Delta-like 1 and lateral inhibition during hair cell formation in the chicken inner ear: evidence against cis-inhibition

Lateral inhibition is a key mechanism for cell diversification and patterning in a wide variety of embryonic and adult tissues. In this process, cells adopting a given developmental fate inhibit their neighbours from doing likewise through direct cell-to-cell communication mediated by the Notch signalling pathway (reviewed by Chitnis, 1995; Lewis, 1998; Artavanis-Tsakonas et al., 1999). The signal-sending cells that adopt the primary fate express ligands of the Delta/Serrate (also named Jagged in mammals)/Lag2 (DSL) family that bind to and activate Notch receptors in the neighbouring signal-receiving cells. This triggers a series of γ-secretase-dependent cleavages that release the intracellular domain of Notch (NICD). The NICD then translocates to the nucleus and induces expression of transcriptional repressors of the Hairy and Enhancer of Split (HES) family, which inhibit adoption of the primary fate (reviewed by Bray, 2006). Hence the signal-sending and signal-receiving cells end up with low and high levels of Notch activity, respectively, and adopt opposite fates.

In the standard model of lateral inhibition with feedback, the expression of DSL ligands is repressed by Notch activity (Heitzler and Simpson, 1991; Heitzler et al., 1996); consequently, a signal-receiving cell will lose over time its ability to act as a signal-sending cell. By amplifying small and random variations in the expression of DSL ligands, this intercellular negative-feedback loop is thought to be sufficient to drive initially equivalent cells into either high or low Notch states (Collier et al., 1996). However, in many cases lateral inhibition is biased by intrinsic factors that impact on the ability of cells to send or receive Notch signals (reviewed by Schweisguth, 2004). There is also growing evidence that, in addition to their effects in trans, DSL ligands can inhibit Notch activity cell-autonomously (in cis) (de Celis and Bray, 1997; Klein et al., 1997; Micchelli et al., 1997; Jacobsen et al., 1998) (reviewed by del Álamo et al., 2011). The great variety of cellular contexts in which lateral inhibition operates could be matched by equal diversity in its molecular mechanisms.

One striking outcome of lateral inhibition is found in epithelial sheets, in which interacting cells differentiate into a salt-and-pepper mosaic of alternate cell types. Examples of such epithelia are found in the lung endoderm (Morimoto et al., 2010), the Xenopus epidermis (Deblandre et al., 1999), the zebrafish pronephros (Liu et al., 2007) and in the inner ear. The inner ear comprises several pseudostratified sensory epithelia that contain thousands of mechanosensory hair cells that are activated by the displacement of their apical bundle of stereocilia. Hair cells are located in the apical layer and are interspersed by the apical processes of several non-sensory supporting cells (Fig. 1A,B) with cell bodies that rest on the basal lamina. The formation of this cellular mosaic from bipotent progenitor cells is regulated by lateral inhibition. Newly formed hair cells express delta1-like 1 (Dll1) and jagged 2 (Jag2) ligands that activate Notch1 receptors in neighbouring progenitor cells; in the latter, high levels of Notch activity induce Hes1/5 expression, which in turn suppress expression of the pro-hair cell transcription factor Atoh1 (Fig. 1C) (Bermingham et al., 1999). When Notch signalling is disrupted at the time of hair cell formation, overproduction of hair cells occurs (Haddon et al., 1998; Lanford et al., 1999; Zine et al., 2000; Zine et al., 2001; Kiernan et al., 2005; Brooker et al., 2006). Modelling studies suggest that the emergent properties of lateral inhibition with feedback are sufficient to pattern inner ear epithelia (Fig. 1D) (Collier et al., 1996; Webb and Owen, 2004). However, nothing is known about the actual dynamics of this process in vivo, and the way that Dll1 functions to regulate hair cell fate decisions remains unclear. In fact, it has been reported that artificial overexpression of Dll1 does not affect hair cell formation (Eddison et al., 2000) and that transient contacts between immature hair cells can occur in the embryonic chick inner ear (Goodyear and Richardson, 1997) (supplementary material Fig. S1). These observations suggest that patterning of inner ear epithelia by lateral inhibition is more complicated than the standard model proposes.

