NL2-KO mice were purchased from The Jackson Laboratory (Maine, United States). Both wild-type and homozygote NL2-KO littermates of both sexes from the interbreeding of heterozygote pairs were used. In experiments with in utero electroporation, C57BL/6 mice were used (SLC, Japan). All experiments were conducted following the guidelines for the care and use of laboratory animals of the University of Tokyo, Hokkaido University, and the Japan Neuroscience Society.

Pregnant C57BL/6 mice at embryonic day (E) 11.5 or E12.5 (SLC, Japan) were deeply anesthetized via intraperitoneal injection (50 μg/g) of sodium pentobarbital (Somnopentyl, Kyoritsu Seiyaku Co., Tokyo, Japan). One microliter of a plasmid DNA (1.5 μg/μL) in phosphate-buffered saline containing fast green was injected into the fourth ventricle of embryonic brains. When we transfected two or three different plasmid DNA constructs into PCs, we first prepared individual constructs in phophosphate-buffered saline containing fast green at the concentration of 1.5 μg/μL. We took 1 μL of the individual DNA-containing solutions and mixed them. Then we injected 1 μL of the mixed solution into the fourth ventricle of embryonic brains. The embryo was held in uterus with a forceps-type electrode, and electrical pulses (40 V, with a duration of 50 ms, at intervals of 950 ms per cycle) were delivered five times with an electroporator (Nepa Gene, Japan). After electroporation, the uterus was repositioned in the abdominal cavity, the abdominal wall and skin were closed, and the embryos were allowed to continue developing normally.

Plasmid vectors were designed for PC-specific expressions under the control of a truncated L7 promoter (pCL20c-L7). These vectors were constructed to express GFP and/or microRNA (miRNA) directed against NL2 or P/Q-type voltage-gated calcium channels (P/Q-VDCC) under the control of a truncated L7 promoter. The plasmid with mouse NL2 construct was obtained from Addgene (plasmid 15246) (Chih et al., 2006). The NL2 sequence was cloned into the pCAGGS vector. Engineered microRNAs targeting NL2 were designed according to the BLOCK-iT Pol II miR RNAi Expression Vector Kit guidelines (Invitrogen), using the following sequences: 5’-TGCTGTGTACATCCTGGTCCACTAGCGTTTTGGCCACTGACTGACGCTAGTGGCAGGATGTACA-3′ and 5’-CCTGTGTACATCCTGCCACTAGCGTCAGTCAGTGGCCAAAACGCTAGTGGACCAGGATGTACAC-3′. For P/Q-VDCC, we used the same microRNA sequence as reported by Mikuni et al. (2013).

Mice at P4 to P35 were decapitated following CO₂ anesthesia, and brains were rapidly removed and placed in chilled normal ACSF (0–4°C) containing (in mM) 125 NaCl, 2.5 KCl, 2CaCl₂, 1MgSO₄, 1.25 NaH₂PO₄, 26 NaHCO₃, and 20 glucose, bubbled with 95% O₂ and 5% CO₂ (pH 7.4). Parasagittal cerebellar slices (250 μm) were prepared by using a vibratome slicer (VT-1200S, Leica, Germany). Whole-cell patch-clamp recordings were made from visually identified PCs by using an upright microscope (BX50W1, Olympus, Japan). Internal solution composed of (in mM): 60 CsCl, 10 D-gluconate, 20 TEA-Cl, 20 BAPTA, 4 MgCl₂, 4 ATP, 0.4 GTP, and 30 HEPES (pH 7.3, adjusted with CsOH) for recording excitatory postsynaptic currents (EPSCs) and 124 CsCl, 10 HEPES, 10 BAPTA, 1 CaCl2, 4.6 MgCl₂, 4 ATP, 0.4 GTP (pH 7.3, adjusted with CsOH) for recording inhibitory postsynaptic currents (IPSCs). Picotoxin (100 μM) was added to block inhibitory synaptic transmission for recording stimulation-evoked EPSCs. Picotoxin (100 μM) and tetrodotoxin (0.5 μM) were added for recording miniature EPSCs (mEPSCs). NBQX (10 μM), _R-CPP (5 μM) and tetrodotoxin (0.5 μM) were added for recording miniature IPSCs (mIPSCs). Stimulation pipettes (5–10 μm tip diameter) were filled with the normal ACSF and used to apply square pulses for focal stimulation (duration of 100 ms, amplitude of 0 V to 100 V). CFs were stimulated in the granular layer around the PC soma under investigation. All electrophysiological recordings were performed at 32°C. Ionic currents were recorded with an EPC10 patch clamp amplifier (HEKA). Online data acquisition and offline data analysis were performed using PULSE and PULSE FIT software (HEKA) or Mini analysis Program (ver. 4.0.1, Synaptosoft Inc.). Liquid junction potential was corrected. All drugs were obtained from Tocris.

