Requirement of TrkB for synapse elimination in developing cerebellar Purkinje cells

Abstract

The receptor tyrosine kinase TrkB and its ligands, brain-derived neurotrophic factor (BDNF) and neurotrophin-4/5 (NT-4/5), are critically important for growth, survival and activity-dependent synaptic strengthening in the central nervous system. These TrkB-mediated actions occur in a highly cell-type specific manner. Here we report that cerebellar Purkinje cells, which are richly endowed with TrkB receptors, develop a normal morphology in trkB-deficient mice. Thus, in contrast to other types of neurons, Purkinje cells do not need TrkB for dendritic growth and spine formation. Instead, we find a moderate delay in the maturation of GABAergic synapses and, more importantly, an abnormal multiple climbing fiber innervation in Purkinje cells in trkB-deficient mice. Thus, our results demonstrate an involvement of TrkB receptors in synapse elimination and reveal a new role for receptor tyrosine kinases in the brain.

Introduction

Activation of the receptor tyrosine kinase TrkB by its ligands, the neurotrophins brain-derived neurotrophic factor (BDNF) and neurotrophin 4/5 (NT-4), is of decisive importance for the development, survival and plasticity of the nervous system (Bibel and Barde, 2000; Huang and Reichardt, 2001). Prominent functions of TrkB signaling include the control of neurite outgrowth (Segal et al., 1995; Yacoubian and Lo, 2000; McAllister, 2001) and synapse formation (Martínez et al., 1998; Murphy et al., 1998). For example, TrkB is involved in the refinement of neuronal networks, ensuring the specificity of synaptic connections in the visual system (Cabelli et al., 1995).

The importance of TrkB and/or BDNF for neuronal development differs between brain regions (Klein et al., 1993; Chakravarthy et al., 2006). From the analysis of mutant mice it is known that the axonal and dendritic arborizations of granule and Purkinje cells (PCs) are reduced by ~20% in the absence of TrkB (Minichiello and Klein, 1996; Rico et al., 2002) and even more in the absence of BDNF (Schwartz et al., 1997; Carter et al., 2002). By contrast, the gross morphology of the cerebellum is not disturbed and the total number of PCs and granule cells is not substantially altered in mice with a general or cerebellum-specific deletion of trkB (Minichiello and Klein, 1996; Rico et al., 2002). In the latter, the number of GABAergic synapses is dramatically reduced, while the morphology of glutamatergic synapses appears to be normal (Rico et al., 2002).

The development of PCs consists of four partially overlapping phases: mitosis, migration, apical dendrite formation and synapse formation. Mitosis of precursor cells is completed during the second week of the embryonic phase (Uzman, 1960; Miale and Sidman, 1961). The newly formed PCs migrate from the ventricular zone along processes of radial glial cells towards their definite location between the molecular and internal granular layer (Miale and Sidman, 1961). Initially, they form a multi-layered structure (Mason et al., 1990). By the end of the first postnatal week, however, the PCs are aligned in a strict monolayer and then form a single apical dendrite which is oriented towards the pial surface (Altman, 1972). Finally, synapses are formed and segregate for the different kinds of excitatory afferents (Larramendi and Victor, 1967; Altman, 1972). As in other types of neurons, elimination of inappropriate synapses is an important step in establishing the mature pattern of innervation of PCs (Crépel et al., 1981). While the role of TrkB in synapse formation and maintenance is well established (Seil and Drake-Baumann, 2000; Luikart et al., 2005), it is not clear whether it is also involved in synapse elimination.

The cerebellar function critically depends on the elimination of redundant climbing fiber (CF) inputs to PCs during development (Aiba et al., 1994; Kano et al., 1995; Kano et al., 1997). At birth, PCs are innervated by several CFs. Around postnatal day 7 (P7), synapse elimination starts and two weeks later virtually all PCs are mono-innervated (Crépel et al., 1981; Hashimoto and Kano, 2003). This process requires the establishment of intact parallel fiber (PF)/PC synapses (Crépel et al., 1981).

Here we demonstrate that the morphology of the cerebellum, in particular that of the PCs develops normally in mice lacking all isoforms of TrkB, although TrkB is abundantly expressed in PCs and granule cells (Yan et al., 1997) and in the inferior olive (Riva-Depaty et al., 1998). Surprisingly, the developmental CF/PC synapse elimination is impaired in the trkB^-/- mice.

Results

TrkB receptors exist as full-length (TrkB^FL) and truncated isoforms (TrkB-T1 and TrkB-T2) (Middlemas et al., 1991). Although only the TrkB^FL isoform has tyrosine kinase-activity, truncated TrkB receptors are also biologically active (Baxter et al., 1997). For the present study, we have used a mouse line devoid of all TrkB isoforms throughout the whole body (Rohrer et al., 1999), that is distinct from a mouse line lacking only TrkB^FL (Klein et al., 1993; Minichiello and Klein, 1996) and from a cerebellum-specific trkB mutant (Rico et al., 2002). The general trkB mutants develop a severe phenotype (Klein et al., 1993; Rohrer et al., 1999). Most mutant mice die during the first two postnatal weeks. At birth, their body weight is nearly normal, but the developmental weight gain is reduced in the absence of trkB. As a result, the weight of the mutant mice is often reduced compared to their control litter mates. Remarkably, the gross morphology of the cerebellum has been shown to be relatively normal in trkB^FL and cerebellum-specific trkB mutants (Minichiello and Klein, 1996; Rico et al., 2002). Consistently, foliation of the cerebellar cortex appears to be normal in the trkB^-/- mouse line that we used here.

