The Prosensory Function of Sox2 in the Chicken Inner Ear Relies on the Direct Regulation of Atoh1

The proneural gene Atoh1 is crucial for the development of inner ear hair cells and it requires the function of the transcription factor Sox2 through yet unknown mechanisms. In the present work, we used the chicken embryo and HEK293T cells to explore the regulation of Atoh1 by Sox2. The results show that hair cells derive from Sox2-positive otic progenitors and that Sox2 directly activates Atoh1 through a transcriptional activator function that requires the integrity of Sox2 DNA binding domain. Atoh1 activation depends on Sox transcription factor binding sites (SoxTFBS) present in the Atoh1 3′ enhancer where Sox2 directly binds, as shown by site directed mutagenesis and chromatin immunoprecipitation (ChIP). In the inner ear, Atoh1 enhancer activity is detected in the neurosensory domain and it depends on Sox2. Dominant negative competition (Sox2HMG-Engrailed) and mutation of the SoxTFBS abolish the reporter activity in vivo. Moreover, ChIP assay in isolated otic vesicles shows that Sox2 is bound to the Atoh1 enhancer in vivo. However, besides activating Atoh1, Sox2 also promotes the expression of Atoh1 negative regulators and the temporal profile of Atoh1 activation by Sox2 is transient suggesting that Sox2 triggers an incoherent feed-forward loop. These results provide a mechanism for the prosensory function of Sox2 in the inner ear. We suggest that sensory competence is established early in otic development through the activation of Atoh1 by Sox2, however, hair cell differentiation is prevented until later stages by the parallel activation of negative regulators of Atoh1 function.


Introduction
The inner ear provides the brain with accurate information on mechanical perturbations that result in the perception of sound and balance. Mechano-electrical transduction is initiated by the highly specialized hair cells, which transmit electrical signals to the primary afferent neurons that convey this information to the brain. There is good evidence that hair cell fate depends on the function of the proneural factor Atoh1, that behaves as a master gene for hair cell differentiation [1,2,3]. Atoh1 is an Helix-Loop-Helix (HLH) transcription factor regulated through a positive autoregulatory loop that maintains its expression in the sensory precursors [4,5], and through the negative regulation of other HLH proteins that prevent Atoh1 expression and function [5,6,7,8,9,10,11,12]. Yet, the molecular mechanisms underlying the onset of Atoh1 expression remain obscure.
Sox2 is a High Mobility Group (HMG) box domain transcription factor that belongs to the B1 subfamily of Sox proteins [13] and it behaves as a transcriptional activator [14]. Sox2 shows two seemingly contradictory functions in the developing inner ear. On one hand, it is expressed in neurogenic and sensory progenitors [15,16,17] and it is necessary for hair cell development [18]. Misexpression of Sox2 results in an increased number of neurons and ectopic hair cells [19,20]. On the other hand, Sox2 counteracts Atoh1 function and prevents hair cell formation when over-expressed in sensory precursors [21]. This is reminiscent of the function of SoxB1 genes in the Central Nervous System (CNS), where they promote neural competence but prevent neuronal differentiation [22,23,24]. Since neural commitment depends ultimately on the expression of proneural genes, the general question arises as to how Sox2 regulates proneural gene function.
In the present work, we show that Sox2 directly activates Atoh1 transcription in the early otic vesicle, providing a molecular mechanism for the prosensory function of Sox2 in the inner ear. Besides, we found that Sox2 regulates Atoh1 through an incoherent logic that promotes the expression of both Atoh1 and Atoh1 negative regulators. We suggest that as a result of this dual interaction, otic progenitors are committed to sensory fate early in development, but their differentiation deferred until later stages.

