Pax6 marks Reelin⁺ astrocytes

Our microarray analysis also indicated that Pax6 mRNA was up-regulated, together with Reelin and Slit1, in the Olig1,2^−/− population. Double-labeling experiments indicated that Pax6 marks a subpopulation of astrocytes in the ventro-lateral white matter (Fig. 2A–D), whose distribution is similar to that of Reelin⁺ astrocytes (Fig. 2E–F). Pax6⁺ cells constitute approximately 40% of all GFAP⁺ and NF1A⁺ WMAs (Fig. 2A–D, L). Double labeling for Reelin and Pax6 confirmed that 100% of Reelin⁺ astrocytes co-express Pax6, suggesting that Pax6 marks the VA1 and VA2 subpopulations (Fig. 2E, F, arrows; L). Since Slit1⁺ astrocytes are equally distributed between the VA3 and VA2 populations, ~50% of Slit1⁺ astrocytes (VA2) should be Pax6⁺, and this was indeed the case (Fig. 2H, arrow vs. arrowhead; L). Pax6 was excluded from the ventro-medial domain of Slit1 expression (Fig. 2G), corresponding to the VA3 subset. In the VZ at E13.5, triple-labeling for Pax6, Reelin and Slit1 indicated that the spatial relationship between expression of the HD factor and the two astrocyte markers was similar to that observed in the white matter at E18.5: Pax6 overlapped with all of the Reelin-expressing progenitors (Fig. 2I), and a subset of the Slit1-expressing progenitors (Fig. 2J).

Open in a separate window

Figure 2

Expression of Pax6 in Reelin⁺ astrocytes and their precursors

(A–H) Double-labeling for Pax6 (red) and the indicated markers (green) at E18.5. (B, D, F, and H) are higher magnification views of the areas indicated by arrows in (A, C, E and G), respectively. (I–K) Triple labeling for Pax6, Reelin, and Slit1-GFP in the E13.5 ventral VZ, displayed as pairwise comparisons from the same section. (L) Quantification of Pax6 expression among total (GFAP⁺ or NFIA⁺) astrocytes, and Reelin⁺ or Slit1⁺ WMAs.

To investigate whether Pax6⁺ cells in the VZ are precursors of Pax6⁺ WMAs, we pulse-labeled embryos at E13.5 with BrdU, and chased to E15.5 or E18.5. We observed a progressive shift in the distribution of both total Pax6⁺ cells, and of BrdU-labeled Pax6⁺ cells, from the VZ to the mantle zone (MZ) to the WM, during the chase period (Supplemental Figure S2J, K). The proportion of Pax6⁺ cells in the WM increased ~2.5-fold from E15.5 to E18.5, and this increase could be quantitatively accounted for by assuming that, during the same interval, Pax6⁺ cells in the VZ migrate to the MZ, and from the MZ to the WM (see Supplemental Fig. S2, legend). The most parsimonious interpretation of these data is that at least some Pax6⁺ cells in the E13.5 VZ are precursors of Pax6⁺ WMAs.

Pax6 is required for Reelin expression in astrocytes

Pax6 has been shown to control the positional identity of a subset of INs in the ventral spinal cord (Burrill et al., 1997; Ericson et al., 1997). We therefore asked whether it might also play a role in the determination of astrocyte positional identity. Pax6^lacZ/lacZ homozygous mutants (St-Onge et al., 1997) exhibited a striking reduction in Reelin mRNA expression in the ventro-lateral white matter at E18.5, while expression in the gray matter appeared only modestly reduced (Fig. 3A–B, arrows vs. E–F, arrowheads). Immunostaining confirmed a loss of Reelin expression in GFAP⁺ and NFIA⁺ astrocytes (Fig. 3C–D, arrows vs. G–H, arrowheads). Quantification indicated a strong and statistically significant reduction in both the percentage (Fig. 3I) and absolute number (Fig. 3J) of Reelin⁺ astrocytes, using either GFAP or NFIA as counter-stains.

