We decided to define the complete repertoire of miRNAs that are expressed in any of four different eye tissues, i.e. retina, lens, cornea and RPE. As a first step, we collected retinas, lenses, corneas, RPE tissues and entire eyes from 8-week old C57BL/6 mice. We chose to use the C57BL/6 strain that has a pigmented RPE in order to limit the carry-over contamination of the retina samples with RPE cells during tissue dissection. Following total RNA extraction from these eye tissues, we carried out miRNA profiling using the miRCURY LNA™arrays provided by Exiqon (see Methods). Total RNA from entire eyes was included as a reference to determine differential expression. The miRNAs that showed, in one or more tissues, a difference in Log2 median ratios (ΔLMR) equal or superior to 0.5 (n = 285 out of 597 analyzed) compared to the 'entire eye' reference were considered for further analysis.

To demonstrate whether miRNAs show tissue-enriched expression in a specific eye compartment, the 285 miRNAs that passed the above threshold were used in a two-way hierarchical clustering analysis. This analysis revealed distinct miRNA clusters that are differentially expressed in one or more of the tissues analyzed (Additional file 1). The resulting dendrogram could be divided in five main clusters (A-E) (Figure 1A-E; Green brackets in Additional file 1) enriched in miRNAs predominantly expressed in retina (Cluster A; Figure 1A), RPE (Cluster B; Figure 1B), retina and lens (Cluster C; Figure 1C), lens and cornea (Cluster D; Figure 1D), and cornea and RPE (Cluster E; Figure 1E). Within these five clusters, we defined 25 smaller groups (sub-clusters) of miRNAs with highly similar expression profiles (Blue lines in Figure 1A-E and Additional file 1). The reliability of the clustering analysis of microarray data was confirmed by assessing the overall relationships among the analyzed tissues, which were in line with their developmental origin (Additional file 1B). The significant number of miRNAs that show differential expression profiles among the ocular tissues and their organization in distinct clusters suggest that the regulation of eye development and function by miRNAs requires an intricate biological network. Our microarray results are overall concordant with similar global expression studies on single eye tissues previously reported in mouse or other vertebrates (reviewed in 36 ). However, this is the first example of a simultaneous analysis of the main ocular structures, which therefore allows a direct semi-quantitative assessment of the differential miRNA expression levels in the major components of the mouse eye.

To identify whether groups of miRNAs with comparable expression profiles may be involved in the regulation of specific biological functions, we performed an in silico functional annotation analysis of their predicted target genes using the 25 above- described co-expression sub-clusters (Blue lines in Figure 1A-E and Additional file 1). For each miRNA present within these sub-clusters, we compiled a non-redundant list of predicted targets combining the output of miRanda, TargetScan and Pictar 37 38 39 , three commonly used miRNA target prediction software. Subsequently, we performed a Gene Ontology (restricted to Biological Processes) and KEGG pathway annotation analysis for the target lists of each single miRNA, as described in Methods. We then compared the annotation term lists of all predicted target genes for each miRNA belonging to each sub-cluster and counted the occurrence of shared individual GO/KEGG terms (Additional file 2). An arbitrary cut-off value of 70% was set and only terms enriched in the predicted target gene lists of at least 70% of the miRNAs within a cluster were retained (see Additional file 3). As a first result, we found a significant enrichment in several GO/KEGG terms related to neuronal development and function (Additional files 2 and 3), which, considering the neural origin of the eye, is a good indication of the reliability of our analysis. The above-mentioned enrichment in neuron-related terms was more prominent in sub-clusters that contained miRNAs strongly expressed in retina (Additional files 2 and 3). Another interesting example was sub-cluster 24, strongly expressed in cornea and RPE, which showed enrichment for the GO term 'epithelium development' (Additional files 2 and 3). Overall, we found enrichment for specific processes, consistent with the physiology and function of the ocular tissues in which the analyzed miRNAs were predominantly expressed. This suggests that the above-described cluster analysis of miRNA expression data, although requiring further experimental validation, can be used to gain insight into the specific functional processes regulated by miRNA-target networks within the eye.

