Alternative promoter use in eye development: the complex role and regulation of the transcription factor MITF

Alternative promoter use is a common principle by which genes generate transcript diversity and contribute to an organism's phenotypic complexity. Often, promoter choice does not affect the sequence of the corresponding proteins because translation is invariably initiated from a single downstream start site, but the untranslated 5′ regions can influence mRNA stability and translational efficiency and so refine gene regulation beyond the spatial and temporal control afforded at the transcriptional level. With some genes,however, variant 5′ exons encode unique amino acid sequences and hence provide an additional level of regulation as protein isoforms with distinct N-termini can differ in their function(Carninci et al., 2006; Sandelin et al., 2007). Nevertheless, despite a wealth of sequence information for many species,relatively little is known about the role and regulation of alternative promoter use in vivo, both during development and in adulthood.

A prime example of a gene with multiple promoters is Mitf(microphthalmia-associated transcription factor), which encodes a basic helix-loop-helix leucine-zipper protein that is crucial for mammalian eye development (Hodgkinson et al.,1993; Nakayama et al.,1998; Nguyen and Arnheiter,2000). In mice and humans, the gene contains nine distinct promoters, six of which are linked to different coding exons and three to non-coding exons (Fig. 1)(Hallsson et al., 2000; Steingrimsson et al., 2004; Hershey and Fisher, 2005; Arnheiter et al., 2006; Hallsson et al., 2007). Using probes that do not distinguish between the corresponding isoforms, it has been shown that Mitf is initially expressed in mice throughout the budding optic vesicle and retained in the future retinal pigment epithelium (RPE), the ciliary body and the iris, but is downregulated in the future retina and optic stalk (Bora et al., 1998; Nguyen and Arnheiter, 2000). In Mitf mutants in which all isoforms are equally affected, the RPE can develop as a laminated second retina(Müller, 1950; Bumsted and Barnstable, 2000; Nguyen and Arnheiter, 2000),and in mutants in which Mitf is abnormally upregulated in the retina,the retina can develop as a single-layered RPE(Rowan et al., 2004; Horsford et al., 2005; Bharti et al., 2006). From these results, however, it remains unclear which isoforms are normally expressed during RPE and retina formation and whether any of these isoforms have selective roles or are entirely interchangeable.

The question of isoform-specific function and expression also touches on the problem of whether distinct isoforms are co-regulated or regulated separately. In fact, two lines of evidence argue in favor of co-regulation. One is the observation, made in mice, that the combined genetic lack of the eye transcription factors PAX2 and PAX6 is associated with a total absence of all MITF protein. This finding is supported by the notion that the human A-MITF promoter, which, like the mouse A-Mitf promoter, is located at the 5′ end of the gene, is stimulated by PAX2/PAX6 in vitro, and hence might serve as a control region for the entire MITF locus in the eye in vivo (Bäumer et al.,2003). The other is the recent observation that the paired-like homeodomain transcription factor CHX10 (also known as VSX2 - Mouse Genome Informatics), which is involved in the retinal downregulation of Mitf(Liu et al., 1994; Burmeister et al., 1996; Rowan et al., 2004; Horsford et al., 2005), may stimulate a cluster of genes encoding microRNAs that serve to suppress Mitf mRNAs by recognizing their common 3′UTRs(Xu et al., 2007). Without more-detailed information about the developmental expression pattern of each Mitf isoform, however, an isoform-selective regulation cannot be excluded a priori.

Here, we use a combination of tissue dissection, reverse transcriptase-PCR and isoform-specific antibodies to show that some Mitf isoforms are ubiquitously expressed throughout development in both ocular and extraocular tissues, that others are more restricted to the optic neuroepithelium, and others still are retained exclusively in the RPE. We then analyze two different mouse mutations associated with over- or underexpression of Mitf to assess the functional relevance of some of these isoforms in both retina and RPE. The results indicate that contrary to previous assumptions, the regulation of Mitf during eye development is not simply global, affecting all transcriptional isoforms indiscriminately, but in fact is isoform-selective. They also suggest that some isoforms are more crucial than others for early cell-fate decisions in the developing eye, and hence for the formation of a functional adult eye.

C57BL/6J control mice were from Jackson Laboratories (Bar Harbor, ME) and albino Swiss Webster (SW-Tyr^c) mice were from Taconic(Rockville, MD). The following mutant mice were bred locally and kept according to NINDS-approved animal study proposals:C57BL/6J-Mitf^mi-rw, mixed(C57BL/6J;C3H/HeJ)-Mitf^mi-bw,C57BL/6J^*C3H/HeJ-Mitf^mi-vga9, used as N7F2, and C57BL/6J^*129S1/Sv-Chx10^orJ, used as N10F12-F16.

