Glutamatergic axon-derived BDNF controls GABAergic synaptic differentiation in the cerebellum

To study mechanisms that regulate the construction of inhibitory circuits, we examined the role of brain-derived neurotrophic factor (BDNF) in the assembly of GABAergic inhibitory synapses in the mouse cerebellar cortex. We show that within the cerebellum, BDNF-expressing cells are restricted to the internal granular layer (IGL), but that the BDNF protein is present within mossy fibers which originate from cells located outside of the cerebellum. In contrast to deletion of TrkB, the cognate receptor for BDNF, deletion of Bdnf from cerebellar cell bodies alone did not perturb the localization of pre- or postsynaptic constituents at the GABAergic synapses formed by Golgi cell axons on granule cell dendrites within the IGL. Instead, we found that BDNF derived from excitatory mossy fiber endings controls their differentiation. Our findings thus indicate that cerebellar BDNF is derived primarily from excitatory neurons—precerebellar nuclei/spinal cord neurons that give rise to mossy fibers—and promotes GABAergic synapse formation as a result of release from axons. Thus, within the cerebellum the preferential localization of BDNF to axons enhances the specificity through which BDNF promotes GABAergic synaptic differentiation.


Results
Bdnf expression is restricted to the internal granular layer and the protein is localized on granule cell axons and mossy fiber endings. To begin to assess the involvement of BDNF signaling in regulating cerebellar GABAergic synapses, we first examined the expression of Bdnf mRNA and BDNF protein. Consistent with studies by others, we observed prominent expression of Bdnf, detected with a lacZ knock-in, in the internal granular layer of the cerebellum 15,32 (Fig. 1A). LacZ expression was not detected in the molecular layer, Purkinje cell layer or white matter layer (Fig. 1A). We used a previously described mouse monoclonal anti-BDNF antibody (Mab#9) directed against the mature domain of BDNF 31 to show that the BDNF protein is much more widely distributed with localization in the molecular layer, Purkinje cell layer and internal granular layer of the cerebellar cortex ( Fig. 1B-C), consistent with recent observations by others 14 . To confirm the specificity of the BDNF antibody, we used conventional Bdnf knockout mice as a negative control. As expected, BDNF protein appeared to be absent in the cerebella of mice lacking Bdnf (Fig. 1C).
To analyze the expression of BDNF protein in more detail, we compared the expression of BDNF with that of cell-type specific molecular markers using immunohistochemistry. We observed that within the internal granular layer, little or no BDNF localized to Golgi cell axons (Fig. 1D-F). In contrast, substantial BDNF was found on vGluT1 + mossy fiber terminals (Fig. G-I). Additionally, relatively little BDNF, but detectable amount, was found in the soma or dendrites of granule cells (Fig. 1J-L). Quantification of the area coverage of BDNF on specific cell compartments indicate that BDNF protein overlapped with less than 5% of the area of Golgi cell axons or granule cell dendrites/soma. In contrast, BDNF protein overlapped with ~43% of the area occupied by vGluT1 + mossy fiber terminals in the internal granular layer (Fig. 1M). Considering the 15-25 μ m thickness of the sections used for these studies, the low overlap of BDNF with Golgi cell axons or granule cell dendrites/soma may reflect presence of BDNF protein outside of these structures.
In order to confirm the localization of BDNF protein, electron micrographs of ultrathin sections of the internal granular layer were examined after labeling with an antibody against BDNF and visualization using secondary antibody-coated gold particles. The cerebellar glomeruli and major components of the glomeruli were identified based on their typical morphological characteristics previously described 9,33,34 . We found the highest density of BDNF within mossy fiber terminals (9.41 ± 1.1 particles/μ m 2 ) with a substantially lower density within Golgi cell axons (1.32 ± 0.31 particles/μ m 2 ) ( Fig. 2A-F). Only a low density (4.08 ± 0.73 particles/μ m 2 ) of gold particles was present within granule cell dendrites (Fig. 2B,F). These results are consistent with the results shown in Fig. 1. Together, the data indicate that BDNF is primarily localized within mossy fiber terminals with much lower presence within Golgi cell axons and granule cell dendrites. Since both granule and Golgi cells express TrkB 12 , this could also be a result of receptor binding and endocytosis. Our data also show that even though granule cells express the Bdnf gene, BDNF expressed within granule cells is mostly directed to their axons, not their dendrites (see also ref. 14). This is consistent with a preferential localization of BDNF to axons, but not dendrites of excitatory hippocampal and cortical neurons 6,30,31 . Finally, the most important source of BDNF protein in the IGL appears to be the axons and terminals of mossy fibers.
