Retinoic Acid Functions as a Key GABAergic Differentiation Signal in the Basal Ganglia

Retinoic acid (RA) is essential for the generation of GABAergic inhibitory neurons in the mouse forebrain, and RA treatment of embryonic stem cells induces the production of GABAergic neurons.


Introduction
The embryonic forebrain, deriving from the most anterior part of the neural tube, comprises a complex set of structures in the developing brain. This complexity arises mainly due to the heterogeneity of the neurons comprising it in terms of morphology, structure, function, and genetic specification. During forebrain development, the dorsal domain (pallium) gives rise to the cortex while the ventral region (subpallium) generates the basal ganglia, i.e. the pallidum and the striatum, which, respectively, originate from the medial and lateral ganglionic eminences (MGE, LGE) [1]. The progenitor zones of the subpallial ganglionic eminences are the origin of chemically diverse populations of gamma-aminobutyric acid (GABA)ergic interneurons and projection neurons. GABAergic interneurons are inhibitory local circuit neurons modulating neuronal activity and synaptic plasticity. GABAergic neurons comprise ,20% of all neurons within the cortex and hippocampus and ,95% of the neurons within the striatum [2][3][4]. Whereas GABAergic projection neurons generated in the germinal zones of the LGE migrate radially to the adjacent striatum, GABAergic interneurons arise from both the MGE and LGE and migrate using multiple tangential routes to the olfactory bulb, cortex, and hippocampus [5][6][7][8]. Disturbed GABAergic neuron function has been associated with several neurological disorders including Huntington's disease, autism, schizophrenia, bipolar depression, and epilepsy [9][10][11][12]. Thus, a source of GABAergic neurons for cell replacement therapy may be useful for treatment of these neurological diseases.
GABAergic neuronal diversity emerges during embryogenesis and depends on both the timing and the creation of specific anteroposterior and dorsoventral progenitor domains by the coordinated action of several transcription factors expressed by distinct progenitor populations [13,14]. In contrast, little is known about the extrinsic signaling pathways coordinating GABAergic specification in the basal ganglia. Retinoic acid (RA) functions as an extrinsic signal that regulates patterning of rhombomeres in the hindbrain and neuronal differentiation in the spinal cord [15][16][17], but the role of RA in forebrain development remains unclear. RA is derived from vitamin A through a two-step enzymatic process, employing retinol dehydrogenase (Rdh10) for oxidation of retinol to retinaldehyde, and retinaldehyde dehydrogenases Raldh1 (Aldh1a1), Raldh2 (Aldh1a2), and Raldh3 (Aldh1a3) for oxidation of retinaldehyde to RA, which then functions as a ligand for nuclear RA receptors [18]. A role for RA signaling during mouse striatal development is evident after E12.5 when Raldh3, expressed in the subventricular zone of the LGE [19], plays a required role in the up-regulation of dopamine receptor D2 expression [20]. Consistent with this finding, loss of RA receptor-beta (RARb) in null mutant mice is associated with defects in striatal dopaminergic neurogenesis after E13.5 resulting in motor behavioral defects [21]. A recent study using Rdh10 mutant embryos with reduced RA synthesis in the meninges suggested that RA is required for normal radial expansion of the dorsal cortex [22]. However, other studies have suggested that RA may not act in the embryonic cortex, as RA activity was detected in the LGE but not the cortex [23].
Here, we employ null mutants for Raldh3 (Aldh1a3) and Raldh2 (Aldh1a2) to ascertain the anatomical sites, cellular targets, and consequences of RA signaling in the embryonic forebrain. Our results provide evidence that Raldh3 expression in the LGE is a major source of RA production in the embryonic forebrain and is required for GABAergic differentiation from LGE-derived progenitors in the basal ganglia. Furthermore, our findings suggest that RA generated in the meninges by Raldh2 is not required to stimulate radial expansion of the cortex as previously suggested. We also report that RA induces GABAergic differentiation in neurons generated from LGE-derived neurospheres and human embryonic stem cells, thus implicating a role for RA as a GABAergic differentiation factor both in vivo and in vitro.

