Comprehensive Transcriptomics Profiling of microRNA Reveals Plasma Circulating Biomarkers of Hypertrophic Cardiomyopathy and Dysregulated Signaling Pathways

Abstract

Background:

Hypertrophic cardiomyopathy (HCM) is caused by mutations in genes coding for proteins essential for myocardial contraction. However, it remains unclear through which signaling pathways these gene mutations mediate HCM pathogenesis. Growing evidence indicates that microRNAs (miRNAs) play an important role in the regulation of gene expression. We hypothesized that transcriptomics profiling of plasma miRNAs would reveal circulating biomarkers and dysregulated signaling pathways in HCM.

Methods:

We conducted a multi-center case-control study of cases with HCM and controls with hypertensive left ventricular hypertrophy. We carried out plasma transcriptomics profiling of miRNAs using RNA-sequencing. We developed a transcriptomics-based discrimination model using samples retrieved during the first two-thirds of the study period at one institution (training set). We prospectively tested its discriminative ability in samples collected thereafter from the same institution (prospective test set). We also externally validated the model by applying it to samples collected from the other institutions (external test set). We executed pathway analysis of dysregulated miRNAs with univariable P<0.05.

Results:

This study included 555 patients (392 cases and 163 controls). 1,141 miRNAs passed our quality control filters. The area under the receiver-operating-characteristic curve of the transcriptomics-based model derived from the training set was 0.86 (95% confidence interval [CI], 0.79–0.93) in the prospective test set and 0.94 (95% CI, 0.90–0.97) in the external test set. Pathway analysis revealed dysregulation of the Ras-MAPK pathway and pathways related to inflammation in HCM.

Conclusions:

This study utilized comprehensive transcriptomics profiling with RNA-sequencing in HCM, revealing circulating miRNA biomarkers and dysregulated pathways.

INTRODUCTION

Hypertrophic cardiomyopathy (HCM) is one of the most common inherited cardiovascular diseases, affecting approximately one in 500 to one in 200 individuals in the United States.¹ HCM is caused by mutations in genes encoding sarcomere proteins.² However, despite the high prevalence, it remains unclear through which signaling pathways these genetic mutations mediate HCM pathogenesis and progression to more severe disease manifestations.³ Moreover, it is sometimes challenging, even for experts, to make a diagnosis of HCM based solely on clinical and genotyping data, as genetic testing identifies a positive genotype in only 30%−60% of patients with HCM.²

Research into the human genome has shown that genomic DNA encoding proteins accounts for less than 2% of the entire sequence, and up to 90% of the genome is involved in biological functions related to the regulation of gene expression.⁴ Among the functionally important areas of the genome, the most numerous are the genes for regulatory small non-coding RNAs (sncRNAs), which have significant variety in their structure and function.⁵ Transcriptomics profiling involves a comprehensive analysis of RNA transcripts in the tissue or fluid using high-throughput methods such as RNA-sequencing (RNA-seq).⁶ Comparison of transcriptomes allows for the identification of genes that are differentially regulated in different cell populations, disease states, or in response to different treatments. Thus, transcriptomics profiling has the potential to refine diagnostic accuracy and to elucidate underlying molecular mechanisms of pathogenesis in a variety of diseases, including HCM.

A number of studies have shown that sncRNAs involved in signaling pathways in the heart can be non-invasively detected in plasma.^7,8 Such studies have provided novel insights into the pathogenesis of various cardiomyopathies. In particular, microRNAs (miRNAs; a subset of sncRNAs) have been the most extensively studied in cardiovascular disease, and have been found to be useful in a number of contexts, including in detecting and categorizing heart failure subtypes, diagnosing acute myocarditis, and acting as a biomarker for disease progression in dilated cardiomyopathy.^9–11 Their stability in stored blood makes them particularly attractive candidates as biomarkers.¹² In addition, curated databases linking miRNAs with their target mRNAs are available, along with tools for pathway analysis to derive biological inferences.¹³ However, thus far, no prior studies have applied comprehensive plasma transcriptomics profiling with RNA-seq to patients with HCM. Therefore, our aims in the present study were to (1) identify plasma circulating miRNA biomarkers that distinguish HCM from hypertensive left ventricular hypertrophy (LVH) and (2) specify signaling pathways that are differentially regulated in HCM and correlate with clinical markers of disease severity.

