Pathophysiology of Dyt1 dystonia is mediated by spinal cord dysfunction

Dystonia, a neurological disorder defined by abnormal postures and disorganised movements, is considered to be a neural circuit disorder with dysfunction arising within and between multiple brain regions. Given that spinal circuits constitute the final pathway for motor control, we sought to determine their contribution to the movement disorder. Focusing on the most common inherited dystonia, DYT1-TOR1A, we confined a conditional knockout of Tor1a to the spinal cord and dorsal root ganglia (DRG) and found that these mice recapitulated the phenotype of the human condition, developing early onset generalised torsional dystonia. Physiologically, these mice bore the hallmark features of dystonia: spontaneous contractions at rest, excessive sustained contractions during voluntary movements including co-contractions of motor antagonists, and altered sensory-motor reflexes. Furthermore, spinal locomotor circuits were impaired. Together, these data challenge current understanding of dystonia, and lead to broader insights into spinal cord function and movement disorder pathophysiology.


Introduction 14
Neural circuits that control movement are distributed across the neuraxis and are comprised of 15 multiple interconnected loops involving the cerebral cortex 1 , basal ganglia 2,3 , thalamus 4 , 16 cerebellum 5 , brain stem 6 , and spinal cord 7 . While each of these loops has its own function, it is the 17 collaboration of the ensemble that ultimately produces functional movement and hence behaviour. 18 When dysfunction develops within or between these loops, movement disorders arise. 19 The dystonias comprise the third-most common of the movement disorders (after Parkinson's 20 disease and essential tremor), and are characterised by involuntary sustained muscle contractions 21 across multiple muscle groups, manifesting as abnormal posture and disorganised movements 8 . 22 The irregular muscle activity leading to these hallmark postures and movements bears 3 main 23 neurophysiological biosignatures: (a) spontaneous muscle contraction at rest 9 ; (b) excessive, 24 sustained contractions during voluntary movements often involving co-contractions of 25 antagonistic muscles, which may lead to pain in addition to dysfunctional movement 9,10 ; and (c) 26 altered involuntary sensory-motor reflexes 11,12 . 27 The first link between motor control, movement disorders, and the basal ganglia -a cluster of 28 subcortical nuclei -was drawn in the 1600s 13 . Thereafter, multiple movement disorders were 29 subsequently classified as basal ganglia syndromes throughout the 1800s and early 1900s, 30 including Parkinson's disease, Huntington's disease, and dystonia [14][15][16] . In dystonia, however, 31 limited pathology has been found in the basal ganglia in either humans 17 or animals 18 . Moreover, 32 there is a significant time delay in the development -and response to treatment -of dystonia when 33 the basal ganglia are involved. Specifically, there is a multi-month lag between acquired injury 34 (such as stroke) of the nuclei and development of dystonia 19 , and a similar lag between deep brain 35 stimulation of the basal ganglia and alleviation of symptoms 20 . Furthermore, not all basal ganglia 36 movement disorganisation 9 , we sought to determine whether spinal cord dysfunction could be 48 responsible for the pathogenesis of the clinical signs of dystonia. 49 In this study, we focused on the most prevalent genetic form of dystonia: early onset generalised 50 torsional dystonia, or DYT1-TOR1A, which is commonly caused by an in-frame deletion of 3 base 51 pairs in exon 5 of the TorsinA (TOR1A) gene 23 . We made a mouse model of DYT1-TOR1A 52 dystonia that confines Tor1a knockout to spinal cord and dorsal root ganglion neurons. We show 53 that these mice develop functional and physiological signs that mirror those seen in human DYT1-54 TOR1A dystonia, and use these mice to map spinal motor circuit dysfunction. Confining the 55 knockout to dorsal root ganglion neurons does not reproduce the phenotype. We conclude that 56 spinal-restricted knockout of Tor1a reproduces the pathophysiology of the human condition. This 57 knowledge opens up a new target -the spinal cord -for the development of strategies aimed at 58 treating DYT1-Tor1a dystonia. Our findings also highlight the importance of considering spinal 59 cord circuits in the pathophysiology of movement disorders. 60 11 frequency bands and plotted on circular graphs wherein 0° denotes in-phase synchrony and 180° 237 reflects out-of-phase alternation. 