Paroxysmal dystonia results from the loss of RIM4 in Purkinje cells

Abstract

Full-length RIM1 and 2 are key components of the presynaptic active zone that ubiquitously control excitatory and inhibitory neurotransmitter release. Here, we report that the function of the small RIM isoform RIM4, consisting of a single C2 domain, is strikingly different from that of the long isoforms.

RIM4 is dispensable for neurotransmitter release but plays a postsynaptic, cell type-specific role in cerebellar Purkinje cells that is essential for normal motor function. In the absence of RIM4, Purkinje cell intrinsic firing is reduced and caffeine-sensitive, and dendritic integration of climbing fibre input is disturbed. Mice lacking RIM4, but not mice lacking RIM1/2, selectively in Purkinje cells exhibit a severe, hours-long paroxysmal dystonia. These episodes can also be induced by caffeine, ethanol or stress and closely resemble the deficits seen with mutations of the PNKD (paroxysmal non-kinesigenic dystonia) gene.

Our data reveal essential postsynaptic functions of RIM proteins and show non-overlapping specialized functions of a small isoform despite high homology to a single domain in the full-length proteins.

Introduction

The RIM family of proteins is known for its ubiquitous and important function in synaptic transmission and synaptic plasticity.¹ Not only glutamatergic and GABAergic synapses depend on RIM proteins, but also varicosities that release neuromodulators, such as dopamine.² Knockout (KO) mice of RIM1α, the full-length variant of the RIM1 gene, show deficits in spatial memory and social interaction as well as a decrease in prepulse inhibition.^3,4 Mice deficient for both full-length variants, RIM1α and RIM2α, are not viable.⁵ Within the synapse, the long RIM family members RIM1 and RIM2 maintain a high density of presynaptic calcium channels and keep synaptic vesicles in a release-ready state.^5-10 RIM1α and RIM2α comprise several well characterized protein domains through which they interact with most other active zone (AZ) proteins, the synaptic vesicle protein Rab3 and voltage-gated calcium channels (VGCCs).¹¹ The C-terminal C2 domain of RIM1 and RIM2, the so-called C2B domain, allows membrane anchoring of RIM1/2 and is likely to be particularly important for the spatial coordination of AZ proteins, VGCCs and synaptic vesicles.^12,13

The RIM protein family also includes two shorter members, RIM3 and RIM4,¹¹ which are widely expressed throughout the CNS.¹⁴ These short RIM proteins comprise only a single C2-domain, which is highly homologous to the C2B domain of RIM1/2.¹¹ As such, single domain proteins RIM3 and RIM4 are unlikely to have similar AZ-organizing functions as their long RIM family counterparts. We have previously shown, using a knockdown strategy, that RIM3 and RIM4 are important for dendritic outgrowth.¹⁴ It has also been described that RIM3 and RIM4,¹⁵ as well as the C2B domains of RIM1/2,¹⁶ can interact with β-subunits of VGCCs to suppress calcium-induced inactivation. However, it is not clear whether suppression of inactivation by RIM3/4 alters action potential-mediated vesicle release from classical synapses, where calcium-induced inactivation may only play a minor role. Furthermore, such suppression may be dominated by RIM1/2, which are abundant in presynapses and have a much higher affinity for β-subunits.¹⁵ Taken together, the exact role of RIM3 and RIM4 in CNS function remains largely unresolved.

The results presented here show that RIM4 plays a unique role in the CNS that is very different from the well studied function of RIM1/2. RIM4 acts in a strictly pathway-specific manner and is primarily required by Purkinje cells in the cerebellum. RIM4 does not play a role in transmitter release but acts postsynaptically to regulate somatodendritic excitability of Purkinje cells. During development, RIM4 is required for dendrite outgrowth of Purkinje cells and for the correct localization of synapses along their dendrites. Mice lacking RIM4 develop a severe motor phenotype in the form of a paroxysmal dystonia, not known to result from deletion of the long RIM isoforms. This dystonia shares many features with Ca_v2.1 channelopathies, such as tottering mice, but most closely resembles the paroxysmal non-kinesigenic dystonia (PNKD) caused by mutations in the PNKD protein that abolish binding to the C2B domain of RIM1/2.¹⁷

Materials and methods

A full description of the experimental procedures is provided in the Supplementary material.

Experimental model and subject details

Generation of mouse models and source of cell cultures used in this study are reported in the Supplementary material.

Immunoblotting

Immunoblotting of lysates from brain tissue and HEK293T cells was performed as previously described.^14,18,19 Information about antibodies is reported in the Supplementary material.

Plasmids

The AAV-CMV-mRFP and AAV-CMV-mRFP-T2A-Flag-RIM4 plasmids were generated by restriction cloning. Details about restriction enzymes and primers are provided in the Supplementary material.

Recombinant AAV production

Recombinant AAV1/2 particles were generated by triple transfection of HEK293T cells and purified as described previously.^19-21

Primary neuronal cultures

Primary cortical neurons were prepared from E16–E19 C57Bl/6N mouse brains, as previously described.^19-21

Data analysis

Data are presented as means ± standard errors of the mean (SEM), unless otherwise noted. Statistical significance was determined as indicated in the figure legends. Values were considered significant at P < 0.05. Statistical analyses were performed using Igor Pro 8 or Prism GraphPad (version 6.02).

