Pan-cancer and COAD-specific analyses were performed using systematically curated public omics datasets. RNA sequencing (RNA-seq) profiles and annotated clinical metadata were acquired from TCGA database (https://portal.gdc.cancer.gov/) and GEO repositories (https://www.ncbi.nlm.nih.gov/geo/). Pan-cancer cohorts included multiple datasets, including GSE103479/GSE104645/TCGA-GTEx/GSE71187/GSE9893/GSE107943/GSE63624/GSE10846/GSE181063/GSE54236/GSE28735/GSE17679/GSE190113/GSE98394/GSE99898/GSE28541/GSE84426 (33). Non-neoplastic tissues and samples with censored survival outcomes (overall survival < 30 days or missing clinical data > 20%) were excluded. All gene annotations were standardized to gene symbols to ensure analytical consistency.

Single-cell RNA sequencing (scRNA-seq) data were employed to characterize tumor microenvironment heterogeneity and cellular ecosystem dynamics. Two datasets (EMTAB8107 and GSE166555) were analyzed using standardized preprocessing workflows (cells retained if 200 < nFeature_RNA < 6,000 and mitochondrial gene percentage < 20%), which included quality control, dimensionality reduction, cluster annotation, and identification of malignant cells. All analytical procedures adhered to TISCH2 standards to ensure reproducibility (34).

Spatially resolved transcriptomic profiles were acquired using the 10X Visium Spatial Transcriptomics platform (10x Genomics, available at: https://www.10xgenomics.com/resources/datasets). The data were derived from FFPE-preserved colorectal carcinoma sections (samples CRC1, CRC2, and CRC3), facilitating the integrated analysis of histopathological structure and molecular signatures within the tumor microenvironment.

All data analyzed in this study were obtained from publicly available sources, including previously published studies and established databases.

A curated set of 160 lactylation-related genes (LRGs) was compiled through systematic integration of GSEA Molecular Signatures Database, KEGG pathway annotations, and manual curation of recent literature (Supplementary Dataset 1). Within the COAD cohort, univariate Cox proportional hazards regression analysis was initially performed on all 160 genes without multiple testing correction as an exploratory screening step to generate a broad candidate pool for subsequent rigorous validation. This approach was deliberately adopted to preserve biologically plausible candidates that might be excluded under stringent correction at the screening stage. Core prognostic candidates were subsequently refined through a multi-algorithm consensus framework specifically applied to the COAD cohort. Five independent machine learning approaches—Lasso regression, Ridge regression, Elastic net, Stepwise Cox regression, and CoxBoost—were applied to identify genes with consistent prognostic significance. Risk scores were calculated as the linear combination of gene expression values weighted by algorithm-derived coefficients. Model performance was evaluated using time-dependent receiver operating characteristic (ROC) analysis at 1-, 3-, and 5-year survival time points. An optimal risk stratification cutoff was determined using maximally selected rank statistics (via the survminer R package), with statistical significance of survival differences assessed by log-rank testing. Differential expression patterns of selected genes between risk groups were visualized using heatmaps (pheatmap R package). All analyses were performed in R with packages survival, glmnet, CoxBoost, and timeROC.

Expression matrices were processed using tumor-type-specific normalization pipelines. For pan-cancer analyses, z-scores were calculated based on cohort-specific means and standard deviations. Differential expression between tumor and normal samples was assessed using non-parametric tests with false discovery rate control (limma v3.56.2; |log₂FC| > 1, P < 0.05).

Diagnostic accuracy was evaluated via receiver operating characteristic (ROC) analysis. Area under the curve (AUC) with 95% confidence intervals and smoothed ROC curves were generated to determine discriminatory power.

Prognostic associations were assessed using Cox proportional hazards models, with verification of proportional hazards assumptions. Univariate analysis provided hazard ratios (HR) and 95% confidence intervals (95% CI). External validation was conducted on independent GEO cohorts using the same modeling framework.

For the COAD meta-analysis, a fixed-effects model was employed to combine HR, with inverse variance weighting. Genes significantly associated with improved survival (HR < 1) were considered potential tumor suppressors, while those associated with poorer survival (HR > 1) were considered potential oncogenes. Multivariable Cox models incorporating clinical covariates were developed and visualized using forest plots.