Here, we used a Hes5 reporter to monitor Notch activity in the developing chicken inner ear. We then used different gain-of-function strategies to investigate Dll1 function during hair cell formation and its impact on Notch activity. In agreement with the standard model, our results show that Dll1 promotes hair cell fate cell-autonomously and represses hair cell formation in trans. However, some progenitor cells are not prevented from becoming hair cells by neighbouring Dll1-expressing cells. This is unlikely to result from cis-inhibition of Notch activity by Dll1 itself, given that overexpression of Dll1 does not prevent activation of the Hes5 reporter cell-autonomously. Altogether, our data show that Dll1 is a key determinant of hair cell fate decisions and that it functions primarily in trans. Our findings also support the idea that additional mechanisms must operate downstream of lateral inhibition to eliminate patterning errors in the sensory epithelia of the inner ear.

Fertilised White Leghorn chicken (Gallus gallus) eggs were obtained from Henry Stewart UK and incubated at 37.8°C for the designated times. Embryonic stages are either from Hamburger-Hamilton (HH) tables (Hamburger and Hamilton, 1992) or refer to embryonic days (E), with E1 corresponding to 24 hours of incubation. Embryos older than E5 were sacrificed by decapitation. All procedures were approved by University College London and by the UK Home Office.

The following plasmids were used: pHes5::d2EGFP (Takebayashi et al., 1995); RCAN(B) (Petropoulos and Hughes, 1991); RCAS-cDelta1 (Henrique et al., 1997); pCAS-Delta1-IRES-EGFP (Fior and Henrique, 2005); pDSRED2-C1 (thereafter named pDsRed; Clontech); TOPO-nTomato (Collins et al., 2010); pCAGGS-T2TP encoding the Tol2 transposase; pT2K-CAGGS; pT2K-CAGGS-rtTA²M2, a Tol2 construct encoding the tet-on activator; pT2K-TRE-B1-eGFP (herein named pTRE-EGFP), which consists of a cassette of bidirectional transcriptional units (one controlling transcription of EGFP, the other empty) under the control of a tetracycline-responsive element (TRE) between the left and right ends of Tol2 (Sato et al., 2007; Watanabe et al., 2007); pTurbo-FP635 (Evrogen).

The following plasmids were generated using standard cloning procedures: pHA-NICD-IRES-mRFP1, driving constitutive expression of an HA-tagged version of the intracellular domain of chicken Notch1 and mRFP1; pT2K, a promoterless version of pT2K-CAGGS; pT2K-Hes5::d2EGFP and RCAN(B)-Hes5::d2EGFP; pT2K-Hes5::nd2EGFP; pT2K-CAGGS-Delta1-IRES-EGFP (thereafter named pT2K-Delta1-GFP); pT2K-Hes5::Delta1-IRES-EGFP (thereafter named pT2K-Hes5::Delta1-GFP); pT2K-CAGGS-nTomato (pT2K-nTomato); pT2K-TRE-B1-FP635 (herein named pTRE-FP635), which is a modified version of pTRE-EGFP in which Turbo-FP635 has been cloned in place of EGFP; pT2K-TRE-B1-FP635-Delta1 (pTRE-FP635-Delta1), which contains the coding sequence of chicken Dll1 in the second transcriptional unit of pTRE-FP635.

Microelectroporation of the otic cup of E2 chick embryos was performed using a BTX ECM 830 Electro Square Porator as previously described (Daudet and Lewis, 2005). Doxycycline (Dox) treatments were performed in ovo by injecting around the embryos ~500 μl 20 μg/ml Dox in PBS.

The sensory epithelium was dissected at room temperature in DMEM/F12 medium and immobilised at the bottom of Mattek dishes using either a platinum harp or 1% low-gelling-temperature agarose. Approximately 2-3 ml of DMEM/F12 medium without Phenol Red (Invitrogen) was added and the specimens were maintained for a minimum of 3 hours in a tissue culture incubator (5% CO₂, 37°C) before live imaging. The γ-secretase inhibitor DAPT {N-[N-(3,5-difluorophenacetyl)-1-alanyl]-S-phenylglycine t-butyl ester; Calbiochem} was used at 20 μM. Time-lapse confocal microscopy was performed using a Nipkow spinning-disc confocal microscope and a Zeiss LSM510 inverted confocal microscope equipped with an environmental chamber. Frame intervals for time-lapse experiments ranged from 10 to 20 minutes. Confocal stacks were analysed using ImageJ or iQ (Andor). For quantification of fluorescence levels in the DAPT experiments, average projections of the z-stack (12-bit images) were made at each time point. At least two regions of interest (ROIs) were selected within each sample. Arbitrary fluorescence units were obtained by subtracting background levels of fluorescence from ROI values at each time point.

