|Home | About | Journals | Submit | Contact Us | Français|
To understand the role of specific FGFRs in cortical development, we conditionally inactivated Fgfr2 or both Fgfr1 and Fgfr2 (Fgfr2 cKO or DKO mice, respectively) in radial glial cells of the dorsal telencephalon. Fgfr1 and Fgfr2 are necessary for the attainment of a normal number of excitatory neurons in the cerebral cortex. The action of FGF receptors appears to be through increasing self-renewal of neuronal precursors within the ventricular zone (VZ). Volume measurements, assessments of excitatory neuron number and areal marker expression suggested that the proper formation of the medial prefrontal cortex (mPFC) depends upon the function of Fgfr2, whereas Fgfr1 together with Fgfr2 control excitatory cortical neuron development within the entire cerebral cortex. Fgfr2 cKO mice had fewer and smaller glutamate synaptic terminals in the bed nuclei of the stria terminalis (BST), a projection area for mPFC cortical neurons. Furthermore, Fgfr2 cKO mice showed secondary decreases in GABAergic neurons in the BST and septum. These data demonstrate that FGFR2 signaling expands the number of excitatory neurons in the mPFC and secondarily influences target neurons in subcortical stations of the limbic system.
The cerebral cortex is assembled during embryogenesis through the contribution of two germinal layers, the neuroepithelium abutting the ventricular lumen, or ventricular zone (VZ), and the more superficial subventricular zone (SVZ). Radial glial cells in the VZ generate cortical excitatory pyramidal neurons directly and support their migration to the cortical plate. Radial glia also generate “intermediate” or “basal” progenitors that migrate to the SVZ, divide locally, and also generate neurons that migrate to the cortex (Noctor et al., 2001; Gal et al., 2006; Noctor et al., 2007; Kowalczyk et al., 2009). Thus, the production of cortical pyramidal neurons by both radial glia and the intermediate progenitors within the SVZ contribute to expanded cortical regions and layers that have occurred during cortical evolution (Martinez-Cerdeno et al., 2006).
The prefrontal cortex (PFC) is arguably the cortical region that has undergone the most dramatic expansion during mammalian evolution, albeit by unknown mechanisms. Several members of the Fibroblast growth factor (FGF) family, including FGF2, 8, 15, 17 and 18, are enriched in the anterior telencephalon from which the PFC primordium is derived. Amongst the 3 known FGF receptors (FGFRs) expressed in the brain, Fgfr1 and Fgfr2 are expressed in partially overlapping anterior-posterior decreasing gradient in the cortical neuroepithelium, whereas Fgfr3 is expressed posteriorly (Vaccarino et al., 1999; Muzio et al., 2002; Hasegawa et al., 2004). Some of the FGF ligands are crucial for the specification of anterior cortical areas. For example, Fgf8 gene expression is necessary and sufficient to expand PFC regions and germline Fgf17 knockout mice exhibit reduced the medial PFC (mPFC) territory (Fukuchi-Shimogori and Grove, 2001; Garel et al., 2003; Cholfin and Rubenstein, 2007). In addition to early patterning, Fgf genes have a profound impact on neurogenesis in a region specific manner. For example, the deletion of the Fgf2 gene reduces dividing precursor cells, excitatory cortical neurons and astroglia more strongly in the PFC (Vaccarino et al., 1999; Raballo et al., 2000; Korada et al., 2002) and, similarly, Fgf8 is required for cortical neural stem cells self-renewal, whereas Fgf15 appears to be acting in an antagonistic manner (Borello et al., 2008).
FGF receptor signaling is required at early neural plate stages for the formation of the telencephalon, apparently promoting telencephalic cell survival (Paek et al., 2009). Disruption of Fgfr1 and Fgfr2 at early (pre-neurogenic) stages of telencephalic development demonstrated altered proliferation and cell death, particularly in midline regions (Gutin et al., 2006; Ever et al., 2008). Mice transiently expressing a dominant-negative Fgf receptor transgene able to antagonize all FGFRs in the embryonic cortical neuroepithelium exhibited decreased proliferation, cortical atrophy and loss of TBR1+ excitatory neurons in mPFC and temporal cortex (Shin et al., 2004); consistent with this, mice lacking all three receptors in the dorsal telencephalon demonstrate impairment in normal production of cortical stem and progenitor cells (Kang et al., 2009). However, the role of each FGF receptor type in cortical morphogenesis is still unclear. The disruption Fgfr1 is not sufficient to cause significant abnormalities in cortical neurogenesis or patterning (Hebert et al., 2003; Ohkubo et al., 2004). Fgfr2 is expressed in a partially overlapping fashion with Fgfr1 by cortical radial glia in anterior regions, and may control, alone or together with Fgfr1, cell proliferation and differentiation in response to the anterior source of FGF ligands.
To understand the contribution of Fgfr1 and Fgfr2 to mPFC morphogenesis and their roles in VZ/SVZ precursors in anteromedial regions, we characterized populations of cortical neuron subtypes and precursor cells in the dorsal neuroepithelium of conditional knockout mice with a deficiency of Fgfr2, with or without Fgfr1, in radial glial cells. We examined anteromedial regions of the forebrain suggested to be more affected in dominant negative Fgfr1 mutant mice (Shin et al., 2004) and previous manipulations of Fgfr2 (Gutin et al., 2006) and extended our characterization to include synaptic projections and neuronal populations in key subcortical structures that receive connections from mPFC.
The conditional Fgfr2 null allele harbors loxP recombination sites flanking regions encoding the Ig III binding and transmembrane domains of the Fgfr2 gene (Fgfr2f) (Yu et al., 2003). The conditional Fgfr1 null allele harbors loxP recombination sites flanking the transmembrane and kinase domains of the Fgfr1 gene (Fgfr1f) (Pirvola et al., 2002). Mice homozygous for Fgfr2f alleles, or for both Fgfr2f and Fgfr1f alleles, were crossed with mice expressing the Cre recombinase transgene under the control of the human GFAP promoter (hGFAP) (Zhuo et al., 2001). The hGFAP-Cre transgene targets Cre recombination to radial glia progenitors of the dorsal telencephalon starting at E13.5 (Fig 1A,B; Ohkubo et al., 2004). Cre negative animals, littermates when possible, were used as control animals. To assess early time points of inhibitory neuron development, mice homozygous for both Fgfr2f and Fgfr1f alleles were crossed with GAD67-GFP (delta-neo) mice (Tamamaki et al., 2003). All experimental procedures involving animals were performed in accordance with the Yale Animal Resources Center and Institutional Animal Care and Use Committee (IACUC) policies.
