Decoding chromatin regulator-LAIR1+ M2 like macrophage patterns in atherosclerosis and clinical outcomes of Lung adenocarcinoma patients: evidence from artificial intelligence-driven multi omics and in vitro validation Authors: Chuanqi Zhang, Bo Chen, Zhe Li, Xianwei Li, Weiliang Jiang Published: 2026 ABSTRACT Atherosclerosis (AS) is a common complication of lung adenocarcinoma (LUAD), but its underlying mechanisms in LUAD remain unclear. This study aimed to decipher the role of chromatin regulators in AS pathogenesis and their association with clinical outcomes in LUAD patients. AS- and chromatin regulator (CR)-associated prognostic indicators were identified in LUAD and analyzed in the TCGA-LUAD cohort using Cox regression and Disease Ontology analysis. CR-related risk subgroups were defined by NMF in the GSE26939 cohort. GSVA and WGCNA, combined with interpretable machine learning, were used to construct a predictive model and identify key genes, which were validated in GSE68465. The molecular features of key genes and their roles in AS were further evaluated. Besides, mechanisms of key genes in M2-like macrophages were assessed at the single-cell level in LUAD patients using cutting-edge analytical frameworks. A deep learning framework and molecular docking were used to screen natural compounds. Finally, co-culture experiments were conducted for validation. CR signatures can guide the LUAD patients LUAD patients into high- and low-AS risk groups and were associated with clinical outcomes. LAIR1 can be considered as an AS-related factor enriched in M2-like macrophages and involved in LUAD progression. Quercetin was identified as potential preventive agent for AS in LUAD patients. CR-associated signatures play an important role in AS pathogenesis and LUAD progression.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1007/s10238-026-02217-0. Introduction Lung adenocarcinoma (LUAD), the most common histological subtype of non-small cell lung cancer, is a leading cause of cancer-related mortality worldwide [1]. Its pathogenesis involves the accumulation of genetic and epigenetic alterations in lung epithelial cells, driven by factors such as tobacco smoke and environmental carcinogenesis [2]. Besides, epidemiologically, LUAD incidence remains high, and patients often present with or develop significant comorbidities. Among these, atherosclerosis (AS), a chronic inflammatory disease of the arterial wall characterized by plaque formation, is a major contributor to cardiovascular morbidity and mortality in cancer patients [3, 4]. A growing body of evidence suggests a bidirectional relationship, for example, systemic inflammation and metabolic dysregulation in cancer can accelerate AS, while AS-related chronic inflammation may foster a pro-tumorigenic microenvironment, though the precise molecular links are not fully elucidated. Chromatin regulators, including writers, erasers, and readers of histone modifications, are fundamental arbiters of cellular identity and function by controlling gene accessibility [5]. Dysregulation of chromatin remodeling is a hallmark of cancer, contributing to oncogene activation, tumor suppressor silencing, and therapy resistance in LUAD [6]. Concurrently, epigenetic mechanisms are increasingly recognized as critical players in AS, modulating endothelial dysfunction, vascular smooth muscle cell proliferation, and macrophage-driven inflammation within plaques [7]. This convergence suggests that aberrant chromatin regulator activity may represent a shared pathogenic node, potentially orchestrating immune cell behavior, particularly the polarization of tumor-promoting M2-like macrophages in both diseases [8]. However, a systematic investigation into a unified chromatin regulator signature that connects AS pathology to LUAD clinical outcomes is lacking. To address this gap, we conducted an integrative study combining artificial intelligence(AI)-driven multi-omics analysis with experimental validation. We aimed to identify a chromatin regulator signature associated with AS pathogenesis and LUAD prognosis. Besides, we constructed and validated a robust prognostic