Immunocytochemistry and in situ hybridisation were performed as described previously (Daudet and Lewis, 2005). The following antibodies were used: monoclonal mouse IgG1 anti-HCA (hair cell antigen; supernatant used at 1:1000-1:2000) (Bartolami et al., 1991); monoclonal mouse IgG2b anti-HCS1 (recognising otoferlin; 1:250) (Goodyear et al., 2010); rabbit anti-Serrate1 (Adam et al., 1998); and rabbit anti-Delta1 (Henrique et al., 1997). Goat anti-mouse IgG or anti-rabbit IgG secondary antibodies conjugated to Alexa 405, 488, 546, 633 or 647 (Invitrogen) were used at 1:1000. Specimens were examined on a Zeiss LSM510 inverted confocal microscope.

For comparison of Hes5::nd2EGFP levels in cells transfected with either pTRE-FP635 or pTRE-Delta1-FP635, the inner ear of E5.5 embryos treated with Dox for 21 hours was dissected and imaged on a spinning-disc Nipkow confocal microscope equipped with a 40× oil-immersion objective. Measurements of mean intensity values for the red (FP635) and green (EGFP) channels (12-bit confocal stacks of 40-50 optical slices) were made on randomly selected Hes5::nd2EGFP-positive nuclei using ImageJ. FP635-positive nuclei were defined by a mean value of FP635 fluorescence that exceeded 400, a threshold value corresponding approximately to the mean FP635_background + (10 × s.d. of FP635_background). The mean EGFP levels of each nucleus were standardised using Z-scoring [z=(x–mean)/s.d.] then pooled across each experimental condition (FP635 or Delta1-FP635 transfected). Similar analysis of Hes5::nd2EGFP levels was performed on five samples electroporated with pTRE-Delta1-FP635 and treated at E6 with Dox for 14 hours, then immunostained for Delta1 expression to identify Dll1-induced and non-induced cells. Statistical analyses were computed using SPSS 19 for Mac. The data did not follow a normal distribution, so nonparametric tests were used; all P-values are two-tailed.

To investigate the spatiotemporal pattern of Notch activity during chicken inner ear development, we used a reporter of Notch activity that contains a 0.8 kb fragment of the promoter of the mouse Hes5 gene regulating the expression of a destabilised form of EGFP, pHes5::d2EGFP (Takebayashi et al., 1995). When analysed 16-20 hours after in ovo electroporation of the otic placode/cup at E2 (Fig. 2A,B), pHes5::d2EGFP elicited a pattern of EGFP fluorescence very similar to the endogenous pattern of Notch activity. At this developmental stage, otic neurogenesis occurs in a restricted, anterior region of the otic cup, in which high levels of Dll1 and Hes5 expression are present (Fig. 2B) (see Adam et al., 1998; Abelló et al., 2007; Daudet et al., 2007). In specimens (n=20) transfected with pHes5::d2EGFP, strong fluorescence was restricted to the neurogenic patch despite widespread transfection of the otic cup as shown by pDsRed expression (Fig. 2D). Furthermore, co-electroporation of pHes5::d2EGFP with a plasmid driving constitutive expression of either an active form of the chicken Notch1 receptor (HA-tagged NICD, which activates Notch signalling in cis and independently of ligand binding) or the Notch ligand chicken Dll1 (which activates Notch signalling in trans) induced strong and widespread EGFP fluorescence in the otocyst (Fig. 2E; n=6; data not shown). This indicated that the Hes5::d2EGFP reporter is sensitive to canonical Notch activity in the chicken inner ear.

Hair cell formation starts at around E4 in vestibular patches, peaks at E9-E12 in the utricle and carries on at a slower pace until post-hatch stages (Goodyear et al., 1999). In the basilar papilla, hair cell formation occurs between E5 and E10 only (Katayama and Corwin, 1989; Goodyear and Richardson, 1997). In order to monitor Notch activity with the Hes5 reporter at these late developmental stages, the Hes5 reporter was cloned into the RCAN retroviral construct (RCAN-Hes5::d2EGFP) and into a promoterless version of the Tol2 transposon (PT2K-Hes5::d2EGFP). Both types of vector enable stable integration of transgenes into the genome of chicken cells (Petropoulos and Hughes, 1991; Sato et al., 2007). Following in ovo electroporation of either construct at E2, strong EGFP fluorescence was detected in the developing sensory patches at E5 and E10 (Fig. 3). At E5, the levels of Hes5::d2EGFP fluorescence were more intense in the developing crista than in the other sensory patches (Fig. 3B,C); this fits with the endogenous pattern of Hes5 expression (Fig. 3A) and the temporal progression of hair cell production in the embryonic chick inner ear. At E10, many hair cells identifiable by HCA and otoferlin (HCS1) immunostaining had formed within sensory patches, where high levels of Hes5 gene expression (Fig. 3D,E) and Hes5::d2EGFP fluorescence (Fig. 3F,G) were observed.