Pan-specific rabbit antibodies against MITF have been described(Nakayama et al., 1998). An MITF exon 1B1b-specific antibody was prepared by immunizing rabbits with a synthetic peptide (SRILLRQQLMREQMQEQERR) and affinity purifying the resulting serum using a corresponding peptide column. A mouse monoclonal antibody to the C-terminus of MITF was prepared by standard procedures after immunizing mice with a recombinant mouse MITF fragment corresponding to residues 298-419 (exon 9) tagged with GST. Clone 6A5 (IgG1, κ) was used as ascites fluid made from ICR SCID mice. A rabbit anti-mouse-tyrosinase serum was a gift from Dr Vincent Hearing (National Cancer Institute, National Institutes of Health,Bethesda, MD). The following antibodies were obtained from commercial sources:rabbit anti-PAX6 and mouse monoclonal anti-TUJ1 (Covance Research); rabbit anti-cyclin D1 and rabbit anti-phosphohistone (Upstate Biotechnology); mouse monoclonal against V5 tag (Invitrogen); sheep anti-N-terminus of CHX10(Abcam); sheep anti-C-terminus of CHX10 (Exalpha Biologicals).

Immunolabeling was performed on 14-μm cryostat sections as described previously (Nakayama et al.,1998), except for the exon-specific anti-1B1b antibody, which required antigen retrieval by boiling the sections for 30 minutes.

In situ hybridization was performed using a pan-specific Mitfprobe as described (Nakayama et al.,1998), or a PCR-generated exon 1B1b-specific probe (primer sequences available upon request).

RNA was isolated by two means. (1) RNA extractions using the RNeasy Mini Kit (Qiagen). E9.5 and 10.5 eye primordia were manually dissected along with the surrounding tissue. E11.5-P0 retinal tissue was separated manually from the surrounding RPE/mesenchyme, except where indicated. cDNA was prepared using the Superscript RT-PCR Kit (Invitrogen), using 1 μg of total RNA, and cDNA corresponding to 25 ng of total template RNA was used for each PCR reaction. (2) RNA was isolated using the Picopure RNA Extraction Kit(Arcturus) from 100 cells that were microscopically selected after trypsinization, using micromanipulator-controlled glass microinjection pipettes. In these cases, the amount of cDNA used for amplification with two rounds of PCR using nested primer pairs was estimated to correspond to approximately 1-10 pg of template RNA.

Real-time PCR was performed using an ABI Prism 7000 real-time PCR machine(Applied Biosystems). All RT-PCR products were sequenced at least once to confirm primer specificity. In the eye, sequencing provided no evidence for alternative splicing events linking exon 1A with exon 1J1b, nor exon 1J1b with exon 1C1b, as reported for cultured cell lines(Hershey and Fisher, 2005). Therefore, exons 1J and 1C are here considered as single exons. Primer sets used for RT-PCR and real-time PCR are available upon request.

For Fig. 3B, eyes from 20 E11.5 and 12 E15.5 wild-type embryos were manually dissected to give separate RPE/mesenchymal and retinal fractions. RNA was isolated using the RNeasy Mini Kit. The following procedure was used to generate standard curves and assay the test samples (Fig. 3B). First, isoform-specific cDNAs were generated by PCR using isoform-specific forward primers and a common reverse primer in exon 1B1b. Second, the cDNAs were quantitated spectrophotometrically and quantitation was confirmed by agarose gel electrophoresis. To generate standard curves, appropriate concentrations of the quantitated cDNAs were diluted into cDNA prepared from hearts of Mitf^mi-rw/mi-rw animals. The rw cDNA lacks exon 1B corresponding to the common reverse primer but mimics the cDNA complexity of the test samples, thereby providing an amplification environment for the standards that is similar to that provided by the test samples. Real-time PCR of the cDNA standards was performed with exon-specific forward primers and a common reverse primer in exon 1B1b. Standard curves were generated for each isoform, and separately for each repeat assay performed with the test sample. Real-time PCR of the test samples was performed (four independent assays, each in triplicate), using appropriate primers. Absolute amounts of cDNAs were determined using the appropriate isoform-specific standard curves. The calculated amounts represent the levels of the respective isoforms in cDNA prepared from 25 ng of total RNA and confirmed as giving equal amplification of Gapdh and β-actin cDNAs. Considering that mRNAs represent approximately 3% of total RNA, and RT efficiency is approximately 10%, the isoform-specific levels shown in Fig. 3B correspond to approximately 75 pg of mRNA.

