Multi-Omics Characterization of Lactate-Associated Molecular Subtypes in Lung Cancer Suggests a Role for DKK1 in Lactate-Linked Migration, Invasion, and Lactylation Programs

Simple Summary

Metabolic reprogramming is a common feature of various cancers, and lactate metabolism disorder plays a crucial role in lung cancer progression. This study aims to explore the molecular subtypes related to lactate metabolism in lung cancer and the key genes involved. By integrating multi-omics data and functional experiments, we identified lactate-based subtypes with different prognostic outcomes and a lactate-related prognostic gene signature. We also found that the lactate-regulated gene Dickkopf-1 (DKK1) promotes lung cancer metastasis. These findings provide new insights for lung cancer prognosis assessment and targeted therapy.

Abstract

Background: Lactate accumulation is increasingly recognized as a feature of tumor metabolic reprogramming that can coincide with immune dysregulation and aggressive phenotypes. The prognostic and immunologic relevance of lactate-associated heterogeneity in lung cancer remains to be clarified. Methods: We curated lactate-related genes and identified prognostic candidates in lung cancer cohorts. Consensus clustering was applied to define lactate-associated molecular subtypes, followed by characterization of survival and tumor microenvironment features. A LASSO-based gene signature was developed to generate an individual-level risk score and an integrated nomogram. Multi-omics analyses were used to evaluate concordance between transcriptomic and proteomic alterations. Single-cell transcriptomic data were analyzed to explore cellular heterogeneity in lactate-related programs. In vitro assays evaluated the response of candidate genes to lactate exposure and assessed cell migration and invasion under proliferation-inhibited conditions after genetic perturbation. Results: Two lactate-associated molecular subtypes were identified with distinct overall survival and divergent immune microenvironment features. Subtype 1 was associated with better outcomes and a more immune-inflamed profile, whereas Subtype 2 was associated with poorer outcomes and a myeloid-enriched, immunosuppressive contexture. Pathway analyses indicated subtype-associated differences in extracellular matrix-related processes and apoptosis-associated signaling. We developed an 11-gene prognostic signature and nomogram that stratified patients by risk across TCGA and GEO cohorts. Multi-omics integration highlighted ANLN, FGA, and DKK1 as consistently dysregulated at both transcript and protein levels. Among these candidates, DKK1 showed lactate-responsive induction in vitro. DKK1 perturbation altered lactate-enhanced migratory and invasive phenotypes and was accompanied by changes in intracellular lactate levels and global protein lactylation, supporting a potential feedforward relationship between lactate exposure, DKK1 expression, and lactylation. Conclusions: This study characterizes lactate-associated molecular heterogeneity in lung cancer and provides a lactate-related subtype framework and prognostic risk model for patient stratification. The findings nominate DKK1 as a lactate-responsive candidate linked to migration/invasion phenotypes and lactate/lactylation changes in vitro.

1. Introduction

Lung cancer remains the leading cause of global cancer-related mortality, accounting for approximately 12.4% of incident cancer cases and 18.7% of cancer-associated deaths, thereby remaining the predominant cancer burden worldwide [1]. Non-small cell lung cancer (NSCLC) constitutes approximately 85% of all lung cancer cases, with adenocarcinoma and squamous cell carcinoma being the two predominant histological subtypes. In contrast, small cell lung cancer (SCLC) comprises the remaining 15% of cases and is distinguished by its characteristically aggressive clinicopathological features [2]. Despite paradigm-shifting therapeutic innovations over the past decade—including video-assisted thoracic surgery platforms, platinum-based combination chemotherapy, and programmed cell death protein 1/programmed death-ligand 1 (PD-1/PD-L1) immune checkpoint inhibitors—metastatic lung cancer outcomes remain grave, as evidenced by persistently low (20%) 5-year survival rates [3]. Even molecularly targeted therapies directed against oncogenic drivers (e.g., EGFR, ALK, ROS1) demonstrate a median progression-free survival of merely 9–15 months, primarily due to the inevitable resistance mechanisms mediated through bypass signaling pathways and tumor heterogeneity [4,5,6]. Consequently, the identification of robust predictive biomarkers for therapeutic stratification and the development of mechanistically innovative approaches to overcome treatment resistance represent urgent priorities in translational oncology research.

Lactate, once regarded merely as a glycolytic metabolic product, has emerged as a critical oncometabolite that promotes tumor progression and facilitates immune evasion in lung cancer [7]. NSCLC, characterized by a hypermetabolic phenotype, exhibits heightened aerobic glycolysis (Warburg effect) under both hypoxic and normoxic conditions, resulting in significant lactate accumulation within the tumor microenvironment (TME) [8]. Elevated intratumoral lactate levels show a strong positive correlation with advanced tumor stage and higher T-classification in NSCLC patients, indicating unfavorable prognosis [7]. Mechanistically, lactate dehydrogenase A (LDHA), a key enzyme catalyzing the conversion of pyruvate to lactate, is commonly overexpressed in NSCLC. It promotes tumor proliferation and invasion through HIF-1α-mediated metabolic reprogramming [9]. For instance, in vitro studies demonstrate that silencing LDHA suppresses NSCLC cell migration via targeting the miR-7-5p/c-Myc axis [10]. Clinically, elevated serum LDHA levels are associated with reduced efficacy of platinum-based chemotherapy and resistance to immune checkpoint inhibitors, highlighting its dual role as a metabolic driver and a mediator of therapeutic resistance [11]. These findings position lactate-centric pathways as promising targets for overcoming therapeutic resistance and reshaping the immunosuppressive TME in lung cancer.

