BDNF and its pro-peptide are stored in presynaptic dense core vesicles in brain neurons

Polarized cells use well-conserved mechanisms to sort proteins into specific compartments (Mellman and Nelson, 2008), providing them with a directionality that is critically important for both their function and meaningful integration into tissues. Central nervous system (CNS) neurons are prototypically polarized cells with specialized axonal and dendritic compartments that play essential roles in intercellular signaling. Although neurons typically communicate by releasing low–molecular weight neurotransmitters accumulated in synaptic vesicles, they also store and release peptides or small proteins such as brain-derived neurotrophic factor (BDNF), a member of the neurotrophin family. BDNF is known to regulate a wide variety of brain functions in humans, ranging from food intake (Gray et al., 2006) to memory retention (Egan et al., 2003). Indeed, a single amino acid replacement in pro-BDNF has been convincingly shown to correlate with a diminished capacity to remember words and sentences (Egan et al., 2003; Cathomas et al., 2010). Furthermore, in animal models of disease, reduced levels of BDNF have been associated with several conditions, including depression (Calabrese et al., 2007), Rett syndrome (Chang et al., 2006), and Huntington’s disease (Zuccato et al., 2010).

Although many aspects of BDNF biology in the adult brain are thus beginning to be well appreciated, the subcellular localization of this secreted protein in neurons of the adult CNS is still very unclear, in large part because of the very low levels of endogenous BDNF. To address this question, several studies have resorted to neuronal cultures prepared from the embryonic rodent hippocampus (Goodman et al., 1996; Hartmann et al., 2001; Kojima et al., 2001; Egan et al., 2003; Adachi et al., 2005; Dean et al., 2009; Matsuda et al., 2009; Jakawich et al., 2010). In most cases, conclusions about the localization of BDNF were inferred from visualization experiments using transfected tagged BDNF constructs, with recent studies concluding that BDNF is transported in and released from both axons and dendrites (Adachi et al., 2005; Dean et al., 2009; Matsuda et al., 2009; Jakawich et al., 2010). As firmly established by in situ hybridization studies, the Bdnf gene is expressed in an activity-dependent fashion by numerous excitatory neurons (Zafra et al., 1990), with protein levels increasing by ∼10-fold during the first 3 wk after birth, in parallel with the development of synaptic activity (Tao et al., 1998; Kolbeck et al., 1999). Nonetheless, BDNF remains, even in the adult brain, an extremely rare protein, making its unambiguous detection in vivo a challenging task that is further complicated by the early death of Bdnf-null mutant animals, thus precluding their use as age-matched controls (CONs; Ernfors et al., 1994; Jones et al., 1994).

Given that an understanding of the mode of action of secreted proteins depends on a detailed knowledge of their subcellular localization, the present study investigates the distribution of BDNF in the adult hippocampus at both light microscopic and ultrastructural levels using three lines of transgenic animals: (1) a BDNF knockin line allowing the use of antibodies directed against a tagged version of the BDNF gene (Matsumoto et al., 2008), (2) a line conditionally lacking BDNF in adult neurons allowing for the specificity control of BDNF antibodies (Rauskolb et al., 2010), and (3) a mutant line displaying seizure episodes that are accompanied by markedly elevated levels of BDNF (Heyden et al., 2011). Antibodies against the BDNF prodomain were also used as an independent means of localizing BDNF as well as to generate information about the status of its cleaved prodomain (described hereon as the pro-peptide).

Hippocampal sections prepared from 8-wk-old animals were incubated either with the monoclonal BDNF antibody Mab#9 (anti-BDNF; Fig. 1 B) or with polyclonal antibodies recognizing the BDNF prodomain (anti–pro-BDNF; Fig. 1 H). Tissues from age-matched mice engineered to delete BDNF from neurons (cbdnf ko; Rauskolb et al., 2010) were used as a negative CON (Fig. 1, C, D, I, and J). In addition, sections from a knockin mouse line expressing Bdnf-Myc (Fig. 1 E; Matsumoto et al., 2008) were incubated with Myc antibodies (anti-Myc), with wild-type (WT) tissue used as a CON (Fig. 1, F and G). These three unrelated antibodies yielded strikingly similar staining patterns (Fig. 1, B, E, and H). In particular, BDNF-, pro-BDNF–, and Myc-immunoreactivity (IR) were each most prominently distributed in the cell bodies and axon terminals of the mossy fiber projection pathway, whereas the layers comprising the CA1 area were only weakly stained, especially in the septal hippocampus.

Low-power examination of BDNF-IR in the dentate gyrus (DG) revealed a subset of immunopositive granule cells in the supra- and infrapyramidal blades (Fig. 2 A). BDNF-IR varied in intensity among the labeled cells, with staining concentrated in the somal apex (Fig. 2 A, inset). Anti–pro-BDNF staining was confined to exactly the same subset of granule cells containing BDNF-IR (Fig. 2, B and C), with somal pro-BDNF–IR also concentrated at the apex (Fig. 2 B, inset). In addition, the hilar region, which contains mossy fiber collateral axons of the granule cells, was intensely stained (Fig. 2, A–C). In the CA3 region, a subset of pyramidal neurons also showed both BDNF-IR (Fig. 2, D and F) and pro-BDNF–IR (Fig. 2, E and F). High-resolution examination of sections labeled with anti-Myc and anti–pro-BDNF, along with antibodies against the Golgi matrix protein GM130, revealed Myc-IR throughout the soma and the initial dendritic segments (Fig. 2 G), whereas pro-BDNF–IR showed a similar, albeit more punctate, distribution (Fig. 2 H). Comparison with GM130-IR showed pro-BDNF–positive puncta closely associated with the Golgi apparatus (Fig. 2, I and J). In line with this, anti–pro-BDNF immunogold labeling localized the protein to the Golgi complex of CA3 somata (Fig. 2 K). In CA1, BDNF and pro-BDNF costaining was also detected in a small number of pyramidal neurons in temporal hippocampus sections (unpublished data).

