Selective gene expression in IONs and CFs using AAV and knock-in mice

To test the role of C1ql1-Bai3 signaling in structural changes in mature CF PC synapses, we first examined the effect of increased expression of C1ql1 on CF synapses. To avoid possible indirect effects of misexpression of C1ql1 in mossy fibers [24], we used Htr5b-tTA knock-in mice in which IONs, the origin of CFs, specifically express tTA [25]. We delivered adeno-associated virus (AAV) encoding the tetracycline response element (TRE) followed by a channel rhodopsin-2 yellow fluorescent fusion protein (ChR2-YFP) and human influenza hemagglutinin (HA)-tagged C1ql1. ChR2 was introduced to directly assess the function of CFs in the later experiments (Fig. 1A). Three weeks after injection into 3–4-week-old mice, HA-C1ql1 and YFP were detected in IONs (Fig. 1B). C1ql1 immunostaining indicates that C1ql1 expression in the soma of IONs was doubled by AAV-based expression (Fig. 1C). In the cerebellar cortex, high levels of YFP signal were detected in the molecular layer, with no evidence of misexpression in mossy fibers with rosette-like structures in the granular layer (Fig. 1D). Immunohistochemical staining for calbindin, a PC marker, and vesicular glutamate transporter 2 (vGluT2), a presynaptic marker for CFs, showed that HA-C1ql1 was localized to CF terminals along PC dendrites (Fig. 1E). These results indicate that AAV-based delivery of C1ql1 to Htr5b-tTA knock-in mice specifically and moderately increased the amount of C1ql1 in CF terminals.

Fig. 1

CF selective gene delivery using Htr5B-tTA mice and AAV

A Experimental scheme. AAVs encoding ChR2-YFP with or without C1ql1 were injected into the ION. The right panel shows YFP signals in the ION. Scale bar, 500 μm. B Expression levels of C1ql1 in the ION. Immunohistochemical staining shows total C1ql1, exogenous HA-C1ql1 and YFP in the ION infected with AAV-CTRL and AAV-C1ql1. DAPI staining shows the nucleus. Scale bar, 20 μm. C Quantification of C1ql1 immunoreactivity in the soma of IONs (B2). p = 0.0389, Two-tailed Welch’s t-test; n = 4 mice each. D Selective expression of YFP and C1ql1 in the CFs. No YFP signals were detected in the mossy fibers in the granular layer. ML, molecular layer; PCL, Purkinje cell layer; GL, granular layer. Scale bar, 40 μm. E Immunohistochemical staining of vGluT2 (magenta), YFP or HA-C1ql1 (green) and calbindin (blue) indicates accumulation of HA-C1ql1 at CF synapses. Scale bar, 10 μm. Bars represent mean ± SEM. *p < 0.05

Increased C1ql1 levels in CFs induce re-innervation of mature PCs by multiple CFs

Next, we examined the effect of increased C1ql1 expression on the function of CF-PC synapses using whole-cell patch-clamp recordings from PCs in acute slices. AAV-TRE-ChR2-YFP (control) or AAV-TRE-ChR2-YFP-P2A-C1ql1 was injected into the ION of Htr5b-tTA knock-in mice at 3–4 weeks of age (Fig. 2A). Application of two light stimuli with an interval of 100 ms evoked paired-pulse depression of excitatory postsynaptic currents (EPSCs), a result consistent with a high release probability of CF terminals. Furthermore, increasing light intensity elicited EPSCs in an all-or-none manner in control slices (Fig. 2B1), indicating that PCs are innervated by a single CF input with a single excitation threshold. In contrast, overexpression of C1ql1 in CFs not only increased the amplitude of EPSCs, but also led to the appearance of CFs with two to three activation thresholds (Fig. 2B2, C, D). In addition, EPSCs evoked by different activation thresholds had a slower rise time than EPSCs evoked by most CFs (Fig. 2B2, E). These results suggest that increased expression of C1ql1 in CFs not only enhanced the functions of existing CF-PC synapses, but also induced new CF synapses with distinct properties.

