Characterization of natural killer (NK) cells in lung adenocarcinoma and construction of an NK risk signature based on single-cell and macromolecular RNA-seg data

Abstract

Background/aims

Natural killer (NK) cells play a crucial role in tumor cell apoptosis, immune milieu regulation, and angiogenesis inhibition. This study aims to analyze the NK signature in lung adenocarcinoma (LUAD) and establish an NK cell-based risk signature for predicting the prognosis of LUAD patients.

Methods

Single-cell RNA sequencing (scRNA-seq) data were obtained from the GEO database, while RNA-seq and microarray data from LUAD were simultaneously obtained from the TCGA and GEO databases. The scRNA-seq data were processed using the Seurat R package to identify NK clusters based on NK markers. Differentially expressed genes (DEGs) between normal and tumor samples were identified through differential expression analysis of LUAD-related data. Pearson correlation analysis was used to identify DEGs associated with NK clusters, followed by one-way Cox regression analysis to identify NK cell-related prognostic genes. Subsequently, Lasso regression analysis was employed to construct a risk signature based on NK cell-related prognostic genes. Finally, a column-line diagram model was constructed based on the risk signature and clinicopathological features.

Results

Based on the scRNA-seq data, we identified five Natural killer (NK)cells clusters in lung adenocarcinoma (LUAD), with four of them showing associations with prognosis in LUAD. Out of 19,495 differentially expressed genes (DEGs), a total of 725 genes significantly associated with NK clusters were pinpointed and further narrowed down to form a risk profile comprising 13 genes. These 13 genes were primarily linked to 21 signaling pathways, including vascular smooth muscle contraction, RNA polymerase, and pyrimidine metabolism. Additionally, the risk profile exhibited significant associations with stromal and immune scores, as well as various immune cells. Multifactorial analysis indicated that the risk profile served as an independent prognostic factor for LUAD, and its efficacy in predicting the outcome of immunotherapy was validated. Furthermore, a novel column-line diagram integrating staging and NK-based risk profiles was developed, demonstrating strong predictability and reliability in prognostic forecasting for LUAD.

Conclusion

The NK cell-based risk signature proves to be a valuable tool for predicting the prognosis of patients with LUAD. Furthermore, a comprehensive understanding of NK cell characterization in LUAD could potentially unveil insights into the response of LUAD to immunotherapies and offer novel strategies for cancer treatment.

Introduction

Lung cancer is a lethal disease that causes the most frequent cancer-related deaths worldwide [1], with adenocarcinoma accounting for more than half of all cases [2]. Despite advances in surgery, radiotherapy, chemotherapy, and other novel therapeutics, the incidence and mortality of lung adenocarcinoma (LUAD) continues to increase [3]. Therefore, it is essential to identify the high-risk pathological factors for predicting the prognosis of LUAD. Omics technology is regarded as a effective tool for understanding the molecular pathogenesis of LUAD [4–6]. Whole-genome sequencing and transcriptome sequencing has provided valuable insights into the gene amplification of RNA methyltransferase in LUAD, suggesting the activation mechanism of tumorigenesis [4]. Single-cell RNA sequencing techniques were applied to investigate the distinct immune signatures between the LUAD and lung squamous cell carcinoma (LUSC) [5]. Furthermore, bioinformatics approaches have been instrumental in dissecting specific molecular targets and pathways in LUAD, such as elucidating the role of nuclear-localized insulin receptors through gene-knockdown strategies [7].

Natural killer (NK) cells, a type of tumor immune cells, are known for their ability to recognize and eliminate cancer cells [8]. Previous studies have shown that NK cells detect and lyse LUAD cells by engaging activating receptors with stress-induced ligands that are overexpressed on cancer cells’ surfaces [9, 10]. NK cells display an potent antitumor immune cytotoxicity by regulating the tumor microenvironment via MEK/ERK and PI3K/Akt/mTOR pathways [11, 12]. NK cells have the potential to interact with other immune cells to shape the adaptive immune response against LUAD [13]. Growing evidence have suggested that the presence and status of NK cells within LUAD are correlated with patient prognosis [14–16]. Increased NK cell infiltration and activity are frequently associated with improved survival and treatment outcomes [17].

