Ligand-dependent Notch signaling strength orchestrates lateral induction and lateral inhibition in the developing inner ear

Notch signaling plays a dual role during the sensory development of the inner ear, where it is required during early stages of prosensory specification and also for hair cell determination (Neves et al., 2013). The prosensory function of Notch relies on lateral induction, which is defined as the process by which a ligand-expressing cell stimulates those nearby to upregulate ligand expression, promoting ligand propagation and a coordinated cell behavior (Eddison et al., 2000). Hair cell determination results from lateral inhibition, whereby a ligand-expressing cell inhibits the expression of the ligand in the neighbors, thereby preventing them from adopting the same fate and generating a fine-grained cellular pattern (Bray, 2006). These two modes of operation rely, in each case, on the associated gene regulatory circuit. The prosensory function of Notch is mediated by the Notch ligand jagged 1 [Jag1; also known as serrate 1 (Ser1) in chick] (Eddison et al., 2000; Brooker et al., 2006; Kiernan et al., 2006; Daudet et al., 2007; Hartman et al., 2010; Pan et al., 2010; Neves et al., 2011). In the developing inner ear, Jag1 induces its own expression in adjacent cells and the expression of prosensory genes such as Sox2, resulting in the homogenous commitment of otic progenitors to the prosensory fate (Neves et al., 2011). By contrast, hair cell determination is driven by the Notch ligand delta 1 [Dl1; also known as delta-like 1 (Dll1) in chick] (Haddon et al., 1998; Daudet and Lewis, 2005; Kiernan et al., 2005; Brooker et al., 2006). Dl1 expression is thought to be induced by the transcription factor Atoh1, which initiates hair cell development (Mulvaney and Dabdoub, 2012). Atoh1 activates its own transcription (Helms et al., 2000) and is inhibited by Notch signaling (Takebayashi et al., 2007; Doetzlhofer et al., 2009). Lateral inhibition between cells mediated by Dl1 is resolved into a fine-grained pattern of hair and supporting cells (Collier et al., 1996). Notch target genes of the Hes family (Hairy and Enhancer of split paralogs) and Hes-related factors (Hesr, Hey/Herp genes) function as transcriptional repressors (Iso et al., 2003; Fischer and Gessler, 2007); they repress Atoh1 and are crucial for hair cell development (Zheng et al., 2000; Zine et al., 2001; Tateya et al., 2011; Du et al., 2013).

During the developmental window in which prosensory precursors transit to hair cell fate commitment, both Jag1 and Dl1 are expressed in the otic epithelium. This raises the question of how sensory precursors deal with the simultaneous presence of the two ligands, and how the final pattern is resolved. Lateral induction and lateral inhibition have been described as distinct circuits leading to specific cellular outputs of ligand propagation versus fine-grained patterning, respectively. Yet, there is no description of how these circuits cooperate and/or antagonize each other during development.

The present work shows that, in the chick inner ear, Jag1 induces weaker levels of Notch activity than Dl1, resulting in the differential expression of the target genes Hey1 and Hes5 as functional readouts. When both ligands signal together, competition arises and Jag1 decreases the overall signaling. Modeling indicates that, under such conditions, Jag1 operates as a partial agonist of Notch, effectively acting as a competitive inhibitor of Dl1/Notch signaling. In consequence, Jag1 inhibits its own expression in neighboring cells, thereby facilitating lateral inhibition and hair cell patterning. Experiments and modeling show that hair cell patterning is highly robust and that Jag1 biases, but does not determine, the supporting cell fate. The model attributes this robustness to Atoh1 autoactivation upstream of Dl1.

