Tumor-infiltrating natural killer cell profiling for therapeutic stratification in patients with resectable non-small cell lung cancer.

Introduction
The introduction of immune checkpoint inhibitors (ICIs) has fundamentally reshaped the treatment landscape for advanced non-small cell lung cancer (NSCLC). Multiple predictive biomarkers, including tumor programmed cell death 1 ligand 1 (PD-L1) expression, tumor mutational burden (TMB), and smoking status, have become helpful indicators for clinical decision-making.1–24 As neoadjuvant immunotherapy or chemoimmunotherapy becomes increasingly adopted as a promising strategy for early-stage, resectable NSCLC,25–29 a clinical focus is shifting beyond palliative treatment for advanced disease toward curative-intent surgery, with maximizing long-term survival outcomes. Recent clinical trials, including CheckMate 816 and AEGEAN, have shown that neoadjuvant or perioperative chemoimmunotherapy significantly reduces the risk of relapse or death by 23% to 37% compared to chemotherapy alone.25–27 Notably, pathologic complete response (pCR) rates among current or former smokers receiving neoadjuvant chemoimmunotherapy ranged from 17% to 26%, whereas non-smokers achieved rates between 0% and 10%, likely attributable to their elevated TMB and increased tumor immunogenicity.25–27 However, the tumor-immune microenvironment (TiME) in early-stage NSCLC is biologically distinct from that of advanced disease,30, 31 limiting the direct applicability of biomarkers established in the metastatic setting. This discrepancy underscores an urgent need to identify biomarkers specifically tailored to the neoadjuvant setting in early-stage NSCLC.
From a surgical perspective, neoadjuvant chemoimmunotherapy presents several logistical and ethical challenges. Its real-world implementation often encounters multiple practical barriers, including surgical delays, cancellations, technical difficulties of surgery due to immune-mediated fibrotic changes, and increased postoperative complication rates.32 After neoadjuvant chemoimmunotherapy, 16% to 22% of patients did not undergo planned surgical resection due to disease progression, treatment-related adverse events, or new medical ineligibility.25, 26, 28, 33 Although smokers demonstrate superior responses to neoadjuvant chemoimmunotherapy, they also face higher risks of postoperative complications.34–37 This paradox highlights an unmet need for biomarkers that can stratify patients not only by therapeutic benefit but also by surgical risk.
Amid recent clinical advances, the identification and validation of reliable biomarkers in smokers is essential for optimizing neoadjuvant chemoimmunotherapy strategies and improving outcomes in patients with potentially resectable NSCLC. Therefore, we aimed to identify smoking-associated immune cell determinants by analyzing single-cell RNA sequencing (scRNAseq) data from the lung microenvironment of smokers and NSCLC tissues. Leveraging these scRNAseq-derived cellular signatures, we performed in silico deconvolution of bulk mRNA sequencing data to investigate the relationships between smoking-related immune cell compositions and clinical outcomes in patients with NSCLC who underwent either upfront surgical resection or neoadjuvant chemoimmunotherapy followed by surgery.

Materials and Methods

Study cohorts
This study was conducted under protocols approved by the Institutional Review Board at Baylor College of Medicine (BCM) (H-35782: approval date – 10/22/2021) and the University of Arizona (UA) (1811124026). Written informed consent was obtained from all participants for the collection of clinical data and biospecimens and the publication of the study data. At BCM, a prospectively maintained, single-institution database was reviewed retrospectively. The BCM cohort included 102 patients with histologically confirmed NSCLC who underwent complete pulmonary resection and mediastinal lymph node dissection. Staging was determined according to the 8th edition of the TNM classification.38 At UA (Drs. Polverino and Malo), lung tissue samples were obtained from patients undergoing lung volume reduction surgery or lung transplantation for severe emphysema, or from patients undergoing lung resection for a solitary peripheral nodule, with sampled tissue taken from a region at least 10 cm away from the nodule. The UA cohort (N=61) included 13 non-smokers, 21 ever-smoker control subjects, 15 patients with mild chronic obstructive pulmonary disease (COPD)(GOLD stage 1–2), and 12 patients with severe COPD (GOLD stage 3–4). COPD was defined as a pre-bronchodilator forced expiratory volume in 1 second (FEV1)/forced vital capacity (FVC) ratio of less than 0.7 (70%). Disease severity was classified according to the GOLD criteria as follows: GOLD 1 (FEV1 ≥ 80%), GOLD 2 (50% ≤ FEV1 < 80%), GOLD 3 (30% ≤ FEV1 < 50%), and GOLD 4 (FEV1 < 30%).

