Hypothermia-induced dystonia and abnormal cerebellar activity in a mouse model with a single disease-mutation in the sodium-potassium pump

Mutations in the neuron-specific α3 isoform of the Na+/K+-ATPase are found in patients suffering from Rapid onset Dystonia Parkinsonism and Alternating Hemiplegia of Childhood, two closely related movement disorders. We show that mice harboring a heterozygous hot spot disease mutation, D801Y (α3+/D801Y), suffer abrupt hypothermia-induced dystonia identified by electromyographic recordings. Single-neuron in vivo recordings in awake α3+/D801Y mice revealed irregular firing of Purkinje cells and their synaptic targets, the deep cerebellar nuclei neurons, which was further exacerbated during dystonia and evolved into abnormal high-frequency burst-like firing. Biophysically, we show that the D-to-Y mutation abolished pump-mediated Na+/K+ exchange, but allowed the pumps to bind Na+ and become phosphorylated. These findings implicate aberrant cerebellar activity in α3 isoform-related dystonia and add to the functional understanding of the scarce and severe mutations in the α3 isoform Na+/K+-ATPase.


Introduction
Dystonia is a movement disorder characterized by involuntary sustained or repetitive muscle contractions, causing twisting movements and abnormal postures [1,2]. It is usually caused by head injuries, drug side effects, metabolic insult, or genetic alterations, and is thought to involve the neuroanatomic circuitry of the basal ganglia, sensorimotor cortex, brainstem, and cerebellum [3].
While most dystonias are idiopathic, some are familial, and modifications of more than 25 designated genes associated with dystonia (DYTs) have been described [4]. Several mutations in the ATP1A3 (DYT12) gene, encoding the neuron-specific α 3 isoform of the Na + /K + -ATPase, can cause rapid-onset dystonia-parkinsonism (RDP) characterized by an abrupt onset of dystonia and parkinsonian motor-related features [5], or alternating hemiplegia of childhood (AHC) characterized by fluctuating spells of tonic, dystonic, hemiplegic and oculomotor abnormalities [6][7][8]. A separate mutation in ATP1A3 is responsible for the Cerebellar ataxia, Areflexia, Pes cavus, Optic atrophy, and Sensorineural hearing loss (CAPOS) syndrome [9]. All three ATP1A3 disorders are typically triggered by environmental and/or physiological events such as physical exhaustion, temperature changes, emotional stress, or infections, pointing to a broad spectrum of often distinct, yet overlapping, neurological disorders [10].
Intriguingly, missense mutations that cause different amino acid substitutions at the same position in the ATP1A3 gene can cause distinct diseases. A prominent example is amino acid position 801, where different mutations cause RDP or AHC (D801Y) [5,11,12] or AHC (D801N, D801E and D801V) [6,[13][14][15][16]. That position 801 is a hotspot for disease-causing mutations correlates with its crucial role in Na + /K + -ATPase pump function; the aspartate residue there is conserved in all Na + /K + -ATPase isoforms of all animal species, where it alternately coordinates both K + ions [17,18] and two of the three Na + ions transported [19,20], and is required for enclosure of the K + -ions [21].
In the present study we show that a mouse model with the D801Y disease-mutation (α 3 +/D801Y mice) [22] displayed severe hypothermia-induced dystonia, which correlated with abnormal cerebellar neuronal activity in vivo. In vitro pump characterization revealed that Dto-Y mutant pumps failed to carry out Na + /K + exchange, but retained the ability to bind Na + . These data thus provide a heretofore unknown link between hypothermia and dystonia that implicates aberrant cerebellar activity in α 3 isoform-related dystonias and provides functional insight into the disease-causing effects of the underlying Na + /K + -ATPase dysfunction.

