Integrative GEO and Mendelian randomization analysis reveals transcriptomic and lipidomic features of esophageal adenocarcinoma

Abstract

Background:

The incidence of esophageal adenocarcinoma (EA) has significantly increased in developed Western countries. Despite medical advancements, the prognosis remains poor, with a 5-year survival rate of less than 20%. By 2024, the global incidence is expected to reach 141,300 new cases annually, underscoring the urgent need to elucidate the mechanisms underlying EA pathogenesis to develop effective preventive and therapeutic strategies.

Methods:

To identify differentially expressed genes (DEGs) linked to EA, microarray datasets sourced from the Gene Expression Omnibus (GEO) database were scrutinized, incorporating 4 datasets that met the defined criteria. Using expression quantitative trait loci and Mendelian randomization (MR) analyses, the contribution of genetic factors to EA development was evaluated. Functional pathways were explored using Gene Ontology, Kyoto Encyclopedia of Genes and Genomes, and Gene Set Enrichment Analysis, which revealed enrichment in lipid metabolism. Consequently, Bayesian-weighted MR analysis was performed on 179 plasma lipid subgroups.

Results:

We identified 492 DEGs, 211 of which were downregulated and 281 were upregulated. The MR analysis identified 178 genes with significant causal effects on EA. Four co-expressed genes were ultimately determined: FZD2, KRT23, and CES1 were significantly upregulated in EA and positively associated with its occurrence, whereas ALDOC (aldolase, fructose-bisphosphate C) was inversely associated with EA risk. Elevated levels of sphingomyelins, sterol esters, diacylglycerols, and triacylglycerols were linked to a reduced risk of EA, whereas high levels of phosphatidylethanolamine correlated with a heightened risk.

Conclusions:

Integration of DEGs, expression quantitative trait loci, and lipidomics data provides robust insights into the molecular mechanisms of EA. These findings provide a promising foundation for the development of novel targeted therapies.

1. Introduction

Esophageal adenocarcinoma (EA) is a rapidly progressing and highly aggressive malignancy of the digestive tract that predominantly affects the distal esophagus. It originates from the dysplastic transformation of columnar epithelium into malignant tumors.^[1] Developed Western countries have witnessed a significant rise in EA incidence in recent years, coupled with a dismal prognosis evidenced by a 5-year survival rate of <20%.^[2] By 2024, the global incidence of EA is projected to increase to 141,300 new cases annually, posing significant challenges for its treatment.^[3] Hence, there is an urgent need to elucidate the mechanisms underlying EA pathogenesis in order to develop effective preventive or therapeutic strategies.

The existing body of research on etiological mechanisms and chemopreventive strategies for EA remains controversial. Gastroesophageal reflux disease and Barrett esophagus (BE) are recognized as the primary risk factors for EA. Chronic exposure to reflux of bile acids and gastric acid can induce oxidative DNA damage and cytokine storms, exacerbating esophageal mucosal injury.^[4] Proton pump inhibitors mitigate the damage caused by gastric acid reflux to some extent, but their efficacy in EA prevention remains debated.^[5,6] Moreover, nonsteroidal anti-inflammatory drugs (NSAIDs) inhibit the overexpression of COX-2 in BE/EA, curtailing chronic inflammatory responses and dysplastic transformation of epithelial cells. Aspirin has also been shown to modulate the oncogenic activation of nuclear factor-kappaB and the expression of CDX2 in BE patients, thereby inhibiting apoptosis and carcinogenesis.^[7] However, it is crucial to note that long-term NSAID use increases the risk of gastrointestinal ulcers, and the efficacy of different types and frequencies of NSAID use in preventing EA varies.^[8] Statins may reduce the incidence of EA by enhancing COX-2 activity, inhibiting Ras farnesylation, and curbing the activation of extracellular signal-regulated kinase and protein kinase B.^[9] Given the inherent confounding biases and serendipity of observational studies, further prospective research is required to substantiate these findings.