For Fig. 4, 14 whole eyes each from E12.5 wild-type and Mitf^mi-rw/mi-rw embryos, and from P0 wild-type and Mitf^mi-rw/mi-rw mice were used. RNA isolation, RT reactions, and real-time PCR were as above except that: (1)single-exon-specific primers were used; (2) the measurements were repeated three times, each time in triplicate; and (3) the results are given as the ratio between the amounts measured in Mitf^mi-rw/mi-rw and in wild-type cDNAs. This means that unlike for Fig. 3B, there was no need for isoform-specific standard curves. However, to correct for variations between samples, wild-type and mutant results were first normalized to the internalβ-actin and Gapdh levels.

The Mitf^mi-rw mutation was characterized by PCR and Southern blots (see Fig. S1 in the supplementary material). The various Mitf isoforms were cloned and their activities assayed in vitro (see Fig. S2 in the supplementary material).

For details of ChIP assays using anti-CHX10 antibodies and amplification of Mitf promoter DNA see Figs S3, S4 in the supplementary material. Primer sequences are available upon request.

Previous studies using pan-specific in situ hybridization probes and antibodies have shown that Mitf expression in the developing eye begins at around embryonic day (E) 9.5 and lasts until postnatal day (P) 0(Nakayama et al., 1998; Nguyen and Arnheiter, 2000). To analyze the isoform composition of this pan-specific signal, we first used an in situ hybridization probe specific for exon 1B1b, which is common to all isoforms except M-Mitf (Fig. 1), a major component of neural crest-derived choroidal and iris melanocytes. As shown in Fig. 2A, exon 1B1b is expressed in the optic vesicle at E9.5, is downregulated in the distal optic vesicle at E10.0, and is then concentrated in the developing RPE as previously reported for a single time point(Amae et al., 1998). As expected, no labeling was seen in choroidal and iris melanocytes, which start to accumulate at around E11.5 on the proximal side of the RPE and normally are labeled with pan-specific probes (Nakayama et al., 1998). Also, no labeling was seen on sections from Mitf^mi-rw (Mitf red-eyed white) embryos(Fig. 2A, right-most panel),which carry a genomic deletion of exon 1B(Hallsson et al., 2000) but express Mitf, as detectable with pan-specific probes (see below, Fig. 5F).

To determine whether exon 1B1b-containing mRNAs are translated into corresponding proteins, we performed a series of immunoblot,immunoprecipitation and immunohistochemical analyses. The specificity of an antibody for exon 1B1b-containing MITF (reactivity with A-, J- and D-MITF, but not M-MITF) is shown in Fig. 2B, along with the pan-specific reactivity of a mouse monoclonal antibody to the MITF C-terminal domain (CTD) and the absence of reactivity of either antibody with normal fibroblast extracts. Using these antibodies, we then performed a combined immunoprecipitation/immunoblot assay of extracts of RPE/choroids and retinas that were manually dissected and separately pooled from approximately 100 eyes of wild-type E12.5 embryos. The results revealed at least three bands in RPE/choroid and one band of lower electrophoretic mobility in retina (Fig. 2B,bottom right panel, arrowheads). This finding indicates that RPE/choroid and retina express distinct exon 1B1b-containing isoforms, but, based on electrophoretic mobility alone, their precise identification is not feasible because post-translational modifications might potentially modify their apparent molecular weights.

In immunohistochemical assays (Fig. 2C), both exon 1B1b- and pan-specific rabbit antibodies gave overlapping expression patterns in the neuroepithelial parts of the developing eye, starting from E9.5, when MITF is expressed in the entire optic vesicle,through E17.5 and beyond (not shown), when it is expressed predominantly in the RPE. Neural crest-derived choroidal melanocytes, however, were only labeled with the pan-specific antibody (arrows in Fig. 2C, panel labeled wt E15.5). Both exon 1B1b- and pan-specific antibodies gave a low-level labeling in the retina and surrounding mesenchyme when compared with equivalently treated sections from Mitf^mi-rw/mi-rw embryos, which lack exon 1B (Hallsson et al.,2000), or with Mitf^{mi-vga-9/mi-vga-9} embryos,which lack MITF entirely (Hodgkinson et al., 1993; Nakayama et al.,1998) (Fig. 2C,panels labeled Mitf^mi-rw/mi-rw or Mitf^{mi-vga-9/mi-vga-9}). This result is consistent with the observation presented in Fig. 2B, showing that E12.5 retinas continue to express at least some isoform(s) of MITF.