This pursuit of prognostic biomarkers is reflected in the development of new predictive models based on multi omics studies such as methylation, metabolic reprogramming, and immune regulation [18], to identify key metabolic determinants such as amino acid metabolism related features [19], ribonucleoside diphosphate reductase subunit M2 (RRM2) [20], mitochondrial pathway activity[Mitochondrial Pathway Signature (MitoPS) predicts immunotherapy response and reveals NDUFB10 as a key immune regulator in lung adenocarcinoma], which to some extent demonstrate potential as prognostic biomarkers for LUAD patients. Therefore, NK cells interact with lung cancer in a complex and dynamic manner, and understanding these interactions is critical for predicting patients’ prognosis.

Despite some studies on NK cells in LUAD have been conducted [21], the systematic characteristics of NK cells, as well as their association with LUAD prognosis and immunotherapy response, remain limited known. While several molecular and metabolic biomarkers have shown prognostic value [18–20], a comprehensive, NK cell-centric signature that integrates its functional state with clinical outcome is still lacking. In this study, we collected LUAD single-cell RNA-sequencing (scRNA-seq) and transcriptome data from publicly available databases. We differentiated LUAD subclusters and identified NK cell-related risk signature for LUAD. The clinical relevance of the NK cell-based signature was identified, and the immunological landscape and response to immunotherapy that underpin the NK cell-based signature were further investigated. Finally, we created a unique nomogram that combined the NK cell-based risk signature with clinicopathological features to make it easier to use NK cell-based features in the prognosis of LUAD. It might bring new insights into the biology of LUAD, resulting in more targeted therapy and a better prognosis for those with LUAD.

Materials and methods

Data acquisition and organisation

ScRNA-seq data from GSE131907 were obtained from the Gene Expression Omnibus (GEO) database, comprising 2 LUAD and 2 normal control samples. Cells expressing each gene in at least 3 cells and with a minimum of 250 genes expressed per cell were selected. Mitochondria to rRNA ratio was evaluated using the PercentageFeatureSet function in the Seurat R package. Cells with at least 6000 genes expressed, UMI > 100 were retained, totaling 23,673 cells.

Single-cell transcriptomic and copy number variation (CNV) data were sourced from The Cancer Genome Atlas (TCGA) database, along with relevant LUAD clinical information. Samples lacking survival and outcome status data were excluded, resulting in 500 tumor and 59 paraneoplastic samples. The GSE31210 cohort of 226 LUAD samples was utilized as a validation cohort, with normal and tumor samples lacking follow-up and outcome status information removed. Subsequently, 10 cancer-related pathways were identified from the literature, including the WNT, PI3K, NOTCH, RAS, cell cycle, MYC, TGF-Beta, HIPPO, and NRF1 pathways [22].

Definition of NK

We re-analyzed LUAD single-cell RNA sequencing data using the Seurat package to characterize NK cells comprehensively [23]. Cells with over 6000 or under 250 expressed genes were filtered out. The gene expression data was log-normalized, and batch effects from 4 samples were removed. PCA was employed for non-linear dimensionality reduction with 20 principal components and a resolution of 0.25. Cells were clustered into subgroups using FindNeighbors and FindClusters functions (dim = 40, resolution = 0.2), followed by t-distributed stochastic neighbor embedding (t-SNE) with the RunTSNE function. NK cells were identified using marker genes (NKG7, KLRD1, KLRB1, GNLY, and GZMB). Clustering and t-SNE dimensionality reduction were then performed specifically on NK cells. Marker genes for each NK cluster were identified using the FindAllMarkers function, with inter-cluster comparisons conducted using logFC = 0.5, minpct = 0.35, and corrected P-value < 0.05. Gene ontology enrichment analysis (KEGG enrichment analysis) [24] on NK cluster marker genes was carried out using the clusterProfiler software package. CNV features in NK clusters were analyzed using the CopyKAT R software package to distinguish between tumor and normal cells in each sample [25].