The molecular genetic mechanisms of EA primarily involve mutations that inactivate the p53 protein family, disrupt the cell cycle, and activate oncogenic signaling pathways, including receptor tyrosine kinase and transforming growth factor-beta (TGF-β) pathways.^[10] Targeted therapies such as trastuzumab and lapatinib have demonstrated potential in delaying disease progression in some patients, but the overall survival benefit remains modest, warranting further investigation.^[11]

Given the limitations and gaps in current research, this study sought to uncover differentially expressed genes (DEGs) linked to EA through the analysis of microarray datasets and plasma lipid phenotypes. Expression quantitative trait loci (eQTL) and Mendelian randomization (MR) analyses will be employed to evaluate the associations and causal relationships between these genes in EA pathogenesis. These methodologies offer robust statistical power and the ability to infer causality and reduce the confounding biases commonly observed in observational studies. Through Gene Ontology (GO) enrichment analysis, Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis, and Gene Set Enrichment Analysis (GSEA), we explored the potential functional pathways and mechanisms linked to these DEGs. We hope that these methods will elucidate the molecular underpinnings of EA, offering new insights for etiological research and identification of therapeutic targets.

2. Materials and methods

2.1. Data sources

In this study, gene expression datasets relevant to “esophageal adenocarcinoma” and “Homo sapiens” were retrieved from the Gene Expression Omnibus (GEO) database using microarray data analysis. The datasets were selected based on the following criteria: inclusion of both normal and EA samples, absence of any prior treatments or interventions (such as chemotherapy or immunotherapy), and the availability of raw data for public access. Four datasets that satisfied these criteria were included in the analysis.

2.2. Identification of DEGs

Preprocessing the datasets of esophageal tissue samples included in the study: GSE1420, GSE13898, GSE74553, and GSE26886 (Table 1).^[12–15] Following merging and batch correction of the datasets, differential analysis was conducted on 64 normal and 157 EA samples. The significance threshold for filtering DEGs using classical Bayesian data analysis was set at P < .05 and LogFC > 0.585.^[16] Data normalization and standardization were performed using the platform files in the respective datasets. Principal component analysis (PCA) with the “prcomp” function was performed to eliminate batch effects, thereby evaluating and validating key genes that distinguish EA samples from healthy controls.

Table 1.

Characteristics of the 4 GEO datasets.

GSE ID	Samples	Tissues	Platform	Experiment type	Last update date
GSE1420	9 cases and 9 controls	Esophagus	GPL96	Array	August 10, 2018
GSE13898	75 cases and 28 controls	Esophagus	GPL6102	Array	Febraury 15, 2013
GSE74553	52 cases and 8 controls	Esophagus	GPL17692	Array	December 06, 2018
GSE26886	21 cases and 19 controls	Esophagus	GPL570	Array	March 25, 2019

GEO = Gene Expression Omnibus.

2.3. eQTL and plasma lipidome exposure data

The eQTL localization of 5311 European individuals’ peripheral blood samples by the Westra HJ team provided comprehensive genetic data related to transcription and gene expression.^[17] Human plasma lipidomics data were obtained from the GWAS catalog database (id: GCST90277238–GCST90277416), covering genome-wide features of 7174 European individuals and encompassing 179 lipid species across 13 lipid categories.^[18] The P-value threshold for eQTL instrumental variables (IVs) was set at P < 5e-8, while for plasma lipid IVs, the P-value threshold was set at P < 1e-05 to obtain sufficient SNPs. Subsequent clumping removed the linkage disequilibrium (r² < 0.001, clumping distance = 10,000 kb) of the IVs. SNPs with low explanatory power and expression levels were excluded using F-statistics to avoid bias from weak IVs.^[19]

2.4. EA outcome data

Summary data for EA from GWAS were sourced from the MRC Integrative Epidemiology Unit database.^[20] The GWAS ID used was ebi-a-GCST003739, covering 21,271 European individuals from multiple centers (ncase = 4112, ncontrol = 17,159) with 12,911,041SNPs.^[21]

2.5. MR analysis

Five MR assessment methods based on different assumptions were applied, including MR-Egger, weighted median, inverse variance weighted (IVW), simple mode, and weighted mode. Heterogeneity was assessed using the Cochran Q test, and pleiotropy was examined using the MR-Egger intercept test. Additionally, MR-Pleiotropy Residual Sum and Outliers (MR-PRESSO) and MR-PRESSO Global tests were used for the secondary testing of outliers and horizontal pleiotropy during the analysis.^[22] Under the assumption of no pleiotropy, the IVW method had the most accurate causal estimation capability; thus, it was used as the primary method for interpreting the results.^[23] Genes meeting stringent criteria from the 3 methods (MR-Egger, weighted median, and IVW) with consistency in effect direction were considered to have significant causal effects. Leave-one-out analysis (LOO), scatter plots, and forest plots of single SNPs provided visual sensitivity analysis and assessed the robustness of the results. Subsequently, an intersection between DEGs results and MR (eQTL) results identified co-expressed upregulated and downregulated genes. All MR analyses adhered to assumptions of relevance, independence, and exclusivity.

