Abstract
All nonmammalian vertebrates studied can regenerate inner ear mechanosensory receptors (i.e., hair cells) (Corwin and Cotanche, 1988; Lombarte et al., 1993; Baird et al., 1996), but mammals possess only a very limited capacity for regeneration after birth (Roberson and Rubel, 1994). As a result, mammals experience permanent deficiencies in hearing and balance once their inner ear hair cells are lost. The mechanisms of hair cell regeneration are poorly understood. Because the inner ear sensory epithelium is highly conserved in all vertebrates (Fritzsch et al., 2007), we chose to study hair cell regeneration mechanism in adult zebrafish, hoping the results would be transferrable to inducing hair cell regeneration in mammals. We defined the comprehensive network of genes involved in hair cell regeneration in the inner ear of adult zebrafish with the powerful transcriptional profiling technique digital gene expression, which leverages the power of next-generation sequencing (‘t Hoen et al., 2008). We also identified a key pathway, stat3/socs3, and demonstrated its role in promoting hair cell regeneration through stem cell activation, cell division, and differentiation. In addition, transient pharmacological inhibition of stat3 signaling accelerated hair cell regeneration without overproducing cells. Taking other published datasets into account (Sano et al., 1999; Schebesta et al., 2006; Dierssen et al., 2008; Riehle et al., 2008; Zhu et al., 2008; Qin et al., 2009), we propose that the stat3/socs3 pathway is a key response in all tissue regeneration and thus an important therapeutic target for a broad application in tissue repair and injury healing.
Introduction
The sensory epithelium of the inner ear is mainly composed of two types of cells: hair cells and supporting cells (Fritzsch et al., 2006). Inner ear hair cells are the basic mechanosensory receptors for hearing and balance (Vollrath et al., 2007), while supporting cells provide a variety of functions including being the stem cells for replacing hair cells in most vertebrates (Balak et al., 1990; Baird et al., 1996; Jones and Corwin, 1996). All nonmammalian vertebrates studied show the ability to regenerate their inner ear hair cells (Cruz et al., 1987; Corwin and Cotanche, 1988; Lombarte et al., 1993; Baird et al., 1996). However, in mammals, loss of inner ear hair cells caused by acoustic overexposure (McGill and Schuknecht, 1976), otoxic drugs (Lim, 1976) or aging (Soucek et al., 1986) is the major cause of permanent auditory and vestibular deficiencies because mammals lose regenerative ability after birth (Roberson and Rubel, 1994). Because the inner ear sensory epithelium is highly conserved in all vertebrates (Fritzsch et al., 2007), numerous studies have been done in nonmammalian vertebrates to understand the mechanism of hair cell regeneration (Brignull et al., 2009). However, our understanding of the mechanisms involved is still very limited because it is a complex, multistaged process (Stone and Cotanche, 2007).
In this study, we used the powerful profiling technique digital gene expression (DGE) (‘t Hoen et al., 2008; Morrissy et al., 2009) to study the hair cell regeneration in zebrafish at high resolution to get a more comprehensive view of the process. In zebrafish, spontaneous and damage-induced hair cell production has been demonstrated in both the inner ear (Bang et al., 2001; Higgs et al., 2002; Schuck and Smith, 2009) and the neuromasts (Harris et al., 2003), a mechanosensory structure highly similar to the sensory epithelia of the inner ear (Nicolson, 2005) and thus an excellent model for studying hair cell regeneration (Harris et al., 2003; Hernández et al., 2007; Ma et al., 2008; Behra et al., 2009). In addition, zebrafish are commonly used as a genetic/genomic model organism, making zebrafish a valuable system for studying the molecular mechanisms of hair cell regeneration in adult vertebrates in a systematic fashion.
By analyzing the expression profiles from inner ear tissues during regeneration, we identified a key pathway, stat3/socs3, and demonstrated its role in promoting regeneration. Signal transducer and activator of transcription 3 (stat3) is a transcription factor. Its downstream targets include stat3 itself and suppressor of cytokine signaling 3 (socs3) (Leonard and O'Shea, 1998). Socs3 protein antagonizes the activation of Stat3 in the cytosol as a negative feedback (Leonard and O'Shea, 1998). The self-restrictive pathway is known to be involved in the following various biological processes: cell proliferation, cell migration, immune responses, cell survival (Yoshimura, 2009; Yu et al., 2009), as well as regeneration in skin (Sano et al., 1999; Zhu et al., 2008), liver (Dierssen et al., 2008; Riehle et al., 2008), fins (Schebesta et al., 2006), and retinas (Qin et al., 2009). Comparing our data with those in other publications, we propose that the stat3/socs3 pathway is a key response in all tissue regeneration and thus a potential therapeutic target for tissue repair.
