Circulating microRNAs Identify Early Phenotypic Changes in Sarcomeric Hypertrophic Cardiomyopathy

Abstract

Background.

Hypertrophic cardiomyopathy (HCM) is the most common genetic cardiomyopathy. Pathogenic germline variation in genes encoding the sarcomere is the predominant etiology of disease. However diagnostic features, including unexplained left ventricular hypertrophy (LVH), typically do not develop until late adolescence or after. The early stages of disease pathogenesis and the mechanisms underlying the transition to a clinically overt phenotype are not well understood. In this study we investigated if circulating microRNAs (miRNAs) could stratify disease stage in sarcomeric HCM.

Methods.

We performed arrays for 381 miRNAs using serum from HCM sarcomere variant carriers with and without a diagnosis of HCM and healthy controls. To identify differentially expressed circulating miRNAs between groups, multiple approaches were used including Random Forest, Wilcoxon rank sum test, and logistic regression. The abundance of all miRNAs was normalized to miRNA-320.

Results.

Of 57 sarcomere variant carriers, 25 had clinical HCM and 32 had subclinical HCM with normal LV wall thickness (21 with early phenotypic manifestations and 11 with no discernible phenotypic manifestations). Circulating miRNA profile differentiated healthy controls from sarcomere variant carriers with subclinical and clinical disease. Additionally, circulating miRNAs differentiated clinical HCM from subclinical HCM without early phenotypic changes; and subclinical HCM with and without early phenotypic changes. Circulating miRNA profiles did not differentiate clinical HCM from subclinical HCM with early phenotypic changes, suggesting biologic similarity between these groups.

Conclusions.

Circulating miRNAs may augment the clinical stratification of HCM and improve understanding of the transition from health to disease in sarcomere gene variant carriers.

INTRODUCTION

Hypertrophic cardiomyopathy (HCM) is a disease of the heart muscle that results in left ventricular hypertrophy (LVH) which is not explained by abnormal loading conditions (e.g., hypertension, aortic stenosis) or infiltrative/storage processes. HCM is the most common genetic cardiomyopathy and an important cause of sudden cardiac death in young persons¹. While there are varied etiologies of HCM and clinical manifestations are highly heterogeneous, variants in sarcomere genes are the most common identifiable cause of disease and are responsible for at least 60% of familial cases². Sarcomere gene variants result in autosomal dominant disease with variable and age-dependent clinical expression and penetrance.

Individuals with LVH who have a pathogenic or likely pathogenic sarcomere variant (G+) are referred to as having “clinical” sarcomeric HCM (G+LVH+), while those with a pathogenic or likely pathogenic sarcomere variant, but who do not have LVH on cardiac imaging, and therefore do not have a diagnosis of HCM, are referred to as having “subclinical” sarcomeric HCM (G+LVH-). Subclinical variant carriers are at risk for developing HCM and require lifelong surveillance for disease development^{3, 4}. Our current approach to treating HCM is reactive, beginning once LVH is evident on echocardiogram, and emphasizing symptom relief. Due to genetic testing, those with subclinical HCM (G+LVH-) present a unique opportunity in the cardiovascular field to develop proactive, disease preventing or modifying approaches. Because only a subset of those with subclinical HCM eventually develop disease, precise risk stratification tools that identify those at highest risk to transition to overt HCM are essential and could improve outcomes.

microRNAs (miRNAs) are small noncoding ~22 nucleotide RNAs capable of modulating the expression of many genes. At the most simplistic level, miRNAs bind to target messenger RNA (mRNAs) by recognizing reverse complementary 6 – 8 nucleotide “seed” sequences, most frequently located within 3’ untranslated regions. miRNAs can cause translational repression or RNA destabilization and several array-based studies have detailed changes in miRNA expression in normal versus failing adult human hearts⁵.

Circulating miRNAs are stable in blood and can be useful diagnostic biomarkers in a broad range of cardiovascular diseases⁶. We have published studies of circulating miRNAs as prognostic and diagnostic biomarkers in children with heart failure⁷, Kawasaki Disease⁸, and in association with twin-twin transfusion cardiomyopathy^{9, 10}. There have been several studies investigating circulating miRNAs in HCM, but these studies are limited by small numbers of study participants, are performed in highly heterogenous HCM populations, describe expression of a small number of pre-selected miRNAs, and do not consider the influence of sarcomere gene variants ^11–15.

The purpose of this study was to use a non-biased array approach to investigate whether circulating miRNAs could serve as a clinical stratification biomarker for sarcomeric variant carriers with subclinical and clinical HCM.

METHODS

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

Study Participants

We studied a subset of participants from the Hypertrophic Cardiomyopathy Network (HCMNet) study¹⁶ and the Valsartan for Attenuating Disease Evolution in Early Sarcomeric HCM (VANISH) trial^17–19, focusing on clinical HCM patients <20 years of age at enrollment and subclinical HCM individuals of similar age and sex. Serum samples and clinical data obtained at baseline evaluation (prior to any intervention) were used for this study. Both studies were IRB approved and consent included use of banked serum specimens for future investigation. All participants provided informed consent or youth assent as appropriate. Serum from healthy controls were obtained from the IRB approved University of Colorado Human Tissue Biobank. Healthy controls were recruited from the cardiology or other outpatient clinics at Children’s Hospital Colorado. Controls did not have a family history of HCM and had a normal cardiac exam and/or normal cardiac structure and function on a transthoracic echocardiogram.