Cerebellar morphology and TrkB-expression

A clear and early hallmark of cerebellar development is the arrangement of the PCs in a strict mono-layer by the end of the first postnatal week (Altman, 1972). We found that already at P7, the formation of the PC monolayer was almost completed in both wild type and in trkB^-/- mice (Fig. 1b). We studied the anatomical distribution of Trk receptors at P7 using immunohistochemistry. We employed pan-Trk antibodies, because the commercially available antibodies against TrkB were not specific in our hands (Suppl. Fig. 1). As shown in Fig. 1b, calbindin is present in PCs and in the white matter. Trk receptors are present in the granular and PC layer in the wild type mice. The diffuse staining in trkB^-/- mice is probably caused by TrkA and TrkC, which are both expressed in the neonatal cerebellum (Moore et al., 2004). There is no evidence for any substantial upregulation of TrkA and/or TrkC upon trkB deletion. No differences between wild type and trkB^-/- mice were found regarding the number and general morphology of PCs (Fig. 1b).

Fig. 1.

Normal layer formation in cerebella of trkB^-/- mice. (a) Genotyping of experimental animals by PCR. (b) Immunohistochemistry of Trk receptors and calbindin in cerebella from P7 trkB^+/+ and trkB^-/- mice. Trk receptors were stained using a rabbit pan-Trk antibody and calbindin using a monoclonal calbindin antibody. The pan-Trk antibody demonstrated the low abundance of other Trk receptors in trkB^-/- mice. Scale bar = 50 μm.

PC morphology

Once PCs are settled in a mono-layer, they undergo a dramatic change in appearance. This development is particularly prominent between P10 and P14, when PCs retract their perisomatic dendrites and simultaneously develop a very large, single apical dendritic tree (Altman, 1972). We performed high-resolution two-photon imaging of PCs filled with the fluorescent dye Alexa Fluor-488 (Fig. 2). Wild type PCs were studied at P10 and P14 in order to determine which morphological features are under developmental control. Wild type PCs at both ages were compared with P14 trkB^-/- PCs to determine the effect of TrkB-signaling on the development of the PC morphology. Earlier, qualitative observations of PCs in trkB^FL-/- mice (Minichiello and Klein, 1996) and in cerebellum-specific trkB^-/- mice (Rico et al., 2002) showed that these PCs formed single, well-developed apical dendritic trees. We extend these findings and demonstrate that neither the truncated TrkB-isoforms nor TrkB expressed outside the cerebellum, e.g. in the CFs, influence the qualitative appearance of PC dendritic trees (Fig. 2a).

Fig. 2.

Normal PC morphology in trkB^-/- mice. (a) Two-photon images of PCs filled with Alexa Fluor-488 in acute slices. P10 PCs have a small apical dendritic tree and several perisomatic dendrites (left). At P14, PCs display extensive branching in both trkB^+/+ (middle) and ^-/- mice (right). No perisomatic dendrites can be observed anymore. (b) Digitized reconstructions of PCs from the images in (a) that were used for the quantitative analysis of the PC morphology. The colors reflect the centrifugal branch order analysis, each order having its own color. (c) Bar graphs depicting mean values of morphometric parameters resulting from the analysis of the reconstructed PCs. From left to right: number of stem dendrites originating at the soma, apical length, total dendritic length, somatodendritic area, number of bifurcations, tree asymmetry index. (d) Circles with increasing diameter were placed on the dendritic trees starting in the center of the soma (top). The number of intersections with dendritic elements was calculated for each circle (bottom). Shown are the mean values per circle. The error bars depict the SEM per 10 μm. The Sholl distribution of the P14 trkB^+/+ PCs reflects the larger dendritic size of these cells compared to the P10 trkB^+/+ PCs and also the P14 trkB^-/- PCs (two-way ANOVA). (e) Cluster analysis using 17 cell morphological parameters according to Ward's method revealed that the P10 trkB^+/+ PCs formed their own cluster (light blue). The P14 trkB ^+/+ (dark blue) and ^-/- (red) PCs did not segregate into separate clusters. Unless otherwise stated, the mean ± s.d. is shown. Significant difference with the P10 trkB^+/+ PCs is depicted by * and between P14 trkB^+/+ and trkB^-/- PCs by #. Unless otherwise stated, we used ANOVA with Tukey's post-test. The threshold for significance was set at 0.05. For this analysis, 10 P10 trkB^+/+ PCs, 13 P14 trkB^-/- and 7 trkB^-/- PCs were used.

In order to reveal possible quantitative differences in the size, complexity and shape of PCs between trkB^+/+ and ^-/- mice we explored the PC morphology in both genotypes in more detail. The results of this analysis are summarized in Table I. The size of the cell body is slightly smaller in P14 trkB ^-/- PCs than in those of their wild type litter mates (Table I). However, PC somata of trkB-deficient mice at P14 display a spherical shape similar to wild type P14 somata and thus have similar values of circularity (Table I). In sagittal cerebellar slices PC axons rarely are located in the plane of sectioning. For a characteristic parameter of axonal size we chose the average of the axon diameter measured at 5, 10 and 15 μm distance from the soma. No significant difference in axonal thickness was found between the three groups of PCs tested (Table I).