The granule cells give rise to mossy fiber axons, whose targets include the complex spines on proximal dendrites of CA3 neurons. Mossy fibers project through and terminate in stratum lucidum (SL) and are characterized by prominent specialized endings known as mossy fiber boutons (MFBs). Accordingly, strong BDNF-IR and pro-BDNF–IR were observed within SL (Figs. 1 [B, E, and H] and 2 [D–F]). Using high-resolution confocal microscopy, both BDNF-IR and pro-BDNF–IR were found to be colocalized to the same subset of MFBs (Fig. 3, A–C).

Additional markers were then applied to identify the type of vesicles containing BDNF-IR and to compare the distribution of BDNF with other peptides known to be anterogradely transported by granule cells. As expected, BDNF-IR did not colocalize with the synaptic vesicle markers synaptophysin (SYP; Fig. 3, D and E) or VGLUT-1 (Fig. 3, F and G). We then tested possible colocalization with Met-enkephalin (Met-enk), an opioid peptide also derived from a larger precursor protein and stored in dense core vesicles (DCVs; Cheng et al., 1995). In a small proportion of granule cells and their axons, Met-enk–IR was detected throughout the soma and initial dendritic segment (Fig. 3 I). Although Met-enk–positive granule cells invariably coexpressed BDNF-IR (Fig. 3 H), the immunoreactive signals of these two precursor-derived molecules remained separate, suggesting that they do not reside together in the same secretory vesicles (Fig. 3, J and K–M). Similar conclusions were reached with cholecystokinin (CCK), a neuropeptide transported along the mossy fiber projection pathway of the ventral mouse hippocampus (Gall et al., 1986). Double labeling with anti-CCK and -BDNF revealed complete segregation of the two peptides within MFBs (Fig. 3, N–P).

Presynaptic BDNF labeling was also observed in the CA1 region, whereby in stratum radiatum (SR), thin varicose processes labeled with anti-BDNF (Fig. 3, Q–S) and anti–pro-BDNF (not depicted) were sparsely distributed throughout the neuropil, likely corresponding to presynaptic Schaffer collateral axons originating from BDNF-positive CA3 neurons. Importantly, BDNF-IR did not colocalize with the postsynaptic marker synaptopodin (synpo; Fig. 3, R and S).

Ultrathin sections of SL were then examined at 2,000-fold magnification after immunogold labeling with BDNF and pro-BDNF antibodies. MFBs were identified based on their typical morphological characteristics, namely a high density of synaptic vesicles, numerous synaptic contacts with CA3 complex spines, nonsynaptic puncta adherentia at dendritic shafts, and a relatively large surface area. A subset of MFB profiles labeled with anti-BDNF (Fig. 4 A), anti-Myc (not depicted), or anti–pro-BDNF (Fig. 4 B) contained distinct aggregates of gold grains; at 10,000-fold magnification, these gold clusters were found to be associated with large vesicles encompassed by an electron-dense membrane. Although the vesicles were sometimes masked by gold grains, fortuitous grain distribution occasionally revealed an electron-dense core (Fig. 4, A and B, insets). Within MFB profiles of cbdnf ko sections, gold grains were never specifically associated with any type of organelle (Fig. 4, C and D). To assess the density of immunogold particles in MFB profiles, gold clusters and single gold grains were quantified in anti-BDNF and anti–pro-BDNF–labeled tissues and compared between sections from pooled WT/Bdnf-Myc mice and cbdnf ko mice. Anti-BDNF–labeled MFB profiles showed a mean density of 2.67 ± 0.35 clusters/µm² in WT/Bdnf-Myc animals compared with 0.38 ± 0.12 clusters/µm² (P < 0.005) in cbdnf ko profiles (Fig. 4 E). Moreover, single gold grain densities were significantly reduced in cbdnf ko MFB profiles, with 5.63 ± 0.94 grains/µm² compared with 21.99 ± 2.57 grains/µm² in WT/Bdnf-Myc (P < 0.005; Fig. 4 F). Similarly, pro-BDNF immunogold labeling showed a mean density of 2.01 ± 0.43 clusters/µm² and 27.44 ± 8.21 single grains/µm² in MFB profiles from WT/Bdnf-Myc animals, whereas in comparison, MFB profiles from cbdnf ko animals showed significant reductions in cluster (0.14 ± 0.10/µm²; P < 0.05; Fig. 4 E) and single grain (5.82 ± 2.06/µm²; P < 0.05; Fig. 4 F) densities, respectively.

Although BDNF immunogold labeling was mostly concentrated in presynaptic terminals, labeled vesicles were also occasionally observed within unmyelinated axon segments in SL (Fig. 5 A), which in fortuitous sections could be seen to give rise to giant MFBs (Fig. 5 B).

Ultrathin sections of SR (CA1) labeled with anti-BDNF and anti–pro-BDNF immunogold were also examined, and, as expected, large cluster-labeled secretory vesicles were observed within small axon terminals (Fig. 5 C). Depending on the proximal distal level of SR, these infrequent labeled boutons likely correspond to Schaffer collateral terminals or entorhinal terminals. No such labeled terminals were observed in cbdnf ko sections (Fig. 5 D).