METHODS

The data that support the findings of the present study are available from the corresponding author upon reasonable request.

Study design and sample

In this case-control study, case subjects were patients over the age of 18 years with HCM. Case subjects were recruited via the HCM Programs at Massachusetts General Hospital (MGH) and Columbia University Irving Medical Center (CUIMC) between 2015 and 2021. Control subjects were patients over the age of 18 years with hypertensive LVH. Control subjects were recruited from general cardiology clinics at MGH and other institutions participating in the Partners Biobank. The Partners Biobank, currently known as the Mass General Brigham Biobank, was established in 2010 and is a large (>120,000 consented participants) clinical data and sample repository for research with high-quality samples.¹⁴

The diagnosis of HCM was established based on left ventricular (LV) maximum wall thickness ≥15 mm that was out of proportion to systemic loading conditions.¹⁵ Clinical phenotypes were confirmed via direct patient examination in clinic and chart review. Patients with HCM phenocopies (e.g., Fabry disease, Danon disease, and cardiac amyloidosis) were excluded by extensive clinical evaluations and by performing additional testing (e.g., genetic testing, cardiac magnetic resonance imaging [CMRI]) when needed.¹⁵ Baseline characteristics of study participants were collected at the time of enrollment. Echocardiographic parameters were reported based on the echocardiogram acquired closest to the time of enrollment. Echocardiographic images were obtained using commercially available ultrasound systems (iE33, Philips Healthcare, Amsterdam, The Netherlands). Echocardiographic parameters displayed in Table 1 were obtained using standard protocols. Genetic testing was performed using commercial gene panels for HCM from Clinical Laboratory Improvement Amendments approved labs whose use has previously been validated in HCM.^16–18 Variants categorized as “definitely pathogenic” or “likely pathogenic” were considered to be genotype positive and those classified as “variant of uncertain significance”, “likely benign”, and “benign” were considered to be genotype negative.² The study protocol was approved by the Mass General Brigham Human Research Office/Institutional Review Board and the Institutional Review Board of Columbia University Irving Medical Center. All participants provided written informed consent.

Table 1.

Baseline clinical characteristics of the study sample

Characteristics*	HCM (n=392)	Hypertensive LVH (n=163)	P value
Demographics
Age (year)	60 ± 16	67 ± 11	<0.001
Male	255 (65)	103 (63)	0.65
NYHA class ≥2	190 (49)	15 (9)	<0.001
Race/ethnicity			0.001
Caucasian	305 (81)	117 (72)
African-American	25 (7)	7 (4)
Asian	8 (2)	13 (8)
Other or unidentified	54 (14)	26 (16)
Medical and family history
Prior AF	107 (27)	43 (26)	0.81
Prior VT/VF	22 (6)	3 (2)	0.05
Prior syncope	82 (21)	8 (5)	<0.001
Prior septal myectomy	32 (8)	-	-
Time from septal myectomy to enrollment (year)	3.6 [1.0–7.3]	-	-
Prior alcohol septal ablation	27 (7)	-	-
Time from septal ablation to enrollment (year)	7.0 [2.9–13.5]	-	-
Family history of sudden cardiac death	36 (9)	6 (4)	0.02
Family history of HCM	92 (24)	0 (0)	<0.001
Medications
β-blocker	261 (67)	116 (71)	0.33
Calcium channel blocker	79 (20)	21 (13)	0.04
ACE inhibitor	36 (9)	67 (41)	<0.001
ARB	58 (15)	36 (22)	0.04
Diuretic
Loop diuretic	44 (11)	47 (29)	<0.001
Thiazide	34 (9)	38 (23)	<0.001
Potassium sparing diuretic	25 (6)	23 (14)	0.003
Disopyramide	27 (7)	1 (1)	0.002
Amiodarone	14 (4)	8 (5)	0.47
Blood pressure
Systolic blood pressure (mmHg)	124 ± 16	133 ± 18	<0.001
Diastolic blood pressure (mmHg)	74 ± 9	74 ± 11	0.41
Echocardiographic measurements
Left atrial size (mm)	42 ± 7	42 ± 7	0.81
Interventricular septum thickness (mm)	17 ± 4	13 ± 2	<0.001
Posterior wall thickness (mm)	12 ± 2	12 ± 1	0.44
Left ventricular outflow tract gradient at rest (mmHg)	11 [0–42]	-	-
Left ventricular outflow tract gradient with Valsalva maneuver (mmHg)	19 [0–70]	-	-
Left ventricular ejection fraction (%)	69 ± 10	62 ± 11	<0.001
Left ventricular end-diastolic diameter (mm)	43 ± 6	47 ± 7	<0.001
Left ventricular end-systolic diameter (mm)	27 ± 5	32 ± 7	<0.001
Systolic anterior motion of mitral valve leaflet	177 (46)	1 (1)	<0.001
Degree of mitral regurgitation†	2 [1–2.5]	1 [1–2]	<0.001
Genetic testing (n=216)
Pathogenic or likely pathogenic	55 (25)	-	-