The immune microenvironment was characterized using three complementary approaches: single-sample Gene Set Enrichment Analysis (ssGSEA; GSVA v1.48.2, kcdf=“Gaussian”), xCell deconvolution, and CIBERSORTx (LM22 signature, 1,000 permutations) (35, 36). Patients were stratified into expression-based quartiles for further analysis. Infiltration scores were filtered and subsequently clustered using unsupervised hierarchical clustering, which revealed conserved immune subtypes across malignancies.

Functional annotation was carried out using integrated Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), and Gene Set Enrichment Analysis (GSEA; clusterProfiler v4.8.2, Org.Hs.eg.db v3.17.0; 1,000 permutations, gene set size 15–500, q < 0.05) (37–39). The TCGA-COAD cohort was divided into high- and low-expression groups based on median expression of target genes for subsequent functional investigations. Differential gene expression analysis was performed using the “limma” package, with pre-defined thresholds for fold-change and statistical significance (40).

Functional states of malignant cells were further characterized using the CancerSEA database to explore various oncogenic phenotypes (41). And, pathway activity was quantified using z-score-based algorithms, and functional state scores were computed via the “GSVA” package (42). Normalized scores were standardized and subjected to correlation analysis to evaluate associations between gene expression and functional states.

Single-cell transcriptomic profiling was performed using the Seurat package following standard analytical workflows (43). Cells were subjected to quality control based on thresholds for unique molecular identifiers and mitochondrial gene content. Preprocessing included normalization, dimensionality reduction via principal component analysis, and batch effect correction. Cells were stratified into high- and low-expression groups based on median expression levels, and cluster proportions were compared across conditions (44).

Cell-cell communication analysis was performed to infer ligand-receptor interaction networks. Interaction probabilities and communication strengths were quantified between cellular subpopulations. Directional signaling flux and core pathways were identified based on signal input/output patterns and centrality metrics (45).

Spatially resolved transcriptomic profiling was employed to investigate the topographic distribution and functional relevance of target genes within the tumor microenvironment (46). Cell type deconvolution of spatial transcriptomics data was performed using a reference-based Bayesian framework. Quality control was conducted according to standard scRNA-seq processing criteria.

A gene signature score matrix was generated by averaging expression of top marker genes from single-cell references. Cellular abundance gradients across spatial domains were quantitatively assessed and visualized using color gradients to represent enrichment intensity. Histological regions were classified as malignant based on predefined cellular composition thresholds, facilitating patient stratification into normal-like and mixed subgroups. Differential expression between groups was evaluated using non-parametric tests with multiple testing correction.

The Human Protein Atlas database (https://www.proteinatlas.org) provides comprehensive mRNA and protein expression data across 44 human tissues, along with antibody-based protein profiles that include expression levels and subcellular localization information (47).

To identify and predict interactions among genes or proteins, we utilized the STRING database (http://string-db.org/). Additionally, BioGRID version 4.4 (https://thebiogrid.org/) was employed as a key resource for accessing high-quality, curated, and non-redundant physical and genetic protein-protein interactions (PPIs).

A protein-protein interaction network was constructed using a combined score cutoff of > 0.9. Based on this network, the expression of AARS2-related proteins was further analyzed across multiple cancer types.

Immunohistochemical analysis was performed to assess AARS2 protein expression in COAD tissue specimens. The study protocol received approval from the Ethics Committee of the Second Hospital of Lanzhou University (Approval No. 2025A-1084). Tissue samples were collected from 12 patients diagnosed with COAD, with adjacent normal colon mucosa serving as controls. All specimens were promptly snap-frozen in liquid nitrogen and transferred to the laboratory for subsequent analysis. Comprehensive clinical metadata for the cohort—including gender/age distribution, TNM staging, histological grade, prior treatment history, and MSI status (MSI-H versus MSS)—along with correlation analyses evaluating associations between AARS2 protein expression levels and key clinicopathological parameters (T stage, N stage, M stage, and tumor grade), are fully documented in the Supplementary Materials (Supplementary Table 1).

For IHC, a primary antibodys AARS2 (Proteintech, 22696-1-AP) and cGAS (Proteintech, 26416-1-AP) were applied at a dilution of 1:50. Protein expression was quantitatively evaluated using the ImageJ IHC profiler plugin. The assessment integrated both staining intensity and the proportion of immunopositive areas. The percentage of positive staining relative to the total tissue area (distribution) was calculated, and a final expression score was derived by multiplying the intensity value by the distribution percentage.

