Unveiling miRNA Biomarkers for Hypertrophic Cardiomyopathy through Integrated Bioinformatics and Machine Learning Analysis

doi:10.21203/rs.3.rs-4816996/v1

Download PDF

Research Article

Unveiling miRNA Biomarkers for Hypertrophic Cardiomyopathy through Integrated Bioinformatics and Machine Learning Analysis

https://doi.org/10.21203/rs.3.rs-4816996/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

This study explores microRNAs (miRNAs) as biomarkers for hypertrophic cardiomyopathy (HCM), an inherited cardiac disease with clinical diversity and sudden death risk. Using bioinformatics and machine learning (ML), Gene Expression Omnibus (GEO) datasets were analyzed to identify miRNA signatures for early detection, risk assessment, and personalized treatment of HCM. Differential expression analysis of three GEO datasets identified 155 differentially expressed genes (DEGs) and 5 differentially expressed miRNAs (DE-miRNAs). Functional annotation and pathway analysis revealed their roles in inflammatory responses, extracellular matrix organization, and cellular stress responses. Notably, upregulated (COL21A1, PROM1) and downregulated (FOS, BTG2, ELL2, PDK4, SERPINE1, SRGN, TIPARP) genes were detected as potential DE-miRNA targets. Validation highlighted importance of ELL2 and PDK4 in HCM pathology. Support Vector Machine (SVM) and Random Forest (RF) models demonstrated high predictive accuracy for HCM using DE-miRNAs, suggesting new paths for early diagnosis and personalized therapy.

Hypertrophic Cardiomyopathy

microRNAs

Biomarkers

Machine Learning

Gene Expression Analysis

Hypertrophic cardiomyopathy (HCM) is a genetic disorder characterized by the thickening of the heart muscle (myocardium), which can lead to a range of clinical manifestations from asymptomatic to severe heart failure or sudden cardiac death [1]. HCM is the most common inherited cardiac disease, with a prevalence of 0.2–0.5% individuals globally [2, 3]. Despite its frequency, the diagnosis, prognosis, and management of HCM pose significant challenges due to the clinical heterogeneity of the condition and the complexity of its underlying genetic architecture. The traditional diagnostic criteria, based on imaging and family history, often result in late detection, after significant myocardial remodeling has occurred [1]. Therefore, there is a pressing need for novel biomarkers that can facilitate early detection, improve risk stratification, and guide personalized therapeutic interventions.

In the growing area of heart disease research, finding new biomarkers for HCM has attracted a lot of interest [4]. MicroRNAs (miRNAs) have emerged as crucial regulators of gene expression, influencing a wide array of physiological and pathological processes, including cardiac development and disease [5]. These small, non-coding RNA molecules can modulate the expression of target genes post-transcriptionally, thereby participating in the complex genetic networks underlying HCM [6]. Recent studies have highlighted the potential of miRNAs as biomarkers for various cardiovascular diseases, given their stability in biological fluids and their critical roles in disease pathophysiology [4, 7–9]. This body of work provides a foundation for our investigation, aiming to identify unique miRNA signatures that could serve as reliable markers for HCM diagnosis, prognosis, and therapeutic targeting. One pivotal study by Thum et al. [10] demonstrated the significant role of miR-21 in cardiac fibrosis, suggesting a broader implication of miRNAs in cardiac remodeling processes characteristic of HCM. A study carried out by Tian et al. [11] suggests that miRNA-27a*, derived from cardiac fibroblasts and selectively incorporated into extracellular vesicles, plays a key role in paracrine signaling during cardiac remodeling, offering potential as a biomarker and therapeutic target for chronic heart failure. Another study conducted by Guan et al. [12] highlights miR-106a as a key contributor to the advancement of pathological hypertrophy, suggesting its potential as a novel target for therapeutic intervention in this condition. These studies illuminated the potential for miRNAs to serve as both biomarkers and therapeutic targets in heart disease. Some studies have suggested a potential role for miRNAs in HCM, highlighting their association with the expression of myocardial hypertrophy and/or fibrosis [13–15]. These findings highlight the capacity of miRNAs to reflect the underlying pathophysiological changes in HCM, offering insights into the molecular basis and potential avenues for early detection of the disease. Building upon these foundations, a comprehensive analysis by Lin et al. [16] utilized high-throughput sequencing to delineate the miRNA expression profile in HCM patients. They identified miRNA-208b-3p as being significantly associated with the pathological features of HCM. Moreover, a study by Guo et al. [8] utilized three ML methods —Random Forest (RF), Support Vector Machines (SVM), and logistic regression—to identify potential biomarkers using miRNAs from the HCM co-expression network. Their work not only reinforced the diagnostic potential of miRNAs but also demonstrated the power of computational models in uncovering disease biomarkers. These studies collectively underscore the promise of miRNAs as biomarkers for HCM, reflecting a shift towards molecularly informed approaches in diagnosing and managing cardiovascular diseases. However, gaps remain in our understanding of the specific miRNAs involved in HCM and their interactions with target genes. Our study aims to bridge these gaps by employing an integrated bioinformatics and ML approach to identify novel miRNA biomarkers for HCM.

We utilized publicly available Gene Expression Omnibus (GEO) datasets: GSE36946, which profiles miRNA expression in surgical myectomy tissue from HCM patients; GSE36961, which focuses on transcriptome profiling in a similar cohort; and GSE149010, sourced from The Myocardial Applied Genomics Network (MAGNet; www.med.upenn.edu/magnet), is dedicated to collecting and storing human cardiac tissue for genomic research purposes [17]. Our analysis began with the identification of differentially expressed genes (DEGs) from GSE36961 and differentially expressed miRNAs (DE-miRNAs) from GSE36946. We then identified target genes for the detected DE-miRNAs and discovered common genes between DEGs and target genes of DE-miRNAs. Further validating these findings, we utilized the GSE149010 dataset to confirm the relevance of the common genes between DEGs and target genes of DE-miRNAs. The discriminatory power of these DE-miRNAs was assessed by applying receiver operating characteristic (ROC) analysis and calculating the area under the curve (AUC) values, aiming to determine their potential as diagnostic biomarkers. Additionally, we applied SVM and RF models using these DE-miRNAs to evaluate their effectiveness as biomarkers for HCM.

