Retinal expression of small non-coding RNAs in a murine model of proliferative retinopathy

Ocular neovascularization is a leading cause of blindness in proliferative retinopathy. Small non-coding RNAs (sncRNAs) play critical roles in both vascular and neuronal development of the retina through post-transcriptional regulation of target gene expression. To identify the function and therapeutic potential of sncRNAs in retinopathy, we assessed the expression profile of retinal sncRNAs in a mouse model of oxygen-induced retinopathy (OIR) with pathologic proliferation of neovessels. Approximately 2% of all analyzed sncRNAs were significantly altered in OIR retinas compared with normoxic controls. Twenty three microRNAs with substantial up- or down-regulation were identified, including miR-351, -762, -210, 145, -155, -129-5p, -150, -203, and -375, which were further analyzed for their potential target genes in angiogenic, hypoxic, and immune response-related pathways. In addition, nineteen small nucleolar RNAs also revealed differential expression in OIR retinas compared with control retinas. A decrease of overall microRNA expression in OIR retinas was consistent with reduced microRNA processing enzyme Dicer, and increased expression of Alu element in OIR. Together, our findings elucidated a group of differentially expressed sncRNAs in a murine model of proliferative retinopathy. These sncRNAs may exert critical post-transcriptional regulatory roles in regulating pathological neovascularization in eye diseases.

Scientific RepoRts | 6:33947 | DOI: 10.1038/srep33947 are the most extensively studied sncRNAs; these are highly conserved molecules of ∼ 21-25 nucleotides that function mainly in mRNA silencing capacity. Post-transcriptional regulation of gene expression by miRNAs plays a key role in ocular tissue development and in angiogenesis 12,13 , suggesting a potential role of miRNAs in proliferative retinal neovascularization. SnoRNAs are another class of newly discovered yet less studied sncRNAs varying from 60 to 300 nucleotides in length. SnoRNAs play important roles in post-transcriptional RNA modification and may impact oncogenesis as well as pathogenesis of some hereditary and autoimmune diseases, such as Prader-Willi syndrome and multiple sclerosis 14,15 , yet their roles in retinopathy is not clear.
In this study, we used a mouse model of OIR to identify sncRNAs responsive to perturbations associated with pathologic retinal angiogenesis. Analysis in the mouse model is important as there are few human sncRNA retinal expression data available from patients with proliferative retinopathy. SncRNAs significantly regulated in this murine model may be potential mediators of hypoxia-driven regulation of retinal neovascularization. Our data revealed twenty three miRNAs and nineteen snoRNAs with significantly altered expression levels in OIR mouse retinas as compared with those from normoxic control mice. Pathway analyses identified hypoxia and angiogenesis related genes as putative miRNA targets in OIR retinas, shown in a graphic network of miRNA-gene interaction. In addition, our analysis in the transcriptomic profiles characterizing pathologic angiogenesis in the human vascular retina and choriocapillaris also strengthened inferences on the influence of miRNAs as key regulators in human vascular eye diseases. Moreover, we observed decreased overall miRNA expression in OIR retinas consistent with reduced miRNA processing enzyme Dicer, and increased expression of Alu element. Overall, this study provides a detailed analysis of sncRNA expression in a murine model of proliferative retinopathy. Further investigation of these candidates may help elucidate the post-transcriptional regulatory events leading to pathologic neovascularization in proliferative retinopathy.

