Genetic silencing of olivocerebellar synapses causes dystonia-like behaviour in mice

Theories of cerebellar function place the inferior olive to cerebellum connection at the centre of motor behaviour. One possible implication of this is that disruption of olivocerebellar signalling could play a major role in initiating motor disease. To test this, we devised a mouse genetics approach to silence glutamatergic signalling only at olivocerebellar synapses. The resulting mice had a severe neurological condition that mimicked the early-onset twisting, stiff limbs and tremor that is observed in dystonia, a debilitating movement disease. By blocking olivocerebellar excitatory neurotransmission, we eliminated Purkinje cell complex spikes and induced aberrant cerebellar nuclear activity. Pharmacologically inhibiting the erratic output of the cerebellar nuclei in the mutant mice improved movement. Furthermore, deep brain stimulation directed to the interposed cerebellar nuclei reduced dystonia-like postures in these mice. Collectively, our data uncover a neural mechanism by which olivocerebellar dysfunction promotes motor disease phenotypes and identify the cerebellar nuclei as a therapeutic target for surgical intervention.

D ystonia is an incurable neurological disorder that is defined by abnormal muscle contractions and repetitive twisting of affected body parts. These symptoms intensify during movement 1 . Dystonia can occur either as an independent disease or as a comorbid condition with other movement disorders including ataxia, tremor and Parkinson's disease 2 . The age of onset is variable. Hereditary, primary dystonia is common in young children and teens, whereas focal dystonia often affects adults. It is becoming clear that dystonia is a result of an aberrant motor network, and recent work points to the cerebellum via the basal ganglia as capable of instigating dystonia 3 .
The inferior olive projects its axons exclusively to the cerebellum. Among its targets are direct contacts with the Purkinje cell dendrites via projections called climbing fibres. Climbing fibres induce a unique action potential called the complex spike 4 . The climbing fibre-Purkinje cell synapse mediates the predominant mode of olivocerebellar communication 5 . It coordinates the precise timing of motor commands, although it may also control motor learning and error correction during movement 4 . Climbing fibres are excitatory; they release glutamate and modulate Purkinje cell activity. Accordingly, in vivo, the complex spike action potentials are powerful enough to control the intrinsic firing of Purkinje cells 6 that control the pattern of firing in the cerebellar nuclei 7,8 .
In specific cases of human dystonia, the inferior olive was found to have metabolic defects and it was infiltrated by microglia and macrophages 9,10 . In rodent models, electrophysiological analyses revealed defects in cerebellar activity and olive communication 11 . In particular, in the dystonic rat dt, complex spike firing rate is lower than normal and the Purkinje cells and cerebellar nuclei exhibit abnormal burst firing 12 . Previous work using lesioning methods to destroy the inferior olive also suggested its potential for initiating dystonic postures 13 .
To further address the role of the inferior olive in dystonia, here we sought to manipulate olivocerebellar activity, and then test whether motor behaviours that resemble dystonia arise. Our goal was to develop a noninvasive in vivo approach for altering olivocerebellar function. We therefore devised a conditional mouse genetic model to test the hypothesis that loss of olivocerebellar function triggers cerebellar defects that cause dystonia-like behaviour. In the model, eliminating excitatory synaptic neurotransmission between the inferior olive and cerebellum selectively targeted olivocerebellar communication.
To examine if and how dystonia-like behaviours emerge we combined our genetic approach with behavioural paradigms, in vivo electrophysiology, molecular expression analysis and anatomical examination. We also tested whether cerebellar neuromodulation is an effective approach for alleviating dystonia using reversible chemical lesions and deep brain stimulation (DBS). The data uncovered a cerebellar circuit origin of dystonialike behaviour and its mechanism of action. The data also provided new opportunities for restoring movement with cerebellar directed DBS.

Results
A genetic strategy for targeting olivocerebellar activity. We exploited a unique intersection between the expression patterns of two genes to conditionally block excitatory neurotransmission from the inferior olive to the cerebellum in mice. Specifically, we removed vesicular glutamate transporter 2 (Vglut2) from the olive using a Ptf1a Cre driver allele (Fig. 1a,b; see Methods). Ptf1a encodes a bHLH transcription factor that is primarily expressed in inhibitory neurons in which it controls cell lineage in discrete structures in the brain (hypothalamus and cerebellum 14 ; retina 15 ; spinal cord 16 ), except for in the inferior olive where it controls excitatory neuron development 17 (Fig. 1c). Vglut2 encodes a transporter protein required for loading the neurotransmitter glutamate into presynaptic vesicles for chemical synaptic signalling in excitatory cells 18 and is the vesicular transporter expressed in axon terminals of inferior olivary neurons. We postulated that Ptf1a and Vglut2 expression should intersect in the inferior olive, since the olive is the only glutamatergic brain structure known to express Ptf1a 17 . We first crossed the Ptf1a Cre mice to a Rosa TdTomato reporter to test the recombination efficiency in the inferior olive ( Supplementary  Fig. 1). By examining red fluorescent protein-positive cells in combination with NeuN expression in the olive, we found that Ptf1a Cre ;Rosa lox-stop-lox-tdTomato mice exhibited B90% recombination (90 ± 3.5%; n ¼ 27 sections from 3 animals; Fig. 1d,e). Moreover, consistent with previous data, we found limited recombination anterior to the hindbrain 14 (cerebral cortex and hypothalamus; Supplementary Fig. 1a-i). Next, we crossed the Ptf1a Cre mice to mice with floxed alleles of Vglut2 (Vglut2 flox ¼ Vglut2 fx ). Based on when the two genes turn on, we anticipated that Vglut2 would be deleted from inferior olivary cells as they migrate towards the base of the brainstem-thus, climbing fibres should not express VGLUT2 and therefore would be unable to signal with glutamate ( Fig. 1j). Analysis of mRNA expression in neonatal (P0) and adult mice using in situ hybridization showed that in the Ptf1a Cre ;Vglut2 fx/fx mutant mice Vglut2 was substantially reduced in the olive, but not in the surrounding brainstem nuclei that project mossy fibres to the cerebellum (Fig. 1c). Accordingly, in the cerebellar cortex, VGLUT2 protein was eliminated from the molecular layer, the site of termination of climbing fibres (Fig. 1g,h), with an efficiency of B91% across the cerebellum ( Fig. 1f and Supplementary Table 1). In contrast, mossy fibre terminals in the granular layer still expressed VGLUT2 with a similar distribution between Vglut2 fx/fx and Ptf1a Cre ;Vglut2 fx/fx mice ( Fig. 1g and Supplementary Table 1). CART, which is heavily expressed in a subset of climbing fibres in lobules IX and X, 19 and transmission electron microscopy revealed that deleting Vglut2 in the inferior olive did not disrupt the termination of climbing fibres in the molecular layer or their targeting onto the shafts of Purkinje cell dendrites (Fig. 1h,i).