Materials and Methods
Animal husbandry.
Zebrafish were maintained under approved animal protocols as previously described (Westerfield, 2000) and in compliance with guidelines for animal care from NIH and the University of Maryland.
Noise exposure of adult zebrafish.
Adult wild-type TAB-5 (Amsterdam et al., 1999) mixed-sex zebrafish (∼1 year old) were exposed to white noise (100–10,000 Hz, 150–170 dB re 1 μPa) for 48 h at 28–29°C according to a protocol modified from Smith et al. (2006). After exposure, the fish were maintained under regular husbandry conditions until killed. The control fish were not exposed to noise.
Tag profiling and tag mapping.
The tag profiling data of inner ear tissues were generated by Illumina. Tags detected only once were discarded. The tag sequences were mapped against transcriptome and genomic sequence databases: Refseq RNAs, UniGene, Ensembl RNAs, Ensembl RNA ab initio, and the zebrafish genome (Zv.8). UniGene IDs were used as the primary index (except for those Refseq RNAs without corresponding UniGene IDs). The expression level of a certain gene (identified by a unique UniGene ID) was the summation of counts of the tags that were “unambiguously mapped” to the gene normalized by the total counts of all tags obtained in the profile in the expression unit called transcripts per million.
Analyses of expression profiling data.
The sample clustering analysis was done using Genesifter software (Geospiza). Candidate genes were identified for further analysis if they showed significant differences in expression level during regeneration compared with the control sample (i.e., ≥1.5-fold increase or ≥2-fold decrease in expression level with a p value <0.01, as determined by χ2 test or Fisher's exact test. Pathway analysis was done with MetaCore software (GeneGo).
Quantitative RT-PCR.
Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) and bactin1 were used as reference genes. The relative change in expression level of a gene X in an experimental sample compared with the control sample was calculated as 2 × (ΔCtEXPx − ΔCtCONx), where ΔCtEXPx = (CtEXPref − CtEXPx) and ΔCtCONx = (CtCONref − CtCONx). Ct is the cycle number at which amplification rises above the background threshold. All primers are listed in Table 1.
Whole-mount in situ hybridization.
Regular whole-mount in situ hybridization (ISH) of zebrafish embryos was done as previously described (Oxtoby and Jowett, 1993). Fluorescent whole-mount in situ hybridization was performed according to the regular in situ hybridization protocol, except that the TSA Cyanine 3 system (PerkinElmer) was used for developing fluorescent signals. All primers for probe synthesis are listed in Table 1.
Morpholino and mature mRNA injection.
The injections were done in fertilized eggs of TAB-5 (Amsterdam et al., 1999) or Tg(cldnb:GFP) (Haas and Gilmour, 2006) lines. Morpholinos (MOs) against stat3, socs3a, and tp53 (Table 1) were injected at a concentration of 500 μm. The mature mRNA of socs3a was injected at a concentration of 37.8 ng/μl.
Induction of lateral line hair cell death in larval zebrafish.
The 5 d postfertilization (dpf)-old larval fish were treated with 10 μm CuSO4 for 2 h at 28.5°C as previously described (Hernández et al., 2007) to induce complete lateral line hair cell loss.
Quantification of hair cell numbers and mitotic events during lateral line hair cell regeneration.
After CuSO4 treatment to induce lateral line hair cell death, 5-dpf-old Tg(pou4f3:GFP) larvae (Xiao et al., 2005) were then raised at 28.5°C in the Holtfreter's buffer containing S3I-201 (400 μm, EMD Biosciences) or DMSO (40 μl/10 ml buffer, Sigma). Lateral line hair cells (GFP-positive cells) in living larvae were quantified at 24, 48, and 72 h post-treatment (hpt). The mitotic events during lateral line hair cell regeneration were quantified by a bromodeoxyuridine (BrdU) incorporation assay (Ma et al., 2008).
Immunohistochemistry.
The primary and secondary antibodies used include the rabbit myosin VI and myosin VIIa antibodies (1:200 dilution, Proteus Biosciences), rabbit anti-stat3pS727 (1:50 dilution, Abcam), mouse anti-BrdU with Alexa Fluor 546 conjugate (1:200 dilution, Invitrogen), goat GFP antibody (FITC conjugated, 1:200 dilution, Abcam), and Alexa Fluor 568 (or 488) goat anti-rabbit IgG (1:1000 dilution, Invitrogen). The inner ear sensory epithelia were dissected as previously described (Liang and Burgess, 2009).
Microscopy and image analysis.