To quantitatively evaluate the disparity among the amplitudes of multiple CF-mediated EPSCs (CF-EPSCs) in individual PCs, we calculated the disparity ratio and the disparity index as shown previously (Hashimoto and Kano, 2003).

Disparity ratio= A 1 A N + A 2 A N + ⋯ + A N − 1 A N N − 1 .

Disparity index = S . D . M

S.D. = ∑ A i − M 2 N − 1

To calculate the disparity ratio and disparity index, the amplitudes of individual CF-EPSCs in a given PC with multiple CF innervation were measured at the same holding potential and numbered in the order of their amplitudes (A1, A2, …, AN, N ≥ 2; N is the number of CFs innervating a given PC. AN represents the largest CF-EPSC) (Hashimoto and Kano, 2003). The smaller the difference in the amplitude between AN and other weaker CF-EPSCs, the larger the value of disparity ratio. If all CFs innervating a PC exhibit similar amplitudes of CF-EPSCs, the disparity ratio approaches 1. The disparity index is the coefficient of variation for all CF-EPSC amplitudes recorded in a PC (Hashimoto and Kano, 2003). The larger the difference in the amplitude of CF-EPSCs, the greater the value of the disparity index.

Vermal cerebellar slices (250 μm thickness) were prepared from mice at P10 to P13. Recording pipettes were filled with internal solutions containing (in mM): 135 K-gluconate, 10 Na-gluconate, 5 KCl, 0.5 EGTA, 10 HEPES, 4 Mg-ATP, 0.4 Na₃-GTP (pH 7.3, adjusted with NaOH) and 0.1 OGB-1, which was used for imaging CF-induced calcium signals. After establishing whole-cell recording configurations, PCs were loaded for at least 20 min with the calcium indicator OGB-1. In the normal ACSF without blockers of excitatory or inhibitory synaptic transmission, fluorescence images were acquired at 30 Hz using a high-speed confocal laser scanning microscope (CSU22, Yokogawa). The calcium signals were obtained from the soma or proximal dendritic regions of PCs. The calcium-dependent fluorescence signals from selected regions of interest (ROI) were background subtracted and expressed as increases in fluorescence divided by the prestimulus fluorescence values (ΔF/F₀) using ImageJ.¹ The calcium imaging was performed at 32°C.

For immunofluorescence, mice were deeply anesthetized with pentobarbital (100 mg/g of body weight, i.p.), perfused with 4% paraformaldehyde in 0.1 M sodium phosphate buffer, pH 7.2, and processed for parasagittal microslicer sections (50 μm in thickness; VT1000S; Lecia, Nussloch, Germany). After permeabilization and blocking for nonspecific binding of antibodies, the sections were incubated overnight with affinity-purified primary antibodies with 2% normal donkey serum and 0.3% Triton X-100 in phosphate-buffered saline (PBS) for 48 h at 4°C. Subsequently, the sections were rinsed in PBS and then incubated for 2 h at room temperature with a mixture of fluorochrome-conjugated secondary antibodies (species-specific indocarbocyanine (Cy3)- or indodicarbocyanine (Cy5)-conjugated secondary IgG; 1:200, Jackson ImmunoResearch, West Grove, PA) in the dark. Incubation was finally stopped by washing three times with PBS. After washing, the sections were mounted in the mounting medium (DAKO, Carpinteria, CA, United States) and then examined under a laser scanning confocal microscope (FV1000, Olympus).

For the morphological analysis of CF innervation, mice were anesthetized with chloral hydrate (350 mg/kg of body weight, i.p.). A glass pipette filled with 2–3 μL of 10% solution of dextran Alexa Fluor-594 (DA-594, Invitrogen) in PBS was inserted stereotaxically into the inferior olive by the dorsal approach, as described previously (Miyazaki et al., 2004; Miyazaki and Watanabe, 2011). Tracers were injected by air pressure (Pneumatic Picopump; World Precision Instruments). After 4 days of survival, mice were fixed by transcardial perfusion. For combined labeling by tracer and immunofluorescence, DA-594-labeled microslicer sections were incubated with a mixture of calbindin and VGluT2 antibodies followed by 2 h incubation with a mixture of Alexa 488- (Invitrogen) and Cy5-labeled species-specific secondary antibodies. Images of triple labeling were taken with a confocal laser-scanning microscope (FV1000, Olympus).