Based on the median expression score across all samples, specimens were classified into high- and low-expression groups: scores above the median were defined as high expression, while those equal to or below the median were considered low expression. This classification was utilized in subsequent univariate and multivariate regression analyses. To ensure objectivity, three independent pathologists performed the evaluations blinded to all clinical data.

To investigate the functional role of AARS2 in COAD at the cellular level, human COAD cell lines (HCT 116) were used in this study. All cell lines were obtained from authenticated cell banks and routinely tested to exclude mycoplasma contamination.

Cells were cultured in the recommended complete growth medium (DMEM) supplemented with 10% fetal bovine serum and 1% penicillin–streptomycin, and maintained at 37°C in a humidified incubator with 5% CO₂. Cells within 20 passages were used for all subsequent experiments to ensure reproducibility and stability of cellular phenotypes.

Western blot analysis was performed to evaluate the protein expression levels of AARS2 in COAD cells, thereby supporting the transcriptomic findings at the protein level.

Cells were washed twice with ice-cold phosphate-buffered saline (PBS) and lysed in RIPA buffer supplemented with protease and phosphatase inhibitors. Cell lysates were clarified by centrifugation at 12,000 × g for 15 min at 4 °C. Protein concentrations were determined using a BCA protein assay kit. Equal amounts of protein (20–40 μg) were separated by SDS–PAGE and transferred onto PVDF membranes.

Membranes were blocked with 5% non-fat milk and incubated with primary antibodies overnight at 4 °C. After incubation with horseradish peroxidase-conjugated secondary antibodies, protein bands were visualized using enhanced chemiluminescence (ECL), with GAPDH and Tubulin used as an internal loading control.

Total RNA was extracted from HCT116 cells using TRIzol reagent. RNA concentration and purity were assessed spectrophotometrically (A260/A280 ratio 1.8–2.0). First-strand cDNA was synthesized from 1 μg total RNA using a commercial reverse transcription kit. Quantitative real-time PCR was performed in triplicate with SYBR Green chemistry on a real-time PCR system. The mRNA expression levels of AARS2, CCL5, CXCL10, and IFNB1 were quantified using gene-specific primers (sequences provided in Supplementary Table 2). Amplification conditions: 95°C for 30 s; 40 cycles of 95 °C for 10 s and 60°C for 30 s. GAPDH was used as the endogenous reference gene for normalization of target gene expression. Relative expression was calculated using the 2^−ΔΔCt method. Amplification specificity was confirmed by melting curve analysis. All procedures adhered to MIQE guidelines.

Extracellular lactate levels in cell culture supernatants were measured to evaluate lactate accumulation in COAD cells associated with AARS2 expression.

Cells were seeded in 6-well plates and maintained under standard culture conditions. Culture supernatants were harvested at 24 h and 48 h post-seeding, centrifuged (e.g., 300 ×g, 5 min, 4°C) to remove cellular debris, and lactate concentrations were quantified using a L-Lactic Acid (LA) Colorimetric Assay Kit (Elabscience, catalogue no.: E-BC-K130-M) following the manufacturer’s protocol. Lactate levels were calculated from a standard curve and normalized to total protein content (determined by BCA assay) or cell number.

Statistical analyses were conducted using R software (version 4.3.1) with appropriate packages. Differences in gene expression among spatially defined regions (tumor, normal, and mixed) were assessed using Wilcoxon rank-sum tests. For single-cell transcriptomic data, the Kruskal–Wallis H test was used to compare gene expression across cell types, and Spearman’s correlation was applied to evaluate associations between immune cell proportions and gene expression levels. Survival analysis was performed using Kaplan–Meier curves with log-rank tests. Univariate and multivariate Cox proportional hazards models were used to identify independent prognostic factors, with HR and 95% confidence intervals reported. A p-value < 0.05 was considered statistically significant.