Results
Hypothermia induce convulsion-like movements and dystonia-like postures in α 3 +/D801Y mice such conditions ( Fig 1A). However, stress tests that included restraining, tail suspension, randomly timed electric foot shocks, exposure to fox urine, hyperthermia resulting in elevation of body temperature to 40.4 ± 0.3˚C, and forced swimming in warm 35˚C water did not result in genotype-specific abnormal symptoms. Even a 2-week chronic unpredictable stress protocol failed to provoke symptoms. In contrast, forced swimming in 5-10˚C cold water for as little as 4 min consistently caused severe dystonia-like postures with hyperextended limbs (Fig  1B left ), and long periods of body twisting and convulsion-like movements (Fig 1B right ) in the α 3 +/D801Y mice (Fig 1A and S1 Video). These symptoms were never observed in WT littermates ( Fig 1B left and Supplementary movie 1). The attacks were pulsating in severity and lasted on average 39.2 ± 2.5 min (n = 6), after which α 3 +/D801Y mice fully recovered, with no apparent residual or persisting symptoms. As a result of the 4 min cold-water swim, core body temperature dropped significantly to about 20˚C in both α 3 +/D801Y mice and WT littermates (α 3 +/D801Y : 36.5 ± 0.2˚C to 19.5 ± 0.6˚C (p<0.0001, two-way ANOVA with genotype (WT versus α 3 ) and condition (baseline versus hypothermia) as main factors followed by Tukey's multiple comparisons test); WT: 36.5 ± 0.1˚C to 19.8 ± 0.2˚C (p<0.0001)) ( Fig 1C) without any difference in either baseline or post swim body temperature between α 3 +/D801Y and WT mice (p>0.9999 and p = 0.9873, respectively). Since only swimming in cold, and not in temperate water resulted in attacks, we speculated that lower body temperature, rather than the stress of forced swimming, was causative for the attacks. To address this, mice were placed in a freezing cold, -20˚C, environment and kept there until attacks developed, or until core body temperature dropped below 20˚C.
Remarkably, all six α 3 +/D801Y mice (but none of the six WT mice) developed identical attacks to those caused by the cold-water swim ( Fig 1A) as their body temperature dropped below 20˚C (Fig 1C), strongly indicating that lowered body temperature alone induced the attacks.
To further rule out a stress aspect of the attacks, α 3 +/D801Y mice were treated with the alphaadrenergic blocker prazosin before being subjected to cold-water swim. All prazosin-treated α 3 +/D801Y mice still developed attacks ( Fig 1A) with durations and severity identical to those untreated. If hypothermia indeed caused the attacks, rewarming the animal during an attack might be expected to reduce the duration and/or severity of the attack. Indeed, placing α 3 +/D801Y mice on a 33.3˚C heating pad immediately attack symptoms began, after induction by a cold-water swim, significantly diminished the duration of the attacks compared to the average time for recovery at room temperature (heating pad: 11.8 ± 0.5 min; room temperature: 39.2 ± 2.5 min (p = 0.0002, paired t-test)) ( Fig 1D).

Hypothermia-induced attacks are dystonic in nature
To explore the electrophysiological nature of these attacks, electrocorticographical (ECoG) recordings and hind limb electromyography (EMG) recordings were performed on five α 3 +/D801Y mice before (baseline) and during hypothermia-induced attacks.
ECoGs were recorded bilaterally from the primary motor cortex in freely moving animals (Fig 2A and 2B). None of the α 3 +/D801Y mice showed any epileptic activity during the attacks, and there was no evidence of genotype-related differences in ECoG signals or power spectra between baseline recordings and recordings during attacks (Fig 2C and 2D, respectively). As an experimental control, seizures were induced using a lithium-pilocarpine protocol, which resulted in the expected dramatically different ECoG activity and corresponding power spectra ( Fig 2E). Simultaneous EMG recordings from the anterior tibialis and gastrocnemius muscles of the hind limb ( Fig 2F) revealed a pronounced increase in co-contraction of these muscles during +/D801Y mice, following restraining for 10 min (n = 5), tail suspension for 6 min (n = 6), randomly timed electric foot shocks (n = 5), exposure to fox urine (n = 5), warm incubator (43˚C) (n = 5), temperate water swim (35˚C) (n = 6), chronic variable stress protocol (n = 11), cold water swim (5-10˚C) (n = 10), cold environment (-20˚C) (n = 6) and Prazosin treatment before cold water swim (n = 5). Only hypothermia, caused by cold water swim or cold environment exposure, consistently induced attacks in the α 3 +/D801Y mice (n = 15 for cold water and n = 6 for cold environment). (B) Example of dystonic-like posture with hind limbs hyperextended caudally (left picture, arrow) and a period of convulsion with abnormal postures and twisting movements (right picture) in α 3 +/D801Y mice after cold water swim. WT mice never displayed similar abnormal symptoms (left picture). (C) Core body temperature measured by rectal probe at onset of attack induced by exposure to cold water or cold environment. Both methods induced a significant drop in body temperature just below about 20˚C before symptoms occurred in α 3 +/D801Y mice. WT mice displayed identical drops in body temperature (n = 6 for both WT and α 3 ). (D) Attack duration after induction by cold water when α 3 +/D801Y mice were left to recuperate at room temperature or on a 33.3˚C heating pad (n = 6). https://doi.org/10.1371/journal.pgen.1006763.g001 Dystonia and abnormal cerebellar activity related to the sodium-potassium pump the attacks compared to baseline (Fig 2G and 2H), characteristic of dystonia. This was quantified with representative cross correlograms (Fig 2I). From the absence of epileptic activity in ECoG, and the evidence for co-contractions in the EMG recording, we conclude that the attacks were dystonic in nature.

