Aberrant DNA methylation of miRNAs in Fuchs endothelial corneal dystrophy

Homeostatic maintenance of corneal endothelial cells is essential for maintenance of corneal deturgescence and transparency. In Fuchs endothelial corneal dystrophy (FECD), an accelerated loss and dysfunction of endothelial cells leads to progressively severe visual impairment. An abnormal accumulation of extracellular matrix (ECM) is a distinctive hallmark of the disease, however the molecular pathogenic mechanisms underlying this phenomenon are not fully understood. Here, we investigate genome-wide and sequence-specific DNA methylation changes of miRNA genes in corneal endothelial samples from FECD patients. We discover that miRNA gene promoters are frequent targets of aberrant DNA methylation in FECD. More specifically, miR-199B is extensively hypermethylated and its mature transcript miR-199b-5p was previously found to be almost completely silenced in FECD. Furthermore, we find that miR-199b-5p directly and negatively regulates Snai1 and ZEB1, two zinc finger transcription factors that lead to increased ECM deposition in FECD. Taken together, these findings suggest a novel epigenetic regulatory mechanism of matrix protein production by corneal endothelial cells in which miR-199B hypermethylation leads to miR-199b-5p downregulation and thereby the increased expression of its target genes, including Snai1 and ZEB1. Our results support miR-199b-5p as a potential therapeutic target to prevent or slow down the progression of FECD disease.


Results
Global DnA methylation patterns of miRnA sequences are altered in fecD. Our prior genomescale analysis of the DNA methylation landscape of corneal endothelial tissue found a significant difference between FECD and normal control patients 30 . In particular, we identified a high number of differentially methylated miRNA sequences 30 . Because DNA methylation plays a central role in regulating miRNA expression 34,35 and widespread miRNA downregulation has been observed in FECD 36 , we performed a subanalysis of the Illumina Infinium HumanMethylation450 (HM450) array data to focus on the 2,227 miRNA probes (targeting 463 miRNA genes in total). Sample pairwise correlation and hierarchical clustering analyses revealed differential genome-wide miRNA DNA methylation patterns in FECD samples compared with normal control samples ( Supplementary Fig. 1). Further nonparametric principle component analyses showed that this variance was not attributable to the clinical variables of age, sex, pachymetry, or guttata grading (data not shown).
The majority of differentially methylated miRNAs display promoter DNA hypermethylation in fecD. We next examined and compared the DNA methylation levels of individual miRNA sequence between FECD cases and controls by comparing single HM450 probe. Of the 2,227 miRNA-associated probes (targeting 463 miRNA genes), 216 probes (targeting 156 miRNA genes) were differentially methylated in the FECD samples (p < 0.05; Fig. 1a). Of the 216 probes, the large majority (154 probes; 71%) were hypermethylated in the FECD samples, and a small minority (62 probes; 29%) of sequences were hypomethylated (Fig. 1b). Almost all of the differentially methylated probes (208 probes; 96%) targeted miRNA promoter sequences (148/154 hypermethylated probes and 60/62 hypomethylated probes; Fig. 1b). None of the miRNA probes were significantly differentially methylated with respect to age or sex, suggesting that these parameters are not major drivers of DNA methylation changes in miRNA genes in FECD patients (data not shown). Table 1 shows details on the 20 top ranking differentially methylated miRNA genes in the FECD samples as compared to controls. miR199-B was found to be the most extensively hypermethylated miRNA gene and miR-1182 was the most extensively hypomethylated miRNA gene in FECD samples (Table 1).

Genomic locations of differentially methylated miRNA genes with respect to their host genes.
