A subset of SMN complex members have a specific role in tissue regeneration via ERBB pathway-mediated proliferation

Spinal muscular atrophy (SMA) is the most common genetic disease in children. SMA is generally caused by mutations in the gene SMN1. The survival of motor neurons (SMN) complex consists of SMN1, Gemins (2–8), and Strap/Unrip. We previously demonstrated smn1 and gemin5 inhibited tissue regeneration in zebrafish. Here we investigated each individual SMN complex member and identified gemin3 as another regeneration-essential gene. These three genes are likely pan-regenerative, since they affect the regeneration of hair cells, liver, and caudal fin. RNA-Seq analysis reveals that smn1, gemin3, and gemin5 are linked to a common set of genetic pathways, including the tp53 and ErbB pathways. Additional studies indicated all three genes facilitate regeneration by inhibiting the ErbB pathway, thereby allowing cell proliferation in the injured neuromasts. This study provides a new understanding of the SMN complex and a potential etiology for SMA and potentially other rare unidentified genetic diseases with similar symptoms.


Introduction
Spinal Muscular Atrophy (SMA) is the leading hereditary cause of infant mortality (1,2). The majority of SMA cases are caused by mutations in the survival of motor neuron 1 (SMN1) gene. SMN1 is ubiquitously expressed and a reduction of SMN1 protein leads to motor neuron death in patients afflicted with SMA. Although the incidence of SMA is approximately 1:6000 in live births, the carrier frequency for a heterozygous SMN1 mutation can approach 1:40 in adults. Many important questions remain regarding the pathology of the disease, including why the ubiquitously expressed SMN1 protein primarily impacts motor neurons, which other organs are potentially affected by SMN1 deficiencies, and whether SMA is a developmental or degenerative disease (or both).
The SMN1 protein is part of the SMN complex, responsible for the assembly of small nuclear ribonucleoproteins (snRNP) needed for pre-mRNA splicing (3,4). The SMN complex consists of nine proteins, however the majority of research on the complex has focused on the characterization of SMN1 and its role in SMA. In addition to its role in the SMN complex, SMN1 plays a role in many other biological processes, including axon growth, mRNA transport, and translational control (5,6). Although there is some evidence showing that other SMN complex members, such as GEMIN3 and GEMIN5, also have functions independent of the SMN complex (7,8), it remains largely unknown how the other SMN complex members relate to SMA, and whether other members have functions beyond the SMN complex. A reduction of the SMN1 protein in SMA results in the reduction of other SMN complex members (9), suggesting that there is a functional inter-dependence among the nine genes.
In a previous study (10), we showed that mutations in smn1 and gemin5 negatively impacted the ability of zebrafish to regenerate different tissues after injury. Regenerative medicine is a rapidly expanding field of science that focuses on replacing or regenerating organs damaged by injury, aging, or degenerative conditions. An active area of research within regenerative medicine is the restoration of hearing by replacing the lost mechanosensory receptors of the inner ear known as hair cells. Age-related hair cell death impairs the hearing of hundreds of millions of the elderly and hearing loss as a side-effect of therapeutic medications remains a major concern (11). In general, mammals have very limited regenerative capability, however many non-mammal animal models including the zebrafish have been used extensively because they possess a much broader capacity for wound healing, including the capacity to regenerate hearing after damage. Zebrafish are particularly well suited for studying the regeneration of hair cells because of a second organ that fish and amphibians possess on their skin known as the lateral line, which consists of clusters of hair cells in structures known as "neuromasts" (12). Similar to the case in the mammalian inner ear, hair cells in the neuromasts are surrounded by support cells, which in fish and amphibians can replace the lost hair cells through either mitotic division or trans-differentiation. Support cells in the zebrafish neuromast are further surrounded by mantle cells, which resemble quiescent stem or progenitor cells (13). Hair cell regeneration studies in zebrafish have uncovered numerous genetic factors and molecular pathways that are associated with the regeneration of hair cells (14); conversely, random mutagenesis studies revealed that mutations in only a small number of genes actually significantly alter hair cell regeneration specifically (10,13,15,16). There is consistently a gap between the number of genes transcriptionally associated with regeneration and the number of genes essential for regeneration in other tissues as well (17,18). Therefore, seeking novel genes essential for tissue regeneration is pivotal in understanding the core molecular mechanisms of wound healing, and for ultimately advancing regenerative medicine.
In this study, we systemically mutated all nine genes encoding SMN complex proteins. Using hair cell regeneration in the zebrafish lateral line as an assay, we identified three SMN complex members as essential factors that regulate regeneration through ErbB pathway-mediated cell proliferation.
Additional studies revealed that these regenerative members were also essential for the regeneration of other tissues and all shared common transcriptional pathways altered in the mutant larvae. Our findings demonstrated a subset of the SMN complex proteins had separate functional roles involved in tissue regeneration.