Clinical Characteristics of the HCM cohorts

Clinical characteristics were ascertained during participation in the parent HCMNet and VANISH studies^16–19. Genetic testing confirmed that all study participants carried a pathogenic or likely pathogenic variant in a sarcomere gene, including MYH7, MYBPC3, TNNT2, TPM1, MYL2, MYL3, TNNI3, ACTC1. Variant pathogenicity was determined using standard criteria accounting for segregation, conservation, published information and public databases, and absence or very low frequency in appropriate control populations. VANISH investigators with expertise in genetics reviewed questionable variants to determine eligibility by consensus^{20, 21}. Analyses of standard 2-dimensional and tissue Doppler imaging of echocardiograms were performed by blinded personnel associated with the study’s Echo Core Laboratory. LVH was defined by left ventricular wall thickness (LVWT) ≥13 mm or z-score adjusted for body surface area (BSA) ≥ +3. Participants meeting this threshold were designated as clinical HCM (G+LVH+). Participants without evidence of LVH were designated as subclinical HCM (G+LVH-). The subclinical HCM cohort was further classified based on electrocardiographic (ECG) findings and echocardiography with tissue Doppler imaging. Those with ECG abnormalities (Q waves or repolarization abnormalities)²², LV wall thickness to cavity dimension ratio ≥1.9, or an age-adjusted lateral or septal tissue Doppler diastolic velocity (E’) z-score ≤ −1.5 were designated as subclinical HCM with early phenotypic changes (G+LVH-P+). Those with subclinical HCM and normal echocardiograms and ECGs were designated as subclinical HCM without early phenotypic changes (G+LVH-P-). Results for serum N-terminal pro B-type natriuretic peptide (NT-proBNP, pg/mL) and New York Heart Association (NYHA) classification were also obtained.

miRNA Extraction and Array Analysis

10 μl of serum from each participant were submitted to 3 freeze/heat cycles to ensure miRNA release from microvesicles or interaction with serum proteins as previously described by our group²³. 3 μl of the resulting serum was used for array studies. miRNAs were reverse transcribed using a pool of primers specific for each miRNA. Real-time PCR reactions were performed in a 384 well plate containing sequence-specific primers and TaqMan probes in the ABI7900.

Array data were analyzed using Expression Suite Software v1.1 (ThermoFisher). Our group has evaluated miRNA expression by miRNA array in 413 pediatric subjects, including healthy children and children with various medical diagnoses including dilated cardiomyopathy, Kawasaki Disease, single ventricle congenital heart disease, acute respiratory distress syndrome, pulmonary hypertension, and pediatric heart transplant recipients. We identified miR-320 as the least variable miRNA across all groups studied, providing the basis of its selection as the normalizing miRNA for this study. Unless otherwise specifically stated, classification of miRNAs was performed using randomForest package in R version 4.0.0 using the default parameter and 50,000 trees (www.R-project.org). The primary purpose of the analysis plan was to identify the circulating miRNAs that differentiated the clinical groups of interest. Therefore, we used 3 independent methods, random forest, Wilcoxon rank sum test, and logistic regression to identify the top differentiating miRNAs.

Unsupervised classification with Random Forest (RF) analysis⁸.

RF analysis is presented as a multidimensional scaling plot in 2 dimensions (Dimension 1 [Dim 1] on the x-axis and Dimension 2 [Dim 2] on the y-axis). Separation of groups by miRNAs in Dim 1 are noted in the figures. The RF variable importance plots including the Mean decrease accuracy and Mean decrease Gini for all analyses are included in figures. A plateau in the Mean decrease accuracy and the Mean decrease Gini is noted after the top few miRNAs, demonstrating that the top 3 miRNAs represented the largest contributors to the between-group differences and were therefore the focus of the analysis. Hierarchical clustering was performed using hclust function (Ward’s method) and the ClassDiscovery package in R version 4.0.0. Wilcoxon rank sum test was performed between groups and differentiating miRNAs had a q-value (or false discovery rate) of ≤0.15, while accounting for multiple comparisons. Fold change represent data in log₂ scale and fold change was calculated by subtracting the means of the comparison groups. Logistic regression analysis was performed using R version 4.0.0. To control false discovery rate (FDR), raw p-values were adjusted (q-values) by the Benjamini and Hochberg method using p.adjust function in R version 4.0.0 (stats package). The area under the receiver operating characteristic curves (AUC) were calculated using the pROC package in R version 4.0.0. The area under the receiver operating characteristic curves (AUC) were calculated using the pROC package in R version 4.0.0. Because the AUCs are based on RF, which is unsupervised, 95% confidence intervals are not available for the AUC. Receiver operating characteristic (ROC) curves were calculated to determine the sensitivity and specificity for the differentiating miRNAs as described in each figure based on the top 3 miRNAs identified by RF (as described above), and confirmed to be significantly different by Wilcoxon rank-sum test and logistic regression. On the ROC curves, the x-axis (1 – specificity) represents the false positive (FP) fraction [FP/(FP+True Negative)] and the y-axis (sensitivity) represents the true positive (TP) fraction [TP/(TP+ False Negative)]. Clinical variables and outcome data were compared using Wilcoxon rank sum test for continuous variables and Fisher’s exact test for categorical variables, and significance was set a priori at p<0.05.

Pathway Analysis

Associations between miRNAs and canonical pathways from the Kyoto Encyclopedia of Genes and Genomes database (KEGG) were determined using the miRNA enrichment analysis and annotation tool (miEAA)^{24, 25}. Specifically, an over-representation analysis (ORA) with miEAA was used. The alternate approach to ORA, gene set enrichment analysis (GSEA), by default ignores gene sets that contain fewer than 15 genes or more than 500 genes and therefore could not be used for this dataset²⁴. A minimum of four out of the five relevant miRNAs and a q≤0.15 for predicted pathways were necessary for pathways to be considered relevant.