To determine the expression profile of individual isoforms, we exclusively used RT-PCR, primarily because it provides the sensitivity to detect low-abundance isoforms. For the early time points (E9.5-10.5), RNA was obtained from whole-eye primordia including the surrounding tissue, and for the later time points (E11.5-P0), separately from RPE/choroidal and retinal fractions. cDNAs were amplified using either the respective 5′exon-specific forward primers in combination with a reverse primer in exon 1B1b, or, in the case of M-Mitf, a reverse primer in exon 2A. Pan-specific primers corresponded to exons 2A and 4. The number of PCR cycles was adapted for each primer pair to give amplification in the linear range. As shown in Fig. 3A, the first signal was seen at E9.5 with pan-specific primers and weakly with A- and J-Mitf-specific primers, whereas H- and D-Mitf-specific amplification gave the first signals at E10.5. No other isoforms were detected at these early stages.

During the subsequent developmental time points, the different isoforms could be grouped into four distinct expression patterns(Fig. 3A). First, throughout the period from E11.5 to P0, A- and J-Mitf showed similar expression profiles,both in the retina and the RPE/choroid. Second, H-Mitf was preferentially found in the RPE/choroid but was present at low levels in the retina as well,particularly after E15.5. Third, D- and M-Mitf were found only in the RPE/choroid and not the retina. Fourth, C-, MC-, E- and B-Mitf were undetectable, or only barely detectable, throughout development in either retina or RPE/choroid (not shown). In support of these results, from E11.5 to P0, pan-specific amplification showed higher expression levels in RPE/choroid compared with retina. Additional assays indicated that the Mitf-related genes Tfe3, Tfeb and Tfec (also known as Tcfe3, Tcfeb and Tcfec, respectively - Mouse Genome Informatics) (Hallsson et al.,2007), were also expressed in the developing eye, and amplification for Chx10, which in the eye is retina-specific(Liu et al., 1994), gave a signal only in the retinal fraction. This latter finding, together with the fact that D- and M-Mitf were only seen in the RPE/choroidal fraction,indicates that efficient tissue separation had been achieved.

The above analysis established a temporal expression profile for each isoform, but did not address the question of which isoform is expressed specifically in the RPE because the RPE/choroidal fraction contains other cells, including choroidal melanocytes, that might contribute exon 1B1b-containing isoforms detectable by PCR. Therefore, we repeated the above analysis by specifically selecting retinal, mesenchymal and RPE cells. For these experiments, we used eyes from Mitf^mi-bw (Mitf black-eyed white) embryos, whose only pigmented cells are RPE cells as they lack choroidal melanocytes (Yajima et al., 1999). In postnatal eyes of such mice, M-Mitf mRNAs are missing, full-length exon 1B1b-containing mRNAs are reduced, and skipping of common downstream exons is increased(Yajima et al., 1999). At E11.5 through E17.5, however, we found no differences between Mitf^mi-bw and wild-type eyes in Mitf expression,except that M-Mitf was lacking in the mutant (not shown). Hence, the use of separately trypsinized RPE/mesenchymal and retinal fractions of E15.5 Mitf^mi-bw/mi-bw allowed us to select microscopically small populations (100 cells) of pure mesenchymal, RPE or retinal cells (for details see Materials and methods). The respective cDNAs were amplified in two rounds of PCR using nested primer sets. This analysis confirmed that A-Mitf is ubiquitous and showed that H- and D-Mitf are not only present but in fact enriched in the RPE, compared with retina and mesenchyme. No signals were found for the low-abundance J-Mitf and, as expected, for M-Mitf(Fig. 3B, lanes 1-3). The results are consistent with those obtained from eye tissues dissected using standard techniques (Fig. 3B,lanes 4-6), where A-Mitf and low levels of J-Mitf were found in all fractions(lanes 4-6), H- and D-Mitf were enriched in the fraction that contains RPE and mesenchyme (lane 5), and M-Mitf was absent from all fractions.

