We collected compliant LUAD sequencing datasets from two different databases. First, the count matrix of the TCGA-LUAD dataset, the maf format files of genomic mutant sites, and the corresponding clinical information were collected from the UCSC-Xena database (https://xena.ucsc.edu/). We included only patients without missing clinical information, containing a total of 492 LUAD patients, and used the count matrix as a development cohort after normalizing it to a transcripts per kilobase million (TPM) matrix. In addition, we collected the pancancer dataset from UCSC Xena. Three compliant LUAD datasets were retrieved from the GEO database (http://www.ncbi.nlm.nih.gov/geo/) and convened: GSE30219 (GPL570 Platform), GSE42127 (GPL6884 Platform), and GSE72094 (GPL15048 Platform). After including patients diagnosed with LUAD among them, the “sva” package was used to integrate the three datasets to remove batch effects between platforms (13), resulting in a meta-GEO dataset of 615 patients with LUAD for external validation. Finally, the single-cell dataset GSE131907 of LUAD was collected without additional processing and analyzed according to the original parameters (14). We used the “seurat” package to preprocess and analyze the single-cell data. In short, we log-normalized the data by “NormalizeData”. Subsequently, 1500 feature variables in the dataset were identified by “FindVariableFeatures”. PCA analysis was performed on the data set based on the 1500 feature variables. Finally, neighbors and cell clusters were identified by the first 40 principal components.

We obtained two lung cancer cell lines (HCC827 and A849) and a normal lung epithelial cell line (16HBE) from GAINING-Bioscience (China). These cells were cultured in DMEM (Biological, Israel) supplemented with 10% heat-inactivated FBS, 1% penicillin, and 1% streptomycin. The cells were grown at a constant temperature of 37°C and with 5% CO2. We used Lipofectamine 8000 reagent (Invitrogen, USA) for transient transfection of siRNA according to the manufacturer’s instructions, aiming to suppress the expression of specific genes. The siRNAs used and their corresponding blank controls were as follows: si-FADD: GGAAGAAGACCTGTGTGCAGCATTT; siNC: GGAAGAAGTCCGTGTCGACAGATTT.

Whole RNA was extracted using TRIzol reagent (Invitrogen, CA, USA) according to the instructions. cDNA of FADD was amplified according to the following primers: Forward: CATCTACCTCCGAAGCGTCC; Reverse: GGGCTACCTTCCTGGAGAGA. The 7500 Fast Real-time PCR system (Thermo Fisher, USA) was used for gene quantification according to the instructions, and the internal reference was GAPDH.

Following the operating manual of the kit, we applied the Cell Counting Kit-8 (CCK-8, Bioss, China) to detect the growth activity of various LUAD cell lines. An enzyme-linked immuno-absorbance assay (BioTek, USA) was used to reflect the cell numbers by measuring the absorbance at 450 nm during cell growth.

To assess the significant biological pathways of subgroups with different levels of FADD, we first identified significantly differentially expressed genes (DEGs) among FADD subgroups based on a threshold of fold change>2 and adjusted p value<0.05 utilizing the “limma” package. Functional annotation and enrichment analysis of DEGs were achieved through the Metascape online database (https://metascape.or). The differentially enriched pathways among different FADD subgroups were subsequently analyzed and identified by GSEA software (version 4.0.1) based on the KEGG background database.

We systematically analyzed the heterogeneity of the immune microenvironment in different FADD subgroups from different perspectives. First, we evaluated the differences in cell types among different FADD subgroups from a single-cell perspective and assessed the cellular interaction pathways between cells with different FADD levels using the “CellChat” R package (15). The immune scores between different FADD subgroups were generated by the “ESTIMATE” algorithm (16). Subsequently, ssGSEA was executed by the “gsva” package to assess the relative activity of tumor immune-related pathways among different FADD subgroups (17). Finally, the abundance of immune cell infiltration among FADD subgroups was assessed by the “Cibersort” algorithm containing the reference expression profiles of 22 immune cells (18).

We collected homologous recombination defect (HRD) scores and MSI scores of different LUAD patients in the TCGA cohort from previous studies and assessed the differences between different FADD subgroups by “ggpubr” (19). We processed the raw maf files using “maftools” and calculated the tumor mutation burden (TMB) of individual patients after excluding nonsense mutated fragments (20). We classified significantly mutated genes according to a threshold of mutation number >45 and evaluated their differences and mutational cooccurrence among different FADD subgroups.