Next, we examined the possible localization of BDNF in dendrites by analyzing sections double labeled with anti-Myc and anti–microtubule-associated protein–2 (MAP-2). Both in BDNF-positive granule cells and CA3 neurons, Myc-IR only extended as far as the initial dendritic segments, and there was no evidence of colocalization in SL (Fig. 6, A–C). Sections were alternatively labeled with antibodies to Arc/Arg3.1, an immediate early gene product up-regulated in somata and dendrites during elevated synaptic activity (Lyford et al., 1995). Confocal scanning revealed that the majority of granule cells labeled with BDNF-IR and pro-BDNF–IR also expressed Arc/Arg3.1-IR (Fig. 6, D–G). Although Arc/Arg3.1-IR was seen throughout the soma and dendritic arbor, strong coexpression of BDNF-IR and pro-BDNF–IR was confined to the cell soma (Fig. 6, D–G).

In ultrathin sections prelabeled either with anti-BDNF or anti–pro-BDNF immunogold, gold grains were sparsely distributed within dendritic profiles (Fig. 6, H and K). When profiles were thoroughly scrutinized for specifically labeled vesicles or endosomes, none was found to be stained above the background levels observed in dendrites from cbdnf ko sections (Fig. 6, I and L). Gold grain quantification of anti-BDNF–labeled sections (Fig. 6 J) revealed a mean density of 3.56 ± 0.45 grains/µm² in WT/Bdnf-Myc mice versus 4.72 ± 0.91 grains/µm² in cbdnf ko mice (P = 0.81). Similarly, in anti–pro-BDNF–stained sections, mean densities were comparable between WT/Bdnf-Myc mice (4.69 ± 1.09 grains/µm²) and cbdnf ko mice (4.08 ± 2.02 grains/µm²; P = 0.31; Fig. 6 M).

Next, we examined the hippocampus of mice lacking the presynaptic protein Bassoon, as these mutants develop episodic generalized seizures (Altrock et al., 2003) and have enlarged cortices and hippocampi (Angenstein et al., 2007). Concurrent with the development of seizures, BDNF protein levels become significantly higher than those measured in adult CON littermates (see Fig. 10 C; Heyden et al., 2011). Whereas Bassoon (Bsn) mutants showed the typical distribution pattern of BDNF-IR and pro-BDNF–IR, a dramatic increase in staining intensity largely confined to the neuropil was observed (compare CON in Fig. 7 [A–D] with Bsn in Fig. 7 [E–H]). In contrast, granule cell bodies from Bsn mutants did not show increases in anti-BDNF (Fig. 7 I) or anti–pro-BDNF (Fig. 7 J) staining intensities, although a much higher proportion of cells was labeled in comparison with CON tissues. Granule cell dendrites in the molecular layer remained unlabeled in Bsn mutants, whereas in the CA3 region, the stark increase in BDNF-IR (Fig. 7 G) and pro-BDNF–IR (Fig. 7 H) was confined to SL. Closer examination revealed intense presynaptic labeling in MFB profiles (Fig. 7, K and L), which was confirmed by a lack of colocalization with the postsynaptic markers synpo (Fig. 7 K) and MAP-2 (Fig. 7 L), respectively.

Enhanced BDNF staining in Bsn mutants was not only confined to the granule cell–CA3 projection pathway. In the CA1 region, where BDNF-IR can usually only be detected at high magnification (compare Fig. 1 with Figs. 3 Q and 5 D), a conspicuous band of punctate BDNF-IR and pro-BDNF–IR was observed at the border of SR and stratum lacunosum–moleculare (SL-M; Fig. S1, A–C), corresponding to a region known to harbor fibers from the entorhinal cortex (Amaral and Lavanex, 2006). Importantly, these increased signals colocalized (Fig. S1 D), showing no overlap with either synpo-IR (Fig. S1 E) or glial fibrillary acidic protein (GFAP)–IR (Fig. S1 F), which label dendritic spines and reactive astrocytes, respectively. This suggests that entorhinal neurons also represent a possible presynaptic source of BDNF for CA1 neurons.

Immunogold-labeled sections from both WT and Bsn mutants were then examined at 2,000-fold magnification. In comparison with CON animals (Fig. 8, A and B), a higher number of labeled MFB profiles containing more BDNF-positive DCVs was observed in sections from Bsn mutants (Fig. 8, C and D), with clusters accumulated at the synaptic membrane. This was confirmed by quantification, with anti-BDNF–labeled MFBs containing a significantly higher density of gold clusters (1.7 ± 0.3 clusters/µm²; Fig. 8 E) and grains (14.4 ± 2.5 grains/µm²; Fig. 8 F) compared with CONs (0.8 ± 0.01 clusters/µm² with P < 0.05 and 6.9 ± 0.5 grains/µm² with P < 0.05, respectively). Similarly, anti–pro-BDNF–labeled profiles from Bsn mutants also displayed a higher density of gold clusters (2.02 ± 0.27 clusters/µm²; Fig. 8 E) and grains (17.2 ± 2.8 grains/µm²; Fig. 8 F) compared with CON profiles (0.96 ± 0.04 clusters/µm² with P < 0.005 and 10.1 ± 0.9 grains/µm² with P < 0.05, respectively); these relatively low increases in mean cluster densities are a result of the larger areas of MFB profiles in Bsn mutants.

