Comprehensive analysis of the autophagy-dependent ferroptosis-related gene FANCD2 in lung adenocarcinoma

A multi-step approach was used to identify and analyze the autophagy-dependent ferroptosis-related key gene in LUAD. The transcriptome and clinical information were downloaded from The Cancer Genome Atlas (TCGA) project and Gene Expression Omnibus (GEO) data. Autophagy-dependent ferroptosis-related genes were screened by the published articles. Differentially expressed genes (DEGs) related to autophagy-dependent ferroptosis were identified. Univariate and multivariate Cox analyses were applied to screen out the independent prognosis genes related to overall survival (OS). The key gene was identified by the intersection of the DEGs and the prognostic genes. The LUAD patients were classified into the high-risk and low-risk groups based on the key gene expression level. Kaplan-Meier (K-M) analysis and receiver operating characteristic (ROC) curve were conducted to analyze the survival prognosis of patients in TCGA and GEO cohorts. Chemotherapy sensitivity was predicted between different risk groups. Gene ontology (GO) and Kyoto encyclopedia of genes and genomes (KEGG) were conducted to investigate the potential bio-function of the key gene. ImmuneScore was calculated using the Tumor Immune Estimation Resource (TIMER) algorithm, and the TMB was counted as the total number of mutations per megabyte of tumor tissue.

All the RNA-Seq data were normalized as fragments per kilobase of transcript per million mapped reads. mRNAs ensemble gene identities were derived from the HUGO Gene Nomenclature Committee (HGNC) database. The corresponding clinical information includes age, gender, tumor grade, lymph node metastasis, AJCC TNM stages, and survival outcomes. Patients with insufficient clinical data were excluded. OS was estimated as the primary endpoint.

Autophagy-dependent ferroptosis-related genes were retrieved from the literature published before January 2021. After combining the related mRNA expression and the clinical data, the gene expression files were obtained. The DEGs between LUAD and normal lung tissues were identified with a false discovery rate (FDR) < 0.05 in the TCGA cohort. Univariate and multivariate Cox analyses of OS were performed to screen the genes with prognostic values in TCGA-LUAD cohort. The key gene was identified by the intersection of the DEGs and the prognostic genes in the TCGA cohort. The cut-off score was defined as the median expression level of the key gene in the LUAD cohort. Patients were stratified into high-risk and low-risk groups based on the cut-off score. To choose appropriate matching cohorts to perform survival analysis before selecting prognostic-related genes, we performed propensity score matching to reduce the selection bias between the high- and low-risk groups. Propensity scores were estimated using age, gender (male versus female), TNM stage (I, II, III, IV), Tumor stage (T1, T2, T3, T4), Lymph Node stage (N0, N1, N2, N3) and Metastasis stage (M0, M1) in TCGA-LUAD cohort. In the same way, propensity scores were estimated using age, gender (male versus female), TNM stage (I, II, III, IV), and TP53 (Wild versus Mutant) in the GEO cohort. The prognostic value and the clinical correlation of the key gene were both validated between the high- and low-risk groups in TCGA and GEO cohort (GSE116959). The time-dependent ROC curve analyses were conducted to evaluate the predictive power of the key gene. The mRNA expression level of the key gene in various types of cancers was identified in the Oncomine database [16]. The mRNA and protein expression of the key gene in LUAD were determined using the Gene Expression Profiling Interactive Analysis (GEPIA) and The Human Protein Atlas (HPA) database [17, 18]. To verify the correlation between ferroptosis and LUAD outcome, we also analyze the survival value of GPX4 in the LUAD cohort, which is the master regulator of ferroptosis [19].

We analyzed the commonly used chemotherapy drugs, including pemetrexed, cisplatin, gemcitabine, paclitaxel, vinorelbine, docetaxel, doxorubicin, etoposide, erlotinib, and gefitinib [20]. The chemotherapeutic response prediction was made based on the TCGA-LUAD cohort using the “pRRophetic” R package [21]. The half maximal inhibitory concentration (IC50) of patients in different risk groups were compared.

