A microRNA Expression Profile as Non-Invasive Biomarker in a Large Arrhythmogenic Cardiomyopathy Cohort

Arrhythmogenic Cardiomyopathy (AC) is a clinically and genetically heterogeneous myocardial disease. Half of AC patients harbour private desmosomal gene variants. Although microRNAs (miRNAs) have emerged as key regulator molecules in cardiovascular diseases and their involvement, correlated to phenotypic variability or to non-invasive biomarkers, has been advanced also in AC, no data are available in larger disease cohorts. Here, we propose the largest AC cohort unbiased by technical and biological factors. MiRNA profiling on nine right ventricular tissue, nine blood samples of AC patients, and four controls highlighted 10 differentially expressed miRNAs in common. Six of these were validated in a 90-AC patient cohort independent from genetic status: miR-122-5p, miR-133a-3p, miR-133b, miR-142-3p, miR-182-5p, and miR-183-5p. This six-miRNA set showed high discriminatory diagnostic power in AC patients when compared to controls (AUC-0.995), non-affected family members of AC probands carrying a desmosomal pathogenic variant (AUC-0.825), and other cardiomyopathy groups (Hypertrophic Cardiomyopathy: AUC-0.804, Dilated Cardiomyopathy: AUC-0.917, Brugada Syndrome: AUC-0.981, myocarditis: AUC-0.978). AC-related signalling pathways were targeted by this set of miRNAs. A unique set of six-miRNAs was found both in heart-tissue and blood samples of AC probands, supporting its involvement in disease pathogenesis and its possible role as a non-invasive AC diagnostic biomarker.


84 miRNA Cardiac-Related Array Screening in the Blood of AC Patients
The 84 miRNA cardiac-related array screening was also performed in whole blood samples of nine AC gen+ patients from the discovery cohort (Table 1, ID = AC1, AC2, AC10-AC16), of whom two (ID = AC1, AC2) also had tissue-profiling (84 miRNA cardiac-related array and Small RNA-seq), and 4 ctrl, in order to investigate the circulating miRNA profile of AC gen+ patients. We detected 15 DE miRNAs compared to ctrl (AC-blood profile), six significantly overexpressed and nine significantly underexpressed (Log2FC cut-off >1 and <−1, respectively) (Supplementary Material, Figure S1). No gene-dependent miRNA expression profile was observed in circulation.

84 miRNA Cardiac-Related Array Screening in the Blood of AC Patients
The 84 miRNA cardiac-related array screening was also performed in whole blood samples of nine AC gen+ patients from the discovery cohort (Table 1, ID = AC1, AC2, AC10-AC16), of whom two (ID = AC1, AC2) also had tissue-profiling (84 miRNA cardiac-related array and Small RNA-seq), and 4 ctrl, in order to investigate the circulating miRNA profile of AC gen+ patients. We detected 15 DE miRNAs compared to ctrl (AC-blood profile), six significantly overexpressed and nine significantly underexpressed (Log 2 FC cut-off >1 and <−1, respectively) (Supplementary Material, Figure S1). No gene-dependent miRNA expression profile was observed in circulation.

Validation of DE miRNA in a Larger AC Gen+ Cohort
In a larger AC cohort comprising 46 gen+ index cases fulfilling TFC (Supplementary Material, Table S1, #1-46) and 20 ctrl, 10 DE miRNAs shared between tissue and blood profiles were validated. In this setting, we investigated the expression level of three more miRNAs associated with AC in previous studies: miR-21-5p, miR-320a, and miR-184 [22,23,27], as were also found in the AC-tissue profile. Specifically, miR-21-5p was overexpressed in myocardial tissue AC samples (Log 2 FC 2.986 ± 0.83), but not in blood, similarly miR-320a was only underexpressed in myocardial tissue AC samples (Log 2 FC −1.210 ± 0.43) and miR-184 was only detected by NGS screening (Log 2 FC 0.52 ± 0.30).