2.6. GO/KEGG enrichment analysis

Functional annotation and pathway enrichment analysis of the identified genes were conducted using GO and KEGG to explore the potential mechanisms influencing the occurrence and development of EA, with a significance threshold set at P < .05.

2.7. GSEA enrichment analysis

GSEA analysis supplemented the above enrichment analyses, indicating the overall trend of co-expressed genes in relevant functional expression or pathways, rather than individual gene enrichment. Statistical significance was defined at P < .05.

2.8. Statistical software

All analyses were conducted using R software (version 4.3.3; R Foundation for Statistical Computing, Vienna, Austria). The “limma” package version 3.58.1 (for DEGs filtering), “pheatmap” package version 1.0.12 (for volcano plots and heatmaps). The “TwoSampleMR” version 0.5.11 (for MR analysis; University of Bristol, Bristol, United Kingdom). The “clusterProfiler” package v4.10.1 (for GO/KEGG enrichment analysis; Southern Medical University, Guangzhou, China).^[24]

3. Results

3.1. PCA analysis: batch effect correction

Figure 1A (before batch correction) shows the PCA results with distinct clustering, indicating significant batch effects primarily driven by technical variability. Figure 1B (after batch correction) demonstrates successful batch effect mitigation, as evidenced by the substantial dataset overlap, enabling accurate biological interpretation and meaningful cross-study comparisons.

Figure 1.

Normalization and standardization of GEO data. (A) Before batch correction; (B) after batch correction. GEO = Gene Expression Omnibus.

3.2. DEGs identification

After Bayesian P-value correction, a total of 492 DEGs were identified, including 211 downregulated DEGs and 281 upregulated DEGs (Table S1, Supplemental Digital Content, https://links.lww.com/MD/P822). The heatmap of DEGs illustrates the top 50 most significantly upregulated and downregulated DEGs after correction (Fig. 2A). The volcano plot in Figure 2B shows the overall performance of the differential gene dataset.

Figure 2.

Visualization of differentially expressed genes. (A) Heatmap of differentially expressed genes. (B) Volcano plot of differentially expressed genes. Blue represents downregulated gene expression, while red represents upregulated gene expression.

3.3. MR analysis of eQTL

The Table S2, Supplemental Digital Content, https://links.lww.com/MD/P822 presents eQTL data that meet the criteria for IVs, with all F-statistics for IVs exceeding 10. A total of 5422 eQTL phenotypes were included in the valid analysis (Table S3, Supplemental Digital Content, https://links.lww.com/MD/P822), and underwent tests for pleiotropy and heterogeneity (Tables S4 and S5, Supplemental Digital Content, https://links.lww.com/MD/P822). Following filtration for consistent effect direction across multiple MR methods and exclusion of pleiotropy, 178 genes were identified as having significant causal effects on EA (Table S6, Supplemental Digital Content, https://links.lww.com/MD/P822). Figures 3A and B illustrates the DEGs and 4 co-expressed genes from the MR analysis.

Figure 3.

Co-expressed genes. (A) Three upregulated co-expressed genes. (B) One downregulated co-expressed gene.

The IVW results indicated that Frizzled Class Receptor 2 (FZD2) (OR = 1.136; 95% CI: 1.011–1.278; P = .032), Keratin 23 (KRT23) (OR = 1.158; 95% CI: 1.041–1.288; P = .040), and carboxylesterase 1 (CES1) (OR = 1.188; 95% CI: 1.027–1.374; P = .020) were co-expressed genes that were upregulated in EA patients and significantly positively associated with the occurrence of EA. Conversely, aldolase, fructose-bisphosphate C (ALDOC) (OR = 0.819; 95% CI: 0.693–0.968; P = .019) is a downregulated co-expressed gene significantly negatively correlated with EA occurrence (Fig. 4).

Figure 4.

MR forest plot of co-expressed genes. nSNP = number of single nucleotide polymorphism; OR = odds ratio; 95% CI = 95% confidence interval.

