microRNA regulatory circuits in a mouse model of inherited retinal degeneration

miRNA dysregulation is a hallmark of many neurodegenerative disorders, including those involving the retina. Up-regulation of miR-1/133 and miR-142, and down-regulation of miR-183/96/182 has been described in the RHO-P347S mouse retina, a model for a common form of inherited blindness. High-throughput LC-MS/MS was employed to analyse the protein expression of predicted targets for these miRNAs in RHO-P347S mouse retinas; 133 potential target genes were identified. Pathway over-representation analysis suggests G-protein signaling/visual transduction, and synaptic transmission for miR-1, and transmembrane transport, cell-adhesion, signal transduction and apoptosis for miR-183/96/182 as regulated functions in retina. Validation of miRNA-target mRNA interactions for miR-1, miR-96/182 and miR-96 targeting Ctbp2, Rac1 and Slc6a9, respectively, was demonstrated in vitro. In vivo interaction of miR-183/96/182 and Rac1 mRNA in retina was confirmed using miR-CATCH. Additional miRNAs (including miR-103-3p, miR-9-5p) were both predicted to target Rac1 mRNA and enriched by Rac1-miR-CATCH. Other Rac1-miR-CATCH-enriched miRNAs (including miR-125a/b-5p, miR-378a-3p) were not predicted to target Rac1. Furthermore, levels of ~25% of the retinal Rac1 interactors were determined by LC-MS/MS; expression of Rap1gds1 and Cav1 was elevated. Our data suggest significant utilisation of miRNA-based regulation in retina. Possibly more than 30 miRNAs interact with Rac1 in retina, targeting both UTRs and coding regions.

The genes encoding miR-183/96/182 are clustered within 4 kb on mouse chr6qA3. This cluster is referred to as 'sensory organ-specific' , is highly expressed in retina and regulated by light 8,15 . The functional significance of the miR-183/96/182 cluster in retina has been explored. Inactivation of the cluster led to photoreceptor synaptic defects, electroretinography (ERG) abnormalities and progressive retinal degeneration (RD) in mice 16 . Depletion of miRNAs from cones resulted in loss of cone outer segments but was reversed by re-expression of just two miRNAs; miR-183 and miR-182 17 . While the role of miR-1, miR-133, and miR-142 in retina and RD is ill defined, previous work in RP mouse models 12 and a Müller cell ablation/RD model 10 found significant up-regulation of these miRNAs.
There is clear evidence of a link between retinal dysfunction (disease or experimentally induced) and altered miRNA expression. Far less is known about the target genes, which are post-transcriptionally regulated by these miRNAs in retina and indeed elsewhere. As mentioned above, prior work highlighted miR-1, miR-133, miR-142 and miR-183/96/182 dysregulation in RP models 12 including the RHO-347+ /− Rho+ /− (R347) mouse 18,19 . In the current study, the R347 mouse 18,19 is further explored to probe the miRNA regulatory pathways in RP. The strategy adopted involved analysing the retinal proteome in R347 versus 129 wild type (wt) retinas via high-throughput liquid chromatography-mass spectrometry (LC-MS/MS) and predicting candidate miRNA-target mRNA pairs by matching proteins with significantly (and inversely) altered expression to in silico computed targets for the above six miRNAs. An interesting validated target, i.e. Rac1, previously implicated in retinal degeneration [20][21][22] , was further probed utilizing miRNA capture affinity technology (miR-CATCH) and analysis of the retinal Rac1 interactome.