Open in a separate window

Figure 3

Astrocyte subtype conversion in Pax6^−/− mice

(A–B, E–F) In situ hybridization for Reelin mRNA in E18.5 spinal cord of wild-type (A–B) and Pax6 mutant (E–F) embryos. (B) and (F) are higher-magnification views of (A) and (E), respectively. Arrows in (A, B) indicate Reelin⁺ cells in the white matter. (C–D, G–H) Double-labeling for Reelin and GFAP (C, G) or NFIA (D, H) in wild-type (C–D) and Pax6 mutant (G–H) embryos. (I–J) Quantification of Reelin expression by WMAs in E18.5 wild type (blue bars) and Pax6^−/− (red bars) spinal cord. . (I) The percentage of total GFAP⁺ or NFIA⁺ WMAs expressing Reelin is significantly reduced in the Pax6^lacZ/lacZ mutant (***, p<.001). (J) The absolute number of GFAP⁺ and NFIA⁺ astrocytes in the white matter is not changed in the Pax6^lacZ/lacZ spinal cord, nor is the average number of Pax6⁺ GFAP⁺ astrocytes. (*** and **, p<.001 and .002, respectively). . The data are derived from 5 wild type and 6 mutant embryos from 3 independent litters. (K–P) Spinal cord sections from Pax6^+/+; Slit1^GFP/+ (K–M) and Pax6^lacZ/lacZ; Slit1^GFP/+ (N–P) embryos double-labeled for the indicated markers. Arrows indicate VA1 astrocytes (K–M) that exhibit de-repression of Slit1 in the mutant (N–P). Red staining in (M, P) represents anti-Pax6 (M) or anti-βgal (P). (Q) Slit1 expression among total astrocytes (GFAP⁺ or NFIA⁺) is significantly increased in Pax6^−/− embryos (*** and *, p = .0003, and .012, respectively). (H) Schematic illustrating the changes in Reelin and Slit1 expression by ventral astrocytes in the Pax6 mutant.

We detected no reduction in the total number of astrocytes in the spinal cord of E18.5 Pax6 mutant embryos (Fig. 3J, GFAP⁺ and NFIA⁺), consistent with an earlier analysis using GLAST as a generic astrocyte marker (Ogawa et al., 2005). Since Reelin⁺ cells constitute ~60% of all ventral WMAs, this argues against the idea that Reelin⁺ astrocytes selectively die in the absence of Pax6, otherwise there should be a measureable reduction in overall astrocyte number. However, it is possible that the mutation could cause the selective death of Pax6⁺ astrocytes, which is then compensated by the expansion of other, Pax6⁻, astrocyte subpopulations. To address this possibility, we employed the marker gene lacZ, present in the Pax6 knockout allele (St-Onge et al., 1997), to trace the fate of Pax6⁺ cells in the absence of Pax6 function. The percentage of Pax6⁺ or βgal⁺ cells that expressed GFAP or NFIA was identical in wild-type and mutant embryos, respectively (Fig. 3I; GFAP⁺/[Pax6⁺ or βgal⁺]; NFIA⁺/[Pax6⁺ or βgal⁺]). Thus, the loss of Pax6 did not produce a failure of generic astrocyte differentiation by Pax6⁺ cells. Furthermore, the absolute number of GFAP⁺ Pax6⁺ (or GFAP⁺ βgal⁺) cells was unaffected in the mutant (Fig. 3J; [Pax6⁺ or βgal⁺], GFAP⁺), confirming that there was no selective death of Pax6⁺ astrocytes. However, there was a strong reduction in the percentage of Pax6-βgal⁺ cells that were Reelin⁺ in the mutant, compared to wild-type (Fig 3I, Reelin⁺/[Pax6⁺ or βgal⁺]). Taken together, these data indicate that Pax6 is required for the expression of Reelin in VA1 and VA2 astrocytes, but not for their generic differentiation to astrocytes, migration to the white matter or survival.