Immune checkpoint blockade (ICB), a cornerstone of contemporary cancer immunotherapy, has demonstrated remarkable clinical efficacy across multiple malignancies. However, its therapeutic effectiveness remains limited in NSCLC, with objective response rates typically below 20% [12,13]. Macintyre et al. first reported that prolonged lactate exposure (10–20 mM) drives dysfunction in cytotoxic T lymphocytes, characterized by suppressed interferon-γ (IFN-γ) production and impaired proliferation in both oncologic and inflammatory disease models [14]. This metabolic reprogramming is further amplified through PKM2-mediated mechanisms. As the terminal glycolytic enzyme, PKM2 not only accelerates the Warburg effect but also promotes immunosuppressive TME remodeling by facilitating chemotactic recruitment of tumor-associated macrophages and myeloid-derived suppressor cells [15]. Notably, our mechanistic understanding has recently expanded to include PD-L1 upregulation driven by the lactate-GPR81 signaling axis. In NSCLC cells, lactate stimulation triggers GPR81-mediated transcriptional activation, resulting in increased PD-L1 protein expression and enhanced membrane localization. Conversely, genetic silencing of GPR81 abolishes this immune checkpoint induction by inhibiting STAT3 phosphorylation [16]. These findings collectively highlight the critical need for comprehensive characterization of lactate-mediated immune evasion mechanisms in NSCLC. Although metabolic gene signatures have been investigated for prognostic stratification, the clinical relevance of lactate-related genes (LRGs) remains largely unexplored.

In this study, we systematically analyzed multi-cohort transcriptomic data from The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) to characterize lactate-related gene programs in lung adenocarcinoma (LUAD) and squamous cell carcinoma (LUSC). Using consensus clustering, we defined two lactate-associated molecular subtypes with distinct clinical outcomes and tumor microenvironment contexture: Subtype 1 showed improved survival with an immune-inflamed profile, whereas Subtype 2 showed poorer survival with a myeloid-enriched, immunosuppressive contexture. We further examined pathway-level differences between subtypes, highlighting extracellular matrix remodeling–related processes and apoptosis-associated signaling as candidate biological axes linked to divergent pathogenic behaviors. We subsequently developed and validated a LASSO-derived prognostic model across independent cohorts to quantify individual risk. Finally, guided by multi-omics prioritization and in vitro perturbation assays, we investigated a plausible mechanistic link between lactate exposure, DKK1 induction, and lactylation-associated programs as a candidate explanation for subtype-associated malignant phenotypes, nominating the lactate–DKK1 axis as a potential therapeutic vulnerability.

2. Materials and Methods

2.1. Data Acquisition

Transcriptomic profiles and corresponding clinical data were acquired from The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) databases. The training cohort (TCGA-LUNG) and validation cohort (GSE50081) were obtained from TCGA and GEO, respectively. A curated list of 207 lactate-related genes (LRGs) was compiled from published literature for subsequent analyses.

2.2. Identification of Lactate-Driven Molecular Subtypes

Consensus clustering analysis was performed using the “ConsensusClusterPlus” R package to stratify LUNG patients into lactate-associated subtypes based on the expression profiles of 207 LRGs.

2.3. Differential Gene Expression and Functional Enrichment Analysis

Differentially expressed genes (DEGs) between lactate-related subtypes were identified using the “limma” R package with thresholds of |log₂ fold change (FC)| ≥ 1.5 and adjusted p-value ≤ 0.05. A total of 1222 significant DEGs (602 upregulated, 620 downregulated) were subjected to Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses. Visualization was implemented using “ggplot2” and “clusterProfiler”.

2.4. Prognostic Model Construction via Cox Regression

Univariate Cox regression analysis identified survival-associated LRGs (p < 0.05). Significant candidates were subsequently refined using least absolute shrinkage and selection operator (LASSO)-penalized multivariate Cox regression to prevent overfitting. The final prognostic signature was constructed using optimized coefficients.

2.5. Model Validation and Gene Expression Profiling

The predictive efficacy of the model was evaluated in both TCGA-LUNG and GSE50081 cohorts. Overall survival differences between high- and low-risk groups were assessed using Kaplan–Meier survival analysis with log-rank tests. The accuracy of the prognostic model was evaluated using receiver operating characteristic (ROC) curves analysis. Expression differences in signature genes (p < 0.05) were visualized using boxplots generated in Origin software(version 2023, OriginLab Corporation, Northampton, MA, USA).

2.6. Immune Infiltration and Tumor Microenvironment (TME) Characterization

The relative proportions of 22 immune cell types were quantified using CIBERSORTx (https://cibersortx.stanford.edu/) [17]. TME composition was assessed via ESTIMATE algorithm (“estimate” R package), which generated ImmuneScore, StromalScore, and ESTIMATEScore. Violin plots (“vioplot” R package) illustrated subtype-specific differences in immune infiltration (p ≤ 0.05).

