Genomic non-redundancy of the mir-183/96/182 cluster and its requirement for hair cell maintenance

microRNAs are important regulators of gene expression. In the retina, the mir-183/96/182 cluster is of particular interest due to its robust expression and studies in which loss of the cluster caused photoreceptor degeneration. However, it is unclear which of the three miRNAs in the cluster are ultimately required in photoreceptors, whether each may have independent, contributory roles, or whether a single miRNA from the cluster compensates for the loss of another. These are important questions that will not only help us to understand the role of these particular miRNAs in the retina, but will deepen our understanding of how clustered microRNAs evolve and operate. To that end, we have developed a complete panel of single, double, and triple mir-183/96/182 mutant zebrafish. While the retinas of all mutant animals were normal, the triple mutants exhibited acute hair cell degeneration which corresponded with impaired swimming and death at a young age. By measuring the penetrance of this phenotype in each mutant line, we determine which of the three miRNAs in the cluster are necessary and/or sufficient to ensure normal hair cell development and function.

after sequence divergence occurs, clustered miRNAs can continue to evolve in parallel and maintain overlapping sets of target mRNAs due to the constraints of target availability and common selective pressures, thereby preserving those redundancies 15 . An example of this is the mouse mir-17~92a cluster, which contains 6 miRNAs that together regulate the expression of seven different subunits of voltage-gated potassium channels in spinal neurons. While overexpression of single miR-17~92a component miRNAs effectively downregulated target genes, the downregulation was enhanced when the entire cluster was overexpressed. Furthermore, downregulation of the entire cluster was more effective than single miRNAs at resolving allodynia 16 . These experiments indicate that although the seeds of these miRNAs have diverged and their targets have diversified, they cooperate by targeting different genes in a common pathway and thereby amplify a downstream effect. This cooperativity of multiple miRNAs to target multiple functionally related genes is considered a mechanism for the coordinated control of gene networks 17 .
Several studies have alluded to important roles that the mir-183/96/182 cluster may play in the retina, as well as in other tissues. A gene trap that inactivated the entire cluster resulted in progressive photoreceptor degeneration and increased susceptibility to light damage 18 , and targeted deletion of mir-183/96 had a similar effect 19 . A mouse expressing a "miRNA sponge, " in which miR-183, miR-96, and miR-182 were all depleted by overexpression of a transgene encoding multiple miRNA binding sites (MBS) for each miRNA was unaffected under normal conditions but was also sensitized to high-intensity light 20 . Loss of only miR-182 had a similar effect on light damage sensitivity, caused a decrease in ERG amplitude, and resulted in downregulation of several photoreceptor-specific genes 21,22 . This cluster is also highly expressed in hair cells 11,23 , and mutations in mir-96 have been linked to hearing disorders in humans 24,25 , as well as in the diminuendo mouse 26 . This phenotype is also apparent in two mouse models in which the entire cluster was deleted 27,28 . In the zebrafish, hair cells in the ear permit perception of acceleration and rotation, as well as sound. In addition to the ear, bundles of hair cells that form on the head and trunk, called neuromasts, sense water movement along the body of the fish, and are important for schooling, predation, and predator avoidance 29,30 . Additional studies in zebrafish have implicated the mir-183/96/182 cluster in the development of these organs 31 .
The relatively mild retinal phenotype of mir-182 −/− mice (compared to whole-cluster knockouts) suggests that those remaining miRNAs may partially compensate for the loss of miR-182. Indeed, of the hundreds of predicted miR-182 targets, the study only identified five candidates that were significantly upregulated in the retina of mir-182 −/− mice that could be correlated to retinal disease 22 . Furthermore, computational analysis with Primary sequence of each miRNA in human, mouse, and zebrafish. The minimal seed sequence is displayed in red text. Nucleotides within the seed region with 100% conservation between component miRNAs and across species are highlighted in yellow. Nucleotides highlighted in black are similarly conserved, but are outside the seed sequence. (c) miRNA expression in zebrafish head tissue during early development was measured by qPCR. For all data points, n = 3. **p < 0.01, ***p < 0.001, ANOVA with Dunnet's multiple comparisons test, relative to 1dpf. TargetScan 32 indicates that each of these five candidate genes harbor target sites for miR-183 and/or miR-96, suggesting that loss of additional cluster components could exacerbate their misregulation and the accompanying retinal phenotype.