The majority of miRNAs are located within intronic or exonic regions of protein-coding genes (host genes), and increasing evidence suggests a functional relationship between miRNAs and their host genes 37 . Therefore, we next sought to examine the spatial relationship of the differentially methylated miRNA probes with their host genes. We mapped all 216 differentially methylated miRNA probes to their host genes using the Ensembl Genome Browser (https://uswest.ensembl.org/index.html). Of the 154 hypermethylated miRNA probes, 74% (113 probes) corresponded to intragenic sequences and 27% (41 probes) occurred in intergenic sequences (Fig. 1c). Of the intragenic probes, 41% (63 probes) occurred within intronic sequences, 23% (36 probes) within exonic sequences, and 9% (14 probes) in intron/exon boundaries (Fig. 1c). Similar to the hypermethylated probes, the majority of hypomethylated miRNA probes occurred in intragenic sequences (71%, 44 out of 62 probes) and intronic sequences (44% intronic, 24% exonic, 3% in intron/exon boundary sequences) (Fig. 1d). Together, these data show that the majority of differentially methylated miRNA probes occur in intragenic and intronic sequences of their host genes.
Gene body regions of miRnA host genes are frequent targets of aberrant DnA methylation in fecD. Emerging evidence has revealed mutually regulatory roles between particular miRNAs and their host genes [38][39][40][41] . Therefore, to decipher the function of DNA methylation in the epigenetic regulation of miRNAs and their host genes, we next assessed the methylation status of host genes for the 156 differentially methylated miRNA genes in FECD. The list of 156 host genes was curated using publicly available miRNA databases, including Ensembl Genome Browser (https://uswest.ensembl.org/index.html), miRIAD (http://www.miriad-database.org), and miRStart (http://mirstart.mbc.nctu.edu.tw/about.php). A total of 1,823 probes mapped to CpG sites in the www.nature.com/scientificreports www.nature.com/scientificreports/ set of 156 host genes. A volcano plot display of the DNA methylation status of the 1,823 probes showed that 239 probes were differentially methylated in FECD samples compared to control samples (p < 0.05; Fig. 2a). In fact, most (188; 79%) of the 239 differentially methylated CpG sites were hypermethylated in FECD samples, with only a minority (51; 21%) of them hypomethylated. Alignment of the probe sequences to the human genome database revealed that the vast majority (192; 80%) of the 239 differentially methylated CpG sites were located in the gene body regions of miRNA host genes (Fig. 2b). A substantial proportion (161; 86%) of the 188 hypermethylated probes targeted CpG sites within gene bodies, whereas only 14% of them were mapped to the promoter regions of corresponding miRNA host genes. Similarly, a high percentage (31; 61%) of the 51 hypomethylated probes also mapped to gene body sequences of miRNA host genes. Table 2 provides detailed information on the top 20 differentially methylated miRNA host genes identified in the FECD samples as compared to controls. Given that miRNAs may be co-regulated with their host genes, we further analyzed the DNA methylation data for co-methylation patterns. We identified a subset of miRNAs and their host genes to be co-methylated at their corresponding promoter CpG sites, suggesting that DNA methylation may play an important role in regulating the co-expression of miRNAs and their host genes in FECD (Fig. 2c).
Validation of miRnA DnA methylation changes using MethyLight. To validate and quantify miRNA DNA hypermethylation changes identified by the global array 30 , MethyLight analysis was performed on an additional cohort of control and FECD patient corneal samples (Supplementary Table 2) 42 . We assessed the promoter DNA methylation status of following miRNAs: miR-199A1, miR-874, miR-140, miR-23B, and miR-1306 ( Supplementary Fig. 2, Supplementary Table 1). These miRNAs were selected for validation because of their important roles in the pathogenesis of FECD or in the regulation of key cellular processes (e.g. cell survival, oxidative stress, inflammation, fibrosis, and deposition of extracellular matrix) [43][44][45][46][47][48][49][50] . The MethyLight results confirmed DNA hypermethylation in the FECD samples compared with control samples (Fig. 3) and verified our array   www.nature.com/scientificreports www.nature.com/scientificreports/ findings. All five MethyLight assays gave higher mean DNA methylation values (Percent of Methylated Reference, PMR) in the FECD samples compared to the control samples. Among these five miRNA genes, miR-199A1 and miR-23B were found to be methylated in the FECD samples but unmethylated in the control samples (Fig. 3). For miR-199A1, the average PMR values were 20 and 3 for FECD and control samples, respectively (p = 0.039; Fig. 3). For miR-23B, the mean PMR values were 34 for FECD samples and 5 for control samples (p = 0.038; Fig. 3). There are considerable inter-individual variations in DNA methylation levels of each miRNA among FECD patients, as reflected by PMR values (Fig. 3). We observed higher DNA methylation levels of miRNAs in patients with a more severe disease state. Recent studies have demonstrated that DNA methylation levels can be associated with disease severity 51,52 . Taken together, these MethyLight results confirmed hypermethylation of miRNA sequences in FECD tissue compared with normal control samples found in the genome-wide array using an independent set of patient samples.
