Cerebellar Neurodegeneration and Neuronal Circuit Remodeling in Golgi pH Regulator-Deficient Mice

Abstract The Golgi apparatus plays an indispensable role in posttranslational modification and transport of proteins to their target destinations. Although it is well established that the Golgi apparatus requires an acidic luminal pH for optimal activity, morphological and functional abnormalities at the neuronal circuit level because of perturbations in Golgi pH are not fully understood. In addition, morphological alteration of the Golgi apparatus is associated with several neurodegenerative diseases, including Parkinson’s disease, Alzheimer’s disease, and amyotrophic lateral sclerosis. Here, we used anatomical and electrophysiological approaches to characterize morphological and functional abnormalities of neuronal circuits in Golgi pH regulator (GPHR) conditional knock-out mice. Purkinje cells (PCs) from the mutant mice exhibited vesiculation and fragmentation of the Golgi apparatus, followed by axonal degeneration and progressive cell loss. Morphological analysis provided evidence for the disruption of basket cell (BC) terminals around PC soma, and electrophysiological recordings showed selective loss of large amplitude responses, suggesting BC terminal disassembly. In addition, the innervation of mutant PCs was altered such that climbing fiber (CF) terminals abnormally synapsed on the somatic spines of mutant PCs in the mature cerebellum. The combined results describe an essential role for luminal acidification of the Golgi apparatus in maintaining proper neuronal morphology and neuronal circuitry.


Introduction
The flattened, cisternal structure of the Golgi apparatus is essential for ensuring that proteins are correctly glycosylated and sorted to the appropriate destination (Klumperman, 2011;Papanikou and Glick, 2014). The optimal activity of intracellular organelles, such as the Golgi apparatus, is strictly regulated by the acidic pH of the lumen (Weisz, 2003). Perturbation of the luminal pH not only has marked effects on posttranslational modification and transport of proteins but also causes morphological abnormalities in the Golgi apparatus (Palokangas et al., 1994;Kellokumpu et al., 2002). Similar morphologic abnormalities have also been observed in cerebellar Purkinje cells (PCs; Baloyannis, 2014) and motor neurons in patients with neurodegenerative disorders (Gonatas et al., 1992;Mourelatos et al., 1996;Haase and Rabouille, 2015;Ayala and Colanzi, 2017). However, the causative link between impaired luminal acidification of the Golgi apparatus, fragmentation of the Golgi apparatus and neurodegeneration remains unresolved.
Several hypotheses have been proposed to explain how fragmentation of the Golgi apparatus can cause neurodegeneration. Some of these include microtubule depolymerization, pathogenic protein misfolding, mislocalization of aggregate-prone proteins and phosphorylation of GRASP65, a structural protein of the Golgi apparatus (Alvarez-Miranda et al., 2015). In addition to these explanations, we hypothesized that the impairment of luminal acidification of the Golgi apparatus causes fragmentation, which results in neurodegeneration. Ubiquitin protein ligase E3A (Ube3a), a causative gene product of Angelman's syndrome, is involved in the acidification of the Golgi lumen (Condon et al., 2013). Interestingly, the Golgi-associated protein, ATP6V0A2, is involved in congenital disorders of glycosylation (CDG) because of a pH imbalance in the Golgi apparatus (Kornak et al., 2008;Demaegd et al., 2013). Additionally, patients with several types of CDG as a result of mutations in genes encoding Golgi-associated proteins show cerebellar atrophy and ataxia (Climer et al., 2015).
We previously identified an anion channel that is localized to the Golgi apparatus and is a Golgi pH regulator (GPHR). The GPHR is essential for maintaining Golgi luminal acidification (Maeda et al., 2008). Using keratinocyte-specific GPHR knock-out mice, the GPHR was found to be essential for proper skin barrier function (Tarutani et al., 2012). Here, we took advantage of the GPHR conditional knock-out mice to better understand the morphologic and functional abnormalities in cerebellar circuits. We show that deletion of GPHR in cerebellar inhibitory circuits [PCs and basket cells (BCs)] leads to cerebellar ataxia due to PC degeneration. Impaired luminal acidification of the Golgi apparatus leads to vesiculation and fragmentation in PCs. Importantly, we found that axonal and synaptic abnormalities precede the loss of PCs. Moreover, the loss of GPHR in BCs causes disruption of inhibitory input to PC soma and further leads to climbing fiber (CF) innervation onto the soma of mutant PCs. These results indicate that GPHR-dependent Golgi luminal acidification plays an essential role in the structural and functional maintenance of PCs and BCs.
the peptide Cys-AHKQAPEKHMAP as the antigen. Rat hybridoma cells (Clone 35C5) were obtained by standard approaches.

Plasmids
The plasmid containing the open reading frame of the pH sensor protein, pHluorin, was kindly provided by Dr. James Edward Rothman (Miesenböck et al., 1998). We used the N-terminal 81 amino acids of human beta 1,4galactosyltransferase as a targeting signal for the Golgi apparatus. The plasmids, pHluorin-Golgi and pmKate2-Golgi, were constructed based on pEYFP-Golgi (TA-KARA/Clontech, catalog #6909-1) by replacing the sequence encoding EYFP with the sequences encoding pHluorin and mKate2, respectively (Tanida et al., 2014). The DNA fragment encoding pHluorin-Golgi was introduced into the NheI-EcoRI sites of pIRES, and the DNA fragment encoding mKate2-Golgi was introduced into the SalI-NotI sites of pIRES (TAKARA/Clontech, catalog #Z1605N). The resultant plasmid was named as pHluorin-Golgi-IRES-mKate2-Golgi.

Primary culture of cortical neurons and adenovirus infection
Primary cortical neurons were prepared according to a general protocol. Briefly, the cortex was dissected and dissociated from E14 mice embryos. Dissociated cortical cells were plated on poly-L-Ornithine (Sigma-Aldrich) coated 14-mm coverslips. Neurobasal medium (Thermo Fisher Scientific) supplemented with 2% B27 (Thermo Fisher Scientific) and 2 mmol L-glutamine (Nacalai Tesque) was used as the culture medium. After 2 d, 10 mmol/L cytosine arabinoside (Nacalai Tesque) was added to inhibit non-neuronal growth. Two days after this step, with the aim of achieving equal infection, the culture medium containing the adenovirus with 10 multiplicity of infection (MOI) was added to the medium. The lysates from cultured neurons were subjected to immunoblotting to detect GPHR and actin. Adenovirus vectors: Ad-CMV-iCre (#1045) and Ad-GFP (#1060) were purchased from Vector Biolabs.