Single-cell isolation and scRNAseq
Lung tissues were collected for single-cell isolation as previously described.30, 39 In brief, tissues were minced and enzymatically digested using a human tissue dissociation kit (Miltenyi Biotec Inc., Auburn, CA, USA, Cat# 130-095-929). The digested mixture was sequentially filtered and washed twice. Then, cells were cryopreserved in a solution of 10% DMSO and 90% fetal bovine serum at −80°C. For long-term storage, cells were placed in the vapor phase of a liquid nitrogen storage system.
Single-cell suspensions were processed to isolate nuclei following the 10X Genomics protocol. Isolated nuclei were loaded into separate wells of a Chromium Next GEM Chip J (10X Genomics, Cat# 2000264) to generate gel beads-in-emulsion (GEMs), with each GEM containing uniquely barcoded beads for downstream single-cell capture and indexing. Nuclei concentration was adjusted to approximately 1,000 cells/μL to target recovery of up to 10,000 nuclei per sample. Sequencing libraries for simultaneous assay for transposase-accessible chromatin (ATAC) and 3’ gene expression were prepared using the Chromium Next GEM Single Cell Multiome GEM Kit A (10X Genomics, Cat# 1000282) according to the manufacturer’s protocol. All gene expression libraries were sequenced on the Illumina NovaSeq 6000 system using paired-end reads with dual indexing: 28 cycles for Read 1, 10 cycles for i7 Index, 10 cycles for i5 Index, and 90 cycles for Read 2. A sequencing depth of approximately 20,000 read pairs per cell was targeted. In this study, we used only gene expression data.
Sequenced reads were aligned to the human reference transcriptome (GRCh38) using the Cell Ranger (10x Genomics, v9.0.1)40 with default setting and generate gene-by-cell count matrices. Downstream quality control and analysis were performed in R using the Seurat package (v.5.3.0).41 Cells with fewer than 200 detected genes, greater than 20% mitochondrial gene content, or an abnormally high number of genes as doublets were excluded. Data normalization was performed using LogNormalize, with a scaling factor of 10,000. Dimensionality reduction was performed using uniform manifold approximation and projection (UMAP) for visualization.42, 43 Clustering was conducted using the Louvain algorithm with a resolution parameter set to 1.0.44 Initial cluster annotations were guided by differential gene expression analysis and confirmed using canonical marker genes to define major immune populations.

RNA extraction, mRNA sequencing, and data analysis
Total RNA was extracted from fresh-frozen human NSCLC tissues using mirVana™ miRNA Isolation Kit (Thermo Fisher Scientific, Cat# AM1560). Ribosomal RNA was removed, and cDNA libraries were prepared using Illumina Stranded Total RNA Prep with Ribo-Zero Plus (Illumina, Cat# 20040529). Libraries were sequenced on the Illumina NovaSeq 6000 with 100 bp paired-end reads, generating a mean of 80 million reads per sample at the Human Genome Sequencing Core at BCM. Raw sequencing reads were processed according to established bioinformatic protocols. Initial quality assessment was performed using FastQC, followed by adapter trimming and quality filtering with Trimmomatic. The human reference genome (GRCh38) and corresponding annotation files were retrieved from Gencode (https://www.gencodegenes.org/) and utilized to generate the STAR aligner index. High-quality reads were subsequently aligned to the reference genome using STAR. Gene- and isoform-level quantification was performed with transcripts per million (TPM).

