Drug metabolism‐related eight‐gene signature can predict the prognosis of gastric adenocarcinoma

Abstract

Background

Metabolic abnormalities in patients with gastric adenocarcinoma lead to drug resistance and poor prognosis. Therefore, this study aimed to explore biomarkers that can predict the prognostic risk of gastric adenocarcinoma by analyzing drug metabolism‐related genes.

Methods

The RNA‐seq and clinical information on gastric adenocarcinoma were downloaded from the UCSC and gene expression omnibus databases. Univariate and least absolute shrinkage and selection operator regression analyses were used to identify the prognostic gene signature of gastric adenocarcinoma. The relationships between gastric adenocarcinoma prognostic risk and tumor microenvironment were assessed using CIBERSORT, EPIC, QUANTISEQ, MCPCounter, xCell, and TIMER algorithms. The potential drugs that could target the gene signatures were predicted in WebGestalt, and molecular docking analysis verified their binding stabilities.

Results

Combined with clinical information, an eight‐gene signature, including GPX3, ABCA1, NNMT, NOS3, SLCO4A1, ADH4, DHRS7, and TAP1, was identified from the drug metabolism‐related gene set. Based on their expressions, risk scores were calculated, and patients were divided into high‐ and low‐risk groups, which had significant differences in survival status and immune infiltrations. Risk group was also identified as an independent prognostic factor of gastric adenocarcinoma, and the established prognostic and nomogram models exhibited excellent capacities for predicting prognosis. Finally, miconazole and niacin were predicted as potential therapeutic drugs for gastric adenocarcinoma that bond stably with NOS3 and NNMT through hydrogen interactions.

Conclusions

This study proposed a drug metabolism‐related eight‐gene signature as a potential biomarker to predict the gastric adenocarcinoma prognosis risks.

Objective: Metabolic abnormalities in patients with gastric adenocarcinoma (GA) lead to the development of drug resistance and a poor prognosis. Therefore, this study aims to explore biomarkers that can predict the prognostic risk of GA from the perspective of drug metabolism. Methods: RNA‐seq and clinical information of GA were downloaded from UCSC and GEO. The univariate and LASSO regression analyses were used to identify prognostic gene signatures of GA. The relationships between GA prognostic risk and tumor microenvironment were assessed by CIBERSORT, EPIC, QUANTISEQ, MCPCounter, Xcell and TIMER algorithms. The potential drugs of feature genes were predicted in Webgestal and molecular docking analysis was generated to verify their binding stabilities. Results: Combined with clinical information, 8 gene signatures including GPX3, ABCA1, NNMT, NOS3, SLCO4A1, ADH4, DHRS7 and TAP1 were identified from drug‐metabolism‐related gene‐set. Based on their expressions, risk scores were calculated and patients were grouped into high‐risk and low‐risk. Survival analyses showed that patients in the high‐risk group had poorer outcomes. The risk group was also detected as an independent prognostic factor of GA, and the established prognostic and nomogram models both showed excellent capacities in predicting GA prognosis. Furthermore, we also revealed significant differences in immune cell infiltrations between risk groups. Finally, Miconazole and Niacin were predicted as potential therapeutic drugs of GA that bond stably with NOS3 and NNMT through hydrogen interactions. Conclusions: This study proposed 8 gene signatures related with drug metabolism as potential biomarkers to predict the prognostic risk of patients with GA.

1. INTRODUCTION

Gastric adenocarcinoma is a life‐threatening malignancy of the gastrointestinal tract and has become the third leading cause of cancer death globally. Of all the gastric cancers, approximately 90%–95% are gastric adenocarcinomas. ¹ In 2018, more than one million new cases were confirmed, the majority of which were locally advanced at the time of diagnosis. ² The 5‐year survival rate of advanced or metastatic gastric adenocarcinoma is less than 30%. ³ The incidence of local recurrence or distant metastasis of gastric cancer after surgery remains at 40%–70%, even with surgical intervention, radiotherapy, chemotherapy, and other treatment strategies, along with a certain degree of side effects after radiotherapy and chemotherapy. ⁴ Therefore, numerous studies have been conducted to explore prognostic biomarkers in an attempt to improve the clinical outcome of patients with gastric adenocarcinoma. Among them, Yao and Ren et al. evaluated the importance of immune microenvironment‐related genes in gastric adenocarcinoma prognosis by mining public databases ⁵ , ⁶ . However, a more comprehensive understanding of tumorigenesis mechanisms and the exploration of potential prognostic biomarkers from multiple perspectives are still required.