To capture images of the entire saccular epithelia, several overlapping pictures were taken for each epithelium with Zeiss Vision software and tiled with Photoshop 7.0 software (Adobe). The imaging in the living larvae was done using an Axiovert200M and an Apotome Grid Confocal (Zeiss). The antibody staining and fluorescent in situ hybridization results of embryos/larvae were visualized under AxiovertNLO confocal microscope (Zeiss) with Zeiss AIM software. The regular in situ hybridization results were captured using a Leica MZ16F microscope with Leica FireCam software.
Statistical analysis.
χ2 tests (genes with tag count >5) and Fisher's exact tests (genes with tag count ≤5) were used to compare the expression levels of the genes quantified by tag profiling. The calculation was done using R software. Expression data from inner ear samples collected at different time points during regeneration were compared with those from the control sample. One-way ANOVA, post hoc test, and Student's t test were used to compare hair cell numbers. The calculations were done using Excel software (Microsoft).
Results
Inner ear hair cell regeneration in adult zebrafish
We induced inner ear hair cell loss in adult zebrafish by modifying a previously published noise exposure protocol (Smith et al., 2006) (Fig. 1A). Using this setup, we could consistently induce hair cell loss in the anterior-medial region of the saccular macula (Fig. 1B), but not in the utricular or lagenar macula (data not shown), which is consistent with the proposed function of saccule as the major auditory end organ in otophysians (Fay, 1978). We confirmed complete elimination of the hair cells (as opposed to hair cell bundle damage) with phalloidin (staining hair cell bundles) and myosin VI/VIIa antibodies (staining hair cell bodies) (Fig. 1C). When we quantified the hair cells in a 20 × 20 μm area at 40% of the total length of the anterior–posterior axis of the epithelium at 24 h intervals from 0 to 120 h postexposure (hpe), we found that the hair cells regrew rapidly and returned to control levels in 96 h (Fig. 1C,D), consistent with a previous publication (Schuck and Smith, 2009).
Candidate gene/pathway identification by DGE
Based on the time line of hair cell regeneration in adult fish (Fig. 1C,D), we collected inner ear tissue samples at 0, 24, 48, and 96 hpe and generated the expression profiles from the samples using the DGE tag-profiling technique (Morrissy et al., 2009). The raw data from profiling included sequences of 944,347 unique tag sequences (20 nt) and a “count” associated with each tag sequence (Table 2). The count is a direct quantification of the copy number of the transcript that the tag represents. The total count number of all the tag sequences was 15,744,948. After filtering the raw data (see Materials and Methods), we mapped 303,342 unique tags from a total of 15,103,943 sequences to transcriptome and genome databases (Table 2). Approximately 60% of the tags were mapped to at least one known/predicted transcript or at least one location on the genome (Fig. 2A). Although ∼40% of the unique tag sequences failed to be mapped, the total proportion of unmapped tags only accounted for ∼7.2% of the total tag counts. When one mismatch was allowed during the mapping, only ∼7% of the tag sequences failed to be mapped, suggesting that the failure in mapping most likely resulted from sequencing errors and/or single-nucleotide polymorphisms. In this study, we adopted conservative criteria, only analyzing tags with no mismatch in mapping and matching to a single UniGene entry.
A total of 80,604 unique tags were unambiguously mapped to a known/predicted transcript (Fig. 2A) (i.e., the tag could be assigned to one and only one of the known/predicted transcripts). Those 80,640 tags were used for gene identification and expression level calculations. Clustering analysis of the five expression profiles with Genesifter (Geospiza) software based on the calculation of the overall differences between profiles showed that the 96 hpe profile shared the highest similarity with the control profile, followed by the 24 hpe and the 48 hpe profiles, which were highly similar to each other, and then by a 0 hpe profile that differed the most from the control profile (Fig. 2C, top). Essentially the transcriptional profiles were most different immediately after noise exposure and slowly returned to normal over the course of 4 d.
We used χ2 tests and Fisher's exact tests to compare the expression level of each gene during regeneration to the control level. A total of 2269 genes with significant changes in their expression levels at one or more time points during regeneration were identified as candidate genes (data available upon request). Measureable changes in expression levels detected by DGE were confirmed in several candidate genes (selected because of already known functions in hair cell development or because of direct interest in the genes) with quantitative RT-PCR (qRT-PCR) (Fig. 2B). We performed a pathway analysis for the 2269 candidate genes using the Metacore (GeneGo) software package. The functions of the most enriched interaction network calculated from the candidate genes in each experimental time point are listed in Figure 2C (bottom). While the enriched pathways identified by Metacore were generally broad, the fact that the different collected time points had very different enrichment profiles, indicated distinct phases in the regeneration process. The results of profile distance calculation and enrichment analysis corresponded nicely with the distinct morphological changes in the saccular epithelium during regeneration, as well as with the known cellular mechanisms during hair regeneration in other nonmammalian vertebrates (Stone and Cotanche, 2007).
