Multiple functions of precursor BDNF to CNS neurons: negative regulation of neurite growth, spine formation and cell survival

Background Proneurotrophins and mature neurotrophins elicit opposite effects via the p75 neurotrophin receptor (p75NTR) and Trk tyrosine kinase receptors, respectively; however the molecular roles of proneurotrophins in the CNS are not fully understood. Results Based on two rare single nucleotide polymorphisms (SNPs) of the human brain-derived neurotrophic factor (BDNF) gene, we generated R125M-, R127L- and R125M/R127L-BDNF, which have amino acid substitution(s) near the cleavage site between the pro- and mature-domain of BDNF. Western blot analyses demonstrated that these BDNF variants are poorly cleaved and result in the predominant secretion of proBDNF. Using these cleavage-resistant proBDNF (CR-proBDNF) variants, the molecular and cellular roles of proBDNF on the CNS neurons were examined. First, CR-proBDNF showed normal intracellular distribution and secretion in cultured hippocampal neurons, suggesting that inhibition of proBDNF cleavage does not affect intracellular transportation and secretion of BDNF. Second, we purified recombinant CR-proBDNF and tested its biological effects using cultured CNS neurons. Treatment with CR-proBDNF elicited apoptosis of cultured cerebellar granule neurons (CGNs), while treatment with mature BDNF (matBDNF) promoted cell survival. Third, we examined the effects of CR-proBDNF on neuronal morphology using more than 2-week cultures of basal forebrain cholinergic neurons (BFCNs) and hippocampal neurons. Interestingly, in marked contrast to the action of matBDNF, which increased the number of cholinergic fibers and hippocampal dendritic spines, CR-proBDNF dramatically reduced the number of cholinergic fibers and hippocampal dendritic spines, without affecting the survival of these neurons. Conclusion These results suggest that proBDNF has distinct functions in different populations of CNS neurons and might be responsible for specific physiological cellular processes in the brain.


Background
The development and functioning of the mammalian nervous system are regulated by neurotrophins, that are a family of neurotrophic factors which includes nerve growth factor (NGF), BDNF, neurotrophin-3 (NT-3), and NT-4/5 [1]. Like many peptide hormones and growth factors, neurotrophins are first synthesized as precursors and are subsequently cleaved either intracellularly by prohormone convertases (PCs) and/or furin, or extracellularly by plasmin and matrix metalloproteases (MMPs) to form mature proteins [2]. These proteins elicit their biological actions by binding to the Trk family of receptor-type tyrosine kinases [3]. For decades, proneurotrophins were thought to be biologically inactive; the dogma was changed when Lee et al. [4] reported that proNGF preferentially interacted with p75 NTR instead of Trk receptors. This interaction leads to apoptosis of peripheral neurons, an effect opposite to the pro-survival action of mature NGF. Since then, proapoptotic effects of proNGF and proBDNF mediated by p75 NTR have been demonstrated in a number of model systems [4][5][6][7][8]. Therefore, proteolytic cleavage of proneurotrophins is thought to be an important regulatory step for the direction of neurotrophin function, given the diametrically opposed functions of proneurotrophins and mature neurotrophins, which are elicited via the p75 NTR and Trk receptors, respectively [9,10].
The cleavage of proBDNF ( 125 RVRR 128 ↓ HS in Fig. 1A), [2] is thought to take place either intracellularly by serine proteases such as PC1/3 and/or furin, or extracellularly by extracellular proteinases such as plasmin activated by tissue plasminogen activator (tPA), and/or MMPs. The importance of proBDNF cleavage has been reported recently [11,12]. These two reports suggest that proBDNF cleavage in the nervous system is regulated in a more specific manner and depends on the cellular context. However, little is known about the function of proBDNF when compared with the numerous reports on the mechanisms of mature BDNF (matBDNF) action [10,13]. In vitro experiments have shown that proBDNF enhances apoptosis of sympathetic neurons and basal forebrain neurons [7,8,14]. Application of exogenous proBDNF facilitates long-term depression (LTD), through an interaction with p75 NTR [15], suggesting that proBDNF may exert multiple biological actions in the nervous system.
Another important question is whether proBDNF has a physiological function [10,13]. The two rare human SNPs at nucleotides 373 (G/T) and 379 (G/T) (Reference SNP cluster id: rs1048220 and rs1048221, respectively) present in the coding region of the human BDNF gene are located near the cleavage site (Fig. 1A) and theoretically lead to the substitution of the arginine residues (R) by methionine (R125M) and leucine (R127L), respectively (Fig. 1A). Although the precise validation for these SNPs has not been fully determined, cDNA analysis on R125M suggests that one out of 92 persons from Utah State carried M-type BDNF gene http://www.ncbi.nlm.nih.gov/ sites/entrez. Thus, these polymorphisms may affect the cellular mechanisms and biological functions of BDNF in human brain.
In the present study, taking advantage of the cleavage resistance of these BDNF variants (CR-proBDNF), various cultured CNS neurons including hippocampal neurons, CGNs, and BFCNs were treated with recombinant CR-proBDNF protein to address the molecular functions of proBDNF on CNS neurons. Our studies suggested that CR-proBDNF showed normal intracellular distribution and secretion in cultured hippocampal neurons and that CR-proBDNF exerts distinct functions in cerebellar granule neurons (CGNs), basal forebrain cholinergic neurons (BFCNs) and hippocampal neurons.
BDNF polymorphic substitutions inhibit the conversion of proBDNF to matBDNF and lead to predominant secretion of proBDNF in CNS neurons . Doublet bands of 29-and 32-kDa were detected with the monoclonal anti-proBDNF antibody (IB: proBDNF), whereas an additional band of 14-kDa (matBDNF) was detected using an antibody against the mature domain (IB: BDNF). The levels of bicistronically expressed GFP, which correlated with the viral infection levels, were similar in all cultures. Note that proBDNF bands were predominant in the lysates of cells expressing the M, L, and ML constructs, while matBDNF expression was residual. In C and F, 1 ng recombinant matBDNF and proBDNF (R125M/R127L-BDNF) were used as positive controls. (D) The ratio of proBDNF to matBDNF was quantified by densitometric analysis on BDNF, and proBDNF bands. **P < 0.01 (Student's t-test) when compared with matBDNF (mat-BDNF RR + proBDNF RR = 100% as control). n = 3 independent experiments. (E) The entire amount of BDNF was quantified by densitometric analysis on matBDNF, proBDNF, and GFP bands. *P < 0.05 (t-test) when compared with RR (100% as control). n = 3 independent experiments. (F) Secretion of poorly cleaved proBDNF. Cultured cerebral cortical neurons were incubated with Sindbis viruses for 12 h and maintained in serum-containing medium for 3 days. Supernatants were collected and immunoprecipitated using anti-Myc antibody-conjugated agarose beads, followed by Western blot analysis using rabbit antibody against the mature domain (IB: BDNF) and mouse antibody recognizing the prodomain (IB: proBDNF). Cell lysates were blotted with anti-GFP antibody to normalize viral infection (IB: GFP). that expressed both BDNF and green fluorescent protein (GFP) [19]. The multiplicity of infection (MOI) was 10 and 1 for Western blotting and immunostaining, respectively [18]. Such precise controls of Sindbis virus infection did not promote cell death of the neurons [18]. Western blotting was performed to compare the processing of these exogenous proteins. In control neurons expressing BDNF RR , we detected a doublet band of 29-and 32-kDa (proBDNF) and a 14-kDa band (matBDNF), using either anti-BDNF or anti-Myc antibodies (RR in Fig. 1C, IB: BDNF and Myc). Exogenously expressed BDNF RR appeared to be cleaved in a fashion similar to endogenous BDNF [20]. Interestingly, the proBDNF bands (29-and 32-kDa) were predominant in neurons expressing BDNF M , BDNF L , or BDNF ML , which showed residual levels of matBDNF (14-kDa) (M, L, and ML in Fig. 1C, IB: BDNF and Myc). This finding was confirmed using a quantitative analysis (Fig. 1D). Bands corresponding to precursor BDNF (Fig. 1C, IB: proBDNF) were also detected using anti-prodomain monoclonal antibody [Additional file 1B, mouse]. We ascertained that the apparent difference in the ratio of proBDNF to matBDNF between the wild-type and BDNF variants was due to differences in proteolytic processing and not to variability in the efficiency of Sindbis virus infection or BDNF expression levels, as all samples exhibited similar levels of GFP expression from the bicistronic vector used in this study (Fig. 1C). However, the total amount of BDNF {(mat-BDNF + proBDNF)/GFP} was slightly, yet significantly, reduced in neurons expressing the mutant BDNF constructs (Fig. 1E), which suggests that inhibition of proBDNF cleavage may affect the overall levels of expression of BDNF. These results showed that the amino acid substitutions associated with the SNPs (R125M, R127L, and R125M/R127L) inhibit the proteolytic cleavage of proBDNF to matBDNF in CNS neurons.
To examine whether the cleavage resistance affected BDNF secretion, we expressed the constructs encoding Myctagged BDNF RR , BDNF M , BDNF L , or BDNF ML in cultured cortical neurons using the transfection protocol depicted in Fig. 1C. Supernatants collected 3 days later were immunoprecipitated using an anti-Myc antibody conjugated to agarose beads. Secreted BDNF isoforms were detected by Western blot analysis using an anti-mature domain antibody that detects both proBDNF and matBDNF (Fig. 1F). Neurons expressing wild-type BDNF (RR) secreted several BDNF species: 30-and 32-kDa precursor forms, a 26-kDa intermediate form, and a 14-kDa mature form (RR in Fig.  1F, IB: BDNF). In contrast, supernatants collected from neurons expressing the BDNF polymorphic variants (M, L, and ML) contained predominantly the precursor forms with negligible, but detectable, levels of matBDNF (M, L, and ML in Fig. 1F, IB: BDNF). These results were confirmed using an anti-prodomain monoclonal antibody (M, L, and ML in Fig. 1F, IB: proBDNF). These results suggest that the BDNF polymorphic variants markedly inhibit the conversion of proBDNF to matBDNF, either intracellularly or extracellularly, and that they are mainly secreted as proBDNF.