Identification of NK hub genes

In the analysis, we initially identified differentially expressed genes (DEGs) between tumor and normal tissues using the limma software package, with a screening threshold of FDR < 0.05 and |log2(Fold Change)|>1.5 [26]. Subsequently, we evaluated the correlation between these DEGs and NK cell clustering, identifying key NK cell-related genes with p < 0.001 and cor > 0.4.

For prognostic gene screening, univariate Cox regression analysis was conducted using the survival package to identify genes with p < 0.05. Further gene selection was performed through least absolute shrinkage and selection operator (lasso) Cox regression analyses, followed by multivariate Cox regression analyses using stepwise regression methods. Based on the multivariate Cox model results, a risk signature was constructed using the formula: risk score = Σβi*Expi, where i represents genes in the risk signature, expi denotes gene expression, and βi signifies the coefficient from the multivariate Cox model. Patients were then categorized into high-risk and low-risk groups after zero-mean normalization. To assess the predictive performance of the risk signatures, receiver operating characteristic curve (ROC) analyses were carried out using the timeROC software package. Similar analyses were conducted in the validation cohort to validate the findings.

Immune landscape analysis

We assessed the proportion of 22 immune cell subtypes in the TCGA cohort using the CIBERSORT algorithm and calculated an immunity score and stroma score using the ESTIMATE algorithm [27](https://sourceforge.net/projects/estimateproject/) to further explore the tumour microenvironment (TME).

Constructing risk characteristics and column line plots

First, we performed univariate and multivariate Cox regression analyses of clinicopathological and risk characteristics. Variables with P < 0.05 in the multivariate Cox model were used to construct column-line plots for predicting the prognosis of LUAD patients using the rms package [28]. Also, calibration curves were generated to assess the predictive accuracy of the models, and the reliability of the models was assessed using decision curve analysis (DCA).

Responsiveness to immune checkpoint blocks

We downloaded transcriptomic and clinical data from the IMvigor210 cohort (http://research-pub.gene.com/IMvigor210CoreBiologies) of LUAD patients treated with anti-PD-L1 drugs (atezolizumab) [29]. In parallel, we also downloaded transcriptomic data from the GSE78220 cohort for pre-existing melanomas treated with anti-PD-1 checkpoint inhibition and used it to determine the potential value of risk marker scores in predicting responsiveness to immune checkpoint blockade therapy (ICB) [30].

Statistical analyses

All statistical analyses were performed using R software (v3.6.3). Correlation matrices were performed using Pearson or Spearman correlation. Wilcoxon test was performed to compare the two groups. Survival differences were compared using K-M curves and log rank test (log rank test ). p-value < 0.05 was considered statistically significant.

Results

Screening of NK clusters in scRNA-seq data

The design flowchart of this study is shown in Fig. 1. We obtained 23,673 cells after preliminary screening of scRNA-seq data. After log-normalisation and dimensionality reduction, 20 subclusters were obtained, and 680 NK populations were identified based on five marker genes, including NKG7, KLRD1, KLRB1GNLY, and GZMB (Fig. S1A, B). We further clustered and downscaled the cells of the 680 NK populations and finally identified five NK clusters (Figure S1C, D). We utilised the TSNE plots of the 4 sample distributions (Fig. 2A) and finally generated 5 NK clusters which were used for subsequent analysis (Fig. 2B). A total of 737 DEGs were identified in the 5 NK clusters, of which the expression of the top 5 DEGs (identified as marker genes in the NKF clusters) is shown in Fig. 2C. The ratio of the 5 clusters in each cohort is shown in Fig. 2D. KEGG analysis showed (Fig. 2E) that these DEGs were enriched in a variety of pathways, including TNF signalling pathway, T-cell receptor signalling pathway, and NK cell-mediated cytotoxicity. In addition, the five NK clusters consisted of 680 tumour cells and normal cells according to CNV characteristics (Fig. 2F).