This comprehensive approach underscores the potential of miRNAs as biomarkers in HCM, offering a new avenue for early diagnosis, risk assessment, and the development of targeted therapeutic strategies. Through the integration of bioinformatics analysis with ML models, this study aims to unveil the complex molecular landscape of HCM and advance the frontier of personalized medicine in cardiac care.

Data Collection and preprocessing

Our study utilized publicly available datasets from the GEO database: GSE36946, GSE36961, and GSE149010. The GSE36946 and GSE36961 datasets include miRNA and mRNA profiling analyses of surgical myectomy tissue from patients diagnosed with HCM and control individuals without cardiac diseases. The GSE149010 dataset comprises mRNA profiles collected from left ventricular free-wall tissue during cardiac surgery from heart failure patients and unused donor hearts with normal function. The GSE36946 dataset includes miRNA expression profiles obtained using Illumina Human v2 MicroRNA Expression Beadchips (GPL8179), comparing 107 HCM patients with 20 control subjects. GSE36961 comprises mRNA expression data generated by Illumina HumanHT-12 V3.0 Expression Beadchips (GPL15389), focusing on the comparison between 106 HCM patients and 39 controls. GSE149010, generated by Illumina HiSeq 2500 (GPL16791), offers transcriptional expression data, includes 28 HCM patients and 166 healthy controls.

The raw expression data and corresponding group labels for HCM versus control groups were imported using the GEOquery package [18]. To stabilize variance across expression values, a log2 transformation was applied. The expression data from different arrays were normalized through the quantile normalization method to ensure comparability of expression levels between samples, utilizing the limma package [19]. The transformative impact of this preprocessing step is evident when comparing the data distributions before and after normalization. Specifically, Figs. 1A and 1B demonstrate the normalization effect on the GSE36961 dataset, showcasing a more uniform expression pattern post-normalization. Similarly, for the GSE36946 dataset, Figs. 1C and 1D highlight the role of normalization in correcting the distribution skewness and aligning the datasets on a comparable scale.

Additionally, to refine our analysis further, we implemented limma precision weights [20]. This enhanced the accuracy of the differential expression analysis by accounting for variability in individual gene measurements.

Differential expression analysis

Differential expression analysis was conducted using the limma package in R [19]. For each dataset, we constructed a design matrix to model the relationship between expression levels and group variables (HCM vs. Control). We identified differentially expressed genes (DEGs) and miRNAs (DE-miRNAs) based on a False Discovery Rate (FDR) of less than 0.05 (using the Benjamini-Hochberg method) and an absolute log2 fold change (|log2FC|) of greater than 1. To visualize our differential expression analysis, we utilized volcano plots to highlight genes with statistically significant changes in expression and large fold-changes.

Functional annotation and pathway analysis

We conducted functional annotation and pathway analysis using the clusterProfiler package [21] to elucidate the biological implications of the DEGs identified from GSE36961. We mapped the gene symbols to their corresponding Entrez IDs using org.Hs.eg.db [22] and carried out Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment and Gene Ontology (GO) term enrichment analyses for biological processes (BP), cellular components (CC), and molecular functions (MF). To visualize the significant pathways and GO terms, we created dot plots using enrichplot package [23], thereby facilitating the identification of biological processes and pathways potentially involved in HCM.

miRNA target prediction

We predicted the target genes of significantly differentially expressed miRNAs using the multiMiR package [24], which offers both predicted and validated miRNA-target interactions. We focused exclusively on validated miRNA target databases, including miRecords [25], miRTarBase [26], and TarBase [27]. To identify common genes between the DEGs from GSE36961 and the target genes of DE-miRNAs from GSE36946, we conducted a Venn diagram analysis using the ggvenn package [28].

Functional network analysis

To explore the biological functions associated with miRNA targets, we utilized functional network analysis through GeneMANIA (http://www.genemania.org), an online platform designed for inferring functions of genes or proteins [29]. This tool generates predictions by analyzing proteins with shared functions and uses a network approach to forecast interactions and functionalities of genes. GeneMANIA's networks, which can be tailored to highlight specific functions, compile data from various databases including GEO, BioGRID, EMBL-EBI, Pfam, Ensembl, Mouse Genome Informatics, the National Center for Biotechnology Information, InParanoid, and Pathway Commons.

Validation of common miRNA target genes using an independent dataset

We further validated genes identified as differentially expressed from previous analyses, along with target genes of differentially expressed miRNAs, using an independent dataset, GSE141910. This dataset served as a validation dataset to confirm the expression patterns of these genes in a separate cohort of HCM patients and controls. The validation aimed to affirm their potential involvement in HCM pathogenesis and evaluate their utility as biomarkers for the disease.

For validation, we selected candidate genes that were either upregulated or downregulated in the context of HCM. We then cross-referenced these genes with the GSE141910 dataset to assess their expression levels in an independent cohort. The focus of this analysis was to identify significant differences in gene expression that would further substantiate their association with HCM.

Machine learning models for biomarker identification

To explore the potential of DE-miRNAs as biomarkers for HCM, we conducted a machine learning (ML) analysis using two predictive models: SVM with a radial basis function kernel [30] and RF [31]. The goal of this analysis was to assess the ability of DE-miRNAs to distinguish between HCM patients and control subjects.

First, we performed ROC analysis for each DE-miRNA to evaluate their discriminatory capacity to classify HCM patients and healthy controls. Subsequently, we prepared two datasets for the ML models: the first dataset comprised the expression levels of these DE-miRNAs, while the second, an expanded dataset, included additional miRNAs with an AUC greater than 0.80. Each dataset was divided into training (70%) and testing (30%) sets to validate the predictive accuracy of the models.

For both SVM and RF models, we utilized a 5-fold cross-validation method during the training phase to optimize the parameters and ensure generalizability of the models. The models were trained to classify samples into either HCM or control groups, using preprocessing steps such as centering and scaling of data to improve performance.