RESULTS

Clinical Characteristics

Clinical characteristics of the cohort are shown in Table 1. Of the 57 sarcomere variant carriers, 32 had subclinical HCM (G+LVH-, 53% female, mean age of 14.2 ± 3.6 years) and 25 had clinical HCM (G+LVH+, 38% female, mean age 14.2 ± 4.6 years). Of the subclinical HCM cohort, 91% were under the age of 18 years, while for the G+LVH+ cohort, 88% were under 18 years. Healthy controls were younger than sarcomere variant carriers (mean age 7.9 ± 5.3 years (p<0.05), all < 18 years old) and 42% were female. Of the 14 healthy controls, 13 had echocardiograms demonstrating normal cardiac structure and function. The one control lacking echocardiographic data had a normal cardiovascular examination and no cardiac symptoms. The myosin heavy chain (MYH7) gene was most commonly involved (53% and 58% in the G+LVH- and G+LVH+ cohorts respectively), followed by myosin binding protein C (MYBPC3; 41% and 27% respectively). Variants in troponin T (TNNT2) and myosin light chains (MYL2, and MYL3) were present in the remaining participants. The mean z-score adjusted to BSA for maximum LVWT was +2.0 ± 1.0 for the subclinical HCM group and +13.3 ± 7.9 for the clinical HCM group (p<0.001). Clinical HCM participants had lower mean age-adjusted E’ z-scores (lateral E’ −1.9 ± − 1.5 vs – 0.6 ± 1.0; septal E’ −2.2 ± 1.6 vs −0.6 ± 1.0, p<0.001 for each) and higher mean NT-proBNP levels (257.2pg/mL [interquartile ratio, IQR 91, 652] vs 41.2 [25, 61], p<0.001). All but one clinical HCM participant was NYHA functional class I. Two sets of siblings are present in this analysis. One is a brother-sister pair with subclinical HCM without early phenotypic changes (G+LVH-P-) who carry a MYH7 variant. The other sibling set is a brother-sister pair classified as clinical HCM (G+LVH+) who also carry a MYH7 variant. All other participants are unrelated.

Table 1:

Study subject demographics.

	G+LVH-	G+LVH- P-	G+LVH- P+	G+LVH+	Control
	N= 32	N= 11	N=21	N=25	N=14
Mean Age, years	14.2 (3.6)	13.4(3.6)	14.6 (3.7)	14.2 (4.6)	7.9 (5.3)
18 year of age or under, n, (%)	29 (91%)	11 (100%)	18 (86%)	23 (88%)	14.0(100%)
Female, n, (%)	17 (53%)	6 (55%)	11 (52%)	10 (38%)	6.0 (42%)
Sarcomeric Gene*, n (%)
MYH7	17 (53%)	6 (55%)	11 (52%)	15 (58%)	NA
MYBPC3	13 (41%)	5 (45%)	8 (38%)	7 (27%)
TNNT2	1 (3%)	0 (0%)	1 (5%)	2 (8%)
MYL2	0 (0%)	0 (0%)	0 (0%)	1 (4%)
MYL3	1 (3%)	0 (0%)	1 (5%)	1 (4%)
BSA-adjusted z-score for maximum LV Wall thickness	2.0 (1.0)	1.7 (0.9)	2.1 (1.0)	13.3 (7.9)	Normal^
E’ lateral age-adjusted z-score	−0.6 (1.0)	−0.3 (0.6)	−0.8 (1.1)	−1.9 (1.5)
E’ septal age-adjusted z-score	−0.6 (1.0)	−0.3 (0.9)	−0.7 (1.0)	−2.2 (1.6)
LVEF, %	67.7 (6.8)	71.0 (8.3)	66.0 (5.3)	71.7 (8.5)
NT proBNP, pg/mL					NA
Geometric mean [IQR]	41.2 [25, 61]	52.2 [31, 78]	36.4 [22, 50]	257.2 [91, 652]
NYHA:
Class I, n (%)	32 (100%)	11 (100%)	21 (100%)	24 (96%)	14 (100%)
Class II, n (%)	0 (0%)	0(0%)	0 (0%)	1 (4%)	0 (0%)

Numbers are mean and standard deviation, unless otherwise indicated

^*Sample had 1 preclinical and 3 overt HCM with 2 variants

^†13 of 14 controls had an echocardiogram performed that demonstrated normal cardiac structure and function. One control did not have an echo performed.

NTproBNP values <5 pg/ml (lower limit of detection) are set to 4 pg/ml

Two G+LVH+ participants were missing NTproBNP values

BSA, body surface area; NYHA, New York Heart Association; LV, left ventricle; LVEF, left ventricular ejection fraction; NTproBNP, N-terminal pro-B-type natriuretic protein; IQR, interquartile range; E’, tissue Doppler diastolic velocity

Circulating miRNAs distinguish sarcomere gene variant carriers from healthy controls

Of 381 RNAs analyzed, 106 miRNAs differentiated the 57 sarcomere variant carriers from the 14 healthy controls (Table S1 and Figure S1A) with a robust AUC for the top 3 miRNAs identified by RF of 0.88. One hundred eight miRNAs differentiated clinical HCM (G+LVH+, n=25) from healthy controls (n=14) (Table S2 and Figure S1B) with an AUC for the top 3 miRNAs identified by RF of 0.94 (of note, removal of one of the siblings from the G+LVH- group did not alter the AUC). There were 50 miRNAs with q≤0.15 (Table S3 and Figure S1C) that differentiated subclinical HCM (G+LVH-, n=32) from healthy controls (n=14) with an AUC for the top 3 miRNAs identified by RF of 0.93 (Figure 1A–B; of note, removal of one of the siblings from the G+LVH+ group did not alter the AUC).