Dorsal expansion of astrocytic Slit1 expression in the absence of Pax6

The foregoing analysis left open the question of whether the loss of Pax6 function simply caused a failure of Reelin expression by astrocytes, or rather a change in the positional identity of VA1 astrocytes. In the latter case, one might expect that presumptive VA1 astrocytes would acquire a Slit1⁺ phenotype. Due to a lack of adequate antibodies to Slit1, we inter-crossed the Pax6-lacZ mice with Slit1-GFP mice, and analyzed the expression of Slit1-GFP in Pax6 mutants (Slit^GFP/+; Pax6^lacZ/lacZ). In Pax6 mutant embryos, there was a significant increase in the percentage of GFAP⁺ or NFIA⁺ astrocytes that expressed Slit1-GFP (Fig. 3Q, red bars). Moreover, the spatial domain of Slit1-GFP expression appeared to expand dorsally in the mutant (Fig. 3K-M vs. N–P; arrows). Most importantly, the percentage of Pax6⁺ astrocytes that expressed Slit1 increased from ~50–60% in wild-type embryos, to virtually 100% in the mutant (Fig. 3Q, Slit1⁺/Pax6⁺ or βgal⁺), reflecting the acquisition of Slit1 expression by dorsal Pax6-βgal⁺ cells (Fig. 3M, P, arrows). Taken together, these data indicate that in the absence of Pax6 function, not only is Reelin expression lost, but Slit1 expression is de-repressed, in VA1-type astrocytes (Fig. 3R). This suggests that loss of Pax6 causes a ventralization of the VA1 domain, such that supernumerary VA3 astrocytes are now found in a more dorsal, ectopic location, analogous to the phenotypes of other spinal cord patterning mutations that change the number and distribution of interneuron subtypes (Jessell, 2000; Shirasaki and Pfaff, 2002).

Pax6 is sufficient to promote Reelin and repress Slit1 expression in astrocytes

To investigate whether Pax6 plays an instructive role in regulating the expression of markers of astrocyte positional identity, we performed in vivo gain of function experiments, in the embryonic chick neural tube. We first examined Reelin and Slit1 expression in the unmanipulated E12 chick spinal cord, to confirm that these genes labeled subsets of WMAs, as in the mouse. We observed white matter domains containing Reelin⁺/Slit1⁻, Reelin⁺/Slit1⁺ and Reelin⁻/Slit1⁺ VA1–3, astrocytes, as in the mouse (Supplemental Figure S3). We therefore examined the effect of Pax6 mis-expression on these ventral astrocyte populations. The spinal cord of E2 chick embryos was electroporated with replication competent RCAS(B) retroviruses carrying either the chick Pax6 or GFP genes. Electroporated embryos were harvested and analyzed at E12, following 10 days of development in ovo. Pax6 mis-expression significantly increased the percentage of Reelin⁺ cells among NFIA⁺ cells in the white matter, compared with control GFP mis-expressing embryos (Fig. 4F vs. H, “WM;” CC, “Reelin,” “E”). The total number of NFIA⁺ astrocytes was not affected by Pax6 electroporation (data not shown). These data indicate that Pax6 is sufficient, as well as necessary, for Reelin expression by ventral astrocytes in vivo.

Open in a separate window

Figure 4

Pax6 is sufficient to promote Reelin expression and repress Slit1 expression in astrocytes

(A–D) Immunostaining for RCAS retroviral coat protein AMV in E12 spinal cord of chick embryos electroporated with either RCAS-GFP (A, B) or RCAS-Pax6 (C, D “WM,” white matter; “GM,” gray matter. (E-BB) Antibody staining for NFIA combined with fluorescent in situ hybridization for Reelin (E–P) or Slit1 (Q-BB) mRNAs. (I–L) and (U–X) are the same panels as in (E–H) and (Q–T), respectively, but include NF1A expression. (M–P) and (Y-BB) are higher power images of the boxed areas indicated in (I–L) and (U–X), respectively. Arrows indicate double-positive cells, arrowheads NFIA single-positive cells. (CC) The percentage of Reelin⁺ NFIA⁺ cells is significantly different between the electroporated (E) sides of Pax6 and GFP embryos (***, p <.001), as well as between the E and contralateral (C) sides of Pax6 embryos (*, p=.02). The reduction in Slit1 expression on the E side of Pax6 embryos is significant with respect to both the C side (**, p = .002), and the E side of GFP controls (*, p = .009). mean±S.E.M., n= 5 embryos per condition, two independent experiments.

Because we observed Pax6⁺, Reelin⁺ cells in the VZ as well as in the white matter (Fig. 2I), we also asked whether mis-expression of Pax6 could promote a ventral expansion of the domain of Reelin expression, in this germinal layer. We therefore examined embryos at E5, a stage at which ventral VZ progenitors are specified for a glial fate (Shibata et al., 1997; Deneen et al., 2006). Pax6 mis-expression indeed promoted a ventral expansion of Reelin expression within the VZ (Supplemental Figure S4C, arrow; M). Importantly, a similar result was obtained when electroporation was performed at E4.5, by which time neurogenesis has largely ceased in the ventral region of the VZ (Deneen et al., 2006) (Supplemental Figure S4I, arrow; M). This ventral expansion of Reelin expression occurred in NFIA⁺ cells (Fig. S4B, H, arrows), which are likely glial precursors (Deneen et al., 2006). The fact that this phenotype can be observed in embryos electroporated after ventral neurogenesis has terminated suggests a direct role for Pax6 to control Reelin expression in glial precursors, rather than an indirect role mediated by earlier effects on neuronal precursors. In support of this conclusion, we were able to observe changes in Reelin expression, caused mis-expression of Pax6, at axial levels where there was no change in the number or distribution of ventral interneuron subtypes whose development is Pax6-dependent (Burrill et al., 1997; Ericson et al., 1997; Osumi et al., 1997) (data not shown).