Inhibition of proBDNF cleavage does not affect intracellular transportation and secretion of BDNF in hippocampal neurons
Previous studies suggest that inhibition of intracellular cleavage by furin alters the subcellular distribution and secretion of neurotrophins, particularly of NT-3 [21]. To determine whether the two SNPs affect the intracellular distribution of BDNF, we expressed transgenes encoding GFP-tagged (C-terminus) wild-type BDNF (BDNF-GFP) or cleavage-resistant R125M/R127L-BDNF (CR-proBDNF) in cultured hippocampal neurons according to a previous report [18] (Fig. 2A). The two fusion proteins exhibited similar distribution patterns in the cell body and neuronal processes ( Fig. 2A, arrows). Immunocytochemical analysis demonstrated that both BDNF-GFP and CR-proBDNF-GFP colocalized with a trans-Golgi network marker, TGN38 (Fig. 2B, arrows) and a secretory vesicle marker, secretogranin II (SgII) (Fig. 2C, arrows). An immunostaining study further showed that the signals of BDNF-Myc or CR-proBDNF-Myc were largely colocalized with the proBDNF signals (Fig. 2D, arrows). These findings, together with the result of Fig. 1F, suggest that inhibition of BDNF cleavage by the amino acid substitutions did not greatly influence the subcellular distribution of BDNF and that intracellular cleavage of proBDNF is not a critical step for intracellular transportation and secretion of BDNF in hippocampal neurons.