Measurement of pHluorin intensity
pHluorin-Golgi and mKate2-Golgi were expressed by adenovirus-mediated transfection of pHluorin-Golgi-IRES-mKate2-Golgi in primary cultured neurons using Lipofectamine 2000 (Thermo Fisher Scientific). Seven days after transfection, live-cell images of primary cultured neurons expressing pHluroin-Golgi-IRES-mKate2-Golgi were obtained using a FV1000 confocal laser scanning microscope (Olympus). The intensities of pHluorin and mKate2 in the same ROI were measured by the histogram analysis plugin in ImageJ (Schneider et al., 2012). The mean of the intensity ratio (pHluorin/mKate2) was calculated and processed for statistical analysis.

Histological examination
Deeply anesthetized mice were fixed by cardiac perfusion with 4% paraformaldehyde in 0.1 M phosphate buffer (pH 7.4). Tissues were embedded in paraffin or 4583;Sakura Fintek). Threemicron paraffin sections were stained with 0.1% cresyl violet for Nissl staining. Meyer's hematoxylin and eosin (H & E) staining was performed on 20-m-thick cryosections. For immunofluorescence, frozen sections were subjected to antigen retrieval by incubating in target retrieval solution (S1699, Dako) for 10 min at 105°C and blocked with TNB buffer (0.1 M Tris-HCl, pH 7.5, 0.15 M NaCl, and 0.5% TSA blocking reagent; FP102; PerkinElmer) at room temperature for 1 h. The sections were then incubated with a mixture of primary antibodies at 4°C overnight, followed by incubation with fluorescently labeled secondary antibodies [Alexa Fluor 488 (Thermo Fisher Scientific) for Cy3 or Cy5 (Jackson ImmunoResearch)] containing DAPI at room temperature for 1 h. Fluorescence images were obtained using a FV1000 confocal laser scanning microscope (Olympus).

Stereological sectioning and quantification of PCs
The total number of PCs in the mouse cerebellum was estimated by stereological quantification and optical disector counting methods, as described previously (Chintawar et al., 2009(Chintawar et al., , 2012Ip et al., 2017). Sagittal cerebellar cryosections from control and GPHR F/F ;GluD2-Cre mice (three mice per group) were serially cut at 20-m thicknesses to cover the entire cerebellum. Systematic random sampling of every 20th section collected was performed for a total of 18 sections per mice. Sections were immunostained with the anti-calbindin antibody to reveal PC soma and their dendrite in the molecular layer. Calbindin immunostaining images were visualized by using a fluorescence microscope BZ-X710 with a 20ϫ objective lens (Keyence). The optical disector method was applied to identify and count PC tops and thus to calculate average PC densities (Nv). The reference volume (Vref) from PCs and molecular layers was obtained using Cavalieri's principle. Finally, the total number of PCs (Npc) was estimated according to the formula: Npc ϭ Vref ϫ Nv. The images were analyzed using ImageJ cell counter and Grid layout plugins (Schneider et al., 2012).

Electron microscopy (EM)
Mice were fixed by cardiac perfusion with 4% paraformaldehyde in 0.1 M phosphate buffer (pH 7.4). Cerebellar sections were post fixed with 2% paraformaldehyde and 2% glutaraldehyde in 0.1 M phosphate buffer (pH 7.4) overnight followed by post fixation with 1% OsO 4 , dehydration with a graded series of ethanol and embedding in Epon812 (Oken Shoji). Ultrathin sections were cut with an ultramicrotome UC6 (Leica Microsystems), stained with uranyl acetate and lead citrate and examined with a transmission EM HT7700 (Hitachi). For volume EM, images were taken with a focused ion beam scanning EM (FIB-SEM) Helios Nanolab 660 (FEI Company). Epon blocks were milled by a focused ion beam and a serial block-face for every 20 nm was imaged using a backscattered electron detector (MD detector) at an acceleration voltage of 2.0 kV and a current of 0.4 nA. 3D reconstruction was performed using Amira software (FEI Company).

Measurement of the cerebellar area, PC density, and microglia density
Measurement of the cerebellar area was conducted as described previously (Xu et al., 2018). Sagittal cerebellar cryosections (20 m thick) from control and GPHR F/F ; GluD2-Cre mice (three mice per group) were serially cut and every 10th section was collected for H & E staining. Cerebellar size was measured from five sections corresponding to positions 102-104 of the mouse brain atlas (Franklin and Paxinos, 2007). Sections were visualized by using a fluorescence microscope BZ-X710 with a 4ϫ objective lens. Images were stitched into a whole cerebellum by the BZ-X image analyzer (Keyence) and the area of the cerebellum was measured. The mean area was calculated and processed for statistical analysis. For quantification of PC densities, cerebellar sections from control and GPHR F/F ;GluD2-Cre mice (three mice per group) were immunostained with the anti-calbindin antibody. The number of calbindin-positive PCs from each cerebellum was counted from a total of five sections, corresponding to positions 102-104 of the mouse brain atlas (Franklin and Paxinos, 2007). The PC density was calculated as the total number of PC somata divided by the sum of the area of PC and molecular layers. The mean PC densities were calculated and processed for statistical analysis. To measure microglia density, sections containing cerebellar lobule IX from control and GPHR F/F ;GluD2-Cre mice (three mice per group) were immunostained with the anti-IBA1 antibody, and nuclei were stained with DAPI. The number of total IBA1-positive cells was counted on five sections. The microglia density was obtained by dividing the IBA1-positive cells, with clearly visible nuclei, by the total area of the lobule. The mean density was obtained from all the microglia analyzed and processed for statistical analysis. All images were analyzed using ImageJ.

