The tumor immune estimation resource database (TIMER; https://cistrome.shinyapps.io/timer/) (9) and the UALCAN database (http://ualcan.path.uab.edu/index.html) (10) were used for the differential expression analysis of ACADSB between tumor tissues and corresponding normal tissues. TIMER was also applied to explore the association between ACADSB expression and overall survival (OS) in different types of cancers.

The RNA sequencing (RNA-seq) data and clinical information of patients with ccRCC (TCGA-KIRC) were downloaded from The Cancer Genome Atlas (TCGA) using the “TCGAbiolinks” R package (11). Only patients with both RNA-seq data and valid clinical information were included in this study, and duplicated samples were excluded. The expression profiling microarray data of GSE36895 (with 29 ccRCC samples and 23 adjacent normal samples) and GSE53757 (with 72 paired ccRCC and adjacent normal samples) were downloaded from Gene Expression Omnibus (GEO) using the “GEOquery” R package (12).

A tissue chip with 90 pairs of ccRCC and corresponding normal tissues was purchased from Outdo Biotech Co., Ltd. (HKidE180Su03, Shanghai, China). The experiment received ethical approval for sample use from Shanghai Outdo Biotech Co., Ltd. (Barcode: YB M‐05‐02). IHC was performed on tissues fixed with formaldehyde and embedded in paraffin wax. After deparaffinization and rehydration, the endogenous peroxidase activity was blocked and antigen retrieval was performed. The ACADSB antibody (13122‐1‐AP, 1:12000; Proteintech, CA) was incubated overnight at 4°C. After careful washing and incubation with the specified horseradish peroxidase (HRP)‐conjugated secondary antibody, ACADSB expression was detected using 3,3N‐ diaminobenzidine tetrahydrochloride (DAB).

The intensity and extent of ACADSB staining were evaluated by two experienced pathologists. The method for calculating the score of ACADSB staining was as follows: the extent of staining in an ×200 field was scored as 0, 0%; 1, 1–25%; 2, 26–50%; and 3, 51–100%. The intensity of staining was scored as 0, no signal; 1, light brown; 2, brown; and 3, dark brown. The final score of each field was the average obtained from the two pathologists by multiplying the extent score by the intent score. The scores of ACADSB staining were categorized as follows: low expression (−/+) for scores 0–1 (−) and 2–3 (+) and high expression (++/+++) for scores 4–6 (++) and 7–9 (+++). All evaluations were performed using a Leica DM4000 M microscope.

Gene-set enrichment analysis (GSEA) (13) was performed with the GSEA software (version 4.0.3). Samples from TCGA were divided into two groups based on the expression of ACADSB. The Broad Molecular Signatures Database (MSigDB v6.0) (14) set H (hallmark gene sets, 50 gene sets) were used, which summarize and represent specific well-defined biological states and pathway processes. Enrichment analysis was performed by default weighted enrichment statistics, with the random combinatorial count set as 1,000. Gene sets were judged as significantly enriched by P<0.05 as well as false discovery rates (FDR) < 0.25.

The genes which had greater than 0.4 Spearman correlation coefficient with ACADSB in expression level were defined as ACADSB-related genes. To explore the functional annotation and involved pathways of ACADSB-related genes, the gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were executed via the “clusterprofiler” package (15).

All statistical analyses were performed in R version 4.1.0. For RNA-seq data, raw counts data were used for differential expression analysis via the “DESeq2” package (16). Fragments per kilobase million (FPKM) values were converted to Transcripts per million (TPM) and log2 transformed for further analysis. For microarray data, differential expression analysis was performed using the “limma” package (17). Genes with an adjusted p value of less than 0.05 and fold change (FC) greater than 2 were regarded as differentially expressed genes (DEGs). The receiver operating characteristic (ROC) curves were plotted and areas under the curve (AUC) were calculated to investigate the diagnostic performance of ACADSB. Patients were stratified into low or high groups based on ACADSB expression, using the median expression as the cut-off value. The relationships between clinical pathologic characteristics and ACADSB expression were analyzed with Chi-squared test or Wilcoxon test. Univariate and multivariate Cox proportional hazards models were used to compare the influence of ACADSB expression on OS along with other clinicopathological parameters. P< 0.05 was considered to indicate a statistically significant difference.

Differential expression analysis using TIMER showed that ACADSB is down-regulated in multiple types of cancer, including bladder cancer (BLCA), breast cancer (BRCA), cholangiocarcinoma (CHOL), colon adenocarcinoma (COAD), esophageal carcinoma (ESCA), head and neck squamous cell carcinoma (HNSC), kidney chromophobe (KICH), kidney renal clear cell carcinoma (KIRC), kidney renal papillary cell carcinoma (KIRP), liver hepatocellular carcinoma (LIHC), lung adenocarcinoma (LUAD), lung squamous cell carcinoma (LUSC), rectum adenocarcinoma (READ), stomach adenocarcinoma (STAD), thyroid carcinoma (THCA), and uterine corpus endometrial carcinoma (UCEC; Figure 1A ). Proteome data derived from the UALCAN database also suggest that ACADSB is down-regulated in breast cancer, colon cancer, ccRCC, and UCEC ( Figure 1B ). In addition, Kaplan–Meier analysis indicated that low expression of ACADSB is associated with poor prognosis in BRCA luminal subtype, COAD, KIRC, brain lower grade glioma (LGG), and mesothelioma (MESO) ( Figures 1C–E and Supplementary Figures 1A, B ).