Recombinant CR-proBDNF elicits apoptosis of cerebellar granule neurons cultured in low K + containing medium
To test the biological effects of the CR-proBDNF in vitro, we purified the recombinant mutant protein using either the Baculovirus [22] or E. coli expression systems [23]. The identity of the purified proteins was confirmed using silver staining via analysis of their comigration with a molecular weight marker (~30-kDa) and detection on Western blot using a proBDNF-specific antibody (Fig. 3A). Baculoviruses expressed two BDNF isoforms that migrated at approximately 30-and 32-kDa; the latter was likely to be a glycosylated form (Fig. 3A, left). The CR-proBDNF protein expressed in E. coli formed inclusion bodies; therefore, we used urea denaturation for purification followed by refolding of the protein. The refolded CR-proBDNF exhibited a secondary structure, as assessed using circular dichroism (CD) spectroscopy [Additional file 2A].
To address the role of proBDNF on cell viability of CNS neurons, we first examined the function of E. coli-derived CR-proBDNF protein in cultured cerebellar granule neu-rons (CGNs), which express both p75 NTR and TrkB receptors [24]. CGNs require high levels of extracellular K + (HK, 26 mM) for survival. In addition, a low extracellular K + concentration (LK, 5 mM) induces apoptosis [25]. Cell viability of CGNs was assessed 48 h later using DAPI staining, which allows the visualization of apoptosis-induced chromatin condensation in the nucleus. Consistently with previous reports [26,27], matBDNF promoted the survival of CGNs, even when cell death was induced by LK ( Fig.  3B, matBDNF, arrows). However, CR-proBDNF failed to prevent apoptosis induced by LK (Fig. 3B, CR-proBDNF, arrowheads). Quantitative analysis (Fig. 3C) revealed that 100 ng/ml CR-proBDNF promoted LK-induced apoptosis in CGNs and increased the number of dying cells by 57.0 ± 0.2% when compared with the mock culture (ANOVA followed by post-hoc analysis; **P < 0.01). In contrast, 100 ng/ml matBDNF significantly inhibited LK-induced apoptosis in CGNs when compared with the mock culture (39.0 ± 0.0%, **P < 0.01) [26,27]. These results suggest that CR-proBDNF promotes LK-induced apoptosis of cultured CGNs and that CR-proBDNF and matBDNF exert the opposite biological effect on the survival of CGNs.
A previous study demonstrated that recombinant cleavage-resistant proBDNF, at subnanomoler concentration, elicited apoptosis of NGF-deprived superior cervical ganglion (SCG) neurons [7]. We therefore investigated the effect of E. coli-derived CR-proBDNF on apoptosis of LK-Intracellular proteolytic cleavage of proBDNF is not a crucial step for intracellular transportation and secretion of BDNF in hippocampal neurons Figure 2 Intracellular proteolytic cleavage of proBDNF is not a crucial step for intracellular transportation and secretion of BDNF in hippocampal neurons. (A) Distribution of GFP-tagged wild-type BDNF (BDNF-GFP) and cleavage-resistant (CR)-proBDNF (R125M/R127L-BDNF-GFP) in the cell body and processes. Cultured neurons were subjected to Sindbis viruses expressing the indicated constructs for 3 h. Three days later, the neurons were fixed for imaging of GFP fluorescence using confocal microscope. Representative low-magnification (upper) and high-magnification images of the cell body (middle) and neuronal processes (bottom) are shown. (B-C) The infected cells were immunostained with the indicated antibodies for confocal imaging. Note that CR-proBDNF-GFP greatly co-localized with TGN38 (B) and SgII (C), similarly to wild-type BDNF-GFP (arrows). Scale bar in all images, 10 μm. (D) Expression of wild-type proBDNF-Myc or CR-proBDNF-Myc. Hippocampal neurons were introduced with constructs encoding GFP and wild-type BDNF-Myc or CR-proBDNF-Myc using Sindbis virus expression system. Three-days after a brief (3 h) infection, cells were double-stained using anti-Myc and anti-proBDNF antibodies. BDNF-Myc signals largely co-localized with proBDNF signals (arrows) indicating that proBDNF is the predominant intracellular isoform. Scale bar, 5 μm.
treated CGNs at a similar concentration. A significant increase in the number of apoptotic CGN neurons was found even at 10 ng/ml (0.2 nM) CR-proBDNF, to a similar extent to that of 100 ng/ml (Fig. 3D). In addition, E. coli-derived CR-proBDNF failed to induce TrkB phosphorylation for up to 360 min after the application [Additional file 2B] and CR-proBDNF-induced cell death was not significantly found in CGNs derived from p75 NTR-/mice [Additional file 2C, p75 NTR-/-, t-test, P = 0.145, compared to Mock]. Finally, a proapoptotic effect was evident for CR-proBDNF derived from both Baculovirus and E. coli, whereas heat-denatured CR-proBDNF had no effect [Additional file 2D]. These results together suggest that E. coli-derived CR-proBDNF has a proBDNF biological activity to the similar to mammalian cell-derived proBDNF [7].
We also examined whether CR-proBDNF could affect the survival of CGNs cultured in HK medium, which promotes the survival of these cells [25]. Quantitative analysis of cell death using DAPI staining revealed that neither CR-proBDNF nor matBDNF had a significant effect on the number of CGNs in this survival-promoting medium [Additional file 2E], suggesting that proBDNF is proapoptotic in vulnerable neurons rather than in healthy neurons and that compared to peripheral neurons CNS neurons may not be so sensitive to proBDNF. Finally, it is notable that the BDNF polymorphisms, which lead to an inefficient cleavage of proBDNF, might be risk factors for defects in CGNs development and/or survival.

