Multi-omics and Mendelian randomization identify S1PR5 as a causal protective gene and NK cell-mediated prognostic biomarker in lung adenocarcinoma

Abstract

Background

Lung adenocarcinoma (LUAD) sustains an immunosuppressive tumor microenvironment (TME) via stromal-immune interactions. Efferocytosis regulates immune suppression and tissue homeostasis, yet biomarkers stratifying its TME states are lacking in LUAD, hindering precision therapy. This study aimed to investigate efferocytosis-associated immune regulation with both causal and prognostic relevance in LUAD, and to identify key biomarkers with potential implications for tumor stratification and therapeutic guidance.

Methods

Gene expression profiles from The Cancer Genome Atlas (TCGA) LUAD cohort and Genotype-Tissue Expression (GTEx) underwent differential expression analysis. Efferocytosis-related genes (ERGs) from GeneCards were intersected to identify LUAD-associated candidates. Mendelian randomization (MR) and colocalization evaluated causal ERG-LUAD relationships. Risk-related ERGs were systematically analyzed for expression, biological functions, prognosis, and immune interactions. Single-cell RNA sequencing (scRNA-seq) mapped cellular expression specificity of core ERGs. Machine learning prognostic models (S1PR5-enriched NK cell-related prognostic signature, SENRPS) were developed and validated across independent cohorts.

Results

S1PR5 expression was significantly lower in tumor tissues from LUAD patients compared to healthy lung tissue, at both the transcript and protein levels. We identified S1PR5 as a dual biomarker serving both as a protective factor against LUAD pathogenesis and a prognostic marker for survival outcomes, linked to favorable prognosis and enhanced therapy sensitivity. ScRNA-seq localized S1PR5 to natural killer (NK) cells, enhancing the anti-tumor activity of CD16⁺ NK cells and mediating interactions with antigen-presenting CEACAM8⁺ macrophages. The SENRPS model integrates molecular and cellular features for risk stratification and clinical decision-making.

Conclusions

S1PR5 serves as a causal protective gene and prognostic biomarker governing cytotoxic immunity. SENRPS bridges TME dynamics to clinical risk prediction and therapeutic optimization, advancing LUAD precision oncology.

Highlight box.

Key findings

• This study integrated multi-omics, Mendelian randomization, and machine learning to identify S1PR5 as a causal protective factor against lung adenocarcinoma (LUAD). Higher S1PR5 expression correlated with reduced tumor progression, improved therapeutic response, and enhanced natural killer (NK) cell-mediated cytotoxicity and tumor immunosurveillance via single-cell profiling. The SENRPS machine learning model, derived from these insights, enables robust cross-cohort risk stratification and treatment prediction.

What is known and what is new?

• LUAD maintains an immunosuppressive tumor microenvironment through stromal-immune interactions. The role of efferocytosis in shaping immune responses and promoting tumor progression has been observed in multiple cancers, but its specific role in LUAD remains unclear. Identifying efferocytosis-related biomarkers is crucial for stratifying LUAD patients by immune function and therapeutic response.

• This study pioneers an integrative multi-omics framework that integrates transcriptomics, Mendelian randomization, and machine learning to identify S1PR5 as a causal LUAD suppressor associated with cytotoxic NK cell activation and macrophage crosstalk. Additionally, a prognostic model was constructed to enhance patient stratification and outcome prediction.

What is the implication, and what should change now?

• This study elucidates S1PR5’s dual protective genetic and prognostic biomarker in LUAD pathogenesis, establishing its mechanistic link to cytotoxic NK cell activation. The functional status of S1PR5-enriched NK cells in antitumor immunity further underscores the potential of this signature for guiding personalized therapeutic strategies. The SENRPS model enables clinically actionable risk stratification and therapeutic optimization, advancing LUAD precision oncology.

Introduction

Lung cancer remains an immense global health burden, accounting for 12.4% of incident malignancies and 18.7% of cancer-related deaths worldwide (1). Among non-small cell lung cancer (NSCLC) subtypes, lung adenocarcinoma (LUAD) exhibits remarkable tumor microenvironment (TME) complexity, a dynamic ecosystem comprising malignant, stromal, and immune cell populations that collectively influence therapeutic responses through intricate cross-talk (2,3). Such intricate TME complexity underscores the urgent need for precise stratification systems and reliable prognostic models for optimizing clinical management strategies (4,5).

Efferocytosis, the coordinated clearance of dying cells, serves dual roles in LUAD progression. While critical for maintaining tissue homeostasis, this phagocyte-driven process also promotes immunosuppression (6-9). Apoptotic cells release “Find-me” signals like sphingosine-1-phosphate (S1P) and CX3CL1, which guide phagocyte recruitment. Monocytes and macrophages detect these signals through CX3CR1 receptors for CX3CL1 and S1P receptors S1PR1–5 for S1P (10). This coordination ensures efficient apoptotic clearance while preventing excessive immune infiltration. Beyond phagocyte recruitment, “Find-me” signals exhibit broader immune functions. For instance, S1P gradients guide natural killer (NK) cell infiltration into tissues, where these cells trigger apoptosis through antibody-mediated cytotoxicity (ADCC) (11-13). This reveals a sophisticated cellular network where apoptotic signals simultaneously coordinate debris clearance and innate immune responses. However, in LUAD, the functional mediators involved in efferocytosis and the phenotypic heterogeneity of associated immune cell subsets remain insufficiently defined. Understanding these dynamics within the TME could uncover novel prognostic biomarkers for LUAD.

Recent technological advancements now enable a more systematic interrogation of TME complexity. Bulk RNA sequencing (bulk RNA-seq) provides population-level transcriptional profiles, while single-cell RNA sequencing (scRNA-seq) offers high-resolution insights into diverse immune cell states and their dynamic interactions (14). In LUAD, multiple studies have proposed TME-based biomarkers for prognostic stratification (15-18). These include models constructed from global TME features or specific immune cell subsets (19-22), as well as functionally relevant gene signatures, such as those related to metabolism (23), circadian rhythm (24), and programmed cell death (25). Such biomarker analyses have facilitated the classification of LUAD into distinct molecular subtypes, enabling more precise prognosis prediction and treatment selection. However, despite substantial efforts to identify TME-based biomarkers, few have simultaneously established causal relevance to LUAD pathogenesis and exhibited consistent prognostic significance in clinical settings. Mendelian randomization (MR) and summary-data-based Mendelian randomization (SMR) provide powerful frameworks for causal inference by using genetic variants as instrumental variables to assess the unconfounded, directional effects of molecular traits on clinical outcomes (26-28). Integrating MR with transcriptomic and single-cell data transcends traditional association-based methods by enabling the discovery of biomarkers with dual value, encompassing both causal relevance in tumorigenesis and predictive value for clinical outcomes.