Gene Set Enrichment Analysis
Gene set enrichment analysis (GSEA) was conducted using the GSEA 4.3.3 Desktop Application (http://www.gsea-msigdb.org/gsea/index.jsp).45, 46 Hallmark and BioCarta gene sets were obtained from the Molecular Signatures Database (MSigDB). This analysis involved 1000 random permutations for the gene sets and weighted enrichment statistics. The normalized enrichment score (NES) was used to assess enrichment, and significance was determined at a false discovery rate (FDR) < 0.05.

Generation of scRNAseq-derived NK cell signature and quantification of NK cell Abundance
To generate a custom NK cell signature, we identified differentially expressed genes from our scRNA-seq dataset (P<10−5 with |fold change|>2). This signature was then incorporated into CIBERSORTx47 for digital cytometry analysis of bulk mRNA sequencing data. To infer the relative proportions of immune cell types, we utilized CIBERSORTx in relative mode, applying the LM22 reference matrix consisting of 22 immune cell phenotypes, including NK cells. Baseline NK cell composition in LM22 was replaced with new NK cell compositions based on scRNAseq-derived NK cell signature. Furthermore, to quantify absolute NK cell abundance, we employed CIBERSORTx in absolute mode, which calculates cell-type scores adjusted by the reference immune cell panel following quantile normalization. Unlike relative mode that reflects proportional composition, absolute mode provides a score representing the inferred abundance of each immune cell type, enabling more direct comparison across samples.

Statistics
Cell compositions were expressed as percentages and summarized using means with standard deviations (SD). Pearson’s correlation analysis was performed to identify cell subsets positively or negatively associated with smoking status or COPD severity. Categorical variables were compared using Fisher’s exact test or Chi-square test, as appropriate. Continuous variables were analyzed using Student’s t-test for two-group comparisons and one-way ANOVA for comparisons involving more than two groups. For comparisons between tumors and their matched adjacent normal lung tissues, a paired t-test and Pearson correlation analysis were used. Survival analyses were conducted using the Kaplan-Meier method, and differences between groups were assessed using the log-rank test. Overall survival (OS) was defined as the time from surgery to death from any cause or last follow-up, while disease-free survival (DFS) was defined as the time from surgery to the first documented recurrence or death. To determine the cutoff value for stratifying patients by NK cell subset abundance, we employed a data-driven approach based on hazard ratios (HRs) using Cutoff Finder.48 Specifically, we plotted HRs across a range of potential cutoff values and selected the peak of the HR as the optimal threshold for survival analysis. Statistical significance was defined as P<0.05. Unless otherwise noted, all tests were two-sided. For comparisons of cell-type abundances between smokers and non-smokers in the limited scRNAseq datasets, one-sided tests were employed to assess the a priori hypothesis of higher abundance in smokers. All statistical analyses and graphic visualizations were performed using IBM SPSS 29.0 Statistics (IBM Corp) and GraphPad Prism 10.0 (GraphPad Software).

Data Availability
The data generated in this study are available within the articles and its supplementary data files and next-generation sequencing data for all samples in this study have been deposited in the Gene Expression Omnibus accessions (GSE283245, GSE300685, and GSE302339). Additional information is available upon request to the corresponding author.