Despite the significant number of mutations generated and experimental manipulations performed on this cluster, no study has systematically evaluated the potential for either cooperativity or redundancy among the three component miRNAs, and synthesizing a unifying theory of the cluster's function is difficult given the different models and methods employed. Given their high degree of similarity and their conservation throughout the evolutionary lineage, it is probable that these miRNAs work together in some way. Our objective in this study is therefore to understand which miRNAs in the mir-183/96/182 cluster contribute to development and maintenance of sensory cells in the visual and auditory systems, and how they might cooperate with each other to do so.
The zebrafish is a well-established animal model for the study of human retinal disease [33][34][35] . Although zebrafish lack a fovea, its retina is very cone-rich relative to rodent models, making it a particularly useful model for the study of this cell type. Furthermore, the primary sequences of miR-183, miR-96, and miR-182 are 100% conserved between zebrafish and mammals, suggesting that their functions are similarly conserved. The high fecundity of the zebrafish combined with the facility with which one can now generate targeted mutations make it well suited for experiments where multiple mutant lines need to be produced simultaneously. These qualities make the zebrafish the optimal model system in which to address our question.

Results
miR-183/96/182 expression in zebrafish larvae. miR-183/96/182 is highly expressed in mammalian sensory cells, including those in the retina, inner ear, and dorsal root ganglia 11,18,[36][37][38][39][40] . Within the mouse retina, they are expressed robustly in both photoreceptors and retinal interneurons 11 . To confirm the expression of these miRNAs in developing zebrafish larvae, we used qPCR assays designed to detect their mature products. In agreement with previous reports 31 , we found that each miRNA from the cluster is detectable in larvae as early as 1 day post-fertilization (dpf), and that their expression increases relative to a control snRNA through day 5 (Fig. 1c). We also observed persistent expression in adult zebrafish retina (not shown). We therefore concluded that the zebrafish is an appropriate system in which to proceed with our studies. miR-183, miR-96, and miR-182 are predicted to have a high degree of target overlap. To visualize potential target overlaps of these miRNAs in a broad context, we used CSmirTar (Condition-Specific miRNA Targets database) 41 to identify human miR-183/96/182 targets associated with the term "retinitis pigmentosa, " a blinding disease characterized by progressive photoreceptor degeneration that is similar to the phenotype observed in mutant mice. This yielded 125 unique genes, of which 70% had target sites for more than one miRNA in this cluster, and 44% had target sites for all three. As a comparison, we did an identical analysis using miR-191, miR-26a, and miR-181b, which are unclustered but are also highly expressed in photoreceptors 9 . These three miRNAs also had 125 unique targets linked to retinitis pigmentosa, but only 46% of them had target sites for more than one of the miRNAs, and only 13% had target sites for all three. (Supplementary Fig. S1). The substantially greater degree of overlap among miR-183/96/182 targets is further evidence that they are likely to have functional overlap, which is a common feature of clustered miRNAs. products of mutant miR-183/96/182 alleles are expressed at extremely low levels and have impaired function. In mice, the loss of miR-183 and miR-96 in addition to miR-182 produces a more severe retinal phenotype than loss of miR-182 alone 18,22 , but the contribution of each miRNA remains unclear. To resolve this in-vivo, we generated a full panel of mutant zebrafish. Using a CRISPR/Cas9-based gene editing approach, we mutated each miRNA from the cluster both independently as well as in all possible combinations. This approach would allow us to systematically, in a single model organism, determine which of these three miRNAs are necessary and/or sufficient to ensure photoreceptor survival. We predicted that if miR-183, miR-96, and miR-182 were completely redundant, only the triple mutant would exhibit any phenotype, as any single intact miRNA would compensate for the other two. Alternatively, if they act cooperatively, then single and double mutants would exhibit varying degrees of phenotypic severity or penetrance.
