Compartmentalized Notch signaling sustains epithelial mirror symmetry

The three-dimensional organization of tissues is essential for the efficient function of organs. It must also be maintained during the entire life of the individual and be recovered during organ repair because its loss can generate devastating pathologies (Wodarz and Nathke, 2007; Zallen, 2007). The pervasive planar cell polarity has emerged as an architectural property of tissues that allows investigations of the link between form and function (Axelrod, 2009; Strutt and Strutt, 2009). One group of organs that relies on planar cell polarity for coherent sensory function is the acusticolateralis system that comprises the inner ear and lateral line, the shared plane-polarized elements of which are the mechanosensory hair cells (Lewis and Davies, 2002; Rida and Chen, 2009). The planar polarization of the hair cells allows animals to detect and interpret the direction of propagation of a sound (Hudspeth, 1985). Hair cells are substantially similar in their development and physiology across species. However, although their loss in mammals is irreversible leading to permanent deafness, other vertebrates are endowed with a hair-cell regenerative capacity during their entire lives (Corwin and Cotanche, 1988; Ryals and Rubel, 1988; López-Schier, 2004; Collado et al., 2008). During organ repair, cell fate and tissue architecture are often acquired concurrently to allow functional recovery before full anatomical repair, which may be essential for organs on which animals depend for survival. In the sensory organs of the zebrafish lateral line, called neuromasts, the complete loss of hair cells triggers a rapid and precise regeneration process (Williams and Holder, 2000; López-Schier and Hudspeth, 2006; Brignull et al., 2009), during which cell fate and epithelial planar polarity are recovered progressively along three phases (Fig. 1A). Phase I commences when hair cells begin to regenerate. During this phase, the central part of the neuromast epithelium (the macula) becomes increasingly oval because it elongates along a single axis (Fig. 1C,D) (López-Schier and Hudspeth, 2006). Phase II begins when the neuromast macula expands symmetrically to regain its circular shape. During this phase epithelial planar polarity propagates laterally (Fig. 1A,E). Phases I and II last circa 30 hours each. A Phase III of homeostasis begins after around 60 hours from the onset of regeneration. During Phase I, the neuromast becomes bilaterally symmetric because a vertical midline separates the epithelium in two halves. Each half contains hair cells of identical orientation but opposite to that of the confronting half (Fig. 1D). This occurs because three processes integrate with remarkable spatiotemporal precision. First, hair cells regenerate in pairs and their soma localize adjacent to each other along the direction of epithelial planar polarity. Second, each hair cell of the pair orients opposite to its sibling within the plane of the macula. Third, all the hair cells at each side of the midline orient identically. It is not known how these processes integrate to generate bilateral symmetry. In the present study we use high-resolution live imaging in transgenic zebrafish, genetic and pharmacological perturbations of regeneration, and fluorescent sensors to address this question.

Zebrafish used were maintained under standardized conditions and experiments were performed in accordance with protocols approved by the PRBB Ethical Committee of Animal Experimentation. To show different cell populations in the zebrafish neuromasts, SqET4 and SqET20 were used. For time-lapse imaging SqET4 fish were crossed to Tg[CldB:lynGFP] (Haas and Gilmour, 2006) to generate double transgenic larvae. The Tg[atoh1a:tdTomato] line was constructed from zC247L22 BAC and dTomato cDNA, and was used to analyze the expression profile of Atoh1a during regeneration. The Tp1[bglob:hmgb1-mCherry] line (Parsons et al., 2009) was used to report the Notch activity. Tg[hsp70:gal4;UAS:Nicd-myc] line, a gift from P. Chapouton (Helmholtz, Zentrum, München, Germany), was used to express a constitutively active form of Notch by heat shock during regeneration.

Selected larvae for fixation and live imaging were treated in E3 medium containing 250 μM neomycin for 1 hour at room temperature. After treatments, larvae were washed several times using small diameter strainer to wash off residual neomycin. Larvae were allowed to recover from neomycin treatment for couple of hours before live imaging and treatments with other drugs to avoid high mortality.