The de-repression of Slit1 observed in Pax6^−/− embryos predicted that mis-expression of Pax6 might down-regulate Slit1 expression in WMAs. Indeed, we observed a significant reduction in the percentage of Slit1⁺; NFIA⁺ astrocytes in the ventral region of Pax6- electroporated embryos, compared to GFP-electroporated controls (Fig. 4R vs. T, “WM;” CC, “Slit1”). Pax6 mis-expression also caused a mild, but detectable, down-regulation of Slit1 mRNA in the VZ at E5 (data not shown). Thus, Pax6 is sufficient, as well as necessary, for Slit1 repression in a subset of astrocytes and their presumptive precursors. Taken together, these data suggest that Pax6 plays an instructive role in regulating astrocyte positional identity, through positive and negative regulation of Reelin and Slit1, respectively.

Expression of Nkx6.1 by Slit1⁺ astrocytes and their precursors

The observation that Pax6 promotes expression of Reelin, and represses that of Slit1, explained the VA1 phenotype, but raised the question of how VA2 astrocytes can express Slit1, given that they are also Pax6⁺. One possibility is that repression of Slit1 by Pax6 requires a cofactor, which is present in VA1 astrocytes but lacking in VA2 astrocytes. This seemed unlikely, since Pax6 was able to repress Slit1 in ventral (VA3) WMAs, in gain-of-function experiments. Alternatively, VA2 astrocytes may contain a factor absent in VA1 astrocytes, which promotes Slit1 expression and/or neutralizes the repressive activity of Pax6 towards Slit1. One candidate for the latter activity is Nkx6.1, a HD transcription factor expressed in the ventral VZ at both neurogenic and gliogenic (Fu et al., 2003) stages. Expression of Nkx6.1 partially overlaps that of Pax6, in the p2 and pMN domains, while the more dorsal p0 and p1 domains express Pax6 but not Nkx6.1 (Briscoe et al., 2000; Jessell, 2000; Briscoe and Ericson, 2001). We therefore investigated the possibility that expression of Slit1 in WMAs is under the control of Nkx6.1.

We first compared the expression of Nkx6.1 and Slit1 in E18.5 spinal cord WMAs. Triple-labeling for GFAP, Nkx6.1 and Slit1-GFP revealed that Nkx6.1 is expressed in WMAs (Fig. 5A–C, arrows), and that the dorso-ventral domain of its expression is approximately co-extensive with that of Slit1 (Fig. 5A–B, D). Quantification indicated that >90% of Slit1⁺ WMAs are Nkx6.1⁺ (Fig. 5J, green bar). We next asked whether Pax6 and Nkx6.1 are co-expressed in the region of the white matter where VA2 astrocytes (Pax6⁺, Slit11⁺) are located. Triple labeling revealed an overlap of Pax6 and Nkx6.1 expression in the ventro-lateral white matter, within Slit1⁺ cells (Fig. 5E–I, arrows). Quantification indicated that virtually 100% of Pax6⁺, Slit1⁺ cells were Nkx6.1⁺ (Fig. 5I, arrows; J, turquoise bar). These data support the idea that VA2 astrocytes are Nkx6.1⁺, Pax6⁺ and Slit1⁺. (It was not possible to directly compare Nkx6.1 and Reelin expression in these experiments, due to antibody incompatibility).