DE miRNA in Independent Validation Cohorts
The expression level of the 6-miRNA panel was then measured in five independent groups: 24 AC index cases with no rare variants in AC-correlated genes (gen−) (Supplementary Material, Table S1

Validation of DE miRNA in a Larger AC Gen+ Cohort
In a larger AC cohort comprising 46 gen+ index cases fulfilling TFC (Supplementary Material, Table S1, #1-46) and 20 ctrl, 10 DE miRNAs shared between tissue and blood profiles were validated. In this setting, we investigated the expression level of three more miRNAs associated with AC in previous studies: miR-21-5p, miR-320a, and miR-184 [22,23,27], as were also found in the AC-tissue profile. Specifically, miR-21-5p was overexpressed in myocardial tissue AC samples (Log2FC 2.986 ± 0.83), but not in blood, similarly miR-320a was only underexpressed in myocardial tissue AC samples (Log2FC −1.210 ± 0.43) and miR-184 was only detected by NGS screening (Log2FC 0.52 ± 0.30).
Multiple logistic regression analysis demonstrated that the entire six-miRNA panel is able to differentiate AC patients when compared to healthy subjects (AUC = 0.995 and a p value < 0.001) with a sensitivity of 96.67% and a specificity of 100%. This set of six miRNAs was also able to distinguish patients with ongoing disease from silent family pathogenic-variant carriers, AC gen+phen−, with AUC of 0.825 (sensitivity: 88.33% and specificity: 70.59%). Finally, this six-miRNA panel was also able to distinguish AC affected patients from patients with other cardiac entities. The AC group, independent from the genotype, was differentiated from DCM group showing an AUC of 0.917 (sensitivity: 81.67% and specificity: 95%), from the HCM group displaying an AUC of 0.804 (sensitivity: 75% and specificity: 75%), from BrS showing an AUC of 0.981 (sensitivity: 90% and specificity: 100%) and from myocarditis showing an AUC of 0.978 (sensitivity: 93.33% and specificity: 100%) (Figure 4).  (Figure 4).     (Figure 4).
Of note, microT-CDS target prediction algorithm targeted AC pathway predicting 16 genes that were linked to this six-miRNA panel, with a p value of 0.02, among which DSG2 or JUP and TCF7L2, a transcription factor that is involved in the Wnt/β-catenin pathway [29].
Further in silico validation analyses showed 1200 possible genes targeted by our miRNAs with experimental evidences of interaction (miRTarBase analysis) and different potentially involved pathways (KEGG analysis). Wnt signalling pathway emerged among the signalling pathways that were correlated to disease pathogenesis with 4.21 × 10 −31 p value. Adherens junction and TGFβ pathway were confirmed with p values 2.12 × 10 −8 and 1.87 × 10 −7 , respectively. Finally, the AC pathway was also targeted, as previously predicted, with a p value of 0.0153 showing the interaction of our miRNA panel with 12 genes: MITF, EP300, ITGB1, BTRC, PTPN1, SMAD4, CCND1, ACTN4, LRP6, MAPK1, ERBB2, and CSNK1D.

Discussion
MiRNA profiling on myocardial tissue samples of AC patients demonstrated a specific miRNA signature. Six DE miRNAs were in common and validated in a larger AC cohort of blood samples both from Italy and UK, despite differences in the circulating versus myocardial tissue miRNA-expression. This characteristic miRNA AC profile provides non-invasive means to distinguish patients with classic AC phenotype from gen+ family members without AC phenotype and patients that are affected by other forms of cardiac disease.