Results
Microarray analysis of miRNA expression profile in mouse OIR model. The mouse OIR model exposes mouse pups to 75% oxygen from postnatal day (P) 7 to 12. After returning to room air at P12, the relative hypoxia stimulates retinal ischemia and pathologic neovascularization. Maximal pathologic retinal angiogenesis can be observed at P17, mimicking pathologic retinal vessels as seen in ROP and DR (Fig. 1a) 10 . To identify the difference in miRNA expression profiles, we performed microarray analysis of miRNAs using total RNA isolated from P17 mouse OIR retinas and age-matched normoxic controls. The array chip contains probes for 1,412 mouse mature and precursor (stem-loop) miRNAs, all of which were included in the significance analysis. MiRNAs with significantly altered levels (P < 0.05) in OIR retinas compared with controls were identified. Among all miRNAs analyzed, approximately 1% of them (13 miRNAs) were up-regulated in OIR retinas with fold change greater than 1.25-fold, and 0.71% of them (10 miRNAs) were down-regulated at less than 0.75-fold (Fig. 1b).
Expression profile of snoRNAs was changed in OIR retinas. In addition to miRNAs, we also analyzed in OIR retinas the differential expression profiles of the snoRNAs, another class of small non-coding RNAs associated with nucleotide chemical modification, such as methylation and pseudouridylation, by guiding and tethering partner enzymes to specific sites on RNA targets [16][17][18] . The array chip has 2,302 probes for snoRNA and 19 probes in the array revealed significant altered expression in P17 OIR retinas compared with normoxic controls ( Fig. 3a and Table 2). Fourteen of them were up-regulated and five were down-regulated in OIR retinas. Based on the nucleotide sequence characteristics, ten of these altered snoRNAs were in the C/D box family (Fig. 3b), with two conserved motifs, C (consensus RUGAUGA) and D (CUGA), at the 5′ and 3′ ends respectively. Two snoRNAs were recognized as H/ACA box snoRNAs with conserved motifs H box (ANANNA) and the ACA box (ACA). C/D box and H/ACA box snoRNAs guide 2′ -O-methylation and pseudouridylation, respectively 16,17 . The rest seven were unclassified snoRNAs (Fig. 3c). Among the 19 snoRNAs, seven of them have reported target RNAs, whereas twelve snoRNAs have no known targets and hence are orphan snoRNAs (Table 2). Although these snoRNA probes are of human sequence, mouse orthologs of them largely exist and may harbor unknown function to be explored.
We found that the most up-regulated snoRNA is 14q(II-10), at ∼ 1.4-fold change (log 2 ratio was ∼ 0.5) in OIR retinas compared with normoxic controls (Fig. 3a,b and Table 2). 14q(II-10) (SNORD114-10) is a C/D box snoRNA, whose human form is located on chromosome 14q32. The mouse SNORD114-10 ortholog gene, Gm25224, shares 68% of identity and 100% of coverage with the human form, and is encoded in homologous loci on mouse chromosome 12. Another C/D box snoRNA, HBII-296B (SNORD91B), revealed up to 1.27-fold change and 90% of identity to its mouse ortholog Gm22771. In addition, U19 (SNORA74), the most substantially altered H/ACA snoRNA in the array, showed 20% increase in OIR retinas compared to normoxic retinas; moreover, its mouse ortholog, Gm23946, shares 78% of identity and 97% of coverage. Our findings indicated that these altered snoRNAs may exert post-translational co-regulatory roles in mediating OIR response, potentially through altering ribosome biogenesis pathways 16,17 . MiRNA biogenesis was decreased in OIR retinas. From our analysis of OIR miRNA expression profile, we observed global miRNA down-regulation in OIR retinas (Fig. 4a). For all mouse miRNA probes, including mature and precursor miRNAs (pre-miRNAs), more than half (∼ 56%) of probes showed decreased expression levels in OIR retinas compared with the average fold change of normoxic controls. Among all significantly changed mature miRNAs, 34.6% were up-regulated and 65.4% were down-regulated, suggesting that miRNA processing pathway might be affected in OIR.
