bHLH-dependent and -independent modes of Ath5 gene regulation during retinal development

Transcription factors of the basic helix-loop-helix family (bHLH) are involved in the specification of a wide range of cell fates throughout the nervous system. Traditionally known for their roles in determining neural fate within the nervous system, individual bHLH factors are required for the differentiation of specific subsets of cells(Guillemot, 1999). The exact function of each bHLH factor and the type of cell that differentiates in response to their activity depends in part upon the timing and location of activity (Moore et al., 2002). Thus, proper temporal and spatial regulation of bHLH factor expression is essential for normal function.

Ath5, a vertebrate homolog of Drosophila atonal, is expressed in the developing retina in all vertebrates studied and plays a crucial role in regulating retinal neurogenesis(Vetter and Brown, 2001). In Xenopus, Xath5 is expressed in a tightly restricted set of cells in the developing neural retina, the olfactory placodes and the pineal gland. In the retina, Xath5 expression commences in retinal progenitors just prior to cell cycle exit and onset of differentiation, but expression is downregulated before cells become fully mature retinal neurons(Kanekar et al., 1997). In mouse, zebrafish and chick, Ath5 is also expressed in a similar restricted manner within the retina immediately preceding the onset of retinal ganglion cell (RGC) differentiation (Brown et al., 1998; Kay et al.,2001; Matter-Sadzinski et al.,2001). Ath5 is specifically required for the differentiation of the RGC cell type as in both mouse and zebrafish Ath5 loss-of-function mutants, RGCs are either drastically reduced or missing altogether (Brown et al.,2001; Kay et al.,2001; Wang et al.,2001).

As Ath5 plays such a key role in retinal development, it is important to understand how its expression is regulated. Analysis of the chicken Ath5 (Cath5) promoter has suggested that it can be regulated by multiple bHLH factors in retinal cell culture, and that both Cath5 and Ngn2 are bound to the Cath5 promoter at specific stages of retinal development (Matter-Sadzinski et al., 2001; Skowronska-Krawczyk et al., 2004). However, it is not yet known what is required for correct tissue-specific expression in vivo. Recent genome sequencing efforts have led to the development of phylogenetic footprinting strategies, whereby cross-species sequence comparison of noncoding regions from homologous genes can identify candidate enhancers that may play a role in regulating gene expression. This approach provides excellent predictive power for functional enhancer elements (Bulyk,2003); however, all candidate elements must ultimately be tested in vivo to determine whether they can regulate gene expression and assess how they contribute to the normal pattern of expression during development(Muller et al., 2002).

In this study, we have analyzed the regulation of Ath5 expression in Xenopus, as well as in mouse. First, we identified a highly conserved proximal non-coding region, and showed that it mediates bHLH-dependent regulation of Ath5 expression. This region was sufficient to promote transgene expression in Xenopus, but not in mouse. Second, we show that a longer transgene, that includes additional 5′ cis-regulatory sequence, promotes bHLH-independent transgene expression in Xenopus. Longer transgenes from either the mouse or Xenopus Ath5 cis-regulatory region were also sufficient to promote expression in mouse. For the Math5 transgene (Atoh7 –Mouse Genome Informatics), expression did not depend upon Math5 itself. Thus,there exist both bHLH-dependent and-independent modes of Ath5 gene regulation; however, the importance of bHLH-dependent expression mediated through the conserved proximal region may be species dependent.

A 14.3 kb Xath5a genomic clone was isolated from a Xenopus laevis genomic library in LambdaGEM-11 (a gift from Dr Michael King). The 3.3 kb of 5′ Xath5 genomic sequence immediately 5′ of the coding region was cloned into the promoterless GFP reporter construct pG1 (a gift from C.-B. Chien) to create the pG1X5-3.3 kb construct. Three transgenic constructs, pG1X5ΔPstI, pG1X5ΔHindIII, and pG1X5ΔDraII, were made by digesting the pG1X5-3.3 kb construct with the indicated enzyme and religating. All other deletion constructs were made using PCR amplification from existing plasmids and the PCR fragments were then cloned into the pG1X5-TATAA construct. The minimal mouse Fospromoter was derived from the TOP:dGFP construct (a gift from R. Dorsky)(Dorsky et al., 2002) and cloned into pG1. Constructs with point mutations were made using the QuikChange II kit (Stratagene). All E-box sequences were mutated from CAnnTG to ATTnTG, which has been shown to destroy the ability of bHLH factors to bind to regulatory E-box sequences (Helms et al., 2000). The mutation introduced into the α-box site was based on a previous study (Culi and Modolell, 1998). The primer sequences used to generate constructs created with PCR amplification are available upon request. All constructs were verified by sequencing.