A 10 mM 5-bromo-2′-deoxyuridine (Sigma, St Louis, MO, USA) stock solution in DMSO was diluted to 10 μM in E3 medium and used to soak SqET4 larvae. Larvae were allowed to swim in this solution for 8, 24 or 48 hours, and fixed in 4% paraformaldehyde (PFA) overnight at 4°C, and then processed for immunostaining. N-[N-(3,5-Difluorophenacetyl)-L-alanyl]-S-phenylglycine t-butyl ester (DAPT, Sigma) in DMSO were used at the final concentration of 50-100 μM in E3 medium to inhibit the γ-secretase. Larvae were allowed to swim in this solution for desired time points at 28°C. As a control, larvae were treated with 1% DMSO.

For immunohistochemistry and fluorescent in situ hybridization, larvae were fixed in 4% PFA overnight at 4°C and washed several times with 0.1% Tween-20-containing phosphate-buffered saline (0.1% PBSTw). For immunohistochemistry, larvae were immediately blocked in 10% bovine serum albumin (BSA) for at least 2 hours. For fluorescent in situ hybridization, samples were washed in gradual 25-100% PBSTw:methanol and stored at −20°C for at least 1 day. Primary antibodies and monoclonal antibodies were used at the following dilutions: mouse monoclonal antibody anti-BrdU, 1:100 (Upstate/Millipore, Billerica, MA, USA); mouse monoclonal anti-cMyc, 1:150 (NewMarkers, Fremont, CA, USA); rabbit anti-claudinB, 1:500; mouse anti-HCS1, 1:100; and rabbit anti-parvalbumin 3, 1/2000. Texas Red-labeled donkey anti-mouse and -rabbit and Cy5-labeled donkey anti-mouse and -rabbit immunoglobulin secondary antibodies (Jackson ImmunoResearch, West Grove, PA, USA) were used at 1/150. Antisense digoxigenin- and fluorescein-labeled riboprobes were synthesized according to manufacturer's instructions (Roche) by using T7/SP6/T3 RNA Polymerases. Anti-DIG and -fluorescein POD antibodies (Roche) and Tyramide Signal Amplification (TSA, PerkinElmer, Waltham, MA, USA) were used to detect antisense riboprobes. Probes used were notch3a, deltaA and atoh1a. For vital labeling of hair cells, larvae were immersed in 5 μg/ml of DiASP (Invitrogen, Carlsbad, CA, USA) in E3 medium for 2 minutes at room temperature and then washed several times to remove the excess of dye. For phalloidin staining, samples were fixed in 4% PFA overnight at 4°C then washed several times in 0.1% PBSTw and incubated in phalloidin-Alexa 568 or Alexa 488 (Invitrogen) diluted 1:20 in 0.1% PBSTw overnight at 4°C. Samples were washed several times in 0.1% PBSTw before being mounted.

Fluorescent images were obtained using either a Leica SP5 or SPE or Andor Spinning disk using a 40× oil immersion objective. All images were processed using ImageJ software package. For time-lapse video-microscopy, larvae were anesthetized in 0.02% tricaine (ethyl 3-aminobenzoate methanesulfonate salt, A5040, Sigma) and mounted in 1% low melting point agarose (Agarose, low melting point, A9414, Sigma) in E3 medium. Samples were imaged using either Leica SPE Leica SP5 or Andor spinning disk using a 40× oil immersion objective. Embryos were maintained at 21-25°C and z-stacks were collected at 2- to 10-minute intervals. Fixed samples were mounted in 0.1% PBSTw with Vectashield with DAPI (1/100, Vector Laboratories, Burlingame, CA, USA) (Faucherre et al., 2009).

Compounds for cell cycle inhibitors were used as described before (Murphy et al., 2006). Larvae were treated for the desired period with aphidicolin at 100 μM, genistein at 25 μM, nocodazole at 150 μM or colchicine at 1 mM in E3 medium supplemented with 1% DMSO.