2.7. Proteomics Sample Preparation, Mass Spectrometry, and Data Analysis

All proteomics experiments were performed by Shanghai Baiqu Biomedical Technology Co., Ltd, Shanghai, China. Lung cancer cells in logarithmic growth phase were assigned to a control group or a sodium lactate–treated group (n = 6). Cells were washed with ice-cold PBS, lysed in SDT buffer, and sonicated to extract total proteins. Protein concentration was measured by BCA assay. For each sample, 100 μg protein was reduced, alkylated, and digested with trypsin at 37 °C overnight. Peptides were desalted using C18 StageTips, lyophilized, and stored at −80 °C. Peptides were reconstituted and analyzed on an EASY-nLC 1200 system coupled to a Q Exactive HF-X mass spectrometer (Thermo Scientific). Peptides were separated on a reversed-phase C18 column using solvent A (0.1% formic acid in water) and solvent B (0.1% formic acid in 80% acetonitrile) with a 60 min linear gradient. Data were acquired in DIA mode with MS1 scans (m/z 350–1500, resolution 120,000) and HCD MS2 scans (resolution 30,000). Raw data were processed in MaxQuant (v2.0.3.0) against the UniProt human database. Carbamidomethylation (Cys) was set as a fixed modification, and methionine oxidation and protein N-terminal acetylation as variable modifications. Trypsin was specified with up to two missed cleavages; FDR was controlled at 1% at peptide and protein levels. Differentially expressed proteins were defined as FC ≥ 1.5 or ≤0.667 with p < 0.05, followed by GO and KEGG enrichment analyses.

2.8. Cell Proliferation Assay (CCK-8)

Cell proliferation was evaluated using the Cell Counting Kit-8 (CCK-8; Dojindo LaboratoriesKumamoto, Japan). Briefly, cells (5 × 10³/well) were seeded into 96-well plates and cultured for 24 h before treatment under the designated conditions. CCK-8 reagent (10 μL/well) was added, followed by incubation for 1–4 h at 37 °C. The absorbance at 450 nm was measured using a microplate reader (BioTek, Winooski, VT, USA), with untreated cells serving as controls. The proliferation rates were normalized to baseline values.

2.9. Western Blot

Protein expression was analyzed by Western blot. Cells were lysed in RIPA buffer (Beyotime, Shanghai, China), and protein concentrations were quantified using the BCA assay (Beyotime, China). Equal amounts of protein (20–40 μg) were resolved by SDS-PAGE, transferred onto nitrocellulose membranes (Millipore, Billerica, MA, USA), and blocked with 5% non-fat milk for 1 h at room temperature (RT). The membranes were incubated overnight at 4 °C with the following primary antibodies: anti-DKK1 (1:1000, Abcam, ab307367), anti-β-actin (1:1000, Abcam, ab8226), and anti-Pan Lactic acid-Lysine (1:1000, ABclonal, A23004). After three 10 min PBST washes, membranes were probed with species-matched IRDye^® fluorescent secondary antibodies (1:10,000, LI-COR Biosciences, Lincoln, NE, USA) for 1 h at 25 °C. Following additional PBST washes, immunoreactive bands were visualized using an Odyssey CLx Imaging System (LI-COR), with β-actin serving as the loading control.

2.10. Quantitative Real-Time PCR (qPCR)

Total RNA was extracted using TRIzol reagent (Invitrogen, Carlsbad, CA, USA), and RNA purity and concentration were assessed using a NanoDrop 2000 spectrophotometer (Thermo Fisher, Waltham, MA, USA). cDNA was synthesized with 1 μg of total RNA using the PrimeScript RT Reagent Kit (Takara, Kusatsu, Shiga, Japan). qPCR was performed on an ABI 7500 Fast system (Applied Biosystems, Foster City, CA, USA) using SYBR Green Master Mix (Roche, Basel, Switzerland). The cycling protocol consisted of an initial denaturation at 95 °C for 30 s, followed by 40 cycles of 95 °C for 5 s and 60 °C for 34 s. Relative gene expression levels were calculated using the 2^(-ΔΔCt) method with β-actin as the endogenous control.

2.11. Cell Invasion Assay

Transwell chambers (8 μm pore size; Corning, NY, USA) were coated with 60 μL of Matrigel matrix (BD BiosciencesFranklin Lakes, NJ, USA) diluted 1:8 in serum-free medium and polymerized at 37 °C for 4 hr to establish a basement membrane barrier. Calu-1 and A549 cells were pretreated for 24 h with 25 μM sodium lactate (or DKK1-knockdown/overexpression models). Then, cells were serum-starved for 12 h, harvested, and resuspended in serum-free medium at a density of 2 × 10⁵ cells/mL. 200 μL of cell suspension was seeded in the upper chamber, while the lower chamber contained complete medium supplemented with 10% FBS as a chemoattractant. After 24 h of incubation at 37 °C in 5% CO₂, non-invasion cells on the upper membrane surface were gently removed with a cotton swab. Invasive cells on the lower surface were fixed with 4% paraformaldehyde for 15 min, stained with 0.1% crystal violet for 20 min, and rinsed with phosphate-buffered saline (PBS). Five randomly selected fields per chamber were imaged using an inverted phase-contrast microscope (Nikon Eclipse TS2, Tokyo, Japan), and cell quantification was performed. Invasion capacity was assessed by counting the crystal violet-positive cells in each field.

2.12. Wound-Healing Assay

Cells were seeded in 6-well plates and cultured to 90–100% confluence. To inhibit proliferation, cells were pretreated with mitomycin C (5 μg/mL) for 2 h, washed three times with PBS, and scratched with a sterile 200-μL pipette tip. After removing detached cells by PBS wash, cells were cultured in serum-free medium with the indicated treatments. Images were captured at 0 h and 24 h, and wound closure was quantified using ImageJ (version 1.53t, National Institutes of Health, Bethesda, MD, USA). Experiments were repeated at least three times.