Figure 1.

miRNAs differentiate healthy controls from subclinical hypertrophic cardiomyopathy (HCM) sarcomere variant carriers, both with and without early phenotypic manifestations. (A) Multidimensional scale plot (Dimension 1 on the x-axis and Dimension 2 on the y-axis) depicts results of the Random Forest (RF) analysis and demonstrates circulating miRNAs-31-5p, −523-3p, and −26b-5p differentiate healthy controls (HC, black squares, n=14) from subclinical HCM (subclinical sarcomeric HCM [G+LVH-], green triangles, n=32) in Dimension 1 (Dim 1). miR-26b-5p fold change-2.92, q=0.0001; miR-523-3p fold change 1.21, q=0.0005; miR-31-5p fold change −2.18, q=4.61 E-06. (B) Receiver operator curve (ROC) of these 3 miRNAs show an area under the curve (AUC) of 0.93. (C) RF analysis demonstrates circulating miRNAs-523-3p, −31-5p, and −518f-3p differentiate healthy controls (HC, black squares, n=14) from subclinical HCM with early phenotypic changes (G+LVH-P+, blue triangles, n=21). miR-523-3p fold change 1.35, q=0.0007; miR-518f-3p fold change 1.21, q=0.0021; miR-31-5p fold change −2.08, q=0.0007. (D) ROC of these 3 miRNAs show an AUC of 0.94. (E) RF analysis demonstrates circulating miRNAs-26b-5p, 301a-3p, and −31-5p differentiate healthy controls (HC, black squares, n=14) from subclinical HCM without early phenotypic changes (G+LVH-P-, yellow triangles, n=11). miR-26b-5p fold change −3.73, q=0.0009; miR-301a-3p fold change −2.62, q=0.0009; miR-31-5p fold change −2.35, q=0.0009. (F) ROC of these 3 miRNAs show an AUC of 1.0.

Participants with subclinical HCM were further subdivided into those with and those without evidence of early phenotypic manifestations of sarcomere variants. Those subclinical HCM participants without early phenotypic changes (G+LVH-P-, n=11) lacked LVH (LVWT z-score < +3) and had normal ECGs and age-adjusted z-scores for E’ > −1.5. Subclinical HCM with early phenotypic changes (G+LVH-P+, n=21) did not have evidence of LVH, but had ECG abnormalities or age-adjusted z-scores for E’ ≤ −1.5. There were 74 miRNAs with a q≤0.15 that differentiated healthy controls from subclinical HCM with early phenotypic changes (Table S4 and Figure S1D) with an AUC for the top 3 miRNAs identified by RF of 0.94 (Figure 1C–D). Finally, there were 116 miRNAs with q≤0.15 that differentiated healthy controls from subclinical HCM without early phenotypic changes (Table S5 and Figure S1E) with an AUC for the top 3miRNAs identified by RF of 1.0 (Figure 1E–F). The AUC for this comparison demonstrates that despite morphological similarities between subclinical HCM and healthy controls (i.e. normal ECG and echocardiogram), circulating miRNAs strongly discriminated the cohorts. These data suggest that sarcomeric variants are resulting in subtle changes that cannot be detected by standard echocardiographic or ECG imaging.

Subclinical HCM with (G+LVH-P+) versus without (G+LVH-P-) early phenotypic manifestations

There were 39 miRNAs with q≤0.15 that differentiated subclinical HCM participants based on the presence or absence of early phenotypic changes (Table S6). RF analysis identified miRNAs-181a-5p, −181c-5p, −328–3p as the top 3 miRNAs differentiating G+LVH-P- participants (n=11, yellow triangles) from G+LVH-P+ participants (n=21, blue triangles) (Figure 2A). The variable importance plot (Figure 2B), determined from the RF analysis, demonstrates the relative importance of each individual miRNA in differentiating between the 2 groups, with miRNAs listed in order of decreasing importance. Hierarchical clustering shows the G+LVH-P- and the G+LVH-P+ participants clustering separately (Figure 2C), while box plots show differences in the expression of these 3 miRNAs by Wilcoxon rank-sum test (Figure 2D). The ROC shows an AUC of 0.83 (Figure 2E), consistent with high specificity/sensitivity of these 3 miRNAs to differentiate sarcomere variant carriers with normal LV wall thickness based on the presence or absence of early phenotypic manifestations of ECG changes and/or impaired LV relaxation by echocardiogram.

Figure 2.

miRNAs differentiate subclinical hypertrophic cardiomyopathy (HCM) variant carriers with and without early phenotypic manifestations. (A) Random Forest (RF) analysis demonstrates circulating miRNAs-181a-5p, −181c-5p, and −328-3p, differentiate subclinical HCM with early phenotypic changes (G+LVH-P+, blue triangles, n=21) from subclinical HCM without early phenotypic changes (G+LVH-P-, yellow triangles, n=11). (B) RF variable importance plot including the Mean decrease accuracy and the Mean decrease Gini for each between group comparison of miRNA expression. The further to the right a representative miRNA point falls within the grid, the greater the contribution of that specific miRNA in differentiating the groups of interest. miRNAs are listed from top to bottom in order of decreasing importance in contributing to the difference between the with early phenotypic changes (P+) and without early phenotypic changes (P-) groups. (C) Hierarchical clustering using these 3 miRNAs shows good separation between groups. Siblings (brother, sister pair) are denoted by *. (D) Box plots show that there is statistically significant differences between subclinical HCM with and without early phenotypic changes. miR-181a-5p fold change 1.10, q=0.0071; miR-181c-5p fold change 0.86, q=0.0071; miR-328-3p fold change 1.10, q=0.0248. (E) Receiver operator curve (ROC) of these 3 miRNAs show an area under the curve (AUC) of 0.83.