Univariate Cox survival analysis was performed on the 160 lactylation-related genes across pan-cancer cohorts as an initial screening step to generate a candidate pool. Forest plots of hazard ratios (HR) and 95% confidence intervals identified 25 genes associated with overall survival in COAD (Figure 1A). Critically, these 25 genes represented an exploratory candidate pool rather than definitive prognostic markers. The complete statistical results for all 33 genes reaching nominal significance (p < 0.05) in the univariate Cox analysis—including gene ID, HR, HR.95L, HR.95H, and raw p-value—are provided in Supplementary Table 3 for full transparency. Furthermore, we systematically characterized the RNA expression profiles of the aforementioned 160 genes across 30 distinct cancer types, leveraging a comprehensive integration of TCGA and GTEx datasets (Figure 1B). After successfully identifying 25 candidate genes associated with poorer OS in COAD, we further systematically evaluated the prognostic value of this gene set across a pan-cancer cohort encompassing 30 different cancer types. The analysis revealed that these candidate genes exhibit consistent prognostic patterns in several malignancies, including ACC, COAD, KIRC, KIRP, and LGG, where they were significantly associated with poorer outcomes in OS, DFI, DSS, and PFI, indicating a role as risk factors. In contrast, in tumors such as CESC and MESO, the same genes appeared to function as protective factors, correlating with more favorable clinical outcomes (Figures 1C–F). We next examined the expression patterns of these genes by comparing cancerous tissues with normal tissues across each cancer type, as well as by evaluating differences between tumor tissues and adjacent non-tumor tissues (Figures 1G, H). We also employed four distinct algorithms (Z-score, GSVA, ssGSEA) and PLAGE) to evaluate pathway activity associated with these core genes. The results indicated significantly altered pathway activity in a substantial number of tumor tissues compared to normal tissues (Figures 1I–L). Subsequently, we investigated the statistical correlation between promoter methylation levels and expression levels of these candidate genes associated with poorer OS in COAD across various cancer types, as well as compared their differences between tumor samples and normal controls (Figures 1M, N).

To develop robust prognostic models, we evaluated multiple algorithms—including regularized regression methods (Ridge, Lasso, Elastic_net with alpha values spanning 0.1–0.9), stepwise Cox models (Stepcox_forward, Stepcox_backward, Stepcox_both), and CoxBoost. We assessed their performance using the average area under the curve (AUC) at 1-, 3-, and 5-year time points across cancer types.Among these algorithms, the Stepcox_forward model demonstrated competitive performance: it showed relatively higher average AUC values in most cancer types and at most time points (Figure 2A). For example, the overall average AUC of Stepcox_forward across cancer types was 0.653—exceeding the average AUCs of other methods (e.g., CoxBoost: 0.639, Ridge: 0.632, Lasso: 0.627, and Elastic_net variants). Later, we utilized regression coefficient heatmaps to clarify the relative contributions of the 25 genes across distinct models (Figure 2B). Remarkably, AARS2 was identified as the gene with the largest regression coefficient across a wide range of algorithms, including Lasso, Ridge, Elastic_net with alpha values spanning from 0.1 to 0.9, stepwise Cox regression models (Stepcox_forward, Stepcox_backward, Stepcox_both), and CoxBoost. AARS2 was subsequently identified as the key gene for further analysis. Specifically, meta-analysis revealed that high AARS2 expression was significantly associated with poorer survival in COAD patients (pooled HR = 1.17, 95% CI: 1.03–1.33, p < 0.01; Supplementary Figure 1A).

To explore the potential role of AARS2 in carcinogenesis, we analyzed its mRNA expression in COAD using datasets from TCGA and GEO (GSE71187). Analysis of TCGA data indicated that AARS2 expression was significantly upregulated in COAD tissues relative to normal samples (P = 0.004; Figure 2C). This result was further supported by paired analysis, which also demonstrated elevated AARS2 expression in tumor tissues compared with matched adjacent non-tumor tissues (P = 0.029; Figure 2D). Consistent upregulation of AARS2 was also observed in independent validation cohorts from GEO and TCGA-GTEx (all P < 0.01; Figures 2E, F). These findings imply that AARS2 overexpression may contribute to the development of COAD.