Fig. 2

Increased C1ql1 levels in CFs allow adult PCs to be innervated by multiple CFs

A Experimental scheme for recording CF-EPSCs. B Representative CF-EPSC traces from wild-type PCs. The blue bar indicates the timing of light stimulation. Increasing the light intensity elicited a single EPSC in control slices (B1) in an all-or-none manner, but multiple EPSCs with a slower rise time in slices overexpressing C1ql1 (B2). C Total CF-EPSC amplitude. The graph shows the sum of the peak amplitudes of single CF-EPSCs or multiple CF-EPSCs. p = 0.0372, two-tailed Welch’s t-test; n = 28 cells from 4 mice (CTRL); n = 29 cells from 5 mice (+ C1ql1). D The percentage of the number of CFs innervating single PCs. The number of EPSCs evoked by distinct CF activation thresholds (number of steps) is shown. p = 0.0234, Mann–Whitney U test; n = 51 cells from 4 mice (CTRL); n = 49 cells from 5 mice (+ C1ql1). E Average of the 10–90% rise time of CF-EPSCs. p = 0.0028, two-tailed Welch’s t-test; n = 28 responses from 4 mice (CTRL); n = 46 responses from 5 mice (+ C1ql1). F Representative CF-EPSC traces from Bai3 knockout PCs (left: CTRL, right: +C1ql1). G Total CF-EPSC amplitude. The graph shows the sum of peak amplitudes of single CF-EPSCs or multiple CF-EPSCs. p = 0.8534, two-tailed Welch’s t-test; n = 31 cells from 3 mice (CTRL); n = 42 cells from 4 mice (+ C1ql1). H The percentage of the number of CFs innervating single PCs. p = 0.6159, Mann–Whitney U test; n = 38 cells from 3 mice (CTRL); n = 50 cells from 4 mice (+ C1ql1). I Average of the 10–90% rise time of CF-EPSCs. p = 0.0609, two-tailed Welch’s t-test; n = 33 responses from 3 mice (CTRL); n = 46 responses from 4 mice (+ C1ql1). Bars represent mean ± SEM. **p < 0.01; *p < 0.05; ns, not significant

PCs establish a mature innervation pattern with a single CF by postnatal day 20 in rodents [7]. To rule out the effect of C1ql1 overexpression on PC development, we examined PC dendritic arborization, which could affect CF synapse formation. Immunohistochemical staining of PCs with calbindin revealed no gross differences in the dendritic arborization between PCs innervated by control and C1ql1-overexpressing CFs (Supplementary Fig. 1A). In addition, C1ql1 overexpression in CFs did not affect the membrane capacitance of PCs, an electrophysiological estimate of the total surface area (Supplementary Fig. 1B). Furthermore, AAV-based overexpression of C1ql1 in 6-week-old mice increased the percentage of PCs innervated by multiple CFs in the same manner as in 3-week-old mice (Supplementary Fig. 1C, D). These results indicate that the effect of C1ql1 overexpression was not confounded by the developmental stage of the PCs.

C1ql1 regulates CF-PC synapse formation by binding to the CUB domain of Bai3 during development [18, 19]. To determine whether the effect of C1ql1 requires Bai3, we overexpressed C1ql1 in Bai3 knockout mice at 3–4 weeks of age. CF-evoked EPSC amplitudes were much smaller in Bai3 knockout than in wild-type mice (Fig. 2C vs. 2G), a result consistent with previous reports [18, 19]. In contrast to conditional knockout mice in which the Bai3 gene was postnatally deleted [18], PCs in constitutional Bai3 knockout mice did not show a multiple innervation pattern by CFs (Fig. 2H). Importantly, overexpression of C1ql1 in CFs did not result in an increase in the amplitude of CF-evoked EPSCs in Bai3 knockout PCs (Fig. 2G). In addition, overexpression of C1ql1 did not affect the percentage of PCs innervated by CFs with multiple excitation thresholds (Fig. 2F2, H). Similarly, we did not detect CF-evoked EPSCs with a slower rise time (Fig. 2F2, I). Taken together, these results suggest that C1ql1 overexpression, likely via binding to Bai3, could induce the formation of new CF synapses and increase the proportion of PCs innervated by multiple CFs in mature PCs.