238 The classic locomotor profile of out-of-phase bursting activity between bilateral extensors, 239 bilateral flexors, and ipsilateral flexor-extensors 7 was observed in the littermate controls, with 240 phase data concentrated at 180° (Fig. 3P-R). For the most part, the burst coordination observed in 241 postnatal day 1 spinal Tor1a d-cko mice was broadly similar to littermate controls (Fig. S3L). 242 However, this normal bursting profile became disrupted at postnatal day 2-5 wherein there was a 243 predominant shift in the bursting activity towards in-phase synchrony ( Fig. 3P-R). Cross-root 244 coherence remained above 0.8 for all root pairs examined ( Fig. S3I-K), suggesting that disruption 245 to rhythmic bursting observed in one root was largely related or predictive of the disrupted bursting 246 activity observed in the other root. Together, these data reveal that Tor1a-deleted spinal circuits 247 directly produce excessive spontaneous activity at rest and disorganised motor output during 248 locomotion. 249

Spinal monosynaptic reflexes are impaired in spinal Tor1a conditional knockouts 250
Given the spinal locomotor circuit dysfunction, we next looked at the most basic spinal circuit, 251 one that can also be readily studied in humans: the monosynaptic (myotatic) reflex. Case reports 252 indicate that individuals with generalised dystonia, including genetically-confirmed DYT1-253 TOR1A, show diminished monosynaptic reflex amplitudes 11,39 , increased variability in the evoked 254 response amplitude 11,12 , and the infiltration of aberrant asynchronous activity 11 . We thus 255 systematically assessed the monosynaptic reflex across space (L1-L5) and time (P7-P13, an age 256 range where the phenotype is fully penetrant) in spinal Tor1a d-cko mice. 257 Graded stimuli were applied to the dorsal root and the evoked monosynaptic reflex was recorded 258 from the ventral root (Fig. 4A, bottom). Plotting representative monosynaptic reflexes revealed a 259 spatiotemporal pattern parallel to the dystonic-like phenotype. Compared to age-matched 260 littermate controls, spinal reflexes in Tor1a d-cko mice were reduced in the caudal-most root at 261 postnatal day 7, but otherwise broadly normal in the rostral roots (L1, L3) ( asynchronous peaks (Fig. S4A-E). In addition to impairments in the reflex response itself, the 267 latency to onset -a measure largely dependent on afferent conduction -was significantly increased 268 in spinal Tor1a d-cko mice compared to littermate controls ( Fig. 4G-I; Fig. S4H). Together, these 269 data suggest that there is a clear caudal-to-rostral progression in monosynaptic spinal reflex 270 impairments during postnatal maturation, with a dispersion of the reflex across time. 271 Spinal Tor1a deletion leads to distributed pathophysiology in the monosynaptic reflex 272 We next sought to gain mechanistic insights into reflex dysfunction by interrogating the four 273 constituent components of this reflex arc: (a) proprioceptive afferents in the dorsal roots, (b) 274 synapses with motoneurons, (c) the motoneurons themselves, and (d) efferent transmission in the 275 ventral root. We focused on the caudal lumbar motor pools (L4-L5) as they are the earliest affected. 276 We used a ventral horn ablated preparation (Fig. 5A, left) to determine whether motoneurons were 277 intrinsically affected by the spinal-restricted Tor1a deletion. Motoneurons in spinal Tor1a d-cko 278 mice appeared smaller than those in littermate controls (Fig. 5B)

. While motoneurons in spinal 279
Tor1a d-cko mice had similar resting membrane potentials as control motoneurons (Fig. 5C), there 280 was a significant reduction in whole cell capacitance (Fig. 5D) and a ~350% increase in input 281 resistance (Fig. 5E), consistent with the smaller cell size. Together, these data indicate that lumbar 282 motoneurons are directly affected by Tor1a deletion. 283 But smaller motoneurons alone could not explain all the changes in the monosynaptic reflex, so 284 we next focussed on the afferent limb of the reflex using low-threshold stimulation of the dorsal 285 roots. Compared to littermate controls, afferent-evoked excitatory post-synaptic currents (EPSCs) 286 in spinal Tor1a d-cko mice had reduced amplitude, prolonged duration, and multiple asynchronous 287 peaks (Fig. 5F) -outcomes that were corroborated when we activated a subset of afferent fibres 288 via discrete microstimulations ( Fig. S5C-D). At all ages assessed, the EPSC area -a measure of 289 charge carried -was decreased in motoneurons in spinal Tor1a d-cko mice compared to controls 290 ( Fig. 5G; Fig. S5A). There was a decrease in EPSC conductance (Fig. 5H), but when scaled to 291 input conductance there was no difference (Fig. 5I), suggesting that the monosynaptic effects of 292 afferent inputs to motoneurons are similar for littermate control and spinal Tor1a d-cko mice. Of 293 note, at all ages tested, there was a significant increase in the latency to dorsal root-evoked 294 13 monosynaptic EPSC in spinal Tor1a d-cko mice as compared to controls ( Fig. 5J; Fig. S5B), a 295 finding that parallels the increased latencies observed in extracellular recordings. 