Site directed mutagenesis
The mutated reporter constructs Atoh1enhmut-BG-EGFP and Atoh1enhmut-BG-ZA were generated using the QuickChangeH Site-Directed Mutagenesis Kit (Stratagene). Briefly, mutually complementary primers (Invitrogen, sequence available upon request) aligning with the region of the Atoh1 enhancer containing the SoxTFBS were designed according to the manufacturer's instructions to create three point mutations. The mutated reporter construct was replicated in a PCR reaction and the parental DNA digested with DpnI. Undigested mutated constructs were amplified in bacterial hosts and sequenced to detect the insertion of the desired mutation before using in subsequent functional assays.

Chicken (Gallus gallus) embryos and in ovo electroporation
Fertilized hens' eggs (Granja Gibert, Tarragona, Spain) were incubated at 38uC for designated times and embryos were staged according to Hamburger and Hamilton [26]. HH12-14 chicken embryos were electroporated in ovo with the desired vector (1 mg/ ml, for Sox2 expression vectors, 1,5 mg/ml for Atoh1 reporter; 2 mg/ ml for Sox2 reporter) mixed with fast green (0.4 mg/ml) that were injected onto the otic cup by gentle air pressure through a fine micropipette. Square pulses (8 pulses of 10 V, 50 Hz, 250 ms) were generated by an electroporator Square CUY-21 (BEX Co., LTd, Tokiwasaiensu, Japan). Focal electroporation of HH20-21 otic vesicles was performed in ovo, using a method modified from Chang et al. [27].

HEK293T cell transfection
HEK293T cells were cultured in DMEM supplemented with glutamine, antibiotics and 10% fetal bovine serum. Before transfection, cells were cultured in serum and antibiotics-free medium. For transfection, the DNA was mixed with Polyethylenimine 1 mg/ml (PEI, Polysciences Inc, PA, USA) at the ratio of 4 ml of PEI/mg of DNA, incubated twenty minutes at room temperature and finally added to the cell culture. For Atoh1 enhancer activity assays, 1 mg of Sox2 expression vector (or Sox2HMG-VP16 or Sox2DHMG) was co-transfected with 0,5 mg of Atoh1eh-BG-ZA and 0,2 mg of pCMV-Luciferase for bgal activity assays, or 0,5 mg of Atoh1eh-BG-EGFP and 0,2 mg of pDsRed for direct fluorescence assays. For Western blot and qRT-PCR analysis, 1 mg of Sox2 expression vector was co-transfected with 0,2 mg of pCIG-EGFP.

Quantitative real time PCR (qRT-PCR)
Eight to twelve otic vesicles were dissected and total RNA isolated using RNeasy Mini kit (Qiagen). For HEK293T cells, total RNA from 6-well plates was isolated with a standard Trizol extraction (Invitrogen). Retrotranscription of 15 ng (chicken samples) or 1 mg (HEK293T samples) of purified mRNA was used to synthesize cDNA with Superscript III DNA polymerase (Invitrogen) and random primers (Invitrogen). Real time PCR was carried out using SybrGreen master mix (Roche), 1 ml of retrotranscribed cDNA and specific primers sets for each gene (Invitrogen, primer sequences are available upon request), in LightCycler480 (Roche). cGAPDH and hPum1 were used as calibrator genes for chicken and HEK293T samples, respectively. Expression levels of each gene were normalized to the calibrator gene and then referred to the levels in control samples, which were arbitrarily set to 1. Transcription levels were further normalized to co-transfected GFP. Quantitative real-time PCR experiments were performed with cDNA from three independent biological replicates.
bGal and luciferase enzymatic assays Protein extracts from cells were prepared using Reporter Lysis buffer (Promega) according to the manufacturer's instructions. For bGal activity, triplicates of each protein extract (10 ml) was mixed with 90 ml bGal staining solution (100 mM PBS, 100 mM MgCl 2 , 4 mg/ml ONPG, 4,5 M bmercaptoethanol) in a 96-well ELISA plate and incubated for 2-20 h at 37uC. bGal activity was determined by the absorbance at 420 nm in a microplate reader (VERSAmax, Molecular Devices, Cape Cod). For luciferase activity, 10 ml of each protein extract was mixed with 20 ml of Luciferase Assay Reagent (Promega) and activity was determined with a Luminescence Microplate Reader (Clarity, BioTek). For each well, bGal activity was normalized for the level of transfection using luciferase activity and then the values in transfected samples were referred to the corresponding control, which was arbitrarily set to 1. Enzymatic activity was measured with protein extracts from three independent biological replicates.