Morphometry of PCs
To quantify immunofluorescence images, five sections containing cerebellar lobule IX from each mouse were immunostained and analyzed using ImageJ. Measurements were obtained from three mice in each group. To calculate the ratio of PCs in which the Golgi apparatus is abnormally distributed, the number of PCs with such abnormality was divided by the total number of PCs. The abnormal distribution of the Golgi apparatus in PCs was defined by perinuclear accumulation of GM130 and TGN38, as revealed by immunostaining. The mean ratio of PCs with abnormal distribution of the Golgi apparatus was calculated and processed for statistical analysis. Cal-bindin immunostaining images were analyzed by ImageJ to quantify axon swelling, which enabled measurement of the size and number of axon swellings in the immunostained sections. An axonal torpedo area exceeding 15 m 2 was defined as swelling axons. The number of swelling axons was divided by the area of granule cell layer or deep cerebellar nuclei (DCNs) to calculate the density of axonal swelling. To count Pinceau structures, sections were immunostained with anti-calbindin and antiparvalbumin antibodies to elucidate BC axon terminals. The ratio of PCs with Pinceau structures was calculated as the total number of PCs with Pinceau structures divided by the total number of PCs. The ratio obtained was processed for statistical analysis. To quantify the frequency of mGluR1␣ and vGlut2 single-positive or doublepositive puncta, analysis was restricted to the bottom two-thirds of the PC soma to avoid contamination of the analysis by puncta on dendrites. The ratio of vGlut2 puncta on PC soma was calculated by dividing the positive cells with the total number of cells analyzed. The mean obtained was processed for statistical analysis. For quantification of lamellar bodies, defined as smooth ERs with more than three stacks, in the cytoplasm of PCs, electron micrographs taken from control (n ϭ 5, P60) and GPHR F/F ;GluD2-Cre (n ϭ 5, P60) mice were analyzed. The number of lamellar bodies was counted and divided by the total area of the PC cytoplasm to show lamellar body densities. The mean of lamellar body densities was calculated and processed for statistical analysis.

Behavioral tests
The accelerating rotarod test was performed on a rotarod machine with automatic timers and falling sensors (MK-660D, Muromachi Kikai). Control and GPHR F/F ; GluD2-Cre mice (n ϭ 5 for each genotype of females) were placed on a rotating rod, rotation was accelerated to 40 rpm for 300 s, and the latency period for each mouse to fall off was measured. If a mouse remained on the rod for Ͼ300 s, the latency period was recorded as 300 s. Mice underwent three trials per day and the mean latency period of the three trials was considered for statistical analysis. Cerebellar ataxic phenotypes of control and GPHR F/F ;GluD2-Cre mice (three mice per group) were evaluated at P45, P60, P90, P150, and P300 using the composite phenotype scoring system based on hindlimb clasping, open field gait and ledge tests (Guyenet et al., 2010). Each test was repeated three times. All tests were scored on a scale of 0 -3, with a combined total of 0 -9 for all three tests. For the limb clasping test, mice were lifted for 20 s by grasping their tail and movement of the hind limbs was scored as follows. A score of 0 indicates a mouse spreading their hind limbs away from the abdomen. A score of 1 indicates a mouse pulling their hindlimb partially toward their abdomen for Ͼ5 s. A score of 2 indicates a mouse pulling both hind limbs partially toward their abdomen for Ͼ5 s. A score of 3 indicates a mouse retracting hind limbs and touching the abdomen for Ͼ50% of the observation time. For the open field gait test, mice were placed on a flat surface and oriented away from the researcher. A score was recorded as follows. A score of 0 indicates that a mouse moved normally and its abdomen did not touch the surface. A score of 1 indicates that tremor was observed or the mouse appeared to limp while walking. A score of 2 indicates a mouse showing severe tremor, severe limp, lowered pelvis or feet pointing away from the body. A score of 3 indicates that a mouse had difficulty to move forward and dragged their abdomen along the surface. For the ledge test, mice were lifted from a cage and placed on the ledge of the cage. A score was recorded as follows. A score of 0 indicates that a mouse walked along the ledge without losing its balance and went back into the cage using its paws. A score of 1 indicates that a mouse lost its footing while walking on the ledge. A score of 2 indicates that a mouse was unable to walk effectively on or let itself down from the ledge. A score of 3 indicates that a mouse was unable to walk, get down from the ledge, or simply fell off. All behavioral tests were performed by researchers blinded to the genotype of the mice.

Statistical analysis
See Table 1 for a summary of statistical results relating to this work. Actual p values and the number of samples are stated in the text, figure legends, and Table 1. All statistical analyses were performed using GraphPad Prism 6 (GraphPad Software Inc.; RRID: SCR_002798). At least three independent samples were used for assessing statistical significance. Normality of the datasets was determined by the Kolmogorov-Smirnov test. Comparisons between multiple experimental groups for the accelerating Rotarod test, measurement of cerebellar area, quantitative analysis of PC density, Pinceau structure counting and quantification of the frequency of vGlut2 were made using one-way ANOVA followed by Sidak's multiple comparisons test. For comparing two experimental groups for quantitative real-time PCR, quantitative analysis of the pHluorin intensity, density of microglia, abnormal distribution of the Golgi apparatus, quantitative analysis of axonal swelling, electrophysiological analysis and quantitative analysis of the lamellar body density were conducted using the unpaired Student's t test. For combined scores of the composite phenotype scoring system, which did not follow a normal distribution according to the Kolmogorov-Smirnov test, a non-parametric test was performed using the Kruskal-Wallis rank sum test followed by the Dunn's test. The data are presented as the mean Ϯ SEM; p Ͻ 0.05 was considered statistically significant and p Ͻ 0.01 was considered highly significant.