Results

RIM4 deficiency causes severe paroxysmal dyskinesia and dystonia

To investigate the functional role of RIM4 in the CNS, we generated constitutive RIM4 knockout (KO_const) mice by crossing conditional RIM4^fl/fl mice with the ubiquitously expressed PGK Cre driver line (Fig. 1A). Ablation of RIM4 protein expression was verified in the hippocampus, cortex and cerebellum of RIM4 KO_const mice by quantitative immunoblotting using an isoform-specific antibody raised against the RIM4 N-terminus (Fig. 1B and C). RIM4 KO_const mice were born at the expected Mendelian ratio (data not shown) and showed no obvious abnormalities during the first two postnatal weeks. However, starting around postnatal Day 21 (P21), RIM4 KO_const mice gained less weight than wild-type (WT) littermates (Fig. 1D). Around this time, RIM4 KO_const mice began to develop hours-long spontaneous and stress-induced (e.g. handling) dyskinetic and dystonic episodes. RIM4 KO_const mice and their wild-type littermates were continuously EEG (cortex) and video monitored (24/7) in their home cage for five consecutive days. We roughly divided dyskinetic-dystonic episodes into three phases (I–III, Fig. 1E), which correspond well to the severity scales 2–4 (mild to severe) proposed for grading dystonia symptoms.²² Phase I (Fig. 1E and F) lasted up to 10 min and consisted of seconds-long repetitive jerking of the forelimbs and dystonic upward extensions of one of the hindlimbs. Ambulation was largely maintained and mice only occasionally fell to the side. Phase II (∼10–15 min) was characterized by additional repetitive and sustained neck flexion together with extension of the trunk. Abnormal postures and falls were more frequent and a typical movement was to push the head into the bedding while the hindlimbs performed a pedalling motion. Phase III lasted ∼3 h and was associated with severe immobility and persistent abnormal postures with almost no ambulation. RIM4 KO_const mice developed dystonic episodes mostly once per day (1 ± 0.98 per day; Fig. 1G). The most frequent onset of motor episodes was in the early morning hours, suggesting a circadian-like pattern (Fig. 1G). In the EEG recordings we did not detect spike-and-wave discharges associated with periods of immobility during the episodes.^29-31 Both ethanol and caffeine also induced severe dyskinetic-dystonic episodes in RIM4 KO_const mice, phenocopying the spontaneously occurring episodes (90% and 100% of RIM4 KO_const mice, respectively; Fig. 1H).

Automated gait analysis using a video-based CatWalk system revealed that even outside of dyskinetic-dystonic episodes, RIM4 KO_const mice displayed deficiencies in motor coordination compared to the wild-type mice, such as gait abnormalities, as evidenced by an irregular footprint pattern (Fig. 1I and J). RIM4 KO_const mice also performed worse in the accelerating rotarod test and showed a significantly reduced latency to fall compared to wild-type littermates (Fig. 1K). RIM4 KO_const mice also performed worse in the ledge test (Supplementary Fig. 1A) and the hind limb clasping assay (Supplementary Fig. 1B). RIM4 wild-type and KO_const mice were given the opportunity to use a regular or complex running wheel in their home cage and were then observed for 24 h/day for four consecutive days (2 days regular wheel and 2 days complex wheel, Fig. 1L). Consistent with only minor deficits in regular walking, no difference in mean running speed was observed between RIM4 wild-type and KO_const mice when running on wheels equipped with rungs in a regular pattern. However, RIM4 KO_const mice performed significantly worse when the pattern of the rungs in the wheel was made more complex by removing individual rungs (16 of 38) (Fig. 1L and M). While RIM4 wild-type mice showed a small increase in their instantaneous average running speed even in the complex wheel, RIM4 KO_const mice became significantly slower. RIM4 KO_const mice had fewer running sessions per time, not taking the duration of episodes into account [wild-type, 36.1 ± 3.7 in running wheel (RW), 37.4 ± 2.8 in complex wheel (CW); KO_const 15.2 ± 3.2 in RW, 19.1 ± 2.2 in CW] than their wild-type littermates. Nevertheless, RIM KO_const mice occasionally attempted to use the wheel even when they were in a dystonic episode (black strokes on red bars in Fig. 1L). Taken together, the absence of RIM4 causes deficits in motor coordination and a circadian, long-lasting episodic motor impairment.

Synaptic transmission is not affected by RIM4 deficiency

To test whether ablation of RIM4 affects synaptic transmission, we performed whole-cell voltage clamp recordings from hippocampal CA1 pyramidal neurons in RIM4 wild-type and KO_const mice. Both the mean amplitude and frequency of miniature excitatory postsynaptic currents (mEPSCs) were not altered in RIM4 KO_const mice (Fig. 2A–C, 1 µM tetrodotoxin, 20 µM CNQX, 100 µM D-APV). Paired-pulse facilitation of field excitatory postsynaptic potentials (fEPSPs) recorded in the CA1 stratum radiatum and evoked by Schaffer collateral stimulation was unaffected in RIM4 KO_const mice (Fig. 2D). Furthermore, quantitative immunoblotting showed no significant changes in the expression of other components of the presynaptic release machinery, such as RIM1 and Liprin-α, in hippocampal lysates from RIM4 KO_const mice compared to wild-type (Fig. 2E). These findings were further supported by quantitative transcriptomic (total RNA sequencing, Supplementary Fig. 2A) and proteomic (Supplementary Fig. 2B) analyses, which did not reveal any significant changes in the levels of synaptic proteins in RIM4 KO_const mice. Levels of the other short RIM variant, RIM3, appeared increased in hippocampus (Fig. 2E), but not in cerebellum and cortex (Supplementary Fig. 2B, proteomics data). In conclusion, RIM4 appears to be dispensable for synaptic transmitter release in CA1 of the hippocampus.

RIM4 deletion specifically causes morphological changes in the cerebellum

The cerebellum constitutes a critical circuit for motor coordination and control and RIM4 is robustly expressed in the cerebellum (Fig. 1B and Supplementary Fig. 3). We analysed the size of the cerebellum in Nissl-stained sagittal sections of RIM4 wild-type and KO_const mice (Fig. 3A) and found a significant reduction in the cerebellar area in mice lacking RIM4 (Fig. 3B). In contrast, the size of the cerebral cortex and hippocampus was unchanged in RIM4 KO_const mice (Fig. 3B), despite robust RIM4 expression in those regions (Supplementary Fig. 3). Quantification of the size of the individual cerebellar layers revealed a significantly reduced area of the molecular layer (ML) in RIM4 KO_const mice, whereas the size of the granule cell layer and the white matter did not differ between wild-type and KO_const (Fig. 3C and D). In contrast to the reduced thickness of the molecular layer in RIM4 KO_const mice (Fig. 3E), the total length of the Purkinje cell layer (Fig. 3F) and the density of Purkinje cells along this length (Fig. 3G) were unchanged, as was the estimated number of Purkinje cells (Fig. 3H). We labelled the dendritic tree of individual Purkinje cells by single cell electroporation of TMR-Dextran and performed a 3D reconstruction of Purkinje cells in confocal image stacks (Fig. 3I). The area covered by the dendritic tree of Purkinje cells was significantly reduced in RIM4 KO_const mice compared to wild-type mice (Fig. 3J), and a 3D-Sholl analysis revealed a significantly reduced dendritic arborization starting at a distance of ∼50 µm from the soma (Fig. 3K). A detailed analysis showed that, in particular, fourth to sixth order dendritic segments were significantly shorter in Purkinje cells of RIM4 KO_const mice (Fig. 3L).