Previous Studies on miRNA in AC
AC is caused by heterozygous pathogenic variants in genes encoding for proteins of the desmosomal complex, identified in nearly 50% of patients and exhibiting low and age-dependent disease penetrance [6]. Thus, it has been advocated that clinical heterogeneity and incomplete penetrance may be caused by other factors than genetics, which synergistically contribute in the expressivity of the disease, such as DE miRNAs [12]. Recent studies in 24 HTx AC patients assessed 1078 miRNAs on myocardial tissue samples and showed that 21 of them were significantly DE, establishing miR-21-5p and miR-135b as the modulators of Wnt and Hippo pathways [22]. Other studies that were performed in PKP2-deficient cardiomyocytes demonstrated that, among 750 miRNAs analysed, only miR-184 downregulation might be implicated in the disease pathogenesis [26]. Instead, transcriptome analysis in cardiac stromal cells derived from three AC patients and three controls showed three DE miRNAs linked to AC, among which miR-29b-3p [25]. Previously, plasma samples of three AC patients and three controls were screened for 377 miRNAs, identifying miR-320a as a possible AC biomarker. The underexpression of miR-320a was then validated in a larger cohort (36 AC, 53 controls, and 21 idiopathic ventricular tachycardia-IVT) showing an AUC of 0.69 [23]. A subsequent study evaluated 11 circulating miRNAs on 28 AC patients with definite diagnosis, 11 with borderline/possible AC diagnosis, and 23 had IVT, showing that four miRNAs (miR-144-3p, miR-145-5p, miR-185-5p, and miR-494) were significantly overexpressed in definite AC patients; specifically, miR-494 showed higher circulating levels in definite AC patients with recurrent ventricular arrhythmias after ablation (AUC = 0.832) [24]. More recently, the miRNA expression profile of a transgenic mouse model with the human DSG2 Q558* truncated protein showed three DE miRNAs when comparing the left ventricle and RV; specifically, miR-217-5p and miR-708-5p were upregulated and miR-499-5p were downregulated in the mouse RV [30].
Taken together, literature and data from our experiments on myocardial tissue and blood samples that were derived from AC patients, tissue-related miRNA expression profiling showed miR-21-5p, miR-29b-3p, and miR-144-3p overexpression, which was not observed in blood samples from AC patients. Similarly, miR-320a and miR-494 were significantly underexpressed in the myocardium, but no reduction of circulating levels was detected in the larger AC cohort. The other previously reported miRNAs were either normally expressed in tissue and in circulation or they were not detected by small RNA- Seq (miR-184, miR-135b, miR-145-5p, miR-185-5p, miR-217-5p, miR-708-5p, and miR-499-5p). The above findings demonstrated that the use of different technical approaches, analysis strategies, and specimens (animal models, cells, plasma) for miRNA detection, may deeply influence the congruence among studies.
On the contrary, our miRNA expression profiles that were obtained from the RV of HTx AC patients and blood samples from AC patients fulfilling TFC showed an unbiased six DE miRNAs signature obtained after a thoughtful experimental design and deep filtering of miRNAs through the validation process from a wide microRNA screening with two different techniques to the six-miRNA panel validated in the largest cohort reported so far. We performed our analysis, including only AC patients with definite TFC, as demonstrated by the same miRNA profile that was obtained both in gen+ and gen− patients as well as an independent AC-UK cohort with gen+ and gen− patients. This miRNA profile is independent from the genotype and it is likely a consequence of the disease phenotype.
Further, AC-related miRNA signature specificity was supported by an in silico correlation with signalling pathways that were linked to disease pathogenesis. Based on the target microT-CDS predictor, our six miRNA panel directly targeted 16 genes that were related to AC, among which the components of the desmosomal complex or transcription factors ie JUP, DSG2, or TCF7L2. Additionally, the experimentally validated miRNA-gene interactions showed a correlation with Wnt, Hippo, and/or TGFβ signalling pathways [31,32]. However, other studies will be necessary to determine the role of these six DE miRNAs as the precursors or consequence of the disease phenotype.