Open in a separate window

Figure 5

Nkx6.1 is co-expressed by Slit1⁺ astrocytes and their precursors

(A–I) Triple-labeling for Nkx6.1, GFAP and Slit1-GFP (A–D) or Nkx6.1, Pax6 and Slit1-GFP (E–I). (I) is a higher power image of the boxed area in (E). Note that all Slit1⁺ cells in this domain (boxed area, E) are Nkx6.1⁺ and Pax6⁺ (E–G, I, arrows, white nuclei), while all Pax6⁺ cells dorsal to the boundary of Slit1 expression are Nkx6.1⁻ (E–G, I, arrowheads). (J) Quantification of Nkx6.1 expression in astrocyte sub-populations. Nkx6.1 is expressed in > 90% of Slit1⁺ astrocytes (green bar). (K–M) Triple antibody labeling for Nkx6.1, Slit1-GFP and Pax6 in the E13.5 VZ, displayed as pairwise comparisons from the same section. The three progenitor domains are indicated (L, M, arrows). “GM,” gray matter. (N) Composite schematic illustrating relationship between the domains of Reelin and Slit1 expression, and those of Pax6 and Nkx6.1 expression, in the white matter at E18.5 and in the VZ at E13.5.

In addition to their co-expression in the white matter, Nkx6.1 and Slit1 are co-expressed in the VZ, where the dorsal boundaries of their expression domains are co-extensive (Fig. 5K). Triple-labeling for Pax6, Nkx6.1 and Slit1-GFP indicated that the domain of Pax6 expression that overlaps that of Slit1 (Fig. 5M, pA2) also overlaps that of Nkx6.1 (Fig. 5L, pA2). Taken together, these data suggest that Nkx6.1 is expressed by the VA3 (Pax6⁻) and VA2 (Pax6⁺) astrocyte subpopulations, and by their presumptive precursors (pA3 and pA2, respectively), but not by the VA1 astrocytes (Fig. 5N).

Astrocytes in the spinal cord white matter exhibit distinct positional identities

The recognition that positional differences underlie neuronal subtype diversity has aided both the identification and functional characterization of different neuronal subpopulations, as well as analysis of the developmental mechanisms that control their generation (Jessell, 2000; Briscoe and Ericson, 2001; Shirasaki and Pfaff, 2002). Here we provide evidence for molecular differences in the positional identities of WMA subpopulations within a specific region of the CNS. Previous studies have reported differences in the expression of markers such as ephrin-B1 and BLBP, by radial glial precursor cells at different locations along the dorso-ventral axis of the VZ (Ogawa et al., 2005). However these differences are not maintained by differentiated WMAs. The transcription factor SCL was shown to control generic aspects of astrocyte differentiation, in a position-specific manner (Muroyama et al., 2005), but no evidence was provided for any role in controlling position-specific phenotypic properties of WMAs.

The functional significance of the VA1–3 astrocyte subpopulations revealed by our studies is not yet clear. Reelin and Slit1 are secreted molecules involved in cell migration (Rice and Curran, 2001) and axon guidance (Brose et al., 1999; Plump et al., 2002), respectively. Slits in particular are expressed by astrocytic glia in the floorplate, a mid-line structure ((Holmes et al., 1998; Brose et al., 1999); reviewed in (Lemke, 2001)). The combinatorial expression of these molecules by subsets of WMAs therefore suggests a potential role for this heterogeneity in guiding axonal trajectory and/or cell migration (Powell et al., 1997). Interestingly, Reelin is expressed not only by VA1 and VA2 astrocytes, but also by V1 and V2 interneurons (Yip et al., 2004b). The dorsal restriction of Reelin expression is thought to be important in guiding the migration of sympathetic pre-ganglionic motoneurons, within the spinal cord (Yip et al., 2000; Yip et al., 2004b; Yip et al., 2007). Perhaps the expression of Reelin by VA1 and VA2 astrocytes contributes to this guidance process. In addition to their role in axon guidance, astrocytes are important for enhancing synapse formation and synaptic efficacy (Pfrieger and Barres, 1997; Ullian et al., 2001; Ullian et al., 2004). Perhaps astrocytes derived from a particular progenitor domain preferentially enhance synapse formation by neuronal subtypes derived from the same progenitor domain.

A transcriptional code for astrocyte positional identity

Our data support a combinatorial model for the specification of astrocyte positional identity, in which expression of Pax6 in the absence of Nkx6.1 specifies a VA1 phenotype (Reelin⁺, Slit1⁻), co-expression of Pax6 and Nkx6.1 specifies a VA2 phenotype (Reelin⁺, Slit1⁺), and expression of Nkx6.1 in the absence of Pax6 specifies a VA3 phenotype (Slit1⁺, Reelin⁻) (Fig. 7A). Nkx2.2 may also participate in this code, by repressing Reelin in pA3 VZ precursors (Supplemental Figure S5) via repression of Pax6 (Briscoe et al., 1999), an activity that Nkx6.1 lacks (Briscoe et al., 2000). Consistent with this, mis-expression of Nkx2.2 strongly repressed Reelin, and up-regulated Slit1, in the VZ at E5 (data not shown). However Nkx2.2, unlike Nkx6.1, is not expressed in differentiated Slit1⁺ WMAs (Supplemental Figure S6), but rather in oligodendrocytes (Qi et al., 2001; Zhou et al., 2001). Therefore, the maintenance of Slit1 expression in VA3 WMAs is more likely to involve Nkx6.1.