C1ql1-Bai3 signaling induces synapse formation by transverse CF branches

How can a PC in which excess CFs have already been pruned except for a dominant single CF be innervated again by other CFs? A previous in vivo time-lapse imaging study showed that while ascending branches of CFs formed stable synapses with proximal dendrites of PCs, the thin transverse branches were highly dynamic and did not make synapses in adult wild-type mice [26]. Thus, to clarify the contribution by transverse CF branches, we traced GFP-positive CFs in coronal cerebellar sections from wild-type mice to which AAV-TRE-GFP (control) or AAV-TRE-GFP-P2A-C1ql1 was injected at 3–4 weeks of age (Fig. 3A, B1). Co-immunostaining of GFP and vGluT2 revealed that the transverse CF branches were observed at various locations along the PC dendrites, but they mostly lacked vGluT2 in control sections (Fig. 3B2, C, D), indicating their inability to form functional synapses as reported previously [20, 26, 27]. Interestingly, when C1ql1 was overexpressed in CFs, transverse branches elongated and often became positive for vGluT2 (Fig. 3B2, C, D). The elongation of the transverse branch occurred mostly in distal dendrites (80–160 μm from the soma) (Fig. 3 C, E). Transverse branches that were positive for vGluT2 were longer than those negative for vGluT2 (Fig. 3F). Since EPSCs at synapses farther electrotonic distance from the recording site show a slower rise time, these results suggest that an increased proportion of PCs innervated by multiple CFs is at least partly caused by the transverse CF branches forming synapses on distal dendrites.

Fig. 3

C1ql1-Bai3 signaling induces synapse formation by transverse CF branches

A Experimental scheme. B Coronal cerebellar sections. GFP (CTRL) or GFP plus C1ql1 (+ C1ql1) was overexpressed in CFs. Maximum intensity z-projection images are shown. Dotted lines, upper and lower boundaries of the molecular layer (B1). Enlarged views of representative CF branches (B2). Arrowheads indicate vGluT2-negative and positive branches in CTRL and + C1ql1, respectively. Scale bars, 20 μm. C Height and length of CF transverse branches in the molecular layer. Branch height was measured from the apical pole of PC somata. Transverse branches negative (-) and positive (+) for vGluT2 are indicated by white and red circles, respectively. n = 152 branches (CTRL), n = 132 branches (+ C1ql1). D Percentage of vGluT2-positive CF transverse branches in CTRL or + C1ql1. p = 1.352 × 10− 5, n = 152 (CTRL); n = 132 (+ C1ql1). Mann–Whitney U test. E Histogram showing the mean length of CF transverse branches as a function of their height in the molecular layer. Black and orange bars represent the cerebellum in CTRL and + C1ql1, respectively. 0–40 μm: p = 0.8328, n = 3 (CTRL), n = 18 (+ C1ql1); 40–80 μm: p = 0.1731, n = 34 (CTRL), n = 36 (+ C1ql1); 80–120 μm: p = 2.648 × 10− 5, n = 63 branches (CTRL), n = 58 (+ C1ql1); 120–160 μm: p = 7.370 × 10− 5, n = 49 (CTRL), n = 20 (+ C1ql1); >160 μm: p = 0.1321, n = 3 (CTRL), n = 9 (+ C1ql1). Two-tailed Welch’s t-test. F Histogram showing the mean length of CF transverse branches with the presence or absence of vGluT2. CTRL: p = 0.4338; vGluT2(-), n = 137; vGluT2(+), n = 15. +C1ql1: p = 0.003482; vGluT2(-), n = 93; vGluT2(+), n = 39. Two-tailed Welch’s t-test. All datasets are from 3 mice per group (CTRL and + C1ql1). Bars represent mean ± SEM. **p < 0.01; *p < 0.05; ns, not significant

To determine whether the effect of C1ql1 on CF transverse branches required Bai3, we next traced CFs in coronal cerebellar sections from Bai3 knockout mice (Supplementary Fig. 2A) to which AAV-TRE-GFP (control) or AAV-TRE-GFP-P2A-C1ql1 was injected. In Bai3 knockout mice, the overexpression of C1ql1 in CFs did not elongate the transverse CF branch or increase the percentage of vGluT2-positive terminals (Supplementary Fig. 2B-F). These results indicate that overexpression of C1ql1, likely via interaction with Bai3, could induce the growth and synapse formation by of transverse CF branches, resulting in an increased number of PCs re-innervated by multiple CFs after CF-PC synapses mature.