The biological functions and pathways of the key gene were elucidated through the DEGs between the high-risk and low-risk groups. GO enrichment and KEGG pathway analyses [22] were then assessed in DAVID database. The correlation analysis of the key gene with tumor proliferation and cell cycle markers was conducted in GEPIA database [17].

The enrichment levels of immune cells were quantified by the Tumor Purity, Estimate Score, Immune Score, and Stromal Score in each sample. The tumor immune cell infiltration was calculated by Single Sample Gene Set Enrichment Analysis (ssGSEA). Then we analyzed the correlation between the key gene expression and the abundance of infiltrating immune cells (B cells, CD8+ T cells, CD4+ T cells, macrophages, neutrophils, and dendritic cells) via The Tumor IMmune Estimation Resource (TIMER) database [23].

The somatic mutation profiles of LUAD patients were downloaded from TCGA database. The mutation frequency with the number of variants/the length of exons (38 million) were calculated for each sample. The OncoPlot of the top 10 mutated genes was plotted. The detailed mutational information, including the variant classification, the number of variant type, and the single-nucleotide variant (SNV) class, were displayed. Then we assessed the correlation between the key gene expression and the TMB levels.

The mRNA and protein expression of the key gene in LUAD were determined using quantitative real-time PCR (qRT-PCR) and immunohistochemistry. The clinical tissue samples of LUAD were obtained from patients who received surgery in Thoracic Oncology Department of Sun Yat-sen University Cancer Center, which was approved by the Institutional Review Committee of Sun Yat-sen University Cancer Center. The detailed procedure was performed according to strict adherence to the manufacturers’ instructions.

For qRT-PCR, 24 paired LUAD and normal tissues were resected and stored in RNAlater immediately. Total RNA was extracted using TRIzol reagent. cDNA was synthesized from total RNA using cDNA reverse transcription kit (Thermo Fisher Scientific). qRT-PCR was performed using the SYBR Green PCR kit (Thermo Fisher Scientific). The housekeeping gene GAPDH was used as an endogenous control. Primer information: FANCD2: 5′- AAAACGGGAGAGAGTCAGAATCA-3′ (forward) and 5′- ACGCTCACAAGACAAAAGGCA-3′ (reverse); GAPDH: 5′- GGAGCGAGATCCCTCCAAAAT-3′ (forward) and 5′- GGCTGTTGTCATACTTCTCATGG-3′ (reverse). The cycle threshold (Ct) of each gene in samples was recorded. Relative quantification was calculated as 2-ΔCt (ΔCt values = target gene mean Ct value - control gene mean Ct value).

Twenty pairs of LUAD immunohistochemistry samples were fixed using 10% formalin and embedded in paraffin. Immunohistochemistry was carried out using the processed 5 μm continuous sections. Samples were dewaxed with decreasing concentrations of 100, 95, 75, and 50% ethanol and washed in deionized water. The sections were heated in a microwave with TE buffer pH 9.0 to retrieve antigens. Endogenous peroxidase was inhibited by incubation in goat serum. Then they were incubated in rabbit anti-FANCD2 (Proteintech, 204006–1-AP, 1:200) overnight at 4 °C. Next, the sections were incubated with horseradish peroxidase-coupled goat anti-rabbit secondary antibody and stained using DAB Detection Kit (Polymer). The following process is cell nucleus staining, dehydration, xylene infusion, and mounting [24]. The immunohistochemical scores were scored by two independent pathologists. The intensity of FANCD2 expression was scored as zero, negative; one point, weak staining; two points, mild staining; three points, strong staining. The positive stained area percentage (PSAP) of FANCD2 expression was scored as 1, 0–25%; 2, 25–50%; 3, 50–75% and 4, 75–100%. FANCD2 IHC score = Intensity score × PSAP score.

