|Home | About | Journals | Submit | Contact Us | Français|
Stroke is a leading cause of disability; but no pharmacological therapy is currently available for promoting recovery. The brain region adjacent to stroke damage, the peri-infarct zone, is critical for rehabilitation, as it exhibits heightened neuroplasticity, allowing sensorimotor functions to re-map from damaged areas1–3. Thus, understanding the neuronal properties constraining this plasticity is important to developing new treatments. Here we show that after a stroke in mice, tonic neuronal inhibition is increased in the peri-infarct zone. This increased tonic inhibition is mediated by extrasynaptic GABAA receptors (GABAARs) and is caused by an impairment in GABA transporter (GAT-3/4) function. To counteract the heightened inhibition, we administered in vivo a benzodiazepine inverse agonist specific for the α5-subunit-containing extrasynaptic GABAARs at a delay after stroke. This treatment produced an early and sustained recovery of motor function. Genetically lowering the number of α5 or δ-subunit-containing GABAARs responsible for tonic inhibition also proved beneficial for post-stroke recovery, consistent with the therapeutic potential of diminishing extrasynaptic GABAAR function. Together, our results identify new pharmacological targets and provide the rationale for a novel strategy to promote recovery after stroke and possibly other brain injuries.
Stroke is a major source of disability, confining one-third of stroke survivors to nursing homes or institutional settings4. Recent studies have shown that the brain has a limited capacity for repair after stroke. Neural repair after stroke involves re-mapping of cognitive functions in tissue adjacent to or connected with the stroke5,6. Functional recovery in this peri-infarct tissue involves changes in neuronal excitability that alter the brain's representation of motor and sensory functions. Stimulation of peri-infarct cortex enhances local neuronal excitability through a process that involves long-term potentiation (LTP), alters sensorimotor maps, and improves use of affected limbs5–8. The inhibitory neurotransmitter GABA is critical for cortical plasticity and sensory mapping. Altering GABAergic transmission changes sensory maps during the critical period of cortical development9 and produces rapid alterations in adult cortical maps that resemble changes occurring after stroke10,11. Alterations in cortical maps through blockade of GABAergic signaling are associated with fundamental changes in cellular excitability including LTP12. In a similar manner to normal cortical plasticity, GABAergic mechanisms may mediate changes in neuronal excitability that play a central role in functional recovery of peri-infarct cortex after stroke.
Cortical GABAergic signaling through GABAARs is divided into synaptic (phasic) and extrasynaptic (tonic) components. Tonically active extrasynaptic GABAARs set an excitability threshold for neurons13,14. Extrasynaptic GABAARs primarily consist of α5 or δ-subunit-containing receptors13,14. Pharmacological and genetic knockdown of α5-GABAARs enhance LTP and improve performance on learning and memory tasks15,16. The selective effects of extrasynaptic GABAARs on cellular excitability and plasticity, and the evidence that changes in neuronal excitability underlie functional reorganization in peri-infarct cortex, suggest that this system may play a role in post-stroke recovery. We find that stroke increases tonic GABAergic transmission in peri-infarct cortex and dampening this tonic inhibition produces an early and robust gain of motor recovery post-stroke (Supplementary Fig. 1, schematic summary).
We examined neuronal excitability in the peri-infarct cortex of mice during the period of recovery and reorganization after a photothrombotic stroke to forelimb motor cortex. Whole-cell voltage-clamp recordings in in vitro brain slices prepared at 3-, 7-, and 14-days post-stroke (Fig. 1a) showed a significant increase in GABAAR-mediated tonic inhibition (Itonic) in layer 2/3 pyramidal neurons, compared to neurons from sham controls (control: 8.05±0.80 pA/pF, n=24, vs. post-stroke: 13.6±1.41 pA/pF, n=45, Mann-Whitney U-test, P<0.05; Fig. 1b). Itonic remained elevated from 3- to 14-days post-stroke (Supplementary Fig. 2a). The mean phasic excitation remained unchanged over the 2-week period after stroke (Supplementary Fig. 3a, b). The mean phasic inhibition was unchanged except for a transient decrease at 7-days post-stroke (Supplementary Fig. 3c, d). The resting membrane and GABA reversal potentials were both unchanged (Supplementary Fig. 3e, f).
