The vertebrate retina is highly ordered along both its dorsoventral (DV) and nasotemporal (NT) axes, and this order is topographically maintained in its axonal connections to the superior colliculus of the midbrain. Although the graded axon guidance cues that mediate the topographic mapping of retinocollicular connections are increasingly well understood, the transcriptional regulators that set the DV and NT gradients of these cues are not. We now provide genetic evidence that Vax2, a homeodomain protein expressed in the ventral retina, is one such regulator. We demonstrate that in Vax2 mutant mice, retinocollicular projections from the ventral temporal retina are dorsalized relative to wild type. Remarkably, however, this dorsalization becomes systematically less severe in progressively more nasal regions of the ventral retina. Vax2 mutants also exhibit flattened DV and NT gradients of the EphA5, EphB2, EphB3, ephrin-B1 and ephrin-B2 axon guidance cues. Together, these data identify Vax2 as a fundamental regulator of axial polarization in the mammalian retina.
Although axial polarization of the retina is essential to perception of the visual world, the molecules that specify the retinal axes during development have not been intensively studied. For the most part, analyses of embryonic retinal organization and topography have instead focused on one important read-out of retinal polarization – the cell surface receptors and ligands that guide the precise wiring of retinal ganglion cell (RGC) neurons to their synaptic targets in the superior colliculus (SC) (O’Leary et al., 1999). These receptors and ligands act in concert to organize retinocollicular connections into a topographic map – a spatial ordering of axonal connections in which the Cartesian coordinates of a two-dimensional sheet of projecting RGC neurons are mapped onto the coordinates of a second two-dimensional sheet of target neurons in the SC. In the retinocollicular map, axons from RGCs in the nasal retina project to and form synapses in the caudal end of the SC, while RGCs in the temporal retina project to targets in the rostral SC. RGCs located at intermediate NT retinal positions project to correspondingly intermediate caudal-rostral positions in the SC. The dorsoventral (DV) axis of the retina is similarly mapped onto the lateral-medial axis of the SC.
The proteins most closely tied to the topography of retinocollicular mapping are the receptor tyrosine kinases of the EphA family, together with their ligands, the ephrin-A proteins (Flanagan and Vanderhaeghen, 1998; O’Leary and Wilkinson, 1999). Two key lines of recent genetic evidence (Frisén et al., 1998; Brown et al., 2000; Feldheim et al., 2000), together with a large body of earlier in vitro membrane stripe and in vivo misexpression studies, have demonstrated that a low-nasal-to-high-temporal retinal EphA receptor gradient, combined with a reciprocal low-rostral-to-high-caudal collicular ephrin-A gradient, serve to order the mapping of the NT axis of the retina onto the caudal-rostral axis of the SC. The cell-surface molecules that mediate mapping of the orthogonal retinocollicular axes are less well understood. Although clear DV gradients of EphB2 and EphB3 receptor expression have been observed in the vertebrate retina, and medial-lateral gradients of ephrin-B ligands have been detected in the SC (Marcus et al., 1996; Braisted et al., 1997; Holash et al., 1997; Schulte et al., 1999), these gradients are not reciprocally configured, as would be required for a chemorepellent action of the ephrin-Bs, and genetic tests of the importance of the EphB/ephrin-B signaling system to retinocollicular mapping have yet to be reported.