Elevated luminal pH of the Golgi apparatus in primary cultured neurons lacking GPHR
Initially, we assessed the disruption of the GPHR gene in primary cultured neuron from GPHR floxed embryos by using an adenovirus expression system. Deletion of the GPHR protein was confirmed by immunoblot analysis using an anti-GPHR antibody in primary cultured neurons from GPHR floxed mice expressing Cre recombinase (Fig.  1A). As expected, exogenous expression of Cre recombinase suppressed GPHR mRNA levels in primary cultured neurons, as examined by quantitative real-time PCR ( Fig. 1B; Table 1; GPHR F/F ; AV: GFP: 0.992 Ϯ 0.025, n ϭ 6; GPHR F/F ; AV: Cre: 0.044 Ϯ 0.011, n ϭ 6; unpaired Student's t test, t (10) ϭ 35.41, ‫‪p‬ءء‬ Ͻ 0.0001). The function of GPHR for maintaining the luminal pH of the Golgi apparatus in neurons was examined by constructing a signal localized at the Golgi apparatus to the pH sensor protein pHluorin and red fluorescence protein mKate2, respectively (Miesenböck et al., 1998;Tanida et al., 2014). Using the constructed tandem expression plasmid (i.e., pHluorin-Golgi-IRES-mKate2-Golgi), alkalization of the Golgi apparatus lumen in response to deletion of GPHR in primary cultured neurons was monitored. The fluorescence intensity of pHulorin markedly increased, indicating that deletion of GPHR elevated the luminal pH of the Golgi apparatus in primary cultured neurons (Fig. 1C). Quantitative measurements confirmed that the pHluorin/mKate2 fluorescent ratio of primary cultured neurons lacking GPHR (GPHR F/F ; AV: Cre) is higher than that of the control (GPHR F/ϩ ; AV: Cre) neurons ( Fig. 1D; Table 1; GPHR F/ϩ ; AV: Cre: 1.015 Ϯ 0.038, n ϭ 5; GPHR F/F ; AV: Cre: 1.390 Ϯ 0.068, n ϭ 5; unpaired Student's t test, t (8) ϭ 4.827, ‫‪p‬ءء‬ ϭ 0.0013). These results suggest that GPHR is responsible for maintaining normal luminal acidic conditions of the Golgi apparatus in neurons.
H & E staining at P60 revealed difference in the size of the cerebellum between control and GPHR F/F ;GluD2-Cre mice, and at P90 the difference was more pronounced (Fig. 3A,B; Table 1; one-way ANOVA, three mice per group; five sections per mouse; n ϭ 15 section for each group, F (3,56) ϭ 42.26, p Ͻ 0.0001; and Sidak's post hoc test, control vs GPHR F/F ;GluD2-Cre, ‫‪p‬ءء‬ ϭ 0.003 at P60, control vs GPHR F/F ;GluD2-Cre, ‫‪p‬ءء‬ Ͻ 0.0001 at P90). The histological feature of PC lost was detected in GPHR F/F ; GluD2-Cre mice at P60 and was more severe at P90, as examined by Nissl staining (Fig. 3C). Consistent with Nissl staining, loss of PCs was also detected by calbindin immunostaining. GPHR F/F ;GluD2-Cre mice exhibited loss of PCs especially in lobule IX at P60 and a decrease in PC density was observed in whole cerebellar lobules at P90, as examined by calbindin immunostaining (Fig. 3D,E; Table 1; one-way ANOVA, three mice per group; five sections per mouse; n ϭ 15 sections for each group, F (5,96) ϭ 112.2, p Ͻ 0.0001; and Sidak's post hoc test, control vs GPHR F/F ;GluD2-Cre, p ϭ 0.9213 at P45, ‫‪p‬ءء‬ ϭ 0.002 at P60, ‫‪p‬ءء‬ Ͻ 0.0001 at P90). The number of PCs in control and GPHR F/F ;GluD2-Cre mice by stereological methods were estimated to confirm these histological results ( Table 2). Comparison of the number of PCs in the cerebellum between control and GPHR F/F ;GluD2-Cre mice at P90 revealed that there were only ϳ20% of the PCs remaining in GPHR F/F ;GluD2-Cre mice. These results suggest that the number of PCs in GPHR F/F ;GluD2-Cre mice decreases in an age-dependent manner. Corresponding to the degenerative changes in PCs, activated microglia immunopositive for IBA1 and/or MAC2 were abundant in the cerebellum of GPHR F/F ;GluD2-Cre mice (Fig. 3F). Quantification confirmed that the density of IBA1immunopositive microglia in lobule IX increased significantly in GPHR F/F ;GluD2-Cre mice when compared with that of control mice (

GPHR loss causes abnormal morphology of the Golgi apparatus
Alkalization of the Golgi luminal pH causes morphological abnormalities of the Golgi apparatus in cultured cells (Tartakoff, 1983;Sakaguchi et al., 1996;Maeda et al., 2008). However, very little information can be gleaned from in vivo literature. In normal PCs, the Golgi apparatus is properly distributed by forming a perinuclear strand Figure 2. GPHR F/F ;GluD2-Cre mice lack motor coordination. A, GPHR is deleted in PCs. Double immunofluorescence for GPHR (green) and parvalbumin (Parv; red) in the PC layer from GPHR F/ϩ ;GluD2-Cre mice at P60 and GPHR F/F ;GluD2-Cre mice at P30, P45, and P60. Dotted immunoreactivity for GPHR is not discernable in PCs from GPHR F/F ;GluD2-Cre mice after P45. Scale bars: 20 m. B, Accelerating rotarod test for GPHR F/ϩ ;GluD2-Cre and GPHR F/F ;GluD2-Cre mice. The average retention time on the rotarod is shown (n ϭ 5 mice, female); ‫‪p‬ءء‬ Ͻ 0.001 (one-way ANOVA with Sidak's post hoc test). C, Limb clasping reflex in GPHR F/F ;GluD2-Cre mice at P90. When lifted by the tail, GPHR F/ϩ ;GluD2-Cre mice extended their hind limbs, whereas GPHR F/F ;GluD2-Cre mice moved their legs to the trunk. D, Behavioral analyses of GPHR F/F ;GluD2-Cre mice relative to GPHR F/ϩ ;GluD2-Cre mice. Graph showing cerebellar ataxic phenotypes score of GPHR F/ϩ ;GluD2-Cre and GPHR F/F ;GluD2-Cre mice at P45, P60, P90, P150, and P300 (three mice per group). Average composite score for each genotype at each age was calculated. Data are presented as the mean Ϯ SEM; n ϭ 9 trial for each group, ‫‪p‬ءء‬ Ͻ 0.01 (Kruskal-Wallis test followed by the Dunn's test). For A, three mice per genotype; n ϭ 5 sections per mouse; representative images are shown. See also Movie 1. (Palay and Chan-Palay, 1974;Liu et al., 2017). To characterize the Golgi apparatus in PCs of GPHR F/F ;GluD2-Cre mice, we performed double staining for GM130 and TGN38, which are markers for cis-and trans-Golgi, respectively. In control PCs, GM130-and TGN38-positive Golgi signals were distributed as perinuclear small patches, which were enriched at the apical pole facing the primary dendrite. Each Golgi stack was designated as cisand trans-cisterns. In contrast, GPHR deficiency caused perinuclear accumulation of the immunopositive areas and some were shifted to the basolateral perikarya at P45, indicating an abnormal distribution of the Golgi apparatus (Fig. 4A). After P45, the Golgi apparatus appeared as a mixture of GM130-and TGN38-positive areas (Fig. 4A). Quantification confirmed that the ratio of the abnormal Golgi apparatus distribution in PCs of GPHR F/F ;GluD2-Cre mice increased significantly in an age-dependent manner (Table 1; GPHR F/F ;GluD2-Cre at P45: 66.85 Ϯ 3.13; GPHR F/F ;GluD2-Cre at P60: 90.58 Ϯ 2.23, three mice per group; five sections per mouse; n ϭ 15 sections for each group; GPHR F/F ;GluD2-Cre at P45: 185 PCs; GPHR F/F ;GluD2-Cre at P60: 163 PCs; unpaired Student's t test, t (28) ϭ 6.174, ‫‪p‬ءء‬ Ͻ 0.0001). EM showed that instead of stacks of flattened cisterns, clusters of vacuolar structures were observed in GPHR-deficient PCs at P45, suggesting fragmentation of the Golgi apparatus (Fig. 4B). Furthermore, volumetric reconstruction of the Golgi apparatus by FIB-SEM confirmed fragmentation of the Golgi apparatus with clusters of vacuoles in PCs from GPHR F/F ; GluD2-Cre mice ( Fig. 4C; Movies 2, 3). These data indicated that acidification of the luminal pH of the Golgi apparatus is important for maintaining Golgi apparatus morphology and distribution in PCs.

