An externally validated lactylation-associated prognostic signature for overall survival prediction in lung adenocarcinoma identifies TUBA1C as a candidate gene

Introduction

Non-small cell lung cancer (NSCLC) remains one of the leading causes of cancer-related mortality worldwide, and lung adenocarcinoma (LUAD) is the most common histological subtype of NSCLC. LUAD exhibits substantial histological and molecular heterogeneity, which contributes to marked differences in prognosis among patients (1-3). Current treatment strategies include surgery, radiotherapy, chemotherapy, targeted therapy, and immunotherapy; however, survival outcomes remain unsatisfactory, particularly because many patients are diagnosed at advanced stages (4,5). Although low-dose computed tomography (LDCT) has improved the detection of early pulmonary nodules, imaging parameters, pathological classification, and tumor-node-metastasis (TNM) staging alone do not fully capture the biological heterogeneity underlying survival differences in LUAD (6). Therefore, transcriptome-based molecular signatures that complement conventional clinicopathological stratification may help improve individualized prognostic assessment.

Machine learning has been widely applied in oncology research to develop prognostic models, and identify candidate biomarkers from high-dimensional molecular data. Integrative analyses of single-cell and bulk transcriptomic profiles can further connect patient-level prognostic signatures with cell-type-specific transcriptional states. Previous studies have demonstrated the utility of machine learning-based molecular signatures in colorectal cancer, prostate cancer, esophageal squamous cell carcinoma, and osteosarcoma (7-10). However, many existing prognostic models mainly focus on immune-, metabolic-, or pathway-related signatures, whereas the prognostic relevance of lactylation-related transcriptional programs in LUAD remains insufficiently characterized.

Since the discovery of the Warburg effect, metabolic reprogramming has been recognized as a hallmark of cancer. Cancer cells frequently display increased aerobic glycolysis, leading to lactate accumulation and reshaping of the tumor metabolic microenvironment. Lactate participates not only in energy homeostasis, redox buffering, and fatty acid metabolism but also in tumor growth, metastasis, and epigenetic regulation (11). A landmark study demonstrated that metabolically generated lactate can serve as a precursor for histone lysine lactylation, thereby linking cellular metabolic state with transcriptional regulation (12). Subsequent studies have implicated lactylation-related processes in multiple malignancies, including ocular melanoma, breast cancer, prostate cancer, and lung cancer, where they have been associated with tumorigenesis, metabolic reprogramming, angiogenesis, invasion, and gene-expression regulation (13-17). These findings suggest that lactate metabolism and lactylation-related programs may represent biologically relevant axes in cancer progression. In parallel, dysregulated glycolysis has attracted substantial therapeutic interest, and pharmacologic strategies targeting glycolytic pathways have been explored to disrupt the energy and biosynthetic supply required for tumor proliferation (18,19).

The tumor microenvironment (TME) of LUAD is a complex ecosystem composed of malignant cells, immune cells, stromal cells, endothelial cells, and extracellular matrix components. Single-cell studies have shown that LUAD progression is accompanied by substantial tumor-cell heterogeneity and dynamic interactions between malignant cells and the surrounding microenvironment (20). In EGFR-mutant NSCLC, immunosuppressive TME features, including altered lymphocyte infiltration, recruitment of suppressive immune cells, and upregulation of immune checkpoints, have been proposed as contributors to limited responses to immune checkpoint inhibitor monotherapy (21). More broadly, covalent histone modifications and related epigenetic programs have long been linked to cancer biology (22). These observations support the possibility that lactate metabolism and lactylation-related transcriptional programs may interact with immune and stromal features in LUAD. However, transcriptomic analyses alone cannot directly prove lactylation-driven immune remodeling, and direct lactylation assays are required for mechanistic confirmation.

Despite these advances, the clinical relevance of lactylation-related genes (LRGs) in LUAD remains incompletely defined. In particular, it remains unclear whether lactylation-related transcriptional programs can be used to stratify prognosis, characterize tumor metabolic and immune states, and identify candidate genes for further biological investigation. Based on curated LRGs and public transcriptomic datasets, we developed a machine learning-based lactylation-related gene signature (LRGS) for overall survival (OS) prediction in LUAD and evaluated its performance in TCGA-LUAD, GSE31210, and GSE72094. We further investigated LRGS-associated pathway activity, genomic alterations, immune infiltration, immunotherapy-related features, drug sensitivity, and single-cell intercellular communication patterns. Among the 15 signature genes, TUBA1C was prioritized for preliminary expression validation because it encodes an α-tubulin isoform related to cytoskeletal organization and mitotic processes, while its functional role in LUAD requires further experimental validation (23). This study aims to provide a transcriptome-based framework for prognostic stratification and hypothesis generation, rather than direct evidence of lactylation activity or causal mechanisms linking lactylation-related transcriptional programs to LUAD progression. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0963/rc).

Methods

The overall study design and analytical workflow are shown in Figure 1.

Figure 1 Flow chart. CI, confidence interval; HR, hazard ratio; M, metastasis; MATH, mutant-allele tumor heterogeneity; N, node; RSF, random survival forest; T, tumor.

Data sources and processing

RNA-seq data and corresponding clinical information for LUAD were retrieved from The Cancer Genome Atlas (TCGA) portal (https://portal.gdc.cancer.gov/). All public datasets were downloaded on October 26, 2024. This study was designed as a retrospective prediction model development and external validation study based on public transcriptomic datasets. TCGA-LUAD was used as the model development cohort, whereas GSE31210 and GSE72094 were downloaded from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/) and used as independent external validation cohorts. Transcript per million (TPM) values were extracted from TCGA-LUAD for downstream analyses and transformed as log₂ (TPM + 1). For GEO cohorts, normalized expression matrices were processed according to the corresponding platform annotation files and log2-transformed when appropriate. Patients were included if transcriptomic data and OS information were available. For TCGA-LUAD, only primary tumor samples were retained for survival modeling. For GSE31210, samples annotated as “frozen tissue of primary lung tumors” were retained. For GSE72094, samples annotated as “lung adenocarcinoma” were retained. Duplicated sequencing profiles and samples with incomplete survival or essential clinical information were excluded, yielding a final cohort of 503 TCGA-LUAD patients. The final sample sizes of GSE31210 and GSE72094 were 226 and 398, respectively. Somatic mutation and copy number variation (CNV) data were also obtained from TCGA. The primary outcome was OS, which was defined as the time from initial diagnosis or follow-up entry to death from any cause. Patients who were alive at the last follow-up were censored. Because outcome and predictor data were extracted from de-identified public datasets, no additional blinded assessment of outcomes or predictors was performed. No formal sample-size calculation was performed because this study used all eligible samples available in the public cohorts. Because the cohorts were derived from public repositories, cohort-specific accrual periods and follow-up information were extracted from available dataset annotations and original publications when available; these details, together with dataset download date and median follow-up duration, are summarized in Table S1.

To define LRGs, we curated 22 genes from published studies (24,25), and retrieved 332 genes from the GeneCards database (https://www.genecards.org/). After merging and removing duplicates, 337 LRGs were retained (Table S2). In addition, the IMvigor210 cohort was included as an exploratory immunotherapy cohort to evaluate the association between LRGS and response to immune checkpoint blockade (ICB) (26).

Single-cell data collection and processing

Single-cell RNA-seq data from 11 LUAD patients were obtained from the GSE131907 dataset. Quality control was performed by retaining cells with <10% mitochondrial gene content and 100–6,000 detected genes; genes detected in fewer than three cells were excluded. Highly variable genes (n=2,000) were identified for downstream analyses, and batch effects across the 11 samples were corrected using Harmony (27). Cells were clustered using Seurat’s FindNeighbors and FindClusters functions and visualized using t-distributed stochastic neighbor embedding (t-SNE). Cell types were annotated based on canonical marker genes (Table S3).

Single-sample gene set enrichment analysis (ssGSEA) scores were computed using the gene set variation analysis (GSVA) package to quantify LRG-set enrichment in each cell. These scores were interpreted as relative enrichment of lactylation-related transcriptional programs rather than direct measurements of lactylation activity. Differentially expressed genes (DEGs) between cells with high and low LRG-set scores were identified using Seurat’s FindMarkers function with min.pct =0.35, and lactylation-related single-cell marker genes were subsequently obtained.

ssGSEA and weighted gene co-expression network analysis (WGCNA)

ssGSEA quantifies the enrichment of a predefined gene set in each sample, and reflects the relative activity of the corresponding transcriptional program. In this study, ssGSEA scores were calculated using the GSVA package to quantify LRG-set score in 503 TCGA-LUAD samples. The resulting score was termed the LRG-set score. Because direct lactylome, histone lactylation, global protein lactylation, or metabolomic lactate measurements were not available in these public cohorts, this score was interpreted as a transcriptomic proxy of LRG programs rather than a direct measure of lactylation activity, protein lactylation, histone lactylation, or lactate abundance. WGCNA was used to characterize co-expression patterns across samples and to identify modules of highly co-expressed genes. The WGCNA package was applied to RNA-seq data from 503 TCGA-LUAD samples, and an appropriate soft-thresholding power was selected to approximate a scale-free network topology. The weighted adjacency matrix was transformed into a topological overlap matrix (TOM), and the corresponding dissimilarity matrix (dissTOM) was computed. The minimum module size was set to 50 genes, and the module merging threshold was set to 0.15. Dynamic tree cutting was used for hierarchical clustering and module detection, and the module most correlated with the LRG-set score was selected for downstream analyses. The sample clustering and WGCNA module dendrogram are shown in Figure S1A,S1B.