CR-proBDNF inhibits neurite growth in basal forebrain cholinergic neurons
To examine whether proBDNF-induced apoptosis is a general phenomenon in all CNS neurons that express p75 NTR , we chose to use basal forebrain cholinergic neurons (BFCNs), which are known to express high levels of p75 NTR [24]. A large body of evidence demonstrates that matBDNF promotes survival and differentiation of BFCNs [1]. A previous, yet initial, report characterized biological response of the BFCNs to matBDNF [28]: (1) the pro-survival effect was relatively more striking in low-density cultures, while the differentiated effects were more pronounced in high-density culture and (2) delayed application of matBDNF diminished the pro-survival effect even in low-density culture. In the present study, we sought to apply CR-proBDNF to rat BF neurons cultured in serum-containing medium for 2 weeks in order to see the biological action of proBDNF on BFCNs.
Consistently with previous reports [28,29], 4-day treatment with matBDNF (100 ng/ml) markedly increased the number and complexity of acetylcholine esterase (AChE)positive neurites of the 2-week cultured BFCNs ( Fig. 4A and 4C, matBDNF). The cell viability of the AChE-positive BFCNs was assessed by DAPI staining (Fig. 4A). Unlike what was observed in CGNs, neither matBDNF nor CR-proBDNF (E. Coli-derived recombinant protein) has a significant effect on the number of the survived BFCNs and on the nuclear morphology ( Fig. 4A, arrows, and 4B). However, CR-proBDNF caused the opposite effect on fiber density in BFCNs: treatment with 100 ng/ml CR-proBDNF for 4 days led to a marked reduction in the neurite number of AChE-positive neurites extending outwards from the cells (Fig. 4D, CR-proBDNF), without affecting the survival of these cells (Fig. 4C, CR-proBDNF). Quantitative analysis showed that CR-proBDNF decreased the number of primary fibers by more than 45% (Fig. 4D, CR-proBDNF). In contrast, matBDNF increased the number of the AChE-positive fibers by approximately 17% (Fig. 4D, mBDNF). We performed several methods to evaluate of neurite growth. Sholl analysis, which is used to assess the number and complexity of fibers extending out from the cell body [30], revealed that CR-proBDNF significantly reduced the number of cholinergic fibers in the first 10 μm from the cell body by 42% (Fig. 4E, CR-proBDNF). We also determined the neurite density of BFCNs in culture dishes (intensity of cholinergic fibers/number of cholinergic neurons). The opposing effects of matBDNF and CR-proBDNF on the fiber density were confirmed by this quantitative method (see Methods): Mock 100 ± 13.6% (as control); matBDNF, 149 ± 9.3%**; CR-proBDNF, 37.3 ± 9.4%** (t-test, **P < 0.01, compared to Mock; n = 6 independent culture dishes) (Fig. 4F). These results together suggest that CR-proBDNF inhibits neurite growth of BFCNs while matBDNF enhances this growth.
Finally, we examined whether CR-proBDNF had the similar effect on fiber density of BFCNs in serum-free condition. To this end, we cultured BFCNs in serum-free medium containing 5 μg/ml transferrin, 5 μg/ml insulin, 20 nM progesterone and 10 μM cytosine arabinoside (AraC) [29] for 2 weeks and applied CR-proBDNF to the neurons in the same medium. Four-day treatment with CR-proBDNF significantly decreased the fiber density, but not cell viability, of BFCNs, and even at subnanomoler concentration the effect of CR-proBDNF was significant (t-test, **P < 0.01) compared to that of un-treated cultures (Fig. 4G). These data suggest that proBDNF acts to BFCNs in a direct manner for negative regulation of the neurite density.

CR-proBDNF reduces dendritic spine density of hippocampal neurons in primary and slice cultures
As proBDNF facilitates LTD [15], proBDNF may have distinct biological roles in synaptic plasticity [10]. To test this possibility, we cultured hippocampal neurons (6-8 × 10 4 cells/cm 2 ) for 3-4 weeks and then applied 50 ng/ml of matBDNF or proBDNF for 2 days. We first performed DiI Recombinant CR-proBDNF induces apoptosis of cerebellar granule neurons cultured in low K + -containing medium Figure 3 Recombinant CR-proBDNF induces apoptosis of cerebellar granule neurons cultured in low K + -containing medium. (A) Preparation of recombinant CR-proBDNF (R125M/R127L-BDNF: proBDNF ML ) using the Baculovirus or E. coli expression systems. Protein products were analyzed by silver staining (left) and Western blotting using rabbit antibody against BDNF prodomain. (B-C) Opposite actions of CR-proBDNF and matBDNF revealed by apoptosis assay. After 4 days of culture in HK medium, CGNs were treated with E. coli-derived CR-proBDNF or matBDNF in LK medium. Cell death was assessed using DAPI staining 48 h after treatment with the indicated drug in LK medium. (B) Arrows and arrowheads indicate representative living and dead cells, respectively. Scale bar, 50 μm. (C) For quantification of cell death, 300-400 cells were counted in four independent fields per coverslip. (D) Dose-dependent test of pro-apoptotic activity of CR-proBDNF. Cell death was assessed using DAPI staining 48 h after treatment with CR-proBDNF in LK medium. DAPI data were expressed as the percentage of mock cultures. In the cell death assay (C-D), n = 4 independent culture dishes. ANOVA followed by post-hoc analysis, *P < 0.05; **P < 0.01. Results were replicated in at least three independent experiments.
We further visualized spine protrusions using the F-actinbinding dye phalloidin [34]. Similarly to the results obtained in Fig. 5A, matBDNF and CR-proBDNF had opposite effects on dendritic spine formation (Fig. 5B, arrows and arrowheads). Quantitative analysis in a similar manner to Figure 5A the similar CR-proBDNF decreased the density of phalloidin-labeled spines by 41.6 ± 9.0% (Fig. 5B, right, CR-proBDNF, nonparametric test; *P < 0.05 when compared with Mock). We further tested whether a functionally blocking antibody against p75 NTR , AB1554 [35,36] inhibited the effect of CR-proBDNF on spine density, because it was reported that, after 14 d in vitro, hippocampal neurons allowed the significant colocalization of p75 NTR with PSD95 [15]. Treatment with AB1554 (1/100 dilution) significantly rescued the effect of CR-proBDNF on spine density (Fig. 5B, right, CR-proBDNF + AB1554), suggesting that proBDNF involves p75 NTR on the neuronal surface to reduce the spine density. We next examined the effect of CR-proBDNF (50 ng/ ml, 2 days) on the number of MAP-2-positive neurons in culture. Quantitative analysis revealed that CR-proBDNF did not affect the survival nor the overall dendritic morphology of the cultured mature hippocampal neurons used in this experiment [Additional file 3], suggesting that this effect of CR-proBDNF on spine density may not be due to the proapoptotic effect of proBDNF.
Additionally, we examined whether CR-proBDNF affected spine density in serum-free and low-density cultures of hippocampal neurons, as described in Methods. In this culture, the percentage of contaminated glial cells was 7.0 ± 1.4% (n = 4 independent culture dishes). Treatment of CR-proBDNF for 2 days led to a significant decrease in spine density and 10 ng/ml (0.2 nM) was sufficient for the negative action of CR-proBDNF (Fig. 5C). These data support the notion that proBDNF reduces spine density of hippocampal neurons in a direct manner.
Finally, to evaluate this novel effect of CR-proBDNF in physiological contexts, we carried out two additional experiments. First, we investigated the effect of CR-proBDNF on dendritic spine density in hippocampal slice cultures. Interestingly, CR-proBDNF (50 ng/ml, 2 days) reduced the density of spines in hippocampal slice cultures (Fig. 5D, CR-proBDNF). Second, we recorded spontaneous miniature excitatory postsynaptic currents (mEPSCs) [37,38] in cultured hippocampal neurons (21 days in vitro) after 2-day treatment with matBDNF or CR-proBDNF (50 ng/ml). Hippocampal neurons treated with CR-proBDNF exhibited basal synaptic transmission (Fig.  6A), which support the notion that CR-proBDNF does not affect the survival of more than 3 week-cultured hippocampal neurons [Additional file 3]. However, proBDNF treatment significantly decreased the amplitude of mEP-SCs (Fig. 6B, left, nonparametric test, *P < 0.05 when compared with the Mock group). As reported previously [39], the amplitude of mEPSCs did not increase significantly after matBDNF treatment. Neither matBDNF nor CR-proBDNF significantly affected the frequency of mEP-SCs (Fig. 6B, right). These results together suggest that proBDNF reduces dendritic spine density of CNS neurons in slice culture and attenuates the amplitude of functional synaptic transmission.