The generation of transgenics was carried out as described(Kroll and Amaya, 1996), with variations (Hutcheson and Vetter,2002). Whole-mount embryos were analyzed for GFP expression between stages 28 and 42, when the majority of retinal neurons are born. Embryos were scored as retinal positive if there was specific retinal expression that was similar in pattern and timing to endogenous Xath5expression. For constructs with weak or no detectable GFP expression by fluorescence, expression or lack of expression was confirmed either by in situ hybridization or by whole-mount antibody staining using a polyclonal anti-GFP antibody (Torrey Pines) and an Alexa-Fluor 488-conjugated secondary antibody(Molecular Probes). For analysis on sections, embryos were sectioned on a cryostat at a thickness of 14 μm.

Embryos were processed for in situ hybridization as described previously(Kanekar et al., 1997). Double in situ hybridization on sections was performed using a digoxigenin-labeled GFP probe and a fluorescein-labeled Xath5 probe as previously described (Hutcheson and Vetter,2001).

Sequence analysis was performed using VISTA(http://www-gsd.lbl.gov/vista/)(Frazer et al., 2004; Mayor et al., 2000). Alignment criteria used were 80% nucleotide identity over 30 bp blocks. Candidate transcription factor binding sites were identified using the Genomatix MatInspector module(http://www.genomatix.de/)(Quandt et al., 1995), the Transfac database(www.gene-regulation.com)(Wingender et al., 1996) or ClusterBuster(http://zlab.bu.edu/cluster-buster/)(Frith et al., 2003).

Other vertebrate Ath5 genomic sequences were identified as follows: X. tropicalis(http://genome.jgi-psf.org/xenopus0/xenopus0.home.html);mouse and human Ath5 (Atoh7) have been previously isolated and aligned (Brown et al.,2002); chick (GenBank Number AJ630209)(Skowronska-Krawczyk et al.,2004); Fugu (scaffold 1775; http://genome.jgi-sf.org/fugu3/fugu3.home.html);and zebrafish (clone AL627094; http://trace.ensembl.org/perl/ssahaview). The genomic sequence for Drosophila atonal was derived from BAC clone#AC008094(http://flybase.bio.indiana.edu/).

0.6 kb and 2.3 kb sequences found immediately upstream from the Math5 start ATG codon (nucleotides 2472 to 3072 and 772 to 3072 from AF418923) were cloned into the pG1 GFP reporter construct. These Math5 (M5) enhancer-promoters and the GFP-coding sequence were purified away from vector sequences and used to generate independent mouse lines transgenic with either the M5-0.6 kb or M5-2.3 kb transgene. Similarly,the Xenopus pG1X5-proximal or pG1X5-3.3 kb constructs were used to generate transgenic mice. All transgenic strains are viable and fertile and maintained on a CD1 background. For some experiments the M5-2.3 transgenic line was crossed with the Math5 mutant(Brown et al., 2001) maintained on CD-1. E12.5 or E13.5 mouse embryos from timed pregnancies were harvested,kept on ice in PBS for whole mount imaging. The genotype of M5-2.3Tg/+ or M5-2.3Tg/+;Math5^–/–embryos was determined by PCR genotyping of embryonic or adult tail DNA using primers specific for GFP coding sequences or as described previously(Brown et al., 2001).

In order to investigate the mechanisms that regulate the expression of Xath5, we first sought to identify genomic sequences sufficient to drive tissue-specific expression of a transgene in vivo. Roughly 15 kb of genomic Xath5a sequence was isolated from a X. laevisgenomic library. Like most genes encoding proneural bHLH factors, Xath5 has one exon containing the coding region and no introns. In addition, the genomic clone contained 3.3 kb of 5′ and 11.8 kb of 3′ sequence flanking the Xath5-coding region (data not shown). Based upon analysis of several full-length Xath5 cDNA clones(Kanekar et al., 1997), the putative transcription start site was determined to lie 205 bp 5′ of the translation start site. The highly conserved TATAA basal promoter element is 226 bp 5′ from the translation start site(Fig. 1A).