In whole-mount preparations examined at high magnification (Fig. 3H), high levels of EGFP were restricted to elongated cells with small polygonal apical surfaces intercalated in between hair cells, which are typical features of progenitor cells and supporting cells. A few EGFP-positive cells were undergoing mitotic division, as demonstrated by DAPI staining of their condensed nuclear DNA in apical planes of the epithelium. Preliminary time-lapse observations of transfected inner ear samples (Fig. 3I; supplementary material Movies 1, 2; data not shown) confirmed that a large number of Hes5::d2EGFP-positive cells were indeed mitotic progenitor cells. Their nuclei underwent a characteristic interkinetic migration, with mitosis occurring exclusively at the apical surface and generating transient movements of neighbouring cells during anaphase and telophase. By contrast, cells with very reduced Hes5::d2EGFP fluorescence, a rounded and constricted apical surface, and a more steady and apically located nucleus (visualised by co-transfecting with pT2K-nTomato in some samples) were not proliferating; these were likely to correspond to the hair cells previously identified in fixed specimens.

We next analysed Dll1 expression in relation to hair cell markers and Notch activity. Immunostaining for Dll1 was visible in small intracellular vesicles – a subcellular localisation that is due to the internalisation of DSL ligands in signalling cells (Itoh et al., 2003; Matsuda and Chitnis, 2009). However, the cells with highest levels of Dll1 protein were not the HCA/HCS1-positive hair cells, but cells located in between these well-differentiated hair cells (Fig. 3J,K). In Hes5::d2EGFP-transfected samples, the Dll1-positive cells had low levels of EGFP fluorescence, a rounded and constricted apical surface and no HCA/HCS1 expression (Fig. 3L,L′) or HCA expression only (supplementary material Fig. S2); these characteristics suggested that Dll1-expressing cells were prospective hair cells. Interestingly, at least some of these cells appeared to contact more mature-looking hair cells (yellow arrowhead in Fig. 3L,L′ and supplementary material Fig. S2), an observation that is in line with that of transient contacts between mature and immature-looking hair cells reported by Goodyear and Richardson (Goodyear and Richardson, 1997).

To directly test Dll1 function during hair cell formation, we used two different Tol2 constructs. In the first, a constitutively active promoter drives chicken Dll1 in all transfected cells (pT2K-Delta1-EGFP; Fig. 4A). In the second construct, Dll1 expression is regulated by the mouse Hes5 promoter (pT2K-Hes5::Delta1-EGFP; Fig. 5A) so as to investigate specifically the effects of Dll1 in ‘Notch-active’ progenitor cells, which are normally prevented from differentiating into hair cells. Both constructs included an IRES-EGFP sequence to identify transfected cells. Following electroporation at E2, the inner ear was analysed at stages when hair cells have differentiated, between E7 and E12. Highly transfected ears exhibited morphological defects, possibly owing to interference of the artificial Dll1 expression with the early function of Notch activity in prosensory specification (data not shown). For clarity, we focus here on the effects of Dll1 on hair cell differentiation only. These were more clearly seen in the basilar papilla than in vestibular epithelia, as hair cell production is almost complete by E10 in the former tissue (Katayama and Corwin, 1989; Goodyear and Richardson, 1997).