Discussion
Since the discovery that proNGF binds p75 NTR with a higher affinity than matNGF and that it induces apoptosis rather than enhancing cell survival in sympathetic neurons [4], proneurotrophin-p75 NTR signaling and function have returned to the center stage in the field of neurotrophin research. There is a pressing need to understand the cleavage of proneurotrophins, as well as their biological functions in different brain regions. A crucial, yet poorly solved, issue is whether proBDNF has a physiological function [13]. In the present study, based on a predictive study of BDNF protein variants, we first provided molecular and cellular insights on the possible consequences that the rare human SNPs have near the cleavage site of proBDNF ( Figs. 1 and 2). Second, we demonstrated that the poorly cleaved proBDNF (CR-proBDNF) enhances apoptosis in cultured CGNs (Fig. 3). Finally, we demonstrated the novel actions of proBDNF, which include the inhibition of neurite outgrowth in BFCNs (Fig. 4) and the reduction of dendritic spine density in mature hippocampal neurons (Fig. 5). These results suggest that proBDNF cleavage is a crucial step for the nega-Neurite growth of basal forebrain cholinergic neurons is inhibited by CR-proBDNF, but elicited by matBDNF , and 30 (CR-proBDNF) cells from three independent coverslips. In multi-bar graphs, ANOVA followed by post-hoc analysis was used. **P < 0.01. Results were replicated in at least three independent experiments. (F) The opposing effects of matBDNF and CR-proBDNF on neurite fiber density were confirmed by a distinct quantitative method. The maximal threshold of AChEpositive cholinergic fiber intensity was defined as 70% above the background. The total intensity of the fibers was determined in an optical field and was divided by the number of AChE-positive neurons in the same field. Data were collected from four independent fields in a single chamber. t-test, **P < 0.01, compared to Mock (100% as control); n = 6 independent culture dishes. Scale bar, 5 μm (A and C). (G) Effect of CR-proBDNF on neurite density of BFCNs in serum-free condition and a dosedependency test of CR-proBDNF. Cell survival (left) and the neurite number of BFCNs (right) were determined. Note that proBDNF negatively regulates the neurite density of BFCNs in serum-free conditions and at subnanomoler concentration. n = 3 independent coverslips. t-test, **P < 0.01, significantly different from Mock. tive regulation of neurotrophic actions in the brain and that proBDNF exerts multiple biological actions on CNS neurons.
Although the rare human BDNF polymorphisms studied here are validated by the National Center for Biotechnology Information (NCBI) and the mutations cause amino acid substitutions near the cleavage site of proBDNF, the functional and physiological impacts remain ill-defined. Applying a bioinformatic analysis (PONDR), we predicted that the mutants would undergo disordered-toordered transitions around the proteinase cleavage site of proBDNF (Fig. 1B). The inhibition of proteolytic cleavage of the BDNF polymorphic variants was demonstrated by Western blot analysis of cultured neurons transfected with constructs encoding the variant forms of proBDNF (Fig.  1C) and an in vitro protease digestion assay [Additional file 4]. Interestingly, it was reported that a specific plasmin-cleavage could occur after Arg125 rather than Arg128 of the RVRR sequence [40]. Nakayama (1997) also described in a review article that since furin cleavage is essential for the production of a wide variety of biologically active proteins it is possible that mutation of the furin cleavage site of the precursors may result in genetic disorders [41]. Additionally, we have recently generated a knock-in mouse line by replacing the endogenous BDNF allele with a CR-proBDNF-encoding construct and found inefficient processing of the mutant proBDNF protein (data not shown), suggesting that the human BDNF polymorphisms would cause reduced the conversion of proBDNF into matBDNF in vivo.
We further demonstrated that the intracellular distribution (trans-Golgi and secretory granules) and subcellular localization (dendrites and soma) of CR-proBDNF were normal (Fig. 2), which suggests that the amino acid substitutions per se do not alter intracellular sorting and trafficking and supports the notion that endogenous proBDNF is transported [12] and secreted in an activitydependent manner [42]. In this context, these findings are in marked contrast to those obtained for the Val66Met variation of BDNF, which leads to the striking alteration of the sorting of BDNF to the secretory granules and of its subsequent activity-dependent secretion [18,43].
Two recent reports described the mechanisms of BDNF secretion [11,12]. In mixed cultures of neurons and glial cells, proBDNF was rapidly converted into matBDNF and was secreted as matBDNF. In neuron-rich cultures, however, proBDNF was the predominant form released. These two reports suggest that proBDNF cleavage in the nervous system is regulated in a more specific manner and on cellular context [13]. Although our results stemmed from overexpression experiments, the inhibition of proBDNF cleavage significantly reduced the content of BDNF in cells and resulted in the predominant secretion of the proBDNF when compared with wild-type BDNF. Our in silico prediction study suggested that the prodomain was structurally disordered thus rendering it likely to be degradation (Fig. 1B). Together, these results raise questions about the roles of the BDNF prodomain and the stable presence of proBDNF inside and outside of neurons.
Based on studies on proNGF, the concept is emerging that mature neurotrophins promote cell survival, while proneurotrophins facilitate apoptosis [9]. A recent study showed that proBDNF also promotes the death of sympathetic neurons, which extends the pro-apoptotic role to a second member of the neurotrophin family [7]; however, little is known about the inhibitory action of proBDNF on the survival of CNS neurons [8]. Using primary cultures of CGNs derived from rat and p75 NTR -knockout mice, we demonstrate that CR-proBDNF enhances LK-induced apoptosis in CGNs (Fig. 3, [Additional file 2]). It is also noteworthy that we have developed a simple protocol to generate recombinant proBDNF from E. coli. (Fig. 3A, [Additional file 2A]). First, the recombinant proBDNF, which, even at subnanomolar concentrations, significantly promoted apoptosis of cultured CGNs (Fig. 3D), [Additional file 2]. Second, E. coli-derived CR-proBDNF failed to induce TrkB phosphorylation for up to 360 min after the application [Additional file 2B]. Third, similar to the previous reports [7,8], in the absence of p75NTR, E. coli-derived CR-proBDNF failed to promote apoptosis of CGNs [Additional file. 2B]. We also showed that silkworm-derived CR-proBDNF is capable of promoting apoptosis of cultured CGNs. These results are conceptually parallel to previous reports using recombinant proBDNF derived from mammalian cells [7,8].