To determine whether the miRNA biogenesis process (Fig. 4b) is affected in OIR, we next analyzed the expression of pre-miRNAs and mature-miRNAs from the significantly altered miRNAs ( Table 1). The results showed remarkable down-regulation of mature RNAs, compared with their respective pre-miRNAs in OIR (Fig. 4c). The decrease of overall miRNA expression and particularly mature miRNAs might be caused by altered levels of key RNA processing enzymes, such as Drosha and Dicer (Fig. 4b), both ribonuclease III responsible for processing primary miRNAs (pri-miRNAs) and cleaving pre-miRNAs respectively. We thereby examined the expression of Drosha and Dicer1 mRNAs in OIR and normoxic retinas using q-PCR. Drosha was slightly increased in OIR retinas; whereas expression of Dicer1 was significant decreased (∼ 50%) in the OIR retinas (Fig. 4d), indicating that insufficient Dicer expression may explain in part the decreased mature miRNA expression levels.
Previous studies in human age-related macular degeneration indicated that DICER deficiency caused accumulation of Alu RNA, leading to retinal cell death 19 . Alu elements, derived from the small cytoplasmic 7SL RNA, are also post-transcriptional regulators of gene expression through alternative splicing and RNA editing 20 . Since we observed the decrease of Dicer1 in OIR, we next evaluated whether the expression levels of mouse Alu-equivalent (a) Schematic diagram of the OIR model. Neonatal mice were exposed to 75% oxygen from postnatal day (P) 7 to P12 and returned to room air from P12 to P17 to induce maximum pathologic neovascularization at P17. Total RNAs were isolated from OIR-exposed or age-matched normoxic control mouse retinas at P17 (n = 3 per group) and subsequently applied to Affymetrix GeneChip miRNA array chips for analyzing small non-coding RNAs (sncRNAs), including mature miRNAs, pre-miRNAs (stem-loops), and snoRNAs. There were 1,412 mouse (Mus musculus) miRNAs on the array chip. (b) A dot plot illustrates the mean fold changes of miRNA expression in OIR compared with normoxic retinas versus their P-value. Each dot represents one miRNA (including pre-miRNA). Among all analyzed miRNAs, 0.92% of them (13 miRNAs, as red dots) were upregulated at more than 1.25-fold, and 0.71% (10 miRNAs, as green dots) were down-regulated in OIR retinas at less than 0.75-fold.
Scientific RepoRts | 6:33947 | DOI: 10.1038/srep33947  B1 and B2 elements were changed in OIR. We found that both B1 and B2 RNAs were significantly increased in OIR retinas ( Fig. 4e), suggesting the possibility of coordinated regulation of Alu RNA and their potential roles in proliferative retinopathy.
In Silico miRNA-to-target networks in regulating pathologic neovascularization in OIR. Having identified multiple miRNA alteration in OIR, we next evaluated their potential mRNA targets that may impact neovascularization. Specifically we composed a molecular network spanning direct miRNA-mRNA targets in OIR by using bioinformatics databases TargetScan and miRanada-mirSVR, following by singular enrichment analysis performed by GeneCodis3 program [21][22][23] . A total of 7,998 genes, which might be targeted by those miR-NAs with significant differential expression in OIR (identified in Table 1), were found with gene ontology (GO) annotation in biological process categories. Over 90% (7,289) of the putative target genes were classified in the top 25 GO categories (Fig. 5a). Most of the targets were shown related to transcription regulation, cellular component transport, cell cycle and development. Moreover, the enrichment analysis showed that 137 target genes were annotated in angiogenic-and hypoxic-related GO categories. Among them, 105 genes were related to angiogenesis, including both positive and negative regulation of angiogenesis; and 41 genes were annotated with response to hypoxia. After combining the prediction of miRNA-gene interaction with the functional GO analysis for the differentially expressed miRNAs in a graphic form, we obtained a visual illustration of the miRNA-target gene network associated with angiogenesis and hypoxia in the OIR (Fig. 5b). This integration illustrated a complex cross-interacting network of identified miRNAs and their predicted targets which may participate in angiogenesis and be involved in pathogenesis of proliferative retinopathy. Some of the putative target genes may be regulated by several miRNAs, and involved in multiple GO categories.