2.13. Lactate Measurement

Lactate concentrations in cell culture supernatants and tissue lysates were quantified using a Lactate Assay Kit (ADS-W-T009-96, Abcam, Cambridge, UK) according to the manufacturer’s instructions. The absorbance was measured at 570 nm, and lactate concentrations were calculated from a standard curve.

2.14. Cell Culture and Treatment

A549 and Calu-1 cell lines (ATCC, Manassas, VA, USA) were cultured in RPMI-1640 medium (Gibco, NY, USA) supplemented with 10% FBS (Gibco, New York, NY, USA) and 1% penicillin-streptomycin at 37 °C in 5% CO₂. For experimental treatments, cells were seeded into culture plates and allowed to adhere for 24 h. Cells were then treated with indicated concentrations (0, 6.25, 12.5, 25, 50, and 100 μM) of compounds for 24 h. All cell culture procedures were performed under strict aseptic conditions.

Cells at 60–70% confluence were transfected with siRNA or expression plasmids using Lipofectamine 3000 transfection reagent (Invitrogen, Carlsbad, CA, USA). The transfection complexes were prepared by incubating siRNA/plasmids with Lipofectamine 3000 in Opti-MEM (Gibco, Shanghai, China) according to the manufacturer’s protocol and followed by 6 h of incubation with cells at 37 °C before replacing with complete growth medium. The sequences of siRNAs for human DKK1 mRNA were used (Hysigen Bioscience, Suzhou, China): siDKK1-1: 5′-GCCGGAUACAGAAAGAUCATT-3′ (sense), siDKK1-2: 5′-UGAUCUUUCUGUAUCCGGCTT -3′ (antisense). Full-length human DKK1 and human androgen receptor plasmids were obtained from GenScript (Nanjing, China). Transfection efficiency was verified 24 h post-transfection by either Western blot or qPCR analysis.

2.15. Cell Culture

A549 and Calu-1 cells were cultured in DMEM (Cat # 11965092, Invitrogen, (Carlsbad, CA, USA) supplemented with 0.1% penicillin-streptomycin (P/S, Cat #G1236, Solarbio, Beijing, China) and 10% FBS (Cat # SV30208.03, HyClone, Logan, UT, USA) in a 37 °C incubator with humidified air and 5% CO₂. The experiments were performed after 2–4 generations of cell transmission.

2.16. Extracellular Acidification Rate (ECAR) Assay

ECAR was measured using a fluorometric ECAR assay kit (Cat # E-BC-F069, Elabscience, China) according to the manufacturer’s instructions. A549 and Calu-1 cells with DKK1 knockdown or overexpression were seeded in black clear-bottom 96-well plates. After treatment, the medium was replaced with kit-provided saline solution and incubated at 37 °C in the dark for 30 min. Probe working solution (saline:probe stock = 95:5, v/v) was added, and fluorescence was monitored at Ex/Em = 490/535 nm every 2–3 min for 100–120 min at 37 °C. ECAR was calculated from the linear phase as (ΔF_sample − ΔF_blank)/ΔT, with blank wells containing saline solution only.

2.17. Statistical Analysis

Data were analyzed using GraphPad Prism 8.0 software (GraphPad Software, San Diego, CA, USA). Data are presented as mean ± SEM. Student’s t-test (two-group comparisons) or one-way ANOVA analysis with Tukey’s post hoc test (multiple groups) was employed. Statistical significance was defined as p < 0.05.

3. Results

3.1. Molecular Subtyping and Functional Characterization of Lactate-Driven Phenotypes in LUAD

3.1.1. Single-Cell Sequencing Analysis of Tumor Microenvironment Heterogeneity and Epithelial Cell Metabolic Reprogramming in Lung Cancer

To investigate the cellular composition and molecular characteristics of the lung cancer microenvironment, we analyzed single-cell RNA sequencing data from the GSE131907 dataset, which included 11 lung cancer samples and 11 normal control samples. After initial quality control, 82,570 cells from 22 samples were retained for subsequent analyses (Figure 1A). To further dissect the cellular landscape of the lung cancer microenvironment, we performed UMAP dimensionality reduction and clustering on these 22 samples, resulting in 24 distinct cell clusters. Subsequent systematic annotation of cell types was conducted by analyzing the expression of specific marker genes in each cluster (Figure 1B and Figure S1), and these clusters were classified into eight major cell subtypes, namely B cells, endothelial cells, epithelial cells, fibroblasts, mast cells, myeloid cells, T and NK cells, and undetermined cells.

We then identified differentially expressed genes (DEGs) among these eight cell subtypes and visualized their expression patterns using a heatmap (Figure 1C). Based on the expression profiles of these DEGs, KEGG functional enrichment analysis was performed for different cell subtypes. Notably, in epithelial cell subtypes, compared with normal tissues, the HIF signaling pathway and central carbon metabolism pathway showed significant functional alterations (Figure 1D). This suggests a marked upregulation of glycolysis in lung epithelial cells, which may lead to substantial lactate accumulation, a phenomenon that is likely closely associated with the initiation and progression of lung cancer.