model based on this signature. In addition, we dissected the cellular and functional role of the key driver within the tumor microenvironment(TME) using multi-omic genomics with exploration of potential therapeutic agents targeting this axis. This work provides a novel framework for understanding the AS-LUAD interplay and identifies actionable targets for risk stratification and therapy. Materials and methods Source of bulk profile Transcriptomic and clinical data for LUAD were retrieved from The Cancer Genome Atlas (TCGA-LUAD, n = 593) and the GEO database: GSE26939 (LUAD patients = 116, platform GPL9053), GSE68465 (LUAD patients = 462, platform GPL96) [9, 10]. AS-related datasets GSE23746(Control = 19 and AS = 76, platform GPL2700) and GSE154851(Control = 32 and AS = 38, platform GPL16699) were obtained from GEO [11, 12]. TCGA-LUAD transcriptomic data from the TCGA-LUAD cohort were log2(x + 1) transformed to FPKM. GEO microarray data were normalized using the limma R package with background correction and quantile normalization [13]. A comprehensive list of chromatin regulator genes was compiled from GeneCards (relevance score > 1) [14]. AS-related target genes were extracted from the Comparative Toxicogenomic Database (CTD) [15]. Recognition of AR-associated patterns In the TCGA-LUAD cohort, univariate Cox proportional hazards regression analysis (p < 0.05) was performed to identify genes significantly associated with overall survival (OS) [16]. The intersection of chromatin regulator genes, AS targets, and LUAD prognostic genes yielded the initial set of chromatin regulator-AS (AR)-associated prognostic indicators. Their biological functions were analyzed via GO and KEGG enrichment using the clusterProfiler R package [17]. Somatic mutation profiles were visualized with the maftools R package [18]. Disease ontology (DO) analysis was performed to identify AR-associated indicators specifically linked to LUAD pathogenesis via DOSE package in R [19]. Next, AR-associated indicators specifically linked to LUAD correlated relationship and importance were estimated by circlize and GOSemsim package in R [20]. NMF analysis Based on the expression of the LUAD-pathogenesis-related AR indicators, NMF clustering was performed on the GSE26939 cohort using the NMF R package [21]. The optimal number of clusters (k) was determined by evaluating the cophenetic correlation coefficient and dispersion. Patients were stratified into distinct subgroups (C1 and C2) [21]. KM survival analysis compared OS between subgroups. Differences in immune cell infiltration (CIBERSORT algorithm in R), GSEA pathway activity (GSEA analysis using the clusterProfiler R package ), expression of the 11 AR indicators, tumor stemness index (One-class logistic regression, OCLR algorithm in R), predicted response to immunotherapy (TIDE algorithm in R), and immune checkpoint expression were compared between the high-risk (C1) and low-risk (C2) subgroups [17, 22–24]. WGCNA analysis GSVA was performed to calculate AS and chromatin regulator pathway activity scores in the C1 and C2 subgroups according to AS and chromatin regulator gene set [25]. WGCNA was conducted on GSE26939 to identify modules correlated with the AS score, chromatin regulator score, and patient subgroup(C1 and C2) [26]. The soft-thresholding power (β) was chosen to achieve appropriately β a scale-free topology fit. Genes within the most relevant module were extracted [26]. Univariate Cox regression (p < 0.05) on these genes in the TCGA-LUAD cohort identified candidate prognostic genes. Explainable machine learning model construction We systematically trained and tested 101 machine learning algorithm combinations using the caret and glmnet packages in R, with TCGA-LUAD as the training set and GSE68465 as the independent validation set [27]. The optimal model was selected based on the highest average concordance index (C-index) [27]. Model performance was assessed via KM survival analysis and time-dependent ROC analysis [27]. A nomogram integrating the model risk score and clinical stage was constructed, and calibration was evaluated with 1000 bootstrap resamples [27]. SHAP