In zebrafish, the conventional viewpoint has been that CRISPR/Cas9-induced double strand breaks are repaired by the non-homologous end joining pathway 42 , although recent evidence suggests that an alternative repair mechanism might be involved 43 . Either mechanism generates small insertions, deletions, and combinations thereof. The mutations we recovered varied in size from 2-15 base-pairs, plus one large deletion that eliminated both mir-183 and mir-96 ( Supplementary Fig. S2, Fig. 2a). CRISPR/Cas9 targeting requires the presence of a 3-nucleotide PAM (protospacer adjacent motif) site in the target DNA sequence which is required for Cas9 binding, and therefore places restrictions on exactly where mutations can be generated 44,45 . This resulted in most of the mutations being slightly upstream of the mature miRNA sequence. However, we predicted that the mutations would nonetheless modify the hairpin structure in the primary transcript, and thus prevent normal miRNA expression by inhibiting or altering transcript processing by Drosha/Dgcr8. RNA structural models generated with CentroidFold 46 supported this prediction (Fig. 2b). While the basic hairpin structure was preserved in each case, the stem was shortened such that the sequence of the mature miRNA was closer to the branch point. Importantly, the location of the branch point is the most critical factor in determining the place at which the Drosha/Dgcr8 complex cleaves the hairpin from the primary transcript 47 . Therefore, even if these mutant transcripts are indeed viable substrates for further processing, the pre-miRNAs generated from them would likely be truncated at both the 5′ and 3′ ends and would have dramatically altered targeting properties.
We tested the processing and function of these mutant miRNAs with two approaches. First, we tested the hypothesis that these mutations would decrease their expression by using qPCR to measure the relative abundance of the mature miRNAs in mutant zebrafish. We were unable to detect any miR-183 or miR-96 from the www.nature.com/scientificreports www.nature.com/scientificreports/ homozygous mir-183/96 lri69 and mir-183/96/182 lri78 mutants, each of which carry a large mutation that deletes both mir-183 and mir-96 in their entirety, confirming the specificity of the reaction. In all other cases, we detected low amounts of product, not exceeding 24 percent of that in wild-type animals ( Fig. 3a-c). The degree to which these miRNAs were depleted in the mutants provided reasonable assurance that additional paralogs are not present in the zebrafish genome. It is notable that the mir-96 mutation in the mir-96 lri60 and mir-96/182 lri79 alleles, which is a net insertion of only 2 nucleotides and decreased expression by 90 percent and 88 percent, respectively, was nearly as disruptive as the much larger mir-182 lri61 mutation (11bp deletion). Because there was still detectable expression from most mutant alleles, it raised questions about whether they retained function. To address this, we used cultured HEK293 cells to test the functionality of a subset of these mutant miRNAs by co-transfecting them with a cognate reporter construct and measuring luciferase activity relative to a control miRNA. While wild-type miRNAs each decreased reporter activity by 38-42 percent relative to the control, mutant miRNAs did not ( Fig. 3d-f). Taken together, these results suggest that although small amounts of miRNA-like molecules may still be processed from some of these mutant transcripts, their function is severely impaired. While we cannot rule out residual function of mutant miRNAs that had detectable expression, we suggest that they are at best severe hypomorphs.

mir-183/96/182 mutants have normal retinal structure and function.
Because we observed expression of these miRNAs in larval animals, and because their deletion has been shown to cause photoreceptor degeneration in other species 9,18,19,27 , we hypothesized that mutant zebrafish larvae would exhibit similar photoreceptor loss. We further predicted that if these miRNAs were functionally redundant, only the triple mutant would be affected, as the presence of any single component would compensate for the loss of the other two. We used peanut agglutinin lectin (PNA) to label cone outer segments, and the monoclonal antibodies zpr1 and zpr3 to label red/ green double cone inner segments and rod outer segments, respectively, in 5 dpf fish. Although miR-183 and miR-182 were shown to be critical for proper cone outer segment development in mice 9 , we found no difference in the number of cone outer segments (Fig. 4a, Supplementary Fig. S3a) and only a modest shortening of mir-183/96/182 lri78 cone inner segments (WT, 11.7 ± 0.4 μm; lri78, 10.3 ± 0.1 μm, Fig. 4b, Supplementary Fig. S3b). Rod outer segments, which are immature and have a non-uniform distribution in 5 dpf fish, were similarly unaffected (Fig. 4c). Loss of miR-182 in mice caused a deficit in retina function but did not affect its structure 22 , so we also tested the mir-183/96 lri69 , mir-96/182 lri79 , and mir-183/182 lri81 double mutants as well as the mir-183/96/182 lri78 triple mutant for functional deficits in vision by evaluating the optokinetic response at 5 days post fertilization while varying the pattern contrast or spatial frequency. While the contrast response was normal for each mutant (Fig. 5a), we did see a slight, but significant, increase in the spatial frequency OKR gain of the mir-183/96 lri69 mutant relative to wild-type larvae (Fig. 5b).