Difference analysis and univariate Cox regression analysis

Differential expression analysis was performed on TCGA-LUAD TPM data using the limma package, with thresholds of |log2 fold change| ≥ log2(1.5) and P<0.05 to identify DEGs. Candidate genes were defined as the intersection of single-cell marker genes, WGCNA module genes, and DEGs. To evaluate associations with OS, univariate Cox regression was conducted for the intersected genes, and genes with P<0.05 were considered significant.

Machine learning to build prognostic models

We evaluated 10 machine learning algorithms—least absolute shrinkage and selection operator (LASSO), support vector machine (SVM), gradient boosting machine (GBM), random survival forest (RSF), elastic net (Enet), stepwise Cox, ridge regression, CoxBoost, SuperPC, and PLSR-Cox—resulting in 101 model combinations (28). TCGA-LUAD was used for model training and model selection, whereas GSE31210 and GSE72094 were used only for external validation.

Candidate genes identified from the preceding screening steps were entered into the machine learning framework. For each model combination, the concordance index (C-index) was calculated in the model development cohort and in the two external validation cohorts. Models were ranked according to the mean C-index across the development and external validation cohorts to prioritize model combinations with stable cross-cohort performance, and the model with the highest mean C-index was selected as the optimal model.

The final optimal model was StepCox (forward) + RSF. The final LRGS predictors were the expression values of 15 genes: GAPDH, LDHA, ALDOA, SEC61G, CCT6A, YWHAZ, PGK1, FKBP3, C15orf48, HMGA1, CTSH, NPC2, NME1, TUBA1C, and PDCD5.

For each cohort, the model-derived risk score was extracted for subsequent survival analysis. In the external validation cohorts, risk scores were calculated using the model trained in TCGA-LUAD without recalibration, refitting, or model updating. Model performance was assessed using the C-index, Kaplan-Meier survival analysis, and time-dependent receiver operating characteristic (ROC) analysis. Time-dependent area under the curve (AUC) values were calculated at 1, 2, and 3 years.

Survival analysis and construction of predictive nomograms

Patients in TCGA-LUAD, GSE31210, and GSE72094 were stratified into high- and low-risk groups based on the standardized LRGS risk score. Specifically, the LRGS risk score was converted into a z-score within each cohort. Patients with a z-score >0 were assigned to the high-risk group, whereas those with a z-score ≤0 were assigned to the low-risk group.

Kaplan-Meier survival curves were generated using the survminer package, and differences in OS between risk groups were assessed using the log-rank test. Time-dependent ROC curves and the corresponding AUC were computed for each cohort using the timeROC package. The prognostic performance of LRGS was further evaluated using the concordance index.

Associations between LRGS and clinicopathological variables, including sex, clinical stage, and TNM classification, were further examined. A nomogram integrating LRGS and clinical variables was constructed to estimate individualized survival probabilities for patients with LUAD. Calibration curves and the C-index were used to assess calibration and discrimination of the nomogram, and decision curve analysis (DCA) was performed to evaluate its net clinical benefit.

Genomic alterations across LRGS-defined risk subgroups

Mutant-allele tumor heterogeneity (MATH) is a method for quantifying intratumoral heterogeneity (ITH) based on the distribution of mutant allele fractions. Previous studies have demonstrated the prognostic value of MATH in multiple tumor types, including breast and colon cancers (29,30). Accordingly, MATH scores were calculated for each LUAD patient using a previously described approach (31).

To characterize LRGS-associated somatic alterations, the maftools package was used to generate waterfall plots depicting the mutational landscapes of the high- and low-risk LUAD groups. The top 20 genes showing the largest between-group differences were further selected for CNV analysis.

Comprehensive immune profiling and prediction of response to ICB

To investigate associations between LRGS and immune infiltration in the LUAD TME, the relative proportions of 22 immune cell types were estimated using CIBERSORT. To further corroborate the CIBERSORT-based estimates, ESTIMATE, ssGSEA, and xCell were applied.

The cancer-immunity cycle is a central framework for antitumor immunity and immunotherapy and comprises seven steps: cancer antigen release, antigen presentation, priming and activation, trafficking of immune cells to tumors, infiltration into tumors, recognition of cancer cells by T cells, and killing of cancer cells. TIP was used to infer immune activity, and step-wise activity scores were generated for TCGA-LUAD samples across the seven steps.

To estimate the potential benefit from ICB based on immunomodulatory factors, immunosuppressive cell signatures, MHC-related molecules, and effector-cell signatures, we used the Immunophenoscore (IPS). IPS is computed using a machine learning framework based on gene-expression profiles of representative immune cell types. Higher IPS values indicate a greater likelihood of benefit from immunotherapy. IPS scores for TCGA-LUAD samples were obtained from The Cancer Immunome Atlas (TCIA) database (https://www.tcia.at/home).

LRGS-associated drug sensitivity prediction

To support personalized treatment, the oncoPredict package was used to predict chemotherapy sensitivity in LUAD patients stratified into high- and low-risk groups based on LRGS scores (32). Using drug-response data from the Genomics of Drug Sensitivity in Cancer (GDSC2) database (https://www.cancerrxgene.org/) (33), oncoPredict was applied to estimate drug sensitivity (IC50) for a panel of therapeutic agents in LUAD samples, with P<0.05 considered statistically significant.

Protein-protein interaction (PPI) network analysis and candidate-gene prioritization

The 15 genes from the optimal LRGS model were submitted to the STRING database (https://cn.string-db.org/) to construct a PPI network in Homo sapiens. The minimum required interaction score was set to 0.400. The STRING-derived interaction network was imported into R, and degree centrality and betweenness centrality were calculated using the igraph package. Degree centrality was defined as the number of direct interactions connected to each node, whereas betweenness centrality was used to estimate the bridging role of each node within the network. The centrality results are provided in Table S4. TUBA1C was prioritized for preliminary expression validation based on its inclusion in the final LRGS, prognostic relevance, biological plausibility as an α-tubulin isoform involved in cytoskeletal organization and mitotic regulation, prior evidence in LUAD/NSCLC or other malignancies, and its network connectivity in the LRGS PPI network.

Clinical specimens and ethical approval

Three paired LUAD tumors and matched adjacent normal tissues were collected at The Second Affiliated Hospital of Nanchang University, fixed in 4% paraformaldehyde, paraffin-embedded, and processed for immunohistochemistry (IHC). All specimens were pathologically verified by board-certified pathologists. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Institutional Ethics Committee of The Second Affiliated Hospital of Nanchang University [No. (2018) CDEFYYLK(3-05)]; written informed consent was obtained from all participants.

Cell culture

The normal human bronchial epithelial cell line BEAS-2B was obtained from the National Collection of Authenticated Cell Cultures (NCACC), and human NSCLC cell lines (A549, PC9, H1299, H1650, and H1944) were purchased from Shanghai Zhong Qiao Xin Zhou Biotechnology Co., Ltd. Cell line identity was verified by STR profiling, and cultures were routinely confirmed to be mycoplasma-free. Cells were maintained at 37 ℃ in a humidified incubator with 5% CO₂ in DMEM (BEAS-2B, A549) or RPMI-1640 (PC9, H1299, H1650, H1944) supplemented with 10% FBS, 100 U/mL penicillin, and 100 µg/mL streptomycin; H1944 medium additionally contained 1% GlutaMAX™ and 1 mL of 100 mM sodium pyruvate.

IHC

IHC was performed on paraffin-embedded sections prepared from three paired LUAD tumors and matched adjacent non-tumor tissues. Sections were processed using standard protocols, including deparaffinization and rehydration, antigen retrieval, blocking, incubation with antibodies, signal development, and counterstaining and mounting. TUBA1C immunoreactivity was evaluated in tumor and matched normal tissues, and staining was semi-quantitatively scored using the H-score method.

Reverse transcription quantitative polymerase chain reaction (RT-qPCR)

Total RNA was extracted using an RNA Extraction Kit (Vazyme, China) according to the manufacturer’s instructions. Extracted RNA was reverse-transcribed into cDNA using the HiScript II Q Select RT SuperMix for qPCR (+gDNA wiper) kit (Vazyme, China). Primer sequences for TUBA1C and β-actin are provided in Table S5.

Statistical analysis

All statistical analyses were performed in R software (version 4.4.1). Before model development and validation, duplicated sequencing profiles and samples with missing survival information or essential clinical variables were excluded. Therefore, all analyses were conducted using a complete-case approach, and no statistical imputation was performed.

The chi-square test was used to compare clinical characteristics between the training and validation cohorts. The Wilcoxon rank-sum test was applied to compare two groups with non-normally distributed variables. For differential expression analysis, genes were considered differentially expressed based on log2 fold change and adjusted P values. Kaplan-Meier survival curves and log-rank tests were performed using the survival package to compare OS between risk groups. Univariate and multivariate Cox regression analyses were conducted to identify independent prognostic factors. Time-dependent ROC curves and the corresponding AUC values were calculated using the timeROC package. A two-sided P<0.05 was considered statistically significant. Significance was annotated as follows: ns, P≥0.05; *, P<0.05; **, P<0.01; ***, P<0.001.

Results

Identification of LRGs by integrating single-cell and bulk transcriptomic data

Cells from GSE131907 were annotated into 10 major populations, including B cells, dendritic cells, endothelial cells, epithelial cells, fibroblasts, mast cells, myeloid cells, plasmacytoid dendritic cells, plasma cells, and T/NK cells, based on canonical marker genes. The corresponding marker-gene expression patterns and top marker genes for each population are shown in Figure 2A-2C.