RA Signaling in the Embryonic Forebrain
Although Raldh3 expression in the LGE from E12.5 to early postnatal stages suggests the LGE is a major site of RA action in the embryonic forebrain [19,24], Raldh2 and Rdh10 are expressed in the meninges beginning at E12.5-E13.5, suggesting that RA synthesized there may regulate corticogenesis [22,25]. In order to better define the timing and location of RA signaling in the developing forebrain from E12.5-E14.5, we employed a tissue explant RA reporter cell assay [26]. Cortex and LGE tissues were dissected from E12.5 to E14.5 embryos and grown as explants in co-culture with the RA-reporter cells. As positive controls, eye (E12.5 to E13.5), which expresses Raldh1 and Raldh3, and meninges (E14.5), which expresses Raldh2, were dissected from the same embryos. Reporter cells co-cultured with cortex or LGE from E12.5 wild-type embryos displayed no RA activity, whereas eye explants did ( Figure 1A-C); lack of RA activity in E12. 5 LGE explants may be due to low initial Raldh3 expression. In accordance with the increase in Raldh3 expression in the LGE after E12.5 [19], LGE explants from both E13.5 and E14.5 induced strong RA activity in the surrounding reporter cells ( Figure 1F,I). In contrast, E13.5 and E14.5 cortical explants remained unable to induce RA activity, whereas meninges and eye explants at these stages exhibited RA activity (Figure 1D,E,G,H). To verify that RA activity detected in the LGE is due to Raldh3 expression, we found that loss of RA synthesis by Raldh3 resulted in lack of RA activity in Raldh32/2 LGE explants but had no effect on RA activity in meninges ( Figure 1M-O). Using an Raldh22/2 mutant model we found that all RA activity detectable in wild-type meninges at E14.5 was eliminated in Raldh22/2 meninges ( Figure 1G,J). Raldh22/2 cortical explants contained no RA activity as observed in wild-type ( Figure 1H,K), while RA activity was still observed in Raldh22/2 LGE ( Figure 1I,L). Together, the above findings demonstrate that RA produced by Raldh3 in the LGE can activate transcription in the basal ganglia, whereas RA produced by Raldh2 in the meninges does not activate transcription in the adjacent cortex.
Exposure of the reporter cells to a range of RA concentrations between 1 nM and 1 mM provided a dose-response for RA activity ( Figure S1); a concentration of 1 nM was sufficient to activate the reporter line as previously reported [26]. Thus, the fact that RA activity was not detected in cortical explants from E12.5 to E14.5 indicates that RA is present at very low levels in the cortex. However, recent studies proposed a role for RA in corticogenesis and additionally reported a value for the concentration of RA in the mouse E14.5 cortex (0.28 mmole/ mg) [22], which is seven orders of magnitude higher than that previously reported for mouse E13.5 forebrain (12 pmol/g) [27]. The former value (presented as mmole/mg rather than pmol/g) is most likely in error as other studies reported RA concentrations in adult mouse cortex as 16 pmol/g and adult striatum as 78 pmol/g [28], but this leaves in doubt how much RA was actually detected in E14.5 cortex.

LGE-Derived Progenitors Maintain Their RA Activity In Vitro
Our observation that RA activity during forebrain development is due primarily to Raldh3 expression in the LGE prompted us to investigate if neural precursors isolated from E14. 5 LGE maintain their Raldh3 expression and RA activity when expanded in vitro under mitogen stimulation. Hence, we employed immunocytochemistry with a Raldh3 antibody together with the tissue explant RA bioassay using neurospheres generated from E14. 5 LGE and cortex of wild-type and Raldh32/2 embryos. RA activity and Raldh3 immunostaining were detected in neurospheres derived from wild-type E14.5 LGE ( Figure 2C-D). In contrast, both Raldh3 immunostaining and RA activity were eliminated in LGEderived neurospheres from Raldh32/2 embryos (Figure 2A-B). Neither Raldh3 immunostaining nor RA activity were found in neurospheres derived from the cortex of either wild-type or Raldh32/2 embryos ( Figure 2E-H). These results further confirm that Raldh3 expression in the LGE is responsible for RA synthesis and additionally showed that neurospheres expanded from LGE cells maintain their RA activity.

GABAergic Differentiation of LGE-Derived Neurospheres Depends on RA
We investigated whether RA has an effect on the differentiation potential of regionally derived neurosphere cultures. Neurospheres from LGE and cortex of E14.5 wild-type and