Quantifications of BrdU-labeled cells and hair cells in whole-mounts were carried out from 10 larval-stage samples and seven neuromasts per sample. Hair cells were counted both based on GFP marker of SqET4 and apical part each hair cell as an indicator of mature hair cell. Anti-BrdU antibody was used to detect cells or hair cells incorporated BrdU. Proliferation of hair cells that appeared with or without regeneration/neomycin was calculated according to this formula: BrdU-labeled hair-cell nuclei/total number of hair cell nuclei. Total number of cells in neuromasts was calculated from DAPI staining of the nuclei. All data of BrdU incorporation are represented as percentage of the standard deviation. For experiments with the presence of DAPT, averages for controls and treated larvae were compared within each group. The same statistical analysis was used to compare between groups of controls and cell-cycle inhibitor-treated larvae. To compare the distribution of hair cell polarities in the macula of regenerating and mature neuromasts, we performed, for each condition, a one-way ANOVA that tests the null hypothesis that samples in two or more groups are drawn from the same population. For each polarity we assumed the null hypothesis that polarity distribution is homogeneous in the neuromast, comparing distributions between the ventral and dorsal halves of a neuromast and between the rostral and caudal halves.

Previous work has revealed that the mature neuromast is formed by the mechanosensory hair cells and by two types of supporting cells: peripheral mantle cells and central sustentacular cells (Chezar, 1930; Ghysen and Dambly-Chaudière, 2007). Lateralis hair cells regenerate in pairs and sequentially on the dorsal or ventral aspect of parallel neuromasts (Fig. 1B) (López-Schier and Hudspeth, 2006). The spatiotemporal precision of the regenerative process suggested further heterogeneity in the neuromast. We decided to interrogate cellular identity in mature organs at 7 days post fertilization (7 dpf) by characterizing transgenic lines that express the green-fluorescent protein (GFP) in the lateral line (Parinov et al., 2004; López-Schier and Hudspeth, 2006). The SqET20 line expresses the GFP in supporting cells (Hernández et al., 2007). The staining of SqET20 animals with an antibody to the pan-supporting cell marker Claudin B revealed that GFP expression was heterogeneous, showing areas with low levels of green fluorescence located at both poles of a line perpendicular to the axis of planar cell polarity of the neuromast (Fig. 1F). Strikingly, this line shifts through 90° in perpendicular neuromasts to remain perpendicular to the direction of epithelial planar polarity (Fig. 1H). The GFP^low areas in SqET20 also coincide with those expressing high levels of alkaline phosphatase (Fig. 1G) (Ghysen and Dambly-Chaudière, 2007; Villablanca et al., 2006). These SqET20^GFP-low/AP⁺ areas demonstrate previously unknown compartments located at the poles of the midline of bilateral symmetry. We call them ‘polar compartments’. We have previously shown that the SqET4 transgenic line expresses high levels of GFP in postmitotic hair cells, as revealed by their incorporation of the fluorescent dye DiASP, which permeates through the mechanotransduction channels (López-Schier and Hudspeth, 2006). Every DiASP(+) cell was also GFP(+), excluding the possibility that SqET4 reveals only a subset of the hair cells in the neuromast. In addition to hair cells, live imaging of SqET4 animals also identified individual cells expressing lower levels of GFP (Fig. 1B,C) (López-Schier and Hudspeth, 2006). Live imaging showed that SqET4^GFP-low cells invariably undergo a single mitosis to produce a pair of hair cells (Fig. 1B) (López-Schier and Hudspeth, 2006). SqET4^GFP-low cells never generated cell types other than hair cells, suggesting that they are ‘unipotent hair-cell progenitors’ (UHCPs). From these results, we conclude that the neuromast is compartmentalized and that it contains at least five distinct cell populations: mantle cells, sustentacular cells, polar-compartment cells, transient UHCPs and the hair cells.

