Targeting NLRC4 for potential therapeutic strategies in lung adenocarcinoma

Abstract

Background

Autophagy, a vital cellular process, plays a significant role in the development of a spectrum of diseases, notably cancer. The objective of this study was to assess the prognostic significance and explore the possible roles of autophagy-related genes (ARGs) in lung adenocarcinoma (LUAD) and in 33 cancer types.

Methods

In this study, ARGs were sourced from the Human Autophagy Database (HADb), with gene expression data retrieved from The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) database. The LASSO Cox and multivariate Cox regression models were employed to identify ARGs with prognostic significance, leading to the development of the Autophagy-Related Gene Prognostic Signature (ARGPS), which was compared with previously established prognostic models. The associations between the ARGPS and clinical parameters were examined to identify independent prognostic factors. Additionally, a pan-cancer analysis underscored the role of the ARGPS in immune subtyping, the tumor immune context, survival outcomes, stemness scores, and sensitivity to antitumor drugs. Finally, a virtual drug screening was performed to predict potential target interactions.

Results

In the GSE116959 dataset, we identified 14 DEARGs with statistical significance (p < 0.05 and |logFC|> 1). Using LASSO Cox and multivariate Cox regression, we developed an independent prognostic signature, identifying seven ARGs to form the ARGPS. LUAD patients were stratified into low-risk and high-risk groups. Pan-cancer analysis highlighted significant heterogeneity in ARGPS expression among various cancers. ARGPS members were significantly correlated with immune infiltration, drug resistance, and stemness in tumors. Virtual screening identified five potential NLRC4-targeting drugs for LUAD treatment.

Conclusions

In this study, we developed a predictive risk model based on seven ARGs. We comprehensively examined ARGPS expression and its correlation with immune infiltration and the tumor microenvironment. These findings could inform targeted immunotherapy and chemotherapy for LUAD and other cancers. The low expression of NLRC4 in LUAD indicated its potential as a therapeutic target.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12672-025-02839-y.

Introduction

Lung cancer is the leading cause of cancer-related death world-wide and non-small cell lung cancer (NSCLC), one of the major subtypes, accounts for approximately 85% of all lung cancers [1, 2]. Lung adenocarcinoma (LUAD) is the most common histological subtype of NSCLC, accounting for more than 40% of all lung cancers [3]. Despite considerable efforts in the diagnosis and treatment of LUAD, the overall five-year survival rate is still relatively low at 17% [4, 5]. The poor prognosis is largely due to the lack of reliable biomarkers that can predict the prognosis of patients early. Therefore, the discovery of new molecular biomarkers that can improve the prognosis and treatment strategies of LUAD patients is urgently needed.

Autophagy is an important catabolic process that plays an important role in maintaining cell homeostasis [6–9]. It has been reported to be involved in many biological and pathological processes, including pathogenic inflammation, neurodegenerative diseases, aging and cancer [9, 10]. The level of autophagy proteins plays an important role in regulating tumor cell growth and progression, inhibiting or promoting tumor development. In the early stages of cancer, autophagy usually functions as a tumor suppressor. However, once cancer has developed, autophagy promotes tumor progression [11]. Therefore, exploring appropriate molecular biomarkers based on autophagy can not only aid in predicting the prognosis of the disease, but also in providing a theoretical basis for clinicians to carry out autophagy therapy for LUAD.

Many studies have shown that autophagy palys an important role in the progression of NSCLC, with an important role [12, 13].A study of NSCLC patients with stage I/II disease revealed that those with higher levels of the autophagy marker LC3 had better prognosis [14]. In LUAD, the expression of Beclin-1 is negatively correlated with tumor size and stage [15]. In addition, the Beclin-1 coding gene BECN1 has been confirmed to play a carcinogenic role in NSCLC by regulating vimentin ubiquitination. These results show that the upregulation of BECN1 promotes cell migration, but that its downregulation significantly inhibits cell migration [16]. In one study, a risk model was constructed based on the expression levels of five autophagy related genes, allowing for the prognosis prediction and serving as a prognostic biomarker for LUAD [17]. The above results confirm the role of autophagy in lung cancer, suggesting that ARGs can be used as prognostic biomarkers. Although the application of autophagy in the treatment of lung cancer is controversial, the combination of autophagy inhibitors and chemotherapy seems promising. In general, autophagy appears to play an important role in lung cancer. Although some differences exist among previous studies, autophagy is an interesting research direction for the treatment of lung cancer.

In this study, we established a scoring system called the autophagy-related gene prognostic signature (ARGPS) through PCA and LASSO Cox regression analysis. Combined with the risk score, we established and verified the prognostic nomogram. Finally, we analyzed the tumor microenvironment, immune subtypes and drug sensitivity of autophagy related genes in 33 cancers. Relationships among the ARGPS, tumor stem cell score and immune subtypes were further analyzed, offering a new way to reveal new mechanisms and therapeutic targets of autophagy related to LUAD.

Materials and methods

Data acquisition

A total of 232 autophagy-related genes (ARGs) were obtained from the Human Autophagy Database (HADb; http://www.autophagy.lu/clustering/index.html) [18]. GSE116959 data were obtained from the GEO database (https://www.ncbi.nlm.nih.gov/geo/). Differentially expressed ARGs were obtained from among differentially expressed genes and ARGs by using a Venn diagram. Gene expression data and clinical feature data of LUAD patients were obtained from TCGA database (https://portal.gdc.cancer.gov/).

Functional enrichment analysis and construction of the PPI network

Gene Ontology (GO) annotation and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis were performed via the"ClusterProfiler"software package, and the results were visualized by the"ggplot2"software package in R software [19]. We used the Protein Interaction Network Analysis (PINA) platform to build a PPI network, and filtered and analyzed the network through built-in tools (Filter and Cancer Context) to obtain biological and functional insight [20]. Moreover, we performed pathway analysis on the genes using the Reactome database to reveal their potential roles in LUAD [21].

Consensus cluster analysis based on 14 DEARGs

The differential expression of 14 autophagy related genes in LUAD was identified by using the"limma"software package in R software. Gene expression levels and their correlations with clinicopathological features were visualized by the"pheatmap"and “vioplot” software packages. The"corrplot"package was used to reveal the correlations among autophagy related genes. The Kaplan–Meier method was used to analyze overall survival in different clusters. The pheatmap R software package was employed for clinical correlation analysis.

Construction and verification of the risk prediction model for DEARGs

To obtain a more practical model, we used LASSO regression to screen the most powerful prognostic genes with regression coefficients, constructed an autophagy-related gene prognostic signature (ARGPS), and calculated the risk score. Univariate Cox regression was applied to analyze the clinical characteristics and risk score related to OS. Multivariate Cox regression analysis revealed its independent prognostic value. We considered p < 0.05 as significant for all comparisons. To further explore the potential application value of the ARGPS in LUAD, especially its association with immunotherapy, we utilized the GEPIA database to analyze the correlation between the ARGPS and key immune checkpoint biomarkers (PD-1 and CTLA-4) [22].

Three LUAD cell lines (A549, H1299, and H1975) and normal bronchial epithelial cells (16HBE) were purchased from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China). Total RNA was extracted from 16HBE, A549, H1299, and H1975 cells using TRIzol reagent (Invitrogen). After the purity and concentration of the total RNA were determined, it was reverse transcribed to cDNA using a PrimeScript RT Kit, which was used for qRT-PCR with SYBR Green Premix Ex Taq II. Finally, the relative mRNA expression levels of the 7 DEARGs were calculated by using 2^−ΔΔCT. The sequences of primers used were synthesized by BioSune Biotech (Shanghai, China) (Table 1).