Identification of OIR-responsive miRNA targets in transcriptome profile of human neovascular eye diseases. Our findings accentuated the potential regulatory roles of miRNAs in retinopathy disease progression, suggesting that miRNAs would be critical regulators of transcriptomic profiles and biomarkers linked to the human vascular eye diseases. In order to identify such promising miRNA-based biomarkers, we used information on the targets of the OIR-responsive miRNAs (Table 1) supported by strong experimental evidence in miRTarBase 6.1 to filter disease-defining constituents of transcriptomic profiles from human retinal specimens containing: 1) active and quiescent fibrovascular membranes (FVMs) of people with proliferative diabetic retinopathy (PDR) 24 ; and, 2) choroidal neovascular (CNV) membranes of people with neovascular age-related macular degeneration (AMD) 25 (Table 3). Fourteen of the OIR-responsive miRNAs yielded findings in the analyses. Eight of them showed validated gene targets differentially expressed in retinas of people with PDR, whereas six OIR-responsive miRNAs had targets in active FVMs. In patient samples with neovascular AMD, six of the OIR-responsive miRNAs had experimentally validated targets in genes differentially expressed in neural retinas, and six of the altered miRNAs in OIR showed gene targets in RPE/choroid complexes. After probing potential targets of OIR-responsive miRNAs in the context of comparative neovascular AMD and PDR transcriptome analyses, we investigated functional enrichment of the 38 pertinent targets identified both as validated OIR-responsive miRNA targets and dysregulated molecules in PDR and neovascular AMD by using GeneCodis3. The strongest enrichment from the GO biological processes database existed for transcriptional regulation, signal transduction, development, and blood coagulation, with 7 of the 38 pertinent constructs defining these processes (Table S1). Focal adhesion and p53 signaling showed strong enrichment in human KEGG pathways from the GeneCodis3 database, with 5 targets of OIR-responsive miRNAs differentially expressed in human retinal specimens with PDR, and 4 targets for neovascular AMD (Table S2). Together our analysis indicated that miRNAs may define gene expression signature in human retinal tissues manifesting microvascular pathophysiology, offering promise for development of global biomarkers and/or prognostic indicators.

Discussion
MiRNAs play crucial roles in the biological development and cellular function, such as neurogenesis, angiogenesis, inflammation, and metabolism [26][27][28][29] . Dysregulation of miRNA impacts many diseases, including cancers, neurological and cardiovascular diseases; thereby miRNAs are being developed as important biomarkers for these diseases 28,30,31 . In this study, we demonstrated differential expression profiles of miRNA in an oxygen-induced proliferative retinopathy model by microarray analyses (Fig. 1). Our data identified numerous miRNAs that were significantly altered in mouse retinas with induced proliferative retinopathy (Table 1 and Fig. 2), suggesting that these miRNAs may be potential regulators of abnormal pathologic vascular formation in proliferative retinopathy. Furthermore, our data revealed a global miRNA down-regulation in the mouse OIR retinas likely caused by decreased expression of Dicer1 (Fig. 4a-d). Previous studies showed that retinal Dicer-knockout mice exhibit global miRNA down-regulation which compromises the normal retinal development [32][33][34] , indicating that our observation of Dicer deficiency may partially explain impaired vascular and neuronal development in OIR retinas. Consistent with our findings, trends of decreased overall retinal miRNA expression can be derived from previous works utilizing different rodent models of oxygen-induced ischemic retinopathy associated with retinal hypoxia 35,36 . Hypoxia is a key regulator in altering miRNA expression profiles in human cancers via decreasing miRNA biogenesis and suppression of Dicer through an epigenetic regulation mechanism 37 . Together these findings suggest that miRNA biogenesis is substantially dysregulated in retinopathy, likely modified by retinal hypoxia and resultant Dicer suppression. In addition to processing miRNAs, DICER1 also processes degradation of Alu RNA element. Alu RNA accumulation and DICER1 deficit stimulates retinal pigmented epithelium degeneration and geography atrophy in age-related macular degeneration 19 . In diabetic patients and murine diabetic models, Dicer exhibited reduced expression and Alu RNA was up-regulated 38 , implicating that disrupted miRNA biogenesis and Alu toxicity may be associated with the development of diabetic retinopathy. Apart from the OIR-resulted global changes of miRNA levels, we found that murine Alu-like B1/B2 elements were up-regulated in the OIR retinas associated with decreased Dicer1 (Fig. 4e), indicating a potential role of Dicer-deficiency-induced Alu element accumulation in retinal neovascularization and degeneration in proliferative retinopathy.