Divergent roles for SMN complex members in embryo development and hair cell regeneration
Hearing loss is one of the common sensory disorders negatively affecting the quality of life for hundreds of millions of people worldwide (11). In a search for novel genes involved in hearing regeneration, we previously performed a large-scale mutagenesis screen in zebrafish and identified smn1 and gemin5 as essential genes for hair cell regeneration (10). Both Smn1 and Gemin5 belong to the SMN complex, a multiprotein complex functioning in the biosynthesis of small nuclear ribonucleoproteins (snRNP). To investigate whether the regenerative abilities of Smn1 and Gemin5 are linked to the SMN complex activity, we independently mutated all nine genes in the SMN complex (Suppl. Table 1) and examined their involvement in hair cell regeneration. We found that in addition to smn1 and gemin5, mutations in gemin3 altered hair cell regeneration but had no effect on initial hair cell development (Fig. 1A. Suppl. Fig. 1A. We also found that mutations in the other six SMN complex genes, gemin2, gemin4, gemin6, gemin7, gemin8 and strap/unrip, had no impact on hair cell regeneration ( Fig. 1D-I). None of the nine mutants showed an overt morphological phenotype in early larvae (data not shown), but all mutants except gemin8 and strap failed to survive to adulthood (Suppl. Table 2). Altogether, these data classified the functions of the nine members of the SMN complex into three categories: three (Smn1, Gemin3 and Gemin5) are essential for hair cell regeneration and adult survival; four (Gemin2, Gemin4, Gemin6 and Gemin7) are essential for adult survival but not for hair cell regeneration; two (Gemin8 and Strap) are required for neither.
We generated an additional mutation allele for the three genes involved in regeneration, smn1, gemin3, and gemin5 to verify their role in hair cell regeneration. The second mutation alleles all recapitulated the deficits in hair cell regeneration ( Fig. 1B-C, Suppl. Fig. 1B). To examine whether the regeneration was specific to the ablation of hair cells using CuSO4, we performed the ablation using neomycin and observed similar regeneration deficits (Suppl. Fig. 1C-E).
In support of their divergent phenotypes in hair cell regeneration and adult survival, wholemount in situ hybridization analysis showed that the SMN complex genes possessed some common but also some different expression patterns at 3 days post fertilization (dpf) (Suppl. Fig. 2A-H). The brain expression of these genes in particular revealed both shared and specific expression patterns: five of these genes were restricted to a stripe at approximately the mid-hindbrain boundary. smn1 was more enriched in the eye regions and gemin5 was more condensed at the midline of the brain; gemin3 expression was relatively weaker than the others; gemin7 and strap showed a ubiquitous expression which was different from the other gemins. At 1 dpf, a stronger similarity was detected in the brain expression between smn1 and gemin5 (Suppl. Fig. 2I). Both genes were enriched in the eyes, brain and midline area.
Our whole-mount in situ hybridization analysis showed that none of the SMN complex genes were particularly enriched in the lateral line neuromasts (data not shown). However, single cell RNAsequencing analysis conducted by the Piotrowski group demonstrated that all these SMN complex genes are expressed at detectable levels in lateral line neuromasts, and different genes in the complex are expressed in different neuromast cell types at different levels (Suppl. Fig. 2J) (13). The non-identical and complex patterns of gene expression for the different SMN subunits (as well as the different phenotypes) suggest that the various roles for each protein may not be simply as co-expressed subunits, but that the composition of the SMN complex and potentially alternate functions of the subunits may vary based on expression levels and cellular context. Maternal mRNA and protein deposition allows zebrafish embryos to grow rapidly during the first few hours after birth and some maternal proteins can persist for days after fertilization. Although regeneration was analyzed at 7 dpf, we still examined whether the hair cell regeneration phenotype could be associated with the initial maternal deposition or a difference in the stability of mutant mRNAs.
We analyzed the expression level of two regeneration genes (smn1 and gemin5) and three nonregeneration genes (gemin4, gemin6 and strap) at different stages of embryonic development by semiquantitative PCR and found no clear difference between these two groups of genes (Suppl. Fig. 2K), suggesting mRNA destabilization does not explain hair cell regeneration phenotypes or eventual larval death.
Genetic interactions have been observed among SMN complex genes (19,20). To study whether there is a synergy among the three genes involved in regeneration, we generated an smn1 and gemin5 double mutant and studied the effect of simultaneous depletion of two genes on morphology and hair cell regeneration. We found the smn1 and gemin5 double mutant had a normal embryonic morphology and normal hair cell development (data not shown), as observed in both the smn1 and gemin5 single mutants. The double mutant showed the expected deficiency in hair cell regeneration; however, the level of deficiency was comparable to that of the gemin5 mutant (Suppl. Fig. 3A-C). Taken together, these data suggest there is no functional synergy among these regeneration genes, and smn1 and gemin5 appear to both be necessary and fall in the same regenerative pathway as the phenotypes in double mutants were neither synergistic nor additive.