Table I.

PC morphometry

	trkB+/+ P10	trkB+/+ P14	trkB-/- P14
n	10	13	7
Decreasing during development
No. of primary dendrites	6.2 ± 3.5	1.2 ± 0.6*	1.1 ± 0.4*
Constant during development
Axon diameter (μm)	1.1 ± 0.1	1.5 ± 0.2	1.2 ± 0.1
Soma perimeter (μm)	61.5 ± 6.6	63.8 ± 10.3	60.8 ± 8.1
Tree asymmetry index	0.47 ± 0.12	0.48 ± 0.05	0.51 ± 0.06
Distance of max. ramification	0.62 ± 0.21	0.62 ± 0.15	0.66 ± 0.10
Rostrocaudality	0.82 ± 0.14	0.90 ± 0.15	0.92 ± 0.13
Mediolaterality	0.71 ± 0.18	0.72 ± 0.10	0.71 ± 0.21
Increasing during development
Soma area (μm2)	200 ± 30	272 ± 82*	240 ± 38
Soma circularity	0.67 ± 0.08	0.82 ± 0.03*	0.82 ± 0.09*
Total dendritic length (mm)	0.43 ± 0.13	2.64 ± 0.92*	2.07 ± 0.64*
Somatodendritic area (μm2)	627 ± 209	4787 ± 2445*	5159 ± 1727*
Apical length (μm)	43.5 ± 6.2	132.2 ± 62.2*	107.9 ± 21.0*
No. of bifurcations	36 ± 9	179 ± 50*	163 ± 43*
No. of dendritic elements	78 ± 15	360 ± 99*	328 ± 86*
No. of dendritic endings	42 ± 7	328 ± 86*	164 ± 43*
Max. angular branch order	5.1 ± 1.0	7.8 ± 1.0*	7.4 ± 0.8*
Max. centrifugal branch order	13.2 ± 4.3	18.4 ± 2.7*	20.6 ± 3.9*
SA max. no. of intersections	11.8 ± 3.8	26.6 ± 7.5*	23.6 ± 4.6*
SA average no. of intersections	5.6 ± 1.6	12.0 ± 3.4*	11.1 ± 2.7*
SA sum of intersections	226 ± 73	1623 ± 1092*	1127 ± 314*

PC morphometric parameter values. Shown are the means ± s.d. (except axon diameter: means ± SEM).

^*p < 0.05 as compared to wild type P10. None of the parameters differed significantly between trkB^+/+ and ^-/- (ANOVA with Tukey post-test). SA = Sholl analysis.

For a detailed characterization of dendritic trees in the presence and absence of TrkB we traced the images of Purkinje cells (Fig. 2b) and obtained reconstructions suitable for automated analysis (Figs. 2c-e). At P10, the PCs had 5.2 ± 3.5 (mean ± s.d.) perisomatic dendrites. Neither the trkB^+/+ nor the trkB^-/- PCs show perisomatic dendrites at P14 (Figs. 2a-c). Instead, almost all PCs at P14 have one stem dendrite, irrespective of the genotype. 1/13 PCs in trkB^+/+ mice had three and 1/7 PCs in trkB^-/- had two apical dendrites. Owing to a pronounced growth of the apical dendrite between P10 and P14, the length of the PC increases more than 3-fold during this period, whereas the total dendritic length increases even more than 6-fold (Fig. 2c). A similar growth is observed in the trkB mutants, although both parameters are on average ~20% smaller than in wild type mice at P14 (Fig. 2c), which is in line with previous reports (Minichiello and Klein, 1996; Rico et al., 2002). The somatodendritic projection area calculated from the number of pixels in the z-projections (Fig. 2a) is similar for trkB^+/+ and ^-/- PCs at P14 (4787 ± 2445 vs. 5159 ± 1727 μm²; Fig.2c).

Concurrent with the growth of the dendritic tree is an increase in its complexity. From P10 to P14 the number of dendritic branch points increases from 36 ± 9 to 179 ± 50 at P14 in wild type PCs. The latter value is similar in P14 trkB^-/- PCs (163 ± 43 (means ± s.d., n.s. (ANOVA)), Fig. 2c). The analysis of the dendritic branching pattern revealed little difference between P14 ^+/+ and ^-/- PCs. The maximal concentric branch order of PC dendrites at P14 was 18.4 ± 2.7 and 20.6 ± 3.9 in wild type and trkB^-/- mice, respectively (n.s., ANOVA). These values are in clear contrast to the maximal concentric branch order at P10, which was 13.2 ± 4.3 (Fig. 2b, Table I). The lateral distribution of dendritic elements along the apical dendritic tree was estimated using the tree asymmetry index (Van Pelt et al., 1992). We found no significant changes in the tree asymmetry index, neither during normal development (P10 trkB^+/+: 0.47 ± 0.12; P14 trkB^+/+: 0.48 ± 0.05, respectively) nor as a result of trkB deletion (P14 trkB^-/-: 0.51 ± 0.06 (means ± s.d., tested with ANOVA, Table I).