Metabolic abnormalities are the primary cause of drug resistance in patients with gastric adenocarcinoma, and current studies have provided profound insights into the metabolic changes of gastric adenocarcinoma and discussed their possible regulatory mechanisms. ⁷ A related genome‐wide profile analysis identified several important gastric cancer biomarkers that were significantly associated with drug metabolism pathways. ⁸ The long non‐coding RNA MACC1‐AS1, a biomarker related to gastric cancer prognosis, was reported to regulate disease metabolism through enhanced glycolysis and antioxidant capacity. ⁹ Furthermore, ectopic expression of S100P has been found to correlate with proliferation and increased drug resistance in gastric cancer cells. ¹⁰ A bioinformatics study found that the differentially expressed genes (DEGs), including ASPN, COL1A1, FN1, VCAN, and MUC5AC, in gastric cancer were significantly associated with survival prognoses of patients and were predominantly enriched in drug metabolism pathways. ¹¹ Although these genes were found to be involved in drug metabolism pathways, few studies have systematically reported drug metabolism‐related drugs and thoroughly explored their prognostic values.

Therefore, the current study aimed to identify genes that are significantly associated with gastric adenocarcinoma prognosis from drug metabolism‐related gene sets. Based on the expression of these genes, we constructed a prognostic model and explored the predictive performance of the model through internal and external validation. Furthermore, we predicted the drugs that could target these genes and performed a molecular docking analysis to verify their binding. Additionally, multiple databases were used to investigate the relationship between the prognostic risk of gastric adenocarcinoma and the immune microenvironment. The workflow of the study is shown in Figure S1. The feature genes proposed in this study may be potential therapeutic markers, resulting in improved clinical outcomes in patients.

2. MATERIALS AND METHODS

2.1. Data acquisition

RNA‐seq data of gastric adenocarcinoma and related clinical information, including tumor stage, family history, lymph node examined count, neoplasm histologic grade, primary diagnosis, resection or biopsy site, disease type, overall survival (OS), and OS duration were downloaded from the UCSC Xena (https://toil.xenahubs.net) platform. ¹² A total of 32 paracancerous samples and 375 tumor samples were included in this study, and 348 gastric adenocarcinoma samples with prognostic information were enrolled to develop a prognostic model. Furthermore, expression data and clinical information of 65 gastric adenocarcinoma samples in the microarray dataset GSE13861 were obtained from the Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo/) database. ¹³ GSE13861 was detected using the GPL6884 Illumina HumanWG‐6 v3.0 Expression BeadChip and was used as the validation set.

2.2. Analysis of drug metabolism‐related genes

Based on published articles ¹⁴ and The Cancer Genome Atlas (TCGA) database, a total of 228 drug metabolism‐related genes were matched. Using the empirical Bayes method provided by the limma package (v3.10.3, http://www.bioconductor.org/packages/2.9/bioc/html/limma.html), ¹⁵ drug metabolism‐related DEGs between tumor and paracancerous samples were identified with p‐values adjusted by Benjamini & Hochberg method < 0.05, and |log fold‐change| >0.5 as thresholds.