Tonic inhibition is effectively controlled by the degree of extracellular GABA uptake through neuronal and astrocytic GABA transporters (GATs)14. We applied a GAT-1-selective antagonist, NO-711 (10μM), and found a significantly greater effect (% Itonic increase after GAT blockade) in post-stroke neurons (94.0±16.3%, n=10) than in controls (34.3±11.4%, n=6; P<0.05; Fig. 2a). Co-application of NO-711 and the GAT-3/4-selective antagonist SNAP-5114 (40 μM) produced a substantial increase in Itonic in controls (300.6±46.0%, n=4; Fig. 2a), revealing the synergistic actions of GATs in the cortex as previously proposed17. In post-stroke neurons, co-application only produced an effect (110.7±32.0%, n=5) similar to GAT-1 blockade alone (P=0.68; Fig. 2a), indicating a dysfunction in GAT-3/4 after stroke. Sequential blockade of the two GATs confirmed the post-stroke impairment, as peri-infarct Itonic showed no further response to GAT-3/4 blockade after the initial GAT-1 block, in contrast to responses shown in controls (Fig. 2b, c). This effect was not due to receptor saturation, as Itonic showed further increase in response to raised [GABA] under GAT blockade (Supplementary Fig. 4a). Western-blot analysis confirmed a reduced GAT-3/4 level in peri-infarct cortex, whereas GAT-1 levels were unchanged (Supplementary Fig. 5).
We hypothesized that the chronically elevated tonic inhibition in the peri-infarct region may antagonize the neuronal plasticity required for functional recovery after stroke. Therefore, we tested whether reducing the excessive tonic inhibition would improve function recovery. Of the two GABAARs subtypes shown to underlie tonic inhibition in cortical neurons, the α5-GABAARs can be antagonized specifically by L655,708, a benzodiazepine inverse agonist16, while no specific antagonist exists for δ-GABAARs. L655,708 (100 nM) decreased Itonic in control neurons by −13.3±5.2% (n=4), but produced a significantly greater decrease in post-stroke neurons (−30.0±4.1%, n=13; P<0.05; Fig. 2d, e), which reverted Itonic back to control level (control: see above vs. post-stroke + L655,708: 140.8±18.5pA, n=13; P=0.702; Fig. 2f). L655,708 produced only minimal effects on phasic inhibitory currents in both post-stroke and control conditions (Supplementary Fig. 4b).
We next tested the effects of reducing tonic inhibition on functional recovery after stroke, using measures of fore- and hindlimb motor control. Stroke produced an increase in the number of foot-faults in grid-walking task, and a decrease in forelimb asymmetry in the cylinder task from 7-days post-stroke. Chronic treatment with L655,708 starting 3-days post-stroke resulted in a dose-dependent maximal gain of function beginning from 7-days post-stroke in both tasks (P<0.001; Fig. 3a–c). Acute treatment with L655,708 just prior to behavioral testing had a minimal effect on stroke recovery (Supplementary Fig. 7). To assess the necessity of long-term administration, we discontinued L655,708 treatment after 2 weeks and found a decrease in functional gains, although these mice still performed better than vehicle-treated stroke controls (Supplementary Fig. 6).
To further corroborate the role of reduced tonic inhibition in enhancing stroke recovery, we tested mice with deletions of either α5 or δ-subunit-containing GABAARs (Gabra5−/− and Gabrd−/−, Methods)18. Gabra5−/− animals showed significantly better motor recovery post-stroke, comparable to L655,708-treated wild-type animals (Fig. 3d–f). In addition, Gabra5−/− animals displayed a significant reduction in hindlimb foot-faults (Fig. 3e). Gabrd−/− animals also showed significant improvements in motor recovery (Fig. 3d–f), but to a lesser extent than the Gabra5−/− mice. Thus, modulation of α5GABAARs affords greater functional gains in motor recovery than δGABAARs, and genetic removal of α5GABAARs produces a more widespread increase in motor recovery than pharmacological antagonism. Administration of L655,708 to Gabrd−/− mice produced an even greater recovery, confirming the beneficial effect of reducing peri-infarct tonic inhibition (Fig. 3a, c).
