Identification and validation of mitochondrial metabolism-ralated biomarkers in coronary heart disease

Abstract

Coronary heart disease (CHD) affects life quality of patients by impaired coronary artery blood supply. We were planning to study the molecular mechanisms of mitochondrial metabolism-related genes (MMRGs) in CHD. The following data were sourced from public databases: transcriptome data of CHD and MMRGs. The candidate genes were obtained by differential expression analysis and MMRGs. The identification of biomarkers was facilitated by machine learning algorithms and gene expression analyses. Of particular significance was the utilization of the nomogram for the evaluation of the diagnostic efficacy of the biomarkers. Following this, enrichment analysis, immune infiltration analysis, compound prediction and molecular docking were performed. Expression levels of biomarkers were tested in vitro by reverse transcription quantitative polymerase chain reaction (RT-qPCR). Palmitoyl protein thioesterase 2 (PPT2) and Mediator complex subunit21(MED21) were validated as biomarkers. A nomogram developed utilizing these biomarkers demonstrated a satisfactory capacity for differentiating among various sample types. The neuroactive ligand-receptor interaction, polycomb repressive complex, protein processing in endoplasmic reticulum, and calcium signaling pathway were the pathway co-enriched by biomarkers. Immune infiltration analysis indicated that PPT2 and MED21 were anticorrelated with regulatory T cells and central memory CD4 T cells, respectively. In addition, 20 compounds targeting PPT2 and MED21 were identified, respectively. Notably, molecular docking studies demonstrated a strong binding affinity between PPT2 and benzo(a)pyrene. The RT-qPCR analyses confirmed the downregulation of PPT2 and MED21 in CHD. PPT2 and MED21 were identified associated with mitochondrial metabolism in CHD, providing effective support for clinical diagnosis of CHD.

1. Introduction

Coronary heart disease (CHD) is a common chronic noncommunicable cardiovascular disease characterized by impaired blood flow to the coronary arteries, with high morbidity and mortality worldwide.^[1] Due to an aging population and unhealthy lifestyles, its morbidity and mortality rates continue to rise and it represents a leading cause of death and poses a significant global public health concern, accounting for approximately 33% of deaths worldwide.^[2,3] Coronary atherosclerosis originates from the intima of the coronary artery. Its pathological process involves early endothelial dysfunction and lipid deposition, medium-term atherosclerotic plaques and fibrous plaques, and late composite plaques such as calcium salt deposition and calcified plaque formation. Over time, these atherosclerotic plaques will lead to coronary artery stenosis or occlusion, reducing blood flow capacity, and ultimately leading to CHD.^[4–6] In addition, if coronary atherosclerosis develops into plaque rupture or thrombus formation, blood clots can completely block blood flow to the heart, leading to adverse cardiovascular events such as myocardial ischemia, heart failure, myocardial infarction, stroke, hypoxia, or cardiac necrosis.^[7] A variety of risk factors, including hypertension, hyperlipidemia, diabetes, obesity and family genetics factors, are involved in the complicated pathopsysiology of CHD.^[8–10] Although the existing drugs and surgical treatments have improved the prognosis of patients to a certain extent, there are still limitations such as difficulty in curing the disease, poor treatment of high-risk patients, and postoperative complications. In recent years, studies have shown that the modification of multiple genes and signaling pathways significantly contributes to the progression of CHD, but the functional significance of these genes and their potential clinical utility remain to be further explored.^[10,11] Therefore, finding potential biomarkers and achieving a comprehensive understanding of the pathophysiology of CHD could be essential for both diagnosis and treatment.

Mitochondria are organelles responsible for cell metabolism and energy production in eukaryotic cells, and play a key role in maintaining cell homeostasis. Mitochondria produce adenosine triphosphate (ATP) through oxidative phosphorylation as the main energy source of cells. In addition, mitochondria also play a crucial role in essential processes like apoptosis, calcium signaling and reactive oxygen species production. Mitochondrial dysfunction and metabolic disorders are closely related to numerous diseases, such as metabolic disorders, cancer and neurodegenerative diseases.^[12–14] In cardiovascular diseases, the reprogramming of mitochondrial metabolism has been extensively studied, especially in cardiomyocytes with high energy consumption. Studies have found that the normal operation of mitochondrial metabolism is the basis of heart health, and metabolic dysfunction often leads to the occurrence and development of cardiovascular disease (CVD).^[15] In CVD such as atherosclerosis and heart failure, the change of metabolic pathway is manifested as the transformation from glucose oxidative phosphorylation to glycolysis and fatty acid oxidation. This metabolic reprogramming is crucial in disease progression.^[16,17] Mitochondrial metabolic abnormalities may significantly contribute to the development of cardiovascular disease (CVD), but the mechanism of action regarding the onset and progression of CHD remains unclear.^[15] Therefore, further exploring the mechanism of mitochondrial metabolism-related genes (MMRGs) in CHD will provide new strategies for understanding the etiology of the disease.

In summary, in order to completely comprehend the potential mechanism of MMRGs in CHD, this study aims to identify candidate genes related to MMRGs in CHD patients by transcriptome analysis, and combine machine learning algorithms, ROC curve and expression level verification to obtain palmitoyl protein thioesterase 2 (PPT2) and MED21 as MMRGs-related biomarkers in CHD. PPT2 may regulate the activity and function of related proteins by hydrolyzing palmitoylation modification, thus further participate in the regulation of various physiological processes in cells, such as membrane transport, signal transduction, etc.^[18] During embryonic development, MED21 is involved in regulating the expression of genes related to cell differentiation and tissue formation, which is essential for normal embryonic development.^[19] Furthermore, We conducted in-depth analysis to explore the influence of the 2 biomarkers on the pathogenesis of CHD and possible mechanisms, offering a fresh theoretical basis for the diagnosis and prevention of the disease.