To test for regulatory activity, we fused 3.3 kb of Xath5 5′non-coding sequence to a promoterless vector containing a GFP reporter gene(pG1; C.-B. Chien and D. Gilmour, unpublished) to create the pG1X5-3.3 kb construct (Fig. 1A), which was used to generated transgenic frog embryos(Kroll and Amaya, 1996). At stage 17, when Xath5 expression first commences in the developing olfactory placodes, two faint spots of GFP were visible in the anterior end of transgenic embryos (data not shown). Like endogenous Xath5(Fig. 1B)(Kanekar et al., 1997),pG1X5-3.3 kb transgene expression was apparent in the neural retina, the olfactory placodes and the pineal gland at stage 28(Fig. 1C,D).

To define which regions within the 3.3 kb Xath5 cis-regulatory region are crucial for retinal expression, we generated a series of deletion constructs (pG1X5-ΔHindIII, pG1X5-ΔDraII, and pG1X5-ΔPstI) by restriction digest(Fig. 2A). When each construct was tested in transgenic Xenopus embryos, expression of the GFP transgene was comparable with that seen with the full pG1X5-3.3 kb construct(Fig. 2A). Thus, the 464 bp of sequence closest to the Xath5-coding region are sufficient to promote transgene expression in the retina. To verify this, the pG1X5-proximal construct containing only 427 bp of upstream sequence was used to generate transgenic embryos (Fig. 2B). These embryos consistently expressed GFP in the retina(Fig. 2C). Expression in the olfactory placodes and pineal was also observed, but less frequently (data not shown).

In the Xath5 genomic sequence, the TATAA box, a highly conserved basal promoter element (Smale and Kadonaga, 2003), is located 226 bp upstream of the translation start site. The TATAA region alone (pG1X5-TATAA) was not sufficient to drive retinal GFP expression (Fig. 2B), suggesting that the critical elements in the Xath5proximal regulatory region lie between positions –427 and –226. To test this idea, the sequence from –427 to –226 was fused to a minimal mouse Fos promoter cloned into the pG1 vector. The Fos basal promoter alone did not drive retinal expression(Fig. 2B). When 201 bp of Xath5 sequence (–427 to –226) was fused to the Fos basal promoter (Fig. 2B), weak but specific transgene expression in the retina was detectable (Fig. 2D). Together,these results indicate that elements sufficient for retinal expression are located between positions –427 and –226 relative to the Xath5 translation start site.

To assess whether the transgene is expressed in the appropriate cells within the retina, we performed double in situ hybridization on sections from stage 42 pG1X5-proximal transgenic embryos, and found that the expression of GFP RNA (Fig. 3A,E)was restricted to the ciliary marginal zone (CMZ) in a pattern identical to endogenous Xath5 mRNA (Fig. 3C,G) (Kanekar et al.,1997). Interestingly, at stage 42, GFP fluorescence was visible throughout the central retina in all three neuronal layers and in most neuronal cell types, probably owing to perdurance of the GFP protein(Fig. 3D). This is consistent with recent lineage analysis of Math5-expressing cells in the mouse retina showing that they give rise to all major classes of retinal cell types that occupy the three cell layers (Yang et al., 2003) (J. Brzezinski and T. Glaser, unpublished).

All known vertebrate Ath5 genes are expressed in the developing retina coincident with the onset of retinal ganglion cell genesis, suggesting that mechanisms regulating Ath5 expression may be conserved as well. We therefore compared Xath5 noncoding sequence to other Ath5genomic sequences to identify conserved regulatory regions. Pairwise mVISTA analysis (Frazer et al., 2004; Mayor et al., 2000) between Xenopus laevis and mouse identified a peak of homology in the 5′ noncoding region (Fig. 4A) within the 427 bp region that was sufficient for tissue-specific expression in Xenopus(Fig. 2). Alignment of this region across X. laevis, X. tropicalis, mouse, human and chick showed a 65 bp stretch with significant nucleotide identity among all of these vertebrate species. Within this sequence, two E-boxes, which are binding sites for bHLH factors (Fig. 4B,positions –270 and –260, referred to as E2 and E1), and the flanking bases were almost entirely conserved (18/21 bp identical, Fig. 4B). Additionally, both zebrafish and fugu have the conserved E1E2 cassette, although there is little conservation elsewhere in the 65 bp region(Fig. 4B).