Because we were interested in identifying early signaling events that trigger regeneration, we prioritized our candidates for functional analysis from the highly enriched networks that were apparent at the earliest measured time point (0 hpe, Fig. 2D). Metacore and Pathway Studio both identified a large interconnected pathway with the hub of that pathway centering on stat3 (Fig. 2D). Confirming the potential importance, the gene socs3a (a zebrafish homolog of human SOCS3), a downstream target of stat3 (and a negative-feedback inhibitor for stat3 activity) was one of the largest transcriptional changes at 0 hpe (≈10-fold increase). By 24 hpe, stat3 and socs3a signaling return to normal levels, suggesting that this response is an important early regenerative response but is not necessary in later stages of regeneration. With the help of the analysis, we identified the stat3/socs3 pathway as the dominant signaling pathway to become activated at the earliest time point of recovery with >50 relevant genes in the pathway responding to hair cell damage (Fig. 2D). No other known signaling pathway had as many connected nodes as the stat3/socs3a node in the 0 hpe responses. Related genes in the stat3/socs3 pathway, such as socs3b (a paralog of socs3a in zebrafish), Janus kinase 1 (jak1), and matrix metalloproteinase 9 (mmp9), also showed significant changes in their expression levels (Fig. 2B; data available on request).
Self-restrictive feedback between stat3 and socs3a in zebrafish
In mammals, there is a self-restrictive feedback mechanism between stat3 and socs3: stat3 activates the transcription of socs3, which in turn, antagonizes stat3 activation (Leonard and O'Shea, 1998). To test whether there is a similar feedback mechanism between stat3 and socs3a in zebrafish, we temporarily knocked down stat3 or socs3a in zebrafish embryos using MO injection. We detected a decrease in the mRNA levels of both stat3 and socs3a in stat3 morphants but an increase in the mRNA levels of both genes in socs3a morphants (Table 3). This is the expected result for a negative-feedback loop. To further confirm the knock-down studies, we injected mature mRNA of socs3a to zebrafish embryos and found a reduction in stat3 expression level, mimicking the effects of stat3-MO injection (Table 3). These data demonstrate that the self-restrictive feedback relation between stat3 and socs3a in mammals is maintained in zebrafish.
The stat3/socs3a pathway is involved in hair cell production in zebrafish development
In addition to the hair cells in their inner ears, zebrafish, like other fishes and amphibians, also possess the mechanosensory lateral line organ for detecting water movement over the body (Montgomery et al., 2000; McHenry et al., 2009). The neuromasts in the lateral line are composed of hair cells and supporting cells that are highly similar to those in the inner ear sensory epithelium (Nicolson, 2005). In addition, the hair cell regeneration in the lateral line neuromasts shares a similar molecular mechanism to that in the inner ear (Harris et al., 2003; Ma et al., 2008; Behra et al., 2009).
Both stat3 (Oates et al., 1999; Thisse and Thisse, 2004) and socs3a (Fig. 3A) are expressed in the neuromasts at 5 dpf, implying a normal function in those tissues for either development or homeostasis. In addition, stat3 is known to be expressed in the anterior region of the otic vesicle at earlier developmental stages (Oates et al., 1999; Thisse and Thisse, 2004). As an initial test of the roles for stat3 and socs3 in hair cell regeneration, we tested the effects of morpholino inhibition of each gene on normal inner ear development. Testing two different morpholinos for socs3a and a previously published and verified one for stat3 (Yamashita et al., 2002), knocking down stat3 or socs3a had a clear inhibitory effect on hair cell production (Fig. 3B–E). In stat3 morphants, the anterior otolith in the otic vesicles was often missing while the posterior otolith appeared approximately normal (Fig. 3B). Correspondingly, a closer examination of the maculae in the otic vesicle revealed a significantly decreased number of hair cells in the anterior macula in those morphants while the hair cell number in the posterior macula remained normal (Fig. 3C,D). Coinjection with tp53 MO resulted in an identical phenotype (data not shown), suggesting that the hair cell-related phenotypes are gene specific. The phenotypes of the morphants corresponded nicely with the expression pattern of both genes detected by in situ hybridization (Oates et al., 1999; Thisse and Thisse, 2004) (Fig. 3A). The phenotypes also serve as an indirect confirmation of the involvement stat3/socs3 pathway in hair cell regeneration, as hair cell production during development and regeneration will, to some extent, share similar genetic programs (Stone and Rubel, 1999; Cafaro et al., 2007; Ma et al., 2008; Daudet et al., 2009).