The overall neuromast size is smaller in mutants with regenerative phenotypes
We examined the neuromast cell patterning in the mutants and control siblings at 2 days post hair cell ablation to see if we could detect a difference in neuromast size in mutants using both transgenic labeling and immunohistochemical staining approaches. Double transgenic labeling of support cells by Tg(tnks1bp1:EGFP) and hair cells by Tg(atoh1a:dTomato) in gemin5 mutants revealed that support cells in the mutant occupied a reduced area likely because of fewer cells and hair cells were fewer when compared to that of the control siblings ( Fig. 2A). Whole neuromast labeling using Tg(cldnb:EGFP) showed that the size of the neuromast in the mutant was smaller than that of the control siblings (Fig. 2B). Alkaline phosphatase staining revealed that the structure of lateral line neuromasts were much more reduced in the mutant (Fig. 2C). Co-staining with anti-hair cell antibody and nuclear dye DAPI revealed a reduction in the number of hair cells and neuromast cells (Fig. 2D-E).
We also used transgenes and immunohistochemical staining to examine the neuromast cells at 2 days post hair cell ablation in smn1 and gemin3 mutants. Consistent with the results of the gemin5 mutation, mutations in smn1 and gemin3 also caused a reduced area of support cells, impaired regeneration of hair cells and smaller neuromasts (Suppl. Fig. 4). All the data suggest the proliferative capacity in the neuromasts is reduced preventing organ growth.

Regenerative deficient mutants show reduced proliferation after injury
To directly test proliferative capacity of the support cells in the mutant neuromasts, we used an EdU incorporation assay to label the proliferation of neuromast cells after hair cell ablation. Compared to the control siblings, all three mutants possessed a reduced number of EdU positive cells (Fig. 3), suggesting that after hair cell ablation, the mutants lack the capacity to effectively proliferate either their support cells or mantle cells.

Regeneration deficient mutants are less sensitive to the ErbB pathway inhibitor AG1478
We conducted numerous tests to find pathways affected by gemin5 mutations. Most conditions were negative (Suppl. Table 3), with only AG1478 (inhibiting ErbB signaling) showing a specific phenotype. Treatment with 2 µM AG1478 caused a dramatic increase in lateral line neuromasts of control siblings, but only a mild increase in the gemin5 mutant larvae (Fig. 4A-B), indicating that the gemin5 mutant is resistant to ErbB pathway inhibition. To determine whether the reduced sensitivity was common to all three regeneration gene mutations, we also treated smn1 and gemin3 mutants with AG1478. Consistent with the findings from the gemin5 mutant, the smn1 mutant and the gemin3 mutants also showed a reduced responsiveness to AG1478 ( Fig. 4C and D). To test if the ErbB pathway responded normally in other mutants in the complex, we treated gemin6 mutants with AG1478. In contrast to the mutants that disrupted regeneration, the gemin6 mutant responded to AG1478 comparable to that of their control siblings (Fig. 4E). Altogether, these results point out that the inability to respond to AG1478 inhibition specifically in the mutants that inhibited regeneration, suggesting a mechanistic link.