296 While the longer latencies could result from impairments at the Ia-motoneuron synapse, they could 297 also simply be due to deficits in afferent conduction itself. Thus, we recorded L4 and L5 dorsal 298 root volleys in response to root stimulation (Fig. 5K), and found slower afferent conduction 299 velocities in spinal Tor1a d-cko mice compared to controls (Fig. 5L-M), suggesting that the longer 300 latencies to EPSCs resulted from slower conduction velocities. But increased latencies alone 301 cannot account for the asynchronous peaks observed in EPSCs. To this end, we microstimulated 302 the dorsal roots at various sites, activating small subsets of fibres ( Fig. 5N-O). After scaling the 303 conduction time by distance, we discovered that spinal Tor1a d-cko mice showed longer and 304 variable conduction times as compared to controls (Fig. 5P), suggesting that the multiple peaks in 305 the EPSCs result from time dispersion of the incoming afferent action potentials. That is, two 306 effects occur in the dorsal roots of the spinal Tor1a d-cko mice: slower conduction velocities and 307 increased variance of these velocities across fibres. 308 Given the conduction impairments in dorsal roots, we turned to the ventral roots to determine 309 whether motor axons are also affected (Fig. S5E). We found a modest, but significant decrease in 310 efferent conduction velocity in spinal Tor1a d-cko mice as compared to controls ( Fig. S5F-G). 311 Responses to microstimulation also revealed increased scaled conduction times and variances in 312 the ventral roots ( Fig. S5H-I). In summary, all compartments of the monosynaptic reflex arc -from 313 action potential conduction of sensory afferents to motoneurons to efferent output in the motor 314 roots themselves -are vulnerable to Tor1a dysfunction and contribute to impaired sensory-motor 315 integration in Tor1a d-cko mice. 316

Discussion 317
Dysfunctional spinal circuits: a neural substrate for early onset generalised torsional 318 dystonia 319 We have uncovered that spinal circuit dysfunction is a key contributor to the pathophysiology of 320 DYT1-TOR1A dystonia. By confining a Tor1a gene mutation to the spinal cord and dorsal root 321 ganglion neurons while leaving expression in the brain normal, mice phenotypically express a 322 generalised torsional dystonia, have an ultrastructural signature indicating loss of function of 323 torsinA 26,29,40 (with unknown relevance to altered neurotransmission, abnormal muscle 324 contractions, and disorganised movements) in spinal but not brain neurons, have spinal locomotor 325 circuit dysfunction, and have abnormal monosynaptic sensorimotor reflexes. 326 Co-existing with the motor impairments were signs of sensory dysfunction, with increased 327 variance in the conduction velocities of the fastest dorsal root fibers. While not a major feature of 328 Physiologically, we show that spinal Tor1a d-cko mice bear the three primary pathophysiological 359 signatures of DYT1-TOR1A dystonia: (a) spontaneous muscle contractions at rest 9 , (b) excessive, 360 sustained contractions during voluntary movements 9,10 , and (c) altered sensory-motor reflexes 11,12 . 361 To date, there has been limited study of the pathological mechanisms underlying these signatures. 362 Equipped with a fully-penetrant model that consistently and reproducibly develops dystonia and a 363 suite of spinal cord preparations to probe sensory-motor dysfunction, we systematically 364 interrogated the precipitating pathophysiological changes of early onset generalised torsional 365 dystonia. Recording from isolated hindlimb motoneuron pools revealed that excessive 366 spontaneous muscle contractions -including co-activation of motor antagonists -can be directly 367 produced by dysfunctional spinal circuits. Much like the in vivo phenotype, there are clear caudo-368 rostral generalisations in spinal circuit dysfunction over postnatal development. 369 In summary, we identified a dysfunctional neural substrate that phenotypically and physiologically 370 recapitulates early onset generalised torsional dystonia: spinal cord circuits. 371

Revisiting evidence in support of spinal circuit dysfunction in DYT1-TOR1A 372
