the CYP1B1 knockdown plasmid, shRNA targeting CYP1B1 (sh-CYP1B1 group) and negative control shRNA (sh-Ctrl group) were synthesized and cloned into the pCDH-CMV-MCS-EF1-copGFR plasmid, and GC cells were transfected with the plasmid according to the manufacturer's instructions. The target sequences of shRNAs are shown in Table S2.

2.17 Animal Models

All animal experiments were approved by the Committee on the Ethics of Animal Experiments of Hebei Medical University. Six weeks old male balb-c nude mice were purchased from Beijing Hufukang Biotechnology Co., LTD., and randomly divided into knockdown control group (sh-Ctrl group) and CYP1B1 knockdown group (sh-CYP1B1 group), with 5 mice in each group, to evaluate the effect of CYP1B1 on mice. 2 × 10⁶ MKN45 cells were injected into the right side of the mice. Tumor growth was monitored every 3 days, and the volume was calculated using the formula 0.5 × length × width². After 24 days, the mice were euthanized through intravenous injection of sodium pentobarbital, following the “Laboratory animal—Guidelines for euthanasia”, and the tumors were removed for further analysis.

3.1 Identification of Differentially Expressed Genes

The flow chart of the study is shown in Figure 1. To evaluate the expression patterns of ARGs in gastric cancer, clinical information and RNA-seq expression data were obtained from the TCGA database in 375 gastric cancer patients and 32 normal subjects. 338 ARGs were obtained from the GeneCards database. We found that a total of 333 genes were differentially expressed between normal tissues and gastric cancer tissues (Table S5), among which 74 genes were up-regulated and 49 were down-regulated. Only the DEGs of |Fold Change| > 2 and FDR < 0.01 were shown in this paper. We used the “Pheatmap” package to generate heat maps of differential genes through hierarchical clustering, in which red represents up-regulated gene expression and blue represents down-regulated gene expression (Figure 2A). “ggplot2” package is used to generate the volcano map; the horizontal coordinate represents Log2 (Fold Change), and the vertical coordinate represents −Log10(False Discovery Rate); each point represents a gene; red represents the up-regulated differential gene, blue represents the down-regulated differential gene, and black represents genes that are not significantly differentially expressed (Figure 2B). At the same time, we constructed the differential gene expression network and queried the miRNA expression regulation corresponding to each differential gene (Figure 2C).

3.2 Identification of Gene Subtypes Related to Anoikis in Gastric Cancer

In order to better study the expression characteristics of ARGs in gastric cancer, we calculated the ssGSEA score of differentially expressed ARGs in each sample and categorized the gastric cancer samples by consensus clustering method (Figure 3A). We evaluated k = 3 as the best number of groups by using the Calinski-Harabaz index (Figure 3C). The horizontal coordinate represents the number of groups, and the vertical coordinate represents the evaluation index. The larger the corresponding ratio of the vertical coordinate, the smaller the covariance of the data within the categories, the larger the covariance between the categories, and the better the clustering effect. K-M analysis showed no significant difference in survival time between the three subtypes(p = 0.096) (Figure 3B).

3.3 Weighted Gene Correlation Network Analysis

To determine the relationship between each subtype and clinical features, we performed WGCNA for each of the three subtypes. The optimal soft threshold β for cluster1, cluster2, and cluster3 was calculated to be 5, 3, and 7, respectively. Based on the optimal soft threshold β, cluster1, cluster2, and cluster3 were classified into 4, 3, and 3 modules, respectively. Among them, the genes in the gray module are not clustered into any module, so they belong to the invalid module. In cluster1, we calculated the scale-free fit index under different soft thresholds, and we chose 0.8 as the value criterion, so the minimum soft threshold of cluster1 is 5 (Figure 4A), and we also calculated the average connectivity under different soft thresholds (Figure 4B). Firstly, we plotted the gene clustering map (Figure 4C); the upper part is the tree diagram of hierarchical clustering of genes, and the lower part is the gene module, which corresponds to the top and bottom. Secondly, we plotted the correlation heatmap of the genes; the darker the color, the stronger the interactions between genes, and since the diagonal line indicates the interactions between genes within the module, the darkest color is found on the diagonal line (Figure 4E). Finally, we plotted a heatmap of the correlation between modules and clinical traits within each subtype; the darker the color the higher the correlation, red represents positive correlation, green represents negative correlation, and the number above each cell represents correlation while the number below represents significance (Figure 4D), in which the blue module was significantly correlated with clinical T stage(p = 0.006), while the correlation with age was the poorest (p = 0.0004). Meanwhile, we determined the optimal soft threshold β in cluster2 and cluster3, respectively, and plotted gene clustering, correlation heatmap between genes, and clinical trait correlation heatmap. The turquoise module were significantly correlated with clinical T stage in cluster2 (p = 0.0004) (Figure 5D). Both the blue module and the turquoise module were significantly correlated with clinical T stage in cluster3 (p = 0.001) (Figure 6D), while the turquoise module was the least correlated with age in cluster2 (p = 0.0009) (Figure 5D) and cluster3 (p = 0.0004) (Figure 6D).