To further understand the function of the stat3/socs3 pathway in hair cell production, we measured the mRNA level of atonal homolog 1a (atoh1a) in stat3 and socs3a morphants. atoh1a is one of the zebrafish homologs of atoh1, an essential gene required for hair cell fate commitment (Bermingham et al., 1999; Woods et al., 2004; Millimaki et al., 2007). Both the knocking down of stat3 and the overexpression of socs3a resulted in a reduction in atoh1a mRNA level (Table 3) but did not change the overall distribution or timing of expression based on in situ hybridization (data not shown). In contrast, a strong increase in atoh1a mRNA level was observed in socs3a morphants where stat3 was hyperactivated (Table 3), demonstrating that stat3/socs3a are positive/negative regulators of atoh1a expression. In addition, the cross talk between stat3/socs3 and atoh1 was further confirmed by in situ hybridization of the socs3a morphants (Fig. 3E). At 24 and 36 hpf, there is little detectable difference in atoh1a expression between socs3a morphants and wild-type fish by in situ hybridization (data not shown). But at 48 and 72 hpf, there is a clear expansion of atoh1a expression in the otic vesicle (Fig. 3E, arrowhead), the hindbrain, and the spinal chord (Fig. 3E, bracket), suggesting the stat3–atoh1a regulatory relationship is not unique to the sensory epithelia of the inner ear and lateral line, but is a more general pathway for regulation. These results suggest that the stat3/socs3 pathway is involved in the hair cell differentiation process possibly by directly or indirectly expanding atoh1a expression.
Stat3 activation during hair cell regeneration and development in lateral line neuromasts
Because the lateral line neuromasts are located on the surface of the body, it makes the lateral line a convenient model for testing hair cell regeneration. Expression of both stat3 (Thisse and Thisse, 2004) and socs3a (Fig. 3A) was detected in the neuromasts at 5 dpf, so we continued with additional functional studies of the stat3/socs3a pathway during regeneration in larval neuromasts. At 5 dpf, the number of hair cells in the lateral line neuromasts reaches a relatively stable level (Ma et al., 2008; Behra et al., 2009), which enables us to differentiate regeneration of hair cells from normal growth and development.
We induced hair cell death in the lateral line neuromasts with CuSO4, a protocol ensuring complete lateral line hair cell elimination within 2 h in 5 dpf larvae (Hernández et al., 2007), which is followed subsequently by regeneration of hair cells to control levels within 72 h (Behra et al., 2009). We performed in situ hybridization in larval fish after 2 h of CuSO4 treatment and found an increase in the level of stat3 expression in the lateral line neuromasts when hair cell death was induced (Fig. 4A). In addition, we also found changes in the location of stat3 expression induced by CuSO4-triggered hair cell death (Fig. 4A), which was further confirmed by fluorescent whole-mount in situ hybridization in ET20 larvae that marks the mantle cells of the neuromasts with GFP expression the outermost layer of nonsensory cells in the neuromast (Parinov et al., 2004) (Fig. 4B). RNA probes against stat3 were mostly detected in the GFP-positive cells of the neuromasts in untreated ET20 larvae (Fig. 4B, top). When hair cell death was induced, stat3 expression was no longer limited to the mantle cells, but was expressed strongly throughout the supporting cells of the neuromasts. The in situ hybridization data confirmed the activation of stat3 induced by hair cell death in the lateral line neuromasts was in the supporting cell population, matching expression increases observed by DGE during inner ear hair cell regeneration.
The activation of mammalian stat3 protein is mostly characterized by the phosphorylation of either Y705 (Yoshimura, 2009; Yu et al., 2009) or S727, which functions concurrently with Y705 (Wen et al., 1995) or independently (Liu et al., 2003; Qin et al., 2008). The corresponding amino acid (S751) and the flanking amino acid sequences are highly conserved in zebrafish stat3 when compared with the human protein, so we first tracked the activation of stat3 in the neuromasts after CuSO4 treatment with an antibody that recognizes STAT3 S727 phosphorylation (STAT3pS727). We found a significant increase in STAT3pS727-positive nuclei in the neuromasts at 12 hpt with CuSO4 in the CuSO4-treated larvae relative to control (Fig. 4A,B). Double labeling with GFP in Tg(scm1:GFP) (Behra et al., 2009) demonstrates that the cells with nuclear staining of STAT3pS727 are the supporting cells (Fig. 4A). The activation and the nuclear import of stat3 protein in the supporting cells subsequent to hair cell death indicate the activation of stat3 signaling during hair cell regeneration.