Golgi abnormality is accompanied by axonal and synaptic impairment in PCs
We reported previously that the regulation of luminal pH of the Golgi apparatus is indispensable for ensuring the normal function of the Golgi apparatus, such as glycosylation and vesicular transport in cultured cells (Maeda et al., 2008). In addition, the Golgi apparatus is important for axonal transport of neurotransmitters in mature neurons (Campenot et al., 2003;Horton et al., 2005;Matsuki et al., 2010;Liu et al., 2017). Thus, morphological alterations in PC axons on perturbations in luminal pH of the Golgi apparatus were investigated. PC axons traverse the granule cell layer to innervate neurons in DCN. First, we focused on PC axons in the granule cell layer. Different from control axons at P60 (Fig. 5A), calbindin immunostaining revealed that axon swellings appeared at P45 in GPHR F/F ;GluD2-Cre mice, and these areas increased in size and number at P60 (Table 1; size: GPHR F/F ;GluD2-Cre at P45: 40.04 Ϯ 2.99 m 2 ; GPHR F/F ;GluD2-Cre at P60: 70.22 Ϯ 4.63 m 2 , unpaired Student's t test, t (28) ϭ 6.405, ‫‪p‬ءء‬ Ͻ 0.0001; density of GPHR F/F ;GluD2-Cre mice at P45: 56.54 Ϯ 4.94/nm 2 ; GPHR F/F ;GluD2-Cre at P60: 80.19 Ϯ 8.56/nm 2 , unpaired Student's t test, t (28) ϭ 2.384, ‫ء‬p ϭ 0.0241; three mice per group; five sections per mouse; n ϭ 15 sections for each group). EM observations revealed myelinated swelling axons in the granular cell layer at P45 in GPHR F/F ;GluD2-Cre mice (Fig. 5B). In the DCN region of GPHR F/F ;GluD2-Cre mice, swelling axons also appeared at P45 and increased prominently in number at P60 ( Fig. 5C; Table 1; density of GPHR F/F ;GluD2-Cre mice at P45: 261.1 Ϯ 32.54/nm 2 ; GPHR F/F ;GluD2-Cre at P60: 880.6 Ϯ 46.07/nm 2 , three mice per group; five sections per mouse; n ϭ 15 sections for each group, unpaired Student's t test, t (28) ϭ 10.98, ‫ء‬p Ͻ 0.0001). Many of these swelling axons were positive for calbindin but negative for the vGAT, which would reflect swelling axons containing materials other than synaptic vesicles, as shown in the EM observations (Fig. 5C,D). Interestingly, some presynaptic PC terminals exhibited an abnormal ultrastructural profile and showed symmetrical synaptic contacts with electron-dense postsynaptic density (Fig.  5E). In addition, we found that axosomatic synapses on individual DCN neurons were reduced as early as P45, leading to intercellular spaces being created around the DCN neurons (Fig. 5D-F). These results indicated that the impairment of acidic conditions in the Golgi apparatus causes axonal and synaptic abnormalities in PCs.