Gold grain distribution and density in dendrites were also compared between Bsn mutant (Fig. 9, C and D) and CON tissues (Fig. 9, A and B). Dendritic profiles were again scrutinized for evidence of labeled vesicles, but none was found in tissues from either group. Density measurements revealed background values in Bsn mutant mice similar to those observed in CON mice, both in tissues labeled with anti-BDNF (5.4 ± 0.4 grains/µm² for Bsn mutant vs. 5.7 ± 0.5 grains/µm² for CON; P = 0.66; Fig. 9 E, left) and anti–pro-BDNF (5.1 ± 0.1 grains/µm² for Bsn mutant vs. 6.3 ± 0.3 grains/µm² for CON; P = 0.12; Fig. 9 E, right). Increased labeling density was not detected in the extracellular space or in nonneuronal cell types such as astrocytes.

To determine whether background gold labeling is evenly distributed over different subcellular compartments, we extended the quantitative gold grain analysis to dendritic spine profiles and myelinated axon profiles in SL, as they are devoid of cluster-labeled organelles. Quantification of single gold grains overlying these profiles revealed that the density of background gold labeling depends on the type of subcellular compartment (Table 1). The mean values for gold grain densities in spine and myelinated axon profiles did not differ between WT/Myc versus cbdnf ko or WT versus Bsn mutant tissues. Therefore, spines and myelinated axons from WT mice do not exhibit anti-BDNF and anti–pro-BDNF immunogold labeling above the background levels observed in cbdnf ko tissues, nor do they contain specifically labeled organelles.

The identity of the molecules recognized by the BDNF antibodies in the immunochemistry experiments was then determined using hippocampal lysates from WT and mutant mice (cbdnf ko and Bsn). Immunoprecipitates were analyzed by Western blotting (WB) and probed with either the BDNF polyclonal antibody N-20 or the monoclonal antibody 5H8, which recognizes an epitope in the prodomain of BDNF (Fig. 1 A). These experiments revealed the presence not only of pro-BDNF (∼32 kD) but also of the much more abundant pro-peptide (∼17 kD; Fig. 10 A; also see Fig. 1 A). Both signals were absent in lysates from cbdnf ko animals (Fig. 10 A). Quantification of the corresponding signal intensities revealed that the ratio of pro-peptide versus pro-BDNF is ∼10.3 ± 2.0 (n = 3; Fig. 10 B), similar to that of BDNF versus pro-BDNF (11 ± 2.0; n = 3; Fig. 10, B and C). A similar analysis of hippocampal lysates from Bsn mutants confirmed an approximate threefold increase each in BDNF, its pro-peptide, and pro-BDNF in mutant tissues (Fig. 10 C). These results also indicate that the respective ratios of BDNF and the pro-peptide versus pro-BDNF are similar to what is observed in WT animals. Of note is that the consistent detection of BDNF pro-peptide was only possible after glutaraldehyde fixation of the transfer membrane (see Materials and methods). Furthermore, under the lysate and incubation conditions used, no measurable proteolysis of recombinant pro-BDNF (500 pg) added to the lysates at the beginning of the extraction procedure could be detected (Fig. S2). The recovery of added recombinant pro-BDNF was 102.2 ± 5.8% (n = 4).

Our study reveals that in the adult hippocampus, BDNF and its cleaved pro-peptide are stored in large DCVs located in the presynaptic terminals of excitatory neurons. Both in WT and Bsn mutant mice, BDNF and its pro-peptide are stored at roughly equimolar ratios and are ∼10-fold more abundant than pro-BDNF in hippocampal lysates. Together with the lack of any detectable BDNF staining in dendrites or spines, our results provide direct support for an anterograde mode of action for BDNF in the intact CNS. They also offer a morphological substrate for recent findings indicating that the release of endogenous BDNF accounts for some of the rapid calcium transients observed at synaptic sites on dendrites of CA3 neurons (Lang et al., 2007).

The specificity of BDNF immunostaining in the adult brain is very difficult to ascertain, as Bdnf knockout animals die before reaching maturity, a fact that complicates the interpretation of previous immunohistochemical studies (Wetmore et al., 1991; Dugich-Djordjevic et al., 1995; Schmidt-Kastner et al., 1996; Altar et al., 1997; Conner et al., 1997; Yan et al., 1997; Luo et al., 2001; Danzer and McNamara, 2004; Tongiorgi et al., 2004; Zhou et al., 2004; Agassandian et al., 2006; Avwenagha et al., 2006, Salio et al., 2007). Our study deals with the mature hippocampus, as much of the current interest in BDNF relates to its role in the adult brain, and the staining specificity of our BDNF monoclonal antibody was verified with tissues from age-matched mice lacking BDNF in the CNS (Kolbeck et al., 1999; Rauskolb et al., 2010). The choice of BDNF antibody was also important, as none of the commercially available BDNF antibodies we tested yielded specific staining (unpublished data). In addition, a mouse line was used in which Bdnf was replaced with a Myc-tagged version of the gene (Matsumoto et al., 2008), allowing additional control experiments to be performed with anti–human Myc antibodies on WT tissue sections. We note that a recent light microscopy study that also used a knockin strategy to tag endogenous BDNF came to a conclusion very similar to our own, revealing identical labeling of the mossy fiber pathway with antibodies directed against either the HA epitope used to tag BDNF or the BDNF prodomain (Yang et al., 2009). However, our biochemical results (see later part of Discussion) now indicate that pro-BDNF antibodies primarily detect the BDNF pro-peptide, which is present at much higher levels than uncleaved pro-BDNF. The pro-BDNF antibodies further validate our localization data, as they recognize an epitope unrelated to those detected by the antibodies to BDNF or Myc but similarly label presynaptic large DCVs. These organelles are well known to be involved in the regulated secretion of many other neuronal signaling peptides including Met-enk (Cheng et al., 1995), which was also used here as a presynaptic marker. DCVs are typically found in presynaptic terminals but rarely in postsynaptic structures such as dendrites or spines, with the notable exception of the DCVs containing the neuropeptides vasopressin and oxytocin, found in the dendrites of specialized neurosecretory cells (Kennedy and Ehlers, 2011).