Deeply anesthetized animals were intracardially perfused first with phosphate buffered saline (PBS) followed by 4% paraformaldehyde in PBS. Brains were cryoprotected and cryostat (Leica, CM1900, Bannockburn, Illinois) sectioned at 20 μm and at 50 μm thickness respectively for embryonic ages—embryonic day 15.5 (E15.5) and 16.5 (E16.5)—and adult ages—six weeks and three months. Free floating immunostaining was used for adult brain tissue and slide mounted processing was used for embryonic brain tissue. Sections were processed for staining first by blocking for 1 hour with 10% goat serum in PBS with 0.025% TritonX-100, 0.0125% Tween20 (PBS++) and then incubating for 24-48 hours at 4°C with 5% goat serum/PBS++ containing primary antibodies as follows: β-galactosidase (βgal) (1:10,000; Cappel, 56028, #05036), brain-1 (1:500; M. Wegner), caspase-3 (1:500; Cell Signaling, #9661), Cux1 (1:100; Santa Cruz, #sc-6327; used with unmasking reagent; Vector Laboratories, #H-3300) neuronal class III β–tubulin (Tuj1) (1:1000; Promega, G7121, #131959), reelin (1:1000; Chemicon, #MAB5364), bromodeoxyuridine (BrdU) (1:500; Accurate Chemicals, OBT0030, #H8913), proliferating cell nuclear antigen (PCNA) (1:100; Upstate, 05-347, #32555), phosphohistone 3 (pH3) (1:1000; Sigma, H9908), T-brain-1 (TBR1) (1:1000; Abcam, AB31940), paired box gene 6 (Pax6) (1:1000; Abcam, AB78545), green fluorescent protein (GFP) (1:1000; Abcam, AB13970, #660556), T-brain-2 (TBR2) (1:1000; Chemicon, AB9616), neurofilament heavy (SMI-32) (1:1000; Covance, #14864801), glutamic acid decarboxylase 67 (GAD67) (1:2000; Chemicon, MAB5406, #0606034244), calretinin (CR) (1:2000; Chemicon, MAB1586, #17120001), parvalbumin (PV) (1:2000; Sigma, #P3088), somatostatin (1:1000; Abcam, AB64053), neuronal nuclei (NeuN) (1:500; Chemicon, MAB377), gamma-aminobutyric acid (GABA) (1:5000; Sigma, #A2052), and SATB homeobox 2 (SATB2) (1:1000; V. Tarabykin). Brain sections were then washed three times in PBS followed by an incubation in 5% goat serum/PBS++ containing Alexa dye-conjugated secondary antibodies (1:250-1000; Molecular Probes) or AMCA (1:100; Vector Laboratories, #CL-100) for fluorescent detection. Sections were coverslipped using mounting medium with DAPI (Vector Laboratories, #H-1200).
To evaluate the timing and specificity of Cre-recombination affecting Fgfr2 gene expression, females with timed pregnancies were deeply anesthetized and euthanized by cervical dislocation on E13.5 and E15.5. Embryos were removed and the dorsal and ventral regions of the telencephalon were dissected out from the embryonic brains. The cerebral cortex and hippocampus were pooled in one sample from each embryo and the ganglionic eminences for a second sample, immediately homogenized in Trizol reagent (Sigma) and frozen. After thawing, RNA was isolated using chloroform phase separation, precipitated with 75% alcohol, and re-suspended in water. RNA concentrations were determined using a Nanodrop Spectrophotometer (Thermo Scientific). cDNA was synthesized using Superscript III First Strand Synthesis Kit (Invitrogen). Quantitative PCR was carried out using Taqman Gene Assays (Applied Biosystems) for FGFR2 IIIc isoform (ID Mm01269938) and Beta-actin with GeneAmp Fast PCR Mastermix (Applied Biosystems) in a StepOne™ Instrument (Applied Biosystems).
To evaluate FGFR2 protein expression, two month old mice were anesthetized and decapitated. The cerebral cortex was dissected from other brain regions and homogenized on ice in protein lysis buffer, containing, in mmol/L: 50 Tris–HCl, pH 8.0, 150 NaCl, 50 NaF, 1 Na3VO4, 0.1 PMSF, 1 benzamidine and in g/L: 10 Nonidet P-40, 5 sodium deoxycholate, 0.002 aprotinin,1 ×10−5 leupeptin and 1×10−5 pepstatin. Homogenates were centrifuged at 13,200 rpm for 20 min at 4°C. Supernatants were stored at −80°C until used. For western blotting, equal amounts of homogenate (5 μl) were resolved in 10% sodium dodecyl sulfate–polyacrylamide gel electrophoressis and transferred onto a nitrocellulose membrane. Membranes were blocked at room temperature for 60 minutes in 5% nonfat dry milk in Tris-buffered saline with Tween 20 (TBST) (containing, in mmol/L:10 Tris-HCl buffer, pH 8.0, 150 NaCl and 0.1% Tween-20), and incubated overnight at 4°C with rabbit anti-FGFR2 antibody (1:1500; Sigma-Aldrich #F0300) in 5% nonfat dry milk in TBST. Blots were washed three times in TBST and incubated with horseradish peroxidase-conjugated donkey anti-rabbit secondary-antibody (1:5000; GE Healthcare #NA934) in 5% nonfat dry milk in TBST. Bands were detected using enhanced chemiluminescence (Pierce Biotechnology). Blots were stripped and then reincubated for one hour at room temperature with mouse anti-βactin (1:4000; Sigma-Aldrich #115K4825), washed and incubated with horseradish peroxidase-conjugated sheep anti-mouse secondary-antibody (1:10,000; GE Healthcare #NA931(V)) using the same procedure as above.
BrdU was dissolved in 0.07N NaOH at the concentration of 5 mg/ml, sterile filtered and injected i.p. in the proportion of 100 mg/kg at E14.5, 24 or 48 hours prior to animal perfusion.