Spontaneous firing of Purkinje cells is important for motor performance^23,24 and is maintained in Purkinje cells in acute slices. We pharmacologically blocked both excitatory and inhibitory inputs and performed juxtacellular recordings of spontaneous activity of Purkinje cells (Fig. 3M). The mean firing frequency was strongly reduced in RIM4 KO_const compared to wild-type mice (Fig. 3N). Furthermore, application of caffeine, which induces episodes of dystonia in vivo (Fig. 1G), inhibited spontaneous firing of Purkinje cells in RIM4 KO_const mice but not in wild-type mice (Fig. 3O).

Altered synaptic integration by Purkinje cells

The firing activity of Purkinje cells in vivo is strongly modulated by synaptic input.²⁵ Climbing fibre (CF) synaptic input from neurons in the inferior olive (IO) generates pronounced dendritic calcium signals,²⁶ which have previously been studied in great detail in vivo.^27-29 We performed GCaMP6f-based in vivo calcium imaging of Purkinje cells in the lobules V or VI of the cerebellar vermis in anaesthetized and head-fixed RIM4 KO_const and wild-type mice to examine synaptic activation of Purkinje cells (Fig. 4A). Wild-type mice displayed spontaneous calcium signals across the lateral extent of dendrites of Purkinje cells in vivo similar in amplitude, kinetics and frequency, as previously reported^30,31 (Fig. 4B and Supplementary Fig. 4A and B). While calcium signals occurred with similar kinetics in RIM4 KO_const mice (rise time 20–80%, WT 55.9 ± 1.4 ms, KO_const 54.7 ± 3.5 ms; decay time, WT 288.5 ± 17.2 ms, KO_const 313 ± 23.1 ms, Supplementary Fig. 4A and B), their amplitudes were substantially reduced (Fig. 4C and D). Calcium signal amplitude distributions broadly followed Poisson distributions (dashed lines) with smaller values and reduced amplitude variability in RIM4 KO_const mice (Fig. 4D). Calcium signals occurred at a similar mean rate in wild-type and RIM4 KO_const mice (Fig. 4E), but their temporal pattern strikingly differed. The intervals between two consecutive calcium signals in Purkinje cells of wild-type mice showed three characteristic values (peaks in the histogram, bullet points in Fig. 4F) at ∼0.3, 2 and 5 s that clearly deviated from a random distribution. These characteristics were not observed in RIM4 KO_const mice, and the distribution of intervals resembled that of a random process (dashed line, Fig. 4F), suggesting that the rhythms of synaptic signals generated in the olivo-cerebellar loop and arriving at Purkinje cells are severely diminished. Next, we compared the temporal correlation of calcium signal activity between neighbouring Purkinje cells by calculating the Pearson's correlation coefficient of calcium signal event times. As shown in Fig. 4G, calcium activity was highly correlated for a group of Purkinje cells within a neighbourhood of approximately ±20 µm in wild-type mice. The degree of synchronicity of neighbouring Purkinje cells in RIM4 KO_const mice was greatly reduced, and the region of increased synchronicity was narrowed to approximately ±10 µm, indicating a pronounced alteration in the function of Purkinje cell ensembles (Fig. 4H).

The reduced amplitude of dendritic calcium signals in Purkinje cells in vivo prompted us to compare CF signalling in acute slices (confocal imaging). To this end, we examined calcium entry in patch-clamped Purkinje cells in current-clamp mode using the calcium indicator OGB-1 (200 µM) and confocal microscopy (Fig. 4I and J). CF stimulation induced clear complex spikes in Purkinje cells from wild-type and KO mice (Fig. 4K and Supplementary Fig. 4C and D). Unexpectedly, and in contrast to the in vivo recordings, CF stimulation elicited calcium transients in both groups that were similar in amplitude along the dendrites (Fig. 4K and L) as well as in kinetics (Supplementary Fig. 4E and F). However, we found that Purkinje cells in RIM4 KO_const integrated a series of CF stimuli at the most frequently observed interval of ∼0.3 s slightly less effectively than Purkinje cells in wild-type mice (Fig. 4M and N). We further compared the synaptic conductance of CF EPSCs by recording Purkinje cells in voltage-clamp mode during stimulation of a CF (Fig. 4O). CF EPSCs were significantly increased (∼1.6-fold) in Purkinje cells from RIM4 KO_const mice, whereas paired pulse depression (Fig. 4O) and the kinetics of synaptic currents (Supplementary Fig. 4G and H) were not changed.

Ablation of RIM4 in Purkinje cells recapitulates cerebellar pathology and motor dysfunction of the constitutive KO

To examine if ablation of RIM4 only in Purkinje cells is sufficient to recapitulate the spontaneous episodes of dystonia observed in RIM4 KO_const mice, we crossed conditional RIM4^fl/fl mice with mice expressing Cre recombinase under the Purkinje cell-protein 2 (PCP2) promoter,³² resulting in a Purkinje cell-specific RIM4 knock-out (KO_PCP2). We verified the specificity of RIM4 ablation in KO_PCP2 mice by performing fluorescent in situ hybridization (RNAscope) experiments, which revealed the selective reduction of RIM4 mRNA in cerebellar Purkinje cells without changes in other principal neurons in the cortex and hippocampus, as well as other cell types such as granule cells (GCs) and molecular layer interneurons (MLIs) in the cerebellar cortex (Fig. 5A and B). As observed in the RIM4 KO_const line, RIM4 KO_PCP2 mice exhibited a similarly reduced body weight compared to their wild-type littermates (Fig. 5C). Video monitoring revealed that RIM4 KO_PCP2 mice also exhibited spontaneous dyskinetic-dystonic episodes (Fig. 5D), which were indistinguishable from that of RIM4 KO_const mice. Furthermore, as observed in RIM4 KO_const mice, caffeine administration induced dyskinetic-dystonic episodes in all RIM4 KO_PCP2 mice tested, whereas ethanol was slightly less effective, inducing episodes in only 3/6 mice compared to 9/10 in RIM4 KO_const mice (Fig. 5E). Finally, RIM4 KO_PCP2 mice also reproduced the poor baseline (non-episode time) performance of RIM4 KO_const mice on the accelerating rotarod (Fig. 5F) and the complex running wheel (Fig. 5G and H).