Low/sub-seizure dosing of picrotoxin (PTX: 0.1mg/kg, i.p.), a use-dependent GABAAR antagonist, enhances learning and memory in transgenic mouse models of Alzheimer's and other cognitive impairments by reversing an increased GABAergic inhibitory tone, acting at both synaptic and extrasynaptic GABAARs19,20. The pharmacological effects of PTX on reducing phasic and tonic inhibition were not altered after stroke (Supplementary Table II). PTX given to animals from 3-days post-stroke resulted in a significant gain of forelimb function on the grid-walking task compared to vehicle-treated stroke controls (P<0.05; Supplementary Fig. 9a). No significant changes were observed in hindlimb function or forelimb asymmetry (Supplementary Fig. 9b, c). Combined L655,708 + PTX treatment showed similar initial functional gains compared to stroke + L655,708 alone; however, prolonged PTX + L655,708 treatment produced a deterioration in motor function such that the performance progressively worsened at late periods after stroke (Supplementary Fig. 9). These data suggest that increasing cortical excitability too far or reducing phasic inhibition negatively impact functional recovery.
An important element in stroke treatment is the timing of drug delivery. GABAAR agonists administered at the time of stroke decrease stroke size20. Therefore, dampening tonic inhibition too early after stroke may produce an opposite effect, i.e. increased cell death. To test this, we assessed stroke volume at 7-days post-stroke, in animals treated with 1) vehicle, 2) L655,708 from stroke onset, and 3) L655,708 from day-3 post-stroke. Stroke volumes were similar between mice treated with vehicle and L655,708 from day-3 (Fig. 4). In contrast, stroke volume was significantly increased in animals treated with L655,708 from stroke onset (P<0.05; Fig. 4). These data indicate a critical timeframe for therapeutically dampening tonic inhibition post-stroke: reduction too early would exacerbate stroke damage, while delaying treatment by 3-days would promote functional recovery without altering stroke size. Genetic deletion of α5- or δ- GABAARs did not affect infarct size or neuronal number in peri-infarct cortex (Supplementary Fig. 8). Unlike pharmacological antagonism of α5-GABAAR-mediated inhibition, in Gabra5−/− and Gabrd−/− mice, the genomic absence of one of the extrasynaptic GABAARs may trigger compensatory upregulation of the other receptor13 and thus obscuring their roles in neuroprotection immediately after stroke.
Current therapies that promote functional recovery after stroke are limited to physical rehabilitation4. Here, by identifying an excessive tonic inhibition after stroke, we have found promising new targets for pharmacological interventions to promote recovery. The elevated tonic inhibition in cortical pyramidal neurons occurs during precisely the same time-period important for cortical map plasticity and recovery1–3. Alterations in other aspects of cortical signaling have also been described during this period, including altered GABAAR subunits, glutamate receptor expression and neuronal network properties21–24. Protein levels of GAT-1 and GAT-3/4 were shown to be decreased in peri-infarct cortex in some rodent stroke models, and reactive astrocytes exhibit reduced uptake of other neurotransmitters23. However, there are conflicting data on GABAAR levels after stroke23–26. We found decreased protein level and compromised function of GAT-3/4 in peri-infarct cortex. The elevated tonic inhibition may curtail cortical plasticity and spontaneous recovery after stroke, and is consistent with tonic GABAergic inhibition exerting a causal role in limiting motor recovery in stroke.
Non-selectively decreasing GABAergic tone facilitates neuronal plasticity in genetic models of cognitive diseases19,20. We show for the first time that antagonizing an elevated tonic inhibition enhances motor recovery after stroke, consistent with the idea that molecular and cellular events of neuronal plasticity are dampened in the peri-infarct zone, and promoting this plasticity facilitates functional recovery. Together, our results have identified novel pharmacological targets and provide a rational basis for developing future therapies to promote recovery after stroke and possibly other brain injuries.
Focal stroke was induced by photothrombosis in adult male C57BL/6 mice (age 2–4 month) as described by27.
Following decapitation, brains were rapidly removed and placed into a N-methyl-D-glucamine (NMDG)-based cutting solution to enhance neuronal viability28. Coronal slices (350μm) were cut and transferred to an interface-style chamber containing artificial cerebrospinal fluid as previously described13. Recordings were made from intact peri-infarct cortical layer-2/3 pyramidal neurons and analyzed as previously described13.
L655,708 was dissolved in DMSO and then diluted 1:1 in 0.9% saline. L655,708-filled ALZET-1002 pumps were implanted at 3-days post-stroke and replaced every two weeks. In acute administration studies, 5mg/kg L655,708 was administered (i.p.) 30 minutes prior to testing. The concentration in one minipump, 5mM, delivers a 200ug/kg/day dose in mice. With one or two minipumps implanted, this provides a dose escalation. PTX (0.1mg/kg i.p. bi-daily) starting 3-days post-stroke was administered alone or in concert with L655,708.