After training, we applied the models to the testing dataset to predict the group labels. The performance of each model was assessed through ROC analysis, with a focus on the AUC metric to measure the capability of models to differentiate between HCM and control samples.

Statistical analysis

All statistical analyses were conducted using R software (version 4.3.2). The Student's t-test wit Bonferroni correction were performed to compare the expression levels between HCM patients and controls for the selected genes within the GSE141910 dataset. A two-tailed p-value less than 0.05 considered statistically significant.

Differential expression analysis

Based on selection criteria of FDR < 0.05 and |log2FC| > 1, we identified a total of 155 DEGs when comparing HCM samples to healthy controls within the GSE36961 dataset (S1 Table). Among them, 51 were upregulated, and 104 were downregulated in the HCM group compared to the healthy group. Furthermore, analysis of the GSE36946 dataset revealed 5 DE-miRNAs associated with HCM, with 2 DE-miRNAs upregulated (hsa-miR-373 and hsa-miR-514) and 3 downregulated (hsa-miR-10a, hsa-miR-10a* and hsa-miR-144*) in the HCM samples (S2 Table). Volcano plots are used to visually represent the DEGS and DE-miRNAs in Fig. 2A and 2B, respectively.

GO and KEGG enrichment analyses

To investigate the functional categories of the DEGs in GSE36961 between HCM and healthy samples, we conducted GO and KEGG pathway enrichment analyses. The GO enrichment analysis for Biological Process (BP) revealed significant enrichment of DEGs in processes including the regulation of inflammatory response, regulation of endopeptidase activity, regulation of myeloid cell differentiation, and regulation of intracellular zinc ion homeostasis (Fig. 3A). Within the Cellular Component (CC) domain, DEGs were predominantly associated with collagen-containing extracellular matrix, secretory granule lumen, cytoplasmic vesicle lumen, blood microparticle, and platelet alpha granule (Fig. 3B). For Molecular Function (MF), the enriched functions included integrin binding, peptidase regulator activity, heat shock protein binding, endopeptidase inhibitor activity, peptidase inhibitor activity, and molecular sequestering activity (Fig. 3C). KEGG pathway analysis highlighted involvement in pathways such as coronavirus disease, phagosome, complement and coagulation cascades, pertussis, and apoptosis (Fig. 3D).

Identification of common miRNA targets

Using multiMiR package and three validated databases, we predicted a total of 3362 target genes of DE-miRNAs. A venn diagram analysis showed that there are 2 common upregulated (Fig. 4A) and 6 common downregulated (Fig. 4B) genes between DEGs from GSE36961 and target genes of DE-miRNAs from GSE36946 (S3 Table).

The common upregulated genes were COL21A1 and PROM1. COL21A1 encodes the alpha chain of type XXI collagen, a FACIT family member that supports extracellular matrix integrity in tissues with type I collagen, with alternative splicing producing various transcript variants [32]. PROM1 produces a pentaspan transmembrane glycoprotein found in membrane protrusions, associated with adult stem cell maintenance, cancer, retinitis pigmentosa, and Stargardt disease, with expression driven by tissue-specific alternative promoters and resulting in multiple isoforms [33].

The common downregulated genes were BTG2, ELL2, FOS, PDK4, SRGN, SERPINE1 and TIPARP. The BTG2 protein, part of the antiproliferative BTG/Tob family, regulates the cell cycle G1/S transition [34]. ELL2, part of the transcription elongation factor complex, is involved in snRNA transcription by RNA polymerase II and is located in the nucleoplasm [35]. The Fos gene family, comprising FOS, FOSB, FOSL1, and FOSL2, encodes leucine zipper proteins that dimerize with JUN family proteins to form the AP-1 transcription factor complex, playing key roles in cell proliferation, differentiation, transformation, and in some instances, apoptotic cell death [36]. PDK4, part of the PDK/BCKDK kinase family, encodes a mitochondrial protein that regulates glucose metabolism by inhibiting the pyruvate dehydrogenase complex through phosphorylation, with its expression controlled by glucocorticoids, retinoic acid, and insulin [37]. SRGN encodes a proteoglycan found in hematopoietic cell granules, involved in neutralizing hydrolytic enzymes and possibly mediating granule-mediated apoptosis through its association with granzymes and perforin [38]. The SERPINE1 gene encodes a serine proteinase inhibitor that primarily blocks tissue plasminogen activator and urokinase, playing a role in fibrinolysis inhibition and innate antiviral immunity, with mutations in this gene causing defects in plasminogen activation [39]. TIPARP encodes a protein in the poly(ADP-ribose) polymerase family, involved in histone poly(ADP-ribosyl)ation and potentially T-cell function, with multiple transcript variants due to alternative splicing [40].

Validation of the common miRNA targets

The GSE141910 dataset, which includes transcriptional expression data from 194 samples—28 from patients with HCM and 166 from controls without any cardiac conditions—was utilized to validate the two commonly upregulated and seven commonly downregulated genes identified in the previous section. In the genes targeted by upregulated miRNAs, neither COL21A1 nor PROM1 showed a significant increase in expression in the HCM samples (Fig. 5A). Conversely, among the genes targeted by downregulated miRNAs, ELL2 and PDK4 exhibited a significant decrease in expression in HCM patients, while FOS, SERPINE1, and TIPARP did not demonstrate significant changes in expression (Fig. 5B). Although our bioinformatics analysis indicated that BTG2 and SRGN were downregulated in HCM samples, these results could not be confirmed as both genes actually showed upregulation in the HCM samples in the validation dataset.

Functional interaction

The GeneMANIA online webtool was used to explore the functional interactions of the DE-miRNA target genes with each other. We included four genes (BTG2, ELL2, PDK4, and SRGN) in the functional interaction analysis, as their expressions differed significantly between HCM and control samples in the GSE141910 validation set. The interaction network included 20 other related genes in addition to the four targets that were entered, and there were a total of 230 links. Seven interaction types were involved, with physical interactions being the most frequent type of interaction (77.64%). The most covered functions shared by genes were regulation of sulfur metabolic process, acetyl-CoA biosynthetic process from pyruvate, acetyl-CoA biosynthetic process, acetyl-CoA metabolic process, regulation of purine nucleotide biosynthetic process, regulation of nucleotide biosynthetic process, and thioester biosynthetic process. The interaction network is illustrated in Fig. 6.