Selective absence of RIM4 from Purkinje cells was sufficient to reproduce the morphological changes: we observed a similar reduction in cerebellar size in RIM4 KO_PCP2 mice (WT 1 ± 0.02, KO_PCP2 0.78 ± 0.037, area normalized to WT mean) (Fig. 5I andSupplementary Fig. 5A–F), while the cortex and hippocampus were unaffected (Fig. 5I). No obvious loss of Purkinje cells was observed (Supplementary Fig. 5G). Selective RIM4 ablation in Purkinje cells altered the distribution of climbing fibre synapses along their dendrites: immunolabelling against the vesicular glutamate transporter 2 (VGlut2) and calbindin revealed a pronouncedly more proximal location of climbing fibre synapses compare to wild-type mice (Fig. 5J and K).

The electrophysiological properties of Purkinje cells in RIM4 KO_PCP2 mice also phenocopied those identified in RIM4 KO_const Purkinje cells. Spontaneous activity of Purkinje cells in slices in the presence of both excitatory and inhibitory blockers was significantly reduced in juxtacellular recordings and showed a significantly increased sensitivity to bath application of 1 mM caffeine (Fig. 5L). Current clamp recordings from patch-clamped Purkinje cells in RIM4 KO_PCP2 mice revealed broadened action potentials and an increased and delayed fast afterhyperpolarization when compared to RIM4 wild-type mice (Fig. 5M and N), whereas action potential amplitude, resting membrane potential and input resistance were unchanged.

To address if neurotransmitter release from synapses of Purkinje cells is affected by RIM4 ablation, we evaluated GABA release at the Purkinje cell-deep cerebellar nuclei (DCN) synapse (Fig. 5O). We performed voltage clamp recordings from neurons in the DCN. Inhibitory postsynaptic currents were isolated by 20 µM CNQX and 100 µM D-APV and we electrically stimulated Purkinje cell axons in nearby white matter (200–300 µm distance from the recorded soma). Paired-pulse ratios (PPRs) were very similar between wild-type and RIM4 KO_PCP2 mice (Fig. 5O, right), suggesting that deletion of RIM4 from Purkinje cells does not affect transmitter release at their presynapses. PPR of parallel fibre (PF)-EPSCs in Purkinje cells of RIM4 KO_PCP2 mice, in which RIM4 was absent postsynaptically, was also not altered (Fig. 5P).

We repeated in vivo calcium imaging experiments in wild-type and RIM4 KO_PCP2 mice under the same conditions but injected conditional GCaMP6f virus (AAV2/1-pAAV.CAG.Flex.GCaMP6f.WPRE.SV40) to label Purkinje cells specifically (Supplementary Fig. 5H). Spontaneous calcium signals in dendrites of Purkinje cells occurred at a similar frequency as in the constitutive RIM4 WT/KO groups. As observed in the constitutive RIM4 KO line, the amplitudes of calcium signals in Purkinje cells of RIM4 KO_PCP2 mice were substantially smaller (WT 0.42 ± 0.06 ΔF/F, KO_PCP2 0.12 ± 0.02 ΔF/F) and the spatial synchrony of calcium signals across neighbouring Purkinje cells in a micro-zone was reduced when compared to wild-type mice (Supplementary Fig. 5I–K). These experiments show that the loss of RIM4 from Purkinje cells alone is sufficient to trigger alterations of synaptic integration in cerebellar circuits.

At the ultrastructural level of Purkinje cells in RIM4 KO_PCP2 mice, their somata, dendrites and synapses appeared overall normal with no signs of pathology or degeneration (Fig. 6). However, we found PF synapses in RIM4 KO_PCP2 mice to occasionally merge to large multi-synapse varicosities contacting multiple spines of Purkinje cells (Fig. 6A and B). Large multi-synapse varicosities with three and more release sites were almost exclusively observed in RIM4 KO_PCP2 mice (Fig. 6C). While the size (area) of single synapse boutons in both wild-type and RIM4 KO_PCP2 mice was smaller than 0.5 µm², almost all large multi-synapse boutons showed cross-sections of >0.5 µm² (Supplementary Fig. 6). Similar structures have previously been observed in the dystonic tottering mouse line,³³ suggesting commonalities in the mechanisms of cerebellar pathology. However, large multi-synapse varicosities (at least three synapses) were observed in only 31 of 922 counted PF synapses (Supplementary Fig. 6) of RIM4 KO_PCP2 mice, making it difficult to judge their pathophysiological relevance.

In the same samples, we also assessed the density of PF and CF synapses in the proximal (0–30 µm), middle (30–60 µm) and distal (60–90 µm) parts of the molecular layer. In each part of the molecular layer, we counted individual release sites [only asymmetric, glutamatergic, defined by vesicles, narrow synaptic cleft and postsynaptic density (PSD)] and classified them as either PF or CF synapses if they terminated on spines (Fig. 6A and B) or large dendritic shafts (Fig. 6D), respectively. We observed opposite trends in the density of PF and CF synapses (release sites). While the density of PF synapses in the proximal and medial part of the molecular layer was reduced by ∼20% in RIM4 KO_PCP2 mice (Fig. 6E), the density of CF synapses in these proximal parts did not differ. Conversely, in the distal part of the molecular layer (60–90 µm) PF synapse density was comparable between groups but there were significantly fewer CF synapses in RIM4 KO_PCP2 mice. This is consistent with the idea that the developmental process of territory assignment between CF and PF synapses is mutually linked³⁴ and further supports our confocal data, which also indicate that CF synapses in RIM4 KO_PCP2 mice extend less into the distal region of Purkinje cell dendrites (Fig. 5J and K).