Figure 2d demonstrates implementation and results of the concentric Sholl analysis (Sholl, 1953) performed on PCs at P10 and P14 in the wild type and P14 in trkB^-/- mice. In addition to the large difference between wild type PCs at P10 and P14 owed to the rapid growth and ramification during this time interval there is also a small difference between P14 wild type and trkB^-/- PCs in the distribution of intersections as a function of the radius of the Sholl circles. This difference vanishes when a parameter that is independent of cell size is used for comparison. We divided the radius of the circle with the largest number of intersections by the radius of the most distal circle. This normalized distance of maximal ramification is similar for all three groups analyzed (Fig. 2d, Table I). The segmental variation of the Sholl analysis describes the radial distribution of dendritic elements around the soma, which was found to be highly similar between wild type and trkB^-/- PCs (data not shown). The only exception proved to be a small fraction of the dendrites still located at the basal side of the soma at P10. Accordingly, the proportion of dendritic elements in the medial regions of the dendritic tree reflected by its mediolaterality is not significantly different (Table I). Neither was the distribution of dendritic branches between the apical and basal side of the PC soma, represented by the rostrocaudality, significantly different between wild type and trkB^-/- PCs at P14. Dendritic branches in all three groups of cells are similarly distributed between apical and rostral parts of the dendritic tree as demonstrated by their values of rostrocaudality (Table I). Finally, we employed cluster analysis on the 17 morphometric parameters listed in Table I (Fig. 2e). The results clearly demonstrate that the PCs in wild type mice at P10 are morphologically very different from the P14 PCs. In contrast, P14 PCs do not segregate into two different groups according to their genotype. We conclude that P14 wild type and trkB^-/- PCs are morphologically highly similar, despite the finding that the PCs of P14 trkB^-/- mice are up to 20% smaller than those of their wild type litter mates. Thus, trkB deletion does not interfere significantly with normal PC development.

Dendritic spines

We selected dendrites from the 2-photon recordings of the PCs according to the criteria mentioned in Methods. The experimentator was unaware of the origin of the dendrites to avoid experimental bias. We found that the density of putative CF spines showed a developmental decrease between P10 and P14 in PCs from wild type mice (Table II). At P14, however, no significant difference between wild type and trkB^-/- PCs could be detected (Fig. 3). With respect to the density of PF spines, we found similar values for all three groups of PCs tested (Table II). This indicates that trkB deletion does not play a major role in the formation of PC dendritic spines, neither for CF nor PF spines.

Table II.

Dendritic spine density

	trkB+/+ P10	trkB+/+ P14	trkB-/- P14
n	9	5	6
Proximal spines (μm-1)	0.9 ± 0.3	0.3 ± 0.2*	0.3 ± 0.1*
Distal spines (μm-1)	1.4 ± 0.1	1.0 ± 0.2	1.5 ± 0.3

Presented are the mean ± s.d. (proximal spines) or the mean ± SEM (distal spines).

^*p < 0.05 (as compared to trkB+/+ P10; ANOVA with Tukey post-test).

Fig. 3.

Normal spine densities in trkB^-/- mice. Spiny branchlets are covered with spines that are mainly innervated by PFs. The density of the PF spines is similar in P14 trkB^+/+ (a) and ^-/- PCs. in trkB^+/+ PCs (b). Top: z-projections of three or four planes. Bottom: surface renderings of the complete z-stacks.

Development of GABAergic synapses

In order to explore how the deficiency of TrkB affects the establishment of GABAergic synapses onto PCs, we recorded mIPSCs from PCs in acute slices taken at P7 (Fig. 4). Two mice of each genotype were used for the analysis of the mIPSCs. The mIPSCs could be completely blocked by the application of bicucculine (10 μM, n = 4, data not shown) and can therefore be considered as mediated by GABA_A receptors. We found a reduction in the amplitude of GABAergic mIPSCs together with a decrease in frequency. In addition, we found that the mIPSC-decay time constant (τ_decay) is longer in the absence of TrkB (Fig. 4c, Table III). Our results therefore suggest that the absence of TrkB-signaling slows the formation and maturation of GABAergic synapses (Bao et al., 1999; Seil and Drake-Baumann, 2000).

Fig. 4.

Delayed maturation of GABAergic synapses in trkB ^-/- mice. (a) Recordings of mIPSCs in PCs in cerebellar slices from P7 trkB^+/+ and ^-/- mice in the presence of tetrodotoxin (500 nM), CNQX (20 μM) and APV (25 μM). (b) Superimposed example mIPSCs at a more expanded time scale. (c) Cumulative histograms, constructed from 75 mIPSCs per experiment, depicting the 10-90% rise time, the amplitude, the ^τdecay and the interval of the mIPSCs. All distributions were significantly different (two-way ANOVA). The mIPSCs in the trkB^-/- PCs were smaller and decayed slower, while they occurred less frequently than the trkB^+/+ mIPSCs. From both genotypes, seven PCs were used for the recordings of mIPSCs.

Table III.