It has been previously reported that mitotically active supporting cells are the main contributors to hair-cell regeneration in the zebrafish lateral line (Ma et al., 2008). However, other data suggested a role of supporting-cell transdifferentiation for the regeneration of hair cells (Hernández et al., 2007). Therefore, the relative contribution of supporting-cell proliferation versus transdifferentiation remains unknown. We decided to use the SqET4 line to resolve this issue. We started by treating 5 dpf SqET4 animals with 250 μM of neomycin for 1 hour. We have previously demonstrated that this regime eliminates the hair cells of the lateral line (López-Schier and Hudspeth, 2006). Neuromasts achieve near complete regeneration 48 hours after hair-cell ablation. After neomycin treatment, regenerating animals were left to recover in water containing the DNA synthesis marker bromo-deoxy-uridine (BrdU) (Fig. 2A). We found that 59% of the hair cells were BrdU(+) at 48 hours post treatment (hpt) (Fig. 2B,C), suggesting that proliferation and transdifferentiation could contribute to hair-cell regeneration. However, neomycin cannot kill young non-mechanoreceptive hair cells because this aminoglycoside antibiotic permeates through the mechanotranduction channels. Thus, the presence of immature neomycin-resistant hair cells in the neuromast could explain why just over half of the hair cells were BrdU(+) at 48 hpt. Another explanation for this result is that hair-cell precursors remain arrested in a stage of the cell cycle beyond the S phase, and would progress to mitosis upon hair-cell loss without incorporating BrdU. To discriminate between these possibilities, we exposed fish to BrdU for 8 hours at 4 dpf before neomycin treatment at 5 dpf, in order to incorporate the BrdU in UHCPs and non-mechanoreceptive hair cells that may be present in the neuromast (Fig. 2D). These animals were also left to recover for 48 hours, again in the presence of BrdU. Under this condition, almost 100% of the hair cells were BrdU(+) at 48 hpt (Fig. 2E,F). This result strongly suggests that cell division is the main or the obligatory process underlying hair-cell regeneration.

To further test the extent to which mitotic activity is essential for hair-cell regeneration, we ablated hair cells in SqET4 animals with neomycin, and let them to recover in the presence of several inhibitors of the cell cycle (Fig. 2G,J). Treatments with nocodazole, aphidicolin, genistein and colchicine revealed that genistein and colchicine were the most effective suppressors of regeneration (Fig. 2H). However, control animals showed that genistein and colchicine were toxic to hair cells (Fig. 2G). Aphidicolin and nocodazole, by contrast, did not affect the viability of hair cells over a 48-hour period (Fig. 2G,J). Therefore, the nearly complete lack of hair-cell regenerates in genistein- and colchicine-treated animals may represent a compound effect of these drugs on UHCPs mitoses and hair-cell viability. This assertion is further supported by the progressive reduction of the number of hair cells at 48 hpt compared with 24 hpt in animals not treated with neomycin (Fig. 2G). Aphidicolin and nocodazole strongly suppressed hair-cell regeneration but did not completely prevent the production of GFP(+) cells. In control animals not treated with neomycin, the number of GFP(+) cells in neuromasts did not vary between 0, 24 and 48 hpt under aphidicolin and nocodazole treatment (Fig. 2G). Thus, the few GFP(+) cells present in inhibitor-treated neuromasts are likely to be hair cells that were immature at the time of neomycin treatment. Alternatively, under cell-cycle inhibition, the UHCPs that may have been already present by the time of neomycin treatment could transdifferentiate into hair cells without undergoing mitosis. To address this possibility, we directly visualized regeneration using SqET4 fish, and compared hair-cell production between control animals and those under nocodazole. Live imaging revealed that over a period of 10 hours, control animals developed 6 hair cells (Fig. 2I), whereas nocodazole-treated animals developed two hair cells and sometimes a single UHCP that could never divide or produce hair cells (Fig. 2J; see Movie 1 in the supplementary material), with the consequent arrest of regeneration. The pair of hair cells that frequently developed in nocodazole-treated fish could have been immature at the time of neomycin treatments, which may also explain why we did not achieved 100% of BrdU(+) hair cells after regeneration. We interpret these results as an indication that supporting-cell transdifferentiation does not contribute to hair-cell regeneration in the lateral line.