Lysates of cerebella from 2-month-old NL2-KO and wild-type mice were prepared using standard protocols. Briefly, cerebella were lysed with RIPA buffer (125 mM Tris–HCl, pH 6.8; 10% Mercaptoethanol, 0.004% Bromophenol Blue, 10% Sucrose, and 4% SDS; Nacalai tesque, Kyoto, Japan), and then the lysates were left on ice for 30–60 min and centrifuged at 10,000 g at 4°C for 10 min. Equal amounts of lysates were resolved on 10% polyacrylamide-SDS gel by SDS-PAGE (at 196 V and 40 mA for 90 min) and transferred to polyvinylidene difluoride membranes. The blot was performed at 25 V and 100 mA for 2 h. Subsequently, these membranes were probed with the indicated primary antibodies overnight at 4°C, followed by incubation with HRP-conjugated secondary antibodies for 1 h at room temperature. Proteins were visualized by HRP reaction using ECL SuperSignal West Pico kit (Thermo Scientific, Waltham, MA, USA). Data acquisition and analysis were performed using an ImageQuant LAS 4000 image analyzer (GE Healthcare, Chicago, IL, United States).

Affinity-purified primary antibodies raised against the following molecules (host species, final concentration) were used: anti-NL2 (rabbit, 1 μg/mL; Synaptic Systems, Germany), anti-NL3 (rabbit, 1 μg/mL; Synaptic Systems, Germany), anti-calbindin (goat, 1 mg/mL) (Miura et al., 2006), anti-vesicular glutamate transporter 1 (VGluT1) (1 mg/mL, Frontier Institute), anti-vesicular glutamate transporter 2 (VGluT2) (1 mg/mL, Frontier Institute), anti-VGAT (rabbit, 1 mg/mL) (Miura et al., 2006), anti-GABA_A receptor α1 (guinea pig, 1:1000; Synaptic Systems, Germany), anti-GABA_A receptor α2 (guinea pig, 1:1000; Synaptic Systems, Germany), anti-PSD 95 (rabbit, 1:250; Invitrogen), and anti-Gephyrin (mouse, 1:1000, Synaptic Systems, Germany).

All statistical values are presented as mean ± SD unless indicated otherwise. First, normality was checked for individual datasets by using the Shapiro–Wilk test. Since normality was often not assured, we employed nonparametric statistical tests in the present study. We used the Mann–Whitney U test to compare two independent datasets, and the Kruskal-Wallis test followed by Dunn’s test for multiple comparisons or the Mann–Whitney test with False Discovery Rate (FDR) correction using the Benjamini-Hochberg procedure for multiple comparisons. We used the Kolmogorov–Smirnov test for datasets with cumulative distributions. Differences were considered statistically significant if the p-value was smaller than 0.05. All statistical analyses were conducted with GraphPad Prism 10 Software (GraphPad Software, La Solla, CA, USA).

We first checked whether global knockout (KO) of NL2 affected the expression of synaptic proteins in the cerebellum by quantitative western blot analysis of cerebellar lysates (Figure 1). We found that the expression of NL2 was almost undetectable (Figures 1A,B) (> 90% decrease, p = 0.002, Mann–Whitney U test) and that the expression of GABA_A receptor α1 was moderately decreased to approximately 33% (p = 0.002, Mann–Whitney U test) in NL2-KO mice compared to those in wild-type mice (Figures 1C,D). In contrast, we observed that the expression of NL3 was significantly increased in NL2-KO mice compared to that in wild-type mice (73% increase, p = 0.017, Mann–Whitney U test), presumably because of compensation for the lack of NL2 (Figures 1C,D). We found no significant difference in the expression of GABA_A receptor α2 (p = 0.093, Mann–Whitney U test), gephyrin (p = 0.937, Mann–Whitney U test), vesicular inhibitory amino acid transporter (VIAAT; p = 0.24, Mann–Whitney U test), or PSD95 (p = 0.589, Mann–Whitney U test) (Figures 1C,D). These results suggest that GABA_A receptor-mediated inhibitory synaptic transmission may be attenuated in NL2-KO mice.