The spinal cord is comprised of neural circuits that control the basic syllables of movement -e.g. 373 reciprocal inhibition to change a joint angle, co-excitation of flexor and extensor motor neurons to 374 stabilise a joint, and co-inhibition of these motoneurons to allow the joint to move freely in 375 16 biomechanical space 53 . These syllables are concatenated across time to form functional 376 movement 22 . In dystonia, there is abnormal control of these fundamental syllables, akin to a 377 paraphasia of movement. Thus, there was clear logic in considering that spinal circuit dysfunction 378 leads to the signs of DYT1-TOR1A. 379 In fact, there are seeds of substantiation sprinkled through the literature that point to spinal circuit 380 dysfunction. For example, Dyt1-Tor1a animal studies have shown nuclear envelope 381 malformations in spinal neurons 26,29 , spinal motoneuron loss 27 , and reduced spinal GABAergic 382 inputs to primary afferent fibres 54 . In non-Dyt1 dystonia models, Lamb1t mice have coincident 383 EMG activity between opposing muscles, a phenomenon that persists post-spinal transection and 384 thus directly implicates dysfunctional spinal circuits 55 . And in people affected by DYT1-TOR1A, 385 analyses of spinal reflexes indicate that dystonic individuals may have impairments in 386 monosynaptic stretch reflexes 11,12 and reciprocal inhibition 9,56-58 . Even though these reflexes are 387 mediated by spinal circuits, the impairments observed have been attributed to dysfunction of 388 descending systems 56,59 . But spinal circuits are complex, and form specialised, multi-layered 389 networks that integrate supraspinal, spinal, and sensory inputs to organise motor output 60 . Thus, in 390 dystonia pathophysiology, it is logical to consider spinal circuits as a critical nexus for neurological 391 dysfunction and movement disorganisation in dystonia. 392 Reconciling spinal circuit dysfunction with prevailing models of dystonia: movement 393 (dis)organisation at many levels 394 We have shown that spinal circuit dysfunction can recapitulate one of the most severe forms of 395 primary dystonia. That is, in the homozygous condition, descending command signals cannot 396 override or compensate for spinal circuit dysfunction such that generalised torsional dystonia 397 manifests over postnatal time. Yet one of the most effective treatment options for DYT1-TOR1A 398 is deep brain stimulation of a site in the basal ganglia, the globus pallidum interna 61 . If spinal 399 circuit dysfunction leads to disorganised movements, then why is DBS an effective treatment for 400 dystonia? Testing DBS in spinal Tor1a d-cko mice is technically unfeasible, due to the 401 combination of rapid onset and progression of motor signs in pre-weaned, undersized pups, the 402 size of the necessary hardware, and the expected duration of stimulation needed for alleviation of 403 dystonic signs. As such, addressing how DBS might be effective if the key pathophysiology is 404 spinal requires revisiting prevailing models of dystonia in the context of motor control at large. 405 Compared to other DBS-treated movement disorders such as essential tremor or Parkinson's 406 disease, wherein stimulation offers rapid symptom relief within seconds to hours 62 , many weeks 407 of continuous stimulation is required before tonic dystonic movements show improvement 61 . This 408 delay to symptom amelioration directly implicates neuroplastic mechanisms: a long-term process 409 with adaptive effects that can be localised or distributed via interconnected circuits 62 . In fact, 410 maladaptive neuroplasticity is a widely recognised contributing factor to dystonia 63 with mis-wired 411 circuitry implicated in the local motor planning ensemble (basal ganglia loops) 64  To conclude, with this model of DYT1-TOR1A dystonia, we have a newfound entry point into 429 investigating the complex pathophysiology of the disease. As a circuitopathy, dystonia can be 430 considered as a process that affects motor circuits throughout the central nervous system, including 431 those in the spinal cord. The notion that spinal motor circuits are simple relays between the brain 432 and muscles has long been dispelled. Yet spinal circuit dysfunction is rarely considered in 433 movement disorder pathophysiology. We would suggest that new treatment strategies for DYT1-434 TOR1a dystonia could be aimed at addressing the pathophysiology underlying symptoms, the 435 circuits of which are largely resident in the spinal cord. 436

Acknowledgements: 437