Comparison between Tissue and Blood miRNA Profiles
Several studies advanced a possible role of circulating miRNAs in distant cell-cell communication through their secretion into protective vesicles (exosomes), bound to RNA-binding proteins, or even released within apoptotic bodies [12]. In an acute myocardial infarction model, it was demonstrated that increased plasma miRNA levels were the consequence of myocardial cell damage [33], and in a liver injury model, decreased hepatic levels of miRNAs reflected their increase in patients' plasma as if they were released from the injured liver to circulation [34]. Further, the miRNA expression profiles in tumor cells demonstrated the release of these miRNAs from the tissue in the blood, highlighting that circulating DE miRNAs may be the consequence of tumour tissue changes detectable in blood as non-invasive biomarkers [35].
Accordingly, four of our six miRNA-panel lacked directionality consistency in tissue and blood profiles. Specifically, we observed three miRNAs (miR-122-5p, miR-182-5p, miR-182-3p) that were underexpressed in tissue and overexpressed in the circulation, supporting the hypothesis that the high blood levels of miRNAs might be released from apoptotic or necrotic cardiomyocytes, indicating progression of tissue damage. As such, we presume that myocardial damage is reflected in the circulation and that these miRNAs can be used as non-invasive biomarkers of the disease, enabling differential diagnosis. Whether the release of these specific miRNAs in the circulation is due to active secretion by macrovesicles or passive release/leakage by apoptotic cardiomyocytes needs to be elucidated.
On the other hand, miRNA overexpression in tissue and underexpression in the circulation might indicate an active uptake of miRNAs by different cell types, as demonstrated in Acute Myeloid Leukemia, in which the miR-92a levels were decreased in plasma and increased in cells [36]. As such, we found miR-142-3p overexpression in myocardial tissue and underexpression in circulation, supporting the idea of miRNAs active uptake from the bloodstream into the myocardial tissue. Additional studies should determine the cell origin of miR-142-3p and its role in disease pathogenesis.

Discriminatory Power: A New Biomarker for AC?
Current diagnosis rests on fulfilling a set of clinical criteria and on genetic testing, which are relatively specific, but not highly sensitive. In the largest AC cohort with definite TFC so far, we demonstrated, by a comprehensive miRNA expression analysis, that the expression level of six DE miRNAs is a consistent feature in patients with AC, and it is not found in other forms of heart-muscle disease. The strength of this six-miRNA signature is based on its ability to differentiate patients affected by AC from healthy subjects with approximately 97% sensibility and 100% specificity. In this study, gen+ and gen− AC TFC patients from Padua and UCL research centers shared the same miRNA profile as to indicate that expression alterations might be a consequence of the disease phenotype. The lack of known biomarkers in AC makes its validation to be used as biomarker difficult and challenging.
Further, the same six-miRNA dataset differentiated AC index cases from patients affected by other primary cardiomyopathies or phenocopies, such as DCM, HCM, BrS, and myocarditis. In this setting, the AC patients were discerned from DCM patients with 82% sensitivity and 95% specificity, from HCM patients with 75% sensitivity and 75% specificity, from BrS patients with 90% sensitivity and 100% specificity, and from myocarditis patients with 93% sensitivity and 100% specificity. Their ability as a whole set to differentiate the disease phenotype from other cardiac diseases led to the hypothesis that this six miRNA signature might be used as a non-invasive biomarker of the disease. Although additional validation studies will be required before this new test can be used clinically, this is a potentially important distinction, which hints at disease-specific mechanisms and suggests that DE miRNAs may cause clinical presentation variability. Finally, this potential biomarker was able to differentiate AC patients with disease phenotype from AC gen+phen− family members, so as to underline the fact that the genetic status alone is not sufficient to determine fully penetrant disease. As such, this six-miRNA set is a potential tool for establishing AC diagnosis in family members of an index patient without extensive clinical workup, which often leads to equivocal results. A longer follow up period and a larger cohort are needed to demonstrate its predictive value for disease prognosis.