2. Material and methods

2.1. Data collection

The data were obtained from the GEO database (https://www.ncbi.nlm.nih.gov/geo/). The GSE113079 dataset (Sequencing platform: GPL20115) was employed as a training set, which included 93 CHD complete blood specimens and 48 complete blood specimens (Table S1, Supplemental Digital Content, https://links.lww.com/MD/R342). The GSE42148 dataset (Sequencing platform: GPL13607) served as a validation set, which comprised 13 CHD complete blood specimens and 11 complete blood specimens(Table S2, Supplemental Digital Content, https://links.lww.com/MD/R342). Furthermore, the Molecular Signatures Database (MSigDB, https://www.gsea-msigdb.org/gsea/msigdb/) was employed to identify 1234 MMRGs(Table S3, Supplemental Digital Content, https://links.lww.com/MD/R342). The study workflow can be found in Figure S1, Supplemental Digital Content, https://links.lww.com/MD/R343

2.2. Differential expression analysis

To determine which differentially expressed genes (DEGs) are expressed differently in CHD and control groups, the analysis of DEGs was carried out by employing the limma package (v 3.54.0)^[20] based on training set GSE113079 for both groups of samples (CHD vs control) (|log₂fold change| >1, adj.P-value <.05). Correction was performed using the Benjamini-Hochberg method. Moreover, the ggplot2 package (v 3.4.1)^[21] was utilized to construct a volcano plot of DEGs, and the top 10 genes (sorted by |log₂ fold change| from high to low) with significant up- and down-regulation differences were labeled, and then the heatmap of DEGs was drawn using the ComplexHeatmap package (v 2.14.0).^[22]

2.3. Identification of candidate genes

In order to identify candidate genes for further study, the ggvenn package (v 0.1.9) (https://CRAN.R-project.org/package=ggvenn) was utilized in order to obtain the intersection genes between DEGs and MMRGs. Following this, gene ontology and kyoto encyclopedia of genes and genomes analyses were conducted to determine the potential biological functions and mechanisms associated with the candidate genes, employing the clusterProfiler package (v 4.2.2)^[23] (P <.05). Furthermore, a confidence score of ≥0.15 was employed in the upload of the candidate genes into the STRING database (https://string-db.org/). A protein-protein interaction network was developed and visualized utilizing Cytoscape software (v 3.8.2)^[24] for the candidate genes.

2.4. Machine learning, ROC analysis and expression validation

In order to further refine the selection of candidate genes, we performed Least absolute shrinkage and selection operator (LASSO), Boruta and support vector machine-recursive feature elimination (SVM-RFE) analysis on training set GSE113079. The glmnet package (v 4.1-8)^[25] was employed to perform LASSO regression of candidate genes, thereby identifying LASSO feature genes. To determine the optimal regularization strength, we implemented 10-fold cross-validation. Each iteration utilized 9 folds of data as the training set, with the remaining 1-fold serving as the validation set. This process was repeated 10 times, recording the deviance for each fold. Based on the cross-validationresults, both lambda.min and lambda.1se thresholds were ultimately output. This approach balances precision and simplicity while mitigating overfitting. The Boruta package (v 8.0.0)^[26] was used to extract Boruta-signature genes through creating confirmed features. This method employs an out-of-bag evaluation based on random forests, utilizing approximately one-third of untrained samples for validation in each iteration. This inherently achieves the effect of 3-fold cross-validation. Parameters are set to 500 iterations with doTrace = 2. Through repeated resampling of both the sample and feature spaces, combined with the bagging mechanism, it effectively enhances model stability. This approach thus achieves highly robust feature selection whilst avoiding overfitting. An SVM-RFE of candidate genes was conducted using the e1071 package (v 1.7-16).^[27] A nested cross-validation strategy was employed: the outer layer divided the samples into 10 folds, with 9 folds selected for each round of inner 10-fold RFE feature screening, while the remaining fold served as validation. The inner RFE iteratively removes features based on SVM weights while calculating error rates. After synthesizing results from ten validation rounds, the feature subset with the lowest average error rate is selected as the optimal gene combination. This approach effectively reduces model volatility through dual cross-validation and multiple averaging, ensuring robustness and generalisability in feature selection outcomes. Furthermore, ggvenn package (v 0.1.9) was applied to obtain the intersection genes between 3 signature genes, which were defined as signature genes. Subsequently, based on the GSE113079 and GSE42148 datasets, ROC curves were plotted using the pROC software package (v. 1.18.0).^[28] The confidence intervals for the area under the curve (AUC) were calculated via the DeLong test to evaluate the target genes’ ability to distinguish CHD samples from control samples. The AUC values were derived, genes exhibiting an AUC value exceeding 0.7 were identified as potential biomarkers. An investigation was conducted to compare the expression of candidate biomarkers in the CHD and conotrol groups, utilizing the GSE113079 and GSE42148 datasets. The Wilcoxon rank-sum test was employed to analyze the datasets (P < .05). Biomarkers were identified as genes exhibiting significant inter-group differences and consistent expression trends across both datasets.