2.3. Enrichment analysis and protein‐protein‐interaction (PPI) network construction

Gene ontology (GO) functions and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways of drug metabolism‐related DEGs were enriched in DAVID v6.8 (http://david.ncifcrf.gov/) ¹⁶ , ¹⁷ . GO and KEGG terms with p‐value < 0.05, and gene count ≥2 were selected with significant correlation. Moreover, the STRING database (v11.0, http://string‐db.org/) ¹⁸ was used to analyze the relationship between proteins coded by drug metabolism‐related DEGs, and the PPI network was visualized using Cytoscape (v3.6.1, http://cytoscape.org/). ¹⁹

2.4. Screening of drug metabolism‐related prognostic DEGs

Based on the drug metabolism‐related DEGs obtained above and the survival information of gastric adenocarcinoma samples, univariate Cox regression analysis was performed to select drug metabolism‐related prognostic DEGs using the survival package (v2.41–1, http://bioconductor.org/packages/survivalr/) in R3.6.1. ²⁰ DEGs with a p‐value < 0.05 were determined to be significantly correlated with prognosis.

2.5. Construction and validation of the prognostic model

By utilizing the prognostic information of gastric adenocarcinoma samples in the training set and the expression values of drug metabolism‐related prognostic DEGs in each sample, genes were further selected as the optimized gene set using the least absolute shrinkage and selection operator (LASSO) regression analysis in the glmnet package (v2.0–18, http://cran.r‐project.org/web/packages/glmnet/index.html) ²¹ , ²² . The risk score of each sample was then calculated as follows:

$$Riskscore=\sum {\beta }_{\mathit{gene}}\times Ex{p}_{\mathit{gene}}$$

where β_gene indicates the LASSO regression coefficient of gene signature, and Exp_gene indicates their expression levels in gastric adenocarcinoma samples. To verify the effectiveness of the prognostic model, the risk scores of the samples in GSE13861 were calculated. The samples were then grouped into high‐ and low‐risk groups based on the median risk score. Kaplan‐Meier (KM) curves were created to analyze the difference in survival status between the two groups.

2.6. Feature analysis of gene signature

The expression cutoff value of each gene signature in the training set was obtained to determine the optimal cutoff point calculated by the Survminer package of R3.6.1 (v0.4.3). KM analysis was then performed to evaluate the difference in survival prognosis between samples in the high‐ and low‐expression groups. A heatmap was created to observe the relationships between the expression level, risk score, and clinical characteristics of each gastric adenocarcinoma sample.

2.7. Statistical analysis of clinical features among risk groups

In the training set, the chi‐squared test in R3.6.1 was used for statistical analysis and comparison of categorical variables, including TNM classification, tumor stage, family history, neoplasm histologic grade, primary diagnosis, resection or biopsy site, and disease type, between risk groups. The t test was used for continuous variables, including age and lymph node examined count.

2.8. Immune microenvironment analysis between risk groups

In the current study, CIBERSORT (https://cibersort.stanford.edu/index.php) ²³ , ²⁴ , EPIC (https://gfellerlab.shinyapps.io/EPIC_1‐1/), ²⁵ QUANTISEQ, ²⁶ MCP‐counter (https://github.com/ebecht/MCPcounter), ²⁷ xCell (https://xcell.ucsf.edu/), ²⁸ and TIMER (https://cistrome.shinyapps.io/timer) ²⁹ were used to estimate immune cell infiltration among risk groups. The Wilcoxon test was used to analyze the difference between the two groups and a heatmap was created accordingly. Furthermore, the ESTIMATE algorithm ³⁰ was used to assess the stromal and immune scores of gastric adenocarcinoma samples, and the t test was applied to analyze the difference, followed by generation of a box plot.

2.9. Differential pathway analysis between risk groups

Gene Set Enrichment Analysis (GSEA, v3.0) software was used for pathway enrichment with c2.cp.kegg.v7.1.symbols.gmt in MSigDB v7.1 (http://software.broadinstitute.org/gsea/msigdb/index.jsp) ³¹ as an enrichment background. Then, the differences in enriched pathways were analyzed between risk groups with a false discovery rate (FDR) < 0.05.