Table 1.

Primer sequences for qRT-PCR analysis of 7 ARGs

Primer name	Primer sequence (5ʹ → 3ʹ)
ATIC-F	CGGCCAGCTCGCCTTATTT
ATIC-R	AGAGCTTTTGCAGTCCCTCC
BIRC-F	TTTCTCAAGGACCACCGCATC
BIRC-R	CAAGTCTGGCTCGTTCTCAG
DAPK2-F	GACTTTGGTCTGGCTCACGA
DAPK2-R	GATGACGCCTATGCTCCACA
DLC1-F	CTCACTCTGGAAGCACTCGG
DLC1-R	GCTCCGAAGTGGAGTAGCTG
DRAM1-F	CTCGACATGCCACATACGGA
DRAM1-R	TCACAGATCGCACTCACTACG
ITGB4-F	AGTGTGTCCGTGTGGATAAGG
ITGB4-R	GTGGTGTCAATCTGGGTCTCC
NLRC4-F	CTGACATTGGAGAGGGAATGGA
NLRC4-R	CAGACAAGCAGCAGGAGACTA
GAPDH-F	GCACCGTCAAGGCTGAGAAC
GAPDH-R	TGGTGAAGACGCCAGTGGA

F = forward, R = reverse

Comparison of prognostic models

To verify the accuracy of our predictive model, we conducted a comparative analysis with four alternative models. To compare these five models, we utilized Kaplan–Meier survival curves. Additionally, we employed the “survcomp” R package to calculate the concordance index (C-index) and the restricted mean survival time (RMS) for each model. The concordance index (C-index) was a metric used to assess the predictive accuracy of a model, with values ranging from 0.5 to 1.0. A higher value indicated a stronger predictive ability of the model [23].

TCGA pan-cancer data acquisition and differential expression analysis

The TCGA pan-cancer data were downloaded from the UCSC Xena browser (https://xenabrowser.net) [24]. By using the transcriptional data of 33 TCGA tumors, we generated a box plot of the expression levels of 7 DEARGs (ATIC, BIRC5, DAPK2, DLC1, DRAM1, ITGB4 and NLRC4). The linear mixed effect model was subsequently used to statistically compare gene expression between normal and tumor tissues for 18 cancer types (more than five adjacent normal tissues) [25]. Finally, we analyzed the differential expression levels of seven differentially expressed autophagy related genes in 18 tumors (BLCA, BRCA, CHOL, COAD, ESCA, GBM, HNSC, KICH, KIRC, KIRP, LIHC, LUAD, LUSC, PRAD, READ, STAD, THCA and UCEC). A P value less than 0.05 was considered statistically significant.

Immune subtype, tumor microenvironment and drug sensitivity analysisof 33 TCGA pan-cancers and LUAD datasets

The “limma”, “ggplot2”, and “reshape2” R packages were used to perform immune subtype analysis of the 7 DEARGs. We used the"estimate"R package and"limma"R package to obtain the immune score, stromal score and estimate score of 33 TCGA cancer samples [26]. We defined six immune subtypes to detect immune infiltration in the tumor environment. For the immune subtypes obtained from the TCGA pan-cancer data was used ANOVA models to detect the correlation between DEARG expression and immune infiltration type in the tumor microenvironment. The tumor stemness characteristics extracted from the transcriptome and epigenetics of TCGA cancer samples were used to measure the stem cell-like characteristics of tumor cells. We conducted the Wilcoxon test for differential analysis of age, fustat and gender. A p-value < 0.05 was considered statistically significant.

Virtual screening of potential drugs targeting NLRC4

The crystal structure of the NLRC4 protein was obtained from the Protein Data Bank's online database (https://www.rcsb.org/). The Protein Preparation module of the Schrodinger software suite was then employed to meticulously remove water molecules and any extraneous ions, restore any absent side chains and loop regions, and execute energy minimization to refine the structure. The Sitemaps module within the Schrodinger suite was adeptly utilized to delineate potential binding sites on the NLRC4 protein. The Glide Grid module was subsequently engaged to fabricate the necessary docking grid files for further analysis.

A repository of small molecule compounds from Traditional Chinese Medicine (TCM), curated by the Laboratory of Medicinal Chemistry and Natural Product Chemistry at Nankai University (http://cadd.pharmacy.nankai.edu.cn/yatcm/home), was accessed and downloaded [27]. The LigPrep module of Schrodinger was meticulously applied to preprocess the molecules within this library, encompassing protonation, desalting, hydrogenation, and the generation of a multitude of tautomers and stereoisomers, alongside conformational sampling complemented by energy minimization. The QuickProp module and Glide module of the Schrodinger suite were strategically deployed for ADMET property screening, high-throughput virtual screening (HTVS), standard precision screening (SP), and extra precision screening (XP). In conclusion, the quintet of potential drugs with the most distinguished scores were culled from the screening results, and a comprehensive analysis of their binding modes was subsequently conducted.

Results

Identification of differentially expressed ARGs and functional enrichment analysis

We analyzed and obtained differentially expressed genes in GSE116959 (including 57 LUAD samples and 11 peritumoral normal lung tissue samples) via the"limma"software package in R software (R 4.1.0). Figure 1a displays the volcano plot, whereas Fig. 1b presents the heatmap of DEARGs between normal and tumor tissues. We identified a total of 14 DEARGs from among the 232 autophagy genes (Table S1) using the Venn Diagram package. This set included 6 up-regulated genes (ATIC, BIRC5, ERO1L, GAPDH, CDKN2 A, and ITGB4) and 8 down-regulated genes (NLRC4, DAPK2, DLC1, CASP1, DRAM1, PPP1R15 A, FKBP1B, and FOS) (Fig. 1c, d).

Fig. 1.

Differentially expressed genes in GSE116959. a In the volcano map, red dots represent the up-regulated DEGs, green dots represent the down-regulated DEGs, and the black dots represent the DEGs whose expression was not differentially expressed. b Heatmap showing the differentially expressed genes between LUAD and normal lung tissues. Red represents high expression and green represents low expression. c Up-regulated autophagy related genes (ARGs) in GSE116959. d Down-regulated autophagy related genes (ARGs) in GSE116959

In addition, to explore the biological interpretation of these DEARGs, we performed functional enrichment pathway analysis. According to the results of the GO enrichment analysis, these DEARGs are mainly involved mainly in autophagy, cell-substrate adhesion and apoptosis, which are related to biological process (BP), molecular function (MF) and cell composition (CC) (Fig. 2a, Table S2). KEGG enrichment analysis showed that these genes are mainly involved mainly in infection, cancer and platinum drug resistance (Fig. 2b). The construction of the gene-pathway network was based on important enrichment pathways and genes regulating these pathways, as shown in Fig. 2c.

Fig. 2.