Consequences of loss of BDNF on the localization of cerebellar GABAergic synaptic proteins. We next examined the effects of the loss of BDNF on the localization of GABAergic synaptic proteins in the internal granular layer of the cerebellum. In P14 conventional Bdnf −/− mice, we observed an approximately 46% reduction in the localization of gephyrin (a postsynaptic marker for GABAergic synapses), but no impairment of the localization of GAD65 or GAD67 (presynaptic markers for GABAergic synapses) within the developing IGL ( In prior work, we demonstrated that absence of TrkB does not affect total expression of gephyrin within the cerebellum 9 . Consequently, the absence of BDNF, similar to that of TrkB, likely controls gephyrin localization at synapses, but not its total expression level. At this developmental stage (P14), the localization of GAD67 was also not detectably reduced in mice lacking TrkB in the cerebellum 9 . As the conventional Bdnf −/− mice die around 2 weeks after birth, we have not examined the cerebella of older mice. We have, however, deleted Bdnf more specifically within cerebellar precursors, using Wnt1::Cre 35 and were surprised to observe that these mice had normal localization of GAD67 and gephyrin at inhibitory synapses in the IGL (Fig. 3G-L). The apposition of GAD67 with gephyrin ( Fig. 3M-R) also remained comparable in mice lacking BDNF compared to controls. Similar results were obtained following deletion of Bdnf specifically in granule cells using mα6::Cre (data not shown). These observations suggest that within the IGL, there is either compensation for BDNF due to presence of other neurotrophins with the ability to activate TrkB (e.g. NT3, NT4) or that BDNF is supplied by cells located outside the cerebellum.
Mossy fiber-derived BDNF regulates localization of GABAergic synaptic proteins in the internal granular layer. In addition to granule cells, BDNF is expressed by neurons located outside the cerebellum that give rise to mossy fibers 36 . To explore the possibility that mossy fibers are the source of BDNF that controls gephyrin localization at Golgi cell-granule cell GABAergic synapses, we used mice that express Cre from the endogenous Shh locus 37 , which mediates recombination in neurons that give rise to a subset of cerebellar mossy fiber endings (Fig. 4A). YFP expression from the ROSA::YFP reporter was used to identify mossy fibers from neurons that have been recombined. Only 26 ± 3% of the mossy fibers in the IGL were recombined by Shh::Cre mice ( Fig. 4A; data not shown). Mossy fibers terminals were recombined in lobules I-V, but not in lobules VI-IX ( Fig. 4B-E). As expected, immunohistochemical analysis using anti-BDNF antibody showed that Shh::Cre efficiently deletes the Bdnf as well as the ROSA loci in mossy fibers ( Fig. 4F-I).
Results showed that deletion of Bdnf from YFP + mossy fibers did not reduce the localization of GAD67 at GABAergic synapses, but did result in reduced synaptic localization of gephyrin in P21 Shh::Cre; Bdnf fl/ fl ; ROSA::YFP mice ( Fig. 4A-F). The mean area ratio of GAD67:vGluT1 expression was comparable between Shh::Cre; Bdnf fl/fl ; ROSA::YFP mice and controls (Fig. 4C), but the mean area ratio of gephyrin:vGluT1 was reduced by ~54% in Shh::Cre; Bdnf fl/fl ; ROSA::YFP mice compared to control (Fig. 4F). The requirement for BDNF is spatially restricted, as gephyrin localization is normal around mossy fibers originating from neurons that have not been recombined ( Fig. 4G-L).