RIM4 in excitatory cortical, hippocampal or inferior olive neurons is dispensable for normal motor function

By performing histochemical stainings for the activity of the enzyme β-galactosidase (lacZ) in mice expressing lacZ under the endogenous RIM4 promoter, we observed very strong labelling in IO neurons (Fig. 7A). IO neurons play an important role in motor control by the olivo-cerebellar loop. RIM4 was also found to be expressed, although at substantially lower levels, in hippocampus and cortex. To test for the potential contribution of the IO and the hippocampus/cortex to dyskinesia and dystonia, we also aimed to delete RIM4 specifically from these regions only. We crossed conditional RIM4^fl/fl mice with two lines expressing Cre recombinase under the control of either the corticotropin releasing factor (CRF; Fig. 7B)³⁵ or the NeuroD subfamily 6 (NEX; Fig. 7C)³⁶ promoter to generate RIM4 KO_CRF and RIM4 KO_NEX mice, respectively. We verified cell type-specific knockout by quantifying RIM4 mRNA levels in the respective brain regions and cerebellum using fluorescent in situ hybridization (RNAScope) (Fig. 7D and E). In contrast to RIM4 KO_const and KO_PCP2 mice, the overall brain size and the size of various subregions were unaltered in both RIM4 KO_CRF and KO_NEX mice (Fig. 7F and G). Five consecutive days of video monitoring revealed that neither of these conditional RIM4 KO lines exhibited spontaneous (Fig. 7H) nor caffeine-induced episodes of dyskinesia/dystonia or any other motor impairments (Fig. 7I). Furthermore, the motor performance of RIM4 KO_CRF and RIM4 KO_NEX mice on an accelerating rotarod (Fig. 7J) and on a simple or complex running wheel (data not shown) was comparable to that of their wild-type littermates. These results strongly suggest a highly Purkinje cell-specific role of RIM4.

To explore if RIM1/2 share functions in cerebellar motor circuits with RIM4, we used the same PCP2-Cre driver line that fully reproduced the symptoms of the RIM4 KO_const and deleted RIM1 and RIM2 from Purkinje cells. However, RIM1/2 KO_PCP2 mice did not show episodes of motor impairments during video monitoring in the home cage (Fig. 7H), nor could they be induced by caffeine injection (Fig. 7I). In addition, their performance on the accelerating rotarod (Fig. 7J) and their spontaneous firing activity were unaltered (Fig. 7K). Thus, RIM4 plays a unique role in Purkinje cells, which is not replicated by the long RIM family members.

Differential developmental requirement of RIM4

Our RIM4 ablation strategies resulted in a loss of the protein from early development on (constitutive KO) and around postnatal day (P) 6 for PCP2-Cre.³² Thus, it is not clear whether the structural and functional changes observed in adult mice are due to an acute deficiency of RIM4 in adulthood or whether there is an earlier and transient requirement for RIM4 that results in irreversible changes that persist into adulthood. To address this, we delivered RIM4 to Purkinje cells in RIM4-deficient mice by injecting recombinant adeno-associated virus (rAAV) particles expressing a mRFP1-T2A-Flag-RIM4 cassette (Fig. 8A) at either P0 (RIM4_P0 based on RIM4 KO_const) or 3 months of age (RIM4_Adult based on RIM4 KO_PCP2) and examined whether key structural and functional alterations could be reversed. RIM4 wild-type and KO transduced with rAAV-mRFP served as controls (Fig. 8A). Viral delivery of RIM4 at P0 reversed the dendritic tree abnormalities, reduced the mean firing frequency and reverted the increased caffeine sensitivity of Purkinje cell firing observed in RIM4 KO_const mice (Fig. 8B, C, F and G). In contrast, when we virally delivered RIM4 to adult RIM4 KO_PCP2 mice, we observed a divergent effect: while dendritic arborization was not restored (Fig. 8D and E), both the mean frequency and the caffeine sensitivity of Purkinje cell firing were restored to levels seen in wild-type mice (see values earlier) (Fig. 8F and G). Delivery of mRFP-RIM4 to wild-type (OE_P0, OE_Adult) mice did not affect dendrite branching or firing behaviour of Purkinje cells (Supplementary Fig. 7). Taken together, these results suggest that there is a critical window of time during early development, which requires RIM4 for the correct establishment of the dendritic tree of Purkinje cells. The absence of RIM4 during this critical period results in an irreversible alteration of the dendritic branching pattern. In contrast, the functional alterations observed in adult mice result from the acute absence of RIM4 and can be restored by providing RIM4 protein.

Discussion

We have identified an unexpected function for the short RIM family member, RIM4, which differs from the presynaptic organizing role of the long RIMs, RIM1/2, in the CNS. RIM4 does not play a significant presynaptic role, but is crucial for somatodendritic signal integration, and exerts this function specifically in Purkinje cells. Furthermore, the deletion of RIM4 causes a severe motor disorder that is not present in knockout mice of any other RIM family member, highlighting the systemic importance of the distinct cellular function of RIM4.

RIM4 does not play a role in the regulation of synaptic vesicle release

Paired-pulse plasticity of synaptic responses is a robust indicator of transmitter release probability. Mice deficient for the long RIMs (RIM1/2) exhibit changes in paired-pulse plasticity of synaptic responses and a decreased release probability.^37-40 We found that paired-pulse plasticity was not altered in the absence of RIM4 at the PC-DCN synapse (RIM4 KO_PCP2), the PF-PC, the CF-PC synapses and the Schaffer collateral synapses (both RIM4 KO_const). Furthermore, calcium sensitivity at the PF-PC synapse and the frequency of mEPSCs in the hippocampus were not affected by RIM4 deletion. Taken together, these results strongly suggest that RIM4 has no apparent presynaptic function. This is consistent with the view that the presynaptic role of the long RIMs in organizing voltage-gated calcium channels and synaptic vesicles in the AZ requires their multi-domain structure, whereas RIM4 comprises only a single C2 domain and a short isoform-specific N-terminal sequence.¹⁴

Somatodendritic role of RIM4

While we found little evidence for a presynaptic role of RIM4, we observed a number of structural and functional changes suggesting that RIM4 exerts important somatodendritic functions in Purkinje cell excitability. Unlike our previous study, which did not allow us to conclude whether the observed reduced dendritic tree¹⁴ might be secondary to a presynaptic effect, our current dataset now provides strong evidence that RIM4 in Purkinje cells is required for proper dendritic development: deletion of RIM4 specifically in Purkinje cells resulted in reduced dendritic branching, and this branching could be restored by viral delivery of RIM4 only to Purkinje cells. Our functional measurements clearly demonstrate a role for RIM4 in the organization of somatic excitability. In the absence of synaptic input (pharmacologically blocked), Purkinje cells displayed decreased firing rates, broadened action potentials and heightened sensitivity to inhibition of firing by caffeine due to the absence of RIM4. Firing deficits were replicated through selective RIM4 ablation in Purkinje cells. Furthermore, by reintroducing RIM4 via viral expression in adult RIM4 KO_PCP2 mice, firing rates and caffeine sensitivity were restored to wild-type levels.