We sought to define a smaller region that still promotes transgene expression in the retina by generating sequential deletions of the pG1X5-proximal construct (Fig. 5A). pG1X5-proximal-401 showed retinal expression, as did pG1X5-proximal-389, but more weakly so. This suggests that elements necessary for robust expression, but not tissue specificity, lie between –401 and–389. Transgenic constructs containing further deletions of the proximal region up to –337 similarly promoted weak retinal expression of GFP, but the pG1X5-proximal-250 construct, in which the conserved E-boxes E1 and E2 are missing, did not (Fig. 5A).

To determine whether E1 and E2 are required elements in the proximal cis-regulatory region, we mutated each E-box individually (ΔE1 orΔE2) or in combination (ΔE1ΔE2) in the pG1X5-proximal transgene (Fig. 5B). Single E-box mutations reduced but did not eliminate retinal expression of GFP(Fig. 5B). When both E-boxes were mutated, GFP was not expressed (Fig. 5B), indicating that these sites are critical for the activity of the proximal Xath5 cis-regulatory region.

To further test whether expression of the pG1X5-proximal transgene depends upon bHLH activity, we generated transgenic embryos then injected RNA for RFP (red fluorescent protein; 1 ng) and a dominant-negative form of Xath5 (500pg) into a dorsal animal blastomere at the eight-cell stage. Dominant-negative Xath5 was created by replacing the putative Xath5 activation domain with the repressor domain from Drosophila Engrailed to create Xath5-EnR. This dominant-negative protein interferes with the proneural activity of multiple atonal-related bHLH factors, including Xath5, Xath3,NeuroD and XNgnR1 (T. VanRaay, M. Logan and M.L.V., unpublished). Expression of Xath5-EnR was able to suppress pG1X5-proximal transgene expression on the injected side (6/13; Fig. 5C-E), arguing that proneural bHLH activity is required for expression driven by the proximal Xath5 cis-regulatory region.

Although E1 and E2 are necessary elements in the proximal promoter, we sought to define the minimal set of elements sufficient to drive retinal specific expression of the transgene. To test the E-boxes alone, two copies of E1 and E2 (–272 to –253) were fused to the pG1X5-TATAA construct(Fig. 5F). The resulting transgene pG1X5-TATAA+2xE1E2 never promoted detectable GFP expression. Thus,the highly conserved E-boxes alone are insufficient for transgene expression. We then created pG1X5-TATAA+33 bp (Fig. 5F), which includes one copy of E1 and E2 as well as an adjacent motif that resembles the α-box previously identified in the Drosophila scute sensory mother cell (SMC) enhancer(Culi and Modolell, 1998). In Xenopus embryos transgenic for pG1X5-TATAA+33 bp, GFP expression was not specific and was found in a wide range of neural and muscle tissues(Fig. 5G).

Next we created pG1X5-TATAA+48 bp, which includes an additional 15 bp of conserved upstream sequence (Fig. 5F). Embryos transgenic for pG1X5-TATAA+48 bp showed GFP expression that was strikingly similar to endogenous Xath5expression, and also included consistent weak expression in axial muscles(Fig. 5F,H), with a small percentage of embryos showing only axial muscle expression. We conclude that addition of this 15 bp region is sufficient to confer almost complete tissue specificity to transgene expression. Interestingly, this region is highly conserved across species (11/15 nucleotides identical between Xenopus laevis and mouse), but contains no known transcription factor binding sites.

This 48 bp fragment contains two E boxes, the α-box and the additional 15 bp fragment required for tissue-specific expression. To assess whether the α-box contributes to the strength or specificity of expression, we mutated four residues within the α-box sequence(pG1X5-TATAA+48 bp-αmut). We found that this construct yielded a significantly reduced fraction of GFP positive transgenic embryos, but the overall pattern was unchanged (Fig. 5F). Thus, the α-box appears to contribute to the strength but not specificity of expression promoted by the short 48 bp fragment.