Hes/Hey factors are well-known targets of Notch signaling and are required for inner ear development (Zheng et al., 2000; Zine et al., 2001; Hayashi et al., 2008; Li et al., 2008; Tateya et al., 2011). We asked whether there is a relationship between the expression of the ligands Dl1 and Jag1 and the different Hes/Hey genes. In the chicken inner ear, Jag1 is expressed during both prosensory and hair cell differentiation stages, whereas Dl1 is only expressed during the latter (Adam et al., 1998; Morrison et al., 1999; Cole et al., 2000). From several Hes and Hey genes screened, only Hes5 and Hey1 mapped to the sensory domains. Sensory development follows a dorsal-to-ventral sequence and, at the stage shown in Fig. 1A (E5), the dorsalmost patches (cristae and macula utricularis) had initiated differentiation, whereas the ventralmost domains (macula sacularis and basilar papilla) were still prosensory. This allows different developmental stages to be observed in a single specimen. Hey1 was expressed along with Jag1 in the sensory epithelium, from the dorsalmost cristae to the maculae and the basilar papilla (pc, ms and bp in Fig. 1Aa-f). The expression of Hes5 differed from that of Hey1 and did not parallel Jag1. Hes5 was expressed only in the dorsalmost patches (pc, ac and mu in Fig. 1Ag,p,q), but not in the ventral domains (ms and bp in Fig. 1Ah,i,r) that exhibited a strong Jag1 signal (Fig. 1Ab,c,l). Interestingly, Hes5 expression matched well with that of Dl1 (Fig. 1A, compare m,n with p,q).

The expression patterns of Jag1, Dl1, Hes5 and Hey1 genes during prosensory and hair cell determination are illustrated in Fig. 1B. During prosensory stages, Jag1 and Hey1 showed a homogeneous cellular expression pattern (Fig. 1Ba,b), whereas during hair cell determination Dl1 showed a speckled pattern similar to that of Atoh1 expression in nascent hair cells (Fig. 1Bc,g,h). By contrast, Jag1, Hey1 and Hes5 were expressed in the basal layer that corresponds to supporting cells (Fig. 1Bd-f), complementary to the luminal layer occupied by MyoVIIa-positive hair cells (Fig. 1Bh).

These observations suggest that, throughout development, different Notch ligands result in the activation of different downstream genes. To test this, we analyzed the effects of Dl1 or Jag1 gain-of-function on Hey1 and Hes5 expression in the otic vesicle.

The timecourse of Hey1 and Hes5 mRNA levels after human JAG1 (hJag1) or chick Dl1 (cDl1) electroporation is shown in Fig. 2A,B. Hey1 was significantly induced by both ligands at all time points examined. Hes5 was also induced by Dl1 (Fig. 2B), but only weakly by Jag1, and delayed with respect to Hey1 (Fig. 2A). Hey1 induction was always stronger than that of Hes5 after Jag1 overexpression, whereas the effects of Dl1 on Hey1 and Hes5 were not significantly different after 6 h. This is illustrated by a comparison of the relative mRNA increase (fold increase) for each condition (Fig. 2C,D; supplementary material Fig. S1). In these experiments, the amount of transcribed hJag1 and cDl1 was checked to be equivalent (supplementary material Fig. S1Aa). The results suggest that Hey1 and Hes5 are differentially regulated by Jag1 and Dl1.

Given that both Dl1 and Jag1 activate the same signaling pathway, we sought possible explanations for their different actions on target genes. Since Notch1 is the only Notch receptor expressed in the inner ear of the chick embryo at the developmental stages under study (Adam et al., 1998; Abelló et al., 2007), the activation of different intracellular cascades for Dl1 and Jag1 is unlikely. One possible explanation is that the different ligands induce different strengths of Notch signal, which in turn result in different outputs. In the experiments that follow, we first explored whether different levels of active Notch differentially regulate Notch target genes and, second, whether Dl1 and Jag1 induce different levels of Notch activity.

We tested the effects of the loss and gain of function of Notch on Hey1 and Hes5 to then further analyze the quantitative relationship between Notch activity and target activation. We examined the expression of Notch ligands (Fig. 3A) and targets (Fig. 3B) after incubation with the γ-secretase inhibitor LY411575 (Ferjentsik et al., 2009). As expected from the different regulatory circuits (Notch inhibition or activation of the ligand), Notch blockade showed opposite effects on Dl1 and Jag1 (Fig. 3A). Also as expected for direct target genes, Hey1 and Hes5 expression was strongly repressed after Notch inhibition (Fig. 3B). Note that ∼20% of Hey1 expression was refractory to Notch inhibition, suggesting the presence of alternative regulatory mechanisms.