We then looked for age-related effects in older animals, as the existence of incomplete redundancies may cause a delay in disease progression. Each of the single mutants, as well as the mir-183/96 lri69 double mutant, had  48 , which could potentially mask a slow degeneration. We looked for evidence of regeneration by immunostaining for PCNA, a marker of proliferating cells. PCNA staining was limited to 3-5 cells per section in all animals observed at 12 months, but was more extensive in all fish observed at 6 months, most likely due to more active retinal expansion in the younger animals. We quantified PCNA staining in the 6-month old mir-183/96 lri69 , mir-96/182 lri79 , and mir-183/182 lri81 double mutants, but found that they were not significantly different from age-matched wild-type control animals (Fig. 6c). We also looked for changes in Müller cell activity by immunostaining for Gfap (Glial fibrillary acidic protein). These cells span the entire thickness of the retina and are important regulators of retinal homeostasis, and also play an integral role during the response to retinal injury in zebrafish 48 . This antibody typically stains the end feet of quiescent Müller cells along the vitreal surface of the retina, but that staining pattern will become more irregular in response to retinal damage 49 . Again, with each mutant we saw no notable morphological changes relative to control animals at any timepoint.
Of all of the mutants generated, only the mir-183/96/182 lri78 triple mutant exhibited reduced viability, with many of these animals dying within several weeks post-fertilization. By raising fry from a mir-183/96/182 lri78/+ incross at low density (3 fry/L) to minimize competition from wild-type and heterozygote siblings, we were able to generate a survival curve (Fig. 6d). While only a single animal was lost from a parallel cross of wild-type animals over the course of the 7-week study, animals from the mutant clutch, which was predicted to contain wild-type, heterozygous, and homozygous fish in a 1:2:1 ratio, gradually died until only 78% of the original population remained. This is in agreement with the expected Mendelian ratio if the homozygous mutation were lethal. The surviving fish were genotyped at the end of the study and while wild-type and heterozygous fish were found in the expected 1:2 ratio (n = 9 and 27, respectively; p = 1.125, χ 2 goodness of fit test), we were surprised to also find four surviving homozygote animals. A pair of mir-183/96/182 lri78 homozygotes that survived to 15 weeks were examined histologically with the same antibody panel that was used for the other mutants. Although the high mortality rate of the homozygotes precluded a more robust analysis, we again found nothing indicative of retinal degeneration beyond somewhat shorter rod and cone outer segments, which we attributed to the overall stunted growth of the mutants (Fig. 6e, Supplemental Video S1). www.nature.com/scientificreports www.nature.com/scientificreports/ Variable penetrance of hair cell degeneration in mir-183/96/182 mutants is evidence of non-redundancy. We noted that the surviving mir-183/96/182 lri78 homozygotes exhibited highly irregular swimming patterns. While their unaffected siblings typically swam in a 2-dimensional plane along the bottom of the tank, the mutants spun around and frequently changed direction (see Supplemental Video S1). Furthermore, adult mir-183/96 lri69 mutants also exhibited this phenotype, although the age of onset was delayed by comparison and quite variable. We performed a cross-sectional survey of our mir-183/96 lri69 mutant colony to identify animals with irregular swimming behavior and found that the penetrance of this phenotype within a clutch of animals was positively correlated with age (Fig. 7a, Supplemental Video S2). Given that this miRNA cluster is highly expressed in zebrafish hair cells and has been proposed to regulate their development 31 , and because these cells are critical for normal swimming behavior 50 , we suspected that the animals may be suffering from a hair cell disorder. To investigate this further, we characterized the structure and function of neuromast hair cells within the lateral line system in 5 dpf fish. These cells comprise the principal component of a mechanosensory organ that allows the fish to sense and respond to vibrations and changes in water velocity [51][52][53][54] . To assess hair cell morphology and their distribution within neuromasts, larvae were stained with phalloidin to label actin-based stereocilia, and the monoclonal antibody HCS-1, which recognizes an epitope on the hair cell plasma membrane 55 . We also measured their sensitivity to gentamicin, an ototoxic aminoglycoside antibiotic that causes neuromast hair cell death in a dose-dependent manner 56 , as some miRNA mutant phenotypes only become evident when the animals are challenged with