To study the spatiotemporal distribution of the above-mentioned miRNAs at the cellular level, we performed RNA ISH profiling on sections of murine eyes at different developmental stages. To this purpose, we carried out high-throughput robotic RNA ISH on serial sections of whole heads and eyes (at postnatal stages) using LNA-modified DNA oligonucleotide probes (Figure 2). The ISH analysis was performed in the albino CD1 wild-type mouse strain in order to enable signal detection in the non-pigmented RPE. The selection of miRNAs to analyze mostly relied on the evaluation of microarray profiling data described above. We prioritized for the analysis 212 miRNAs that displayed the highest levels of differential expression in the adult eye by microarray. Moreover, we selected 36 miRNAs that were not detected in a reliable manner by microarray. Given the different resolution and sensitivity of the two technologies, we decided to analyze also the latter ones to verify the ability of RNA ISH to detect miRNAs expressed at very low or single-cell levels. Indeed, the ISH protocol we used includes a signal amplification step which can reveal low abundance transcripts that would have been otherwise undetectable 40 . As a final result, taking also into account technical failures and probe availability, we obtained reliable ISH data for 221 miRNAs (Additional file 4).

For this study, we looked at time points that correspond to milestones of development and differentiation of the major eye structures. In particular, we analyzed miRNA expression at a prenatal stage (E16.5), when the major structures of the eye are already formed, although not fully differentiated; at the date of birth, postnatal day (P) 0, when the newborn mouse eye is still developmentally immature; at P8, when the majority of retinal cells have exited the cell-cycle and occupy their final positions within the retina and, finally, at the P60 stage that corresponds to full adulthood and maturation of the eye is complete. The expression patterns obtained were annotated based on the overall signal distribution (Figure 2) and on the anatomical structures and/or cell-types that stained positive for miRNA expression. When a uniform signal was visible across the entire section, the pattern was defined 'ubiquitous' (ubi) (e.g. miR-127 in P0 head and miR-770-3p in P60 eye; Figure 2). Whenever the signal was restricted to specific cells or structures the expression pattern was defined 'regional' (reg) (e.g. miR-200b staining of the olfactory epithelium at P0 and miR-96 specific staining of the adult retina; Figure 2), while in the absence of a visible signal, the attribute 'not detected' (ND) was assigned (e.g. miR-122 in P60 eye; Figure 2). Finally, an additional category, 'ubiquitous with pattern' (uwp), was defined for head sections at E16.5 and P0 in cases where a signal was detected in most cells with some structures showing stronger staining (e.g. miR-188-5p in P0 head; Figure 2).

Overall, we identified a range of different spatially and/or temporally restricted expression profiles for 178 miRNAs, while 43 did not yield any signal in the eye at any developmental stage. We found a number of miRNAs (n = 30) that despite displaying significant differential expression by microarray, were not detected by RNA ISH. This could be explained either by differences in sensitivity between the two approaches or by possible miRNA expression level differences across mouse strains. On the other hand, by RNA ISH we obtained a clear regional hybridization signal for 14 miRNAs that were not detected in a reliable manner by microarray profiling (e.g. miR-188-5p, miR-296-5p, miR-680, miR-681; see database). These observations highlight the advantage of employing diverse approaches in determining miRNA expression profiles. A summary of the ISH signal distributions observed at each developmental stage is shown in Table 1. A detailed annotation of the expression patterns at all time points is listed in Additional file 4. Although broadly in agreement with previously published work, in some cases the ISH images revealed differences in miRNA distribution (e.g. miR-182). We believe that these discrepancies could be attributed to differences in the experimental conditions and, most importantly, to the improved sensitivity in the detection of low-abundance transcripts offered by the signal amplification step of the automated approach.

The ISH data from the adult eye could be directly compared to the microarray profiling performed at the same stage (see above and Table 2). A summary of miRNA distribution in the different anatomical structures of the adult eye as well as in the retinal cell layers is shown in Figure 3. Overall, the comparison between the RNA ISH annotation and the microarray data revealed a high degree of consistency and the combination of the two datasets largely contributes to the definition of the tissue distribution of miRNAs in the eye.

To make these data readily accessible we constructed miRNeye, a web-searchable atlas of miRNA expression patterns in the mouse eye that can be accessed at http://mirneye.tigem.it. The atlas offers high-resolution images of the cellular distribution of miRNAs in the mammalian eye and represents a useful tool in the study of miRNA contribution in eye formation, maintenance and function. Combined with target prediction software, knowledge of the miRNA cellular distribution, as well as of the miRNA clusters of co-expression, can greatly aid in unraveling their role in the eye. Below we provide a more detailed description of the spatiotemporal miRNA expression profiles we detected in each of the analyzed murine eye tissues.

We found that the majority of miRNAs showing, by RNA ISH, regional distribution in the adult eye (89 out of 106) were detected in the retina, in one or more of its cell layers. Among them, we both found miRNAs with previously reported retinal expression domains e.g. miR-124a, miR-29c, miR-9, miR-182, miR-181a 32 , among others, but we also identified at least 70 miRNAs, which, to our knowledge, were not previously described to have a regional expression in the retina in any vertebrate species (Figures 3 and 4).