analysis was used to interpret the model and identify the feature with the greatest contribution, designated as the hub gene in R [27]. The association of the hub gene with immune cell infiltration was estimated by CIBERSORT and ESTIMATE packages in R [27]. Its diagnostic value for LUAD in TCGA-LUAD cohort, and AS (GSE23746 andGSE154851) was evaluated using ROC curves with pROC package in R [28]. Mutation frequency was queried via cBioPortal database [29]. Single-cell analysis The GSE203360 dataset was processed using Seurat in R [30]. Cells with unique feature counts < 200 or > 6000, or mitochondrial gene percentage > 20% were filtered out [30, 31]. Data were normalized, scaled, and principal components were calculated. Cells were clustered and visualized using UMAP and t-SNE [30]. Cell types were annotated with SingleR using a human immune cell reference in R [32]. Cell-cell communication networks were analyzed with CellChat package in R [33]. Metabolic heterogeneity was assessed using SCENITH package in R [34]. The BayesPrism tool in R was used to deconvolute the M2-like macrophage abundance from single-cell data into the TCGA-LUAD bulk cohort, and its correlation with survival was analyzed [35]. Within the annotated macrophage cluster, virtual knockout of LAIR1 was simulated using scTenifoldKnk in R to analyze its impact on intercellular communication and intracellular signaling [36]. Pseudotime trajectory analysis of macrophages was performed with Monocle2 in R [37]. Bayes-prisms were performed according to human lung cancer GWAS data and Bayes-prism package in R [35, 38]. Cell lines and culture condition Human LUAD cell line A549, monocytic cell line THP-1, and human umbilical vein endothelial cells (HUVECs) were purchased from SUNCELL company(China) and cultured in RPMI-1640 or DMEM with 10% FBS(Gibico, Germany) and 100 U/mL of penicillin-streptomycin (Sigma-Aldrich, USA) for 24 h. Macrophage polarization was induced by treatment with IL4/IL13 (10 ng/mL, Vazyme, China) or LPS/IFNγ (10 ng/mL, Vazyme, China) for 48 h. Subsequently, the polarized THP-1 cell was cultured in serum-free medium for 24 h. The supernatant was then collected as conditioned medium and stored at -80 °C. Next, we added 3 mL supernatant to A549 for simulation macrophage-A549 communication. HEK293T cell line were purchased from SUNCELL company(China), which was cultured in high-glucose DMEM (Gibico, Germany) was supplemented with 10% FBS(Gibico, Germany) and 100 U/mL of penicillin-streptomycin (Sigma-Aldrich, USA). q-RT-PCR Total RNA was extracted using TRIzol reagent (TaKaRa, China), and its concentration and purity were determined. Subsequently, 1 µg of total RNA was reverse-transcribed into cDNA with the HiScript II Q RT SuperMix kit (Vazyme, China), which incorporates a gDNA removal step. qRT-PCR was performed for 40 cycles using SYBR Green MasterMix (YEASEN, China), with each sample run in triplicate. Gene expression levels were quantified via the 2^(-ΔΔCt) method, using GAPDH as the endogenous reference gene for normalization. The specific primer sequences used are provided as below: LAIR1: 5’-ATCGGGGTCTCAGTGGTCTTC-3’. 5’-TGCTTTATCTGATTCTGGCGATG-3’. GAPDH: 5’- CATCACTGCCACCCAGAAGAC − 3’. 5’- CCAGTGAGCTTCCCGTTCAG − 3’. shRNA for silencing A lentiviral-mediated approach was used to establish stable knockdown cell lines. Briefly, shRNA sequences targeting ICAR1 were individually cloned into the pLKO.1 lentiviral vector. Each construct was then co-transfected with the psPAX2 packaging plasmid and the VSV-G envelope plasmid into HEK293T cells using a standard transfection reagent to produce lentiviral particles. The viral supernatant was collected 72 h post-transfection. For infection, A549 cells in the logarithmic growth phase were seeded at 5 × 10⁴ cells per well and infected with the harvested viral supernatant supplemented with 4 µg/ml polybrene. Following a 72-hour incubation, the cells were trypsinized and replated at a low density (500 cells per 10-cm dish) to initiate puromycin selection, which was maintained for three weeks to eliminate non-transduced cells. Surviving monoclonal colonies were isolated