Unbiased estimates of TBR1, pH3, TBR2, SMI-32, GAD67, CR, PV, NeuN and GABA immunopositive cells were obtained with a computer running the StereoInvestigator software (Microbrightfield, Colchester, Vermont) and coupled to a Zeiss Axioskope 2 Mot Plus (Carl Zeiss, Oberkochen, Germany) equipped with a lucivid and calibrated motorized stage controller that allows precise control of y-, x-, and z-axes. The use of the stereological approach produced data containing consistent cell counts with variances within each sample and across sampling sites that were in a statistically acceptable range. Using stereology to comprehensively assess brain regions of interest eliminates potential chance differences and bias in the process of selecting sites to evaluate (Peterson, 1999).
Using the optical fractionator, nuclear profiles were counted in 3-dimensional counting boxes. For embryonic stereological TBR2 and pH3 counts, every 10th section was used beginning with the most anterior section that showed a clear ventricular and subventricular zone and continuing posteriorly to the last section before the hippocampus. Randomly placed sampling grids of 250×250 μm were used with a counting frame of 50×50×5 μm. For adult stereological results, counting frames measured 100×100×5 μm and sampling grids varied according to the region of interest (800×800 μm in neocortical areas, 400×400 μm in the medial PFC, ventral orbital frontal cortex (vOFC), cingulate cortex (CC) and septum, and 200×200 μm in the bed nuclei of the stria terminalis (BST). For counts in the neocortex and cingulate cortex, 50 μm coronal sections were counted at a frequency of one every 20th section. Cortical contours were drawn at the lower extent of layer 6 and excluded the archicortex with a boundary located at border with the entorhinal/insular cortex. Boundaries of the CC were defined anteriorly by the rostral extent of the corpus callosum or Probst bundles, ventrally by the cortical white matter and dorsally by changes in Layer II which becomes less densely-packed and more homogenous in the non-cingulate cortical region. For mPFC and vOFC, 50 μm coronal sections were counted at a frequency of one every 10th section. The mPFC in rodents has been shown to encompass portions of the anterior cingulate, prelimbic and infralimbic cortices; the mPFC boundaries (Fig. 3) were therefore drawn using previously described cytoarchitectonic differences for these regions (Krettek and Price, 1977) including homogeneity and density of Layer V (high in mPFC, low in precentral cortex), definition of overall lamination (high in mPFC, low in infralimbic), and evenness of Layer II (high in mPFC, low in infralimbic). For the BST and the septum, 50 μm coronal sections were counted at a frequency of one every 5th section. Boundaries of the dorsal portion of the BST nuclei were defined by the posterior anterior commissure, laterally by the striatum, ventrally by densities of hypothalamic nuclei, and dorso-medially by the septal nuclei. Boundaries for the septal nuclei were determined dorsally by the corpus callosum or, when absent, cortex, and ventro-laterally by the striatum. Variations between counts was assessed by the Gunderson coefficients of error and data were only used when cell count error values were sufficiently low (<0.15).
In order to evaluate densities of PCNA and PH3 cells within embryonic regions of interest, a blind method of counting cells within standardized regions of VZ and SVZ was adopted. For PCNA counts, 40x Apotome Z-stack micrographs obtained on an ApoTome equipped Axiovert 200M with Axiovision 4.5 software (Carl Zeiss, Thornwood, NY, USA) were examined in one to three representative anterior and posterior sections in each brain by an experimenter blind to the animal genotype. For PH3 counts, 5x micrographs of dorsal telencephalon were examined in two to three representative anterior and posterior sections in each brain by an experimenter blind to the animal genotype. For evaluations of cell cycle re-entry, three sections in each brain triple stained with Pax6, PCNA and BrdU were used to count nuclear profiles within the VZ and SVZ. The density of double labeled PCNA/BrdU cells was assessed within 40x Z-stacks (total volume: 7.0 ×104 μm3) and the percentage of PCNA/BrdU double immunostained cells versus the total number of BrdU labeled cells was calculated.
Collection and sectioning of postnatal day 0 brain tissue was performed as described above for immunocytochemistry using RNAse free solutions. In situ hybridization was performed in frozen sections using the digoxigenin system as described previously (Ohkubo et al., 2004) with some modification. Briefly, a riboprobe incorporating digoxygenin-labeled nucleotides was synthesized from linearized plasmids for Cadherin 6 derived from the subcloned PCR fragments. Sections were fixed in 4% paraformaldehyde, washed in PBS, and dehydrated through grades of ethanol. Sections were then air dried, treated with proteinase K in buffer, fixed again, rinsed, acetylated, rinsed again in water, and air dried. Prehybridization was done with incubation in hybridization buffer followed by washes in 2X SSCand RNAase treatment. Tissue then underwent washes in 2X SSC, 50% heat-inactivated sheep serum in PBS. Sections were incubated overnight in a solution containing anti-digoxigenin Fab fragments, conjugated to alkaline phosphatase, preabsorbed with chick embryo powder to reduce background labeling. Following antibody incubation, sections were washed and incubated in developing buffer. The color reaction was carried out by incubation in developing buffer with NBT/BCIP. Finally, sections were washed in PBS, dehydrated through grades of ethanol, and mounted in Permount.
Fifty μm thick sections containing the dorsal portion of the BST (as defined in the preceding section) were washed in 0.1 M phosphate buffer (PB). To eliminate unbound aldehydes, sections were incubated in 1% sodium-borohydride and then rinsed in PB. Subsequently, sections were incubated with GAD-67 or vGlut2 antibodies (Chemicon), followed by incubation in biotinylated secondary immunoglobulin. Then, sections were incubated in avidin-biotin-complex and the tissue-bound peroxidase was visualized by a diaminobenzidine reaction. After immunostaining, unstained cortical perikarya (10 from each mouse) in the BST in flat embedded blocks were randomly selected. The blocks were osmicated (in 1% osmium tetroxide in PB), and dehydrated in increasing ethanol concentrations. During the dehydration, 1% uranyl-acetate was added to the 70% ethanol to enhance ultrastuctural membrane contrast. Dehydration was followed by flat-embedding in Araldite. Ultrathin sections were cut on a Leica ultra microtome, collected on Formvar-coated single-slot grids and analyzed with a Tecnai 12 Biotwin (FEI Company, Hillsboro, Oregon) electron microscope.