Less clear still are the transcriptional control mechanisms through which the NT and DV retinal gradients of the EphA and EphB receptors are established during development. Recently, two candidates for transcriptional regulators of DV polarization of the retina – Tbx5 and Vax2 – have been identified (Barbieri et al., 1999; Koshiba-Takeuchi et al., 2000; Schulte et al., 1999). Vax2 (for ventral anterior homeobox 2) is one of two vertebrate-specific homeobox genes that are structurally related to, and that have almost certainly evolved from the duplication of, the vertebrate Emx genes (Barbieri et al., 1999; Hallonet et al., 1998; Ohsaki et al., 1999). Loss-of-function experiments for Vax1 have demonstrated that its product plays essential roles in axon guidance and major tract formation in the developing forebrain (Bertuzzi et al., 1999; Hallonet et al., 1999). Vax2, which carries a homeodomain identical to that of Vax1, is a candidate regulator of the retinal DV axis for two reasons. First, it is steeply graded in its expression along this axis in chick and frog embryos, with highest expression ventrally. And second, dominant gain-of-function studies in these embryos have shown that, when misexpressed in the dorsal retina, Vax2 is capable of ventralizing this tissue, as assessed by (1) the altered expression of DV marker genes, both putative guidance cues such as EphB2/B3 and putative transcriptional regulators such as Tbx5, and (2) the altered projection of RGC axons to the midbrain. These observations have been interpreted as indicating that Vax2 may function as a global ‘ventralizing’ regulator of the developing eye. In this report, we describe the generation of Vax2–/– mice, and the use of these mutants to perform loss-of-function tests of this hypothesis.
MATERIALS AND METHODS
The mouse Vax2 gene was cloned from a strain 129/sv genomic library and was inactivated by deletion/replacement (see Fig. 1G). Vax2–/– mice were generated by standard procedures (Tybulewicz et al., 1991). Electroporated W95 embryonic stem (ES) cell clones were doubly selected with G418 and FIAU, and were screened for homologous recombination by Southern blot, using the 230 bp 3′-external probe indicated in Fig. 1. Four positive ES cell clones were microinjected into C57/Bl6 blastocysts to generate chimeric mice, which were then mated with C57/Bl6 to produce heterozygous knockouts, all according to standard procedures (Capecchi, 1989). Although most Vax2–/– mice did not exhibit overt postnatal phenotypes, ∼7% of homozygotes were abnormal, in that they either were unusually small in size, exhibited a trembling phenotype or displayed small eyes.
In situ hybridization with 33P-radiolabeled or digoxigenin-labeled riboprobes were performed as described previously (Braisted et al., 1997; Zhadanov et al., 1995). Dissected mouse embryos were anesthetized and immersion fixed in 4% paraformaldehyde/PBS overnight at 4°C, followed by overnight infiltration with 20% sucrose. Cryoprotected tissues were then embedded in OCT medium (Miles). Hybridization was performed with 20 μm coronal and transverse sections of retina at birth (P0), and with embryos at E9.5, E11.5 and E14.
In vivo anterograde labeling was performed between postnatal day 7 (P7) and P9 by focal injection of 1,1′-dioctadecyl-3,3,3′3′-tetramethylindocarbocyanine perchlorate (DiI) (Molecular Probes; 8% w/v in N,N-dimethylformamide) into the left eye, as described previously (Brown et al., 2000; Simon and O’Leary, 1992). Mice were sacrificed 1 day later, and termination zone (TZ) number and location in the contralateral superior colliculi were analyzed using a Zeiss LSM510 confocal microscope. The size and location of focal retinal DiI injection sites were identified post hoc.
Intraocular injection of fluorescently tagged cholera toxin B-subunit tracers (CTB) into one eye allowed for visualization of both ipsilateral and contralateral retinal projections. For anterograde tracing of the RGC axons, CTB conjugated to a fluorescein (FITC) fluorophore (List Biological Laboratories, Campbell, CA) was made in a 0.2% solution in 1% DMSO. Whole eye fill injections were carried out at P8 and the animals were perfused with 4% paraformaldehyde 24 hours later. Brains were cut coronally in 200 μm sections and subsequent imaging was performed using a confocal microscope (Zeiss).