RIM4 plays a unique key role in Purkinje cells

Constitutive deletion of RIM4 from all cells, starting early during development, and selective deletion of RIM4 only from Purkinje cells during the second postnatal week resulted in qualitatively and quantitatively the same phenotype. This strongly suggests that RIM4 is of paramount importance for Purkinje cells only from adolescence onward. Our data suggest that the absence of RIM4 throughout development and in other neuron types leads to either no changes or subtle effects. However, it could be argued that constitutive deletion of RIM4 may cause additional deficits that would be masked if RIM4 played similar roles in functionally opposing neurons, such as excitatory and inhibitory neurons. We tested this idea by generating RIM4 KO_NEX mice in which RIM4 is primarily absent from cortical and hippocampal excitatory neurons, but which show no signs of motor deficits. The finding that deletion of RIM4 from neurons of the IO, which express RIM4 at exceptionally high levels compared to all other neurons in the brain, did not result in any obvious changes in the RIM4 KO_CRF mice is unexpected and surprising. It should be noted that many of our assays were primarily focused on analysing motor performance, and it may well be that less severe disturbances caused by RIM4 deficiency in non-Purkinje cells remained undetected.

Postnatal deletion of RIM4 only from Purkinje cells was sufficient to induce (i) baseline motor dysfunction (wheel, rotarod); (ii) spontaneous and caffeine/ethanol induction of dyskinetic/dystonic episodes; (iii) reduced spontaneous firing in Purkinje cells; (iv) enhanced suppression of firing by caffeine; and (v) alterations of in vivo calcium signals in Purkinje cells. These findings strongly suggest that cerebellar Purkinje cells are responsible not only for reduced baseline motor coordination, but also for the induction of dyskinetic/dystonic episodes. In the absence of RIM4 in Purkinje cells, caffeine induces a pronounced weakening of tonic firing of Purkinje cells, leading to unbalanced activity in the cerebellar nucleo-olivary loop via disinhibition of the deep cerebellar nuclei, the target of Purkinje cells. The important and primary role of RIM4 in Purkinje cells in this process is also exemplified by the fact that the firing deficit and increased caffeine responsiveness can be restored by viral delivery of RIM4 to Purkinje cells. A causal role of Purkinje cells in ataxia and the initiation of paroxysmal dystonia is consistent with the finding that deletion of P/Q-type VGCCs only from Purkinje cells also mimics key aspects of tottering mice.^41,42

Aberrant synaptic integration in Purkinje cells

The three most prominent features of abnormal signal integration were (i) loss of characteristic frequencies of calcium signals; (ii) substantially reduced amplitude of calcium signals; and (iii) reduced spatial extent of correlation of calcium activity of neighbouring Purkinje cells. In our view, it is unlikely that these changes are mechanistically attributable to alterations intrinsic to Purkinje cells. Rather, as discussed later, a loss of firing rhythms and conversion to random firing of IO neurons could explain the three hallmark changes in calcium signalling in Purkinje cells.

The appearance of spontaneous dendritic calcium signals in Purkinje cells reflects CF activity and thus action potential firing of IO neurons. The loss of the three characteristic frequencies of calcium signals in Purkinje cells of KO mice in vivo (2, 0.5, 0.2 Hz, Fig. 4F) clearly indicates altered firing patterns of IO neurons. In KO mice, the occurrence of calcium signals in Purkinje cells appeared almost random, suggesting that the action potential firing pattern of the presynaptic IO neuron is also random in KO mice.

A disturbed firing pattern of IO neurons may also explain the greatly reduced in vivo calcium (GCaMP) signal amplitude in Purkinje cells of RIM4 KO mice. Such dendritic calcium signals are typically not caused by single action potentials, but rather by bursts of action potentials from IO neurons.⁴³ The more action potentials IO neurons fire per burst, the larger the dendritic calcium transients in Purkinje cells. Because the intervals between action potentials within an IO burst are short (∼5 ms),⁴³ action potentials do not appear as individual calcium transients but accumulate into a relatively slow composite GCaMP signal (rise time ∼50 ms, decay time constant ∼300 ms) (Supplementary Fig. 4A and B). If the accumulation was perfectly linear, the amplitude of the GCaMP signal would directly reflect how many action potentials the IO neuron fired per burst. Random fluctuations in the number of action potentials per burst in IO neurons should follow a Poisson distribution, and so should then, if accumulation is more or less linear, the calcium signal amplitudes in Purkinje cells. Indeed, the distribution of calcium signal amplitudes in both wild-type and RIM4 KO mice could be fit reasonably well by a Poisson distribution. The fit parameters suggested that, on average, IO neurons in RIM4 KO mice fire only ∼60% of action potentials per burst compared to wild-type littermates (lambda was reduced to 60%). Thus, action potentials in IO neurons of RIM4 KO mice do not coincide as much as in wild-type mice and do not show the clustered occurrence of high frequency firing (∼200 Hz)⁴³ such that there are fewer action potentials per bursts. This means that not only the three slow characteristic frequencies that can be measured directly with GCaMP imaging are lost, but also a high frequency component that underlies the formation of longer and more pronounced bursts, supporting the view of an overall loss of coordination of action potential firing towards a random firing pattern of IO neurons.

Neighbouring Purkinje cells form so-called micro-zones, which are defined by commonalities in afferents and in the target areas of their projections. CF activity in Purkinje cells within such micro-zones has been reported to be temporally correlated due to the synchronized action potential activity of IO neurons projecting to that micro-zone.⁴⁴ In our dataset, we also found this spatiotemporal correlation of dendritic calcium activity amongst a dozen Purkinje cells, most likely reflecting CF activity, within ∼20 µm in wild-tye mice (Fig. 4G and H). We observed this correlation to a significantly lesser degree in RIM4 KO and RIM4 KO_PCP2 mice. These findings suggest that not only have individual IO neurons lost their typical firing pattern, but also that IO neurons that belong to a micro-zone of Purkinje cells demonstrate weakened temporal correlation with each other. Taken together, the modifications of in vivo calcium signals seen in Purkinje cells of RIM4 KO and RIM4 KO_PCP2 mice can be attributed to altered firing patterns of IO neurons. This could explain why calcium transients recorded in response to experimental electrical stimulation of Purkinje cells in acute brain slices did not appear to differ between the two groups (Fig. 4K and L). Of course, it should be noted that our conclusions are indirect. We inferred alterations in IO neuron firing from changes in calcium signals that we recorded in Purkinje cells without direct recordings from IO neurons.