To obtain a quantification of synaptic number, unbiased for possible changes in synaptic size, the disector technique was used. On consecutive 90-nm-thick sections, we determined the average projected height of the synapses and used about 30% of this value as the distance between the disectors. On the basis of this calculation, the number of synapses was counted in two consecutive serial sections about 270 nm apart (“reference” and “look-up” sections) of 10 immunolabeled perikarya profiles in each animal. Synapse characterization was performed at a magnification of 20,000. Symmetric and asymmetric synapses were counted on all selected neurons only if the pre- and/or postsynaptic membrane specializations were seen and synaptic vesicles were present in the presynaptic bouton. Immunolabeling for vGlut2 was associated with asymmetrical, excitatory synapses, and immunolabeling for GAD67 was associated with symmetrical, inhibitory connections. Synapses with neither clearly symmetric nor asymmetric membrane specializations were excluded from the assessment. The plasma membranes of selected cells were outlined on photomicrographs and their length was measured. Plasma membrane length values measured in the individual animals were added and the total length was corrected to the magnification applied. Synaptic densities were evaluated according to the formula NV=Q-/Vdis, where Q- represented the number of synapses present in the “reference” section that disappeared in the “look-up” section. Vdis is the disector volume (volume of reference), which is the area of the perikarya profile multiplied by the distance between the upper faces of the reference and look-up sections, i.e., the data are expressed as numbers of synaptic contacts per unit length (100 μm) of perikaryon membrane. Section thickness was determined by using the minimal fold method.
Univariate analysis of variance (ANOVA) was performed with SPSS to identify whether cell count deficits in KO animals determined by stereology were different across regions, determined by interactions between two variables: brain regions and genotype. ANOVA was also performed on zone-counted data in the VZ and SVZ to identify interactions between age and genotype. Student’s t tests were performed with Microsoft Excel to compare specific brain regions and numbers of cells for stereological cell counts and densitites.
Targeted inactivation of the Fgfr2 gene or both the Fgfr2 and Fgfr1 genes in radial glia was accomplished via Cre-dependent recombination of the loxP-flanked sequences in the Fgfr2f and Fgfr1f alleles. FGF receptors are expressed in the developing VZ and are enriched in cells located at the apical surface, likely radial glia (Fig. 1D) (Vaccarino et al., 1999; Vaccarino, 2000; Vaccarino et al., 2001; Yoon et al., 2004). Cre was driven by a human GFAP-Cre transgene (hGFAP-Cre) active in telencephalic radial glial cells beginning at E13.5–E14.5 5 (Fig 1A-C) (Zhuo et al., 2001; Ohkubo et al., 2004, Kang et al., 2009). Mice homozygous for the recombined Fgfr2 or both Fgfr1 and Fgfr2 null alleles were referred to as Fgfr2 cKO and Fgfr1;Fgfr2 DKO (their genotypes being, respectively, Fgfr2f/f;hGfapCre and Fgfr1f/f; Fgfr2f/f;hGfapCre). The hGFAP-Cre transgene was targeted to the dorsal telencephalon beginning at embryonic day 13.5, and had virtually no expression in the subcortical telencephalon, as shown by the β-galactosidase reporter (Fig.1A-C). Quantitative PCR confirmed previous reports (Kang et al., 2009) on this mouse line that Fgfr2 mRNA was significantly decreased only in the E13.5 and E15.5 dorsal telencephalon (Supplemental Table 1). Western blot demonstrated a clear reduction in the wild-type FGFR2 protein in the cortex of the dorsal telencephalon of adult Fgfr2 cKO mice, with appearance of a 25% smaller protein, which likely represents the Cre-deleted mutant form (Fig. 1E).
The inactivation of all FGFRs in cortical precursor cells using a dominant negative Fgfr1 transiently expressed using the Otx1 promoter caused a 45% reduction in cortical pyramidal neurons expressing TBR1 (Shin et al., 2004), a transcription factor expressed by early-born excitatory neurons of the cerebral cortex (Hevner et al., 2001). Consistent with these results, TBR1 staining was decreased throughout the cortical plate of Fgfr1;Fgfr2 DKO mice at E16.5 as compared to control mice (Fig. 2A-F), suggesting that Fgfr2 and Fgfr1 might be contributing to this defect.
To understand the mechanism of the deficiency in TBR1+ cells, we used an unbiased stereological approach to quantify proliferating precursors in the developing brain. We immunostained coronal sections for TBR2, a transcription factor whose expression is enriched in intermediate precursors as they become committed to this cell fate in the VZ and then migrate into SVZ where they proliferate. TBR2 expression temporally precedes TBR1 expression in neuronal precursor cells (Englund et al., 2005). The density of TBR2-immunostained cells in the anterior cortical wall, throughout VZ, SVZ and the intermediate zone, was reduced in Fgfr2 cKO and Fgfr1;Fgfr2 DKO mice at E16.5 compared to controls (Fig. 2G,H,J). The 27% TBR2 deficit found in this sample of Fgfr1;Fgfr2 DKO mice (n=2 per group) was confirmed in another sample of Fgfr1;Fgfr2 DKO mice (n=2 per group) in an independent experiment, which demonstrated a 21% deficit in TBR2 density.
To understand the mechanism by which TBR2+ intermediate precursors were depleted in Fgfr2 cKO and Fgfr1;Fgfr2 DKO animals, we first evaluated whether cell proliferation was affected by using the cell cycle proteins phosphohistone 3 (pH3) and proliferating cell nuclear antigen (PCNA), which are constitutively expressed by dividing cells during M phase and throughout the cell cycle, respectively. Consistent with the TBR2 data, the density of PCNA+ cells was reduced in the SVZ of Fgfr1;Fgfr2 DKO at E16.5 as compared to control mice (Supplemental Fig. 1). Stereological assessment of the number of pH3+ cells demonstrated a 40% reduction in the number of mitotic cells in the VZ in Fgfr1;Fgfr2 DKO (Fig.2 G-I). Furthermore, the distribution of pH3+ cells in basal vs apical regions of VZ was altered. Dividing cells within basal regions (closer to the pial surface) reflect the proliferative activity of intermediate precursors of the SVZ, while dividing cells within apical regions (apposing the ventricular surface) reflect proliferative activity of primary progenitors of the VZ; examining the ratio of basal to apical pH3+ cells within specific areas of developing cortex examines effects on each population while controlling for overall alterations in proliferation. The ratio of basal to apical pH3+ nuclei in the dorsal telencephalon was significantly increased in Fgfr1;Fgfr2 DKO compared to control littermates. The basal:apical ratio was more significantly affected in anterior (0.81 ± 0.23 vs 1.47 ± 0.16 in controls vs Fgfr1;Fgfr2 DKO) than posterior regions (0.68 ± 0.06 vs 0.89 ± 0.16 in controls vs Fgfr1;Fgfr2 DKO). Alterations in the pH3+ basal:apical ratio were due to disproportionate reductions of pH3+ apical cells in mutant mice. Indeed, the density of apical pH3+ cells was 36 ± 2.6 cells/field in controls vs 21 ± 1.8 cells/field in Fgfr1;Fgfr2 DKO, while the density of basal pH3+ cells was 27 ± 5.5 in controls vs 24 ± 2.6 cells/field in Fgfr1;Fgfr2 DKO). Again, this difference was almost three fold greater in anterior sections, with 55% fewer apical pH3+ cells (p<0.05), compared to posterior sections which showed only 21% fewer apical pH3+ cells. The specificity of this loss to the apical side suggested that FGFR signaling affected radial glial progenitors in VZ primarily.