Machine learning approach

Our initial investigation involved conducting ROC curve analyses and computing the AUC values for each differentially expressed miRNA, which highlighted their substantial capacity to distinguish between HCM and healthy controls. The AUC values demonstrated strong predictive performances: 0.884 (95% CI: 0.815–0.952) for hsa-mir-373, 0.873 (95% CI: 0.809–0.937) for hsa-mir-144*, 0.865 (95% CI: 0.794–0.937) for hsa-mir-514, 0.825 (95% CI: 0.700–0.950) for hsa-mir-10a, and 0.821 (95% CI: 0.715–0.928) for hsa-mir-10a*. The corresponding ROC curves for each DE-miRNA are illustrated in Fig. 7A.

Following this, we constructed two ML models, SVM and RF, utilizing the five differentially expressed miRNAs as predictive features. These models significantly outperformed the analysis based on individual miRNAs, achieving noteworthy AUC values of 0.953 (95% CI: 0.882-1) for the SVM and 0.922 (95% CI: 0.835-1) for the RF in the testing phase, as depicted in Fig. 7B. Notably, the SVM model delivered an accuracy of 92.1%, a recall of 96.9%, and a precision of 93.9%, in contrast to the RF model, which achieved an 86.8% accuracy, matched the 96.9% recall, and obtained an 88.6% precision. The confusion matrices elucidating the performance of both SVM and RF models are detailed in Figs. 8A and 8B.

To extend our exploration for potent miRNA biomarkers capable of discriminating HCM from healthy samples, we employed an additional selection criterion, targeting miRNAs with AUC values exceeding 0.80. This approach revealed another 21 miRNAs, enriching our biomarker panel (S4 Table). With this expanded miRNA set, we re-optimized the SVM and RF models, which consequently showcased remarkable predictive capabilities, each securing an AUC of 0.995 (95% CI: 0.98-1) on the test dataset, as shown in Fig. 7C. This improvement significantly bolstered the diagnostic accuracy, with the SVM model achieving a 97.4% accuracy, 100% recall, and 97.0% precision, and the RF model demonstrating a 94.7% accuracy, still with a perfect recall and a precision of 94.1%. The corresponding confusion matrices for the enhanced SVM and RF models are illustrated in Figs. 8C and 8D, affirming the utility of incorporating a more comprehensive miRNA profile for the diagnosis of HCM.

Our study embarked on a comprehensive bioinformatics and ML analysis to describe the transcriptomic landscapes and identify novel miRNA biomarkers distinguishing HCM samples from healthy controls. We implemented a rigorous preprocessing step, including log2 transformation and quantile normalization, and generated all GEO datasets using Illumina chips, with GSE36946 generated on Illumina Human v2 MicroRNA expression beadchip, GSE36961 on Illumina HumanHT-12 V3.0 expression beadchip, and GSE141910 on Illumina HiSeq 2500 expression beadchip. Thus, we effectively minimized technical biases, enhancing the reliability and comparability of gene expression data across samples.

We identified 155 DEGs between HCM and healthy samples, with 51 upregulated and 104 downregulated. Our analysis also facilitated the identification of common target genes between DEGs and DE-miRNAs, revealing key upregulated genes COL21A1, PROM1, and downregulated genes FOS, BTG2, ELL2, PDK4, SERPINE1, SRGN, and TIPARP. This intersection not only validates the regulatory impact of identified miRNAs on gene expression but also accentuates their potential as biomarkers for HCM.

The GO and KEGG pathway enrichment analyses revealed significant biological insights. The GO analysis showed that DEGs were significantly enriched in biological processes such as regulation of inflammatory response and endopeptidase activity, myeloid cell differentiation, and intracellular zinc ion homeostasis, suggesting a complex interplay of immune response and metabolic regulation in HCM pathology. Furthermore, the KEGG pathway analysis elucidated involvement of DEGs in critical pathways including coronavirus disease, phagosome, complement and coagulation cascades, pertussis, and apoptosis, indicating potential mechanisms through which HCM influences cardiovascular health and disease progression. Studies, such as those on atherosclerosis, have shown that DEGs are notably enriched in pathways related to immune system interactions, metabolic processes, and specific diseases like dilated cardiomyopathy, highlighting the complex interplay between genetic factors and cardiovascular health [41]. These findings underscore the multifaceted nature of cardiovascular pathologies, including HCM, and suggest that similar pathways might be implicated in HCM pathology, warranting further investigation into the roles of infectious agents, immune responses, and thrombotic processes

Our analysis of the GSE141910 validation dataset, which includes transcriptional expression data from 194 samples—28 from patients with HCM and 166 from controls—revealed intriguing discrepancies and validations that highlight the complexity of gene expression regulation in HCM. Notably, our analysis aimed to validate expressions of two genes targeted by upregulated microRNAs, namely COL21A1 and PROM1. Despite expectations based on their upstream miRNA regulation, neither gene exhibited a statistically significant increase in expression within the HCM patient samples as illustrated in Fig. 5A. This outcome could suggest a possible post-transcriptional modulation or involvement of additional regulatory mechanisms that override the miRNA effect, a concept supported by various studies demonstrating the complexity of miRNA-target interactions in cardiac diseases. Research has shown that miRNAs can engage in intricate regulatory networks, impacting gene expression through multiple pathways, including inhibiting translation or promoting mRNA degradation. These interactions are influenced by the degree of complementarity between the miRNA and mRNA and can be modulated by alternative biogenesis pathways and the presence of other regulatory molecules [42]. Furthermore, studies have highlighted the role of miRNAs in various aspects of cardiac physiology and pathology, including cardiac remodeling and heart failure, where miRNAs regulate key genes and pathways involved in these processes. These interactions can be influenced by the cellular context, the presence of other regulatory proteins, and even changes in the physical structure of the myocardium, which can all alter the functional impact of miRNAs [5, 43]. Moreover, studies underscore the importance of miRNAs in regulating mitochondrial function in cardiomyocytes, revealing yet another layer of complexity in how miRNAs influence cardiac pathology and health [44]. These insights collectively enhance our understanding of the regulatory landscape in which miRNAs operate, particularly in cardiac function and disease, and underscore the potential for diverse regulatory mechanisms to modulate or even override miRNA effects.

