The nucleic acid-sensing (NAS) pathways-related genes were acquired from GSEA gene sets, path cards (https://pathcards.genecards.org/), and published articles. By eliminating duplicated genes, 371 NAS-related genes were included in this study. Three independent SCLC cohorts were enrolled in this study, including 79 SCLC patients collected from cBioPortal, 21 samples from GSE30219, and our cohort including 14 SCLC patients from Sichuan Provincial People’s Hospital, University of Electronic Science and Technology of China. This study was approved by the Ethics Committee of Sichuan Province People’s Hospital, School of Medicine, University of Electronic Science and Technology of China (No.20240122). The clinicopathological characteristics of the 14 patients were provided in Supplementary Table 1 . The raw bulk transcriptome counts data, normalized and log2 converted RNA-sequencing (RNA-seq) profiles FPKM and normalized RMA of SCLC patients, and normal samples were acquired from cBioPortal (https://www.cBioPortal.org/) and GEO (https://www.ncbi.nlm.nih.gov/geo/). Normal samples and SCLC samples without complete clinical information were excluded from this study. The “GeoTcgaData” R package was applied to convert ensemble IDs to gene symbols.

To identify the differentially expressed genes (DEGs) between SCLC and normal samples, the “limma” package was used with the criterion of the adjust. P < 0.05 and |log2FC| > 1.0.

Based on the expression profiles of 371 NAS-related genes, unsupervised classes of SCLC datasets were estimated with the consensus clustering method using the ‘Consensus Cluster-Plus 1.60.0’ package in R, and two clusters were obtained. Principal Component Analysis (PCA) was used to verify the clusters based on the expression profiles of the above genes.

To explore the differences in overall survival (OS) of the two clusters identified by consensus clustering, we conducted a Kaplan–Meier analysis generated by the ‘Survival’ and ‘Survminer’ packages of R.

The least absolute shrinkage and selection operator (LASSO) method was used to screen out the genes that make a difference to the OS of SCLC patients, and a total of 12 genes were regarded as candidate genes. Based on the 12 genes, multivariate Cox regression was utilized to construct the final signature. Ultimately, the risk score of SCLC patients was calculated as: Risk score= ∑Coefficient of(i)× Expression of the gene(i). The patients could be divided into high-risk and low-risk groups according to the median risk score.

The tissue slides of SCLC patients (n = 14) were stained with multiplex fluorescence using PanoPANEL mIHC Kits (#0004100100). The slides were dewaxed by xylene and rehydrated by ethanol, and then the slides were heated in a microwave with antigen retrieval. The slides were incubated with primary antibody overnight at 4°C for specific antigen binding including anti-IRF1 (#8478, CST, USA), anti-CD45 (#13917, CST, USA), anti-CD68 (#97778. CST, USA), anti-CD8 (#85336, CST, USA), anti-panCK (#4545, CST, USA), anti-FoxP3 (#12653, CST, USA). The corresponding secondary antibodies should be pipetted onto the slides, and incubated at room temperature for 1 hour. DAPI was used to stain cell nuclei and at last the slides were sealed. The staining was scored by image J software based on the intensity and degree of staining. The degree of staining was compared using the Wilcoxon rank-sum test.

To evaluate the clinical significance of the risk score, we compared the associations between risk scores and clinicopathological features. The characteristics of the patients included survival status and subtypes (SCLC-A, SCLC-N, SCLC-P, and SCLC-Y). We used the “clusterProfiler” R package to identify differences in biological function between the high-risk and low-risk groups. Univariate and multivariate Cox regression analyses were performed to determine whether the risk model could be an independent prognostic factor for SCLC patients. We constructed a nomogram to predict the 1-, 2-, and 3-year OS of SCLC patients by the “RMS” R package.

To explore the landscape of the immune microenvironment, the CIBERSORT algorithm was used to measure the infiltration of immune cells.

The TIDE (Tumor Immune Dysfunction and Exclusion) score was calculated in TIDE (http://tide.dfci.harvard.edu/), which is usually used to predict the response to immunotherapy for cancer patients. The ‘oncoPredict’ package was used to predict the sensitivity of SCLC patients at different risks to chemotherapy drugs.

Statistical analyses were performed using GraphPad Prism (v.9.0) (for experimental data), and R (v.4.2.1), and RStudio (v.3.5.3) (for sequencing data and matched clinical variables). Comparisons between groups were conducted using χ2 tests for categorical variables. Student’s t test was used for continuous variables. All the tests were two tailed, and P < 0.05 was considered statistically significant.

Before analysis, we adjusted the transcriptome data from public datasets and our dataset ( Supplementary Figure S1A ). The GSE30219 cohort yielded 271 differentially expressed genes (DEGs) compared to adjacent normal tissues ( Figure 1A ). Among the 271 DEGs, 164 genes are upregulated, and 107 genes are downregulated, including several NAS related genes, such as TLR3, PTGS2, IL-6, PRKDC, and MMP9 ( Figure 1B ). The UMAP analysis revealed that NAS related genes performed well in distinguishing cancer tissues and adjacent normal tissues ( Figure 1C ). We then conducted consensus clustering analysis to identify novel subtypes of SCLC based on NAS related genes. It demonstrated that the SCLC patients could be well classified into two clusters when k = 2 ( Figures 1D–F ; Supplementary Figures S1B–D ). The PCA analysis, used to validate the clusters with respect to the expression profiles of NAS-related genes, showed a precise classification of the two clusters ( Supplementary Figures S1E-G ). More importantly, a notable disparity was observed in the OS time between the two clusters. Patients in Cluster 1 exhibited a favorable prognosis, whereas those in Cluster 2 showed a poor prognosis ( Figure 1G ). Similar results were validated in two additional SCLC cohorts, including the GEO cohort and our own cohort ( Figures 1H, I ). Overall, SCLC patients can be effectively classified into two groups based on the expression of NAS-related genes and the patients in cluster 1 have a better prognosis than patients in cluster 2.