To further understand the mechanism of these deficits in radial glial and intermediate progenitors, we evaluated cell renewal and cell death amongst neural precursor cells of the VZ and SVZ. Specifically we quantified the number of dividing cells at E14.5 (labeled with BrdU) that re-entered the cell cycle by E15.5 or E16.5 (as determined by double-labeling with PCNA). The fraction of BrdU-labeled cells that re-entered the cell cycle (PCNA/BrdU double labeled cells) was reduced in the VZ of Fgfr1;Fgfr2 DKO animals as compared to their littermate controls (ANOVA for genotype F(1,9) = 24.83; p=0.002) (Fig. 3A-P, quantification in Q,R). Virtually all of these PCNA/BrdU double labeled cells in the VZ were also positive for Pax6, a transcription factor expressed by radial glial cells. In contrast, the proportion of cells re-entering the cell cycle was not significantly different in mutants versus controls in the SVZ (Pax6-negative cells), although a trend to a decrease was noted 48h after BrdU labeling (i.e., at E16.5) (Fig. 3R) Furthermore, at both time points (i.e., E15.5 and E16.5) the thickness of the VZ as assessed by Pax6 immunostaining, was markedly reduced in Fgfr1;Fgfr2 DKO compared to control littermates (Fig. 3, compare B,J with F,N) .Small increases in the number of dying cells were observed at E15.5 and E16.5 in Fgfr1;Fgfr2 DKO compared to littermate controls (Supplemental Fig. 2). However, the quantity of embryonic cell death was insufficient to account for the deficit in neuronal precursors. Hence, the loss of normal FGFR signaling in dorsal telencephalic cells produces a deficient self-renewal of Pax6+ precursors in the VZ.
The Fgfr2 cKO and Fgfr1;Fgfr2 DKO lacked the corpus callosum and hippocampal commissure approximately 66% and 90% of the time respectively, a more frequent occurrence of this deficit than described for the Fgfr1 cKO mice (Smith et al., 2006). Brain volume in the Fgfr2 cKO, Fgfr1;Fgfr2 DKO and respective control animals was assessed by stereological methods. Total cortical volume was decreased by 25.8 % in the Fgfr1;Fgfr2 DKO and by 9.2% in Fgfr2 cKO animals compared to control littermates (Table 1, Fig. 4A-D). Cortical surface area and thickness were both affected to the same degree in Fgfr2 cKO animals as compared to controls (thickness decreased by 7.9%, surface area decreased by 6.5%).
Based on previous findings that some FGF ligands binding to FGFR1 and FGFR2 are expressed in gradients emanating from the anterior midline (Maruoka et al., 1998; Xu et al., 1999; Crossley et al., 2001; Storm et al., 2006) and findings in the current study showing greater reductions in apical pH3+ cells anteriorly, anterior and medial cortical regions were subdivided on the basis of cellular architecture, as described in the rat by Krettek and Price (1977) in order to parcellate midline integrative areas from sensory and motor regions in more dorsal and lateral zones and from more ventral limbic cortices (Fig. 4E-L). Stereology was used to estimate the volume and neuron number in the dorsomedial frontal cortical region, also called medial prefrontal cortex (mPFC), as compared to the ventral orbital frontal cortex (vOFC) and cingulate cortex (CC).
As noted above, Fgfr2 cKO animals showed a decrease in cortical volume across regions compared to controls, (F(1,35) = 3.900; p= 0.05) but interestingly, they also showed a genotype x region interaction F(2,35) = 3.544; p=0.042). Indeed, there was a 21% decrease in volume of mPFC in the Fgfr2 cKO animals which was more than two fold as affected as the overall cortex (Table 1; Fig. 4). For the Fgfr1;Fgfr2 DKO animals, decreases in cortical volume were comparable across regions (ANOVA genotype x region nonsignificant). These animals showed a 36.1% decrease in mPFC volume, a deficit equivalent to that in overall cortex (Table 1). Within the mPFC, the surface area in its dorsal-ventral extent was observed to be more affected than its thickness. The other frontal and medial cortical regions, vOFC and CC, showed no significant decreases in volume in either of the mutant animals (Table 1) and were also not observed to have reductions in the surface area or thickness compared to controls.
This volume decrease in mPFC was confirmed with in situ hybridization for cadherin 6 (Cad6). Cad6, a gene which is expressed in lateral dorsal cortex (the somatosensory region) undergoes a medial shift in its expression pattern in Fgf17 mutant mice, which have a contraction of mPFC areas and a corresponding expansion of the adjacent somatosensory territory (Cholfin and Rubenstein 2007). In Fgfr2 cKO and Fgfr1;Fgfr2 DKO mice, Cad6 expression extended more medially, consistent with a reduction in the mPFC territory without a corresponding expansion of adjacent cortical areas (Fig. 4M-Q). This is consistent with an FGFR-mediated change in neurogenesis affecting the size of mPFC neuronal population, rather than global cortical re-patterning.