Fig. 1.

The flow chart of this study

Fig. 2.

Identification of NK clusters based on scRNA seq data fromLUAD patients. A Distribution map of 21 samples; B Distribution map of 5 natural killer cells after clustering; C Expression point map of the top 5 marker genes in the subgroups; D Proportion and cell number of the subgroups and adjacent tissues in cancerous tissues; E KEGG enrichment analysis of the cellular subgroups of the 5 natural killer cells; F Distribution map of the tsne of malignant and non-malignant cells predicted by the copykat package

Expression of cancer-related pathways in NK

To further investigate the correlation between NK clusters and tumour progression, we performed GSVA scoring of 10 tumour-related pathways in the 5 NK clusters (Fig. 3A).The proportion of malignant cells in NK_0, NK_1 and NK_3 clusters was significantly higher than that in the other 2 clusters (Fig. 3B), with significant differences between each cluster. In addition, we analysed the GSVA scores of 10 tumour-associated pathways between malignant and non-malignant cells in each NK cluster, and the RAS pathway in the NK_0 cluster and the NOTCH pathway in the NK_1 cluster showed significant differences (Fig. 3C–G).

Fig. 3.

Characterisation of tumour-associated pathways in NK clusters. A Heatmap of the scores of 10 tumour-associated pathways enriched by NK cells; B Comparison of NK clusters in malignant and non-malignant cells; Comparison of GSVA scores of each pathway between malignant and non-malignant cells (C) NK_0; D NK_1; E NK_2; F NK_3; G NK_4;(wilcox.test, *P < 0.05; **P < 0.01; ***P < 0.001; and ****P < 0.0001). ns not significant

To determine the association between NK clusters and prognosis, we first calculated the ssGSEA scores of marker genes for each NK cluster based on the TCGA cohort. The results showed that all five NKC clusters had higher scores in normal samples than in tumour samples (Fig. 4A). The LUAD samples from the TCGA dataset were divided into high and low NK score groups according to the optimal cut-off values analysed by the survminer R package. The samples in the NK_1, NK_2, NK_3, and NK_4 groups of the high-NK score group had a better prognosis than the low-NK score group, whereas NK_0 was not associated with the prognosis of LUAD (Fig. 4B–F). The above results suggest that the contribution of NK_0 clusters to the progression of LUAD is small, despite the differences in NK_0 enrichment in LUAD and normal samples.

Fig. 4.

Association of 5 NK clusters with prognosis of LUAD patients: A Comparison of 5 NK scores in cancerous and normal tissues; K-M curves of high and low NK score groups (B) NK_0 cluster; C NK_1 cluster; D NK_2 cluster; E NK_3 cluster; F NK_4 cluster. P < 0.01, ****P < 0.0001

Identification of NK cell-related core genes

We further constructed the risk signature by screening DEGs between tumour tissues and normal tissues. A total of 19,495 DEGs were obtained, of which 12,525 were up-regulated DEGs and 6,970 were down-regulated DEGs. among them, 725 genes showed significant correlation with prognostically relevant NK clusters. In addition, the prognostic value of each gene was assessed by one-way Cox regression analysis, in which 133 genes showed prognostic value (Fig. 5A, B), which in turn led to enrichment analysis (Fig. 5C), which showed that biological processes (BP) included et al.; cellular components (CC) were mainly enriched in et al.; molecular functions (MF) were mainly involved in et al.; and KEGG pathways were enriched in mainly associated with et al. pathways. (This need to write the discussion first, because I write the first few do not necessarily discuss the results, and so the discussion pick some results, and then I fill in this piece of content.). We further used Lasso Cox regression analysis to narrow down the number of genes, filtering out 13 genes with lambda value = 0.0311 (Fig. 5D, E). Finally, 13 genes, including ANOS1, ADGRE3, ADAMTS8, CCT3, CD79A, CX3CR1, DPEP2, FMO3, FBP1, OASL, RRAS, SELENOP, and SFTPC, were included in the risk signature after multifactorial Cox regression analysis using stepwise regression (Fig. 5F). The final 13-gene signature formula is as follows RiskScore=”-0.121*CD79A-0.056*DPEP2 + 0.094*OASL + 0.363*RRAS + 0.052*ADAMTS8-0.282*ADGRE3-0.039*ANOS1 + 0.155*CCT3 + 0.075*CX3CR1-0.145*FBP1-0.115*FMO3-0.087*SELENOP-0.021*SFTPC”. We calculated risk scores for each sample and classified them into high and low risk groups after z-mean normalisation.