In this study, we established an integrative multi-omics paradigm to identify efferocytosis-related prognostic determinants in LUAD. MR and SMR analyses were implemented to validate causal relationships between bulk transcriptomic findings and LUAD pathogenesis, aiding in the identification of functionally relevant genes. This analytical pipeline suggested S1PR5 as a potential genetically modulated protective factor against LUAD progression. Single-cell mapping localized S1PR5 expression to cytotoxic CD16⁺ NK cell subsets, with functional characterization revealing their prognostic significance. A machine learning model incorporating these signatures demonstrated clinical translation potential. This framework connects fundamental insights into efferocytosis-related molecular mechanisms with clinically actionable prognostic tools. Multi-omics analysis systematically characterized the function and prognostic significance of S1PR5, while the functional profiling of S1PR5-enriched NK cell subsets enabled precise stratification of LUAD patients. We present this article in accordance with the REMARK and STROBE-MR reporting checklists (29-31) (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-294/rc).

Methods

Study design and data collection

Transcriptomic and clinical data from 1,539 samples were aggregated for analysis; 524 LUAD samples were obtained from The Cancer Genome Atlas (TCGA) LUAD cohort via UCSC Xena (https://xenabrowser.net/), and 288 normal lung samples were retrieved from Genotype-Tissue Expression (GTEx). External validation utilized three Gene Expression Omnibus (GEO) datasets: GSE13213 (n=117) (32), GSE42127 (n=133) (33), and GSE72094 (n=398) (34). Samples with incomplete clinical data were excluded to ensure cohort comparability. ScRNA-seq data from 11 LUAD and 11 distant normal lung samples were retrieved from GSE131907 (35). To further explore the association between S1PR5 expression and immunotherapy response, we analyzed scRNA-seq data from neoadjuvant immunotherapy-treated NSCLC patients available in dataset GSE207422 (36), along with bulk RNA-seq data from two additional NSCLC immunotherapy cohorts, GSE126044 (n=16) (37) and GSE190265 (n=26) (38). The integrated workflow is summarized in Figure 1A. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Figure 1.

Workflow and identification of differentially expressed efferocytosis-related genes. (A) The study workflow, including data acquisition, integration, and analysis steps. (B) Venn diagram illustrating the overlap between DEGs identified in bulk transcriptomic analysis and ERGs from GeneCards. (C) Volcano plot of 64 differentially expressed ERGs, with the Y-axis showing −log₁₀(adj. P value) and the X-axis showing log₂(fold change). Genes are color-coded based on statistical significance and fold change: pink for upregulated, green for downregulated, and gray for non-significant. DEG, differentially expressed gene; ERG, efferocytosis-related gene; GEO, Gene Expression Omnibus; LUAD, lung adenocarcinoma; NK, natural killer; TCGA, the Cancer Genome Atlas.

Bulk RNA-seq data preprocessing

For all bulk RNA-seq datasets, expression matrices were obtained in either fragments per kilobase of transcript per million mapped reads (FPKM) or transcripts per million (TPM) formats, with FPKM data for TCGA and GTEx, and TPM data for GSE13213, GSE42127, GSE72094, GSE126044, and GSE190265. To reduce batch effects and ensure cross-cohort comparability, all datasets were normalized using the normalizeBetweenArrays function from the “limma” R package (v3.58.1). Expression values were then log₂-transformed using the formula log₂(FPKM/TPM +1). Clinical information was matched to each sample, and only those with complete annotations were retained for downstream analyses.

Identification of differentially expressed efferocytosis-related genes (DEERGs)

ERGs were initially curated from the GeneCards database (https://www.genecards.org/) (Table S1). Differential expression analysis was performed between TCGA-LUAD tumors and GTEx normal tissues using the “limma” R package (v3.58.1). ERGs with an absolute log₂ fold change (FC) >0.32 [log₂(1.25)] and false discovery rate (FDR)-adjusted P<0.05 were classified as DEERGs. Protein expression levels of normal lung tissue and LUAD tissues were compared using immunohistochemistry data retrieved from the Human Protein Atlas (HPA) database (https://www.proteinatlas.org/).

MR and colocalization analyses

Two-sample MR and SMR were restricted to European-ancestry populations to mitigate population stratification. “TwoSampleMR” (v0.6.3) was used to evaluate causal effects of cis-expression quantitative trait loci (cis-eQTLs) on LUAD risk, drawing cis-eQTLs from eQTLGen (https://www.eqtlgen.org/cis-eqtls.html) and GTEx dataset. LUAD genome-wide association study (GWAS) summary statistics (GCST004744, n=66,756) were obtained from the GWAS Catalog(https://www.ebi.ac.uk/gwas/) (39). Genetic instruments were selected based on stringent criteria: genome-wide significance (P<5×10⁻⁶), linkage disequilibrium independence (r²<0.001; 10,000 kb clumping), and instrumental strength [F-statistic (F = β_exposure² / SE_exposure²) >10] (40,41). Allele harmonization was performed for both exposure and outcome SNPs, with cis-eQTL data serving as the exposure and LUAD GWAS data as the outcome. The Wald ratio was used for genes with a single eQTL, while the inverse variance-weighted (IVW) method was applied for multiple instruments, with Cochran’s Q testing heterogeneity (40,41). Benjamini-Hochberg correction defined significance (FDR<0.05) and suggestive associations (nominal P<0.05, FDR≥0.05). Steiger filtering was used for directional validation in cases where insufficient SNPs precluded bidirectional MR (42,43). SMR analysis (http://cnsgenomics.com/software/smr/) integrated LUAD GWAS and cis-eQTLs with HEIDI testing to evaluate heterogeneity (27,44). Bayesian colocalization analysis using the “coloc” R package (v5.2.3) evaluated five causal scenarios (H₀–H₄) through posterior probability (PP) calculations, with PP.H₄ (shared causal variant) ≥0.80 providing strong evidence for colocalization of gene expression and LUAD risk signals (45).

Functional annotation of differentially expressed genes (DEGs)

Functional annotation of DEGs was performed using Gene Ontology (GO) enrichment analysis, which included biological processes (BP), cellular components (CC), and molecular functions (MF) categories, as well as Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis. Both analyses were conducted with the “clusterProfiler” R package (version 4.10.0), applying multiple testing correction using the Benjamini-Hochberg method to adjust P values. Additionally, gene set enrichment analysis (GSEA) was implemented using the “GSEA” R package (version 1.2) to further interrogate pathway-level perturbations.

Genetic alteration profiling

The cBioPortal platform (https://www.cbioportal.org/) was systematically interrogated to retrieve and analyze mutation profiles of candidate proteins in NSCLC, including alteration frequency, mutational spectrum, locus-specific variants, and their clinical correlations.

Immune microenvironment characterization

Immune infiltration in TCGA-LUAD was quantified using a multi-algorithmic deconvolution approach, including XCELL (46), TIMER (47), QUANTISEQ (48), MCPCOUNTER (49), EPIC (50), CIBERSORT-ABS (51), and CIBERSORT (52). Pearson correlations between S1PR5 expression and immune cell scores were computed (P<0.05).