Neuromast cell proliferation is not induced by AG1478 in gemin5 mutants
To understand why gemin5 mutants responded differently to AG1478, we used embryos with a double transgene Tg(pou4f3:GAP-GFP);(SqET20:EGFP) to label neuromast cells, and used an EdU incorporation assay to mark proliferating cells. We exposed the double transgenic embryos either to a mock treatment or to AG1478, and the resulting embryos were stained with the nuclear dye DAPI. In each neuromast, DAPI labeled all neuromast cells, the Tg(pou4f3:GAP-GFP) labeled hair cells, Tg(SqET20:EGFP) labeled mantle cells that demarcate the outer periphery of neuromast, and the GFP negative and DAPI positive cells in between were support cells. Quantification results showed that AG1478 promoted the proliferation of all three types of neuromast cells in wild-type larvae (Suppl. Fig.   5A-B), consistent with previous finding that AG1478 promotes cell proliferation (21).
We then applied AG1478 to gemin5 embryos possessing the same Tg(pou4f3:GAP-GFP);(SqET20:EGFP) transgenes and monitored the neuromast cells in the gemin5 mutant. When compared to the control siblings, gemin5 mutants possessed a significantly reduced number of neuromast hair cells (as visualized by Tg(pou4f3:GAP-GFP) and mantle cells (as visualized by Tg(SqET20:GFP) (Fig. 4F), indicating that the gemin5 mutation impaired neuromast cell proliferation in response to ErbB inhibition.
Several studies have indicated that AG1478 regulates neuromast cell proliferation through modulating the cell signaling activity between the Schwann cells, interneuromast cells, and the axons via WNT (21)(22)(23)(24). We tested whether we could detect disruptions in Schwann cells in gemin5 mutant embryos. Schwann cell morphology and quantity were evaluated using the Tg(foxd3:GFP) transgene or by the expression of myelin basic protein a (mbpa). Neither revealed a noticeable difference between the control siblings and mutants (Suppl. Fig. 6A-B). Lateral line axons were labeled with an antibody targeting acetylated tubulin and appeared to be comparable between the control and mutant embryos (Suppl. Fig. 6C). Wnt pathway activity was manipulated with both Wnt pathway activator BIO and inhibitor IWR1 and neither showed any differences between the mutant and control siblings (Suppl. Table 3). Altogether these data suggest the disruptions in myelination by the Schwann cells was not associated with the failure of AG1478 to induce supernumerary neuromasts in the gemin5 mutant.

Genetic mutations of ErbB pathway genes recapitulate the AG1478 effect
Similar to inhibition by AG1478, mutations in ErbB pathway genes, such as erbb2, erbb3b and nrg1, lead to an increase in lateral line neuromasts (21,23,24). Mutations in erbb3b and nrg1 appear to have no other significant impact on embryo axis patterning nor on adult survival. We therefore generated stable genetic mutations for both erbb3b and nrg1 and compared the effect of these mutations to the AG1478 effect on lateral line neuromast formation. As expected, both erbb3b and nrg1 mutations caused a dramatic increase in the number of lateral line neuromasts, however, the increase was consistently lower than that of AG1478 treatment ( Fig. 5A-B), suggesting AG1478 inhibits ErbB signaling more broadly than that mediated by either erbb3b or nrg1 alone and that there may be some redundant signaling from other related genes.
We then examined how mutations of the genes in the ErbB pathway impact neuromast formation in the gemin5 mutant background. We generated double mutants of erbb3b/gemin5 or nrg1/gemin5. Consistent with our previous observations, a homozygous mutation in either erbb3b or nrg1 alone caused an increase in the number of neuromasts, a homozygous mutation for gemin5 alone caused no alteration, and a heterozygous mutation of either gene alone or together produced no change ( Fig. 5C-D). Both double mutants displayed a lower level of increase of lateral line neuromasts when compared to erbb3b or nrg1 mutant, however, the number of neuromasts in the double mutants was significantly higher when compared to the gemin5 mutant alone, indicating disruption of the ErbB pathway could partially rescued the deficiency of neuromast formation in the mutant. The partial rescue in the double mutants suggest that AG1478 was failing to sufficiently inhibit ErbB signaling in gemin5 mutants instead of the alternative explanation that interneuromast cells were unable to respond properly to release of ErbB signaling.
Rescue was also attempted by mutating the erbb2 gene in the gemin5 mutant background.
Since erbb2 loss of function is early embryonically lethal, we generated a mosaic knockdown of erbb2 by injecting multiple CRISPR guide RNAs into the embryos from a gemin5 heterozygous incross, and then used the injected embryos to quantify lateral line neuromast formation. Mutation frequency analysis showed these erbb2 CRISPR guide RNAs resulted in a near-complete mutagenesis of the erbb2 gene ( Fig.   5E). Neuromast number quantification showed the erbb2 knockdown promoted more neuromasts in the control sibling than in gemin5 mutant (Fig. 5F).