Because differentiating P- and P+ subclinical HCM is of high clinical relevance as those with early phenotypic changes are more likely to progress to overt disease, a comparative analysis of miRNA expression between these 2 groups was performed using 3 independent statistical methods, RF, Wilcoxon rank-sum test, and logistic regression. Figure 3A demonstrates the top 10 differentiating miRNAs by these 3 methodologies; miRNAs are listed from top to bottom in order of decreasing significance, with all miRNAs having a q-value<0.05. As can be seen by the overlapping segments of the Venn diagram (Figure 3B), 6 of these 10 miRNAs are noted to be significantly different between G+LVH-P- and G+LVH-P+ by all 3 analytic methods. The top 3 miRNAs identified by RF analysis that were used to develop the AUC for differentiating P- from P+ (Figure 2E) are included among the 6 miRNAs that were also different by Wilcoxon rank sum and logistic regression. The differentiating miRNAs and their coefficients with a q-value <0.05 by logistic regression are shown in Table 2 (miRNAs are listed in order of increasing q-value).

Figure 3.

Comparative analysis of the 3 independent methods used to determine the top differentially expressed miRNAs between the without early phenotypic changes (P-) and with early phenotypic changes (P+) groups. (A) Presented in the table are the top 10 miRNAs that differentiate P- from P+ based on analysis by Random Forest (RF), Wilcoxon rank sum test, and logistic regression q-values. miRNAs are listed in order of increasing q-values (q-values for all depicted miRNAs are <0.05.) The top 3 miRNAs by RF analysis that were used to generate the hierarchical cluster and receiver operator curve (ROC) shown in Figure 1 are denoted by * in the table. (B) The overlapping areas of the Venn diagram correspond to miRNAs that differentiate P- and P+ groups based on more than one analytic method. The 6 miRNAs that are significantly different between the P- and P+ groups across all statistical analyses are also highlighted in the table.

Table 2:

Logistic regression was performed for circulating miRNA expression in the without early phenotypic changes (P-) compared to the with early phenotypic changes (P+) subclinical hypertrophic cardiomyopathy (HCM) group. miRNAs with a q-value < 0.05 are shown here along with their corresponding coefficient. miRNAs are listed in order of increasing q-value. 95% confidence intervals (C.I.) and likelihood rations (LR) for each miRNA are also displayed.

miRNA	Coefficient	q-value	95% C.I.	LR
miR-301a-3p	2.548	0.016	[0.819, 6.20]	12.179
miR-181a-5p	3.017	0.016	[0.954, 6.578]	12.700
miR-181c-5p	8.755	0.016	[2.033, 9.454]	13.121
miR-532-5p	1.418	0.023	[0.445, 2.759]	9.533
miR-142-3p	1.503	0.023	[0.454, 3.051]	10.008
miR-193b-3p	1.672	0.023	[0.562, 3.237]	10.549
miR-328-3p	1.760	0.023	[0.526, 3.481]	9.721
miR-491-5p	1.782	0.023	[0.556, 3.572]	10.000
miR-127-3p	3.054	0.023	[0.643, 6.758]	9.733
miR-92a-3p	1.684	0.025	[0.525, 3.284]	9.180
miR-101-3p	1.478	0.025	[0.418, 3.133]	8.953
miR-28-3p	1.601	0.027	[0.454, 3.329]	8.674
miR-324-3p	1.465	0.032	[0.427, 2.821]	8.218
miR-140-5p	1.773	0.033	[0.486, 3.482]	8.068
miR-145-5p	1.107	0.036	[0.260, 2.531]	7.736
miR-106b-5p	0.839	0.037	[0.218, 1.680]	7.588
miR-25-3p	1.178	0.040	[0.302, 2.299]	7.363

In order to determine whether the siblings in the G+LVH-P- group influenced the results as a consequence of their familial relationship we randomly removed one sibling and repeated the analysis (Figure S2). After removal of one sibling, miRNAs-181a-5p, −181c-5p, and −328–3p still separated the groups well by RF analysis (Figure S2A and S2B) and hierarchical clustering (Figure S2C), expression of these miRNAs remained different by Wilcoxon rank-sum test (Figure S2D), and the AUC was 0.81 (Figure S2E). Additionally, when one sibling was randomly removed from the analysis, the 6 overlapping miRNAs in the middle of the Venn diagram in Figure 3 remain different between G+LVH-P- and G+LVH-P+ (q-value <0.05). Logistic regression was minimally affected by the removal of one sibling (Table S7), whereas t-test showed the same number of miRNAs were differentially regulated (Table S8).

Clinical HCM versus subclinical HCM

In contrast to the robust ability of circulating miRNAs to discriminate all other cohort comparisons, miRNAs did not differentiate clinical HCM from subclinical HCM. The RF analysis demonstrates no discrimination between the groups (Figure S3A), there were no miRNAs with q≤0.15, and the AUC was 0.54. Additionally, when looking at subclinical HCM with and without early phenotypic manifestations separately, circulating miRNAs did not differentiate clinical HCM from subclinical HCM with early phenotypic manifestations (Figure S3B). For the 3 miRNAs with q≤0.15 between the G+LVH+ and G+LVH-P+ groups (Table S9) the AUC was poor at 0.67.

However, circulating miRNAs did differentiate clinical HCM from subclinical HCM without early phenotypic manifestations (Table S10). RF analysis identified miRNAs-193b-3p, 301a-3p, and −181a-5p as the top 3 miRNAs differentiating clinical HCM (G+LVH+, n=25, purple circles) from subclinical HCM without early phenotypic changes (G+LVH-P-, n=11, yellow triangles) (Figure 4A). The variable importance plot (Figure 4B) demonstrates the importance of these 3 miRNAs in differentiating G+LVH+ from G+LVH-P-. Hierarchical clustering shows the G+LVH+ and the G+LVH-P- participants clustering separately (Figure 4C), while box plots show differences by Wilcoxon rank-sum test in the expression of these 3 miRNAs (Figure 4D). The ROC shows a high AUC of 0.80 (Figure 4E). In this analysis there were 2 sibling pairs, one pair in the G+LVH-P- and one pair in the G+LVH+ group. In order to determine if the presence of sibling pairs influenced the results, one sibling from each group was randomly removed and the analysis repeated. The same 3 miRNAs, miRNA-193b-3p, −301a-3p, and −181-5p separated groups well by RF and hierarchical clustering (Figure S4A, S4B, and S4C), expression remained different by Wilcoxon rank-sum test (Figure S4D and Table S11), and the AUC remained high at 0.83 (Figure S4E).