Sensitivity analyses further corroborate the reliability of these results (Table S8, Supplemental Digital Content, https://links.lww.com/MD/P822), with MR-Egger intercept tests and MR-PRESSO Global tests revealing no evidence of pleiotropy. Additionally, the Cochran Q test revealed no evidence of heterogeneity. Robust evidence was provided by visual inspections from scatter plots, LOO analyses, and single SNP forest plots (Figures S1–S4, Supplemental Digital Content, https://links.lww.com/MD/P821). Figure 5 depicts the chromosomal location information of the co-expressed genes.

Figure 5.

Circos plot of co-expressed genes. Chr = chromosome.

3.4. GO and KEGG enrichment analysis of co-expressed genes

The GO enrichment analysis revealed significant enrichment of processes related to lipid metabolism, cholesterol regulation, and neural development. Key cellular components include vesicle membranes and immune response granules, whereas crucial molecular functions involve various enzymatic activities and protein interactions, particularly within the wingless/integrated (Wnt) signaling pathway (Fig. 6A, Table S9, Supplemental Digital Content, https://links.lww.com/MD/P822). KEGG enrichment analysis revealed the significant enrichment of multiple metabolic and disease-related pathways within the dataset. Notably, the key gene ALDOC was significantly enriched in carbohydrate metabolism pathways, including the pentose phosphate pathway, fructose and mannose metabolism, and glycolysis/gluconeogenesis. Additionally, enrichment was identified in pathways associated with basal cell carcinoma, amino acid biosynthesis, drug metabolism, Staphylococcus aureus infection, and the hypoxia-inducible factor 1 (HIF-1) signaling pathway (Fig. 6B, Table S10, Supplemental Digital Content, https://links.lww.com/MD/P822).

Figure 6.

(A) GO enrichment analysis of candidate hub genes. (B) KEGG enrichment analysis of candidate hub genes. GO = Gene Ontology, KEGG = Kyoto Encyclopedia of Genes and Genomes.

3.5. GSEA enrichment analysis

The aforementioned study suggests that the co-expressed upregulated CES1 gene is associated with lipid metabolism pathways in EA. Therefore, GSEA enrichment analysis was used to further investigate its overall expression and pathway characteristics. In the high CES1 expression group, the top 5 biological activities were arachidonic acid metabolism, insulin signaling pathway, linoleic acid metabolism, olfactory transduction, and oocyte meiosis (Fig. 7C), and the top 5 pathways were epidermal cell differentiation, epidermis development, keratinization, keratinocyte differentiation, and cornified envelope (Fig. 8C). In the low CES1 expression group, the top 5 significant biological activities were complement and coagulation cascades, extracellular matrix (ECM)-receptor interaction, focal adhesion, N-glycan biosynthesis, and systemic lupus erythematosus (Fig. 7D). The top 5 pathways were ECM organization, collagen-containing ECM, endoplasmic reticulum lumen, extracellular structure organization, and ECM structural constituents (Fig. 8D). GSEA enrichment analysis of the other 3 co-expressed genes is shown in Figures 7 and 8.

Figure 7.

GSEA of expression profiles. (A) High expression of ALDOC; (B) low expression of ALDOC; (C) high expression of CES1; (D) low expression of CES1; (E) low expression of KRT23; (F) high expression of FZD2. ALDOC = aldolase, fructose-bisphosphate C, CES1 = carboxylesterase 1, GSEA = Gene Set Enrichment Analysis, KRT23 = keratin 23.

Figure 8.

Gene Set Enrichment Analysis (GSEA) pathway enrichment analysis. (A) ALDOC high expression group; (B) ALDOC low expression group; (C) CES1 high expression group; (D) CES1 low expression group; (E) FZD2 high expression group; (F) FZD2 low expression group; (G) KRT23 high expression group; (H) KRT23 low expression group. ALDOC = aldolase, fructose-bisphosphate C, CES1 = carboxylesterase 1, FZD2 = frizzled class receptor 2, KRT23 = keratin 23.