Figure 2 Identification of 48 lactylation-related genes at single-cell and bulk RNA levels. (A) 10 cell types identified in GSE131907 via marker gene analysis; (B) expression of marker genes of each cell population; (C) heatmap map of the top 5 marker genes in each cell population; (D) lactylation-related gene-set score in each cell population; (E) distribution of lactylation-related gene-set scores across different cell types; (F) module-trait heatmap showing the correlation between the MEblue module and the lactylation-related gene-set score; (G) positive correlation between GS and MM in the blue module; (H) volcano plot showing the results of differential analysis of TCGA-LUAD tumor samples and normal samples, with the top 5 up-regulated and down-regulated genes specifically annotated; (I) 48 overlapping genes identified among MEblue module, scRNA-seq marker genes, and bulk RNA-seq DEGs; (J) GO enrichment analysis of the 48 LRGs reveals significant terms in protein folding, pyruvate metabolism (BP); secretory granule lumen (CC); and unfolded protein binding (MF); (K) univariate Cox regression analysis results of LRG genes and the correlation between these genes: key candidate genes include protective genes SFTPB and CTSH (HR <1) and risk genes GAPDH, HMGA1, LDHA, CCT6A, and SEC61G (HR >1); (L) genes showing predominant CNV patterns: ILF2, DAP3, CCT3 (mainly gain); C15orf48, FCGRT, HMGB2 (mainly loss). BP, biological process; CC, cellular component; CN, copy number; DEG, differentially expressed gene; FC, fold change; GS, gene significance; GO, Gene Ontology; LRG, lactylation-related gene; LUAD, lung adenocarcinoma; MF, molecular function; MM, module membership; scRNA-seq, single-cell RNA sequencing; TCGA, The Cancer Genome Atlas; t-SNE, t-distributed stochastic neighbor embedding.

Among these cell types, LRG-set scores were relatively higher in epithelial cells and myeloid cells (Figure 2D,2E). Based on ssGSEA scores, single cells were stratified into cells with high and low LRG-set scores, and 1,552 DEGs were identified for downstream analysis. To identify co-expression modules associated with lactylation-related features in bulk RNA-seq data, WGCNA was performed in the TCGA-LUAD cohort. A total of 23 modules were identified, and the blue module showed the strongest correlation with the LRG-set score (cor =0.67) (Figure 2F). In this module, module membership was strongly correlated with gene significance (cor =0.77, P<1×10⁻²⁰⁰), suggesting that these genes were closely associated with lactylation-related transcriptional characteristics.

Differential expression analysis of TCGA-LUAD bulk transcriptomic data identified 5,173 DEGs between tumor and normal tissues (|log2FC| >0.585 and adj.P.Val <0.05). We then intersected genes from the blue module, bulk RNA-seq DEGs, and single-cell DEGs. This analysis yielded 48 LRGs that were consistently implicated across both bulk and single-cell datasets (Figure 2G-2I).

GO enrichment analysis showed that these 48 LRGs were mainly enriched in biological processes related to protein folding and pyruvate metabolism, cellular components such as secretory granule lumen and chaperone complexes, and molecular functions including unfolded protein binding. Univariate Cox regression further identified 39 of the 48 genes as significantly associated with prognosis (Table S6). In addition, CNV analysis suggested frequent amplifications in genes such as ILF2, DAP3, CCT3, CCT5, YWHAZ, and PDCD5, whereas deletions were more common in genes such as ALDH2, LDHA, C15orf48, FCGRT, HMGB2, and OSTC (Figure 2J-2L).

Machine learning-based construction of the LRGS prognostic signature

To ensure that the prognostic genes were comparable across datasets, we intersected the 39 prognosis-related LRGs with the expression matrices of TCGA-LUAD, GSE31210, and GSE72094. This step retained 38 genes for model construction. These genes were subsequently entered into machine learning analysis, with TCGA-LUAD serving as the model development cohort and GSE31210 and GSE72094 serving as external validation cohorts. Model performance was assessed using the C-index in each cohort (Figure 3A).

Figure 3 StepCox[forward] + RSF model (LRGS) shows an association between lactylation-related transcriptional features and overall survival in LUAD across three independent datasets. (A) LRGS model construction and C-index evaluation across TCGA-LUAD, GSE31210, and GSE72094; (B-D) in the TCGA, GSE31210, and GSE72094 cohorts, LRGS is consistently associated with poor overall survival; (E) Univariate Cox analysis of 15 genes in the optimal model: CTSH and NPC2 are protective factors (HR <1), while the remaining 13 genes are risk factors (HR >1). AUC, area under the curve; CI, confidence interval; HR, hazard ratio; LRGS, lactylation-related gene signature; LUAD, lung adenocarcinoma; RSF, random survival forest; TCGA, The Cancer Genome Atlas.

Among the 101 evaluated model combinations, StepCox (forward) + RSF ranked first according to the mean C-index across the development and validation cohorts and was selected as the final model. The C-index values of this model were 0.935 in TCGA-LUAD, 0.668 in GSE31210, and 0.637 in GSE72094, with a mean C-index of 0.747 (Figure 3A). After applying the eligibility criteria, 503 patients from TCGA-LUAD were included in the model development cohort, while 226 patients from GSE31210 and 398 patients from GSE72094 were included as external validation cohorts. TCGA-LUAD contained transcriptomic, survival, clinicopathological, somatic mutation, and CNV data, whereas GSE31210 and GSE72094 were used mainly for external validation based on transcriptomic and survival information. Because the three cohorts were derived from different public repositories and platforms, available clinicopathological variables were not completely identical across datasets. The numbers of death events, censored cases, and median follow-up time in each cohort are summarized in Table S1. Available baseline characteristics of the development and external validation cohorts are shown in Table S7.

Based on the cohort-specific z-score cutoff, 177 and 326 patients in TCGA-LUAD were assigned to the high- and low-risk groups, respectively. In GSE31210, 120 and 106 patients were classified as high and low risk, whereas the corresponding numbers in GSE72094 were 245 and 153. Kaplan-Meier analyses showed that patients in the high-risk group had significantly worse OS than those in the low-risk group in TCGA-LUAD (P<0.0001), GSE31210 (P<0.001), and GSE72094 (P<0.001).

Time-dependent ROC analyses further supported the predictive performance of the LRGS. The AUC values were 0.96, 0.98, and 0.99 at 1, 2, and 3 years in TCGA-LUAD, respectively. The corresponding AUC values were 0.69, 0.74, and 0.63 in GSE31210, and 0.64, 0.67, and 0.66 in GSE72094 (Figure 3B-3D). These findings indicate that LRGS retained prognostic stratification ability across multiple cohorts, although its discriminative performance was lower in the external validation cohorts than in TCGA-LUAD.

The final StepCox (forward) + RSF model included 15 genes. Among them, CTSH and NPC2 were protective factors, whereas the remaining genes were associated with increased risk (Figure 3E).

Clinical relevance of LRGS and development of a clinicopathologic nomogram

Because clinicopathological characteristics are widely used in prognostic assessment, we next evaluated the relationship between LRGS and conventional clinical variables. In the TCGA-LUAD cohort, significant differences between the high- and low-risk groups were observed for clinical stage (P=3×10⁻¹⁰), T stage (P=2.8×10⁻⁵), N stage (P=1.7×10⁻⁶), and M stage (P=0.0086) (Figure 4A-4D). These findings suggest that a higher LRGS risk score is associated with more adverse clinicopathological features. In addition, the distribution of M stage differed between LRGS-defined subgroups, and ROC analysis yielded an AUC of 0.661 for discriminating M stage, indicating a modest association between LRGS and metastasis-related status in LUAD (Figure S1C,S1D).

Figure 4 LRGS effectively stratifies LUAD patients by clinicopathological stages. (A) Risk score increases with advancing T stage; (B) differences in risk scores among patients grouped by T, N, M, and stage; (C) clinical characteristics and expression distribution of model genes according to LRGS risk score; (D) the LRGS high-risk group exhibits more adverse clinical characteristics (P<0.05); (E,F) both univariate and multivariate analyses identify LRGS as an independent adverse prognostic factor in LUAD. ns, P≥0.05; *, P<0.05; ***, P<0.001; ****, P<0.0001. CI, confidence interval; HR, hazard ratio; LRGS, lactylation-related gene signature; LUAD, lung adenocarcinoma; M, metastasis; N, node; ns, not significant; T, tumor.

To determine whether LRGS is an independent prognostic factor, univariate and multivariate Cox regression analyses were performed in TCGA-LUAD (Figure 4E,4F). LRGS was significantly associated with OS in univariate analysis (HR =6.304, P<0.001). This association remained significant in multivariate analysis (HR =6.239, P<0.001), supporting the independent prognostic value of LRGS in LUAD.

To further facilitate clinical interpretation, we constructed a nomogram that integrated LRGS with clinicopathological variables. Calibration curves showed acceptable agreement between predicted and observed outcomes. The C-index suggested that the LRGS-integrated nomogram provided better discrimination than individual clinicopathological variables in the TCGA-LUAD cohort. DCA further suggested potential net clinical benefit compared with conventional clinicopathological variables alone (Figure S1E-S1H).