Interestingly, the expression of the vesicular glutamate transporter vGluT1 was also reduced within mossy fibers lacking BDNF ( Fig. 5J-K) compared to control ( Fig. 5G-H). We did not pursue this observation further to determine whether it is a cell-autonomous effect or whether it reflects alterations of innervations of the cell soma and dendrites of these neurons in the pons and hindbrain.
To explore the impact of mossy fibers on GABAergic synapse formation in the IGL, we examined cerebellar slices cultured from P14-21 in vitro and compared these to P21 slices acutely prepared from 5 mice. Not surprisingly, the deafferented cerebellar slices cultured for 7 days in vitro (DIV) exhibited significantly reduced vGluT1 + expression compared to their age-matched controls ( Fig. 6A-D,F-I). The localization of GAD67 surrounding granule cell dendrites was comparable between the cultured and control slices, but the localization of gephyrin was reduced by ~69% in cultured slices compared to the P21 controls ( Fig. 6A-J). Together, these findings indicate that the presence of mossy fibers is required to direct localization of the postsynaptic scaffold protein, gephyrin, presumably because the mossy fibers are required for the delivery of BDNF.

Addition of recombinant BDNF in primary cerebellar granule cells promotes clustering of key postsynaptic GABAergic proteins.
To explore potential effector proteins that could mediate the activities of BDNF in postsynaptic differentiation, we assessed the consequences of adding recombinant human BDNF to cultured cerebellar granule cells. We previously showed that the localization of a cell adhesion molecule, Contactin-1, in the cerebellum is dependent on TrkB 9 and others have shown that BDNF controls the translocation of Contactin-1 to the cell surface of hippocampal neurons 38 . We first examined the expression of a key postsynaptic GABAergic protein, the α 6 subunit of GABA A receptors in cerebellar granule cells obtained from P8 pups cultured for 8 days in vitro. We treated the cultured neurons with 20 ng of recombinant BDNF for 5 days starting at 3 days in vitro. We observed a ~2.1-fold increase in the expression of GABA A R α 6 subunit in granule cells treated with BDNF along with an 8% increase in intensity of the protein expression ( Fig. 7A-J). We next examined the expression of Contactin-1 and observed a ~1.5-fold increase in Contactin-1 on MAP2 + portions of the granule cells and a ~2.4-fold increase in Contactin-1 on Tau + portions of the granule cells in cultures treated with BDNF ( Fig. 7K-T). Consistent with this finding, our previous work showed that loss of Contactin-1 results in a reduction in the localization postsynaptic GABAergic proteins in the internal granular layer 9 . Taken together, we provide evidence that BDNF promotes the localization of key GABAergic postsynaptic proteins and supports the importance of Contactin-1 in the assembly and maintenance of cerebellar GABAergic synapses .

Discussion
Interference with BDNF expression results in defects in the development and function of GABAergic synapses 23,39,40 . Despite well documented roles of granule cell-expressed BDNF for migration of granule cell precursors and organization of synaptic vesicles within glutamatergic parallel fiber-GABAergic Purkinje cell dendrite synapses 15,41 , we found that deletion of Bdnf from granule cells or from all cells resident within the cerebellum did not impair the localization of GABAergic synaptic proteins within the internal granular layer (summarized in Figure 8). Interestingly, we found that loss of BDNF from a subset of mossy fibers resulted in reduced synaptic localization of gephyrin without influencing that of GAD65/67 (Figure 8). Since Golgi cells innervate multiple glomeruli 33 , GAD65/67 localization may be unaffected because the Golgi cells receive trophic BDNF support from adjacent glomeruli where mossy fibers express BDNF. Since Bdnf was only deleted from ~26% of mossy fiber terminals with the Shh::Cre used in the present work, it seems quite unlikely that individual Golgi cells were completely deprived of mossy fiber-derived BDNF. Thus, BDNF may control postsynaptic differentiation through local activation of TrkB, while BDNF controls presynaptic differentiation through more global mechanisms. Consistent with prior observations, our data indicate that granule cells are the major, probably the only cell type within the cerebellar cortex that express BDNF. Additionally, similar to a recent report 14 , we found that BDNF is preferentially localized within axons, but not dendrites of granule cells. Granule cell axon BDNF controls the apposition of pre-and postsynaptic specializations formed by GABAergic stellate and basket cells with Purkinje cell dendrites in the molecular layer. This observation is consistent with our prior work that demonstrated an important role for TrkB in promoting GABAergic synapse formation in the molecular layer 12 .