Author Summary
The vitamin A metabolite retinoic acid is an important signaling molecule needed for development of the central nervous system. Previous studies have shown a role for retinoic acid in regulating genes involved in the generation of motor neurons both in the hindbrain and spinal cord, but the role of retinoic acid in the forebrain has remained elusive. Here, we investigated mice that lack the ability to metabolize vitamin A into retinoic acid in the forebrain. Although no defects were observed in the generation of forebrain cortical neurons, we did observe a serious deficiency in GABAergic neurons, which provide inhibitory input to cortical neurons. Specifically, our results reveal that retinoic acid is required for forebrain neurons to activate an enzyme that converts glutamate to the inhibitory neurotransmitter GABA. We also find that retinoic acid treatment of human embryonic stem cells could stimulate production of GABAergic neurons. Deficiencies in GABAergic neurons have been associated with several neurological disorders, including Huntington's disease, autism, schizophrenia, and epilepsy. Knowledge of how GABAergic neurons are generated may aid efforts to treat these diseases.
Raldh32/2 embryos were differentiated for 7 d and subsequently analyzed immunocytochemically with antibodies against the pan-neuronal marker b-tubulin-III (Tuj1), the GABAsynthesizing enzyme glutamic acid decarboxylase (Gad67), the neural progenitor marker nestin, and the astrocytic marker glial fibrillary acidic protein (GFAP). Many wild-type LGE neurospheres untreated with RA were found to co-express Tuj1 and Gad67, indicating they have differentiated and matured into a GABAergic phenotype (44.3%613.0%), however very few Tuj1/ Gad67-positive cells were detected in differentiated cultures of Raldh32/2 LGE neurospheres (13.2%64.1%) ( Figure 3A-B,K). Tuj1-expressing neurons that differentiated from cortical Raldh32/2 neurospheres appeared to have a similar morphology to those generated from cortical wild-type neurospheres, and Gad67 was never detected ( Figure 3C-D). Nestin-positive progenitors and cells with an astrocytic morphology expressing GFAP appeared similar in LGE and cortical differentiated cultures derived from wild-type and Raldh32/2 neurospheres ( Figure 3E-H).
LGE neurospheres from both wild-type and Raldh32/2 embryos were differentiated in the presence of RA in order to further test the effect of RA on GABAergic neuronal differentiation. After 1 wk of differentiation in the presence of 100 nM RA, the majority of the generated neurons in Raldh32/2 and wild-type cultures were GABAergic, as observed by double staining for Tuj1 and Gad67 ( Figure 3I-J). However, when cortical neurospheres were differentiated in the presence of RA, Gad67 was never detected in either wild-type or Raldh32/2 differentiating cultures ( Figure 3K-L). Quantification of GABAergic neuron differentiation from wild-type and Raldh32/2 LGE neurospheres with or without added RA showed a significant increase of Gad67-positive neurons in the presence of RA. The proportion of GABAergic cells derived from Raldh32/2 LGE neurospheres increased significantly from 13.2%64.1% under control conditions to 83.37%611.75% when the neurospheres had been differentiated in 100 nM RA ( Figure 3M). The proportion of GABAergic neurons in cultures of wild-type LGE neurospheres increased from 44.3%613.0% under control conditions to 76.45%611.75% after treatment with 100 nM RA ( Figure 3M). Thus, E14. 5 LGE derived cells from Raldh32/2 embryos can be expanded as neurospheres and are able to differentiate into neurons and glia, but they are unable to differentiate into GABAergic neurons unless RA is added. Raldh3 is responsible for RA activity in neurospheres derived from the LGE but cortex-derived neurospheres lack RA activity. (A-D) Neurospheres generated from the LGE of E14.5 wild-type (WT) embryos exhibit Raldh3 immunoreactivity and they induce RA activity when co-cultured with F9-RARE-lacZ RA-reporter cells (n = 4); neurospheres derived from Raldh32/2 LGE always lacked both Raldh3 detection and RA activity (n = 4). (E-H) Neither Raldh3 nor RA activity were detected in cortex-derived neurospheres from either wild-type or Raldh32/2 (KO) embryos (n = 4). doi:10.1371/journal.pbio.1000609.g002 Quantification of the percentage of LGE neurosphere-derived Tuj1-positive neurons also positive for Gad67 showed that in differentiating medium containing only fetal calf serum (FCS) with no added RA, the percentage of Gad67+/Tuj1+ cells derived from Raldh32/2 LGE neurospheres was significantly lower than that of wild-type LGE neurospheres. Following addition of 100 nM RA to the differentiation medium for 1 wk, the majority of Tuj1-positive neurons derived from both wild-type and Raldh32/2 LGE neurospheres were also Gad67-positive. The percentage was calculated by dividing the immunopositive cell number with the total number of DAPI-stained nuclei. Values are listed as mean 6 SEM; * p,0.05; ** p,0.01; *** p,0.001. doi:10.1371/journal.pbio.1000609.g003

Raldh3 Expression in Basal Ganglia Is Required for GABAergic Differentiation in the LGE
The observation that RA generated by Raldh3 induces GABAergic differentiation of neural precursors in vitro prompted us to investigate if RA signaling is required for GABAergic differentiation in the developing forebrain. We examined a panel of markers for both neural progenitors and differentiated neurons in forebrains from E14.5 wild-type and Raldh32/2 embryos. Raldh3 protein was observed at high levels in the SVZ of the LGE; Raldh3 was not detected in the Raldh32/2 forebrain ( Figure 4A-B). Nestin and RC2 immunoreactivity was not significantly changed in Raldh32/2 versus wild-type basal ganglia at E14.5, suggesting that generation of neural progenitors is not affected when RA signaling is disrupted ( Figure 4C-F). In order to determine the proliferative capacity of these neural progenitors, double immunohistochemistry was performed with the proliferation marker Ki67 and the radial glia/progenitor marker nestin ( Figure S2A-B). We observed no reduction in the number of LGE proliferating progenitors in Raldh32/2 embryos compared to control embryos, showing that both generation and proliferation of neural progenitors is not affected when RA signaling is disrupted ( Figure S2C).
MAP2 immunostaining marking postmitotic neurons [29] was unaffected in Raldh32/2 basal ganglia ( Figure 4G-H). The numbers of MAP2-expressing neurons were quantified in the striatum, cortex, and septum from wild-type and Raldh32/2 embryos. MAP2-expressing cells did not change in number in these regions of the forebrain in Raldh32/2 embryos, confirming that neurogenesis was not affected ( Figure S2D).
A defect in GABAergic differentiation was observed when RA signaling is lost in the basal ganglia. Gad67-positive cells are normally present along the SVZ from the LGE to the septum ( Figure 4K), and the pattern of GABA immunoreactivity is normally similar to Gad67 although GABA-positive cells extend into the ventricular zone ( Figure 4I); this is probably due to the fact that while Gad67 immunoreactivity marks only cells with GABA production (i.e. GABAergic cells), GABA immunoreactivity could mark cells synthesizing GABA plus cells that uptake GABA released by Gad67-positive cells. In Raldh32/2 embryos, detection of both Gad67 and GABA was nearly eliminated in the LGE and septum ( Figure 4J,L). At E12.5, when Raldh3 expression in the LGE has just initiated, we observed that GABA was detected in the MGE and LGE in a pattern that was not significantly different between wild-type and Raldh32/2 forebrain; GABA detection in the LGE at E12.5 was at a lower level than that seen at E14.5 ( Figure S3A-B). As RA activity is not yet detected in the LGE at E12.5 but is seen by E13.5 ( Figure 1C,F), these findings demonstrate that RA signaling initiating after E12.5 is required to stimulate the high level of GABAergic differentiation normally observed in the LGE by E14.5.
The cellular source of RA in the LGE has previously been associated with newly born neurons in the subventricular zone expressing Raldh3 [24]. We analyzed Raldh3 immunoreactivity in two distinct types of cells, radial glia and postmitotic neurons of the LGE at E14.5. Double-labeling studies demonstrated that none of the RC2-positive radial glia exhibit Raldh3 detection, although the radial processes of these cells were observed next to Raldh3positive cells localized in the SVZ that did not possess radial processes ( Figure S4A-C). In contrast, most Raldh3-expressing cells were also labeled with neuronal marker MAP2 ( Figure S4D-F). The above results provide further evidence that Raldh3expressing cells in the LGE are newly born neurons defining a discrete region of the SVZ.