3.6. MR analysis of plasma lipids

Analysis of 179 plasma lipids revealed that 11 lipids had a significant causal relationship with EA. After Bayesian weighted MR correction, 7 plasma lipid phenotypes retained their significance (Fig. 9). Six lipids showed a negative association with EA: sterol ester (SE) (27:1/22:6) levels (OR = 0.773; 95% CI: 0.675–0.884; adjusted-P < .001), diacylglycerol (DAG) (18:1_18:1) levels (OR = 0.809; 95% CI: 0.699–0.936; adjusted-P = .036), DAG (18:1_18:3) levels (OR = 0.765; 95% CI: 0.646–0.906; adjusted-P = .018), sphingomyelin (SM) (d34:0) levels (OR = 0.872; 95% CI: 0.769–0.987; adjusted-P = .021), triacylglycerol (TAG) (50:5) levels (OR = 0.837; 95% CI: 0.736–0.952; adjusted-P = .007), and TAG (56:6) levels (OR = 0.836; 95% CI: 0.739–0.945; adjusted-P = .046). Phosphatidylethanolamine (PE) (O-18:1_18:2) levels (OR = 1.243; 95% CI: 1.060–1.457; adjusted-P = .004) showed a positive association with the risk of EA. Table S11, Supplemental Digital Content, https://links.lww.com/MD/P822 presents the results of the sensitivity analysis, where MR-PRESSO identified an outlier only in PE (O-18:1_18:2) levels (rs118048608). After excluding the outlier, no significant results showed evidence of heterogeneity or pleiotropy. The robustness of the results was further supported by visualizations in scatter plots, LOO, and single SNP forest plots (Figures S5–S15, Supplemental Digital Content, https://links.lww.com/MD/P821).

Figure 9.

MR forest plot of plasma lipidomics. MR = Mendelian randomization, nSNP = number of single nucleotide polymorphisms; OR = odds ratio; 95% CI = 95% confidence interval.

4. Discussion

The incidence of EA, a gastrointestinal malignancy with low survival rates, has rapidly increase in Western countries.^[3] Notably, current preventive measures have demonstrated suboptimal effectiveness, and the molecular mechanisms underlying the disease remain incompletely understood. This study analyzed 4 microarray datasets from the GEO database (64 healthy controls and 157 EA patients)^[12–15] alongside eQTL data.^[17] By comparing DEGs with the MR analysis results, we identified 4 co-expressed target genes causally linked to EA. Among them, CES1, FZD2, and KRT23 are upregulated genes positively associated with EA risk, while ALDOC is a downregulated gene negatively correlated with EA risk.

GO enrichment analysis revealed significant involvement in lipid metabolism, cholesterol regulation, and neural development, with a particular emphasis on vesicle membranes and Wnt signaling pathways. KEGG analysis highlighted crucial pathways such as carbohydrate metabolism, basal cell carcinoma, amino acid biosynthesis, drug metabolism, and the HIF-1 signaling pathway. Our research focuses on these gene targets and underscores the pivotal role of lipid metabolic pathways in the pathogenesis of EA.

The CES1 gene, located on chromosome 16, is a member of the α, β-hydrolase fold family and is predominantly found in the liver and gastrointestinal organs.^[25] Its primary functions include lipid metabolism, drug metabolism, and enzyme induction, which align with the findings of our enrichment analysis.^[26] Capece D et al^[27] discovered that CES1 in colorectal cancer (CRC) with consensus molecular subtype 4 can be driven by the nuclear factor-kappaB signaling pathway, interfering with TAG metabolism. Inhibiting the CES1 target can effectively reduce the risk of obesity-related tumors; CRC mice treated with CES1 inhibitors showed significantly reduced tumor growth, a conclusion also confirmed in liver cancer.^[28] However, the mechanism of CES1 in EA is not yet clear. This might be related to increased bile acid reflux, which activates the NF-ΚB signaling pathway, driving CES1 overexpression and inducing mucin overexpression in the esophageal mucosa, leading to carcinogenesis.^[7,29]

FZD2, located on chromosome 17, is a crucial protein involved in Wnt signal transduction, with primary biological functions in embryonic development and tumor microenvironment regulation. Notably, abnormal activation of the Wnt signaling pathway occurs as early as the BE stage, with nuclear accumulation of β-catenin contributing to the progression from BE to EA.^[10,30] Continuous activation of Wnt also promotes EA metastasis, consistent with our findings that FZD2 is highly expressed in EA and primarily involved in Wnt signaling and neural system development. FZD2 may also regulate cell proliferation or apoptosis by activating the Notch signaling pathway, driving the self-renewal and tumorigenicity of EA stem cells.^[10,31,32]

KRT23, also located on chromosome 17, is involved in the regulation of the TGF-β and SMAD signaling pathways.^[33] Our results also emphasize its importance in intermediate filaments and intracellular structures. Limited research suggests that KRT23 expression is associated with CRC and gastric cancer. Zhang et al^[34] found that KRT23 overexpression promotes CRC through human telomerase reverse transcriptase, while KRT23 deficiency enhances melatonin’s inhibitory effect on gastric cancer cells.^[35] However, the impact of KRT23 on EA remains unclear, potentially linked to disruptions in the TGF-β signaling pathway, which interfere with cell cycle arrest and impair SMAD-dependent transcriptional regulation, exacerbating EA cell proliferation and invasion.^[36]