Dual gradients of Vax2 expression in the mouse retina
Although the Vax2 homeodomain is identical to that of Vax1 (Barbieri et al., 1999; Schulte et al., 1999), the two Vax genes nonetheless exhibit distinct and largely complementary patterns of expression during embryogenesis in the mouse. At embryonic day 10.5 (E10.5) and thereafter, Vax1 mRNA is expressed in the ventral diencephalon and telencephalon, and in the optic stalk and disk, but is not prominently expressed in the developing neural retina (Bertuzzi et al., 1999; Hallonet et al., 1999). By contrast, the expression of mouse Vax2 mRNA is largely confined to the ventral half of the developing eye itself (Fig. 1A). These distinct embryonic expression domains in the mouse contrast with the situation in the chick, where a single gene (VAX) exhibits a hybrid Vax1/Vax2-like expression profile throughout the ventral diencephalon, optic stalk, and ventral retina (Schulte et al., 1999). At E11.5, mouse Vax2 mRNA is expressed in all cells of the ventral neural retina (Fig. 1B), but by birth becomes restricted to ventral RGCs (Fig. 1E). In addition to the previously reported high-ventral-to-low-dorsal retinal gradient of Vax2 mRNA (Fig. 1B,E), we have observed a shallower Vax2 gradient along the NT axis, with highest expression nasally (Fig. 1B-D,F). This dually polarized expression is evident throughout the neural retina at E14 (Fig. 1C,D), when RGC axons first reach the SC, and persists at least until birth in the mouse, when the DV patterning of RGC projections is well under way. The double Vax2 retinal gradient in the mouse is similar in orientation and configuration to the recently described dual retinal expression gradient of the bone morphogenetic protein 4 (BMP4) antagonist ventroptin in the chick (Sakuta et al., 2001).
Inactivation of the mouse Vax2 gene
In order to test the hypothesis that Vax2 may specify positional identity in the retina along the DV axis (Barbieri et al., 1999; Schulte et al., 1999), we generated Vax2 mutant mice (Fig. 1G-I). We replaced exon 2 of the mouse Vax2 gene, which encodes the first two essential α helices of the Vax2 homeodomain, with a G418 resistance (PGK-neo) cassette. In addition to deleting exon 2, this mutation introduces a frame shift that eliminates all Vax2-coding sequence downstream of exon 2 (see Fig. 1G, Materials and Methods).
Retinocollicular mapping in the Vax2 mutants
We first analyzed the projection of RGC axons from the retina to the SC in the Vax2 mutants. Normally, the DV and NT axes of the retina are precisely mapped onto the correspondingly orthogonal lateral-medial and caudal-rostral axes of the SC (Brown et al., 2000). If Vax2 is required for ventral patterning of the retina, then ventral RGCs that lack Vax2 should be dorsalized in terms of their projection to the SC; that is, they should project to lateral rather than to medial SC. In mice, the topography of the retinocollicular map is mature by postnatal day 7 (P7) (O’Leary et al., 1986). We found that most of the Vax2–/– mice (94 out of 101) were healthy and superficially normal for many months after birth; in marked contrast to Vax1 mutants (Bertuzzi et al., 1999), no postnatal colobomata were observed. We therefore labeled discrete loci of RGCs and their projecting axons across the full extent of the NT and DV axes of the mutant retina at P7-9. We performed focal injections of the lipophilic axon tracer DiI (Fig. 2), and then analyzed labeled projections to the contralateral SC 1 day later (see Materials and Methods). In total, we analyzed 40 injections (from 40 mutant mice). The Vax2 mutant retinae that we selected for study were histologically indistinguishable from wild type in terms of retinal lamination, cellular density, organization of plexiform layers and closure of the optic disk (data not shown).