Relationship between miRnA DnA methylation status and expression. The majority of the miRNA DNA methylation changes observed in FECD tissues occurred in miRNA promoter sequences. Because promoter DNA methylation is often inversely correlated with gene expression levels 53,54 , we next sought to determine how DNA methylation might affect miRNA expression in FECD tissue. Matthaei et al. previously compared miRNA expression profiles of corneal endothelial samples obtained from FECD patients and from normal donors using transcriptome analysis 36 . Their results demonstrated downregulation of 87 miRNAs in FECD compared with normal endothelium and suggested that altered miRNA expression may play an important role in the pathogenesis of FECD disease 36 . Therefore, we integrated our DNA methylation data with their miRNA expression data and generated a Venn diagram showing all differentially methylated and differentially expressed miRNAs (Fig. 4a). Of 156 miRNAs that are hypermethylated and 87 miRNAs that have down-regulated expression in FECD, 18 miRNAs have concurrent hypermethylation and decreased expression in FECD compared to the control samples (Fig. 4a,b). In particular, miR-199b-5p expression was almost completely silenced 36 and it was the miRNA with the highest level of promoter hypermethylation (Fig. 4b). This strong correlation between down-regulated miR-199b-5p expression and its high level of promoter hypermethylation in FECD suggests that miR-199b-5p directed pathways may have an important role in FECD pathogenesis.

miR-199b-5p negatively regulates Snai1 and ZEB1 expression in corneal endothelial cells.
MiRNAs can negatively regulate gene expression by directly binding to specific sequences in the 3′-UTR of target mRNAs and inducing mRNA cleavage or translation inhibition 55 . In mammals, there is a high-degree of Watson-Crick base-pairing between miRNA and target mRNA at nucleotides 2-7 at the 5′ end of miRNA, termed the "seed match" 56 . Mismatches in the miRNA-mRNA duplex were found to be ineffective in repressing gene expression 57 . To further delineate the functional role of miR-199b-5p in FECD pathogenesis, we performed in silico analysis to predict putative target genes and corresponding binding sites using two computational prediction algorithms (Targetscan and miRmap). More than one thousand targets of miR-199b-5p were predicted from these programs. Snai1 and ZEB1 were of particular interest because their overexpression leads to excessive extracellular matrix production in FECD 58 . Both prediction tools independently gave Snai1 and ZEB1 high scores (97 and 83.4 respectively). Sequence alignment analyses revealed a highly conserved miRNA-199b-5p binding motif www.nature.com/scientificreports www.nature.com/scientificreports/ in the 3′-UTR of both Snai1 and ZEB1 across many species (Fig. 5a). In particular, this predicted binding site was located in the 3′-UTR of human Snai1 (positions 725-731; NM_005985.3) and ZEB1 (positions 1023-1029; NM_001128128.2) (Fig. 5b).
To further evaluate the effect of miR-199b-5p on Snai1 and ZEB1 expression in human corneal endothelial cells, we transfected miR-199b-5p mimic, inhibitor, or negative controls into human corneal endothelial cells and measured the expression levels of Snai1 and ZEB1 by qRT-PCR. We found that the miR-199b-5p mimic significantly inhibited Snai1 and ZEB1 expression by 50%, compared to negative control group (p = 0.0002 and p = 0.002, respectively; Fig. 5d). In contrast, the miR-199b-5p inhibitor had the opposite effect and resulted in increased Snai1 and ZEB1 expression (~1.3 fold, p = 0.012 and p = 0.005, respectively; Fig. 5d). These results demonstrate that miR-199b-5p can directly bind to and negatively regulate Snai1 and ZEB1 in human corneal endothelial cells.