3.1.2. Molecular Subtyping of Lung Adenocarcinoma Based on Lactate-Related Genes

Univariate Cox regression analysis of 207 lactate metabolism-related genes identified 27 genes significantly associated with prognosis in the LUNG cohort (n = 998; Figure 2A). Kaplan–Meier (KM) survival curves for these 27 candidate lactate-related genes are shown in Figure S2. Consensus clustering determined the optimal cluster number (k) based on the relative change in the area under the cumulative distribution function (CDF) curve, with k = 2, yielding two molecularly distinct subtypes: Subgroup 1 (n = 424) and Subgroup 2 (n = 574) (Figure 2B, C). Striking disparities in lactate-metabolic gene expression profiles were observed between these subtypes (Figure 2D). Kaplan–Meier (KM) survival curves revealed significant prognostic divergence, with Subgroup 1 demonstrating significantly better overall survival compared to Subgroup 2 (Figure 2E). This robust stratification confirms the existence of biologically distinct lactate-driven molecular subtypes in lung adenocarcinoma, providing clinically relevant stratification markers for prognosis prediction.

The TCGA-LUNG cohort dataset was obtained from the UCSC Xena (UCSC Xena) and used as the training set. Following quality control and initial sample filtration, we performed differential gene expression analysis between the two lactate-related subtypes using the limma (V3.62.2) and edgeR (V4.4.2) packages in R (V4.4.2). Applying a threshold of |logFC| ≥ 1.5 with adjusted p < 0.05, we identified 1222 differentially expressed genes (DEGs), including 602 upregulated and 620 downregulated genes. Functional enrichment analysis revealed significant involvement of these DEGs in cancer-related pathways, including p53 signaling pathway, Cell cycle regulation, Chemical carcinogenesis and other related pathways (Figure 3A, B). CIBERSORTx deconvolution analysis demonstrated distinct immune microenvironment profiles between subtypes, with Subgroup 1 exhibiting significantly higher infiltration levels of B cells and CD8+ T cells, whereas Subgroup 2 showed elevated infiltration of M0 macrophages and M1-polarized macrophages (Figure 3C). Based on the 27 lactate-related genes used for molecular subtyping, we calculated a GSVA enrichment/activity score in the T/NK cell compartment and stratified cells into low- and high-lactate accumulation subtypes. The GSVA score boxplot showed that T/NK cell subclusters could be clearly stratified into low- and high-lactate accumulation subtypes, validating the successful classification of these T-cell subtypes (Figure 3D). We identified differentially expressed genes (DEGs) between high- and low-lactate–accumulating T cells and performed KEGG pathway enrichment analysis. The DEGs were significantly enriched in multiple tumor-relevant pathways, predominantly involving metabolic reprogramming and hypoxia adaptation, including Glycolysis/Gluconeogenesis, Carbon metabolism, Biosynthesis of amino acids, and Central carbon metabolism in cancer, as well as the HIF-1 signaling pathway. In addition, pathways related to the tumor immune microenvironment and proteostasis, such as Cytokine–cytokine receptor interaction and Proteasome, were also over-represented, indicating substantial functional differences between the two lactate-associated T-cell subtypes (Figure 3E). Overall, the 27 lactate-metabolism genes defined in this study not only stratify patients into prognostically distinct subtypes but also enable precise single-cell identification of a T-cell subset with a distinct “metabolic–dysfunctional” phenotype. This subset is characterized by hypoxia-related metabolic reprogramming, heightened inflammatory signaling, and disrupted proteostasis and may represent a critical link between the lactate-associated molecular subtype and poor prognosis, potentially contributing to immune evasion.

3.1.3. Development of a Lactate Metabolism-Associated Prognostic Signature Using LASSO Regression

LASSO-Cox regression analysis was performed to identify candidate prognostic genes from the DEGs and construct an optimized prognostic signature. Through this analytical approach, eleven core prognostic genes—Anillin (ANLN), Dickkopf-1 (DKK1), Transcription factor AP-2 alpha (TFAP2A), Fibrinogen alpha chain (FGA), Pleckstrin homology domain containing A6 (PLEKHA6), Kallikrein 8 (KLK8), Cellular repressor of E1A stimulated genes 2 (CREG2), Regulator of G protein signaling 20 (RGS20), Lymphocyte antigen 6 family member K (LY6K), Transmembrane protein 130 (TMEM130), and KIAA1324—were identified as significant prognostic indicators (Figure 4A, B). The identified genes were subsequently integrated into a lactate metabolism-associated prognostic signature, where the LASSO regression coefficients served as optimal weights for gene expression levels in the risk score calculation.

3.1.4. Validation of the Prognostic Signature in the TCGA-LUAD Cohort

The lactate metabolism-derived prognostic model demonstrated robust stratification capability in the TCGA-LUAD cohort. Using median risk scores as the stratification threshold, patients were effectively classified into distinct high-risk (n = 252) and low-risk (n = 252) subgroups. Kaplan–Meier analysis revealed superior overall survival (OS) in the low-risk group compared to high-risk counterparts (Figure 4C). Temporal predictive performance was quantitatively validated through ROC curve analysis, with area under the curve (AUC) values of 0.66, 0.674, and 0.677 for 1-, 2-, and 3-year survival predictions, respectively (Figure 4D). Notably, the developed prognostic model maintained predictive accuracy across all evaluated timepoints, suggesting consistent temporal stability for clinical stratification.

To evaluate the cross-institutional generalizability of our prognostic signature, we performed external independent validation using the GSE50081 cohort (n = 181) from the GEO database. Applying identical risk stratification criteria, patients were effectively categorized into high-risk (n = 91) and low-risk (n = 90) subgroups (Figure 4E). Subsequent survival analysis demonstrated maintained prognostic discrimination, with the low-risk group showing significantly prolonged overall survival. Temporal validation through ROC analysis revealed preserved predictive capacity, achieving an AUC of 0.606 for OS prediction (Figure 4F). This external validation across patient populations and platform-dependent expression profiles demonstrates the signature’s robust clinical applicability.