ErbB pathway inhibition partially rescues hair cell regeneration
Activation of ErbB signaling has been implicated in the regeneration of other tissues (25, 26), so we investigated whether ErbB pathway inhibition could improve hair cell regeneration in the three mutants that disrupt regeneration. In performing the hair cell regeneration assay in the presence of the ErbB inhibitor, AG1478 had no obvious effect on the regeneration of hair cells in control siblings, however, it did exhibit a dose-dependent rescue of regeneration in all three mutants (Fig. 6). Our interpretation of the data from Figures 5 and 6 is that ErbB signaling in the smn1, gemin3, and gemin5 mutants was hyperactive, such that the increased ErbB activity was blocking AG1478 induction of ectopic neuromasts. Similarly, too much ErbB signaling was blocking the initiation of hair cell regeneration, but now AG1478 inhibition was sufficient to partially release the block in regeneration, presumably because ErbB signaling levels were generally lower in the regenerating neuromast compared to the interneuromast cells, or the level of reduction needed to see rescue was lower in the case of hair cell regeneration compared to neuromast induction.

smn1, gemin3, and gemin5 mutations affect the regeneration of multiple tissues
Regeneration of different tissues can be achieved by utilizing a common set of molecular pathways and many genes are pan-regenerative in that they are induced and essential regardless of the specific injured tissue (17). Both smn1 and gemin5 genes were involved in regulating the regeneration of multiple tissues including neuromasts, caudal fins, and livers (10). To determine whether gemin3 had similar phenotypes, we examined the regeneration of caudal fin and liver in gemin3 mutants. Like in smn1 and gemin5 mutants, we found gemin3 mutations did not alter the normal development of caudal fins or livers (data not shown), however, upon damage the mutant exhibited a deficiency in restoring the damaged tissues as was seen with the other mutants. After caudal fin amputation, the restored fin in the gemin3 mutant was significantly smaller and often missing the pigment gap ( Fig. 7A-B). Similarly, following chemical-mediated liver ablation in the Tg(fabp10a:CFP-NTR) transgenic background, the gemin3 mutant displayed a clear deficiency in liver regeneration compared to the control siblings ( Fig.   7C-D). As a control, gemin6 mutants were also examined for a role in the regeneration of caudal fin and liver. In contrast to the regeneration mutants, gemin6 mutants showed normal regeneration of both the caudal fin and the liver (Suppl. Fig. 7). These data suggest that gemin3, like smn1 and gemin5, is generally involved in regeneration, regardless of the tissue tested.

RNA-Seq reveals shared downstream targets among the genes involved in regeneration
To identify the pathways shared amongst the mutants blocking regeneration, we conducted RNA-sequencing (RNA-Seq) and miRNA-sequencing (miRNA-Seq) using the mutants of both regeneration genes and non-regeneration genes from the SMN complex. We found that the three genes impacting regeneration, regulated a common set of downstream targets which were distinct from the gemin mutations that did not affect regeneration (Fig. 8A-B). Significantly, we found erbb3b was upregulated in the three non-regenerative mutants (Fig. 8C), consistent with our hypothesis that ErbB signaling was hyperactive in these mutants (Fig. 5C). In addition, RNA-Seq data revealed that a mutation in one of the "regeneration genes" had no effect on the expression of the other two genes (Suppl. Fig. 8), suggesting there is no inter-regulation between the genes at the transcriptional level.

Observed upregulation of the tp53/Mdm2 pathway was not the major cause of the regeneration phenotype
RNA-Seq data identified an increase in expression for both the tp53 and mdm2 genes specifically in the mutants inhibiting regeneration (Suppl. Fig. 9A). Several lines of published evidence indicate that p53 interacts with Mdm2 and activation of the p53/Mdm2 pathway is associated with SMN complex activity and SMA (27)(28)(29). To investigate a potential role for the tp53/Mdm2 pathway in hair cell regeneration, we depleted tp53 genetically in both the smn1 and gemin5 mutant backgrounds. For the gemin5 study, we crossed the gemin5 mutant with the tp53 M214K mutation (30), and found that the gemin5/tp53 double mutants showed no improvement in regeneration or adult survival when compared to the gemin5 mutant alone (Suppl. Fig. 9B, data not shown). For the smn1 mutation, the tp53 and smn1 genes in zebrafish were both on chromosome 5, so we obtained double mutants carrying homozygous mutations for an smn1 2 bp insertion and a tp53 7 bp deletion, by injecting smn1 CRISPR guide RNAs into embryos harboring a homozygous 7 bp tp53 deletion mutation (31) and raised those fish for inbreeding. Consistent with the results of the gemin5/tp53 double mutant, tp53 mutants did not rescue the regeneration deficiency or the adult survival of smn1 mutants (data not shown).
RNA-Seq data showed that Mdm2 was also significantly induced in the three genes involved in regeneration. Since mdm2 mutations cause early embryonic lethality, we created a partial knockdown of Mdm2 by injecting mdm2 CRISPR guide RNAs into the gemin5 mutant background and found the resulting mosaic mutations in mdm2 did not rescue the hair cell regeneration in the gemin5 mutants (data not shown). Altogether, these data indicate that despite strong induction of tp53 and mdm2 in all three mutants blocking regeneration, the tp53/Mdm2 pathway is not a major contributor to the regeneration phenotype of the gemin5 or smn1 mutants although it does suggest that all three genes are involved in a common subset of pathways not shared by the other SMN complex proteins, and those pathways are involved in injury responses and tp53 stress responses.