Single-cell RNA sequencing data integration, quality control, and cell type annotation

Single-cell transcriptomic analysis was performed on the GSE131907 dataset using the Seurat package (v5.0.3). Quality control steps were implemented: (I) gene-level filtering retained genes detected in ≥3 cells and expressed in ≥300 cells; (II) cell-level filtering excluded cells with <200 or >7,500 detected genes or >5% mitochondrial gene content; (III) biological filtering removed erythrocyte-associated genes. Counts were normalized per cell, scaled to 10,000 transcripts. Principal component analysis was conducted using variable genes, followed by batch correction via Harmony integration. Dimensionality reduction and clustering utilized the top 17 principal components, and cell type annotation was done manually using marker genes. DEGs of each cluster were identified using Seurat’s FindAllMarkers function with a cutoff of adjusted P value <0.05 and absolute log₂ FC >1.

Due to differences in sequencing protocols across studies, tailored quality control strategies were applied for each dataset. For GSE207422, preprocessing followed the general pipeline with adjusted quality control thresholds. Specifically, cells were excluded if they met any of the following criteria: (I) cells with fewer than 500 detected genes; (II) cells with >20% of UMIs mapping to mitochondrial genes; and (III) cells with >50% of UMIs mapping to ribosomal genes were excluded. Using the top 18 principal components, dimensionality reduction and clustering were performed, followed by manual cell type annotation using marker genes.

NK and myeloid cells were isolated for dimensionality reduction and clustering, following the standard Seurat workflow. Subclassifications within these cell types were identified based on specific marker genes, and the results were visualized using t-SNE plots (53). Macrophage functional scores were quantified using the Seurat AddModuleScore function, integrating gene signatures derived from prior publications (54).

Pseudotime trajectory analysis of NK cells

NK cell trajectories were reconstructed using Monocle (v2.30.1). Genes differentially expressed across pseudotime (q value <1×10⁻²⁰⁰) were identified, with trajectories visualized via DDRTree. Differentiation states and direction of the cells were inferred using CytoTRACE (v0.3.3).

Cell-cell interaction network analysis

Ligand-receptor interactions across cell types were analyzed using the “CellChat” R package (v1.6.1). Interactions between all cell populations were systematically inferred and visualized using CellChat’s built-in functions.

Construction of prognostic signature by integrative machine learning approaches

A survival prediction model for LUAD was constructed using marker genes from CD16⁺S1PR5⁺ NK cells in the scRNA-seq data. Ten machine learning algorithms, along with their unique combinations, were evaluated using the “Mime1” R package (v0.0.0.9000) (55). The algorithms included Random Survival Forest (RSF), Elastic Net (Enet), Stepwise Cox (StepCox), CoxBoost, plsRcox, Supervised Principal Components (superPC), GBM, Survival-SVM, Ridge, and Lasso. Prior to machine learning, univariate Cox regression analysis identified significant prognostic genes (P<0.05).

To avoid overfitting, the TCGA-LUAD dataset was used as the training cohort, while three independent GEO datasets, GSE13213, GSE42127, and GSE72094, served as external validation cohorts. Risk scores were calculated for each sample using candidate models, and the model with the highest concordance index (C-index) across both training and validation sets was selected. In addition, model performance was quantitatively evaluated using receiver operating characteristic (ROC) curve analysis. We calculated both the overall area under the curve (AUC) and time-dependent AUCs at 1-, 3-, and 5-year time points to assess the model’s discriminative ability over time. Patients were stratified into high- and low-risk groups based on median risk scores within each cohort. Prognostic performance was assessed using Kaplan-Meier survival analysis via the rs_sur function in Mime1.

To further validate the robustness and comparative advantage of the model, both overall and time-dependent AUC values were systematically compared with the similar prognostic models in the literature (56).

Survival analysis

Survival analysis for S1PR5 expression was performed using the GEPIA web platform (http://gepia.cancer-pku.cn), with disease-free survival (DFS) defined as the interval from initial treatment to disease recurrence or death from any cause. For NK cell subset marker signatures, we computed enrichment scores through Gene Set Variation Analysis (GSVA) and subsequently evaluated overall survival (OS) using the survival R package (v3.5-8), where OS was defined as the time from diagnosis to mortality from any cause. In both analyses, the median expression value within each cohort served as the stratification threshold for comparative survival outcomes.

Therapeutic response prediction

Immunotherapy response was evaluated using the Tumor Immune Dysfunction and Exclusion (TIDE) platform (http://tide.dfci.harvard.edu/) by analyzing the TCGA-LUAD cohort. To quantify chemotherapy sensitivity, we estimated the half-maximal inhibitory concentration (IC₅₀) values of standard chemotherapeutic agents using transcriptomic data-driven predictions via the “oncoPredict” R package (v0.2), which employs a ridge regression framework trained using the Genomics of Drug Sensitivity in Cancer (GDSC) pharmacogenomic profiles (57). For analytical standardization, the S1PR5 expression stratification threshold was defined by the median expression level within the cohort.

Correlation analysis between S1PR5 and immune checkpoint gene expression

The correlation between S1PR5 expression and representative immune checkpoint molecules was analyzed using the GEPIA platform (http://gepia.cancer-pku.cn). Pearson correlation analysis was conducted to quantify the linear relationships between S1PR5 and selected immune checkpoint genes. Results were presented as scatter plots with corresponding Pearson correlation coefficients (R) and P values indicated. The strength and direction of correlation were interpreted based on the absolute value of R: |R| >0.5 was considered strong, 0.3–0.5 moderate, and <0.3 weak; positive and negative values indicated positive or negative correlation, respectively (58).

Statistical analysis

All statistical analyses and graphical representations were performed using R software version 4.3.2 (http://www.R-project.org). Between-group comparisons were analyzed using Student’s t-test for two-group analyses and Wilcoxon rank-sum test for multi-group comparisons. Statistical significance was defined as P<0.05 across all tests.

Results

Identification of DEERGs in LUAD

Initially, bulk transcriptomic analysis identified 11,067 DEGs between GTEx-derived normal lung tissues and TCGA-LUAD tumor samples. Intersection with 142 ERGs from GeneCards identified 64 DEERGs, including 42 downregulated and 22 upregulated genes in LUAD compared to non-neoplastic tissues (Figure 1B,1C; Figure S1A). KEGG pathway analysis demonstrated DEERG enrichment in TME-linked immune processes, particularly neutrophil, phagosome, and tumorigenesis-related pathways (Figure S1B-S1D). These immune-modulatory signatures position DEERGs as potential TME regulators and prognostic biomarkers in LUAD.

MR prioritizes S1PR5 as a causal protective factor

To elucidate causal relationships between DEERGs and LUAD, two-sample MR analysis was implemented using GWAS data from the LUAD cohort and cis-eQTL data from GTEx whole blood. Initial analysis identified three candidate loci meeting nominal significance (P<0.05): S1PR5 (OR =0.47; 95% CI: 0.32–0.70), CALR (OR =1.62; 95% CI: 1.10–2.41), and SCARB1 (OR =1.15; 95% CI: 1.02–1.30). After Benjamini-Hochberg multiple testing correction, only S1PR5 retained significance (FDR =0.005, Figure 2A). Bayesian colocalization supported shared causal variants between S1PR5 expression and LUAD risk in blood tissue (PP.H4 =83.1%; Figure 2B). Replication using GTEx lung cis-eQTL data corroborated the protective effect (OR =0.72; 95% CI: 0.59–0.88; FDR=0.04), though colocalization support attenuated (PP.H4<80%; Figure S2A,S2B). To confirm the directionality of these causal relationships, we applied Steiger filtering. The results from both GTEx blood cis-eQTL (P=3.57×10⁻⁹) and GTEx lung cis-eQTL (P=1.08×10⁻⁵) data as exposures provided strong evidence against reverse causality (Table S2).