2.10. Analysis of independent prognostic factors and construction of a nomogram prediction model

To determine whether the prognostic model could be used as an independent prognostic factor, univariate Cox regression analysis was performed on age, sex, tumor stage, family history, neoplasm histologic grade, disease type, and lymph node examined count. Variables with p‐values < 0.05 were selected for multivariate Cox regression analysis, followed by a further selection of statistical significance at p < 0.05. A nomogram model was created to predict the 1‐, 2‐, 3‐, and 5‐year survival probabilities of gastric adenocarcinoma patients according to the multivariate Cox regression analysis results. Calibration curves were generated to verify model accuracy.

2.11. Drug enrichment prediction of gene signature

Based on the obtained genes, the WebGestalt database (http://www.webgestalt.org/option.php) ³² was used for drug enrichment prediction using over‐representation analysis. Drugs with p < 0.05 were selected as candidate drugs that could bind with the genes. Based on the prognostic effect of the gene signature, drugs that function as inhibitors or inducers were selected. Finally, the binding stabilities between candidate inhibitors or induced drugs and related target genes were validated by molecular docking using AutoDock 4.2.6.

3. RESULTS

3.1. Analysis of drug metabolism‐related DEGs of gastric adenocarcinoma

Differential analysis was performed between gastric adenocarcinoma samples and paracancerous samples based on 228 matched drug metabolism‐related genes. The volcano plot (Figure 1A) shows the 77 obtained drug metabolism‐related DEGs. Then, enrichment analysis was performed on these drug metabolism‐related DEGs, and 64 biological processes (BP), 13 cellular components (CC), 47 molecular functions (MF), and 21 KEGG pathways were obtained. The bubble charts of Figure 1B display the top 10 GO and KEGG terms, ranked by p‐values. The results showed that these DEGs were primarily enriched in GO‐BP of several metabolic processes, GO‐CC of TAP complex and host cell, GO‐MF of aldehyde dehydrogenase activity, and in KEGG pathways of drug metabolism, xenobiotic metabolism, and chemical carcinogenesis. The interactions of these proteins coded by DEGs were then analyzed, and a PPI network was created, as shown in Figure 1C. This PPI network contained 66 drug metabolism‐related DEGs and 309 relation pairs. Among these nodes, UGT1A1, CYP2B6, and CYP3A4 had a larger degree.

FIGURE 1.

Analysis of drug metabolism‐related DEGs of gastric adenocarcinoma. (A) The volcano plot shows drug metabolism‐related DEGs between gastric adenocarcinoma samples and paracancerous samples; (B) GO functions and KEGG pathway enrichment analyses of drug metabolism‐related DEGs; (C) The protein‐protein‐interaction network created based on proteins coded by drug metabolism‐related DEGs

3.2. Screening of gene signature of gastric adenocarcinoma

Combined with the prognostic information of gastric adenocarcinoma samples, a univariate Cox regression analysis was performed, and 11 drug metabolism‐related prognostic DEGs were selected with p < 0.05, as shown in Figure 2A. Then, LASSO regression analysis was performed to select the optimized gene set, and an eight‐gene signature was identified, including GPX3, ABCA1, NNMT, NOS3, SLCO4A1, ADH4, DHRS7, and TAP1. Their prognostic effects are shown in Figure 2B, and only SLCO4A1 and TAP1 were considered protective factors with hazard ratios <1. Based on the expression cutoff of these genes in the training set, samples were separated into high‐ and low‐expression groups. The KM curves in Figure 2C illustrated that patients with high GPX3, ABCA1, NNMT, NOS3, ADH4, and DHRS7 expressions had worse survival status, whereas high SLCO4A1 and TAP1 expressions were significantly associated with better prognoses.

FIGURE 2.