Figure 4.

miRNAs differentiate clinical hypertrophic cardiomyopathy (HCM) from subclinical HCM variant carriers without early phenotypic manifestations. (A) Random Forest (RF) analysis demonstrates circulating miRNAs-193b-3p, −301a-3p, and −181a-5p, differentiate subjects with clinical HCM (clinical HCM (G+LVH+), purple circles, n=25) from subclinical HCM who lack early phenotypic changes (G+LVH-P-, yellow triangles, n=11). (B) RF variable importance plot including the Mean decrease accuracy and the Mean decrease Gini for each between group comparison of miRNA expression. The further to the right a representative miRNA point falls within the grid, the greater the contribution of that specific miRNA in differentiating the groups of interest. miRNAs are listed from top to bottom in order of decreasing importance in contributing to the difference between the G+LVH+ and the without early phenotypic changes (P-) groups. (C) Hierarchical clustering using these 3 miRNAs shows good separation between group. (D) Box plots show that there is statistically significant differences between clinical HCM and subclinical HCM without early phenotypic changes. miR-301a-3p fold change 1.1, q=0.11; miR-328-3p fold change 0.91, q=0.11; miR-181a-5p fold change 0.93, q=0.15. There are 2 sets of siblings in this analysis. One set is denoted by * and the other set by # along the x-axis of the cluster analysis. (E) Reciever operator curve (ROC) of these 3 miRNAs show an area under the curve (AUC) of 0.8.

Pathway Analysis suggest miRNAs are involved in the regulation of pathways known to be altered in HCM

The results above demonstrate 5 overlapping miRNAs (miRNAs-181a-5p, −181c-5p, −193b-3p, −301a-3p, and −328-3p, Tables S6 and S8) differentially regulated in clinical HCM and subclinical HCM with early phenotypic changes compared to subclinical HCM without any phenotypic manifestations. To define the possible pathways dysregulated in response to these cardiomyopathic miRNAs, KEGG pathway analysis was done using miEAA. KEGG is a database resource for pathway mapping. Through evaluation of putative targets for these miRNAs, miEAA allows integration of miRNA expression with associated KEGG biological systems or processes. As shown in Table 3, 13 KEGG pathways are predicted to be affected by these specific dysregulated miRNAs, including several metabolism-related pathways. Degradation of amino acids (valine, leucine, isoleucine) and fatty acids, and metabolism of a wide range of substances including ascorbate, aldarate, histidine, lipids and amino acids are predicted to be affected by these cardiomyopathic miRNAs.

Table 3:

Predicted Kyoto Encyclopedia of Genes and Genomes database (KEGG) pathways using over-representation analysis (ORA) targeted by the five cardiomyopathic miRNAs (miRNA-181a-5p, −181c-5p, −193b-3p, −301a-3p, and −328-3p).

KEGG Pathways	P-value	q-value	miRNAs/precursors
Valine, leucine and isoleucine degradation	6.79e-4	0.08869	miRs-181a-5p, −181c-5p, −193-3p, −301a-3p, −328-3p
Fatty acid degradation	8.01e-4	0.08869	miRs-181a-5p, −181c-5p, −193-3p, −301a-3p, −328-3p
Ascorbate and aldarate metabolism	8.67e-4	0.08869	miRs-181a-5p, −181c-5p, −193-3p, −328-3p
Histidine metabolism	0.001393	0.095578	miRs-181a-5p, −181c-5p, −193-3p, −328-3p
Ferroptosis	0.002862	0.095578	miRs-181a-5p, −181c-5p, −193-3p, −301a-3p, −328-3p
African trypanosomiasis	0.002903	0.095578	miRs-181a-5p, −181c-5p, −193-3p, −328-3p
Type II diabetes mellitus	0.003105	0.095578	miRs-181a-5p, −181c-5p, −193-3p, −301a-3p, −328-3p
Steroid biosynthesis	0.003425	0.095578	miRs-181a-5p, −181c-5p, −193-3p, −328-3p
Sphingolipid metabolism	0.003618	0.095578	miRs-181a-5p, −181c-5p, −193-3p, −301a-3p, −328-3p
beta-Alanine metabolism	0.003842	0.095578	miRs-181a-5p, −181c-5p, −193-3p, −328-3p
Tryptophan metabolism	0.003909	0.095578	miRs-181a-5p, −181c-5p, −193-3p, −328-3p
Bile secretion	0.004123	0.095578	miRs-181a-5p, −181c-5p, −193-3p, −301a-3p, −328-3p
Glycerolipid metabolism	0.004321	0.095578	miRs-181a-5p, −181c-5p, −193-3p, −301a-3p, −328-3p

DISCUSSION

Using an unbiased array approach investigating 381 RNAs, we identified circulating miRNAs that robustly discriminated stages of sarcomeric HCM. By comparing comprehensively characterized sarcomere variant carriers to each other and to healthy controls, we found different circulating miRNA profiles in different stages of disease (Figure 5).

Figure 5.

miRNA arrays were performed on healthy controls, sarcomeric hypertrophic cardiomyopathy (HCM), and subclinical sarcomeric HCM. Participants with subclinical HCM do not have left ventricular hypertrophy (LVH) and were further divided into those with early phenotypic changes (P+) or those without early phenotypic changes (P-) on echo and electrocardiogram (ECG). As shown here, the circulating miRNA profile (black miRNA graphic) differs between healthy controls and all sarcomere gene carriers. But the miRNA profile does not differentiate between HCM and P+ (both with red miRNA graphic).