Discussion
Our previous large-scale mutagenesis screen showed smn1 and gemin5, two SMN complex members, were essential for tissue regeneration (10). In this study, we expanded our mutagenesis screen to systemically examine the potential role of each of the nine members of the SMN complex in tissue regeneration. Consistent with the findings reported from other groups (32)(33)(34), we found that mutations in most SMN complex members were essential for adult survival (Suppl. Table 2). However, our genetic data suggest the nine SMN complex members can be categorized into three separate groups: smn1, gemin3 and gemin5 are required for both overall survival and regeneration after injury; gemin2, gemin4, gemin6, and gemin7 are required for survival but not for regeneration; gemin8 and strap/unrip appear to be non-essential factors for either regeneration or survival. The three regeneration members (smn1, gemin3 and gemin5) are regulating regeneration through ErbB pathwaymediated cell proliferation, and they are essential for regeneration of multiple (if not all) tissues.
Studies of the SMN complex has been ongoing for more than two decades (35) with the largest focus on SMN1 because mutations in this gene are responsible for the human disease spinal muscular atrophy (1,2). However, the association of SMN complex members with tissue regeneration was not recognized until our prior study (10) and expanded here. Our work strongly suggests that some of the SMN complex members have separate, independent functions unrelated to snRNP assembly, or that the complex isn't always functioning with all nine genes in a stable, stochiometric ratio. For example, we found transcripts were not expressed uniformly and ubiquitously, but expression varied in different brain regions and in different neuromast cells (Suppl. Fig. 2). However, no additive or synergistic interactions were observed between the three genes involved in regeneration (Suppl. Fig. 3). All three regeneration members functioned through ErbB-pathway-mediated cell proliferation ( Fig. 4 and 6), all three possessed an ability to regulate regeneration in multiple tissues (Fig. 7) (10), and none of the three appeared to be epistatic to the other two. All these findings argue that these regenerative members work together in a shared molecular mechanism. Our findings suggest that the three SMN complex members involved in regeneration possess functions separate from snRNP biosynthesis that are essential for tissue regeneration and are also related to tp53 regulation/activation, although our genetic evidence in these two functions are not directly related.
In line with our findings, previous studies have reported apparently independent activities of SMN complex members. For example, SMN1 has been implicated in motor neuron growth (36, 37), SMN1's function in motor neurons appears to be independent of snRNP biosynthesis (38), and SMN1 has a specific role in axonal mRNA regulation and axonogenesis (5,6). Furthermore, GEMIN3, an RNA helicase, is involved in cell proliferation and microRNA regulation of signal transduction (7). Gemin5 regulates smn1 expression (20), and Gemin5's C-terminus can regulate protein synthesis (8,39). Future studies should be able to evaluate SMN complex-dependent and independent functions more precisely through detailed analysis of splicing isoforms in different genes and/or cell types, under natural, diseased or regenerative conditions.
In this study, we demonstrated a link between the ErbB pathway and three of the SMN complex's proteins. Chemical inhibition of the ErbB pathway with AG1748 in smn1, gemin3, and gemin5 mutants failed to induce ectopic neuromasts or rescue hair cell regeneration, while genetic ablation of erbb3b or nrg1 was able to partially rescue the neuromast induction and hair cell regeneration, suggesting the ErbB pathway is hyperactive in the mutants. Because the ErbB pathway is associated with various neurological diseases (40), it suggests future investigation is warranted to address whether the upregulation of the ErbB pathway in the three SMN mutants is specific to injury responses or if it is also one of the underlying mechanisms in the neurodegenerative pathology of SMA.
Several studies have demonstrated that the ErbB pathway plays a promotive role in the regeneration of other tissues. For example, mutations in erbb2 or erbb3 cause a deficiency in caudal fin regeneration (26), and AG1478 treatment inhibits the regenerative proliferation of cardiomyocytes (25).
Our data indicate that the role of the ErbB pathway in regeneration differs based on tissue type. It remains unclear how the ErbB signaling is integrated into the different roles it plays in different tissues.
Our RNA-Seq data revealed erbb3b is upregulated in smn1, gemin3, and gemin5 mutants (Fig.   8C) and we found that inhibition of ErbB pathway contributes to a partial rescue of their regeneration phenotype (Fig. 6), suggesting that the ErbB pathway is, at least in part, the underlying mechanism for the deficient regeneration, and likely it is only one of many pathways affected during the regeneration.
Besides erbb3b, p53 and mdm2 were also upregulated (Suppl. Fig. 9A). The p53/Mdm2 pathway has long been documented to interact with the SMN complex. p53 has a direct physical interaction with both SMN1 and Gemin3 (29,41). p53 depletion rescues mdm2 mutant phenotypes (27). Abnormal mdm2 splicing and p53 activation are associated with the death of motor neurons in SMA (28). We found inhibition of the p53/Mdm2 pathway brought no alteration to survival or regeneration in the gemin5 mutants (Suppl. Fig. 9. Data not shown), suggesting this pathway is not the major cause of the mutant regeneration phenotypes.
In conclusion, this study provides insight into the SMN complex and potential roles for the complex in wound healing and ErbB signaling. Although SMN1 is the causative gene in the majority of SMA patients, there are still cases of SMA where the causative gene is unknown. Because we see phenotypes cluster with smn1, gemin3, and gemin5, it is possible that a fraction of undiagnosed SMA cases or related neurodegenerative diseases could be caused by variants in either GEMIN3 or GEMIN5. It is also possible that the functions of the three SMN complex members outside of snRNP assembly are somehow linked to SMA pathology and deficient regeneration is an underlying mechanism for SMA and even for other neurological diseases.