^*Data are expressed as number (percentage), mean ± standard deviation, or median [25^th percentile – 75^th percentile].

^†Degree of mitral regurgitation was converted to numerical values according to the following rule: none=0, trace=1, trace to mild=1.5, mild=2, mild to moderate=2.5, moderate=3, moderate to severe=3.5, severe=4.

Abbreviations: ACE, angiotensin converting enzyme; AF, atrial fibrillation; ARB, angiotensin II receptor blocker; HCM, hypertrophic cardiomyopathy; LVH = left ventricular hypertrophy; NYHA, New York Heart Association; VT/VF, ventricular tachycardia or ventricular fibrillation

Venous blood specimens were collected in K₂EDTA-treated tubes at the time of enrollment. All blood samples were centrifuged and the supernatant plasma was aliquoted and immediately frozen at −80°C.

Transcriptomics profiling using RNA-seq

A number of technologies have been developed for transcriptomics profiling, including RNA microarrays, quantitative reverse transcriptase polymerase chain reaction (qRT-PCR), and RNA-seq.⁶ RNA microarrays lack sensitivity for genes expressed at low or very high levels and have a small dynamic range.¹⁹ qRT-PCR provides a wide dynamic range but is low-throughput and expensive.¹⁹ RNA-seq is the first sequencing-based method that allows the entire transcriptome to be profiled in a high-throughput and quantitative manner, with wide dynamic range, and was thus the technique chosen for this study.¹⁹

Total RNA was extracted from 200 μL of plasma using the Zymo Quick-cfDNA/cfRNA Serum and Plasma kit following manufacturer recommendations (Zymo Research Corporation, Irvine, California). After RNA extraction, all RNA samples were quantified using an RNA ScreenTape on an Agilent TapeStation 2200 (Agilent, Santa Clara, California). A total of 10 μL eluate was used to generate RealSeq-Dual libraries using a single-adapter and circularization method to decrease microRNA (miRNA) sequencing bias (RealSeq Biosciences, Santa Clara, California).^20,21 The libraries were amplified for 20 cycles, pooled to equal nanomolarity, and size-selected using Pippin Prep (Sage Bioscience, Beverly, Massachusetts). The size-selected pool was concentrated using Zymo DNA Clean and Concentrator and profiled using Tapestation (Agilent, Santa Clara, California) and Qubit (Thermo Fisher Scientific, Waltham, Massachusetts) before sequencing on the NextSeq 550 (Illumina, San Diego, California). Sequencing was performed with single 75 bp reads.