In pT2K-Delta1-EGFP-transfected basilar papilla (n=7), levels of EGFP fluorescence were variable but transfected hair cells and supporting cells were easily recognisable by their shape as well as by immunostaining with HCA/HCS1. Within the same sample, two contrasting types of transfected region were found: those with a clear reduction in hair cell density (outlined in Fig. 4B), and those that appeared relatively unaffected (arrow in Fig. 4B). The inhibition of hair cell formation only occurred in regions in which relatively large clusters of cells (~10 or more) were transfected and contacted one another (outlined in Fig. 4C). In EGFP-positive regions without any reduction in hair cell density, the majority of hair cells were in fact transfected; although these appeared densely packed, the EGFP-positive hair cell bodies were in general surrounded by a darker outline, indicating the presence of non-transfected supporting cells separating them. This was confirmed by examination of the surface of the epithelium stained with fluorescent phalloidin. Hence, the majority of isolated transfected cells were hair cells, a trend that was more obvious in regions containing fewer transfected cells (Fig. 4D). Despite this, some hair cells, transfected (e.g. arrowheads in Fig. 4C) or otherwise (Fig. 4E,E′), were also found in direct contact with Dll1-overexpressing cells of supporting cell morphology. This suggests that some progenitor cells can differentiate into hair cells despite contacting neighbouring cells that express Dll1.

Similar observations were made in pT2K-Hes5::Delta1-EGFP-transfected samples, in which Dll1 expression was artificially induced in Notch-active cells only (Fig. 5). The reduction in hair cell density was only seen in regions containing large clusters of transfected cells (Fig. 5B,C), whereas a large proportion of isolated transfected cells differentiated into hair cells (Fig. 5D). Out of 171 isolated EGFP-positive cells scored in the basilar papilla (n=6; E11-E12), 114 were hair cells (66%). By comparison, only 30% of hair cells were scored among cells transfected with pT2K-nTomato in control experiments (n=355 cells; hair cell to supporting cell ratio of 1:2.45). This value fits within the 1:1.71 to 1:3.9 range of hair cell to supporting cell ratio determined in the basilar papilla of E12 chicken embryos (Goodyear and Richardson, 1997).

With the Hes5::Delta1-EGFP construct, the intensity of EGFP fluorescence was lower in hair cells than in supporting cells (Fig. 5D), suggesting that transfected hair cells were no longer expressing the Hes5::Delta1-EGFP transgene at the time of analysis. To verify this, we performed immunostaining for Dll1 in Hes5::Delta1-EGFP-transfected samples. None of the weakly EGFP-positive hair cells (arrowheads in Fig. 5E) expressed detectable levels of Dll1 in their cytoplasm. This confirmed that transfected progenitor cells that differentiate into hair cells stop expressing the Delta1-EGFP transgene once they no longer experience Notch activity. By contrast, isolated cells or clusters with high levels of EGFP fluorescence and a supporting cell-like morphology expressed high levels of Dll1 protein (arrow in Fig. 5E). In such clusters, high levels of Notch activity must be present in order to maintain expression of the Delta1-EGFP transgene. Finally, as previously reported with the constitutively expressed Delta1-EGFP transgene, we noted that some hair cells (transfected or otherwise) were contacting Dll1-expressing cells.

Altogether, these results showed that Dll1 represses hair cell differentiation in trans and promotes adoption of the hair cell fate in cis, including for progenitor cells that experienced Notch activity. However, the trans-inhibition of hair cell formation by Dll1 was not systematic; the occurrence of contacts between hair cells and Dll1-overexpressing cells as well as the formation of hair cells within highly transfected regions suggested that some progenitor cells might be insensitive to Dll1-mediated lateral inhibition.

The occurrence of reciprocal contacts between hair cells and the differentiation of hair cells next to Dll1-expressing cells could in theory be explained by cis-inhibition, i.e. if Dll1 inhibited Notch activity cell-autonomously, in addition to its effects in trans. The persistence of high levels of expression of Delta1-EGFP when driven by the Hes5 promoter in clusters of transfected cells suggests that if Dll1-mediated cis-inhibition occurs then it is not as efficient as the trans-activation of Notch receptors. However, Hes5-regulated gene expression is dynamically regulated by Notch activity; therefore, Hes5::Delta1-transfected cells might in fact be oscillating between ON and OFF phases of Dll1 expression and cis-inhibition.

In order to determine whether Dll1 functions by cis-inhibition in the inner ear, we investigated the consequences of a time-controlled induction of Dll1 expression for the activity of the Hes5 reporter. If Dll1 were able to inhibit Notch activity cell-autonomously, we expect that cells in which Dll1 is induced would have reduced levels of Hes5::nd2EGFP fluorescence ~8-12 hours after the onset of inhibition – a delay that is necessary for reduction of Hes5::d2EGFP fluorescence to basal levels after artificial blockade of Notch activity with DAPT in organotypic cultures (see supplementary material Fig. S3 and Movie 3). To test this, we co-transfected inner ear samples with the pT2K-Hes5::nd2EGFP reporter (which includes a nuclear-localised variant of d2EGFP to facilitate analysis of fluorescent signals) and two additional Tol2 constructs: pTRE-Delta1-FP635, in which the expression of Dll1 (and the red fluorescent protein FP635) is regulated by a tetracycline-responsive element (Fig. 6A), and pT2K-CAGGS-rtTA²M2, which encodes the tet-on transactivator (Sato et al., 2007).