using cloning rings and subsequently expanded. To ensure clonal purity, these populations were further subcloned via the limiting dilution method, ultimately yielding stable monoclonal cell lines with knockdown of the respective target genes. Flow cytometry Cells (approximately 1 × 10⁶) were resuspended in phosphate-buffered saline (PBS) and immune-stained first with 5 µL of a mouse anti-human CD163 monoclonal antibody (222441, ZEN-BIOSCIENCE, China) for 30 min, followed by a PE-conjugated secondary antibody (MA1-10377, Thermo Fisher, USA) for 20 min. After washing, the cells were resuspended in fresh PBS and analyzed on a CytoFLEX flow cytometer (Beckman, USA). Data were analyzed using appropriate gating strategies. CCK-8 assays Cells in the logarithmic growth phase were harvested, counted, and plated into 96-well plates at a density of 3,000 cells per well (n = 6 wells per group). At the indicated time points (24, 48, 72, and 96 h), 10 µL of CCK-8 reagent was added to each well, and the plates were incubated for 2 h. The absorbance at 450 nm was then measured using a microplate reader. The cell proliferation rate was calculated by normalizing the absorbance at each day to the 4-hour time point. All experiments were conducted with three independent replicates. Wound healing assays Cells were washed with ice-cold PBS (Hyclone, USA) and harvested by gentle scraping. Total protein extraction was performed using RIPA lysis buffer (Beyotime, China) containing phosphatase and protease inhibitors. After centrifugation at 14,000 × g for 15 min at 4 °C, the supernatant was mixed with 5× SDS-PAGE loading buffer (Beyotime, China) and denatured at 100 °C for 10 min. Proteins were resolved by SDS-PAGE and transferred onto PVDF membranes (Beyotime, China). The membranes were blocked with NcmBlot blocking buffer (NCM Biotech, China) for 10 min and then incubated overnight at 4 °C with primary antibodies diluted in 5% BSA (Solarbio, China). Following washes, membranes were incubated with HRP-conjugated secondary antibodies (ThermoFisher, USA) diluted 1:1000 in a commercial antibody diluent (Beyotime, China) for 2 h at room temperature. Protein bands were visualized using an ECL substrate (Thermo Fisher, USA) and quantified with ImageJ software (version 1.57), with GAPDH serving as the loading control for normalization. The following primary antibodies from Abcam (USA) were used: anti-LAIR1 (ab189412, 1:1000) and anti-GAPDH (ab181602, 1:10000). Statistical analysis All statistical analyses were conducted using R software (version 4.2.2) for the bioinformatics component. A p-value or false discovery rate (FDR) of < 0.05 was defined as statistically significant. Data are presented as mean ± standard deviation (SD), with significance levels indicated as *p < 0.05, **p < 0.01, and ***p < 0.001. For the experimental data, statistical evaluation was performed using GraphPad Prism (version 8.0.2). Each experiment included a minimum of three biological replicates, and results are expressed as mean ± SD. Differences between two groups were assessed using two-way analysis of variance (ANOVA) or unpaired Student’s t-test, with p < 0.05 considered significant. Results Identification of AR-associated patterns in LUAD The intersection of chromatin regulator genes, AS targets from CTD, and LUAD prognostic genes from TCGA yielded 26 AR-associated prognostic indicators (Fig. 1A). GO and KEGG analysis revealed their enrichment in pathways like cell cycle and chromosomal instability(Fig. 1B). Mutation analysis showed a low to moderate mutation frequency for these genes in LUAD (Fig. 1C). DO analysis further refined this list to 11 genes significantly associated with LUAD pathogenesis (Fig. 1D-F). Identification of AR-associated risk group for LUAD patients NMF clustering based on the 11 genes in GSE26939 robustly classified patients into two subgroups (C1: high-risk, C2: low-risk) (Fig. 2A, B). The C1 subgroup exhibited significantly worse overall survival (Fig. 2C). Comparative analysis showed that C1 had a more immunosuppressive microenvironment (e.g., higher M2 macrophage infiltration), elevated tumor stemness, a