To assess whether the reduction in cortical volumes were attributable to a change in total neuron number, excitatory neurons in the cortex were evaluated by stereological sampling of serial sections using antibodies against two pyramidal cell markers, the transcription factor TBR1 and the neurofilament SMI-32. TBR1 is enriched in pyramidal neurons populating lower cortical layers, while SMI-32 is expressed by a subset of pyramidal neurons in all cortical layers (layers II, V and VI). TBR1 neuron number in Fgfr1;Fgfr2 DKO animals was decreased to the same extent in the overall neocortex and in mPFC (33% and 31%, respectively; nonsignificant interaction of genotype and brain region by ANOVA) (Table 2). TBR1 appeared reduced in both upper and lower cortical layers (Fig. 5K,L and O,P). In contrast to TBR1, evaluation of SMI-32+ neurons demonstrated more severe reductions in number and density that were greater in mPFC (ANOVA for genotype F(1,23) = 7.191; p= 0.016; genotype x region interaction F(2,23) = 3.692; p=0.016). Indeed, the number of SMI-32+ cortical neurons of Fgfr1;Fgfr2 DKO adult mice were reduced by 36% and 62% in overall neocortex and mPFC, respectively, as compared to control mice (Table 3). Like TBR1, SMI32 appeared reduced in both upper and lower cortical layers (Fig. 5I,J and M,N). In the mPFC, the density of SMI-32+ neurons was also reduced: there were 4.31 ± 0.52 (SEM) SMI-32+ neurons in control and 1.78 ± 0.24 × 105 cells/mm3 (p<0.01) in Fgfr1;Fgfr2 DKO mice. The CC exhibited a lesser decrease in excitatory neuron number (19.8% and 15% for TBR1 and SMI-32, respectively) and the vOFC showed a variable effect (no significant effect for TBR1 and 52% decrease for SMI-32) (Tables (Tables22 and and33).
We next analyzed the phenotype of Fgfr2 cKO mutants. The Fgfr2 cKO animals also showed significant deficits in TBR1+ neuron numbers in cortex compared to controls (ANOVA for genotype F(1,29) = 4.373; p=0.047). Interestingly, the Fgfr2 cKO animals showed a 37% deficit in TBR1 neuron number in mPFC, comparable to that observed in the Fgf1;Fgfr2 DKO mice (Table 2), however, the interaction between genotype and brain region only approached significance (F(2,29) = 3.124; p=0.062). The density of TBR1+ neurons in mPFC was also decreased by 27% (0.74 ± 0.05 and 0.54 ± 0.05 × 105 cells/mm3 in control and Fgfr2 cKO mice, respectively; p<0.05). In contrast, neither the vOFC, CC nor the neocortex of Fgfr2 cKO animals had significant decreases in TBR1 number and density compared to controls (Table 2). In order to assess any gradient of effect on TBR1+ cell density in cortex, cell density was quantified separately in three regions: rostral, intermediate and caudal cortex. A gradient of the TBR1+ cell deficiency demonstrated that the lack of Fgfr2 affected rostral cortical regions to a greater extent than caudal regions (Supplemental Fig. 3).
The more widely distributed excitatory neurons labeled with SMI-32 were decreased to a greater extent in the Fgfr2 cKO mutant cortex than TBR1 neurons (Fig. 5A-H; Table 3; ANOVA F(1,17) = 6.948; p=0.022). In overall neocortex, there was a 35% decrease in both number and density of SMI-32+ neurons in Fgfr2 cKO animals compared to controls (Table 3). However, there was a stronger 52% decrease in SMI-32 neuron number within the mPFC in Fgfr2 cKO animals (Table 3). The interaction of cortical SMI-32 number between genotype and region was statistically significant (F(2,17) = 5.797; p=0.017). The density of SMI-32+ neurons in overall neocortex was also decreased, i.e., 1.53 ± 0.28 and 1.01 ± 0.02 × 105 cells/mm3 in control and Fgfr2 cKO mice, respectively (p<0.05). In vOFC and CC, SMI-32 counts were lower in mutants than in controls, but differences were not shown to be significant. The similarity in pyramidal neuron loss in mPFC among Fgfr1;Fgfr2 DKO and Fgfr2 cKO animals (respectively, −62% and −52% for SMI-32 and −31% and −37% for TBR1) suggest that the Fgfr2 gene, but not the Fgfr1 gene, is important for the acquisition of an appropriate number of pyramidal cells and volume expansion of the mPFC. Despite the overall reduction in cortical thickness, layer distribution did not appear significantly different as assessed by SATB2 and SMI-32 staining (Fig. 5C,D and G,H).
The loss of TBR1+ and SMI-32+ neurons was consistent with the volume losses documented for the overall cortex and prefrontal regions. In addition, decreases in the density of excitatory neurons in mutant mice might also suggest that a compensatory expansion of some non-excitatory neuron elements was enacted in cortical development. However, DAPI cell density within mPFC demonstrated a deficit in Fgfr2 cKO mice (18.24 × 10−5 cells/mm3) compared to controls (22.35 × 10−5 cells/mm3), which was higher than the cumulative loss of SMI-32 and Tbr1+ neuron density in mPFC. This suggests that mutant mice had no compensatory proliferation of cells within cortex and that the decrease in excitatory neuron density is attributable to an expansion of the neuropil.
Lastly, despite the overall reduction in cortical thickness, layer distribution did not appear significantly different. Fgfr2 cKO and Fgfr1;Fgfr2 DKO mice were similar to control littermates in the distribution of lower layers, assessed by SATB2 and SMI-32 staining (Fig. 5C,D and G,H), as well as upper layers, assessed by Brain-1 (Brn-1), Cut-like homeobox 1 (Cux-1), expressed in cortical layers II and III, and Reelin, a marker for cortical layer I (Supplemental Fig. 4).
Consistent with the pyramidal neuron deficits, adult animals lacking either Fgfr2 (Fig. 6) or both Fgfr1 and Fgfr2 (Table 1) showed significant reductions in cortical white matter compared to controls. White matter volume in Fgfr2 cKO animals was reduced by half and white matter tracts were clearly reduced. Fgfr2 cKO animals frequently lacked a corpus callosum which contributed to deficits in white matter found in medial compartments of cortical white matter (−68%; p<0.005). However, white matter volume was also reduced in lateral compartments (−37%; p<0.01). S100β+ glial cells populating white matter regions showed no deficit in the brains of either Fgf receptor knockout animal lines (Table 1). As no deficit in oligodendrocytes or myelin has been observed in mice lacking Fgfr2 in oligodendrocyte progenitors (Kaga et al., 2006), and Cre recombination does not affect the diencephalon in our mutants, the data suggest that the deficit in white matter volume is likely attributable to a decrease in number, caliber and/or branching of efferent cortical neuronal fibers of pyramidal cells.