The α 3 isoform in cerebellum
Purkinje cells are thought to express solely the α 3 isoform, and not the otherwise ubiquitouslyexpressed α 1 isoform [24,25], and are therefore suggested to be particular highly sensitive to modifications of α 3 isoform function [26,27]. We therefore investigated whether aberrant cerebellar function could be the cause of the observed motor deficits and inducible dystonia. To specifically address the expression of α 1 and α 3 isoforms in Purkinje cells, immunofluorescence staining of cerebellar slices with antibodies against the α 1 and α 3 isoforms was studied ( Fig 4B). WT Purkinje cells showed no expression of the α 1 isoform, whereas strong staining of the α 3 isoform was noted in all Purkinje cells. Notably, the same pattern of absence of expression of the α 1 isoform was observed in Purkinje cells in the α 3 +/D801Y mice, suggesting that mechanisms to compensate for loss of α 3 isoform activity are lacking in Purkinje cells.
Normal gross cerebellar morphology in α 3 +/D801Y mice Loss of cerebellar neurons has been reported in some patients with ATP1A3 mutations [28,29]. We therefore compared gross cerebellar morphology in the α 3 +/D801Y mice and WT littermates. No visible loss of cerebellar mass was evident ( Fig 4C) and cerebellar morphology appeared normal in stained slices from both α 3 +/D801Y mice and control WT littermates ( Fig   4D). Furthermore, quantification of Purkinje cells positive for calbindin revealed comparable numbers in α 3 +/D801Y mice and in WT littermates ( Fig 4E) (p = 0.59, t-test). α 3 +/D801Y mice exhibit irregular cerebellar activity and abnormal highfrequency burst-like firing Next, to investigate if cerebellar neuronal activity was affected, we performed in vivo singleunit extracellular recordings of Purkinje cells (Fig 5A) in awake head-restrained α 3 +/D801Y mice and WT littermates (representative raw traces are in Fig 5B).
The predominant firing rate and CV ISI for α 3 +/D801Y mice during dystonia were also significantly higher compared to the same α 3 +/D801Y mice at baseline (p = 0.0002 and p = 0.0152), whereas there was no significant difference in these parameters between WT mice exposed to the cold water (control WT) and baseline WT mice (p = 0.9959 and p = 0.8977) demonstrating 0.81 ± 0.06; WT: 0.52 ± 0.03 (p = 0.0297)) (Fig 5F lower ), indicating that cerebellar output was more irregular in α 3 +/D801Y mice compared to WT littermates.
The D-to-Y mutation impairs Na + /K + exchange To elucidate the molecular mechanistic consequences of the RDP/AHC-causing D801Y mutation and to compare it to the AHC-causing D801N mutation, Na + /K + -ATPase-mediated currents were recorded in oocytes expressing Xenopus laevis orthologs of α subunit ATP1A1, encoding either the homologous WT aspartate, D813, or the D813Y or D813N mutation that are equivalent to WT D801, D801Y and D801N, respectively, in the human and rodent α 3 isoform, in combination with ATP1B3 β subunit. Xenopus ATP1A1/ATP1B3 pumps were studied because these are believed to be the native isoforms in Xenopus oocytes [30,31], an established system for high-resolution measurements of Na + /K + pump function. Moreover, because the D801-equivalent aspartate is absolutely conserved in all Na + /K + -ATPase α subunits of all species and plays a crucial role in K + -ion binding, the D-to-Y and D-to-N substitutions may be expected to cause comparable disruptions of function in all Na + /K + -ATPase isoforms. Unlike WT Na + /K + pumps (Fig 6A, 6D and 6G), neither D-to-Y ( Fig 6B, 6E and 6H) nor D-to-N ( Fig  6C, 6F and 6I) mutant Na + /K + -ATPases were able to generate the outward (positive) current on exposure to high external K + (K + o ) that signifies normal electrogenic extrusion of 3 Na + i in exchange for import of 2 K + o in each ATPase transport cycle. D-to-Y or D-to-N mutant Na + / K + -ATPases similarly failed to demonstrate any K + o -activated outward current when competing external Na + was absent (Supplementary S2 Fig). However, in the presence of external Na + but absence of K + o , thereby precluding Na + /K + exchange even in WT Na + /K + pumps, both Dto-Y (Fig 6K and 6N) and D-to-N ( Fig 6L and 6O) Na + /K + -ATPases, like WT Na + /K + -ATPase ( Fig 6J and 6M), generated robust transient currents in response to membrane potential jumps. Those ouabain-inhibited Na + charge movements reveal the time course of the major conformational change of Na + -bound phosphorylated Na + /K + -ATPases that, in one direction encloses the three Na + ions and, in the other, releases them to the cell exterior [32]. Also like WT Na + /K + -ATPases (Fig 6G and 6J), both D-to-Y (Fig 6H and 6K) and D-to-N (Fig 6I and  6L) Na + /K + -ATPases still generated the small steady inward currents at large negative potentials that reflect pump-mediated import of protons [32]; in WT, that proton current is seen clearly only without K + o (Fig 6G), when overlapping outward Na + /K + exchange current was absent, but in D-to-Y ( Fig 6H) and D-to-N (Fig 6I), which are both incapable of generating outward Na + /K + exchange current, that inward current was evident with or without K + o . Indeed, because these mutant Na + /K + -ATPases are incapable of tightly binding external K + , they are effectively permanently trapped in the phosphorylated conformations that reversibly release the three bound Na + one at a time to the extracellular medium; it is precisely those conformations that carry out proton import. Importantly, not all effects of D-to-Y and D-to-N mutations were identical under all conditions, with Fig 6H and 6I (Supp. S2H and S2I Fig) suggesting that D-to-Y Na + /K + -ATPases possibly support larger proton influx than D-to-N Na + /K + -ATPases at negative resting potentials.