Hair cell production during development and regeneration will, to some extent, share similar genetic programs (Stone and Rubel, 1999; Cafaro et al., 2007; Ma et al., 2008; Daudet et al., 2009), so we checked whether stat3 activation in the lateral line neuromasts is similar during development and regeneration. In 3 dpf larvae, most cells in the neuromasts showed homogeneous STAT3pS727 labeling only in their nuclei (Fig. 5A′, top). Double-labeling with scm1:GFP demonstrates that those cells are supporting cells (Fig. 5A′, top). In addition, in differentiating hair cells, the nuclear labeling of STAT3pS727 antibody decreased and became spotty, with cytoplasmic labeling becoming predominant, suggesting that the stat3 was becoming inactive in these cells (Fig. 5A′, top, close-ups). The nuclear import of stat3 in supporting cells during hair cell regeneration in 5 dpf and older larvae seems to be recapitulating the stat3 activation during immature neuromast development in younger larvae.
It is noteworthy that in older larvae, a significant level of labeling was still detectable in a subgroup of mantle cells that are marked by GFP in the ET20 transgenic line (Fig. 5A′, bottom). The significance of this cell population having activated stat3 is not yet clear, but it seems to represent a distinct cellular compartment in the lateral line neuromasts.
Inhibition of stat3-promoted hair cell regeneration
To further clarify the function of stat3, we compared the hair cell regeneration processes with and without the presence of a stat3 inhibitor. The majority of the chemical inhibitors were ototoxic and could not be used for regeneration studies (data not shown). One published inhibitor, S3I-201, was not toxic on its own and was used to test the role of stat3 in hair cell regeneration. S3I-201 is a cell-permeable chemical that binds to the SH2 domain of mammalian stat3 protein and reportedly blocks the dimerization of phosphorylated (activated) stat3 molecules (Siddiquee et al., 2007). After exposing Tg(pou4f3:GFP) larvae (in which the hair cells are labeled with GFP) (Xiao et al., 2005) to CuSO4, we quantified the regenerating hair cells and performed a BrdU incorporation assay (Ma et al., 2008). The S3I-201-treated larvae showed a significant increase in BrdU incorporation in the neuromasts compared with DMSO-treated larvae (control) at 24 hpt, but not at 48 or 72 hpt (Fig. 6A). Accordingly, the S3I-201-treated larvae had more hair cells per neuromast at 48 hpt, but not at 24 or 72 hpt (Fig. 6A), which is consistent with an expected lag between supporting cell division and subsequent hair cell differentiation. In essence, the hair cells regenerated faster in the S3I-201-treated embryos than in controls. Similar results were observed when hair cell quantification was done by myosin VI antibody staining (data not shown). S3I-201 did not have a significant impact on hair cell numbers in larvae that had not been treated with CuSO4 (data not shown).
To confirm the inhibitory function of S3I-201 in hair cell regeneration triggered by CuSO4 ototoxicity, we performed ISH to examine the mRNA levels of stat3 and socs3a in the lateral line neuromasts after 1 h of CuSO4 treatment with or without the presence of S3I-201. The ISH data demonstrated a reduction in both stat3 (Fig. 6B) and socs3a (Fig. 6B′) expression in the neuromasts cotreated with both CuSO4 and S3I-201 compared with those treated only with CuSO4. Our qRT-PCR results were in accordance with the ISH results in that stat3 expression in larvae treated with both CuSO4 and S3I-201 was only 81% of the level in larvae treated only with CuSO4 (p = 0.0014, n = 3 for both groups). Since stat3 activates its own expression, the data suggest that S3I-201 stimulated hair cell regeneration by moderately inhibiting stat3 activity.
Discussion
Recent studies have demonstrated the power of using the DGE platform for measuring transcriptional changes (‘t Hoen et al., 2008; Morrissy et al., 2009). Using DGE, we generated a comprehensive list of candidate genes involved in the inner ear hair cell regeneration in adult zebrafish (data available upon request). One major pathway that is initiated in the earliest responses centers on stat3 and socs3a. This pathway is broadly implicated in the initial signaling responses of hair cell regeneration in the zebrafish inner ear. We went on to show that stat3/socs3 signaling played an important role in cell division and hair cell differentiation during both normal development and tissue regeneration.
Stat3 signaling: activated during development and regeneration
To some extent, hair cell production during both development and regeneration share similar genetic programs (Stone and Rubel, 1999; Cafaro et al., 2007; Ma et al., 2008; Daudet et al., 2009). In this study, we identified stat3 and socs3a as important regulators in zebrafish hair cell regeneration in the inner ear (data available upon request) and lateral line system (Fig. 5A). Correspondingly, we also found them required for normal hair cell production during the larval development (Fig. 3B–E). The examination of stat3 activity in lateral line neuromasts during development and regeneration further confirms the similarity shared between the two processes. Activated, nuclear-localized stat3 protein was detected in a large number of supporting cells and differentiating hair cells in 3 dpf larvae (Fig. 5A′) and was downregulated in those same cells in the neuromasts of more mature 5 dpf larvae (Fig. 5A). However, when hair cell death was triggered in 5 dpf larvae, the stat3 protein was again activated and imported into the nuclei of the supporting cells (Fig. 5A), recapitulating the active stat3 observed during neuromast development (Fig. 5A′).