We then checked the gross anatomy of the cerebellum and the morphology of its major cellular and synaptic components by immunostaining for their markers. We found that the foliation and layer structure of the cerebellum and the morphology and density of Purkinje cell (PC) immunostained for its marker calbindin were normal in NL2-KO mice (Figures 2A–D). We found no significant differences between the genotypes in the pattern and density of immunofluorescence for a parallel fiber (PF) terminal marker, type 1 vesicular glutamate transporter (VGluT1), and that for a climbing fiber (CF) terminal marker, type 2 vesicular glutamate transporter (VGluT2) (Figures 2E–G,I). Immunofluorescence for an inhibitory terminal marker, VIAAT, and that for an inhibitory interneuron marker, parvalbumin (PA), were similar between wild-type and NL2-KO mice (Figures 2H,J–K,M). Furthermore, no appreciable difference was noted between the genotypes in the morphology of Bergmann glia immunostained for its marker glial fibrillary acidic protein (GFAP) (Figures 2L,N).

Zhang et al. (2015) reported previously that PC-specific double NL2- and NL3-KO and single NL2-KO severely impaired inhibitory synaptic transmission in PCs at P21-P25 (Zhang et al., 2015). We, therefore, examined whether synaptic transmission is altered in the PCs of our global NL2-KO mice. We performed whole-cell recordings from PCs in acute cerebellar slices and measured miniature excitatory postsynaptic currents (mEPSCs) and miniature inhibitory postsynaptic currents (mIPSCs) (Yamasaki et al., 2006). We found no significant difference in the amplitude (P7-9: p = 0.563; P10-12: p = 0.928; P13-15: p = 0.796; P16-20: p = 0.315; P22-30: p = 0.776, Mann–Whitney test with FDR correction using the Benjamini-Hochberg procedure) and the frequency (P7-9: p = 0.743; P10-12: p = 0.254; P13-15: p = 0.312; P16-20: p = 0.848; P22-30: p = 0.864, Mann–Whitney test with FDR correction using the Benjamini-Hochberg procedure) of mEPSCs between wild-type and NL2-KO PCs throughout the age ranges during postnatal development (Figures 3A–C). In contrast, the amplitude of mIPSCs was consistently smaller in NL2-KO mice than in wild-type mice throughout postnatal development from P7-P9 to P22-P30 (P7-9: p < 0.001; P10-12: p = 0.008; P13-15: p = 0.017; P16-20: p = 0.028; P22-30: p = 0.001, Mann–Whitney test with FDR correction using the Benjamini-Hochberg procedure; Figures 3D,E). The frequency of mIPSCs was lower at P10-P12 (wild-type: 3.42 ± 1.029 Hz; NL2-KO: 2.17 ± 1.18 Hz, p = 0.015) but showed no difference in the other age ranges between wild-type and NL2-KO PCs (P7-9: p = 0.683; P13-15: p = 0.661; P16-20: p = 0.688; P22-30: p = 0.427, Mann–Whitney test with FDR correction using the Benjamini-Hochberg procedure; Figure 3F). These results indicate that GABAergic inhibition to PCs is attenuated in NL2-KO mice particularly from P10 to P12 when massive formation of inhibitory synapses occurs from molecular layer interneurons to PCs (Yamanaka et al., 2004). We also recorded IPSCs from PCs elicited by extracellular stimulation (eIPSCs) at both P10-P12 and P20-P35 (Figures 3G–I). We observed a significant decrease in the eIPSC amplitude (~50%, Figure 3H) in NL2-KO PCs (P10-P12: p = 0.018, and P20-P35: p = 0.004, Mann–Whitney test with FDR correction using the Benjamini-Hochberg procedure), with no alteration in the paired-pulse ratio (P10-P12: p = 0.21, and P20-P35: p = 0.247, Mann–Whitney test with FDR correction using the Benjamini-Hochberg procedure; Figure 3I); suggesting that the reduction of the IPSC amplitude is caused by postsynaptic changes.

We further classified mIPSCs into two categories (small and large) at the amplitude of 100 pA and scrutinized their kinetics at P10 to P12 (Nakayama et al., 2012). The rise time of large mIPSCs was significantly faster than that of small mIPSCs (Figure 3J), supporting that large mIPSCs arise mainly from synapses on the soma (presumably from basket cells) and small mIPSCs arise mainly from synapses on dendrites (presumably from stellate cells) to PCs (Nakayama et al., 2012). We found that the amplitude of small mIPSCs was slightly smaller (wild-type: 49.72 ± 4.124 pA, n = 10; NL2-KO: 44.7 ± 6.195 pA, n = 13, p = 0.018, Mann–Whitney test with FDR correction using the Benjamini-Hochberg procedure), whereas the frequency of large mIPSCs was significantly reduced (wild-type: 1.44 ± 0.549 Hz, n = 10; NL2-KO: 0.71 ± 0.457 Hz, n = 13, p = 0.002, Mann–Whitney test with FDR correction using the Benjamini-Hochberg procedure) in NL2-KO mice than in wild-type mice (Figures 3K,L). These results suggest that inhibitory synaptic inputs from basket cells were reduced particularly at P10-P12 in PCs of NL2-KO mice. Taken together, the E/I balance of PCs is considered to be shifted to enhanced excitation in NL2-KO mice compared with wild-type mice.