Our data suggest that the co-expression of Slit1 and Reelin in VA2 astrocytes reflects a role for Nkx6.1 to override repression of Slit1 by Pax6, without interfering with its activation of Reelin. Consistent with this view, Nkx6.1 is able to prevent repression of Slit1 by Pax6 in co-electroporation experiments, while permitting Pax6 enhancement of Reelin expression (Fig. 6RR). Since Nkx6.1 can act either as an activator or as a repressor, depending on context (Taylor et al., 2005), it could either directly activate Slit1 transcription in VA2 precursors, or do so indirectly, by repressing an as-yet unidentified repressor (Fig. 7B, “R”). The latter model seems more likely, because in Pax6 mutants, Slit1 is de-repressed in Pax6-lacZ⁺ VA1 astrocytes without de-repression of Nkx6.1 (Supplemental Fig. S9). Thus, Nkx6.1 is not essential for Slit1 expression in ventral WMAs under some conditions. This in turn suggests that, in normal embryos, the absence of Slit1 in VA1 astrocytes reflects either direct repression of Slit1 by Pax6, or else a Pax6-dependent Slit1 repressor. We favor the latter explanation, because in the former case, one might expect that Slit1 expression would only occur in ventral astrocytes lacking Pax6, and this is not observed (Fig. 5I, J). We suggest that in VA2 astrocytes, Nkx6.1 may repress a Pax6-dependent repressor of Slit1 (Fig. 7B, VA2). A similar layered-repression model has been proposed to explain the role of Nkx6.1 in controlling IN identity (Vallstedt et al., 2001). Interestingly, it has recently been shown that Nkx6.1 dominantly antagonizes the activation of glucagon transcription by Pax6 in pancreatic α cells, via competition for binding to common promoter elements (Gauthier et al., 2007). A similar gene-specific antagonism may allow co-expression of Pax6 and Nkx6.1 in VA2 astrocytes to permit expression of both Reelin and Slit1.

FACS and Microarray experiments

Olig2-GFP-expressing cells were FACS isolated from spinal cords of Olig1,2 +/− and −/− embryos as previously described (Mukouyama et al., 2006). RNA was extracted from the isolated cells, amplified, biotinylated and hybridized to Affymetrix mU74v2 A, B and C chips (Deneen et al., 2006). Analysis of microarray data was performed using Affymetrix Microarray Suite and Rosetta Resolver software.

In situ hybridization

Non-radioactive in situ hybridization using DIG-labelled (Zhou et al., 2000) or fluorescent probes (Choi et al., 2005) was performed on frozen sections of mouse or chick spinal cord as described. The following probes were used: mouse Reelin (gift of Gabriella D’Arcangelo), mouse Slit1 (gift of Marc Tessier-Lavigne), chick Reelin, chick Slit1 (gift of Ed Laufer). Antibodies used are listed in Supplementary Information.

We thank A. Stoykova, P. Gruss and G. Lanuza for sharing Pax6-LacZ mice and embryos, M. Tessier-Lavigne for the Slit1-GFP mice, and K. McCarthy for pilot experiments with GFAP-CreER mice. We also thank T. Jessell for helpful discussions and antibodies, and E. Laufer and G. D’Arcangelo for in situ probes and antibody reagents. We thank S. Pease, J. Alex and the staff of the Caltech animal facility for assistance with mouse breeding and timed matings; R. Diamond for assistance with FACS; E. Zuo for Affymetrix microarray hybridization; P. Lwigale for advice and assistance with electroporation and egg husbandry techniques; M. Martinez for genotyping mouse lines; J. Chow, M. Lee, R. Ho and J. S. Chang for technical assistance; G. Mosconi for laboratory management and G. Mancuso for administrative assistance. This work was supported in part by NIH grant 1RO1-NS23476. D.J.A. is an Investigator of the Howard Hughes Medical Institute.