To test whether bHLH factors in Xenopus can regulate Xath5 transgene expression in vivo, we overexpressed proneural bHLH factors in pG1X5-3.3 kb transgenic embryos by fertilizing eggs from an adult transgenic female and injecting RNA at the eight-cell stage encoding for RFP(1 ng) and for either Xath5, XngnR1, Xath3, XNeuroD, Xash1 or Xash3 (500 pg each). No change in transgene expression was observed with overexpression of RFP alone (21/21; Fig. 6A-C). We observed significant ectopic GFP expression with overexpression of the atonal-related bHLH factors Xath5 (30/30), XngnR1 (26/28), Xath3 (9/11) or NeuroD (12/12; Xath5 shown in Fig. 6D-F, and data not shown), but not with overexpression of either Xash1 (50/50) or Xash3 (14/15), which are Achaete-Scute-related factors (data not shown). Thus, the pG1X5-3.3 kb transgene, which includes the conserved proximal region, can be recognized and activated by Atonal-related bHLH factors.

Although either E1 or E2 are required for expression of the pG1X5-proximal transgene, we found that they are not required for expression of the full pG1X5-3.3 kb transgene (Fig. 6G), suggesting that additional more distal elements may also contribute to expression. We noticed another weaker region of homology between Xenopus, mouse, human and chick (–1597 to –1669, Fig. 4A) that included two well-conserved E-boxes at –1600 and –1629 (E3 and E4,respectively; data not shown). However, mutation of all four conserved E boxes(ΔE1-4) did not abolish pG1X5-3.3 kb transgene expression(Fig. 6G).

As there are additional non-conserved E-boxes, we wanted to further test whether bHLH activity is required for pG1X5-3.3 kb transgene expression using Xath5-EnR. We fertilized eggs from an adult Xenopus laevis female transgenic for the pG1X5-3.3 kb transgene, and injected RNA at the eight-cell stage encoding for Xath5-EnR (500 pg) and for RFP (1 ng). Overexpression of Xath5-EnR did not suppress pG1X5-3.3 kb transgene expression (63/63; Fig. 6H-J), and also did not suppress expression of endogenous Xath5 (15/16; data not shown). Together, these data argue for a bHLH-independent component to Xath5 gene regulation in vivo, probably mediated by more distal cis-regulatory sequences.

To test whether distal cis-regulatory sequences alone have enhancer activity, we fused a distal PstI fragment that lacks the conserved proximal region (–3162 to –464) to pG1X5-TATAA. We observed specific retinal GFP expression in 39% of transgenic embryos(Fig. 7A,B), with much less frequent olfactory and/or pineal expression as well (data not shown). This distal region also promoted weak but specific retinal expression when coupled to the mouse Fos basal promoter (Fig. 7A). Thus, the distal Xath5 cis-regulatory region contains elements sufficient for retinal expression that are distinct from the conserved proximal region.

As sequence alignments revealed conservation in the Ath5cis-regulatory region between Xenopus and mammals, we tested 0.6 kb and 2.3 kb pieces of 5′ Atoh7 genomic sequence in transgenic Xenopus embryos. Tailbud stage embryos were analyzed for GFP mRNA expression by in situ hybridization. The 0.6 kb Atoh7 cis-regulatory region (pG1M5-0.6 kb) promoted retinal expression in Xenopus,although expression was much weaker and at a lower rate than for the pG1X5-proximal construct (8/144 embryos retinal GFP-positive; Fig. 8A). The weak activity of the pG1M5-0.6 transgene indicates that despite high sequence conservation,crucial elements may have diverged between Xenopus and mouse. By contrast, the pG1M5-2.3 kb construct promoted robust expression of GFP in the retina (19/56 embryos retinal GFP-positive; Fig. 8B), but also in other neural tissues such as the midbrain, hindbrain and cranial ganglia. Within the mature Xenopus retina, GFP mRNA was restricted to the CMZ, similar to that seen in pG1X5-proximal embryos (Fig. 8C, compare with Fig. 3).