Mice were videotaped during walking and exploratory behavior in the grid-walking and cylinder/rearing tasks, tested at approximately the same time each day during the nocturnal period29. Baseline behavioral measurements were obtained one week prior to surgery. Post-stroke animals were assessed at weeks 1, 2, 4, and 6.
For the histological assessment of infarct size, brains were processed at 7-days post-stroke using cresyl violet as previously described30.
All data are expressed as mean ± s.e.m. For electrophysiological comparisons between control vs. post-stroke, Mann-Whitney non-parametric test was used. For multiple comparisons across post-stroke days, one-way analysis of variances (ANOVA) and Newman–Keuls' multiple pair-wise comparisons for post-hoc comparisons were used. For behavioral testing, differences between treatment groups were analyzed using two-way ANOVA with repeated measures and Newman–Keuls' multiple pair-wise comparisons. The level of significance was set at P<0.05.
Under isoflurane anesthesia (2–2.5% in a 70% N2O / 30% O2 mixture), 2–4 month-old adult C57Bl6 (Charles River, Wilmington, MA) male mice were placed in a stereotactic apparatus, the skull exposed through a midline incision, cleared of connective tissue and dried. A cold light source (KL1500 LCD, Zeiss) attached to a 40× objective giving a 2mm diameter illumination was positioned 1.5mm lateral from Bregma, and 0.2mL of Rose Bengal solution (Sigma; 10 g/L in normal saline, i.p.) was administered. After 5-min, the brain was illuminated through the intact skull for 15-min. Rose-bengal produces singlet oxygen under light excitation, which damages and occludes vascular endothelium, resulting in focal cortical stroke under the region of illumination (Fig. 4), circumscribed by peri-infarct tissue with normal neuronal cell number (Supplementary Fig. 8). Two to four month-old adult male Gabra5−/− and Gabard−/− mice1 received stroke as above. These mice had been back-crossed to C57Bl6 in excess of 15 generations, and were compared in behavioral studies to wild-type C57Bl6. Body temperature was maintained at 36.9 ± 0.4°C with a heating pad throughout the operation and did not vary by drug or genetic condition. This stroke method produces a small stroke in the mouse forelimb region of the motor cortex (Fig. 4). Sample size was 10 per group for Gabara5−/− and Gabard−/− in stroke/behavioral studies. Sample size was 8 per group for each condition in dosing of L655,708 (Fig. 3).
Blood pressure (systolic and diastolic) and heart rate were measured in separate cohorts of mice in wild-type (C57Bl6), with and without L655,708 administration via ALZET minipumps from 3-days post-stroke, and in Gabra5−/− and Gabrd−/− mice, before during and after stroke, using a standard non-invasive tail-cuff method (Coda, Kent Scientific, Torrington, CT). There were no significant differences in heart rate or blood pressure by treatment or genotype (Supplementary Table I). All studies in this manuscript complied with the STAIR (Stroke Therapy Academic Industry Roundtable) criteria for stroke investigations in measuring physiological parameters, monitoring treatment effects for at least one month, analyzing treatment effects blinded to conditions, utilizing dose-response studies, and use of a drug administration route with blood brain barrier penetration.
Slices were submerged in the recording chamber and continuously perfused (5–8 ml/min) with oxygenated ACSF (32–34°C). Visualized patch-clamp recordings from layer-2/3 pyramidal neurons were performed at 40x using infra-red oblique-illumination (Leica DM-LFS; Hamamatsu CCD camera C3077–78).
Control recordings were made from cells of sham-operated animals at similar locations as those recorded in post-stroke animals. Microelectrodes (3–5 MΩ) were filled with a cesiummethylsulfonate (CsMeSO4)-based internal pipette solution, containing (in mM) 120 CsMeSO4, 10 CsCl, 5 TEA-Cl, 1.5 MgCl2, 10 HEPES, 0.1 EGTA, 2 Na-ATP, 0.5 Na-GTP, and 5 QX-314, pH 7.25–7.30 with CsOH, 275–285 mOsm. The recording ACSF was supplemented with 5 μM GABA to replenish the extracellular GABA concentration reduced by the high-flow perfusion of the slices1. For recording Itonic at −70 mV, a high-CsCl-based internal solution was used, containing (in mM) 140 CsCl, 1 MgCl2, 10 HEPES, 0.1 EGTA, 4 NaCl, 2 Mg-ATP, 0.3 Na-GTP, and 5 QX-314, pH ~7.3, ~275 mOsm/l, with ACSF containing 3 mM kynurenic acid to block glutamatergic currents.
