Retinal miR expression was initially evaluated using microarray analyses. Comparison of the retina versus brain samples (Figure 1a) or the retina versus mouse platform samples (the latter prepared by pooling total RNA from eight different mouse organs; Figure 1b) resulted in large differences in miR expression profiles (Additional data file 1). Utilizing Exiqon microarrays (Exiqon, Vedbaek, Denmark), 104 out of 224 probes between the retina versus brain and 152 out of 222 probes between the retina versus mouse platform exhibited statistically significant (P < 0.05) differences in miR expression. More specifically, expression of 47 miRs in the retina versus brain and 81 miRs in the retina versus mouse platform changed by more than 2-fold (P < 0.05). In fact, the variance in relative expression was in excess of ± 6 on a log₂ scale (Figure 1a,b). Note that Exiqon's microarray contains 488 mouse miR probes, but the probes that did not detect corresponding miRs in the above RNA samples were omitted from the plots; thus, the actual numbers of miRs included in Figure 1a and 1b were 222 and 224, respectively.

Based on our miR microarray data, we undertook a semi-quantitative comparison of relative expression levels of some known retinal miRs (retinal specificity based on the work reported by Karali 26 and Ryan 27 and their colleagues) in retina, brain, and mouse platform (Figure 2a). Substantial variations in miR relative expression levels between retina and mouse platform were detected, ranging from a value of more than 6 (for miR-183 and miR-96) down to about 1 (miR-125a) on a log₂ scale. Note, however, that these values are relative and therefore do not provide information about absolute miR levels. For example, miR-125a has a similar level of expression in retina, brain, and mouse platform, whereas miR-183 exhibits remarkable specificity for retina. Relative expression levels of additional miRs are given in Figure 2b, in a similar manner to those given in Figure 2a. Differences between relative miR expression levels in the retina versus mouse platform of up to 4 on a log₂ scale were detected (Figure 2b). For example, miR-9*, miR-335, miR-31, miR-106b, miR-129-3p, miR-691, and miR-26b exhibited a relatively high level of expression in the retina when compared with the brain or the mouse platform. On the other hand, the relative levels of miR-376a, miR-138, miR-338 and miR-136 were high in the retina compared with the mouse platform, but even higher in the brain. Let-7d was used as a control to indicate ubiquitous miR expression in the retina, brain, and mouse platform (Figure 2b).

Selected miRs depicted in Figure 2a,b were chosen, and their relative expression levels quantified using qPCR in the retina, brain, and mouse platform (Figure 2c). Notably, a close correlation between qPCR and microarray data was found but, because of the sensitivity of PCR, data from qPCR analysis exhibited a higher dynamic range. For example, a difference in miR-183 expression between retina and platform samples was determined to be approximately 11 on a log₂ scale by qPCR, as compared with about 6 on a log₂ scale by microarray analysis. In case of miR-184 the disparity was more significant, with corresponding log₂ values of approximately 9 (qPCR) versus 2 (microarray). Transformation of the qPCR log₂ values into fold differences suggested that highly retinal specific miRs (for instance, miR-183 and miR-96) are expressed at more than a 1,000-fold greater degree in the retina than in the mouse platform. Recently described retinal miRs, such as miR-129-3p, also exhibited remarkable preference, with expressed being more than 250 times higher in the retina than in the mouse platform (Figure 2c).

Expressions of miR-1, miR-9*, miR-26b, miR-96, miR-129-3p, miR-133, miR-138, miR-181a, miR-182, miR-335 and let7-d were explored by in situ hybridization (ISH) using locked nucleic acid (LNA) probes (Exiqon). It is notable that only the analysis of let-7, miR-181a, and miR-182 produced detectable signals (Figure 3). Let-7 was expressed uniformly in the inner nuclear layer (INL) and labeling was also apparent in the ganglion cell layer (Figure 3a). MiR-181a was strongest in expression among these three miRs and was detected in the inner part of the INL, probably corresponding to amacrine cells and in the ganglion cell layer (Figure 3b). MiR-182 was expressed in the photoreceptor cells in the outer nuclear layer (ONL, Figure 3c). Both let-7 and miR-181a were mainly localized in the nuclear layers (Figure 3a,b), in contrast, miR-182 labeling was weaker in the ONL (cell bodies) but was strongly localized in the photoreceptor inner segments and between the ONL and INL, possibly in photoreceptor synapses (Figure 3c,d). Additionally, miR-182 labeling was also observed in the outer part of the INL. Labeling patterns depicted by ISH indicate cell type specific expression and possible differential intracellular targeting of these miRs, namely to the cell body or, in case of photoreceptor cells, to the photoreceptor inner segments and synapse.