Using bioinformatics tools, we analyzed putative mRNA targets of miRNAs identified in OIR retinas. Our analyses showed that these significant altered miRNAs may target multiple genes associated with several angiogenic-and hypoxic-related GO categories and angiogenic signaling pathways, including VEGF, ANGPT-TIE, and HIF pathways (Fig. 5). In addition, among the differentially expressed miRNAs, miR-145, -155, -210, -375 and -449a are known as hypoxia-regulated miRNAs (hypoxamiRs), implicated in regulating several hypoxia-sensitive cellular responses by targeting genes involved in angiogenesis, cell cycle, and inflammation [39][40][41] . In line with our findings, up-regulation of miR-210, a master hypoxamiR regulated by HIF, can target ephrin-A3 (Efna3) to promote angiogenesis during hypoxia 39 . Similarly, increased expression of another HIF-dependent miRNA-miR-155-in mouse OIR model can target CCN1, a cysteine-rich and integrin-binding matricellular  The significantly altered miRNAs in OIR were subjected to target prediction by TargetScan and miRanad-mirSVR, with GO annotation and singular functional enrichment analysis using GeneCodis3. (a) Top 25 GO terms included 91% of the putative target genes. The vertical axis represents the GO categories; the horizontal axis indicates the number of genes per GO. (b) Summary of miRNA-target gene network shows the miRNAs and their potential target genes which may response to hypoxia and direct angiogenesis in OIR. Up-(red) or down-regulated (green) miRNAs in OIR were subjected to GO annotation and singular functional enrichment analysis using TargetScan, miRanad-mirSVR and GeneCodis3 and plotted with their potential targets using Cytoscape (version 3.2.1). Angiogenesisrelated genes, such as those relevant to VEGF (Vegfa, Flt1, and Kdr) and ANGPT-TIE (Angpt1, Angpt2 and Tek) signaling, and genes in hypoxia response (Hif1a and Ep300), may be regulated by miRNAs altered in OIR to impact pathologic retinal neovascularization.
Scientific RepoRts | 6:33947 | DOI: 10.1038/srep33947 protein, to disturb the normal retinal vessel growth 35 . Consistently, our in silico analysis also showed that miR-155 was up-regulated in OIR retinas and may target several genes associated with angiogenesis, inflammation and metabolism in response to hypoxia. MiR-145 was reported to regulate differentiation and cytoskeletal dynamics of smooth muscle cells during vascular diseases 42 , and control of apoptosis in tumors 43,44 . In experimental models of cerebral ischemia, miR-145 showed significant increase 45,46 , suggesting that hypoxia may play roles in promoting miR-145 expression.