Neurons were voltage-clamped in whole-cell configuration using a MultiClamp-700A amplifier (Molecular Devices); all recordings were low-pass-filtered at 3 kHz (8-pole Bessel) and digitized online at 10 kHz (National Instruments PCI-MIO-16E-4 board). Series resistance and whole-cell capacitance were estimated from fast transients evoked by a 5mV step and compensated to 75%. EPSCs and IPSCs were recorded by voltage-clamping sequentially at −70mV and +10mV, respectively. All drugs were purchased from Sigma or Tocris. L-655,708 and SNAP-5114 were dissolved in DMSO then diluted 1:1000 in H2O. NO-711, Gabazine and GABA were dissolved in H2O.
Custom-written macros running under IGOR Pro v.6.0 (WaveMetrics, Inc.) were used to analyze the digitized recordings to determine the values of tonic currents and mean phasic currents, as previously described1. Itonic was recorded as the reduction in baseline holding currents (Ihold) after bath-applying a saturating amount (>100μM) of the GABAAR antagonist SR-95531 (gabazine), while voltage-clamping at +10mV. NO-711, SNAP-5114 and L-655,708 were added to the recording ACSF via perfusion and their effects on Itonic were recorded as the post-drug shift in Ihold. Drug perfusion was continued until the shifting Ihold remained steady for 1–2 min.
To determine the mean phasic current (Imean), a 60-s segment containing either EPSCs or IPSCs was selected, and an all-point histogram was plotted for every 10,000 points (every 1s), smoothed, and fitted with a Gaussian to obtain the mean baseline current. All baseline mean values were then plotted and linear trends subtracted to normalize the mean baseline current to 0pA. After baseline normalization, the values of each 10,000 points (each 1s) were averaged to yield the value of Imean (in pA/s) for each 1s epoch. The averaged Imean value of a 60s segment was reported as the phasic Imean value for either the spontaneous EPSC or IPSC. Synaptic event kinetics (i.e. frequency, peak amplitude, 10–90% rise time, and weighted decay time constant) are analyzed by custom-written LabView-based software (EVAN), as previously described1. For comparison of the IPSC peak amplitudes under control and PTX-treated conditions (Supplementary Table II), the largest-amplitude count-matched method was used, whereby the amplitude values in a given recording were sorted and the largest x number of events under control condition were averaged and taken for comparison with the average of an equally-matched x number of events under the PTX condition, with x determined by the number of events detected under the 10 μM PTX condition. This method circumvents the erroneous comparison of average amplitudes when considering the effects of a receptor antagonist that reduces the smaller events (in control condition) below the noise level.
To estimate Vrest, the cell-attached recording technique2 was used. Briefly, depolarizing voltage ramps (−100 to +200 mV) were applied to cell-attached patches to activate voltage-gated K+ channels and establish the K+ current reversal potential, which provides a measure of the Vrest, given near equimolar K+ inside the cell and the pipette. EGABA was estimated by measuring the K+ reversal potential after activating GABAARs with 50μM muscimol. Recordings were made using a solution containing the following (in mM): K+ gluconate (135), KCl (5), MgCl2 (2), HEPES (10), EGTA (0.1), Na-ATP (4), Na-GTP (0.3), pH 7.3, 273mOsm/l. A junction potential of 9mV was measured and then subtracted from voltage values of all measurements.
The fitting of multiple distributions to a cumulative probability plot (Supplementary Fig. 2) was done as follows. Cumulative probabilities of the variable x (i.e., P(x)) were calculated and fitted by one or more normal curves approximated by the logistic equation3:
where R1, …,Rn are the ratios of the n normal distributions (such as ), , are the individual means, and p1,…,pn are steepness factors related to the n standard deviations (SD1,. . SDn).
The grid-walking apparatus was manufactured as previously described4, using 12mm square wire mesh with a grid area 32cm / 20cm / 50cm (length / width / height). A mirror was placed beneath the apparatus to allow video footage in order to assess the animals' stepping errors (i.e. `footfaults'). Each mouse was placed individually atop of the elevated wire grid and allowed to freely walk for a period of 5min. Video footage was analyzed offline by raters blind as to the treatment groups. The total number of footfaults for each limb, along with the total number of non-footfault steps, was counted, and a ratio between footfaults and total-steps-taken calculated. Percent footfaults were calculated by: [#footfaults / (#footfaults + #non-footfault steps) * 100]. A ratio between footfaults and total steps taken was used to take into account differences in the degree of locomotion between animals and trials. A step was considered a footfault if it was not providing support and the foot went through the grid hole. Further, if an animal was resting with the grid at the level of the wrist, this was also considered a fault. If the grid was anywhere forward of the wrist area then this was considered as a normal step.