Given that Vax2 is not expressed in the extreme dorsal retina, axon projections from dorsal RGCs should not be altered in the Vax2 mutants. This was the case. The dorsal RGC axons of mutant mice (n=8 dorsal injections, at varying NT positions, in 8 mice) entered the SC at its rostral lateral edge and formed tight termination zones (TZs) at expected locations in the lateral SC (Fig. 2S, and data not shown). In marked contrast, axon projections from most Vax2–/– ventral RGCs did not, with RGC axons from the ventral temporal retina being the most strongly affected. These ventral temporal axons entered the SC at the lateral rostral edge, and always formed single, well-circumscribed TZs in the lateral rostral SC (n=10/10) (Fig. 2A, lower panel, Fig. 2D-F). As both wild-type and Vax2+/– ventral temporal axons project to medial rostral SC (Fig. 2A, middle panel), the projection pattern of these Vax2–/– RGCs represents a complete retinal dorsalization of their mapping behavior. This dorsalization was robust and fully penetrant (Fig. 2D-F), and was reflected not only in the position of the TZs, but also in the site at which RGC axons entered the rostral colliculus (compare middle and lower panels in Fig. 2A).
Unexpectedly, however, the dorsalization of RGC projections became progressively less severe as DiI injections were moved to increasingly nasal regions of the ventral retina. At 80-85% of the retinal NT axis (where extreme temporal=100%), ventral RGC TZs remained aberrantly lateralized in their SC projection, but more than one lateral TZ was typically observed (Fig. 2G,H). As injections were moved to 60-80% of the NT axis, the size of ectopic lateralized TZs became progressively smaller, and an appropriate TZ appeared very near the expected medial position (Fig. 2I-K). For mid-ventral injections into the Vax2 mutant retinae (30-60% of the NT axis), collicular TZs typically appeared near the expected medial location, occasionally as single, well-formed TZs (Fig. 2M). When ectopic lateral TZs were observed, they were most frequently small (Fig. 2B, lower panel and Fig. 2N-R). Finally, ventral DiI injections into the most nasal regions of the Vax2 mutant retina (0-30% of the retinal NT axis) almost always yielded single tight TZs at the expected wild-type location in the medial SC (Fig. 2C, Fig. 2T-W). Innervation space in the rostral medial quadrant of the Vax2–/– SC vacated by lateralized ventral temporal RGC axons appeared to be occupied by TZs from RGCs whose axons would normally occupy extreme medial locations near the midpoint of the RC axis (Fig. 2L). Thus, the Vax2–/– axon projection phenotype progressed from extremely strong and fully penetrant in the ventral temporal retina (Fig. 2A,D-F) to almost non-existent in the ventral nasal retina (Fig. 2C,U-W). Note that this NT phenotypic progression is the inverse of the shallow gradient of Vax2 mRNA (see Discussion).
Maintenance of ipsilateral projections
At maturity, the vast majority (>95%) of RGC axons in the mouse cross at the optic chiasm and innervate the contralateral SC; the rare ipsilateral projections originate primarily in the ventral temporal quadrant of the retina (Sretavan and Kruger, 1998). Although ventral temporal RGC projections are those that are most strongly perturbed in the contralateral SCs of the Vax2 mutants, we nonetheless consistently detected ipsilateral RGC projections after full eye fills with fluorescent anterograde axonal tracers in these mice (Fig. 3; n=5). Indeed, in some mice, these ipsilateral projections appeared to be atypically abundant.