Results
In Silico Target Selection. Altered expression of miR-1, miR-133, miR-142 and miR-183/96/182 in the R347 mouse model has been observed 12 . Notably, the genotype of the R347 mouse used in this study, i.e. RHO-347+ /− Rho+ /− (one mutant and one wt rhodopsin allele) reflects a typical genotype of a patient with autosomal dominant RP. Our miRNA target prediction pipeline used three miRNA target site prediction alogrithms (Table 1) and we filtered for targets predicted by at least two of these prediction methods (Table 1). Additionally, we accepted only those miR targets sites where the prediction tools predicted the given site at the same location (Overlapping predicted target sites in Table 1). We identified 5301 candidate targets for the above miRNAs (Table 1); 3721 unique genes, as some genes were targeted by multiple miRNAs (Table 1). To estimate the number of predicted targets, which were expressed in retina, a wt mouse retinal transcriptome library (Supplementary Table S1) was collated 11,23 . Expression values ranged between 0 and 23554 FPKM (fragments per kb of transcript per million reads) and RPKM (reads per kb of transcript per million reads). 14335 (of 22788) genes had an expression value ≥ 0.5 FPKM/RPKM in at least one dataset and were arbitrarily deemed as being expressed in retina; however, only 12758 of these coded for proteins (BioMart, Ensembl version 79, http://www.biomart.org). The intersection between the retinal transcriptome and predicted miRNA targets was 4718 genes (3262 unique genes; Table 1).
Pathway over-representation analysis 24 was performed on the 1895 identified retinal proteins. First, proteins with increased (> 2.0-fold) or decreased (< 0.5-fold) expression between R347 and wt samples were identified (p < 0.05). Then the two groups of proteins were probed against the complete list of identified proteins; 13 (37 proteins) and 15 (44 proteins) enriched pathway-based sets were identified (Fig. 2a). Pathways with up-regulated proteins included semaphorin, TGF-beta-and TNF-alpha/NF-kB signaling, NMDA receptor-/postsynaptic activation and axon guidance, amongst others (Fig. 2a). Most but not all pathways displaying down-regulated proteins were involved in visual transduction (Fig. 2a).
A subset of potential targets was selected for further analysis. Retinal expression of Rac1, Ctbp2 and Slc6a9 ( Fig. 3a-f), as well as, Api5, Arcn1, Cav1, Dgke, Flot2, Igf1r, Negr1 and Tagln3 (Supplementary Figure S1), was analysed using immunohistochemistry in R347 and wt mice. Given that the R347 phenotype involves photoreceptors, the objective was to select targets with expression in this cell type. Some targets were not localized to photoreceptors; e.g. Cav1 was expressed largely in glial and vascular cells, Igf1r in immune cells and Dgke in the inner retina (Supplementary Figure S1). Three target proteins were selected for further exploration; Rac1, Ctbp2 and Slc6a9.
Immunohistochemistry revealed that Rac1, in line with prior observations 25 , was expressed in all retinal layers and had similar pattern of expression in R347 and wt retinas (Fig. 3a,b). Of the six miRNAs modulated in R347 retinas 12 the Rac1 3′ UTR contains a predicted combined target site for miR-96/182 (Fig. 3g) and a site for miR-142 (Fig. 3g). miR-96 and miR-182 levels were ~50% lower ( Table 2) 12 , while Rac1 protein levels were markedly higher (~3.4-fold, LC-MS/MS) in whole retina samples in R347 versus wt mice ( Table 2), suggesting a potential miR-96/miR-182-Rac1 mRNA regulatory axis. In contrast, levels of Rac1 protein in retinal membrane protein extracts were reduced by ~40% (Table 2), perhaps due to regulation via miR-142. As the increase in Rac1 protein level in whole retina samples was more significant, we opted to further analyse the miR-96/miR-182-Rac1 mRNA regulatory axis.
Ctbp2 was expressed in all retinal layers; in photoreceptors, Ctbp2 was largely confined to synapses in the outer plexiform layer (Fig. 3c,d). There was an apparent decrease in Ctbp2 in the plexiform layer in R347 versus wt retina (Fig. 3c,d). Of the six miRNAs of interest, the Ctbp2 3′ UTR contains predicted target sites for miR-1 and miR-133 (Fig. 3g), the levels of which were significantly increased in R347 versus wt retinas ( Table 2)    protein in R347 versus wt retinas was decreased by ~50% ( Table 2, LC-MS/MS) suggesting that miR-1 and miR-133 may target Ctbp2; the potential miR-1-Ctbp2 mRNA interaction was further tested in the study.