Significance analysis of microarrays was used to screen the differentially expressed genes between the LUAD and normal lung tissues. Univariate Cox proportional hazards model was used to analyze the association between gene expression level and prognosis. Kaplan-Meier method and Log-rank test were used to evaluate the difference between survival curves. The continuous and categorical variables between the two risk subtypes were compared using the two-sided Wilcoxon rank-sum test and chi-square test, respectively. Benjamini-Hochberg method was used to adjust for multiple hypothesis testing. All P values were 2-sided, and P < 0.05 was considered statistically significant.

Statistical analyses and result visualization were performed using R software v3.6.3, v4.0.5 and v4.1.2 (“pheatmap v4.0.5”, “limma v3.6.3 [25]”, “survival v3.6.3”, “survminer v3.6.3”, “ggpubr v3.6.3”, “survivalROC v3.6.3”, “car v3.6.3”, “ggridges v3.6.3”, “genefilter v4.1.2”, “ggpubr v3.6.3”, “pRRophetic v4.1.2”, “ggplot2 v3.6.3”, “colorspace v4.0.5”, “stringi v4.0.5”, “clusterProfiler v4.1.2 [26]”, “enrichplot v4.1.2” and “maftools v4.1.2” R package).

The flow chart of this study is shown in Fig. 1. 316 LUAD patients from the TCGA cohort and 381 LUAD patients from the GEO (GSE116959) cohort were finally enrolled. The detailed clinical and tumor characteristics of the LUAD cohorts are summarized in Table 1. A total of 70 autophagy-dependent ferroptosis-related genes were identified from literatures and showed in Fig. 2 [27‐31]. These genes were classified into ferritinophagy, lipophagy, clockophagy, chaperone-mediated autophagy, and others according to the autophagy type.

Three steps were carried out to screen the key gene. First, 7 DEGs in Fig. 3a were selected. Among the 7 DEGs, CBS, CHAC1, DPP4, FANCD2 and GCLC are highly expressed in LUAD tissues (logFC > 1, P < 0.05). However, ALOX15 and ALOX5 are lowly expressed (logFC < 1, P < 0.05). Second, 8 genes with prognostic values for LUAD were selected (Fig. 3b). Among the eight prognosis genes, the high expression level of KRAS, FANCD2, COPZ1, and CISD1 were related to poor survival of LUAD (Hazard ratio > 1 and P < 0.05). TMEM173, PEBP1, ARNTL and AGER were identified as prognosis protective genes (Hazard ratio < 1 and P < 0.05). Third, the key gene was obtained as the intersection between DEGs and prognosis-related genes using Venn diagrams (Fig. 3c). As a result, FANCD2 was identified as the key gene, which worked as a prognosis-related differentially expressed gene in LUAD.

The FANCD2 expression in different tumors was evaluated using TCGA RNA-sequencing data (Fig. 4). FANCD2 expression was significantly higher in various tumors than adjacent normal tissues, and the consistent findings were shown in LUAD (Fig. 5a). qPCR results provided the biological evidence by using 24 pairs of LUAD and sample lung tissues (Fig. 6a). After examining the mRNA expression level of FANCD2 in LUAD, the protein expression level was further explored by immunohistochemistry. The grading standards of the IHC score are shown in Supplementary Fig. 1. There is a higher IHC score of FANCD2 in LUAD tissues than normal lung tissues (Fig. 6b, c), which is in line with HPA database (Fig. 5b). In summary, the present results indicated that both transcriptional and translational expression levels of FANCD2 were overexpressed in patients with LUAD, which may be involved in the pathogenesis of LUAD.

The LUAD patients were stratified into high and low-risk groups by the median FANCD2 expression level. The low-risk group is well matched by propensity score matching to the high-risk group both in TCGA and GEO cohort (Supplementary Table 3 and Supplementary Table 4). Table 2 shows the association of FANCD2 expression and the clinical features. A high expression of FANCD2 achieved a significant correlation with a high TNM stage (P < 0.05). In the TCGA cohort, the FANCD2 expression was defined as an independent prognostic factor after the univariate and multivariate Cox regression analyses (Fig. 7a). The patients in high-risk groups have a poor survival than the low-risk group in the TCGA cohort.