RESULTS

scRNAseq profiling revealed smoking-associated cellular alterations in the lung microenvironment.
The study design is illustrated in Figure 1. Briefly, we identified cell populations associated with smoking status and COPD severity using lung scRNA-seq data (GSE302339). We then validated the presence of these smoking-related populations within the tumor-immune milieu using lung cancer scRNAseq datasets (GSE300685 and GSE13190749). To evaluate their clinical relevance, we derived cell-type signatures and applied them to NSCLC bulk RNA-seq cohorts via in silico cytometry (CIBERSORTx). Finally, we assessed the association between smoking-associated cell abundance and clinical outcomes in 102 completely resected lung cancers with 96 matched adjacent normal lungs (GSE283245) and examined response to neoadjuvant chemoimmunotherapy in an independent, publicly available cohort of 24 pre-treatment NSCLC tumors (GSE207422).
We first analyzed lung scRNAseq data from a total of 61 patients, including 13 non-smokers, 21 smokers without COPD, 15 patients with mild COPD, and 12 patients with severe COPD, to identify the smoking-associated cellular alterations. The clinicopathological characteristics of this population are summarized in Supplemental Table 1. We analyzed 198,814 cells, which were classified into immune cells, including lymphoid and myeloid cells, and non-immune cells, including stromal, mesothelial, epithelial, and endothelial cells. The cellular composition was visualized using UMAP (Figure 2A), and individual patient-level distributions of major cell phenotypes were shown in Figure 2B. Comparisons across non-smokers, smokers, and patients with mild or severe COPD revealed no significant differences in the overall proportions of major cell types (Figure 2C).
To gain deeper insight into smoking-induced cell alterations in the lung microenvironment, we extended our analysis with more detailed cell phenotypes. Lymphoid cells were classified into CD8 T cells, CD4 T cells, NK cells, B cells, and innate lymphoid cells (ILC), while myeloid cells were classified into macrophages, dendritic cells, monocytes, and neutrophils. Unsupervised clustering within each cell lineage revealed a total of 135 distinct cell phenotypes (Supplementary Figure 1). Correlation analysis was performed to identify cell phenotypes associated with smoking and COPD severity (Figure 2D). Six clusters demonstrated positive correlations with increasing COPD severity, including NK.10, NK.12, ILC.08 (AREG(+) ILC2), CD4.13 (Naïve CD4 T), Epithelial.10, and Endothelium.08. In contrast, six clusters, including NK.08, ILC.14, CD8.04 (Effector CD8 T), Epithelial.12 (ciliated cells), Endothelium.01, and Endothelium.02, showed negative correlations with COPD severity. Notably, among 12 significant clusters, three NK cell phenotypes, including NK.10, NK.12, and NK.08, exhibited significant associations with COPD severity. This finding suggests that NK cells may play a particularly pivotal role in COPD pathogenesis in the lung microenvironment. Given their strong association with COPD severity, NK cell populations were selected for in-depth analysis.