Bai3 overexpression in PCs induces re-innervation by surplus CFs through C1ql1 binding

While new CF-PC synapses were induced by overexpression of C1ql1 in CFs, it was unclear whether PCs overexpressing Bai3 could form new synapses with CFs with normal levels of C1ql1. To address this question, we used lentivirus with the murine stem cell virus (MSCV) promoter [28] to preferentially express EGFP and Bai3 in PCs of mice at 3–4 weeks of age (Fig. 4A). Bai3 expression levels were estimated to be increased by approximately 1.8-fold (Supplementary Fig. 3A, B). We recorded CF-evoked EPSCs from whole-cell patch-clamped PCs by placing the stimulating electrode in the granular layer near the PC soma. CF-EPSCs, which were confirmed by the paired-pulse depression, were elicited in an all-or-none manner in PCs expressing EGFP only, confirming that approximately 90% of wild-type PCs are innervated by a single CF input (Fig. 4 C, D). In contrast, EPSCs were evoked by two or three thresholds of stimulation in PCs overexpressing wild-type Bai3 (Fig. 4 C, D), suggesting that Bai3 overexpression in PCs induces re-innervation by surplus CFs.

Fig. 4

Bai3 overexpression in PCs induces re-innervation by CFs at distal dendrites by binding to C1ql1

A Experimental scheme. B Diagram of the functional domains of Bai3 and its mutants. C Representative CF-EPSC traces recorded from adult wild-type PCs overexpressing the indicated constructs. Paired-pulse stimulation with 50-ms interstimulus interval was applied. D The percentages of the number of CFs innervating single PCs. The number of EPSCs evoked by distinct CF activation thresholds (step numbers) is shown. Bai3-WT: p = 0.0100, n = 69 cells from 11 mice; Bai3-AAA: p = 0.0278, n = 30 cells from 4 mice; Bai3-ΔCT7: p = 0.0007, n = 60 cells from 8 mice; Bai3-S832A: p = 0.0186, n = 22 cells from 3 mice; Bai3-ΔCUB: p = 0.9981, n = 63 cells from 8 mice. Kruskal-Wallis test followed by Steel test vs. CTRL: n = 57 cells from 7 mice. E Average of the 10–90% rise time of CF-EPSCs in PCs overexpressing Bai3-WT. p = 0.0018, Two-tailed Welch’s t-test, n = 24 traces (main), n = 14 traces (surplus) from 11 mice. F Time course of CF-evoked Ca2+ changes associated with main and surplus EPSPs. Changes in the fluorescence (ΔF) were normalized by the averaged fluorescence (F0) before the CF stimulation (arrowhead). Inset, representative CF-EPSPs during Ca2+ imaging. G Representative CF-evoked Ca2+ changes associated with main and surplus EPSPs. S, PC soma. Scale bar, 20 μm. H The mean area of Ca2+ elevation associated with main or surplus CF-EPSPs. The area of large Ca2+ elevation (ΔF/F0 within 30% of the peak value) was measured. p = 2.763 × 10− 5, two-tailed Welch’s t-test, n = 7 responses (main), n = 7 (surplus) from 5 mice. I The closest distance between the site of large Ca2+ elevation and the PC soma was measured (see Methods). p = 0.0075, two-tailed Welch’s t-test, n = 7 responses (main), n = 7 (surplus) from 5 mice. Bars represent mean ± SEM. **p < 0.01; *p < 0.05; ns, not significant

To rule out an effect of Bai3 overexpression on PC development, we examined the membrane capacitance of PCs. As in the case of C1ql1 overexpression in CFs, overexpression of Bai3 in PCs did not affect the membrane capacitance of PCs (Supplementary Fig. 4A), suggesting no gross differences in the total surface area of PCs. Furthermore, lentivirus-based overexpression of Bai3 in 6-week-old mice increased the percentage of PCs innervated by multiple CFs in the same manner as injection into 3-week-old mice (Supplementary Fig. 4B, C). These results indicate that the effect of Bai3 overexpression was not confounded by the developmental stage of the PCs.

Bai3 belongs to the adhesion G protein-coupled receptor family, which mediates intracellular signaling through distinct functional domains [29,30,31,32]. To gain insight into the signaling mechanism mediated by Bai3, we expressed Bai3 with mutations in these functional domains (Fig. 4B). Expression of Bai3-AAA, disabling the ELMO binding motif [29, 30], Bai3-ΔCT7, which lacked the PDZ binding motif [32] and Bai3-S832A, which disrupted the proteolysis sequence in the GPCR auto-proteolysis-inducing (GAIN) domain [31], had similar effects as wild-type Bai3 in inducing re-innervation of PCs by surplus CFs (Fig. 4 C, D). In contrast, the expression of Bai3-ΔCUB, which lacked the binding site for C1ql1 [18], did not affect the pattern of CF innervation in mature PCs (Fig. 4 C, D). These results indicate that overexpression of Bai3 in mature PCs induced innervation by additional CFs by binding to C1ql1, but independently of ELMO, PDZ proteins or proteolysis at the GAIN domain.