We examined whether the postnatal development of CF to PC synapses was affected by the loss of NL2. We recorded whole-cell membrane currents from PCs in acute cerebellar slices and stimulated CFs in the granule cell layer to induce CF-mediated excitatory postsynaptic currents (CF-EPSCs). We systematically moved the location of the CF stimulation pipette and gradually increased the stimulus intensity at each position. We estimated the number of CFs innervating the recorded PC from discrete CF-EPSC steps elicited in that PC (Hashimoto et al., 2009a; Hashimoto and Kano, 2003; Hashimoto et al., 2011). In the majority of wild-type PCs (Figure 4A), a clearly discernible CF-EPSC was elicited in an all-or-none fashion, indicating that such PCs were innervated by single CFs. By contrast, in NL2-KO mice aged from P20 to P35, only 60% of PCs were innervated by single CFs. The frequency distribution of PCs against the number of CF-EPSC steps demonstrates that a significantly higher proportion of PCs were innervated by multiple CFs in NL2-KO mice than in wild-type mice (Figure 4B, p = 0.029, Mann–Whitney U test). In both genotypes, each PC was either mono-innervated by a strong CF (termed “CF-mono”) or multiply innervated by a strong CF plus one or two weaker CFs (termed “CF-multi-S” and “CF-multi-W,” respectively) (Hashimoto et al., 2009a; Hashimoto and Kano, 2003). We compared the kinetics of these three types of CF inputs between wild-type and NL2-KO mice. We found no significant difference in the amplitude, the 10-90% rise time, and the paired-pulse ratio of CF-EPSCs (Supplementary Table S1), although the decay time constant for CF- mono and that for CF-multi-S were slightly shorter in NL2-KO PCs. These results suggest that the deletion of NL2 impaired CF synapse elimination with little effect on the basic properties of CF-EPSCs. Since proper PF synaptogenesis and normal PF to PC synaptic transmission are prerequisites for developmental CF synapse elimination (Hashimoto et al., 2001; Hashimoto et al., 2009b; Hirai et al., 2005; Ichikawa et al., 2002), we examined the nature of PF-mediated EPSCs in PCs of NL2-KO and wild-type mice during P22–P30. No difference was noted between the two mouse strains in the input/output relations and the extent of paired-pulse facilitation (Supplementary Figure S1). These results indicate that PF to PC synapses function normally in NL2-KO mice and suggest that the persistent multiple CF innervation of NL2-KO PCs does not result from impaired PF to PC synaptic function.

To examine CF innervation of PCs morphologically, we stained a subset of CFs by injecting a small amount of an anterograde tracer, Dextran Alexa Fluor 594 (DA594), into the inferior olive and then performed triple fluorescent labeling for calbindin (a PC marker), VGluT2 (a CF terminal marker), and DA594 (Miyazaki et al., 2004; Miyazaki and Watanabe, 2011). In both wild-type and NL2-KO mice, DA594-positive CFs followed the proximal dendrites of PCs and climbed up to around 80% of the molecular layer (Figures 4C,D2,E,F2). Presynaptic terminals of these DA594-labelled CFs along PC dendritic trees mostly overlapped with VGluT2 immunoreactivity (Figures 4C,D1,E,F1). We also confirmed that the molecular layer thickness (Figure 4G; wild-type: 168.2 ± 2.4 μm; NL2-KO: 168.9 ± 2.2 μm; p = 0.8277, Mann–Whitney U test) and the relative height of VGluT2-positive CFs to the molecular layer thickness (Figure 4H; wild-type: 75.5% ± 0.5%; NL2-KO: 76.1% ± 0.6%; p = 0.4279, Mann–Whitney U test) were normal in NL2-KO mice. These results show that dendritic translocation of single strong CF is normal in NL2-KO mice. Although no significant difference was found in the density of vGluT2-positive terminals on the soma between wild-type and NL2-KO mice (Figure 4I, wild-type: 0.077 ± 0.011; NL2-KO: 0.125 ± 0.02, p = 0.2543, Mann–Whitney U test), DA594-labeled/VGluT2-positive terminals and DA594-unlabeled/VGluT2-positive terminals were frequently colocalized on the PC soma of NL2-KO mice (Figure 4F1). These results demonstrate that multiple CF innervation in NL2-KO PCs reflects the presence of redundant synapses on the soma arising from CFs distinct from the main CF innervating the dendrites of the same PC.