The present study focused on the second proliferative phase of OIR with neovascularization, with retinal samples collected at P17, the time point showing maximal pathologic neovessels 47 . Our miRNA findings at P17 were compared with a previous study in OIR at P15 48 in order to evaluate the dynamics of differential miRNA expression (Table S3). This comparison demonstrated different groups of miRNAs preferentially altered at different time points in OIR, yet several miRNAs, including miR-150, miR-375, miR-129-5p and miR-129-3p showed consistent pattern of down-regulation at both P15 and P17. MiR-150 is selectively expressed in mature B and T cells as well as abundant in vascular endothelial cells 49,50 . Besides, in mouse pups undergone prolong hyperoxic condition, miR-150 showed down-regulated in the lungs [51][52][53] . Our previous study showed that endothelial miR-150 targets angiogenic genes, such as Fzd4 and Dll4 as validated by 3′ UTR luciferase assays, to exert an intrinsic suppressive role in pathologic retinal angiogenesis 50 . In addition, miR-375 was observed down-regulated in the murine retinal ischemic model 35,48 . Mir-375 is required for normal pancreatic development and also influences glucose-stimulated insulin secretion 54,55 , indicting its crucial roles in diabetes. Our data of decreased miR-375 in OIR retinas suggest a potential additional role of miR-375 in proliferative retinopathy.
The retina is a complex tissue consisting of several cell types, including neurons, glia, and endothelial cells. Different type of retinal cells may have specific sncRNA expression profiles. For instance, we have identified miR-150 as an endothelial-enriched miRNA in retinal vessels and its down-regulation leads to pathologic neovascularization 50 . Moreover, a recent study characterizing the miRNome of the human retina identified retinal enrichment of three miRNAs hsa-miR-125a-5p, -145-5p, and -486-5p 56 , which are the human homologs of mouse miR-351, -145, and -486 respectively, all significantly altered miRNAs in our study. Among them, hsa-miR-125a-5p and -145-5p revealed association with DR and AMD according to our analysis (Table 3), strengthening their retinal-enriched features. In addition, Arora et al. 56 reported that mmu-miR-125, which contain identical seed sequence region as mmu-miR-351, was strongly expressed in the inner plexiform layer in developing mouse retina, and in the inner and outer nuclear layers in adult retinas, suggesting that mmu-miR-351 might be expressed in retinal neurons, including photoreceptors. Investigation of Müller glia and photoreceptors in developing mouse retinas also highlighted miR-145 as one of Müller glia-enriched miRNAs 57 .  By examining the significantly altered miRNAs of OIR retinas, we have identified 38 genes with the capacity to act as targets of the OIR-responsive miRNAs in a perturbation that leads to pathologic angiogenesis in human specimens from neovascular eye diseases (Table 3). Our findings are aligned with miRNA target enrichment in GO biologic processes of transcriptional regulation and development (Tables S1 and S2). Progression from the animal model of OIR-responsive miRNAs, to the experimentally confirmed targets of the homologous human miRNAs with differential expression profile in retinal and choroidal vascular specimens, has given us a basis to infer that human homologs of these OIR-responsive miRNAs are likely to act in pathologic retinal angiogenesis in clinical populations.
Besides miRNAs, we also observed differential expression of nineteen snoRNAs in OIR mouse retinas ( Fig. 3 and Table 2). To date, most snoRNAs have no verified targets, yet some snoRNAs display up-regulation in cancers and are involved in tumorigenicity 58 . On the other hand, loss of snoRNA could also be pathogenic 14,15 . For instance, Prader-Willi syndrome, a severe hereditary disorder, is caused by deletion of the chromosome 15q11-q13 containing C/D box snoRNA SNORD115 and SNORD116 14 . Furthermore, snoRNAs may also function as miRNA precursors, and are processed to miRNA-liked fragments to regulate gene expression 59 . In our study, while we found a globally up-regulated snoRNA expressing pattern in OIR mouse retinas, the exact role of each altered snoRNA in retinopathy still awaits further investigation.
In conclusion, our study elucidates two groups of sncRNAs-miRNAs and snoRNAs-whose expression pattern is altered in the mouse model of proliferative retinopathy. We present global miRNA reduction along with Dicer1 decrease and Alu RNA increase, all of which may be potentially attributable to retinal hypoxia and ischemia in OIR. MiRNAs are key mediators of post-transcriptional regulation for fine-tuning gene expression; whereas snoRNAs are crucial in regulating protein synthesis and cell behavior, dysregulation of either or both groups of sncRNAs may exacerbate proliferative retinopathy. Our findings provide practical guiding information for deciphering the specific contribution of sncRNA processing pathway and each identified sncRNA in regulating pathological ocular angiogenesis, and potential RNA-based biomarkers and future treatments for eye diseases.