GABAergic mIPSCs

	trkB+/+ P7	trkB-/- P7
n	7	7
Rise time (ms)	0.8 ± 0.1	0.8 ± 0.2
Amplitude (pA)	145.5 ± 19.4	119.7 ± 19.3
τdecay (ms)	3.9 ± 0.2	5.1 ± 0.6
Interval time (s)	1.2 ± 0.2	3.5 ± 1.3

Presented are the means ± SEM. None of the parameter values differed significantly between trkB^+/+ and ^-/- mice (Student's t-test).

Excitatory synaptic input to PCs

We tested also the functionality of excitatory synapses by characterizing the fast glutamatergic inputs to the PCs in trkB mutants at P14. As in wild type mice (data not shown), mutant PCs receive both PF and CF inputs, which could be discriminated on the basis of their characteristic short-term plasticity (Konnerth et al., 1990) (Fig. 5a). CF EPSCs have similar kinetics in wild type and trkB^-/- mice (Table IV). In current clamp conditions, CF activation is recorded as a complex spike in the PC, which is accompanied by a transient rise of [Ca²⁺]_i (Fig. 5b). In both genotypes, CF EPSCs are completely blocked by 20 μM CNQX (Fig. 5c), indicating that they are mediated exclusively by AMPA receptors. Thus, basal synaptic transmission at PF and CF synapses is largely normal in trkB^-/- mice.

Fig. 5.

Synaptic responses to parallel and climbing fiber inputs in P14 trkB^-/- mice. (a) Paired-pulse facilitation of PF-EPSCs (left) an paired-pulse depression of CF-EPSCs (right)in trkB^-/- mice. PFs were electrically stimulated in the molecular layer and CFs in the granular layer. (b) CF activation elicits a complex spike in PCs from trkB^-/- mice. This complex spike is accompanied by a rise in [Ca²⁺]_i. The soma was kept out of focus to reduce phototoxic damage. Bar = 100 μm. (c) The CF-EPSC is mediated exclusively by AMPA receptors, since it could be blocked by the addition of CNQX (20 μM). After washing out CNQX for ~10 minutes, the CF-EPSC partially recovered.

Table IV.

CF characterization

	trkB+/+ P14	trkB-/- P14
n	5	7
Rise time (ms)	0.8 ± 0.1	0.7 ± 0.1
τdecay (ms)	5.2 ± 0.5	6.9 ± 1.8
PPD (%)	38 ± 7	51 ± 5

Presented are the mean ± SEM. τ_decay is the weighted decay time constant. None of the parameter values differed significantly between trkB^+/+ and ^-/- PCs (Student's t-test, p>0.1).

Defect in CF synapse elimination

We investigated the degree of multiple CF innervation of PCs in wild type and TrkB-deficient mice. We placed a stimulation electrode in the granular layer and applied stimuli of increasing strength. Because CF EPSCs are evoked in an all-or-none fashion, we interpreted the number of discrete steps in the amplitude of the postsynaptic response as the number of CFs impinging on the PC (Kano et al., 1995; Hashimoto and Kano, 2003). We systematically checked the number of CF synapses at ≥ 2 different stimulation locations in the granular layer. We identified individual CF synapses on the basis of their amplitude under voltage clamp and the complex spike waveform under current clamp. Hence, the same EPSC amplitude stimulated at both locations was interpreted as being due to one afferent CF.

We found a clear reduction in CF elimination between P10 and P14 in wild type mice. In P10 trkB^+/+ mice, we found that only 6/17 (35%) PCs tested where mono-innervated. This fraction increased to 15/18 (83%) PCs at P14 (Fig. 6a). The remaining three PCs at P14 (17%), were innervated by two CFs. In contrast, only 5/19 (26%) PCs from trkB^-/- mice were singly innervated. Most trkB^-/- PCs received two individual CF inputs (53%; Fig. 6b). Occasionally three (3/19) or even four (1/19) afferent CFs were detected. The difference in ratio between mono- and multiple-innervated PCs was significantly different between trkB^+/+ and ^-/- mice at P14 (p < 0.001; Fisher's exact test), but not between P10 trkB^+/+ and P14 trkB^-/- (p> 0.7; Fisher's exact test). Thus, the degree of multiple CF innervation in trkB ^-/- PCs at P14 resembles that of P10 wild type PCs. These results indicate that the process of developmental CF synapse elimination is strongly impaired in the trkB mutants.

Fig. 6.

Abnormal multiple CF innervation in trkB^-/- mice at P14. In P14 trkB^+/+ mice, most PCs are innervated by a single, large CF. (a) Representative recording from a P10 trkB^+/+ PC, showing no response to a weak stimulus (left), a moderately large response to an intermediate stimulus (middle) and a large response to a strong stimulus (right). Analysis of a large range of stimulation intensities reveals that this PC receives input from two individual CFs (bottom). (b) Representative recording from a P14 trkB^+/+ PC, receiving one CF input and of a P14 trkB^-/- PC receiving two individual CF inputs (c). (d) At P10, the large majority of trkB^+/+ PCs is innervated by more than 1 CF. In contrast, at P14, most trkB^+/+ PCs are already mono-innervated. The phenotype of P14 trkB^-/- PCs is very similar to that of P10 trkB^+/+ PCs, demonstrating a clear reduction in the rate of developmental CF synapse elimination.