Screening of gene signature of gastric adenocarcinoma. (A) The forest map shows the prognostic DEGs selected from the univariate Cox regression analysis; (B) the forest map shows the gene signature of gastric adenocarcinoma screened by the multivariate regression analysis; (C) KM curves show the survival differences between patients with different expression levels of these genes

3.3. Establishment and validation of the prognostic model

Using the LASSO regression coefficients of each gene obtained above, the risk scores of each sample were calculated, and a prognostic model was created. Based on the median value of the risk score, the samples were grouped into high‐ and low‐risk groups. The expression distributions of these genes in the two groups were visualized using a heatmap (Figure 3A). The results suggested that the expression distribution of genes in the risk groups was different. To validate the model efficiency in the training set, we found that the prognostic risk of patients increased along with increased risk scores (Figure 3B). The KM curve (Figure 3C) demonstrated that patients in the high‐risk groups had a worse survival status. Receiver operator characteristic (ROC) curves (Figure 3D) were created to verify model accuracy, and it was observed that the prognostic model had superior performance in predicting 1‐, 2‐, 3‐, and 5‐year survival prognoses with area under the curves (AUCs) of 0.67, 0.679, 0.711, and 0.728, respectively. The prognostic model was further validated using the GSE13861 dataset (Figure 3E–G). Similarly, as the risk score increased, the number of deaths increased. Meanwhile, patients in the high‐ and low‐risk groups had significant differences in survival time, and the model accuracy was proven in the validation set with all AUCs in ROC curves over 0.65. These results indicated a significant association between the risk grouping and actual outcomes.

FIGURE 3.

Construction and validation of the prognostic model. (A) The heatmap shows the expression distribution of genes in high‐ and low‐risk groups; Model validation in the training set (B–D) and the GSE13861 validation dataset (E–G); (B and E) the distribution of survival time on risk scores; (C and F) KM curves show the survival difference between the two groups; (D and G) Receiver operator characteristic curves show the model accuracy in predicting 1‐, 2‐, 3‐, and 5‐year prognoses

3.4. Difference of clinical characteristics between risk groups

The clinical information of the high‐ and low‐risk groups was compared, as shown in Table 1. The results suggested that the two groups had significant differences in lymph node examined count, tumor stage, neoplasm histologic grade, pathologic N, pathologic T, and primary diagnosis.

TABLE 1.

The statistics of clinical features in the high‐risk and low‐risk groups

Subgroups	Low risk	High risk	p‐Value
Age (mean ± SD)	66.0 ± 10.1	64.6 ± 10.6	2.058E−01
Lymph node examined count	26.5 ± 20.9	19.2 ± 15.1	4.264E−04
Family history			5.088E−01
Yes	9	6
No	133	129
NA	32	39
Tumor stage			3.544E−04
I	1	12
II	34	54
III	56	74
IV	70	22
NA	12	12
Neoplasm histologic grade			3.463E−03
G1	3	6
G2	77	46
G3	89	118
GX	5	4
Pathologic M			8.204E−02
M0	161	150
M1	6	16
MX	7	8
Pathologic N			3.294E−03
N0	65	38
N1	39	53
N2	39	32
N3	29	42
NX	2	7
NA	0	2
Pathologic T			6.183E−04
T1	15	1
T2	38	36
T3	79	80
T4	42	53
TX	0	4
Primary diagnosis			1.223E−03
Adenocarcinoma	55	62
Adenocarcinoma with mixed subtypes	0	1
Adenocarcinoma, intestinal type	40	33
Carcinoma, diffuse type	20	39
Mucinous adenocarcinoma	6	13
Papillary adenocarcinoma	4	1
Signet ring cell carcinoma	5	7
Tubular adenocarcinoma	44	18
Site of resection or biopsy			7.366E−01
Body of stomach	46	38
Cardia	44	40
Fundus of stomach	18	22
Gastric antrum	59	68
Lesser curvature of stomach	1	0
Stomach	6	6
Disease type			1.941E−01
Adenomas and Adenocarcinomas	162	154
Cystic, Mucinous and Serous Neoplasms	12	20

Bold p‐value < 0.05 indicates statistical significance.

Abbreviations: NA, not available; SD, standard division.