higher TIDE score (indicating immunotherapy resistance), and increased expression of immune checkpoints like PD-L1 (Fig. 2D-G). Identification of AR-associated correlated gene signature for LUAD patients GSVA confirmed that both AS and chromatin regulator pathway activities were significantly upregulated in the C1 subgroup (Fig. 3D). WGCNA identified the “blue” module as most positively correlated with AS score, chromatin regulator score, and the risk subgroups (Fig. 3A-C). From this module, 33 genes were extracted. AR-associated explainable predictive model construction for LUAD patients Univariate Cox analysis in TCGA-LUAD identified 5 of these as significant prognostic genes (Fig. 4B). Among 101 machine learning models, the RSF + Ridge regression model demonstrated optimal and stable performance across training (TCGA) and validation (GSE68465) cohorts (Fig. 4A). The model effectively stratified patients into high- and low-risk groups with significant survival differences (Fig. 4C-D). The nomogram showed good predictive accuracy, confirmed by calibration plots (Fig. 4E-G). SHAP analysis identified LAIR1 as the feature contributing most to the model predictions (Fig. 4H). The characters of AR-associated molecular patterns in AS and LUAD pathogenesis In TCGA-LUAD, high LAIR1 expression correlated with increased infiltration of M2 macrophages and regulatory T cells, a higher ESTIMATE stromal score (Fig. 5C-D). Mutation analysis via cBioPortal database revealed a low mutation frequency in LUAD (Fig. 5A-B). ROC analysis confirmed its high diagnostic value for LUAD (AUC = 0.91) (Fig. 5F). LAIR1 also showed significant diagnostic power in AS datasets GSE23746 and GSE154851 (AUC = 0.725 and 0.898, respectively) (Fig. 5G-H). Besides, LAIR1 was also up-regulated in LUAD patients compared to normal samples (Fig. 5E). Macrophage single-cell implications in LUAD Analysis of 6 LUAD samples (GSE203360) at single-cell level identified 19 clusters annotated as malignant cells, T cells, B cells, macrophages, DC, monocytes, and epithelial cells (Fig. 6A-D). LAIR1 expression was predominantly localized to a subset of macrophages (Fig. 6E). This macrophage subset expressed high levels of M2 markers (CD163 and CD209) and was therefore defined as M2-like (Fig. 6F). BayesPrism deconvolution revealed that a high abundance of this specific LAIR1 + M2-like macrophage signature in the TCGA-LUAD bulk cohort was associated with significantly worse patient survival (Fig. 6G-H). AR-associated key gene molecular mechanisms in LAIR1 + M2-like macrophage of LUAD patients CellChat analysis revealed extensive communication between macrophages and other cell types, particularly malignant cells (Fig. 7A). A significantly altered metabolic pattern was observed in this macrophage subset (Fig. 7B).Virtual knockout of LAIR1 within the macrophage cluster significantly perturbed key signaling pathways, particularly those involved in macrophage-T cell and macrophage-other myeloid cell interactions (Fig. 7B). Pseudotime trajectory analysis showed that macrophages followed 11 distinct differentiation branches, with LAIR1 acting a stable functional role independent of differentiation stage. (Fig. 7F-G). The association of M2-like macrophage distributed AR-associated key gene with LUAD progression and AS pathogenesis q-RT-PCR and western blot confirmed that LAIR1 expression was upregulated in an AS cell model (ox-LDL treated HUVECs) and in IL-4/IL-13 induced M2-like THP-1 macrophages (Fig. 8A, B). Immunohistochemistry (IHC) data from the HPA database supported higher LAIR1 protein levels in LUAD tissues (Fig. 8C). Knockdown of LAIR1 in M2-like macrophages (sh-LAIR1) significantly reduced the expression of the M2 marker CD206 (Fig. 8D, E). Conditioned medium from LAIR1-knockdown macrophages significantly inhibited the proliferation and migration of A549 cells compared to the control group (Fig. 8F, G). Preventative therapeutic agent enrichment and validation DrugReflector screening nominated the flavonoid quercetin as the top natural compound capable of reversing the high-risk LUAD expression signature (Fig. 9A). Molecular docking demonstrated