Electroporated embryos were incubated until E9, then treated with doxycycline (Dox) in ovo and sacrificed 6, 24 and 48 hours later (n=3 for each time point). The basilar papillae were dissected and immunostained for Dll1 expression and hair cell markers. Six hours after Dox treatment (Fig. 6B,C), Dll1-induced cells were recognisable after immunostaining by their elevated levels of intracellular and membrane-localised Dll1 protein. In contrast to previous experiments, the majority of Dll1-induced cells had supporting cell characteristics (no HCA staining and elongated cell bodies) and the pattern of hair cell differentiation was normal in transfected samples (Fig. 6C,C′), indicating that the Dll1 transgene was not expressed in the absence of Dox. The pattern of transfection was mosaic and the cells overexpressing Dll1 represented only a subset of Hes5::nd2EGFP-positive cells. This allowed a comparison of the levels of Hes5::nd2EGFP within cells that overexpressed Dll1 with those that did not.

The cells with the highest levels of nuclear-localised Hes5::nd2EGFP fluorescence were located in regions where the Dll1 protein was overexpressed, confirming the efficient induction of Notch activity by ectopic Dll1. The levels of Hes5::nd2EGFP fluorescence varied greatly from cell to cell; however, high levels of Notch activity were detected in both Dll1-induced and non-induced cells. At 24 (Fig. 6D) and 48 (not shown) hours after Dox treatment, there was still very strong expression of Dll1 within induced cells, the majority of which had supporting cell morphology. Once again, the fluorescence of the Hes5::nd2EGFP reporter varied from cell to cell, being stronger in Dll1-induced regions, and was particularly high in some of the Dll1-induced cells.

Similar results were obtained in experiments performed at an earlier developmental stage in vestibular sensory patches. The Dox treatment was started at E5.5 and the inner ear tissue analysed 21 hours later. As previously noted, there was strong induction of Hes5::nd2EGFP fluorescence around Delta1-FP635-induced cells, but some Delta1-FP635-expressing cells were also Hes5::nd2EGFP positive (Fig. 6E). Despite some caveats linked to the mosaicism of transfection, the effects of Dll1 in trans, and some uncertainties about the differentiation status of FP635-positive cells at these early developmental stages, we tried to quantify the effect of ectopic Dll1 on the activity of the Hes5 reporter. We measured and compared the mean intensity of Hes5::nd2EGFP fluorescence within individual FP635-positive cells (see Materials and methods) that were either transfected with pTRE-Delta1-FP635 (n=461 out of a total of 1416 nuclei in three samples) or with pTRE-FP635 (n=959 out of a total of 1697 nuclei in five samples) as a control. Statistical analysis of the data showed that the distribution of the normalised values of mean fluorescence intensities (Fig. 6F) was not significantly different between the two categories of transfected cells (Mann-Whitney U=212,380; P=0.231), suggesting that the induction of Dll1 did not significantly alter the intrinsic levels of Notch activity. As a further test, we repeated this analysis on E6 samples treated in ovo with Dox for 14 hours (n=5), and immunostained for Dll1 to identify Dll1-overexpressing cells (Fig. 6G,H). The mean levels of Hes5::nd2EGFP activity varied greatly from cell to cell and were not reduced in Dll1-induced cells. On the contrary, statistical analysis of the normalised values of mean nd2EGFP fluorescence showed that Hes5::nd2EGFP activity was significantly elevated in Dll1-induced cells (n=327) compared with non-induced cells (n=305; Mann-Whitney U=56,318; P=0.05).

Altogether, these data indicate that Dll1 acts primarily in trans to activate Notch receptors, and strongly suggest that expression of Dll1 does not inhibit Notch activity cell-autonomously in the developing inner ear.