Given the upregulated expression of AARS2, we further investigated its clinical relevance in COAD. Prognostic assessment and univariate Cox regression analyses were conducted using the TCGA cohort. Kaplan-Meier curves demonstrated that patients with high AARS2 expression exhibited significantly worse outcomes in overall survival (OS, P = 0.004), progression-free interval (PFI, P < 0.001), and disease-specific survival (DSS, P = 0.004) compared to those with low expression (Figures 2G–I). Consistent findings were observed across multiple independent validation datasets (Supplementary Figures 1B–G). Furthermore, univariate Cox regression confirmed AARS2 expression, T stage, and overall stage as independent prognostic factors for COAD (all P < 0.05; Figure 2J), underscoring the role of AARS2 as an independent predictor of patient survival. Subsequently, analysis of the GSE103479 and GSE104645 datasets revealed numerous differentially expressed genes, particularly those upregulated, which may be associated with AARS2-related functions and warrant further investigation (Figures 2K, L).

Through pan-cancer analysis of TCGA data, we observed significant differences in AARS2 expression between tumor and normal tissues across most cancer types (P < 0.05; Figure 3A). Overall, AARS2 expression showed substantial variation with a general trend toward overexpression in tumors. Specifically, AARS2 was significantly upregulated in BLCA, BRCA, CESC, CHOL, COAD, ESCA, KICH, KIRC, KIRP, LIHC, LUAD, LUSC, PCPG, PRAD, READ, STAD, and UCEC, while it was downregulated in THCA.

To assess the clinical relevance of AARS2, univariate Cox regression analysis was performed to estimate HR and 95% CI for overall survival (OS), disease-free interval (DFI), and progression-free interval (PFI) (Figures 3B–D). The results indicated that elevated AARS2 expression was associated with poorer survival outcomes across these endpoints. Moreover, as summarized in Figure 3E, AARS2 consistently functioned as a risk factor across multiple external datasets, further supporting that high AARS2 expression correlates with reduced survival. Together with ROC curve analysis demonstrating its accuracy in discriminating tumor from normal samples (Figure 3F), exemplified in COAD where AARS2 achieved an AUC of 0.746 (Figure 3G). These findings underscore the potential of AARS2 as both a prognostic and diagnostic biomarker. By integrating single-cell RNA sequencing data from the TISCH database (datasets EMTAB8107 and GSE166555), quantitative analysis demonstrated that the AARS2 gene displays pronounced cellular heterogeneity within COAD. This proliferation marker was specifically upregulated in the malignant proliferation subpopulation, with lower but detectable expression levels also observed in endothelial cells and fibroblasts (Figure 3H). Survival analysis summarized by heatmaps indicated that AARS2 serves as a risk factor in multiple external datasets, with patients exhibiting high AARS2 expression demonstrating poorer survival outcomes in ACC, COAD, KIRC, and KIRP. However, in certain tumor types, we observed the opposite trend, which may be attributable to their distinct biological contexts. For instance, elevated AARS2 expression was associated with a protective effect in PAAD and UVM (p < 0.001), while in several other cancers, no significant correlation with survival was detected. These findings underscore the context-dependent heterogeneity of AARS2’s prognostic impact across cancer types and suggest that further investigation is warranted to elucidate its precise mechanistic roles in different tumor environments (Figures 3I, J). Further links AARS2 expression to immunogenicity and DNA damage scores, providing more evidence for the role of AARS2 in immune regulation and genome stability.

Finally, single-cell and spatial transcriptomic analyses showed that AARS2 was predominantly expressed in malignant cells and was enriched in malignant microregions of tumor tissues (Supplementary Figures 2A, B). These findings highlight its potential role in shaping the tumor microenvironment and further highlight its importance in cancer biology.

Next, we explored its expression profile in CRC (colorectal cancer) from both transcriptomic and spatial transcriptomics perspectives.