Another interesting observation is the specific overlap of STAT3pS727-positive cells and the GFP-positive mantle cells of ET20 larvae (Fig. 5A′), which is further supported by the enrichment of stat3 mRNA in those GFP-positive cells (Fig. 4A). This suggests that the mantle cells have a specific function in the neuromast that requires activated stat3. Previous publications suggest two types of stem cell-like cells in the neuromasts, one type involved in long-term cell replacement (i.e., the true stem cells), and the other in an amplification/differentiation role (Williams and Holder, 2000; Ma et al., 2008). The role of the ET20:GFP-labeled (and stat3-activated) mantle cells may regulate the cell division of adjacent stem cells through a nonautonomous mechanism similar to the role of stat92E in escort cells for regulating the stem cell division in the Drosophila ovary (Decotto and Spradling, 2005). These ET20:GFP and STAT3pS721-positive cells would represent part of the niche environment necessary to promote normal cell turnover as well as tissue regeneration. Alternatively, the mantle cells might directly divide asymmetrically to repopulate the supporting cells of the neuromasts after damage or cell turnover. The triggering of cell differentiation seemed to be achieved, at least in part, through activation of atoh1 (Fig. 3E), a key regulator of hair cell differentiation (Fig. 7). This pathway seems to be used in more than one tissue type as socs3a inhibition results in an expansion of atoh1a expression in both the otic vesicle and in the brain. It is noteworthy that while atoh1a expression is expanded in the otic vesicle with a hyperactive stat3, the number of hair cells was not (data not shown), suggesting that more factors are necessary for hair cell differentiation.
A previous microarray study did show basal-level expression of stat3 in newborn rat cochlea that could be elevated by stress in explant cultures (Gross et al., 2008). It was also reported that stat3 showed a temporal expression pattern similar to gata2 and C/EBP during the in vitro differentiation of conditional cell lines derived from mouse otocyst (Holley et al., 2007). Little is known about the functions of stat3 and socs3 in the inner ear in mammals other than that activated stat3 protein is detected in the outer hair cells in mouse cochlea late in embryonic development (Hertzano et al., 2004). The persistence of stat3 expression in a subset of supporting cells in zebrafish that is not detected in the mammalian cochlea might represent a key difference that enables nonmammalian vertebrates to regenerate hair cells. A more detailed comparison study focusing on the stat3/socs3 signaling pathway in the inner ears in mice and zebrafish would help us further understand key regulators of hair cell regeneration. In addition, it is also intriguing to explore whether cell cycle progression can be triggered in supporting cells by genetic and/or pharmacological enhancement of stat3 activity in damaged mammalian sensory epithelia.
The stat3/socs3 pathway may serve as a common initiator in a variety of regenerative processes
The stat3/socs3 pathway is best known as the mediator of a wide variety of biological processes, including cell proliferation, cell migration, immune response, and cell survival (Yoshimura, 2009; Yu et al., 2009). Studies in mouse skin (Sano et al., 1999; Zhu et al., 2008) and liver (Dierssen et al., 2008; Riehle et al., 2008) further confirm the importance of the stat3/socs3 pathway for proper wound healing. For example, in microarray datasets from other zebrafish regeneration experiments, a strong induction of socs3 expression was consistently observed during the regeneration of fins (Schebesta et al., 2006) and retinas (Qin et al., 2009). stat3 expression was also found to be upregulated in dividing cells during zebrafish retinal photoreceptor regeneration (Kassen et al., 2009). The early activation of stat3 in the supporting cells subsequent to hair cell death is consistent with these previous studies and suggests that this is a common pathway for healing. stat–jak signaling has been found to play a critical role in the development and regeneration of the epithelial cells in the Drosophila midgut by initiating cell division and inducing differentiation (Jiang et al., 2009), a dual process that is highly similar to what we demonstrated during regeneration of the inner ear sensory epithelia. In zebrafish inner ear and lateral line regeneration, the activation of stat3 signaling in supporting cells precedes the cell division, suggesting that the gene is playing a role in cell cycle progression of those cells. A recent study supports this idea as it reported leukemia inhibitory factor promoted cell division in a cochlea-derived cell line, HEI-OC1, through the stat3/jak2 signaling pathway (Chen et al., 2010). This suggests that stat signaling is a fundamental initiating mechanism of tissue regeneration that operates through cell division and differentiation and has been conserved for over the 300 million years of evolution from invertebrates to vertebrates. Together, all these data strengthen our hypothesis that stat3/socs3 is a central activating mechanism of stem cell populations in most, if not all, forms of regeneration.