Zebrafish husbandry and embryology
Zebrafish husbandry and embryo staging were performed according to Kimmel (42). All experiments were in compliance with NIH guidelines for animal handling and research and approved by the NHGRI Animal care and Use Committee (protocol G-01-3). Adult fish survival was examined at 3 months post fertilization. Quantitative PCR (qPCR) was conducted by extracting total RNA with Trizol (Invitrogen, Cat#: 15596026), synthesizing cDNA with SuperScript first-strand synthesis system (Thermo Fisher Scientific. Cat#: 11904018), and then running qPCR with SYBR™ Green PCR Master Mix (Thermo Fisher Scientific, Cat#: 4344463). Beta-actin was used as an internal reference. Semi-qPCR analysis was conducted similarly as qPCR excepted no use of SYBR Green and amplicons analyzed on an agarose gel.
CRISPR/Cas9 mutagenesis was performed as previously described (43). The CRISPR targets and primers used for mutation detection are listed in the CRISPRz database (44) https://research.nhgri.nih.gov/CRISPRz/). CRISPR mutation rates for founder embryos were analyzed by calculating the percentage of mutant signal over the total signal (45).

Hair cell and neuromast quantification
Hair cell staining and quantification were as described (53). Briefly, for analyzing hair cell development, embryos from heterozygotic incrosses were cultured until 5 dpf, and then placed in a cell

Quantifying development and regeneration of caudal fin
Caudal fin development and regeneration were analyzed as previously described (10). In brief, embryos were obtained from a pair of heterozygous parents. Fin development was measured at 5 dpf, using the posterior of pigment gap as a positional reference. For the regeneration analysis, amputation was performed at 3 dpf, at the posterior end of ventral pigment break. The regeneration was measured at 7 dpf, continuing to use the anterior end of pigment gap as a positional reference. ImageJ was used for quantifying the fin areas. All analyzed embryos were genotyped. Graph shows the mean and s. e. m, based the quantification data from approximately 10 embryos per genotype.

Quantification of development and regeneration of liver
Liver development and regeneration was tested using the transgene Tg(fabp10a:CFP-NTR (52).
The embryos used for the analysis were the CFP-positive embryos obtained from a pair of parents with one carrying the heterozygous gene mutation and the other carrying both the heterozygous mutation and an allele of Tg(fabp10a:CFP-NTR). Liver size was measured at 5 dpf. For liver regeneration analysis, the embryos were treated with 10 mM metronidazole for 1.5 days at 3 dpf and analyzed for regeneration at 7 dpf. All analyzed embryos were imaged at a lateral view with head facing right under a Zeiss Axiophot fluorescent microscope, and afterwards genotyped. ImageJ was used to measure the liver areas. Approximately 45 CFP-positive embryos were used for each analysis. Graph shows the mean and s. e. m.