Loss of VGLUT2 in climbing fibres should eliminate Purkinje cell complex spike activity. We tested this using extracellular, in vivo electrophysiology recordings (Fig. 2a-c). The recordings revealed the cellular-level precision of the genetic approach as complex spikes were absent in most Purkinje cells whereas simple spikes persisted (Fig. 2e). We then used a loose-patch, juxtacellular recording approach in which we used glass electrodes to fill a subset of Purkinje cells with neurobiotin after recording to confirm cellular identity after isolating single units in vivo (Fig. 2d-f). From a total of 119 recorded neurons in Ptf1a Cre ;Vglut2 fx/fx mice, we only found 20 mutant Purkinje cells with complex spikes (16.8%; Fig. 2g). However, the histological confirmation of recorded Purkinje cells was not possible in the metal electrode recordings (Fig. 2e). We therefore limited the potential for erroneous classification of neurons first using stereotaxic coordinates; cerebellar nuclear neurons are located deep within the core of control and mutant cerebella and therefore should not be easily confused with more superficially recorded Purkinje cells. Second, we assessed each cell for spike waveform; Purkinje cell simple spikes, which were preserved in the mutants, had a distinct waveform. Still, even with these reliable criteria we cannot guarantee the correct identification of every recorded cell. Severing the interaction between climbing fibres and Purkinje cells did not cause regression of climbing fibres, cell loss, changes in cell distribution ( Supplementary Fig. 2a-n) or a compensatory increase in Vglut1 mRNA in the inferior olive or VGLUT1 protein in the molecular layer as measured by mean pixel intensity in fluorescence images ( Supplementary Fig. 3a-  . We used a dystonia rating scale (see Methods) to quantify the behaviour and found that the mutants scored in the severe range. No significant body weight differences were found at any age measured (Fig. 3b). Performance on rotarod was impaired in Ptf1a Cre ;Vglut2 fx/fx mice ( Fig. 3c and Supplementary Table 1). Open field analysis over 30 min showed that Ptf1a Cre ;Vglut2 fx/fx mice were significantly less mobile compared with Vglut2 fx/fx mice (Fig. 3d  Deletion of Vglut2 alters Purkinje cell firing in vivo. To understand how these motor deficits arose, we examined the electrophysiological properties of Purkinje cells, interneurons and cerebellar nuclei neurons because abnormal cerebellar function is implicated in dystonia 12,24 . We recorded Purkinje cell firing in anaesthetized mice to examine spontaneous activity in P19-P21 juvenile mice at weaning age (Fig. 4a). We measured their frequency (spikes s À 1 ; Hz), coefficient of variance (CV; overall regularity) and CV2 (local regularity). Compared with control littermates, frequency (Fig. 4b,c) and CV (Fig. 4b,c) were significantly lower in the mutants, although CV2 was not changed (Fig. 4b,c). At 5 months of age, when the mice reached adulthood, Purkinje cell firing normalized in the mutants and was no longer abnormal with the exception of the absence of complex spikes (Supplementary Table 1). It is interesting that Purkinje cell function also recovers after the olive is lesioned with 3-acetylpyridine 25 . Purkinje cells are not the only cerebellar cortical neurons to receive input from climbing fibres. Climbing fibres signal to molecular layer interneurons by glutamate spillover 26 . Loss of VGLUT2 in climbing fibres could therefore alter interneuron firing. We found similar firing properties in putative stellate cell and basket cell interneurons in We next used an awake recording approach in head-fixed, behaving mice to examine how Purkinje cells in the Ptf1a Cre ;Vglut2 fx/fx mice fired during active behaviour ( Fig. 4d-h). Similar to what we observed in the anaesthetized adult mice, we found no difference between awake, adult Vglut2 fx/fx and Ptf1a Cre ;Vglut2 fx/fx Purkinje cells (Fig. 4e-h and  Supplementary Table 1).