additional stressors 57,58 . All mutants had neuromasts with the appropriate complement of hair cells, and none were excessively sensitive to gentamicin when we used a dose that was sufficient to produce a www.nature.com/scientificreports www.nature.com/scientificreports/ moderate amount of hair cell death in wild-type fish (Fig. 7b). This finding is in contrast with a previous report that found that knockdown of these miRNAs with morpholinos reduced the number of posterior neuromasts as well as the number of HCS-1 + hair cells in the neuromasts that remained 31 , but we suggest that our genetic ablation approach is a more reliable method for studying gene function, as it is difficult to control for off-target effects of morpholinos 59 . We then looked more directly at neuromast hair cell stereocilia with phalloidin staining, and found that all fish had the expected number of bundles per neuromast (ANOVA, p = 0.28, Fig. 7c). However, while wildtype animals had stereocilia that formed exquisitely organized semicircular bundles, those of the mir-183/96 lri69 and mir-183/96/182 lri78 mutants had an abnormally large proportion of hair cells whose stereocilia were either reduced to a single spot or lacked that characteristic arrangement, and are possibly a combination of immature, degenerating, and incompletely polarized cells. (WT, 7.9 ± 2.0% vs mir-183/96 lri69 , 25.9 ± 4.9% and mir-183/96/182 lri78 , 25.4 ± 3.2%, Fig. 7d,e). Not surprisingly, these were the same mutants that exhibited circular swimming behavior that is consistent with a hair cell disorder. To determine if these structural changes resulted in compromised hair cell function, we recorded microphonic potentials from lateral line neuromasts in response to fluid jet stimulation of wild type, mir-183/96 lri69 , and mir-183/96/182 lri78 zebrafish larvae at 5-6 dpf. This potential is an extracellular voltage change resulting from current flow through activated mechanotransduction channels of hair cells. Deflection of kinocilia by fluid jets evoked a potential at double the frequency of stimulus deflections (Fig. 8a,b), alternately exciting hair cells of opposite orientation. Recordings from mir-183/96/182 lri78 heterozygotes showed stimulus-evoked microphonic potentials equal in average to wild-type (8.01 ± 0.51 and 8.83 ± 0.53 µV, respectively), while homozygous mir-183/96/182 lri78 larvae neuromasts had average potential of 3.19 ± 0.49 µV (Fig. 8f). The number of stereocilia bundles in the neuromasts of homozygous mir-183/96 lri69 and mir-183/96/182 lri78 fish that were observable with brightfield imaging was moderately reduced in comparison to wild type or heterozygous larvae (Fig. 8d,g,i,j), consistent with our observation of significantly more structurally abnormal stereocilia in these animals. Further analysis shows that the microphonic potential per bundle is significantly reduced in neuromasts of homozygous mir-183/96 lri69 (Fig. 8e, 0.410 ± 0.052 µV vs. 0.607 ± 0.038 µV) and mir-183/96/182 lri78 fish (Fig. 8h, 0.321 ± 0.066 µV vs. 0.648 ± 0.036 µV). Therefore, the reduction of microphonic potentials and the corresponding behavioral phenotype in these two mutants is likely due to nonfunctional or degenerating mechanosensitive hair cell bundles in lateral line neuromasts.

Discussion
There is considerable evidence that mir-183/96/182 is critical for photoreceptor maintenance in the mouse. Thus far, there have been two reports of independently generated lines in which all three of these miRNAs have been eliminated. Both lines exhibit some form of retinal pathology 18,27 . It is surprising that the zebrafish mutants we generated did not, but we can speculate about why this is the case. Importantly, we cannot rule out that the mutant miRNAs described here retain some residual function, as our reporter assays may lack the sensitivity to detect that. It is unlikely that additional paralogs are compensating for their loss, as BLAST searches for any of the individual miRNAs yielded no candidates with matching seeds that would be predicted to form hairpins. Furthermore, our own analysis of miRNA expression in mutant zebrafish makes us confident that we targeted the sole copies of these miRNAs in the genome, and the hair cell phenotype in mir-183/96 lri69 and mir-183/96/182 lri78 mutants corroborates that as well. It is possible that the early lethality of the mir-183/96/182 lri78 mutant precluded our ability to observe any late-onset retinal phenotypes in those animals. This could be resolved by re-expressing one or more of the miRNAs in hair cells so as to permit longer survival. Minor or late-onset retinal phenotypes, such as that in the mir-183/96/182 "sponge" mouse 20 , may require magnification or acceleration by stressors such as high intensity light before becoming apparent 57,58 . Finally, it is also possible that teleosts and tetrapods have evolved alternative mechanisms