Much of our understanding of tissue patterning by lateral inhibition derives from mathematical modelling and in vitro studies, but comparatively little is known about the actual dynamics of this process in vivo, in particular in vertebrates. In this study, we tested some of the fundamental assumptions of the lateral inhibition model in the sensory epithelia of the chicken inner ear. In agreement with the standard model, we show that elevation of Dll1 expression in progenitor cells is crucial for adoption of the hair cell fate, and that Dll1 functions primarily in trans to activate Notch activity and to repress hair cell formation. However, we also found that some cells can differentiate into hair cells while contacting other Dll1-expressing cells. Although this indicates that some progenitor cells are not sensitive to lateral inhibition, we argue that this is unlikely to result from the cis-inhibition of Notch activity by Dll1 itself.

Although hair cells express several DSL ligands, Dll1 is thought to play the most significant role in the lateral inhibition of hair cell formation in the inner ear. In fact, the absence of Dll1 produces a much more severe overproduction of hair cells (Kiernan et al., 2005; Brooker et al., 2006) than that of Jag2 (Lanford et al., 1999), whereas Dll3 mutant mice have no discernible ear phenotype (Hartman et al., 2007). However, some questions remain about the mode of action of Dll1 in hair cell fate decisions. Are hair cell fate decisions dictated by progenitor cell-to-cell competition for Dll1 expression? Alternatively, does Dll1 operate only after hair cells are committed in order to protract hair cell formation? Is Dll1 able to inhibit the reception of Notch activity in cis within progenitor cells of the inner ear?

Here, our gain-of-function studies showed that the formation of hair cells was reduced within clusters of Dll1-expressing cells, but that a large proportion of isolated Dll1-expressing cells differentiated into hair cells. This confirms that Dll1 represses hair cell fate in trans, and that cells that express Dll1 have a greater chance of adopting the primary fate than those that do not. In standard lateral inhibition, ‘signal-sending’ and ‘signal-receiving’ cells compete for Dll1 expression, which is repressed by Notch activity in signal-receiving cells. We found that reversing this negative-feedback loop into a positive one using a Hes5 promoter-regulated Dll1 construct can convert the majority of signal-receiving into signal-sending cells. This shows that cells in which Notch is active can differentiate into hair cells if they succeed in elevating their endogenous levels of Dll1, and that the intercellular competition for Dll1 expression is one of the key determinants of hair cell fate decisions.

However, some data did not seem to fit with the standard model of lateral inhibition. First, not all isolated Dll1-expressing cells differentiated into hair cells. One possible explanation is that some progenitor cells are not competent to adopt the hair cell fate; ultimately, it is the expression of proneural genes, such as Atoh1 in the case of hair cells (Bermingham et al., 1999), that determines the adoption of the primary fate during lateral inhibition. Second, some cells (transfected with Dll1 or not) could differentiate into hair cells despite contacting other Dll1-expressing cells. The complete inhibition of hair cell formation was only observed within relatively large clusters of Dll1-expressing cells, which suggests that efficient repression of the hair cell fate is only achieved when several signal-sending cells cooperate to activate Notch in any given signal-receiving cell. It is also possible that some of the progenitor cells are refractory to Notch activity induced by Dll1. The underlying mechanisms could be extremely diverse, ranging, for example, from the reduction of cell surface levels of Notch receptors to interference with downstream components of the Notch signalling cascade. In addition, recent studies have highlighted the capacity of DSL ligands to cell-autonomously inhibit Notch activity; however, as we discuss below, several results suggests that Dll1 does not act in this way in the inner ear.

Gain-of-function studies have revealed that, in some developmental contexts, DSL ligands can inhibit Notch activity cell-autonomously (reviewed by del Álamo et al., 2011). Recent in vitro experiments using cell lines transfected with a Notch reporter have also shown that varying the levels of Dll1 in cis can indeed affect their response to Dll1 in trans (Sprinzak et al., 2010). Although the underlying mechanisms remain unclear, the commonly accepted model is that, when high levels of DSL ligands are present at the surface of a cell, they can bind in cis to Notch receptors, thereby preventing their activation in trans by other cells – a titration effect. Modelling studies have suggested that cis-inhibition could improve the robustness of cell fate decisions and might be essential for efficient patterning of epithelial sheets by lateral inhibition (Barad et al., 2010; Sprinzak et al., 2010), but is this the case in the inner ear?