The extent of the opposite actions of proBDNF and mat-BDNF in the CNS has yet to be determined extensively. Intriguingly, the present study demonstrated that proBDNF is capable of reducing the extent of neurite branching in BFCNs (Fig. 4), which are a well-known population of p75 NTR -highly expressing neurons in the adult brain [24]. However, the morphological action is inconsistent with that of a previous report demonstrating that BF neurons exhibited apoptosis in response to proNGF and proBDNF in culture [8]. One explanation for such a discrepancy is that there were likely differences in maturation of the neurons in culture dishes. We applied CR-proBDNF to BFCNs cultured for more than 2 weeks. A previous report demonstrated that delayed application of matBDNF promoted neuronal differentiation rather than cell survival of the neurons [28]. Since this effect of CR-proBDNF was observed both in serum-containing and serum-free medium (Fig. 4G), we have reasoned that proBDNF may act to the BFCNs in a direct manner. Another interesting finding was that proBDNF reduced proBDNF reduces dendritic spine density in hippocampal neurons Figure 5 proBDNF reduces dendritic spine density in hippocampal neurons. Cultured neurons were maintained for 3-4 weeks and treated with the indicated reagents (50 ng/ml) for 2 (A, B, and D) or 3 days (C). In A and B, neurons were cultured in serum-containing medium. (A) CR-proBDNF reduced the density of DiI-labeled spines. Representative images of DiI-labeled neurons treated with the indicated reagents for 2 days (left). Summary of spine density (middle) and length (right). Data were collected from 52 (Mock), 59 (matBDNF), and 38 (CR-proBDNF) independent cells (non-parametric test, *P < 0.05, compared to Mock). Note that matBDNF increased spine density, while proBDNF reduced spine density (arrows and arrowheads). (B) proBDNF promoted the shrinkage of phalloidin-labeled spines and p75 NTR was involved in the proBDNF action. Application of a functionally blocking antibody against p75 NTR , AB1554 was performed and fixed neurons were stained with an anti-MAP2 antibody and FITC-labeled phalloidin. Representative images of the double-stained neurons (left) and a quantitative analysis of phalloidin-labeled spiny structures (right) are shown, n = 30-40 independent cells from three independent coverslips. t-test, *P < 0.05, **P < 0.01, compared to Mock (100% as control). (C) CR-proBDNF decreases spine density in low-density hippocampal neurons cultured in serum-free medium. The cultures were incubated with CR-proBDNF in serum-free medium for 2 days and double-stained with an anti-MAP2 antibody and FITC-labeled phalloidin. n = 24-32 cells from three independent coverslips. t-test, *P < 0.05, **P < 0.01, compared to Mock. (D) CR-proBDNF decreased spine density in hippocampal slices. Representative images of DiI-labeled neurons at low and high magnification are shown (left). A decrease in spine density was observed in CR-proBDNF-treated slices (arrow heads). Quantification of spine density (right). Data were collected from 7 (Mock) and 6 (CR-proBDNF) independent slices. Note that CR-proBDNF-treated slices showed a decrease in spine density when compared with matBDNF. t-test, *P < 0.05. Scale bar, 10 μm (A, B, and D). dendritic spine density in mature hippocampal neurons: hippocampal neurons treated with CR-proBDNF exhibited a marked shrinkage of spines, both in dissociated culture and in slice culture. This new finding leads us to formulate several important suggestions. First, matBDNF and proBDNF may play opposite roles in structural plasticity: matBDNF and proBDNF may contribute to the augmentation and elimination of synaptic connections, respectively (Fig. 5). Second, it was recently demonstrated that low-frequency stimulation induced predominantly the secretion of endogenous proBDNF from cultured hippocampal neurons [42]. Short-term exposure of CA1 synapses to proBDNF facilitates LTD [15]. We showed that CR-proBDNF, which decreased the density of dendritic spines in hippocampal neurons, also reduced the amplitude of mEPSCs (Fig. 6). In addition to these studies using dissociated hippocampal neurons on the feeder layer of glial cells, proBDNF decreased spine density in a low-density culture and the effect was significant even at subnanomolar concentration (Fig. 5C). The treatment with CR-proBDNF did not affect the number of MAP2-positive neurons in the culture dishes [Additional file 3]. These additional data support the notion that proBDNF acts on spine structures in a direct manner.
However, the mechanism underlying proBDNF-mediated induction of dendritic spine shrinkage remains unclear. Immunohistochemical study demonstrated that, after 14 d in vitro, p75NTR immunoreactivity as punctate staining around the cell body as well as dendrites, with distribution very similar to that of PSD95 immunoreactivity [15]. Zagrebelsky et al. reported a crucial role of p75 NTR on spine formation by gain-of function and loss-of function studies [44]. In the present study, the inhibitory action of CR-proBDNF on spine density was rescued by a p75 NTR functional blocking antibody (Fig. 5B), suggesting that proBDNF/p75 NTR signaling reduces spine density of hippocampal neurons. It was reported that the transport of the signaling complex that contains the p75 NTR receptor and activated Rap1 plays a role in the effect of myelinassociated glycoprotein (MAG), which is a well-characterized axon growth inhibitor [45]. During LTD, Rap mediates the NMDA-R-dependent removal of synaptic AMPA receptors [46]. Furthermore, LTD induces the shrinkage of dendritic spines via the action of cofilin [47]. We also found that the CR-proBDNF-dependent morphological change of dendritic spines in cultured hippocampal neurons appeared in 30 min (data not shown). These recent reports, together with our findings, may provide several clues for the elucidation of the molecular mechanism underlying the proBDNF-dependent synaptic depression.