Raw sequencing files were merged for each sample to generate a single fastq file per sample. The adapter sequence was trimmed and all reads with <15 nucleotides were removed. Trimmed reads were then aligned to a reference: the miRbase dataset for human miRs. Samtools 1.7 was used to convert the SAM files into sorted BAM files. Bam files were converted to reads files using the BuildBamIndex picard-tool via the PicardCommandLine. A count matrix was made using featureCounts. After obtaining a counts matrix of sample vs. aligned object, differential expression was calculated using the DESeq2 package in R.²² DESeq2 uses shrinkage estimation for dispersions and fold changes to improve stability, reproducibility, and interpretability of estimates. DESeq2 also implements a median-of-ratios method to normalize counts. For quality control purposes, samples with <8 million passing filter reads were excluded (four cases and two controls), and miRNAs with <100 counts were removed from the analysis. Additional details of the protocol have been published elsewhere.²¹ Hemolysis ratios were calculated between miRNAs known to be high in red blood cells compared to those known to be high in plasma in order to measure sample contamination by red blood cells.²³

Univariable analysis

All statistical analyses were performed using R (version 4.0.3). In Table 1, values for continuous variables were presented as mean ± standard deviation when normally distributed and median [25^th percentile – 75^th percentile] when non-normally distributed. Comparisons of characteristics between patients with HCM and hypertensive LVH were made using the paired Student t test, the Mann-Whitney-Wilcoxon test, the χ² test, or Fisher exact test where appropriate. A p value of <0.05 was considered statistically significant.

Development and validation of a transcriptomics-based model to distinguish HCM from hypertensive LVH

The discrimination model was developed in 254 patients who were enrolled through MGH in the first two-thirds of the study period (the training set; 180 cases and 74 controls) (Figure 1). The sparse partial least squares discriminant analysis (sPLS-DA) was performed to develop a transcriptomics-based model to distinguish between HCM and hypertensive LVH in the training set. The sPLS-DA is a form of supervised machine learning which is often used to determine the predictive ability of multi-dimensional data, such as transcriptomics data.²⁴ Transcriptomics data tend to be sparse, such that only a small proportion of analyzed RNA transcripts contribute meaningfully to prediction of the outcome. Furthermore, transcriptomics profiling data are typically highly collinear, with multiple RNAs correlating with one another. The sPLS-DA model was thus selected for its ability to handle both sparse and collinear data.²⁴

Figure 1. Patient allocation into training set, prospective test set, and external test set.

Abbreviations: CUIMC, Columbia University Irving Medical Center; MGH, Massachusetts General Hospital; PB, Partners Biobank

The discriminative accuracy of the model was validated in two different test sets (Figure 1). First, the model was prospectively tested in the test set of 128 patients who were enrolled most recently through MGH (the prospective test set; 91 cases and 37 controls). The discrimination model was further externally validated in a second test set of 173 patients in which cases were enrolled through CUIMC and controls were enrolled through institutions participating in the Partners Biobank other than MGH (the external test set; 121 cases and 52 controls).

A hyperparameter tuning grid was created using the caret package to identify the parameters that optimized accuracy of the sPLS-DA model in the training set using 10-fold cross-validation. The accuracy of sPLS-DA model to distinguish between HCM and hypertensive LVH was then validated in the two test sets. The ROCR package in R was used to generate receiver-operating-characteristic (ROC) curves for the sPLS-DA model performance in the test sets. We also conducted two subgroup analyses by stratifying the cases by (1) the presence of hypertension and (2) genotype positivity. Sensitivity and specificity were calculated using the Youden Index as the cutoff point.

Predicted targets of miRNAs included in the discriminative model were obtained using miRDB, a comprehensive online database.²⁵ All targets in miRDB were predicted by a bioinformatics tool, MirTarget, which was developed by analyzing thousands of miRNA-target interactions from high-throughput sequencing experiments.