Bulk transcriptomic pathway characterization and genomic alteration profiling across LRGS risk subgroups

To further characterize the biological features associated with LRGS-defined risk subgroups, we performed pathway enrichment analyses in the TCGA-LUAD cohort. GSEA showed that the low-risk group was enriched in ALLOGRAFT_REJECTION, BILE_ACID_METABOLISM, INFLAMMATORY_RESPONSE, and KRAS_SIGNALING_DN. In contrast, the high-risk group was enriched in GLYCOLYSIS, HYPOXIA, MTORC1_SIGNALING, PI3K_AKT_MTOR_SIGNALING, REACTIVE_OXYGEN_SPECIES_PATHWAY, and TNFA_SIGNALING_VIA_NFKB (Figure 5A,5B). These findings suggest that LRGS-high tumors are associated with more active metabolic and stress-related programs.

Figure 5 The genomic landscapes are distinctly different between the LRGS high- and low-risk groups. (A) The low-risk group is enriched in hallmark pathways such as allograft rejection, bile acid metabolism, inflammatory response, and KRAS signaling downregulation; (B) the high-risk group is characterized by enrichment in glycolysis, hypoxia, mTORC1 signaling, PI3K-AKT-mTOR signaling, reactive oxygen species, and TNF-α/NF-κB signaling pathways; (C,D) the waterfall plots reveal a higher frequency of somatic mutations in high-risk patients; (E) Somatic Mutation interaction analysis of the top 20 mutated genes shows more co-mutations in the low-risk group; (F) the MATH score is significantly higher in the high-risk group. CNV, copy number variation; LRGS, lactylation-related gene signature; MATH, mutant-allele tumor heterogeneity; NF-κB, nuclear factor kappa-B; TNF-α, tumor necrosis factor-alpha.

To further compare pathway activity between the two groups, GSVA was performed. Consistent with the GSEA results, the high-risk group showed increased activity of G2M_CHECKPOINT, E2F_TARGETS, GLYCOLYSIS, HYPOXIA, PI3K_AKT_MTOR_SIGNALING, and REACTIVE_OXYGEN_SPECIES_PATHWAY. By contrast, pathways such as NOTCH_SIGNALING, COAGULATION, IL2_STAT5_SIGNALING, ALLOGRAFT_REJECTION, KRAS_SIGNALING_DN, IL6_JAK_STAT3_SIGNALING, MYOGENESIS, BILE_ACID_METABOLISM, and HEME_METABOLISM were more active in the low-risk group (Figure S2A). In addition, correlation analyses between LRGS risk scores and representative pathway signatures further supported these differences. Pathways positively correlated with LRGS, including HYPOXIA and GLYCOLYSIS, were associated with worse OS, whereas negatively correlated pathways such as IL6_JAK_STAT3_SIGNALING showed the opposite trend (Figure S2B).

We next examined genomic alterations across LRGS-defined subgroups. Somatic mutation analysis revealed distinct mutational landscapes between the two groups. In the high-risk group, the most frequently mutated genes were TP53 (57%), TTN (52%), and CSMD3 (43%). In the low-risk group, the most frequently mutated genes were TP53 (45%), MUC16 (39%), and TTN (39%) (Figure 5C,5D). These data indicate that the two LRGS subgroups differ not only in survival outcome but also in mutation profiles.

To further assess structural genomic alterations, we analyzed CNV patterns in the top 20 genes showing the largest differences between the two risk groups. The low-risk group showed a higher co-occurrence frequency of alterations, whereas the high-risk group displayed distinct amplification and deletion patterns (Figure 5E). The overall CNV frequency distributions in the two groups are shown in Figure S2C. Amplifications were more frequently observed in FLG, KRAS, SPTA1, ZNF536, CSMD3, ZFHX4, PAPPA2, USH2A, NAV3, TTN, XIRP2, and PCDH15, whereas deletions were more common in LRP1B, COL11A1, MUC16, ANK2, and TP53. In addition, ITH was quantified using MATH. The high-risk group showed significantly higher MATH scores than the low-risk group (Figure 5F), suggesting increased genomic heterogeneity in LRGS-high tumors.

Association of LRGS with single-cell transcriptional programs and cell-cell communication

To investigate LRGS-related features at single-cell resolution, we examined the expression patterns of the 15 LRGS genes across annotated cell populations. These genes, including GAPDH, LDHA, ALDOA, SEC61G, CCT6A, YWHAZ, PGK1, FKBP3, C15orf48, HMGA1, CTSH, NPC2, NME1, TUBA1C, and PDCD5, were mainly expressed in epithelial cells, myeloid cells, and T/NK cells. Based on the LRGS model, a risk score was calculated for each cell (Figure 6A).

Figure 6 Single-cell LRGS analysis inferred altered TGF-β, VEGF, and CXCL communication patterns. (A) Single-cell RNA-seq analysis revealed the expression of 15 genes in LRGS in each cell population; (B) KEGG enrichment analysis reveals significant immune and lysosomal pathway activation in high-risk versus low-risk cells; (C,D) more complex ligand-receptor interaction networks/spectra are observed in high-risk versus low-risk tumor cells; (E) GSEA analysis reveals hallmark pathways enriched in high-risk tumor cells, including allograft rejection, MYC targets V1, and reactive oxygen species pathway; (F-H) a circular diagram showing the TGF-β, VEGF, and CXCL signaling pathway network, and a heatmap showing the role of different cell types in the pathway network. CXCL, C-X-C motif chemokine ligand; GSEA, gene set enrichment analysis; KEGG, Kyoto Encyclopedia of Genes and Genomes; LRGS, lactylation-related gene signature; TGF-β, transforming growth factor-beta; t-SNE, t-distributed stochastic neighbor embedding; VEGF, vascular endothelial growth factor.

Cells were then stratified into LRGS-high and LRGS-low groups according to the single-cell risk score, and differential expression analysis was performed. KEGG enrichment analysis showed that genes upregulated in the high-risk state were mainly enriched in phagosome, lysosome, antigen processing and presentation, and autoimmune-related pathways (Figure 6B). These findings suggest that LRGS-associated transcriptional differences are accompanied by changes in immune-related and vesicle-associated cellular programs.

To further assess tumor cell-centered interactions, epithelial cells were stratified according to LRGS risk score, and their communication patterns with other cell types were analyzed. Epithelial cells in the high- and low-risk states showed distinct ligand–receptor interaction patterns, indicating differences in intercellular communication within the TME (Figure 6C,6D). In addition, GSEA of LRGS-high cells showed enrichment of pathways including ALLOGRAFT_REJECTION, MYC_TARGETS, and REACTIVE_OXYGEN_SPECIES_PATHWAY (Figure 6E). These data suggest that the LRGS-high state is associated with altered proliferative, immune-related, and stress-response programs at the single-cell level.

We next characterized signaling pathways involved in cell-cell communication. transforming growth factor-beta (TGF-β), vascular endothelial growth factor (VEGF), and C-X-C motif chemokine ligand (CXCL) signaling were inferred to be relatively more active in the high-risk group (Figure 6F-6H). Specifically, high-risk epithelial cells appeared to be major senders of TGF-β signals and mainly targeted endothelial cells and dendritic cells. In parallel, high-risk epithelial cells, fibroblasts, and myeloid cells showed stronger inferred VEGF signaling toward endothelial cells, suggesting an angiogenesis-related communication pattern. In addition, fibroblasts and high-risk epithelial cells were dominant sources of CXCL signaling directed toward immune cells. These findings may indicate altered chemokine-related immune-cell recruitment patterns in the high-risk state.

Association of LRGS with the tumor immune microenvironment, cancer-immunity cycle, and immunotherapy-related features

To evaluate the relationship between LRGS and the tumor immune microenvironment, we first compared immune score, stromal score, and ESTIMATE score between the two risk groups. The high-risk group showed significantly lower immune score (P=3.5×10⁻⁶), stromal score (P=0.00046), and ESTIMATE score (P=7×10⁻⁶) than the low-risk group. In addition, ssGSEA showed that several immune- and inflammation-related pathways were more active in the low-risk group (Figure 7A,7B). These findings suggest that LRGS-low tumors are associated with a relatively more immune-active microenvironment.

Figure 7 LRGS-high tumors are associated with lower inferred immune and stromal activity. (A) The low-risk group demonstrates higher Immune, Stromal, and ESTIMATE scores compared to the high-risk group; (B) the low-risk group shows significant enrichment in immune-related pathways, including complement and coagulation cascades, leukocyte migration, and adaptive immune signaling (T cell and B cell receptor pathways); (C,D) both CIBERSORT and ssGSEA analyses indicate a higher abundance of immune cells within the TME of the low-risk group compared to the high-risk group; (E) Venn diagram identifying key TME-infiltrating cell types through the intersection of differential expression, correlation, and survival analyses: Dendritic_cells_resting, Macrophages_M0, Mast_cells_resting, Monocytes, and T_cells_CD4_memory_resting; (F) higher CD27, CTLA4, and TIGIT expression in the low-risk group is consistent with a more immune-active transcriptomic state. ns, P≥0.05; *, P<0.05; **, P<0.01; ***, P<0.001. LRGS, lactylation-related gene signature; ns, not significant; ssGSEA, single-sample gene set enrichment analysis; TME, tumor microenvironment.

To further characterize immune-cell infiltration, we used CIBERSORT to estimate the relative abundance of 22 immune cell subsets. The high-risk group showed higher fractions of resting NK cells and activated mast cells. In contrast, the low-risk group showed higher fractions of memory B cells, resting CD4 memory T cells, monocytes, M0/M1 macrophages, resting dendritic cells, and resting mast cells. Similar trends were also observed using ssGSEA and xCell (Figure 7C,7D), supporting the robustness of the immune infiltration differences between the two groups.