Based on the ridge regression algorithm in the R package “pRRhetic”, we evaluated the differences in sensitivity of five first-line lung adenocarcinoma chemotherapeutic agents among different FADD subgroups (21). The results were generated as the patient’s half-maximal inhibitory concentration (IC50) values for a given drug, with lower IC50 values being more sensitive to treatment. We then assessed the sensitivity of patients to immunotherapy based on the level of immune cell exhaustion using the TIDE algorithm (22). Finally, we assessed the similarity of the expression profiles of the TCGA and Meta-GEO cohorts to the immunotherapy cohort by the subclass mapping algorithm and generated the sensitivity to anti-PD-1 and anti-CTLA-4 treatments.

We used R software (Version 4.1.0) and GraphPad Prism 9.0 to analyze the data and draw figures. Appropriate t tests or Wilcoxon tests were applied to compare differences between the two subgroups according to the data structure. χ2 tests were used to compare differences in percentages. The significance of survival prediction was assessed by the log-rank test and univariate and multivariate Cox regression. The correlation of continuous variables was examined by Pearson’s coefficient. The threshold of significance was set at P<0.05.

In the TCGA database, the mRNA levels of FADD were significantly overexpressed in 27 solid malignancy tissues, including LUAD (P < 0.05, Figure 1A ). Survival analysis from a pancancer perspective showed that FADD was a significant risk indicator for seven malignancies, including LUAD ( Figure 1B ). We further confirmed by qRT−PCR that the mRNA levels of FADD were significantly higher in two LUAD cell lines (A549 and HCC827) than in the normal lung tissue cell line 16HBE ( Figure 1C ). By CCK8 assay, we confirmed that FADD played a role in promoting tumor cell proliferation in LUAD cell lines ( Figure 1D ). At the protein level, we found significantly higher FADD levels in LUAD tumor tissues than in normal alveolar epithelium by two FADD antibody staining results (CAB010209 and HPA001464) in the HPA database (https://www.proteinatlas.org/) ( Figure 1E ).

Survival analysis showed significantly worse survival in both TCGA-LUAD and GEO-LUAD datasets in patients with high FADD than in patients with low FADD (P<0.001, P=0.034, respectively; Figures 2A, B ). Both univariate and multifactorial Cox regression analyses corrected for clinical characteristics showed that FADD could be an independent risk factor for OS in patients with LUAD (P<0.05, Figures 2C, D ). Furthermore, by analyzing clinical characteristics, we found a significant positive correlation between FADD and stage in the TCGA dataset (P<0.001, Figure 2E ) and a positive trend in the GEO dataset (P=0.087, Figure 2F ). Comprehensive meta-analysis in the small cell lung cancer database (https://lce.biohpc.swmed.edu/lungcancer) showed that FADD was significantly elevated in tumor tissues ( Figure 2G ) and was an unfavorable prognostic indicator for OS in lung adenocarcinoma patients ( Figure 2H ). In addition, subgroup survival analysis showed excellent prognostic efficacy of FADD in patients with early-stage LUAD in the TCGA cohort, especially in patients with stage II LUAD and age >= 70 years ( Figure S1A ), whereas FADD in the GEO cohort showed good prognostic performance in patients younger than 70 years and in male patients ( Figure S1B ).

We identified a total of 201 DEGs, of which 91 DEGs were upregulated in the high FADD subtype and 110 DEGs were upregulated in the low FADD subtype. Functional enrichment analysis showed that the upregulated genes in the high FADD subtype mainly modulated cell growth, cell junctions, and epidermal development ( Figure 3A ), while the upregulated genes in the low FADD subtype mainly modulated biological oxidation and cellular morphogenesis ( Figure 3B ). GSEA revealed that the pathways enriched in the high FADD subtype were cell cycle, pyrimidine metabolism, P53 signaling pathway, and apoptosis ( Figure 3C ). In contrast, the pathways enriched in the low FADD subtype were primary immune defense, ABC transport, and asthma ( Figure 3D ). In conclusion, these results suggest that patients with low FADD subtypes have a stronger oxidative stress response and immune activity, whereas tumor cells with high FADD subtypes have a dysregulated cell cycle with hyperactive cell division and cell proliferation.