a strong binding affinity between quercetin and the LAIR1 protein (Fig. 9B). The IC50 of quercetin for THP-1 cells was determined to be 60 µM (Fig. 9C). Treatment of THP-1 cells with quercetin at this concentration, followed by collection of the conditioned medium, significantly suppressed the proliferation of A549 cells compared to control CM (Fig. 9D). Discussion and conclusion This study establishes a novel molecular link between AS and LUAD through the lens of chromatin regulator biology. We identified a convergent gene signature reflective of chromatin regulator and AS pathway activity, which effectively stratifies LUAD patients into prognostically distinct subgroups. Through an explainable AI framework, we pinpointed LAIR1 as the central hub of this signature. We demonstrated that LAIR1 is specifically expressed in tumor-associated M2-like macrophages, where it fosters an immunosuppressive microenvironment and is associated with poor prognosis in both computational and experimental models. Furthermore, we validated the natural compound quercetin as a potential therapeutic agent capable of disrupting this LAIR1-mediated axis. LAIR1 (Leukocyte-Associated Immunoglobulin-Like Receptor 1) is an inhibitory immunoreceptor known to transmit negative signals upon collagen binding, thereby modulating immune cell activation [39]. In AS, LAIR1 on macrophages modulates inflammatory responses. Some studies suggest it may dampen excessive inflammation, but it could also promote an M2-like, pro-resolutive state, which may contribute to plaque stability or progression in a context-dependent manner [40]. In cancer, LAIR1 has been reported to be expressed on various immune cells and some tumor cells, often associated with immune suppression and poor outcomes in cancers like gastric cancer and acute myeloid leukemia [41]. Besides, TME also plays an important role in LUAD patients. For example, TME of LUAD, low-density lipoprotein was also associated with the detrimental prognosis of LUAD patients, and higher infiltration of M2 macrophage [42]. In conclusion, our integrated multi-omics and experimental approach deciphered a novel LAIR1-centered chromatin regulator-M2 macrophage axis, linking AS-related pathology to aggressive LUAD. This work provides a reliable prognostic signature and nominates quercetin, a compound with known anti-inflammatory and anti-cancer properties, as a promising repurposing candidate for targeting this axis. However, this study has several limitations. First, the prognostic model requires prospective validation in multi-center clinical trials to assess its real-world utility and integration with existing clinical parameters. Second, the causal link between chromatin regulator dysregulation, LAIR1 upregulation, and M2 macrophage polarization, while supported by in vitro knockdown data, requires validation using in vivo models, such as Lair1-knockout mice in a LUAD metastasis model. The specific upstream signals that induce LAIR1 in the TME remain to be defined. Third, although quercetin showed efficacy in our co-culture system, its low bioavailability is a well-known challenge. Addressing these limitations will be crucial for translating these findings into clinical strategies for managing LUAD patients with or at risk of cardiovascular comorbidities. Fig. 1Identification of chromatin regulator and AS-associated prognostic indicators in LUAD. (A) Venn diagram showing the intersection of chromatin regulator genes, AS target genes, and LUAD prognostic genes, yielding 26 AR-associated indicators. (B) Bubble plot of top enriched GO and KEGG pathways for the 26 indicators.(C) The mutation landscape of the 26 indicators in the TCGA-LUAD cohort.(D-F) DO enrichment analysis identifying 11 indicators specifically linked to LUAD Identification of chromatin regulator and AS-associated prognostic indicators in LUAD. (A) Venn diagram showing the intersection of chromatin regulator genes, AS target genes, and LUAD prognostic genes, yielding 26 AR-associated indicators. (B) Bubble plot of top enriched GO and KEGG pathways for the 26 indicators.(C) The mutation landscape of the 26 indicators in the TCGA-LUAD cohort.