Study Cohort
The study population (discovery and validation cohorts, Figure 1A In this regard, we analysed nine frozen right ventricle (RV) myocardial samples from heart transplanted (HTx, disease end-stage) AC index cases (AC1-AC9) and 99 whole blood samples of AC definite TFC patients in both discovery and validation cohorts (AC1, AC2, AC10-AC16, #1-70, 20 AC-UK, Figure 1A). HTx AC cases presented an end-stage disease with biventricular involvement. The 99 blood samples were collected from index cases with a definite AC diagnosis according to TFC that was mostly enrolled for arrhythmic clinical presentation and without heart failure symptoms. Exception made for 2 HTx patients (AC1, AC2) who underwent analysis for both blood and heart tissue specimens. The ctrl group comprised four RV myocardial tissue samples from subjects who died from conditions other than cardiac diseases and 24 blood samples of healthy subjects. Finally, only whole blood samples were analysed in the independent validation cohort ( Figure 1A). HCM, DCM, and BrS affected patients were diagnosed according to the current recommendations for each cardiac disease [37][38][39] based on their symptoms and clinical signs, while myocarditis affected patients were diagnosed by endomyocardial biopsy [40].
Next Generation Sequencing previously screened the entire study population (TruSight Cardio Sequencing Kit, Illumina, San Diego, CA, USA) and written informed consent was provided for all samples, conforming to the principles that were outlined in the Declaration of Helsinki. In particular, 62 of the 86 AC-PD carried a rare variant on a desmosomal gene (AC gen+) and 24 were genotype-negative (AC gen−) ( Table 1, Supplementary Material Table S1). Ten of 20 AC-UK index cases carried at least a rare variant on a desmosomal gene: 5 patients carried a rare variant in DSP, 3 in PKP2 and 2 were compound heterozygous, DSP/PKP2 and PKP2/JUP. All 20 HCM, 20 DCM, and 13 BrS index cases carried a rare variant in a disease-associated gene and 13 myocarditis cases were free of pathogenic variants in cardiac-related genes. The variants were classified according to American College of American Genetics (ACMG) [26].

miRNA Isolation and Quantification
Total RNA, including small RNAs, was isolated from 20 mg frozen RV myocardial tissue and 50 µL of whole blood samples while using RNeasy Mini Kit (Qiagen, Hilden, Germany), according to manufacturer's instructions (Supplementary Material).
Library preparation while using TruSeq Small RNA Library Preparation kit (Illumina, San Diego, CA, USA) was used on tissue samples of three AC (AC1-AC3) probands and two ctrl whereas Next Flex Small RNASeq kit (PerkinElmer, Waltham, MA, USA was used on blood samples of 4 AC (AC1, AC2, AC10-11) probands and 4 ctrl, according to manufacturer's instructions. Equimolar pooled samples (10pM) were sequenced on a MiSeq platform (Illumina, San Diego, CA, USA). BaseSpace Sequence Hub was used for alignment and subsequent pairwise differential expression analysis (DESeq2) and it led to the identification of DE miRNAs, precursor groups, miRNAs families, and piRNAs for each pair of sample groups (Supplementary Material).

In Silico Target Prediction
In silico target prediction was performed while using DIANA mirPath v3 [28], a web-miRNA targets prediction server based on experimentally validated (TarBase v7.0) and theoretical (microT-CDS) miRNA:gene interactions, providing pathways that were enriched with targeted genes of DE miRNAs. Additionally, miRTarBase v2018 [30] was interrogated for experimentally validated targeted genes of miRNAs, and they were subsequently analysed on NetworkAnalyst.