RA Is Required to Stimulate GABAergic Differentiation in the Striatum
In E18.5 wild-type embryos, Raldh3 detection remains strongest along the SVZ of the LGE (particularly high in the dorsal LGE) with weaker detection further ventrally along the septum; Raldh3 immunoreactivity was eliminated in Raldh32/2 forebrain ( Figure 5A-B). As observed at E14.5, MAP2 immunostaining was unaffected in Raldh32/2 versus wild-type forebrain at 18.5 ( Figure 5G-H). At E18.5, GABAergic differentiation in the striatum was nearly eliminated in Raldh32/2 forebrain as monitored by Gad67 immunoreactivity (Figure 5C-D); GABA detection was reduced in striatum but less so than Gad67 possibly due to diffusion of GABA still generated ventral of the striatum ( Figure 5E-F). Thus, not all regions of the basal ganglia were affected by the disruption of RA signaling, as both Gad67 and GABA immunoreactivity appear relatively normal in the pallidum and septum at E18.5 ( Figure 5C-F).
The subcortical telencephalon is known to be the source of GABAergic projection neurons that migrate radially from the ventricular progenitor zone to reach their final destination. The LGE gives rise to GABAergic striatal projection neurons [1,[30][31][32][33][34][35], while the MGE gives rise to GABAergic projection neurons of the pallidum, septum, and nucleus basalis [1,30]. In order to determine if differentiation of striatal projection neurons is affected in E18.5 Raldh32/2 embryos, we examined Foxp1, a marker for these neurons [36], and found that Raldh32/2 embryos display normal Foxp1 immunoreactivity in the striatum ( Figure 5I-J). This observation supports our previous findings demonstrating that Raldh3 is not required to generate DARPP32-positive neurons, another marker of striatal medium-sized spiny projection neurons [20]. Instead, our findings demonstrate that RA is required for striatal projection neurons to acquire a GABAergic fate.
Our results indicate that RA is required to stimulate GABA synthesis in LGE-derived progenitors. To characterize other aspects of the GABAergic phenotype in Raldh32/2 mutants, we analyzed expression of the vesicular GABA transporter (VGAT, Viaat), a transporter that mediates accumulation of GABA into the synaptic vesicles before exocytotic release to the synaptic cleft [37]. Expression of VGAT appeared normal in Raldh32/2 forebrain ( Figure S5A-B), indicating that RA is not required for this aspect of GABAergic differentiation.
Next, we wanted to investigate whether disruption of RA signaling could induce defects in the specification of other neuronal populations. Previous studies have shown that loss of Raldh3 or RARb in the striatum results in down-regulation of dopamine receptor D2 in the nucleus accumbens [20,21]. However, in Raldh32/2 forebrain no difference was found in the expression of tyrosine hydroxylase (TH), a marker of dopaminergic neurons ( Figure S5C-D). Also, we observed no difference in vesicular glutamate transporter (VGLUT), a marker of glutamatergic neurons ( Figure S5E-F).
In addition to neuronal markers, we examined whether loss of RA affects glia. We showed above that radial glia differentiation is not affected by loss of Raldh3 ( Figure 4E-F). We also analyzed astrocyte differentiation by analyzing expression of the astrocytic marker GFAP. We did not detect any difference in GFAP immunoreactivity in E18.5 Raldh32/2 forebrain compared to wild-type controls ( Figure 5K-L). Our findings thus suggest that RA is not required for gliogenesis or generation of radial projection neurons.
Taken together, our observations at E12.5-E18.5 demonstrate that RA is required to stimulate a high level of GABAergic differentiation first in the LGE and then later in the striatum but that a Raldh3-independent mechanism for GABAergic differentiation occurs in the MGE/pallidum and septum.