The observed decrease in cortical excitatory neurons, particularly in mPFC, and cortical white matter suggested that projections to subcortical regions would be affected in mice lacking Fgfr2. Glutamatergic projections were evaluated with electron microscopy in the the bed nuclei of the stria terminalis (BST), a known site of projection for the mPFC (Heidbreder and Groenewegen, 2003). A 73% reduction in excitatory synaptic terminals immunostained with the glutamate transporter vGlut2 was identified in the BST of Fgfr2 cKO mice compared to controls (p<0.05) (Fig. 7). Inhibitory synapses within the same brain region were not reduced, demonstrating a specific deficit in efferent synapses likely attributable to reduced projections from mPFC.
In order to evaluate whether the loss of mPFC neurons and their synaptic terminals secondarily affect neurons in target regions, we examined two sub-cortical regions that receive projections from mPFC, the BST and the septal nuclei. As compared to control mice, adult Fgfr2 cKO mice exhibited strong decreases of 80% and 62% in inhibitory neurons immunoreactive for GABA within the BST and septum, respectively (Table 4, Fig. 8C-D). Similarly, significant decreases of 54% and 38% were found in the densities of calretinin (CR)+ neurons in BST and septum, respectively, of Fgfr2 cKO mice compared to controls. A 48% deficit in somatostatin+ cells was found restricted to the BST (Table 4). Neurons immunoreactive for calbindin showed a trend to decreased densities in both the BST and Septum. Notably, Cre recombination did not affect these regions (Fig.1A) suggesting that cell deficiences were due to secondary effects. Examinations of a pan-neuronal marker, NeuN, in the septum of adult animals demonstrated deficits of neuronal densities of 23.6% in septum and 33.0% in BST (n=2,2). GABA+ cells were approximately 60% of total NeuN+ cell numbers in both regions, suggesting that the loss of GABA+ and CR+ cells was a true cellular deficit, accounted for by a proportional loss in total neuron number.
To explore the mechanism of the inhibitory neuron losses, we examined neonatal and P7 animals. The septum of Fgfr2 cKO and Fgfr1;Fgfr2 DKO mice and control littermates demonstrated comparable densities of GAD67+ and GABA+ cells at P0 and P7 respectively (Supplemental Fig. 5A-D, K-L). Calretinin+ cells, which were decreased in the septum of adult animals, were not altered in the septum of control and mutant animals at P0 and P7 (Supplemental Fig. 5E-F, I-J). We also analyzed Fgfr1;Fgfr2 DKO animals carrying a GAD67-GFP transgene, which labels all GABAergic interneurons from the time they are generated during embryogenesis (Tamamaki et al., 2003). GAD-67-GFP+ cells showed no deficiency in the developing septum of Fgfr1;Fgfr2 DKO animals at P0 (Supplemental Fig. 5G-H). Together, the data suggest that the loss of GABAergic cells develops sometimes in the postnatal period.
We then examined potential relationships between the cerebral cortical and subcortical phenotypes by statistical analyses. The density of GABA+ cells in BST and septum in WT and Fgfr2 cKO mice correlated with the number of TBR1+ cells in mPFC (r= 0.86, p<0.05 for BST and r= 0.93, p<0.01 for septum) (Fig. 8F,H) but not with the number of TBR1 positive cells in total cortex (Fig. 8E,G). The density of GABA positive cells in BST but not in septum also correlated with cortical white matter volume (respectively r=0.92, p<0.01 for BST and r=0.78 for septum).
The present results demonstrate that the inactivation of Fgfr2 or both Fgfr1/Fgfr2 caused a reduction in excitatory cortical neurons within the neocortex, with a stronger reduction in mPFC. The decrease in excitatory cortical neurons is attributable to FGFRs being required for upregulating the self-renewal of radial glial precursor cells in the VZ, particularly in anterior regions of the VZ. Neurons in mPFC were reduced in Fgfr2 cKO mice to an extent comparable to that of mice lacking both receptors, suggesting that Fgfr2 by itself plays the major role in mPFC development. Excitatory glutamatergic projections within the dorsal BST, a subcortical structure that receives axon terminals from the mPFC, were also significantly reduced in Fgfr2 cKO mice. Furthermore, both the dorsal BST and septum, which also receives projections from the mPFC, showed decreased numbers of inhibitory neurons, which can be attributed to impaired postnatal survival secondary to the lack of excitatory mPFC afferents.
Given that the hGFAP-Cre driver induces recombination of the Fgfr1 and Fgfr2 genes beginning at E13.5, after most of cortical layer VI is already generated, this mouse model allows us to assess the role of FGF signaling in the development of layers II-V. In accordance, we observed that Fgfr2 cKO and DKO mice have greater deficiencies in SMI-32+ neurons, which localize to many layers of cortex, than Tbr1+ neurons, which is primarily, albeit not exclusively, expressed by layer VI cells. Nevertheless, the lack of FGFR1 and FGFR2 did not affect the overall layer structure of the cerebral cortex.
Multiple analyses demonstrated a pronounced shrinkage of mPFC volume in Fgfr2 cKO as well as Fgfr1;Fgfr2 DKO mice. While a relatively modest, albeit significant, 35% decrease in SMI-32+ cells was observed in the overall neocortex of Fgfr2 cKO and Fgfr1;Fgfr2 DKO mice, the number of SMI-32+ excitatory neurons was more strongly reduced (50-60%) in the mPFC of these mutant animals as compared to littermate controls. Furthermore, Fgfr2 cKO and double KO mice had comparable deficits in SMI-32+ excitatory neurons in mPFC as well as neocortex, suggesting that defects in cortical pyramidal cells of layers V-II of these mutants are primarily attributable to the loss of Fgfr2. Equivalent defects in TBR1+ cells were found in the mPFC of both mutant lines, confirming that Fgfr2 contributes more to mPFC development. The more marked decrease in TBR1+ neurons in the neocortex of DKO mice suggested that Fgfr1 and Fgfr2 together are required for the development of TBR1+ neurons in more posterior cortical regions. Together with previous findings of midline deficits with earlier disruptions of Fgfr2 (Gutin et al 2006), these data implicate Fgfr2 uniquely in the growth of anteromedial cortex, whereas recent data implicate Fgfr3 in the growth of caudolateral (occipitotemporal) cortex (Thomson et al., 2009). The data also suggest that the FGF ligands, including FGF8 and FGF17, which are expressed in the rostromedial portion of the telencephalon (Bachler and Neubuser, 2001; MacArthur et al., 1995; Chellaiah et al., 1999) interact preferentially with FGFR2 to affect neurogenesis in mPFC.