The significant decrease in expression of the genes ELL2 and PDK4 in HCM samples, as highlighted by our findings, aligns well with the established roles of these genes in cardiac muscle function and energy metabolism. Notably, studies have shown that PDK4 plays a crucial role in the regulation of glucose and fatty acid metabolism within the heart, indicating its potential involvement in the metabolic adaptations seen in cardiac diseases including hypertrophic cardiomyopathy [45, 46]. Moreover, PDK4 is known to be sensitive to changes in nutritional and hormonal conditions, which can influence its expression and activity, affecting the metabolic state of cardiac tissues [47]. This sensitivity to metabolic conditions supports the observed downregulation in HCM and provides a potential link to the altered metabolic demands in hypertrophic tissues. On the other hand, ELL2, while not as extensively studied in the context of cardiac function, is generally involved in transcriptional processes [35]. Its decreased expression might reflect broader disruptions in gene regulation that could contribute to the pathological landscape of HCM.

Interestingly, the bioinformatics prediction of downregulation for BTG2 and SRGN in HCM samples was contradicted by the validation dataset, where both genes were actually found to be upregulated. This discrepancy could indicate a compensatory mechanism within the cardiac tissue in response to hypertrophic stress or might reflect the involvement of these genes in other aspects of cardiac remodeling that are not solely dependent on miRNA regulation. This observation is particularly significant as it suggests that cardiac pathology may involve multifaceted gene expression responses that could be potential targets for therapeutic intervention. The expression trends of BTG2 and SRGN, therefore, may represent a physiological adaptation or maladaptation in the progression of HCM, warranting further investigation into their specific roles and regulation in the heart.

These results collectively contribute to the intricate mosaic of gene-miRNA interactions within the HCM landscape, illustrating that while some findings confirm theoretical predictions, others reveal the limitations of current models and the need for a more nuanced understanding of the molecular underpinnings of this complex disease.

The investigation into the interplay between DE-miRNAs and their gene targets provides crucial insights into the underlying mechanisms of HCM. The functional interaction analysis performed using the GeneMANIA online webtool has elucidated complex networks that underline the pathophysiology of HCM. By incorporating genes like BTG2, ELL2, PDK4, and SRGN, whose expression levels significantly differ between HCM and control samples as per the GSE141910 validation dataset, a detailed interaction network was revealed. This network, comprising 24 genes and 230 links with seven types of interactions, primarily highlighted physical interactions, which accounted for over three-quarters of all connections. This predominance suggests a significant role of protein-protein interactions in the molecular dynamics of HCM. The network also emphasized key metabolic and biosynthetic processes such as the regulation of sulfur metabolic [48] and various acetyl-CoA related processes [49], pointing to metabolic alterations as crucial elements in HCM pathology. These findings suggest potential therapeutic targets and warrant further investigation into how these metabolic pathways might be manipulated to mitigate HCM progression.

The crucial role of miRNAs in HCM pathology is highlighted by their ability to regulate gene expression post-transcriptionally [50]. Our analysis points to miRNAs like hsa-miR-373 and hsa-miR-514 as vital players in the molecular mechanisms of HCM. Recent research has expanded on identifying miRNA biomarkers for HCM, applying ML algorithms such as RF, SVM, and LR for HCM classification based on miRNA profiles [8]. Such biomarkers are also under investigation for other cardiovascular conditions like myocardial infarction [51–53], myocardial fibrosis [54], coronary artery disease [9], heart failure [55], ischemia [56], stroke [57] and dilated cardiomyopathy [58]. Our use of ROC analysis assessed five DE-miRNAs for HCM detection, with AUC values ranging from 0.821 (hsa-mir-10a*) to 0.884 (hsa-mir-373). Additionally, we enhanced DE-miRNA classification with traditional ML algorithms, SVM and RF. Our dual strategy first utilized five DE-miRNAs as ML model features; then we included additional miRNAs with AUCs over 0.80, totaling 26, for model training.

The first method improved the AUC to 0.953 (95% CI [0.882–1.000]) for SVM, marginally surpassing RF with an AUC of 0.922 (95% CI [0.835–1.000]). Our extended approach further increased discriminatory power to an AUC of 0.995 (95% CI [0.980–1.000]) for both SVM and RF. Although both algorithms showed identical AUC values, SVM slightly outperformed RF in terms of accuracy (97.4% vs. 94.7%) and precision (97.0% vs. 94.1%).

In this study, we identified two key genes and five miRNAs linked to HCM, confirming their diagnostic value through an external dataset and machine learning techniques. While our results offer a hopeful method for diagnosing HCM, certain limitations must be considered. The foundation of our investigation is bioinformatic analysis; hence, additional experimental and clinical studies are necessary for confirmation. Moreover, future research should explore the underlying roles of the identified genes and miRNAs in HCM.

Our study provides a comprehensive view of the transcriptomic alterations in HCM and underscores the potential of miRNAs as biomarkers for this complex cardiac condition. Key genes regulated by differentially expressed miRNAs, such as ELL2 and PDK4, emerged as significant in our analysis, and their relevance was supported by validation in an independent dataset. These findings highlight the intricate molecular interplay at work in HCM and confirm the utility of these genetic markers in understanding pathophysiology of the disease. Enrichment analyses further revealed detailed picture of coexpression dominance, which may drive gene expression landscape of HCM, implicating miRNAs as central players in post-transcriptional regulation of HCM. This emphasizes their potential as both biomarkers and therapeutic targets. The use of ML models, specifically SVM and RF, enhanced the predictive accuracy of miRNA biomarkers, with our models achieving high AUC values, indicating strong discriminatory power between HCM and healthy controls. Our dual approach, initially utilizing a select group of DE-miRNAs and subsequently expanding the biomarker set based on AUC performance, allowed us to refine the classification ability of the ML models. This strategy led to an impressive increase in the AUC, with SVM slightly outperforming RF in both accuracy and precision. This investigation into the transcriptomic landscape and miRNA profiles presents a promising step towards the early detection and personalized treatment of HCM. The biomarkers identified hold potential for clinical application, subject to further validation, emphasizing the role of integrated bioinformatics and ML approaches in advancing cardiac care.