Structural abnormalities are often subtle in dystonia. However, the length of the primary Purkinje cell dendrite is reduced in a knock-in mouse model of Dyt1 (ref. 27). To test whether there are similar defects in the Ptf1a Cre ;Vglut2 fx/fx mice, we measured molecular layer thickness as a proxy for dendrite span. Molecular layer thickness is a sensitive and straightforward measure for developmental and disease-associated defects that disrupt Purkinje cell dendrite size and/or the placement of Purkinje cells into a perfect monolayer 28 . At birth and during postnatal development, molecular layer thickness was significantly less in Ptf1a Cre ;Vglut2 fx/fx compared with Vglut2 fx/fx mice (Fig. 5a-d and Supplementary Table 1). However, by adulthood there was no difference between the genotypes (Supplementary Table 1). We next checked for ectopic upregulation of tyrosine hydroxylase (TH) in Ptf1a Cre ;Vglut2 fx/fx Purkinje cells ( Supplementary Fig. 6). TH is a precursor for dopamine, norepinephrine and epinephrine. Cerebellar dysfunction is often associated with ectopic upregulation of TH in Purkinje cells; its presence is thought to result from changes in excitability and/or Ca 2 þ dysregulation 11 . However, we did not find an ectopic upregulation of TH in the Ptf1a Cre ;Vglut2 fx/fx mice (n ¼ 4 animals per genotype; Supplementary Fig. 6a-d), despite the presence of dystonia.
Vglut2 mutants have irregular firing in cerebellar nuclei. Why do Ptf1a Cre ;Vglut2 fx/fx mice remain dystonic after Purkinje cell structure and simple spike firing are corrected? Purkinje cells project to three pairs of cerebellar nuclei: the fastigial nuclei (medial), interposed nuclei (middle) and dentate nuclei (lateral). Cerebellar nuclear neurons exhibit abnormal irregular firing in several models of dystonia (dystonic rat 12 , tottering 29 , rapid-onset dystonia-parkinsonism 24 and mutation of Lamb1 in mouse 30 ). These previous data indicate that regardless of how the dystonia is caused, one common consequence might be irregular activity in the cerebellar nuclei. To test this hypothesis, we recorded the activity of the cerebellar nuclear neurons in anaesthetized weanling and awake adult Ptf1a Cre ;Vglut2 fx/fx mice (Fig. 6a-c).
In accordance with the impact on Purkinje cells, blocking olivocerebellar excitatory neurotransmission induced defects in the firing of developing cerebellar nuclear neurons. In P19-21 mice, we found that the cells in the Ptf1a Cre ;Vglut2 fx/fx mice fired at a higher rate than Vglut2 fx/fx mice, although there was no difference in the pattern of firing (Fig. 6d-f and Supplementary  Table 1). However, in contrast to the recovery of firing we observed in adult Purkinje cell simple spikes, the loss of climbing fibre input induced permanent defects in the frequency and pattern of firing of cerebellar nuclear neurons (cerebellar nuclei firing is also permanently altered after 3-acetylpyridine 25 ). Ptf1a Cre ;Vglut2 fx/fx cerebellar nuclei had a significantly lower firing frequency, higher CV and higher CV2 (Fig. 6g-i and Supplementary Table 1). The relatively high CV and CV2 indicated that the loss of climbing fibre neurotransmission caused highly irregular cerebellar output at the level of overall firing and spike-to-spike activity. The highly irregular output pattern in adult mutants was reminiscent of the irregular firing in the young mice of both genotypes (Fig. 6d-i), indicating that whereas the Vglut2 fx/fx cells became more regular over time, Ptf1a Cre ;Vglut2 fx/fx cells in juvenile mice fired erratically and remained so into adulthood. Analysis of VGAT and VGLUT2 expression revealed that these firing defects were not accompanied by obvious changes in the expression of inhibitory and excitatory presynaptic markers that label terminals in the cerebellar nuclei ( Supplementary Fig. 7a-d).
Abnormal cerebellar nuclei activity drives motor dysfunction.
Removal of the entire cerebellum in the dystonic rat, dt, eliminates the motor signs that are characteristic of the disease 12 .
Additionally, electrolytic and chemical lesions of the cerebellar nuclei significantly improve motor function in dt 12 . These studies imply that having no cerebellum is better than having a poorly functioning cerebellum. We reasoned that if abnormal cerebellar nuclei activity was driving the dystonia-like motor abnormalities in Ptf1a Cre ;Vglut2 fx/fx mice, then silencing of the  cerebellar nuclei output should improve motor function. To reversibly silence the cerebellar nuclei, action potential generation was blocked by infusing the voltage-gated sodium channel blocker lidocaine into the cerebellar nuclei of adult dystonic Ptf1a Cre ;Vglut2 fx/fx mice. Lidocaine effectively blocks cerebellar firing in vivo (see Discussion and Methods). We used tremor as a quantitative measure of abnormal behaviour because of its specific relevance to dystonia 21 . Tremor was measured before, during and after lidocaine infusion using osmotic pumps (Fig. 7a-d). The cannula of the pump was set to deliver 4% lidocaine bilaterally to the centre of the cerebellar nuclei (Fig. 7b,c). We measured the spread of lidocaine indirectly by examining methylene blue that was included in the infusion solution (Fig. 7b 0 ) or directly by using an anti-lidocaine antibody for immunodetection (Fig. 7b 00 ). The lidocaine remained mainly localized to the intended target, the interposed NS ** ** *** * * *** NS ** *** *   ARTICLE nuclei, with only limited spread into the adjacent fastigial and dentate nuclei. Dystonic tremor in Ptf1a Cre ;Vglut2 fx/fx mice was significantly reduced upon lidocaine infusion (Fig. 7e,f). The overall movement of the lidocaine-treated mutant mice improved ( Fig. 7d and Supplementary Movie 7). After the pump was depleted of lidocaine, tremor in the mutants returned to a level significantly different from that during lidocaine level (Fig. 7e,f). It is intriguing that infusion of saline into the mutants had a significant effect on tremor (Fig. 7f). However, at the end of saline infusion the tremor did not rise above the during treatment power, which is in contrast to the lidocaine condition, suggesting the basis for tremor reduction with saline infusion is the cannula lesion. Interestingly, lidocaine infusion lowered the physiological tremor in Vglut2 fx/fx mice, although the effect was not statistically significant (Fig. 7f). These findings suggested that the main neural deficit driving the dystonia-like motor abnormalities in the Ptf1a Cre ;Vglut2 fx/fx mice was erratic cerebellar nuclear firing. Furthermore, these data raised the possibility that modulating the abnormal output activity of the cerebellar nuclei could have therapeutic benefits.