Following quality control, batch effect correction, and data normalization, the UMAP visualization clearly delineated cell clusters obtained from 21 COAD patients. The cells were classified into three major categories: malignant, stromal, and immune cells (Figure 4A). Subsequent subclustering identified distinct populations, including malignant cells, epithelial cells, fibroblasts, endothelial cells, B cells, exhausted CD8⁺ T cells (CD8Tex), helper T cells (Th), myofibroblasts, enteric glial cells, M1 and M2 macrophages, monocytes, mast cells, and plasma cells (Figure 4B). The UMAP visualization revealed prominent PAQR5 expression primarily within malignant tumor cells, macrophages, and monocytes (Figures 4C, D). We further quantified the proportions of each cell type in PAQR5-positive and AARS2-negative populations, observing a significantly higher percentage of malignant cells in the AARS2-positive group relative to the AARS2-negative group (Figure 4E). Moreover, differential expression analysis across cell types confirmed that AARS2 expression was largely restricted to malignant cells, fibroblasts, epithelial cells, and M1 macrophages (Figure 4F). This cell type-specific expression profile implies that AARS2 may promote COAD progression and influence macrophage immune functions. In the spatial transcriptomic analysis, we investigated the association between AARS2 and the TME. In all three samples (CRC1–3), AARS2 expression showed strong spatial co-localization with tumor cells, indicating its predominant presence within malignant regions (Figures 4G–I). By comparing predefined malignant and normal regions, we detected significantly elevated AARS2 expression in tumor areas, consistent with previous observations (Figures 4J–L). These findings suggest that AARS2 is closely linked to malignant cell populations in the TME and may contribute to colorectal cancer progression and local immune modulation.

We also used multiple algorithms to further explore the relationship between AARS2 expression and immune infiltration. The heatmap reveals distinct variations in immune cell composition across the sample groups, with markedly different distribution patterns between high- and low-expression cohorts (Figure 5A). Spearman correlation analysis revealed an inverse correlation between AARS2 and a wide range of immune-infiltrating cells in cancer, some of which were particularly strong (Figure 5B).

Spatial transcriptomic analysis was performed to assess the Spearman correlation between AARS2 expression and tumor microenvironment components at spatial resolution (Figure 5C). Consistent with prior localization findings, AARS2 expression correlated positively with malignant cell abundance and negatively with multiple anti-tumor immune cells—including CD4+ T cells, NK cells, and B cells in CRC1, as well as macrophages across CRC1–3 samples.

Further evaluation using XCELL and CIBERSORT algorithms revealed significant negative correlations between AARS2 expression and common lymphoid progenitor cells (R = −0.17, P = 0.00037), M2 macrophages (R = −0.20, P = 2.4e−05), and CD4⁺ memory T cells (R = −0.21, P = 5.1e−06) (Figure 5D), suggesting that elevated AARS2 expression is associated with reduced infiltration of lymphoid progenitors, immunosuppressive M2 macrophages, and adaptive immune memory cells—collectively indicating a potential role for AARS2 in shaping an immunosuppressive tumor microenvironment.

We further explored the relationship of AARS2 with immunogenicity and DNA damage levels to elucidate its potential effects on immune evasion and genomic instability (Figure 5E). Given the relevance of immunomodulatory molecules in cancer therapy, we also analyzed their association with AARS2 to delineate an immune landscape linked to AARS2 activity (Figure 5F).