AUC: Area Under the Curve
BP: Biological Process
CC: Cellular Component
DE: Differential Expression
DEGs: Differentially Expressed Genes
GEO: Gene Expression Omnibus
GO: Gene Ontology
HCM: Hypertrophic Cardiomyopathy
KEGG: Kyoto Encyclopedia of Genes and Genomes
ML: Machine Learning
RF: Random Forest
ROC: Receiver Operating Characteristic
SVM: Support Vector Machine

Data and code availability

The datasets analysed during this study are available in the Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo/), reference numbers GSE36946, GSE36961, and GSE141910.

All the R codes and datasets are publicly available at https://github.com/selcukorkmaz/miRNA_HCM_Biomarkers.

Funding: The authors did not receive support from any organization for the submitted work.

Conflict of Interest: The authors declare that they have no conflict of interest.

Author Contribution

B.E.Y. was responsible for conceptualization, investigation, methodology, validation, and writing - original draft, review & editing. S.K. conducted bioinformatics analysis, formal analysis, and participated in writing - original draft, review & editing. All authors reviewed the manuscript.

Marian AJ, Braunwald E. Hypertrophic Cardiomyopathy: Genetics, Pathogenesis, Clinical Manifestations, Diagnosis, and Therapy. Circ Res. 2017;121(7):749-70. doi: 10.1161/CIRCRESAHA.117.311059.
Maron BJ, Desai MY, Nishimura RA, Spirito P, Rakowski H, Towbin JA, et al. Diagnosis and Evaluation of Hypertrophic Cardiomyopathy: JACC State-of-the-Art Review. J Am Coll Cardiol. 2022;79(4):372-89. doi: 10.1016/j.jacc.2021.12.002.
Wolf CM. Hypertrophic cardiomyopathy: genetics and clinical perspectives. Cardiovasc Diagn Ther. 2019;9(Suppl 2):S388-S415. doi: 10.21037/cdt.2019.02.01.
Luo F, Liu W, Bu H. MicroRNAs in hypertrophic cardiomyopathy: pathogenesis, diagnosis, treatment potential and roles as clinical biomarkers. Heart Fail Rev. 2022;27(6):2211-21. doi: 10.1007/s10741-022-10231-z.
Thum T, Catalucci D, Bauersachs J. MicroRNAs: novel regulators in cardiac development and disease. Cardiovasc Res. 2008;79(4):562-70. doi: 10.1093/cvr/cvn137.
Osmak G, Baulina N, Kiselev I, Favorova O. MiRNA-Regulated Pathways for Hypertrophic Cardiomyopathy: Network-Based Approach to Insight into Pathogenesis. Genes (Basel). 2021;12(12). doi: 10.3390/genes12122016.
Wang Z, Xia Q, Su W, Cao M, Sun Y, Zhang M, et al. Exploring the Communal Pathogenesis, Ferroptosis Mechanism, and Potential Therapeutic Targets of Dilated Cardiomyopathy and Hypertrophic Cardiomyopathy via a Microarray Data Analysis. Front Cardiovasc Med. 2022;9:824756. doi: 10.3389/fcvm.2022.824756.
Guo L, Cai Y, Wang B, Zhang F, Zhao H, Liu L, et al. Characterization of the circulating transcriptome expression profile and identification of novel miRNA biomarkers in hypertrophic cardiomyopathy. Eur J Med Res. 2023;28(1):205. doi: 10.1186/s40001-023-01159-7.
Kumar D, Narang R, Sreenivas V, Rastogi V, Bhatia J, Saluja D, et al. Circulatory miR-133b and miR-21 as Novel Biomarkers in Early Prediction and Diagnosis of Coronary Artery Disease. Genes (Basel). 2020;11(2). doi: 10.3390/genes11020164.
Thum T, Gross C, Fiedler J, Fischer T, Kissler S, Bussen M, et al. MicroRNA-21 contributes to myocardial disease by stimulating MAP kinase signalling in fibroblasts. Nature. 2008;456(7224):980-4. doi: 10.1038/nature07511.
Tian C, Hu G, Gao L, Hackfort BT, Zucker IH. Extracellular vesicular MicroRNA-27a* contributes to cardiac hypertrophy in chronic heart failure. J Mol Cell Cardiol. 2020;143:120-31. doi: 10.1016/j.yjmcc.2020.04.032.
Guan X, Wang L, Liu Z, Guo X, Jiang Y, Lu Y, et al. miR-106a promotes cardiac hypertrophy by targeting mitofusin 2. J Mol Cell Cardiol. 2016;99:207-17. doi: 10.1016/j.yjmcc.2016.08.016.
Roncarati R, Viviani Anselmi C, Losi MA, Papa L, Cavarretta E, Da Costa Martins P, et al. Circulating miR-29a, among other up-regulated microRNAs, is the only biomarker for both hypertrophy and fibrosis in patients with hypertrophic cardiomyopathy. J Am Coll Cardiol. 2014;63(9):920-7. doi: 10.1016/j.jacc.2013.09.041.
Derda AA, Thum S, Lorenzen JM, Bavendiek U, Heineke J, Keyser B, et al. Blood-based microRNA signatures differentiate various forms of cardiac hypertrophy. Int J Cardiol. 2015;196:115-22. doi: 10.1016/j.ijcard.2015.05.185.
Fang L, Ellims AH, Moore XL, White DA, Taylor AJ, Chin-Dusting J, et al. Circulating microRNAs as biomarkers for diffuse myocardial fibrosis in patients with hypertrophic cardiomyopathy. J Transl Med. 2015;13:314. doi: 10.1186/s12967-015-0672-0.
Lin LR, Hu XQ, Lu LH, Dai JZ, Lin NN, Wang RH, et al. MicroRNA expression profiles in familial hypertrophic cardiomyopathy with myosin-binding protein C3 (MYBPC3) gene mutations. BMC Cardiovasc Disord. 2022;22(1):278. doi: 10.1186/s12872-022-02714-6.