And we showed the expression changes of each cluster gene (Figure 6F). The upper layer is the WGCNA clustering results; each point represents a gene, and the color of each point corresponds to the color within the module. We can visualize the number of genes in each cluster and the number of genes contained in the module. The lower layer is the gene expression changes; we can visualize the up-regulation and down-regulation of genes within each cluster. Red indicates up-regulation of gene expression, green indicates gene expression down-regulation; the darker the color, the higher the degree of gene up-regulation or down-regulation.

3.4 Enrichment Analysis and Immune Infiltration

GO and KEGG enrichment analyses showed that cluster1 was significantly enriched in response to xenobiotic stimulus, chemical carcinogenesis; cluster 2 and cluster 3 were significantly enriched in cell–cell adhesion, extracellular matrix (Figure 7A). Reactom enrichment analyses showed that cluster1 was enriched in the fatty acids signaling pathway, cluster 2, and cluster 3 were both significantly enriched in the G protein-coupled receptor(GPCR) signaling pathway (Figure 7A). We found that the gene function-enriched regions of cluster1 were different from those of cluster 2 and cluster 3, and the gene function-enriched regions of cluster 2 and cluster 3 were almost the same, which was also consistent with the results of the analysis of WGCNA. Cluster 2 and cluster 3 were mainly clustered on intercellular adhesion and extracellular matrix, which was consistent with the apoptotic properties of anoikis, which is also closely related to the clinical staging. The immune cell infiltration levels of the three subtypes were calculated separately by the “Cibersort” package, and the results showed that the abundance of activated NK cells (p = 0.021), memory B cells (p = 0.025), and M2 macrophages (p = 0.039) differed significantly among the three subtypes, with a p < 0.05 (Figure 7B) (Table S3). The correlation between ssGSEA scores and immune cell infiltration levels was calculated by the Cibersort, spearman's correlation, and the results showed that except for memory B cells (p = 3.6e-11), T cells CD4 naive (p < 0.001), T cells gamma delta (p = 6.7e-14), M1 macrophages (p = 7e-10), eosinophils (p < 0.001), and neutrophils (p = 4.3e-11) were all positively correlated with the ARGs, and the correlation coefficients, R, were all greater than 0.3, showing moderate correlation (Figure 7C), and p was less than 0.05, which was statistically significant. In conclusion, these results suggest that ARGs can provide insight into the immune response and immune infiltration of GC.

3.5 Screening for Differential Genes Between Different Subtypes

In order to more comprehensively study the expression of differentially expressed genes among the three different subtypes, we analyzed each subtype. By ANOVA we identified four sets of genes that were differentially expressed in different subtypes (Figure 8A), and enriched each of the four sets of genes (Figure 8B). We found that cluster1 and cluster3 were both enriched in cell adhesion. At the same time, we screened for differential genes between the three subtypes, and when differential genes were identified for one of the subtypes, genes from the other two subtypes served as controls. |Fold Change| > 1.5 and FDR < 0.05 were used as screening criteria. Eventually, we identified 983, 849, and 1877 differentially expressed genes in cluster1, cluster2, and cluster3, respectively (Figure 8C). Among them, 917 genes were differentially expressed in at least two subtypes (Figure 8D); five genes were differentially expressed between all three subtypes (p < 0.05) (Figure 8E,G), namely CYP1B1, EQTN, NRXN2, TBC1D3E, and TCEAL5. Among them, CYP1B1, NRXN2, and TCEAL5 were found in gene cluster1; EQTN and TBC1D3E are in gene cluster3. Finally, we have performed TIMER algorithm analysis on 5 genes that were differentially expressed among the three subtypes (Figure 8F). We have found that among the five genes, CYP1B1 shows the strongest correlation and the most significant statistical significance (p < 0.001), providing a theoretical basis for future research.