3.1.5. Individual Prognostic Capacity of Signature Genes

To evaluate the independent prognostic contribution of individual signature genes, we performed univariate Cox regression analysis in the LASSO-derived 11-gene panel in the lung cancer cohort. Ten genes—ANLN, DKK1, TFAP2A, FGA, KLK8, CREG2, RGS20, LY6K, TMEM130, and KIAA1324-demonstrated significant independent prognostic capacity for overall survival (Figure 5). This molecular heterogeneity analysis revealed that 90.9% of signature components showed statistically significant associations with patient outcomes, validating the biological relevance of the selected signature and reinforcing the prognostic robustness of our composite model.

3.1.6. Pathway Activity Profiling Across Prognostic Subgroups

To delineate pathway-level heterogeneity between prognostic subgroups, we systematically conducted pathway enrichment analysis of 83 cancer-related pathways (Table S1). Pathway activity quantification was performed through Gene Set Variation Analysis (GSVA) using the GSVA R package with default parameters, generating enrichment scores that reflect pathway activation states. Comparative analysis between high-risk and low-risk subgroups identified significant dysregulation of oncogenic pathways in high-risk patients, with pronounced hyperactivation of Release of cancer antigens, Pyroptosis and TGF−β signaling pathway, etc., and suppression of Killing of cancer cells, Cancer antigen presentation, B cell recruiting, etc. (Figure 6). These pathway-level aberrations showed a significant correlation with adverse clinical outcomes, providing biological validation for the prognostic stratification achieved in our model.

3.1.7. Distinct Immune Microenvironments Characterize Prognostic Subgroups

Comprehensive immune profiling analysis using CIBERSORTx (https://cibersortx.stanford.edu/) algorithm revealed distinct tumor microenvironment profiles between risk subgroups. The low-risk group demonstrated significantly elevated infiltration of anti-tumor effector lymphocytes, including CD8+ T cells, CD4+ resting memory T cells, naïve B cells, and activated memory B cells, along with increased mast cell recruitment (Figure 7A). Immune checkpoint analysis identified 23 differentially expressed regulators, with inhibitory checkpoints (e.g., CD274 (PD-L1), CD276 (B7-H3), VTCN1) showing marked upregulation in high-risk patients, while stimulatory checkpoints (e.g., CD28, ICOS) were enriched in low-risk counterparts (Figure 7B). ESTIMATE algorithm-based scoring confirmed enhanced immunogenicity in low-risk tumors, demonstrating significantly higher immune scores that correlated with favorable clinical outcomes, suggesting an immunologically active microenvironment may underlie their survival advantage (Figure 7C).

3.1.8. Development and Validation of a Clinically Applicable Prognostic Nomogram

To facilitate clinical translation of the prognostic model, we developed an integrated nomogram that combines the LASSO-derived risk scores with key clinicopathological variables, including patient age, TNM stage, and histological grade (Figure 8A). This visual predictive tool provides individualized prediction of 1-, 3-, and 5-year OS probabilities. The calibration curves showed remarkable concordance between predicted and observed survival outcomes at 1-, 3-, 5-year timepoints, confirming the nomogram’s excellent prediction accuracy (Figure 8B). Meanwhile, patients with a low nomogram score have a significantly better prognosis (Figure 8C). Moreover, the developed nomogram exhibited significantly enhanced predictive accuracy compared to conventional clinical indicators, as evidenced by significantly higher AUC values of 0.682, relative to individual prognostic factors including the riskScore (AUC = 0.662), patient age (AUC = 0.529), gender (AUC = 0.533), nodal status (N, AUC = 0.579), tumor size (T, AUC = 0.613) and stage (AUC = 0.631) (Figure 8D).

3.1.9. Proteomic Profiling of Molecular Subgroups

Proteomic validation was performed with stringent differential expression thresholds (|log₂(fold change)| ≥ 1, p < 0.05), identifying 455 differentially expressed proteins (DEPs), comprising 283 upregulated and 172 downregulated proteins (Figure 9A, B). Unsupervised hierarchical clustering revealed distinct proteomic stratification between subgroups, with clear separation patterns in the heatmap visualization (Figure 9C). Protein–protein interaction (PPI) analysis of DEPs using the STRING database highlighted key interacting proteins, including Multiple endocrine neoplasia type 1 (MEN1), Tissue inhibitor of metalloproteinase 3 (TIMP3), Mixed lineage leukemia translocated to 1 (MLLT1), Apolipoprotein E (APOE), and Neuroblastoma RAS viral oncogene homolog (NRAS), which are implicated in tumor development and suppression as well as metabolic regulation (Figure 9D). The complete interaction networks are provided inTable S2.

Integrated KEGG and Gene Ontology (GO) analyses of the 455 DEPs revealed significant multi-level dysregulation of oncogenic pathways. Pathway enrichment demonstrated significant activation of glycosaminoglycan degradation and p53 signaling, coupled with suppression of apoptosis-related signaling and mitochondrial oxidative phosphorylation (Figure 9E, F). These proteomic signatures collectively highlight the metabolic shift characteristic of tumor progression, strongly corroborating the tumor-promoting metabolic reprogramming predicted by our transcriptomic risk stratification.