The Fgfr2 cKO examined here starts during neurogenesis, after patterning of frontal and fronto-medial cortex involving FGF8 and FGF17 is thought to have already occurred (Fukuchi-Shimogori and Grove, 2001; Garel et al., 2003; Storm et al., 2006; Cholfin and Rubenstein, 2007). Our data suggest that FGF ligand/receptor interactions may continue to regulate cortical area formation by different,mechanisms at later stages of development (Hoch et al., 2009).(Hoch et al., 2009). We propose that FGF8 and FGF17 first specify the anterior cortical territory by acting on patterning genes and then, through interaction with FGFR2, expand this territory by promoting progenitor self-renewal and the addition of more excitatory neurons to anterior regions of the cortical plate. FGF receptors are required for self-renewal of radial glial cells in the VZ rather than SVZ precursors, since (1) proliferating VZ cells, but not SVZ intermediate precursors, were less likely to re-enter the cell cycle in Fgfr1;Fgf2 DKO mice; (2) DKO mice also demonstrated a disproportionate loss of apical mitoses, which involve primarily radial glial cells. A recent report (Kang et al., 2009) suggests that the loss of all three FGF receptors at E13.5 decreases radial glial cell self-renewal by increasing their differentiation into TBR2+ intermediate progenitors.
Previous results have suggested that Fgfr3 overexpression leads to an increase of TBR2+ cells in the SVZ of caudal cortical regions (Thomson et al., 2009). SVZ Tbr2+ cells are thought to be precursors for TBR1+ excitatory pyramidal cortical neurons (Pontious et al., 2008) and TBR2 expression is necessary for appropriate cortical thickness, surface area and neuron number (Sessa et al., 2008). Notwithstanding the primary cell renewal defect we found in the VZ, defects in mitotically active cells in the SVZ and TBR2+ intermediate precursors that we observed likely represent a deficit in progenitors contributed to SVZ from VZ. Entirely ruling out a specific defect in the SVZ will require fate mapping of SVZ cells.
The data presented here do not address the possibility that cell cycle length was affected by the conditional elimination of Fgfr2 and Fgfr1;Fgfr2. In our previous reports on FGF2, thought to bind and signal through FGFR1 and FGFR2, a decrease in the total population of proliferating progenitors but no cell cycle length alterations were observed (Vaccarino et al., 1999; Zheng et al., 2004), consistent with a recent report on hGFAP-Cre Fgfr1;Fgfr2 flox, Fgfr3 null mice (Kang et al., 2009).
Our data do not exclude that FGFRs may also be involved in the generation of the earliest contingent of TBR1+ excitatory neurons from the VZ, which are thought to migrate to the cortical plate without sojourning in the SVZ (Miyata et al., 2001); future experiments using earlier expressed Cre drivers are required to demonstrate this. A much earlier cKO of Fgfr2, produced using the FoxG1::Cre line, however, did not produce gross abnormalities, although mPFC or neocortical volume and neuron number were not quantitatively assessed; rather, these mutants exhibited an increase in the density of proliferative markers in the VZ (Ever et al., 2008). This may be explained both by differences in the methodology used for morphometric analyses and by functional redundancy of Fgfr2 with other Fgfr at this early stage of development.
In summary, the expression of Fgfr1 and Fgfr2 within radial glial cells of the telencephalic VZ (Hasegawa et al., 2004; Yoon et al., 2004; Gregg and Weiss, 2005; Smith et al., 2006) combined with the results presented here reinforce the idea that FGFR may induce cell cycle re-entry of Pax6+ cells in the VZ and therefore contribute to radial glia self-renewal rather than differentiation in the embryonic cerebral cortex. Although deficits in the density of TBR2+ at E16.5 (25%-40%) are similar to deficits in SMI-32+ cells at 3 months of age, a continuing role of these receptors on the postnatal maturation of the cortex cannot be excluded.
Deficiencies in subcortical inhibitory neurons were also observed in Fgfr2 cKO. The inhibitory neuron defect is likely not induced by cell autonomous means. Precursors for these neurons in mouse do not arise from dorsal telencephalic radial glia, but from the ventral ganglionic eminences, which are not targeted by the hGFAP-Cre transgene (Fig. 1) (Ohkubo et al 2004). Consistently, no differences in septal inhibitory neuron populations were evident at birth, suggesting that the prenatal development of these cells was likely not affected. Our observation that the density of inhibitory neurons in the BST and septum is correlated with TBR1 neuron number in mPFC and cortical white matter volume supports an indirect mechanism for the effects of FGF signaling, through appropriate cortical development and innervation of subcortical regions. The distinct deficit in glutamatergic terminals in the BST of mice lacking Fgfr2 also supports this hypothesis as a substantial proportion of non-GABAergic inputs to the BST arise from frontal cortex (Dong et al., 2001). Furthermore, NeuN+ cell counts confirmed that this decrease of inhibitory neurons was due to a loss of cells, not down-regulation of immunohistochemical marker expression. Together, the data suggest that the cells themselves are likely absent due to non-survival contributed in part by loss of afferents. Thus, the survival of inhibitory neuron phenotypes in subcortical stations of the limbic system may require FGF signaling in the mPFC. Similar processes have been shown to be involved in olfactory pathways and intra-hippocampal networks, in which appropriate cell survival and maturation depends on the normal development of afferent projections (Capurso et al., 1997; Marques-Mari et al., 2007). These findings demonstrate the crucial role of Fgfr2 for mPFC development, and indirectly, for the development of inhibitory neurons in subcortical regions.
Our findings provide a rationale for variations in Fgf gene transcripts that have been discovered in affective disorders (Evans et al., 2004; Gaughran et al., 2006) and suggest a cellular mechanism that may be involved in psychiatric illnesses. The disproportionate decrease in PFC growth found in Fgf2 cKO mice mimic phenotypes in human structural studies of schizophrenia and bipolar disorder (Selemon et al., 1998; Hirayasu et al., 1999; Ananth et al., 2002; MacDonald and Carter, 2003). The changes within subcortical limbic structures point to the crucial role of PFC in controlling both development and function of the limbic system which is likely to be altered in psychiatric illness (Hains and Arnsten, 2008).
The authors wish to acknowledge Shawna Ellis and Rosie Deegan for laboratory and animal care assistance and Yasushi Ohkubo and Brian Rash for their intellectual contribution to this work. We also thank Yuchio Yanagawa for the GAD67-GFP (delta-neo) mice. This work was supported by R01 MH067715, NARSAD, T32 MH018268, R25 MH071584, and R25 MH077823.