Flattened DV gradients of axon guidance cues
The aberrant RGC projections seen in the Vax2 mutants suggest that this transcription factor normally patterns the mouse retina beginning around E10, and thereby directly or indirectly controls expression of later position-dependent axon guidance cues. Vax2 misexpression in the dorsal chick and frog retina results in (1) the upregulated expression of mRNAs encoding the EphB2 and EphB3 receptor tyrosine kinases, which are normally ventrally restricted, and (2) the downregulated expression of mRNAs encoding their ligands ephrin-B1 and ephrin-B2, which are normally dorsally restricted (Barbieri et al., 1999; Schulte et al., 1999). Similarly, mRNA encoding a dorsally restricted transcription factor and presumed DV regulator – Tbx5 – is downregulated by these gain-of-function manipulations. Consistent with a subset of these earlier observations, we found that ventral expression of both the Ephb2 and Ephb3 mRNAs was lost in the Vax2–/– ventral retina (Fig. 4A and data not shown), and that conversely, ventral expression of the ephrin-B1 and ephrin-B2 mRNAs (Efnb1 and Efnb2 – Mouse Genome Informatics) was acquired (Fig. 4B and data not shown). In addition to these DV alterations, and consistent with our detection of a shallow NT Vax2 gradient, we also observed a partial reordering of the retinal NT axis in the ventral retina of the Vax2–/– mutants. For example, Epha5 mRNA, which is normally distributed in a low-nasal-to-high-temporal gradient throughout the DV axis (Brown et al., 2000), was upregulated in the ventral nasal mutant retina (Fig. 4C). Thus, each of the above changes leads to flattened or abolished gradients of guidance cue mRNAs that are normally graded along both the DV and NT axes (compare left and right panels in Fig. 4A-C). This global regulation of multiple guidance cues is apparently crucial, as preliminary analysis of mice doubly mutant for the Ephb2 and Ephb3 genes indicates that medial-lateral RGC axon targeting defects in these double EphB mutants are much less pronounced than those detailed above for the Vax2 single mutants (R. H., T. McLaughlin and D. D. M. O., unpublished).
Maintained DV gradients of transcriptional regulators
While Vax2 orchestrates the expression of several genes that encode graded axon guidance cues, the gradients of transcription factors that have been hypothesized to define retinal axial polarity, and thereby set up the guidance cue gradients, remained unchanged in the Vax2 mutants. Notably, the expression of Pax2, an early embryonic marker of the ventral retina (Koshiba-Takeuchi et al., 2000; Schulte et al., 1999) remained graded, low-dorsal-to-high-ventral, in early Vax2 mutant embryos (data not shown). Similarly, the inverse high-dorsal-to-low-ventral gradient of mRNA encoding the T box transcription factor Tbx5, which is the earliest known dorsal retinal marker (Koshiba-Takeuchi et al., 2000), was, unlike ephrin-B1 and -B2, unchanged in the Vax2 mutants at E10.5 and P0 (Fig. 4D, and data not shown). Although gain-of-function Vax2 misexpression experiments have suggested that Tbx5 is normally repressed by Vax2 (Schulte et al., 1999), our results demonstrate that this is not the case: loss of Vax2 in the ventral retina does not lead to Tbx5 upregulation. As Tbx5 misexpression experiments have similarly suggested that Vax2 may be repressed by Tbx5 (Koshiba-Takeuchi et al., 2000), analysis of retinal polarity in Tbx5 loss-of-function mutants will be required in order to clarify the relationship between these transcription factors. Nonetheless, our results indicate that the highly polarized expression of Tbx5 does not require Vax2. Similarly, the homeodomain protein Six6 and the winged-helix transcription factor Bf1 (Foxg1 – Mouse Genome Informatics) (Huh et al., 1999; Lopez-Rios et al., 1999), which are also hypothesized regulators of retinal axial polarity, and which at E9.5 exhibit expression in the ventral nasal (Bf1) and ventral temporal (Six6) mouse retina, respectively (Fig. 4E,F), were unchanged in their expression domains in the Vax2 mutants (Fig. 4G, and data not shown). These domains are dynamic for both Bf1 and Six6 from E9.5 to birth: Bf1 is first expressed in the ventral nasal retina (Fig. 4E) and by E12.5 moves to the dorsal nasal retina; and Six6 is first detected in the ventral temporal retina (Fig. 4F), but later occupies the entirety of the retina. These dynamic expression domains were unaffected in the Vax2 mutants, as assessed by in situ hybridization at E10.5, E12.5, E14.5, E16.5 and P0 (Fig. 4G, and data not shown).
Our loss-of-function analyses definitively demonstrate that Vax2 controls DV polarization of the mouse retina. This transcription factor directly or indirectly activates expression of ventral axon guidance cues such as EphB2 and EphB3, and at the same time represses dorsal cues such as ephrin-B1 and ephrin-B2. In the temporal regions of the ventral retina, the loss of Vax2 results in RGC projections to the contralateral SC that are completely dorsalized relative to their wild-type counterparts.