Deleting RIM4 specifically in IO neurons does not cause spontaneous or induced motor episodes, coordination problems or structural abnormalities of the cerebellum (Fig. 7). On the other hand, deleting RIM4 only in Purkinje cells reproduced the in vivo calcium signalling deficits of Purkinje cells in the RIM4 KO_const mice showing that the presumed alterations of the firing patterns of IO neurons are likely not due to the loss of RIM4 from IO neurons. Firing patterns of IO neurons could be altered because the RIM4 deficiency of Purkinje cells leads to an alteration of some cell-to-cell signalling between Purkinje cells and IO neurons. This would necessitate an unusual, long-range signalling from climbing fibre synapses back to the somata of IO neurons in the brain stem to regulate their excitability. We believe it is more likely that the altered firing pattern of IO neurons rather is a consequence of chronically altered network activity in olivo-cerebellar circuits. Almost every night, mice experience episodes of severe motor impairments and this will be accompanied by aberrant network activity, not only in Purkinje cells. Purkinje cells are connected to IO neurons via the cerebellar nucleo-olivary projection and an adaptation of this tri-synaptic circuit to altered somatodendritic physiology and aberrant activity of Purkinje cells in RIM4 KO mice would not be unexpected.

If the calcium signal abnormalities in Purkinje cells may be secondary to aberrant network activity, the question arises as to which of the other alterations observed in the Purkinje cells of RIM4 KO_const or KO_PCP2 mice are directly caused by the absence of RIM4. As explained earlier, the reduced firing frequency and the enhanced caffeine effect are most likely directly due to the loss of RIM4 in Purkinje cells. Furthermore, loss of RIM4 in Purkinje cells also caused an aberrant distribution and density of CF synapses (Figs 5K and 6F). Normally, there is a developmental shift of CF synapses to more distal locations,⁴⁵ resulting in more widespread activation of Purkinje cell dendrites. As we found that CF synapses preferentially accumulate on proximal dendrites in RIM4 KO mice, it appears that this developmental repositioning as well as the usual segregation of PF synapses (Fig. 6E) do not fully occur in the absence of RIM4. We did not show a reversal of those phenomena by viral delivery of RIM4 and therefore it could be argued that these changes are secondary to the periodic episodes of motor problems. However, it is known that the redistribution of CFs along the dendritic tree is completed around the end of the third postnatal week,^46,47 whereas the episodes of dyskinesia and dystonia begin thereafter, with the beginning of the third postnatal week. For this reason, we propose that the CF redistribution is not secondary to the motor episodes but a consequence of the loss of RIM4 from Purkinje cells. A similar proximal distribution of CF synapses was observed in KO mice lacking the auxiliary VGCC subunit α2δ2.⁴⁸ Purkinje cells in these mice also showed increased CF EPSC amplitudes and computational modelling suggested that the more proximal location of the synapses can largely explain the increased amplitude. This effect may be even stronger in our case because not only are the synapses more proximal, but the dendrites are also shorter and less branched, making the dendritic tree more electrically compact and facilitating the spread of dendritic charge into the somatic region sampled by the recording pipette.

Taken together, we propose that RIM4 deletion from Purkinje cells leads to disturbed cellular signalling and causes aberrant dendritic arborization, climbing fibre innervation and firing, and that the resulting abnormal inhibitory output of Purkinje cells is responsible for the deteriorated motor coordination, the caffeine-induced and the nocturnal episodes of motor impairment.

What might be the molecular mechanism of RIM4's performance?

Our experiments do not allow us to identify a direct interaction partner of RIM4 or to pinpoint the exact molecular mechanism. However, comparison with other mouse lines with similar phenotypes and with functions of the RIM family of proteins may allow us to narrow down the list of possible candidate mechanisms.

The paroxysmal occurrence of dyskinetic and dystonic attacks lasting hours and their inducibility by caffeine, ethanol and stress are hallmarks of a group of mouse lines with mutations in or loss of the CACNA1A gene, which encodes the alpha subunit of the P/Q-type VGCC. This group includes the Tottering, Rocker and Rolling Nagoya mice, among others, and represents a number of genetic models of the human disease episodic ataxia 2 (EA2).^49,50 Many of these mutants are associated with decreased calcium entry through P/Q-type VGCCs, which are the most prominent calcium channel type in Purkinje cells. We did not observe a reduction in calcium entry in situ, which could be taken as evidence for altered function of P/Q-type calcium channels. However, considering that CF-induced calcium signals in Purkinje cells in acute slices (OGB-1) were of similar magnitude in RIM4 KO mice (Fig. 4K and L), although CF EPSCs were greatly increased (Fig. 4O), it appears that recruitment of calcium entry by CF synapses may be less effective in RIM4 KO mice and it is possible that a loss-of-function of P/Q VGCCs is functionally compensated by a regulation of the AMPA-R component of CF synaptic responses in Purkinje cells to restore the levels of calcium entry seen in wild-type mice. RIM family proteins, particularly RIM1/2, are required for maintaining normal levels of presynaptic calcium channels at the AZ.^9,10 This function of RIM1/2 is thought to involve a direct interaction of the RIM1/2 PDZ domain with the C-terminus of P/Q, N and R-type VGCCs.⁹ A PDZ domain is not present in RIM4. In addition to the PDZ domain, the C-terminal C2B domain of RIM1/2 and RIM3/4 has been reported to interact with the β-subunits of VGCCs.^15,16 This interaction weakens the calcium-dependent inactivation of VGCCs, thereby increasing calcium entry. This weakening of calcium-dependent inactivation may not significantly affect the brief period of calcium entry mediated by single action potentials, as they occur in presynapses, and therefore not be important for transmitter release (cf. Figs 2 and 5O). Nevertheless, it could substantially elevate calcium entry during longer lasting depolarizations such as complex spike signals (∼30 ms). Thus, the absence of RIM4 may lead to enhanced calcium-dependent inactivation and reduced current through VGCCs in Purkinje cells, causing dyskinesia/dystonia/ataxia as observed in mice with mutated CACNA1A genes. Notably, mutations in the β4 subunit (CACNB4, lethargic mice) also result in paroxysmal dyskinesias, similar to what we observed in RIM4 KO mice.⁵¹ However, despite many phenotypic similarities between strains with mutated CACNA1A or CACNB4 genes and RIM4 KO mice, there are some important differences. First, these mouse lines exhibit epileptic seizures (absence) in addition to dyskinesia, even when VGCCs are deleted only in Purkinje cells.⁴¹ In RIM4 mice, we never observed signs of such seizures during continuous video and EEG monitoring. Second, the paroxysmal motor episodes of tottering mutant mice respond to treatment with acetazolamide,^52,53 whereas this drug had no effect in our RIM4 KO_const mouse line (Supplementary Fig. 9). Third, in RIM4 KO mice, Purkinje cell action potential firing is characterized by heightened regularity (and reduced frequency) and rarely exhibits bursting behaviour (Supplementary Fig. 10). Conversely, in tottering mice, Purkinje cells fire with greater irregularity,^54,55 indicating increased bursting. These differences between dystonic/ataxic syndromes associated with CACNA1 and our RIM4 KO mouse lines may raise scepticism regarding the notion that deletion of RIM4 in Purkinje cells acts through altered VGCC signalling.