Chromatin Immunoprecipitation (ChIP)
HEK293T cells or dissected otic vesicles were processed for ChIP as previously described [28]. Briefly, formaldehyde crosslinked cell or tissue extracts were sonicated in a Bioruptor (Diagenode), and the chromatin fraction incubated overnight with 5 mg of either Goat IgG (Purified Immunoglobulin, Sigma, I9140) or a-Sox2 goat polyclonal antibody (Santa Cruz Biotechnology, Inc., sc-17320, Y-17) in RIPA buffer, and precipitated with protein A/G-Sepharose (Amersham). Cross-linkage of the co-precipitated DNA-protein complexes was reversed, and DNA was analyzed by qRT-PCR as described above. Primers used to detect the different regions of chromatin are available upon request.

Results analysis and statistics
qRT-PCR analysis, reporter enzymatic activity and in vitro ChIP assays were performed with three independent biological replicates. In vivo ChIP assays were performed with two independent biological replicates. The results are shown as mean6SE for one typical experiment, and statistical significance was assessed using Students' t test applied to the three independent experiments. p,0,001 is labeled with ***, p,0,005 is labeled with ** and p,0,05 is labeled with *. n.s., non significant.

Hair cells and neurons derive from Sox2-positive progenitors
Previous work suggests that Sox2 promotes the competence to generate neurons and hair cells in the otic vesicle [19,20]. This predicts that in the embryo, both cell types derive from Sox2positive progenitors. To analyze this possibility, we electroporated the NOP-2-EGFP in HH12 chicken embryos and followed the fate of the progeny with specific markers. The NOP-2-EGFP construct contains the EGFP reporter gene under the control of a Sox2 enhancer that drives expression specifically in otic and nasal placodes [25]. The stability of EGFP provides a cumulative labeling of cells that expressed Sox2 throughout the experiment and, hence, the lineage of Sox2-expressing progenitors ( Fig. 1A-F). In 11 samples, EGFP-positive cells were detected both in the prosensory domain (compare B and C) and in the cochleovestibular ganglion (dotted line, B). Neuronal fate of the Sox2 progeny was confirmed by co-labeling with Islet1 antibody (D, n = 4), and that of hair cells by co-labeling with MyoVIIa and Hair Cell Specific (HCA) antibodies (E and F, n = 4). The results indicate that both hair cells and neurons derive from Sox2-positive progenitors.

Sox2 induces the transcription of Atoh1
Hair cell formation depends on the function of the proneural gene Atoh1 [3], but it is not known which factors regulate the onset of Atoh1 expression in the ear. Since Sox2 function is required for Atoh1 expression and hair cell formation, we asked whether Sox2 was able to induce Atoh1 expression. HEK293T cells were used as a convenient model system for analysis of molecular interactions before testing their biological significance in vivo. HEK293T cells endogenously expressed Atoh1 and Sox2 mRNAs and proteins ( Fig. 2A upper). Accordingly, Atoh1 transcriptional activity was detected after transfection with either EGFP or LacZ Atoh1 reporter constructs ( Fig. 2A, middle photograph and bar diagram, respectively). They contain the reporter genes under the control of Atoh1 enhancer elements that reside 39 of the Atoh1 coding sequence and are sufficient to recapitulate the endogenous Atoh1 expression in several species, including the chicken [4,29,30]. Overexpression of Sox2 increased Atoh1 enhancer reporter activity as measured either by bGal activity on cell extracts (Fig. 2B, left bar diagram) or by EGFP fluorescence (Fig. 2B, photographs on the bottom left), confirming previous observations by Neves et al. [20]. Similarly, Sox2 transfection resulted in an increase in Since Sox2 is an activator transcription factor [14], the effects of Sox2 on Atoh1 transcription should be dependent on both DNAbinding and transcriptional activator function. HEK293T cells were co-transfected with the Atoh1 reporter and with either Sox2HMG-VP16 or Sox2DHMG (Fig. 2C, left diagram). The Sox2DHMG lacks the DNA binding domain and its cotransfection had no effect on Atoh1 reporter activity (Fig. 2C, grey bar). This shows that the regulation of Atoh1 requires the binding of Sox2 to DNA. The Sox2HMG-VP16 construct contains the Sox2 DNA binding domain fused to a potent trans-activator domain. The co-transfection with Sox2HMG-VP16 reproduced the effects of Sox2 on Atoh1 (Fig. 2C, blue bar). These experiments show that Sox2 is able to induce Atoh1, that this depends on the function of Sox2 as an activator transcription factor, and that it requires Sox2 binding to DNA.