Numerous previous studies have used cultured hippocampal neurons to study the storage and release of BDNF (Haubensak et al., 1998; Hartmann et al., 2001; Wu et al., 2004; Adachi et al., 2005; Brigadski et al., 2005; An et al., 2008; Dean et al., 2009; Matsuda et al., 2009). They all come to the conclusion that BDNF is stored in and released from both axonal and dendritic compartments, raising the question of what could underlie the apparent discrepancy with our findings. Although the majority of these studies utilize neurons transfected with a cDNA encoding a fluorescently tagged form of BDNF, some of the conclusions are also based on the detection of endogenous BDNF in vitro. However, cultured embryonic neurons grow in a very different environment compared with neurons in vivo, with fewer and less intimate cell–cell interactions between neighboring neurons and astrocytes, an incomplete ECM, and a lack of laminar input. In vivo, the development of neuronal circuits is spatially tightly regulated (Frotscher et al., 2000), whereas in vitro, synapses seem to form randomly. As a result of these cell-extrinsic differences, it is conceivable that cultured neurons fail to sort cargo-loaded vesicles as strictly as they do in vivo (Mellman and Nelson, 2008). In addition, we note that cultured neurons express significantly higher levels of BDNF than are detectable even in adult hippocampal extracts (unpublished data).

Although in situ hybridization experiments have long established that BDNF mRNA can be unambiguously detected in the cell bodies of excitatory neurons (Hofer et al., 1990; Isackson et al., 1991), some studies have also suggested its presence in dendrites (Tongiorgi et al., 1997, 2004; An et al., 2008) and its active transport upon stimulation (Tongiorgi et al., 1997, 2004; Chiaruttini et al., 2009; Baj et al., 2011; Louhivuori et al., 2011; Wu et al., 2011). However, these experiments revealed that BDNF mRNA is predominantly detected in the most proximal part of the dendrite, whereas other actively transported mRNAs can be detected in dendrites hundreds of micrometers away from the soma, as is the case for the Arc/Arg3.1 (Lyford et al., 1995), CaMKII-α (Burgin et al., 1990; Miller et al., 2002), or dendrin (Link et al., 1995) mRNAs. This leaves the possibility that BDNF mRNA localizes to proximal dendrites by diffusion rather than active transport. Selective trafficking has also been difficult to assess, as activity-dependent transcription augments the intensity of the proximal mRNA signal, allowing more distal diffusion of the in situ hybridization enzymatic reaction product (Tongiorgi et al., 1997, 2004). Although elegant experiments aimed at blocking protein synthesis and interfering with BDNF signaling have inferred the translation of BDNF mRNA in dendrites (Tanaka et al., 2008), it has not been possible thus far to assess directly whether dendritic BDNF mRNAs are translated locally. Thus, it is of particular significance that our experiments revealed a complete lack of specific BDNF protein signals in dendrites not only in WT but also in Bsn mutant animals. Although our experiments cannot rule out that functionally relevant quantities of BDNF may be released from dendrites, they indicate that, at present, there is no structural basis for this speculation.

With regard to the biochemistry of BDNF, initial studies suggested that the most abundant, or indeed the only detectable, translation product in adult brain extracts was uncleaved pro-BDNF (Chen et al., 2005; An et al., 2008). However, a signal of similar size and intensity to pro-BDNF can readily be detected in mice lacking BDNF in the CNS (Matsumoto et al., 2008). Furthermore, pulse-chase experiments showed that pro-BDNF is a short-lived intermediate that is rapidly processed intracellularly (Matsumoto et al., 2008). BDNF, like all other neurotrophins, is initially translated as a glycosylated precursor protein, presumably allowing the proper folding and disulfide bridging of the mature protein (Leibrock et al., 1989; Rattenholl et al., 2001). Accordingly, we detected specific pro-BDNF–IR in the Golgi apparatus of dentate granule cells and CA3 neurons but not in other organelles such as lysosomes or endosomes. Although the intracellular site of pro-BDNF cleavage and the participating proteases are yet to be identified, the existence of the BDNF pro-peptide in brain lysates has not been previously reported (see Fig. 10). This pro-peptide has hitherto escaped detection presumably because it is readily washed away from blotting membranes. This loss could be prevented by fixing the membranes with glutaraldehyde after electrophoretic transfer (Karey and Sirbasku, 1989). Although we observed that prolonged stimulation of cultured neurons results in the cosecretion of BDNF and its pro-peptide (unpublished data), the biological significance of this corelease is unclear.