3.5. Difference of immune microenvironment between risk groups

Based on CIBERSORT, EPIC, QUANTISEQ, MCPCounter, xCell, and TIMER algorithms, the relative infiltration abundances of immune and stromal cells were estimated. The infiltration abundance of immune cells is shown in Figure 4A. Then, by comparing the differences between risk groups, 59 immune microenvironment‐related cells (p < 0.05) were screened out, as shown in Figure 4B. Immune and stromal scores were calculated for each sample, and significant differences between risk groups were visualized using box plots (Figure 4C,D).

FIGURE 4.

Estimation of immune cell infiltration abundance in high‐ and low‐risk groups. (A) The infiltration abundances of immune and stromal cells estimated by CIBERSORT, EPIC, QUANTISEQ, MCPCounter, xCell, and TIMER algorithms; (B) the heatmap shows 59 immune microenvironment‐related cells with p < 0.05; (C) the difference in immune scores between high‐ and low‐risk groups; (D) the difference in stromal scores between high‐ and low‐risk groups

3.6. GSEA pathway enrichment analysis among risk groups

The KEGG pathways of high‐ and low‐risk groups were analyzed using GSEA. By setting the threshold of FDR < 0.05, the high‐risk group significantly enriched 25 KEGG pathways with a normalized enrichment score (NES) >0. Furthermore, 11 significant pathways were enriched in the low‐risk group, with an NES < 0. The enriched pathways of these two groups are shown in Table 2.

TABLE 2.

Enrichment GSEA pathways of high‐ and low‐risk groups

Terms (High‐risk)	NES	FDR p‐Value
Focal adhesion	2.068	0.040
Regulation of actin cytoskeleton	2.054	0.024
Hypertrophic cardiomyopathy HCM	2.031	0.021
Complement and coagulation cascades	2.023	0.018
Vascular smooth muscle contraction	2.004	0.018
Glycosphingolipid biosynthesis ganglio series	1.992	0.018
Calcium signaling pathway	1.976	0.018
KEGG dilated cardiomyopathy	1.965	0.019
Arrhythmogenic right ventricular cardiomyopathy ARVC	1.964	0.017
ECM receptor interaction	1.934	0.023
GAP junction	1.923	0.025
Leukocyte transendothelial migration	1.920	0.024
Vasopressin regulated water reabsorption	1.861	0.040
TGF‐β signaling pathway	1.851	0.041
Neuroactive ligand‐receptor interaction	1.851	0.039
Melanogenesis	1.846	0.038
Pathogenic escherichia coli infection	1.833	0.041
Cell adhesion molecules cams	1.829	0.040
MAPK signaling pathway	1.821	0.041
Prion diseases	1.817	0.041
Adherens junction	1.802	0.045
Glycosaminoglycan biosynthesis chondroitin sulfate	1.801	0.043
Melanoma	1.788	0.046
Hematopoietic cell lineage	1.775	0.049
Axon guidance	1.770	0.049

Terms (Low‐risk)	NES	FDR p‐Value
Spliceosome	−2.117	0.015
DNA replication	−1.966	0.039
RNA degradation	−1.946	0.032
Base excision repair	−1.896	0.040
Homologous recombination	−1.891	0.035
Aminoacyl tRNA biosynthesis	−1.886	0.031
One carbon pool by folate	−1.876	0.029
Pyrimidine metabolism	−1.870	0.027
Proteasome	−1.868	0.025
Cell cycle	−1.862	0.023
Nucleotide excision repair	−1.800	0.039

Abbreviations: FDR, false discovery rate; GSEA, gene set enrichment analysis; NES, normalized enrichment score.

3.7. Analysis of independent prognostic factors and establishment of a nomogram model

To further identify the independent prognostic factors, age, sex, tumor stage, family history, neoplasm histologic grade, TNM classification, disease type, and lymph node examined count were incorporated into the univariate Cox regression analysis (Figure 5A). Variables with p‐values < 0.05 were then selected for multivariate Cox regression analysis. Finally, the risk group, pathologic N, and pathologic M were identified as independent prognostic factors (Figure 5B). A nomogram model was established to predict 1‐, 2‐, 3‐, and 5‐year survival probabilities (Figure 5C), and calibration curves were created to verify model accuracy. Figure 5D shows that the predicted probabilities of 2‐, 3‐, and 5‐year OS were similar to the actual OS, thereby suggesting an excellent prediction accuracy of the nomogram model.