Tan WLW, Anene-Nzelu CG, Wong E, Lee CJM, Tan HS, Tang SJ, et al. Epigenomes of Human Hearts Reveal New Genetic Variants Relevant for Cardiac Disease and Phenotype. Circ Res. 2020;127(6):761-77. doi: 10.1161/CIRCRESAHA.120.317254.
Davis S, Meltzer PS. GEOquery: a bridge between the Gene Expression Omnibus (GEO) and BioConductor. Bioinformatics. 2007;23(14):1846-7. doi: 10.1093/bioinformatics/btm254.
Ritchie ME, Phipson B, Wu D, Hu Y, Law CW, Shi W, et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res. 2015;43(7):e47. doi: 10.1093/nar/gkv007.
Law CW, Chen Y, Shi W, Smyth GK. voom: Precision weights unlock linear model analysis tools for RNA-seq read counts. Genome Biol. 2014;15(2):R29. doi: 10.1186/gb-2014-15-2-r29.
Wu T, Hu E, Xu S, Chen M, Guo P, Dai Z, et al. clusterProfiler 4.0: A universal enrichment tool for interpreting omics data. Innovation (Camb). 2021;2(3):100141. doi: 10.1016/j.xinn.2021.100141.
Carlson M, Falcon S, Pages H, Li N. org. Hs. eg. db: Genome wide annotation for Human. R package version. 2019;3(2):3.
Yu G. enrichplot: Visualization of Functional Enrichment Result. R package version 1.20. 3. 2023.
Ru Y, Kechris KJ, Tabakoff B, Hoffman P, Radcliffe RA, Bowler R, et al. The multiMiR R package and database: integration of microRNA-target interactions along with their disease and drug associations. Nucleic Acids Res. 2014;42(17):e133. doi: 10.1093/nar/gku631.
Xiao F, Zuo Z, Cai G, Kang S, Gao X, Li T. miRecords: an integrated resource for microRNA-target interactions. Nucleic Acids Res. 2009;37(Database issue):D105-10. doi: 10.1093/nar/gkn851.
Huang HY, Lin YC, Cui S, Huang Y, Tang Y, Xu J, et al. miRTarBase update 2022: an informative resource for experimentally validated miRNA-target interactions. Nucleic Acids Res. 2022;50(D1):D222-D30. doi: 10.1093/nar/gkab1079.
Sethupathy P, Corda B, Hatzigeorgiou AG. TarBase: A comprehensive database of experimentally supported animal microRNA targets. RNA. 2006;12(2):192-7. doi: 10.1261/rna.2239606.
Yan L. ggvenn: Draw Venn Diagram by ‘ggplot2’. R Package Version. 2021;19.
Warde-Farley D, Donaldson SL, Comes O, Zuberi K, Badrawi R, Chao P, et al. The GeneMANIA prediction server: biological network integration for gene prioritization and predicting gene function. Nucleic Acids Res. 2010;38(Web Server issue):W214-20. doi: 10.1093/nar/gkq537.
Cortes C, Vapnik V. Support-Vector Networks. Mach Learn. 1995;20(3):273-97. doi: Doi 10.1007/Bf00994018.
Breiman L. Random forests. Mach Learn. 2001;45(1):5-32. doi: Doi 10.1023/A:1010933404324.
Fitzgerald J, Bateman JF. A new FACIT of the collagen family: COL21A1. FEBS Lett. 2001;505(2):275-80. doi: 10.1016/s0014-5793(01)02754-5.
Weigmann A, Corbeil D, Hellwig A, Huttner WB. Prominin, a novel microvilli-specific polytopic membrane protein of the apical surface of epithelial cells, is targeted to plasmalemmal protrusions of non-epithelial cells. Proc Natl Acad Sci U S A. 1997;94(23):12425-30. doi: 10.1073/pnas.94.23.12425.
Rouault JP, Falette N, Guehenneux F, Guillot C, Rimokh R, Wang Q, et al. Identification of BTG2, an antiproliferative p53-dependent component of the DNA damage cellular response pathway. Nat Genet. 1996;14(4):482-6. doi: 10.1038/ng1296-482.
Shilatifard A, Duan DR, Haque D, Florence C, Schubach WH, Conaway JW, et al. ELL2, a new member of an ELL family of RNA polymerase II elongation factors. Proc Natl Acad Sci U S A. 1997;94(8):3639-43. doi: 10.1073/pnas.94.8.3639.
Tulchinsky E. Fos family members: regulation, structure and role in oncogenic transformation. Histol Histopathol. 2000;15(3):921-8. doi: 10.14670/HH-15.921.
Gudi R, Bowker-Kinley MM, Kedishvili NY, Zhao Y, Popov KM. Diversity of the pyruvate dehydrogenase kinase gene family in humans. J Biol Chem. 1995;270(48):28989-94. doi: 10.1074/jbc.270.48.28989.
Toyama-Sorimachi N, Sorimachi H, Tobita Y, Kitamura F, Yagita H, Suzuki K, et al. A novel ligand for CD44 is serglycin, a hematopoietic cell lineage-specific proteoglycan. Possible involvement in lymphoid cell adherence and activation. J Biol Chem. 1995;270(13):7437-44. doi: 10.1074/jbc.270.13.7437.
Declerck PJ, Gils A. Three decades of research on plasminogen activator inhibitor-1: a multifaceted serpin. Semin Thromb Hemost. 2013;39(4):356-64. doi: 10.1055/s-0033-1334487.
Katoh M, Katoh M. Identification and characterization of human TIPARP gene within the CCNL amplicon at human chromosome 3q25.31. Int J Oncol. 2003;23(2):541-7.
Li F, Du TY, Wu LD, Zhang L, Liu HH, Zhang ZY, et al. Identification of key immune-related genes in dilated cardiomyopathy using bioinformatics analysis. Sci Rep. 2023;13(1):1820. doi: 10.1038/s41598-022-26277-w.