In the present experiments, formation of hair cells within highly Dll1-transfected regions could be interpreted as evidence for cis-inhibition; however, hair cells within such regions were, in general, surrounded by untransfected supporting cells. Hence, mosaicism of transfection as well as cell mixing could explain why hair cell formation was not reduced in such regions. The reduction of hair cell density within Hes5::Delta1-transfected clusters provides additional evidence that, if cis-inhibition occurs, it does not prevent the reception of trans-inhibition; in fact, if Notch activity were blocked by Dll1 expression cell-autonomously, Hes5-regulated Dll1 expression would not be occurring in such clusters. Finally, the use of a tet-on inducible system confirmed that the strong elevation of Dll1 expression does not significantly reduce the activity of the Hes5 reporter in progenitor and supporting cells. Although this does not exclude the possibility that other intrinsic factors or DSL ligands expressed in progenitor cells inhibit Notch activity cell-autonomously, the present results provide strong evidence that Dll1 does not. Our data suggest instead that the ability of Dll1 to promote hair cell differentiation in cis results from its inhibitory impact on Dll1 expression in trans, according to the standard model of lateral inhibition with intercellular feedback.

Expression of Dll1 protein is absent or very low in Notch-active cells, peaks in cells with characteristics of very immature hair cells, but is downregulated in differentiated hair cells expressing HCA and otoferlin. This is in agreement with other studies showing temporally restricted waves of expression of Dll1, Jag2 and Dll3 in the auditory hair cells of mammals (e.g. Hartman et al., 2007) and suggests that Dll1/Notch-mediated lateral inhibition is short lived, occurring primarily between progenitor cells and immature hair cells. As hair cells differentiate, they might not deliver sufficient inhibition to prevent neighbouring cells from becoming hair cells. This would explain the transient contacts between mature-looking and immature hair cells in the immature basilar papilla (Goodyear and Richardson, 1997). One advantage of short-lived lateral inhibition is that it could facilitate the addition of new hair cells to inner ear epithelia over an extended developmental period, or even throughout life, as is the case in the avian vestibular system (Jørgensen and Mathiesen, 1988). However, this implies that additional mechanisms must operate downstream of lateral inhibition to eliminate patterning errors such as reciprocal hair cell contacts.

One attractive hypothesis is that differential adhesion properties of hair cell and supporting cells could lead to a progressive refinement of their relative position through homo- and heterotypic interactions (Goodyear and Richardson, 1997; Podgorski et al., 2007). In support of this idea, a recent study has shown that the cell adhesion molecules of the nectin family regulate cell patterning in the organ of Corti (Togashi et al., 2011). Other cell adhesion molecules, including Eph/ephrins and cadherin family members (see Warchol, 2007), can mediate cell sorting in other tissues (Steinberg and Takeichi, 1994; Xu et al., 1999) and may also contribute to the fine-grained patterning of the inner ear epithelia. A two-step model such as this would resemble the mechanism described for patterning of the Xenopus epidermis: lateral inhibition within the inner layer of the non-neural ectoderm establishes a pre-pattern of differentiation, which is then refined as ciliated cells migrate and intercalate into the superior layers of the epidermis (Deblandre et al., 1999). The mechanistic separation of cell fate decisions and fine-grained patterning events could also be advantageous during epithelial regeneration, for example after hair cell loss in the avian inner ear. In the damaged adult basilar papilla, Dll1 expression is upregulated in the newly formed hair cells, but not in the surviving ones (Stone and Rubel, 1999; Daudet et al., 2009). In the case of limited hair cell loss and regeneration, fine-grained patterning via differential cell affinities could provide a solution for the accurate positioning of a limited number of new cells within a pre-established cellular mosaic independently of Notch signalling.

A simultaneous occurrence of inductive (such as lateral inhibition) and morphogenetic (such as cell proliferation or cell adhesion) mechanisms, or ‘morphodynamic’ patterning strategy (Salazar-Ciudad et al., 2003), operates in the inner ear sensory epithelia. This raises new questions regarding the influences of one set of mechanisms over the other(s), in the inner ear as well as in other tissues in which Notch signalling operates. To answer these questions, live-imaging approaches as well as computational models in which the interaction of inductive and morphogenetic processes can be investigated (Podgorski et al., 2007) will be instrumental.

We thank Russ Collins, Jeff Corwin, Domingos Henrique, Koichi Kawakami, Ryoichiro Kageyama, Guy Richardson and Yoshiko Takahashi for sharing reagents; Brandon Nelson for helpful advice; and Julian Lewis for critical comments on the manuscript.