FIGURE 5.

Construction and validation of the nomogram model. (A and B) Identification of the independent prognostic factors using univariate and multivariate Cox regression analyses; (C) the establishment of a nomogram model to predict 1‐, 2‐, 3‐, and 5‐year survival probabilities; (D) calibration curves were created to verify the accuracy of the nomogram model

3.8. Molecular docking analysis of predicted drugs and related targets

Drug prediction enrichment analysis was performed on an eight‐gene signature, and 30 drugs were predicted to target 7 genes. Among them, high ABCA1, NNMT, and NOS3 expression levels were correlated with poor prognoses, and their related inhibitors or binder's glyburide, niacin, and miconazole, respectively, were selected for molecular docking analyses. The binding results of glyburide‐ABCA1, miconazole‐NOS3, and niacin‐NNMT complexes are shown in Figure 6A–C, respectively. As a result, miconazole bonded with the PHE‐473 residue of NOS3 by a hydrogen bond with a length of 2.5 Å. Moreover, the small molecular ligand niacin bonded with ARG‐30 and SER‐32 residues of the receptor NNMT through hydrogen bond interactions. These findings illustrated that these ligand‐receptor complexes were in a stable state of binding.

FIGURE 6.

Molecular docking analyses of Glyburide‐ABCA1 (A), miconazole‐NOS3 (B), and niacin‐NNMT (C) complexes. Yellow and pink indicate small molecular ligand and protein receptors, respectively. Green indicates the binding residues, while the red line indicates hydrogen bond interaction

4. DISCUSSION

Gastric adenocarcinoma ranks fifth among the most prevalent malignancies worldwide, and the development of drug resistance because of metabolic disorders is one of the reasons for poor prognosis. ¹ Therefore, by mining TCGA and GEO, we identified eight prognostic genes, namely GPX3, ABCA1, NNMT, NOS3, SLCO4A1, ADH4, DHRS7, and TAP1 from the drug metabolism‐related gene set using univariate and LASSO regression analyses. Then, a prognostic model was constructed, and patients were grouped into high‐ and low‐risk groups. Survival analyses showed that patients in the high‐risk group had worse prognoses, while the ROC curves showed that the prognostic model had good predictive performance in both the training and validation sets with AUCs > 0.65. Furthermore, the risk group was identified as an independent prognostic factor, and the established nomogram model exhibited good accuracy in predicting 1‐, 2‐, 3‐, and 5‐year survival probabilities.

Among the eight genes, we found that SLCO4A1 and TAP1 were protective factors for gastric adenocarcinoma prognosis, whereas high GPX3, ABCA1, NNMT, NOS3, ADH4, and DHRS7 expressions were significantly associated with poor prognoses. It has been reported that in gastric cancer patients over 60 years of age, GPX3 hypermethylation was significantly correlated with a shorter time to tumor recurrence. ³³ Meanwhile, Wang et al. believed that GPX3 is a risk factor for gastric cancer, and the intron single nucleotide polymorphism of GPX3 may alter gastric cancer risk by affecting gene expression levels. ³⁴ Therefore, we speculated that the risk role of GPX3 in gastric adenocarcinoma prognosis may be closely related to changes in gene epigenetics. As for NNMT, related studies found that high NNMT expression in stromal cells may predict an unfavorable postoperative prognosis for gastric carcinoma. ³⁵ NNMT is known to act as a negative predictor for gastric carcinoma prognosis and is correlated with immune infiltrates ³⁶ , ³⁷ . These findings confirmed our results, and upregulated NNMT in gastric cancer cells may promote the occurrence of epithelial‐mesenchymal transition by activating TGF‐β1/SMAD signaling, ³⁸ thereby leading to tumor recurrence and metastasis. A pan‐cancer analysis found that increased NOS3 expression in gastric adenocarcinoma may lead to poor prognosis through several typical cancer‐related pathways. ³⁹ This finding is consistent with our results; however, how NOS3 influences gastric adenocarcinoma prognosis through cancer‐related pathways requires investigation.