Bai3 overexpression in PCs induces re-innervation by CFs at distal dendrites

The largest EPSCs observed in PCs overexpressing Bai3, which we termed “main CF-EPSC”, had similar kinetics to EPSCs seen in control PCs, but the smaller EPSCs (surplus CF-EPSCs) elicited by distinct stimulus thresholds had slower rise times (Fig. 4E). Since overexpression of C1ql1 in CFs also caused the appearance of small CF-evoked EPSCs with slower kinetics (Fig. 2E) and synapse formation at distal dendrites by transverse CF branches (Fig. 3C), we hypothesized that Bai3 overexpression in PCs similarly induced new CF synapse formation on distal dendrites.

To test this hypothesis morphologically, we expressed EGFP and Bai3 in PCs by lentivirus and sparsely labeled IONs by injecting AAV-TRE-tdTomato into Htr5b-tTA knock-in mice (Supplemental Fig. 3C). We found a few PCs that expressed Bai3 and were selectively innervated by transverse CF branches expressing tdTomato without labeled main CF inputs (Supplemental Fig. 3D). However, since the identification of surplus CF branches relies on the coincidental sparse labeling of PCs and CFs, it was difficult to quantify the effect of Bai3 on the formation of surplus CF synapses by the immunohistochemical method.

To clarify the location of surplus CF synapses that gave rise to EPSCs with slow kinetics, we next used the electrophysiological mapping method. We systematically moved the stimulating electrode every 10 μm in the XY direction in the granular layer (Supplementary Fig. 5A). We found that at some locations, main and surplus EPSCs could be evoked by varying the stimulus intensity, while at other locations, only main or surplus EPSCs were selectively evoked. Overall, the location of the stimulating electrode that evoked surplus EPSCs was farther from the PC soma than that elicited main EPSCs (Supplementary Fig. 5B). These results suggest that in PCs overexpressing Bai3, surplus EPSCs are evoked by CFs that travel farther from the cell body than the main CF.

To directly visualize where surplus CFs formed functional synapses with PCs overexpressing Bai3, we loaded PCs with a Ca2+ indicator (Oregon green BAPTA-1) through a patch electrode. We first identified the sites where only main or surplus CF-EPSCs were selectively evoked (Supplementary Fig. 5C) and then performed Ca2+ imaging under the current-clamp mode (Fig. 4F-I). Stimulation of sites where main EPSPs were selectively elicited caused greater increases in Ca2+ concentrations from a larger dendritic area than stimulation of sites where surplus EPSPs were selectively elicited (Fig. 4G, H). Furthermore, Ca2+ elevations associated with surplus EPSPs were observed in dendrites more distal to the PC soma than those associated with the main EPSPs (Fig. 4G, I). These results further support the hypothesis that overexpression of Bai3 in PCs causes the formation of new CF synapses on dendrites more distal to the main CFs, resulting in multiple CF innervation of mature PCs.

During development, the inhibitory inputs from molecular layer interneurons and CFs compete for synapses on PC somata [33]. In GluD2 knockout mice, PFs and CFs compete for synapses on distal dendrites of PCs [20]. Therefore, to explore the possibility that Bai3 may affect other types of PC synapses, we recorded PF-evoked EPSCs, which were confirmed by paired-pulse facilitation, in PCs overexpressing Bai3 (Supplementary Fig. 6A). The amplitudes of PF-EPSCs in response to increasing stimulus intensities were similar between PCs expressing EGFP only (control) and EGFP plus Bai3 (Supplementary Fig. 6B, C). Miniature inhibitory postsynaptic currents (mIPSCs) recorded from PCs overexpressing Bai3 Supplementary Fig. 6D) and control showed similar amplitudes and frequencies (Supplementary Fig. 6E, F). Although local competition may be missed because PF and inhibitory synapses outnumber CF synapses, these results indicate that overexpression of Bai3 in mature PCs preferentially induces CF synapses without significantly altering the number of other synapses.