Discussion
In this study, we found that the α 3 +/D801Y mouse model exhibited prolonged episodes of hypothermia-induced dystonia, which usually began abruptly with hyper-extension of limbs and developed into abnormal postures and twisting movements, characteristic of dystonia [1,2,33]. To our knowledge, hypothermia has not previously been shown to trigger dystonia using animal models. Nevertheless, that altered temperature can induce phenotypes in the α 3 mice is not completely unexpected, as several clinically reported triggers for RDP, AHC, and CAPOS, involve a change in body temperature [8,10]. Prolonged exercise, alcohol consumption, and fever, all of which raise body temperature, are among the most frequently reported triggers of RDP [23]. Likewise, CAPOS can be induced by fever [9]. Furthermore, exposure to both cold and warm temperatures has been reported to trigger attacks in AHC patients [10,15]. Nevertheless, given that hyperthermia seems to be a common trigger in ATP1A3 patients, we also tested this in the α 3 +/D801Y mice by exposing them to a heated environment that raised their body temperature to 40.4 ± 0.3˚C, a physiologically relevant fever level. However, this experimentally induced hyperthermia failed to provoke any attacks in the α 3 +/D801Y mice despite the facts (i) that it induced symptoms of hyperthermia, also seen in WT littermates, including immobility or circling [34], and (ii) that 10-to 14-day-old WT mice have been reported to undergo febrile seizures at body temperatures averaging 41.3˚C [34]. We found that both Purkinje cells and their synaptic target, the DCN neurons, fire significantly more irregularly in the α 3 +/D801Y mice compared to WT animals, a finding that correlates with observed motor deficits in the α 3 +/D801Y mice, which are similarly found in all ATP1A3-related disorders [8]. The irregular firing was further exacerbated, and evolved into abnormal high-frequency burst-like firing, when dystonia was induced in the α 3 +/D801Y mice, similar to changes in cerebellar activity noted during dystonia induced in WT mice by brain perfusion with low-dose ouabain [35,36]. In further support of cerebellar involvement in α 3 isoform-related dystonia, heterozygous α 3 knock-out mice developed increased symptoms of dystonia after cerebellar perfusion with the excitatory glutamatergic agonist, kainate, and they exhibited enhanced inhibitory neurotransmission in the cerebellar cortex compared to WT mice [24]. Why exactly the cerebellum appears to be highly susceptible to dysfunction upon alterations of α 3 isoform activity remains to be firmly established. But a likely explanation is that Purkinje cells, as also shown here, express only the α 3 isoform of the Na + /K + pump and not the otherwise ubiquitously expressed α 1 isoform [24,25], and also do not appear to be able equivalent, held at -20 mV, exposed to 125 mM Na + solution at pH 7.6 containing 1 μM ouabain (to silence endogenous Na + /K + -ATPases), with 15 mM K + added as indicated by horizontal bars (Ko); the vertical lines are responses to 50-ms steps to other potentials. (D, E, F) Steady-state current levels plotted against voltage, from the recordings shown in (A, B, C) (filled symbols), in the presence (red) or absence (black) of K + , and from subsequent recordings in the same oocyte after inhibition of exogenously expressed pumps by 10 mM ouabain (empty symbols). (G, H, I) Average ± SEM 10 mM ouabain-sensitive steady currents (I ouab-sens) in 125 mM Na + , obtained by subtraction, at 0 mM K + (black circle) or 15 mM K + (red triangle), normalized to the maximum Na + charge movement in each oocyte (J-O, below), a measure of the number of Na + /K + -ATPases; wild type (n = 4 oocytes), D-to-Y (n = 3 with K + , n = 6 without), D-to-N (n = 3). (J, K, L) 10 mM ouabain-sensitive pre-steady-state Na + currents for wild type (J), D-to-Y (K), and D-to-N (L) Na + /K + -ATPases in 125 mM Na + and 0 mM K + solution obtained by subtraction of traces before and after pump inhibition; superimposed traces are from steps to voltages between -180 mV and +60 mV, and back to the holding potential, -20 mV. to express the α 1 isoform as a compensatory response to α 3 isoform dysfunction. This special characteristic of Purkinje cells is supported by the finding that shRNA-mediated knock-down of the α 3 isoform led to disruption of the intrinsic firing of Purkinje cells, but not of DCN neurons which express both α 1 and α 3 isoforms, when synaptic inputs were inhibited in vitro [27]. How hypothermia affects the firing of the