Given the emerging roles played by miRs in various diseases, we hypothesized that perturbed miR expression might contribute to some of the cellular events that underlie the pathology observed in RP. To seek experimental evidence to support this theory, miR expression profiles in retinas from an RP transgenic mouse model (P347S) 36 and c57 and 129 wild-type mice were compared by microarray analyses (Figure 4a,b,c and Additional data files 1 and 2). To reflect the adult miR expression pattern and to allow valid comparison of retinas from P347S and wild-type mice (the former with a progressive retinal degeneration and associated photoreceptor cell loss 36), animals at age 1 month were chosen for the study. Figure 5 illustrates representative retinal histology of P347S (Figure 5a) versus wild-type c57 mice (Figure 5b) at 1 month of age. Compromised photoreceptor outer segments and a slightly decreased thickness of ONL (by ≤25%) were apparent in P347S mice (Figure 4a) when compared with wild-type control animals (Figure 5b). As a result, alterations in the retinal miR profile should be similar in magnitude to that of photoreceptor cell loss (approximately ± 25%). In contrast, larger changes in intracellular miR levels should reflect changes that have occurred because of altered regulation of miR expression in the P347S mutant retina. A 2-fold change threshold was set (+100% and -50%) to screen for miRs that differed in expression between P347S and wild-type mice.

In order to account for the mixed c57/129 genetic background of P347S mice, miR expression profiles in the retinas of P347S mice were compared with those in both c57 (Figure 4b) and 129 wild-type mice (Figure 4c); additionally, miR expression profiles of wild-type c57 versus 129 strains were directly compared (Figure 4a and Additional data file 2). In the c57 versus 129 comparison, minor variations in miR expression profiles were detected; out of 640 probes on the Ambion microarray, 25 gave significant (P < 0.05) but lower than 2-fold deviations between the two strains (Figure 4a). In contrast, the P347S versus c57 retina (Figure 4b) and the P347S versus 129 retina (Figure 4c) plots demonstrated marked alterations between the P347S and wild-type mouse miR profiles. Figure 4 parts b and c are almost identical and reveal statistically significant (P < 0.05) changes of 63 and 75 out of 640 miRs respectively, with only eight and nine miRs exhibiting greater than 2-fold (P < 0.05) changes between the P347S and wild-type c57 or 129 mouse retinal miR expression profiles. Using Exiqon LNA microarray technology, 16 probes had greater than 2-fold alterations (P < 0.05) between the P347S and c57 miR profiles (Additional data file 1). Note that for a number of miRs (for example, miR-1, miR-133, and miR-96), both Ambion and Exiqon microarrays detected similar alterations in expression between the P347S mutant and wild-type retinas.

For qPCR validation, miRs with greater than 2-fold differences (P < 0.05) in expression between the P347S and wild-type mice were selected. Further criteria were that their signal values were above background for all samples and replicates, and probes corresponded to valid entries in the Sanger miR Database 3738. The above conditions were met by miR-1, miR-96, miR-133, and miR-183 (highlighted in red in Figure 4b,c); these miRs were therefore selected for qPCR quantification. Note, that in case of miR-96 greater than a 2-fold difference (P < 0.05) between the P347S and c57 mice was obtained with Exiqon microarrays only (Additional data file 1), while values from Ambion microarray analysis fell just below threshold. Some probes with greater than 2-fold changes (P < 0.05) represented unspecified Ambion or Exiqon miR sequences and thus were excluded from qPCR validation. Figure 6 displays corresponding data from the two different microarrays and qPCR analyses for miR-96, miR-183, miR-1, and miR-133. In general, a good correlation among data from qPCR and the two microarrays was found, with the exception of miR-183, for which the Exiqon microarray did not pick up the differential expression between mutant and wild-type retinas that was observed by qPCR (Figure 6). In summary, expression of miR-96 and miR-183 decreased by more than 2.5-fold (P < 0.001) in mutant retinas, whereas miR-1 and miR-133 increased by more than 3-fold (P < 0.001), as measured using qPCR. These results provide the first evidence for an altered miR expression profile in retinal disease.