The above experiments showed which isoform was expressed in which cell type but did not address the question of their expression levels. In fact, because different primer pairs amplify cDNAs with distinct efficiencies, differences in band intensities cannot be used for comparisons between isoforms. We therefore subjected cDNAs generated from separately pooled RPE/mesenchymal and retinal RNA fractions to real-time PCR using isoform-specific standard curves(see Materials and methods). Fig. 3B (right panel) shows the mean amounts of cDNAs obtained from four repeat assays each for E11.5 and E15.5. Consistent with the results described above, A- and J-Mitf accumulated to similar levels in RPE and retina, and H- and D-Mitf reached higher levels in RPE than in retina. We conclude, therefore, that both the future retina and the future RPE continue to express A- and J-Mitf and that the major isoforms in RPE are D-Mitf and,increasingly with developmental time, H-Mitf.

To address the question of whether the major RPE isoforms, H- and D-Mitf,are functionally relevant, we analyzed mice homozygous for the Mitf^mi-rw (Mitf red-eyed white, for short, rw) allele which is characterized by a genomic deletion that encompasses the 1H-, 1D- and 1B-exons(Steingrimsson et al., 1994; Hallsson et al., 2000). These mice retain a patch of pigmented coat on the head and occasionally on the belly and have eyes, the sizes of which can vary both between individuals and between the left and right side of the same individual (example shown in Fig. 4A, left side). We first determined the exact boundaries of the deletion and found that it comprises a stretch of 86,346 bp starting downstream of exon 1E and ending upstream of exon 1M (Fig. 4A and see Fig. S1 in the supplementary material). The deletion has the consequence that the upstream exons, whether coding or not, splice into the next available downstream exon, exon 2A, rather than the normal exon 1B1b(Fig. 4A). Consequently, if translation initiates at the normal start codons in exon 1A, 1C and 1MC, no open reading frames are maintained beyond exon 2 (marked by red dotted lines in Fig. 4A). Novel open reading frames, however, are generated within the normally non-coding exons 1J and 1E(blue lines in Fig. 4A), but the potential AUG start codons do not conform to Kozak rules for efficient translation initiation (Kozak,1984; Kozak,1986).

To analyze the transcriptional profile of Mitf in eyes of rw embryos, we again used RT-PCR as described above, except that whole-eye tissue served as the source of RNA because dissection into RPE/choroid and retina was not feasible with the mutant eyes. Moreover, we used primers covering only the respective single exons to allow for equal product sizes in wild type and rw. The results(Fig. 4B) indicate that the expression levels of the upstream exons 1A, 1J and 1E, were increased, whereas those of the downstream exon 1M and the pan-specific exon 9 were decreased. These findings were confirmed by real-time PCR(Fig. 4C). The results showed that at E12.5, for instance, an approximately twofold increase in A-Mitf and a ninefold increase in the minor J-Mitf could not compensate for the lack of the major isoforms, H- and D-Mitf, and for the reduction in M-Mitf that was detectable in these whole-eye extracts(Fig. 4B). At P0, however,isoform-specific increases and decreases cancelled each other out, so that pan-specific amplification of exon 9 gave a rw/wild type ratio of∼1.0. Using primers in the specific upstream exons and exon 4, we further confirmed that the overexpressed upstream exons were indeed part of spliced RNAs (Fig. 4D). Note, however,that even though the exon 1E-specific forward and reverse primers gave a visible band in wild type (Fig. 4B), no correctly sized band was seen in wild type with primers in exons 1E and 4 (Fig. 4D). This confirms the finding mentioned above that spliced transcripts containing exon 1E are below the detection threshold in the developing eye.

The lack of normal open reading frames in the internally deleted mRNAs suggested that Mitf^mi-rw is a functional null allele. Intriguingly, however, when we tested corresponding expression constructs in transfection or in vitro transcription/translation assays, we found that MITF proteins were generated that reacted with the antibody to the MITF C-terminal domain. Further analyses showed that the translation of these proteins is initiated predominantly from an internal AUG codon corresponding to Met62 in exon 2B, and that the N-terminally truncated proteins accumulate in the cell nucleus and are transcriptionally active on target gene promoters, including the promoter of the melanin biosynthetic enzyme gene tyrosinase (see Fig. S2 in the supplementary material). These results suggest that rw mice might retain Mitf activity associated with the novel protein isoforms despite the lack of H- and D-Mitf.