GPHR is required for maintaining the Pinceau structure
Next, we investigated the innervation of BCs, one of the GABAergic interneurons, to the somatodendritic compartment of PCs. The axons of BCs extend toward the bottom of the PC soma and their branches are densely terminated around the axon initial segment (AIS), forming a structure known as the Pinceau (Palay and Chan-Palay, 1974;Gianola et al., 2003;Ango et al., 2004;Sotelo, 2008). We performed double immunostaining of parvalbumin, which labels both PCs and GABAergic interneurons in the cerebellar molecular layer and calbindin that selectively labels PCs. While control BC axons completely surrounded the PC somata forming Pinceau structures around AIS at P60, GPHR F/F ;GluD2-Cre mice exhibited markedly lower numbers of Pinceau structures at P60 (Fig. 6A). To confirm whether the loss of Pinceau structures is because of GPHR-deficiency in BCs, we further analyzed this phenomenon using the PC-specific GPHR knock-out mice, GPHR F/F ;L7-Cre mice. Interestingly, the Pinceau structures remained intact in GPHR F/F ;L7-Cre mice at P60 and   (Fig. 6A, arrows). Quantification confirmed that the mean Pinceau frequency was significantly reduced in GPHR F/F ;GluD2-Cre mice when compared with that of control and GPHR F/F ;L7-Cre mice ( ;L7-Cre, ‫‪p‬ءء‬ Ͻ 0.0001). These data suggest that the physical integrity of Pinceau structures depends strongly on the axon status of BCs. Moreover, in GPHR F/F ;GluD2-Cre mice, Pinceau structures were still observed at P30, whereas they were lost at P60 (Fig. 6B), indicating that Pinceau structures were formed until stabilization of cerebellar circuits.
Double immunostaining for the GABA A R␣1 and vGAT to label postsynapses and presynapses, respectively, was performed to investigate whether BC axons in GPHR F/F ; GluD2-Cre mice still form synapses with PC soma at P60. GABA A R␣1-and vGAT-immunopositive punctae overlapped almost completely along the soma of PCs in control mice at P60, displaying the typical pattern of BC axon terminals (Fig. 6C). In contrast, punctae that were doubly positive for GABA A R␣1 and vGAT were only sporadically detected in PC somata of GPHR F/F ;GluD2-Cre mice (Fig.  6C). Unlike BCs, there were no apparent changes in the distribution of stellate cell axon terminals (Fig. 6C). We further performed EM analysis to determine the ultrastructural changes in the AIS and soma of PCs of GPHR F/F ; GluD2-Cre mice. In control mice, BC axons formed synaptic contacts on the surfaces of PC soma with typical ultrastructural features of asymmetrical synapses, such as synaptic vesicle accumulation in presynaptic terminals, rigid alignment of pre-and postsynaptic partners and electron dense postsynaptic membranes (Fig. 6D-F). However, the PC soma of the GPHR F/F ;GluD2-Cre mice was predominantly enwrapped with electron-lucent processes of Bergmann glia (BG) instead of synaptic contacts with BCs, suggesting that BG processes filled the space caused by BC degeneration (Fig. 6G-I).
To further examine whether the elimination of BC contacts at PCs is concomitantly accompanied by a decrease in synaptic transmission, we recorded mIPSCs from PCs as a readout of the neurotransmission from BCs. mIPSCs in PCs have been previously classified into two types: small mIPSCs from SCs and large mIPSCs from BCs (Nakayama et al., 2012). The occurrence of large amplitude mIPSCs was dramatically reduced in GPHR F/F ; GluD2-Cre mice (n ϭ 4) when compared with that of control mice (n ϭ 4; Fig. 7A,B). Accordingly, the mean frequency of large mIPSCs in GPHR F/F ;GluD2-Cre mice was also significantly lower than that of control mice ( Fig.  7C; Table 1; control: 0.44 Ϯ 0.08 Hz, n ϭ 4; GPHR F/F ; GluD2-Cre: 0.17 Ϯ 0.07 Hz, n ϭ 4; unpaired Student's t test, t (6) ϭ 2.518, ‫ء‬p ϭ 0.0454). Furthermore, the mean amplitude of whole mIPSCs was reduced slightly in GPHR F/F ;GluD2-Cre mice ( Fig. 7D; Table 1; control: 99.9 Ϯ 3.6 pA, n ϭ 4; GPHR F/F ;GluD2-Cre: 73.2 Ϯ 7.8 pA, n ϭ 4; unpaired Student's t test, t (6) ϭ 3.105, ‫ء‬p ϭ 0.021). In contrast, the mean frequencies of small and large amplitude mIPSCs were not different, indicating that the specific reduction of large amplitude events was due to the disruption of BC axons from the PC soma. Taken together, these results indicated that BC synapses on PCs were dramatically reduced, due largely to the axonal degeneration of BCs caused by the dysfunction of Golgi apparatus in GPHR F/F ;GluD2-Cre mice.

Formation of excitatory synapses on PC soma in GPHR F/F ;GluD2-Cre mice
Reduced GABAergic transmission from BCs impairs elimination of CF synapses, which leads to abnormal innervation of CF synapses onto PCs (Nakayama et al., 2012). Our morphological and electrophysiological analy-continued stained with DAPI (blue). Cerebellar lobules are indicated by Roman numerals. Scale bar: 300 m. E, The density of PCs was measured following immunofluorescence staining for calbindin GPHR F/ϩ ;GluD2-Cre and GPHR F/F ;GluD2-Cre mice at P45, P60, and P90. Five sections from each mouse corresponding to the cerebellar region as shown in D were measured. Data are presented as the mean Ϯ SEM; n ϭ 15 sections for each group; ‫‪p‬ءء‬ Ͻ 0.01 (one-way ANOVA with Sidak's post hoc test). F, Double immunofluorescence for IBA1 (green) and MAC2 (red) in the cerebellum of GPHR F/ϩ ;GluD2-Cre and GPHR F/F ;GluD2-Cre mice at P60. Nuclei are stained with DAPI (blue). Scale bar: 50 m. For A, C, D, F, three mice per genotype; n ϭ 5 sections per mouse; representative images are shown. ses confirmed the disruption of inhibitory inputs from BCs. Thus, we hypothesized that CF innervation was altered in GPHR F/F ;GluD2-Cre mice. To test this hypothesis, immunostaining for calbindin and vGlut2, a marker for CF axon terminals, was performed. In normal mice, vGlut2-positive CF terminals were distributed in the molecular layer and associated with PC dendrites after elimination (Miyazaki et al., 2003). In both GPHR F/F ;L7-Cre and control mice, vGlut2-positive CF terminals exclusively innervated spino-dendritic domains in the molecular layer. However, in GPHR F/F ;GluD2-Cre mice, CF terminals were frequently associated with PCs soma in addition to dendrites (Fig. 8A, arrowheads). Quantification of the vGlut2 puncta on the PC somata revealed a significant increase in GPHR F/F ;GluD2-Cre mice when compared with that of control and GPHR F/F ;L7-Cre mice (Table 1; control: Next, to investigate whether somatic spines contacting with CF terminals occur on the surface of PCs in GPHR F/F ;GluD2-Cre mice, we performed triple immunofluorescence staining for metabotropic glutamate receptor 1␣ (mGluR1␣), vGlut2 and calbindin. At the CF-PC synapses, vGlut2 and mGluR1␣ are markers for presyn-aptic and postsynaptic proteins, respectively. As hypothesized, mGluR1␣-positive punctae overlapped with vGlut2-positive punctae in perisomatic regions of PCs from GPHR F/F ;GluD2-Cre mice, indicating CF innervation to somatic spines (Fig. 8B). Quantification confirmed that mGluR1 and vGlut2 double-positive puncta on the PC somata were increased significantly in GPHR F/F ;GluD2-Cre mice when compared with that of control mice (Table 1; GPHR F/ϩ ;GluD2-Cre at P60: 9.34 Ϯ 1.27%; GPHR F/F ; GluD2-Cre at P60: 45.73 Ϯ 4.016%; three mice per group; five sections per mouse; n ϭ 15 sections for each group; GPHR F/ϩ ;GluD2-Cre at P60: 172 PCs; GPHR F/F ;GluD2-Cre at P60: 141 PCs; unpaired Student's t test, t (28) ϭ 8.648, ‫ء‬p Ͻ 0.0001). In contrast, there was no apparent change in the distribution of vGlut1-positive parallel fiber terminals (Fig. 8C).