To construct a prognostic model based on a signature of NAS related genes, we conducted LASSO regression analysis and identified a total of 12 genes ( Figures 2A, B ). The multivariate Cox regression analysis demonstrated that there are 7 genes are related to the prognosis of patients, including APOH, IL23A, IRF1, MAPKAPK2, POLR2E, TF, and UBA1( Supplementary Figures S2A–G ). Using the expression levels and coefficients of the 7 model genes, we constructed a prognostic gene signature (7NAS risk model) and calculated the risk score for each patient using the provided formula: Risk score = (-0.05773 * APOH) + (-0.01659 * IL23A) + (-0.01181 * IRF1) + (-0.01023 * MAPKAPK2) + (-0.01169 * POLR2E) + (-0.03609 * TF) + (-0.00676 * UBA1). Our results demonstrated that the risk score correlated with clinical characteristics such as alive or dead survival status but not with the subtypes of SCLC patients ( Figures 2C, D ). The low-risk groups significantly exhibit enrichment of multiple biological processes, particularly immune-related functions, such as adaptive immune response, antigen processing and presentation, and MHC protein complex binding ( Figures 2E, F ). We further analyzed the 7NAS risk model in predicting OS in both the training and validation sets ( Figures 2G–I ). Notably, a significant disparity was observed in the OS time between the high-risk and low-risk groups. Patients in the low-risk group demonstrated a tendency towards longer OS time in both the training and validation cohorts (P < 0.05) ( Figures 2J–L ). Collectively, we constructed a 7NAS risk model that could effectively predict the immune function status and OS of SCLC patients.

Next, we conducted univariate and multivariate Cox regression analyses to validate whether the 7NAS risk score could stand as an independent prognostic factor in SCLC. The univariate Cox regression analysis illustrated that the risk score constituted a risk factor in SCLC (HR = 1.262, 95% CI: 1.173-1.357, P < 0.001, Figure 3A ). Following adjustment for confounding factors, the multivariate analysis also indicated that the risk score remained an independent prognostic factor (HR = 1.234, 95% CI: 1.139-1.337, P < 0.001, Figure 3B ) for SCLC patients. A nomogram was developed to estimate the 1-, 2-, and 3-year OS, incorporating sex, tumor stage, metastasis and risk score ( Figure 3C ). Calibration curves demonstrated the predictive accuracy of this model for 1-, 2-, and 3-year survival rates ( Figure 3D ). It can be observed that the concordance index of nomogram is significantly better than clinical stage and metastasis, but slightly lower than the risk score of 7NAS model ( Figure 3E ). We assessed the area under the curve (AUC) values in a merged cohort which demonstrated the high accuracy in predicting 1-, 2-, and 3-year survival among SCLC patients ( Figure 3F ). The high-nomo-risk group commonly has a worse prognosis than the low-nomo-risk group ( Figure 3G ). Collectively, the 7NAS risk model represents an independent prognostic factor for patients with SCLC.

We analyzed the relationships between the risk scores and the immune characteristics of SCLC patients. The results revealed distinct immune microenvironments between the high-risk and low-risk groups. Specifically, SCLC patients in the low-risk group exhibited higher infiltration of immune cells in the merged dataset that combined the public dataset and our dataset ( Figures 4A–C ). The patients with lower risk scores were more sensitive to immunotherapy in IMvigor210 and GSE126044 datasets ( Figures 4D, E ), and the patients with lower risk scores tended to have better OS in IMvigor210 ( Figure 4F ). We further analyzed the correlation between the expression of the 7 NAS-related genes and the infiltration of intra-tumoral immune cells, separately. The results indicated that IRF1 was correlated with the infiltration of M1 macrophages, NK cells, and γδ T cells ( Figure 4G ). Further confirmation through mIHC revealed a positive correlation between high IRF1 expression and the infiltration of CD45⁺ immune cells, CD8⁺ T lymphocytes, and CD68⁺ macrophages ( Figures 4H, I ). These results showed that the 7NAS risk score, especially the IRF1 expression, has a positive relationship with the activated immune microenvironment.

To examine the predictive role of the 7NAS risk score in drug sensitivity of SCLC patients, we determined the half-maximal inhibitory concentration (IC50) value for each drug. The IC50 values of these chemotherapeutic agents exhibited a positive correlation with the risk score ( Figure 5A ), suggesting that patients with a low-risk score generally exhibited better responses to chemotherapy. Furthermore, in the case of Temozolomide (a second-line drug for SCLC patients) and PARP inhibitors, including MN.64_1864, Olaparib, WIKI4, XAV939, and Palbociclib, patients in the low-risk group exhibited greater sensitivity compared to those in the high-risk group ( Figures 5B–G ). Moreover, patients in the low-risk group decreased the TIDE score compared to the high-risk group from cBioPortal ( Figure 5H ). Conclusively, the 7NAS risk score plays a role in predicting patients who are sensitive to chemotherapy and immunotherapy.