RA Is Required for Differentiation of LGE-Derived GABAergic Interneurons Migrating to the Cortex and Olfactory Bulb
In addition to GABAergic projection neurons, progenitor cells in the LGE produce GABAergic interneurons that migrate tangentially mostly within the cortical intermediate zone, whereas GABAergic interneurons migrating from the MGE disperse into the cortical plate [5,[38][39][40]. Also, cells derived from the dorsal SVZ of the LGE generate many olfactory bulb interneurons via a rostral migratory pathway [40][41][42]. At E18.5, Gad67 immunoreactivity normally extends from the striatum into the intermediate zone of the cortex marking a population of LGE-derived interneurons, but this zone of Gad67 detection was markedly reduced in Raldh32/2 cortex ( Figure 5C-D). Dlx2 is an early marker of GABAergic progenitors in the basal ganglia that is required for GABAergic interneuron migration to the cortex [14]. Dlx2 immunoreactivity was not changed in Raldh32/2 forebrain from E12.5-E18.5, demonstrating that interneurons are generated in the basal ganglia and migrate to the cortex in the absence of RA signaling ( Figure S3C-H). Additionally, detection of Gad67 in the Raldh32/2 olfactory bulb was also clearly reduced compared to wild-type ( Figure 5M-N). Apart from the dorsal LGE, recent studies have shown that additional telencephalic areas may also contribute to olfactory bulb interneurons [43][44][45], which may explain our observed partial elimination of Gad67 immunoreactivity in the Raldh32/2 olfactory bulb. The above findings provide evidence that RA synthesis controlled by Raldh3 is required for GABAergic differentiation of interneurons that originate from progenitors in the LGE then migrate tangentially to the cortex and olfactory bulb.

RA Stimulates GABAergic Differentiation of Human Embryonic Stem Cells
We investigated whether our findings may be useful to generate GABAergic neurons from human embryonic stem (ES) cells for potential cell replacement therapies. Following RA treatment of embryoid bodies and propagation of neural rosettes, cultures were processed immunocytochemically with antibodies against Pax6 (a marker for neural progenitors), Doublecortin (DCX; a marker for immature migrating neurons), the pan-neuronal marker Tuj1, plus GABA and Gad67. With no RA added in the differentiation medium a large proportion of the cells were Pax6-positive (79.9%65.01%), indicating they were neural progenitors ( Figure 6J,M), and many cells exhibiting neuronal processes were Tuj1-positive (32.8%610.3%) and colocalized with DCX (31.3%68.6%), suggesting they were immature neurons ( Figure 6G,M). However, very few Gad67 (2.7%60.6%) positive cells were detected in cultures with no RA added, suggesting that GABAergic differentiation is not favored under these conditions ( Figure 6A,D). Treatment of embryoid bodies with 1 mM RA resulted in a significant increase in both the number of Tuji1/DCX+ neurons that were forming extensive neuronal networks and GABAergic neurons detected with Gad67 (16.2%62.1%) ( Figure 6B,E,M). Moreover, the proportion of Pax6-positive cells was reduced to almost half with 1 mM RA (49.9%611.6%), suggesting that more progenitor cells had differentiated to immature migrating neurons co-expressing Tuj1 and DCX ( Figure 6H To gain further insight on the subtype identity of the RAinduced GABAergic neurons generated in our cultures, we examined expression of region-specific transcription factors previously associated with the specification of both GABAergic interneurons and projection neurons. Many Tuj1-positive cells were also immunopositive for Dlx2, which is expressed in GABAergic precursors in the basal ganglia of the telencephalon, differentiating into both interneurons and striatal projection neurons ( Figure S6A). However, no Tuj1-positive cells were found to be positive for Foxp1, a striatal projection neuron marker ( Figure S6B). Lim1/2 is a LIM homeodomain protein marking interneurons of the diencephalon and spinal cord [17,[46][47][48]. Many GABA-positive neurons also expressed Lim1/2, suggesting they acquire a GABAergic interneuron phenotype ( Figure S6C). Islet1 (Isl1), another LIM homeodomain protein, is expressed in the ventral forebrain where it marks telencephalic GABAergic projection neurons; Isl1 also marks diencephalic interneurons when co-expressed with Lim1/2 and GABA [46,48]. No cells in our culture co-expressed GABA and Isl1, further suggesting that GABA-positive neurons in our cultures do not acquire striatal projection neuron or diencephalic interneuron identities ( Figure  S6D). We found that 43.7%611.94% of GABA-positive cells expressed Lim1/2 and 37.5%611.5% expressed Dlx2 (Figure