Discussion
In this study, the crucial role of luminal acidification of the Golgi apparatus in maintaining normal structural integrity of the Golgi apparatus in neurons is reported. Although the relationship between Golgi pathology and neurodegeneration has been suggested based on the genetic analysis of Golgi-related proteins, no studies have thoroughly analyzed abnormalities of cerebellar neuronal circuits on perturbations in luminal pH of the Golgi apparatus in vivo (Alvarez-Miranda et al., 2015). Since available pharmacological blockers lack specificity for manipulation of the luminal pH of the Golgi apparatus, we believe that genetically modified mouse models are currently an ideal tool to study the relationship between luminal pH and morphology of the Golgi apparatus.
We demonstrated previously that the luminal pH of the Golgi apparatus is altered in GPHR mutant Chinese hamster ovary cells (Maeda et al., 2008). In the present study, using a pH-sensitive GFP variant, pHluorin-Golgi, we revealed that the luminal pH of the Golgi apparatus is elevated in primary cultured neurons that no longer express normal levels of GPHR. Furthermore, it has been reported that the pH gradient between the ER and the  Golgi apparatus disappears by RNA silencing of GPHR in HeLa cells (Vavassori et al., 2013). Taken together, these results support the concept that GPHR is both necessary and sufficient for the maintenance of the Golgi luminal acidic pH and no other mechanism can compensate for pH gradient maintenance in neurons.
Our results revealed that loss of GPHR markedly increased the degeneration of PCs. This, in turn, leads to the reduction of cerebellar size and might contribute to the abnormal motor phenotype observed in GPHR F/F ;GluD2-Cre mice. Interestingly, the loss of PCs stands out in the lobule IX. However, this observation is in sharp contrast with other studies that reported a patterned or area selective PC degeneration (Sarna and Hawkes, 2003;Cerminara et al., 2015). In addition, it has also been reported that lobules IX and X exhibit some level of neuroprotection Hawkes, 2003, 2011). In future, it would be interesting to analyze the factors that confer lobule-dependent neuroprotection in different mouse models.
The observation that PCs begin to degenerate at P60 in GPHR F/F ;GluD2-Cre mice demonstrates that the luminal acidic pH of the Golgi apparatus under physiological conditions is essential for PC survival. Impaired luminal acidification of the Golgi apparatus due to deletion of GPHR causes structural disorganization of the Golgi apparatus and results in neurodegeneration. Notably, PCs from GPHR F/F ;GluD2-Cre mice still exhibit normal organization of the Golgi apparatus with GPHR immunoreactivity until at least P30, indicating that maturation in the early developmental stage of the cerebellum occurs normally in these PCs. In other words, the abolishment of GPHR expression was completed only after stabilization of cer-ebellar circuits. The motor phenotypes because of PC loss in GPHR F/F ;GluD2-Cre mice were comparable with other mouse models where Golgi-associated proteins such as Rer1 and GM130 are ablated (Liu et al., 2017;Valkova et al., 2017). Loss of GM130 starts from early developmental stages and results in atrophic dendrites and eventual disruption of PCs. This indicates that the proper function of the Golgi apparatus contributes to PC maturation (Liu et al., 2017). However, our current study also suggests that luminal acidification of the Golgi apparatus is also responsible for PC survival after maturation.
Our previous in vitro study showed that GPHR plays an important role in maintaining the morphological integrity of the Golgi apparatus (Maeda et al., 2008). In addition, morphological abnormalities in the Golgi apparatus have also been reported in fibroblasts derived from patients with mutations in the gene encoding ATPV0A2, which is involved in luminal acidification of the Golgi apparatus (Hucthagowder et al., 2009;Foulquier et al., 2012). Consistent with these studies, our morphological analyses Holding potential (Vh) ϭ Ϫ70 mV. C, The bar graphs show the average frequency of mIPSC for all, small (Ͻ100 pA), and large (Ͼ100 pA) events in GPHR F/ϩ ;GluD2-Cre mice (n ϭ 4) and GPHR F/F ;GluD2-Cre mice (n ϭ 4). Data are presented as the mean Ϯ SEM; ‫ء‬p Ͻ 0.05 (unpaired Student's t test). D, The bar graphs show the average amplitude of mIPSC for all, small (Ͻ100 pA), and large (Ͼ100 pA) events in GPHR F/ϩ ;GluD2-Cre mice (n ϭ 4) and GPHR F/F ;GluD2-Cre mice (n ϭ 4). Data are presented as the mean Ϯ SEM; ‫ء‬p Ͻ 0.05 (unpaired Student's t test).
show structural abnormalities of the Golgi apparatus in GPHR-deficient PCs. Interestingly, vesiculation and fragmentation of the Golgi apparatus are pathological hallmarks of various neurodegenerative diseases including amyotrophic lateral sclerosis, Parkinson's, Alzheimer's, and Huntington's diseases (Fan et al., 2008;Haase and Rabouille, 2015;Ayala and Colanzi, 2017). In addition, various models for human neurodegenerative diseases imply that fragmentation of the Golgi apparatus in neurons occurs as a result of the dysregulation of ER-Golgi transport (Huynh et al., 2003;Joshi et al., 2014). Together, our findings suggest that GPHR plays important roles, including morphological maintenance of the Golgi apparatus in neurons. However, genetic screening has not identified GPHR as a causative gene for human diseases. Nonetheless, our present study suggests a causative link between fragmentation of the Golgi apparatus and neurodegeneration. We cannot exclude the possibility that an abnormality of the ER secondarily leads to the abnormal morphology of the Golgi apparatus due to GPHR deficiency. Further studies will be required to elucidate whether transport between ER and the Golgi apparatus is also impaired in GPHR-deficient neurons.