To test whether differences in Notch signaling strength impact Notch target selection, we measured the expression of Hey1 and Hes5 with different concentrations of intracellular Notch [mouse Notch1 intracellular domain (mNICD1, or NICD)] (Fig. 3C; see supplementary material Fig. S2 for absolute mRNA levels). Low NICD (<0.1 µg/µl) induced Hey1 but not Hes5 expression, whereas intermediate and high NICD levels (1-2.5 µg/µl) induced both Notch targets, with a preference for Hes5. This indicates that the threshold for Hey1 induction by NICD was lower than that for Hes5.

The different sensitivity of Hey1 and Hes5 to Notch signaling levels was further studied by overexpressing two hypomorphic Notch1 gain-of-function constructs (L1601P and L1601PΔP) that harbor a mutation in the heterodimerization domain of the Notch1 receptor (HD mutation). The two constructs are less potent than the wild-type NICD, as evaluated by 4xCSL-luciferase reporter (Chiang et al., 2008). Electroporation of either construct induced Hey1 but not Hes5 (Fig. 3D). L1601P and L1601PΔP phenocopied low levels of NICD (0.01 and 0.1 µg/µl; Fig. 3D). These results reinforce the notion that low levels of Notch activity are sufficient to induce Hey1 but not Hes5 in the inner ear.

The possibility that Dl1 and Jag1 induce different levels of Notch activity was explored by monitoring Notch reporter activity after ligand overexpression. We used two gene constructs in which DsRed or luciferase reporter genes were driven by multimeric CSL binding site repeats (12×CSL-DsRed or 8×CSL-Luc) (Jeffries et al., 2002; Hansson et al., 2006). These constructs responded to different levels of Notch activity in vitro (supplementary material Fig. S3A). They were active in the otic vesicle (supplementary material Fig. S3B-D), induced by NICD transfection (supplementary material Fig. S3C) and repressed by Notch blockade (supplementary material Fig. S3D). The results of cotransfection of the 12×CSL-DsRed or 8×CSL-Luc reporters with Dl1 or Jag1 are shown in Fig. 4A,B. DsRed reporter activity was measured by direct red fluorescence or by quantifying DsRed mRNA levels (Fig. 4A). Luciferase reporter activity was measured by a colorimetric enzymatic assay on protein extracts from otic vesicles (Fig. 4B). In all cases, levels of reporter activity were higher after Dl1 overexpression than after Jag1 by 2- to 6-fold, suggesting that the strength of the Notch signal induced by Jag1 is much lower than that induced by Dl1.

A key question in understanding the combined function of Jag1 and Dl1 in the sensory patches is to resolve which signal is evoked when both ligands are expressed together. To address this we modeled phenomenologically the signal induced by each ligand. We considered that each ligand drives the same type of signal but with different strengths (Fig. 2, Fig. 3, Fig. 4A,B), and that both ligands use common resources, such as the Notch1 receptor (Adam et al., 1998; Abelló et al., 2007) (Eqn 1). According to the model, when resources become limiting, competition between Dl1 and Jag1 results in an overall signal that is lower than the signal driven by Dl1 in the absence of Jag1 (Fig. 4C). This happens because Jag1 is a weak signaling ligand compared with Dl1 (⁠⁠, with being the ratio of Jag1-driven over Dl1-driven Notch saturated signal). In this case, Jag1 becomes an inhibitor of Notch signaling despite it driving signaling (Fig. 4C). This was experimentally confirmed by the co-expression of Jag1 and Dl1 together with Notch reporter constructs (Fig. 4A,B). Jag1 reduced the signal evoked by Dl1, suggesting that Jag1 and Dl1 compete for the activation of Notch. The combined Jag1 plus Dl1 signal was closer to that of Jag1 than to that of Dl1 (Fig. 4A,B), indicating that Jag1 has a higher apparent affinity than Dl1 for Notch (Fig. 4C,D).