Figure 2.

Mendelian randomization and Bayesian colocalization analyses. (A) Mendelian randomization analysis assessing causal relationships between differentially expressed efferocytosis-related genes and LUAD, using GTEx blood cis-eQTL data as the exposure and LUAD GWAS as the outcome. The ORs with 95% CIs are displayed for each variant. Benjamini-Hochberg correction is utilized. (B) Bayesian colocalization evidence between S1PR5 eQTL in blood and LUAD, with a posterior probability for a shared causal variant (PP.H4) of 83.1%. (C,D) Summary data-based Mendelian randomization analyses integrating GTEx blood cis-eQTL data as the exposure and LUAD GWAS as the outcome. CI, confidence interval; cis-eQTL, cis-expression quantitative trait loci; GTEx, Genotype-Tissue Expression; GWAS, genome-wide association study; LUAD, lung adenocarcinoma; OR, odds ratio; PP, posterior probability.

Consistent results emerged from SMR analyses integrating GTEx and eQTLGen data (Figure 2C,2D; Figure S2C,S2D; Tables S3,S4), reinforcing S1PR5 inversely correlated with LUAD risk.

These consistent findings from multiple MR methodologies confirm S1PR5 as a genetically validated causal protector in LUAD pathogenesis, positioning it as promising target for further mechanistic studies and the development of precision biomarker.

Multi-dimensional clinical analysis identifies S1PR5 as a prognostic biomarker

Integrative analysis at both transcriptomic and proteomic levels demonstrated significant downregulation of S1PR5 expression in LUAD compared to normal tissues (Figure 3A,3B). Furthermore, survival analysis in NSCLC cohorts demonstrated improved DFS correlated with elevated S1PR5 expression (Figure 3C). Additionally, using the median expression value of S1PR5 as the stratification threshold, we categorized TCGA-LUAD patients into distinct subgroups for comparative analysis, with clinical characteristics detailed in Table S5. Increased S1PR5 levels are associated with advanced lymph node metastasis, with N3 stage patients exhibiting decreased expression levels (Figure 3D,3E). Notably, TIDE analysis indicated enhanced response to immunotherapy in the high-expression group (Figure 3F). Drug sensitivity analysis revealed enhanced therapeutic sensitivity in S1PR5-high patients, marked by reduced calculated IC₅₀ values for various drug, such as axitinib, alisertib, dasatinib and AZD (Figure 3G).

Figure 3.

Clinical relevance analysis of S1PR5. (A) mRNA expression levels of S1PR5 in normal lung tissues and LUAD samples from TCGA. (B) Representative immunohistochemical staining images of S1PR5 in normal lung (https://images.proteinatlas.org/29683/61222_A_2_4.jpg) and LUAD (https://images.proteinatlas.org/29683/61276_B_2_1.jpg) tissues obtained from the HPA, shown at 20× magnification. (C) Kaplan-Meier curve for disease-free survival with months on the X-axis. Red represents high S1PR5 TPM, and gray represents low S1PR5 TPM. (D) Heatmap showing the correlation between S1PR5 expression levels and variables such as age, gender, T stage, N stage, M stage, stage, race, and subdivision. (E) Boxplot showing the relationship between S1PR5 expression levels and N stage. (F) Violin plot with TIDE predicted scores on the Y-axis and S1PR5 expression levels on the X-axis. Red represents high expression, and blue represents low expression. (G) Boxplot of oncoPredict-predicted drug sensitivity, with half-maximal inhibitory concentration on the Y-axis and drugs on the X-axis. Red represents high expression, and blue represents low expression. *, P<0.05; ***, P<0.001. HPA, Human Protein Atlas; HR, hazard ratio; LUAD, lung adenocarcinoma; TCGA, The Cancer Genome Atlas; TIDE, Tumor Immune Dysfunction and Exclusion; TPM, transcripts per million.

Further analysis of S1PR5 mutation data in LUAD reveals that, despite genomic alterations occurring in only 1.4% of the population, these mutations exhibit potential associations with clinical prognosis and tumor characteristics, while showing statistically significant positive correlations with tumor burden markers (Figure S3).

Collectively, this multi-tiered validation identifies S1PR5 as a prognostic biomarker in LUAD, reflecting its influence on disease progression and therapy sensitivity.

Functional analysis relates S1PR5 to an immunoregulatory modulator

To delineate the functional impact of S1PR5 in LUAD, we categorized TCGA-LUAD cohort into high- and low-expression subgroups (Figure 4A). Functional enrichment analysis identified upregulated DEGs significantly enriched in extracellular matrix organization, cell cycle regulation, immune responses, and metabolic processes (Figure 4B,4C). This regulatory signature positions S1PR5 as a potential modulator within TME. Multialgorithm deconvolution revealed S1PR5 expression positively correlated with cytotoxic activity and infiltration of CD4⁺ T cells, NK cells, and macrophages, but inversely with tumor-promoting neutrophils (Figure 4D). Further investigations are warranted to elucidate the underlying characters by which S1PR5 modulates the TME and its implications for NSCLC prognosis and therapy.

Figure 4.

Functional enrichment analysis and immune infiltration analysis of S1PR5. (A) Volcano plot showing the DEGs. The X-axis represents the log₂(fold change), and the Y-axis represents the −log₁₀(adjusted P value). Pink, green, and gray points represent significantly upregulated, downregulated, and non-significant genes, respectively. Some specific genes are labeled. (B,C) GO and KEGG functional enrichment analysis of downregulated DEGs (B) and upregulated DEGs (C). The upper part displays GO enrichment analysis. The lower part shows KEGG pathway enrichment analysis. (D) Bubble plot of correlation analysis with immune cells on the X-axis and Pearson correlation coefficients on the Y-axis. Different colors represent different deconvolution approaches. BP, biological processes; CC, cellular components; DEG, differentially expressed gene; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; MF, molecular functions.

ScRNA-seq analysis identifies S1PR5 enriched NK cell subset

The integrated analysis of 11 tumor samples and 11 distant normal lung tissues from the GSE131907 dataset resolved 15 transcriptionally distinct cellular clusters through dimensionality reduction and clustering (Figure 5A). Annotation using lineage-specific marker genes identified eight major cell types: epithelial cells (EPCAM, KRT19, KRT18); endothelial cells (CLDN5, PECAM1, RAMP2, GNG11); fibroblasts (DCN, COL1A2, LUM, MGP); B lymphocytes (CD79A, IGHM, IGHG3, IGLC2, IGKC, IGHA1); T lymphocytes (TRAC, CD3D, CD2, CD3E, IL32); NK cells (GNLY, NKG7, PRF1, KLRD1); myeloid cells (ITGAX, CD14, CD68); and mast cells (TPSAB1, TPSB2, CPA3, MS4A2) (Figure 5B,5C). Comparative cellular quantification revealed significant expansion of B lymphocyte populations in LUAD patients, contrasted by reduced endothelial and NK cell fractions relative to the normal group (Figure 5D, Figure S4A). Notably, S1PR5 exhibited predominant NK cell-restricted expression, with significant downregulation in LUAD specimens, corroborated by bulk RNA-seq analyses (Figure 5E,5F).