Discussion
FECD is the most common type of corneal endothelial dystrophy and a leading indication for corneal transplantation in patients in the United States 6,59 . We previously identified global DNA methylation changes that occur in the corneal endothelial tissue of FECD patients and specifically observed a high number of DNA methylation alterations occurring in miRNA sequences 30 . This finding was intriguing because prior reports had demonstrated that miRNAs were differentially expressed in the corneal endothelium during aging 60 , and that widespread  www.nature.com/scientificreports www.nature.com/scientificreports/ downregulation of miRNA levels occurred in the corneal endothelium of patients with late-onset FECD 36,61 . Because DNA methylation has been shown to be a mechanism for regulating miRNA expression 33 , we performed a sub-analysis of the miRNA DNA methylation array data. The most differentially methylated miRNA sequences were further validated by quantitative MethyLight assay using an additional patient cohort. MiR-199B was identified as the most extensively hypermethylated miRNA sequence in FECD and was selected for additional analysis because its expression was almost completely silenced in FECD 36 . In silico analyses identified Snai1 and ZEB1 as potential direct targets of miR-199b-5p. Using a luciferase reporter assay, we confirmed that miR-199b-5p directly targeted the 3′-UTR of both Snai1 and ZEB1 transcripts and negatively regulated their expression. Collectively, these results demonstrate that miR-199b-5p hypermethylation may contribute to late-onset FECD pathogenesis. Our findings suggest that miR-199b-5p hypermethylation leads to its down-regulated expression and consequently results in the increased expression of miR-199b-5p target genes, including Snai1 and ZEB1.
MiRNAs are small non-coding RNAs that negatively regulate gene expression by binding to specific sequences in the 3′-UTR of target mRNAs 60,62 . Such interactions may result in either translation inhibition or induction of mRNA cleavage 62 . Numerous studies have shown that miRNAs are evolutionarily conserved and are key regulators of diverse biological processes such as development, cell proliferation and differentiation, apoptosis and metabolism 63 . MiRNAs also have important regulatory roles in disease progression, including oncogenesis 64,65 . The molecular mechanisms that control miRNA expression are therefore of critical importance in better understanding normal physiologic processes and disease pathogenesis. Recently, DNA methylation has emerged as a key regulatory mechanism of miRNA expression in several different tissues and disease states [64][65][66][67] .
In this study, we have demonstrated aberrant DNA methylation of miRNA sequences in corneal endothelial tissue of FECD patients. Our array dataset included 2,227 probes associated with 463 miRNA genes, with multiple probes targeting single miRNA genes. We identified 216 probes associated with 156 miRNA genes that were differentially methylated between FECD and control samples, and the vast majority were hypermethylated in FECD. Furthermore, we found that the aberrant DNA methylation occurred almost exclusively in the promoter regions of miRNAs. Since promoter methylation and gene expression are usually inversely correlated, these results suggest DNA hypermethylation as a potential mechanism for the widespread downregulation of miRNA levels in FECD 36 . This preferential hypermethylation of miRNA gene promoters was also reported in other studies 33,[68][69][70] .