Sox2 directly binds to the Atoh1 enhancer
In order to test the possible binding of Sox2 to the Atoh1 regulatory regions, the enhancer sequence of Atoh1 was screened using Transfac database in rVista software and two overlapping Sox Transcription Factor Binding Sites (SoxTFBS) were found. They were conserved among human, mouse and chicken, mapping to the 39 end of the Atoh1 enhancer A (Fig. 3A). In order to test the interaction between Sox2 and these binding sites, we performed a ChIP assay. Chromatin from HEK293T cells was immunoprecipitated with a Sox2 antibody and analyzed for the presence of the SoxTFBS with specific primers for the corresponding region of the Atoh1 enhancer. As controls, we used two regions located 5 kb upstream and downstream of the binding sites. Chromatin precipitated with Sox2 antibody was enriched in the SoxTFBS region of Atoh1 enhancer when compared to the chromatin precipitated with a goat IgG antibody (Fig.3B). Furthermore this enrichment was specific for this region of the chromatin and not detected in the control sites (n = 3).
Site-directed mutagenesis was used to evaluate whether the induction of Atoh1 by Sox2 was dependent on binding to these SoxTFBS. Briefly, we introduced three point mutations in the Atoh1 enhancer reporter construct, which destroys the ability of Sox2 to bind to the conserved SoxTFBS (Fig. 3C, left diagram). Co-transfection of Sox2 with the mutated Atoh1 enhancer reporter reduced bGal activity to half of the value obtained after cotransfection with the native Atoh1 enhancer reporter (Fig. 3C, right bar diagram, n = 3, native and mutated reporter activities compared in the same experiment). Interestingly, the mutation of SoxTFBS did not result in the complete reduction Atoh1 reporter activity to control values. This is likely due to the induction of endogenous Atoh1 protein after Sox2 transfection (See Fig.2B). Atoh1 is able to regulate its own expression through the 'E- box' in the Atoh1 enhancer [4], and this site was intact in the construct.
In summary, these experiments show that Sox2 directly binds to the Atoh1 enhancer and that the regulation of Atoh1 by Sox2 is, at least in part, mediated by the SoxTFBS present in the Atoh1 regulatory regions.