Regeneration in neuromasts is always progressive. Live imaging of SqET4 fish showed that UHCPs develop within or in close proximity of the dorsal and ventral polar compartments (Fig. 1B). Consequently, hair-cell regeneration is strongly directional along an axis perpendicular to that of the eventual planar polarity of the epithelium. Hair-cell siblings develop opposite planar orientation and remain adjacent to each other during Phase I. Thus, regeneration anisotropy defines the midline of bilateral planar-polarity symmetry (Fig. 1C,D). These observations led us to ask how regeneration anisotropy is controlled. Oriented cell divisions can direct growth anisotropy in epithelia (Baena-López et al., 2005; Lecuit and Le Goff, 2007). Therefore, the orientation of UHCP divisions could underlie hair-cell regeneration anisotropy and the formation of the midline of bilateral symmetry. We tested this possibility by direct visualization of 23 independent UHCPs in SqET4 animals, which showed that 14 UHCPs divided obliquely (Fig. 3A,B; see Movie 2 in the supplementary material), suggesting that the orientation of UHCP division is not directly translated into bilateral symmetry. We further validated this conclusion by directly revealing the dynamics of UHCPs using the double transgenic line Tg[Cldnb:mGFP; SqET4], which express a membrane-targeted GFP in all supporting cells and a cytoplasmatic GFP in UHCPs and hair cells, allowing the evaluation of cellular behavior in vivo (Haas and Gilmour, 2006). Unexpectedly, we found that ~60% of UHCP divisions produced hair-cell pairs that rotated within the plane of the epithelium after progenitor cytokinesis (n=36). This is clearly demonstrated in the example shown in Fig. 3C, and Movie 3 in the supplementary material, where, 276 minutes into the time series, the immature hair-cell siblings begin to rotate within the plane of the epithelium, transiently and locally breaking the midline of mirror symmetry. This is most evident 306 minutes into the time series. A complete inversion of the hair cells precisely realigns them along the axis of planar polarity of the neuromast to reform mirror symmetry, which is evident 730 minutes into the time series. Therefore, oriented UHCP divisions are not essential for directional regeneration or bilateral symmetry.

To further assess the relationship between regeneration anisotropy and bilateral symmetry, we deemed it essential to determine the mechanism that mediates the localized acquisition of the UHCP fate. We focused on the polar compartments, hypothesizing that their constituent cells either directly differentiate into UHCPs or that they represent a stem-cell niche from where UHCPs originate. An alternative possibility is that a localized stem cell population does not exist in the neuromast. An extension of this proposition is that the polar compartment is not a niche. We tested these hypotheses by continuous imaging of regenerating neuromasts at single-cell resolution for periods ranging between 32 and 48 hours using Tg[Cldnb:mGFP; SqET4] double transgenic fish. We evaluated cellular behavior before and after UHCP development, and directly tracked cellular movement, proliferation and fate acquisition in vivo. Prospective UHCPs were identified retrospectively by playing the time series backwards, and each labeled with a colored dot. Cells were followed forward over time in all 10 successful long-term recordings, which consistently showed that although most neuromast cells do not change their relative position, some individual supporting cells relocated to the areas defined by the polar compartments, where they subsequently became UHCPs (Fig. 3D; see Movie 4 in the supplementary material). This result shows that prospective UHCPs originate elsewhere in the neuromast. It also indicates that the polar compartments are not stem-cell niches. On the contrary, these compartments represent an environment that opposes a niche to permit the differentiation of UHCPs.

Notch signaling is universally required for hair-cell development in vertebrates (Eddison et al., 2000; Haddon et al., 1998; Lanford et al., 1999; Ma et al., 2008). To understand the mechanism that underlies UHCP-fate acquisition, we used as tools two complementary approaches that combined live imaging with genetic and pharmacological alterations of Notch signaling. First, we expressed a constitutively active from of Notch (N^ICD) by heat-shock during regeneration using triple transgenics Tg[hs70p:Gal4; UAS:N^ICD-Myc; SqET4], which blocked the production of UHCPs and hair cells (Fig. 4A). In a converse experiment, we abrogated Notch activity by treating SqET4 fish with the γ-secretase inhibitor DAPT, which generated ectopic and supernumerary UHCPs and hair cells (Fig. 4A,B) (Ma et al., 2008). These results suggest that a majority of supporting cells can become UHCPs upon loss of Notch signaling. To test this possibility directly, we continuously imaged regenerating Tg[Cldnb:mGFP; SqET4] fish under DAPT treatment and found that UHCPs developed ectopically and in excess (Fig. 4B,C; see Movie 5 in the supplementary material). Importantly, loss of Notch broke directional regeneration and led to planar polarity defects (Fig. 4B).