2.5. Developing and assessing a nomogram

Within the confines of the training set GSE113079, a nomogram model was formulated, underpinned by biomarkers, utilizing the rms package (v 6.8-1) (https://CRAN.R-project.org/package=rms) in order to determine the likelihood of the occurrence of CHD. A calibration curve was generated using the regplot program (v 1.1)^[29] and was utilized to assess the accuracy of the nomogram. Finally, with the ggDCA package (v 1.1),^[30] the decision curve analysis (DCA) was carried out (net benefit above 0 indicated good model predictions).

2.6. Chromosome localization, correlation and subcellular localization analyses of biomarkers

In this study, the RCircos package (v 1.2.2)^[31] was used to determine the chromosome localization of biomarkers. The present study employed Spearman correlation coefficient to analyze the relationship between the biomarkers in all samples of the training set, with the correlation matrix plotted using the psych package (v 2.1.6)^[32] (|r| > 0.3, P <.05).

In the field of biological processes, the RNA molecule is observed to manifest specific functional characteristics depending on its subcellular location. This observation underscores the notion that subcellular localization plays a pivotal role in RNA biological functions. In this study, we initially queried the FASTA sequences of biomarkers from the Protein database (https://ncbi.nlm.nih.gov/protein/). Subsequently, these FASTA sequences were input into the CELLO database (http://cello.life.nctu.edu.tw) to obtain predicted scores for their subcellular localizations, and the definitive and precise localization was established based on the highest score.

2.7. Gene set enrichment analysis and gene set variation analysis (GSVA)

The initial step in the research was to calculate the Spearman correlation coefficients between the various biomarkers and every single gene across the entire range of samples from the training set GSE113079, with the psych package (v 2.1.6) being utilized for the purpose. Following that, the sorted genes were arranged in descending order according to their respective Spearman correlation coefficients, and this ranking was used as the gene set that was tested in further analyses. Meanwhile, reference gene set (c2.cp.kegg.v7.0.entrez.gmt) was retrieved from the MSigDB database. Subsequently, the GSEA was performed employing the clusterProfiler package (v 4.2.2), with a threshold of |normalized enrichment score| > 1, adjust P < .05, and false discovery rate < 0.25.

The target gene set, designated as c2.cp.kegg.v7.0.entrez.gmt, was retrieved from the MsigDB database. The ssGSEA algorithm, implemented in the GSVA program (v 1.42.0)^[33] was used to evaluate the GSVA scores for each gene set across different samples. The comparison of GSVA scores between CHD and control groups was conducted utilizing the limma package (v 3.54.0) (|t|>2, P <.05).

2.8. Immune infiltration analysis

The ssGSEA algorithm, as implemented within the GSVA package (v 1.42.0), was employed to calculate the infiltration scores of 28 immune cells^[34] for each sample in the training set. Moreover, the Wilcoxon test was conducted to identify immune cells exhibiting notable disparities between CHD and control groups (P <.05). Next, Spearman correlation analysis was performed on the differential immune cells, as well as between biomarkers and immune cells using the psych package (v 2.1.6), with a threshold set at |r| >0.3 and P <.05.

2.9. Regulatory network construction

To further elucidate the underlying mechanisms of CHD biomarkers, 3 networks were constructed: transcription factor (TF)-mRNA network, TF/microRNA (miRNA)-mRNA network, and long noncoding RNA (lncRNA)-miRNA-mRNA network. The KnockTF database (http://www.licpathway.net/KnockTF/index.html) and CHIPBase database (http://rna.sysu.edu.cn/chipbase/) were utilized to predict the TFs corresponding to biomarkers, and the intersection of the predictions from the 2 databases was taken as the key TFs. The miRNAs that may target the biomarkers were predicted using the miRDB database, with the top 20 miRNAs of each biomarker identified based on the target score. Subsequently, miRNet database (https://www.mirnet.ca/) was applied to forecast lncRNAs that acted upstream of the miRNAs. The above regulatory networks were visualized by Cytoscape software (v 3.8.2).

2.10. Toxicological interactions and molecular docking

In order to identify compounds that might interact with biomarkers, potential compounds candidates targeting these genes were analyzed within the framework of the Comparative Toxicogenomics Database (CTD, https://ctdbase.org/), with the top 20 potential compounds of each biomarker identified based on the interaction count. The result was then analyzed through the utilization of the Cytoscape software (v 3.8.2) to visualize biomarker-compound interaction network. Based on the relationships between biomarker-compound pairs, the compound demonstrating the highest interaction strength score with each biomarker was chosen as the key compound for molecular docking.

To evaluate the binding affinity between key compounds and biomarkers, molecular docking of the aforementioned components was conducted. The structure files of key compounds were sourced from the PubChem database (https://pubchem.ncbi.nlm.nih.gov/) and protein data for the biomarkers were extracted from the protein data bank database (http://www.rcsb.org). Subsequently, molecular docking was performed using the AutoDock software (v 1.6.2).^[35] Furthermore, the binding energy of top 9 binding site of the complexes was determined in order to evaluate the binding mode and the structural stability among the complexes. The binding activity was deemed to be satisfactory if the affinity was less than −5.0 kcal/mol. Finally, the sites with minimal binding energy were visualized using PyMOL software (v 3.0.3).^[36]

2.11. Reverse transcription quantitative polymerase chain reaction

The experimental design comprised the collection of 5 CHD blood samples and 5 control blood samples, with these being obtained respectively from CHD patients and individuals of a healthy demographic. The patients’ clinical information can be found in Table S4, Supplemental Digital Content, https://links.lww.com/MD/R342. In accordance with the manufacturer’s instructions, total RNA extraction was conducted from 10 samples using TRIzol (Ambion, Austin). The SureScript First-Strand cDNA Synthesis Kit (Servicebio, Wuhan, China) was utilized for the reverse transcription of total RNA into cDNA. RT-qPCR was conducted by means of the 2xUniversal Blue SYBR Green qPCR Master Mix (Servicebio, Wuhan, China). The primer sequences utilized for the PCR were delineated in Table 1. The expression of glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was used as an internal reference gene, and 2^-ΔΔCt method was used to measure.^[37] Unpaired t-tests were employed for the purpose of statistical analysis. GraphPad Prism 5 was used to comparing the expression levels of biomarkers between CHD samples and control samples (P <.05).