To further investigate the methylated probes in a broader genomic context, we mapped the 50 bp sequences of all 216 differentially methylation probes associated with 156 miRNA genes to the human genome. We found that approximately three-quarters of these probes were located within introns and/or exons of relevant host genes. The intragenic resident miRNAs on the same strand as their host genes can be co-transcribed by RNA polymerase II and co-regulated with their host genes 71 . A genome-scale DNA methylation analysis specifically on miRNA host genes revealed that miRNA host genes were frequent targets for aberrant DNA methylation and in particular downregulation of miR-10a was correlated with the promoter hypermethylation of its host gene HOXB4 in tumorigenesis 39 . Our data found that the miRNA host genes were differentially methylated in FECD and that their gene body regions were preferential targets of aberrant methylation. Even though gene body methylation is positively correlated with gene expression 72 , we were unable to measure the changes in mRNA levels of miRNA host genes on the same sample cohorts used in the DNA methylation analyses because of the low cellular yield. Therefore, the physiological relevance of DNA methylation changes of miRNA host genes in FECD remains to be further explored. Additionally, a subset of miRNA genes and their host genes shared hypermethylation of their individual promoters, suggesting that DNA methylation may play an important role in repressing the expression of certain miRNAs and their host genes simultaneously in FECD.  www.nature.com/scientificreports www.nature.com/scientificreports/ Using two independent DNA methylation assay technologies and two separate patient cohorts, we identified miR-199B as the most extensively hypermethylated miRNA in the FECD samples. Interestingly, miR-199b-5p has been shown to be almost completely silenced in FECD tissues 36 . We were unable to perform side-by-side comparative miRNA transcriptome analysis on the same sample cohorts used in the DNA methylation analyses because of the low cellular yield from the FECD samples. To delineate the mechanism by which miR-199b-5p may contribute to FECD pathogenesis, we used computational algorithms to search for putative target genes, and identified Snai1 and ZEB1 as having high prediction scores. Further functional analyses using a luciferase reporter assay confirmed both 3′-UTRs of Snai1 and ZEB1 transcripts as being direct targets of miR-199b-5p. Our result is consistent with the prior finding that miR-199a-5p, a close family member of miR-199b-5p, directly binds the 3′-UTR of the Snai1 mRNA and reduces Snai1 protein level via the UGUGACC motif in its seed sequence 73 . Members of the same miRNA family can have similar physiological function and share the same predicted targets because of their conserved sequence and structural configuration 74 . Our finding that the 3′-UTRs of Snai1 and ZEB1 have the same predicted target site recognized by the identical seed sequence for both miR-199a-5p and miR-199b-5p supports the miR-199 family as having an important regulatory role in Snai1 and ZEB1 expression and function.
Snai1 and ZEB1 are zinc finger transcription factors that regulate gene expression in multiple tissues, including the cornea. Okumura et al. showed that immortalized corneal endothelial cells obtained from late-onset FECD patients highly expressed Snai1 and ZEB1 had excessive production of ECM proteins, including type I collagen and fibronectin 58 . Katikireddy et al. found that Snai1 expression level is significantly upregulated in ex vivo FECD specimens as compared to control samples 75 .
A phenotypic clinical feature of FECD is the development of corneal guttae, which are abnormal collagenous excrescences of the corneal endothelial basement membrane (Descemet's membrane). Recent studies have shown that Snail and ZEB1 can also reduce cell adhesion, increase cell migratory capacity [76][77][78] , and promote apoptosis 79,80 . These phenotypic features have also been observed during FECD pathogenesis [81][82][83] .
Our findings support a model in which aberrant promoter hypermethylation of miR-199b-5p in FECD leads to the up-regulated expression of Snai1 and ZEB1 expression and consequent pathologic overproduction of ECM proteins in the cornea (Fig. 6). Dysregulated DNA methylation of miRNA promoters has been found to be a biomarker in the detection, diagnosis, and prognosis of various cancers types including breast 84 , gastrointestinal 85 , and lung 86 . Our results provide a novel mechanistic insight into the function of DNA methylation in the pathogenesis of FECD and support further studies to determine how methylation of miR-199b-5p may be used The luciferase reporter plasmids containing the wild-type 3′-UTR or mutant 3′-UTR of Snail and ZEB1 with the putative binding sites for miR-199b-5p. (c) The direct binding of miR-199b-5p to 3′-UTR of Snai1 and ZEB1 mRNAs was detected by a dual luciferase activity assay. The luciferase reporter plasmids and β-galactosidase expression plasmids were co-transfected into HEK293 or HCEnC-21T cells with either miR-199b-5p mimic, miR-199b-5p mimic negative control (Neg Ctrl), miR-199b-5p inhibitor, and miR-199b-5p inhibitor negative control (Neg Ctrl). Luciferase activity was measured by the dual luciferase reporter assay system. Relative luciferase activities were calculated by normalizing firefly luciferase activity to β-galactosidase activity in the same sample to correct for transfection efficiencies. Data are represented as the mean ± SEM (n = 3; p < 0.05).