We next examined the relationship between LRGS and specific immune-cell populations in more detail. Spearman correlation analysis identified seven immune cell types that were significantly associated with LRGS (P<0.001) (Figure S3A,S3B). Survival analysis further showed that nine immune cell types were significantly associated with OS in LUAD patients (P<0.05). By integrating the results of differential infiltration analysis, correlation analysis, and survival analysis, we identified five overlapping immune cell types: resting dendritic cells, M0 macrophages, resting mast cells, monocytes, and resting CD4 memory T cells (Figure 7E). These results indicate that these cell populations may be particularly relevant to the biological and prognostic differences captured by LRGS.

We also assessed the activity of the seven steps of the cancer-immunity cycle. Steps 1 and 7 were more active in the high-risk group, whereas the low-risk group showed stronger immune-cell recruitment capacity (Figure S3C,S3D). Thus, although some components of the cancer-immunity cycle appeared activated in LRGS-high tumors, the overall immune context of LRGS-low tumors remained more favorable for effective immune infiltration.

To explore immune checkpoint-related features, we compared the expression of multiple checkpoint genes between the two groups. CD27, CTLA4, and TIGIT were significantly upregulated in the low-risk group (Figure 7F). This pattern is consistent with a relatively more immune-active and potentially more regulable antitumor state in LRGS-low tumors. In addition, IPS analysis showed higher scores in the low-risk group under the CTLA4+/PD1+ and CTLA4+/PD1− settings, suggesting that this group may be more favorable in immunotherapy-related transcriptomic analyses (Figure S3E,S3F).

To further assess the relationship between LRGS and immunotherapy-related features, we analyzed the IMvigor210 cohort. The high-risk group showed higher tumor mutational burden, whereas the low-risk group showed a higher proportion of complete or partial responses. Consistently, the high-risk group included more cases with stable or progressive disease, and patients with stable or progressive disease had significantly higher risk scores than those with complete or partial responses. In addition, risk scores differed across patients with complete response, partial response, stable disease, and progressive disease (Figure S3G-S3J). These findings provide additional support for an association between LRGS and immunotherapy-related response patterns.

Drug sensitivity analysis

To explore the potential therapeutic relevance of LRGS, we next performed drug sensitivity analysis using the oncoPredict package. We focused on EGFR tyrosine kinase inhibitors, MET receptor tyrosine kinase inhibitors, and multi-target tyrosine kinase inhibitors. The predicted IC50 values of erlotinib, gefitinib, and savolitinib were significantly lower in the high-risk group. In addition, LRGS risk scores were negatively correlated with the IC50 values of these agents. By contrast, the IC50 of axitinib was lower in the low-risk group, and LRGS risk scores were positively correlated with axitinib IC50 (Figure S4A-S4D).

PPI, IHC and RT-qPCR validation

To further explore the biological relevance of genes included in LRGS, we constructed a PPI network based on the 15 signature genes using the STRING database (Figure 8A). Network centrality analysis showed that GAPDH had the highest degree centrality, whereas CCT6A had the highest betweenness centrality. TUBA1C ranked fifth according to both degree centrality and betweenness centrality, indicating that it was a network-connected candidate gene rather than the dominant hub node in the LRGS PPI network (Table S4). Several higher-ranked genes, such as GAPDH, CCT6A, PGK1, and YWHAZ, have already been implicated in lung cancer progression, tumor metabolism, or oncogenic signaling. Therefore, TUBA1C was selected for preliminary expression validation based on its inclusion in the final LRGS, prognostic relevance, biological plausibility as an α-tubulin isoform involved in cytoskeletal organization and mitotic regulation, prior evidence in LUAD/NSCLC or other malignancies, moderate network connectivity, and its relatively less characterized role in lactylation-related transcriptional programs in LUAD.

Figure 8 Preliminary expression-level validation of TUBA1C in LUAD tissues and NSCLC cell lines. (A) Protein-protein interaction network analysis of the 15 LRGS genes; (B,C) IHC analysis showing higher TUBA1C staining in LUAD tissues than in matched adjacent non-tumor tissues; (D) RT-qPCR analysis showing upregulated TUBA1C mRNA in NSCLC cell lines (A549, PC9, H1299) compared with normal bronchial epithelial cells (BEAS-2B). IHC, immunohistochemistry; LRGS, lactylation-related gene signature; LUAD, lung adenocarcinoma; ns, not significant; NSCLC, non-small cell lung cancer; RT-qPCR, reverse transcription quantitative polymerase chain reaction.

IHC showed that TUBA1C expression was lower in paired distal normal tissues and higher in LUAD tissues. Correspondingly, the H-score was significantly higher in tumor tissues than in matched normal tissues (Figure 8B,8C). In addition, RT-qPCR analysis showed that TUBA1C expression was significantly upregulated in A549, PC9, and H1299 cells compared with BEAS-2B cells (Figure 8D).

Discussion

LUAD is the most common histological subtype of NSCLC and remains a leading cause of cancer-related death worldwide (34). Its progression involves complex molecular events, including metabolic dysregulation, epigenetic alteration, and remodeling of the TME (35). In recent years, lactate has been increasingly recognized as more than a metabolic byproduct. It also functions as a signaling metabolite and participates in epigenetic regulation through lactylation. Previous work has shown that lactate can promote transcriptional activation and regulate key biological processes such as macrophage polarization (12). In parallel, tumor-derived lactate can impair the function of T cells, dendritic cells, and NK cells, thereby contributing to immune escape and tumor progression (36). However, the transcriptomic landscape and clinical relevance of LRGs in LUAD remain incompletely characterized.

In the present study, we integrated single-cell RNA-seq and bulk RNA-seq data to identify 48 LRGs by intersecting single-cell marker genes, phenotype-associated co-expression modules, and tumor-specific DEGs. Based on these data, we constructed a 15-gene LRGS using machine learning strategies. Enrichment analyses suggested that these genes were mainly associated with protein folding, pyruvate metabolism, and secretory granule lumen-related components. These results imply that lactylation-related states in LUAD may be linked not only to metabolic remodeling but also to broader biological programs involving proteostasis and secretory activity. Across multiple cohorts, LRGS was significantly associated with OS and also correlated with tumor stage and other adverse clinicopathological features. Together, these findings suggest that LRGS may capture a more aggressive biological phenotype in LUAD.

Although LRGS showed high discriminative performance in the TCGA-LUAD cohort, its performance was comparatively lower in GSE31210 and GSE72094. This difference may be attributable to several factors. First, the TCGA-LUAD cohort was used for model development, and the apparent performance in this cohort may therefore be partially influenced by model optimism or overfitting. Second, cross-platform differences between RNA-seq and microarray datasets may affect gene-expression distributions and model transferability. Third, clinical heterogeneity among cohorts, including differences in patient composition, disease stage distribution, treatment background, follow-up duration, and endpoint annotation, may also contribute to performance variation. In addition, cohort-specific normalization and cutoff determination may influence risk stratification across datasets. Therefore, although the external validation cohorts support the prognostic relevance of LRGS, its clinical generalizability should be further evaluated in larger, prospective, multicenter LUAD cohorts with standardized assay platforms and prespecified cutoffs.

From a translational perspective, LRGS may complement conventional clinicopathological risk stratification by capturing molecular risk features that are not fully represented by TNM stage or other anatomical parameters. In the TCGA-LUAD cohort, LRGS remained independently associated with OS after adjustment for available clinicopathological variables, and the LRGS-integrated nomogram showed better discriminative performance than individual clinical variables. DCA further suggested potential net clinical benefit compared with clinicopathological variables alone. These findings indicate that LRGS may improve individualized risk estimation when used together with conventional clinical factors.

Nevertheless, the clinical utility of LRGS remains preliminary. Before clinical implementation, the scoring formula, assay platform, normalization strategy, and cutoff should be standardized and externally calibrated. In addition, prospective multicenter validation and decision-impact studies are required to determine whether LRGS-guided stratification can improve treatment selection, follow-up intensity, or clinical outcomes in patients with LUAD.

To further explore the biological differences between LRGS-defined subgroups, we performed GSEA and GSVA analyses. The low-risk group was enriched in immune-activation programs, such as ALLOGRAFT_REJECTION and INFLAMMATORY_RESPONSE, as well as pathways linked to metabolic balance, including BILE_ACID_METABOLISM and KRAS_SIGNALING_DN. The enrichment of KRAS_SIGNALING_DN in the low-risk group is noteworthy. KRAS pathway dysregulation is common across several cancers, including lung cancer, and sustained KRAS activation can suppress apoptosis, enhance glycolysis, and promote angiogenesis and immune dysfunction (37,38). By contrast, the high-risk group was enriched in GLYCOLYSIS, HYPOXIA, MTORC1_SIGNALING, PI3K_AKT_MTOR_SIGNALING, REACTIVE_OXYGEN_SPECIES_PATHWAY, and TNFA_SIGNALING_VIA_NFKB. These pathways are closely linked to proliferative signaling, stress adaptation, metabolic remodeling, and aggressive tumor behavior (39-41). Therefore, the LRGS-high phenotype may reflect a malignant state characterized by enhanced metabolic activity, stronger proliferative drive, and greater tolerance to cellular stress.