Mutant mice expressing a truncated version of the large presynaptic protein Bassoon develop spontaneous epileptic seizures for reasons that are only partially understood (Altrock et al., 2003). In line with the well-established link between excitatory neuronal activity and Bdnf transcription, BDNF protein levels were found to increase in parallel with, but not precede, the development of seizure episodes (Heyden et al., 2011). It seems unlikely that Bassoon is directly involved in the increase in BDNF levels, as a mutation in the potassium channel Kv1.1 that also leads to seizure activity is accompanied by a marked increase in BDNF mRNA and megencephaly (Diez et al., 2003; Lavebratt et al., 2006), as is the case for the Bsn mutation (Angenstein et al., 2007). Blocking activity with carbamazepine in the Kv1.1 mutant prevents the development of megencephaly and restores BDNF mRNA to normal levels (Lavebratt et al., 2006). Despite the marked increase in BDNF and its pro-peptide in the Bsn mutant, both immunoreactive signals remained confined to presynaptic terminals, with no detectable staining in either granule or pyramidal cell dendrites or glial cells (Bergami et al., 2008). In comparison with WT animals, the only remarkable change observed in the Bsn mutant was an increase in the density of BDNF-positive DCVs in MFBs. These observations are in line with earlier studies in the rat, in which chemically induced seizure activity led to an increase in presynaptic BDNF staining (Vezzani et al., 1999; Scharfman et al., 2002; Danzer and McNamara, 2004). Despite the substantially higher BDNF levels, we did not detect any BDNF staining in the extracellular space, suggesting that after its release, BDNF becomes rapidly diluted by diffusion as well as by cellular reuptake, thus preventing accumulation to a degree that would allow immunohistochemical detection.

The hippocampal mossy fiber projection stands out as the most heavily stained structure in the adult CNS using either BDNF or pro-BDNF antibodies (Fig. 1). This pathway is thought to exhibit presynaptic long-term potentiation (LTP; Nicoll and Schmitz, 2005) and to be important for pattern completion as well as for the storage and recall of information (Bischofberger et al., 2006; Kerr and Jonas, 2008). Recent studies have also shown that long-term experience results in rearrangements of presynaptic MFBs (Galimberti et al., 2006, 2010; Rekart et al., 2007) and that LTP induction results in a dramatic increase in MFB profile size accompanied by postsynaptic spine growth and de novo formation of release sites (Zhao et al., 2012). Although it remains to be demonstrated that presynaptic release of BDNF is involved in these morphological changes, endogenous BDNF signaling has been shown to account for some of the spontaneous, rapid calcium transients observed at synaptic sites on CA3 dendrites (Lang et al., 2007). In addition, θ-burst–induced release of endogenous BDNF from mossy fibers is sufficient to induce intracellular changes in postsynaptic CA3 neurons (Li et al., 2010). Furthermore, several previous studies in CA1 have pointed to the need for BDNF in order for LTP to reach maximal values (Korte et al., 1995; Patterson et al., 1996), and our results uncover a presynaptic location of BDNF in this hippocampal area (Figs. 5 and S1).

Our localization data provide a structural substrate for previous functional results, suggesting that BDNF exerts its action on postsynaptic CNS neurons primarily in an anterograde fashion (Altar et al., 1997; Zakharenko et al., 2003; Baquet et al., 2004). This mode of action is the converse of the long-established model based on NGF in the peripheral nervous system, whereby neurotrophins are released from various cell types, including smooth muscle cells, and then act retrogradely on presynaptic neurons. Together with the recent discovery that the NGF receptor TrkA causes neuronal death in the absence of its ligand, whereas the BDNF receptor TrkB does not (Nikoletopoulou et al., 2010), these results suggest profound differences in the biology and directionality of growth factor signaling between the peripheral nervous system and the CNS.

Brains were obtained from the following lines of mice, all of which were used at ∼8 wk of age: WT (C57Bl6), Bdnf-Myc (Matsumoto et al., 2008), cbdnf ko (Rauskolb et al., 2010), and Bsn mutant (BsnΔEX4/5; Altrock et al., 2003). All experiments were performed in accordance with the institutional guidelines of the University of Freiburg and the University of Basel.

Mice (n = 7 for WT, n = 7 for Bdnf-Myc, n = 4 for cbdnf ko, and n = 7 for Bsn mutant) were transcardially perfused with 0.9% NaCl followed by a mixture of 4% PFA, 0.1% glutaraldehyde, and 0.3% picric acid in 0.1 M phosphate buffer (PB), pH 7.4. Brains were removed and postfixed with 4% PFA in 0.1 M PB for 1 h before thorough washing in 0.1 M PB. Coronal sections (50 µm) of the hippocampus were cut on a vibratome (Leica) and collected in 0.1 M PB. Sections along the septotemporal axis of the hippocampus were further processed either for immunofluorescent labeling or preembedding immunogold labeling.

