Gene expression profiles, including GSE68015 (19) and GSE94349 (20) were obtained from the gene expression omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/). Therefore, the GSE68015 database contains 15 ACP samples and 16 normal brain tissue samples. The GSE94349 database contained 9 ACP and 17 normal brain tissue samples. Using the “sva” R package, we merged the two datasets. There are 24 ACP and 33 normal brain tissue samples. The platform was Affymetrix HG-U133plus2 chips (Platform GPL570). Through the “sva” R package, the two data sets were merged and normalized. To eliminate the batch effect, we normalized the gene expression matrices of the two datasets and removed the genes that were missing from each other. The ComBat function of the SVA package in R was used to eliminate the batch effect.

Limma package of R language was used to screen different expressed genes (DEGs) between 24 ACP and 33 normal brain tissue samples with the threshold of |LogFC| > 2 and P < 0.05. Then, the ggVolcano package and TBtools were used to show the heatmap and volcano plot, respectively. We obtained ferroptosis-related factors from the databases of FerrDb V2 (zhounan. org/ferrdb/current/) and GeneCard [GeneCards - Human Genes | Gene Database | Gene Search (weizmann. ac. il)]. Then, the intersection of DEGs and ferroptosis-related factors was obtained as DEFRGs for further analysis, which was shown by the Venn diagram. Furthermore, protein-protein interaction (PPI) analysis of DEFRGs was conducted based on the STRING website (STRING: functional protein association networks (string db. org) to demonstrate the regulatory relationships.

According to the database of DAVID (http://david.ncifcrf.gov/), the candidate DEFRGs were processed for GO annotations. FunRich tool (http://www.funrich.org/) was undertaken for the Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis. Disease Ontology (DO) enrichment was conducted via OmicShare tools (www.omicshare.com) The enrichment list with a P-value < 0.05 was screened with statistical significance. Msigdb database of gene set enrichment analysis (GSEA, gsea-msigdb.org) was used to analyze the enriched KEGG pathways. We used the GSVA package of R software to analyze up- and down-regulated DEGs.

Least absolute shrinkage and selection operator (LASSO) and support vector machine recursive feature elimination (SVM-RFE) are two important algorithms of machine learning techniques. Lasso linear regression (21–23) is capable of feature selection and regularization. The basic principle is to add the absolute value of the model weight coefficients to the loss function. The overfitting of the model is prevented and the generalization ability of the model is increased by reducing the weights and lowering the weight values. The SVM-RFE algorithm (24) is a combination of SVM and RFE algorithms. Importance assessment of features was performed by SVM model and unimportant features were eliminated using the iterative process of RFE. The SVM-RFE algorithm was based on the principle of maximum interval of SVM, and scores of each feature are ranked by model training samples. The features with the smallest feature scores were removed using the RFE algorithm in a step-by-step iterative manner. Then, the remaining features were used to train the model again for the next iteration, and the optimal feature combination was selected. In this study, we used these algorithms to screen DEFRGs. e1071 and glmnet packages were used for SVM-RFE and LASSO algorithms, respectively. The screened genes were characterized by higher discriminative power and variable, respectively. The overlapped genes were obtained as key DEFRGs. Correlation analysis between DEFRGs was undertaken, and visualized by ggplot 2.

The study was approved by the Medical Ethics Committee of The Seventh Affiliated Hospital of Sun Yat-sen University (2019SYSUSH-020). A total of 35 ACP patients were included. All patients were hospitalized and underwent surgery in our hospital from Mar 2018 through May 2022, and they did not receive radiotherapy and chemotherapy treatments. A total of 10 normal brain samples as controls were obtained by autopsies. All patients or their family were informed and consented. ACP and control tissues were obtained and refrigerated at −80°C.

Purlink RNA Mini assay kit (Thermo Fisher Scientific, Waltham, MA, USA) was used for isolating the total RNA. Then, the obtained RNA was reversed and transcribed into complementary DNA (cDNA). The primers were designed and obtained from Sangon Biotech (Shanghai, China). These are shown as follows:

The SuperReal PreMix Color kit (Tiangen, Beijing, China) was applied to detect the expressions of the above genes. The 2^-ΔΔCt was applied to show the expression levels of key DEFRGs. β-actin was used as a reference.

Ferroptosis markers, including SLC40A1, NFE2L2, HSPB1, GPX4, CHAC1, PTGS2, TF, TFRC, and FTH1 were detected using ELISA kits (Multi Sciences Co., Ltd., China) according to the instructions. The steps were as follows. A total of 20 mg of fresh tissue was obtained and used to detect the expression levels of ferroptosis markers, including SLC40A1, NFE2L2, HSPB1, GPX4, CHAC1, PTGS2, TF, TFRC, and FTH1. The homogenizing medium was added and centrifuged at 4°C, 1500×g for 10 min. The middle layer was aspirated for the experiment. Specific anti-mouse antibodies were pre-coated on an enzyme-labeled plate. The wells of the plate were filled with a standard, the sample to be tested, and a biotinylated detection antibody. After incubation, the unbound antibody was removed and horseradish peroxidase-labeled streptavidin (Streptavidin-HRP) was added. TMB was added and the color was developed avoiding light. After color development, samples were detected by Automatic Microplate Reader (MuhiskanMk3, Thermo Labsystems Ltd., USA).

R (version 4.1.2) was used for statistical analyses. Receiver operating characteristic (ROC) curves were applied to validate the accuracy of machine learning algorithms. Student t-test was used for comparison between the two groups. P < 0.05 was the threshold of significance. All experiments were repeated more than 3 times. According to the multivariate Cox proportional hazards regression model, a nomogram was drawn to predict the relationship between key DEFRG expression and risk of disease. The disease groups were clustered by consensus clustering analysis. The correlation analysis between 6 genes with 9 Ferroptosis related markers, including SLC40A1, NFE2L2, HSPB1, GPX4, CHAC1, PTGS2, TF, TFRC, and FTH1 (25), was calculated using the Pearson correlation method. The statistical software used was Statistical Package for the Social Sciences (SPSS) 20.0.