These results notwithstanding, our analyses also demonstrate that the ventralizing influence of Vax2 is in two respects incomplete. First, although Vax2 expression normally extends across the entirety of the ventral retina, the dorsalization of RGC projections in the Vax2 mutant retina does not. As detailed above (Fig. 2), dorsalization is robust only in the extreme ventral temporal retina, and becomes progressively less severe in progressively more nasal regions of the retina. In the extreme ventral nasal retina, mutant RGC projections are indistinguishable from wild type. And second, some of the phenotypes seen upon loss of Vax2 are not consistent with those reported for Vax2 gain-of-function studies. Most notably, loss of Vax2 does not lead to the activation of Tbx5 expression in the ventral retina, or to any other obvious perturbation in the high-dorsal-to-low-ventral gradient of Tbx5 expression. These latter observations suggest that transcriptional specifiers of vertebrate retinal polarity, like many transcriptional specifiers of retinal identity (Marquardt et al., 2001), operate independently.
The surprising finding that the axonal targeting errors of ventral temporal RGCs are much more severe than those of ventral nasal RGCs suggests that either: (1) Vax2 interacts with one or more additional transcriptional regulators that are similarly graded along the DV and NT axes, such that a ‘limiting concentration effect’ of these nuclear proteins is detected in the mutants only in the retinal quadrant where their aggregate level is normally lowest (Fig. 5A); or (2) the Vax2 gene, which is steeply graded along the retinal DV axis, normally interacts genetically with one or more genes that are steeply graded along the NT axis (Fig. 5B). With regard to the second possibility, other homeodomain transcription factor genes, such as SOHO1 and GH6 in the chick (Schulte and Cepko, 2000), have been shown to be graded along the retinal NT axis, and have been hypothesized to control the expression of NT guidance cues such as EphA receptors and their ephrin-A ligands. The homologs of SOHO1 and GH6 remain to be analyzed in the mouse, but the transcription factor Bf1 is specifically expressed, from E8.5-E9.5, in the mouse ventral nasal retina (Fig. 4E). Although the EphA/ephrin-A guidance cues have been analyzed almost exclusively in terms of RGC mapping along the rostral-caudal axis of the SC (Brown et al., 2000; Feldheim et al., 2000), it is interesting to note that medial-lateral mapping anomalies have been consistently observed in ephrin-A2/A5 double mutants (Feldheim et al., 2000). If, as schematized in Fig. 5B, Vax2 normally acts in concert with one or more transcription factors that are graded along the NT axis, then RGCs in the ventral temporal retina would again be those most sensitive to the loss of Vax2. Direct tests of these and related models must await the generation and analysis of a set of single and compound mutants of Vax2 with Bf1, SOHO1, GH6, ephrin-A2 and ephrin-A5, among others.
Together, our results identify Vax2 as an essential specifier of both DV and NT axial polarity in the developing mammalian eye. They demonstrate that the steeply graded expression of Vax2 that first appears near the midpoint of mouse embryogenesis is essential to the ventral specification of the retina, most prominently in its ventral temporal quadrant. Importantly, they further demonstrate that this specification is independent of more widely expressed transcription factors that have also been thought to act during axial determination of the eye, and suggest that these regulators normally act in concert with Vax2.
This work was supported by grants from the NIH and the Italian Telethon Foundation (G. L., D. D. M. O. and S. B.), by postdoctoral fellowships from the Italian Telethon Foundation (S. B.) and the Swiss National Science Foundation (R. H.), and by the Medical Scientist Training Program at UCSD (S. M.). We thank Qingxian Lu and Todd McLaughlin for helpful advice and discussion, Arthur Brown and Todd McLaughlin for in situ hybridization probes, and Patrick Burrola, Darcie Baynes, Dario Strina and Arjay Clemente for technical assistance.