Knowing that Jag1 drives a weaker signal than Dl1 and that it competes for signaling, we examined whether this is relevant for the transition from prosensory to sensory states. We developed a combined model for lateral induction and lateral inhibition driven by Jag1 and Dl1, respectively. As with lateral inhibition, lateral induction relies on Notch signaling but through opposite regulation of the ligand (supplementary material Fig. S4). This was confirmed by the inhibitory effect of the Notch blocker LY411575 on the induction of Jag1 by Jag1 (Fig. 5A and Fig. 3A). The model presented here is based on Notch signaling by Jag1 and Dl1 (Eqn 1), together with the circuits that regulate ligand expression (Eqns 2-4, Fig. 5B, Materials and Methods). We also introduced Atoh1 as the proneural gene that drives the initiation of Dl1 expression, the repression of Atoh1 by Notch (Takebayashi et al., 2007; Doetzlhofer et al., 2009) (Eqn 4), and its autoactivation (Helms et al., 2000). Hey1 and Hes5 were not included in the model for simplicity; they behaved similarly with respect to the regulation of ligands (not shown) and can be taken as simple readouts of the signal level.

First, we tested the model for Jag1 as the only ligand, a situation that occurs in the prosensory state. In this case, Jag1 drives bistability of homogeneous coherent states (Fig. 5C, green area of lateral induction), enabling Jag1 propagation (Fig. 5D; supplementary material Movie 1), in agreement with the work of Matsuda et al. (2012). This mimics the results of the gain of function of Jag1 (Hartman et al., 2010; Pan et al., 2010; Neves et al., 2011) (Fig. 5A). The bistable regime of lateral induction (i.e. the range of Jag1 production) enlarges with the strength of Jag1-driven signaling, and requires a minimal signaling strength (Fig. 5C).

The transition from the prosensory to the hair cell specification stages was simulated by the expression of Dl1 in a background of Jag1. For a weak Jag1-driven signal (⁠⁠), an appropriate fine-grained patterning of lateral inhibition emerged from a Jag1 homogeneous state upon activation of Atoh1 expression (Fig. 5E, blue area). This pattern was formed by cells expressing Atoh1 and Dl1 (hair cells), which were surrounded by cells expressing Jag1 (supporting cells, Fig. 5F). Atoh1 autoactivation conferred robustness to its maintenance (supplementary material Fig. S5). The model, therefore, reproduced correctly the hair cell determination state. In addition, it revealed that the differential signaling strength between Jag1 and Dl1 was crucial for the transition from the prosensory to the sensory states. Equal signaling strength for both ligands (⁠⁠) enabled hair cell patterning only if Jag1 expression was reduced so as to forbid the emergence of supporting cells expressing Jag1 (Fig. 5G, left). By contrast, a weak Jag1 signaling strength facilitated hair cell patterning (Fig. 5G, middle and right). This is because Jag1, as a weak signal inducer, behaves as a partial agonist of Notch, i.e. Jag1 inhibits Notch signaling in the presence of Dl1 (Fig. 4). Under these circumstances, adjacent cells expressing both Jag1 and D1 mutually inhibit each other, even for low Dl1 levels (supplementary material Fig. S6 and Movie 2). This behavior is in contrast to the mutual activation that Jag1 drives between adjacent cells when expressed alone (supplementary material Fig. S6).

These results indicate that weak Jag1 signaling (⁠⁠) may drive the prosensory state (Fig. 5C) and also facilitates hair cell patterning (Fig. 5G). Only a band of weak Jag1 signaling strengths enables the transition between these two states as triggered by Atoh1 (Fig. 5E).