ITH is an important contributor to malignant progression and therapeutic resistance, and higher MATH scores generally indicate increased ITH (42). In our study, the high-risk group showed significantly higher MATH scores, suggesting greater genomic complexity and clonal diversity. Mutation landscape analyses further showed that the two groups differed in mutation frequencies and CNV patterns. For example, COL11A1 mutations were more frequent in the high-risk group, and previous studies have suggested that COL11A1 may contribute to proliferation, colony formation, and migration in multiple cancers (43). In addition, the low-risk group showed more obvious co-occurrence of genomic alterations, whereas the high-risk group displayed distinct CNV patterns. These findings suggest that LRGS-defined subgroups may reflect different molecular states and evolutionary trajectories. However, the current data support association rather than direct causality. In other words, lactylation-related transcriptional states may coexist with broader genomic features linked to aggressive LUAD. This interpretation is also broadly consistent with evidence showing that high CNV burden and chromosomal instability are associated with poor prognosis in other malignancies, including acute myeloid leukemia (44).

Single-cell analyses suggested that LRGS-high-associated cells exhibited immune- and lysosome-related transcriptional shifts and a distinct ligand-receptor interaction landscape. Communication network analysis inferred relatively higher TGF-β, VEGF, and CXCL signaling in LRGS-high cells, suggesting that these pathways may be associated with LRGS-defined microenvironmental differences. From a biological-context perspective, prior studies have implicated TGF-β signaling in immunosuppressive and stromal/matrix remodeling programs, VEGF signaling in angiogenesis and vascular regulation, and CXCL chemokine networks in immune-cell recruitment and spatial organization (45-47). When considered together with the enrichment of hypoxia, glycolysis, and PI3K-AKT-mTOR-related programs in LRGS-high tumors, these findings suggest a transcriptomically inferred state in which metabolic stress, angiogenic signaling, chemokine signaling, and immune-contexture differences coexist. However, because these analyses were based on pathway enrichment and ligand-receptor inference from transcriptomic data, they do not establish a causal sequence linking lactylation-related transcriptional programs, TGF-β/VEGF/CXCL signaling, and immune remodeling. Consistent with this association-based interpretation, LRGS-low tumors exhibited higher Immune, Stromal, and ESTIMATE scores and enrichment of multiple immune-related programs, including complement/coagulation, leukocyte migration, and T/B cell receptor signaling. CIBERSORT and ssGSEA further supported higher inferred immune-cell abundance in LRGS-low tumors, whereas LRGS-high tumors showed lower inferred immune infiltration and lower predicted sensitivity to CTLA4-directed immunotherapy. Immune cells are central determinants of tumor immune escape and immunotherapy response; for example, macrophage polarization states can shape tumor growth and immunosuppressive intensity, and strategies that reprogram macrophages toward an M1 phenotype may enhance antitumor immunity (48,49). Lactate, as a key metabolite within the TME, has been reported to disrupt immune-cell metabolism and suppress T-cell, NK-cell, and dendritic-cell function (50). Prior studies also indicate that histone lactylation can participate in tumor immune regulation and macrophage-state transitions, and that lactylation-associated METTL3-mediated m6A regulation may contribute to immunosuppressive programs in tumor-infiltrating myeloid cells (51,52). These prior mechanistic studies provide biological context for our findings, but the present study did not directly measure lactate accumulation, protein/histone lactylation, or experimentally validate these signaling mechanisms in LUAD.

At the immune-subset level, LRGS-high tumors showed lower proportions of multiple immune populations, including memory B cells, resting CD4 memory T cells, monocytes, M0/M1 macrophages, resting dendritic cells, and resting mast cells. These results support an association between the LRGS-high state and a less immune-active immune contexture, but they do not demonstrate that lactate accumulation or lactylation directly reduces immune infiltration in LUAD. Notably, the LRGS-low group exhibited higher levels of resting mast cells and showed greater predicted sensitivity to CTLA4-directed immunotherapy. Prior studies have reported that enrichment of resting mast cells is associated with improved prognosis in KIRC (53), suggesting that this subset may also represent a potentially protective component in LUAD. The mechanistic roles of resting mast cells in LUAD, however, require further investigation.

Metabolic remodeling provides a fundamental basis for malignant progression. The Warburg effect supplies energy and biosynthetic substrates but also results in lactate accumulation, thereby shaping a microenvironment conducive to tumor development. With accumulating evidence, lactate is increasingly recognized as more than a metabolic byproduct; it acts as a signaling and regulatory metabolite that can suppress T-cell activity, facilitate immune escape, and modulate the TME (54). In our analyses, GAPDH and LDHA expression was significantly associated with poor prognosis in LUAD, supporting a central contribution of heightened glycolysis to disease progression. PPI analyses further suggested that GAPDH and LDHA occupy hub-like positions, connecting with genes such as PGK1, NME1, CCT6A, YWHAZ, and TUBA1C. These observations suggest a systems-level link between glycolytic flux and processes such as protein synthesis and cytoskeletal organization, highlighting the coordinated regulatory architecture of LRGs. Mechanistically, LDHA catalyzes the pyruvate–lactate interconversion and correlates with Warburg intensity (55). Prior work indicates that elevated LDHA can enhance proliferation, invasion, and metastasis in rectal cancer and may promote liver metastasis by increasing glycolysis and histone lactylation through upstream signaling axes (56,57). In LUAD, suppressing LDHA has been reported to attenuate proliferation, invasion, and migration of cancer cells (58). Collectively, these studies support the view that LDHA-driven lactate metabolism is not only an energetic foundation but also a key bridge linking tumor metabolism, epigenetic regulation, and malignant phenotypes. Although immunotherapy has improved outcomes in subsets of LUAD patients, a considerable fraction remains refractory, and lactate-associated immunosuppression may contribute to suboptimal responses (59).

In our study, TUBA1C was prioritized as a candidate gene for further evaluation among the 15 LRGS genes. This prioritization was based on its inclusion in the final prognostic signature, its association with poor prognosis in LUAD, its biological plausibility as an α-tubulin isoform involved in cytoskeletal organization and mitotic regulation, and prior evidence linking TUBA1C or TUBA1C-related transcripts to tumor progression in several malignancies. In the STRING-derived PPI network, TUBA1C ranked fifth according to both degree centrality and betweenness centrality, suggesting that it had network connectivity within the LRGS gene set, although it should not be interpreted as the dominant hub node.

Accumulating evidence indicates that TUBA1C may have context-dependent relevance across tumor types. In breast cancer, TUBA1C expression has been associated with tumor behavior and paclitaxel resistance (60). In low-grade glioma, TUBA1C expression correlates with immune-cell infiltration, and experimental knockdown has been reported to suppress glioma cell growth and viability (61,62). TUBA1C is also an α-tubulin isoform that contributes to microtubule structure, cytoskeletal dynamics, and mitotic processes, which are closely related to cancer cell proliferation and cell-cycle progression (63,64). In breast cancer and hepatocellular carcinoma, TUBA1C has been linked to aerobic glycolysis, cell growth, migration, proliferation, and poor prognosis (65,66). In addition, proteomic profiling has suggested altered TUBA1C expression in astrocytes exposed to lung cancer cell-derived exosomes, supporting a potential relationship between lung cancer-associated intercellular communication and TUBA1C-related protein alterations (67).

Functional evidence from several cancer models further supports the biological plausibility of TUBA1C as a candidate gene. For example, circTUBA1C has been reported to accelerate NSCLC progression by regulating the miR-143-3p axis (68); in pancreatic ductal adenocarcinoma, glioma, and gastric cancer models, TUBA1C knockdown or genetic perturbation has been associated with reduced proliferation, altered cell-cycle progression, increased apoptosis, and impaired migration or invasion, potentially through cell-cycle-related regulators such as cyclins and cyclin-dependent kinases (69-71). These prior studies provide a reasonable biological context for our findings and are consistent with the enrichment of G2M_CHECKPOINT, E2F_TARGETS, and MITOTIC_SPINDLE-related programs in the LRGS-high group.

To further contextualize TUBA1C within broader tumor-biological programs, we considered it together with the pathway features observed in LRGS-defined subgroups. GSEA and GSVA indicated that the LRGS-high subtype was enriched in oncogenic and stress-adaptive programs, including glycolysis, hypoxia, PI3K–AKT–mTOR signaling, G2M_CHECKPOINT, E2F_TARGETS, and MITOTIC_SPINDLE. Clinically, higher activity of glycolysis- and hypoxia-related programs was associated with unfavorable survival, and proliferation-related programs were also more active in the LRGS-high group. This pattern is consistent with prior reports in which TUBA1C and its co-expressed genes were enriched in cell-cycle-related pathways, including the KEGG cell-cycle pathway (70).

These observations suggest that TUBA1C may be embedded in a broader transcriptional context characterized by proliferative activity, cytoskeletal organization, metabolic reprogramming, and stress adaptation. However, because these pathway analyses were performed at the LRGS subgroup level rather than after direct perturbation of TUBA1C, they should not be interpreted as evidence that TUBA1C directly regulates these pathways. Importantly, our tissue- and cell-level validation supported the expression-level relevance of TUBA1C: IHC analysis showed higher TUBA1C protein expression in LUAD tissues than in matched adjacent non-tumor tissues, and RT-qPCR indicated higher TUBA1C expression in NSCLC cell lines (A549, PC9, H1299) compared with BEAS-2B cells. Therefore, TUBA1C may represent a biologically plausible candidate gene associated with aggressive LRGS-related transcriptional states, but its causal function and therapeutic relevance require further validation through dedicated functional experiments.