Atoh1 transcriptional activity in the otic vesicle depends on Sox2
Sox2 is expressed throughout the neurosensory domain of the otic vesicle [15] and we sought to analyze whether this is able to activate Atoh1 transcription, by using the Atoh1enh-BG-EGFP reporter in vivo. Otic vesicles were electroporated with this reporter construct together with the tracer pDsRed (Fig. 4A-D). Atoh1 reporter was active in the otic vesicle but spatially restricted to the anterior-medial domain ( Fig. 4A and C, n = 19 otic vesicles), corresponding to the Sox2-positive expression domain (Fig. 4D, n = 10 otic vesicles). Note that electroporated cells in the surface ectoderm and lateral aspect of the otic vesicle remained GFPnegative (asterisk in Figs. 4A-B and H-I). Reporter activity was also detected in the neuroblasts of the cochleo-vestibular ganglion, which is consistent with the previous observation these neurons derive from Sox2-positive progenitors (see Fig.1) and suggests that Atoh1 transcription is also activated by Sox2 in this type of progenitors (arrows in Fig. 4C). Later in development, the activity of the reporter was restricted to the nascent hair cells within the sensory patches (HH24, Fig. 4E-G, n = 11 otic vesicles).
We next tested whether the observed Atoh1 reporter activity in the otic vesicle indeed depended on Sox2 (Fig. 4H-O). For this purpose we co-electroporated Atoh1enh-BG-EGFP with Sox2HMG-Engrailed, which suppresses Sox2 function as a dominant negative [23]. This resulted in the suppression of the Atoh1 reporter activity (Fig. 4H-K) suggesting that the early activation of Atoh1 transcription is dependent on Sox2. Furthermore, the electroporation of the Atoh1enh-BG-EGFP reporter construct carrying the mutation in the SoxTFBS (Atoh1enhmut-BG-EGFP, see  for mutation) resulted in none (Fig. 4N, n = 17/23 otic vesicles) or very low (n = 6/23 otic vesicles) reporter activity in the otic vesicle. This was evaluated by comparing EGFP expression after electroporation of the native Atoh1 reporter (Fig. 4L, n = 9 otic vesicles) and the mutated Atoh1 reporter (Fig.4N, n = 23 otic vesicles) for otherwise equivalent electroporations (pDsRed in Fig. 4M and O). Together, these experiments suggest that Sox2 switches on Atoh1 transcriptional activity in the early otic vesicle.
Since Atoh1 transcription is active in the neurosensory domain of the otic vesicle, one critical question is whether Sox2 binds to the endogenous Atoh1 enhancer during normal development. In order to test this possibility, we performed ChIP assay in vivo on dissected otic vesicles, as illustrated in Fig. 5 (left). Indeed, there was a significant enrichment in the SoxTFBS region of the Atoh1 enhancer in the chromatin fraction immunoprecipitated with Sox2 when compared to precipitation with IgG (Fig. 5, upper bar  diagram). Furthermore, this enrichment was specific to this region of the genome as the fraction of SoxTFBS precipitated with Sox2 antibody was significantly higher than the fraction of control region precipitated under the same conditions (Fig. 5, lower bar  diagram). This demonstrates that in the early otic vesicle, Sox2 is bound to the Atoh1 enhancer.
In summary, the regulation of Atoh1 by Sox2 in the otic vesicle relies on the direct binding of Sox2 to the SoxTFBS in the 39 regulatory region of Atoh1 enhancer.