Transcriptional and pathway analyses define distinct lung NK cell subsets.
Building on the association of NK subsets with smoking status and COPD severity, we performed comprehensive transcriptional profiling of NK.10, NK.12, and NK.08 subpopulations to elucidate their functional properties (Figure 3A). UMAP analysis demonstrated that the combined NK.10 and NK.12 populations formed a distinct cluster separated from NK.08. Given the similarity of gene expression, NK.10 and NK.12 were collectively designated as NK subset 1, while NK.08 was classified as NK subset 2. Comparative transcriptomic analysis revealed markedly distinct gene expression profiles between NK cell subsets (Figure 3B). The NK subset 1 displayed a transcriptional signature characteristic of acute inflammatory activation, including upregulation of the tissue repair factor AREG and pro-inflammatory chemokines CCL3, CCL4. This subset demonstrated enhanced cytokine responsiveness through elevated expression of multiple cytokine signaling components responsive to IL-1, IL-12, IL-18, and IFN-γ (IL1B, IL12RB2, IL18RAP, IL21R, IL4R, IFNGR1), and a downstream signaling mediator STAT4. Coordinated upregulation of early and ligand-independent transcription factors (NR4A1, NR4A2, NR4A3) and NK-κB pathway components (NFKB1, NFKB2, NFKBID, NFKBIZ) indicated rapid transcriptional responses to inflammatory stimuli. Cellular stress adaptation was enhanced by increased expression of heat shock proteins (HSPA1A, HSPD1, HSPH11, HSP90AA1). This subset co-expressed the conventional NK marker (NCAM1/CD56), implying an immature phenotype,50, 51 as well as a high level of ZBTB16 (PLZF), a transcription factor for invariant NKT (iNKT) cell development,52, 53 and FCER1G, which encodes a signaling adaptor that enhances both cytotoxicity and cytokine response.54, 55 These features collectively define a stress-responsive NK cell subset (NKSR) uniquely adapted to the inflammatory microenvironment of lungs damaged by smoking and capable of rapid effector responses.
In contrast, the NK subset 2 exhibited a transcriptional profile consistent with chronic activation and functional immune regulation (Figure 3B). This population was characterized by elevated expression of multiple immune checkpoint receptors (PDCD1, LAG3, and TIGIT)56, tissue residency markers, including ZNF683 (Hobit)57 and CX3CR158, 59, and upregulation of CD27,60 indicating long-term adaptation to the lung microenvironment with acquisition of a tissue-resident memory-like phenotype. Terminal differentiation was evidenced by upregulation of KLRG161 and KLRC2 (NKG2C)62. This subset displayed enhanced antigen presentation capabilities through upregulation of HLA class II genes (HLA-DPA1, HLA-DPB1, HLA-DQA2, HLA-DRB1, HLA-DRB5)63, 64 and T cell signaling components (LAT, LCK).65, 66 Increased expression of GTPase of immune-associated proteins (GIMAP) family members (GIMAP1, GIMAP4, GIMAP7), associated with lymphocyte survival and activation,67 along with the cytotoxic effector GZMH further supported a state of adaptive differentiation. These features define a population of tissue-resident NK cells shaped by chronic antigen exposure, with enhanced immunosurveillance capabilities but diminished effector function. We designated this subset as adaptive and immunoregulatory NK (NKAIR) cells, reflecting their putative role as immunoregulatory sentinels within chronically inflamed lungs.
Consistent with their transcriptional profiles, pathway enrichment analyses revealed that NKSR cells were significantly enriched in pro-inflammatory signaling pathways, including TNF-α signaling via NF-κB, IL-2/STAT5 signaling, IL-6/JAK/STAT3 signaling, and the inflammatory response pathway (Figure 3C). NKSR cells also demonstrated enhanced activity in IL-12-mediated signaling networks, including NOS2-dependent IL-12 signaling and IL-12/STAT4-dependent signaling pathway (Figure 3D). These pathway profiles indicate that NKSR cells adopt a cytokine-driven activation mechanisms to facilitate robust pro-inflammatory responses and immune modulation. Conversely, NKAIR cells demonstrated significant enrichment in allograft rejection and RAS-independent pathway in NK cell-mediated cytotoxicity, indicating preserved but possibly regulated cytolytic potential. Interestingly, NKAIR cells exhibited enrichment in T cell receptor signaling, LCK and FYN tyrosine kinase pathway, and co-stimulatory signal pathway, indicating that they may retain cytotoxic potential.

Distinct NK cell subsets are recapitulated in lung adenocarcinoma.
To confirm the presence of NK subsets in the tumor immune microenvironment of lung cancer, we analyzed scRNAseq data from nineteen patients with invasive lung adenocarcinoma (LUAD) by integrating our 8 LUAD scRNAseq data (GSE300685) with publicly available 11 LUAD scRNAseq data (GSE131907: Supplementary Figure 2A).49 The smoking status and pathologic stage of these 19 patients are summarized in Supplemental Table 2. Intratumoral NK cells also had distinct clusters, whose differential gene expression patterns are highly consistent with findings from our previous transcriptomic analysis in Figure 3B. scRNAseq data confirmed the presence of both NKSR and NKAIR cell populations within the TiME (Figure 4A). NKSR cells were enriched in tumors from smokers compared to non-smokers, consistent with smoking-associated expansion of this subset (Figure 4B and 4C). Tumors from smokers also exhibited increased activated CD4 T (CD4 Tact) cells (Figure 4D and 4E) and immunosuppressive M2-like tumor-associated macrophages (TAMs) (Figure 4F, 4G, and Supplementary Figure 2B). The concurrent enrichment of cytokine-releasing NKSR cells with activated lymphocytes, and M2-like TAMs suggests a transitional state - poised between acute immune activation and emerging immune evasion.