Slc6a9 was highly expressed in photoreceptor outer segments, while lower expression was determined in the inner retina (Fig. 3e,f). Lectin PNA labeling co-localised with Slc6a9, indicating that Slc6a9 is expressed exclusively in cones (Supplementary Figure S2). A similar pattern of expression of Slc6a9 was observed in R347 and wt retinas; though the photoreceptor outer segments were shorter and less organized in R347 retinas (Fig. 3e,f). Of the six miRNAs of interest, the Slc6a9 3′ UTR contains a predicted combined target site for miR-96/182 (Fig. 3g); UTR sequences in dual firefly luciferase/Renilla luciferase expression vectors and synthetic pre-miRNAs for mmu-miR-1a-3p (miR-1), mmu-miR-96-5p (miR-96), mmu-miR-182-5p (miR-182) and negative control were co-transfected into Hela cells (n = 5). 24 h post-transfection, luciferase activity of the cells was evaluated using a Dual-glow luciferase assay system. Luciferase expression levels in cells co-transfected with the negative control pre-miRNA were taken as 100% * * p < 0.01, * * * p < 0.001.  Table 2. Selected miRNA and candidate target protein levels in R347 versus wt retinas. miRNA targets were predicted using an in silico prediction pipeline employing microT 61 , miRanda 62 and TargerScan 4 tools. Target protein levels were determined using label-free LC-MS/MS in R347 and wt retinas. miRNA levels were taken from 12 .
the level of these miRNAs was decreased in R347 versus wt retinas ( Table 2) 12 . As the protein level of Slc6a9 was increased by ~70% in R347 versus wt retinas (Table 2), the data predicted that miR-96 and miR-182 may target Slc6a9.

Retinal Rac1 Interactome.
To further explore the function of Rac1 in retina, the mouse Rac1 interactome was constructed in InnateDB 31 (Fig. 5); 133 interactions were mapped. Expression profiles were added to the network from Supplementary Table S1 11,23 , which indicated that 114 (86%) of these proteins were expressed in retina. Expression of 29 members (~25%) was detected in our LC-MS/MS analysis (Fig. 5, Supplementary Table S5a). Of the direct interactors of Rac1, Rap1gds1 and Cav1 were significantly upregulated (~2-fold, p < 0.01), while Nckap1 was down-regulated (~50%, p < 0.01) in R347 versus wt retinas (Fig. 5, Supplementary Table S5a). Many of the detected proteins clustered in the IDB-8002 complex 32 (Fig. 5). These proteins connect to the Rac1 network via Mtnr1a, which is expressed at a very low level in retina (0.43 FPKM/RPKM; Supplementary Table S1). A number of IDB-8002 proteins were significantly up-regulated in R347 versus wt retinas including Rab10, Pgrmc1 and Pdia6 ( Fig. 5; Supplementary Table S5a) while others, e.g. Gnb1, were down-regulated ( Fig. 5; Supplementary  Table S5a). Rac1 interactions with 22 proteins in photoreceptor outer segments have been identified 33 . We added the mouse orthologs of these proteins to the interactome ( Fig. 5; Supplementary Table S5b). Nine of the 22 proteins were detected and analysed in our LC-MS/MS dataset (Fig. 5, Supplementary Table S5b). Expression of Eno1 and Sag was decreased (Fig. 5, Supplementary Table S5b) while expression of other proteins, such as Prdx2 and RhoA, did not change significantly in R347 versus wt retinas ( Fig. 5; Supplementary Table S5b).

Discussion
Previously, six miRNAs with altered expression in R347 mouse retinas were identified 12 . In this study, a proteome map of the R347 model was generated using high-throughput LC-MS/MS (Fig. 1). Proteomics data was combined with in silico target predictions for these six miRNAs to identify potential miRNA-target mRNA pairs. Validation of candidate miRNA-target mRNA interactions was undertaken and possible associations between identified miR regulatory circuits and cellular function(s) explored.