Based on our results in Xenopus, we sought to analyze the activity of these regulatory regions in the developing mouse retina. We determined that the pG1M5-0.6 kb transgene exhibited no GFP fluorescence or mRNA in mice (0/3 independent lines, 0/31 embryos), which is consistent with the low frequency of transgenics and weak expression seen with this transgene in Xenopus. The 2.3 kb region of Atoh7 regulatory DNA conferred strong optic cup and retinal GFP expression in E11-13 mouse embryos (four out of four independent lines, Fig. 8D,F) along with ectopic expression in brain and spinal cord,consistent with expression of this transgene in Xenopus (not shown). When the activity of the Xenopus cis-regulatory region was tested in mice, we found that the 427 bp pG1X5-proximal transgene exhibited no GFP fluorescence or mRNA expression (0/4 independent lines, 0/78 embryos), in contrast to the robust expression seen with this transgene in Xenopus. Interestingly the longer 3.3 kb pG1X5 construct showed specific optic cup and retinal expression in E11-13 mouse embryos (1/6 lines,29/29 embryos; Fig. 8E). Thus,the short proximal region from either mouse or Xenopus is not sufficient for transgene expression in mouse, but longer transgenes from either species gave robust retinal expression.

In the developing chick retina, Cath5 is bound to its own promoter(Skowronska-Krawczyk et al.,2004), suggesting that autoregulation may play a role in Ath5 gene regulation. In addition both Atonal and Math1 (Atoh1)regulate their own expression (Baker et al., 1996; Helms et al.,2000). Therefore, we tested for autoregulation of Math5by crossing two independent 2.3 kb M5-GFP transgenic lines to mice carrying the Math5 targeted deletion (Brown et al., 2001). Mice heterozygous for both the transgene and mutant allele were intercrossed and the resulting embryos examined for GFP fluorescence in the optic cup at E13.5, when endogenous Math5(Brown et al., 1998) and the∼2.3 kb M5-GFP transgene are maximally expressed. This experiment was performed four independent times, using two litters for each transgene(n=56 embryos total). In every mutant embryo that possessed the transgene (n=12), robust GFP fluorescence was observed (compare Fig. 8F with 8G). Thus, Math5 is not required for transgene expression, suggesting that autoregulation is not an essential mechanism for regulating Math5 expression during mouse retinal development.

Our analysis has identified two modes of Ath5 gene regulation in the developing Xenopus retina. First, we found a conserved proximal cis-regulatory region in the Ath5 gene and showed that this region was sufficient to promote bHLH-dependent transgene expression in the developing Xenopus but not mouse retina. Proximal transgene expression in Xenopus required two highly conserved E-boxes that function together with adjacent conserved sequences to promote retinal expression. We also revealed a bHLH-independent mode of Xath5 gene expression mediated through more distal sequences and showed that the distal cis-regulatory region alone was sufficient to promote retinal transgene expression in Xenopus. Furthermore, transgenes that included these more distal sequences were efficiently expressed in both Xenopus and mouse. Thus, we found both conserved and non-conserved aspects to Ath5 gene regulation in the developing retina. It remains to be determined whether these two regulatory regions contribute to different spatial or temporal aspects of Ath5 gene expression.

bHLH factors probably play a role in regulating the retinal expression of Xath5 in vivo through the conserved proximal region. There is ample precedence for vertebrate bHLH factor cross-regulation and autoregulation. For example, in Xenopus during primary neurogenesis at open neural plate stages, Xngnr-1 activates the expression of Xath3 and NeuroD, while Xath3 and NeuroD crossactivate each other's expression(Ma et al., 1996; Perron et al., 1999). Expression of a 1 kb Neurod2 promoter in mouse brain was also recently shown to depend upon two E boxes in the proximal promoter region(Lin et al., 2004).

We found that overexpression of multiple atonal-related bHLH factors,including Xath5, Xath3, XNeuroD and X-ngnr-1, could ectopically activate pG1X5-3.3 kb transgene expression in Xenopus embryos. Which of these are likely to play a role in regulating Xath5 expression in vivo? In the CMZ, a number of bHLH factors are expressed in an overlapping but sequential manner in a peripheral to central spatial pattern that reflects the sequence of gene activation during early eye development(Perron et al., 1998). XNeuroD and Xath3 (NeuroM) do not precede Xath5 (Kanekar et al.,1997; Perron et al.,1999), so they are unlikely to initiate Xath5 expression,but could help maintain retinal expression. X-ngnr-1 is expressed in the CMZ in an earlier yet overlapping pattern, so it may help initiate and/or maintain Xath5 expression. In chick, the related gene Ngn2is co-expressed with Cath5 in early retinal progenitors, and Ngn2 protein is bound to the Cath5 promoter during retinal development,and activates expression of a Cath5 transgene in cell culture(Skowronska-Krawczyk et al.,2004; Matter-Sadzinski et al.,2001). Thus, X-ngnr-1 (or Ngn2 in chick) is a strong candidate for regulating Ath5 expression in vivo.