Table 1.

The primer sequences utilized for the PCR.

Primer	Sequence
PPT2 F	GAGTTGGAGGCGGGACTTC	–	–
PPT2 R	GATCGAGCACTGTCACCACA	–	–
MED21 F	CAGCTAACCCTACAGAAGTATGC	–	–
MED21 R	CAGCTAACCCTACAGAAGTATGC	–	–
GAPDH F	ATGGGCAGCCGTTAGGAAAG	–	–
GAPDH R	AGGAAAAGCATCACCCGGAG	–	–

GAPDH = glyceraldehyde-3-phosphate dehydrogenase, MED21 = mediator complex subunit21, PCR = polymerase chain reaction, PPT2 = palmitoyl protein thioesterase2.

2.12. Statistical analysis

Bioinformatics analyses were conducted utilizing the R language (v 4.2.2). A Wilcoxon rank-sum test was employed to compare the disparities between 2 groups. A P-value of <.05 was deemed as statistically significant.

3. Results

3.1. The 20 candidate genes were linked to mitochondria metabolism and CHD

Firstly, differential expression analysis revealed 497 DEGs between CHD and control groups, with 171 upregulated and 326 downregulated in the CHD group (Fig. 1A and B). Moreover, a total of 20 shared genes between DEGs and MMRGs were identified and selected as candidate genes (Fig. 1C). The enrichment analysis of the 20 candidate genes revealed associations with 292 gene ontology terms. These terms covered a wide range of functions, such as positive regulation of transcription elongation by RNA polymerase II, core mediator complex, intramolecular oxidoreductase activity, etc (Fig. 1D, Table S5, Supplemental Digital Content, https://links.lww.com/MD/R3423; https://links.lww.com/MD/R342). Additionally, 9 KEGG pathways were identified, including spinocerebellar ataxia, nonalcoholic fatty liver disease, and arginine biosynthesis (Fig. 1E, Table S6, Supplemental Digital Content, https://links.lww.com/MD/R342). Furthermore, a protein-protein interaction network demonstrated interactions among 20 proteins and 16 edges (Fig. 1F). Such as, MED21, HSD3B1, and PIK3R1 demonstrated close interactions with other genes.

Figure 1.

Volcanic plot of the distribution of DEGs between CHD and Control group. Top 10 up and down-regulated genes were shown in the figure (CHD group: 93 whole blood samples; Control group: 48 whole blood samples. |log₂FC| > 1, adj.P-value <.05). (A) Heatmap of 497 DEGs (including Top 10 up and down-regulated genes) between the 2 groups. Red: up-regulation; blue: downregulation. (B) Venn diagram of DEGs and the MMRGs. Green area: 497 DEGs; purpple area: 1234 MMRGs; cross area: 20 common genes. (C) The GO analysis enriched a total of 292 pathways,the top 10 pathways were shown in the figure (P <.05). (D) The KEGG analysis enriched a total of 9 pathways (P <.05). (E) PPI network diagram. (F) Circle: genes; line: the interaction of genes. CHD = coronary heart disease, DEGs = differentially expressed gene, MMRGs = mitochondrial metabolism-related genes, GO = gene ontology, KEGG = Kyoto encyclopedia of genes and genomes, PPI = protein–protein interaction.

3.2. The PPT2 and MED21 were identified as biomarkers

As for the result of LASSO algorithm, when the minimum lambda value was 0.0001, 15 LASSO-signature genes were screened among the candidate genes (Fig. 2A). The SVM-RFE algorithm was utilized to ascertain the optimal number of features, a process which ultimately identified 7 as the optimal number, thus allowing for the selection of 7 SVM-RFE-signature genes among the identified candidate biomarkers (Fig. 2B). Through theBoruta model, 16 Boruta-signature genes were selected from the 20 candidate genes (Fig. 2C). Subsequently, we intersected the genes obtained from the above-mentioned machine learning algorithms and finally found that PPT2, MED21, GCLM, and PNPLA5 could be indicated by all the algorithms, meaning that these genes could be used as signature genes (Fig. 2D).

Figure 2.

Fifteen genes were screened among the candidate genes by LASSO algorithm (10-fold cross-validation). (A) 7 genes were screened among the candidate genes by SVM-RFE algorithm. (B) 16 genes were screened among the candidate genes by Boruta algorithm. (C) Intersected all genes obtained from the mentioned above machine learning algorithms and finally 4 signature genes were screened. (D) The AUC values for signature genes in GSE113079 (E) and GSE42148 (F) datasets (AUC >0.7 indicates the model possesses good discriminatory capability). According to the GSE113079 (G) and GSE42148 (H) datasets, PPT2 and MED21 significantly down-regulated in CHD group (Wilcoxon rank-sum test, P <.05). AUC = area under the curve, LASSO = least absolute shrinkage and selection operator, SVM-RFE = support vector machine-recursive feature elimination.