BDNF was detected with the mouse monoclonal anti-BDNF antibody Mab#9 (anti-BDNF; Kolbeck et al., 1999) directed against the mature domain of BDNF (see Fig. 1 A). This antibody also recognizes the uncleaved form of BDNF, i.e., pro-BDNF (Fig. 1 A; Matsumoto et al., 2008). BDNF-Myc was detected with goat polyclonal c-Myc antibodies (anti-Myc; SC-789; Santa Cruz Biotechnology, Inc.). Pro-BDNF was detected with a rabbit polyclonal antibody (anti–pro-BDNF; #ANT-006, batch AN-03; Alomone Labs) raised against the prodomain of BDNF protein (see Fig. 1 A). To determine whether the signals obtained with anti–pro-BDNF could also be observed with other antibodies against the prodomain of BDNF, we tested a commercially available mouse monoclonal pro-BDNF antibody (anti–pro-BDNF–Mab5H8; SC-65514; Santa Cruz Biotechnology, Inc.). The staining pattern produced by this antibody (as specified in the figure legends) was comparable with the results obtained with the polyclonal anti–pro-BDNF. Presynaptic components were detected with mouse monoclonal antibodies against the vesicular glutamate transporter 1 (anti–VGLUT-1; #135511; Synaptic Systems) and rabbit polyclonal antibodies against SYP (anti-SYP; Santa Cruz Biotechnology, Inc.); for postsynaptic labeling, mouse monoclonal antibodies against MAP-2 (anti-MAP-2; #M2320; Sigma-Aldrich), rabbit polyclonal antibodies against the spine apparatus protein synpo (anti-synpo; #S9567; Sigma-Aldrich), and guinea pig polyclonal antibodies against activity-related cytoskeletal protein (anti-Arc/Arg3.1; #156005; Synaptic Systems) were used. A mouse monoclonal antibody against the 130-kD Golgi matrix protein (anti-GM130) was used for the detection of the Golgi complex (#610822; BD). For labeling of neuropeptides within the mossy fiber projection, rabbit polyclonal antibodies against Met-enk (anti–Met-enk; Millipore) and CCK (anti-CCK; #P06307; Millipore) were used. Astrocytes were detected with a guinea pig polyclonal antibody serum against GFAP (anti-GFAP; #031223; Advanced ImmunoChemical Inc.).

As preliminary staining experiments with anti-Myc antibodies yielded high levels of nonspecific staining in sections from WT mice, the Myc antibody solution was preincubated with WT sections, and the solution was subsequently used to stain sections from Bdnf-Myc and WT mice. Except for this first blocking step, the staining protocol for all antibodies was identical, and TBS, pH 7.4, was used throughout to wash the sections. For staining with polyclonal antibodies, sections were blocked with 20% normal donkey serum in TBS for 1 h; for staining with monoclonal antibodies, sections were blocked with 3% M.O.M (Mouse-on-Mouse blocking serum; Vector Laboratories) in TBS for 1 h. Primary antibodies were diluted in a solution of 3% BSA, 2% normal donkey serum, and 0.2% Triton X-100 in TBS to yield the following final concentrations/dilutions: 10 µg/ml anti-BDNF, 0.4 µg/ml anti-Myc, 1.33 µg/ml anti–pro-BDNF, 3.3 µg/ml anti–VGLUT-1, 4 µg/ml anti-SYP, 0.9 µg/ml anti-GM130, 2 µg/ml anti-Arc/Arg3.1, 4 µg/ml anti–MAP-2, 1:2,000 anti-synpo, 3 µg/ml anti–Met-enk, 1:2,000 anti-CCK, and 1:1,000 anti-GFAP. Sections were incubated for at least one night at 4°C in either one or more of these primary antibodies. For fluorescent signal detection, the following Alexa Fluor–conjugated secondary antibodies were used: donkey anti–rabbit-488 or -555, donkey anti–goat-555 or -488, donkey anti–mouse-555 or -647, and donkey anti–guinea pig–Cy5 (all purchased from Millipore). Labeled sections were mounted onto glass slides and coverslipped with fluorescent mounting medium (Dako).

Slide-mounted sections of immunolabeled hippocampi were viewed on an inverted microscope (Axiovert 200; Carl Zeiss) equipped with Plan Apochromat 20× (0.75 NA) or 63× (oil differential interference contrast; 1.4 NA) objectives and attached to a spectral confocal laser system (LSM 510; Carl Zeiss) powered by LSM 4.2 Meta software (Carl Zeiss). The tissue was scanned at room temperature with 488-, 543-, and 633-nm laser lines to detect the Alexa fluorophores 488, 555, and 647, respectively, and high-resolution images of optical sections (z slices) were captured using sequential line (mean of four) scanning. Colocalization of two or three fluorophores was qualitatively assessed in the x, y, and z planes of each optical section. Maximal projection images of confocal z series (stacks) were generated where indicated in the figure legends. Minimal adjustments to image contrast and intensity were made in Photoshop CS (version 8.0.1; Adobe) using the levels or contrast/brightness functions. Images were arranged and annotated using CorelDRAW 12 software (Corel Corporation).

50-µm vibratome sections of perfusion-fixed hippocampi from WT, Bdnf-Myc, cbdnf ko, and Bsn mutant mice were thoroughly washed in 50 mM TBS, pH 7.4, and immersed for 2 h in a cryoprotecting solution (25% sucrose and 10% glycerol in 0.05 M PB) before being snap frozen in liquid nitrogen–cooled isopentane. After immediate thawing and washing in 0.1 M PB, sections were incubated for 1 h in blocking solution (3% M.O.M. in TBS for anti-BDNF staining and 20% normal goat serum in TBS for anti–pro-BDNF and anti-Myc staining) followed by one to three nights at 4°C in a 3% BSA/TBS solution containing either 20 µg/ml anti-BDNF, 1.33 µg/ml anti–pro-BDNF (AN-03), or 0.4 µg/ml anti-Myc antibody. After thorough washing in TBS, the sections were incubated overnight at 4°C in 3% BSA/TBS solution containing either 1.4 nm of gold-conjugated goat anti–mouse, goat anti–rabbit, or rabbit anti–goat IgG. Sections were rinsed in TBS and then fixed for 10 min in a 1% glutaraldehyde solution. Tissue-bound gold particles were enlarged using a silver intensification kit (HQ Silver; Nanoprobes). The sections then underwent osmification for 40 min in a solution of 0.5% OsO₄ and 6.86% sucrose in 0.1 M PB.