While it is not surprising that circulating miRNAs differentiated patients with clinical HCM from healthy controls, the presence of a sarcomere variant, even when LV wall thickness was normal, also influenced miRNA expression. This finding is notable because it is challenging to identify pathology in apparently healthy subclinical sarcomere variant carriers. Prior studies have indicated that subtle decreases in LV cavity size and E’ velocities, as well as ECG changes and profibrotic tendencies can discriminate healthy controls from subclinical HCM sarcomere variant carriers^{16, 22, 26–29}. However, the differences have often been intra-normal and associated with only modest sensitivity and specificity for separating the groups. In contrast, the miRNA profiles identified in this study were robustly discriminating, differentiating subclinical HCM from healthy controls with an AUC of 0.93 (Figure 1B).

More importantly, differences in circulating miRNAs could be detected within the subclinical HCM cohort. Circulating miRNAs differentiated subclinical HCM with early phenotypic manifestations (P+) from those without early phenotypic manifestations (P-) (Figure 2). This finding suggests that although LVH is not yet evident on echocardiogram, biological effect from the sarcomere variant is present based on ECG changes, abnormalities in LV relaxation, and alterations in circulating miRNA patterns. Interestingly, although miRNAs could discriminate clinically overt HCM (G+LVH+) from subclinical HCM without early phenotypic manifestations (P-) with high sensitivity and specificity (Figure 4), circulating miRNAs did a poor job of discriminating between clinical HCM (G+LVH+) and subclinical HCM with early phenotypic manifestations (P+), suggesting biologic commonality between these two states. The presence of early phenotypic manifestations and a shifting of the miRNA profile may herald more imminent transition to clinically overt HCM.

While this study was not designed to determine the pathophysiologic role of circulating miRNAs, the targets of some identified miRNAs are of interest. Of particular interest were the 5 miRNAs, miRNAs-181a-5p, −181c-5p, −193b-3p, −301a-3p, and −328-3p, that were significantly upregulated in clinical and subclinical HCM with early phenotypic manifestations compared to subclinical HCM without early phenotypic manifestations. While these 5 miRNAs may represent a cardiomyopathic miRNA profile that could be used to provide additional clinical stratification of HCM, these miRNAs may also play a role in the transition from health to disease in sarcomere gene variant carriers.

There is evidence demonstrating pathologic cardiac effects of upregulation of miRNA-181a and −181c in heart disease. In a chronic heart failure post-myocardial infarction rat model, treatment with sacubitril/valsartan resulted in improved cardiac function, decreased fibrosis, and downregulation of miRNA-181a in plasma-derived exosomes³⁰. Using this same model, post-MI rodents treated with intramyocardial injections of a miRNA-181a antagomir demonstrated improved cardiac function, decreased cardiac fibrosis, and modulation of genes associated with fibrosis and hypertrophy. These data suggest the beneficial effects of sacubitril/valsartan could be mediated by modulation of miRNA-181a. This is particularly interesting in light of the recent VANISH trial demonstrating delayed progression of hypertrophy in those with early clinical sarcomere HCM treated with valsartan¹⁹. With respect to miRNA-181c, nanoparticle systemic delivery of miRNA-181c to rats resulted in remodeling of mitochondrial complex IV through targeting of mt-COX1³¹. Another target of miRNA-181c is the anti-apoptotic gene Bcl-2. miRNA-181c mediated inhibition of Bcl-2 gene expression in cultured myocardial cells resulted in altered levels of caspases and cytochrome C, while decreased miRNA-181c expression protected against apoptosis³². These data support the hypothesis that miRNA-181c-5p could be contributing to impaired mitochondrial function and dysregulation of apoptotic pathways in HCM.

Interestingly, miRNA-301a-3p, upregulated in the clinical HCM and subclinical HCM with early phenotypic changes groups, but downregulated in subclinical without early phenotypic changes (compared to healthy controls), is upregulated in the hearts of children with dilated cardiomyopathy³³. While the study of miRNA-301a-3p in cardiovascular disease is limited, there is some evidence for its role in cardiac development and cardiomyocyte differentiation³⁴, and increased expression of miRNA-301b is associated with atrial fibrillation³⁵. This differential expression based on the presence or absence of a cardiac phenotype in sarcomere gene variant carriers is particularly compelling and suggests miRNA-301a-3p could be an important mediator of disease development.

Overexpression of miRNA-328 resulted in increased fibrosis and increased hypertrophy in transgenic mouse models^{36, 37}. In addition, miRNA-328’s effects were paracrine-mediated, as expression of miRNA-328 was increased in cardiac fibroblasts co-cultured with cardiomyocytes transfected with miRNA-328 mimics. This fibrogenesis effect was reversed when cardiac fibroblasts were transfected with a miRNA-328 inhibitor³⁷.

Of the cardiomyopathic miRNAs, up-regulation of miR-193b-3p (q=0.025), miR-301a-3p (q=0.032), miR-328-3p (q=0.05), miR-181a-5p (q=0.108), and miR-181c-5p (q=0.074) were detected in human cardiac tissue from HCM patients compared to controls³⁸. These cardiomyopathic miRNAs are predicted to affect several pathways (Table 3) which are related to HCM or cardiomyocyte hypertrophy: (1) Valine, leucine and isoleucine degradation, fatty acid degradation, glycerolipid metabolism, and Type II diabetes mellitus are predicted to be dysregulated in cardiac tissue from mouse and human cardiac tissue of HCM models/patients³⁹. (2) Specific to cardiomyocyte hypertrophy, inhibition of ferroptosis blunts hypertrophy exacerbation⁴⁰. (3) Modulation of the sphingosine-1-phosphate receptor results in improved diastolic dysfunction in a mouse model linked to HCM⁴¹. (4) Alterations in fatty acid metabolism, a decrease in amino acid composition, including histidine, alanine and tryptophan, and mitochondrial dysfunction in myocardial tissue from HCM patients is described⁴². The overlap between the predicted pathways from HCM cardiac tissue and circulating cardiomyopathic miRNAs suggest these miRNAs are, in fact, important for the phenotypes associated with HCM, and may be important therapeutic targets as well as disease biomarkers.