Although stat3 activation can be considered as a promoting factor for a regenerative process, we showed that a mild inhibition of stat3 via a chemical antagonist (Fig. 6B,B′) could accelerate hair cell regeneration by promoting supporting cell division (Fig. 6A). Because stat3 is also known to be required for the maintenance of pluripotency of various types of stem cells (Raz et al., 1999), and we did identify a small group of putative stem cells in the neuromasts with persisting stat3 activity under quiescent/undamaged conditions (Fig. 5A′), a possible explanation for the observation that inhibiting stat3 modestly enhanced regeneration at early time points is that S3I-201 forced the premature division and differentiation of these stem cells, resulting in an increase in BrdU incorporation at 24 hpt and in hair cell number at 48 hpt (Fig. 6A). However, we do not believe such a mechanism contributed much, if any, to the difference between S3I-201-treated larvae and control larvae during regeneration. First, we observed an accelerated regeneration without overproduction or underproduction of hair cells (Fig. 6A), suggesting that we were simply accelerating the normal process. Meanwhile, undamaged larvae raised with S3I-201 from 5 to 8 dpf did not show any significant difference in the number of the lateral hair cells compared with control, showing it was not the release of a cell cycle block in the mantle (ET20-positive) cells that explains the accelerated appearance of the hair cells. In addition, the damage-induced stat3 activation was observed in all of the supporting cells (Fig. 4B), the main source of new hair cells during regeneration (Ma et al., 2008). While we cannot completely rule out the contribution of any other mechanism, the data here do support our hypothesis that S3I-201 promoted the division of the supporting cells through mild downregulation of the damage-induced stat3 activity in those cells. In essence, S3I-201 increased the rate of stat3 inactivation that naturally occurs through socs3a activation, allowing the cells to proceed to mitosis and cell differentiation sooner.
We believe that our results emphasize the balance of stat3/socs3 signaling as being crucial for a successful healing process. Two examples in the literature demonstrated that excessive stat3 activity can block cell cycle progression and inhibit regeneration similar to what we observed; prolonged overexpression of IL-6 and the resulting hyperactivation of stat3 inhibited regeneration after hepatectomy in mice (Wüstefeld et al., 2000; Jin et al., 2006). Moreover, excessive stat3 activity in fatty livers was associated with the cell cycle arrest after hepatectomy (Torbenson et al., 2002). Accordingly, keeping the cytokine signaling tempered by the negative feedback of socs3 is known to be crucial for proper regeneration (Taub, 2004). Tissue-specific knockout of socs3 in mice led to an upregulation in stat3 activity but delayed skin wound healing due to the increased neutrophil and macrophage infiltration and elevated chemokine expression (Zhu et al., 2008). In addition, we are not ruling out the possibility that stat3 may play more than one role in more than one type of cell in tissue regeneration. It is known that stat3 is activated in hepatocytes as well as myeloid cells after hepatectomy (Wang et al., 2010). However, while knocking out stat3 in hepatocytes inhibits cell cycle progression and regeneration, knocking out stat3 in myeloid cells leads to the opposite results (Wang et al., 2010). Previous data also suggest a complicated cross talk of stat3 signaling pathway in different cell types during regeneration (Wang et al., 2010). Given the known active involvement of myeloid cells in lateral line hair cell death in zebrafish larvae (d'Alençon et al., 2010), it is very likely that S3I-201 treatment disturbed the stat3 signaling in not only supporting cells but also myeloid cells and even the cross talk between these two types of cells. Our data together with data from many others demonstrate crucial yet complicated roles of the stat3/socs3 pathway during tissue regeneration. We have developed a model for stat3 that requires initial activation of stat3 with a critical negative-feedback role for socs3a. Further studies are necessary to work out the exact roles of each gene in this complex and dynamic process. While stat3 activation could potentially be therapeutic to injuries in a wide variety of tissues, caution needs to be taken for such application, given the multifaceted functions of stat3 in the injury and healing process.
Footnotes
This research was supported by the Intramural Research Program of the National Human Genome Research Institute, National Institutes of Health (S.M.B.). We thank A. Popper for use of the sound exposure equipment and helpful discussions, and D. Bodine and W. Pavan for critical reading of the manuscript.
The authors declare no competing financial interests.
- Correspondence should be addressed to Shawn M. Burgess at the above address. burgess{at}mail.nih.gov
This article is freely available online through the J Neurosci Open Choice option.