3.1.10. Multi-Omics Integration Identifies Core Lactate-Metabolism Regulators

Integrative multi-omics analysis identified three key lactate-metabolism hub genes—ANLN, DKK1, and FGA—demonstrating concurrent transcriptional and translational dysregulation at both the transcriptomic and proteomic levels in RNA-seq and proteomic data (Figure 9G). Violin plots showed significant upregulation in the high-lactate subgroup (Figure 9H–J). The consistent dysregulation of these metabolic effectors across both transcriptomic and proteomic layers identifies them as central coordinators of tumor-associated lactate shuttling, providing a direct mechanistic connection between the prognostic signature and fundamental processes of cancer metabolic reprogramming.

3.2. Dickkopf-1 Drives Lactate-Lactylation Feedback to Promote Metastatic Progression in Lung Cancer via Metabolic Reprogramming

3.2.1. Sodium Lactate-Induced Lactylation Promotes Malignant Progression in Lung Cancer Cells

To elucidate the functional consequences of lactylation in the development and progression of lung cancer, we established an in vitro model using two well-characterized human lung cancer cell lines: A549 (human non-small cell lung cancer cells) and Calu-1 (human lung squamous cell carcinoma cells). Cells were exposed to a gradient concentration of sodium lactate (0–100 μM in PBS) to simulate the tumor microenvironment’s lactate-rich conditions. CCK-8 assays identified 25 μM sodium lactate as the optimal concentration, demonstrating significant proliferative enhancement with 1.66-fold and 1.56-fold increase in A549 and Calu-1 cells, respectively, compared to controls (Figure 10A,B). Furthermore, both wound-healing assays and Transwell invasion assays demonstrated that sodium lactate treatment markedly accelerated wound closure and increased invasive capacity of A549 and Calu-1 cells (Figure 10C,D). Intracellular lactate quantification confirmed effective lactate loading, showing increased lactate levels in A549 and Calu-1 cells after sodium lactate treatment (Figure S3A,B). These results demonstrate that lactate promotes lung cancer progression through lactylation-mediated mechanisms, directly enhancing cellular motility and invasiveness in lung cancer cells.

3.2.2. DKK1 Serves as a Central Lactate-Responsive Regulator of Metastatic Progression in Lung Cancer

Based on bioinformatics analysis and proteomic sequencing, we identified DKK1, ANLN, and FGA as significantly upregulated targets in lactate-treated lung cancer cells. Subsequent qPCR validation in A549 and Calu-1 cell lines demonstrated that 25 μM sodium lactate preferentially increased DKK1 mRNA expression compared to ANLN and FGA (Figure 11A). Western blot analysis further confirmed concordant DKK1 protein elevation following lactate treatment (Figure 11B), establishing DKK1 as a key candidate for mechanistic investigation. To delineate the functional role of DKK1 in lactate-driven malignancy, we generated DKK1-knockdown models in both A549 and Calu-1 cell lines using siRNA-mediated silencing. Target knockdown efficiency was quantitatively validated through the qPCR and Western blot assay, demonstrating a 73% reduction in DKK1 mRNA expression in A549 cells and 48% reduction in Calu-1 cells, with a corresponding decrease of 58% and 49% at protein levels, respectively (Figure S4). Functional characterization revealed that DKK1 knockdown significantly reduced wound closure in the scratch assay and reduced invasion compared to controls (Figure 11C,D).

Conversely, DKK1-overexpressing cells (Figure S5) exhibited accelerated wound closure in the scratch assay and enhanced invasive capacity relative to empty vector controls (Figure 12A,B). These gain-of-function analyses recapitulated the lactate-induced pro-metastatic phenotype, functionally establishing DKK1 as a central effector of lactate-driven metastatic progression.

Collectively, integrative analysis of our genetic perturbation studies demonstrates that DKK1 functions as both a necessary and sufficient effector of lactate-enhanced metastatic potential in lung cancer.

3.2.3. The DKK1–Lactate–Lactylation Axis Drives Lung Cancer Pathogenesis

To delineate the regulatory role of DKK1 in lactate metabolism, we quantified intracellular lactate levels in Calu-1 and A549 cells. Strikingly, DKK1 knockdown reduced intracellular lactate concentrations by 19% and 12% in A549 and Calu-1 cells compared to controls (Figure 13A). Conversely, DKK1 overexpression increased lactate levels 1.22-fold (A549) and 1.11-fold (Calu-1) (Figure 13B), suggesting DKK1 positively regulates lactate biosynthesis or accumulation. To systematically evaluate global lactylation modifications, we performed a Western blot assay using a pan-lactyl lysine antibody on lysates from treated cells. DKK1 silencing displayed a decreased lactylation levels in A549 and Calu-1 cells (Figure 13C), whereas DKK1 overexpression increased lactylation (Figure 13D). To directly connect DKK1 to metabolic output, we measured ECAR as an indicator of extracellular acidification. DKK1 knockdown significantly decreased ECAR in A549 and Calu-1 cells, suggesting reduced glycolysis-associated acid production and a less acidified extracellular milieu (Figure S6A,B). Conversely, DKK1 overexpression significantly increased ECAR in both A549 and Calu-1 cells, indicating enhanced extracellular acidification and glycolysis-associated acid output (Figure S6C,D). This finding aligns with the concomitant reduction in intracellular lactate levels and global lysine lactylation upon DKK1 silencing, supporting a functional role of DKK1 in lactate-centered metabolic reprogramming. These coordinated changes in lactate abundance and lactylation intensity establish DKK1 as a bidirectional regulator of both lactate metabolism and its downstream post-translational modifications. Collectively, these findings establish a novel DKK1–lactate–lactylation signaling axis that molecularly connects glycolytic metabolism to microenvironment-driven tumor progression.