Discussion

We have investigated the role of TrkB for the development of cerebellar PCs. For this purpose we analyzed mutant mice lacking all isoforms of the TrkB receptor. Because these mouse mutants die within 1-2 weeks after birth, our analysis covered a restricted period of early postnatal development. Nonetheless the results revealed surprising and important new insights. First, TrkB has no obvious contribution for the development of dendrites and spines in PCs. Second, the basic features of glutamatergic excitatory transmission are normal, while GABAergic transmission is slightly impaired. Third, and most strikingly, the normally occurring process of developmental synapse elimination is strongly impaired in TrkB deficient mice.

Development of dendrites and spines in TrkB-deficient mice

In many types of central neurons, the BDNFTrkB system decisively controls dendritic growth and development (McAllister, 2001; Dijkhuizen and Ghosh, 2005). It had been suggested that neurotrophins determine also the elaboration of the dendritic tree of PCs (Schwartz et al., 1997), (but see Shimada et al., 1998 and Adcock et al., 2004). We investigated the development of the PCs in acute cerebellar slices by using two-photon imaging for the analysis of cells that were filled through patch pipettes with the fluorescent dye Alexa Fluor 488. We performed the first detailed morphometric analysis of native PCs devoid of all TrkB isoforms. Our analysis involved PC tracing, reconstruction, branch order determination and Sholl analysis.

The results demonstrate that both the complexity and the shape of the trkB^-/- PC dendritic tree are indistinguishable from those in wild types. This is clearly shown by the analysis of the branching pattern and the Sholl analysis. Also the cluster analysis, performed as a meta-analysis, indicates that there are no obvious differences between PCs of P14 in wild type and trkB^-/- mice. The finding that the dendritic length of the P14 PCs was ~20% smaller in the trkB mutants than in the wild type mice fits to the observation that the total cerebellar diameter was reduced by about ~20%. This result is consistent with earlier observations in trkB^FL-/- mice (Minichiello and Klein, 1996; Rico et al., 2002).

The inspection of the spine in the images that were acquired with two-photon microscopy (Fig. 3) did not reveal obvious differences between wild type and trkB^-/- mice. We found that at P14 the spine density is very similar in wild type and mutant mice suggesting that TrkB is not needed for spine formation and maturation in the cerebellum. This is in contrast to results obtained in the hippocampus, where spine formation was shown to be markedly stimulated by BDNF/TrkB-signaling (Ji et al., 2005; Luikart et al., 2005). The difference between cerebellum and hippocampus might be explained by the quite unique way of spine formation in the cerebellar PCs (Yuste and Bonhoeffer, 2004). Overall, our morphometric analysis firmly establishes that the formation and development of PCs dendrites and spines in vivo are not affected by the complete absence of TrkB.

Glutamatergic and GABAergic synaptic transmission in trkB mutants

The basic features of excitatory synaptic transmission seem to be normal in trkB deficient mice. Thus, the characteristic paired-pulse facilitation at parallel fiber synapses and the paired-pulse depression at climbing fiber synapses are qualitatively similar to what is known from wild type animals (Konnerth et al., 1990). Similarly, transmission at both types of synapses is completely blocked by the AMPA receptor antagonist CNQX (Llano et al., 1991) and the climbing fiber evoked complex spike produces the well-known large dendritic Ca²⁺ transient (Ross and Werman, 1987; Tank et al., 1988).

To estimate the functional status of GABAergic synapses we determined the frequency and kinetics of mIPSCs. We found that in trkB^-/- mice, the mIPSC-frequency is somewhat reduced compared to the wild type. This result may either reflect alterations in presynaptic release probability or, more likely, the presence of fewer GABAergic synapses in the trkB mutants. The latter interpretation is consistent with the earlier finding that the number of bou-tons expressing the GABA-synthesizing enzyme GAD65 is reduced in cerebellum-specific trkB mutants (Rico et al., 2002).

The observed longer decay time-constant of mIPSCs in the PCs from mutant mice might reflect a delay of a well known developmental GABA_A receptor subunit switch. Initially, during the migration and differentiation phase, the predominant α subunit is α3 (Takayama and Inoue, 2004), which is replaced by the α1 subunit in the mature PCs (Laurie et al., 1992; Takayama and Inoue, 2004). This switch in subunit composition has been shown to cause faster decaying IPSCs (Bosman et al., 2002). Thus, the prolonged decay time-constants, together with the reduced frequency of mIPSCs, strongly suggest that the development of GABAergic synapses is impaired in the absence of TrkB-signaling. This is consistent with previous work on cerebellar cultures (Seil and Drake-Baumann, 2000) and on granule cells after bdnf-overexpression (Bao et al., 1999).

A novel mechanism of developmental CF synapse elimination

Three distinct postsynaptic molecular mechanisms are known to control the process of developmental CF synapse elimination. The critical signal proteins corresponding to these mechanisms are (i) the mGluR1 and its downstream partners (Kano et al., 1995, 1997, 1998; Offermanns et al., 1997), (ii) the glutamate receptor GluRδ2 (Hirano et al., 1995; Kashiwabuchi et al., 1995; Hashimoto et al., 2001) and (iii) the predominantly nuclearly expressed Ca²⁺/calmodulin kinase IV (CaMKIV) (Ribar et al., 2000). Mutant mice lacking one of these signaling proteins have strong deficits in cerebellar motor coordination. One common cellular phenotype of these mice is the abnormal persistence of multiple climbing fiber innervation.