Our previous results clearly show that Notch signaling controls the differentiation of UHCPs, but do not explain why UHCPs are produced within the polar compartments. The spatiotemporal development of UHCPs shadows the expression pattern of the hair-cell determination factors DeltaA and Atoh1a (Ma et al., 2008), suggesting that these genes are first expressed by the UHCPs. To test this possibility directly, we imaged hair-cell regeneration in a double-transgenic line expressing a red-fluorescent protein under the transcriptional control of the Atoh1a promoter, combined with the UHCPs marker SqET4. Live imaging of Tg[Atoh1a:TdTomato; SqET4] fish showed that Atoh1a is expressed at low levels in some supporting cells that we regard as prospective progenitors (see below), and is strongly upregulated by the UHCPs at the dorsal or ventral part of parallel neuromasts, where the polar compartments are located (Fig. 4D; see Movie 6 in the supplementary material). Cell expressing high levels of Atoh1a:TdTomato were also GFP(+) and divided to produce a pair of hair cells, indicating that Atoh1^strong cells are UHCPs. Atoh1 is a well-characterized negative transcriptional target of Notch (Itoh and Chitnis, 2001), suggesting that either Notch activity is low in the polar compartments, or that prospective progenitors become refractory to Notch signaling. Our previous live imaging analyses of regenerating Tg[SqET4; Cldnb:mGFP] fish under DAPT treatment indicated that the majority of supporting cells can become UHCPs upon loss of Notch signaling. These results, together with the evidence that loss of Notch signaling expands Atoh1a expression (Itoh and Chitnis, 2001; Ma et al., 2008), indicate that supporting cells are not intrinsically refractory to Notch. Thus, some supporting cells are able to express the UHCP fate within the permissive environment of the polar compartments where Notch activity may be low or absent. Notch3 is the main receptor expressed in neuromasts and is dynamically regulated during hair-cell regeneration (Ma et al., 2008). To reveal the spatial expression pattern of Notch3, we used whole-mount fluorescent two-color in situ hybridization in combination with the UHCP markers Atoh1a and DeltaA. We found that Notch3 is strongly enriched in neuromast areas complementary to those expressing DeltaA and Atoh1a (Fig. 4E,F). Importantly, Notch3 was absent from the areas representing the polar compartments. To assess Notch signaling status directly in neuromasts, we used a validated red-fluorescent Notch sensor in combination with SqET4 (Parsons et al., 2009). Live imaging of neuromasts expressing this combination of markers showed that Notch activation occurs outside the polar compartments (Fig. 4G). Collectively, these results indicate that Notch activity governs hair-cell regeneration anisotropy by preventing supporting cells from becoming progenitors outside the polar compartments.

During phase II of regeneration, the neuromast macula loses its oval shape to become rounder because the hair-cell population expands in a direction perpendicular to that of regeneration. Consequently, planar polarity propagates laterally across the epithelium (López-Schier and Hudspeth, 2006) (Fig. 1A,E). This may occur because hair cells move away from the midline, or because UHCPs begin to appear ectopically in the neuromast during this phase. To discriminate between these possibilities, we divided the neuromast by a Cartesian grid and gave positional values to the hair bundle of each hair cell to calculate the spatial distribution of each polarity (Fig. 5A-D). This analysis showed that nearly 100% of the hair cells of each orientation were placed in separate compartments at either side of the vertical midline during phase I (Fig. 5A,B). A strongly biased distribution of polarities is maintained at either side of a vertical midline bisecting fully regenerated neuromasts (Fig. 5C,D). Importantly, comparison of polarity distributions across a horizontal midline showed distributions of 52/48% in each compartment, indicating no polarity distribution bias between the dorsal and ventral halves of parallel neuromasts (Fig. 5D). These observations support the possibility that the macula expands horizontally by a centrifugal movement of hair cells. During hair-cell development, there is a temporal hierarchy of gene expression along which cells transition as follows: SqET4⁻/Atoh1a^weak+ (prospective UHCPs) → SqET4⁺/Atoh1a^strong+ (UHCPs and immature hair cells) → SqET4⁺/Atoh1a⁻ (mature hair cells) (Fig. 5E). We evaluated the expression of these markers in fully regenerated neuromasts using triple transgenics Tg[atoh1a:TdTomato; SqET4; Cldnb:mGFP] and found that the oldest hair cells (SqET4⁺/Atoh1a⁻) were located at the periphery of the neuromast furthest from the midline of bilateral symmetry, whereas the youngest pairs (SqET4⁺/Atoh1a^strong+) were central and described the typical direction of regeneration that connects the dorsal and ventral polar compartments (n=9) (Fig. 5F). DiASP incorporation experiments that discriminate between immature and mature (mechanotransducing) hair cells revealed that older cells described a line perpendicular to that of the direction of regeneration (Fig. 5G). Importantly, DiASP(+) cells were always excluded from the polar compartments where the UHCPs and immature hair cells are preferentially located (Fig. 5G). These results, together with the scarcity of UHCPs outside the polar compartments (Fig. 3D, Fig. 4C), indicate that the horizontal expansion of planar polarity is not caused by the ectopic development of UHCPs, and that it is likely to occur because older hair cells that maintain their original orientation move away from the midline of the neuromast towards its periphery.