To investigate the biological function of AARS2 in COAD, we first divided the COAD samples into two groups based on the median expression of AARS2 and performed differential expression analysis. This analysis identified both upregulated and downregulated DEGs. Using the thresholds of |log2FoldChange|>1 and P< 0.05, we screened the DEGs and visualized the results (Figure 6A). Analysis of the KEGG pathway in multi-level and multi-gene sets revealed significant differences in multiple gene sets related to COAD between the AARS2 high-expression and low-expression groups. Notably, several pathways under Genetic Information Processing were enriched, including DNA replication, homologous recombination, and aminoacyl-tRNA biosynthesis. Furthermore, visualization via a hierarchical clustering diagram intuitively demonstrated the enrichment scores and statistical significance levels of these distinct gene sets (Figure 6B). Subsequent KEGG enrichment analysis revealed that these genes were mainly involved in processes such as the taste transduction, RIG-I-like receptor signaling pathway, olfactory transduction, JAK-STAT signaling pathway, cytokine-cytokine receptor interaction, and autoimmune thyroid disease (Figure 6C). GO analysis revealed that genes highly expressed in ARRS2-positive malignant cells were primarily associated with sensory perception, including detection of chemical stimuli involved in sensory perception of smell, olfactory receptor activity, and odorant binding. Additionally, these genes were enriched in terms related to intermediate filament organization, such as intermediate filament cytoskeleton and keratin filament assembly, as well as type I interferon receptor binding (Figure 6D). Additionally, GSEA using Hallmark gene sets indicated that pathways related to NABA Core Matrisome, NABA ECM Glycoproteins, NABA Proteoglycans, and Molecules Associated with Elastic Fibres were downregulated in the AARS2 high-expression group (Figure 6E). In contrast, pathways associated with Cell Cycle Checkpoints, Mitotic G1 Phase and G1-S Transition, and DNA Replication were upregulated in this group (Figure 6F). GSVA was conducted to evaluate the association between AARS2 expression and COAD-related phenotypes. The results demonstrated that AARS2 expression significantly correlated with multiple malignant processes in COAD, including angiogenesis (R = 0.33, P = 9.5e−16), apoptosis (R = 0.38, P < 2.2e−16), cell cycle (R = 0.40, P < 2.2e−16), DNA repair (R = 0.35, P < 2.2e−16), epithelial-mesenchymal transition (R = 0.45, P < 2.2e−16), hypoxia (R = 0.45, P < 2.2e−16), inflammation (R = 0.20, P = 2.2e−16), invasion (R = 0.20, P = 1.1e−06), metastasis (R = 0.16, P = 0.00014), proliferation (R = 0.20, P = 1.9e−06), quiescence (R = 0.28, P = 1.9e−06), stemness (R = 0.38, P < 2.2e−16), differentiation (R = 0.38, P < 2.2e−16), and DNA damage (R = 0.37, P < 2.2e−16) (Figure 6G). Furthermore, a pan-cancer analysis was conducted to evaluate the correlation between AARS2 expression and multiple cancer-related phenotypes across various tumor types. The results indicated that AARS2 expression was significantly associated with various tumor-related biological processes, including angiogenesis (R = -0.098, P < 2.2e-16), apoptosis (R = -0.074, p = 6.9e-14), cell cycle (R = 0.15, P < 2.2e-16), differentiation (R = 0.12, P < 2.2e-16), DNA damage (R = 0.20, P < 2.2e-16), DNA repair (R = -0.15, p < 2.2e-16), EMT (R = -0.21, P < 2.2e-16), hypoxia (R = -0.18, P < 2.2e-16), inflammation (R = -0.10, P < 2.2e-16), invasion (R = -0.13, P < 2.2e-16), metastasis (R = -0.08, P = 1.2e-09), proliferation (R = 0.14, P < 2.2e-16), quiescence (R = 0.05, p = 0.003), and stemness (R = -0.023, P = 0.018) (Figure 6H).

To further investigate the interactions involving AARS2-encoded protein and to identify hub genes, we constructed a protein-protein interaction (PPI) network using the STRING database and identified proteins potentially interacting with AARS2 (Figure 7A). Using AARS2 as the source protein, screening and analysis via the BioGRID database revealed numerous proteins that interact with AARS2, primarily through physical interactions (Figure 7B). Similarly, when AARS2 was set as the target protein, multiple interacting partners were also identified (Figure 7C). These findings suggest that AARS2 may play a role in the assembly and functional regulation of multiple protein complexes within the cell.

Using protein expression data from the TCPA database, we performed Spearman correlation analysis between AARS2 and activity scores of 10 cancer-related pathways previously established in published studies. The results revealed a significant positive correlation between AARS2 expression and the apoptotic pathway in COAD (Figure 7D). Using the TCPA database and based on published research results, pathway activity scores of 10 cancer-related pathways—including TSC/mTOR, RTK, RAS-MAPK, PI3K-AKT, hormone ER, hormone AR, EMT, DNA damage response, cell cycle, and apoptosis—were calculated for pan-cancer analysis. The analysis was performed in the high AARS2 expression group across 32 tumor types: ACC, BLCA, BRCA, CESC, CHOL, COAD, DLBC, ESCA, GBM, HNSC, KICH, KIRC, KIRP, LGG, LIHC, LUAD, LUSC, MESO, OV, PAAD, PCPG, PRAD, READ, SARC, SKCM, STAD, TGCT, THCA, THYM, UCEC, UCS, and UVM. The results showed that the activities of the Apoptosis and CellCycle pathways were significantly higher in the high AARS2 expression group (Figure 7E).