Subsequently, ROC analysis was performed, which revealed that the AUC values for PPT2 and MED21 both exceeded 0.7 in GSE113079 and GSE42148 datasets (Fig. 2E and F). The results indicated that PPT2 and MED21 exhibited a satisfactory ability to differentiate between CHD patients and healthy people, suggesting their potential diagnostic value. Therefore, PPT2 and MED21 were pinpointed as candidate biomarkers. According to the GSE113079 and GSE42148 datasets, it was found that PPT2 and MED21 were significantly down-regulated in CHD group (P < .05), with consistent expression trends, and these genes were identified as biomarkers (Fig. 2G and H).

3.3. The nomogram model was found to be a more effective tool for predicting CHD

A nomogram was constructed for the purpose of predicting the risk of CHD using the biomarkers (Fig. 3A). It was evident from the calibration curve that there was a minimal discrepancy between the actual and predicted risks of CHD (Fig. 3B). In addition, the DCA revealed that the nomogram demonstrated a high clinical benefit rate (Fig. 3C). The findings indicated that the nomogram had a high predictive accuracy for the progression of CHD.

Figure 3.

The nomogram showed PPT2 and MED21 were used to predict the risk of CHD (A). The discrepancy between the calibration curve and prediction curve (B). The nomogram curve demonstrated a higher clinical benefit rate than the reference curve (C). CHD = coronary heart disease, PPT2 = palmitoyl protein thioesterase2, MED21 = mediator complex subunit21.

3.4. Biomarkers located in cytoplasmic were significantly enriched in multiple signaling pathways

The PPT2 was distributed on chromosome 6, and MED21 was on chromosome 12 (Fig. 4A). An exploration was conducted into the association between the biomarkers, and the results indicated that PPT2 and MED21 were statistically positively correlated (R = 0.58, P <.001) (Fig. 4B). In addition, PPT2 and MED21 were primarily located in the cytoplasm (Fig. 4F). The locational information and correlation of biomarkers facilitated a deeper understanding not only of the distinct impacts on CHD, but also provided a novel perspective for exploring their functional associations.

Figure 4.

Chromosomal location cycle graph of PPT2 and MED21 (A). Pearson correlation analysis showed that the correlation coefficient between PPT2 and MED21 (B). Pathway enrichment analysis by GSVA analysis (|t|>2, P <.05). Top 10 pathways were shown in the figure. Blue: up; red: down (C). The pathways associated with expression of PPT2 and MED21 (|NES| >1, P <.05, and FDR <0.25) (D and E). Subcellular localization scores of PPT2 and MED21 (F). PPT2 = palmitoyl protein thioesterase2, MED21 = mediator complex subunit21, GSVA = gene set variation analysis.

Subsequently, GSEA analysis observed that pathways associated with expression of PPT2 and MED21 were 19 and 5, respectively (Table S7, Supplemental Digital Content, https://links.lww.com/MD/R342). Furthermore, PPT2 and MED21 were both mainly enriched in some pathways, such as polycomb repressive complex and protein processing in endoplasmic reticulum (Fig. 4D and E).

In order to explore the functional distinctions between the CHD and control groups in greater depth, we undertook a GSVA analysis. Pathway enrichment analysis by GSVA revealed significant increase of the common pathway of complement cascade mac formation, regulation of complement cascade mac inhibition, bmp signaling pathway bmp antagonist, and other signaling pathways in CHD group, while fas jnk signaling pathway, gene silencing by methylation of h3k27 and ubiquitination of h2ak119, tnf jnk signaling pathway and other pathways were enhanced in control group (Fig. 4C, Table S8, Supplemental Digital Content, https://links.lww.com/MD/R342).

3.5. A strong correlation was demonstrated between biomarkers and immune cells

The distribution of immunity scores for the 28 immune cell types in CHD and control groups was illustrated in the thermal map, which displayed varying degrees of infiltration for each immune cell type (Fig. 5A). A comparative analysis of immune cell infiltration revealed significant disparities between 21 distinct immune cell types between 2 distinct groups (Fig. 5B). Of these, immature B cells, myeloid-derived suppressor cells (MDSC), and neutrophils were significantly more abundant in the CHD group compared to control group, while activated CD4 T cells and other cells were considerably less prevalent in CHD group than in control group. A further exploration was conducted into the correlation among differentially infiltrated immune cells and biomarkers (Fig. 5C). Moreover, PPT2 was considerably negatively relevant to regulatory T cells (r < −0.3, P <.001), MED21 exhibited a marked positive association with both central memory CD4 T cells and effector memory CD4 T cells (R >0.6, P <.001) (Fig. 5D).

Figure 5.

Twenty eight kinds of immune cell score heatmap (A). The diagram showed 28 kinds of immune cell score.There are significant disparities among 21 distinct immune cell types between 2 distinct groups (Wilcoxon test. Ns indicates nonsignificant, *P <.05, **P <.01, ***P <.001, ****P <.0001) (B). Heatmap of correlation among differentially immune cells and biomarkers, red: positive correlation; blue: negative correlation (C). Bubble chart of correlation among differentially immune cells and biomarkers. Red: positive correlation; blue: negative correlation (Spearman, |r| >0.3 and P <.05) (D).

3.6. Interaction network was identified between biomarkers and TFs, miRNAs and lncRNAs

In the TF-mRNA network, PPT2 was regulated by 11 TFs, and MED21 was regulated by 13 TFs (Fig. 6A). Additionally, PPT2 and MED21 were regulated by ARID2. Subsequently, PPT2 and MED21 were found to be modulated by 20 miRNAs, respectively (Fig. 6B). Among them, PPT2 was modulated by hsa-miR-6793-3p, and MED21 was modulated by hsa-miR-3529-3p. Then, the lncRNA-miRNA-mRNA network comprised 44 nodes, incorporating 2 mRNAs, 15 miRNAs, and 27 lncRNAs, interconnected by a total of 195 edges (Fig. 6C). In detail, lncRNA (XIST) exerted a regulatory influence on PPT2 and MED21 by regulating hsa-mir-141-3p, hsa-mir-15b-5p and hsa-miR-182-5p, respectively.