Figure 5.

Single-cell analysis of S1PR5 expression and NK cell subtypes. (A) t-SNE visualization of single-cell data, colored according to cell clusters. (B) t-SNE visualization of single-cell data, colored according to cell types. (C) Dot plot illustrating marker gene expression within each cluster. The size of the dots represents the percent expression, and the color intensity indicates the average expression level. (D) Bar plot displaying the relative proportions of different cell types in normal and tumor tissues. The Y-axis represents the relative proportion, and the X-axis indicates the tissue source, with different colors denoting various cell types. (E,F) t-SNE visualization of S1PR5 expression levels. (E) Overall expression across all cells, (F) Expression in normal and tumor tissues. The color intensity indicates the expression level, with red representing higher expression. (G) t-SNE visualization of S1PR5 expression within NK cells. The color intensity denotes the expression level, with red indicating higher expression. (H) Violin plot showing the expression levels of NCAM1 (CD56), S1PR5, and FCGR3A (CD16) in NK cell clusters. The X-axis represents the expression level, and the Y-axis corresponds to the NK cell clusters. (I) t-SNE visualization of NK cell subpopulations, colored according to cell clusters. NK, natural killer; t-SNE, t-distributed stochastic neighbor embedding.

To further investigate the characteristics of NK cell subpopulations, we performed subsequent dimensionality reduction and clustering analysis, which identified eight distinct NK cell clusters. S1PR5 primarily localized to clusters 0, 1, 2, 3, 4, and 7 (Figure 5G). Subcluster annotation based on FCGR3A (CD16), NCAM1 (CD56), and S1PR5 expression delineated three subgroups: CD16⁺S1PR5⁺ NK cells, CD16⁺S1PR5⁻ NK cells, and CD16⁻ NK cells (Figure 5H,5I).

We observed a distinct pattern in NK cell distribution: CD16⁺S1PR5⁺ NK cells were predominantly localized in distal non-malignant lung regions, whereas tissue-resident CD16⁻ NK cells dominated these areas. Conversely, tumor tissues showed a significant decrease in CD16⁺S1PR5⁺ NK cells alongside expansion of both CD16⁺S1PR5⁻ and CD16⁻ subsets (Figure S4B). This divergence implies that S1PR5 governs NK cell migration patterns, and hints that tumor-distal normal pulmonary may systemically reprogram NK cell functions through tumor-educated signaling cascades.

Pseudotime trajectory analysis identifies S1PR5-enriched NK cells as terminally differentiated subsets

To investigate the functional impact of S1PR5 expression on NK cells, we employed cytotrace-based pseudotime analysis to delineate the differentiation dynamics of NK cell subpopulations in relation to S1PR5 expression (Figure S4). CD16⁻ NK cells demonstrated the highest differentiation plasticity, occupying an early immature state within the developmental hierarchy, compared to CD16⁺S1PR5⁺ and CD16⁺S1PR5⁻ subsets (Figure S4C,S4D). Monocle2 trajectory analysis corroborated CD16⁻ NK cells as the differentiation origin, bifurcating into CD16⁺S1PR5⁺ and CD16⁺S1PR5⁻ subpopulations (Figure 6A,6B). S1PR5 expression scaled progressively with differentiation advancement, peaking in terminally differentiated CD16⁺S1PR5⁺ cells (Figure 6C). Concomitantly, the proportions of CD16⁻ NK cell declined during differentiation, while CD16⁺S1PR5⁺ and CD16⁺S1PR5⁻ subpopulation increased (Figure 6D). These data position both CD16⁺S1PR5⁺ and CD16⁺S1PR5⁻ NK cells as mature effector subsets.

Figure 6.

Pseudotime analysis and functional enrichment of NK cell subsets. (A,B) Pseudotime analysis of NK cell subtypes using Monocle. (A) Visualization colored by NK cell subtypes. (B) Visualization colored by pseudotime. (C) S1PR5 expression along pseudotime. The X-axis represents pseudotime, and the Y-axis represents relative expression levels. (D) Cell density changes along pseudotime, colored by NK cell subtypes. (E,F) Bubble plots of GO enrichment analysis for upregulated (E) and downregulated (F) genes between NK cell subtypes. The X-axis represents the number of genes in the pathway, and the Y-axis represents significant enrichment pathways. Bubble size indicates the count of genes, and bubble color represents the −log₁₀(P.adjust) value. (G) KEGG enrichment analysis for upregulated and downregulated genes between NK cell subtypes. Red indicates upregulated genes, and blue indicates downregulated genes. The X-axis represents −log₁₀(P.adjust), and the Y-axis represents significant enrichment pathways. Bubble size indicates the count of genes, and bubble color represents the −log₁₀(P.adjust) value. (H,I) GSEA enrichment analysis for (H) CD16⁺S1PR5⁺ NK cells, and (I) CD16⁺S1PR5⁻ NK cells. Ridge plots show significant enrichment pathways on the Y-axis and log₂(fold change) on the X-axis, colored by −log₁₀(P.adjust). Dot size represents the absolute value of NES, and dot color represents its value. BP, biological processes; CC, cellular components; GO, Gene Ontology; GSEA, gene set enrichment analysis; KEGG, Kyoto Encyclopedia of Genes and Genomes; MF, molecular functions; NES, Normalized Enrichment Score; NK, natural killer.

Functional polarization of S1PR5-enriched NK subsets links cytotoxic activation

The CD16⁺S1PR5⁺ subset demonstrated markedly elevated expression of cytotoxic effector molecules, including Granzyme B (GZMB), Granzyme H (GZMH), Perforin 1 (PRF1), and FasL (FASLG), along with increased expression of activating receptors such as FCGR3A (CD16), NCR1 (NKp46), and NCR3 (NKp30), all essential for antibody-dependent cellular cytotoxicity. Conversely, CD16⁺S1PR5⁺ NK cells exhibited significantly lower expression of inhibitory immune checkpoints such as KLRC1 (NKG2A), TIGIT (TIGIT), PDCD1 (PD-1), and CTLA4 (CTLA4) compared to S1PR5⁻ subsets. Additionally, markers of NK cell maturity, including ADGRG1, NKG7 and CX3CR1 were also significantly upregulated in this subgroup, indicating their terminally differentiated and functionally active phenotype (Figure S5). Moreover, S1PR5 expression was negatively correlated with immune checkpoints including PDCD1 (PD-1), CTLA4 (CTLA4), HAVCR2 (TIM-3), and TIGIT (TIGIT), suggesting that high S1PR5 expression may reflect a less exhausted, more cytotoxic NK cell state (Figure S6). Together, these findings point to a mechanistic role for S1PR5 in shaping NK cell cytotoxic programming.