The spontaneous forelimb task encourages the use of forelimbs for vertical wall exploration / press in a cylinder5. When placed in a cylinder, the animal rears to a standing position, whilst supporting its weight with either one or both of its forelimbs on the side of the cylinder wall. Animals were placed inside a Plexiglas cylinder (15cm in height with a diameter of 10cm was used) and videotaped for 5min. Videotape footage of animals in the cylinder were evaluated quantitatively in order to determine forelimb preference during vertical exploratory movements. While the video footage was played in slow motion (1/5th real time speed), the time (sec) during each rear that each animal spent on either the right forelimb, the left forelimb, or on both forelimbs were calculated. Only rears in which both forelimbs could be clearly seen were timed. The percentage of time spent on each limb was calculated and these data were used to derive an SFL asymmetry index (% ipsilateral use- % contralateral use). The `contact time' method of examining the behavior was chosen over the `contact placement' method, as described by5, as it takes into account the slips that often occur during a bilateral wall press post-photothrombosis.
Seven days after stroke mice were decapitated, the brains rapidly removed and peri-infarct cortex microdissected and frozen (n=5). The equivalent region of cortex was taken in control, non-operated mice (n=3). Samples were homogenized in radioimmunoprecipitation (RIPA) buffer (Pierce; Rockford, IL) and centrifuged at 20000×g at 4°C for 10 minutes. Supernatant was collected as protein extract and stored at −80°C. Western blot was performed as described5. 100 ug of protein from each sample was diluted in 7.5 ul of 2× SDS- sample buffer gel (Invitrogen; Carlsbad, CA) containing dithiothreitol (DTT) (Sigma, St. Louis, MO) and brought to a final volume of 15 ul with RIPA buffer. Samples were denatured at 95°C, loaded on to a 4–12 % gradient Tris-Glycine gel (Invitrogen; Carlsbad, CA), separated via SDS-PAGE, and then transferred to HYBOND™-P (pvdf) membrane (Amersham; Piscataway, NJ) at 30 volts for 2 hours. Membranes were rinsed and blocked overnight at 4°C. Membranes probed with antibodies against GABA Transporter 3 (Rbt Anti-GAT-3 1:1000; Millipore; Temcula, CA), and GABA Transporter 1 (Rbt Anti-GAT-1 1:200; Millipore; Temecula, CA). Following successive washes, membranes were incubated in IgG Donkey Anti-Rabbit HRP-labeled secondary (1:6000; Jackson; West Grove, PA) for one hour at room temperature. Membranes were incubated in ECL PLUS (Amersham; Piscataaway, NJ)) and chemiluminescence was detected using Fluorochem (Alpha Innotech, San Leandro, CA). Membranes were then re-probed for one hour at room temperature with GAPDH (1:2500; Abcam; Cambridge, MA) and Donkey Anti-Rbt-HRP (1:10000; Jackson; West Grove, PA) as an endogenous control protein to ensure equal loading. Immunoblotting was performed in triplicate for each antibody. Adobe Photoshop software (Adobe Systems Inc, San Jose, CA) was used for densitometric analysis of all blots.
Supported by The Dr. Miriam and Sheldon G Adelson Medical Research Foundation, the Larry L. Hillblom Foundation (A.N.C., S.T.C.), the Coelho Endowment (I.M.) and NIH/NINDS grant NS30549 (I.M.). This manuscript was completed partially during tenure of an American Heart Association Postdoctoral Fellowship, a Repatriation Fellowship from the New Zealand Neurological Foundation and the Sir Charles Hercus Fellowship from the Health Research Council of New Zealand (A.N.C). We thank E.O. Mann, J. Chu, J.J. Overman, J. Zhong, and R.M. Lazaro for discussion and assistance.
Supplementary Information: Supplementary information includes Supplementary Methods, Figures, including a Supplementary Figure 1 summarizing the main result and accompanies this paper on www.nature.com/nature.
Author Information:: Reprints and permissions information is available at npg.nature.com/reprintsandpermissions. The authors have no competing financial interests to declare.