Functional enrichment analysis and PPI network of differentially expressed autophagy genes (DEARGs). a Gene ontology terms of DEARGs. b KEGG pathway enrichment of DEARGs. c Gene-pathway network of DEARGs and KEGG pathways. d The whole PPI network of the DEARGs in Protein Interaction Network Analysis (PINA). e A subnetwork obtained from d was screened by the filter-direct interaction condition. f The PPI network obtained from d was screened by cancer context condition. g Pathway analysis of DEARGs through the Reactome database. BP biological process, CC cellular assembly, MF molecular function. The color represents the p value and the spot size represents the number of genes

A protein–protein interaction (PPI) network of 14 DEARGs was constructed via the PINA database (Fig. 2d), and the annotation of the PPI network with direct interactions was performed with “Filter” (Fig. 2e). By using the LUAD parameter in Caner-Context, we obtained the PPI network shown in (Fig. 2f). Candidate prognostic biomarkers were represented as orange (poor prognosis) and green (good prognosis) nodes. The blue edge indicated a positive correlation and statistical significance between the expression levels of interacting proteins, and the red edge indicated a negative correlation. The edge width was directly proportional to the correlation coefficient. Pathway analysis of DEARGs through the Reactome database revealed that their potential roles in LUAD mainly involve the following pathways: Immune System, Signal Transduction, Cell Cycle, Programmed Cell Death, Gene Expression (Transcription), Disease, Metabolism of Proteins, and Cellular Responses to Stimuli (Fig. 2g).

Expression and correlation of DEARGs and consensus clustering analysis in LUAD

The mRNA expression levels of DEARGs were analyzed using transcriptome data in the TCGA-LUAD database. A heatmap was generated to show differential expression of 14 DEARGs between LUAD and normal tissues (Fig. 3a), in which ERO1L was not expressed. Compared with those in normal tissues, the mRNA expression levels of 5 genes (ATIC, BIRC5, CDKN2 A, GAPDH, and ITGB4) were significantly increased in LUAD tissues (p < 0.001) (Fig. 3b). To better understand the relationships among the 13 DEARGs, we analyzed their mRNA expression using Pearson correlation analysis (Fig. 3c). This analysis revealed a positive correlation among most DEARGs, particularly a significant correlation between DLC1 and DAPK2 (r = 0.69).

Fig. 3.

Differential expression, correlation analysis and consensus cluster analysis of DEARGs in TCGA-LUAD cohort. a Heatmap demonstrating the differential expression of DEARGs between LUAD and normal lung tissues. Red indicates high expression and green indicates low expression. b The vioplot shows the differential expression of DEARGs between LUAD and normal lung tissue. Red indicates high expression and blue indicates low expression. c Spearman correlation analysis of DEARGs. d Cumulative distribution function (CDF) curves of consensus score (k = 2–9). e Using unsupervised consistent clustering algorithm to classify LUAD patients based on DEARGs mRNAs with two optimal clusters (k = 2). f Relative change in the area under the CDF curve (k = 2–9). ***p < 0.001

We then used the consensus clustering method to divide the LUAD cohort into several clusters according to the consensus of DEARG mRNA expression by using the"ConsumusClusterPlus"package. When the cluster index"K"increases from 2–9, k = 2 was proven to have the greatest advantage in obtaining the largest difference between clusters (Fig. 3d, e); when k = 2 the minimum interference between clusters was minimal (Fig. 3f). Therefore, the LUAD cohort was divided into two clusters (Cluster 1 and Cluster 2).

To verify the effect of this division, we conducted principal component analysis (PCA) and found significant separation between the two clusters, indicating that the clustering was feasible (Fig. 4a). We used Kaplan–Meier survival analysis to compare the overall survival of patients in the two clusters and found that the survival outcome of patients in Cluster 1 was significantly better than that of Cluster 2 (Fig. 4b). Then we compared the clinicopathological features between the two clusters. As shown in the heatmap, Cluster 1 was associated with earlier N phase and TNM stage, FEMALE and alive (Fig. 4c). Next, we examined the prognostic role of the DEARGs in LUAD and performed univariate Cox regression analysis on the expression level in the LUAD dataset (Fig. 4d, Table 2). The results showed that 8 of the 14 tested genes were significantly correlated with OS (p < 0.05). ITGB4, ATIC, GAPDH and BIRC5 were high-risk genes with HRs > 1, whereas DLC1, NLRC4, DAPK2 and DRAM1 were protective genes with HRs < 1.

Fig. 4.

Principal Component Analysis (PCA), survival analysis and clinical correlation analysis. a Principal component analysis of autophagy-related genes in patients with LUAD. b Kaplan–Meier (K–M) analysis in Cluster 1 and Cluster 2. c The expression profiles of 7 ARGs and their correlations with clinicopathological features. d Univariate Cox regression analysis was used to screen prognostic genes. *p < 0.05, **p < 0.01

Table 2.

Univariate Cox analysis of 13 DEARGs

id	HR	HR.95L	HR.95H	p-value
DLC1	0.964996291	0.939416159	0.991272964	0.009337813
PPP1R15 A	0.999010509	0.989946293	1.00815772	0.831419773
CASP1	0.981225111	0.946826738	1.016873182	0.297888181
ITGB4	1.005894702	1.002292225	1.009510127	0.001323924
ATIC	1.020375994	1.005418131	1.035556388	0.007425877
NLRC4	0.735024223	0.588738252	0.917658412	0.006550351
GAPDH	1.000715933	1.000454212	1.000977723	8.20E-08
DAPK2	0.827518675	0.719156491	0.952208826	0.00819759
DRAM1	0.993408151	0.987727012	0.999121966	0.023811643
FOS	0.998981578	0.997540284	1.000424955	0.166600213
BIRC5	1.019936405	1.006490238	1.033562205	0.00355239
FKBP1B	0.990692661	0.936903536	1.047569906	0.742678853
CDKN2 A	1.005796967	0.99497273	1.01673896	0.295086215

In addition, LASSO Cox regression was then used to avoid overfitting problems in risk characteristics. When the optimal lambda value was reached, seven key autophagy related genes (DLC1, ITGB4, ATIC, NLRC4, DAPK2, DRAM1 and BIRC5) were retained (Fig. 5a, b, Table 3). The risk score of each patient was calculated according to the formula, and all LUAD patients were divided into a low-risk group and a high-risk group. The results of survival analysis showed a difference between the groups, with the 5-year survival rate of the low-risk group being higher (p = 3.491e−03) (Fig. 5c). The risk score plot was shown in Fig. 5d.

Fig. 5.

Construction and validation of the autophagy-related gene prognostic signature (ARGPS). a Partial likelihood deviation with corresponding log(λ) values at the minimum deviation of the LASSO model. b The LASSO coefficient of 7 prognostic DEARGs by tenfold cross-validation. c Kaplan–Meier analysis of OS between high-risk and low-risk groups of LUAD patients based on the risk scores of the ARGPS. d Visualization of the risk score plot of the high-risk group and low-risk groups in LUAD patients

Table 3.