RA Is Not Required for Radial Expansion of the Embryonic Cortex
Previous studies using an Rdh10 ethylnitrosourea (ENU) mutant, which reduces retinaldehyde needed for Raldh2 to catalyze RA synthesis in the meninges, suggested that this source of RA is required for radial expansion of the cortex; a reduction in radial expansion of the cortical postmitotic neuronal layer was proposed to result in a concomitant lateral increase in the proliferative progenitor population in the ventricular zone [22]. Raldh22/2 embryos, which completely lack meninges RA activity ( Figure 1G,J), present an excellent model to examine whether RA is required for corticogenesis since Raldh2 catalyzes the final step of RA synthesis in the meninges. At E14.5, the head region of Raldh22/2 embryos appeared to have developed relatively normally while they invariably displayed stunted forelimbs ( Figure 7A-B), which we have previously shown is due to a lack of RA synthesis by Raldh2 in trunk mesoderm [49,50]. Thus, Raldh22/2 mutants do not exhibit massive head deformities like those reported for Rdh10 mutants [51]. We analyzed expression of Tuj1 and the proliferative marker Ki67 in coronal brain sections of both wild-type and Raldh22/2 embryos at E14.5. The mediallateral length of the ventricular zone in the dorsal forebrain of the Raldh22/2 mutant appeared similar to that of the wild-type embryo ( Figure 7C-D). Moreover, double immunostaining for Tuj1 and Ki67 revealed no changes in radial expansion of the postmitotic Tuj1-expressing cortical layer nor the Ki67 proliferative zone in the Raldh22/2 cortex when compared to wild-type ( Figure 7E-J). Examination of MAP2, another marker for postmitotic neurons, also demonstrated no difference in mediallateral length for the mutant dorsal ventricular zone ( Figure S7A-B) and no difference in radial width for the mutant cortex ( Figure  S7C-D). Finally, examination of RC2, a marker of radial glia whose somata reside in the ventricular zone of the cortex and whose radial processes span the entire distance to the pial surface, showed no difference between the Raldh22/2 and wild-type cortex ( Figure S7E-F). The fact that Raldh22/2 embryos retain a normal ratio of cortical progenitors to postmitotic neurons with no apparent morphological defects, in conjunction with a complete lack of RA activity in mutant meninges and cortical explants, suggests that RA is not required for embryonic corticogenesis. The contradiction between our results with Raldh22/2 embryos and the results of others with Rdh10 mutants [22] may be explained by the observation that Rdh10 mutants, unlike Raldh22/2 embryos, exhibit severe craniofacial defects that distort the cranium and forebrain possibly resulting in a thinner cortex [51].

Expression of Retinoic Acid Receptors in the Developing Forebrain
Previous studies have shown that RARa and RARb are expressed during mouse forebrain development, whereas RARc is undetectable [52,53]; RARa was reported to be widespread in the embryonic forebrain, while RARb was detected primarily in the striatum and is induced by RA. We examined expression of RARa and RARb in E18.5 wild-type forebrains by in situ hybridization. RARa mRNA was widespread in the E18.5 forebrain including both the striatum and cortex, but expression was low or undetectable in the ventricular zone ( Figure S8A). RARb mRNA was detected in the striatum but not in the cortex or ventricular zone ( Figure S8B). Taken together with our observation that the LGE/ striatum is a major localized site of RA synthesis during forebrain development due to Raldh3 expression, overlapping expression of both RARa and RARb in the striatum further suggests that this is a major site of local RA-mediated induction and signaling.