To determine the in vivo consequence of the rw deletion, we analyzed the development of rw eyes in detail. As shown in Fig. 5, in situ hybridization with a pan-Mitf-specific probe indicated that Mitf mRNAs are indeed expressed at the expected locations in the developing rw eyes. At E12.5, however, their levels were reduced compared with wild type, but at E17.5, they were close to normal (Fig. 5, compare A with B, E with F). Immunohistochemistry using pan-specific antibodies indicated a reduced signal in rw at both ages, which is likely to reflect a reduced translational efficiency of the rw mRNAs compared with wild-type RNAs(Fig. 5, compare C with D, G with H). Despite the low MITF protein levels, however, a clear signal for tyrosinase was found in rw RPE, particularly at E17.5(Fig. 5I-L), in contrast to developing eyes from Mitf-null mutants, which lack tyrosinase expression (Nakayama et al.,1998). Consistent with this result, pigmentation, although at low levels, could be found in the rw peripheral RPE and iris(Fig. 5M,N) as well as in the neuroepithelial part of the adult iris(Fig. 5O,P). Nevertheless, RPE development was perturbed, either because the overall MITF protein levels were too low or because full-length proteins containing the normal N-terminal sequences were missing. Much as in developing eyes of other Mitfmutants (Müller, 1950; Bumsted and Barnstable, 2000; Nguyen and Arnheiter, 2000),the dorsal RPE became thick (Fig. 5Q,R), retained high-level PAX6 expression(Fig. 5Q-T), and eventually transdifferentiated into a laminated second retina (not shown). In some eyes,however, the RPE abnormalities were characterized not by a general thickening but by foldings (Fig. 5S,T), a much milder abnormality than seen in other rw eyes. The results suggest that the lack of H- and D-MITF and exon 1B leads to RPE abnormalities of variable severity, but that the novel MITF protein isoforms are still capable of inducing pigmentation of the ciliary margin and the iris.

The downregulation of Mitf during retinal development involves a pathway that is initiated by extracellular signals emanating from the surface ectoderm, notably FGF1/2, and includes the paired-like homeodomain transcription factor CHX10 that negatively regulates Mitf expression in the retina (schematically indicated in Fig. 6A). As previously shown in Chx10^orJ/orJ (Chx10-ocular retardation) mutant embryos, which lack functional copies of Chx10, the future retina is hypoplastic (Konyukhov and Sazhina,1966; Burmeister et al.,1996; Green et al.,2003) and Mitf mRNA and protein levels are increased(Rowan et al., 2004; Horsford et al., 2005). To confirm that the mutant retina hypoproliferates, we used phosphohistone H3 labeling to mark the cells at the G2-M transition. As expected, the thinner Chx10^or-J/or-J retina showed substantially fewer labeled cells (Fig. 6B,C). As shown in Fig. 6D,E, this phenotype is correlated with the retinal retention of Mitf RNAs containing exon 1B1b. To dissect the isoform composition of the upregulated RNA, we used limiting cycle RT-PCR of RNA extracted from wild-type and Chx10mutant eyes at E13.5. The results (Fig. 6F) revealed a substantial increase in H- and D-Mitf, but a barely perceptible increase in A- and J-Mitf. Theoretically, a specific retinal upregulation of A- and J-Mitf might have been partially masked by the ubiquitous expression of these isoforms outside the retina. This is unlikely,though, because retinal changes in the expression of cyclin D1, which is also expressed outside the retina, could easily be detected in the whole-eye extracts, and A- and J-Mitf continued to be expressed throughout development in the Chx10-positive wild-type retina (see Fig. 2).

To determine whether CHX10 interacts in vivo with Mitf promoters,we performed chromatin immunoprecipitation (ChIP) assays using extracts of eye primordia at E10.5-11.0, corresponding to the period during which Mitf is downregulated in the retina. The results indicate a weak interaction of CHX10 with a binding site in the A-Mitf promoter, and strong interactions with a site in the H-Mitf promoter and with at least two sites in the D-Mitf promoter (see Figs S3, S4 in the supplementary material). Taken together, the results suggest that H- and D-Mitf are the major, and likely direct, targets of the Chx10-mediated Mitf downregulation in the retina, whereas A- and J-Mitf are not much affected by Chx10.