During the initial phase of hair-cell regeneration in the zebrafish lateral line, a vertical midline bisects the neuromast epithelium into perfect mirror-symmetric plane-polarized halves. Each half contains hair cells of identical planar orientation but opposite to that of the confronting half. Hair-cell regeneration is strongly directional along the axis of bilateral symmetry. Our results explain the reason behind this reproducible behavior. It occurs because the development of UHCPs is spatially restricted to the dorsal or ventral polar compartments. The division of UHCPs into two hair cells, coupled with a consistent opposite planar orientation of the hair-cell siblings along a single axis, eventually defines the midline of mirror symmetry. Live imaging demonstrates that oriented progenitor divisions are not essential for regeneration anisotropy. Therefore, bilateral symmetry is sustained by a strongly anisotropic regeneration process that relies on the stabilization of progenitor identity in permissive polar compartments. An additional important aspect of our results is that they conclusively demonstrate that the SqET4 transgenic line highlights bona fide hair-cell progenitors, the proliferation of which is essential for hair-cell regeneration in the lateral line. This conclusion is further supported by a recent publication reporting that the promoter element of the atp2b1a gene is responsible for the expression pattern of the GFP in the SqET4 transgenic line, and that a morpholino-mediated knockdown of atp2b1a negatively affected the division of the hair-cell progenitors (Go et al., 2010).

One explanation for the compartmentalized acquisition of UHCP identity is that the polar compartments are stem cell niches. Live imaging indicates that this is unlikely because prospective UHCPs originate elsewhere in the neuromast. Thus, the polar compartments appear to be a permissive environment for the acquisition of UHCP identity by supporting cells. One possibility is that an intrinsic cell-fate determinant instructs some supporting cells to become UHCPs, and that intercellular signals prevent these cells from fully executing their differentiation until they reach a polar compartment. Previous results suggest that the source of hair-cell progenitors are the Sox2(+) cells that reside basally in the neuromast epithelium (Hernández et al., 2007). The observation that in the mouse, chick and zebrafish ears, the combinatorial activity of Sox2 and Notch control the development of pro-sensory patches, from where hair cells will eventually develop supports this proposition (Daudet et al., 2009; Millimaki et al., 2010). Future experiments involving loss- and gain-of-function of Sox2 in the lateral line will allow this possibility to be tested directly.

What controls the spatiotemporal development of UHCP? When Notch activity was blocked in the whole organ by DAPT treatments, directional regeneration was broken, indicating that compartmentalized Notch signaling prevents the ectopic development of hair-cell progenitors. Using fluorescent sensors and live imaging, we demonstrate that the polar compartments have low levels of Notch signaling. Because the localization of the polar compartments does not change over time, a consequence of UHCP development within them is the anisotropic regeneration of the hair cells. We have observed that hair-cell regeneration is always progressive. Live imaging of SqET4 fish showed that UHCP development alternate between the dorsal and ventral aspects of the neuromast. One explanation for the progressive development of UHCPs is that Ahtoh1a in the UHCPs and in young hair cells activates the expression of Notch ligands cell-autonomously. In turn, lateral inhibition originating from these cells would prevent the surrounding supporting cells from becoming new progenitors until hair cells have matured and lost Atoh1a expression. This would decrease the number of cells expressing Notch ligands, allowing the organ to re-set and develop new UHCPs. It will be necessary to define the identity of the relevant Notch ligands and to manipulate their activity in a cell-specific manner to test this hypothesis.