Cerebellar DBS recovers mobility. DBS is a promising treatment for a number of motor diseases. There is particularly high optimism for its use in dystonia 31 . However, the use of DBS is restricted to targeting mainly the globus pallidus, thalamus and subthalamic nucleus. Based on data in the previous sections, we reasoned that directing DBS to the cerebellar nuclei would be expected to improve motor function. The general concept of DBS is that high-frequency stimulation modulates erroneous neural activity and entrains it to a pattern that normalizes behaviour. Although the exact mechanism(s) of DBS action is unclear 32 , one perspective is that the pulses produce inhibitory neuronal effects on somata that are proximal to the location of the electrode. The inhibitory action is thought to be the direct result of depolarization block through a mechanism involving sodium channel inactivation and potassium current potentiation. However, DBS might also increase and regularize the output of the stimulated region by activating local axons. At the network level, the end result is that the entrainment overrides pathological oscillatory activity. We postulated that the irregular firing of cerebellar nuclear neurons in the Ptf1a Cre ;Vglut2 fx/fx mice might be an ideal target for restoring movement with cerebellar DBS. To test this possibility, we implanted DBS electrodes bilaterally into adult dystonic Ptf1a Cre ;Vglut2 fx/fx mice and littermate controls (Fig. 8a,d,e,e 0 ). We targeted the interposed nuclei because they project extensively to regions that control ongoing movement, such as the red nucleus 33 . The interposed also projects to several regions of the thalamus 34 that project to the rest of the motor circuit. We tested three conditions: control implanted (stimulation), mutant implanted sham (no stimulation) and mutant implanted (stimulation). We used a 130 Hz biphasic stimulus (30 mA, 60 ms duration) that has previously been used to mimic human DBS parameters in mouse models of disease 35 . The response to DBS was immediate. Stimulated mutants demonstrated improved mobility. Using the dystonia rating scale, we found that the mutants receiving DBS showed decreased twisting and limb stiffness ( Fig. 8b and Supplementary Table 1). There was no significant difference between pre-DBS and post-DBS suggesting limited residual effects once the DBS was turned off. Moreover, based on repeated measures analysis of variance (ANOVA), there was no significant difference between days in the pre-DBS dystonia rating or during DBS measures (Fig. 8c). A one-way ANOVA showed a between-treatment difference based on averages of pre-DBS, during DBS (Fig. 8c) and post-DBS over 5 days of stimulation (Supplementary Movie 8).
The treated mutants groomed themselves, explored the enclosure and stood on their hind legs (Supplementary Movie 8). We did not observe cFos upregulation in the interposed cerebellar nuclear neurons near the site of stimulation (Fig. 8e 00 ), which is in contrast to the high cFos expression observed after pharmacologically activating the nuclei 36 . This implicates an inhibitory effect of DBS on the cerebellar nuclei somata, although it does not rule out a cooperative excitation of axons. The control mice given DBS into the interposed cerebellar nuclei did not develop any noticeable motor dysfunctions at any point during the experiment (Supplementary Movie 8).
The cerebellum and basal ganglia communicate using a disynaptic pathway involving the centrolateral nucleus of the thalamus 37 . This pathway is also responsible for relaying abnormal cerebellar output activity to the basal ganglia to cause dystonia. Given that the centrolateral nuclei are directly innervated by the cerebellar nuclei, we postulated that DBS directed to this thalamic region should also improve mobility ( Supplementary Fig. 8a,b). We found that bilateral stimulation of the centrolateral nuclei in Ptf1a Cre ;Vglut2 fx/fx mice significantly improved movement ( Supplementary Fig. 8c,d and Supplementary Movie 9).

Discussion
Neurological diseases are difficult to model because, in many cases, their manifestation includes wiring defects, firing abnormalities, plus neurodegeneration. This combination of disease aetiologies also makes it challenging to design effective therapies. To overcome these problems, we devised a mouse genetics approach to selectively silence fast chemical neurotransmission at a single type of connection in the brain-the olivocerebellar synapse. In doing so, we produced a severe motor condition that recapitulated multiple features that are observed in the debilitating forms of human dystonia. The mice exhibited stiff limbs, slow intense twisting, hyperextension of the back, limbs and toes (splayed) and sustained muscle contractions. With this model, we uncovered a circuit mechanism whereby transient defects in the structure and function of Purkinje cells plus an ongoing absence of climbing fibre complex spike firing were associated with permanent, erratic cerebellar nuclei activity, which was the major alteration driving multiple dystonia-like behaviours. We used the mice to find a new DBS target for rescuing motor behaviour.