ALDOC is a pivotal enzyme in the glycolytic pathway, predominantly expressed in the brain, and located on human chromosome 17. Research on ALDOC is relatively scarce. Zhu L et al^[37] discovered that alterations in glycolysis are associated with the tumor microenvironment and drug sensitivity in EA. This aligns with our findings that implicate ALDOC in glycolysis, fructose metabolism, and carbohydrate metabolism. Intriguingly, obesity, a known risk factor for EA, is closely linked to these metabolic pathways.^[38] Our study also highlights the role of the HIF-1 signaling pathway in the biological processes involving ALDOC. Sethi N et al^[39] demonstrated that missense mutations in TP53 induce hypoxic signaling of HIF-1α in the early stages of primary gastroesophageal adenocarcinoma, promoting oncogenesis and poor prognosis. These insights may shed light on the potential mechanisms underlying the relationship between ALDOC and EA.

A Bayesian weighted MR analysis of 179 plasma lipid subgroups was conducted to identify plasma lipid characteristics significantly associated with EA. The analysis revealed that elevated levels of SM and SE, as well as DAGs and TAGs, were linked to a reduced risk of EA, while high levels of PE are associated with an increased risk. These findings are consistent with Molendijk J et al^[40] multi-omics study, which reported a progressive decline in TAGs and an increase in PE and phosphatidylcholine in esophageal tissues from BE to EA. Similarly, Abbassi-Ghadi N et al^[41] identified a significant rise in PE levels in EA through their phospholipidomic research.

The direct effects of SEs and DAGs on EA remain unclear, potentially due to their roles in cholesterol homeostasis within TAGs. DAGs have shown potential in inhibiting body fat accumulation and reducing postprandial serum TAG, cholesterol, and glucose levels, thereby exerting anti-obesity effects.^[42] A decrease in fatty acid saturation in EA might be one of the primary carcinogenic mechanisms.^[40] Polyunsaturated lipids play a crucial role in the composition of cell membrane fatty acids, maintaining membrane stability and fluidity,^[43] influencing inflammatory responses,^[44] and signal transduction.^[45] NF-ΚB can directly regulate the expression of lipid desaturases, thereby affecting NF-ΚB signaling, which is pivotal in the pathogenesis of EA.^[29]

This study utilized MR to causally validate findings from DEGs to transcriptomics and lipidomics, uncovering new therapeutic targets and metabolic mechanisms for EA. Compared to previous observational and cross-sectional studies, the integration of multi-omics and genetic analysis effectively addresses the biases due to confounding factors, yielding more reliable results. However, it is important to note that the study population was derived from European cohorts, raising concerns about the generalizability of the findings to other populations. Furthermore, the roles of certain transcriptomic genes and lipid phenotypes in EA remain unclear, underscoring the need for further basic research to corroborate these results.

5. Conclusion

This study integrated genomic and lipid pathway analyses to reveal the driving factors of EA and provides causal inferences. CES1, FZD2, and KRT23 were highly expressed and pathogenic in EA, whereas ALDOC showed a downregulated negative correlation. Various expression and pathway enrichment analyses have indicated that DEGs are involved in glycolysis/gluconeogenesis, lipid metabolism/cholesterol regulation, drug metabolism, HIF-1 hypoxia injury, and Wnt signaling pathways. Plasma lipidomic analysis suggested that elevated levels of 6 lipids (SE (27:1/22:6), DAG (18:1_18:1), DAG (18:1_18:3), SM (d34:0), TAG (50:5), and TAG (56:6)) were protective against EA, while increased levels of PE (O-18:1_18:2) heighten the risk of developing EA. These findings provide new insights into the pathogenic target genes and mechanisms of EA, facilitating the development of preventive and therapeutic strategies.

Acknowledgments

The authors express their gratitude to the staff and participants of eQTLGen, GEO, and the GWAS Catalog for making the GWAS data publicly accessible. Additionally, we extend our thanks to all coauthors for their contributions to this research.

Author contributions

Writing – original draft: Yue Wang.

Writing – review & editing: Hongbin Liu.