The model predicts that Jag1 facilitates lateral inhibition by competing with Dl1 signaling. To test this, we analyzed the effects of manipulating Jag1 levels at the onset of hair cell determination. hJag1 or Jag1 siRNA were expressed in E3.5 chicken otocysts and the sensory domains examined 3 days after electroporation (Fig. 6). In none of these conditions was hair cell patterning substantially altered (Fig. 6A, whole-mount preparations in the top row). The general organization of the patch and the density of hair cells were similar in all conditions (compare control and Jag1 EP in Fig. 6A, bottom row, and 6B). The loss of function of Jag1 reduced Sox2 expression (supplementary material Fig. S7) and slightly decreased the density of hair cells (Fig. 6B). These results suggest, contrary to expectations, that Jag1 is not involved in hair cell patterning.

To clarify this point, we evaluated model predictions further by mimicking gain- and loss-of-function experiments with numerical simulations (Materials and Methods). As in vivo, hair cell patterning persisted in both cases (Fig. 6C; see supplementary material Fig. S8 for an exploration of the range of validity of the model). When the same simulations were performed in the absence of Atoh1 autoactivation, however, hair cell patterning was totally disrupted (supplementary material Fig. S9). This indicates that Atoh1 autoactivation buffers the role of Jag1 by providing robustness to the patterning process.

If patterning is so robust, is there a signature for Jag1 as a facilitator of hair cell patterning? Numerical simulations revealed a consistent trend: adjacent Jag1-electroporated cells and non-electroporated cells tended to become supporting and hair cells, respectively (Fig. 6C). This resulted in a tendency of Jag1-positive cells to avoid the hair cell fate and become supporting cells, which was particularly evident at low electroporation densities, where the probability of contacts between electroporated and non-electroporated cells was high (Fig. 6C). We reasoned that this behavior resulted from the reduction of Notch signaling elicited by Jag1 when competing with Dl1. Cells carrying Jag1 reduced the Notch signal in neighboring non-electroporated cells and released the Notch-mediated Atoh1 inhibition, thereby promoting the hair cell fate choice. This, in turn, favored Jag1-delivering cells to become supporting cells. The opposite trend, although weaker, was found for numerical simulations of loss of function (Fig. 6C; supplementary material Fig. S8). In summary, modeling suggests that weak Notch signaling driven by Jag1 results in a specific bias in hair cell and supporting cell determination, which becomes the signature feature of Jag1 in facilitating hair cell patterning.

We then asked whether this signature is measurable in vivo by a more detailed analysis of the Jag1 gain- and loss-of-function experiments in the chick embryo (Fig. 6D). The experiments show that cells carrying the hJag1 transgene were less likely to differentiate as hair cells, as predicted by the model. The proportion of hair cells carrying hJag1 was always smaller than the total fraction of electroporated cells (Fig. 6D). As in the numerical simulations, this was more evident at moderate electroporation densities (Fig. 6D, low EP). Also in agreement with the model, loss-of-function experiments showed a weak opposite trend (Fig. 6D). Therefore, the signature of Jag1 as a facilitator of hair cell patterning was evident in vivo as a cell fate bias.

The results show that Dl1 and Jag1 drive Notch signaling at different strengths, eliciting differential expression of Hey1 and Hes5. The expression of Hes5 and Hey1/2/L and their sensitivity to gamma-secretase inhibitors has also been reported in the mouse (Hayashi et al., 2008; Doetzlhofer et al., 2009). Interestingly, Hes5 is more sensitive than Hey1 to treatment with DAPT, suggesting that it requires higher levels of intracellular Notch activity (Doetzlhofer et al., 2009). Moreover, Hayashi et al. (2008) showed that the concentration of DAPT required to inhibit Notch signaling during lateral inhibition is lower than for the prosensory phase, suggesting that Hes5 and lateral inhibition share a similar sensitivity to Notch.