As the retina is a complex tissue composed of many cell types, in some cases, it is possible that the determined miRNA 12 and protein changes may have occurred independently in different retinal cell types. Additionally, changes of protein and miRNA levels in R347 versus wild type retinas may have resulted from changes in retinal cell composition due to the progressive retinal degeneration in R347 mice. To minimise potential misinterpretation of data due to the above, cellular colocalisation of the three targets followed up (i.e. Rac1, Ctbp2 and Slc6a9) and their targeting miRNAs was shown ( Fig. 3 and Table 2). For example, both miR-183/96/182 12,15 and Rac1 have a pan-retinal expression pattern, while miR-1 and Ctbp2 are coexpressed in photoreceptor cells (Fig. 3 and Table 2). To minimise changes in cell composition due to degeneration in R347 retinas, one-month old animals were used. By this age, maturation of the retina is complete, while photoreceptor cell loss is still relatively modest, i.e. approximately 25% 12,18 . Additionally, when analyzing miRNA and protein changes we focused on alterations in excess of ± 25% (i.e. < 75% or > 125% of wt levels). For example, in pathway over-representation analysis, we set cut-off values of < 0.5-fold or > 2-fold change (p < 0.05) in individual protein levels between R347 versus wt samples. Notably, while controlling error, we used the miRNA and corresponding protein changes only to predict  24 . First, active mRNA:miRISC complexes are cross-linked using formaldehyde fixation. Cells are lysed and capture oligonucleotide probes (complexed with metal beads) are hybridized to target mRNAs of interest. Next, captured mRNAs with bound miRISC complexes are pulled down using magnetic separation. Unbound mRNAs are washed away resulting in enrichment of target mRNA:miRISCs complexes. Finally, cross-links are reversed and capture oligonucleotides removed enabling evaluation of target mRNA and the captured targeting miRNAs. (b) Rac1-miR-CATCH was performed using C9 plus C10 capture oligonucleotides (Capture; n = 3) or scrambled control oligonucleotide (Scrambled; n = 3). Total RNA was purified from the samples and Rac1, Ttc21b, Folr1, Tmed5 and Plxna4 mRNAs were quantified by RT-qPCR (n = 3); note that the y-axis is in log scale. (c) Expression of miR-96 and miR-182 was analysed using Exiqon rodent miRNA PCR panel (PCR Panel; n = 2) and Applied Biosystem TaqMan microRNA Assays (TaqMan; n = 3). As miR-96 and miR-182 target the same site (TargetScan) 4 , the combined miR-96 and miR-182 levels are also given (miR-96/182 target site). * p < 0.05, * * p < 0.01, * * * p < 0.001. (d) miRNA targeting of Rac1 mRNA was analysed via combination of in vivo Rac1-miR-CATCH miRNA enrichment (Exiqon's rodent miRNA PCR panel) and in silico predictions for miRNAs targeting the Rac1 3′ UTR (miRSystem) 29 and the Rac1 cDNA (RNA22) 30 . Position of target sites corresponding to miRNAs, which were both enriched and in silico predicted to target Rac1 are given. The blocks represent the seed/target regions reported by the different prediction tools used and range in size from 6 bp to 30 bp. CDS: coding sequence.