cerebellar neurons and induces dystonia in the α 3 +/D801Y mice remains to be elucidated. An earlier study found that the intrinsic activity of Purkinje neurons in culture was increasingly slowed as the cells were cooled below 20˚C, with the duration of action potentials increasing as their frequency decreased [37]. This temperature range correlates with the induction of attacks in the α 3 +/D801Y mice we observed when body temperature fell to 20˚C. Furthermore, the spread of the ISI became larger in the cultured Purkinje cells as the temperature was lowered [37], echoing the increase in CV ISI during hypothermia-induced dystonia in the α 3 +/D801Y mice. In cerebellar slices, the firing of Purkinje cells was furthermore shown to be particular affected by lowering the temperature [38], suggesting that Purkinje cells are highly temperature sensitive. Maintenance of normal intrinsic activity of Purkinje cells depends on function of their α 3 isoform Na + /K + -ATPases [26,27]. Although α 3 isoform Na + /K + -ATPases in WT mice are presumably slowed by low temperature, the complete loss of Na + /K + exchange by α 3 isoform Na + /K + -ATPases bearing the D-to-Y mutation in cerebellar Purkinje neurons of α 3 +/D801Y mice may be expected to impair their ability to adequately sustain electrical activity, compared to WT mice, at similarly low body temperatures.
Clinically, AHC and RDP have been considered to be distinct disorders, although with overlapping features [8]. The same appears true for mice, as the α 3 +/D801Y mice in some aspects phenotypically differ from Mashlool (α 3 ) mice [39] and Myshkin (α 3 ) mice [40] that are heterozygous for the AHC mutations, D801N and I810N, respectively (Table 1). Both Mashlool and Myshkin mice exhibited spontaneous recurrent tonic clonic seizures [39,40]. In contrast, although α 3 +/D801Y mice have a lowered threshold for PTZ-induced seizure, they did not develop spontaneous seizures [22]. Mashlool mice, furthermore, effectively modeling AHC closely by exhibiting hemiplegic episodes of relatively long duration, but also short dystonic spells, upon water exposure [6,39]. Here we show that α 3 +/D801Y mice suffer from abrupt hypothermia-inducible dystonia that recapitulates the abrupt triggerable onset of symptoms in ATP1A3 patients [41]. These induced, and EMG-confirmed, dystonic attacks lasted noticeably longer than those observed in Mashlool mice (α 3 +/D801Y : 39.2 ± 2.5 min; Mashlool (α 3 +/D801N ): 0.07 ± 0.005 min) [39]; however, the dystonic attacks in α 3 +/D801Y mice were not persistent as in RDP patients [23], and could thus resemble spells noted in AHC patients, except that the induced attacks did not include EMG hemiplegia activity or ECoG seizure activity that could be expected of such episodic spells of AHC patients [8]. This possibly reflects the intermediate nature of the D801Y mutation that has been associated with RDP and AHC [5,11,12].
To further explore how different mutations might cause this spectrum of neurological features, we made side-by-side comparisons of the effects of the D801Y and D801N mutations on the function of otherwise identical Na + /K + -ATPase isoforms, in the same cells and under the same conditions. We show that the D-to-Y and D-to-N mutations both abolish Na + /K + exchange, in accordance with the known crucial role of this conserved aspartate residue to coordinate both transported K + ions in the pump binding sites [19,20], and with previous reports that the D801N mutation abolishes K + occlusion [21] and K + -activated outward current [42], and reduced forward cycling [43]. Nevertheless, preservation of Na + /K + -ATPasemediated transient Na + charge movements in both D-to-Y and D-to-N Na + /K + -ATPases demonstrates that both mutants remain capable of binding 3 Na + i ions, of consequent Na + -dependent phosphorylation, and of essential Na + -releasing and -rebinding conformational changes [32], as previously inferred for D801N from ouabain-binding measurements [44]. Intriguingly, the Na + -bound phosphorylated conformations that are shown here to be preserved in the Dto-Y and D-to-N Na + /K + -ATPases, and in which their inability to bind K + ions dooms these mutants to spend most of their time, comprise precisely those conformations that support pump-mediated proton import [32]. In other words, whereas D-to-Y and D-to-N mutants must both be viewed as loss-of-function in terms of Na + /K + exchange, they must both be considered gain-of-function in terms of their exclusive occupancy of proton-importing pump states.