RNA-Seq and miRNA-Seq analyses
The embryos used for RNA-sequencing (RNA-Seq) and miRNA-sequencing (miRNA-Seq) were produced from a cross of a single pair of heterozygous parents, exposed to 10 µM copper sulfate for 2 hours at 5 dpf, and then subjected to caudal fin biopsy for genotyping and the body stored in Qiazol (Qiagen. Cat#: 79306) at 7 dpf. Afterwards the wild-type and homozygous mutant embryos were pooled together and used for total RNA extraction by using Qiagen miRNeasy Mini Kit (Cat#: 217004). The total RNA with an integrity score (RIN) over 9 were used for RNA-Seq and miRNA-Seq analyses.

Statistical analyses
A student t-test (two tailed) was used for comparison between two samples. One-way ANOVA was used when comparing multiple samples. A difference was considered significant when P value was less than 0.05. Error bars in the graphs represent mean ± s.e.m. Asterisks and short lines were used to indicate a significant difference between two groups. ns, P >= 0.05; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. Each experiment presented was repeated at least twice, with the replicates showing statistical significance each time.

Acknowledgement
We would like to thank MaryPat Jones, Blake Carrington, Kevin Bishop and Raman Sood from the NHGRI Zebrafish Core for mutation genotyping; Suiyuan Zhang from the NHGRI Bioinformatics Core

Competing interests
No competing interests declared. Hair cell regeneration is impaired by homozygous mutations of gemin3 hg105 (A), smn1 hg104 (B) and gemin5 hg107 (C), but not in gemin2 hg108 (D), gemin4 hg109 (E), gemin6 hg110 (F), gemin7 hg111 (G), gemin8 hg112 (H) or strap hg113 (I). Red line separates the mutations impacting regeneration from those that have no effect on regeneration. wt, wild-type. het, heterozygotes. hom, homozygotes. Error bars in the graphs represent mean ± s.e.m. The difference between wild-type and homozygote is labeled. ns, P > 0.05; ***P < 0.001; ****P < 0.0001. Error bars in the graphs represent mean ± s.e.m. There is a significant reduction in the number of neuromast cells (**** P < 0.0001). The numbers are presented as percentage because they were obtained from quantification of still confocal images.  Tg(pou4f3:GAP-GFP) and Tg(SqET20:EGFP) in the control and gemin5 hg81 mutant at 5 dpf after AG1478 treatment. Images were taken in the areas surrounding the end of yolk extension. White arrow points to the Tg(SqET20:EGFP) signal in the control embryo, which is dramatically increased in the mutant. Scale bar, 50 µm. The embryos used for the above analyses were generated from a pairwise incross of heterozygotic parents, treated with 2 µM AG1478 from 1 -5 dpf, and then used Yopro-1 staining or transgenic fluorescence at 5 dpf to analyze neuromast formation. CRISPR guide RNA injected gemin5 mutant embryos at 5 dpf. Error bars in the graphs indicate mean ± s.e.m. ns, P > 0.05; *P < 0.05; **P < 0.01; ****P < 0.0001. to drug toxicity to this genetic background, since it was not observed in the smn1 fh229 and gemin5 hg81 embryos. Graphs show the mean ± s.e.m. Statistical difference are indicated as: ns, P > 0.05; *P < 0.05; **P < 0.01; ***P < 0.001. The embryos used for the analysis were generated from a single pair of heterozygous carrier parents, ablated hair cells at 5 dpf, and then treated with 0, 2.5 or 5 µM of AG1478 from 5 -7 dpf. Representative images are shown. Liver tissue is labeled by Tg(fabp10a:CFP-NTR). Graphs show mean ± s.e.m. **P < 0.01; ****P < 0.0001.
Suppl. Fig. 2 Expression analysis of the SMN complex genes. Suppl. Fig. 3 Hair cell regeneration in smn1 fh229 and gemin5 hg81 single and double mutants.
The embryos used for the analysis were generated from a single pair of parents, each carrying heterozygous mutations for both smn1 fh229 and gemin5 hg81 . Graphs show the hair cells regenerated from the smn1 fh229 mutant (A), gemin5 hg81 mutant (B), and smn1 fh229 /gemin5 hg81 double mutant (C). The difference is significant between the smn1 wild-type and homozygotes, between the gemin5 wild-type and homozygotes, and between the smn1/gemin5 control and double mutant (**** P < 0.0001 for all three groups). There is no difference between the gemin5 homozygotes and the smn1/gemin5 double homozygotes (ns, P > 0.05. Not labeled in the graphs).