Several mechanisms could explain the specificity of granule cell-versus mossy fiber-derived BDNF in organizing cerebellar circuits. BDNF localization is controlled by mRNA transport, protein incorporation into dense core vesicles, and dense core vesicle transport and localization 14,18,20 . Its secretion is regulated by sortilin, CAPS1/2 and other proteins that control its incorporation into dense core vesicles, proteases that regulate proBDNF processing, and proteins, such as the syntaxins, CAPS2 and the synaptotagmins that regulate dense core vesicle exocytosis 5,14,19,20,24,40 . Processing of proBDNF to matureBDNF is essential for TrkB activation, because proBDNF activates only p75NTR, not TrkB, resulting in distinct consequences such as axon pruning and LTD 42,43 . Of special interest, recent work suggests that CAPS2 is preferentially localized to axons and is essential for axonal localization and normal exocytosis of BDNF 14 . Deletion of exon 3 within the p150Glued interaction domain of CAPS2 prevents localization of CAPS2 and BDNF to axons and results in a deficit in BDNF secretion 14 . CAPS2 transcripts lacking exon 3 have been identified in a few autistic patients and mice expressing CAPS2 lacking this exon exhibit autistic phenotypes, raising the interesting possibility that the specificity with which BDNF organizes CNS circuits described in this paper is important for organizing neural circuits elsewhere within the CNS and for expression of human cognitive and social behaviors that are abnormal in autism 14,44 .
BDNF controls the postsynaptic differentiation of excitatory synapses in the CNS by promoting dendritic spine formation in various brain regions such as the cortex, hippocampus and striatum 4,8,45 . Since BDNF is secreted as a consequence of neuronal activity, BDNF secretion is an important mechanism for controlling morphogenesis of dendritic spines in response to synaptic activity 46 . Data in the present work show that BDNF secreted by excitatory neurons additionally controls inhibitory synapse formation through regulation of gephyrin localization. While we did not examine the impact of BDNF deficiency on the total expression level of gephyrin, in prior work we demonstrated that absence of its receptor, TrkB, did not result in reduced expression of gephyrin within the cerebellum 9 . Thus we believe that the major effect of BDNF is to control gephyrin localization, not expression level in the cerebellum. In prior work, we have shown that activation of the BDNF receptor TrkB is important for both pre-and postsynaptic differentiation and maintenance of GABAergic synapses in the cerebellum 9,12 . In the IGL, the TrkB ligand, BDNF, secreted by mossy fibers regulates the clustering of gephyrin on  A,B). However, the expression of GABA A R α 6 subunit is significantly increased after the treatment of BDNF for 5 days in culture (H, Ii, Iii) compared to control (C, Di, Dii). (E) Quantification of the area coverage of GABA A R α 6 subunit over total vGluT1 in control (0.23 ± 0.13) and in culture after BDNF treatment (0.49 ± 0.17, p < 0.001). (J) Quantification of the intensity of GABA A R α 6 subunit in control (155 ± 13) and in culture after BDNF treatment (168 ± 9.5, p < 0.01). (K-T) The expression of Contactin-1 on granule cell dendrites (MAP2, green, K, P) and axons (Tau, blue, K, P) was assessed following BDNF treatment. The expression of MAP2 is not noticeably different after the addition of BDNF (Q) compared to control (L), though the expression of Tau appears more punctate after the addition of BDNF (R) compared to control (M). However, the expression of Contactin-1 on the dendrites and axons of granule cells is significantly increased following BDNF treatment (S) compared to control (N). (O) Quantification of the area coverage of Contactin-1 