As the failure of Na + /K + exchange by the D-to-Y and D-to-N mutations demonstrated here in Xenopus oocytes at room temperature under optimal [K + ] o , [Na + ] i , and [ATP] conditions is expected to be recapitulated in D801Y and D801N α 3 isoform Na + /K + -ATPases under physiological conditions in mice, then α 3 +/D801Y mice and Mashlool (α 3 +/D801N ) mice should both be effectively haploinsufficient in terms of α 3 isoform Na + /K + -ATPase-mediated Na + extrusion. However, we found not all effects of D-to-Y and D-to-N mutations to be identical under all conditions, suggesting that D-to-Y Na + /K + -ATPases possibly support larger proton influx than D-to-N Na + /K + -ATPases at negative resting potentials, like those of neurons. Whether phenotypical differences attributed to D801Y versus D801N mutation in α 3 isoform pumps, such as clinical diagnoses of RDP rather than AHC in patients, and phenotypical differences between α 3 +/D801Y and Mashlool mice, reflect these relatively small functional differences cannot be concluded at present, and this requires further investigation. The conclusion that both D801Y and D801N are loss-of-function mutations for Na + /K + exchange, but are effectively gain-of-function in that the mutant pumps engage full time in proton import, possibly accounts for the scarcity and severity of these de novo gene alterations and the apparent lack of nonsense mutations (resulting in haploinsufficiency) in the ATP1A3 patient group [6,10]. If this inference is correct, it has significant implications for future therapeutic intervention in ATP1A3-related diseases in which α 3 isoform haploinsufficiency may be less damaging than carrying certain missense gain-of-function mutations. Thus, strategies such as exon skipping and genome editing may be selected to eliminate the mutated disease-causing allele in cases where it is considered more deleterious than loss-of-function alleles.

Environmental and physiological stress conditions
Each condition was tested in at least 5 mice of each genotype and repeated 3 times per mouse. After being subjected to each condition, the mice were place on a cleared table or empty cage where they were video recorded and closely monitored by the experimenter for the occurrence of an attack. Average occurrence of attacks (%) and attack duration were subsequently calculated per mouse, per condition.
Restraining. Mice were restrained in a 60 mL falcon tube for 10 min. The back of the tube was partially closed off but allowed enough air to prevent suffocation.
Tail hanging. Mice were hung from their tail for 6 minutes. Foot shocks. Mice were placed in a custom made plastic box with metal wiring on the bottom connected to a stimulating device. Randomly timed short shocks were given for 5 min.
Exposure to fox urine. Mice were put in a closed cage with a tube containing a tissue drenched in fox urine for 10 min.
Temperate water swim. A clear plastic box (42 x 26 x 18 cm) was filled with 35˚C water to a water height that forced mice to swim without the possibility to touch the box floor. Mice were placed in the water and forced to swim for 10 min.
Chronic variable stress. The mice were subjected to one of three different stressors over the course of two weeks. Mice were suspended by their tail for 6 minutes on days 1, 6 and 9. The mice were placed in glass jars with 30˚C water for 6 minutes on days 2, 8 and 11. Finally, the mice were restrained for 60 minutes using DecapiCones (BrainTree Scientific, Inc., MA, USA) on days 3, 7 and 10.
Cold water swim. A clear plastic box (42 x 26 x 18 cm) was filled with 5-10˚C cold water to a water height that forced mice to swim without the possibility to touch the box floor. Mice were placed in the water and forced to swim for 4 min. Mice that showed difficulty staying afloat before the 4 min mark were immediately removed from the water. Rectal body temperatures were measured just after the animals were removed from the water.
Cold environment. Mice were placed in an empty clear plastic box and placed in a -20˚C environment until they displayed attacks or their body temperature reached <20˚C as measured by rectal body temperature.
Elevation of body temperature. Mice were placed in a large glass beaker placed in a 43˚C warm incubator for 15 min. Rectal body temperature was measured after the animals were removed from the incubator.