Discussion
In this study we demonstrate a novel requirement of RA generated by Raldh3 for GABAergic differentiation in the basal ganglia. In contrast, RA activity is not detected in the cortex at any stages examined despite detection of RA activity in the adjacent meninges, which requires Raldh2. Even if the cortex does receive a small amount of RA from the meninges that we cannot detect, our findings with Raldh22/2 embryos lacking RA activity in the meninges demonstrate that this source of RA is unnecessary for cortical expansion as suggested by a recent study [22]. Thus, unlike the cortex, the LGE represents an unambiguous site of RA action during forebrain development, and loss of RA in the LGE results in a loss of GABAergic differentiation.
Our studies revealed that at E12.5, when Raldh3 expression is barely detectable in the LGE and RA activity is not yet detected, Raldh32/2 embryos maintain expression of the regulatory gene Dlx2 and early aspects of GABAergic differentiation in the progenitor domains of the basal ganglia. However, by E14.5, when Raldh3 expression has intensified in the LGE and RA activity is easily detectable, RA generated by Raldh3 is required to stimulate GABAergic differentiation in the LGE by inducing Gad67 needed for GABA synthesis. At E18.5, Raldh3 is required to maintain GABAergic differentiation in the LGE, whereas a Raldh3-independent mechanism controls GABA synthesis in the MGE and septum. This observation suggests that the LGE is the main site of RA action along the SVZ. We observed that Raldh3 expression in newly generated neurons at the border of the proliferative and postmitotic zones in the LGE coincides with a region that generates both GABAergic striatal projection neurons and GABAergic interneurons [5,[31][32][33][34]40]. Our studies in Raldh32/2 embryos revealed that RA signaling stimulates a GABAergic phenotype in LGE-derived interneurons migrating to the olfactory bulb and cortex and that RA is required for Foxp1positive striatal projection neurons to further differentiate to a GABAergic fate. As Dlx2 and VGAT were still expressed normally in the absence of RA, the role of RA in GABAergic differentiation may be limited to stimulation of Gad67 activity in the LGE to promote GABA synthesis.
As the appearance of GABAergic interneurons in the olfactory bulb and cortex is reduced rather than eliminated, our findings suggest that interneurons can still migrate to these locations in the absence of RA but that less interneurons have matured to a GABAergic phenotype. Other studies have shown that Gsx2 (Gsh2), a homeobox gene specifying ventral character in the forebrain, is required for Raldh3 expression in the LGE [54] and that Gsx2 is required for specification of GABAergic interneurons that migrate from the LGE to the olfactory bulb [13]. In addition, differentiation of DARPP-32 striatal projection neurons is greatly reduced in Gsx2 null embryos but not in conditional Gsx2 mutants when Gsx2 is progressively inactivated from E10.5-E18.5 [13] and also not in Raldh32/2 mutants [20]. Thus, early expression of Gsx2 is required for correct DARPP-32 striatal projection neuron development, a time when there is no Raldh3 expression in the forebrain. Taking into consideration the above, one can conclude that RA signaling exerts a specific role in specifying the GABAergic phenotype both for production of GABAergic interneurons and for further differentiation of striatal projection neurons to a GABAergic fate. Examination of the Gad67 promoter proximal region revealed no evidence of a canonical RA response element (unpublished data), suggesting that Gad67 may be an indirect target or may be controlled post-transcriptionally by RA signaling in the basal ganglia during GABAergic differentiation. As it is clear that RARa and RARb are both expressed in the basal ganglia, null mutants or antagonists for these RA receptors may be useful to further examine the mechanism through which RA functions during stimulation of GABAergic differentiation. Further, as we show that endogenous RA signaling is preserved in primary LGE neurosphere cultures and is required to generate GABAergic neurons in vitro, such cells may prove useful in studying the mechanism of RA action during GABAergic differentiation.
A previous study suggested that Foxc1 mutants fail to form a complete forebrain meninges and exhibit increased lateral expansion of the cortical ventricular zone and reduced neurogenic radial expansion due to the loss of RA produced by Rdh10 and Raldh2 in the meninges [22]. The major conclusions of that study were drawn by comparison of the cortical phenotype of the Foxc1 mutants with that of an Rdh10 ENU mutant [22,55]. However, our studies on Raldh22/2 embryos lacking RA activity in the meninges demonstrate that RA is not required for radial expansion of the embryonic cortex. Additionally, RA receptors were not detected in the ventricular zone of the developing cortex, where RA was proposed to be required to induce neurogenic division of cortical progenitors. Together, these findings suggest that the dorsal forebrain phenotype in Foxc1 mutants is RA-independent. The Rdh10 ENU mutant employed for those forebrain studies [22] exhibits a very similar phenotype to another published Rdh10 ENU mutant, which has severe neural crest-derived craniofacial defects that are responsible for distortion of the cranium as well as forebrain [51]. Thus, reduced radial expansion of the cortex and increased lateral expansion of the ventricular progenitor zone reported for Rdh10 mutants [22] may not be due to a specific effect of RA on corticogenesis but rather a defect in cranial neural crest migration and differentiation that leads to the altered cortical morphology. Indeed, Rdh10 mutants lack all RA activity in the head during the time when cranial neural crest is migrating due to loss of all retinaldehyde synthesis [51], whereas Raldh22/2 embryos still retain most cranial RA synthesis during this time due to expression of Raldh1 and Raldh3 in ocular and olfactory tissues [56]. Thus, head development in Raldh22/2 embryos is not grossly altered, allowing us to conclude that a lack of cranial RA activity specifically in the meninges does not lead to a defect in radial expansion of the cortex. Furthermore, Raldh2 is not expressed in the dorsal meninges until E12.5 [22] or E13.5 [25], while the lengthening of the dorsal forebrain in Foxc1 mutants is already evident at E12.5 [22]. Based on the above, it seems unlikely that RA produced and secreted in the dorsal meninges could be the neurogenic factor inducing the switch from symmetric to asymmetric division in the ventricular zone to affect embryonic cortical expansion. Alternatively, RA generated in the meninges by Rdh10 and Raldh2 might have another function. RA could diffuse in the opposite direction and control development of the skull, which is populated by cranial neural crest cells. Interestingly, a recent study showed that ablation of all three RA receptors (RAR alpha, beta, and gamma) in cranial neural crest cells results in agenesis or malformations of most of the craniofacial skeletal elements including the frontal and parietal bones, which are adjacent to the dorsal meninges [57]. Additionally, Foxc1 hypomorphic mutants also exhibit malformation of the frontal bone [58], providing further evidence that RA generated in the meninges downstream of Foxc1 may function in cranial neural crest differentiation.
RA treatment is known to facilitate terminal differentiation of neural progenitors derived from ES cells [48,[59][60][61][62][63]. Here we demonstrated that exposure of human ES-derived embryoid bodies to high concentrations of RA promotes differentiation of neuronal precursors to a high percentage of immature GABAergic neurons. Interestingly, although a low endogenous concentration of RA is sufficient to stimulate GABAergic differentiation of cells in the LGE at E14.5, a high concentration of RA is needed for GABAergic differentiation in embryoid bodies derived from ES cells. This may be due to the much more primitive nature of cells in an embryoid body (similar to cells in an early gastrula) compared to neuroepithelial cells of the late embryonic forebrain. As RA binds directly to DNA-bound nuclear receptors that interact with co-repressors and coactivators, we suggest that high concentrations of RA may exert tremendous epigenetic effects on embryoid body cells, driving them to both a neuronal and GABAergic fate. In addition, our RA treatment protocol generated GABAergic neurons exhibiting expression of interneuron transcription factors of either anterior (forebrain) or posterior (spinal cord) identity, but not striatal projection neuron identity. A previous study proposed that mouse ES cells differentiating in medium without RA acquired a GABAergic identity of ventral forebrain co-expressing Gad67 and Isl1 (most likely striatal projection neurons), while exposure to RA resulted in acquisition of a spinal cord interneuron identity [48]; those studies differed from ours in that RA treatment occurred at a later window during embryoid body formation and a lower concentration of RA was used. Thus, differences in the effects of RA on GABAergic interneuron identity in various culture systems may be dependent upon the timing and concentration of RA used.
Production of the inhibitory neurotransmitter GABA in the central nervous system depends on local neurons, and disturbed GABAergic neuron function has been associated with numerous neurological disorders including Huntington's disease, autism, schizophrenia, bipolar depression, and epilepsy [9][10][11][12]. GABAergic interneurons are a particularly attractive cell population for cell-based therapies of these disorders due to their ability to migrate, differentiate, and function following transplantation [64][65][66]. GABAergic interneuron precursors derived from mouse ES cells were shown to migrate, survive for several months, and exhibit neurochemical and electrophysiological characteristics of mature interneurons when transplanted into postnatal cortex [67]. Additionally, transplantation of GABAergic interneuron precursors reduced the number of seizures in epileptic mice [68]. Thus, generation of GABAergic interneurons from RA-treated human ES cells as we report here coupled with isolation of cells with forebrain character may provide useful candidate cells in cell replacement therapies for one or more of these neurological conditions.