We first assessed the microenvironmental heterogeneity of different levels of FADD from a single-cell perspective. We identified eight cell subtypes according to the original parameters ( Figure 4A ). We then found that FADD was highly expressed in most malignant and myeloid cells ( Figure 4B ). Specifically, low FADD cells were more predominant in B cells, NK cells, and T cells, while high FADD cells were more predominant in malignant cells and endothelial cells ( Figure 4C ). We identified significant cellular communication pairs based on a threshold of P<0.05, and the results showed that cells with low FADD expression had more cellular communication activity overall, especially myeloid and malignant cells ( Figure 4D ). Figure 4E shows the detailed communication pathways between different FADD cell subgroups, with fewer active pathways in high FADD cells and more active pathways in low FADD cells. Most of the communication pathways are associated with antitumor immunity (e.g., CXCL, CCL, TNF, etc.) ( Figure 4E ). In addition, malignant cells significantly received VEGF signaling ( Figure 4E ).

We then assessed the heterogeneity of immune infiltration due to different FADD levels from a Bulk-seq perspective. As assessed by the ESTIMATE algorithm, low FADD subtypes had higher immune scores, while high FADD subtypes had higher tumor purity ( Figure 5A ). We then examined the expression differences of seven classical immune checkpoints and therapeutic targets (CD8A, CTLA-4, LAG-3, PD-1, PD-L1, TIM-2, TNF) and found that TIM-2 and PD-L1 were significantly upregulated in the low FADD subtype ( Figure 5B ). Subsequently, by the ssGSEA algorithm, we found that the type II interferon response pathway was upregulated in the low FADD subgroup. While angiogenesis, EMT, hypoxia, paracrine immunity, and APC-coinhibitory pathways were upregulated in the high FADD subgroup ( Figure 5C ), the corresponding correlation analysis is shown in Figure 5D . Finally, Cibersort results showed that CD8 T cells, plasma cells, and memory B cells were upregulated in the low FADD subtype, whereas Tregs and M0 macrophages were upregulated in the high FADD subtype ( Figure 5E ), and the corresponding correlation analysis is shown in Figure 5F . Notably, M2 macrophages and FADD expression were significantly positively correlated ( Figure 5F ). In conclusion, these results suggest that low FADD levels may indicate “hot” tumors with active antitumor immunity. In contrast, high FADD levels were positively correlated with Tregs and M2 macrophages, which may represent immunosuppressed “cold” tumors.

We first assessed overall the differences in two indicators of genomic mutations at different FADD levels. The results showed that patients with high FADD had significantly higher HRD scores ( Figure 6A ), but there was no significant difference in MSI scores between the two FADD subtypes ( Figure 6B ). We calculated the TMB for each patient and showed that the high FADD subtype had a higher TMB, but the difference in TMB between the two subtypes was not significant ( Figure 6C ). We processed the raw mutation data with the maftools package, and the oncoplot showed the difference in mutation profiles between the two FADD subtypes for a total of 26 high-frequency mutated genes ( Figure 6D ). The chi-square test showed an increased frequency of TP53 mutations in the high FADD subtype but no significant differences in the other 25 high-frequency mutated genes ( Figure 6E ). The co-occurrence analysis showed that all high-frequency mutated genes were highly co-occurring and showed significant concordance ( Figure 6F ).

We then evaluated the efficacy of FADD in predicting sensitivity to chemotherapy and immunotherapy. Using ridge regression calculations, we found that patients with high FADD expression had lower IC50 values for the five first-line agents for LUAD (cisplatin, docetaxel, gemcitabine, paclitaxel, and vincristine), indicating that patients with high FADD are more sensitive to conventional chemotherapy ( Figure 7A ). The same results were observed in the externally validated Meta-GEO cohort ( Figure 7B ). The heterogeneity of the immune microenvironment due to different FADD levels suggested the existence of differences in immunotherapy sensitivity; therefore, we first evaluated the response rate of patients with different FADD levels to immunotherapy using the TIDE algorithm. The results showed that patients in the low FADD group in the TCGA-LUAD cohort had a higher response rate to immunotherapy (P=<0.001, Figure 7C ). In the GEO-LUAD cohort, patients in the low FADD group also responded more to immunotherapy (P<0.001, Figure 7D ). Finally, after subclass mapping assessment of the transcriptome, patients in the low FADD group in the TCGA and GEO cohorts were predicted to be more sensitive to anti-PD1 therapy (TCGA: FDR=0.008; GEO: FDR=0.042) ( Figures 7E, F ).