Similarly, relevant data from a GEO cohort (GSE35570) was used to validate the prognostic value of FANCD2 expression in LUAD (Fig. 7b). And the high expression of ferroptosis regulator GPX4 is related to a better prognosis in LUAD patients both in TCGA (P = 0.006) and GEO database (P = 0.006) (Fig. 7c, d). Besides the poor survival and a high TNM stage, the increased expression of FANCD2 was also related to a high frequency of TP53 mutation (P < 0.001, Table 2). The sensitivity and specificity of the FANCD2 model were calculated by the area under ROC (TCGA cohort: AUC = 0.736, GEO cohort: AUC = 0.677), suggesting that the FANCD2 signature was adequate for predicting survival of LUAD (Fig. 8). We investigated the response to chemotherapy in high- and low-risk patients with LUAD. We found that 29 chemotherapeutic drugs displayed significant differences in estimated IC50 between high and low-risk patients and that high-risk patients showed increased sensitivity to all 29 chemotherapies (Fig. 9).

GO and KEGG pathway enrichment analyses were used to evaluate the possible functions and pathways of the DEGs. There is a total of 80 DEGs between the high and low expression level of FANCD2 in the TCGA-LUAD cohort, which are shown in Supplementary Table 7. The top five GO terms were “response to reactive oxygen species”, “regulation of peptidase activity”, “neutrophil degranulation”, “neutrophil activation involved in immune response”, and “neutrophil activation” (Fig. 10 a). The top five pathways were “phagosome”, “antigen processing and presentation”, “human T-cell leukemia virus”, “Th17 cell differentiation”, and “salmonella infection” by KEGG enrichment (Fig. 10 b).

In the correlation analysis we can see that FANCD2 is associated with tumor proliferation markers, including KI67 (R = 0.59, P < 0.001), PCNA (R = 0.53, P < 0.001), and cell cycle markers, incluidng CDK1 (R = 0.60, P < 0.001), CDK2 (R = 0.57, P < 0.001), CDK4 (R = 0.21, P < 0.001), CDK6 (R = 0.23, P < 0.001) (Fig. 11).

The potential immune mechanisms of LUAD were further explored through the scoring of tumor immune components (TumorPurity, ESTIMATEScore, ImmuneScore, StromalScore) and immune infiltrating cells counted according to immunity-enriched groups (Fig. 12). Each patient in the LUAD cohort was scored by the above indicators. The immune cell infiltration levels changed along with the FANCD2 gene copy numbers (Fig. 13). The LUAD patients with a high expression level of FANCD2 had a low ESTIMATEScore (Fig. 13a), ImmuneScore (Fig. 13b), and StromalScore (Fig. 13c), but high TumorPurity (Fig. 13d) was found in the high expression of FANCD2. Neutrophil cell infiltration levels seemed to associate with altered FANCD2 gene copy numbers in LUAD positively (cor = 0.15, P < 0.001, Fig. 14), which is consistent with the GO results of neutrophil degranulation, neutrophil activation involved in immune response, and neutrophil activation in Fig. 10a.

The most frequent driver mutations of the LUAD cohort were displayed in oncoplot, where various colors with annotations represented the different mutation types (Fig. 15). Then, the top 5 mutated genes in LUAD with ranked percentages were exhibited, including P53 (47%), TTN (41%), MUC16 (40%), RYR2 (34%), and CSMD3 (34%). These mutations were further classified according to different mutation categories. Findings indicated that missense mutation accounted for the most fraction (Fig. 16a), and single nucleotide polymorphism (SNP) occurred more frequently than insertion or deletion (Fig. 16b), and C > A was the most common single nucleotide variants (SNV) in LUAD (Fig. 16c). The LUAD patients with a high expression level of FANCD2 showed a higher TMB (P < 0.001, Fig. 16d), which suggested that FANCD2 could work as a TMB marker and play a role in prediction of response to immunotherapy.