The transient activation of Atoh1 and the induction of inhibitors: an incoherent logic?
The above results suggest that Atoh1 is directly activated by Sox2 at early developmental stages. However, Atoh1 expression during pre-differentiation stages is very low or negligible [3,31]. Several HLH factors like Hes/Hey, Ids, Neurog1 and NeuroD have been involved in the inhibition of Atoh1 expression during otic development [5,6,7,8,9,10,11,12,32], and sequence analysis reveals the presence of bHLH binding sites in the Atoh1 39 regulatory regions [4]. Therefore, these factors are potential candidates to counteract the induction of Atoh1 by Sox2. Besides, Sox2 has been also associated with the negative regulation of Atoh1 and hair cell formation during ear development [21], a function that is reminiscent of that of SoxB1 genes in CNS development [23,24]. However, the mechanism behind this seemingly paradoxical situation in which Sox2 is able to both induce and counteract Atoh1 is unknown. In order to gain insight into this problem we explored further the regulation of Atoh1 by Sox2. A time course analysis of Atoh1 expression following Sox2 transfection in HEK293 cells revealed that Sox2 counteracts its own activator effect on Atoh1. Sox2 transfection induced only a transient activation of Atoh1 as measured either by Atoh1 bGal reporter activity or by qRT-PCR analysis of Atoh1 mRNA (Fig. 6A). The loss of Atoh1 transcription occurred even though Sox2 levels increased monotonically throughout the time window of the  (Fig. 6A red line in the right graph). Several mechanisms may account for this behavior, but the following data suggest that both activation and inhibition require DNA binding and the transcriptional activator function of Sox2. The cotransfection of Sox2DHMG (Fig. 6B, graph, grey) had no effect on Atoh1 reporter activity, while the co-transfection of Sox2HMG-VP16 (Fig. 6B, graph, blue) reproduced the effects of Sox2, both the early up-regulation of Atoh1 and the delayed return to baseline. This suggests that the inhibition of Atoh1 by Sox2 is indirect and requires intermediate factors that change the sign of the activator function of Sox2. Hence, the concurrent activation of inhibitor factors is a plausible explanation. The transient behavior of the Atoh1 response to Sox2 is well described by a genetic network where a gene triggers parallel opposing effects on its target (Fig. 6B, right  diagram), the Incoherent Feed Forward Loop (I-FFL) as modeled by Allon [33].
The above observations lead us to think that since Atoh1 expression and hair cell formation in vivo correlate with Sox2 down-regulation [15], it is possible that Sox2 cooperates with other signaling pathways that maintain Atoh1 expression tuned down during pre-differentiation stages. If such a mechanism operates in vivo, one would expect the activation of Atoh1 inhibitory factors after the overexpression of Sox2 in the otic vesicle. Therefore, we explored the ability of Sox2 to induce these factors in the otic vesicle. Indeed, Sox2 induced the expression of Id1-3 (Fig. 6C, left bar diagram), Hes5 and Hey1 (middle bar diagram) and Neurog1 and NeuroD (right bar diagram) in the otic placode. This indicates that in parallel to Atoh1 induction, Sox2 activates and/or modulates the expression of other genes that counteract Atoh1. Neurogenin1 is a direct target of Sox2 in other model systems [34,35], but it remains to be explored whether this also the case in the otic placode. Ids are regulated by BMP signaling [9], and Hes5 and Hey1 are downstream targets of Notch [11], but it is unknown whether Sox2 directly regulates these genes, or if it rather cooperates at other steps in the signaling cascades (see Discussion). In summary, these data suggests that in parallel to the activation of Atoh1, Sox2 induces an incoherent response by promoting the expression of Atoh1 negative regulators.