Climbing fibre activity is thought to be a central requirement for cerebellar functions, ranging from ongoing motor behaviour 38 to learning 39,40 to error correction 41 . By extrapolation, climbing fibre function could be predicted to have a major impact on motor disease. For these reasons, there have been many attempts to manipulate climbing fibres. Climbing fibres have been experimentally targeted by lesioning the inferior olive with chemicals 42 or current 43 , cutting the axons in the inferior cerebellar peduncle 44 , electrically activating the fibres 45 , pharmacologically injecting toxins such as harmaline that synchronize inferior olive firing and increase climbing fibre activity 11 , genetically targeting genes that guide axon pathfinding 46 or innervation 47 , behavioural methods that eliminate the instructive value of the climbing fibre signal 48 and, more recently, by optogenetics approaches 49 . However, our intersectional genetics approach has the distinct advantage of silencing climbing fibres in all functional regions of the cerebellum, starting during development, and without causing major circuit rearrangements, axon regression or cell death. There is also the added advantage of being able to analyse the long-term effects of silencing an endogenous synaptic mechanism that revealed how the absence of proper climbing fibre excitatory neurotransmission can cause early-onset dystonia-like motor behaviour. The early postnatal movement defects are intriguing given recent work showing conditional deletion of Tor1a. Tor1a is a ubiquitously expressed AAA þ protein residing in the lumen of the endoplasmic reticulum/nuclear envelope space, the impairment of which is linked to the primary dystonia, DYT1. Deletion of Tors1a in the mouse brain causes some similar phenotypes to our mice, especially considering a loss of neurons in the cerebellar nuclei 50 . The olive and/or its downstream targets in the cerebellum, red nucleus or thalamus (and basal ganglia) may be central loci that are affected by dystonia mutations. Moreover, it is intriguing that loss of olivary input to the cerebellum resulted in an enhanced level of 10 Hz tremor. These data suggest that olive function normally serves to suppress pathological oscillations by maintaining 10 Hz oscillations at a subthreshold level that has been described extensively in vitro using extracellular 51 as well as patch-clamp recordings 52 and in vivo using whole-cell recordings 53 and local field potentials 52 .
Loss of climbing fibre complex spike activity in Purkinje cells altered the firing pattern of simple spikes, but this effect was only transient as adult Purkinje cells surprisingly resumed normal simple spike firing (Fig. 4d-h). Our results are consistent with previous work that showed that acute silencing of climbing fibre function, such as cooling the inferior olive 54 or acute chemical lesion of the inferior olive 55 , alters Purkinje cell firing. However, Purkinje cell simple spike firing returns to normal after long-term lesioning of the inferior olive 25 and after we silenced olivocerebellar axons. However, one major difference between our results and these previous manipulations is that in the adult mutant mice we did not find a change in simple spike firing, whereas they reported an increase in rate. We suspect that this difference is because our genetic manipulation targets developing Purkinje cells starting during embryogenesis, whereas other manipulations only affect adult Purkinje cells. The source of the difference could also be due to the manner in which simple spikes and complex spikes interact. In the adult, complex spikes are thought to directly suppress simple spikes 6,54 . A different relationship probably exists during development, as evidenced by the low firing rate of simple spikes in young mutant Purkinje cells that do not have complex spikes. Several points support this argument. First, the unique trajectories of simple spike and The loss of complex spikes in young Ptf1a Cre ;Vglut2 fx/fx mice would be predicted to substantially drive down Purkinje cell activity-the lower rate of simple spike activity during development increases the contribution of complex spike activity to the overall rate. In addition to the differences in the stages that were manipulated, it is possible that acute drug-and lesion-based approaches cause a fundamentally different cellular response compared to genetic silencing. In addition, our genetic approach specifically targets excitatory neurotransmission, whereas previous methods likely affect multiple (if not all) modes of cell-to-cell signalling.
Regardless of how the inferior olive is manipulated, a common outcome is that Purkinje cell activity recovers to normal, whereas cerebellar nuclear neuron firing remains abnormal 25 (Fig. 6g-i). There are several potential reasons for the persistence of cerebellar nuclei dysfunction, but first we will consider how the abnormal firing might be initiated. Neural tracing of axons and genetic marking demonstrated that Purkinje cell axons invade the territories of migrating cerebellar nuclear neurons as early as E15.5 in mice 57 . The early loss of glutamatergic signalling from the inferior olive to Purkinje cells in Ptf1a Cre ;Vglut2 fx/fx mice raises the interesting possibility that cerebellar nuclear neurons are abnormal starting from before birth. However, the neurofilament heavy (NFH)-positive excitatory cerebellar nuclear neurons in the mutants appear morphologically normal with only modest differences in the molecular profile of afferent termination throughout the nuclei (Supplementary Fig. 7a-d). However, the early deficit of olive input to the cerebellum could alter developmental gene expression and regulation, as well as the intrinsic firing properties in the cerebellar nuclei. Regardless of these changes and despite the recovery of Purkinje cells, complex spike activity is chronically absent in the Ptf1a Cre ;Vglut2 fx/fx mice. This is a critical point to consider because complex spikes can directly affect cerebellar nuclear neuron activity 7 . This in turn is important because complex spikes synchronize between Purkinje cells located in narrow bands of Purkinje cells 58 that converge and terminate within discrete populations of cerebellar nuclear neurons where they might entrain firing 8,59 . This precise organization of structure and function is thought to control different aspects of motor behaviour that are consistent with a role for complex spike dysfunction in dystonia 60 . Given the severity of dystonia and underlying cerebellar nuclei defects in the Ptf1a Cre ;Vglut2 fx/fx mice, it might be surprising that Purkinje cell function recovers. The cerebellar nuclei not only project out of the cerebellum but they also send inhibitory 61 and excitatory 62 feedback projections to the cerebellar cortex, where they synapse on granule cells and Golgi cells to indirectly affect Purkinje cell function. To counter this circuit, however, potentially destructive abnormal feedback from the cerebellar nuclei in Ptf1a Cre ;Vglut2 fx/fx mice could be dampened by homeostatic control of the Purkinje cells via their own local cortical feedback to granule cells 63 or directly to Purkinje cells 64 . Our awake recordings were intended to capture activity during ongoing and quiet behaviour without distinguishing the Purkinje cell dynamics during motion 65 or learning 62 .