Statistical Analysis
Statistical analysis was conducted with GraphPad Prism version 7.0 for Windows (GraphPad Software, La Jolla, CA, USA, www.graphpad.com) and MedCalc Statistical Software version 16.4.3 (MedCalc Software bvba, Ostend, Belgium; https://www.medcalc.org; 2016). Continuous variables are always presented as mean ± standard error of mean (mean ± SEM), unless otherwise specified. The results are expressed in terms of Log 2 Fold-change (Log 2 FC). A miRNA was considered to be underexpressed when the fold-change is lower than 0.5 (Log 2 FC < −1) and overexpressed when the fold-change is greater than 2 (Log 2 FC > 1).
The expression of each miRNA was compared between the disease-groups and a control group with non-parametric Mann-Whitney test since the distribution was not normal and Kruskal-Wallis test was used for multiple comparisons adjusting the p value with the False Discovery Rate method (two-stage step-up Benjamini, Krieger, and Yekutieli).
The area under the curve (AUC) of Receiver Operating Characteristic (ROC) plots are two-dimensional graphs, in which sensitivity is plotted on the Y axis and 1-specificity is plotted on the X axis. The closer the apex of the curve toward the upper left corner, the greater the discriminatory ability of the test (i.e., the true-positive rate is high and the false-positive 1-Specificity rate is low). This is quantitatively measured by the AUC. The maximum value for the AUC is 1.0, thereby indicating a (theoretically) perfect test (i.e., 100% sensitive and 100% specific). An AUC value of 0.5 indicates no discriminative value (i.e., 50% sensitive and 50% specific) and it is represented by a straight, diagonal line extending from the lower left corner to the upper right. There are several scales for AUC value interpretation, but, in general, ROC curves with an AUC ≤0.7 are not clinically useful and an AUC of 0.97 has very high clinical value. However, there are confidence intervals around this value that must be taken into consideration [41]. AUC was estimated with 95% confidence interval to assess the diagnostic accuracy of each selected miRNA. Sensitivity and specificity were estimated with a 95% confidence interval after detecting the cut-off value with the Youden method. Multiple logistic regression analysis was performed on the six miRNAs from the final panel to study the diagnostic power of all siz miRNAs together on AC (gen+, gen−) group in comparison with ctrl, AC gen+phen−, HCM, DCM, BrS and Myocarditis groups. ROC analysis was performed on predicted results.

Conclusions
Our miRNA expression profiles that were obtained from the RV of HTx AC patients and blood samples from AC patients fulfilling TFC, showed an unbiased six DE miRNAs signature identified while using a targeted experimental design and a deep filtering validation process.
The unique six-miRNA panel identified in this study is characteristic and specific for AC, exhibiting a great discriminatory diagnostic power. In this setting, this set of miRNAs is the best potential miRNA panel as non-invasive AC diagnostic biomarker reported so far.

Limitations
In this study, blood and tissue samples from a small AC cohort of genotype-positive index cases were used as a discovery cohort to identify a subset of miRNAs in the blood helpful for the detection of AC in a non-invasive way. In this setting, more miRNAs were DE using Small RNA-seq, which have not been further investigated. The largest AC cohort reported so far was used for validation of this subset of miRNAs to overcome the bias of the small discovery cohort. Although several independent validation cohorts (gen+phen− family members and different cardiac-disease groups) were used to verify the specificity and sensitivity of our data, the number of subjects included in each cohort is small. Larger cardiac-disease cohorts are required to determine the possible clinical use of this miRNA set. Further experiments on targeted signaling pathways by these miRNAs and gene expression analysis are necessary for determining their implication in disease pathogenesis.
Supplementary Materials: The following are available online at http://www.mdpi.com/1422-0067/21/4/1536/s1, Figure S1: DE miRNAs in the AC-tissue profile and blood profile on the discovery cohort by the 84 cardiac-related array (Log 2 FC, mean ±SEM). Figure S2 Table S1: Clinical data of the AC index cases including discovery and validation cohorts. Table S2: Discovery cohort DE miRNAs shared between tissue and blood arrays; Log 2 FC (mean ±SEM) values of the 10 miRNAs assessed by 84 cardiac-related array. Consistent directionality of the DE in tissue versus blood. Table S3: Thirteen miRNAs expression in the validation cohort; Log 2 FC (mean ±SEM) values of the 13 miRNA analysed on 46 AC gen+ and 20 controls (validation cohort); p value calculated by the Mann-Whitney test.