Romaine SP, Tomaszewski M, Condorelli G, Samani NJ. MicroRNAs in cardiovascular disease: an introduction for clinicians. Heart. 2015;101(12):921-8. doi: 10.1136/heartjnl-2013-305402.
Pang JKS, Phua QH, Soh BS. Applications of miRNAs in cardiac development, disease progression and regeneration. Stem Cell Res Ther. 2019;10(1):336. doi: 10.1186/s13287-019-1451-2.
Zhang GQ, Wang SQ, Chen Y, Fu LY, Xu YN, Li L, et al. MicroRNAs Regulating Mitochondrial Function in Cardiac Diseases. Front Pharmacol. 2021;12:663322. doi: 10.3389/fphar.2021.663322.
Ritterhoff J, Tian R. Metabolism in cardiomyopathy: every substrate matters. Cardiovasc Res. 2017;113(4):411-21. doi: 10.1093/cvr/cvx017.
Zhang S, Hulver MW, McMillan RP, Cline MA, Gilbert ER. The pivotal role of pyruvate dehydrogenase kinases in metabolic flexibility. Nutr Metab (Lond). 2014;11(1):10. doi: 10.1186/1743-7075-11-10.
Wang X, Shen X, Yan Y, Li H. Pyruvate dehydrogenase kinases (PDKs): an overview toward clinical applications. Biosci Rep. 2021;41(4). doi: 10.1042/BSR20204402.
Kolluru GK, Shackelford RE, Shen X, Dominic P, Kevil CG. Sulfide regulation of cardiovascular function in health and disease. Nat Rev Cardiol. 2023;20(2):109-25. doi: 10.1038/s41569-022-00741-6.
Abo Alrob O, Lopaschuk GD. Role of CoA and acetyl-CoA in regulating cardiac fatty acid and glucose oxidation. Biochem Soc Trans. 2014;42(4):1043-51. doi: 10.1042/BST20140094.
Gurtan AM, Sharp PA. The role of miRNAs in regulating gene expression networks. J Mol Biol. 2013;425(19):3582-600. doi: 10.1016/j.jmb.2013.03.007.
Samadishadlou M, Rahbarghazi R, Piryaei Z, Esmaeili M, Avci CB, Bani F, et al. Unlocking the potential of microRNAs: machine learning identifies key biomarkers for myocardial infarction diagnosis. Cardiovasc Diabetol. 2023;22(1):247. doi: 10.1186/s12933-023-01957-7.
Horvath M, Horvathova V, Hajek P, Stechovsky C, Honek J, Senolt L, et al. MicroRNA-331 and microRNA-151-3p as biomarkers in patients with ST-segment elevation myocardial infarction. Sci Rep. 2020;10(1):5845. doi: 10.1038/s41598-020-62835-w.
Wang KJ, Zhao X, Liu YZ, Zeng QT, Mao XB, Li SN, et al. Circulating MiR-19b-3p, MiR-134-5p and MiR-186-5p are Promising Novel Biomarkers for Early Diagnosis of Acute Myocardial Infarction. Cell Physiol Biochem. 2016;38(3):1015-29. doi: 10.1159/000443053.
Lin R, Rahtu-Korpela L, Szabo Z, Kemppi A, Skarp S, Kiviniemi AM, et al. MiR-185-5p regulates the development of myocardial fibrosis. J Mol Cell Cardiol. 2022;165:130-40. doi: 10.1016/j.yjmcc.2021.12.011.
Yan H, Ma F, Zhang Y, Wang C, Qiu D, Zhou K, et al. miRNAs as biomarkers for diagnosis of heart failure: A systematic review and meta-analysis. Medicine (Baltimore). 2017;96(22):e6825. doi: 10.1097/MD.0000000000006825.
Martinez EC, Lilyanna S, Wang P, Vardy LA, Jiang X, Armugam A, et al. MicroRNA-31 promotes adverse cardiac remodeling and dysfunction in ischemic heart disease. J Mol Cell Cardiol. 2017;112:27-39. doi: 10.1016/j.yjmcc.2017.08.013.
Eyileten C, Wicik Z, De Rosa S, Mirowska-Guzel D, Soplinska A, Indolfi C, et al. MicroRNAs as Diagnostic and Prognostic Biomarkers in Ischemic Stroke-A Comprehensive Review and Bioinformatic Analysis. Cells. 2018;7(12). doi: 10.3390/cells7120249.
Lin Y, Chen K, Guo J, Chen P, Qian ZR, Zhang T. Identification of cuproptosis-related genes and immune infiltration in dilated cardiomyopathy. Int J Cardiol. 2024;399:131702. doi: 10.1016/j.ijcard.2023.131702.

No competing interests reported.

S1Table.xlsx
S1 Table. Differentially expressed genes (DEGs) in GSE36961. (XLSX)
S2.Table.xlsx
S2 Table. Differentially expressed miRNAs (DE-miRNAs) in GSE36946. (XLSX)
S3Table.xlsx
S3 Table. Common differentially expressed genes between GSE36946 and GSE36961. (XLSX)
S4Table.xlsx
S4 Table. miRNAs that have AUC values greater than 0.80. (XLSX)

Download PDF

Version 1

posted

You are reading this latest preprint version

Unveiling miRNA Biomarkers for Hypertrophic Cardiomyopathy through Integrated Bioinformatics and Machine Learning Analysis

Status:

Version 1

Abstract

Figures

Introduction

Methods

Data Collection and preprocessing

Differential expression analysis

Functional annotation and pathway analysis

miRNA target prediction

Functional network analysis

Validation of common miRNA target genes using an independent dataset

Machine learning models for biomarker identification

Statistical analysis

Results

Differential expression analysis

GO and KEGG enrichment analyses

Identification of common miRNA targets

Validation of the common miRNA targets

Functional interaction

Machine learning approach

Discussion

Conclusion

Abbreviations

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1