Methods
Animals. C57BL/6J mice were obtained from The Jackson Laboratory (stock no. 000064; Bar Harbor, ME, USA).
All animal experimental procedures were approved by the Boston Children's Hospital Institutional Animal Care and Use Committee and conducted in accordance with the Association for Research in Vision and Ophthalmology Statement for the Use of Animals in Ophthalmic and Vision Research.
Mouse pups and their nursing mother were exposed to 75% oxygen from postnatal day (P) 7 to P12, and returned to room air. Age-matched normoxic control mouse pups were kept in room air throughout the experiments. Mice were sacrificed at P16 or P17, followed by retinal dissection and RNA extraction.
Small RNA isolation and miRNA microarray. Total retinal RNA including miRNAs and other small RNAs were isolated from P17 mouse OIR and normoxic retinas with miRNeasy Micro Kit (Qiagen, Chatsworth, MD, USA) according to the manufacturer's instruction (n = 3 per group). 100 ng total RNA per sample was applied to the Affymetrix GeneChip miRNA 2.0 array (Santa Clara, CA, USA), which covers 100% miRBase v15 contents, in order to examine sncRNA expression profiles in OIR retinas and normoxic controls. Each sample is a biological replicate (n = 3 per group). The array study, from hybridization, RNA labeling, to raw data normalization, was performed by the Harvard Medical School Biopolymers Facility. For the array, the mouse miRNA probe set encompassed 722 mature miRNAs and 690 pre-miRNAs. It also contained 2,302 snoRNAs from human. The expression data were analyzed with the Affymetrix miRNA QCTool software for quality control and background analysis. Pre-miRNA, mature miRNA and snoRNA probes were identified according to annotations provided by Affymetrix. A t-test was applied on the mean expression values of sncRNAs for significance calculations. The resulting significant genes were grouped. Heat maps demonstrating differential sncRNA expression were generated by using the Heatmap Builder software as previously described 60 . The array data were deposited in the NCBI Gene Expression Omnibus (GEO) database (http://www.ncbi.nlm.nih.gov/geo/) for public access (accession number: GSE84303).
Identification of putative targets of OIR-responsive miRNAs. Potential target genes of selective miR-NAs were predicted according to the algorithms of TargetScan v.7 (http://www.targetscan.org) 21 and microRNA. org (miRanada-mirSVR) (http://www.microrna.org/microrna/home.do) 22 with incorporation of context score + + < − 0.2 and mirSVR score < − 0.5, respectively. Functional enrichment analysis was conducted using GeneCodis3 23 with ontology class "Biological Process" for gene annotation. Hits of GO terms were considered according to corrected hypergeometric distributions P-values which were less than 0.05. Based on GO categories, miRNA-target gene regulatory networks were illustrated with direct interaction using Cytoscape (version 3.2.1) 61 .
Identification of OIR-responsive miRNA targets in human neovascular eye disease transcriptome profile. For identifying validated targets of selective miRNAs in human eye diseases, we used miR-TarBase 6.1 (mirtarbase.mbc.nctu.edu.tw) to obtain a set of experimentally validated human homologous miRNA targets of the mouse OIR-responsive miRNAs. TargetScan v.7 Human was used as the source for predicted miRNA-transcript interactions, applying a total context+ + score value < − 0.2 as a threshold for strongly predicted sites. Transcriptome analysis were performed in microarray data from human specimen with fibrovascular membranes from proliferative diabetic retinopathy patients 24 (data obtained from GEO, GSE60436); and choroidal neovascular membranes of patients with neovascular age-related macular degeneration (AMD) 25 (GEO:GSE29801).