As the retina is rich in membranes and associated proteins, beside standard protein extraction from whole retinas, proteins were extracted from enriched retinal membranes 34 , thereby increasing number of identified protein IDs by ~82% (Fig. 1b). We estimated that our quantitative LC-MS/MS enabled analysis of ~9.7% of the retinal proteome, indicating superior coverage compared to prior studies of mouse retina, e.g. 35 . 811 proteins exhibited differential expression between R347 and wt mouse retinas. To explore the possible function(s) of these proteins in retina, pathway over-representation analysis was performed 24 . Most up-regulated pathways in R347 versus wt retinas were related to signal transduction and synaptic plasticity (Fig. 2a). A number of up-regulated pathways were associated with semaphorin interactions (Fig. 2a), secreted transmembrane proteins involved in nervous system development, axon guidance, neuronal plasticity and degeneration 36 . Synaptic plasticity and reorganisation may represent functions of these in the R347 retina. For example, the observed alteration in Ctbp2 expression (Fig. 3c,d)  let-7f-5p n/a n/a 1 3′ miR-151-3p n/a n/a 2 C miR-652-3p n/a n/a 1 C 59 miR-672-5p n/a n/a 5 5′ , C, 3′ miR-139-5p n/a n/a 1 3′ b miR-125a-5p 7.38 n/a n/a n/a miR-125b-5p 5.18 0.0822 n/a n/a miR-378a-3p 4.14 n/a n/a n/a miR-204-5p 3.95 n/a n/a n/a miR-181b-5p 3.80 0.0229 n/a n/a miR-30c-5p 3.10 0.0174 n/a n/a 59 miR-328-3p 2.79 0.1867 n/a n/a miR-211-5p 2.76 0.1614 n/a n/a miR-3107-5p 2.38 n/a n/a n/a Table 3. miRNA targeting of Rac1. miRNAs targeting Rac1 were predicted in silico using miRSystem 29 and RNA22 30 , and compared to miRNAs enriched via in vivo retinal Rac1-miR-CATCH. The miRNAs from the intersection of these two lists are given in part a, while miRNAs enriched in Rac1-miR-CATCH but not predicted in silico to target Rac1 are given in part b. miRNAs are listed in order of miR-CATCH enrichment value and corresponding p values (Student's t-Test) are provided. Note, that p values were not calculated if one or both of the scrambled control samples were not amplified (n/a); enrichment values were not calculated if the scrambled control samples were not amplified (n/a). The number of predicted target sites, the location of the predicted target sites (5′ : 5′ UTR, C: coding sequence, 3′ : 3′ UTR) and reference if targeting has previously been reported are given.
regulate the cytoskeleton, cell adhesion, and survival; such functions could also be targeted in degenerating retinas. Another key group of up-regulated pathways involved TGF-beta receptor and TNF-alpha/NF-kB cytokine signaling (Fig. 2a). TGF-beta is involved in regulation of cell proliferation and differentiation, and is known to influence microglia activation 38 and developmental apoptosis in retina 39 . Microglia activation and apoptosis characterize the R347 retina; the TGF-beta receptor may play a role in these processes. TNF-alpha can be pro-or anti-apoptotic depending on the pathway involved. However, it may have an adverse role in retina as TNF-alpha blockers suppressed retinal damage in a retinal ischemia model 40 . Most down-regulated pathways in R347 retina were related to phototransduction (Fig. 2a). As photoreceptor outer segments, the subcellular compartment for phototransduction, are compromised in R347 retinas, it is likely that decreased levels of some of these proteins were due to loss of outer segments rather than actual reduction in expression level. Unless noted, according to miRTarBase 41 , the miRNA-target mRNA interactions reported here have not been described previously. miR-1 and miR-133 form a miRNA cluster and can influence neuronal function 42 . We previously reported a marked up-regulation of miR-1/133 in mouse models of RP 12 . Our data suggest that miR-1 may target three functional axes in the R347 retina; G-protein signaling/visual transduction, mitochondrial function, and synaptic transmission (Fig. 2b, Supplementary Table S3). We validated miR-1 targeting of Ctbp2 in a 3′ UTR assay (Fig. 3h); notably, the human equivalent miRNA (hsa-miR-1-3p) has also been shown to target CTBP2 in HeLa cells 43 . As the Ctbp2 3′ UTR also has a predicted target site for miR-133, miR-1/133 may co-target Ctbp2 (Fig. 3g); however this was not tested in our study. The miR-1/133 cluster and Ctbp2 are co-expressed in photoreceptors; expression of both miR-1 and miR-133 is increased by ~20-fold in R347 versus wt photoreceptors ( Table 2) 12 . Ctbp2 levels were reduced by ~50% (LC-MS/MS) and Ctbp2 immunolabeling was reduced in photoreceptor synaptic regions (Fig. 3c,d) in R347 retinas; a similar reduction of Ctbp2 has been observed in synaptic remodeling following retinal detachment 37 . The data above suggest that miR-1 suppresses Ctbp2 in R347 retinas and that miR-1 (and possibly miR-133) may regulate synaptic remodeling at photoreceptor synapses by targeting Ctbp2.