We are grateful to Nadine Simons-Weidenmaier, Rafaela Fernandez De La Fuente, and Hrista 438 Micheva for assisting in animal husbandry and beta-testing  Table S1. See also 486 Figure S1. 487  Table S1. See also Figure S2 and Movies 1-7. 505  Table S1. See also Figure S3. 520  Table S1. See also Figure S4. 529  Table S1. See also Figure S5. 552 All experiments were performed in pre-weaned male and female mice with date of birth recorded 566 as postnatal day ("P") 0, or P0. Twenty independent breeding pairs were used to generate all 567 offspring for this study, of which all pups were arbitrarily allocated to the different batch 568 experiments. For longitudinal experiments (e.g. video recordings), pups were randomly selected 569 from the litter for testing. For batch experiments that use the full litter (e.g. electrophysiology), 570 pups were arbitrarily selected from the litter on a day-by-day basis, randomizing the age at which 571 pups were allocated to electrophysiology testing. Experiments were performed while blinded to 572 genotypes, where possible. In the event blinding was impossible at point of data collection (e.g. 573 overt phenotype), data were collected and coded post hoc for subsequent blinded analysis. Per 574 experimental design, no spinal Tor1a d-cko animals were weaned during this study. Throughout 575 the course of all experiments, the welfare of spinal Tor1a d-cko mice was monitored. Spinal Tor1a 576 d-cko mice animals could eat (e.g. milk spots), were active, well-groomed, and did not lose weight 577 through the early post-natal period, with most gaining weight (up to ~6g by P19). The most severe 578 phenotype is that shown in Video S6. vector, the homology arms and conditional knockout region (Tor1a exons 3-5) were generated by 597 PCR using BAC clones RP24-231D2 and RP24-170N21 from the C57Bl/6 library. The targeting 598 vector contained a Neo cassette flanked by loxP sites with DTA used for negative selection. Of 599 the 24 positive embryonic stem cell clones detected with PCR, the 1H5 clone was used to establish 600 the line. Embryonic stem cell injections were performed using C57Bl/6J albino embryos that were 601 then implanted into pseudo-pregnant CD1 females. Founder offspring were identified by their coat 602 colour and germline transmission was subsequently confirmed via breeding with C57Bl/6J females 603 and confirmatory genotyping of the resultant F1 offspring. F1 stocks were transferred to UCL and 604 then bred with C57Bl/6J mice (Charles River) to yield F2 stock with the conditional-ready allele 605 established in males and females. Thereafter, the congenic Tor1a-frt colony was maintained in the 606 C57Bl/6J background using heterozygous Tor1a wt/frt males. Tor1a-frt mice will be made available 607 through the Mutant Mouse Resource and Research Centre (MMRRC). 608 The following primers were used to detect the pre-Flp conditional-ready allele: 609 The reaction mix contained the following (total 25 µl): 1.5 µl DNA, 1.0 µl of 10 µM forward 612 primer, 1.0 µl of 10 µM reverse primer, 12.5 µl premix Taq polymerase, and 9.0 µl double distilled 613 H2O. The cycling conditions were as follows (35 cycles): initial denaturation at 94°C for 3 minutes, 614 26 denaturation at 94°C for 30 seconds, annealing at 62°C for 35 seconds, extension at 72°C for 35 615 seconds, and an additional extension step at 72°C for 5 minutes. Genotypes were identified based 616 on the following band sizes (bp): wildtype allele=161, conditional-ready heterozygotes=161 and 617 306, and conditional-ready homozygotes=308. 618 Experimental breeding. Heterozygous Tor1a wt/frt males and Tor1a wt/frt females were bred to 619 produce homozygous Tor1a frt/frt offspring, which were subsequently used to maintain a 620 homozygous line with inbreeding not exceeding 6 filial generations. Homozygous Tor1a frt/frt mice 621 were then used in the following multigenerational breeding strategy to produce offspring with 622 spinal and DRG-restricted biallelic "double" conditional knockout of Tor1a exons 3-5 (spinal 623  Table S1. 643 qPCR 644 In total, N=14 P18 mice (7 littermate control 7 spinal Tor1a d-cko mice) were deeply anaesthetised 645 via i.p. injection of ketamine:xylazine (300 mg/kg : 30 mg/kg; 0.01 ml/g body weight) and 646 decapitated. Whole brains (with cerebellum attached), lumbar spinal cords, dorsal root ganglia 647 (thoracic-lumbar), hearts, and livers were rapidly harvested, snap frozen in liquid nitrogen, and 648 transferred to -80ºC until processing by experimenters blinded to the genotype. 649 Samples were weighed, digested in a 10% w/v solution of Qiazol lysis buffer (Qiagen RNeasy 650 Lipid Tissue Mini Kit, ThermoFisher 74804), and homogenised using the Ika Homogeniser 651 Workstation with disposable probes. 200 ml of chloroform was added to each sample and mixed 652 vigorously. Samples were then centrifuged at 12,000 G for 15 minutes at 4ºC. The top aqueous 653 layer was transferred into a cold falcon tube and an equal volume of chilled 70% ethanol was 654 added. RNA extraction was then performed following manufacturer protocol. A total of 2 µg of 655 total RNA was used for cDNA synthesis (ThermoFisher High-Capacity RNA-to-cDNA kit, 656 4387406) using the following thermocycler conditions: 37ºC (60 minutes), 95ºC (5 minutes), and 657 hold at 4ºC until plate was stored at -20ºC. 658 Separate TaqMan master mixes (ThermoFisher 4370048) were created for each probe: Tor1a 659 (ThermoFisher 4331182, Mm00520052_m1) and reference genes Actb (ThermoFisher 4331182, 660 Mm00607939_s1), Gapdh (ThermoFisher 4331182, Mm99999915_g1), and Hprt (ThermoFisher 661 4331182, Mm00446968_m1). Samples were run in triplicate using the ThermoFisher qPCR 662 QuantStudio. Tor1a fold gene expression values were estimated using ∆∆Ct method with Hprt or 663 Actb serving as the primary housekeeping gene. 664 The same methods were used for quantifying Tor1a expression in the spinal cords and isolated 665 DRGs of N = 6 P58-P59 littermate control (Avil wt/wt ;Tor1a wt/flox ) and N = 5 DRG Tor1a d-cko 666 mice (Avil wt/cre ;Tor1a flox/flox ). Two outliers -one from each group -were excluded from analysis 667 due to potential tissue contamination issues. 