GSE68015 and GSE94349 were obtained from the GEO database (Home - GEO - NCBI (nih.gov)). GSE68015 contains 15 ACP and 16 normal brain tissue samples. GSE94349 contains 9 ACP and 17 normal brain tissue samples. Using the “sva” R package, we merged the two datasets and normalized them. According to the threshold of | LogFC |>2 and P < 0.05, differential analysis was conducted on 24 ACP and 33 normal brain tissue samples. We obtained 2419 DEGs, including 1181 upregulated DEGs and 1238 downregulated DEGs. The heatmap and volcano plot of these DEGs are shown in Figures 1A, B . The intersection of DEGs and ferroptosis-related factors was shown by the Venn diagram. A total of 579 ferroptosis-related factors were obtained from the FerrDb V2 and GeneCard databases ( Figure 1C ). Venn diagram showed that a total of 69 DEFRGs were identified. Then, PPI network analysis was processed to show the regulated relationship of these DEFRGs ( Figure 1D ).

According to the results of GO functional and KEGG pathway enrichment analysis, DEGs were enriched in various functions, such as synapse organization, synaptic membrane, neuronal cell body, and channel activity ( Figures 2A, B ). KEGG pathways enriched by DEGs included circadian entrainment, ECM-receptor interaction, GABAergic synapse, and glutamatergic synapse ( Figures 2C, D ). As shown in Figures 2E, F , these DEGs were enriched into many diseases, including pervasive developmental disorder, ovarian cancer, renal cell carcinoma, and cell-type benign neoplasm. Based on the results of the GSEA analysis, critical pathways were obtained. A total of 5 KEGG pathways were enriched, including calcium signaling pathway, cardiac muscle contraction, long-term depression, long-term potentiation, and neuroactive ligand-receptor interaction. Besides, there were 5 KEGG pathways obtained in the disease, including complement and coagulation cascades, cytokine receptor interaction, ECM receptor interaction, hemopoietic cell uncage, and pathways in cancer ( Figure 2G ). Then, GSVA analysis was processed for up- and down-regulated DEFRGs. The up-regulated DEGs were enriched in various pathways, such as glycine serine and threonine metabolism, alanine aspartate and glutamate metabolism, histidine metabolism, and tight junction. While down-regulated DEGs were involved in different KEGG pathways, including aminoacyl trna biosynthesis, spliceosome, and riboflavin metabolism ( Figure 2H ).

SVM-RFE and LASSO algorithms were used for the identification of variable DEFRGs with higher discriminative power. As the results are shown in Figures 3A, B , a total of 11 and 10 features were obtained by SVM-RFE and LASSO algorithms, respectively. There were 6 overlapped DEFRGs were obtained, including CASP8, GOT1, KRT16, KRT19, TFAP2C, and TP63 ( Figure 3C ). The ROC curves of the six diagnostic markers are shown in Figure 3D . Their AUC values showed the accuracy of the machine learning algorithm. These Overlapped DEFRGs were located on chromosomes 2, 3, 10, 17 and 20 ( Figure 3E ). The correlation between 6 DEFRGs was calculated. GOT1 was negatively correlated with other DEFRGs, while KRT16, KRT19, and TFAP2C were with positive regulate relationships ( Figure 3F ).

These 6 key DEFRGs were differentially expressed in the normal and ACP disease groups based on GSE68015 and GSE94349 profiles. Figures 4A–F showed that all five markers, including CASP8, KRT16, KRT19, TFAP2C, and TP63, were upregulated in the disease group, while GOT1 was downregulated in the ACP group.

In ACP, total key DEFRGs as diagnostic markers were screened by machine learning. The risk profile of the disease was predicted by the expression levels of these six regulators. Based on the results of the nomogram, CASP8, KRT16, KRT19, and TP63 were the protective factors of the disease, while GOT1 and TFAP2C were the risk factors ( Figure 5 ). Except for TFAP2C, the other 5 DEFGRs were composite clinical expression features. CASP8, KRT16, KRT19, TP63, and GOT1 were potential biomarkers for ACP disease.

The samples in the dataset were clustered according to the screened DEFRGs. When the k value was set to 2, the consensus clustering matrix was the most differentiated, the number of clusters was the most stable, and the consistency scores across subtypes were the highest. The samples could be categorized into 2 isoforms and heat mapped, which showed that C2 expressed higher ( Figures 6A–E ).

A total of 5 key DEFRGs were differentially expressed in the normal and ACP disease groups. The results of Figures 7A–F showed that CASP8, KRT16, KRT19, and TP63, were upregulated in the disease group, while GOT1 was downregulated in the ACP group. However, TFAP2C was with no significant difference. CASP8, KRT16, KRT19, TP63, and GOT1 might be the biomarker for ACP diagnosis.

The correlation of the six key DEGs with nine ferroptosis marker genes is shown in Figure 8A . GPX4, HSPB1, NFE2L2, SLC40A1, CHAC1, and HSPB1 were significantly related to key DEGs. Elisa assay was used for detecting the expression levels of 6 ferroptosis marker genes. CHAC1 and GPX4 expressed lower in ACP disease group than control, while HSPB1, NFE2L2, SLC40A1, and HSPB1 were up-regulated in the disease group ( Figure 8B ). A combination of CASP8, KRT16, KRT19, TP63, and CASP8, GOT1 might be the biomarker for ACP diagnosis via participating ferroptosis process.