Pathway analysis

We performed pathway analysis to specify the canonical pathways that were dysregulated (i.e., either upregulated or downregulated) in patients with HCM compared to those with hypertensive LVH. We identified those miRNAs with values that were significantly different between the two groups based on the Mann-Whitney-Wilcoxon test with p<0.05. We then input the significantly different miRNAs and the reference set of all miRNAs included in the study into the miRNA enrichment analysis and annotation (miEAA) tool. We identified associations between these miRNAs and canonical pathways from the Kyoto Encyclopedia of Genes and Genomes database included in the miRPathDB version 2.0 based on over-representation analysis (ORA) in miEAA.^26,27 Most tools that provide ORA for miRNAs first convert them to their target genes and then perform the analysis on the target genes. However, this approach has been shown to be prone to bias.²⁸ Thus, we chose to use miEAA as it performs ORA directly at the level of the miRNAs. miEAA over-representation analysis calculates the significance of categories for a dataset and shows if a specific category is over-represented or under-represented for the dataset with respect to a reference set. P values are computed by applying the Fisher’s exact test. Given a dataset consisting of miRNAs of which $k$ belong to a certain category $C$ and $l$ do not belong to this category and a reference set of which $j$ miRNAs belong to $C$ and $m$ miRNAs do not belong to this category, the expected number of miRNAs is calculated as $k^{\text{'}}=j*\frac{k+l}{j+m}$. Pathways with a Benjamini-Hochberg corrected false discovery rate (FDR) <0.05 with at least two associated miRNAs were considered positive (i.e., differentially regulated).

Correlation with clinical markers of disease severity in patients with HCM

As an exploratory analysis, the concentrations of all miRNAs were individually correlated with LV ejection fraction, interventricular septal (IVS) wall thickness, New York Heart Association (NYHA) functional class, and left atrial diameter in all patients with HCM using the Spearman rank-order correlation test. Pathway analysis with miEAA was performed for each clinical marker of disease severity using significantly (i.e., univariable p<0.05) correlated miRNAs as detailed in the previous subsection.

RESULTS

A total of 561 patients were enrolled and six patients were excluded through the aforementioned quality control process. Analysis of sample hemolysis ratios demonstrated overall low sample contamination (Supplemental Figure 1). A total of 1,141 miRNAs passed the quality control filters and were included in the study. Thus, 555 patients remained in the analytic cohort, including 392 cases with HCM and 163 controls with hypertensive LVH (Table 1). Patients with HCM were younger on average and had lower systolic blood pressures but greater IVS wall thickness. As would be expected, there was less frequent use of angiotensin converting enzyme inhibitors and diuretics in the HCM group compared to the hypertensive LVH group. The majority of genotype positive patients with HCM had pathogenic or likely pathogenic mutations in MYBPC3 (myosin binding protein C) and MYH7 (myosin heavy chain), as has been previously described in the literature (Supplemental Table 1).²

Discrimination of disease status using the sPLS-DA model demonstrated distinct transcriptomics profiles between patients with HCM and hypertensive LVH, with minimal overlap (Figure 2). After hyperparameter tuning, 69 miRNAs contained within the first two principal components were included in the final sPLS-DA model. The sPLS-DA model developed in the training set using these 69 miRNAs demonstrated an area under the ROC curve of 0.86 (95% confidence interval [CI]: 0.79–0.93; Figure 3A) in discriminating which patients had HCM in the prospective test set. The sensitivity was 93% (95% CI: 86%−98%), specificity was 76% (95% CI: 59%−88%), positive predictive value was 90% (95% CI: 81%−96%), and negative predictive value was 82% (95% CI: 67%−92%). The sPLS-DA model also demonstrated excellent discriminative ability in the external test set, with an area under the ROC curve of 0.94 (95% confidence interval [CI]: 0.90–0.97; Figure 3B). The sensitivity was 86% (95% CI: 78%−92%), specificity was 88% (95% CI: 77%−96%), positive predictive value was 95% (95% CI: 88%−97%), and negative predictive value was 73% (95% CI: 62%−89%). Figure 4 displays the 20 most important miRNAs based on performance in the sPLS-DA model. Predicted targets of these miRNAs included genes in the Ras-MAPK pathway, including MAPK1, MAPK6, MAPK8, MAPK9, MAPK14, MAP3K1, and RAP2C.²⁵ The discrimination model remained robust in the subgroup analyses when cases were stratified by the presence of hypertension (Supplemental Figure 2) or genotype positivity (Supplemental Figure 3).

Figure 2. Two-dimensional score plot using the transcriptomics-based model to distinguish between cases with hypertrophic cardiomyopathy and controls with hypertensive left ventricular hypertrophy.

Orange triangles represent the transcriptomic profile of HCM cases, and blue circles are that of controls.