Potential target transcripts for miR-96, miR-183, miR-1 and miR-133 predicted by miRanda 39 were retrieved from the Sanger miR Database 37. In order to select for targets expressed in the retina, the transcripts were screened against seven Unigene mouse retina libraries and three gene lists derived from NEIBank 40 and serial analysis of gene expression (SAGE) studies in the mouse retina 4142. Matches based on gene names were extracted, resulting in a final subset of 1,664 miRanda predicted transcripts that are associated with known genes and are present in at least one retinal library or gene list (Table 1). The resulting miR targets were sorted by miRanda score, P orthologous group value, presence in the seven retinal libraries and three eye related lists (a score of 1 to 10), and predicted miR target sites per transcript (1 to 3). Additional data file 3 lists potential retinal target transcripts with the highest rankings for miR-96, miR-183, miR-1, and miR-133. Notably, transcripts of retinal disease genes, such as Crb1 (encoding Crumbs homolog 1), Abca4 (subfamily-D ATP-binding cassette member 4), Pde6a (phosphodiesterase 6A), Prpf8 (pre-mRNA processing factor 8) and Prpf31 (pre-mRNA processing factor 31 homolog), together with an additional 48 eye disease genes, are predicted to be targeted by these miRs (Additional data file 3). A subset of highly ranked potential targets for miR-96, miR-183, miR-1 and miR-133 are implicated in the visual cycle (for example Abca4, Pitpnm1 [membrane associated phosphatidylinositol 1], and Pde6a), in cytoskeletal polarization (for example, Crb1 and Clasp2 [CLIP associating protein 2]), and in transmembrane and intracellular signaling (for example, Clcn3 [chloride channel 3], Grina [N-methyl-D-aspartate-associated glutamate receptor protein 1], Gnb1 [guanine nucleotide binding protein beta 1 polypeptide] and Gnb2 [guanine nucleotide binding protein beta 2 polypeptide]). Notably, predicted targets of miR-96 and miR-183 also include apoptosis regulators, such as Pdcd6 (programmed cell death 6) and Psen2 (presenilin 2) and transcription factors (for example, Asb6 [ankyrin repeat and SOCS box-containing protein 6] and Ndn [Necdin]). Additionally, target transcripts for miR-1 and miR-133 comprise mRNA processing factors (for example, Syf11 [SYF2 homolog RNA splicing factor], Prpf8, and Hnrpl [heterogeneous nuclear ribonucleoprotein L]), an apoptosis inhibitor (Faim [Fas apoptotic inhibitory molecule]), and proteins that are involved in intracellular trafficking and motility (for example, Ktn1 [Kinectin 1]), Actr10 [ARP10 actin related protein 10 homolog], and Myh9 [non-muscle myosin heavy chain polypeptide 9]; see Additional data file 3).

In summary, it has been demonstrated that miR expression in retinas from two wild-type mouse strains are very similar, and in contrast different patterns of expression between the retina, brain, and mouse platform were determined by miR microarray profiling. The results of the study suggest that the relative magnitude in expression of widely accepted retinal miRs varies remarkably in retina. Furthermore, the preferential expression in the retina of additional miRs, such as miR-376a and miR-691, represents a novel discovery. Retinal ISH analysis suggested cell type specific and intracellularly localized expression for the detected miRs. A comparative analysis between P347S and wild-type mouse retinas revealed a significant alteration in miR expression profiles in mutant mice, as evaluated by microarray analysis and validated by qPCR. More specifically, significant differences in expression of miR-1, miR-96, miR-133, and miR-183 in retina were observed between RHO mutant and wild-type mice. Potential retinal target transcripts for these miRs included, among others, genes implicated in retinal diseases and genes encoding components that are involved in apoptosis and intracellular trafficking.