668 Western blots 669 N=7 P18 mice were deeply anaesthetised via i.p. injection of ketamine:xylazine (300 mg/kg:30 670 mg/kg; 0.01 ml/g body weight) and decapitated. Whole brains (with cerebellum attached) and 671 lumbar spinal cords were rapidly dissected from each animal, snap frozen in liquid nitrogen, and 672 transferred to -80ºC until protein extraction. Samples were homogenised on ice in 10% w/v 673 radioimmunoprecipitation assay (RIPA) buffer, then agitated on a rotator at 4ºC for 30 minutes. TBST, 7% non-fat dry milk, 0.1% sodium azide). The following day, membranes were washed 692 with 0.1% TBST (5 washes at RT, each 5 minutes, gentle agitation), incubated with goat anti-693 rabbit horseradish peroxidase (HRP)-conjugated species anti-rabbit secondary antibody at 1:1,000 694 (same solution as primary, RT for 60 minutes, gentle agitation) (Bio-Rad 1706515; RRID: 695 AB_11125142), then washed with 0.1% TBST (6 washes at RT, each 5 minutes, gentle agitation). 696 Immunoreactive bands were detected using HRP substrate (Millipore WBLUC0500) and imaged 697 using the Chemidoc MP Imaging system (Bio-Rad 17001402). Band density was quantified using 698 the Bio-Rad Image Lab software (v6.0.1). Molecular weights were estimated using the Precision 699 Plus Protein Kaleidoscope Standard (Bio-Rad 161-0375). After chemiluminescent imaging, 700 membranes were stained with Coomassie blue R-250 dye to estimate total protein 71 . Using 701 Coomassie blue as a loading control, torsinA blot images were analysed and formatted using Image 702 Lab software. 703

Ultrastructure 704
In total, N=8 P18 mice (4 littermate control 4 spinal Tor1a d-cko mice) were deeply anaesthetised 705 via i.p. injection of ketamine:xylazine (300 mg/kg:30 mg/kg; 0.01 ml/g body weight) and 706 transcardially perfused with room temperature 0.1 M PBS (ThermoFisher 70011036) followed by 707 EM-grade 4% formaldehyde (TAAB F003). The skull and vertebral column were harvested and 708 stored in 4% PFA for approximately 3-4 weeks before brains, lumbar spinal cords, and dorsal root 709 ganglia were subsequently dissected. 710 Basal ganglia. Whole brains were embedded in 4% agarose (Sigma-Aldrich A4018) and then 711 sectioned coronally using a vibrating microtome (Leica Biosystems VT1200S). Initial sectioning 712 settings of 0.30 mm/s speed, 1 mm amplitude, and 300 µm thickness were used to quickly cut 713 through the brain until the anterior commissure appeared. Thereafter, the cutting thickness was 714 decreased to 200 µm and serial cross-sections were collected until the hippocampal formation 715 became prominent. Slices were stored in a 24-well petri dish filled with chilled 0.1M PBS during 716 collection and then transferred into a 0.1M sodium cacodylate solution containing EM-grade 2% 717 formaldehyde/1.5% glutaraldehyde (TAAB G011) and stored at 4°C overnight. Images of 718 vibratome slices were captured using a stereo microscope to identify which cross-sections 719 contained the globus pallidus and striatum. The globus pallidus and striatum were then bilaterally 720 dissected from two vibratome cross-sections per animal, yielding N=4 globus pallidus and striatum 721 "blocks" per animal, respectively. 722 Samples were prepared for electron microscopy following a modified protocol 72 . Samples were 723 washed in 0.1M sodium cacodylate buffer and post-fixed in 1% OsO4/1.5% potassium 724 ferricyanide for 60 minutes at 4°C. Samples were then washed in distilled H2O and treated with 725 1% thiocarbohydrazide (TCH) for 20 minutes at RT, 2% osmium (OsO4) for 30 minutes at RT, 726 1% uranyl acetate (UA) overnight at 4°C, and lead aspartate for 30 minutes at 60°C, with 727 intermediate washing in distilled H2O between each step. This was followed by dehydration of the 728 samples via increasing concentrations of ethanol: 70%, 90%, and two rounds of 100%. Samples 729 were then infiltrated with 1:1 propylene oxide (PO):epon resin for 60 minutes followed by two 730 rounds of fresh 100% epon for 60-120 minutes before being mounted and polymerised overnight 731 at 60°C. Brain slices were mounted on the flattened top of a pre-polymerised epon stub under a 732 coverslip to maintain the flatness of the tissue/block surface, spinal cord samples were mounted in 733 moulds so that the surface to be sectioned was parallel and adjacent to the final block surface. 