To further elucidate the molecular and functional characteristics of these NK cell subpopulations, we performed differential gene expression analysis and subsequent functional enrichment analysis. Upregulated genes in CD16⁺S1PR5⁺ NK cells were functionally characterized through GO analysis (Figure 6E). BP annotations highlighted extensive crosstalk between myeloid and lymphoid cells, enhanced cytotoxicity, and increased NK cell activation, positioning these cells as key mediators of immune surveillance. CC analysis demonstrated enrichment in intercellular communication machinery, while MF terms emphasized actin-binding and phosphatase activity, indicating that CD16⁺S1PR5⁺ NK cells are functionally primed for immune interactions, cell adhesion, and signal transduction.

In CD16⁺S1PR5⁺ NK cells, downregulated genes exhibited inverse functional profiles (Figure 6F). BP categories showed significant suppression of immunoregulatory pathways, whereas CC terms indicated decreased membrane remodeling and vesicular transport capacity compared to CD16⁺S1PR5⁻ cells. MF profiling revealed deficiencies in epigenetic regulatory complexes and protein interactome networks, suggesting impaired modulatory plasticity during immune adaptation.

KEGG pathway analysis corroborated these functional disparities (Figure 6G). GSEA quantified this divergence, showing positive enrichment of cytotoxic programs in CD16⁺S1PR5⁺ NK cells and preferential enrichment of immune dysfunction pathways in CD16⁺S1PR5⁻ NK cells (Figure 6H,6I). Critically, cytotoxic effector functions within the CD16⁺ compartment were predominantly mediated by S1PR5⁺ rather than S1PR5⁻ NK subsets.

Clinically, survival analyses demonstrated that the CD16⁺S1PR5⁺-derived gene signature was independently associated with prolonged OS in both the TCGA-LUAD and GSE13213 cohorts (Figure S4E,S4F), highlighting its prognostic relevance. To further assess its predictive value in immunotherapy, we analyzed scRNA-seq data GSE207422 from NSCLC patients receiving neoadjuvant treatment. CD16⁺S1PR5⁺ NK cells were identified and enriched in responders, with significantly higher proportions observed in patients achieving partial response (PR), and elevated S1PR5 expression in those with pathological complete response (pCR) (Figure S7A-S7G). Consistently, across multiple independent NSCLC immunotherapy cohorts (GSE126044, GSE190265), high S1PR5 expression tended to be associated with improved treatment response and prolonged survival (Figure S7H-S7J). These findings nominate S1PR5 as a potential prognostic biomarker and a lever to augment NK cell antitumor immunity.

Myeloid cell subclustering and cellular interactions

Functional enrichment analysis of NK subpopulations revealed potential crosstalk with myeloid lineages. To dissect these interactions, we performed dimensionality reduction and subclustering of myeloid cells, identifying 11 transcriptionally distinct subsets: Monocytes (VCAN, FCN1, S100A12, CD14, FCGR3A), FABP4⁺ Macrophages (FABP4, C1QA, C1QB, C1QC, CD68), PLTP⁺ Macrophages (PLTP, C1QA, C1QB, C1QC, CD68), SPP1⁺ Macrophages (SPP1, C1QA, C1QB, C1QC, CD68), CEACAM8⁺ Macrophages (CEACAM8, C1QA, C1QB, C1QC, CD68), BRSK1⁺ Macrophages (BRSK1, C1QA, C1QB, C1QC, CD68), TRAC⁺ Macrophages (TRAC, C1QA, C1QB, C1QC, CD68), plasmacytoid dendritic cells (DCs) (LILRA4, GZMB, CLEC4C, IRF7), DC1s (CLEC9A, XCR1, CADM1, BATF3), DC2s (CD1C, FCER1A, CD1E, ITGAX), and DC3s (LAMP3, CCR7, CCL19, CD83) (Figure 7A, Figure S8A). Further functional profiling demonstrated that SPP1⁺ and PLTP⁺ macrophages exhibited the highest M2 polarization scores, consistent with an immunosuppressive phenotype. Functional scoring revealed that SPP1⁺ macrophages harbored elevated phagocytic activity, while CEACAM8⁺ macrophages showed superior antigen processing and presentation signatures, indicative of their role in adaptive immune priming. Angiogenic potential was notably increased in SPP1⁺ macrophages compared to other subsets (Figure 7B,7C).

Figure 7.

Single-cell analysis of myeloid cell subpopulations and cell communication networks. (A) t-SNE visualization of single-cell data within myeloid cells, colored according to cell clusters. (B,C) Violin plots showing the functional scores of macrophage subpopulations. (B) M1 and M2 polarization scores, (C) scores for Phagocytic, Antigen Processing and Presentation, and Angiogenesis. The X-axis represents macrophage subpopulations, and the Y-axis represents the scores. (D) Circular plot showing cell communication targeting CD16⁺S1PR5⁺ NK cells in the tumor microenvironment. (E) Dot plot showing ligand-receptor interaction network analysis targeting CD16⁺S1PR5⁺ NK cells. Bubble color and size represent the calculated communication probability and P value, respectively. Blank spaces indicate zero communication ability. (F) Circular plot showing cell communication targeting CD16⁺S1PR5⁻ NK cells in the tumor microenvironment. (G) Dot plot showing ligand-receptor interaction network analysis targeting CD16⁺S1PR5⁻ NK cells. Bubble color and size represent the calculated communication probability and P value, respectively. Blank spaces indicate zero communication ability. (H) Chord diagram showing SPP1 signaling pathway networks in the tumor. (I) Heatmap showing the roles of different cell types in SPP1 signaling pathway networks in the tumor. (J) Chord diagram showing IFN-II signaling pathway networks in the tumor. (K) Heatmap showing the roles of different cell types in IFN-II signaling pathway networks in the tumor. DC, dendritic cell; NK, natural killer; t-SNE, t-distributed stochastic neighbor embedding.

CD16⁺S1PR5⁺ and CD16⁺S1PR5⁻ NK subsets exhibited divergent interaction patterns with myeloid cells. CD16⁺S1PR5⁻ NK cells engaged in more extensive myeloid interactions within tumors compared to normal tissues, enriched for pathways such as SPP1 signaling (Figure S8B,S8C). CD16⁺S1PR5⁺ NK cells preferentially communicated with CEACAM8⁺ macrophages, a subset associated with antigen presentation, with enhanced interaction strength within the TME (Figure 7D,7E), primarily mediated through the RETN-CAP1 ligand-receptor axis. In contrast, CD16⁺S1PR5⁻ NK cells exhibited stronger interactions with SPP1⁺ macrophages, which displayed M2-like polarization and pro-angiogenic signatures (Figure 7F,7G), predominantly through the SPP1-CD44 signaling axis. Mechanistically, activation of the SPP1 pathway was largely driven by SPP1⁺ macrophages, with CD16⁺S1PR5⁻ NK cells receiving significantly stronger SPP1 signaling inputs than their S1PR5⁺ counterparts in tumor tissues (Figure 7H,7I). Additionally, CD16⁺S1PR5⁺ NK cells, along with CD16⁻ NK cells, primarily contributed to IFN-γ (type II interferon) signaling, thereby promoting the activation of diverse myeloid subsets within the TME (Figure 7J,7K). These findings indicate that S1PR5 reshapes NK cell-macrophage communication by redirecting NK cell interactions away from immunosuppressive SPP1⁺ macrophages toward CEACAM8⁺ macrophages, which possess potential antigen-presenting and immune-activating capacities. This transition is likely mediated via the RETN-CAP1 signaling axis, which may promote NK cell-driven activation or licensing of CEACAM8⁺ macrophages, thereby amplifying antitumor immune responses within the TME.