The sensory organ-specific miR-183/96/182 cluster has been studied extensively in retina 15 ; e.g., its inactivation results in progressive RD in mice 16 . Remarkably, expression of miR-183/182 is sufficient to maintain outer segments and expression of cone opsins in cone photoreceptors 17 . Notably, the miR-183/96/182 cluster is regulated by light; it is down-regulated in dark-adapted and up-regulated in light-adapted retinas 8 . Pathways potentially regulated by miR-183 include GABA receptor activation, L1cam-mediated interactions and L1 signal transduction (Fig. 2b). A significant overlap between miR-96 and miR-182 targets was established involving enriched pathways for solute carrier-mediated transmembrane transport and Robo receptor signaling. Pathways potentially regulated by miR-96 exclusively included execution of apoptosis and integrin mediated cell adhesion (Fig. 2b); whereas those identified for miR-182 comprised transmembrane transport of small molecules, G-protein signaling, synaptic transmission and cell-adhesion (Fig. 2b). Atp1b3, Paip2b and Slc1a1 have been previously identified as retinal targets for miR-183/96/182 8 . Our data further validates Atp1b3 and possibly Slc1a1 targeting by this miR cluster in R347 retina as their expression increased by 31.6% (p < 0.001) and 38.9% (p > 0.05), respectively; Paip2b was not detected.
One of the miR-96/182 targets identified in this study was Slc6a9; these miRNAs and Slc6a9 are co-expressed in photoreceptors (Fig. 3e,f and Table 2) 12,17,44 . In vitro 3′ UTR assays confirmed miR-96 targeting of Slc6a9, while miR-182 (a target site with lower conservation) did not suppress Slc6a9 (Fig. 3h). Additionally, we have demonstrated that Slc6a9 is exclusively expressed in cone photoreceptors (Supplementary Figure S2). While 'uneven' expression of Slc6a9 in the outer nuclear layer was previously observed 44 , to our knowledge, cone-specific expression of Slc6a9 has not been demonstrated before.
Another interesting target identified for the miR-183/96/182 cluster was Rac1 (Fig. 3a,b and Table 2). Rac1 is an essential component in the CNS where it regulates axon growth, neuronal morphology and survival 45 . In photoreceptor cells, a key subcellular compartment for Rac1 is the outer segment 46,47 . Here, Rac1 is activated by intense light via binding to rhodopsin 45,47 . In a light-induced photoreceptor degeneration model, Rac1 was activated while its mRNA expression also increased 22 . Rac1 activation was also demonstrated in a diabetic retinopathy model 21 . This may be due to Rac1′ s involvement as a component of the NADPH oxidase system 48 , which contributes, for example, to diabetes-induced oxidative stress in the retina 49 . In the present study, we established more than 3-fold increase in Rac1 protein level in whole retina protein extracts in R347 versus wt mouse retinas. We also found a less prominent decrease (~40%) of Rac1 in the retinal membrane protein extracts; we speculated that this reduction in membrane bound Rac1 level could have been caused by either protein relocation and/or the marked loss of rod outer segments (a major membrane component) in R347 retinas, rather than bona fide alteration in Rac1 expression. Other studies also suggest that up-regulation of Rac1 could be a common feature in photoreceptor degenerations. For example, constitutive activation of Rac1 in developing rods results in rod mislocalization, lack of formation of segments, and abnormal synaptic localization 20 . Conditional knockdown of Rac1 in photoreceptors provided protection against light-induced photoreceptor death and did not have negative effects on retinal structure and function 26 . As Rac1 is a component of the NADPH oxidase system that produces reactive oxygen species 48 , protection in this model may relate to modulation of this system.