We also obtained relevant immunohistochemical staining images from the HPA database to evaluate the expression patterns of the AARS2 gene across various tumor types. Our analysis revealed significant differences in AARS2 expression among different cancers. Specifically, colorectal cancer, thyroid cancer, lymphoma, ovarian cancer, and skin cancer exhibited a relatively high proportion of strong staining (denoted in red), accounting for 75%, 75%, 55%, 58%, and 55% of cases, respectively, indicating high AARS2 expression in these malignancies. In contrast, breast cancer, liver cancer, lung cancer, pancreatic cancer, and prostate cancer showed a predominant moderate staining pattern (denoted in blue), with proportions of 83%, 67%, 73%, and 91%, respectively, suggesting intermediate expression levels of AARS2 in these tumors (Figure 7F). Meanwhile, immunohistochemical data obtained from the HPA database demonstrated that AARS2 is highly expressed in carcinoid tumors, colorectal cancer (CRC), and stomach adenocarcinoma (STAD). The staining intensity and distribution patterns further revealed its predominant intracellular localization (Figures 7G–I).

Leveraging a clinically annotated tissue cohort from our institution—comprising surgically resected specimens from 12 COAD patients along with matched adjacent normal mucosa as internal controls (Figure 8A)—we first evaluated AARS2 protein expression via immunohistochemistry (IHC). IHC analysis demonstrated that AARS2 immunoreactivity was significantly stronger in COAD tissues than in adjacent normal mucosa (Figure 8B). Quantitative IHC scoring corroborated this observation, revealing a statistically significant increase in AARS2 expression in tumor tissues (Figure 8C; P < 0.01).

In parallel, immunohistochemical analysis of cGAS revealed significantly elevated protein levels in COAD tumor tissues relative to adjacent normal mucosa (Figure 8D), with quantitative assessment confirming this upregulation (Figure 8E; P < 0.001). Notably, increased cGAS protein abundance does not equate to enhanced functional activity, as cGAS signaling is primarily governed by post-translational modifications (e.g., lysine lactylation) rather than expression quantity. Consequently, the observed elevation in cGAS protein levels in clinical specimens reflects a quantitative change that requires functional validation. Subsequent in vitro experiments were therefore designed to specifically interrogate the activity status of the cGAS–STING pathway and its regulatory relationship with AARS2.

To elucidate the biological function of AARS2 in COAD, we established loss-of-function models in HCT 116 cells using three distinct small interfering RNAs (siAARS2-1, siAARS2-2, and siAARS2-3). The knockdown efficiency was successfully verified at both the transcriptional and translational levels. As shown in Figures 9A, qRT-PCR analysis demonstrated that AARS2 mRNA levels were significantly downregulated in all three knockdown groups compared to the negative control (NC) group (P<0.0001). Consistent with the mRNA data, Western blot analysis confirmed a marked reduction in AARS2 protein expression (Figure 9B).

Given the association between AARS2 and metabolic processes, we further investigated whether AARS2 depletion affects the glycolytic phenotype of COAD cells. Analysis of cell culture supernatants revealed that lactate production was significantly attenuated following AARS2 knockdown. Compared to the NC group, extracellular lactate concentrations were consistently reduced in the siAARS2-1, siAARS2-2, and siAARS2–3 groups at both 24 h (Figure 9C) and 48 h (Figure 9D) time points.

Since intracellular lactate levels are closely linked to protein lactylation, a newly identified post-translational modification, we examined global lactylation levels using a pan-lactylation antibody. Western blot results indicated that AARS2 silencing led to a reduction in total protein lactylation, which was most pronounced in the siAARS2–1 and siAARS2–2 groups (Figure 9E). Although the siAARS2–3 group exhibited a less distinct decrease in pan-lactylation signals despite reduced secreted lactate, the overall trend suggested that AARS2 is essential for maintaining high levels of intracellular lactylation.

Based on the premise that lactylation negatively regulates cGAS activity, we hypothesized that the observed reduction in lactate metabolism and protein lactylation upon AARS2 knockdown would restore cGAS signaling functionality. To validate this, we assessed the mRNA expression of key downstream effectors of the cGAS-STING pathway, including type I interferons and proinflammatory chemokines. Consistent with our hypothesis, AARS2 silencing elicited a robust activation of the innate immune response. As illustrated in Supplementary Figure 3, the expression levels of CCL5 (Supplementary Figure 3A), CXCL10 (Supplementary Figure 3B), and IFNB1 (Supplementary Figure 3C) were significantly upregulated in the siAARS2-1, siAARS2-2, and siAARS2–3 groups compared to controls (all P<0.05).