Alternative cellular behaviors dependent on Notch levels have been reported in relation to the decision between cell proliferative and cell arrest states (Mazzone et al., 2010; Perdigoto et al., 2011; Ninov et al., 2012). In the prosensory patches, sensory progenitors proliferate (Murata et al., 2009), whereas in the differentiating sensory organs the hair cells exit the cell cycle and differentiate (Chen and Segil, 1999) while supporting cells enter a quiescent state (Oesterle and Rubel, 1993). One possibility is that gene regulation and cellular function depend on the different levels of Notch signaling elicited by the different ligands. Recently, Liu et al. (2013) showed that Notch activity is almost undetectable during prosensory stages, but increases during hair cell determination. This fits well with our results and with the notion that the prosensory state is driven by Jag1 and that hair cell patterning involves strong Dl1 signaling.

In the inner ear, expression patterns and functional studies suggest that lateral induction and inhibition are associated with different Notch ligands that initiate signaling, with Jag1 driving lateral induction and Dl1 lateral inhibition (Brooker et al., 2006; our present results). The association of Dl1 with lateral inhibition is a general theme in neural development (Henrique et al., 1995; Adam et al., 1998; Kageyama et al., 2010) and that of Jag1 with lateral induction has been shown also in the lens (Le et al., 2009), developing pancreas (Golson et al., 2009), early hematopoiesis (Robert-Moreno et al., 2008) and angiogenesis (Benedito et al., 2009). However, this correspondence does not seem to hold for all situations; for example, Jag1 selects V1 neuroblasts in the neural tube by lateral inhibition (Ramos et al., 2010). In the inner ear, Jag1 and Dl1 are oppositely regulated by Notch signaling, which readily accounts for their association into the circuits of lateral induction and of lateral inhibition, respectively. The inhibition of Dl1 by Notch has been associated with the repressor effect of Hes/Hey genes on bHLH proneural genes (Kageyama et al., 2010), but the activation of Jag1 by Notch remains poorly understood (Katoh, 2006).

Modeling predicts that the function of Jag1 alone is distinct from that in the presence of Dl1 because of competition for common resources and their different signaling strengths. Therefore, it is likely that context determines the behavior of Dl1 and Jag1 ligands in different tissues. The interaction of different ligands with the Notch receptor is modulated by various factors, such as Fringe glycosylation that potentiates Dl1-induced Notch signaling while hampering Jag1 (Bruckner et al., 2000; Haines and Irvine, 2003; Benedito et al., 2009). Lunatic fringe is expressed in sensory regions of the inner ear (Morsli et al., 1998; Cole et al., 2000) and is a likely candidate to maintain low-level Jag1-driven signaling in sensory domains.

The model presented in this paper suggests that Notch levels play a key role in resolving the confrontation between Jag1 and Dl1 in otic sensory progenitors. But, in addition, the model yields a non-trivial behavior according to which Jag1 actually facilitates lateral inhibition patterning. The specific role of Jag1 results from the competition between Dl1 and Jag1 for Notch signaling through the Notch receptor. When Jag1 signals less than Dl1, Jag1 may act as a dominant-negative or a partial agonist of the Notch receptor (Buchler et al., 2003), reducing overall Notch signaling. This situation resembles that found in cis-inhibition of Notch signaling, in which Dl1 ligand in a cell competes with Dl1 ligand in neighboring cells to bind to Notch receptor (Formosa-Jordan and Ibañes, 2014). In the context of inner ear development, cis-inhibition does not occur (Chrysostomou et al., 2012), but the competition between Dl1 and Jag1 ligands results in a similar effect on the signal. Our results indicate that, upon Atoh1 expression, Jag1 switches from increasing overall signaling and driving lateral induction to effectively decreasing Notch signaling and facilitating hair cell patterning. This facilitation arises from the mutual inhibition between adjacent equivalent cells driven by Jag1 when competing with Dl1 (supplementary material Fig. S6).

Given the robustness of hair cell patterning, the signature of Jag1 is a bias in the cell fate rather than a large disturbance of the pattern. We did not observe the effects described in mouse cochlear explants by Zine et al. (2000), who showed an excess of outer hair cells after a 5-day treatment with antisense oligos targeted against Jag1. Our work focuses instead on chick vestibular maculae, in which, in our hands, Jag1 loss results in the loss of supporting cell identity, although this does not seem sufficient to switch to the hair cell fate. The potential effects of newly added cells during patterning and of other morphogenetic mechanisms remain to be elucidated.