Fig. 5.

Identification of pivotal predictive genes to construct risk profiles. A Volcano plot of genes differentially expressed in cancer and normal tissues in the TCGA cohort; B Volcano plot of prognostic-related genes by one-way Cox regression analysis; C Bar chart of prognostic gene enrichment analysis; D Lasso trajectory plot; E coefficient distribution plot; F Cox coefficients of prognostic genes

The AUC values for 1- to 5-year survival for this model ranged from 0.66 to 0.71 in the TCGA cohort, 0.64 to 0.72 in the GSE3141 cohort,0.65 to 0.69 in the GSE31210 cohort, 0.57 to 0.71 in the GSE37745 cohort, 0.70 to 0.77 in the GSE50081 cohort, 0.60 to 0.68 in the GSE68465 cohort.Kaplan-Meier survival analyses demonstrated a similarity to that in the TCGA cohort and the GEO cohort. 0.70 to 0.77 in the GSE68465 cohort, and ranging from 0.60 to 0.68 in the GSE68465 cohort.Kaplan-Meier survival analyses showed significantly worse survival outcomes in high-risk patients compared with low-risk patients in the TCGA cohort and the GEO cohort. (Fig. 6A–F).

Fig. 6.

The cohort risk model ROC curve and K-M curve. A TCGA; B GSE3141; C GSE31210; D GSE37745; E GSE50081; F GSE68465

Mutation and pathway analysis of key genes

Next, we examined SNV mutations in 13 genes in the risk signature. The results showed that all genes except RRAS had SNV mutations in more samples (Fig. 7A). We analysed the probability of co-occurrence of these key genes and the 10 most mutated genes. As shown in Fig. 7B, the probability of mutation co-occurrence was not significant in these 12 genes, but the probability of mutation co-occurrence was significant in ADGRE3 and TTN, CX3CR1 and MUC16, RYR2 and ZFHX4, ANOS1 and LRP1B, DPEP2 and CSMD3, and RYR2 and ZFHX4. We also used OncogenicPathways to count known oncogenic signalling pathways in the TCGA cohort (Fig. 7C).

Fig. 7.

Risk gene-associated pathways (A) SNV mutation map of key genes; B mutation cosegregation probability map; C Fraction of pathway and samples affected ; D molecular characterisation map of key genes; E gene-pathway correlation heatmap; F heatmap of enriched scores of key pathways. *P < 0.05, **P < 0.01, ***P < 0.001

To further elucidate the association between risk genes and LUAD, we analysed the correlations between these genes and several molecular features of LUAD. The results showed that most of the genes were significantly negatively associated with aneuploidy scores, homologous recombination defects, score alterations, fragment counts and non-silencing mutation rates, while only CCTS was significantly positively associated with aneuploidy scores, homologous recombination defects, score alterations, fragment counts and non-silencing mutation rates (Fig. 7D). In addition, we analysed the potential pathways associated with each risk gene. As shown in Fig. 7E, F and a total of 21 pathways were significantly associated with these 13 genes, including the B cell receptor signalling pathway, the chemokine signalling pathway and leukocyte migration.