Surgery
Mice were anesthetized using isoflurane (4% in 1.5 L/min O 2 for induction and 1-2% in 1.5 L/ min O 2 for maintenance) after which the skull was exposed, cleaned and treated with Opti-Bond All-In-One (Kerr Corporation; Orange, CA, USA) in order to ensure adhesion of the light cured hybrid composite Charisma (Heraeus Kulzer; Hanau, Germany) to later attach implants.
In case of preparation for ECoG recordings, four small holes (0.5 mm in diameter) were subsequently drilled for implantation of the ECoG electrodes; two above bilateral primary motor cortices (+1 mm AP and ± 1 mm ML relative to bregma) and two in the interparietal bone (-1 mm AP and ± 1 mm ML relative to lambda) to accommodate the ground and reference electrodes (Fig 2A). Teflon coated silver ball-tip electrodes (~200 μm), attached to an ECoG headmount (Pinnacle Technology; Lawrence, KS, USA), were carefully inserted into the holes and fixed in place using the hybrid composite Charisma. The rest of the skull was subsequently covered with hybrid composite Charisma to ensure insulation after which the headmount was attached to the skull.
Surgical preparation for EMG recordings was similar to the ECoG procedure. In this case coated stainless steel wires were attached to the headmount and were subcutaneously led to the hind limb. In order to avoid too much pressure on the wires at hind limb level, the wires were sutured to a patch of skin in the back and hip before attaching them to the hind limb muscles. The end of the wires were partially stripped and carefully stitched to the anterior tibialis and gastrocnemius muscles of the right hind limb. Ground and reference electrodes were made shorter than the others, partially stripped and left loose at the level of the hip.
For in vivo electrophysiological recordings a metal bracket was attached to the front part of the skull using light cured hybrid composite Charisma. A recording chamber was constructed on top of cerebellum with dental cement and the cavity was filled with silicone.
Following surgery all animals were given flunixin (2.5 mg/kg) and 500 μl saline, after which they were allowed at least four days to recover before being subjected to experimental procedures. All animals were monitored closely for any complications on a daily basis after surgery.

Motor behavior tests
Mice were transferred to the test room 1 hour prior to testing for acclimation. Behavioral apparatus were cleaned between tests in 70% EtOH.
Gait analysis. Front and hind paws where painted with wet paint and the mouse was allowed to run across a long sheet of white paper. Fore base width, hind base width and stride length were subsequently estimated.
Hind limb clasping. Mice were suspended by their tails for 30 sec. Each mouse was evaluated by video analysis and given a score of either 0 (no abnormal hind limb movement) or 1 (abnormal hind limb movement) in 10 sec intervals allowing a maximum score of 3. Abnormal hind limb movement was defined as the retraction of either one or both hind limb towards the midline [45].
Balance beam. Mice were tested for 3 successive days on a 1 m long, 9 mm wide suspended wooden beam. Number of slips and the time to traverse the center 80 cm of the beam was evaluated by video analysis and calculated as the mean of three trials.
Rope climb. Mice were tested on a 40 cm long and 10 mm in diameter vertical rope. Time to climb was evaluated by video analysis and estimated as the mean of three trials.
Parallel rod floor. Mice were tested in the parallel rod floor apparatus (Stoelting Co, Wood Dale, US) for 15 min. Anymaze software (Stoelting Co, Wood Dale, US) was used to record number of slips and distance traveled.
Grip strength. A grip bar was attached to the grip strength meter (Bioseb, Vitrolles, France) to allow measurements of grip strength. A mouse was picked up by its tail and lowered until it grasped the grip bar at which point the mouse was pulled away horizontally until its grip was released. Readouts of grip strength given in grams were normalized to the body weight of the mouse and calculated as the mean of five trials.

Functional characteristics of mutant Na + /K + -ATPases
Mutagenesis and expression. The substitution C113Y (equivalent to C101Y in human α 3 ) was introduced into Xenopus Na + /K + -ATPase α 1 subunits to render them ouabain resistant so their activity could be isolated after silencing endogenous ouabain-sensitive Xenopus Na + /K + -ATPases by maintained exposure to 1 μM ouabain. C113Y α 1 subunits (here designated "wild type") then served as templates for D813Y and D813N mutations (equivalent to D801Y and D801N, respectively, in human α 3 ); these cysteine and aspartate residues are conserved in all Na + /K + -ATPase α isoforms of all species. All substitutions were by QuikChange (Stratagene, California, US) and were verified by sequencing. cDNAs were transcribed in vitro, and 15-45 ng of Na + /K + -ATPase α 1 subunit cRNA was coinjected with 5-15 ng of wild-type Xenopus β 3 subunit cRNA into defolliculated Xenopus laevis oocytes, which were incubated at 18˚C for 3 days before recording.