Most of the miRNA expression patterns in the retina had a developmental onset. At E16.5, retina-expressed miRNAs stained mainly the inner neuroblastic layer (INBL) where newborn ganglion cell neurons are located (see Additional file 4 and miRNeye database). Likewise, in retinas of newborn mice (P0), miRNA expression was mainly detected in the vitreal cell layer of the retina (i.e. INBL) where differentiated ganglion and amacrine cells reside (see blue arrowheads in Figure 4A, first column). We did not encounter any case in which miRNA expression was detected only in the mitotic progenitors of the outer neuroblastic layer (ONBL) at either E16.5 or at P0. This was expected given that the miRNAs we analyzed by RNA ISH were selected based on a microarray study carried out on fully-differentiated, adult ocular tissues and is also in line with the general view that miRNAs principally promote cell differentiation 15 . In a limited number of cases (e.g. miR-96, miR-184), staining at P0 was extended also to the ONBL, where immature photoreceptors are found (see green arrowheads in Figure 4A). For these miRNAs, photoreceptor-enriched expression was further confirmed at P8 and P60 (green arrows in Figure 4A, second and third column). Interestingly, among the retina-expressed miRNAs, we identified miR-542-3p, that showed an apparent low dorsal to high ventral gradient of expression in the developing (E16.5) and newborn (P0) retina (Figure 4B).

The laminar organization of the mammalian retina - in particular at the adult stage when all differentiation events have been completed - can be exploited to deduce the possible identity of cells expressing a given miRNA. On that basis, we identified miRNAs (e.g. miR-409-5p, miR-433, miR-541, miR-742) that, in the adult retina, are clearly detected in the vitreal part of the Inner Nuclear Layer (INL) and are likely to be expressed in amacrine cells (blue arrows in the third column of Figure 4A; Database). Another class of miRNAs (e.g. miR-29c, miR-30d, miR-96, miR-99b, miR-124a, miR-182, miR-183, miR-184, miR-381, miR-425) also stained, in the postnatal retina, the Outer Nuclear Layer (ONL) where rod and cone photoreceptors reside (green arrows in the third column of Figure 4A; Database). Many of the latter miRNAs were also expressed at P0 in the olfactory epithelium, which, similarly to the retina, is equipped with sensory epithelia (see database). Notably, the vast majority of the regional miRNAs detected in the retina stained preferentially, though to a different extent, the Ganglion Cell Layer (GCL) and the Inner Nuclear Layer (INL) and to a much lesser extent the Outer Nuclear Layer (ONL). This is consistent with the above-described preferential staining of the INBL at E16.5 and P0 and suggests that the neuronal populations that are part of the innermost retinal cell layers need, for their proper function, the expression of a wider catalog of miRNAs compared to photoreceptors cells. Finally, we observed subtle variations in the spatiotemporal expression patterns of polycistronically transcribed miRNAs (e.g. the miR-183/96/182). We believe that they could be explained either by the differential processing of the primary transcript or by the differential affinity of the probes to their respective mature miRNA.

The RPE is a monolayer of pigmented epithelium that lies between the photoreceptors of the neural retina and the choroid. Similarly to the neural retina, the RPE derives from the anterior neural plate (neuroectoderm). Exogenous signals from the surrounding periocular mesenchyme induce RPE specification whereas inputs from the surface ectoderm drive differentiation of the prospective neural retina (reviewed in 41 ). The initial regionalization between the presumptive RPE and neural retina is further consolidated by the expression of distinct sets of regulators, including well-established transcription factors and possibly tissue-specific miRNAs 42 .

By ISH we detected 29 miRNAs in the adult RPE (examples in Figure 4C, arrowheads), 26 of which also stained the neural retina. Instead, from the microarray analysis, we identified 68 miRNAs that are enriched in the RPE. By hierarchical clustering the majority of these miRNAs grouped in cluster B (Figure 1B) and some of them were found to share expression enrichment both in RPE and retina, similar to what we observed by ISH. This similarity in the miRNA content of RPE and neural retinal cells could be attributed to the common origin of the two tissues. Instead, the miRNAs that are differentially expressed between RPE and retina could be those that principally contribute in the establishment and maintenance of RPE identity. Finally, miRNAs that were concurrently enriched in RPE, cornea and/or lens (Figure 1B, D, E) may include miRNA signatures associated with epithelial physiology.