A second form of paroxysmal dyskinesia and dystonia, characterized by episodes of motor dysfunction lasting for hours, but without seizures, is not associated with VGCCs but with mutations in the MR-1/PNKD gene.^56,57 The gene and protein were named after the familial human disease ‘paroxysmal non-kinesigenic dyskinesia’, which occurs in patients carrying mutations (missense p.A7V, p.A9V) in the PNKD gene.⁵⁸ Mice transgenically expressing the mutant form of the CNS-specific L (long) isoform of the PNKD protein exhibit symptoms resembling the human disease and suffer from paroxysmal dyskinetic and dystonic attacks very similar to what we observed in RIM4 KO mice.⁵⁹ Dystonic motor episodes can also be triggered in these mice through caffeine and ethanol injections or stress (such as handling them).⁵⁹ Notably, knockout mice lacking PNKD do not show dystonia episodes.⁵⁹ Thus, motor episodes do not occur in the presence or absence of PNKD, but only when the mutant form of PNKD is expressed. This highlights the importance of the N-terminal mutations in causing the disease and has led to the conclusion that these mutations confer a gain-of-function, albeit a systemically deleterious function, and has sparked interest in the relevance of this mutation to the pathology. Thus, the absence of the PNKD protein, and therefore any function it might normally have, does not lead to the phenotype. The phenotype only occurs if the mutant protein is expressed, making the consequences of these mutations of particular interest. Using an interaction assay, it was found that wild-type PNKD interacts with the C2B domains of RIM1 and RIM2, but the mutant PNKD does not.¹⁷ However, neurons expressing the mutant form of PNKD that does not bind RIM1/2 showed normal synaptic release.¹⁷ This suggests that RIM1/2 binding to PNKD does not play a significant role in the occurrence of the motor phenotype, which is consistent with our observation when RIM1/2 is knocked out in Purkinje cells. Notably, the RIM4 C2 domain shares high homology with the C2B domains of RIM1/2 and strongly binds to PNKD (Supplementary Fig. 8). It could be expected that the A7V and A9V mutations in PNKD also block this interaction. PNKD is highly expressed in cerebellar Purkinje cells,⁵⁹ so it is conceivable that PNKD and RIM4 interact in Purkinje cells in vivo. The similar motor phenotypes resulting from PNDK mutation to a non-RIM binding form and RIM4 deletion, and the lack of such phenotypes following PNKD or RIM1/2 deletion (specifically in Purkinje cells, our study) or reduction of RIM1/2 levels due to PNKD deletion,¹⁷ suggest that the PNKD-RIM4 interaction may be required for normal somatodendritic function of Purkinje cells. If this interaction does not occur in Purkinje cells either due to loss of RIM4 or if it is blocked by the mutations in the N-terminus of L-PNKD, PNKD could negatively impact the normal function of Purkinje cells, leading to paroxysmal dyskinesia and dystonia. The cellular function of the PNKD protein is currently unknown, but wild-type PNKD may be involved in the regulation of cellular oxidative stress.⁵⁶ Mice expressing the mutant form of PNKD show altered dopaminergic signalling in striatal pathways.^59,60 However, research about mice with mutations in other brain regions, particularly the cerebellum, is currently unavailable. If murine PNKD mechanistically intersects with RIM4 deficiency as we propose, then PNKD mutant mice should replicate our key findings of altered Purkinje cell physiology. However, even if our hypothesis is correct, it should be noted that RIM4 may fulfil multiple functions in Purkinje cells. Interaction with β-subunits of VGCCs and possibly other effectors may play additional roles in causing the phenotype. Future research is needed to identify the precise molecular interactions and partners responsible for RIM4's previously unidentified role.

Data availability

Further information and requests for resources should be directed to and will be fulfilled upon reasonable request by Susanne Schoch McGovern ([email protected]) and Dirk Dietrich ([email protected]).

Acknowledgements

We would like to acknowledge the assistance of the Virus (supported in part by SFB1089), Proteomics (M. Sylvester), Genomics, Electron Microscopy (H. Beckert, L. Maus) and Bioinformatics Core Facilities of the University of Bonn Medical Faculty and the Genomics and Bioinformatics platform at LIMES. We thank Sabine Opitz, Pia Trebing, Shayne Gilgenbach and Jessica Klemmer for excellent technical assistance, Polina Gulakova for cloning RIM4 viral vectors, Elizabeth Matthews and Sabrina Delattre for performing initial electrophysiological experiments, Tony Cijsouw for initial immunocytochemical trials and discussions, Julia Wolf for support in preparing tissue for electron microscopy, Khondker Ushna Sameen Islam and Erick Martinez-Chavez for help with RNAScope, Ileana Christea for binding assays, Martin Schwarz for help with clearing of PCP2-Cre::B6.Cg-Gt(ROSA)26Sor^{tm14(CAG−tdTomato)Hze}/J mouse brains, Lena Gschossmann for support in setting up the in vivo imaging analysis and S. Goebbels and K. Nave for kindly providing Neurod6^tm1(cre)Kan mice. Graphical abstract was created with BioRender.com.

Funding

Our work was supported by the German Research Foundation (DFG) (S.S.: SFB1089, SCHO 820/4-1, SCHO 820/6-1, SCHO 820/7-1, SCHO 820/5-2, SPP1757; D.D.: SFB1089, SPP1757, INST1172 15, DI853/3-5&7, INST 217/785-1), the Federal State of Nordrhein-Westfalen (iBehave, S.S.), the BONFOR program of the Medical Faculty of the University of Bonn (S.S., D.D.).

Competing interests

The authors report no competing interests.

Supplementary material

Supplementary material is available at Brain online.

References

Author notes