(D-F) DO enrichment analysis identifying 11 indicators specifically linked to LUAD Fig. 2AR-associated risk stratification of LUAD patients. (A) Consensus matrix from NMF clustering (k = 2) in the GSE26939 cohort. (B) Cophenetic correlation coefficient for selecting the optimal cluster number (k = 2). (C) KM survival curves for the C1 (high-risk) and C2 (low-risk) subgroups. (D) Heatmap comparing of 11 AR-associated prognostic indicators between C1 and C2. (E) Box plots comparing tumor stemness index and TIDE score between subgroups. (F) Bar plot comparing expression levels of key immune checkpoint genes between subgroups.(G) Bar plot of TIDE analysis between C1 and C2. (H) CIBERSORT analysis between C1 and C2. (I) GSEA analysis between C1 and C2 AR-associated risk stratification of LUAD patients. (A) Consensus matrix from NMF clustering (k = 2) in the GSE26939 cohort. (B) Cophenetic correlation coefficient for selecting the optimal cluster number (k = 2). (C) KM survival curves for the C1 (high-risk) and C2 (low-risk) subgroups. (D) Heatmap comparing of 11 AR-associated prognostic indicators between C1 and C2. (E) Box plots comparing tumor stemness index and TIDE score between subgroups. (F) Bar plot comparing expression levels of key immune checkpoint genes between subgroups.(G) Bar plot of TIDE analysis between C1 and C2. (H) CIBERSORT analysis between C1 and C2. (I) GSEA analysis between C1 and C2 Fig. 3Construction and validation of the AR-associated prognostic model. (A) WGCNA cluster dendrogram and module assignment. (B-C) β threshold of WGCNA analysis. (D) GSVA scores for AS and chromatin regulator pathways in C1 and C2 subgroups. (E) Module-trait relationship heatmap highlighting the blue module. (F) WGCNA cluster dendrogram and module assignment. (G-J) Blue module trait Construction and validation of the AR-associated prognostic model. (A) WGCNA cluster dendrogram and module assignment. (B-C) β threshold of WGCNA analysis. (D) GSVA scores for AS and chromatin regulator pathways in C1 and C2 subgroups. (E) Module-trait relationship heatmap highlighting the blue module. (F) WGCNA cluster dendrogram and module assignment. (G-J) Blue module trait Fig. 4Construction and validation of the AR-associated prognostic model. (A) Bar chart comparing the C-index of top-performing machine learning models in training and validation sets. (B) Univariate cox regression analysis for prognostic indicator identification. (C) KM and time-dependent ROC analysis in TCGA-LUAD cohort. (D) KM and time-dependent ROC analysis in GSE68465 cohort.(E-G) Examination of AR-associated model performance. (H) SHAP analysis Construction and validation of the AR-associated prognostic model. (A) Bar chart comparing the C-index of top-performing machine learning models in training and validation sets. (B) Univariate cox regression analysis for prognostic indicator identification. (C) KM and time-dependent ROC analysis in TCGA-LUAD cohort. (D) KM and time-dependent ROC analysis in GSE68465 cohort.(E-G) Examination of AR-associated model performance. (H) SHAP analysis Fig. 5Characterization of LAIR1 in LUAD and AS. (A-B) Genetic variation analysis of LAIR1 in LUAD patients. (C) Scatter plot showing positive correlation between LAIR1 expression and ESTIMATE stromal score. (D) Scatter plot shows positive correlation between LAIR1 expression and CIBERSORT immune score. (E) Expression of LAIR1 in LUAD. (F) LAIR1 diagnostic performance in TCGA-LUAD patients. (G-H) Diagnostic performance of LAIR1 in GSE23746 and GSE154851 Characterization of LAIR1 in LUAD and AS. (A-B) Genetic variation analysis of LAIR1 in LUAD patients. (C) Scatter plot showing positive correlation between LAIR1 expression and ESTIMATE stromal score. (D) Scatter plot shows positive correlation between LAIR1 expression and CIBERSORT immune score. (E) Expression of LAIR1 in LUAD. (F) LAIR1 diagnostic performance in TCGA-LUAD patients. (G-H) Diagnostic performance of LAIR1 in GSE23746 and GSE154851 Fig. 6Single-cell landscape of LUAD and localization of LAIR1. (A-B) UMAP and t-SNE plot of cell clusters. (C) UMAP and t-SNE plot of