After osmification, sections were washed in 0.1 M PB followed by 50% ethanol (EtOH). The tissue was then incubated in 1% uranyl acetate in 70% EtOH for 35 min followed by 10-min dehydration steps in increasing grades of EtOH. After washing in propylene oxide, the tissue was embedded in Durcupan (Fluka). Ultrathin sections (60 nm) of selected hippocampal areas (DG, CA3, and CA1) were cut and mounted on formvar-coated nickel grids. Sections were viewed and examined in an electron microscope (LEO 906E; Carl Zeiss).

Gold-labeled sections from WT, Bdnf-Myc, cbdnf ko, and Bsn mutant animals were first examined in the electron microscope at magnifications ranging from 6,000 to 12,000×, and areas containing labeled structures in SL of CA3 were then photographed using a dual-speed charge-coupled device camera (TRS SharpEye; Troendle) and analySIS acquisition software (Soft Imaging System; Olympus). MFBs were unambiguously identified based on their well-defined morphology (see criteria in Results). An organelle in an MFB was deemed labeled if associated with at least four gold grains (described in this study as a cluster). MFBs containing labeled organelles were then selected for quantitative analysis, in which the number of clusters per bouton profile was manually counted, and the area and perimeter of each profile were simultaneously measured with cell^P software (Soft Imaging System). The number of clusters per bouton was then reexpressed as a density measurement (i.e., the number of clusters per µm² of MFB). Although the rare gold clusters observed in cbdnf ko sections showed no association with any particular organelle, they were still included in the counts. In addition to cluster quantification, the respective number of single gold grains in MFBs from WT, Bdnf-Myc, cbdnf ko, and Bsn mutant tissues was also determined and ultimately expressed as a density measurement (i.e., the number of grains per µm² of MFB).

For quantification of gold labeling in dendritic profiles, an initial qualitative analysis of the tissue revealed a complete lack of cluster-labeled organelles, including large secretory vesicles and endosomes. Therefore, to compare gold grain density in dendrites from different groups of animals, the number of single gold grains per dendritic profile was counted and expressed in relation to the area of the corresponding profile (i.e., the number of grains per µm²).

Finally, to determine whether background gold grain labeling was equally distributed across different subcellular compartments, the quantitative gold grain analysis was extended to two additional compartments in SL, namely dendritic spine profiles and myelinated axon profiles. These two compartments were chosen on the basis of also being devoid of cluster-labeled organelles.

Hippocampi were dissected from 8-wk-old WT, cbdnf ko, and Bsn mutant mice and then weighed and stored at −80°C. 10 vol/wt of radioimmunoprecipitation assay buffer (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1 mM EDTA, 1% Triton X-100, and 0.2% Na deoxycholate) was added. To prevent proteolysis, freshly prepared protease inhibitors including a protease inhibitor cocktail (Roche), 10 µM 1,10-phenanthroline monohydrate, 10 mM 6-aminohexanoic acid, and 10 µg/ml aprotinin were added to the buffer. The tissues were sonicated, the homogenates centrifuged, and the supernatants collected. IP was performed with 500 µg of total protein using either the BDNF monoclonal antibody Mab#9 or a rabbit polyclonal antibody raised against the BDNF pro-peptide (Koshimizu et al., 2009) in the presence of protein G agarose (Roche) according to the manufacturer’s instructions. After blotting, the proteins were fixed to the membrane with 0.5% glutaraldehyde for 30 min at room temperature and probed with either polyclonal rabbit anti-BDNF N20 (sc-546; Santa Cruz Biotechnology, Inc.) or monoclonal mouse anti–pro-BDNF Mab5H8 (1:200) using an IP Western kit (GenScript) to eliminate nonspecific signals from protein G. Signal intensity measurements were performed using ImageJ software (National Institutes of Health). Recombinant WT and cleavage-resistant pro-BDNF (Fayard et al., 2005) as well as the mouse BDNF pro-peptide were produced in COS7 cells transfected with the corresponding cDNAs cloned in plasmid cytomegalovirus vectors. The concentrations of these proteins were then determined against known amounts of purified BDNF pro-peptide produced in Escherichia coli (Koshimizu et al., 2009). The COS7-derived cleavage-resistant pro-BDNF was used as a molecular mass marker, whereas WT pro-BDNF was used in recovery experiments with tissue lysates (Fig. S2)

All values are expressed as mean ± SEM. For BDNF and pro-BDNF immunogold quantification, 10–20 MFBs and 30–40 dendrites per WT, Bdnf-Myc, cbdnf ko, and Bsn mutant mice were analyzed, and mean values for each animal were calculated. Group means (pooled WT/BDNF-Myc vs. cbdnf ko and WT vs. Bsn mutant) were compared using a two-sided Student’s t test. A probability of P < 0.05 was considered to be statistically significant.

Fig. S1 shows an increase in BDNF-IR and pro–BDNF-IR in area CA1 of the Bassoon mutant hippocampus. Fig. S2 shows the complete recovery of WT recombinant pro-BDNF from hippocampal lysates.

We thank Drs. J. Kowalski and M. Sibbe for helpful statistical advice.

This work was supported by the Swiss National Foundation (31003A_124902/1), the German Research Foundation (SFB 780, project A4), and a Postdoctoral Research Fellowship (to S. Dieni) from the Alexander von Humboldt Foundation. M. Frotscher is a Senior Research Professor of the Hertie Foundation.