Discussion
The prosensory function of Sox2: sensory commitment and deferred hair cell differentiation Throughout evolution, the expression and function of the Sox2 correlates with the commitment to neural fate [36]. However, Sox2 prevents proneural gene function and neuronal differentiation [23,24]. This is also the case during ear development: Sox2 is necessary for sensory fate specification [18], and the misexpression of Sox2 results in increased number of neurons and ectopic hair cells [19,20]. However, Sox2 shows also an antagonistic function with Atoh1 that results in the prevention of hair cell differentiation [21]. The aim of this work was to shed light on the mechanism behind this dual function.
The results show that, both in vitro and in vivo, Sox2 is able to directly activate Atoh1 transcription by binding to the SoxTFBS in the 39 Atoh1 enhancer region, as shown by the functional experiments with the mutated reporter and by ChIP analysis. In the early otic vesicle, ChIP assay reveals that Sox2 is bound to the 39 Atoh1 enhancer and, moreover, the mutation of the SoxTFBS in the 39 regulatory region of Atoh1 suppresses the activity of the enhancer in the otic vesicle. This suggests that Atoh1 transcription is switched on early in otic development, well before hair cell differentiation, and that Sox2 may be one of the factors involved in the initiation of Atoh1 expression. Interestingly, this inductive function seems not to be conserved in non amniotes where Sox2 has been reported to have a rather permissive role in respect to Atoh1 [37].
However, during ear development, Atoh1 is not upregulated until differentiation stages [3,31], suggesting that the initial induction of Atoh1 transcription is prevented, in parallel, by specific mechanisms that result in the appropriate timing of hair cell differentiation. Therefore, Atoh1 expression in neurosensory progenitors would be under the regulation of both activator factors (Sox2, present work) and repressor factors (see below). Bivalent states are characterized by the simultaneous presence of active and repressed chromatin markers in the gene regulatory regions. This is characteristic of many genes involved in cell commitment and pluripotency and it has been reported for the Atoh1 gene in neural cell lines [38], and in mammalian otic sensory progenitors (Z. Stojanova, T. Kwan and N. Segil, personal communication).
Several factors may account for the prevention of Atoh1 upregulation before the stages of hair cell differentiation. Like Atoh1, Neurog1 and NeuroD are also bHLH proneural genes of the Atonal family, but they drive neurogenesis in the otic vesicle [19]. Neurog1 inhibits Atoh1 expression in the inner ear [5] and recent data show that the knock-out of NeuroD (a down-stream target of Neurog1) generates heterotopic and precocious activation of Atoh1 and hair cell fates [8]. It is particularly striking that in the cochleo-vestibular ganglion of NeuroD null mice, there is a significant number of ectopic hair cell-like cells expressing high Atoh1 [8]. This is consistent with our observation that Atoh1 transcriptional activity is found in normal neuroblasts and suggests that unless selectively suppressed, the initial state of neurosensory progenitors is indeed multipotent for both neuronal and hair cell phenotypes. It is worth noting that Neurog1 and NeuroD are transcriptional activators, so the mechanism by which they are able to inhibit Atoh1 is unclear. In contrast, other HLH genes, like Hes/Hey and Id family members, are known transcriptional repressors [39,40,41,42]. They are expressed in otic progenitors and their function is associated with the prevention of Atoh1 function and premature differentiation [7,9,10,11,12,43]. In fact, the Atoh1 enhancer contains a series of bHLH binding sites, which may account for the negative regulation exerted by these genes [4]. Taken together, these factors exert multiple and diverse functions in neural development, but they share a common inhibitory action on Atoh1 that results in the maintenance of the undifferentiated state of neurosensory progenitors.

Sox2 activation of Atoh1 inhibitors: an incoherent loop?
Our results show that Sox2 induces the expression of several of the above mentioned inhibitory factors. Although most of them are under the control of specific signaling pathways, Sox2 is nevertheless able to promote their expression. This indicates that Sox2 operates with an incoherent logic with respect to Atoh1: it both activates Atoh1 and promotes its inhibition. Several network motifs have been studied by Alon [33] as a set of recurrent gene regulation patterns that result in predictable functional behaviors. The activation of Atoh1 by Sox2 fits well with the so-called Type1 Incoherent Feed Forward Loop (I1-FFL) in which the two arms of the FFL act in opposition. The result is a transient target gene activation, with amplitude and timing dependent on the thresholds and time constants of the individual interactions, while the final steady-state level depends on the strength of the inhibition [33]. This type of model predicts well the transient nature of the response of Atoh1 in the presence of continuously increasing concentrations of Sox2 mRNA in vitro. Indeed, the fact that the same behavior is induced by the Sox2-HMG VP16 construct indicates that the decay must be induced by intermediate factors that change the sign of the original signal. In our case, Sox2 directly activates Atoh1 transcription but, on the other hand, Sox2 also up-regulates several inhibitors of Atoh1 that include Neurog1, NeuroD, Hes/Hey and Id genes. This probably causes a balance between activation and inhibition that results in the observed profile of transient activation and steady-state down-regulation of Atoh1. This molecular interaction offers a simple explanation for the intriguing dual effects of Sox2: the induction of neural competence and prevention of differentiation. Further experiments will be required to demonstrate the detailed kinetics and the modulation of this genetic network and to describe the detailed mechanisms by which Sox2 modulates the expression of Atoh1 inhibitors.
Neurog1 has been previously described as a direct target of Sox2 in neural crest cells [35] and recent work suggests that this may the case also in the inner ear [34]. This provides support to the operation of a 1I-FFL in which Sox2 directly regulates Atoh1 and also its negative regulator Neurog1. However, it remains to be proven that this direct interaction operates in vivo in the otic vesicle. But nevertheless, the suggestion that the direct regulation of Atoh1 may extend to Neurog1, provides an interesting model for the function of Sox2 in the specification of the neurosensory competence of the otic placode, and the sequential generation of neurons and hair cells (see below).
The other Atoh1 inhibitors regulated by Sox2 are Hes5, Hey1, and Id1-3. They have never been described as primary Sox2 targets, and their regulation during inner ear development is mainly dependent on non-autonomous signaling. Although we cannot exclude the possibility that Sox2 regulates them directly, it is likely that Sox2 cooperates with the signaling pathways that regulate their expression. The regulation of Hes5 and Hey1 in the ear is mostly Notch-dependent [11]. Sox2 misexpression does not affect the expression of Notch ligands in the ear [21]. But in the otic vesicle Sox2 does result in the induction of Notch1 (Neves et al., unpublished data), and Notch1 has been identified as a direct target of Sox2 in the retina [44]. On the other hand, Id genes are regulated by BMP signaling in the inner ear [9]. Apart from Ids, Sox2 electroporation up-regulates several elements of the BMP pathway that are upstream Id transcription. This includes the Smad Interacting Protein 1 (SIP1, Neves et al., unpublished data), which has been identified as a potential Sox2 target by in silico analysis [45]. Taken together, the data suggest that unlike Atoh1 or Neurog1, Sox2 may regulate these other inhibitors by interacting with the signaling pathways that regulate their expression.

