Investigating gene-microRNA networks in atrial fibrillation patients with mitral valve regurgitation.

BACKGROUND
Atrial fibrillation (AF) is predicted to affect around 17.9 million individuals in Europe by 2060. The disease is associated with severe electrical and structural remodelling of the heart, and increased the risk of stroke and heart failure. In order to improve treatment and find new drug targets, the field needs to better comprehend the exact molecular mechanisms in these remodelling processes.


OBJECTIVES
This study aims to identify gene and miRNA networks involved in the remodelling of AF hearts in AF patients with mitral valve regurgitation (MVR).


METHODS
Total RNA was extracted from right atrial biopsies from patients undergoing surgery for mitral valve replacement or repair with AF and without history of AF to test for differentially expressed genes and miRNAs using RNA-sequencing and miRNA microarray. In silico predictions were used to construct a mRNA-miRNA network including differentially expressed mRNAs and miRNAs. Gene and chromosome enrichment analysis were used to identify molecular pathways and high-density AF loci.


RESULTS
We found 644 genes and 43 miRNAs differentially expressed in AF patients compared to controls. From these lists, we identified 905 pairs of putative miRNA-mRNA interactions, including 37 miRNAs and 295 genes. Of particular note, AF-associated miR-130b-3p, miR-338-5p and miR-208a-3p were differentially expressed in our AF tissue samples. These miRNAs are predicted regulators of several differentially expressed genes associated with cardiac conduction and fibrosis. We identified two high-density AF loci in chromosomes 14q11.2 and 6p21.3.


CONCLUSIONS
AF in MVR patients is associated with down-regulation of ion channel genes and up-regulation of extracellular matrix genes. Other AF related genes are dysregulated and several are predicted to be targeted by miRNAs. Our novel miRNA-mRNA regulatory network provides new insights into the mechanisms of AF.


RNA preparation
Total RNA including small RNAs was extracted from RA biopsies of six AF patients and six control patients. Briefly, tissue samples were homogenized in QIAzol reagent (QIAGEN, Maryland, USA) using a Precellys 24 homogenizer (Bertin Technologies, Montigny-le-Bretonneux, France). Samples were DNase treated and RNAs were isolated using the miRNeasy kit (QIAGEN, Hilden, Germany) according to manufacturer's instructions. RNA concentration was measured in a NanoDrop 2000 (ThermoScientific, Wilmington, USA) and quality was assessed using the RNA 6000 Nano RNA assay in a 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA). RNA samples with a RIN > 6 were used in further experiments.

RNA-sequencing
We used a total of 50 ng of extracted RNA per sample to perform RNA-sequencing experiments. First, ribosomal RNAs were depleted from total RNA using the Ribo-Zero rRNA removal kit (Illumina, San Diego, California, USA). Barcoded cDNA libraries were prepared using the TruSeq Stranded Total RNA Library Prep Kit (Illumina, San Diego, California, USA) following manufacturer's protocol. Libraries were then sequenced on an Illumina HiSeq 2500 with four library samples per lane (125 bp pair-end reads). The control sample HS24 was not used due to low amount of input RNA. The average number of reads per sample ranged from 71 to 82 million. This work was performed at the Next Generation Sequencing Service Facility, Center for Genomic Medicine, Righospitalet, Denmark (www.genomic-medicine.dk).

Transcriptome sequencing analysis
The raw paired-end reads were aligned to the human GRCh38 reference transcriptome using the Kallisto pseudoaligner with default options [1]. The Kallisto index was built using the protein-coding cDNA reference combined with the non-protein coding transcripts (ncRNA) reference. Both references were obtained from Ensembl (ftp://ftp.ensembl.org/pub/release-95/fasta/homo_sapiens/). Kallisto quantifies the abundance of reads on transcript level. On average, 54.24% of the reads mapped to the transcriptome yielding an average of 41,868,553 mapped reads per sample. The resulting quantification files were imported to R using tximport (Bioconductor) and transformed to transcript per million (TPM) for downstream analysis. Transcript counts were collapsed to gene level using BioMart (Bioconductor). Genes with TPM < 1 between all samples were discarded, leaving a final gene set of 44,852 ensemble gene IDs. Differentially expressed genes in AF compared to control tissue samples were obtained using the Bioconductor package DEseq2 [2]. DEseq2 uses the Wald test for significant testing and the Benjamini-Hochberg method to control for FDR. Genes with false discovery rate (adj.p) < 0.05 and log2fold-change (FC) above 1 or below -1 were considered differentially expressed genes (DEGs). Principal component analysis (PCA) and unsupervised hierarchical clustering of sample-to-sample distance matrixes were used to analyse sample clustering according to transcriptomic similarities. Heatmaps and volcano plots were used to visualize gene expression differences.

Validation of RNA-sequencing by quantitative PCR
Qualitative polymerase chain reaction (qPCR) was performed to confirm the reliability of the RNAsequencing data. Eight genes were tested using the exact same isolated RNA samples from the RA of five  The raw data from the CEL files generated by the Affymetrix's miRNA array was normalized using the robust multi-array average (RMA) method [3]. Microarray quality metrics showed one control sample (HS24) as an outlier (Supplementary Figure S1). The sample was removed from the analysis. The array data was then filtered to include only human probes in the analysis output with a final number of 4,202 mature human miRNAs. The FC values and p-values (p) of expression changes were calculated using the Limma package in R/Bioconductor project [4]. FC values between the two groups were log2 transformed. A cut-off p lower than 0.01 was used for selection of differentially expressed miRNAs. No FC cut-off was applied.
Samples and differentially expressed miRNAs were subjected to unsupervised hierarchical clustering and plotted as a heat map using the pheatmap package from CRAN.

Validation of microarray by qPCR
qPCR experiments were performed to confirm the reliability of microarray data, including eleven mature miRNAs that were both up-and down-regulated in the microarray assay. Three RNA samples from patients with AF and three from control subjects were tested in triplicates. cDNA was synthesized from 10 ng total  Table 1 contribute to the variation observed in the AF samples (13, 14, 15 and 18, 20, 23). B. Heatmap of sample-to-sample distance matrix showing similarities between sample. Dark blue represents a low distance and therefore high similarity. AF -atrial fibrillation; SR -sinus rhythm A B