Our results provide for the first time evidence that TrkB is required for CF elimination. It is unclear, however, whether TrkB controls CF elimination through a distinctly new postsynaptic pathway. Interestingly, however, CaMKIV is one of the downstream enzymes activated by TrkB-stimulation (Finkbeiner et al., 1997; Minichiello et al., 2002). Thus, it is conceivable that both TrkB and CaMKIV are part of the same signaling pathway controlling the developmental CF synapse elimination. However, TrkB is strongly involved also in other transduction mechanisms (Huang and Reichardt, 2003), while CaMKIV in its turn can be stimulated in several other ways (Soderling, 1999). In addition to such a presumed postsynaptic mechanism also presynaptic factors may determine the disturbed CF synapse elimination. Possibly, the delayed development of GABAergic synapses in the cerebellum (Rico et al., 2002) may perturb the function of the cerebellar circuitry and, thereby, impair the normal process of developmental CF synapse elimination. In fact, our results (Fig. 4) provide direct evidence for abnormalities of the cerebellar GABAergic system.

The main conclusion of our study is that, in cerebellar Purkinje neurons, the deficiency of TrkB has little significance for the general dendritic morphology of the cell, but a critical role in the developmental elimination of redundant climbing fiber synapses. This is a novel function of the TrkB signaling in synaptic rewiring, which adds an important new dimension to the more established roles of neurotrophins in synaptic plasticity and activity-dependent synaptic strengthening.

Methods

Animals

trkB mutants were originally generated on an ICR (Institute for Cancer Research) strain background (Rohrer et al., 1999). In an attempt to improve the survival of trkB^-/- mice, the trkB KO allele was bred to the BALB/c strain for >10 generations. trkB^-/- pups from BALB/c trkB^+/- matings occasionally (10-15%) grew to an age of P7 or older. These mice were used for the present experiments. All experiments were done in compliance with institutional animal welfare guidelines.

Genotyping

DNA used for genotyping was extracted from mouse tail biopsies. PCR reaction was performed over 35 cycles with following primers: trkb-n2: 5′-ATGTCGCCCTGGCTGAAGTG; trkbc8: 5′-ACTGACATCCGTAAGCCAGT; pgk3-1: 5′-GGTTCTAAGTACTGTGGTTTCC. Annealing temperature was set at 60°C. The products of the PCR reaction were visualized using agarose gel electrophoresis as shown in Fig. 1a.

Immunohistochemistry

Immunostaining was performed according to routine methods. Briefly: 30 μm thick free floating frozen sections from P7 mice were blocked with appropriate serum and then incubated overnight at 4°C with primary antibodies. Sections were washed and incubated with secondary antibodies at room temperature for 2 h. Fluorescent images were scanned with a confocal microscope IX70 (Olympus, Tokyo, Japan). Primary antibodies: rabbit anti-pan-Trk (C-14, sc-11, Santa Cruz Biotechnology, Santa Cruz, CA, 1:500), rabbit anti-TrkB (AB5372, Chemicon, Temecula, CA, 1:500), mouse anti-TrkB (BD Transduction Laboratories, San Jose, CA, 1:500), mouse anti-calbindin D-28k (Swant, Bellinzona, Switzerland, 1:500), rabbit anti-calbindin D-28k (AB1778, Chemicon, 1:500). Secondary antibodies: goat anti-mouse and goat anti-rabbit coupled to Alexa Fluor 488 (Molecular Probes, Eugene, OR) and goat anti-mouse and goat anti-rabbit coupled to Cy3 (Jackson ImmunoResearch, West Grove, PA).

Slice preparation

Mice were decapitated at P7, P10 or P14, their cerebella quickly removed and placed in artificial cerebrospinal fluid (ACSF; containing (in mM): 125 NaCl, 2.5 KCl, 2 CaCl₂, 1 MgCl₂, 1.25 NaH₂PO₄ and 20 glucose; carboxygenated with 95% O₂ and 5% CO₂; <4 °C). Slices (300 μm thick) were cut using a vibratome slicer (Leica, Wetzlar, Germany) and stored for 1 h at 35°C and then stored ≤8 h at room temperature in ACSF.

Two-photon microscopy

Slices were placed on the stage of an upright BX50WI microscope (Olympus) and submerged in ACSF (with 4.5 mM KCl; 32°C). PCs were kept under whole-cell voltage-clamp for 2 min at -70 mV, and then the pipette was gently removed (bolus loading). The intracellular solution contained (in mM): 140 K-gluconate, 4 NaCl, 12 KCl, 10 HEPES, 4 Mg-ATP, 0.4 Na-GTP, and 1 Alexa Fluor-488 (Molecular Probes), pH 7.3 (with KOH). Images were collected with a 60×(1 NA)or 100 × (1.1 NA) water immersion objective (Nikon, Tokyo, Japan) using a custom-built two-photon laser-scanning microscope (Nikolenko et al., 2003), consisting of a modified Fluoview (Olympus) confocal microscope and a Ti:sapphire laser providing 100 fs pulses at 80 MHz (Mai Tai, Spectra Physics, Mountain View, CA). While exciting at 785 nm, the emitted fluorescence was detected by an internal photo-multiplier tube in whole-area detection mode. For each cell, 50-100 focal planes were recorded with a nominal z-resolution of 0.5 μm. The z-stacks were deconvolved off-line using Huygens Essential (Scientific Volume Imaging, Hilversum, The Netherlands).