over total MAP2 in control (0.61 ± 0.19) and in culture after BDNF treatment (0.91 ± 0.23, p < 0.001). (T) Quantification of the area coverage of Contactin-1 over total Tau in control (0.39 ± 0.15) and in culture after BDNF treatment (0.92 ± 0.28, p < 0.001). Scale bar = 50 μ m. granule cell dendrites; and in the molecular layer, BDNF secreted from granule cells regulates gephyrin clustering on Purkinje cell dendrites. In each instance, BDNF secreted by one neuron acts to promote GABAergic synaptic contact formation between two separate populations of cells. One potential molecular mechanism by which BDNF promotes gephyrin clustering could be through interactions of BDNF and Wnt signaling. Recent in vitro studies have shown that BDNF promotes the expression of Wnt2 which in turn promotes cortical dendritic growth and spine formation 47 . Both cerebellar granule cells and Purkinje cells express Wnt family members which could potentially work synergistically with BDNF to control GABAergic synapse formation 24 .
Postsynaptic scaffolding proteins such as gephyrin and PSD-95 organize neurotransmitter receptors to regulate the synaptic stabilization and homeostasis 48,49 . Prior work has shown that BDNF regulates PSD-95 at excitatory synapses through several pathways activated through TrkB, PI3Kinase-Akt, phospholipase-Cγ and MAPK/ ERK signaling 50 , which control PSD95 palmitoylation and phosphorylation 50 . In prior studies, we obtained evidence that the PLC-γ binding site on TrkB is required for recruitment of gephyrin to GABAergic postsynaptic sites along granule cell dendrites 9 , implicating calcium-dependent pathways as effectors of TrkB action. Consistent with this, more recent and definitive work implicates calcium-calmodulin-dependent protein kinase II in phosphorylation-dependent regulation of gephyrin clustering 51 . Gephyrin localization is also controlled by collybistin, a phosphatidylinositol-3 phosphate binding protein with Rho family GEF activity, which is activated by binding to Neuroligin-2 as well as by the monomeric Cdc42 homologue CT-10/RhoQ 52 . Since phosphatidylinositide signaling and Cdc42 are activated by BDNF 53 , it seems likely that BDNF-TrkB control inhibitory synapse formation and maintenance through pathways in addition to those downstream of PLC-γ and calcium 52 . ; goat anti-β -galactosidase (1:1000; AbD Serotec). We found that the staining pattern for wildtype and experimental animals for all antibodies used in our studies did not differ between male or female.

Methods
For tissue staining with anti-mature BDNF (Mab#9) (Figs 1 and 4), brain tissues were immersed in an antigen retrieval solution (10 mM sodium citrate buffer, pH 6.0) at 4°C overnight. The tissues were then immersed in boiling antigen retrieval solution (200-500 ml) with stirring using a hot plate for 3-5 min. The tissues were then placed in cold 30% sucrose in PBS and treated with standard immunohistochemistry procedure.

Quantification of synaptic protein expression and synaptic protein contacts.
To determine the area and average pixel intensity of the expression of synaptic proteins surrounding individual mossy fiber endings within the IGL in lobules I-IV of the cerebellum 12 , images of a fixed area in the IGL from different mice were collected on a Zeiss LSM 5 Pascal confocal microscope (63x and 100x objectives, n.a. 1.40, Plan-Apochromat), keeping individual pixel intensities in the linear range for control images. The mean area (sum of pixels above threshold) and fluorescent pixel intensity of the protein of interest surrounding an individual glomerulus within each analyzed region were determined on cryosections at sub-saturating antibody concentrations and calculated using the histogram and color histogram function in NIH ImageJ (version 1.38x; NIH, USA).