The above results prompted us to determine whether the lack of H- and D-MITF and exon 1B, as seen in rw embryos, might be sufficient to correct the Chx10-mediated retinal hypoproliferation phenotype. Such a correction has thus far only been described with dominant-negative Mitf mutations that equally affect all isoforms(Konyukhov and Sazhina, 1966; Horsford et al., 2005; Bharti et al., 2006). Therefore, we intercrossed Chx10^orJ/orJ mice with Mitf^mi-rw/mi-rw mice to generate double homozygotes. Staining for cyclin D1 in conjunction with PAX6 and the neuronal marker TUJ1[β-3-tubulin (TUBB3)] indicated that, as expected, eyes from Mitf^mi-rw newborns(Fig. 7A,E,I) had a retinal thickness and lamination similar to those of wild type(Fig. 7D,H,L), whereas eyes from Chx10^orJ newborns showed the severe retinal hypoproliferation and retina-to-RPE transdifferentiation described previously(Burmeister et al., 1996; Rowan et al., 2004; Horsford et al., 2005)(Fig. 7B,F,J). In the double Mitf^mi-rw;Chx10^orJ homozygotes, however, cyclin D1 staining and retinal thickness were nearly normal, although retinal lamination was still abnormal (Fig. 7C,G,K). This result indicates that in Chx10 mutants, Mitf upregulation, once deprived of the contribution of H- and D-Mitf and exon 1B, no longer leads to retinal hypoproliferation. This is consistent with earlier observations showing that dominant-negative Mitfmutations partially correct the Chx10 mutant phenotype(Konyukhov and Sazhina, 1966; Horsford et al., 2005).

The dissection of the expression patterns of Mitf isoforms has shown that some isoforms, such as H- and D-Mitf, are present in a temporally and spatially restricted manner whereas others, such as A- and J-Mitf, are found ubiquitously both within and outside the optic neuroepithelium. The notion that H- and D-Mitf are more important for eye development than other isoforms rests on several lines of reasoning. First, H- and D-Mitf consistently accumulate to higher levels in the RPE than other isoforms. Second, retinal hypoproliferation in Chx10 mutants is associated with abnormally high levels of H- and D-Mitf, but not of A- and J-Mitf or other isoforms. Third, the lack of H- and D-Mitf in Mitf red-eyed whitemutants, although confounded by complex changes in mRNAs and protein isoforms,is clearly associated with developmental eye defects. Fourth, when H- and D-Mitf are missing in Chx10 mutants, retinal development is restored. It appears, therefore, that H- and D-Mitf need to be precisely regulated during eye development, whereas the absence of other isoforms, or their presence in the retina, is more readily tolerated.

The fact that upstream Mitf regulatory regions contain distinct transcription-factor-binding motifs that are conserved across mammalian and avian species (Hallsson et al.,2007) argues strongly for a transcriptional control of the different Mitf isoforms. For instance, the ubiquitously active A-Mitf promoter has potential binding sites for transcription factors with widespread expression patterns such as MYC/MAX, STAT and SP1(Hallsson et al., 2007). By contrast, the H- and D-Mitf promoters have potential binding sites for the retina-specific transcription factor CHX10(Hallsson et al., 2007) (this paper), precisely the factor involved in the retinal downregulation of these two promoters. In fact, ChIP assays clearly indicate that at least some of these sites are occupied by CHX10 in vivo, suggesting a direct regulation of the respective Mitf promoters by this transcription factor.

A separate control of each promoter does not however exclude the presence of additional mechanisms by which distinct Mitf mRNAs are co-regulated. A transcriptional co-regulatory mechanism has been suggested by the observation, made by immunofluorescence, that Pax2/Pax6 double knock-out mice lack MITF protein expression in the optic neuroepithelium and that the human A-MITF promoter contains binding sites for PAX2/PAX6,suggesting that it acts as a control region for all ocular MITFexpression (Bäumer et al.,2003). These binding sites, however, are not well conserved across mammalian and avian species. Hence, it is possible that even in Pax2/Pax6 double knock-out mice, the ubiquitously expressed A- and J-MITF, which are undetectable by immunofluorescence on histological sections(data not shown), might continue to be present, and only the major and immunofluorescently detectable H- and D-MITF are absent. A different,post-transcriptional co-regulatory mechanism is suggested by the conservation,in the 3′UTR, of potential recognition sites for microRNAs, some of which are thought to be controlled by Chx10(Hallsson et al., 2007; Xu et al., 2007). In fact,microRNAs are excellent candidates to contribute to the retinal downregulation of all Mitf mRNAs that share the same 3′UTR, but the contribution of these microRNAs is likely to be small and might not exceed the minor effects that Chx10 has, for instance, on the levels of A- and J-Mitf.