Univariate and multivariate Cox regression analysis of autophagy-related gene prognostic signature (ARGPS)

ID	Univariate Cox regression analysis				Multivariate Cox regression analysis
	HR	HR.95L	HR.95H	p value	HR	HR.95L	HR.95H	p value
Age	1.010	0.994	1.026	0.233	1.019	1.003	1.035	0.023
Gender	1.044	0.762	1.430	0.789	0.887	0.642	1.225	0.466
Stage	1.667	1.438	1.931	1.12E−11	1.474	1.184	1.834	0.001
T	1.526	1.251	1.861	3.12E−05	1.157	0.929	1.441	0.194
M	0.923	0.769	1.107	0.386	0.987	0.811	1.200	0.894
N	1.725	1.445	2.060	1.82E−09	1.207	0.945	1.540	0.131
Riskscore	2.313	1.755	3.049	2.64E−09	2.331	1.735	3.130	1.89E-08

Clinical significance, prognostic value and experimental validation of the seven-gene prognostic signature

A heatmap was drawn to show the differences between clinicopathological features and the mRNA expression levels of the seven prognostic genes in the high-risk and low-risk groups. The high-risk group was significantly associated with advanced T stage (p < 0.001), N stage (p < 0.001), TNM (p < 0.01), and fustat (p < 0.01), with male sex being more common (p < 0.05) (Fig. 6a). Next, univariate and multivariate Cox analyses were performed to determine independent prognostic predictors in LUAD. Univariate Cox analysis revealed that TNM stage (p < 0.001, HR = 1.667, 95% CI = 1.438–1.931), T stage (p < 0.001, HR = 1.525, 95% CI = 1.251–1.861), N stage (p < 0.001, HR = 1.725, 95% CI = 1.444–2.060) and risk score (p < 0.001, HR = 2.313, 95% CI = 1.755–3.049) were significantly correlated with OS (Fig. 6b). After these parameters were incorporated into the multivariate Cox regression model, only TNM stage (p < 0.001 HR = 1.474, 95% CI = 1.184–1.834) and risk score (p < 0.001, HR = 2.331, 95% CI = 1.735–3.130) were identified as independent prognostic factors (Fig. 6c).

Fig. 6.

Clinical significance and experimental verification of risk grouping. a The clinicopathological features of the high-risk group and low-risk group. b Univariate Cox analysis of the risk score and clinicopathological features. c Multivariate Cox analysis of the risk score and clinicopathological features to determine independent prognostic predictors in the LUAD patients. *p < 0.05, **p < 0.01, ***p < 0.001

Analysis through the GEPIA database revealed correlations between the ARGPS and key immune checkpoint biomarkers (such as PD-1 and CTLA-4). Specifically, the expression of DRAM1, DLC1, and DAPK2 was negatively correlated with CTLA-4 expression, whereas BIRC5 expression was positively correlated with CTLA-4 expression. Additionally, the expression of NLRC4 and BIRC5 was positively correlated with PD-L1 expression (Fig. 7).

Fig. 7.

The correlations between ARGPS and key immune checkpoint biomarkers (such as PD-1 and CTLA-4)(ATIC (a), BIRC5 (b), DAPK2 (c), DLC1 (d), DRAM1 (e), ITGB4 (f), and NLRC4 (g))

qRT-PCR was used to detect the mRNA expression of the seven autophagy related genes in the 16HBE, A549, H1299, and H1975 cell lines. As expected, the results for the seven genes were essentially consistent with the results reported above, in which ITGB4(A), BIRC5(C), and ATIC(B) were highly expressed, whereas NLRC4(E), DRAM1(F), DAPK2(G), and DLC1(D) were expressed at low levels (Fig. 8).

Fig. 8.

The 7 differentially expressed autophagy related genes in LUAD were verified via qRT-PCR. There were significant differences in the expression of 7 genes (ITGB4 (a), ATIC (b), BIRC5 (c), DLC1 (d), NLRC4 (e), DRAM1 (f), and DAPK2 (g)) between the 16HBE and A549 or H1975 cell lines (p < 0.050), whereas the expression of ITGB4 was no significantly different between the 16HBE and H1299 cell lines (p > 0.050). *p < 0.05, **p < 0.01, ***p < 0.001

Comparison of prognostic models

We conducted a thorough assessment of our model's predictive accuracy by generating time-dependent ROC curves and survival curves, and subsequently compared our model's performance with those of prior studies. Figure 9a–e illustrate the ROC curves and survival data for the five predictive models under consideration. As depicted in Fig. 8a, the AUC values for our model at 1, 3, and 5 years are 0.691, 0.649, and 0.572, respectively, which surpass those of models from previous research. Regarding OS analysis, the p-values for the models developed by Bian, Chen, and Liang are 0.160, 0.081, and 0.117, respectively, whereas our model exhibits results in a statistically significant distinction between high-risk and low-risk groups (p = 0.003). As demonstrated in Fig. 9f–g, our model has notable advantages in terms of both the C-index and the RMS error. Specifically, the C-index values for our model and the four previously published risk models are 0.641, 0.578, 0.623, 0.528, and 0.611, respectively.

Fig. 9.

Excellent predictive performance of the ARGPS model. a ROC and OS in the high- and low-risk groups according to the ARG signature. b ROC and OS in the high- and low-risk groups according to the Bian signature. c ROC and OS in the high- and low-risk groups according to the Chen signature. d ROC and OS in the high- and low-risk groups according to the Liang signature. e ROC and OS in the high- and low-risk groups according to the Zhang signature. f Comparison of C-index among risk models. g Observations of RMS among risk models

DEARGs expression in pan-cancer

To understand the internal expression patterns of the seven ARGs, we detected the expression levels of these genes in all 33 cancer types in the TCGA pan-cancer data (Table 4). Figure 10 showed the data for the seven autophagy-related genes in 18 TCGA tumors and normal tissues (more than 5 normal tissues were selected for pan-cancer expression analysis).

Table 4.

Seven autophage-related genes differential expression across 18 cancer types

Cancer	ITGB4	BIRC5	NLRC4	ATIC	DAPK2	DLC1	DRAM1
BLCA	0.000627	9.14E−09	0.461482	4.49E−05	0.053169	1.27E−07	0.832511598
BRCA	3.21E−14	1.22E−57	0.494234	1.16E−47	7.32E−33	6.13E−46	1.37E−06
CHOL	2.26E−09	2.26E−09	0.017317	2.26E−09	0.000332	0.038115	6.77E−08
COAD	0.000487	6.71E−21	1.86E−05	6.54E−23	4.77E−05	1.82E−06	1.00E−11
ESCA	0.001734	7.79E−08	0.060705	0.000169	0.186179	0.113364	0.674544347
GBM	0.105794	0.000155	0.001198	0.018256	0.04875	0.460228	0.00014406
HNSC	3.69E−18	2.54E−24	3.73E−07	4.89E−18	0.006703	0.296516	1.52E−09
KICH	4.50E−10	5.90E−05	0.026817	1.36E−09	0.563391	2.48E−10	3.38E−05
KIRC	0.002579	9.07E−34	9.22E−32	0.001142	0.282792	0.607538	9.36E−20
KIRP	7.00E−07	3.85E−16	2.31E−05	1.67E−09	3.01E−09	3.98E−16	1.18E−11
LIHC	1.85E−10	2.35E−28	0.052874	1.56E−23	1.19E−20	1.55E−10	1.11E−11
LUAD	9.76E−12	2.68E−32	1.32E−32	8.51E−34	5.82E−30	1.76E−30	4.75E−19
LUSC	4.50E−26	1.26E−30	7.77E−30	8.41E−27	5.16E−28	1.08E−30	2.81E−30
PRAD	3.76E−14	2.54E−20	0.384298	3.49E−05	0.447155	0.009781	4.23E−05
READ	0.001995	0.000185	0.053814	2.70E−07	0.165055	0.001607	0.001678706
STAD	3.13E−05	5.16E−13	5.44E−08	1.25E−16	0.715863	0.327452	3.17E−08
THCA	4.40E−18	5.86E−07	0.000155	3.51E−22	1.34E−16	9.59E−08	6.76E−18
UCEC	0.308201	1.90E−21	0.79143	9.83E−17	6.12E−05	2.98E−17	0.354524369

Fig. 10.