To learn about potential Rac1 regulatory circuits in retina, the Rac1 interactome was constructed and interrogated ( Fig. 5 and Supplementary Table S5). We detected ~25% of the interactome members by LC-MS/MS; of the direct interactors, Rap1gds1 and Cav1 were significantly up-regulated (~2-fold, p < 0.01) in R347 retinas. Rap1gds1 is a stimulatory GDP/GTP exchange protein, which activates RhoA and Rac1 50 . Parallel up-regulation of Rap1gds1 and Rac1 therefore suggests a marked activation of Rac1. Cav1 is a key component of caveolae plasma membranes where it interacts with various signaling molecules 51 . Cav1 is primarily expressed in retinal Muller cells (Supplementary Figure S1) 52 suggesting activation of the Rac1 axis in glial cells. Additionally, nine Rac1 photoreceptor outer segment interactors (reported previously) 33 were detected in our LC-MS/MS study ( Fig. 5 and Supplementary Table S5b). Levels of these proteins either decreased (such as Eno1 and Sag) or did not change significantly (e.g. Prdx2 and RhoA). Interpretation of these data was hampered by the significant loss of outer segments in R347 retina, which likely interfered with these protein levels.
Rac1 and miR-183/96/182 are co-expressed in retinal cells (Fig. 3a,b and Table 2) 12 . miR-96/182 targeting of Rac1 was validated for both miRNAs by in vitro 3′ UTR assays (Fig. 3g,h). Additionally, using miR-CATCH (Fig. 4) 27 , we established that both miR-96 and miR-182 interact with Rac1 in vivo in retina. Note, that the Rac1 3′ UTR also contains a predicted miR-142 target site, functionality of which we did not test. As hsa-miR-142 targeting of RAC1 was shown in human hepatocellular carcinoma cell lines 53 , miR-142 targeting of Rac1 is also possible. Our data suggest that miR-183/96/182 cluster has a significant influence on Rac1 expression and that this regulatory circuit may play an important role in both healthy and RP retinas. Linking our results to current knowledge suggests that in retina, parallel to activation of Rac1 46 , light up-regulates expression of the miR-183/ 96/182 cluster 8 , which in turn may provide negative feedback to Rac1 translation. Additionally, reduced expression of miR-183/96/182 cluster in R347 retinas may decrease efficacy of this feedback mechanism and contribute to elevated Rac1 levels.
While the miR-183/96/182 cluster has an essential role in sensory organs, these miRNAs are also implicated in regulation of non-sensory cells and disorders (e.g. cancer) 54 . Previously established functions for miR-183/96/182 include regulation of circadian rhythm, apoptosis 54 , photoreceptor differentiation and synaptic connectivity 15 . Our data confirm apoptosis, signal and synaptic transduction and add two novel categories; transmembrane transport and cell-adhesion (Fig. 2b, Supplementary Table S3). Based on combination of in silico target computation and high throughput proteome analysis, we predicted more than a hundred miRNA-target mRNA interactions in retina (Supplementary Table S3). We tested five of these interactions using miR-CATCH and/or 3′ UTR assays and validated four interactions suggesting that our predictions are robust. As such, many other genes in Scientific RepoRts | 6:31431 | DOI: 10.1038/srep31431 Supplementary Table S3 may also represent genuine targets for these miRNAs, which could be validated in further studies.