Transcript differential expression analysis using DESeq2 revealed that ACADSB was significantly down-regulated in ccRCC tissues compared with normal kidney tissues in TCGA-KIRC (log2FC = -2.0, FDR < 0.001; Figure 2A ), GSE36895 (log2FC = -1.7, FDR < 0.001; Figure 2B ), and GSE53757 (log2FC = -2.2, FDR < 0.001; Figure 2C ). Further, ROC analyses showed that ACADSB expression could be a single significant parameter for discriminating between normal and tumor tissues in TCGA-KIRC (AUC = 0.952, 95% CI = 0.926–0.977, Figure 2D ), GSE36895 (AUC = 0.931, 95% CI = 0.862–1.0, Figure 2E ), and GSE53757 (AUC = 0.966, 95% CI = 0.935–0.998, Figure 2F ). A total of 84 paired tumor–normal samples were included in the IHC analysis, with six pairs of samples excluded due to slices escaping from the glass slide. For normal kidney tissues, most samples (75/84, 89.3%) showed high ACADSB expression, while all ccRCC samples (84/84, 100%) showed low ACADSB expression ( Figure 2G ). Typical pictures of staining in paired tumor and normal samples are shown in Figures 2H, I .

In the TCGA-KIRC cohort, there’re 530 cases with both RNA-seq data and clinical information. Four duplicated cases and 4 cases whose overall survival time was 0 days were excluded. Therefore, a total of 522 patients with both RNA-seq data and valid survival information were included in this study. The detailed clinical characteristics are listed in Table 1 . Patients were divided into a high expression group and a low expression group based on ACADSB expression. Among the 522 participants, 341 (65.3%) were men and 181 (34.7%) were women, with a median age of 61 years at the time of initial diagnosis; 315 (60.7%) patients had stage I or stage II disease pathology, and 204 (39.3%) patients had stage III or stage IV disease pathology. Regarding histological grade, 236 (45.9%) patients were categorized into grade 1 or grade 2, and 278 (54.1%) patients were categorized into grade 3 or grade 4. Follow-up duration was 40 months on average (range, 0.1–151.2 months). At the time of the last follow-up, 344 (68.9%) patients were tumor-free, while 155 (31.1%) patients still had tumors.

The relationship between ACADSB expression and clinicopathological characteristics is shown in Figure 3 . ACADSB expression was significantly associated with pathological T stage (T1-2 vs. T3-4, P< 0.001; Figure 3A ), pathological N stage (N0 vs. N1, P< 0.01; Figure 3B ), pathological M stage (M0 vs. M1, P< 0.001; Figure 3C ), pathological stage (stage I-II vs. stage III-IV, P< 0.001; Figure 3D ), tumor grade (grades 1–2 vs. grades 3–4, P< 0.001;  Figure 3E ), tumor status (tumor free vs. with tumor, P< 0.001; Figure 3F ), and sex (P< 0.01; Figure 3G ). The differences between groups stratified by age (<65 years vs. ≥65 years; Figure 3H ) and race (white vs. black or African American vs. Asian; Figure 3I ) did not attain statistical significance.

Kaplan–Meier survival analysis showed that lower ACADSB expression was associated with a worse prognosis (P< 0.001; Figure 1E ). The univariate Cox regression model revealed that age at diagnosis (HR = 1.03, 95% CI = 1.017–1.044, P< 0.001), pathological stage (HR = 3.927, 95% CI = 2.847–5.417, P< 0.001), histological grade (HR = 2.679, 95% CI = 1.895–3.787, P< 0.001), and ACADSB expression (HR = 0.421, 95% CI = 0.331–0.535, P< 0.001) were associated with the OS of patients with ccRCC. Multivariate Cox regression after adjustment indicated that age at diagnosis (HR = 1.027, 95% CI = 1.013–1.042, P< 0.001), pathological stage (HR = 2.695, 95% CI = 1.877–3.765, P< 0.001), histological grade (HR = 1.639, 95% CI = 1.137–2.364, P< 0.01), and ACADSB expression (HR = 0.577, 95% CI = 0.446–0.746, P< 0.001) were independent prognostic factors for OS in patients with ccRCC ( Figure 4 and Supplementary Table 1 ).

In the hallmark dataset, 10 gene sets were significantly enriched in the ACADSB high expression phenotype, including TGF-β signaling, androgen response, UV response down, heme metabolism, bile acid metabolism, protein secretion, adipogenesis, FA metabolism, PI3K-AKT-mTOR signaling, and mitotic spindle ( Figure 5 and Supplementary Figure 1C ).

In TCGA-KIRC, GSE36895, and GSE53757, 3093, 2111, and 594 respective genes had ACADSB expression with a Spearman correlation coefficient of >0.4. Two hundred and fourteen genes representing the intersection between all three datasets were defined as ACADSB-related genes ( Figure 6A ). At a false discovery rate (FDR) of 0.05, 47 GO biological processes, 13 GO cellular components, 14 GO molecular functions, and 15 KEGG pathways were significantly enriched for these genes ( Figure 6B and Supplementary Table 2 ).

Ferroptosis-related genes were derived from the FerrDb database (18). A total of 108 genes were identified as ferroptosis driver genes. Correlation analysis revealed that the expression of ACADSB was significantly positively correlated with the expression of 62 ferroptosis diver genes in the TCGA-KIRC dataset ( Figure 6C ). Similar results were found in GSE36895 and GSE53757 ( Supplementary Figure 1D, E ).