The fact that each of the nine promoters is linked to a unique exon also leads to the important question of whether multiple promoters exist solely to provide the correct levels of MITF protein at each developmental time point and location, or whether they exist to generate mRNAs or proteins with isoform-specific functions. Indeed, some evidence suggests that sequence differences at the MITF N-terminus can influence the proteins' activities. For instance, ectopic expression of ascidian Mitf, the N-terminal sequence of which resembles A-MITF, induces additional pigment cells in ascidian larvae, as does mouse A-MITF, but not mouse H- or M-MITF(Yajima et al., 2003). Furthermore, A-, MC- and E-MITF regulate distinct sets of target genes in Mitf-null mutant mast cells(Shahlaee et al., 2007). There is also evidence, however, that expression levels are more important than specific sequences in determining MITF activities. For instance, regardless of whether A-MITF or M-MITF was expressed in quail neuroepithelial cells,pigmented colonies arose that consisted of RPE-like epithelial cells, in which MITF was low, and of neural crest-like dendritic cells, in which MITF was high(Planque et al., 2004). Our findings with the natural Mitf^mi-rw deletion suggest, in fact, that low levels of N-terminally truncated MITF proteins are capable of inducing pigmentation in the ciliary margin and the neuroepithelial portions of the iris, whereas in the complete absence of Mitf no such pigmentation is ever seen (Hodgkinson et al., 1998; Nakayama et al.,1998). This finding is consistent with the fact that N-terminally truncated MITFs retain activity on promoters of pigmentation genes, including the promoter of tyrosinase, which encodes the rate-limiting enzyme for melanin synthesis (see Fig. S2 in the supplementary material). On the other hand, the dorsal RPE is still abnormal in Mitf^mi-rw embryos. It is conceivable, therefore, that full RPE development, in contrast to mere pigmentation, depends on the presence of the upstream coding exons, perhaps on exon 1B1b in particular. This exon contains a glutamine-rich domain, thought to be involved in protein-protein interactions, that is conserved in vertebrates and flies (Hallsson et al.,2007) and is reminiscent of similar domains in other transcription factors such as CREB and SP1. That cell differentiation does not require this domain, however, is evident from the fact that M-MITF is sufficient to induce pigmentation upon experimental expression in cultured cells(Bejar et al., 2003), and that Mitf^mi-rw mice show RPE and iris pigmentation as well as some pigmented spots in the coat. In any event, a definitive answer to the question of why there are distinct promoters must await results from targeted promoter/exon knockouts and isoform-specific rescue experiments.

In addition to highlighting the importance of alternative promoter use, our study of the Mitf^mi-rw deletion also draws attention to alternative translation initiation. Alternative translation start sites have been observed in several genes and, much as with alternative promoter use, can contribute to functional diversity in wild type and alter disease in mutants(Sandelin et al., 2007; Scheper et al., 2007; Zhang et al., 2007). The fact that in the rw mutant mice, translation can be initiated from downstream start codons, suggests that these codons might be utilized for initiation even in the context of a wild-type sequence, although at low efficiency. Whether this is biologically relevant, however, remains to be explored.

Finally, the fact that the Mitf^mi-rw deletion is associated with random bilateral asymmetries in eye development merits special consideration. The fact that genetically identical cells, even if exposed to the same environment, can show substantial variations in phenotypic manifestations is thought to result from `stochastic' variations in gene expression that are likely to be due to variant epigenetic chromatin states(for a review, see Kaern et al.,2005). It is reasonable to assume that high levels of gene expression will generally buffer cells against random fluctuations in their molecular composition and so ensure a normal embryonic development. If,however, only threshold levels of gene expression are achieved, as is the case for Mitf in rw mice with their abnormal Mitftranscripts, stochastic fluctuations in epigenetic states might lead to major phenotypic differences between individual cells. Given that epigenetic states can be passed on through cell divisions, cellular variations might lead to tissue variations and so to differences between organs that are normally bilaterally symmetrical. Hence, our study of Mitf isoforms not only touches on the complexities by which a single gene controls organogenesis, but also on the intriguing question of how bilateral symmetries are being generated and maintained.

We thank Dr Vincent Hearing for antibodies to tyrosinase, the NINDS Sequencing Facility and the Animal Health and Care Section for excellent services, and Drs Dubois-Dalcq and Brian Howell and the members of the Mammalian Development Section for helpful comments on the manuscript. This work was supported in part by grants from the Italian Telethon Foundation to S.B. and by the Intramural Research Program of the NIH, NINDS.