Conclusion
The present study suggested that the function of proBDNF in the CNS is not limited to apoptosis and consolidation of LTD [10], and that in a cell type-specific fashion proBDNF affects dendritic growth, spine growth, and cell survival of CNS neurons. Very recently, it has been demonstrated that proBDNF eliminated presynaptic terminals using Xenopus nerve-muscle co-cultures [48]. Hence proBDNF itself and the proteolytic cleavage of proBDNF could play important roles in the regulation of BDNF signaling in the architecture of the nervous system.
Wistar ST rats and C57BL/6 mice were from NIPPON SLC (Hamamatsu, Japan). B6C3F1 mice were from the Charles River Laboratory (Kanagawa, Japan). The p75 NTR knockout mice (ngfr tm1Jae ) were from The Jackson Laboratory (Bar Harbor, ME). All animal experiments were strictly in accord with the protocols approved by the Institutional Animal Care and Use Committee of the AIST.

Primary cultures of CNS neurons and application of the indicated reagents
Primary cultures of dissociated rat and mouse CGNs were prepared from the cerebella of P9 rats and P6-7 p75NTR -/ mice, respectively, according to a previous report [27] with some modifications. Briefly, cells were dissociated using a plastic pipette after digestion with papain (90 units/ml, Worthington, Lakewood, NJ) at 37°C and were then cultured in minimum essential medium (MEM, Gibco, Carlsbad, CA) that contained 5% fetal bovine serum (FBS), 5% horse serum, to a final cell density of 5 × 10 5 cells/cm 2 on polyethyleneimine-coated culture plates. After culture for 1 day in a humidified CO 2 (5%) incubator, the medium was changed to Minimum Essential Medium (MEM) that contained 26 mM potassium (HK) and was supplemented with 5% HS and 20 μM AraC. After 4 days in culture, cells were placed in serum-free MEM containing 5.4 mM potassium (LK) or HK-MEM, for experimentation. In separate cultures of CGNs, the percentage of GFAP-positive glial cells was 4.3 ± 0.72% (n = 4 independent culture dishes).
Dissociated BFCN cultures were prepared using E20 rats (Wistar ST) as described [49,50], with a few modifications. Briefly, the basal forebrain area containing the septal nuclei was dissected and digested by papain/DNaseI. Dissociated neurons were seeded onto polyethyleniminecoated 8-well chamber slide (Nunc, Rochester, NY). To maintain the cell survival, serum-containing medium composed of 5% FBS, 5% heat-inactivated HS, and 90% Dulbecco's minimum essential medium (DMEM, Invitrogen, Carlsbad, CA) was used and cells were cultured to a final density of 3 × 10 5 cells/cm 2 on polyethyleneiminecoated culture plates. The medium was changed every 7 days. After 14 days in culture, the cells were treated with the indicated reagents in serum-containing medium, described above. In this preparation, GFAP-positive glial cells formed a feeder layer under the neurons (data not shown). In separate experiments (Fig. 4G), BFCNs were maintained and treated with the indicated drugs in serumfree defined medium containing 5 μg/ml transferrin, 5 μg/ ml insulin, 20 nM progesterone and 20 μM AraC [29]. The medium was changed every 3 days. Then, in sister cultures, the percentage of GFAP-positive glial cells was calculated to be 11.4 ± 2.1% (n = 3 independent culture dishes).
For quantitative analysis of dendritic spine density, dissociated hippocampal neurons were prepared according to previous reports [37,51] with some modifications. For the experiments of Figs. 5A and 5B, hippocampi were dissected from embryonic day 18-20 rats and treated with 0.125% trypsin for 15 min at room temperature. The dissociated neurons were plated on polyethylenimine (0.01%)-coated glass coverslips (Matsunami, Osaka, Japan) at a density of 6-8 × 10 4 cells/cm 2 . The neurons were first cultured in DMEM containing 5% FBS and 5% HS. After 4 d in culture, AraC (2 μM) was added for 24 h to inhibit proliferation of non-neuronal cells. The neurons were then maintained in DMEM containing 10% HS for up to 3 weeks. Half of the medium was changed twice weekly. GFAP-positive glial cells formed a feeder layer under the neurons (data not shown). In separate experiments (Fig. 5C), dissociated neurons (2 × 10 4 cells/cm 2 ) were plated on polyethylenimine (0.01%)-coated plastic Petri-dishes (Corning Costar, Corning, NY) and co-cultured with confluent glial cells cultured on insert cup but separated by a microporous membrane filter and co-cultured in a common medium (B27/Neurobasal medium, Invitrogen) supplemented with 10 μM AraC. The medium was changed every 3 days. The biological assay of Fig. 5C was performed for 3-4 week cultured neurons. For immunoprecipitation (Fig. 1F), primary cultures of cortical neurons (3 × 10 5 cells/cm 2 ) were prepared from E20 rats [52]. Three-day cultured cerebral cortical neurons were infected with wild-type or mutant BDNF-expressing Sindbis viruses for 12 h and maintained in serum-containing CR-proBDNF attenuates amplitude of functional synaptic transmission in cultured hippocampal neurons Figure 6 CR-proBDNF attenuates amplitude of functional synaptic transmission in cultured hippocampal neurons. Cells were cultured as described in Fig. 5A and treated with the indicated reagents for 3 days. A patch-clamp technique in the whole-cell configuration was used to record mEPSCs from pyramidal-shaped neurons and all data were statistically evaluated. (A) Representative mEPSCs. (B) Amplitude (left) and frequency (right) of mEPSCs. Note that proBDNF significantly decreased the amplitude, but not the frequency, of mEPSCs. Data were collected from 27-31 independent cells (non-parametric test, *P < 0.05, significantly different from Mock).
prodomain of human BDNF. The antibody was affinitypurified against the antigenic peptide (Aves Labs, Tigard, OR).