Analysis of key genes in relation to immunity

The results of our data showed that CCT3 was significantly negatively correlated with the matrix, immune and estimated scores, while the remaining 12 genes were significantly positively correlated with the matrix score, immune score and estimated score (Fig. 8A). After grouping each gene according to its median expression, we compared the three scores in different expression groups. The results showed that for the ANOS1 gene, the three scores in the high-expression group were significantly lower than those in the low-expression group, whereas the three scores in the high-expression group for the remaining genes were significantly higher than those in the low-expression group (Fig. 8B). Correlation analysis showed that the 13 key genes screened were significantly correlated with a variety of immune cells, including B cells, T cells, and NK cells (Fig. 8C). In addition, we also observed significant differences between high and low expression groups of risk genes in several immune cells (Fig. 8D).

Fig. 8.

Key gene immune relationship analysis (A) Heatmap of key gene immune infiltration algorithm (B) Key gene immune score box plot (C) Heatmap of key immune cell relationship (D) Heatmap of key gene immune cell correlation analysis

Sensitivity analysis of risk signature for PD-L1 Blockade immunotherapy

T-cell immunotherapy has emerged as a cancer treatment with synergistic survival benefits. Therefore, we evaluated the prognostic value of risk characteristics for immune checkpoint therapy in the IMvigor210 and GSE78220 cohorts.There were 348 patients in the IMvigor210 cohort with varying degrees of response to anti-PD-L1 receptor blockers, including complete remission (CR), partial remission (PR), stable disease (SD) and progressive disease (PD). The results showed that in the IMvigor210 cohort, patients in the low-risk group demonstrated a significant clinical benefit and significantly longer overall survival compared to the high-risk group (Fig. 9, p = 0.0017). Risk scores were higher in patients with SD/PD than in patients with CR/PR (Fig. 9B). The percentage of SD/PD was higher in the high-risk group than in the low-risk group (Fig. 9C). In addition, there was a significant difference in survival between stage I + II patients (Fig. 9D, p = 0.041) and stage III + IV patients (Fig. 9E, p = 0.018) in the different risk groups, indicating that the risk scores were sensitive in patients with different LUAD stages. Meanwhile, in the GSE78220 cohort, we also found that patients in the low-risk group had significantly longer overall survival than those in the high-risk group (Fig. 9F, p = 0.0067) Patients with PD had higher risk scores than those with CR/PR (Fig. 9G), and the percentage of PD in the high-risk group was higher than that in the low-risk group.

Fig. 9.

Risk score responsiveness to PD-L1 blockade immunotherapy in the IMvigor210 cohort. A Prognostic differences between risk score groups in the IMvigor210 cohort; B Differences in risk scores for response to immunotherapy in the IMvigor210 cohort; C Distribution of response to immunotherapy in risk score groups in the IMvigor210 cohort; D Prognostic differences between risk score groups for early patients in the IMvigor210 cohort; E Prognostic difference between risk scoring groups of late-stage patients in the IMvigor210 cohort; F Prognostic difference between risk scoring groups in the GSE78220 cohort GSE78220 cohort; G Difference in risk scoring of immunotherapy responses in the GSE78220 cohort; (H) Distribution of immunotherapy responses in risk scoring groups in the GSE78220 cohort. p < 0.0001

Independent risk factor identification and histogram construction

To optimise the predictive performance of risk features, we integrated clinicopathological features and risk scores by univariate and multivariate COX regression analyses. Multivariate COX regression analysis showed that risk signature (HR = 2.69, 95% CI: 1.93–3.74, P < 0.001) and metastatic status (HR = 2.06, 95% CI: 1.39–3.05, P = 0.002) were the most important independent prognostic factors for LUAD. Therefore, we constructed a histogram combining staging and risk scores (Fig. 10A–C). The calibration curve showed that the histogram was effective in predicting the actual survival outcome (Fig. 10D). In addition, DCA showed that the discriminative power of the histogram in identifying high-risk patients was superior to that of the risk score and staging, as shown in Fig. 10E. TimeROC analysis showed that the AUC of both the risk score and the histogram was higher than that of other metrics in the TCGA cohort (Fig. 10F).