NK cell-based subtyping stratifies treatment outcomes in resectable NSCLC.
To investigate the infiltration patterns of defined NK cell subsets between tumor and adjacent normal tissues in patients with NSCLC, we derived scRNAseq-based NK cell signatures to identify NKSR and NKAIR from bulk RNA sequencing data (Supplementary Table 3). Lung scRNAseq data showed that the average gene expression of these NK signatures was specific to the corresponding NKSR and NKAIR populations (Supplemental Figure 3). We then applied these NK signatures in CIBERSORTx digital cytometry47 across three datasets: 102 treatment-naïve NSCLC tumor samples (GSE283245), 96 matched adjacent normal lung samples (GSE283245), and publicly available 24 pre-treatment NSCLC tumors from patients who received neoadjuvant chemoimmunotherapy (GSE207422).68 The clinicopathological characteristics of our NSCLC cohort are summarized in Table 1.
Immune deconvolution revealed a significant reduction of NKSR cells in tumors compared to matched adjacent normal lung tissues (P=5.0×10−5 by paired t-test). NKAIR cell levels showed no significant difference between tumor and normal tissues (Figure 5A). While NKAIR cell abundance did not correlate between tumor and normal tissues, tumor NKSR levels showed a positive correlation with NKSR in matched adjacent normal tissues (Figure 5B), indicating that host/tissue context may partly shape the NKSR phenotype within the tumor. Stratification by smoking status (current smokers [CS], former smokers [FS], and non-smokers [NS]) showed that although NKSR cells are significantly enriched in the smoker lung microenvironment, tumor-specific exclusion of NKSR cells was most pronounced in current smokers (P=5.3×10−4 by paired t-test) and modestly observed in non-smokers (P=0.034 by paired t-test) (Figure 5C). Together, these findings suggest smoking-associated modulation of NK-cell infiltration and selective evasion of NKSR cells during tumorigenesis.
Given the more pronounced distributional differences in NKSR cells between tumors and adjacent normal tissues, we next assessed their prognostic relevance. The 102 patients were classified into NKSR HIGH and NKSR LOW groups based on absolute NKSR cell abundance score, using a cutoff value of 1.83 (Supplementary Figure 4A). The NKSR HIGH group demonstrated significantly better OS (P=0.001, HR=0.267, 95% CI: 0.112–0.633) and DFS (P=0.015, HR=0.439, 95% CI: 0.223–0.868) (Figure 5D). Interestingly, this survival benefit of NKSR HIGH group was particularly evident in current smokers, with significantly prolonged OS (P=0.026, HR=0.259, 95% CI:0.072–0.930) and DFS (P=0.034, HR=0.306, 95% CI: 0.096–0.974) (Figure 5D). In contrast, NKAIR abundance did not correlate with survival outcomes (Supplementary Figure 4B).
Univariable Cox regression analysis to assess the prognostic impact of intratumoral NKSR cell abundance showed that female sex, non-smoking status, lower cumulative tobacco exposure (pack-year), LUAD histology, earlier pathologic stage, and higher NKSR cell abundance in the tumor were all significantly associated with better OS (Supplementary Table 4). In the multivariable analysis incorporating these factors, higher NKSR cell abundance in the tumor remained independently associated with favorable outcomes, including better OS (P = 0.019; HR, 0.348; 95% CI, 0.144–0.841) and DFS (P = 0.029; HR, 0.459; 95% CI, 0.228–0.923). Advanced pathological stages (stage II/III) independently predicted worse DFS (P = 0.047; HR, 2.066; 95% CI, 1.010–4.226) (Table 2).