Figure 6.

TF-biomarker regulation network. Red: biomaker; blue: TFs. (A) TF-mRNA-miRNA regulation network. Blue: TFs; red: biomaker; pink: miRNA (B). lncRNA-miRNA-mRNA regulation network. Red: biomaker; pink: miRNA; orange: lncRNA (C). TFs = transcription factors.

3.7. Toxicological interactions and molecular docking

Potential compounds of biomarkers for CHD were obtained according to the CTD database as shown in Figure 7A. These included 20 compounds that targeted PPT2, and 20 compounds that targeted MED21. Notably, the compound exhibiting the highest correlation with PPT2 was benzo(a)pyrene, the compound exhibiting the highest correlation with MED21 was bisphenol A.

Figure 7.

Biomarker-drug/ small molecule compound regulation network. Red: biomaker; purple: drug or small molecule compound (A). Molecular docking binding diagram showed the specific binding of PPT2-bisphenol A with minimal binding energy (B) and MED21-bisphenol A with minimal binding energy (C). PPT2 = palmitoyl protein thioesterase2, MED21 = mediator complex subunit21.

The binding of benzo(a)pyrene and bisphenol A to PPT2 and MED21 was further evaluated. According to the molecular docking results, binding energies of all 9 binding sites of PPT2 to benzo(a)pyrene were less than −5 kca/mol (Table 2), whereas the binding energy of MED21 to bisphenol A was less than −5 kca/mol at only 1 site (Table 3). The specific binding of PPT2-bisphenol A and MED21-bisphenol A with minimal binding energy was shown in Figure 7B and C.

Table 2.

PPT2 binding site with benzo (a) pyrene.

Mode	Affinity (kcal/mol)	Dist from rmsd l.b.	Best mode rmsd l.b.
1	−8.5	0	0
2	−8.4	15.679	18.548
3	−8.4	29.828	30.815
4	−8.3	29.824	30.896
5	−8.3	16.873	19.454
6	−8.2	29.746	31.266
7	−8.2	29.711	31.758
8	−8.2	0.761	5.498
9	−8.2	0.72	2.932

PPT2 = palmitoyl protein thioesterase2.

Table 3.

MED21 binding site with bisphenol A.

Mode	Affinity (kcal/mol)	Dist from rmsd l.b.	Best mode rmsd l.b.
1	−5.1	0	0
2	−4.8	0.582	5.672
3	−4.5	60.463	64.433
4	−3.6	5.767	8.44
5	−3.4	62.064	64.93
6	−3.3	6.223	7.531
7	−3.2	44.934	47.818
8	−3.1	43.173	45.379
9	−3	49.885	51.541

MED21 = mediator complex subunit21.

3.8. Biomarkers were significantly under-expressed in CHD clinical samples

To further validate the expression of biomarkers associated with mitochondrial metabolism, we performed RT-qPCR experiments utilizing clinical samples. When compared to the control group, the results illustrated in Figure 8A and B indicate a significant decrease in MED21 and PPT2 expression levels in the CHD group (P <.001) These results align with our bioinformatic analysis and offer further corroboration for the involvement of mitochondrial metabolism in CHD.

Figure 8.

The results of RT-qPCR experiments indicated a marked decrease in the expression levels of MED21 (A) and PPT2 (B) within the CHD group when contrasted with the control group (n = 5, unpaired t-tests, ***P <.001). CHD = coronary heart disease, MED21 = mediator complex subunit21, PPT2 = palmitoyl protein thioesterase2, RT-qPCR = reverse transcription quantitative polymerase chain reaction.

4. Discussion

CHD is a cardiovascular disease with high morbidity and mortality, and its pathophysiology is intricate.^[1] The newest evidence suggests that metabolic disorders and mitochondrial dysfunction may play significant roles in the onset and progression of CHD.^[15] In our study, we found the possibility of PPT2 and MED21 as CHD biomarkers by bioinformatics analysis, and the they were significantly down regulated in CHD group. From the perspective of diagnostic application value, these 2 biomarkers exhibited a certain capacity for disease differentiation. They can serve as important indicators for distinguishing between different pathological conditions in CHD diagnosis, thereby providing effective support for clinical diagnosis of CHD.

Palmitoyl protein thioesterase (PPT) is a lysosomal enzyme that mainly acts as a thioester linker between fatty acids and cysteine residues in lipid-modified proteins, facilitating the removal of long-chain fatty acids from cysteine residues in proteins.^[38] PPT comes in 2 varieties: PPT1 and PPT2.^[39] PPT2 encodes a lysosomal thioesterase homologous to PPT1.^[40] Both PPTs have significant effects on lysosomal thioester catabolism.^[41] PPT2 protein is mainly located in lysosomes. Under normal physiological conditions, it is involved in regulating the level of palmitoylation of intracellular proteins. Palmitoylation of proteins is an important posttranslational modification that affects the localization, stability and function of proteins.^[42] Emerging epidemiological studies have linked chronically elevated serum palmitic acid levels to an increased risk of CHD.^[43] And PPT2 is associated with hypermetabolism caused by uncoupled mitochondrial oxidative. Studies have shown that PPT2 is targeted to lysosomes via the mannose-6-phosphate receptor pathway. The enzyme exhibits strong activity towards palmitoylated model substrates, suggesting that it may play a role in lysosomal thioester catabolism.^[44] Some studies also revealed that PPT2 deficiency leads to severe neurodegeneration, while overexpression of PPT2 may accelerate tumor growth.^[40,44] Some DEGs were found to be involved in pathways associated with coronary artery disease, of which twenty genes (including PPT2) were identified as optimal features and used to generate a logistic classifier based on the LASSO.^[45] However, research on PPT2 in CHD is very limited.