(d)
The effect of miR-199b-5p on the expression of Snail and ZEB1. HCEnC-21T cells were transfected with miR-199b-5p mimic or miR-199b-5p inhibitor with their corresponding negative controls. 48 hours posttransfection, total RNA was extracted and mRNA expression levels of Snail and ZEB1 were quantified by qRT-PCR. Data are represented as the mean ± SEM (n = 3; p < 0.05).
www.nature.com/scientificreports www.nature.com/scientificreports/ as a clinical biomarker of phenotype expression in FECD. In the past few years, there have been major advances in testing blood, saliva, and cheek swab samples to genetically screen for corneal dystrophies 87,88 . Further studies are needed to determine if DNA methylation changes can be detected in these tissue samples that correspond to corneal methylation changes. Anterior chamber paracentesis samples can potentially be assayed in the future as another alternative. A recent study has shown that DNA methylation changes associated with bladder cancer are currently being screened for in urine samples 89 .
The epigenome is increasingly being recognized as fertile ground for drug development, as DNA methylation is dynamic and pharmacologically reversible 90 . Several drugs currently exist that target and inhibit DNA methylation 90 . Of these, the cytidine analogues 5-azacytidine (5-Aza-CR) and 5-aza-2′-deoxycytidine (5-Aza-CdR) are the two most potent DNMT inhibitors and have been approved by the Food and Drug Administration (FDA) in the USA for the treatment of myeloid malignancies and other solid tumors [91][92][93] . A large number (>30) of different, epigenome-targeting drugs are currently in clinical trials 94 Figure 6. MiR-199B is hypermethylated in CpG island at its promoter region and its mature transcript miR-199b-5p directly inhibits the expression of Snai1 and ZEB1 in FECD. MiR-199B is the most extensively hypermethylated miRNA gene in FECD. Its mature transcript miR-199b-5p functions as a direct negative regulator of two zinc finger transcription factors, Snai1 and ZEB1, which have been shown to lead to increased production of extracellular matrix proteins in FECD.
www.nature.com/scientificreports www.nature.com/scientificreports/ DnA methylation microarray. Array-based DNA methylation data was collected in our prior studies 30 .
The IDAT files are available on the GEO DataSets database [accession number GSE94462; National Center for Biotechnology Information (NCBI), Bethesda, MD, USA].
Dual luciferase reporter assay. Dual luciferase reporter assays were performed to validate Snail1 and ZEB1 as bona fide miRNA target genes. Briefly, 0.3 × 10 6 of HCEnC-21T cells 97 or HEK293 cells were seeded in 24-well plates and then co-transfected with 500 ng of pMIR-REPORT wild-type or mutant plasmid, 100 ng of β-gal plasmid, and 25 nmol miR-199b-5p mimic, 25 nmol scrambled mimic negative control, 50 nmol miR-199b-5p inhibitor, or 50 nmol scrambled inhibitor negative control (Life Technologies, Carlsbad, CA), using Lipofectamine 3000 (Life Technologies, Carlsbad, CA) in OptiMEM (Gibco, CA). A β-galactosidase expression plasmid was used as an internal control for transfection efficiency. Forty-eight hours after transfection, cells were subjected to lysis and firefly luciferase and β-galactosidase enzymatic activities were measured consecutively using a dual-luciferase reporter assay system (Applied Biosystems, Bedford, MA) as per the manufacturer's instructions. Relative firefly luciferase activity (firefly luciferase activity/β-galactosidase enzymatic activities) were expressed as changes relative to that value of the negative control, which was set as 1. Three independent experiments were performed in triplicate.
www.nature.com/scientificreports www.nature.com/scientificreports/ Data are presented as fold change in gene expression normalized to GAPDH. Relative quantification of expression was calculated with the 2 −ΔΔCt method 98 .
Statistical analyses. Statistical analyses on genome-wide methylation data of miRNA genes were executed in R 30 . All other analyses were performed using two-tailed t-test or alternative non-parametric Wilcoxon test to compare mean values using R software. Differences with p-values less than 0.05 were considered statistically significant and the corresponding p values were indicated.