RA-Deficient Mouse Models
Raldh32/2 embryos exhibiting postnatal lethality just after birth were previously described [20]. Raldh22/2 embryos exhibiting midgestation lethality have been described previously [69]. To prevent Raldh22/2 early lethality, the maternal diet was supplemented for a short time with a very low dose of RA as described previously [50]. Briefly, 0.1 mg of all-trans-RA (Sigma Chemical Co.) was added per gram of standard mouse chow and provided fresh to pregnant females at E6.75-E8.75. At E9.25 mice were returned to standard chow until embryos were collected at E14.5; Raldh22/2 mutants obtained with this method invariably exhibit stunted forelimbs and interdigital defects, demonstrating a loss of RA function in regions where Raldh2 is responsible for RA synthesis [50]. Dietary supplementation with this dose of RA is indeed low as HPLC measurements have shown that it provides less RA to embryos than Raldh2 normally generates [70]. Administered RA is cleared within 12-24 h after treatment ends [69], thus allowing one to examine embryos at E10.5-E14.5 that now lack RA activity normally generated by Raldh2. Embryos were genotyped by PCR analysis of yolk sac DNA and were staged by designating noon on the day of the vaginal plug as E0.5. All mouse studies conformed to the regulatory standards adopted by the Animal Research Committee at the Sanford-Burnham Medical Research Institute.

Neurosphere Preparation and Differentiation
Lateral ganglionic eminence (LGE) and cortex were dissected from E14.5 wild-type and Raldh32/2 mutant embryos and incubated in DMEM containing 0.1% trypsin and 0.05% DNase for 15 min at 37uC followed by mechanical dissociation. The cells were spun down and resuspended at a concentration of 100,000 cells/ml in basic medium DMEM-F12 supplemented with B27, 10 ng/ml basic fibroblast growth (bFGF) factor, and 20 ng/ml epidermal growth factor (EGF). No apparent differences in growth rate or appearance were observed for wild-type compared to Raldh32/2 neurosphere cultures. Neurospheres were differentiated by culturing on plates pre-coated with poly-L-ornithine. EGF and bFGF were removed from the expansion medium and 1% serum was added (differentiation medium). The spheres were maintained under differentiation conditions for 7 d in the presence or absence of 100 nM RA before fixation.

Explant Culture and RA Bioassay
Tissue explants or neurospheres from wild-type and mutant embryos were cultured overnight on Sil-15 F9-RARE-lacZ RA reporter cells followed by detection of b-galactosidase activity as previously described [26].

Human Embryonic Stem Cell Culture and Differentiation
Human ES cells (line H9) were cultured and passaged weekly on a feeder of irradiated embryonic mouse fibroblasts as described previously [74]. The protocol for RA-induced GABAergic differentiation was based on previous methods [75]. Briefly, human ES cell-derived embryoid bodies (EBs) were cultured in EB growth medium in non-adherent plates for 3 d, followed by RA treatment for 3 d (RAd3). RA was removed and the RAd3 EBs were plated on culture dishes pre-coated with poly-L-ornithine and fibronectin and cultured for an addition 4 d (RAd7) in serumfree neuronal induction medium, comprised of neurobasal medium supplemented with B27, bFGF, and EGF. At RAd7 neuroepithelial rosettes were isolated mechanically from the differentiation cultures with a 2 ml serological pipette. After isolation, rosettes were replated in the same neuronal induction medium for an additional 14 d (RAd21) before fixation.

Quantitation of Immunoreactivity
Immunopositive cells and total DAPI-stained nuclei were counted to calculate the percentage of immunopositive cells. Five randomly picked areas from three independent experiments were counted for each marker. Data were presented as mean 6 SEM; for pair-wise analysis of treatment conditions and/or genotypes, an ANOVA test was used.