The tumor microenvironment significantly contributes to the occurrence, progression, prognosis, and immunotherapy response of gastric adenocarcinoma. ⁶ Gastric adenocarcinoma is a chronic gastritis caused by Helicobacter pylori and is often characterized by the infiltration of immune cells, including granulocytes, macrophages, and T lymphocytes. ⁴⁰ Therefore, the current study explored the relationship between prognostic risk and immune cell infiltration. Significant differences in the infiltration of 59 immune microenvironment‐related cells, such as B cells, T cells, neutrophils, and macrophages were found between the risk groups. Relevant studies have reported that high infiltration of B lymphocytes is beneficial for gastric cancer prognosis. ⁴¹ Fristedt and Knief et al. suggested that increased infiltration density of B cells or plasma cells was associated with improved prognosis and prolonged OS of esophageal and gastric cancers ⁴² , ⁴³ . As for T cells, increased CD8+T cell infiltration was associated with impaired OS, and PD‐L1 expression was higher in patients with a higher CD8+T cell density, suggesting a possible mechanism of adaptive immune resistance. ⁴⁴ Christina et al. ⁴⁵ reported that the high CD8+T cell density was an adverse prognostic factor for gastric adenocarcinoma patients. However, certain contrasting results suggested that increased CD3+ and CD8+ T lymphocyte infiltrations are associated with improved survival status. ⁴⁶ The upregulation of PD‐1 and PD‐L1 is believed to promote T‐cell apoptosis in gastric adenocarcinoma. ⁴⁷ The association between T cell infiltration and the prognostic risk of gastric adenocarcinoma has been extensively introduced; however, the specific prognostic effects of T cells remain controversial, possibly because of the differences in identified T cell subtype markers and gastric cancer molecular subtypes among different studies.

At the end of this study, we predicted several inhibitors or binders based on prognostic risk genes as potential drugs for gastric adenocarcinoma treatment, and the molecular docking analysis showed that the drug ligand and the protein receptor bond stably in the form of hydrogen bonds. However, our understanding of the therapeutic potential of drugs for gastric adenocarcinoma is limited. In future studies, we will conduct animal experiments to explore whether the molecular mechanism of niacin and miconazole for gastric adenocarcinoma treatment is related to the expression of target genes.

5. CONCLUSIONS

In summary, this study proposed an eight‐gene signature related to drug metabolism as a potential biomarker to predict the prognostic risk of gastric adenocarcinoma patients. The predicted drugs (niacin and miconazole) can stably bind to target genes and have therapeutic potential for gastric adenocarcinoma patients. Additionally, we found a significant correlation between the tumor immune microenvironment and gastric adenocarcinoma prognosis. Our study helps to better understand the relationship between gastric adenocarcinoma prognosis, drug metabolism, and the immune microenvironment.

CONFLICT OF INTEREST

The authors declare that they have no competing interests.

AUTHOR CONTRIBUTION

Hongmei Yin and Qiong He conceived and designed the research, and Zhen Li and Jia Chen participated in the acquisition of data. Wanli Yang performed the analysis and interpretation of data. Jia Chen and Wanli Yang participated in the design of the study and performed the statistical analyses. Hongmei Yin, Qiong He, and Xiaobo Hu conceived the study, participated in its design and coordination, and helped draft the manuscript and revised the manuscript for important intellectual content. All authors read and approved the final manuscript.

Supporting information

Figure S1

Yin H‐M, He Q, Chen J, Li Z, Yang W, Hu X. Drug metabolism‐related eight‐gene signature can predict the prognosis of gastric adenocarcinoma. J Clin Lab Anal. 2021;35:e24085. 10.1002/jcla.24085

DATA AVAILABILITY STATEMENT

The raw data supporting the conclusions of this manuscript will be made available by the authors, without undue reservation, to any qualified researcher.