Several limitations of this study should be acknowledged. First, this was a retrospective prediction model study based on public transcriptomic datasets, and missing data were handled using a complete-case approach. Therefore, potential selection bias cannot be fully excluded. Second, although the LRGS was externally validated in GSE31210 and GSE72094, further validation in larger, independent, multicenter, and ideally prospective LUAD cohorts is still required before clinical application. Third, the LRGS was derived from transcriptomic profiles of LRGs rather than direct lactylome or proteomic lactylation data. Therefore, the signature should be interpreted as a lactylation-associated transcriptional signature. Fourth, although RT-qPCR in NSCLC cell lines and IHC in clinical samples supported increased TUBA1C expression, these assays were limited to expression-level validation, and the number of clinical specimens was small. No functional assays were performed to determine whether TUBA1C causally contributes to LUAD progression, immune modulation, or lactylation-related pathways. Therefore, TUBA1C should be regarded as a candidate gene for further functional investigation rather than a confirmed mechanistic driver. In addition, the immunotherapy and drug-sensitivity analyses were exploratory and require further validation in LUAD-specific treatment cohorts. The biological roles and regulatory mechanisms of the remaining 14 genes in the prognostic signature also merit systematic investigation. Detailed treatment information and cohort-specific accrual/follow-up annotations were not uniformly available or harmonized across all public cohorts, and no model recalibration or updating was performed in the external validation cohorts.

Conclusions

In conclusion, we developed and validated a LRGS comprising 15 genes for prognostic stratification in LUAD. LRGS showed favorable prognostic stratification ability across cohorts and was associated with clinicopathological features, supporting the relevance of lactylation-related transcriptional programs in LUAD. Despite limitations, further validation in larger cohorts and mechanistic studies will be needed to clarify causal links and improve clinical translation. Overall, these findings support the potential utility of LRGs in prognostic assessment of LUAD. In particular, the integrated computational analyses and preliminary experimental validation implicate TUBA1C as a candidate biomarker, while its functional role and therapeutic relevance require further investigation.

Acknowledgments

We would like to express our sincere gratitude to all the colleagues and institutions that provided support for this study. We also thank the reviewers for their valuable comments and suggestions, which have greatly improved the quality of this manuscript. We also acknowledge the TCGA, GEO, TCIA, GDSC2, and STRING databases, as well as the staff of the Department of Thoracic Surgery and the Institutional Ethics Committee of The Second Affiliated Hospital, Jiangxi Medical College, Nanchang University, for their assistance.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0963/rc

Data Sharing Statement: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0963/dss