Tumor-infiltrating NK cell composition predicts response to neoadjuvant chemoimmunotherapy in NSCLC.
Given that smokers appear to benefit from neoadjuvant chemoimmunotherapy,25–27 NKSR-absent tumor is associated with unfavorable survival, and NKAIR cells exhibit immunoregulatory signals, we hypothesized that the intratumoral balance of these NK cell subsets may influence the efficacy of neoadjuvant chemoimmunotherapy. To test this hypothesis, we analyzed mRNA data from 24 NSCLC patients who underwent complete pulmonary resection following neoadjuvant chemoimmunotherapy with PD-1 blockade and chemotherapy (GSE207422).68 Patients were grouped by intratumoral NKSR and NKAIR abundance score. The group with NKSR LOW and NKAIR HIGH had significantly greater treatment responses, including higher rates of pCR and major pathologic response (MPR) (P=0.017), compared to all other NK cell profile combinations (Figure 5E). Specifically, 4 of 8 (50%) in this group achieved a pCR, and an additional 2 patients (25%) exhibited an MPR, resulting in a combined pCR/MPR rate of 75%. In contrast, among patients with other NK cell profile combinations, only 1 of 16 (6.3%) achieved pCR and 2 of 16 (12.5%) exhibited MPR. These findings suggest that tumors with high NKAIR and low NKSR cell abundance are more likely to respond to neoadjuvant chemoimmunotherapy, potentially serving as a predictive biomarker for treatment stratification.

DISCUSSIONS
In this study, we identified two transcriptionally and functionally distinct NK cell subsets, stress-responsive NK cells and adaptive/immunoregulatory NK cells. These NK subsets are differentially associated with smoking status, COPD severity, and clinical outcomes in NSCLC. Using scRNAseq of lung tissue from non-smokers, smokers, and LUAD, we defined immune signatures linked to smoking exposure and disease state. Subsequent digital cytometry of bulk RNA-seq data from completely resected treatment-naïve tumors revealed that high NKSR cell preservation correlated with better survival following surgery, especially in current smokers. In contrast, in neoadjuvant chemoimmunotherapy-treated NSCLC cohort, tumors enriched with NKAIR cells and depleted of NKSR cells demonstrated superior pathologic responses to chemoimmunotherapy. These findings offer a novel framework for NK cell-based stratification in resectable NSCLC - advancing a biologic clock of immune evasion in which progressive NKSR depletion marks movement toward an immune-suppressed, tumor-permissive state (Figure 6).
Neoadjuvant immunotherapy or chemoimmunotherapy offers substantial promise in the treatment of NSCLC by enabling tumor downstaging prior to curative-intent surgery and targeting micro-metastatic disease.69 However, successful implementation requires careful multidisciplinary planning to manage treatment-related adverse events70 and ensure completion of the planned surgery with minimal postoperative complications.32, 71 Current or former smokers often face higher surgical risks due to smoking-related comorbidities and structural lung changes, which are attributable to postoperative complications. Nevertheless, they appear to experience a greater benefit from neoadjuvant chemoimmunotherapy, with approximately 20% achieving a pCR.25–27 This observation presents an opportunity for personalized therapeutic strategies, as smokers, who had high surgical risks but a greater likelihood of achieving pCR or MPR, may be prioritized for neoadjuvant chemoimmunotherapy, while patients predicted to derive limited benefit, based on intratumoral NK cell profiling, may proceed directly to upfront surgery (Figure 7).
Despite the increasing adoption of neoadjuvant chemoimmunotherapy in smokers with early-stage NSCLC, there remains a critical lack of established biomarkers to guide patient selection. Currently, tumor PD-L1 expression is the only clinically validated and FDA-approved biomarker for predicting immunotherapy response in advanced NSCLC.72, 73 However, its clinical utility is hindered by variability in assay platforms, antibodies, and cutoff thresholds.74 Another potential biomarker, TMB, faces similar limitations, including challenges in obtaining adequate tissue samples and the limited representation of tumor heterogeneity from a single biopsy.75, 76 To address these issues, blood-based TMB is being investigated as a non-invasive alternative.77−79 In addition, clearance of circulating tumor DNA (ctDNA) prior to surgery has been associated with improved treatment responses and longer DFS in stage IIIA (N2) NSCLC.77−79 While these biomarkers show promise in advanced disease, no validated biomarkers currently exist for predicting response to neoadjuvant chemoimmunotherapy in early-stage NSCLC, highlighting a significant unmet need to characterize the tumor immune microenvironment in this setting. Our study suggests that the intratumoral balance of NKSR and NKAIR cells may serve as a novel predictive immune signature. Tumors with low NKSR and high NKAIR cells were significantly more likely to achieve pCR or MPR following neoadjuvant PD-1 blockade combined with chemotherapy. In contrast, patients with high NKSR infiltration at baseline exhibited favorable outcomes after upfront surgery, suggesting that these tumors may possess sufficient intrinsic immune surveillance, reducing the added benefit of neoadjuvant therapy.