Mediator complex subunit is essential for gene transcriptional regulation.^[46] As a core subunit of the Mediator complex, MED21 is widely involved in the transcriptional regulation of RNA polymerase II in eukaryotes. It can respond to various intracellular signals, bind to transcriptional activators or inhibitors, and regulate the level of gene expression.^[47] MED21 is related to the development of various diseases. The previous study showed that MED21 is intimately linked to the development of various types of cancer.^[48] In addition, some research validated the certain relevance between MED21 and some genetic diseases,^[49] suggesting the importance of MED21 in maintaining normal physiological functions. Recent studies have revealed subunits of the Mediator complex (like MED1, MED12, MED13, MED14, MED15, and MED23) are increasingly recognized for their roles in cardiovascular diseases (CVDs) and their development. Dysregulation of the Mediator complex in heart conditions can disrupt normal heart function and promote vascular dysfunction, with some subunits specifically implicated in adipogenesis, lipid metabolism, and the progression of various cardiovascular pathologies.^[50–52] Furthermore, there is almost no research on MED21 in CHD.

We found that PPT2 and MED21 were regulated by ARID2. Additionally, PPT2 was modulated by hsa-miR-6793-3p, and MED21 was modulated by hsa-miR-3529-3p. In detail, lncRNA exerted a regulatory influence on PPT2 and MED21 by regulating hsa-mir-141-3p, hsa-mir-15b-5p and hsa-miR-182-5p, respectively. The results fully reflected the complex molecular regulation mechanism of PPT2 and MED21 in disease progression. However, the specific role of PPT2 and MED21 in CHD remains unclear and warrants further investigation. According to our study, PPT2 and MED21 were found to be down-regulated in CHD when compared to normal control clinical samples. Although the link between CHD and downexpression of PPT2 and MED21 is not clear, study showed the reduced expression of PPT2 and MED21 could potentially serve as a novel diagnostic and prognostic biomarker, as well as a target for therapeutic interventions.^[53,54]

GESA pathway enrichment analysis showed that PPT2 and MED21 were co-enriched in the neuroactive ligand-receptor interaction, in the calcium signaling pathway, in the polycomb repressive complex, protein processing in endoplasmic reticulum. Growth hormone secretagogue receptor (GHSR) was involved in the pathway of neuroactive ligand-receptor interaction.^[55] It has been established that the peptide hormone ghrelin serves as an endogenous ligand for the GHSR.^[56] Furthermore, there is increasing evidence that the heart contains both ghrelin and the receptor GHSR-1a, and that the administration of ghrelin enhances cardiovascular health in both humans and animal models.^[57–59] Therefore, we inferred that PPT2 and MED21 may play key roles in cardioprotective effects and the calcium signaling pathway is central to different heart and vascular pathologies.^[60] It influences various cellular functions across many cell types, including fibroblasts, endothelial cells, macrophages, lymphocytes, and vascular smooth muscle cells.^[61,62] A previous study reported that NCX1, the calcium channel gene product of SLC8A1, was expressed in inflammatory and myofibroblast-like cells in the arterial wall and in cardiomyocytes from Kawasaki disease autopsy tissues.^[63] Patients with CHD may experience reduced acute inflammation if the calcium signaling pathway is inhibited. Protein processing in endoplasmic reticulum is involved in the pathogenesis of cardiovascular diseases,such as atherosclerotic plaque, coronary artery disease.^[64] Moreover,a series of studies with endoplasmic reticulum stress have been shown to reduce myocardial infarct size and cardiac hypertrophy in animal models.^[65,66]

The immune infiltration analysis indicated that activated CD4 T cells were considerably less prevalent in CHD group. Growing evidence documented the key role of T cells, especially CD4 + T cells, as drivers and modifiers in cardiovascular disease development,^[67] and they contribute to the complex and multifactorial pathogenesis of atherosclerosis.^[68] In addition, we found significant disparities among 21 distinct immune cell types between CHD group and control group,especially immature B cells, MDSC, and neutrophils were significantly more abundant in CHD group. This is consistent with the fact that monocytes/macrophages,^[69] neutrophils, and lymphocytes mechanistically contribute to the development of CHD.^[70] Further, our study suggested that PPT2 and MED21 were anticorrelated with regulatory T cells, MED21 exhibited a marked positive association with both central memory CD4 T cells and effector memory CD4 T cells. The results are consistent with the extensive evidence that reduced regulatory T cells/effector T cells ratio is closely related with the pathophysiology of coronary atherosclerosis.^[71] Based on these, we inferred that PPT2 and MED21 may affect the behavior or related functions of immune cells and regulate the immune microenvironment in CHD. Emerging evidence suggests that lysosomal lipid metabolism in T cells plays a critical role in shaping their inflammatory phenotype and functional fate. PPT2 has been implicated in maintaining lipid homeostasis within immune cells. Downexpression of PPT2 may impair lysosomal lipid degradation, leading to lipid accumulation and lysosomal stress in T cells. This metabolic perturbation could skew T cell differentiation toward pro-inflammatory subsets, such as Th1 and Th17 cells, thereby exacerbating vascular inflammation and accelerating atherosclerotic progression in CHD. Thus, we hypothesize that PPT2 downexpression in T cells disrupts lysosomal lipid metabolism, promotes T cell-mediated inflammation, and contributes to CHD pathogenesis.^[72–78] Then we will investigate the causal role of PPT2 and MED21 in CRISPR-mediated manipulation of primary CD4⁺ T cells, combined with lipidomics and endothelial cell co-culture experiments, to directly test their roles in CHD-associated inflammation. These experiments will be conducted as follow-up work.