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0963/prf

Funding: This work was supported by the National Natural Science Foundation of China (No. 82160410), the Key Research and Development Program of Jiangxi Province (No. 20223BBG71009), and the Jiangxi Province Graduate Innovation Fund Project (No. YC2024-B053). The funders had no role in study design, data collection, analysis, interpretation, manuscript writing, or the decision to submit for publication.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0963/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Institutional Ethics Committee of The Second Affiliated Hospital of Nanchang University [No. (2018) CDEFYYLK(3-05)]; written informed consent was obtained from all participants.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Relli V, Trerotola M, Guerra E, et al. Abandoning the Notion of Non-Small Cell Lung Cancer. Trends Mol Med 2019;25:585-94. [Crossref] [PubMed]
2. Little AG, Gay EG, Gaspar LE, et al. National survey of non-small cell lung cancer in the United States: epidemiology, pathology and patterns of care. Lung Cancer 2007;57:253-60. [Crossref] [PubMed]
3. de Sousa VML, Carvalho L. Heterogeneity in Lung Cancer. Pathobiology 2018;85:96-107. [Crossref] [PubMed]
4. Brahmer JR, Tykodi SS, Chow LQ, et al. Safety and activity of anti-PD-L1 antibody in patients with advanced cancer. N Engl J Med 2012;366:2455-65. [Crossref] [PubMed]
5. Schabath MB, Cote ML. Cancer Progress and Priorities: Lung Cancer. Cancer Epidemiol Biomarkers Prev 2019;28:1563-79. [Crossref] [PubMed]
6. National Lung Screening Trial Research Team. Results of initial low-dose computed tomographic screening for lung cancer. N Engl J Med 2013;368:1980-91. [Crossref] [PubMed]
7. Chen H, Wang Z, Sun C, et al. MALMPS: A Machine Learning-Based Metabolic Gene Prognostic Signature for Stratifying Clinical Outcomes and Molecular Heterogeneity in Stage II/III Colorectal Cancer. Adv Sci (Weinh) 2025;12:e01333. [Crossref] [PubMed]
8. Wang Y, Zhu H, Ren J, et al. Integrative machine learning models predict prostate cancer diagnosis and biochemical recurrence risk: Advancing precision oncology. NPJ Digit Med 2025;8:524. [Crossref] [PubMed]
9. Yu C, Bian Y, Gao Y, et al. Machine learning-based lactate-related genes signature predicts clinical outcomes and unveils novel therapeutic targets in esophageal squamous cell carcinoma. Cancer Lett 2025;613:217458. [Crossref] [PubMed]
10. Cui L, Zhao S, Teng HL, et al. Integrins identified as potential prognostic markers in osteosarcoma through multi-omics and multi-dataset analysis. NPJ Precis Oncol 2025;9:19. [Crossref] [PubMed]
11. Li X, Yang Y, Zhang B, et al. Lactate metabolism in human health and disease. Signal Transduct Target Ther 2022;7:305. [Crossref] [PubMed]
12. Zhang D, Tang Z, Huang H, et al. Metabolic regulation of gene expression by histone lactylation. Nature 2019;574:575-80. [Crossref] [PubMed]
13. Yu J, Chai P, Xie M, et al. Histone lactylation drives oncogenesis by facilitating m(6)A reader protein YTHDF2 expression in ocular melanoma. Genome Biol 2021;22:85. [Crossref] [PubMed]
14. Deng J, Liao X. Lysine lactylation (Kla) might be a novel therapeutic target for breast cancer. BMC Med Genomics 2023;16:283. [Crossref] [PubMed]
15. Luo Y, Yang Z, Yu Y, et al. HIF1α lactylation enhances KIAA1199 transcription to promote angiogenesis and vasculogenic mimicry in prostate cancer. Int J Biol Macromol 2022;222:2225-43. [Crossref] [PubMed]
16. Jiang J, Huang D, Jiang Y, et al. Lactate Modulates Cellular Metabolism Through Histone Lactylation-Mediated Gene Expression in Non-Small Cell Lung Cancer. Front Oncol 2021;11:647559. [Crossref] [PubMed]
17. Zhang R, Li L, Yu J. Lactate-induced IGF1R protein lactylation promotes proliferation and metabolic reprogramming of lung cancer cells. Open Life Sci 2024;19:20220874. [Crossref] [PubMed]
18. Hanahan D, Weinberg RA. Hallmarks of cancer: the next generation. Cell 2011;144:646-74. [Crossref] [PubMed]
19. Ganapathy-Kanniappan S, Geschwind JF. Tumor glycolysis as a target for cancer therapy: progress and prospects. Mol Cancer 2013;12:152. [Crossref] [PubMed]
20. Wu F, Fan J, He Y, et al. Single-cell profiling of tumor heterogeneity and the microenvironment in advanced non-small cell lung cancer. Nat Commun 2021;12:2540. [Crossref] [PubMed]
21. Madeddu C, Donisi C, Liscia N, et al. EGFR-Mutated Non-Small Cell Lung Cancer and Resistance to Immunotherapy: Role of the Tumor Microenvironment. Int J Mol Sci 2022;23:6489. [Crossref] [PubMed]
22. Hake SB, Xiao A, Allis CD. Linking the epigenetic ‘language’ of covalent histone modifications to cancer. Br J Cancer 2004;90:761-9. [Crossref] [PubMed]
23. Bates D, Eastman A. Microtubule destabilising agents: far more than just antimitotic anticancer drugs. Br J Clin Pharmacol 2017;83:255-68. [Crossref] [PubMed]
24. Cheng Z, Huang H, Li M, et al. Lactylation-Related Gene Signature Effectively Predicts Prognosis and Treatment Responsiveness in Hepatocellular Carcinoma. Pharmaceuticals (Basel) 2023;16:644. [Crossref] [PubMed]
25. Yang H, Zou X, Yang S, et al. Identification of lactylation related model to predict prognostic, tumor infiltrating immunocytes and response of immunotherapy in gastric cancer. Front Immunol 2023;14:1149989. [Crossref] [PubMed]
26. Rosenberg JE, Hoffman-Censits J, Powles T, et al. Atezolizumab in patients with locally advanced and metastatic urothelial carcinoma who have progressed following treatment with platinum-based chemotherapy: a single-arm, multicentre, phase 2 trial. Lancet 2016;387:1909-20. [Crossref] [PubMed]
27. Korsunsky I, Millard N, Fan J, et al. Fast, sensitive and accurate integration of single-cell data with Harmony. Nat Methods 2019;16:1289-96. [Crossref] [PubMed]
28. Liu Z, Liu L, Weng S, et al. Machine learning-based integration develops an immune-derived lncRNA signature for improving outcomes in colorectal cancer. Nat Commun 2022;13:816. [Crossref] [PubMed]
29. Ma D, Jiang YZ, Liu XY, et al. Clinical and molecular relevance of mutant-allele tumor heterogeneity in breast cancer. Breast Cancer Res Treat 2017;162:39-48. [Crossref] [PubMed]
30. Rajput A, Bocklage T, Greenbaum A, et al. Mutant-Allele Tumor Heterogeneity Scores Correlate With Risk of Metastases in Colon Cancer. Clin Colorectal Cancer 2017;16:e165-70. [Crossref] [PubMed]
31. Mroz EA, Tward AD, Hammon RJ, et al. Intra-tumor genetic heterogeneity and mortality in head and neck cancer: analysis of data from the Cancer Genome Atlas. PLoS Med 2015;12:e1001786. [Crossref] [PubMed]
32. Maeser D, Gruener RF, Huang RS. oncoPredict: an R package for predicting in vivo or cancer patient drug response and biomarkers from cell line screening data. Brief Bioinform 2021;22:bbab260. [Crossref] [PubMed]
33. Yang W, Soares J, Greninger P, et al. Genomics of Drug Sensitivity in Cancer (GDSC): a resource for therapeutic biomarker discovery in cancer cells. Nucleic Acids Res 2013;41:D955-61. [Crossref] [PubMed]
34. Sung H, Ferlay J, Siegel RL, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin 2021;71:209-49. [Crossref] [PubMed]
35. Saji H, Tsuboi M, Shimada Y, et al. Gene expression profiling and molecular pathway analysis for the identification of early-stage lung adenocarcinoma patients at risk for early recurrence. Oncol Rep 2013;29:1902-6. [Crossref] [PubMed]
36. Husain Z, Huang Y, Seth P, et al. Tumor-derived lactate modifies antitumor immune response: effect on myeloid-derived suppressor cells and NK cells. J Immunol 2013;191:1486-95. [Crossref] [PubMed]
37. Cox AD, Der CJ. The dark side of Ras: regulation of apoptosis. Oncogene 2003;22:8999-9006. [Crossref] [PubMed]
38. Pylayeva-Gupta Y, Grabocka E, Bar-Sagi D. RAS oncogenes: weaving a tumorigenic web. Nat Rev Cancer 2011;11:761-74. [Crossref] [PubMed]
39. Laplante M, Sabatini DM. mTOR signaling in growth control and disease. Cell 2012;149:274-93. [Crossref] [PubMed]
40. Aggarwal BB, Shishodia S, Sandur SK, et al. Inflammation and cancer: how hot is the link? Biochem Pharmacol 2006;72:1605-21. [Crossref] [PubMed]
41. Peng Y, Wang Y, Zhou C, et al. PI3K/Akt/mTOR Pathway and Its Role in Cancer Therapeutics: Are We Making Headway? Front Oncol 2022;12:819128. [Crossref] [PubMed]
42. Dagogo-Jack I, Shaw AT. Tumour heterogeneity and resistance to cancer therapies. Nat Rev Clin Oncol 2018;15:81-94. [Crossref] [PubMed]
43. Wan X, Deng Q, Chen A, et al. Bioinformatics analysis and experimental validation of the oncogenic role of COL11A1 in pan-cancer. 3 Biotech 2024;14:290.
44. Su Y, Han Z, Ji Y, et al. Patterns and variations of copy number alterations in acute myeloid leukemia: insights from the LeukAtlas database. Leukemia 2025;39:827-36. [Crossref] [PubMed]
45. Hou W. Role of TGFβ-activated cancer-associated fibroblasts in the resistance to checkpoint blockade immunotherapy. Front Oncol 2025;15:1602452. [Crossref] [PubMed]
46. Lanitis E, Irving M, Coukos G. Tumour-associated vasculature in T cell homing and immunity: opportunities for cancer therapy. Nat Rev Immunol 2025;25:831-46. [Crossref] [PubMed]
47. Ozga AJ, Chow MT, Luster AD. Chemokines and the immune response to cancer. Immunity 2021;54:859-74. [Crossref] [PubMed]
48. Gunassekaran GR, Poongkavithai Vadevoo SM, Baek MC, et al. M1 macrophage exosomes engineered to foster M1 polarization and target the IL-4 receptor inhibit tumor growth by reprogramming tumor-associated macrophages into M1-like macrophages. Biomaterials 2021;278:121137. [Crossref] [PubMed]
49. Pan Y, Yu Y, Wang X, et al. Tumor-Associated Macrophages in Tumor Immunity. Front Immunol 2020;11:583084. [Crossref] [PubMed]
50. Brown TP, Ganapathy V. Lactate/GPR81 signaling and proton motive force in cancer: Role in angiogenesis, immune escape, nutrition, and Warburg phenomenon. Pharmacol Ther 2020;206:107451. [Crossref] [PubMed]
51. Irizarry-Caro RA, McDaniel MM, Overcast GR, et al. TLR signaling adapter BCAP regulates inflammatory to reparatory macrophage transition by promoting histone lactylation. Proc Natl Acad Sci U S A 2020;117:30628-38. [Crossref] [PubMed]
52. Xiong J, He J, Zhu J, et al. Lactylation-driven METTL3-mediated RNA m(6)A modification promotes immunosuppression of tumor-infiltrating myeloid cells. Mol Cell 2022;82:1660-1677.e10. [Crossref] [PubMed]
53. Cui Y, Shen T, Xu F, et al. KCNN4 may weaken anti-tumor immune response via raising Tregs and diminishing resting mast cells in clear cell renal cell carcinoma. Cancer Cell Int 2022;22:211. [Crossref] [PubMed]
54. Rabinowitz JD, Enerbäck S. Lactate: the ugly duckling of energy metabolism. Nat Metab 2020;2:566-71. [Crossref] [PubMed]
55. Le A, Cooper CR, Gouw AM, et al. Inhibition of lactate dehydrogenase A induces oxidative stress and inhibits tumor progression. Proc Natl Acad Sci U S A 2010;107:2037-42. [Crossref] [PubMed]
56. Zhou J, Xu W, Wu Y, et al. GPR37 promotes colorectal cancer liver metastases by enhancing the glycolysis and histone lactylation via Hippo pathway. Oncogene 2023;42:3319-30. [Crossref] [PubMed]
57. Cheng CS, Tan HY, Wang N, et al. Functional inhibition of lactate dehydrogenase suppresses pancreatic adenocarcinoma progression. Clin Transl Med 2021;11:e467. [Crossref] [PubMed]
58. Gao M, Wang M, Zhou S, et al. Machine learning-based prognostic model of lactylation-related genes for predicting prognosis and immune infiltration in patients with lung adenocarcinoma. Cancer Cell Int 2024;24:400. [Crossref] [PubMed]
59. Lahiri A, Maji A, Potdar PD, et al. Lung cancer immunotherapy: progress, pitfalls, and promises. Mol Cancer 2023;22:40. [Crossref] [PubMed]
60. Nami B, Wang Z. Genetics and Expression Profile of the Tubulin Gene Superfamily in Breast Cancer Subtypes and Its Relation to Taxane Resistance. Cancers (Basel) 2018;10:274. [Crossref] [PubMed]
61. Sun YF, Zhang LC, Niu RZ, et al. Predictive potentials of glycosylation-related genes in glioma prognosis and their correlation with immune infiltration. Sci Rep 2024;14:4478. [Crossref] [PubMed]
62. Zhu H, Hu X, Gu L, et al. TUBA1C is a Prognostic Marker in Low-grade Glioma and Correlates with Immune Cell Infiltration in the Tumor Microenvironment. Front Genet 2021;12:759953. [Crossref] [PubMed]
63. Tangutur AD, Kumar D, Krishna KV, et al. Microtubule Targeting Agents as Cancer Chemotherapeutics: An Overview of Molecular Hybrids as Stabilizing and Destabilizing Agents. Curr Top Med Chem 2017;17:2523-37. [Crossref] [PubMed]
64. Liu X, Chen Y, Li Y, et al. Targeting mitosis exit: A brake for cancer cell proliferation. Biochim Biophys Acta Rev Cancer 2019;1871:179-91. [Crossref] [PubMed]
65. Wu Z, Sun S, Fan R, et al. Tubulin alpha 1c promotes aerobic glycolysis and cell growth through upregulation of yes association protein expression in breast cancer. Anticancer Drugs 2022;33:132-41. [Crossref] [PubMed]
66. Wang J, Chen W, Wei W, et al. Oncogene TUBA1C promotes migration and proliferation in hepatocellular carcinoma and predicts a poor prognosis. Oncotarget 2017;8:96215-24. [Crossref] [PubMed]
67. Ye L, Wu Y, Zhou J, et al. Influence of Exosomes on Astrocytes in the Pre-Metastatic Niche of Lung Cancer Brain Metastases. Biol Proced Online 2023;25:5. [Crossref] [PubMed]
68. Yang J, Jia Y, Wang B, et al. Circular RNA TUBA1C accelerates the progression of non-small-cell lung cancer by sponging miR-143-3p. Cell Signal 2020;74:109693. [Crossref] [PubMed]
69. Albahde MAH, Zhang P, Zhang Q, et al. Upregulated Expression of TUBA1C Predicts Poor Prognosis and Promotes Oncogenesis in Pancreatic Ductal Adenocarcinoma via Regulating the Cell Cycle. Front Oncol 2020;10:49. [Crossref] [PubMed]
70. Gui S, Chen P, Liu Y, et al. TUBA1C expression promotes proliferation by regulating the cell cycle and indicates poor prognosis in glioma. Biochem Biophys Res Commun 2021;577:130-8. [Crossref] [PubMed]
71. Wang H, Cui H, Yang X, et al. TUBA1C: a new potential target of LncRNA EGFR-AS1 promotes gastric cancer progression. BMC Cancer 2023;23:258. [Crossref] [PubMed]