cell types. (D) Proportional bar chart of cell type composition across samples. (E) Feature plot showing LAIR1 expression. (F) Expression of M2 macrophage markers across cell types. (G-H) Kaplan-Meier curve showing that high inferred abundance of the LAIR1 + M2-like macrophage signature (from BayesPrism deconvolution) correlates with poor survival in the TCGA-LUAD cohort Single-cell landscape of LUAD and localization of LAIR1. (A-B) UMAP and t-SNE plot of cell clusters. (C) UMAP and t-SNE plot of cell types. (D) Proportional bar chart of cell type composition across samples. (E) Feature plot showing LAIR1 expression. (F) Expression of M2 macrophage markers across cell types. (G-H) Kaplan-Meier curve showing that high inferred abundance of the LAIR1 + M2-like macrophage signature (from BayesPrism deconvolution) correlates with poor survival in the TCGA-LUAD cohort Fig. 7Functional analysis of LAIR1 in macrophages at single-cell resolution. (A) Circle plot of significant cell-cell communication pathways inferred by CellChat. (B) Metabolic heterogeneity of various cell types. (C-E) Virtual knockout of LAIR1 in macrophage. (F) Pseudotime trajectory of macrophages inferred by Monocle2. (G) Expression trend of LAIR1 along the macrophage pseudotime trajectory Functional analysis of LAIR1 in macrophages at single-cell resolution. (A) Circle plot of significant cell-cell communication pathways inferred by CellChat. (B) Metabolic heterogeneity of various cell types. (C-E) Virtual knockout of LAIR1 in macrophage. (F) Pseudotime trajectory of macrophages inferred by Monocle2. (G) Expression trend of LAIR1 along the macrophage pseudotime trajectory Fig. 8In vitro validation of LAIR1 function in M2 macrophages. (A-B) qRT-PCR analysis of LAIR1 mRNA levels in HUVECs treated with ox-LDL and in THP-1 cells polarized to M2 phenotype. (C) IHC examination of LAIR1 expression. (D) WB confirming LAIR1 protein knockdown efficacy in M2-like THP-1 cells. (E) Flow cytometry analysis showing reduced CD206 + population in M2-like THP-1 cells after LAIR1 knockdown (sh-LAIR1) compared to control (sh-NC). (F) CCK-8 assay showing the proliferation of A549 cells cultured with conditioned medium from sh-LAIR1 macrophages. (G) Wound healing assay showing migration of A549 cells treated with conditioned medium from sh-LAIR1 macrophages In vitro validation of LAIR1 function in M2 macrophages. (A-B) qRT-PCR analysis of LAIR1 mRNA levels in HUVECs treated with ox-LDL and in THP-1 cells polarized to M2 phenotype. (C) IHC examination of LAIR1 expression. (D) WB confirming LAIR1 protein knockdown efficacy in M2-like THP-1 cells. (E) Flow cytometry analysis showing reduced CD206 + population in M2-like THP-1 cells after LAIR1 knockdown (sh-LAIR1) compared to control (sh-NC). (F) CCK-8 assay showing the proliferation of A549 cells cultured with conditioned medium from sh-LAIR1 macrophages. (G) Wound healing assay showing migration of A549 cells treated with conditioned medium from sh-LAIR1 macrophages Fig. 9Identification and validation of quercetin as a therapeutic candidate. (A) Schematic of the DrugReflector pipeline identifying quercetin as the top candidate to reverse the high-risk LUAD signature. (B) Molecular docking model showing the binding pose of quercetin within the predicted binding pocket of the LAIR1 protein. (C) Dose-response curve of quercetin in THP-1 cells for IC50 determination.(D) CCK-8 assay showing the proliferation of A549 cells cultured with conditioned medium from THP-1 cells treated with DMSO or quercetin at its IC50 Identification and validation of quercetin as a therapeutic candidate. (A) Schematic of the DrugReflector pipeline identifying quercetin as the top candidate to reverse the high-risk LUAD signature. (B) Molecular docking model showing the binding pose of quercetin within the predicted binding pocket of the LAIR1 protein. (C) Dose-response curve of quercetin in THP-1 cells for IC50 determination.(D) CCK-8 assay showing the proliferation of A549 cells cultured with conditioned medium from THP-1 cells treated with DMSO or quercetin at its IC50 Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2