Neurosensory competence and the sequential generation of neurons and hair cells in the inner ear
The problem of cell fate specification is central to neural development. How do different cell types with defined phenotypic characteristics originate from multipotent progenitors? The functional unit of the ear consists of three elements of neural origin: the mechano-transducing hair cells, the supporting cells, and the primary afferent neurons. All three elements derive from the neurosensory competent domain of the otic vesicle and their development follows a stereotyped spatial and temporal pattern, with neurons being specified prior to hair cells [46,47,48]. Neuronal fate is specified by the expression of the proneural genes Neurog1 and NeuroD [6,49,50]. Sensory fate specification occurs after neurogenesis, and commitment to the sensory fate is associated with the expression of Atoh1 [1,3].
The observation that both neurons and hair cells derive from Sox2-positive progenitors fits well with the idea of the common origin of both cell types, as suggested by viral and genetic tracing [5,51]. How does Sox2 specify this dual competence in the otic progenitors? Sox2 is able to induce the expression of proneural genes Neurog1, NeuroD and Atoh1 [34 and present work], which would be sufficient, in principle, to specify neuronal and hair cell fates. But the question then is how these fates are sorted out, and why hair cell fate is delayed with respect to neuronal fate. One possibility is that Sox2 establishes neurosensory competence early in development, by the activation of the major proneural genes Neurog1 and Atoh1. However, the down-regulation of Atoh1 by Neurog1 and NeuroD would allow neurogenesis but not hair cell differentiation. Cell fate decisions would thus depend on selective repression of the initial neurosensory potential, rather that the temporal acquisition of new properties. It is not until Sox2 is counteracted that Atoh1 expression would be permitted in hair cells, but not in supporting cells. Daudet and co-workers have recently shown that Sox21 is expressed during hair cell differentiation and that it is able to inhibit Sox2 expression (N. Daudet, personal communication). A similar interaction between Sox2 and Sox21 was described in the neural tube [52].
In summary, Sox2 promotes sensory fate in the otic vesicle by direct binding to Atoh1 regulatory sequences. However, Atoh1 activation is deferred and Atoh1 up-regulation and hair cell differentiation do not occur until later developmental stages. One possible explanation for this dual effect is that Sox2 triggers an incoherent response that results in a steady-state inhibition of Atoh1. This would provide a simple explanation for the dual function of Sox2 in neural development, i.e.: promotion of neural competence and suppression of differentiation.