To investigate at which stage of postnatal development the abnormality of synapse elimination takes place in NL2-KO mice, we followed the developmental course of CF innervation. At P5-P6, just before the onset of synapse elimination, the majority of PCs were innervated by three or more CFs in both strains of mice with no significant difference in the frequency distribution of PCs in terms of the number of CF-EPSC steps per PC (Figure 5A, p = 0.628, Mann–Whitney U test). At P7-P9, the frequency distributions of PCs markedly shifted to smaller numbers from those at P5-P6 in both strains of mice and there was no significant difference between the two (Figure 5B, p = 0.103, Mann–Whitney U test). After P10, the number of CFs in each PC progressively decreased, but the degree of reduction was significantly smaller in NL2-KO mice than in wild-type mice. Consequently, the frequency distribution of PCs exhibited a significant difference between the two strains of mice at P10-P12, P13-P15, and P16-P19 (Figures 5C–E, P10-P12: p = 0.002; P13–P15: p = 0.044; P16-P19: p = 0.045, Mann–Whitney U test). These results indicate that initial formation of CF synapses and elimination of redundant CF synapses from P5 to P9 are normal but CF synapse elimination after P10 is impaired in NL2-KO mice.

To test whether functional differentiation into single “strong” CF and the other “weak” CFs in each PC is affected in NL2-KO mice, we calculated the two parameters termed disparity index and disparity ratio, which represent the degree of the functional differentiation of CF synapses (Hashimoto et al., 2009a; Hashimoto and Kano, 2003) (Figure 5F). We found that both parameters were similar between NL2-KO mice and wild-type mice throughout postnatal development and in young adulthood (Disparity ratio: P5-6: p = 0.41; P7-9: p = 0.782; P10-12: p = 0.692; P13-15: p = 0.98; P16-20: p = 0.565; P22-35: p = 0.642; Disparity index: P5-6: p = 0.188; P7-9: p = 0.775; P10-12: p = 0.913; P13-15: p = 0.886; P16-20: p = 0.583; P22-35: p = 0.642; Mann–Whitney test with FDR correction using the Benjamini-Hochberg procedure; Figure 5F). Taken together, the results so far indicate that CF synapse elimination after P10 is specifically impaired in PCs of NL2-KO mice, despite normal functional differentiation into single “strong” and the other “weak” CFs, normal CF synapse elimination from P5 to P9, and normal dendritic translocation of a single “strong” CF.

Since NL2 is expressed widely in various cell types in the central nervous system (Patrizi et al., 2008; Varoqueaux et al., 2004), it is not clear whether the reduced GABAergic neurotransmission in PCs and impaired CF synapse elimination resulted from the lack of NL2 in PCs. To address this issue, we first examined whether PC-specific expression of NL2 in NL2-KO mice could rescue their cerebellar phenotypes. We introduced cDNA for NL2 together with that for EGFP into PCs of NL2-KO and wild-type mice at E11.5 or E12.5 by in utero electroporation. We used the same NL2 cDNA construct that was used to rescue the effects of microRNA-mediated NL2 knockdown (KD) in PCs (see Supplementary Figure S2). In this construct, seven nucleotides were mutated without changing the amino acid sequence in the microRNA targeted site so that the expression of NL2 was restored to normal level (Uesaka et al., 2012). We found that a subset of PCs expressed EGFP and no other cell types in the cerebellum did so. We made whole-cell recordings from EGFP-positive PCs that expressed NL2 and from EGFP-negative PCs in the same slices as the control. We found that after PC-specific expression of the NL2 cDNA, the mean amplitude (p = 0.62, Mann–Whitney U test) and frequency (p = 0.259, Mann–Whitney U test) of mIPSCs in PCs of NL2-KO mice were similar to those of wild-type mice (Figures 6A,B). Importantly, after PC-specific expression of NL2 cDNA, CF innervation patterns of EGFP-positive PCs during P21-P35 were identical between NL2-KO mice and wild-type mice, whereas CF innervation patterns were significantly different between EGFP-positive and EGFP-negative PCs in NL2-KO cerebellum (Kruskal-Wallis test: H(2) = 6.3886, p = 0.041; Dunn’s test for multiple comparisons: WT-NL2-RES vs NL2-KO: p = 0.0389; KO-NL2-RES vs NL2-KO: p = 0.0321; WT-NL2-RES vs KO-NL2-RES, p = 0.9321; Figures 6C,D). There were no significant changes in the basic electrophysiological properties of CF-EPSCs including the amplitude, the 10–90% rise time, the decay time constant, and the paired-pulse ratio (Supplementary Table S2). These results suggest that NL2 expression in PCs is sufficient to restore normal GABAergic transmission in PCs and normal CF synapse elimination during postnatal development.