ECoG and EMG recordings
Recordings were performed in awake, freely moving animals using a Pinnacle data acquisition system (Pinnacle Technology; Lawrence, KS, USA). They were given 1-2 accommodation sessions to allow them to get used to the experimental setup after which experiments started. 10-30 min of baseline activity was recorded in each mouse before being subjected to a cold-water swim. ECoG or EMG activity was subsequently recorded during the attack that followed until they were fully recovered (20-50 min later). As a control, tonic-clonic seizures were induced using a Lithium Pilocarpine protocol in the same mice. 18-24 hrs prior to seizure induction, mice were given an IP injection of lithium chloride (3 mmol/kg; Sigma-Aldrich; St. Louis, MO, USA) dissolved in 0.9M NaCl after which 60-120 mg/kg of pilocarpine hydrochloride (dissolved in 0.9M saline; Sigma-Aldrich) was given IP to induce the epileptic seizure. Data were sampled at 400 Hz for ECoG and 600 Hz for EMG recordings and filtered online using a 1-150 Hz (ECoG) or 1-250 Hz (EMG) band pass filter. Signals were visualized and power spectra and cross-correlations were calculated using a custom written LabView algorithm (National Instruments; Austin, TX, USA).

In vivo electrophysiology
A craniotomy was made over the cerebellum and the mice were head restrained via the implanted metal bracket in the recording setup and given one hour to acclimate before recording was initiated.
In order to record from α 3 +/D801Y mice during dystonic attacks, mice were cooled in 6-8˚C cold water for 4 min and immediately head restrained and after which the recording was initiated. As a control, the same experiments were done in wild type littermates. Single-cell activity of Purkinje cells and DCN neurons was recorded extracellularly using a tungsten electrode (Thomas Recording, 2-3MO), which was advanced into the cerebellum until either the Purkinje cell layers or DCNs were reached (All coordinates given from Bregma: Caudal 6.45-6.55 mm, Lateral ±1.2 mm, Ventral 1.4-2.2 mm for Purkinje cell layers and 3-3.5 mm for DCN). The constructed recording chamber was filled with saline and used for ground connection. Signals were band-pass filtered (200 Hz-20 kHz) and amplified (2000×) using a custom built amplifier and then digitized (20 kHz) using a National Instruments PCI MIO 16 XEI (National Instruments Corporation, Austin, US).
Waveforms were sorted offline using characteristics of the spikes such as amplitude and energy and by principal component analysis (Offline Sorter, Plexon, US). Purkinje cells and DCN neurons were identified by location, characteristic firing rate, and the presence of complex spikes for Purkinje cells.
Average firing rate (spikes per sec, sp/s), predominant (mode) firing rate (sp/s), and coefficient of variation of interspike intervals (CV ISI) were calculated using custom made MATLAB software.

Statistical analysis
Statistical analyses were performed in Graphpad Prism software (GraphPad Software Inc, La Jolla, US). Student's t-test (unpaired or paired) was used to determine significance when comparing two groups. Two-way ANOVAs followed by Tukey's multiple comparisons test were used to find statistically significant differences between three or more groups. Statistical analyses for each experiment are indicated in the result section with corresponding p-values. Differences were considered significant if p<0.05. and C) Currents recorded in Na + -loaded oocytes expressing exogenous ouabain-resistant Na + / K + -ATPases without (wild type, A), or with, a D-to-Y (B) or D-to-N (C) mutation at position 801 equivalent, held at -20 mV, exposed to Na + -free solution at pH 7.6 containing 1 μM ouabain (to silence endogenous pumps), with 15 mM K + added as indicated by horizontal bars (Ko); the vertical lines are responses to 50-ms steps to other potentials. These recordings are from the same experiments as in Fig 6A, 6B and 6C. (D,E and F) Steady-state current levels plotted against voltage, from the recordings shown in A,B and C (filled symbols), in the presence (red) or absence (black) of K + , and from subsequent recordings in the same oocyte after inhibition of exogenously expressed pumps by 10 mM ouabain (empty symbols). (G,H and I) Average ± SEM 10 mM ouabain-sensitive steady currents (I ouab-sens), obtained by subtraction, at 0 mM K + (black circle) or 15 mM K + (red triangle), normalized to the maximum Na + charge movement, a measure of Na + /K + -ATPase number, determined for each oocyte from transient currents in 125 mM Na + and 0 mM K + ; wild type (n = 4 oocytes), D-to-Y (n = 5 with K+, 6 without), D-to-N (n = 3); at -180 mV, D-to-Y currents averaged -120 nA/nC in 0 mM Na + with 0 mM K + , and -300 nA/nC in 0 mM Na + with 15 mM K + . (TIF) S1 Video. Cold water-induced attacks.