One outstanding question is the origin of axial references and whether they rely on long-range signals. If this were the case, disruptions of regeneration anisotropy should not affect epithelial planar polarity because hair cells would be properly oriented regardless of their position in the epithelium. It follows that disruptions of Notch that break regeneration anisotropy should not cause planar-polarity defects in the epithelium. However, loss of Notch signaling randomizes hair-cell orientation, arguing against a role of long-range cues in polarizing the epithelium. However, we cannot currently rule out the possibility that Notch signaling directly controls the establishment of axial references for planar polarity by affecting the expression, transport or interpretation of long-range polarizing cues. Future experiments using iterative cycles of hair-cell ablation and regeneration, with an intervening blockade of Notch signaling, may be able to provide evidence for or against a role of Notch in the control of polarizing cues. An alternative possibility that could reconcile the effect of loss of Notch with a role of long-range polarization is that Notch signaling controls the positioning of the organelle that defines planar polarity in this tissue, the kinocilium of the hair cell. Under this scenario, loss of Notch would not allow kinocilia to respond to external polarizing cues. However, we believe that this is not likely because loss of Notch signaling in the mouse and the zebrafish ears produce normally polarized hair cells, which align in similar directions to their neighbors (Haddon et al., 1998; Lewis and Davies, 2002). In addition, a disruption of the internal machinery for planar polarity in the cell by loss of Notch should generate some hair cells with centrally located kinocilia, a phenotype that we have never observed. With our current knowledge, the most likely explanation for the planar polarity defects in neuromasts lacking Notch signaling is that supernumerary hair cells produce mechanical disturbances or packaging defects. Thus, a role for long-range polarizing signals at the origin of axial references remains a possibility.

Although hair-cell regeneration remains strongly anisotropic, the macula eventually expands symmetrically. This could be due to the development of UHCPs outside the polar compartments during Phases II and III. Alternatively, polar compartments may themselves relocate or expand around the entire neuromast. Although these situations are possible, the weight of the evidence is against them. First, the majority of UHCPs develop within the dorsal or ventral aspect of neuromasts in all three phases of regeneration, and the small percentage of UHCPs that could not be unambiguously located within these compartments would be insufficient to account for the rapid and widespread symmetric expansion of planar polarity. Second, we observed that peripheral hair cells were negative for Atoh1a, indicating that they are older that those located centrally or nearby the polar compartments. Therefore, we conclude that planar polarity is likely to propagate symmetrically because hair cells move away from the midline towards the periphery of the neuromast. It follows that the hair cells located peripherally along a line perpendicular to that of regeneration should be the oldest. Our results on DiASP incorporation support this conclusion.

Live imaging demonstrates that oriented progenitor divisions are not essential for regeneration anisotropy. In addition, we observed that the majority of hair-cell sibling rotate around their contact point immediately after UHCP cytokinesis. To the best of our knowledge, our data is the first description of ‘planar cell inversions’. This remarkable cellular behavior could offer further mechanistic insights into the process that sustains bilateral symmetry during regeneration. It may reveal a general process that links epithelial polarity to external mechanical cues (Aigouy et al., 2010). It could also play a role in maintaining cellular orientation in plastic tissues with high cellular turnover, such as in the mammalian kidney, lung or brain ventricles (Fischer et al., 2006; Mirzadeh et al., 2010; Saburi et al., 2008; Yates et al., 2010).

In this study we employed transgenic sensors and markers of hair-cell progenitors, combined with high-resolution long-term in toto continuous live imaging, to demonstrate that hair-cell regeneration is strongly anisotropic, and that regeneration anisotropy is regulated by Notch signaling. The model that emerges from our results (summarized in Fig. 6) suggests that bilateral symmetry is sustained by compartmentalized Notch activity, which governs regeneration anisotropy by permitting the stabilization of UHCP identity in the polar compartments. There are several examples of dynamic or stable compartments that allow cells to loose stemness and progress through a differentiation program (Mathur et al., 2010; Voog et al., 2008). In the specific case of the lateral line, the polar compartments appear to oppose a niche by allowing supporting cells to become UHCPs. Further molecular dissection of this process in the zebrafish lateral line may provide deeper insights into the mechanisms that control the homeostasis of tissue architecture.

We thank V. Korzh, M. Parsons, N. Lawson, M. Allende, S. Burgess, R. Tsien and T. Whitfield for reagents, and many colleagues for comments on the manuscript. This work was supported by an European Research Council Starting Grant to H.L.-S., the Ministry of Science and Innovation of Spain to H.L.-S., the CRG, and the FCT of Portugal through the GABBA PhD program to F.P.-T.