We then examined whether PC-specific knockdown (KD) of NL2 in wild-type mice causes reduced GABAergic transmission in PCs and impaired CF synapse elimination. We deleted NL2 specifically from PCs during postnatal development using the microRNA-mediated KD system (Uesaka et al., 2012; Uesaka et al., 2014). We introduced an NL2-KD vector together with EGFP cDNA into PCs under the control of a PC-specific L7 promoter by in-utero electroporation at E11.5–12.5. As a control, we similarly introduced scrambled NL2 microRNA into PCs (Figure 7A). We found that the amplitude of mIPSC in PCs with NL2-KD was significantly smaller than that of untransfected GFP-negative PCs or that of PCs expressing scrambled NL2 mRNA during P21-P35 (Kruskal-Wallis test: H(2) = 12.62, p < 0.001; Dunn’s test for multiple comparisons: GFP-ve vs. NL2-KD: p = 0.009; scrambled-NL2 vs. NL2-KD: p = 0.006), whereas no significant difference was found in the frequency of mIPSC among the three groups (Kruskal-Wallis test: H(2) = 0.1575, p = 0.929; Figure 7B). In contrast, the amplitude (Kruskal-Wallis test: H(2) = 0.5172, p = 0.786) and frequency (Kruskal-Wallis test: H(2) = 3.067, p = 0.224) of mEPSCs were similar among the three groups of PCs (Figure 7C). We also found that the amplitude of eIPSC in PCs with NL2-KD was significantly smaller than that of untransfected GFP-negative PCs or that of PCs expressing scrambled NL2 mRNA during P21 to P35 (Kruskal-Wallis test: H(2) = 10.18, p = 0.002; Dunn’s test for multiple comparisons: GFP-ve vs. NL2-KD: p = 0.041; scrambled-NL2 vs. NL2-KD: p = 0.008; Figure 7D). In contrast, no difference was found in the paired-pulse ratio of eIPSCs among the three groups of PCs (Kruskal-Wallis test: H(2) = 0.883, p = 0.663; Figure 7D). As for CF innervation, we found that a significantly higher percentage of PCs remain multiply innervated by CFs in PCs with NL2-KD during P22-P35 when compared with untransfected GFP-negative PCs or PCs expressing scrambled NL2 microRNA (Kruskal-Wallis test: (H(2) = 9.278, p = 0.0097; Dunn’s test for multiple comparisons: GFP-ve vs. NL2-KD: p = 0.0146; scrambled-NL2 vs. NL2-KD: p = 0.0468; GFP-ve vs scrambled-NL2: p = 0.724; Figures 7E). Co-expression of RNAi-resistant NL2 cDNA effectively rescued the impaired CF synapse elimination by NL2-KD in PCs (Supplementary Figure S2). There were no significant changes induced by NL2-KD in PCs in the basic electrophysiological properties of CF-EPSCs including the amplitude, the 10–90% rise time, the decay time constant, and the paired-pulse ratio (Supplementary Table S3). These results indicate that NL2 in PCs is required for normal GABAergic transmission in PCs and normal CF synapse elimination during postnatal development.