734 Lumbar spinal cord and dorsal root ganglia. Following dissection from the vertebral column, the 735 L4-L5 lumbar spinal cord segments were identified and the corresponding dorsal root ganglia 736 isolated. Thereafter, each spinal cord was transversely cut to yield N=3, 1 mm blocks (a rostral, 737 central, and caudal block). The rostral and central blocks were treated as described above. All 738 samples were dehydrated via 10-minute incubations of an increasing ethanol gradient: 50%, 70%, 739 80%, and 100% (x2). Samples were embedded as described above in PO (10 minutes  Videos were collected by experimenters blinded to the genotypes from N=4 independent breeding 775 pairs, N=14 mice in total (N=8 littermate control and N=6 spinal Tor1a d-cko mice, respectively). 776 In parallel, age-matched video recordings were performed using the control offspring derived from 777 four different dystonia-related strains in the lab (N=16 mice). These videos served as controls for 778 the training dataset for normal postnatal sensory-motor development in mice. 779 Recordings started at P1-P3 and continued every other day until postnatal day P14-P15, the pre-780 juvenile stage age at which the primitive reflexes have disappeared and the adult-like sensory-781 motor behaviours are refined 73 . Prior to weight assessments and video recordings, the parents were 782 transferred to a separate cage while the home cage with pups was placed on an external heating 783 pad. Individual pups were transferred to the clear plexiglass dish of a weight scale and videos were 784 recorded at 30 fps, 1080p using 8 MP camera (f/2.2 aperture, 1.5µm pixel size, 1/3-inch sensor). 785 Approximately 60-90 seconds of video were recorded per pup. Pups were given unique paw tattoos 786 32 using a hypodermic needle and Ketchum green tattoo ink (F.S.T. 24201-01) following their first 787 recording session to facilitate longitudinal recordings. For each pup, a representative ~30 second 788 segment of video was selected and formatted for external, unbiased peer review. 789 Five external raters who had previous experience with mouse behaviour were selected to provide 790 unbiased phenotype scoring of littermate control and spinal Tor1a d-cko mice. 791 Raters were informed that they would review postnatal videos of a "new mouse model for a 792 movement disorder" and that they would assign the mouse -via a unidirectional online test -to 793 "control or mutant groups." If mutant was selected, then raters would "check off which body 794 region(s) are affected." Thus, raters were blinded to the model (dystonia), the mutation (spinal-795 restricted conditional knockout), the genotype (littermate vs spinal Tor1a d-cko mice), when the 796 phenotype was expected to manifest (postnatal timeline), where signs might emerge along the body 797 axis (e.g. head vs hindlimbs), and how the signs might manifest (e.g. hyperextension, hyperflexion, 798 torsion). Prior to scoring, raters had the option to review a set of annotated training videos that 799 delineated the normal functional milestones typically observed during postnatal maturation in non-800 disabled pups. The sensitivity of the postnatal testing paradigm was defined as the proportion of 801 spinal Tor1a d-cko observations that were accurately classified as "mutant" (N=55 and 115 true 802 positives at P1-P6 and P7-P13, respectively) divided by the total number of true spinal Tor1a d-803 cko observations (P1-P6: N=55 true positives + N=20 false negatives; P7-P13: 115 true positives 804 + N=0 false negatives). The specificity was defined as the proportion of control observations that 805 were correctly classified as "control" (N=101 and N=137 true negatives at P1-P6 and P7-P13, 806 respectively) divided by the total number of control observations (P1-P6: N=101 true negatives + 807 N=14 false positives; P7-P13: N=137 true negatives + N=3 false positives). 808 A similar recording strategy was used for the DRG Tor1a d-cko cohort. In total, N=5 littermate 809 control and N=5 DRG Tor1a d-cko mice generated from two independent breeding pairs were 810 recorded from postnatal day 1-9, an age range that reflects the development of stepping as well as 811 the complete onset and progression of early onset generalised dystonia in the spinal Tor1a  Preparation. For each animal, five ~10 cm long pieces of A-M Systems Wire (#793200) were cut 819 to fabricate two sets of bipolar recording electrodes (gastrocnemius and tibialis anterior, 820 respectively) and one monopolar grounding electrode (base of tail). At one end of each wire, 821 approximately 1 mm of coating was removed to expose the underlying wire. At the other end of 822 each wire, approximately ~1-2 mm of wire coating was removed and then inserted into a 30-gauge 823 hypodermic needle extracted from its plastic hub. The needle was then crimped 3-5 times to ensure 824 stable contact between exposed wire and needle shaft. Colour coded heat shrink was attached to 825 the base of the needle to facilitate electrode identification during the subsequent implantation and 826 recording procedure. 