Endogenous Bai3 and C1ql1 are involved in the re-innervation of CFs in mature PCs

Can CFs form new synapses in mature cerebellar circuits that do not overexpress C1ql1 or Bai3? Loss of PF-PC synapses in conditional GluD2 knockout mice has been reported to trigger re-innervation of PCs through CF transverse branches without exogenous manipulation of C1ql1-Bai3 signaling [20]. Therefore, we investigated the role of endogenous Bai3 in conditional GluD2 knockout mice. Using lentivirus with an MSCV promoter, we sparsely expressed a Cre recombinase and EGFP in PCs of 3–4-week-old wild-type and conditional GluD2 (Grid2f/f) and/or Bai3 (Bai3f/f) knockout mice (Fig. 5A, Supplementary Fig. 7A). Whole-cell patch-clamp recordings from acute cerebellar slices prepared from conditional GluD2 knockout mice two months after Cre introduction revealed that PCs were innervated by multiple CFs with distinct thresholds (Fig. 5B, D), as previously reported [20]. In contrast, CF evoked smaller EPSCs in an all-or-none manner in conditional Bai3 knockout mice (Fig. 5B-D), indicating that the pattern of innervation of PCs by a single CF is unaffected by knocking out Bai3 in adult mice as reported previously [12]. Interestingly, in contrast to GluD2 knockout mice, when both GluD2 and Bai3 were knocked out, CF-EPSCs became smaller (Fig. 5B, C), but many PCs remained innervated by a single CF (Fig. 5D). Furthermore, C1ql1 immunopositive puncta were significantly upregulated in the upper molecular layer of GluD2 knockout mice at 2–3 months of age (Supplemental Fig. 7B, C, D), suggesting the involvement of C1ql1 in CF synapse formation in GluD2 knockout mice. These results suggest that endogenous Bai3, probably together with endogenous C1ql1, is required for re-innervation of mature PCs by CFs in GluD2 knockout mice.

Fig. 5

Endogenous Bai3 is required for synapse formation of re-innervating CFs

A Experimental scheme. B Representative CF-EPSC traces recorded from PCs of the indicated genotypes. C Total CF-EPSC amplitude. The graph shows the sum of peak amplitudes of single CF-EPSCs or multiple CF-EPSCs. CTRL (n = 17 cells from 2 mice) vs. Bai3 knockout (KO), p = 1.272 × 10− 6 (n = 29 cells from 5 mice); vs. GluD2 KO, p = 0.8311 (n = 25 cells from 5 mice); vs. Bai3 KO::GluD2 KO, p = 1.336 × 10− 6 (n = 33 cells from 4 mice). One way ANOVA followed by Dunnett’s test. D The percentages of the number of CFs innervating single PCs. The number of EPSCs evoked by distinct CF activation thresholds (step numbers) is shown. CTRL (n = 29 cells from 2 mice) vs. Bai3 KO, p = 0.9859 (n = 49 cells from 5 mice); vs. GluD2 KO, p = 1.885 × 10− 5 (n = 46 cells from 5 mice); vs. Bai3 KO::GluD2 KO, p = 0.3750 (n = 61 cells from 4 mice). Kruskal-Wallis test followed by Steel test. Bars represent mean ± SEM. **p < 0.01; ns, not significant

Bai3-induced re-innervation of PCs by CFs requires PC activity

Since structural synaptic plasticity occurs in an activity-dependent manner throughout life in the mammalian brain [34, 35], we next investigated whether increased C1ql1-Bai3 levels could bypass neuronal activity to form new CF synapses in mature PCs. Using an AAV-based Cre-DIO (double-floxed inverse open reading frame) system, we expressed EGFP and ESKir2.1, a non-rectifying variant of Kir2.1 potassium channel [36], to specifically suppress levels of intrinsic PC activity (Supplementary Fig. 8A). As a control, we used ESKir2.1AAA, a mutant channel lacking channel activity [36]. Loose-patch recordings in acute cerebellar slices prepared from mice 2–3 weeks after the AAV injection at 3–4 weeks of age confirmed the absence of spontaneous action potentials in PCs expressing ESKir2.1, but not ESKir2.1AAA (Supplementary Fig. 8B, top traces). Whole-cell voltage-clamp recordings revealed that the amplitude of CF-evoked EPCSs was reduced in PCs expressing ESKir2.1, but the number of stimulus thresholds (reflecting the number of CF inputs) was similar to control PCs (Supplementary Fig. 8B-D). Similarly, the application of tetrodotoxin or NBQX is reported to reduce the amplitude of CF-EPSCs and CF synapses in adult PCs [11,12,13]. While the site of action was unclear in these pharmacological studies, our findings indicate that the intrinsic activity of PCs is required to maintain CF synapses in mature PCs.