Fig. 10.

Column line graphs predicting prognosis for LUAD (A, B) Univariate and multivariate Cox analysis of risk score and clinicopathological characteristics; C Nomogram of predictive models combining risk score and staging; D 1-, 3-, and 5-year calibration curves of the Nomogram; E Decision curves of the Nomogram; F analysis using Time-ROC Comparison of clinicopathological characteristics and predictive ability of Nomogram. ***P < 0.001

Discussion

NK cells regulate tumor proliferation, angiogenesis and the immune system of LUAD [31]. Studies on the pathogenesis of NK cells in LUAD have dug into the genetic-molecular area [32]. Nevertheless, the characterization of NK cells in LUAD remains unclear, research on prognostic prediction of LUAD based on NK cells is comparatively limited. Our study focused on the diversity of NK cells and performed a systematic characterization and classification of NK cells in LUAD using scRNA-seq data. Eventually, we eventually identified five NK cell clusters with diverse features that might regulate biology of the LUAD. Recent studies have demonstrated the prognostic significance of the NK cell-related gene signature in the assessment of LUAD outcomes [33, 34]. Consistently, our findings revealed four distinct clusters exhibiting strong associations with LUAD prognosis, as quantified by a composite score derived from DEGs encompassing all five clusters. Moreover, our results suggested that differences in the Notch signaling pathway among the identified clusters might underlie the prognostic significance of NK cells in LUAD. Notch signaling pathway plays a vital role in regulating the progression of LUAD [35]. Suppressing the Notch signaling pathway has demonstrated efficacy in attenuating collagen synthesis and curtailing cellular invasiveness [36]. Notably, high levels of Notch-related genes have been proven to be correlated with poor survival outcomes in LUAD patients [37].

Considering the prognostic relevance of the four NK cell clusters, we constructed a 13-gene risk signature. This signature comprised nine protective genes(ANOS1, ADGRE3, CD79A, CX3CR1, DPEP2, FMO3, FBP1, SELENOP and SFTPC) and four risk genes(ADAMTS8, CCT3, RRAS and OAST). In our study, SNV mutations were detected in ANOS1, ADGRE3, ADAMTS8, CCT3, CD79A, CX3CR1, DPEP2, FMO3, FBP1, OASL, SELENOP, and SFTPC, while no statistically significant co-occurrence patterns were observed among these mutations. SNV mutations hold a significant position in the pathogenesis of LUAD, as they can manifest as “driver mutations” that bestow tumor cells with growth advantages, thereby fueling the onset and progression of the disease [38, 39]. While no prior studies have revealed a connection between SNV mutations in these risk genes and LUAD, our findings indicated such mutations might contribute to the progression of LUAD. We further found that the 13 genes were significantly correlated with 21 pathways. Part of these pathways were proven to be associated with progression and prognosis of LUAD. Vascular smooth muscle contraction, RNA polymerase, pyrimidine metabolism and B cell receptor signaling pathway act as effective predictors for the prognosis and therapeutic responses of LUAD patients [40–42]. Spliceosome-related and lysosome-related genes are suggested to increase the risk of LUAD [43, 44]. Homologous recombination pathway is up-regulated in LUAD, and high expression of related-genes is correlated with worse overall survival [45].