3.6 Survival Analysis and Establishment of Nomogram

Survival analyses were performed for all genes in the three subtypes (Figure 9A), with green dots representing HR < 1, unfavorable for survival; and red dots representing HR > 1, favorable for survival. The “Survival” package in R language was used to screen out all genes with a significance of p < 0.05 in univariate cox regression and differentially expressed in at least two clusters, and multivariate cox regression analyses were performed to screen for independent prognostic factors (Figure 9B). Finally, we screened 32 prognostically relevant genes (ABCA9, ADAMTSL3, ADH1B, AOX1, BCHE, CRYAB, DAAM2, ELOVL2, FAM110B, FLRT2, GALNT16, GAMT, GLP2R, GUCY1A3, ITGB3, ITGBL1, KCNS2, KIAA0408, LAMA2, NAP1L2, NAP1L3, NGFRAP1, NHSL2, NUDT10, PCDH9, PTCH2, SELP, SLC16A7, SPARCL1, TCEAL5, THPO, TNN). We found that TCEAL5 was differentially expressed among the three subtypes and was an independent prognostic factor, which implies that TCEAL5 is likely to be a new target for gastric cancer treatment. A nomogram was constructed by applying 32 independent prognostic factors (Figure 9C), and the calibration graph showed that the prognosis predicted by the nomogram model was in high agreement with the actual situation (Figure 9D).

3.7 CYP1B1 Promotes the Progression of Gastric Cancer In Vivo and In Vitro

We chose CYP1B1 for experimental validation. Clinical tissue samples of gastric cancer liver metastasis were selected, and the results of qRT-PCR (Figure 10A) and WB (Figure 10B) showed that the expression of CYP1B1 increased sequentially in normal tissue, gastric cancer tissue, and liver metastatic tissue. To analyze the impact of CYP1B1 on the malignant behavior of gastric cancer, we quantified its expression in normal gastric cells (GES-1) and GC cell lines (MKN45, AGS, HGC27), with the highest expression of CYP1B1 observed in MKN45 cells (Figure 10C). For modeling the anoikis of gastric cancer cells, gastric cancer cells were cultured in ultra-low attachment plates for 24 h, and the qRT-PCR results showed that CYP1B1 expression was higher than in the non-suspended group (Figure 10D). By constructing CYP1B1 low-expression cells and control groups through lentiviral transfection, after suspension culture, the qRT-PCR results indicated that the CYP1B1 low-expression group had lower CYP1B1 expression than the control group (Figure 10E, Figure S1). WB results showed high expression of E-Cadherin and low expression of CYP1B1, ITGAV, N-Cadherin, Vimentin, and Snail proteins, indicating a non-EMT tendency (Figure 10F). Meanwhile, the expression of the anti-apoptotic protein Bcl-2 decreased in the CYP1B1 low-expression group, while the pro-apoptotic proteins CL-caspase3, CL-caspase7, and Bax increased (Figure 10G). Finally, the effect of CYP1B1 expression on cell invasion and migration was analyzed through Transwell and scratch assays. The results indicated that low expression of CYP1B1 inhibited the migration of gastric cancer cells (Figure 10H,I). To further verify the in vitro experimental results, sh-Ctrl MKN45 cells and sh-CYP1B1 MKN45 cells were subcutaneously injected to establish a subcutaneous tumor model in nude mice (Figure 10J). We observed changes in the volume of subcutaneous tumors and plotted the corresponding growth curves, with results showing that CYP1B1 knockdown significantly inhibited the growth of subcutaneous tumors (Figure 10K). After 24 days, the mice were euthanized, and the tumor weight of the sh-Ctrl group was significantly heavier than that of the CYP1B1 knockdown group (Figure 10L). The above results confirm the carcinogenic role of CYP1B1 in promoting gastric cancer both in vivo and in vitro.