Penetrance and phenotypic progression to clinical HCM are variable and occur over many years. Studies in animal models of sarcomeric HCM^43–46 and the recent results of the VANISH trial¹⁹ suggest that phenotypic progression and disease development may be modifiable. In the VANISH trial, treatment with the angiotensin receptor blocker, valsartan, abrogated disease progression in study participants with early sarcomeric HCM. Therefore, the timing of treatment initiation is essential and most likely to be successful if begun early in the course of disease when cardiac remodeling is minimal and pathophysiology is not yet entrenched and irreversible^{19, 43, 44, 47, 48}. Developing better tools to identify sarcomere gene variant carriers who are at risk for the emergence of overt disease is an important unmet need. The present study suggests that circulating miRNAs have potential as prognostic indicators of disease development. With further refinement, miRNA profiles could augment current imaging-based surveillance protocols to monitor subclinical sarcomere gene variant carriers, assess response to treatment, and improve our ability to identify those who may be imminently transitioning to clinically overt disease. Collectively, these efforts will enable more personalized clinical care, targeting individuals who may benefit most from more aggressive management rather than those who may have delayed or absent penetrance. Validation of these findings in a larger study population is necessary, and mechanistic studies are essential to develop hypotheses related to therapeutic targets based on miRNA-mediated mechanisms of disease.

LIMITATIONS

Limitations to this study should be noted. First, a small number of participants contributed to this study, however this study was larger than previous circulating miRNA studies in patients with HCM. While we know that sex impacts HCM phenotypic expression, with males tending to develop clinically overt disease earlier in life and adult females tending to be more symptomatic and have worse outcomes^{3, 4, 49, 50}, we are unable to evaluate the influence of sex on miRNA expression in this cohort due to inadequate sample size. Second, there was no validation cohort available to confirm the results of the array experiments, and the AUC values we obtained could be an overly optimistic assessment of the differentiating power of the identified miRNAs. Third, there were 2 sets of siblings included in this study. To confirm that the presence of siblings did not affect the overall results, we removed one of the siblings at random from each relevant group and re-ran the analysis as described. Results were unchanged when only one sibling was included in the analysis. And finally, the healthy control group in this study was younger than the HCM cohort and may not be fully comparable.

CONCLUSIONS

Circulating miRNAs have potential to serve as biomarkers that provide precise clinical stratification of sarcomeric HCM that goes beyond genotype and clinical phenotype. Importantly, there are overlapping miRNAs in participants with clinical HCM and subclinical HCM with early phenotypic changes compared to subclinical HCM without early phenotypic changes reflecting a cardiomyopathic miRNA profile. In addition, the miRNA differences between subclinical HCM with and without early phenotypic changes is robust and has potential to be highly impactful and clinically relevant. The ability of miRNAs to augment the stratification of subclinical HCM could increase the likelihood that disease modifying therapies are preferentially directed towards those most expected to develop overt disease. Results of this study are hypothesis-generating with respect to miRNA-mediated mechanisms of HCM pathophysiology, and future mechanistic studies are justified.

What is new?

• Sarcomeric hypertrophic cardiomyopathy (HCM) is a disease that can run in families and is caused by variations in genes of the sarcomere, resulting in abnormal thickening of the heart muscle. Family members of an affected individual who carry the same sarcomere gene variant must be seen by a cardiologist on a regular basis to look for the onset of HCM. This is the first study to demonstrate that small pieces of RNA (microRNAs, miRNAs) in the blood differ between individuals with sarcomeric HCM and those who carry a sarcomere gene variant but have not yet developed disease.

What are the clinical implications?

• Only a subset of individuals that carry sarcomere gene variants develop HCM. There are no existing tools to accurately predict which individuals that carry sarcomere gene variants will develop HCM and which individuals will never develop disease. There are promising studies demonstrating that some medicines could slow down or prevent development of HCM in those with early or mild disease. A biomarker such as miRNA could be used to identify sarcomere gene variant carriers at highest risk for the future development of HCM as these individuals would be most likely to benefit from preventive therapies.

Non-standard Abbreviations and Acronyms

HCM

Hypertrophic cardiomyopathy

LVH

Left ventricular hypertrophy

G+

Pathogenic sarcomere variant

G+LVH+

Clinical sarcomeric HCM

G+LVH-

Subclinical sarcomeric HCM

miRNAs

microRNAs

mRNA

messenger RNA

HCMNet

Hypertrophic Cardiomyopathy Network

VANISH

Valsartan for Attenuating Disease Evolution in Early Sarcomeric HCM

LVWT

Left ventricular wall thickness

BSA

Body surface area

ECG

Electrocardiographic

E’

Tissue Doppler diastolic velocity

G+LVH-P+

Subclinical HCM with early phenotypic changes

G+LVH-P-

Subclinical HCM without early phenotypic changes

NT-proBNP

N-terminal pro B-type natriuretic peptide

NYHA

New York Heart Association

RF

Random Forest

Dim 1

Dimension 1

Dim 2

Dimension 2

FDR

False discovery rate

AUC

Area under the receiver operator curve

C.I.

Confidence intervals

ROC

Receiver operator characteristic

FP

False positive

TP

True positive

KEGG

Kyoto Encyclopedia of Genes and Genomes database

miEAA

miRNA enrichment analysis and annotation

ORA

Over-representation analysis

GSEA

Gene set enrichment analysis

MYH7

Myosin heavy chain

MYBPC3

Myosin binding protein C

TNNT2

Troponin T

MYL

Myosin light chain

IQR

Interquartile ratio