Transgenic P347S 36 and wild-type 129 and c57 mouse strains were used in these experiments. P347S animals are on a mixed c57/129 genetic background and carry a Pro347Ser mutation in the carboxyl terminal of RHO; this mutation has been identified in some autosomal dominant RP families 32. To compensate for the extra RHO transgene, these mice were maintained on a mouse rhodopsin +/- background (Rho^+/-) 30, resulting in a P347^+/-Rho^+/- genotype. Retinal degeneration in these animals is slower than in most other RP models, with little or no photoreceptor cell loss at age 1 month and 50% of photoreceptors remaining at 4 to 5 months of age 36. The spatial expression of RHO is normal and electroretinography amplitudes are comparable to that in the wild-type animals at 1 month of age 36. Mice were maintained under specific pathogen free housing conditions. Animal welfare complied with the Association for Research in Vision and Ophthalmology statement for the Use of Animals in Ophthalmic and Vision Research and the European Communities Regulations 2002 and 2005 (Cruelty to Animals Act). At 1 month of age mice were killed by carbon dioxide asphyxiation.

For in situ hybridization studies, eyes from four animals from each strain were dissected and fixed in 4% paraformaldehyde for 4 hours at 4°C. For total RNA isolation retinas and brains were dissected immediately and extracted using the mirVana™ RNA Isolation kit (Ambion Inc., Austin, TX, USA), in accordance with the manufacturer's procedure. Tissue samples for total RNA were obtained in triplicate. In each sample six retinas were pooled, whereas individual brains were frozen in liquid nitrogen and homogenized over dry ice; 50 to 100 μg of the resulting powder was used for extraction. In order to represent the mouse body, a mouse total RNA platform was prepared by pooling total RNA from eight different mouse organs (liver, thymus, heart, lung, spleen, testicle, ovary, and kidney) from the Mouse Assorted Total RNA kit (Ambion Inc.).

Two different miR microarray technologies (mirVana™ miRNA Bioarray [Ambion Inc.] and miRCURY™ LNA miR Array [Exiqon, Vedbaek, Denmark]) were used. The mirVana technology is single-colored and profiles 640 human, mouse, and rat miRs (including 154 Ambion miRs) using amine-modified DNA probes. The miRCURY microarray is dual-colored (to accommodate parallel hybridization of a reference sample) and contains LNA probes for 342 mouse and 146 Exiqon miRs. Note, that the Ambion and Exiqon company miRs are not entered into the Sanger miR Database 37. P347S, 129 and c57 retinal samples were outsourced to Ambion Inc., and P347S retinal, c57 retinal, c57 brain and mouse platform samples were outsourced to Exiqon for miR profiling. All samples and replicates were analyzed on separate miR microarrays.

The mirVana miRNA Labeling Kit (Ambion Inc.) was used to label the samples with Cy5. The labeled samples were denatured and hybridized to the array for 12 to 16 hours at 42°C. Low stringency washes were followed by a high stringency wash to remove nonspecific binding to the array probes. The arrays were dried and images were acquired using the Axon^® GenePix 4000B scanner and GenePix software (Molecular Devices Ltd., Wokingham, UK). The raw signal for each probe was obtained by subtracting the maximum of the local background and negative control signals from the foreground signal. The data was pre-processed to remove poor-quality spots and normalization was used to remove any systematic bias. Global normalization of the microarrays was undertaken using the variance stabilization normalization 56 method. The resulting generalized log₂ values were used in further data analysis.

Using the miRCURY™ LNA miR Array Labeling kit (Exiqon), experimental samples and a reference sample were labeled in separate reactions with Hy3 and Hy5, respectively. Labeled experimental and the reference sample were combined, denatured, and hybridized to microarrays at 65°C for 16 to 18 hours. Low stringency and high stringency washes were carried out and the microarrays dried. Images were acquired using the Axon^® GenePix 4000B scanner and GenePix software. The data was pre-processed and normalized using the global locally weighted scatterplot smoothing procedure 57. Normalized log₂-transformed Hy3/Hy5 ratios were used for further analysis.

Microarray data from the above studies are available at the public database Array Express 58 using the following accession numbers: E-TABM-329 (miRNA expression in diseased mouse retina) and E-TABM-332 (comparative miRNA profile of retina, brain, and RP).

Two-step qPCR was performed using ABI's TaqMan miR Assay (Applied Biosystems, Foster City, CA, USA), in accordance with the manufacturer's recommendations. Briefly, 10 ng total RNA was reverse transcribed with miR specific primers in 15 μl reaction volumes. Reverse transcription reactions were diluted 60-fold and 5 μl was amplified in triplicates by TaqMan qPCR on a 7300 Real Time PCR System (Applied Biosystems); quantification was performed utilizing the comparative Ct method 59. RNU19 was employed as an internal control; log₂-transformed miR/RNU19 expression ratios were used for further analysis.