of photoreceptor maintenance since their divergence from a common ancestor In the zebrafish genome, approximately 50% of all miRNAs are positioned within 10 kB of another miRNA 15 , a phenomenon that is prevalent in all other species as well, although to varying degrees. Multiple theories have been put forth to explain the emergence and evolution of clustered miRNAs, including de novo birth and local and trans-duplication 12 . Since individual miRNAs within a cluster are so tightly linked, there is little chance of recombination between them, which helps to preserve the cluster's structure and prevents components from being lost 60 . Furthermore, any mutations that randomly arise in any cluster component that interfere with its www.nature.com/scientificreports www.nature.com/scientificreports/ expression will be selected against, as there is always pressure to preserve at least one functional copy of the ancestral miRNA. This provides a safe harbor in which young miRNAs can evolve, and once an advantageous function emerges, positive selective pressure further promotes its retention 60 . Because clustered miRNAs are typically co-expressed, they are exposed to the same pool of target mRNAs, and it is therefore likely that they acquire not only overlapping sets of targets, but also that their discrete target sets include genes with related functions. Indeed, four miRNAs with unique seeds from the mir-17~92a cluster, when transfected into cultured cells, each www.nature.com/scientificreports www.nature.com/scientificreports/ caused similar changes to the transcriptional profile 15 . Additionally, genes affected by individual miRNAs from the cluster could be grouped into discrete functional units 15 . The same study found that genes targeted by multiple miRNAs within a cluster are expressed at lower levels than genes targeted by only one. This lends strength to the theory that clustered miRNAs cooperate to regulate biological pathways in two ways: by co-targeting single mRNAs and by targeting separate genes that participate in a common mechanism, and is in agreement with our own observations in which hair cell phenotype penetrance in zebrafish was correlated with loss of additional components of the mir-183/96/182 cluster. Cooperative activity of these miRNAs likely includes actions on overlapping as well as discrete target sets, although our study cannot distinguish between the two.
In conclusion, we return to our original objective: to define the minimal requirement of these miRNAs for normal retinal development and function. The absence of a retinal phenotype in any of the mutants indicates that, at least in zebrafish, these miRNAs are collectively dispensable for that tissue. Our study is consistent with previous reports describing their role in hair cells [24][25][26]31,61,62 , and expands on those findings by shedding light on their relative contribution to the support of those cells. Because neither the mir-183 lri70 , mir-96 lri60 , nor mir-182 lri61 mutants had compromised hair cells, we can also conclude that there is no absolute requirement for any one of the three miRNAs. Furthermore, both miR-183 and miR-96 are each sufficient to maintain cluster function, as neither the mir-96/182 lri79 nor the mir-183/182 lri81 mutants suffered any deficiencies either. The minor phenotype in the mir-183/96 lri69 mutant, however, indicates that miR-182 is the least capable of maintaining the function of the entire cluster by itself. This suggests that one or more targets regulated by miR-183 and miR-96 but not miR-182 are crucial for normal hair cell function. Again using CSmirTar, we filtered predicted miR-183/96/182 target genes with the disease term "sensorineural hearing loss, " and found two genes that meet these criteria, FGFR1, which may regulate patterning of the developing cochlear duct 63 , and MYO6, which is expressed in mature hair cells and is critical for the integrity of stereocilia 64 . Further insight into the biology of these miRNAs will be gained by comprehensively examining their potential targets and identifying additional common genes and pathways that they may co-regulate. This set of mutant zebrafish will be a useful tool to empirically test new hypotheses that arise from that analysis. Furthermore, to our knowledge we have generated the first complete set of mutant animals for any vertebrate miRNA cluster. They will therefore be a valuable resource for the study of clustered microRNAs in general, and will broaden our understanding of how they evolve and interact.

Methods
Zebrafish husbandry. Adult zebrafish were maintained and raised on an Aquatic Habitats recirculating water system (Pentair; Apopka, FL) in a 14:10-hour light-dark cycle. The Cleveland Clinic Institutional Animal Care and Use Committee (IACUC) approved all experimental procedures and ensured they were performed in accordance with relevant regulations.