The results suggest that the robustness of hair cell patterning to changes in Jag1 expression arises mainly from the autoactivation of Atoh1. Although Atoh1 autoactivation neither facilitates nor promotes pattern initiation, Atoh1 autoactivation maintains patterning once it is established and stabilizes the final pattern. This is in agreement with recent results showing that once sensory progenitors start to highly express Atoh1 and subsequently Dl1, they cannot be prevented from becoming hair cells (Driver et al., 2013). This might also underlie the observation by Chrysostomou et al. (2012) of the generation of hair cells in direct contact with several neighboring cells expressing high levels of Dl1.

In summary, the combination of opposed feedback regulatory mechanisms, differential signaling strength and competition between Jag1 and Dl1 for Notch is crucial for orchestrating lateral induction and lateral inhibition during ear development. Through the positive loop of Jag1, Notch establishes a coherent domain of low Notch activity, in which Notch signaling is expanded by lateral induction. Notch, in turn, induces Sox2 expression and bHLH targets that prevent differentiation. The result is the specification of the prosensory patches. Upon Atoh1 expression, both Dl1 and Jag1 mediate the inhibition of neighbors, generating the hair cell/supporting cell lattice, and patterning follows the rules of lateral inhibition. In other words, Jag1 exhibits a new function in facilitating lateral inhibition and hair cell patterning; this is because, in the presence of Dl1, both ligands compete for the same resources. Since Jag1 is a weaker activator, it acts as a partial agonist of Notch receptor and effectively inhibits signaling. If both Dl1 and Jag1 were to signal with the same strength, lateral induction and lateral inhibition would conflict and the pattern would be disrupted. Instead, weak Jag1 signaling enables Jag1 to drive prosensory patches when acting alone, but then to switch its role upon Dl1 expression.

A DNA mix containing the desired vector(s) mixed with Fast Green was injected onto the otic cup of HH12-14 chick embryos (Neves et al., 2011) or into HH20-21 otic vesicles (Kamaid et al., 2010). Jag1 overexpression in 3-day experiments was confirmed by immunohistochemistry. Animal procedures were approved by the Ethical Committee of the Parc de Recerca Biomèdica de Barcelona.

Stealth cJag1 RNAi or scrambled Stealth controls (Life Technologies; supplementary material Table S1) were mixed with GFP or DsRed and Fast Green and injected at 30 µM into HH20-21 otic vesicles. The effects were assessed 1-3 days after electroporation (Fig. 6). Efficiency of Jag1 silencing was assessed by immunostaining (supplementary material Fig. S7). Stealth RNAi is stabilized against nuclease degradation and stable for at least 3 days in culture (Life Technologies).

Otic vesicles were dissected from HH20 chick embryos, grown in culture (Pujades et al., 2006) and incubated with either DMEM or 100 nM LY411575 (Dr Kim Dale, University of Dundee, UK) for 6 h (Fig. 3A,B). In some experiments, HH12-14 otic cups were electroporated with hJag1 and allowed to develop for 4 h in ovo. Electroporated and control otic vesicles were then dissected and further cultured with DMEM or media supplemented with 100 nM LY411575 for 12 h (Fig. 5A).

Embryos were processed for ISH and IHC as described (Acloque et al., 2008; Neves et al., 2007). Primary antibodies were: anti-Jag1 rabbit polyclonal (Santa Cruz Biotechnology, sc-8303, H-114, 1:50), anti-GFP rabbit polyclonal (Clontech, 632460, 1:400), anti-MyoVIIa mouse monoclonal (Developmental Studies Hybridoma Bank, 138-1, 1:100), anti-Sox2 goat polyclonal (Santa Cruz, Y-17, 1:400) and anti-DsRed rabbit polyclonal (Takara, 632496, 1:400). Secondary antibodies were Alexa Fluor 488-, 568- and 594-conjugated anti-mouse, anti-goat and anti-rabbit (Molecular Probes Invitrogen, 1:400). Sections were counterstained with DAPI (100 ng/ml, Molecular Probes) and mounted in Mowiol (Calbiochem). Fluorescence was analyzed by conventional fluorescence microscopy using a Leica DMRB with Leica CCD camera DC300F and images were processed with Adobe Photoshop. Some otic vesicles were photographed and analyzed without sectioning. Fluorescence of each otic vesicle was measured and corrected for the area electroporated using ImageJ (Fig. 4A): corrected total fluorescence (CTF)=integrated density−(electroporated area×mean background fluorescence).