5'-Digoxigenin (DIG) labeled, LNA-modified oligonucleotide ISH probes were purchased from Exiqon for the following mouse miRs: 1, 9*, 26b, 96, 129-3p, 133, 138, 181a, 182 and 335, and let-7d (including sense-159) as background control. Paraformaldehyde-fixed eyes were cryoprotected, cryosectioned (12 μm), thaw-mounted onto 3-aminopropyltriethoxysilane-coated microscope slides, and stored at -20°C. Sections were post-fixed in 4% paraformaldehyde and treated with diethyl-pyrocarbonate before a 2-hour pre-hybridization step in hybridization solution (50% formamide, 5 × sodium chloride/sodium citrate [SSC; pH 6.0], 0.1% Tween, 50 μg/ml heparin, and 500 mg/ml yeast tRNA). Sections were hybridized with LNA probes at 20 nmol/l concentration at the melting temperature (Tm) minus 21°C in a humidified chamber for 16 to 18 hours. Hybridized sections were then washed with 50% formamide and 2 × SSC at the hybridization temperature. Following 1 hour of blocking in 2% sheep serum, 2 mg/ml bovine serum albumin in phosphate-buffered saline (PBS) with 0.1% Tween, the slides were incubated with anti-DIG/alkaline phosphatase antibody/enzyme conjugate (1:2,000; Roche Diagnostics Ltd, Burgess Hill, UK) overnight at 4°C. Following successive washes in PBS with 0.1% Tween, the sections were incubated with nitroblue tetrazolium and 5-bromo-4-chloro-3-indoyl phosphate substrate (NBT-BCIP; Roche) for up to 48 hours. The reaction was stopped by washes in PBS, nuclei were counterstained with 4',6-diamidine-2-phenylindole-dihydrochloride. Sections were analyzed by bright field normal and phase-contrast as well as fluorescent microscopy using an Axiophot microscope (Carl Zeiss Ltd, Hertfordshire, UK). Corresponding images were overlaid in Adobe Photoshop (Adobe Systems Europe Ltd, Glasgow, UK).

Potential retina specific targets of miR-1, miR-96, miR-133, and miR-183 were generated through computational means. Mouse transcripts predicted to be microRNA targets were retrieved from the Sanger microRNA Database 37. Predictions were computed using microRNAanda version 3 39 and filtered for P orthologous group value < 0.05. IDs of genes expressed in retina were downloaded from seven selected mouse eye libraries in UniGene (build #164) as follows: Lib.8659, NIH_MGC_94 (23,422 expressed sequence tags [ESTs] grouped into 7,638 UniGene entries); Lib.6780, NIH_BMAP_Ret4_S2 (19,072 ESTs grouped into 8,437 UniGene entries); Lib.5390, RIKEN full-length enriched, adult retina (6,089 ESTs grouped into 3,452 UniGene entries); Lib.15224, mouse retina, unamplified: mk/ml (4,658 ESTs grouped into 2,832 UniGene entries); Lib.20873, mouse retina, Y2H (nbk) (1,843 ESTs grouped into 1290 UniGene entries); Lib.12980, mouse adult retina (1,111 ESTs grouped into 872 UniGene entries); and Lib.6773, NIH_BMAP_Ret3 (961 ESTs grouped into 748 UniGene entries).

Additionally, genes were retrieved from two SAGE studies that determined genes expressed in the mouse retina and from an eye disease gene list at NEIBank 40 translated into mouse homologs: 3,516 UniGenes (with tag-level > 3) from Blackshaw and coworkers 41; 3,475 UniGenes from Blackshaw and coworkers 42; and mouse homologs of 488 eye disease genes from NEIBank 40.

Ensembl transcripts from microRNAanda prediction and UniGene entries from mouse retina libraries and lists were linked and extracted whenever gene names were available in both sources.

Data from given sets were pooled and averaged, and standard deviation values calculated. Statistical significance of differences between datasets were determined using either Student's two-tailed t-test or analysis of variance; differences with P < 0.05 were considered statistically significant.