Abbreviations: HCM, hypertrophic cardiomyopathy

Figure 3. Receiver-operating-characteristic curve using the transcriptomics-based model to distinguish between cases with hypertrophic cardiomyopathy and controls with hypertensive left ventricular hypertrophy in the prospective test set (Panel A) and the external test set (Panel B).

Abbreviations: AUC, area under the receiver-operating-characteristic curve; CI, confidence interval

Figure 4. The 20 most discriminant microRNAs to distinguish between cases with hypertrophic cardiomyopathy and controls with hypertensive left ventricular hypertrophy in the sparse partial least squares-discriminant analysis model.

Abbreviations: miRNA, microRNA

A total of 976 miRNAs were found to be dysregulated in patients with HCM compared to those with hypertensive LVH. Supplemental Table 2 displays miRNA expression from the current study in comparison to seven prior studies of circulating miRNAs in HCM and provides corroboration for the majority of the findings from these smaller studies that did not use RNA-seq.^8,29–34 Pathway analysis revealed differential expression of miRNAs involved in the Ras-MAPK pathway (FDR=0.04; Table 2) and pathways downstream of Ras-MAPK, including the mTOR signaling pathway (FDR=0.01).

Table 2.

Pathways differentially regulated in hypertrophic cardiomyopathy compared to controls with hypertensive left ventricular hypertrophy based on microRNA expression

Pathway	FDR	Expected microRNAs	Observed microRNAs
Hedgehog signaling pathway	0.01	479	501
mTOR signaling pathway	0.01	746	766
Glucagon signaling pathway	0.02	574	594
Cellular senescence	0.02	798	814
FoxO signaling pathway	0.02	786	802
Gap junction	0.03	522	541
Thyroid hormone signaling pathway	0.03	712	729
Phosphatidylinositol signaling system	0.03	536	554
Ras signaling pathway	0.03	796	810
Insulin signaling pathway	0.04	695	711
Apelin signaling pathway	0.04	678	693
Cytokine-cytokine receptor interaction	0.04	674	690
Ether lipid metabolism	0.04	307	322
Insulin resistance	0.04	630	646
Longevity regulating pathway	0.04	625	641
mRNA surveillance pathway	0.04	542	558
Hippo signaling pathway	0.04	742	756
MAPK signaling pathway	0.04	855	866
C-type lectin receptor signaling pathway	0.05	602	617

Abbreviations: FDR = false discovery rate; FoxO, forkhead box O; MAPK, mitogen-activated protein kinase; mTOR, mammalian target of rapamycin

The exploratory analysis of correlation with clinical markers of disease severity revealed 565 miRNAs that were significantly correlated with LV ejection fraction. Pathway analysis identified dysregulation of miRNAs involved in the TNF (FDR=0.005) and MAPK signaling pathways (FDR=0.007) within this set of 565 miRNAs (Supplemental Table 3). There were 179 miRNAs that were significantly correlated with IVS wall thickness. Pathway analysis of these 179 miRNAs revealed dysregulation of pathways related to Ras-MAPK (e.g., FcεRI signaling pathway: FDR=0.005, PI3K-Akt signaling pathway: FDR=0.008, JAK-STAT pathway: FDR=0.01, ErbB signaling pathway: FDR=0.02) as well as pathways involved in inflammation (cytokine-cytokine receptor interaction: FDR=0.02 and interleukin 17: FDR=0.02; Supplemental Table 4). In addition, 140 miRNAs were significantly correlated with NYHA class and 19 miRNAs with left atrial diameter but no pathway dysregulation was identified in these miRNAs.

DISCUSSION

In the present case-control study of 392 cases with HCM and 163 controls with hypertensive LVH that applied RNA-seq to plasma for the first time in HCM, comprehensive transcriptomics profiling of 1,141 miRNAs revealed a set of plasma circulating biomarkers that appear to discriminate between HCM cases and controls. Pathway analysis of the miRNAs that were significantly different between cases and controls exhibited dysregulation of the Ras-MAPK signaling pathways in patients with HCM.^30,35 Ras-MAPK and its upstream pathways were also significantly correlated with LV ejection fraction as well as the degree of LVH, as measured by the IVS wall thickness, in patients with HCM.