RNA was isolated using the RNeasy Micro or Mini Kit (Qiagen) including a step of in-column digestion with DNase. cDNA was synthetized from 15 ng RNA with Superscript III DNA polymerase (Invitrogen) and random primers (Invitrogen), and reverse transcribed in duplicate. qRT-PCR was carried out using 1 μl cDNA, SYBR Green Master Mix (Roche) and gene-specific primers (Invitrogen; supplementary material Table S2) in a LightCycler480 (Roche). Each transcribed in duplicate cDNA was used as template for each pair of primers in triplicate PCR reactions.

Protein extracts were prepared using Reporter Lysis Buffer (Promega; 10 µl per otic vesicle). β-Gal activity and luciferase activity were determined as previously described (Neves et al., 2012). Luciferase activity was normalized for the level of transfection using the β-Gal enzymatic reaction.

Results are shown as mean±s.e.m. of n experiments as indicated in figure legends. Statistical significance was assessed using Student's t-test.

Numerical integration of the dynamics (Eqns 1-4) was performed with Runge-Kutta methods of the fourth order with time step 0.1 until the stationary state was reached, using a custom Fortran77 program (Fig. 5D-G and Fig. 6C; supplementary material Figs S5,S6,S8,S9). Supplementary material Simulation 1 is a Mathematica notebook [version 9.0, Wolfram Research; code provided in pdf and Mathematica notebook (.nb) format] also used for integration. Regular and irregular two-dimensional lattices of 12×12 or 18×18 cells with periodic boundary conditions were used. Irregularity parameter value was (Formosa-Jordan et al., 2012). Initial condition for each molecular species for the -th cell was ⁠, with being a uniform random number between −0.5 and 0.5 and a stationary homogeneous solution. LSA results were checked for different parameter values through numerical integration of the dynamics (data not shown).

⁠, ⁠, and values were represented in linear logarithmic grayscale. Cells with ⁠, ⁠, and were represented as white cells. Hair cells were defined as those expressing Atoh1 above a threshold and were represented in red. This threshold was the Atoh1 value at the stationary homogeneous state with intermediate Atoh1 level of the corresponding parameter values unless otherwise stated.

We performed numerical integration of Eqns 1-4 on 3×3 perfectly hexagonal cells with periodic conditions and initial condition ⁠, with being defined as above and a stable fine-grained pattern solution. Pattern maintenance was considered to hold when the final state had between 20% and 80% of the cells in a high Atoh1 state (i.e. high Atoh1 levels were at least 1% higher than the minimal levels of Atoh1 in the same tissue).

Integration of Eqns 2-4,6,7 was performed. Electroporated cells correspond to for and for and ⁠. Non-electroporated cells have ⁠. For loss of function, Eqns 1-4 were integrated with at in electroporated cells. Bar charts show averages of the results and s.e.m. for three random patterns of electroporations. Control electroporations correspond to ⁠. Simulations assume a constant cell number, therefore, the potential effect of newly added cells was not considered.

This work is dedicated to the memory of Julian Lewis, whose work has been an example of scientific merit and a source of inspiration to the authors. We thank Domingos Henrique, Anna Bigas and Saúl Ares for critical reading of the manuscript, and Andrés Kamaid for kindly sharing results. LY411575 was made in the College of Life Sciences, Dundee University, and is a kind gift from Kim Dale. Warren Pear kindly provided NICD mutant constructs. We thank Marta Linares, Andreu Alibés and Miquel Sas for technical assistance.