Boxplot of the 7 DEARGs between cancer and adjacent normal tissues (ATIC (a), BIRC5 (b), DAPK2 (c), DLC1 (d), DRAM1 (e), ITGB4 (f), and NLRC4 (g)). The blue boxplots indicate normal tissues. The red boxplots represent cancer tissues. *p < 0.05, **p < 0.01, ***p < 0.001

As shown in Fig. 10, the expression of ATIC (a), BIRC5 (b) and ITGB4 (f) in LUAD and adjacent normal tissues was significantly different, and with high expression in LUAD (p < 0.001). DAPK2 (c), DLC1 (d), DRAM1 (e) and NLRC4 (g) were highly expressed in normal tissues, and the difference was statistically significant (p < 0.001). Significant heterogeneity in the corresponding gene expression levels within and between tumors was observed for all seven autophagy-related genes (Fig. 11a). Compared with other autophagy related genes, NLRC4 and DAPK2 showed relatively low level of expression averaged in all cancer types, and ITGB4 and ATIC having higher expression (Fig. 11a). These findings suggested that there are inherent differences in the expression of ARGs between different tumor types and between different ARGs within each tumor type. In this study, we investigated the expression levels of all 7 genes in primary patient tumors of 18 cancer types with at least 5 pairs of adjacent normal samples (Fig. 11b). All seven ARGs exhibited significant expression differences in different tumor types, but the direction of change varied, and there were differences in each gene and each cancer type. Among the 18 tumors, ATIC, BIRC5 and ITGB4 were mainly up-regulated, whereas DAPK2, DLC1, DRAM1 and NLRC4, with a few exceptions, were mainly down-regulated. According to the Spearman correlation test, we found that DRAM1 and NLRC4 (r = 0.047, p < 0.05), BIRC5 and ATIC (r = 0.42, p < 0.05), and BIRC5 and DLC1 (r = −0.37, p < 0.05) had the highest correlation among all pairwise correlations of the seven genes, indicating that they may have some common features or functions (Fig. 11c).

Fig. 11.

Expression levels of 7 DEARGs in cancer and adjacent normal tissues. a Boxplot showing the distribution of DEARG gene expression in all 33 cancers. b Heatmap showing the difference in autophagy gene expression between primary tumors and adjacent normal tissues based on log2 (fold change) in 18 tumor types that more than 5 adjacent normal samples. c Correlation plot of gene expression among the 7 DEARGs

Pan-Cancer survival analysis of autophagy-related genes

High expression of BIRC5 and ITGB4 was associated with poorer prognosis than low expression of BIRC5 and ITGB4 in LUAD patients (p = 0.001) (Fig. 12a, d). In LUAD patients, the prognosis related to low DAPK2, DARM1 and NLRC4 expression was worse than that related to high DAPK2, DARM1 and NLRC4 expression (p < 0.05) (Fig. 12b, c, and e). In this study, 33 cancer types were used to study the relationship between autophagy related gene expression and overall survival, as analyzed by a univariate Cox proportional hazards regression model; the forest plots were shown in Fig. 12f. Specifically, high expression of AITC(Fig. 13a) and BIRC5 (Fig. 13b) was associated with poor prognosis in multiple tumor types, such as ACC, LIHC, and KIRC. Additionally, DAPK2 indicated poor prognosis in DLBC, GBM, and KIRC (Fig. 13c). In contrast, DLC1, ITGB4, DRAM1, and NLRC4 exhibited significant heterogeneity in their prognostic roles across different tumors.Elevated expression of DLC1 was primarily associated with a survival advantage, predicting better outcomes in patients with KIRC, UCEC, and UVM, but it was linked to poor prognosis in ACC, LGG, and MESO (Fig. 13d). DRAM1 was associated with poor prognosis in GBM, LUSC, TGCT, and UCEC, while it indicated a survival advantage in SKCM (Fig. 13e).ITGB4 predicts poor prognosis in LGG but was associated with a survival advantage in patients with BRCA, LAML, and SARC (Fig. 13f).NLRC4 indicated poor prognosis in LGG but was associated with a survival advantage in SKCM (Fig. 13g).

Fig. 12.

Survival analysis of 7 DEARGs in LUAD. a–e Survival analysis of LUAD patients. The red line indicates high expression and the blue line indicates low expression. f Forest plots showing the survival advantages and disadvantages of 7 DEARGs for different cancer types

Fig. 13.

Survival analysis of 7 DEARGs in pan-cancer except LUAD (ATIC (a), BIRC5 (b), DAPK2 (c), DLC1 (d), DRAM1 (e), ITGB4 (f), and NLRC4(g)). The red line indicates high expression and the blue line indicates low expression

Correlation analysis of ARGS expression with the immune response and tumor microenvironment and drug sensitivity in pan-cancer

To understand how each of ARGs was associated with immune components, we evaluated the correlation between autophagy related genes and immune infiltration in cancer. Six types of immune infiltration were found in human tumors, corresponding to tumor promotion and tumor inhibition [28]: C1 (Wound Healing), C2 (IFN-gamma Dominant), C3 (Inflammatory), C4 (Lymphocyte Depleted), C5 (Immunologically Quiet), and C6 (TGF-beta Dominant). We analyzed six immune infiltration subtypes were associated with the expression level of autophagy related genes in TCGA pan-cancer data (Fig. 14a). The correlations between higher levels of ITGB4, BIRC5 and ATIC and type 1 and 2 infiltrates (C1 and C2), indicate that a tumor promoting role of these genes, as patients belonging to the category of worse survival, had a higher proliferation rate. In contrast, compared with those of the other infiltration types, the expression of DLC1 and DAPK2 was greater in C3, and the expression of DRAM1 and NLRC4 in C3 and C6 was significantly higher than that in C1 and C2. These findings indicated that higher gene expression was related to good immune composition which suggested that these genes may play a tumor inhibition role.

Fig. 14.

Correlation analysis between autophagy related gene expression and pan-cancer immune and pan-cancer microenvironment. a Association of DEARGs with immune subtypes in 33 cancers (p < 0.001). b–e Correlation between the expressions of 7 DEARGs and Estimates score, Stromal score, Immune score and Tumor Purity. f, g Correlation between the expressions of seven ARGs and tumor stemness scores

Not surprisingly, we also observed significant differences in the degree of association between ARGs and the stromal scores of different cancer types (Fig. 14b). DLC1 had the highest correlation with the stromal scores of different cancer types (cor = 0.60, p < 0. 001). More specifically, we found a significant positive correlation between ITGB4 and stromal scores in KICH patients (p < 0.01), NLRC4 in CHOL patients, DRAM1 in DLBC patients, DAPK2 in TGCT patients and DLC1 in MESO patients (p < 0.001). BIRC5 and ATIC were significantly negatively correlated with the stromal scores of GBM and SARC, respectively (p < 0.001). In addition, we used the ESTIMATE program to detect the correlation between autophagy related genes and immunity; we estimated scores and tumor purity in tumors, and observed results similar to those of the stromal score test (Fig. 14c–e).