Intriguingly, CD16⁺S1PR5⁺ NK cells specifically engaged with endothelial cells in normal lung tissues through CX3CL1-CX3CR1 and MIF-(CD74 + CXCR4) ligand-receptor pairs, a pattern undetected in both CD16⁺S1PR5⁻ NK cells and TME (Figure 7E,7G). This interaction suggests pulmonary endothelia may chemoattract CD16⁺S1PR5⁺ NK cell infiltration. The chemotaxis-associated genes CCRL2, CX3CL1, and SPHK1 exhibited endothelial-predominant expression with significantly higher levels in normal versus tumor tissues (Figure S9). These cellular interaction differences potentially explained the distinct NK cell infiltration signatures distinguishing distal lung parenchyma from TME.

Integrative machine learning establishes an S1PR5-enriched NK cell-related prognostic signature (SENRPS)

To establish a robust SENRPS for LUAD, we integrated ten machine learning algorithms and combinatorial models to analyze 92 prognosis-associated genes identified through univariate Cox regression in the TCGA-LUAD training cohort. External validation was conducted in three independent cohorts (GSE13213, GSE42127, GSE72094), with model performance evaluated by the C-index (Figure 8A). The prognostic performance of the StepCox-Ridge regression model was systematically evaluated using ROC curve analysis. In the TCGA-LUAD discovery cohort, the model yielded an overall AUC of 0.70, with time-dependent AUCs of 0.73, 0.73, and 0.69 at 1-, 3-, and 5-year intervals, respectively. Its predictive accuracy was consistently reproduced across three independent external validation cohorts: GSE13213 (1-/3-/5-year AUCs: 0.87/0.71/0.70), GSE42127 (1-/3-/5-year AUCs: 0.79/0.69/0.63), and GSE72094 (1-/3-/5-year AUCs: 0.65/0.66/0.74). In addition, the StepCox-Ridge regression model yielded the highest C-index among all models evaluated in this study, with a mean C-index of 0.66 across all cohorts and 0.647 in validation datasets (Figure 8B). Risk stratification based on the SENRPS score effectively distinguished patient subgroups with significantly different OS across all cohorts (Figure 8C). To further substantiate its clinical utility, the predictive performance of SENRPS was benchmarked against a previously published efferocytosis-related prognostic model (56). SENRPS consistently achieved higher overall and time-specific AUCs across all datasets, underscoring its superior prognostic accuracy and enhanced applicability in risk stratification and individualized survival prediction (Figure S10).

Figure 8.

Establishment and validation of the prognostic model SENRPS via machine learning. (A) Ten machine learning algorithms and their unique combinations were evaluated, with the C-index calculated for each model across all validation datasets. (B) The 1-, 3-, and 5-year AUC values of the stepCox[forward] + Ridge model in all datasets. (C) Overall survival analysis of all LUAD cohorts with high and low SENRPS. Red represents high SENRPS, and gray represents low SENRPS. (D) Heatmap showing the correlation between SENRPS levels and variables such as age, gender, T stage, N stage, M stage, stage, race, and subdivision. (E) Violin plot with TIDE predicted scores on the Y-axis and SENRPS levels on the X-axis. Red represents high risk, and blue represents low risk. (F) Boxplot of oncoPredict-predicted drug sensitivity, with half-maximal inhibitory concentration on the Y-axis and drugs on the X-axis. *, P<0.05; **, P<0.01; ***, P<0.001. AUC, area under the curve; CI, confidence interval; LUAD, lung adenocarcinoma; SENRPS, S1PR5-enriched NK cell-related prognostic signature; TIDE, Tumor Immune Dysfunction and Exclusion.

Clinically, both univariate and multivariate Cox regression analyses identified the SENRPS model as an independent prognostic factor (Table S6). Additionally, high-risk stratification in TCGA-LUAD correlated with advanced T/N -stage, and clinical stage, while also predicting reduced sensitivity to immunotherapy and elevated resistance to chemotherapy and targeted therapies (Figure 8D-8F, Figure S11). These findings position SENRPS as a dual-purpose biomarker for LUAD prognosis and treatment stratification, distinguishing therapy-resistant, immune-evasive high-risk tumors from therapy-responsive, immune-engaged low-risk subgroups. The functional status of S1PR5-enriched NK cells in antitumor immunity further underscores the potential of this signature for guiding personalized therapeutic strategies.

Discussion

Lung cancer remains the leading cause of cancer mortality worldwide, with therapeutic challenges arising from both tumor heterogeneity and microenvironment-mediated resistance (3,59). Despite advances in molecular targeted therapies, immunotherapy, and precision radiotherapy (60,61), persistent biological heterogeneity continues to undermine treatment predictability. Efferocytosis, an evolutionarily conserved apoptotic clearance mechanism, represents a paradoxical pathway in oncogenic progression. Although its immunosuppressive functions via M2-like macrophage polarization and regulatory T cell expansion are mechanistically mapped in colorectal, pancreatic cancer, and mammary carcinomas (8,62,63), the spatial organization and functional pleiotropy of this process and regulation in LUAD remain inadequately characterized. This gap underscores the necessity to characterize efferocytosis-related signatures in LUAD, which may simultaneously decode mechanistic cross-talk between apoptotic clearance pathways and TME immunosuppression, and enable clinical risk stratification for precision therapy.

We conducted an integrative multi-omics framework combining bulk transcriptomics, MR, single-cell multi-omics, and machine learning. This framework aims to decode the efferocytosis-related molecular landscape in LUAD by identifying ERGs that function dually as genetic safeguards against LUAD pathogenesis and robust prognostic biomarkers for risk stratification.

A total of 64 DEERGs were identified through intersection of bulk transcriptomic datasets. These candidate genes were subsequently interrogated via a two-stage analytical framework combining two sample MR and SMR. MR analysis exploits germline genetic variants as IVs, providing a robust epidemiological framework for causal inference when randomized controlled trials are ethically or practically infeasible. This approach inherently minimizes residual confounding, reverse causation, and measurement bias through three core instrumental variable assumptions: relevance, independence, and exclusion restriction. Specifically, we utilized cis-eQTLs as molecular instruments given their high heritability and tissue-specific stability. Two-sample MR identified S1PR5 as a protective locus against LUAD development. Crucially, SMR demonstrated methodological concordance with MR findings. Bayesian colocalization analysis then found a novel genome-wide significant risk locus at the rs9749078, establishing this variant as a pleiotropic modulator of both efferocytosis-related molecular signatures and malignant progression through transcript-level regulation.