Bisphenol A is a contaminant associated with various health issues, including anomalies in reproductive and developmental systems, impaired brain and neurological functions, cancer, and cardiovascular disease.^[79] By elevating serum levels of proinflammatory cytokines, bisphenol A can also enhance the inflammatory response and ultimately lead to the apoptosis of coronary artery smooth muscle cells.^[80] Moreover, bisphenol A may also contribute to coronary vascular weakness and ventricular hemorrhage by causing changes in endothelial cells within the coronary arteries and hearts of the rats.^[81] The previous research indicated bisphenol A exhibited a strong correlation with the severity of CHD, and experimental studies have linked bisphenol A to the progression of atherosclerosis.^[82] Researchers are investigating a medication or natural substance that can protect against the harmful effects of bisphenol A on health.^[83] The CTD database identified 20 compounds targeting PPT2 and MED21. Notably, molecular docking studies demonstrated that the compound exhibiting the highest correlation with PPT2 was benzo (a) pyrene, and the compound exhibiting the highest correlation with MED21 was bisphenol A. The results of the study indicated that these compounds might promote the development of CHD by influencing the function of biomarkers. However, further in vivo and in vitro experiments alongside epidemiological studies are required to confirm this causal relationship.

In this study, a small sample cohort was used for exploratory verification in the RT-qPCR verification phase, aiming to initially verify the reliability of bioinformatics analysis results and lay a foundation for subsequent large sample research. We hypothesize that PPT2 and MED21 expression levels decrease at different stages of CHD, with levels decreasing in subclinical plaques and further decreasing in patients with acute coronary syndromes. However, due to the small size of the validation sample, the statistical performance may be insufficient, which limits the universality of the results. In the future, it needs to be verified by a large scale clinical sample cohorts including different ages, genders, complications.

Our study firstly validated that PPT2 and MED21 were biomarkers in CHD by machine learning algorithms. Subsequently, a range of analytical procedures were employed to evaluate the function of the biomarkers in the regulatory mechanisms of CHD. Additionally, We used CTD database to screen bisphenol A targeting MED21 and benzo (a) pyrene targeting PPT2. Notably, molecular docking studies demonstrated the potential interaction between them, the results showed these compounds might contribute to the progression of CHD. Finally, RT-qPCR analysis of clinical samples showed the same results as the bioinformatics analysis.

However, we still face multiple challenges in pushing this finding into clinical practice. Firstly, a reliable diagnostic threshold has not been established under the current limited sample size, and its specificity for CHD compared to other cardiovascular or metabolic diseases, and the stability of blood samples under different storage conditions need to be strictly evaluated. Secondly, detection of the 2 biomarkers expression levels can only depend on RT-qPCR at present, when they compare to established CHD biomarkers (like troponins, NT-proBNP, CRP). We need to explore more suitable detection technology (like ELISA, Immunochromatography) for clinical application. But established biomarkers mainly reflect myocardial injury, cardiac load or inflammatory response, and they are not directly related to the pathological mechanism of CHD. In contrast, PPT2 and MED21 were associated with mitochondrial metabolism in CHD. It is expected to provide mechanistic insights at the functional level of disease. In the future, we will carry out Immunohistochemistry experiment to confirm tissue-specific localization and expression patterns of PPT2 and MED21, siRNA-Mediated Gene Knockdown experiment to explore downstream effects of the 2 biomarkers on calcium signaling pathway and endoplasmic reticulum stress response, Co-Immunoprecipitation experiment to establish a direct mechanistic link between biomarkers and chromatin remodeling in the context of CHD, In Vivo Validation in Animal Models to evaluate the physiological effects of biomarkers modulation in vivo, proteomic and metabolomic validation to reveal the signal networks and metabolic pathways regulated by PPT2 and MED21. The follow-up work aim to comprehensively elucidate the core role of PPT2 and MED21 in mitochondrial metabolism of CHD, and lay a theoretical foundation for future clinical transformation.

5. Conclusions

In conclusion, this study not only confirmed that PPT2 and MED21 were associated with mitochondrial metabolism in CHD, but also they were significantly down regulated in CHD group. The 2 biomarkers were found to be involved in pathways such as neuroactive ligand-receptor interaction, calcium signaling pathway and endoplasmic reticulum. We also found there are strong correlations between 2 biomarkers and CD4 T cells. These findings provide new theoretical references for subsequent in-depth research on CHD. Furthermore, We plan to experimentally validate their functions and clinical significance in follow-up studies.

Author contributions

Conceptualization: Xi Zhao.

Data curation: Xi Zhao, Jiayan He.

Formal analysis: Xi Zhao, Jiayan He.

Funding acquisition: Jiayan He.

Methodology: Jiayan He.

Software: Xi Zhao.

Supervision: Jiayan He.

Writing – original draft: Xi Zhao.

Writing – review & editing: Xi Zhao.