Next, we examined whether Bai3 overexpression could induce CF re-innervation in PCs expressing ESKir2.1 by coinfecting L7-Cre mice with AAV-Syn-DIO-ESKir2.1-T2A-EGFP and Lenti-MSCV-mCherry-P2A-Bai3 (Fig. 6A). In PCs overexpressing Bai3 and ESKir2.1AAA, we detected CF-EPSCs with multiple thresholds (Fig. 6B, left traces; Fig. 6D), as observed in the absence of ESKir2.1AAA (Fig. 4 C, D). In contrast, CF stimulation evoked smaller EPSCs with a single threshold in most PCs coexpressing ESKir2.1 and Bai3 (Fig. 6B, middle traces; Fig. 6D). These results indicate that intrinsic PC activity is required for Bai3 to re-innervate mature PCs.

Fig. 6

Bai3-induced re-innervation by CFs requires PC activity

A Experimental scheme. The right panel shows three cases: in all cases recorded PCs overexpress Bai3, but all PCs are active (spiking PCs, left), all PCs are silenced by ESKir2.1 (silent PCs, middle) and recorded PCs are active but neighboring PCs are silenced (right). B Representative action potentials by loose-patch recordings (upper traces) and CF-EPSCs by whole-cell patch-clamp recordings (lower traces) from PCs expressing the indicated constructs. C Total CF-EPSC amplitude. The graph shows the sum of peak amplitudes of single CF-EPSCs or multiple CF-EPSCs. p = 0.0098, CTRL + Bai3 vs. ESKir2.1 + Bai3; p = 0.9500, CTRL + Bai3 vs. ESKir2.1 + Bai3, GFP (-); p = 0.0393, ESKir2.1 + Bai3 vs. ESKir2.1 + Bai3, GFP (-). One way ANOVA followed by Tukey’s test. n = 16 cells from 3 mice (CTRL + Bai3), n = 18 cells from 3 mice (ESKir2.1 + Bai3), n = 12 cells from 3 mice (ESKir2.1 + Bai3, GFP (-)). D The percentages of the number of CFs innervating single PCs. The number of EPSCs evoked by distinct CF activation thresholds (step numbers) is shown. p = 0.0061, CTRL + Bai3 vs. ESKir2.1 + Bai3; p = 0.0182, CTRL + Bai3 vs. ESKir2.1 + Bai3, GFP (-); p = 0.9894, ESKir2.1 + Bai3 vs. ESKir2.1 + Bai3, GFP (-). Kruskal-Wallis test followed by Steel test n = 44 cells from 5 mice (CTRL + Bai3), n = 49 cells from 9 mice (ESKir2.1 + Bai3), n = 37 cells from 6 mice (ESKir2.1 + Bai3, GFP (-)). Bars represent mean ± SEM. **p < 0.01; *p < 0.05; ns, not significant

In these experiments, expression of ESKir2.1 (as detected by EGFP) was widespread in many PCs at the injection site, whereas Bai3 (as detected by mCherry) was detected in only a few PCs (Supplementary Fig. 8E). However, we occasionally found non-silenced PCs expressing only Bai3, surrounded by silent PCs expressing only ESKir2.1 (Fig. 6A, right; Supplementary Fig. 8E, bottom). The amplitude of CF-EPSCs in such non-silenced PCs was similar to that in control PCs expressing ESKir2.1AAA and Bai3 (Fig. 6C). Unexpectedly, however, CF-EPSCs were evoked by a single threshold in these non-silenced PCs (Fig. 6B, right traces; Fig. 6D). These results suggest that multiple innervation by CFs requires neuronal activity not only in the Bai3-expressing PCs but also in the surrounding PCs.

CF activity is required for C1ql1 to induce the innervation of adult PCs by CFs

Finally, we investigated whether increased C1ql1 levels in CFs could induce new CF synapses in the absence of CF activities. To suppress CF activities in vivo, we inject