Recent studies have suggested that the interplay between NK cells and the tumor immune microenvironment (TIME) could contribute to tumor progression [46, 47]. In our study, 9 protective genes and three risk genes were positively and significantly correlated with the immune score, whereas one risk gene exhibited a negative association. These findings suggested potential interactions between these genes and the TIME in LUAD, hinting at their potential as therapeutic targets. The TIME is constituted by a diverse array of immune cells within the tumor islets, collectively dictating the antitumor immunological milieu of TIME [48]. NK cells can engage in complex interactions with these immune cells, fostering an immunosuppressive TIME that allows tumor cells to evade immune surveillance [49]. In the risk signature, most protective genes were negatively correlated with NK cell activation and T cell activation. Meanwhile, regulatory T cells were negatively associated with protective genes and positively associated with risk genes. It was reported that NK cells could modulate T cell apoptosis in tumors [50]. T cells also could interact with NK cells and enhance their viability [51]. These suggested that NK related-genes acted not only on NK cells but also on T cells, therefore regulating the whole TIME.

Despite the fact that many patients have innate or acquired resistance to immunotherapeutic therapy [52], our findings showed that the risk signature efficiently stratified patients based on their likelihood of benefiting from immunotherapies. Additionally, previous studies have reveal that NK cells attract considerable interests for therapy of solid tumors [53, 54]. However, the therapy was less effective for certain patients, which might be attributed to the extremely complex immunosuppressive microenvironment [55]. Fortunately, our findings demonstrated that the NK cell-based signature had the capacity to predict the response to anti-PD-L1 immunotherapy. These findings might provide potential solutions to the issues, and improve the immunotherapy efficacy of NK cells. Nevertheless, further experimental research is warranted to elucidate the mechanisms underlying NK cell-TIME interactions in LUAD and their potential therapeutic implications in LUAD immunotherapy.

Our study has several limitations that should be noted. Firstly, the retrospective use of public data to establish NK cell clusters and the NK cell-based risk signature highlights the need for validation in prospective, multi-center LUAD cohorts in future investigations. Secondly, while a predictive nomogram was developed, further research is necessary to better understand the specific biological relationship between the included NK cell-related genes and clinicopathological characteristics. Finally, we will further conduct experimental verification on relevant research content in the future to improve our research results.

Looking ahead, The NK cell features constructed in this study can serve as a core module for integrated analysis with broader tumor biology information. For example, drawing on the efficient analysis framework of the cloud based GWAS platform [56] can accelerate the validation of the universality of this feature in multiple crowd queues. In addition, the application of cutting-edge computational methods such as dynamic network biomarkers [57] is expected to dynamically capture key network changes in NK cell characteristics during LUAD treatment response or resistance in future longitudinal sample data, thereby revealing their dynamic characteristics as predictive biomarkers.

Conclusion

In summary, our study comprehensively characterized NK cell populations in LUAD, identifying five distinct clusters with different characteristics. The DEGs in the five clusters were abundant in vascular smooth muscle contraction, RNA polymerase, pyrimidine metabolism, the B cell receptor signaling pathway, etc. Four of the clusters showed significant correlations with LUAD prognosis and were applied to create an NK cell-based prognostic risk signature consisting of 13 genes. The NK cell-based gene signature appeared to be associated with the immunological landscape and might be utilized to predict response to anti-PD-L1 immunotherapy. Finally, a novel nomogram incorporating the risk signature and clinicopathological parameters was constructed, and it performed well in predicting the clinical prognosis of LUAD patients.

Author contributions

Yikun Guo organized the data and wrote the manuscript for this article, contributing 4%; Daowen Yang and Chuanxi Tian made revisions to the article, contributing 30% and 30% respectively; All authors have read and approved the final manuscript. The above three are all authors of this article.

Funding

None.

Data availability

This study is a retrospective study. All data in the manuscript are based on public data for secondary data analysis, and do not require ethical approval, informed consent, or other related documents. All data can be accessed on the following websiteGEO database: ((https://www.ncbi.nlm.nih.gov/geo/))IMvigor210 cohort (http://research-pub.gene.com/IMvigor210CoreBiologies)Clinical trial number: not applicable.

Declarations

Ethics approval and consent to participate

Not applicable. This study did not generate any new data. It is a secondary analysis of relevant data from large public databases such as GEO database, IMvigor210 cohort, and therefore does not require ethical approval and related documents. All patient information in the database is anonymous, so informed consent is not required.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.