We demonstrated NK cell diversity with NKSR and NKAIR subsets exhibiting opposite correlations with disease severity. NKSR cells are characterized by transcriptional gene profiles with acute inflammation, cytokine responsiveness, stress responses, and rapid activation. NKSR cells are poised for rapid activation under stress and may contribute to the recruitment and maintenance of activated T cells and proinflammatory macrophages in the tumor microenvironment. Conversely, NKAIR cells display a matured, chronically activated, functionally partially exhausted phenotype with tissue-resident memory cell-like features. This subset also exhibited antigen-presenting capabilities with T cell signaling components, representing preserved cytotoxic potential but reduced effector function in chronically inflamed lungs. NKAIR cells expressed multiple immune checkpoint receptors (PDCD1, LAG3, TIGIT), indicating impaired effector function but potential for therapeutic reinvigoration. Notably, NKAIR enrichment in pre-treatment tumors was associated with enhanced response to chemoimmunotherapy, likely due to partial restoration of cytotoxicity following PD-1 blockade and immune remodeling by chemotherapy.80, 81
Indeed, neoadjuvant chemoimmunotherapy can substantially remodel the TiME. Recent single-cell studies of stage III NSCLC have shown that neoadjuvant treatment is associated with increased intratumoral NK cell infiltration compared to untreated tumors.82 Furthermore, our findings highlight opportunities for therapeutic innovation, such as enhancing NKSR cells via cytokine support (e.g., IL-15 superagonists)83, blocking NK-specific checkpoints (e.g., NKG2A)84, or adoptive NK cell therapy.85, 86 Strategies to prevent stress ligand shedding or recondition the TiME may further potentiate NK cell-mediated antitumor activity.
Several limitations warrant consideration. First, the observational design precludes causal inference. Although we observed strong associations among smoking, NK cell subsets, and clinical outcomes, functional studies are needed to establish mechanistic roles for NKSR and NKAIR cells. Second, NK profiling was performed on resected tumors or bulk RNA-seq, which may not fully reflect the pretreatment immune landscape in small biopsies, although feasibility was demonstrated in a clinical trial.39 Future work should develop noninvasive biomarkers, such as circulating or sputum NK phenotypes or soluble NK ligands, to facilitate clinical application. Third, the scRNAseq LUAD cohort and the neoadjuvant-treated cohort were relatively small, underscoring the preliminary nature of these findings. Validation in larger, independent, prospective cohorts will be required. Finally, the influence of co-occurring immune subsets, stromal interactions, and tumor-intrinsic factors was not comprehensively assessed and may affect NK cell function and treatment response.

CONCLUSIONS
This study explores how smoking may shape the intratumoral immune landscape and examines NK cell heterogeneity in relation to lung immunity and NSCLC outcomes. We describe tumor-infiltrating NK cell profiling as a potential stratification approach for resectable NSCLC. In our data, a higher abundance of stress-responsive NK cells was associated with more favorable surgical outcomes in treatment-naïve NSCLC, whereas tumors enriched for adaptive/immunoregulatory NK cells are more likely to benefit from neoadjuvant chemoimmunotherapy followed by surgery. These observations suggest that intratumoral NK composition could serve as both a prognostic marker and a predictor of response to preoperative therapy. Preoperative tumor NK profiling might help inform individualized decisions to select patients for upfront surgery versus neoadjuvant chemoimmunotherapy. Further studies are needed to validate these NK subsets as clinical biomarkers, develop noninvasive profiling methods, and evaluate strategies aimed at preserving or restoring NKSR or reinvigorating NKAIR function. Together, these efforts have the potential to refine perioperative treatment strategies and improve long-term outcomes for patients with resectable NSCLC by optimizing the balance between therapeutic benefit and risk.

Supplementary Material
Supplement 1