We did not observe ectopic pG1X5-3.3 kb transgene expression with overexpression of either Xash1 or Xash3, suggesting that these factors do not positively regulate Xath5 gene expression. In chick, Ath5and Ash1 are expressed in mutually exclusive domains in the retinal neuroepithelium, and Cash1 expression inhibited activity of the chick Ath5 regulatory region in transfected retinal cells in culture(Matter-Sadzinski et al.,2001). It remains possible that in Xenopus, Xash1 or Xash3 could have similar inhibitory activity on bHLH-dependent expression from the pG1-X5 proximal region alone.

We determined that either E1 or E2 are necessary in the context of the proximal Xath5 promoter to drive retinal expression of a GFP transgene. However these two highly conserved and essential E-boxes alone were insufficient for retinal-specific expression. A 33 bp fragment that included E1 and E2 plus adjacent conserved 5′ sequence drove GFP expression throughout the CNS as well as in muscle, suggesting it may be non-specifically recognized by multiple bHLH factors. This 33 bp sequence includes a motif that is similar to the consensus for an α-box, which was first identified as a critical element in the scute sensory mother cell (SMC) enhancer(Culi and Modolell, 1998), and may be a candidate binding sequence for a winged-helix zinc finger transcription factor (Genomatix MatInspector). In Xenopus, this sequence played a role in robustness but not specificity of transgene expression.

Inclusion of an additional 15 bp of 5′ sequence constrained transgene expression to tissues where endogenous Xath5 is expressed, along with weak expression in the axial muscles. The additional 15 bp of sequence included in this construct are highly conserved, with 11 out of 15 nucleotides identical between Xenopus laevis and mouse; however, no candidate transcription factor binding sites were apparent. It is intriguing to speculate that some novel factor interacts with this sequence and participates in bHLH-dependent Ath5 gene regulation.

In the mouse retina, bHLH factors are unlikely to initiate Math5expression. Ngn2 commences expression at E13, 2 days later than Math5 (Brown et al.,1998), and Math5 is not preceded by expression of genes encoding any other known proneural bHLH factor. Thus, Math5 must be regulated by other factors present in the developing optic vesicle and cup. Consistent with this idea, the short proximal region from either mouse or Xenopus, which contains the conserved E-boxes, was insufficient to promote transgene expression in mouse. Thus, despite strong sequence homology in the proximal cis-regulatory region, the role bHLH factors play in regulating the expression of Ath5 appears to be species specific. This may in part depend upon whether other bHLH factors precede Ath5,as in frog and chick, or whether it is the first retinal bHLH gene expressed,as in the mouse. We also cannot rule out a role for bHLH factors in some other aspect of Ath5 gene regulation, potentially at later stages of development in mouse. It also remains possible that in Xenopus or chick, bHLH factors do not play an essential role in regulating Ath5gene expression as we identified a bHLH-independent distal cis-regulatory region that alone was sufficient to promote retinal transgene expression.

Autoregulation may also play a role in regulating Xath5expression. In Drosophila, Atonal function is crucial for its own expression (Sun et al., 1998),and in vertebrates Math1 has autoregulatory activity in the developing spinal cord, but not in the cerebellum (Gazit et al., 2004; Helms et al.,2000). In chick retinal cultures, Cath5 transgene expression could be activated by chick Ath5 itself and Cath5 was found bound to its promoter in vivo, suggesting autoregulation(Skowronska-Krawczyk et al.,2004; Matter-Sadzinski et al.,2001). Consistent with this, we showed that overexpression of Xath5 could activate ectopic transgene expression in Xenopus. However, in mouse expression of the pG1M5-2.3 kb transgene did not depend upon Math5, arguing that autoregulation is not essential for Ath5 gene expression in the developing mouse retina. This fits with the observation that expression of a β-galactosidase reporter introduced into the Math5 locus by homologous recombination is expressed in Math5^–/– optic cup and retina (E11-birth)(Brown et al., 2001) (data not shown). Consistent with this, expression of the longer pG1X5-3.3 kb transgene in Xenopus, as well as expression of endogenous Xath5, was not blocked by the dominant-negative Xath5-EnR. Thus, there are clearly additional sequences that can mediate Ath5 gene expression in both species, as discussed below. This outcome does not preclude the possibility that autoregulation contributes to some aspect of Ath5 expression during retinal development.