The cornea is an avascular, transparent structure within the anterior eye, of prime importance for light refraction and vision. It consists of three distinct layers: the epithelium, the stroma and the endothelium. The epithelium derives from the surface ectoderm whereas the corneal stroma and endothelium are formed by neural crest and mesoderm-derived mesenchymal cells 43 . Corneal function relies on the self-renewal capacity of the epithelium which in turn is ensured by a population of Limbal Epithelial Stem Cells (LESCs) mostly present in the limbus, which is located at the corneo-scleral junction 44 45 . To date, little is known about the miRNA content of the cornea and on its contribution to corneal differentiation and homeostasis.

By RNA ISH, miRNA expression in the cornea could be detected as early as E16.5 and P0 in 5 and 14 cases, respectively (Figure 5A; Additional file 4; Database). At P8, when the epithelium is actively proliferating and at P60, when the cornea has reached adult thickness, we could detect 22 and 54 regional miRNAs, respectively (Figure 5A; Additional file 4; Database). From the latter group, 11 stained clearly both the corneal epithelium (red arrowheads) and endothelium (blue arrowheads, right column in Figure 5B; Database) while the remaining ones (e.g. miR-106a, miR-130a, miR-132, miR-494, miR-31) were detected only in the epithelium (left column in Figure 5B; Database). Collectively, our results confirm the cornea expression of miRNAs already reported in literature 27 31 32 and reveal the corneal-enrichment of others that had not been previously described to be expressed in the eye (e.g. miR-130a, miR-130b, miR-132, miR-129-3p, the miR-200 family, miR-468, miR-874). Most importantly, the ISH analysis enhances the current scarce knowledge on the cellular distribution of these miRNAs within the cornea. In particular, we can differentiate between miRNAs that are expressed exclusively in the ectoderm-derived epithelium and others that are expressed both in the epithelium and the endothelium. Interestingly, among the miRNAs detected in the corneal epithelium, we could highlight different patterns of distribution in the limbus. For instance, 14 miRNAs clearly stained the entire corneal epithelium, including the limbus (e.g. miR-494; Figure 5B; Database), while at least 9 were not detected in the limbal region despite being expressed in most of the corneal epithelium (e.g. miR-31, Figure 5B; Database). Thus, it is tempting to speculate that such differences in miRNA expression within the limbus may contribute to the establishment of the proper environment required for LESCs maintenance.

The lens is an elastic, transparent structure located in the anterior segment of the eye with a role in light refraction and focal distance adjustment. During embryogenesis, the lens derives from the surface ectoderm following inductions from the neuroectoderm and consists of the capsule, a layer of cuboidal epithelium and the lens fibers.

By RNA ISH we detected 39 regional miRNAs in lens, especially at prenatal (E16.5) and postnatal (P0) stages when the lens structure remains intact after sectioning. In all cases, staining was evident in the anterior lens epithelium, where cells actively proliferate during lifetime and in the equatorial zone, where cells differentiate to lens fiber cells (Figure 6). Interestingly, some of the lens-enriched miRNAs are expressed at significant levels in the cornea (Figure 1D) and by ISH we frequently observe concurrent lens and cornea miRNA expression (e.g. miR-130a, miR-149, miR-883b-5p; red arrowheads in Figure 6), probably reflecting the common origin of these epithelial tissues.

RNA ISH analysis revealed a number (n = 50) of regional miRNAs that stained specifically the uveal tunics of the mouse eye in particular the iris and the ciliary body (CB) (Additional file 4; Figure 7A). Among these, there were miRNAs with a reported role in smooth muscle (e.g. miR-143 and miR-145) 46 and stem cell differentiation (e.g. miR-106a and miR-203) 47 48 (Figure 7A, B). This is in line with the fact that both the iris and ciliary body consist mainly of smooth muscle while their pigmented epithelia are niches of progenitor cells with stem-like properties of self-renewal and multi-potentiality 49 50 . The remarkable plasticity of these cell populations allows them, under appropriate culture conditions, to be trans-differentiated into pigmented neurospheres that express neuronal or glial markers 50 . In that respect, the miRNAs we detected in the iris and ciliary body (Additional file 4; Database) could represent important mediators of stemness and plasticity of pluripotent cells with neurogenic potential. This is particularly true for those that will turn out to be specifically expressed in the pigmented epithelia of these structures, though further experimental validation will be needed to confirm this hypothesis. Similarly, the strong and specific staining for miR-143 and miR-145 in the iris sphincter at all developmental stages analyzed (Figure 7B) is consistent with recent reports that demonstrate the key role of these two miRNAs in modulating smooth muscle cell proliferation and differentiation 46 . By microarray profiling these two miRNAs are highly enriched in the RPE (Figure 1B). That could in part explain the potential of RPE cells to lose epithelial characteristics and transdifferentiate into myofibroblasts during pathogenesis of proliferative vitreoretinopathy 51 .