It is interesting that the loss of complex spikes led to a delay in the structural maturation of Purkinje cell dendrites in early postnatal mice. The recovery of the Purkinje cell dendritic span before adulthood indicates a powerful intrinsic capacity for developmental plasticity in Purkinje cells, and it also suggests that afferent input controls the temporal dynamics of Purkinje cell development. Although the proper formation of the Purkinje cell dendrite and the assembly of the cortical microcircuit are required for proper function 57 , how the delay in the Ptf1a Cre ;Vglut2 fx/fx mutants affects behaviour is unclear. It is possible that abnormal Purkinje cell structure and the altered firing co-contribute to the early-onset twisting and hyperextension observed at P7, whereas the abnormal circuit firing alone maintains the dystonia throughout adulthood. However, it is important to note that all other aspects of adult cerebellar architecture and connectivity that we examined were also normal after olivocerebellar silencing. Putative inhibitory interneuron firing in the cerebellar cortex was unaltered, further suggesting that olive function plays a precisely timed and likely very specific role in establishing the wiring diagram of the cerebellum. The impact on cerebellar nuclear function was in stark contrast-as discussed above, their firing never recovered. The severe firing pattern abnormalities in the cerebellar nuclei of Ptf1a Cre ;Vglut2 fx/fx mice is consistent with a broader role for these cells in different dystonias 3,11,24,30 . If the abnormal cerebellar nuclei output is a major factor driving dystonia-like behaviour, then removing it should reverse these behaviours. To begin addressing this problem, we first asked whether blocking cerebellar nuclei output could improve motor function. We injected lidocaine into the cerebellar nuclei and indeed movement was improved. However, even though it prevents the transmission of neural activity at the soma, one caveat of the approach is that lidocaine can also block activity in axons passing near the nuclei 66,67 . However, the impact on incoming axons is minimized in the Ptf1a Cre ;Vglut2 fx/fx mice because the olivocerebellar fibres, which carry the most powerful input to the cerebellum, are already blocked genetically. Additionally, the other afferent system, mossy fibres, as well as Purkinje cell axons themselves all feed into the very cerebellar nuclei that lidocaine is intended to block. Other drugs such as the GABA-A receptor agonist muscimol could also be used to eliminate cerebellar output in Ptf1a Cre ;Vglut2 fx/fx mice. The inactivation would be different since lidocaine blocks additional cellular components. However, at the behavioural level, lidocaine and muscimol injections into the cerebellar nuclei produce similar consequences [67][68][69] . This equivalence likely exists because blocking the output of the nuclei determines the behavioural outcome, and therefore inhibiting passing axons has little consequence because they eventually feed into the nuclei that are blocked. Regardless, the lidocaine experiment results raised a therapeutic challenge. Which symptom is most problematic, and which one should (and can) be rescued to improve mobility?
Our model presented an opportunity to harness the endogenous potential of cerebellar circuits for functional brain repair of dystonia-like behaviours. We chose to use DBS because of its effectiveness in treating a wide range of neurological problems, including dystonia, Parkinson's disease, essential tremor, epilepsy, Tourette's syndrome and obsessive-compulsive disorder. Recent work even demonstrates improvement in cognitive function in a mouse model of Rett syndrome 35 . The improvement in mobility after DBS into the interposed cerebellar nuclei of the Ptf1a Cre ;Vglut2 fx/fx mice (Supplementary Movie 8) opens the possibility of targeting the cerebellar network in different human motor and nonmotor brain diseases.

Methods
Animals. Dr Chris Wright (Vanderbilt University School of Medicine) kindly provided the Ptf1a Cre mice 14 . We purchased the Vglut2 floxed (Vglut2 fx , #012898) mice 70 from The Jackson Laboratory (Bar Harbor, ME, USA) that contained LoxP sites surrounding exon 2 of the gene encoding vesicular glutamate transporter 2 (VGLUT2). Rosa lox-stop-lox-tdTomato mice were purchased from The Jackson Laboratory (#007908) by Dr Benjamin R. Arenkiel (Baylor College of Medicine) and then kindly provided to us. Neuropeptide Y (Npy-Gfp) BAC transgenic mice were also purchased from The Jackson Laboratory (#008321). We bred the mice using timed pregnancies, and we designated noon on the day a vaginal plug was detected as embryonic day (E) 0.5 and the day of birth as P0. For the developmental stages of the study, mice of ages P0, 7, 9, 14 and 19-21 were used. For mature and adult studies, mice from P30 to P180 were used. Mice of both sexes were studied. All animal studies were carried out under an approved institutional animal care and use committee animal protocol according to the institutional guidelines at the Baylor College of Medicine.
Tissue processing and anatomy. The mice were perfused with 1 M phosphate-buffered saline (PBS) and 4% paraformaldehyde (PFA) and the tissue cut on a cryostat or a microtome. Immunohistochemistry and in situ hybridization were carried out as described previously 28 . For in situ hybridization, mice were anaesthetized with isoflurane, brains rapidly removed from the skull and immersed in OCT (optimal cutting temperature), and immediately flash frozen by placing the tissue moulds into liquid nitrogen. Sagittal sections (25 mm) were cut through the cerebellum and the sectioned slices placed directly onto electrostatically coated glass slides (Probe On Plus Fisher Brand; Fisher Scientific). The tissue was probed with Vglut2 (SLC17A6) or Vglut1 (SLC17A7) digoxigenin-labelled mRNA probes using an automated in situ hybridization procedure termed Genepaint. All reagent incubations, washes and staining were performed by the in situ robot. The signal was detected by colourimetric detection using BCPI/NBT reagents. Upon completion of processing, the slides were removed from the machine and then cover-slipped with permanent mounting medium (Entellan mounting media (Electron Microscopy Sciences, Hatfield, PA, USA)). The mRNA in situ probes were designed against the sequences indicated below.