Cancer stem like cells (CSCs) are derived from different sources, including long-term surviving stem cells or progenitor cells, or can be transformed from non-stem cells to CSCs by deregulation of related signaling pathways [29]. CSCs promote tumor progression through their self-renewal and invasive ability, which is the main reason for treatment induced drug resistance [29]. The correlation between autophagy related genes and tumor stemnesss was assessed by RNAss (Fig. 14f) and DNAss (Fig. 14g). Interestingly, we found that ITGB4, NLRC4, DAPK2, DLC1 and DRAM1 were negatively correlated with most RNAss and DNAss (p < 0.0001). Notably, although all genes were strongly positively or negatively correlated with DNAss and RNAss, they were positively correlated with DNAss and negatively with RNAss in THCA. These conflicting results suggested that in different cancers, DNAss and RNAss can recognize different cancer cell populations with different characteristics or different stemness degrees.

CellMiner is a database for studying molecular and pharmacological data in NCI-60 cancer cell lines [30]. We evaluated the expressions of ARGs in these cell lines and systematically tested the correlation between expression levels and drug sensitivity to more than 200 chemotherapeutic drugs (Table S3). Interestingly, we found that increased expression of DARM1 and IGTB4 was associated with increased resistance of different cell lines to multiple chemotherapeutic drugs (Cor < − 0.4 and p < 0.0001) (Fig. 15). For example, DRAM1 was associated with cell resistance to Vinblastine, Tamoxifen, VINORELBINE, TYROTHRICIN, Eribulin mesylate, Vinorelbine, Paclitaxel, Actinomycin D, Docetaxel, Pipamperone and Crizotinib; ITGB4 was associated with cellular resistance to Bortezomib, Ixazomib citrate, Arsenic trioxide, Carmustine and ARSENIC TRIOXIDE (Fig. 15). We also noticed a few genes to be related to sensitivity to different drugs and noted that different genes may have opposite associations with the same drug. For example, DLC1 and DRAM1 were related to increased sensitivity of cells to Dasatinib, while NLRC4 was related to increased resistance of cells to the same drug (Table S3).

Fig. 15.

Association of seven DEARGs expression with drug sensitivity. The top 16 statistically significant correlations between DEARGs and anticancer drug sensitivity. The scatter plots show the associations between ARGs expression and drug sensitivity

Correlation analysis of autophagy related genes in LUAD

Here, we provided a thorough investigation of autophagy related genes using TCGA-LUAD data. The association between autophagy related gene expression and immune subtypes in LUAD had a similar pattern to that observed in all 33 TCGA tumors, and all seven genes were significantly associated with types of immune infiltration (p < 0.05), though no patient samples belonged to the C5 immune subtype in LUAD (Fig. 16a). The expression levels of NLRC4 and BIRC5 pathology were significantly different with respect to age in LUAD patients (p < 0.05) (Fig. 16b). Figure 16c shows that with regard to fustat, the expression levels of all seven autophagy related genes were significantly different of LUAD patients (p < 0.05). The expression levels of BIRC5, NLRC4 and ATIC in LUAD patients were significantly different in gender (p < 0.05) (Fig. 16d). The expression level of DAPK2 in LUAD patients was significantly different in pathological M stages (p < 0.05) (Fig. 16e). Figure 16f indicated BIRC5, ATIC, DAPK2, DLC1 and DRAM1 expressions in LUAD patients were significantly different according to pathological TNM stage (p < 0.01).

Fig. 16.

Correlation analysis between the expression of seven ARGs and clinical characteristics in LUAD. a Association of autophagy related genes expression with immune subtypes in LUAD (p < 0.001). Correlation between autophagy related gene expression and age b, fustate c, gender d, M stage e, and TNM stage f. ^*p < 0.05, ^**p < 0.01, ^***p < 0.001

Given that stromal cells are a large compartment in the tumor microenvironment especially in LUAD, we further studied the correlation between the expression of ARGs and the stromal score. We found that NLRC4, DLC1 and DRAM1 correlated positively with the LUAD stromal score (p < 0.001), with NLRC4 having the strongest correlation (Cor = 0.6) (Fig. 17). However, there were no significant correlations between ITGB4 or DAPK2 and the stromal score. NLRC4, DLC1 and DRAM1 were also found to be associated with the immune score (p < 0.05) and estimated score (p < 0.001). Except for BIRC5 and ATIC, all ARGs correlated negatively with RNA stemness score (r =  − 0.21 ~  − 0.48, p < 0.001), with a smaller association with DNAss (r =  − 0.086 ~  − 0.23, p < 0.001) (Fig. 17).

Fig. 17.

Correlation between autophagy related gene expression and RNAss, DNAss, Stromal score, Immune score, and Estimate Score in LUAD

In a manner analogous to the findings in LUAD, the genes NLRC4, DLC1, and DRAM1 exhibited a positive correlations with the StromalScore in both BRCA and PRAD (p < 0.05).

However, contrary to the observations in LUAD, the strongest correlation in BRCA and PRAD was noted for DLC1, with correlation coefficients of 0.62 and 0.65, respectively (Figs. 18, 19). In BRCA, ITGB4 expression did not significantly correlation with the StromalScore, whereas in PRAD, no significant correlation was observed for BIRC5 and ATIC expression with the StromalScore. Furthermore, NLRC4, DLC1, and DRAM1 were positively correlated with both the ImmuneScore and the ESTIMATEScore, mirroring the results obtained in LUAD. Our investigation also revealed that the correlation of the same gene with the ImmuneScore varied across different tumor types. For instance, in BRCA, ATIC was negatively correlated with the ImmuneScore, whereas no significant correlation was detected in PRAD (Figs. 18, 19). In PRAD, with the exception of BIRC5 and ATIC, all ARGPS were negatively correlated with the RNAss, with correlation coefficients ranging from − 0.25 to − 0.64 (p < 0.05), and all ARGPS were positively correlated with DNAss, with correlation coefficients ranging from 0.09 to 0.27(p < 0.05).

Fig. 18.

Correlation between ARGs expression and RNAss, DNAss, Stromal score, Immune score, and Estimate Score in BRCA

Fig. 19.

Correlation between autophagy related gene expression and RNAss, DNAss, Stromal score, Immune score, and Estimate Score in PRAD

Virtual screening for potential drugs targeting NLRC4

In the meticulous virtual screening initiative encompassing the PDB online database and the YaTCM database, our focus was honed on identifying potential NLRC4-targeting drugs.

Leveraging the SitemMap module within the Schrodinger software suite, we adeptly predicted the plausible binding sites of the NLRC4 protein. Through a meticulously crafted series of screening protocols, which included ADMET, HTVS, SP, and XP, we selected 26 molecules endowed with flexible docking capabilities (Table 5).From this pool of candidates, we meticulously identified the top five potential drugs with the highest scores, namely Paleatin B (Fig. 20a), Magnolignan C (Fig. 20b), (±)−3ʹ,3ʹʹ-Bisdemethylpinoresinol (Fig. 20c), Muricarpone B (Fig. 20d), and 5-O-Ethyl-hirsutanonol(Fig. 20e), exhibiting affinities of − 12.171, − 11.508, − 11.313, − 11.088, and − 11.081, respectively.

Table 5.