Results in context

Many sncRNAs have been found to play an important role in the regulation of gene transcription, epigenetics, and post-translational messenger RNA processing.^6,36 A number of studies have shown that concentrations of circulating sncRNAs can be measured noninvasively in the blood and change in response to an array of acute and chronic disease states.^7,8,37–39 Among sncRNAs, miRNAs are particularly attractive candidates as biomarkers of disease.³⁷ miRNAs are carried within extracellular vesicles and can also be bound to RNA-binding proteins such as high-density lipoprotein.⁷ These transport mechanisms protect miRNAs from degradation and allow for their stability in plasma and frozen samples. Prior studies have identified circulating miRNA signatures that allow for differentiation between patients with heart failure with preserved ejection fraction from those with heart failure with reduced ejection fraction.⁹ Within the field of HCM, a small case-control study (41 cases vs. 41 controls) assessed the plasma levels of 21 miRNAs using RT-PCR and found that patients with HCM had a distinct miRNA profile compared to healthy controls and patients with LVH due to aortic stenosis.³⁸ Another small (n=55) study identified 14 plasma circulating miRNAs that were upregulated in patients with HCM who had diffuse myocardial fibrosis on CMRI compared to those without.³⁷ Although these studies collectively illustrate the potential of plasma transcriptomics profiling to identify biomarkers of HCM and disease progression, the majority of them have employed RT-PCR or microarray assays, which are susceptible to nonspecific hybridization and saturation biases, and have only focused on a small array of sncRNAs.⁶ In addition, study sample sizes were small and often lacked external validation. Given the extensive number of miRNAs studied (1,141 miRNAs) and the large sample size (n=555) of the present study, we replicated the findings of a number of these prior studies and identified novel miRNA associations with HCM by applying RNA-seq to plasma samples from patients with HCM.²⁹

RNA-seq is an emerging approach for transcriptomics profiling with a greater dynamic range than the conventional methods and enables an unbiased investigation of the entire transcriptome. RNA-seq has been applied to transcriptomics studies using patient-derived myocardial tissue and isogenic induced pluripotent stem cell-derived cardiomyocytes, revealing signaling pathways dysregulated in HCM.^40–42 Despite the usefulness of transcriptomics profiling with RNA-seq to specify biomarkers and dysregulated pathways, this technology has not been applied to the investigation of plasma circulating HCM biomarkers. In this context, the present study with a large number of patients (n=555) and comprehensive transcriptomics profiling (>1100 miRNAs passing the quality control filter) adds to the body of knowledge by identifying plasma circulating miRNA biomarkers of HCM and signaling pathways associated with the disease pathogenesis.

Signaling pathways associated with HCM disease status and clinical markers of severity

In the present study, there was differential expression of miRNAs involved in the Ras-MAPK pathway and the PI3K-Akt pathway (which is upstream of Ras-MAPK) in HCM. The PI3K-Akt pathway was also found to be differentially regulated in the miRNAs that were correlated with degree of LVH among patients with HCM. These observations suggest that the Ras-MAPK and associated pathways are not only involved in HCM pathogenesis but also associated with more severe disease. The Ras-MAPK pathway and related pathways have been previously implicated in the development of cardiac hypertrophy. c-H-ras mRNA expression levels were found to be increased among patients with HCM.⁴³ Subsequently, in a study using a transgenic mouse model with constitutively activated Ras, mice exhibited an HCM phenotype, characterized by cardiac hypertrophy and diastolic dysfunction.⁴⁴ In a separate mouse model with a gain-of-function mutation in the RAF1 gene (a key component of the Ras-MAPK pathway), the mouse myocardium similarly demonstrated cardiomyocyte hypertrophy with increased cardiac contractility, akin to the human HCM phenotype.⁴⁵ Furthermore, blockade of the Ras-MAPK pathway with a specific inhibitor of the pathway led to regression of cardiac hypertrophy.⁴⁵ In humans, germline mutations resulting in aberrant Ras signaling are associated with a group of syndromes called RASopathies, which often feature early-onset cardiac hypertrophy mimicking that of HCM.^46,47 PI3K-Akt-mTOR signaling has also been found to be involved in the pathogenesis of cardiac hypertrophy in RASopathies, and prolonged constitutive activation of PI3K in the heart is known to result in hypertrophy.⁴⁸ In addition, miR-21 upregulation was found to activate the ERK/MAPK pathway in cardiac fibroblasts and was noted to be significantly upregulated in the present study.⁴⁹