Purkinje cell morphometry

Purkinje cell morphometry was performed using z-projections of two-photon image stacks. The circularity of the soma was defined as 4π*area/perimeter². The total somatodendritic area was calculated automatically using custom-written Labview VIs (National Instruments, Austin, TX). The dendrites in deconvoluted z-stacks of two-photon images were traced with the use of the Neuron Morpho plugin (Dr. G. D'Alessandrio, University of Southampton, UK) for ImageJ (NIH, USA). A dendritic element was defined as the connection either between two bifurcations, the soma and the first bifurcation or the last bifurcation and the tip of the dendrite. The asymmetry of the largest dendritic tree of each PC was characterized by the “tree asymmetry index” (Van Pelt et al., 1992), calculated as:

$$A_t={(n-1)}^{-1}\Sigma \mid r-s\mid ∕(r+s-2)$$

with r and s being the number of dendritic elements on the right-hand or left-hand side, respectively. For the centrifugal branch order analysis after each bifurcation, the branch order of both daughter dendritic elements increases by one. For the calculation of the angular branch order, at each bifurcation only the dendritic element with the larger angle with the parent dendrite is assigned a higher branch order.

A concentric Sholl analysis was performed as described (Sholl, 1953) using custom-written software on the 2D-reconstructions of the PCs (Fig. 2e). The number of intersections with dendritic elements was counted for all circles that were virtually drawn on the dendritic tree at a distance of 1 μm. The radius of the largest circle with at least one intersection with the dendritic tree was defined as the apical length of the PC. The “normalized distance of maximal ramification” was calculated as the radius of the circle with the highest number of intersections divided by the apical length of the PC. For segmental Sholl analysis, the images of PCs were covered with equiangular sectors, aligned to the principal axis, and the number of dendritic elements in each sector was counted. Mediolaterality is calculated as the number of dendritic elements in the four octants adjacent to the principal axis divided by the number of dendritic elements in all sectors. Accordingly, rostrocaudality is defined as the number of elements in the four octants on the apical side of the PC divided by the number of elements in all sectors (Galhardo and Lima, 1999).

Cluster analysis according to Ward's method (Milligan, 1980) was performed using the PAST software (Hammer et al., 2001). The parameters used for the analysis are listed in Table I. For spine analysis, we used the z-stacks of the 2-photon images of the PCs. The putative CF spines (Larramendi and Victor, 1967) were counted on the first three concentric branch orders, provided that they were sufficiently thick and sparsely covered with spines. In order to estimate the PF spine density, we selected six terminal branches belonging to the 5th to 12th centrifugal branch order. Spine counting was performed blindly to the genotype of the mouse.

Electrophysiology and confocal Ca²⁺ imaging

PCs were visually identified using an Eclipse E600FN microscope with a 40x or 60x objective (Nikon) and whole-cell recordings were made using either an EPC-8, an EPC-9 (HEKA, Lambrecht, Germany) or a BVC-700A amplifier (Dagan Corporation, Minneapolis, MN). Pipettes (3-4 M Ω) were pulled from borosilicate glass. During recordings, slices were continuously perfused with ACSF (with 4.5 mM KCl).

GABAergic mIPSCs were recorded at 33°C in the continuous mode of Pulse (HEKA) in the presence of 4.5 mM external K⁺, 20 μM CNQX (Sigma, Deisenhofen, Germany), 25 μM APV (Sigma) and 500 nM TTX (Alomone Labs, Jerusalem, Israel) using a sample frequency of 10 kHz and a low-pass filter at 5 kHz. The holding potential was -70 mV. The intracellular solution contained (in mM): 135 CsCl, 10 HEPES, 15 NaCl, 5 tetraethylammonium chloride, 0.16 EGTA, 4 Mg-ATP, 0.4 Na-GTP at pH 7.3 (with CsOH). Recordings were analyzed off-line.

PF and CF EPSCs and EPSPs were recorded at room temperature in the presence of 4.5 mM external K⁺ and 10 μM bicuculline with Pulse using a sample frequency of 10 kHz and a low-pass filter at 3 kHz. The holding potential was -70 mV. The intracellular solution contained (in mM) 148 K-gluconate, 10 HEPES, 10 NaCl, 0.5 MgCl₂, 4 Mg-ATP, 0.4 Na-GTP, and 0.05 Oregon Green BAPTA-1 (Molecular Probes) at pH 7.3 (with KOH).

Confocal Ca²⁺ imaging was performed using a confocal laser-scanning microscope (Odyssey, Noran, Middleton, WI) attached to an upright microscope (Eclipse E600FN, x40 water immersion objective, NA 0.8 (Nikon)). Fluorescence images were obtained at 30 Hz using custom made software (FastAnalysis, Labview). Dendritic regions activated by CF stimulation were identified as described (Hartmann et al., 2004).