The top 26 compounds selected from YaTCM as potential drugs targeting NLRC4

Hits	Name	Docking score
1	Paleatin B	− 12.171
2	Magnolignan C	− 11.508
3	(±)−3ʹ,3ʹʹ-Bisdemethylpinoresinol	− 11.313
4	Muricarpone B	− 11.088
5	5-O-Ethyl-hirsutanonol	− 11.081
6	(2R-trans)−2,3-Dihydro-2-(4-hydroxy-3-methoxyphenyl)−3-(hydroxymethyl)−7-methoxy-5-benzofuran-propanol	− 10.727
7	3-(-alpha,4-Dihydroxy-3-methoxybenzyl)−4-(hydroxy-3-methoxybenzyl)tetrahydrofuran	− 10.716
8	5-O-Methylhirsutanonol	− 10.603
9	Princepin	− 10.591
10	3-(alpha,4-Dihydroxy-3-methoxybenzyl)−4-(hydroxy-3-methoxybenzyl) tetrahydrofuran	− 10.426
11	Licarin A	− 10.319
12	Citrusin A_qt	− 10.187
13	Isoamericanol A	− 10.116
14	5-(3-hydroxy-4-isoacetoxybutyn-1)-bithiophene	− 10.107
15	Polyhalogenated homosesquiterpenic fatty acid D	− 10.048
16	Nordihydroguaiaretic acid	− 9.936
17	(2R-trans) 2,3-Dihydro-2-(4-hydroxy-3-methoxyphenyl)−3-(hydroxymethyl)−7-methoxy-5-benzofuranpropanol acetate	− 9.86
18	5-(3,4-dihydroxybutyn-1)-bithiophene	− 9.374
19	(s)-tryptophan-betaxanthin	− 9.197
20	Isoamericanol A	− 9.197
21	(SC2RC7)-gamma-Glutamyl-S-benzylcysteine	− 8.538
22	(–)-Thujaplicatin trimethyl ether	− 8.23
23	5alpha,6beta,7beta-Trihydroxy-8alpha-methoxy-2-(2-phenylethyl) chromone	− 8.07
24	threo-2,2ʹ-Dimethoxy-4-(3-hydroxy-1-propenyl)−4ʹ-(1,2,3-tri-hydroxypropyl) diphenyl ether	− 7.969
25	Citric acid	− 7.785
26	S-Methylglutathione	− 7.583

Fig. 20.

Virtual screening for potential drugs targeting NLRC4 (a Paleatin B, b Magnolignan C, c (±)-3',3''-Bisdemethylpinoresinol, d Muricarpone B, and e 5-O-Ethyl-hirsutanonol)

Discussion

LUAD is the most common type of lung cancer in the world. It has a poor prognosis, high invasiveness and a low 5-year survival rate in advanced disease. Although the specific etiology of LUAD is unclear, a variety of clinical features are associated with the disease [31, 32]. In recent years, bioinformatics analysis and the discovery of biomarkers have become important tools in disease diagnosis, prognosis prediction, and treatment decision-making [33]. For example, Liu et al. established a prognostic model based on mitophagy-related genes that can significantly improve the prognostic evaluation of LUAD [34]. Immune checkpoint inhibitors (ICIs) represent a new hot spot in tumor therapy. Huang et al. explored a programmed cell death gene prognostic model associated with survival and immunotherapy prediction [35]. Studies have shown that autophagy is associated with tumor-infiltrating lymphocytes (TILs) in early-stage lung adenocarcinoma (LUAD) and can identify early LUAD patients with different OS [36].

In recent years, increasing evidence has shown that autophagy is a"double-edged sword", that inhibits tumors in the early stage, causes tumor progression in the later stage, and subsequently leads to drug resistance [6, 37, 38]. Studies in HNSCC, ovarian cancer, colorectal cancer, non-small cell lung cancer, and prostate cancer have provided abundant support for the link between autophagy and tumorigenesis [31, 39]. However, most researchers were concerned about the role and potential impact of single ARGs, and there were few systematic analyses of autophagy-based signatures for LUAD have been performed.

Based on the GEO transcriptome expression profile data, we screened ARGs and identified 14 differentially expressed ARGs in tumor samples form LUAD patients. Go analysis showed that most of these DEARGs are involved in autophagy, inflammatory complexes, apoptosis, cytokine adhesion and other mechanisms. KEGG analysis showed enrichment in tumor pathways and platinum drug resistance, suggesting that autophagy gene imbalance may be involved in tumor biological processes. Through the analysis of the Reactome database, we further discovered that ARGPS in LUAD were primarily involved in key pathways such as the immune system, signal transduction, programmed cell death, and gene expression (transcription). These pathways played important roles in the development and progression of LUAD. Autophagy was closely related to the proliferation and poor prognosis of LUAD. The expression of Beclin 1 in human lung adenocarcinoma was negatively correlated with tumor size and primary tumor stage, but its expression in non-small cell lung cancer (NSCLC) was lower than that in normal tissues [40]. In NSCLC cells, the induction of autophagy by mTOR inhibitors was associated with radio sensitization [41]. In addition, hydroxychloroquine (an autophagy inhibitor) has been tested for general therapeutic efficacy in NSCLC [8]. Therefore, research on valuable prognostic features was helpful for clinical diagnosis.

We used the TCGA-LUAD dataset for verification. In addition to ERO1L, 13 ARGs were expressed. LUAD patients were divided into Cluster1 and Cluster2 by PCA cluster analysis. Survival analysis showed that Cluster1 had a higher survival rate and was related to clinical characteristics. Then, we found 8 ARGs related to OS in univariate Cox regression analysis. Through LASSO and multivariate Cox regression analysis, we found that 7 genes (ATIC, BIRC5, DAPK2, DLC1, DRAM1, ITGB4, and NLRC4) were significantly correlated with OS in LUAD patients. We established a prognostic model of the autophagy-related gene signature. We then verified the results in cell lines and showed that the expression of the above seven genes was consistent with the analysis results.

An autophagy gene prognostic model was established and verified. The autophagy signature was able to divide LUAD patients into a high-risk group and a low-risk group according to the median risk score. The OS of patients in the high-risk group was significantly shorter than that of patients in the low-risk group. TNM stage and risk score were determined to be independent prognostic factors for LUAD by univariate and multivariate Cox analyses. Compared with previous predictive models revealed that our model has significant advantages in both the C-index and the RMS error (Fig. 9). Therefore, this ARGPS may be used as a prognostic biomarker for clinical application in the future.

In the present study, we investigated the relationships between the expression patterns of ARGs and the overall survival of 33 major tumor types using TCGA pan-cancer data, as well as the relationships between their expression patterns and the tumor microenvironment and pharmacological effects. Pan-cancer analysis has shown that in various types of tumors (including LUAD, BRCA, and PRAD), the expression profiles of ARGPS are significantly correlated with the StromalScore, Immune Score, ESTIMATEScore, RNASS and DNAss. Furthermore, our study has further revealed that the expression of ARGPS was correlated with the expression levels of immune checkpoint markers (PD-1 and CTLA-4). These findings provided important evidence for a deeper understanding of the role of autophagy in immune regulation in LUAD. The expression of DLC1 and NRLC4 decreased continuously in primary tumor samples, compared with adjacent normal samples. All other ARGs were inconsistently up-regulation or down-regulation in different cancer tumors, in which DAPK2 and DRAM1 were mainly down-regulated, but AITC, BIRC5 and ITGB4 were mainly up-regulated in the tested cancer types. In addition, we found that ARGs were associated with immune subtypes and the tumor microenvironment, and that the degree of association varied according to family member and tumor types. More interestingly, our results showed that ITGB4, BIRC5 and ATIC cluster closely and were associated with more aggressive immune subtypes (C1 and C2) and tumor stem cell like characteristics, indicating their role in the immune response and drug resistance. In conclusion, our work may greatly help to reveal their role in tumorigenesis, especially with respect to immune type, the tumor microenvironment and drug resistance, which was very important for the development of personalized cancer therapeutic drugs.