Generation of zebrafish mutants with CRISPR/Cas9. CRISPR targeting reagents were designed following established protocols 44 . Briefly, CRISPR target sequences were designed with ZiFiT and complementary oligos encoding the sequence were cloned into the BsaI site of pDR-274. Constructs were linearized with BsaI and gRNAs were synthesized using the MegaShortScript T7 kit (ThermoFisher). gRNA was purified by phenol:chloroform extraction and ammonium acetate/ethanol precipitation. We diluted the gRNA to 400 ng/ul and mixed it 1:1 with 20 μM recombinant Cas9 protein (New England Biolabs, Ipswich, MA), allowed a 5 minute incubation at room temperature to form complexes, and then injected 1nL into freshly fertilized zebrafish embryos 65 . To generate single mutants, gRNAs targeting each miRNA were injected into wildtype embryos. The mir-183/96 lri69 double mutant was a large deletion spanning both loci that was recovered from a wildtype embryo in which mir-183 was targeted. The remaining double and triple mutants were generated in a second round of mutagenesis by targeting mir-182 in embryos from the previously established mutant lines. F1 fish carrying novel alleles were identified first by high-resolution melt analysis using PrecisionMelt master mix (Bio-Rad), followed by sequencing. Once mutant lines were established, genotyping was carried out by electrophoresis with 3% agarose gels, which gave sufficient resolution to distinguish all mutant alleles except mir-96 lri60 . This small indel was genotyped by high-resolution melt analysis. Sequences for all oligos are included in Supplementary Table S1. With the exception of the mir-183/96/182 lri78 mutant, which was maintained with heterozygous crosses, all lines were maintained as homozygotes. Control animals for all experiments were age-matched WT animals from the same background strain as the mutants.
Luciferase reporter constructs and assays. Reporter constructs were generated by first inserting a SanDI site into p3E-polyA via site directed mutagenesis (GeneArt, ThermoFisher), generating p3E-SanDI. Following the protocol described by Kluiver and colleagues 66 , oligonucleotides encoding a pair of miRNA binding sites (MBS) were inserted into the SanDI site of p3E-SanDI. All constructs used here contained a single insertion encoding two MBS. MBS specificity was confirmed by using the PITA algorithm 67 to match each MBS to a database of 255 zebrafish mature miRNAs annotated in miRBase release 20 68 . We then recombined the MBS vectors with p5E-CMV/Sp6, pENTR-luc, and pDEST-Tol2pA2 using the Gateway three-fragment vector construction kit (ThermoFisher) to generate the completed reporters. To generate miRNA expression constructs, miRNAs were cloned from wildtype and mutant zebrafish genomic DNA and ligated into the EagI/Bsu36I site of pEF-GFP. All luciferase assays were conducted in HEK-293 cells in 24-well plates. Cells were grown to confluence (2019) 9:10302 | https://doi.org/10.1038/s41598-019-46593-y www.nature.com/scientificreports www.nature.com/scientificreports/ and then transferred to serum-free medium. Each well was transfected with 50 ng pRL-SV40, 25 ng reporter, and 65 ng miRNA, in triplicate, using Lipofectamine 2000 (ThermoFisher). We found that reporter output was proportional to the amount of reporter DNA transfected up to a saturation point between 50 and 100 ng. Beyond that threshold, miRNA efficacy was compromised, even when using excess miRNA template. By using only 25 ng reporter plasmid, we ensured that the assays were conducted within the linear range to maintain miRNA sensitivity. 24 hours post-transfection, cells were lysed and assayed for normalized luciferase expression using the dual-luciferase reporter assay system (Promega, Madison, WI) with a Victor X2 plate reader (Perkin Elmer, Waltham, MA). pENTR-LUC (Addgene plasmid # 17473) was a gift from Eric Campeau & Paul Kaufman 69 . pEF-GFP (Addgene plasmid # 11154) was a gift from Connie Cepko 70 . pRL-SV40P (Addgene plasmid # 27163) was a gift from Ron Prywes 71 . p3E-polyA, p5E-CMV/Sp6, and pDestTol2pA-2 are components of the zebrafish Tol2kit 72 . Sequences for all oligos are included in Supplementary Table S1.