The tissue was examined using a JEOL 1010 microscope and the data imported into Adobe Photoshop CS5 for analysis.
Analysis of stained images. Photomicrographs of tissue sections were captured using Zeiss AxioCam MRm (fluorescence) and AxioCam MRc5 (DAB-reacted tissue sections) cameras mounted on a Zeiss Axio Imager.M2 microscope or on a Zeiss Axio Zoom.V16. Images of tissue sections were acquired and analysed using either Zeiss AxioVision software (release 4.8) or Zeiss ZEN software (2012 edition). After imaging, the raw data were imported into Adobe Photoshop CS5 and corrected for brightness and contrast levels. Schematics were drawn in Adobe Illustrator CS5. For the full sagittal image shown in Supplementary Fig. 1, 10 Â images were taken and then overlaid together in Adobe Photoshop CS5. Molecular layer thickness was determined by measuring in ImageJ the distance from the top of the Purkinje cell soma to the top of the Purkinje cell dendritic tree. Measurements were taken from four mice of each genotype, from each of the four cerebellar transverse zones. Age-matched, gender-matched mice of the same genotype were averaged and were compared using Student's t-test. ImageJ was used for counting for the recombination efficiency experiments. Recombination efficiency in the inferior olive was determined by driving expression of tdTomato in the olive using the Ptf1a Cre driver line. The olive was also stained for NeuN. The images were then separated by colour, thresholded and using the manual counting feature in ImageJ, the number of tdTomato-positive cells and the number of NeuN-positive cells were counted. A percentage was calculated using tdTomato-positive/NeuN-positive Â 100. For the analysis of protein puncta in the molecular layer and cerebellar nuclei, tissue sections were stained with calbindin (to label Purkinje cells) and VGLUT2, NFH (to label neurons of the cerebellar nuclei and determine the borders of the nuclei) and VGLUT2 or NFH and VGAT. Images were separated by colour, auto-contrasted, auto-thresholded, run through a built-in watershed and counted using the analysed particles feature. The number of puncta/area were averaged and compared between genotypes using Student's t-test. Fluorescence intensity values were acquired in ImageJ by finding the mean pixel intensity in a region of interest. For WGA-Alexa 555 (diluted to 1% in PBS pH 7.4; Thermo Fisher Scientific, Waltham, MA, USA) marking of the interposed cerebellar nuclei after recording, B10-20 nl of the tracer was injected and then the mice perfused (as described above) after recording. The tissue was cryoprotected, cut at 40 mm, mounted, and the fluorescence signal imaged directly. The Alexa 555 signal was overlaid on brightfield images of the same tissue.
Behavioural analyses. Rotarod performance was quantified by recording the latency to fall or to rotate 3 consecutive times on an accelerating rod (Ugo Basile, Varese, Italy). Mice were tested on the rotarod each day for 3 days with 3 trials separated by 15 min. Tremor amplitude and frequency were analysed on a single trial with a Tremor Monitor 28 (San Diego Instruments, San Diego, CA, USA). Using the provided software from SDI, the raw waveforms were passed through a fast Fourier transform to create a power spectrum graph. The power spectrum was then condensed to 1 Hz bins and averaged across all mice and graphed with error bars indicating s.e.m. For statistical analysis, the peak, centred at 9-13 Hz, was compared by Student's t-test. The open field activity assay was carried out using an Open Field Locomotion system (Omnitech Electronics, Inc., Columbus, OH, USA). Mice were acclimated to the Open Field room for 30 min with white noise provided. Mice were then placed in the Open Field apparatus and tracked for 30 min. Between-group differences were statistically evaluated by Student's t-test. Videos were taken using a Sony 9.2 megapixel HDR-PJ540 camera. Videos were then processed to demonstrate the motor behaviours with PlayMemories and iMovie software. Mice were weighed at several time points and averaged for each age, for each genotype. The data at each time point were gender-matched and then compared with an unpaired Student's t-test.
Lidocaine infusion. Mini-osmotic pumps (Alzet, Cupertino, CA, USA) were primed and filled with 4% lidocaine (Sigma-Aldrich, L7757) diluted in 0.9% saline with 0.01% methylene blue dye or just 0.9% saline with 0.01% methylene blue dye as a sham control. The pumps were connected to custom-built cannulas (Plastics One, Roanoke, VA, USA) that were targeted to release the lidocaine 71 , which blocks cerebellar firing in vivo 72 , to the interposed nuclei for 7 days. Tremor was measured before, during and after treatment (see below). Mice were perfused for anatomical examination shortly after the pump was depleted, but after the final tremor analysis was conducted to verify the surgical targeting of the cannulas. Intrasubject data were tested using a paired t-test and inter-subject data were tested using an unpaired t-test.