A previous study found that ATIC was an effective and yet unrecognized target for chemoradiotherapy sensitization. Broadly, it suggested that purine levels in cells may play an unrecognized role in regulating the efficiency of the DNA damage response, which may be used in chemoradiotherapy sensitization strategies [42]. ATIC polymorphisms were confirmed to be related to the effectiveness and toxicity of pemetrexed in NSCLC [43]. Our study found that high expression of ATIC was associated with a poor prognosis in ACC, LIHC, KIRC and UVM. BIRC5 is an immune related gene that promotes cell proliferation and inhibits apoptosis. It is highly expressed in most tumors and resulted in poor prognosis of cancer patients [44]. BIRC5 may be a promising predictive marker and therapeutic target for breast cancer and pancreatic cancer [45]. Our study also confirmed this, with BIRC being highly expressed in ACC, KIRC, KIRP, LGG, LIHC, MESO and PAAD, and suggesting a poor prognosis. It has been reported that down regulation of DAPK2, a tumor suppressor, enhanced the proliferation and migration of NSCLC cells in vitro and in vivo by significantly activating NF-κB signaling pathway [46]. However, our study revealed that patients with high expression of DAPK2 had a poor prognosis in DLBC, GBM, and KIRC. DLC1 could inhibit the proliferation, migration and invasion of LUAD cells by inhibiting the MAPK signaling pathway [47]. In this study, high DLC1expression was associated with poor prognosis in ACC, LGG and MESO patients. DRAM1 may play an anti-cancer role in NSCLC cells by promoting EGFR lysosomal degradation [48]. A recent study found that expression levels of ITGA11 and ITGB4 were significantly up-regulated in LUAD and LUSC, but that the expression level of ITGB8 was significantly up-regulated in LUSC. In addition, a higher expression level of ITGB4 was associated with worse OS in patients with LUAD [49].

NLR family CARD domain-containing 4 (NLRC4), a pivotal member of the NLR family, plays a crucial role in the innate immune response to bacterial invasion [50]. After inflammasome formation, NLRC4 initiates a cascade of immune reactions through IL-18 production and other pathways, actively engaging in the body's first line of defence against pathogens [50]. Emerging as a promising therapeutic target, NLRC4 is capable of bolstering potent anti-tumor immune responses, thereby combating the invasive tendencies of a variety of cancers [51]. Recent studies have illuminated the role of NLRC4 in epithelial cells, where its expression fosters a robust anti-tumor immune response, potentially curbing the transition of colorectal cancer (CRC) into a more virulent metastatic phase [52]. Furthermore, the presence of NLRC4 is linked to enhanced T-cell infiltration within tumors, a factor that is often indicative of a more favorable patient prognosis [52]. Our findings revealed that while NLRC4 expression was relatively subdued in LUAD, it exhibited a positive correlation with the immune score. Notably, higher levels of NLRC4 were observed in the advanced C3 and C6 stages compared to the earlier C1 and C2 stages. This correlation suggests that elevated NLRC4 expression is tied to a more advantageous immune profile, hinting at a possible tumor-suppressive role for these genes. Given these insights, NLRC4 emerges as a potential therapeutic target for LUAD. Utilizing virtual drug screening techniques, we have pinpointed five compounds with the potential to target NLRC4, opening new avenues for the development of targeted therapies against LUAD.

However, several limitations of this study should to be noted. Firstly, the underlying molecular mechanisms of DEARGs remain to be elucidated, and their roles could be confirmed through additional in vitro or in vivo studies. Secondly, a comprehensive assessment of LUAD prognostic factors is necessary, taking into account variables such as smoking status, tumor dimensions, and the presence of lymph node metastasis. Moving forward, our research team is committed to validating the expression levels of NLRC4 in clinical specimens and to incorporating a broader range of clinical data for further validation and model refinement. Additionally, we intend to delve deeper into the mechanisms by which NLRC4 contributes to the development of LUAD, aiming to enhance our understanding of its role in the disease process.

Conclusions

In conclusion, our study established a seven autophagy gene prognostic signatures that cloud accurately predict OS in LUAD patients. We anticipate that NLRC4 will emerge as a significant prognostic biomarker, shedding new light on the underlying mechanisms of LUAD and offering novel perspectives for its management.

Abbreviations

ACC

Adrenocortical carcinoma

ARGs

Autophagy-related genes

ARGPS

Autophagy-related gene prognostic signature

BLCA

Bladder urothelial carcinoma

BRCA

Breast invasive carcinoma

CESC

Cervical squamous cell carcinoma and endocervical adenocarcinoma

CHOL

Cholangiocarcinoma

COAD

Colon adenocarcinoma

DEARGs

Differentially expressed autophagy-related genes

DLBC

Lymphoid neoplasm diffuse large B-cell lymphoma

ESCA

Esophageal carcinoma

GBM

Glioblastoma multiforme

GO

Gene ontology

HADb

Human Autophagy Database

HNSC

Head and neck squamous cell carcinoma

HRs

Hazard ratios

KICH

Kidney chromophobe

KEGG

Kyoto Encyclopedia of Genes and Genomes

KIRC

Kidney renal clear cell carcinoma

KIRP

Kidney renal papillary cell carcinoma

LAML

Acute myeloid leukemia

LGG

Brain lower grade glioma

LUSC

Lung squamous cell carcinoma

LIHC

Liver hepatocellular carcinoma

LUAD

Lung adenocarcinoma

MESO

Mesothelioma

OS

Overall survival

OV

Ovarian serous cystadenocarcinoma

PRAD

Prostate adenocarcinoma

PAAD

Pancreatic adenocarcinoma

PCPG

Pheochromocytoma and paraganglioma

READ

Rectum adenocarcinoma

STAD

Stomach adenocarcinoma

SARC

Sarcoma

SKCM

Skin cutaneous melanoma

TCGA

The Cancer Genome Atlas

TGCT

Testicular germ cell tumors

THCA

Thyroid carcinoma

THYM

Thymoma

UCEC

Uterine corpus endometrial carcinoma

UVM

Uveal melanoma

UCS

Uterine carcinosarcoma

Author contributions

Xue Xu and Meng-Yu Zhang: Writing-original draft, Visualization, Methodology, Data curation, Conceptualization. Jia-Qi Fan: Visualization, Formal analysis. Guo-Dong Li: Writing-review & editing, Validation. Jun-Yi Li: Writing—review and editing, Visualization. Xiao Chen: Writing—review and editing, Funding acquisition, Conceptualization. All authors reviewed the manuscript.

Funding

This work was supported by the Natural Foundation of Shandong Province (ZR2023QH097) and the Traditional Chinese Medicine Science and Technology Project of Shandong Province (M20240903).

Data availability

The data that supported the findings of this study were openly available in open databases. 232 autophagy-related genes (ARGs) were available in the human autophagy database (HADb) at http://www.autophagy.lu/clustering/index.html.GSE116959 data was openly available in GEO database at https://www.ncbi.nlm.nih.gov/geo/ (10.1038/s41388-019-0935-y), reference number (31417181). The gene expression data and clinical feature data of lung adenocarcinoma (LUAD) were available in the Cancer Genome Atlas (TCGA) database at https://portal.gdc.cancer.gov/. TCGA pan-cancer data, were openly available in UCSC Xena browser at https://xenabrowser.net. The PCR data that supports the findings of this study, including any relevant details needed to reproduce the published results, are available from the corresponding author upon reasonable request.

Declarations

Competing interests

The authors declare no competing interests.