Beyond genetic predisposition, the clinical significance of S1PR5 was corroborated through multidimensional survival analysis. Elevated S1PR5 expression correlated with prolonged DFS in NSCLC cohorts. Drug sensitivity profiling revealed inverse correlations between S1PR5 levels, with enhanced sensitivity to multiple drugs in S1PR5-high patients. Notably, TIDE analysis demonstrated S1PR5-high cases exhibited enhanced likelihood of benefiting from immune checkpoint blockade. Together, these findings establish S1PR5 as both an independent prognostic factor and predictive biomarker guiding therapeutic selection in LUAD.

Collectively, S1PR5 has been identified as a molecular determinant in the pathogenesis of LUAD, with elevated expression associated with favorable prognosis. As a G protein-coupled receptor within the sphingosine-1-phosphate receptor family, S1PR5 primarily couples with Gi/o proteins, regulating downstream signaling pathways involving ERK, Rac, adenylyl cyclase (AC), and phospholipase C (PLC) (64,65). Through its role in lymphocyte homing and migration, S1PR5 is integral to immune regulation, particularly in multiple sclerosis (MS) and experimental autoimmune encephalomyelitis (EAE). In oncology, S1PR5 demonstrates dual effects. Increased expression correlates with poor prognosis in colorectal cancer, suggesting a tumor-promoting role. Conversely, S1PR5 activation suppresses prostate cancer cell proliferation and migration, indicating tumor-suppressive properties in certain malignancies (66,67). The therapeutic relevance of S1PR5 in immune-mediated diseases is exemplified by FDA-approved agents such as siponimod and ozanimod, which selectively target this receptor. These drugs reduce circulating lymphocyte counts and attenuate neuroinflammation in MS through specific polar and hydrophobic interactions with S1PR5 (68). While S1PR5-targeting agents are clinically approved for autoimmune disorders, their antitumour potential remains to be validated in future clinical studies.

Functional enrichment analysis and integration of multi-algorithm deconvolution further consistently linked S1PR5 expression to immune-activating TME. Subsequent single-cell transcriptomic profiling identified S1PR5-enriched terminally differentiated CD16⁺S1PR5⁺ NK clusters, which exhibited upregulated cytotoxic effector programs and correlated with extended survival in LUAD cohort. These subpopulations were positioned as pivotal executors of antitumor immunity. Conversely, S1PR5-deficient CD16⁺ NK subpopulations displayed preferential interactions with tumor-associated macrophage subsets, particularly pro-angiogenic SPP1⁺ macrophages, which represented M2-polarized subsets. This intercellular network mapped to immunosuppressive niche formation, functionally characterizing CD16⁺S1PR5⁻ NK cells as contributors to immune-evasive niche formation. Furthermore, in distant lung tissue, CD16⁺S1PR5⁺ NK recruitment is orchestrated by CX3CL1-CX3CR1 and MIF-(CD74 + CXCR4) pathways, with endothelial-specific overexpression of CCRL2, CX3CL1, and SPHK1. These chemotactic axes account for their scarcity in TMEs and highlighting specific chemotactic adaptations in distant lung tissue.

Under physiological conditions, NK cells stratify into two functionally distinct subsets. The CD56^brightCD16^dim population resides in barrier tissues as tissue-resident NK cells (trNK), maintaining immune homeostasis through cytokine production including IL-12, IFN-γ, and CCL5 that recruits adaptive immune cells. Conversely, the circulating CD56^dimCD16^bright subset constitutes approximately 95% of peripheral blood NK cells and exhibits potent cytotoxicity via perforin and granzyme secretion, with high expression of S1PR5. S1PR5 orchestrates NK cell functionality through two mechanisms (69). By regulating cytotoxic capacity, S1PR5 deficiency directly impairs degranulation and cytotoxic molecule secretion, critically compromising tumoricidal activity (13,70,71). Simultaneously, S1PR5 maintains migratory equilibrium through sphingosine-1-phosphate (S1P) gradient sensing, establishing a dynamic “recruitment-homing” balance. In LUAD pathogenesis, this equilibrium is challenged as pulmonary endothelia recruit circulating NK cells via CCRL2-Chemerin signaling while S1PR5 enriched populations counterbalance through S1P-mediated tissue egress (72). Perturbations in this system produce distinct pathological outcomes: partial S1PR5 downregulation sequesters NK cells in peripheral tissues, whereas complete deficiency traps them in BM/LN reservoirs, depleting circulating pools and accelerating tumor metastatic spread. Mechanistic studies reveal S1PR5 loss concomitantly downregulates homing receptors (CXCR5/CXCR6/CXCL13), creating dual defects in both cytotoxic potential and tumor-directed migration capacity (11). Moreover, primary head and neck squamous carcinoma, gastric cancer, and esophageal cancer exhibit progressive reductions in tumor-infiltrating CD56^dimCD16^bright NK cell proportions during disease progression. In NSCLC, it is further corroborated compromised anti-tumor immunity, characterized by diminished NK cell infiltration, reduced CD16⁺ NK subsets, and upregulated inhibitory receptors (12,73).

Notably, CD16⁺S1PR5⁺ NK cell subsets were identified in both tumor cores and distant lung tissues, indicating tumor-driven phenotypic adaptation. These cells coexisted intratumorally with CD16⁻ NK subsets and may differentiate into CD16⁺S1PR5⁻ immune-evasive phenotypes. Their enrichment in distant tissues implies a tumor-induced shift from tissue-resident to circulating NK cell states via microenvironmental reprogramming.

Clinically, S1PR5 holds dual translational potential. On one hand, S1PR5 agonists may promote bone marrow and lymph node egress, enhancing systemic immunosurveillance. On the other hand, CD16⁺S1PR5⁺-based prognostic models enable refined risk stratification in LUAD. In this context, we developed a machine learning-based prognostic signature (SENRPS) incorporating CD16⁺S1PR5⁺-associated markers, which effectively predicts patient prognosis. Compared with previously published prognostic models based on efferocytosis-related gene signatures, SENRPS demonstrated a higher C-index and consistently yielded superior overall and time-dependent area under the ROC curves (AUCs) across multiple independent cohorts (56). Moreover, unlike prior models, our approach not only provides prognostic value but also highlights S1PR5 as a biologically meaningful therapeutic target. As a membrane receptor, S1PR5 is readily detectable in tumor tissues via RNA-seq or in situ hybridization, and its expression can be dynamically monitored in peripheral blood using flow cytometry or single-cell transcriptomics. These features support its utility as both a prognostic marker and an immune monitoring tool. Furthermore, S1PR5 activation may enhance NK cell maturation and cytotoxic potential, offering a mechanistic avenue to augment adoptive cell therapies such as CAR-NK. Together, these findings position S1PR5 as a biomarker of diagnostic, dynamic, and interventional relevance in LUAD.

Conclusions

In summary, this study systematically delineates the functional landscape of efferocytosis-related genes in LUAD through systematic multi-omics integration, identifying S1PR5 as a key regulator of cytotoxic immunity, serving as both a protective genetic determinant against LUAD pathogenesis and an independent prognostic biomarker. Single-cell resolution revealed that S1PR5 governs NK cell subset functions. The SENRPS signature model integrates molecular and cellular features for risk stratification and clinical decision-making.

Supplementary

The article’s supplementary files as

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. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.