Total RNA from eye tissues of 8-week old adult C57BL/6 mouse was obtained using the miRNeasy kit (Qiagen) according to the manufacturer's instructions. RNA was quantified using the NanoDrop 1000 (Thermo Fisher). RNA quality was assessed by gel electrophoresis.

For the ISH experiments, CD1 embryos were removed by Caesarean section at the developmental stage of E16.5. Heads were surgically removed from E16.5 and P0 mice. P8 and P60 mice were sacrificed by cervical dislocation and the eyes were surgically removed and marked with a burn spot in order to define the dorso-ventral orientation. The optimal conditions for sample pre-treatment prior to automated ISH with LNA probes were set-up in preliminary test experiments. All specimens were embedded in O.C.T. compound (Tissue Tek) and cryosectioned. Serial frontal 18 μm cryostat sections were systematically collected on microscope slides starting at a standardized plane of section to ensure reproducible anatomical coverage. The sections were fixed in 4% paraformaldehyde in PBS, acetylated in 1 × triethanolamine and dried in an ethanol series.

All animal manipulations were done in accordance with the European regulations for the use of laboratory animals and authors confirm adherence to the ARVO (Association of Research for Vision and Ophthalmology) Statement for the Use of Animals in Ophthalmic and Vision Research.

Microarray-based miRNA profiling of total RNA isolated from eye tissues was performed by Exiqon on the miRCURY LNA™Array Version 11.0 annotated to miRBase 12.0 applying a "common reference" design in which each sample in the study is hybridized against a common reference. Microarray hybridizations were performed in duplicates. Briefly, biological duplicates of each tissue (i.e. Retina, Lens, Cornea, RPE; labeled with Hy3) and RNA from entire eyes (labeled with Hy5) were hybridized on the same slide. To determine dye-bias, the set-up included an additional slide on which RNA from the entire eye was labeled with both dyes. All samples were then normalized against the common reference allowing direct comparison among them. The data obtained was normalized using a global LOESS (Locally Weighted Scatterplot Smoothing) regression algorithm 54 . The microarray analysis was performed according to MIAME standards and the results are available from the GEO database with the accession number GSE22882. MiRNA expression of each sample was compared to the reference RNA ('entire eye') and the difference in Log2 median ratios (ΔLMR) was determined (see Additional file 5 for a complete list of differential expression values). MiRNAs lacking reliable read-outs were labeled with the acronym "NA" (e.g. when 3 or more of the 4 replicated measures were flagged by the image analysis software due to the presence of empty spots, saturated spots or spots lacking optimal morphology, or when the normalized signal intensities were comparable to the background level, set as 1.5× the median signal intensity of the given slide). A false discovery rate (FDR) <0.05 was used to assess significant miRNA differential expression (estimated by 1000 permutations and calculated with the freeware dCHIP; http://biosun1.harvard.edu/complab/dchip/).

Cluster analysis was performed in Java using the MultiExperiment Viewer (MEV) packages ver. 4.5 55 . An average linkage clustering analysis of the data was based on Pearson Correlation distance.

Gene Ontology (GO) analyses were performed using the web-tool DAVID at http://david.abcc.ncifcrf.gov/home.jsp and default parameters 56 . GO analyses were performed on the non-redundant list of predicted targets compiled from three target prediction software (MiRanda, Pictar, TargetScan) 37 38 39 . The obtained BP and KEGG categories were filtered for FDR ≤ 5 and FDR ≤ 20, respectively, against the default Mus musculus background (DAVID 6.7). Redundant terms and commonly encountered categories were eliminated. Results are shown in Additional file 2.

Hybridizations were performed in an automated manner using a Tecan Genesis liquid handling platform 40 . Briefly, the slides were placed in flow-through chambers positioned into a temperature-controlled rack. All solutions were then added using a computer-controlled liquid handling system. This system allowed the concurrent use of 15 different probes. For detection of the mature miRNA sequence, 5'DIG pre-labeled miRCURY LNA™ miRNA Detection Probes (Exiqon) were used at a final concentration of 10nM. Following probe reaction with anti-digoxigenin-POD (Roche) an amplification step was performed using the TSA Biotin System (NEN Life Science Products) which enhances the sensitivity to at least 5- to 10-fold. Hybridization temperatures for the different probes had been established in previous experiments. Probes were grouped in 4 sets and hybridized at 40°C, 50°C, 55°C or 65°C (for probe sequence and hybridization temperatures see Additional file 6). After ISH, slides were photographed in a light microscope (Leica DM-RXA2) equipped with a motorized stage, a Leica electronic focusing system, a Hitachi CCD camera and a PC-based controller that drives stage and camera.