While widespread binding of miRNAs to mRNA coding regions has been documented 4,55,57 , most prediction tools focus on 3′ UTRs. Notably, miR-CATCH detects any mRNA-bound miRNAs independent of the location of the site 27 . RNA22 30 analysis of the Rac1 cDNA was used to search for miRNA target sites outside the 3′ UTR. Target sites in the 5′ UTR and coding region, which were both in silico predicted and the corresponding miRNAs enriched in Rac1-miR-CATCH comprised let-7a/e-5p, miR-9-5p, miR-26a-5p, miR-151-3p and miR-652-3p ( Fig. 4d and Table 3a). Two of these miRNAs, i.e. mmu-miR-124-3p 58 and hsa-miR-652-3p 59 , have been shown to target Rac1/RAC1, respectively. Other Rac1-miR-CATCH-captured miRNAs, which were not predicted to target Rac1 included miR-125a/b-5p, miR-378a-3p and miR-204-5p; all of which had ≥ 4-fold enrichment (Table 3b). Non-canonical modes of miRNA binding 60 , not identified by miRNA target site prediction algorithms, may underlie the enrichment for these miRNAs. Our data suggest that more than 30 miRNAs may interact with Rac1 mRNA in retina, targeting the 5′ UTR, coding region and 3′ UTR, using both canonical and non-canonical modes of action ( Fig. 4d and Table 3). miRNA dysregulation is a hallmark of RD. Focusing on predicted targets for modulated miRNAs in R347 retina, including miR-1/133, miR-142 and miR-183/96/182, high-throughput proteome analysis provided a unique opportunity to explore miRNA regulation in a model system for inherited retinopathy. Our results highlight widespread effects of these miRNAs in retina, in particular, miR-1 and the miR-183/96/182 cluster; we validated a number of specific miRNA-target interactions in vitro and in vivo. Key cellular functions identified as being regulated by these miRNAs include signal transduction, synaptic transmission, cell-adhesion and transmembrane transport. Exploiting combination of miR-CATCH and in silico miRNA target predictions, we propose extensive miRNA-Rac1 mRNA interactions in retina, including targeting both coding and non-coding regions. A number of Rac1-miR-CATCH-enriched miRNAs not predicted in silico, suggest that non-canonical miRNA targeting of Rac1 may be common. The high-throughput proteome analysis enabled quantitative evaluation of expression of ~25% of the retinal Rac1 interactome. Significant utilisation of miRNA-based regulation, a number of miRNA targets, and possible miRNA-regulated cellular pathways and functions were identified in retina; some of these interactions may represent potential targets for future therapeutic intervention for RD.
Proteome Analysis. LC-MS-MS and label-free quantification were performed as described 54 . LC-MS/MS analysis was carried out on an Ultimate3000 nano HPLC system (Dionex, Sunnyvale, CA, USA) coupled to a LTQ OrbitrapXL mass spectrometer (Thermo Fisher Scientific Inc., Waltham, MA, USA) 34 . MS spectra were acquired in OrbitrapXL and up to 10 of the most abundant peptide ions selected for fragmentation in the linear ion trap. Peptides were quantified using Progenesis QI (Waters, Milford, MA, USA) and identified with Mascot (version 2.5; Matrix Science, Boston, MA, USA) software. Statistical significance was determined using ANOVA and p < 0.05 values were regarded as statistically significant. Gene Ontology (GO, http://geneontology.org) analysis was carried out using 'membrane' as query term. Pathway enrichment analysis in ConsensusPathDB (Release MM9) 24 was performed using Reactome (http://www.reactome.org), Wikipathways (http://www.wikipathways. org) and MouseCyc (http://mousecyc.jax.org) databases; the minimal overlap with the input list and the p value cut off were set to 2 and 0.05, respectively. The mouse Rac1 interactome was generated in InnateDB 31 . Subcellular localization was confirmed using Uniprot (http://www.uniprot.org) and GeneCards (http://www.genecards.org), and modified for a number of genes, in particular for the ones with 'unknown' localization. Mouse orthologs of previously identified photoreceptor outer segment Rac1 interactors were added manually 33 .