We found that that unlike the proximal Xath5 transgene, expression driven by the full 3.3 kb transgene was not bHLH dependent, as it did not require four conserved E-boxes and could not be suppressed by overexpression of Xath5-EnR. Thus, we conclude that more distal sequences mediate bHLH-independent expression. We found that a distal fragment from the Xath5 cis-regulatory region alone was sufficient to promote retinal transgene expression in Xenopus. Future analysis will focus on defining which elements within this region are required for expression. Interestingly, longer 5′ cis-regulatory regions from either mouse (2.3 kb) or Xenopus (3.3 kb) were sufficient to promote appropriate transgene expression in mouse, consistent with our conclusion that Ath5 gene expression in mouse is largely bHLH-independent. However,there is little sequence homology between mouse and Xenopus in the distal region other than E3 and E4, so it is unclear whether the mechanisms governing Math5 and Xath5 gene regulation through more distal sequences will be conserved.

What are the signals that regulate bHLH-independent Ath5 gene regulation? In Drosophila, the initiation and propagation of atonal expression directly depends upon signals such as hedgehog(Dominguez, 1999; Sun et al., 1998). However in the 3.3 kb Xath5 cis-regulatory sequence, we found no clear binding sites for Gli zinc finger transcription factors, which mediate hedgehog signaling, although it remains possible that regulation is indirect. In the fish retina, Ath5 expression is initiated by a signal from the optic stalk (Masai et al., 2000),and it is possible that this signal acts through the cis-regulatory sequences that we have identified. In support of this idea, the 3.3 kb pG1X5 transgene was expressed in the developing retina in transgenic zebrafish embryos,suggesting that the 3.3 kb Xath5 regulatory sequences are appropriately recognized in zebrafish (A. Pittman and C.-B. Chien, personal communication).

We have identified bHLH-dependent and -independent modes of Ath5gene regulation in Xenopus, raising the issue of how they contribute to endogenous Xath5 expression. It is possible that the distal and proximal cis-regulatory sequences serve overlapping or redundant functions. Alternatively, the distal and proximal regions may regulate different phases of Xath5 expression. For example, during Drosophila eye development atonal gene expression is initiated in a bHLH-independent fashion by factors such as hedgehog, then expression becomes dependent upon Atonal itself (Hsiung and Moses,2002). It is therefore possible that Xath5 gene regulation is similar, with initiation of gene expression being bHLH-independent and maintenance of expression requiring Xath5 and/or other bHLH factors such as X-Ngnr-1. In chick, bHLH factors are clearly involved in regulation of Cath5 (Skowronska-Krawczyk et al., 2004; Matter-Sadzinski et al., 2001), but it remains to be determined whether there is bHLH-independent regulation as well. In mouse, we found no evidence for bHLH-dependent Atoh7 gene regulation, demonstrating that Math5 expression is bHLH independent. Thus, we have shown that although some mechanisms of Ath5 gene regulation are conserved, there are intriguing species-specific differences that remain to be explored.

We are particularly grateful to Rob Grainger (University of Virginia) for training and invaluable technical assistance in the early phases of this project. Jennifer Rasmussen, Harold Rust and Mike Steele cloned the X. laevis and X. tropicalis Ath5 genomic fragments. Meredith Schneider assisted with the double in situ hybridization. Chi-Bin Chien provided the pG1 vector and Richard Dorsky provided the pG1-cfos construct. Dave Blackburn created the 0.6 kb M5GFP and 2.3 kb M5GFP constructs. The transgenic core at Children's Hospital Research Foundation generated founder animals in the mouse transgenic experiments. We also thank Nick Marsh-Armstrong (Johns Hopkins University) for technical advice and control constructs. R. Dorsky and A. Sanchez provided valuable comments on the manuscript. The manuscript was also improved by insightful comments from the reviewers. This work was supported by NIH grants to N.L.B. (EY13612) and M.L.V. (EY12274), and was supported by the National Science Foundation under Grant Number 0075023 to M.L.V. D.H. was supported by an NIH Developmental Biology Training Grant (5T32 HD07491).