The results so far indicate that lack of NL2 in PCs results in persistent multiple CF innervation due to impaired CF synapse elimination after P10. We hypothesize that reduced GABAergic inhibition in PCs with NL2 deletion was the main cause of impaired CF synapse elimination. CF inputs depolarize PCs and induce calcium transients mediated mainly by P/Q-type voltage-gated calcium channels (P/Q-VDCC) in vivo. Our previous studies indicate that calcium transients of PCs by the activation of P/Q-VDCC are crucial for CF synapse elimination (Hashimoto et al., 2011; Kawata et al., 2014; Mikuni et al., 2013). We therefore examined whether CF-induced calcium transients were altered in NL2-lacking PCs with reduced GABAergic inhibition. We made simultaneous whole-cell recordings under current-clamp mode and intracellular calcium measurements from PCs at P10-P12 in normal external solution without GABAergic blockers. We selected PCs that were multiply innervated by a single “strong” CF (CF-multi-S) and one or two “weak” CFs (CF-multi-W) (Figures 8A,B). Activation of CF-multi-S induced characteristic complex spikes and large calcium transients in the soma and proximal dendrites of PCs in both wild-type and NL2-KO mice (Figure 8B). We found no significant differences between the two genotypes in somatic (Figure 8C; p = 0.459, Mann–Whitney test with FDR correction using the Benjamini-Hochberg procedure) and proximal dendritic (Figure 8D; p = 0.905, Mann–Whitney test with FDR correction using the Benjamini-Hochberg procedure) calcium transients. In contrast, stimulation of CF-multi-W induced relatively small EPSPs with one or two spikes and relatively small calcium transients when compared with stimulation of CF-multi-S in wild-type mice. Importantly, stimulation of CF-multi-W in NL2-KO mice induced significantly larger calcium transients in both somata (Figure 8C; p = 0.0098, Mann–Whitney test with FDR correction using the Benjamini-Hochberg procedure) and proximal dendrites (Figure 8D; p = 0.00176, Mann–Whitney test with FDR correction using the Benjamini-Hochberg procedure) than in wild-type mice.

We also investigated whether a similar enhancement of calcium transients induced by weaker CF inputs was seen in PCs with NL2-KD (Figure 9). Activation of CF-multi-S induced characteristic complex spikes and large calcium transients in both somata and proximal dendrites with no significant differences between control and NL2-KD PCs (Figures 9A–D, Somata: p = 0.8; Proximal dendrite: p = 0.912, Mann–Whitney U test). In contrast, stimulation of CF-multi-W induced larger calcium transients in both somata and proximal dendrites in NL2-KD PCs than in control PCs (Figures 9A–D, Somata: p = 0.005; Proximal dendrites: p = 0.075, Mann–Whitney U test). Taken together, these results indicate that deletion of NL2 in PCs attenuates GABAergic inhibition in PCs, which allows weaker CFs to generate relatively large calcium transients in the soma and proximal dendrites. Large calcium transients may enhance the activation of calcium-dependent reinforcement signals for synapses and therefore may permit weak CF synapses to survive on the PC soma.

Previous studies indicate that calcium transients induced by CF inputs are mediated mainly by P/Q-VDCCs (Hashimoto et al., 2011; Otsu et al., 2014). It is therefore possible that enhanced calcium influx through P/Q-VDCCs might facilitate the survival of weak CF inputs in PCs lacking NL2. To check whether P/Q-VDCC is involved in the persistent multiple CF innervation in PCs lacking NL2, we performed double KD of NL2 and P/Q-VDCC in PCs and compared the effect with that of single P/Q-VDCC KD (Figure 10). As reported previously (Kawata et al., 2014; Mikuni et al., 2013; Uesaka et al., 2014), the knockdown of P/Q-VDCC in PCs caused a significant impairment of CF synapse elimination. In PCs expressing microRNA against P/Q-VDCC together with scrambled NL2 microRNA (Scr-NL2 + PQ KD), about 50% of PCs remained multiply innervated by CFs and their frequency distribution in terms of the number of CFs innervating each PC was significantly different from that of untransfected GFP-negative PCs (Figures 10A,B). We then performed double KD of NL2 and P/Q-VDCC in PCs and found that the effect of KD was similar to that of single KD of P/Q-VDCC (Figures 10A,B). Frequency distribution histograms were almost identical between PCs expressing microRNA against NL2 together with that against P/Q-VDCC (NL2-KD + PQ-KD) and PCs with single KD of P/Q-VDCC (Scr-NL2 + PQ-KD) (Kruskal-Wallis test: (H(2) = 11.83, p = 0.0027; Dunn’s test for multiple comparisons: GFP-ve vs. Scr-NL2 + P/Q-VDCC, p = 0.0424; GFP-ve vs. NL2-KD + P/Q-VDCC, p = 0.0025; NL2-KD+P/Q-VDCC vs Scr-NL2 + P/Q-VDCC, p > 0.9999; Figures 10A,B). No significant differences that appeared functionally important were found among the three groups of PCs in the basic electrophysiological properties of CF-EPSCs including the amplitude, the 10-90% rise time, the decay time constant, and the extent of the paired-pulse ratio (Supplementary Table S4). The occlusion of the effect of NL2-KD by simultaneous P/Q-VDCC KD suggests that calcium influx through P/Q-VDCC into PCs is involved in the persistent multiple CF innervations in NL2-deleted PCs.