Loss of GM130 and Rer1 proteins in premature PCs affects dendrite and axon morphology, respectively, indicating that the proper function of the Golgi apparatus is important for intact PC polarization (Liu et al., 2017;Valkova et al., 2017). In our study, there was apparently no change in the morphology of PC dendrites in GPHR F/F ;GluD2-Cre mice, whereas axonal swelling in GPHR-deficient PCs was observed at P45, and the number and size of axon swellings increased with age. Our observation is in accordance with those from previous studies in mouse models where PCs lack Golgiassociated proteins Valkova et al., 2017). However, it is unlikely that the axonal swellings in GPHR-deficient PCs are related to myelin abnormality as observed in PCs deficient in ceramide glucosyltransferase, a Golgi-associated enzyme , because myelination of GPHR-deficient PC axons appears normal. Furthermore, we also noticed that a larger portion of the soma of DCN neurons of GPHR F/F ;GluD2-Cre mice were surrounded with intercellular spaces because of the loss of axosomatic synapses as early as P45. Importantly, the synaptic terminals of BCs in GPHR F/F ;GluD2-Cre mice were also lost at P60, which strengthens our notion that impairment of luminal acidification of the Golgi apparatus is closely associated with axonal and synaptic pathology. Axonal swelling is a hall- Figure 8. CF terminals innervate PC somata in GPHR F/F ;GluD2-Cre mice. A, Double immunofluorescence for calbindin (Calb; green) and vGlut2 (red) in the molecular layer at P60. vGlut2-positive CF terminals were observed on PC somata in addition to the dendrites of GPHR F/F ;GluD2-Cre mice. Arrowheads indicate vGlut2-positive CF terminals on PC soma. Scale bars: 50 m. Boxed regions are enlarged and shown in the right panels. B, Triple immunofluorescence for mGluR1␣ (green), vGlut2 (red), and calbindin (Calb; cyan) in PCs at P60. New Figure 8. continued spines are formed on the surfaces of PC somata in GPHR F/F ;GluD2-Cre mice. Asterisks indicate PC somata. Scale bars: 10 m. C, Distribution of parallel fiber terminals in GPHR F/F ;GluD2-Cre mice. Triple immunofluorescence for mGluR1␣ (green), vGlut1 (red), and calbindin (Calb; cyan) in cerebellar sections of GPHR F/ϩ ;GluD2-Cre mice and GPHR F/F ;GluD2-Cre mice at P60. Asterisks indicate PC somata. Scale bar: 20 m. For AϪC, three mice per genotype were used; n ϭ 5 sections per mouse; representative images are shown. mark of axonal dystrophy, which can lead to neurodegeneration (Coleman, 2005). Major causes for axon degeneration converge in the dysfunctions of axonal transport, the autophagy-lysosome system, mitochondria, and increased intra-axonal calcium (Coleman, 2005;Komatsu et al., 2007;Nishiyama et al., 2007;Liang et al., 2010;Koike et al., 2017;Yamaguchi et al., 2017). Al-though the direct link between impairment of luminal acidification of the Golgi apparatus and axon and synaptic abnormalities remains unclear, it is possible that loss of GPHR is associated with axonal transport dysfunction.
Intriguingly, CF innervation on PC soma has been observed in PC-specific deletion of GluD2 and RORa because of territorial changes in parallel fibers and CFs Figure 9. CF innerved to PC soma in GPHR F/F ;GluD2-Cre mice. Electron micrographs of PC soma in GPHR F/ϩ ;GluD2-Cre (A) and GPHR F/F ;GluD2-Cre mice (BϪF) at P60. Arrowheads indicate active zones. Arrows indicate dense stacks of the cisternae of the smooth endoplasmic reticulum. Asterisks indicate necks of the spine from PC soma. Scale bars: 500 nm. AϪF, Three mice per genotype were used; n ϭ 5 sections per mouse; representative images are shown. (Miyazaki et al., 2010;Chen et al., 2013). Furthermore, reduced GABAergic transmission from BCs leads to abnormal elimination of CF synapses onto PCs during early developmental stages (Nakayama et al., 2012). However, the role of GABAergic transmission in maintaining proper CF innervation in the adult cerebellum has not been examined. In our mouse model, CF innervates the PC soma instead of the BC axonal terminal territory after stabilization of the cerebellar circuits. This novel finding suggests that, in addition to heterosynaptic competition of excitatory innervations, the inhibitory-excitatory imbalance can also result in a redistribution of the innervation territory on mature PCs. Further studies are required to understand the roles of BCs in maintenance of proper CF innervation onto PCs by use of BC-specific Cre mouse lines.
What could be the pathological consequences of altered cerebellar neuronal circuitry in GPHR F/F ;GluD2-Cre mice? Our morphological data showed that the formation of somatic spines in GPHR-deficient PCs and the targeting of CFs to these somatic spines as well as the electrophysiological data suggesting disrupted and nonfunctional inhibitory BCs hint at a possible excitatory-inhibitory imbalance in these PCs. Unlike the normal situation, where CF inputs are created only in dendrites, the direct excitation of the soma by CF implies that voltage changes are minimal and thus result in a dramatic increase in the PC membrane potential. Moreover, the severity of depolarization around the PC soma is maintained because inhibition from BC terminals no longer exists. As calcium permeable channels open because of depolarized membrane potentials, overexcitation may create an unnecessarily large calcium load in the PC cytoplasm, leading to calcium excitotoxicity. This may lead to the eventual disruption and death of PCs. Indeed, PCs from GPHR F/F ;GluD2-Cre mice possess dense stacks of smooth ER, which prevent excessive Ca 2ϩ release from the reservoir (Banno and Kohno, 1996) and respond to calcium excitotoxicity because of the excitatory-inhibitory imbalance. Nevertheless, GPHR-deficient PCs are largely lost with age, suggesting that neurodegeneration is likely caused by impairment of luminal acidification of the Golgi apparatus. Further studies must be conducted to elucidate the function of the Golgi apparatus for maintaining the integrity of neuronal circuits.