Immunohistochemistry. Zebrafish larvae were fixed for 2 hours at 4 °C in PBS containing 4% paraformaldehyde, equilibrated in PBS + 30% sucrose, embedded in Tissue Freezing Medium (Electron Microscopy Sciences, Hatfield, PA), and sectioned at 10 μm thickness. Adult eyes were likewise fixed 2 hours in PBS + 4% paraformaldehyde, but were more gradually equilibrated by first incubating 3 hours in 5% sucrose before transferring them to 30% sucrose overnight. Adult eyes were then washed overnight in a 1:1 mixture of 30% sucrose:TFM before embedding. Transverse sections including or adjacent to the optic nerve were cut at 10 μm thickness and blocked 1 hour in PBS containing 2% BSA, 5% goat serum, 0.1% Tween-20, and 0.1% DMSO. Primary antibody incubation was carried out overnight at 4 °C in blocking buffer. Antibodies used are as follows: zpr1 (1:100, ZIRC, Eugene, OR), zpr3 (1:100, ZIRC), zrf-1 (1:100, ZIRC), PCNA (1:100, Sigma, St. Louis, MO, clone PC-10). PNA lectin conjugated to Alexa-568 (ThermoFisher) was used at 1:100 in blocking buffer. All secondary antibodies were from ThermoFisher and were used at 1:500 in blocking buffer with overnight incubations at 4 °C. Sections were counterstained with DAPI and imaged on a Zeiss Imager Z.2 with Apotome attachment, and post-processed in ImageJ. For cone inner segment analysis, we measured the length of five zpr1 + cells at regular intervals across the retina. Rod outer segment length was measured similarly with zpr3 imaging. Cone outer segments and PCNA + nuclei were counted across the entire retina. All experiments used between 3 and 6 animals per genotype.
optokinetic response (oKR). OKR measurements were recorded with a Visiotracker (TSE Systems, Bad Homburg, Germany) as previously described 73 . Each data point is the mean of measurements from at least 6 different animals. Gentamicin treatment. Live 5 dpf larvae were immersed in fish water containing 10 µg/ml gentamicin (ThermoFisher) for 10 minutes, which we found was sufficient to kill approximately 40% of the hair cells in wild-type animals. They were then rapidly washed 3 times in fish water, immediately fixed in 4% paraformaldehyde and stained and imaged as described below.
Whole mount immunostaining. Zebrafish larvae were raised to 5 dpf, fixed overnight in PBS + 4% paraformaldehyde at 4 °C, and blocked 2 hours in PBS + 5% goat serum + 1% Tween-20. To label stereocilia, samples were stained overnight at 4 °C with Alexa488-phalloidin (ThermoFisher) diluted 1:50 in blocking buffer with gentle agitation followed by five 1-hour washes in blocking buffer. Stereocilia were imaged on a Leica SP8 confocal microscope (Leica Microsystems, Buffalo Grove, IL). To image hair cell somas, similarly prepared larvae were labeled with HCS-1 (DSHB, Iowa City, IA, 1:100) overnight in blocking buffer, washed, and then incubated overnight in anti-mouse-Alexa568 secondary antibody (ThermoFisher, 1:500). Samples were imaged on a Zeiss Axio Imager.Z2 with Apotome.2 attachment for structured illumination. Post-processing was done with ImageJ and Adobe Photoshop. Counts of HCS-1-positive hair cells were averaged from six neuromasts per fish, including three posterior neuromasts and three supra-or sub-orbital neuromasts. All scoring and quantification was performed by two independent individuals on images with identifying information removed.

Measurement of neuromast microphonic potentials.
Recordings of microphonic potentials 50,74 were conducted as previously described 75 . Zebrafish larvae were anesthetized using ethyl 3-aminobenzoate methanesulfonic acid (Sigma-Aldrich) dissolved in a bath solution (120 mM NaCl, 2 mM KCl, 10 mM HEPES, monitored heart rate and circulation of blood cells to ensure the viability of larvae during recording. To view and capture images we used a Grasshopper3 CMOS camera (FLIR) and manufacturer provided software. We recorded microphonic potentials using a PC-505B amplifier (Warner Instruments), SIM983 amplifier (set at 20× , Stanford Research), and a PCI-6221 digitizer (National Instruments). Potentials were recorded from posterior neuromasts of larvae ranging from 5 to 6 dpf at room temperature (~22 °C). We used a borosilicate glass pipette (World Precision Instruments) with a resistance of 3-6 MΩ when filled with bath solution. This recording pipette was placed near the apical edges of the neuromasts. Kinocilia tufts were deflected with a fluid jet delivered via an additional glass pipette with a diameter of approximately 10 µm, which was controlled by jClamp software (SciSoft) and driven by HSPC-1 (ALA Scientific Instruments). A fluid jet pipette was placed approximately 80 µm away from the neuromast and used to deliver sinusoidal stimuli of 50 Hz generated by jClamp. Placement of fluid jet and recording pipettes were controlled by a set of micromanipulators (MPC-325; Sutter Instrument). Microphonic potentials were recorded with a jClamp software in current-clamp mode, and low-pass filtered at 200 Hz. All records represent an average of at least 500 trials.