827 The limited size and mobility of spinal Tor1a d-cko mice permitted an acute preparation for 828 recording hindlimb EMG activity in pre-weaned mice. N=2 P17 and N=4 P19 spinal Tor1a d-cko 829 mice were briefly anaesthetized using isoflurane and ophthalmic ointment was applied to the eyes. 830 The fur overlying the right hindlimb and posterior half of the dorsum was shaved and small 831 incisions were made at the mid-back and at the level of the hip. Underlying adherent tissue between 832 the incision sites was bluntly dissected to create an open channel for routing the recording wires. 833 The 1 mm bared ends of the wires were bent into hook-like shapes using the beveled edge of a 30-834 gauge needle and then two each were percutaneously implanted into right hindlimb gastrocnemius 835 and tibialis anterior. A monopolar ground electrode was inserted near the base of the tail. The hip 836 incision was loosely sutured closed leaving some excess wire external to the animal to provide 837 sufficient slack for accommodating behavioural activity. After suturing the incisions, the animal 838 was transferred to an open field arena and kept warm with external heating pads until fully alert. 839 N=4 P18 C57Bl/6J wildtype mice of both sexes were briefly anesthetized with isoflurane until the 840 paw reflex was abolished. While mice were kept warm, the left or right hindlimb was shaved and 841 skin palpated to identify the underlying medial gastrocnemius and tibialis anterior muscles. Bipolar 842 recording electrode wires (A-M Systems Wire #793200) were inserted through the skin and into 843 the belly of the muscle. Anaesthesia was then removed and the mice were transferred to an empty 844 34 cage enclosed within a custom-made Faraday cage. EMG recordings were performed once the 845 animal recovered from acute anaesthesia. Bouts of stepping and quiet rest were flagged as 846 described below. 847 Recordings. Recordings were made using a custom-made amplifier and headstage. Signals were 848 amplified (100x), bandpass filtered (30 Hz -10 kHz), and digitized (10-20 kHz) using a CED 849 Power3a running Spike2 v9 software, and saved for offline analysis. Concomitant videos were 850 also recorded for a subset of animals using a Basler camera (acA800-510uc) with a Fujinon lens 851 (DF6HA-1B; 1:1.2/6mm). Videos were recorded at 104 fps, 0.5 megapixel and synched with EMG 852 recordings using manual and software cues interspersed throughout the entire recording session. 853 The duration of recordings ranged from 30-45 minutes and included the following behavioural 854 conditions: recovery from anaesthesia, alert and at-rest, locomotion, and tail suspension. Following 855 terminal recordings, animals were humanely culled and ear biopsies collected for post hoc 856 genotyping. 857 Analysis. Using contemporaneous notes, datafile keymarks, and videos as reference, EMG data 858 were segmented into four conditions: at-rest (for baseline EMG thresholding), tremulous-like 859 phenotype at-rest, locomotion, and tail suspension. Noise contamination was removed from the 860 segmented dataset using Spike2 native and custom-written scripts. Contaminants such as glitches 861 ("FixGlitch" script), mains noise (50 Hz notch filter with "HumRemoval" script for 50 Hz 862 harmonics), or motion artefacts (25-50 Hz digital high-pass filter) were eliminated as needed. The 863 resultant clean EMG signals are shown. For locomotor analysis, the cleaned signal was full-wave 864 rectified and subsequently analysed as follows. 865 For each animal, we first calculated the average at-rest baseline EMG activity level. For subsequent 866 detection of the locomotor-related bursts, we set a minimum rising (burst onset) and falling (offset) 867 threshold crossing to baseline average ± 10 S.D. or average ± 20 S.D., depending on the extent of 868 excessive spontaneous EMG activity at rest. These criteria were visually validated on linear 869 envelope and smoothed EMG signals before proceeding with analysis. 870 With the established threshold level, the individual EMG events within each burst were detected 871 using the Spike2 "Events" script and then concatenated into full bursts using the "Bursts" script 872 with a maximal burst onset interval of t=0.0025s and longest within-burst inter-spike interval of 873