Strengths of the present study

We implemented several strategies to minimize false-positive and false-negative findings and to augment the external and internal validity of the study. Firstly, the discriminative ability of the model derived from the training set was prospectively validated in a separate test set of patients who were enrolled later in the study period as well as an external validation cohort of patients enrolled at separate institutions. Secondly, to minimize false-positive findings, we used an adjusted FDR cutoff of 0.05 for significantly dysregulated pathways. The FDR threshold of 0.05 ensures that <1 out of 20 pathways declared positive are false-positive. Thirdly, performing pathway analysis over univariable comparisons of miRNAs also strengthens the biological plausibility of the subsequent findings and lowers the risk of false-positive discovery as the miRNAs are interconnected rather than isolated findings. Lastly, this study included the largest number of patients with HCM with the most comprehensive plasma transcriptomics profiling owing to the use of RNA-seq, thus increasing the internal validity of the study and reducing the risk of missing important biomarkers and signaling pathways (i.e., false-negatives).

Potential Limitations

There are several potential limitations to this study. Firstly, only single measurements of miRNAs were taken for each patient, and these measurements may vary over time. Secondly, patients with hypertension and those with HCM were not matched based on clinical characteristics such as age, sex, and baseline medication usage, which could potentially affect miRNA levels. Thirdly, there is limited information available as to the directionality of effect of these miRNAs on Ras-MAPK, though the majority of evidence points to a primarily activating role. Fourth, although pathway analysis identified statistically significant pathways, the FDR values were modest. Also, the relationship between circulating miRNA with tissue level activation is still unknown. These highlight the need for confirmatory analyses in myocardial tissue. Finally, although patients underwent extensive evaluations at HCM centers, genetic testing was not performed on all patients with a diagnosis of HCM. In addition, genetic testing was performed using commercial gene panels for HCM rather than whole exome sequencing. Patients with hypertensive LVH also did not undergo genetic testing. Thus, it is possible that patients could have been misclassified.

Conclusions

In this study, using comprehensive transcriptomics profiling in HCM, we identified plasma circulating miRNAs that appear to discriminate between patients with HCM and hypertensive LVH. Moreover, we determined signaling pathways that may be dysregulated in HCM and correlate with clinical markers of disease severity.

Short commentary.

• What is new?

• Using transcriptomic profiling, we identified a panel of circulating microRNAs which accurately discriminated between cases with hypertrophic cardiomyopathy compared to controls with hypertensive left ventricular hypertrophy. We also found that certain signaling pathways were dysregulated in patients with hypertrophic cardiomyopathy, including the Ras-MAPK pathway and pathways involved in inflammation.

• What are the clinical implications?

• The present study could serve as an important step in specifying potential plasma circulating RNA biomarkers that can be used in conjunction with clinical data to aid clinicians in more accurately diagnosing patients with hypertrophic cardiomyopathy. It also identified dysregulated signaling pathways which could inform future efforts in developing targeted therapeutics for the personalized treatment of patients with hypertrophic cardiomyopathy.

Non-standard Abbreviations and Acronyms:

CMRI

cardiac magnetic resonance imaging

CUIMC

Columbia University Irving Medical Center

FDR

false discovery rate

HCM

hypertrophic cardiomyopathy

LV

left ventricle

LVH

left ventricular hypertrophy

MGH

Massachusetts General Hospital

miEAA

miRNA enrichment analysis and annotation

miRNA

microRNA

NYHA

New York Heart Association

RNA-seq

RNA-sequencing

ROC

receiver-operating-characteristic

RT-PCR

reverse-transcription polymerase chain reaction

sncRNA

small non-coding RNA

sPLS-DA

sparse partial least squares discriminant analysis