Western blot analyses
The lysates of cultured neurons were prepared as described [52]. Briefly, cultured cells were rinsed three times with ice-cold PBS and were quickly lysed in cold lysis buffer containing 50 mM Tris-HCl (pH 7.4), 1 mM EDTA, 150 mM NaCl, 10 mM NaF, 1 mM Na 3 VO 4 , 1% Triton X-100, 10 mM Na 2 P 2 O 7 , 100 μM phenylarsine oxide, and 1% protease inhibitor cocktail (Complete Mini, Roche Diagnostics, Hertforshire, UK). Lysates were boiled for 5 min at 100°C and were then sonicated. The protein concentration of the supernatants was determined using the bovine serum albumine (BSA) standards from Pierce Biotechnology. The lysates were resolved by SDSpolyacrylamide gel electrophoresis (SDS-PAGE). Western blot protocols were according to Suzuki et al. (2004) [52] but modified for detecting signals more specifically. Samples were sonicated in an equal volume of SDS sample buffer (0.125 M Tris-HCl, pH 6.8, 20% glycerol, 4% SDS, and 10% 2-mercaptoethanol), heated at 100°C for 3 min, and resolved by SDS-PAGE. Samples were electrotransferred to polyvinylidene fluoride membranes (Immobilon P membrane, Millipore, Bedford, MA), which were blocked in Tris-buffered saline (TBS) containing 0.2% Tween-20 (TBS-T) and 3-5% BSA or Block Ace (Dainippon pharmaceutical, Osaka, Japan) and were then incubated with the indicated primary antibodies in TBS-T containing 0.3-0.5% BSA (TBS-TB) or 1-10% Block Ace at room temperature for 60-90 min. After washing 3 times with TBS-T, membranes were incubated with peroxidase-conjugated secondary antibodies in TBS-TB at room temperature for 30 min, and then washed 3 times with TBS-T. The chemical luminescence signals were detected using the ImmunoStar Reagents (Wako) or the SuperSignal West Femto Maximum Sensitivity Substrate (Pierce Biotechnology). The exposure time was adjusted so that the intensities of the bands were within the linear range. For quantification of the relative amounts of proBDNF and matBDNF, the blots were scanned and the images were converted to Tiff files and quantified using the NIH Image software (ImageJ version 1.37v). The relative amounts of proBDNF and matBDNF were obtained by normalizing the values to total BDNF values. Signals for proBDNF and matBDNF in a given lane of the blot (after subtraction of the background in the same lane) were added to obtain the value for total BDNF and were normalized to the signal of the GFP band. The data were averaged and were presented as mean ± standard error of the mean (SEM).

Immunoprecipitation
Protein G Sepharose (15 μl gel, Amersham Pharmacia, Piscataway, NJ) was added to the conditioned culture medium of the dissociated cerebral cortical neurons transduced with Myc-tagged CR-proBDNF (R125M/R127L-BDNF) constructs and the mixture was rotated at 4°C for 60 min. Anti-c-Myc monoclonal antibody-conjugated agarose beads (20 μl gel, BD, San Jose, CA) were added to the supernatant after removal of the Protein G Sepharose. Incubation of the lysates with the beads was performed at 4°C overnight. The immune complexes were pelleted and immunoblotted using antibodies against BDNF or proBDNF.

Cell viability assay
Cell viability was quantified by counting the number of dead cells using DAPI or Hoechst staining [54], and by measuring LDH released into the medium using the kit (Wako) [27]. For the LDH assay, culture media were centrifuged at 10,000 g for 10 min and 50 μl of dye buffer were added to 10-20 μl of each supernatant. The mixtures were incubated at 37°C for 15 min and 50 μl of terminating buffer was added to each sample. The absorbance was photometrically determined at a wavelength of 560 nm. For DAPI and Hoechst staining, cells were fixed using 4% paraformaldehyde/PBS for 20 min and were then stained with 100 ng/ml DAPI or 1 μg/ml Hoechst 33258 in PBS for 15 min. The ratio of condensed to intact nuclei was quantified based on observations using a fluorescence microscope.

AChE histochemistry and quantitative analysis
Fixation and immunohistochemistry of cultured cholinergic neurons were performed according to previously described methods [49,50]. AChE histochemistry was performed as described [55]. The phase-contrast images of AChE-positive cholinergic fibers and the cell body with a diameter of > 2.5 μm were acquired using a cooled CCD camera (MRc5, Carl Zeiss, Swiss) mounted on a Nikon microscope (TE230, Nikon, Japan). Sholl analysis was performed to determine the number of AChE-positive neurites at a distance of 10 μm from the soma [30]. The number of AChE-positive cholinergic fibers that extended from the cell body were counted according to Suzuki et al. (2004) [52]. We also quantified cholinergic fiber density using the following method: images of AChE-positive cholinergic fibers were captured using the optical conditions described above. First, the background intensity was determined by averaging the values obtained from three non-stained areas. The maximal threshold of fiber intensity was defined as 70% above the background intensity. The total intensity of the AChE-positive fibers was determined in an optical field (500 × 500 pixels/field) and was divided by the number of AchE-positive cholinergic neurons in the same field. Data were collected from four independent fields in a single chamber. All experiments were performed in a blind fashion.

Quantitative analysis of dendritic spine density
Dendritic spine density was quantified according to previous reports [31,33,56], with a few modifications. Cultured hippocampal neurons were treated with the indicated reagents for 2-3 days and were then fixed using 4% paraformaldehyde/PBS for 20 min. DiI solution (20 mg/ml) was used for labeling dendritic spines. For visualization of F-actin accumulated in the dendritic spines, fixed cells were immunostained with an anti-MAP2 antibody [52], followed by 1 h incubation with TRITC-conjugated phalloidin (1:750 in PBS). We used confocal microscopy (FV-1000, Olympus, Tokyo, Japan) with a 60 × objective lens to capture stacked images of dendritic spines along the z-axis of the imaged cell. The spatial resolution of captured images was 0.1 μm/pixel. The number of spines was quantitatively evaluated at the proximal side of the second dendrite via manual counting using Metamorph software (Universal Imaging, Downingtown, PA). Dendritic spines in slice cultures were labeled by rhodamine-dextran injection (6 mg/ml in a patch pipette) using a single-cell electroporator [57].

Statistical analyses
Typically, significance was determined using the Student's t-test after testing for normal distribution. In several multi-bar figures, significance was analyzed by ANOVA followed by a post-hoc comparison. In all graphs, data were presented as means ± standard error of the mean (SEM).
MK. Wrote the paper: HK KK BL MK. All authors read and approved the final manuscript.