For analysis shown in Figs 3 and 7, the sum of pixels above threshold was measured in an area of 21374 μ m 2 (63x objective) for each protein of interest and represented as an absolute number as previously described (Rico et al. 2002). For all other analysis, the sum of pixels above threshold was normalized to the sum of pixels of vGluT1 expression in an area of 8600 μ m 2 (100x objective) for each protein of interest and represented as an area ratio. The threshold was determined by measuring the average pixel intensity of an area in the cerebellar cortex that does not express the protein of interest. For analysis of the colocalized expression of BDNF (Fig. 1), colocalization of BDNF on individual cell-type is measured by using the RG2B Colocalization macro in ImageJ and BDNF expression not localized on cell type-specific markers was omitted using the same macro (http://rsbweb.nih.gov/ ij/plugins/rg2bcolocalization.html).
For analysis shown in Figs 3 and 7, the number of contacts between synaptic proteins were calculated using a macro previously described 59 . This macro analyzes aspects of synapses by identifying apposing pre-and postsynaptic puncta of sufficient size and signal intensity, and objectively processes images with the same settings. Cerebellar slice culture. Organotypic cerebellar slices were obtained from P14 mice of either sex and cultured for 1 week based on previously described methods 60 . Brain dissection and slice preparation were carried out in low sodium Artificial Cerebrospinal Fluid (ACSF) containing 1 mM calcium chloride, 10 mM D-Glucose, 4 mM potassium chloride, 5 mM magnesium chloride, 26 mM sodium bicarbonate, 246 mM sucrose and phenol red solution (1:1000), ph 7.3. The solution was sterilized by filtering through a 0.22 μ m filter and stored at 4 °C. Cerebellar slices were cultured in 75% Minimum Essential Medium Eagle (MEM), 25% heat-inactivated horse serum, 25 mM HEPES, 1 mM glutamine, 5 mg/ml glucose, penicillin (100 U/ml) and streptomycin (100 U/ ml). After vibratome sectioning (Leica), the cerebellar slices were transferred to 6-well plates containing 30 mM culture plate inserts with 0.4 μ m pores. 1 ml of culture media was added per well which was pre-conditioned by incubation for at least 2 hours in the cell culture incubator at 37 °C and 5% CO 2 .
Mixed cerebellar neuronal culture. Cerebella were dissected from P5 mice of either sex in cold CMF-PBS, then mechanically minced into smaller pieces. Intact tissue was collected in 15 ml polypropylene tube containing 2.5 ml cold CMF-PBS. 250 μ l of 10X trypsin and 10X DNAse were added to digest the cerebella for 10-15 min. at 37 °C. The cells were centrifuged at 800 x RCF at 4 °C for 5 minutes to remove the trypsin. Cells were resuspended in 2 ml CMF-PBS (with 200 μ l 10X DNAse). The cells were triturated with P1000 tips and the total volume was brought up to 5 ml. Cells were filtered (70 μ m nylon cell strainer) into 50 ml falcon tube and centrifuged at 900 x RCF at 4 °C for 5 minutes to remove debris. Cells were then resuspended in 5 ml medium (+ serum). Cells were plated on glass coverslips pretreated with poly-D-lysine/poly-L-lysine at a density of 1 × 10 6 cells/ml. Medium was changed to serum-free medium next day. Half the medium was changed every 2-3 days. Recombinant human BDNF was generously provided by Amgen (Thousand Oaks, CA).
Analysis and quantification of immunogold labeling with electron microscopy. P60 wildtype mice of either sex were perfused and postfixed overnight using 4% paraformaldehyde. Cerebellar tissues were processed as previously described 9,12 . We analyzed individual glomeruli from 20 electron micrographs (magnification of 11,500X, single 2 μ m sections). Only simple glomeruli were analyzed since large complex glomeruli containing more than one mossy fiber ending extended outside of the micrographs. Individual components of each glomerulus were identified by criteria described previously 9 using those published by others 33,34 .