4. Discussion

Lung cancer remains the leading cause of global cancer-related mortality, largely due to late-stage diagnosis, therapeutic resistance, and intratumoral heterogeneity [18]. Lactate, long considered a terminal glycolysis metabolite, is increasingly recognized as an oncometabolite that can shape tumor progression and immune regulation within the tumor microenvironment (TME) [19]. The Warburg-like glycolytic phenotype can contribute to lactate accumulation and extracellular acidification, which have been linked to angiogenesis, immune evasion, and metastatic potential across tumor types [20]. Consistent with this framework, our data support a lactate–DKK1–lactylation working model in which lactate exposure is associated with DKK1 induction and altered lactylation, and DKK1 contributes to lactate-enhanced wound-closure–related motility (under proliferation-inhibited conditions) and invasion in vitro. Our single-cell analysis further delineated cellular heterogeneity in lactate-associated programs and suggested associations between lactate-related gene activity and T-cell functional states at cellular resolution, which is compatible with prior evidence that lactic acid/lactate can suppress cytotoxic lymphocyte proliferation and effector cytokine production [21,22].

Consensus clustering of lactate-associated prognostic genes identified two molecular subtypes (Subtype 1 and Subtype 2) with divergent clinical outcomes and immune profiles. Subtype 1 was associated with improved overall survival and features consistent with an immune-inflamed microenvironment. Conversely, Subtype 2 was associated with worse outcomes and a myeloid-enriched contexture with higher expression of immunoregulatory markers. These immune differences are directionally consistent with experimental literature showing that tumor-derived lactic acid can influence macrophage polarization and pro-angiogenic signaling programs in the TME [23,24].

The LASSO-derived prognostic signature demonstrated cross-platform robustness in TCGA and GEO cohorts and, together with the integrated nomogram, provides a practical tool for retrospective risk stratification. Immune infiltration patterns and checkpoint-related expression differences across subtypes/risk groups generate hypotheses regarding potential immunotherapy responsiveness, but do not constitute direct prediction of benefit and require validation in immunotherapy-treated cohorts. This framing is consistent with prior mechanistic studies showing that lactate signaling can regulate immune checkpoint biology in lung cancer cells, including PD-L1 upregulation through lactate–GPR81 signaling [16]. In addition, prior work indicates that higher tumor glycolysis/lactate biology can be associated with altered responses to checkpoint therapy in certain contexts, supporting the rationale for further validation rather than definitive clinical claims.

Multi-omics integration highlighted ANLN, FGA, and DKK1 as consistently dysregulated at transcript and protein levels. Among these candidates, DKK1 showed lactate-responsive induction under the tested conditions, and genetic perturbation supported that DKK1 contributes to lactate-enhanced migration and invasion in vitro. Prior reports in NSCLC are compatible with a pro-invasive role for DKK1, including links to EMT-like programs and migration/invasion phenotypes [25]. In addition, clinical studies have reported associations between elevated DKK1 (including serum DKK1) and adverse prognosis in NSCLC, supporting its candidacy as a clinically relevant marker while acknowledging context dependence across cancers [26,27]. Mechanistically, our observations that DKK1 perturbation is accompanied by changes in intracellular lactate and global lysine lactylation are consistent with the broader concept that lactate availability can couple to lactylation as a regulatory layer of gene expression and cellular phenotype [28]. However, the molecular intermediates linking DKK1 to lactate production/transport and to lactylation machinery remain to be defined, and targeted metabolic flux and transporter studies will be required to establish the causal steps.

A related open question is how metabolic reprogramming communicates with distant tissue microenvironments during metastasis. Prior work supports that lactate can participate in metabolic symbiosis via monocarboxylate transporters and can contribute to niche biology in vivo, providing a conceptual basis for investigating lactate-linked inter-tissue effects [29,30]. In this context, our lactate–DKK1–lactylation model may represent one candidate mechanism by which lactate-associated programs co-vary with immune and invasive phenotypes; nonetheless, direct evidence in patient-derived and in vivo models will be required to establish whether and how this axis contributes to metastatic niche formation.

Several limitations should be noted. First, the in vitro models employed cannot fully recapitulate the complexity of in vivo tumor microenvironments, where stromal interactions and hypoxia may significantly modulate DKK1-lactate crosstalk. Second, the precise molecular mechanisms linking DKK1 to lactate metabolism remain to be fully characterized. Future studies mapping lactylome changes under DKK1 perturbation could identify specific effector proteins mediating these phenotypes. Third, the clinical relevance of this axis requires further validation using patient-derived samples, with particular focus on correlation with metastatic potential or survival outcomes. Nevertheless, these findings fundamentally advance our understanding of how tumor cells sustain their metabolic reprogramming and establish a framework for targeting DKK1 or modulating the lactate-rich tumor microenvironment as a novel therapeutic strategy to improve clinical outcomes.

5. Conclusions

In summary, our work characterizes lactate-associated molecular heterogeneity in lung cancer and provides a subtype framework and prognostic risk model for retrospective patient stratification. Our multi-omics and in vitro experiments nominate DKK1 as a lactate-responsive candidate linked to migration/invasion phenotypes and lactate/lactylation changes under the tested conditions. Future studies using immunotherapy-treated cohorts, patient-derived samples, and in vivo models will be essential to validate clinical and mechanistic implications.