Deep brain stimulation. Twisted bipolar electrodes were fabricated out of 50 mm tungsten wire. The spacing between the two electrodes was calculated to bilaterally target the interposed cerebellar nuclei or the centrolateral nuclei of the thalamus. We used a Master 8 pulse generator and an Iso-Flex stimulus isolator (AMPI, Jerusalem, Israel) to deliver a biphasic 130-Hz stimulus with 60 ms duration at 30 mA for 1 h per day for 5 days. Mice were filmed throughout the process. The videos were cut into regular interval 1 min segments, randomized and then analysed in a blinded manner using a dystonia rating scale 3,73 before, during and after DBS. The 5 min of video was analysed for each period, with the NATURE COMMUNICATIONS | DOI: 10.1038/ncomms14912 ARTICLE NATURE COMMUNICATIONS | 8:14912 | DOI: 10.1038/ncomms14912 | www.nature.com/naturecommunications 'during DBS' period starting immediately after the stimulator was turned on and with the 'after/post-DBS' period starting immediately after the stimulator was shut off. Dystonia rating scale from 0 to 5 as in Pizoli et al. 73 was used wherein 0 ¼ no motor abnormalities; 1 ¼ slightly slowed or abnormal motor behaviour, no dystonia; 2 ¼ mild impairment, sometimes limited ambulation, dystonic postures when disturbed; 3 ¼ moderate impairment, frequent spontaneous dystonic postures; 4 ¼ severe impairment, sustained dystonic postures and limited ambulation; and 5 ¼ prolonged immobility in dystonic postures. Mutant mouse ratings were examined using a paired t-test or one-way repeated-measures ANOVA. Implanted but nonstimulated mice were used as sham controls. Mice were perfused after the final day of testing to examine the cerebellum for anatomical verification of surgical targeting of the electrodes. For the cFos experiments the mice were perfused with 4% PFA, as described above, 1 h after DBS.
In vivo electrophysiology. For the anaesthetized in vivo recordings of Purkinje cells, inhibitory interneurons and cerebellar nuclei neurons, the mice were given ketamine/dexmedetomidine (75 and 0.5 mg kg À 1 , respectively) and maintained with B0.25% isoflurane 28 . For head-fixed, awake recordings, mice were implanted with custom-made headplates and a craniotomy made above the cerebellum 56,74 . After 72 h of recovery, the mice were trained for 30 min per day in a head-fixed apparatus for 3 days before recording 56,74 . Cells were recorded and categorized based on standard stereotaxic coordinates measured from Bregma 75 . Control Purkinje cells were identified by the presence of complex spike activity. Analysis of complex spike percentage was taken from the awake recorded and juxtacellular labelled cells, as well from metal electrode recorded cells that were held briefly in anaesthetized mice. We examined the presence or absence of a complex spike for a total of 47 control Purkinje cells and 119 mutant Purkinje cells. For both awake and anaesthetized metal electrode recording procedures, single-unit recordings were attained with 5-8 MO tungsten electrodes (Thomas Recording, Germany) and then the signals digitized into Spike2 (CED, England) where single units were verified with principal component analysis. Analysis of continuous traces of 4300 s was performed with Spike2, MS Excel and MATLAB (Mathworks, Natick, MA, USA). We calculated the total number of simple and complex spikes over a predefined period of recording to obtain simple and complex spike firing frequency (Hz ¼ spikes s À 1 ). Firing pattern variability, or regularity, is defined as a measure of the consistency of time intervals between spikes (interspike time interval or ISI ¼ seconds). To quantify the average variability in firing pattern, CV of the ISI was calculated as the ratio of the s.d. of ISIs to the mean ISI of a given cell. We measured local regularity with CV2. CV2 measures firing pattern variability within a short period of two interspike intervals (CV2 ¼ 2|ISI n þ 1 -ISI n |/ (ISI n þ 1 þ ISI n ) 76 . Spike rates, CV and CV2 were computed and reported as mean ± s.e.m. Statistical analyses were performed with unpaired, two-tailed Student's t-tests or one-way ANOVA. Developmental Purkinje cell recordings were compared using nonparametric statistical analyses with the Kruskall-Wallis test, followed by Dunn's multiple comparison test for post hoc analyses between the two groups. Statistical significance is indicated in the graphs as *Po0.05, **Po0.01 or ***Po0.001. Juxtacellular recording. We labelled Purkinje cells in anaesthetized 3-5-monthold mice using 8-10 MO glass electrodes that were filled with 1% neurobiotin (Vector Laboratories Inc., Burlingame, CA, USA) diluted in 0.9% sterile saline. After identifying and then entraining a single unit, we performed a loose-patch fill 77 by delivering 1-5 nA of positive current for 45-60 min. The mice were then perfused and the neurobiotin labelling of single filled cells detected using the VECTASTAIN Elite ABC method (Vector Laboratories Inc.) and visualized with DAB.
ECoG and EMG recordings. For ECoG, 00-90 Â 1/16 flat point stainless steel machine screws (#B002SG89QQ) were implanted into the skull over the cerebellum (À 6.9 mm from Bregma at the midline) and cerebral cortex ( À 1 mm anterior-posterior from Bregma, 1 mm medial-lateral). For EMG, 76.2 mm silver wire was implanted into the gastrocnemius muscle and the coated portion of the wire directed below the skin into custom-built ports (also receives EEG connections; Pinnacle Technology, Lawrence, KS, USA) that were secured to the skull. We used a custom-built Pinnacle Technology ECoG plus EMG pre-amplifier to amplify the muscle and brain signals before further amplification with a Model 410 Brownlee amplifier. ECoG and EMG signals were recorded for 15 min in each mouse. The signals were digitized into Spike2. Kainic acid (Sigma-Aldrich, St Louis, MO, USA) was delivered once per hour at a dosage of 5 mg kg À 1 to induce seizure activity. The ECoG signal during visually apparent seizure activity was compared with the ECoG during either normal locomotion or dystonia, all normalized to the baseline (quiescent) ECoG signal.
Data availability. Data from the experiments presented in the current study are available from the corresponding author on request.