Canceling actions involves a race between basal ganglia pathways

Transcription

1 Cnceling ctions involves rce etween sl gngli pthwys Roert Schmidt 1, Dniel K Leventhl, Nicols Mllet 1,3, Fujun Chen 1 & Joshu D Berke 1 npg 13 Nture Americ, Inc. All rights reserved. Slient cues cn prompt the rpid interruption of plnned ctions. It hs een proposed tht fst, rective ehviorl inhiition involves specific sl gngli pthwys, nd we tested this y compring ctivity in multiple rt sl gngli structures during performnce of stop-signl tsk. Suthlmic nucleus (STN) neurons exhiited low-ltency responses to cues, irrespective of whether ctions were cnceled or not. By contrst, neurons downstrem in the sustnti nigr prs reticult (SNr) only responded to cues in trils with successful cncelltion. Recordings nd simultions together indicte tht this sensorimotor gting rises from the reltive timing of two distinct inputs to neurons in the SNr dorsolterl core suregion: cue-relted excittion from STN nd movement-relted inhiition from stritum. Our results support rce models of ction cncelltion, with stopping requiring -cue informtion to e trnsmitted from STN to SNr efore incresed stritl input cretes point of no return. The ility to suppress nd cncel ctions is core component of cognitive control, nd impirments in this ility contriute to impulsive nd compulsive ehviors including drug ddiction nd ttentiondeficit hyperctivity disorder 1. Action suppression is often ssessed using the stop-signl tsk, which hs een widely pplied in oth humns nd experimentl nimls 3,6 8. On most trils, sujects re given Go cue tht prompts specific, rpid movement. On the remining trils the sme Go cue is followed y signl, indicting tht the sujects should cncel tht movement efore it egins. The intervl etween Go nd cues is djusted so tht sujects sometimes succeed in stopping nd sometimes fil in stopping. In generl, stop-signl tsk performnce is well descried y theoreticl models in which the Go nd cues respectively initite stochstic go nd stop processes tht rce for completion. The outcome of this rce determines whether stopping is successful 3,4. Extensive evidence for involvement of the sl gngli in ction suppression comes from phrmcologicl mnipultions 9,1, lesions 6,11, stimultion 1,13, imging 1,14 nd computtionl modeling,1. It hs een proposed tht the conceptul rce etween go nd stop processes corresponds to literl rce etween distinct neurl pthwys 1, converging on sl gngli output nuclei tht provide tonic inhiition of ctions 16,17. Specificlly, stritl direct pthwy neurons re thought to promote movements (go) y inhiiting SNr, nd STN neurons serve s rke on ehvior (stop) y exciting the sme SNr cells,18 (Fig. 1). Here we tested this hypothesis y compring the fine timing of ctivity in ech sl gngli structure. Our results support the sic notion of rce etween go nd stop processes tht initilly evolve in seprte neurl circuits, nd lso provide evidence for multiple sl gngli mechnisms in ehviorl inhiition 14,19. RESULTS To ssess the correspondence etween distinct sl gngli pthwys nd hypothesized cognitive processes, we pplied the high sptiotemporl resolution of single-unit electrophysiology to rt stop-signl tsk sed round our prior decision-mking studies,1 (Fig. 1 nd Supplementry Tle 1). We trined ech rt to plce its nose in centrl port until the onset of Go cue (1-kHz or 4-kHz tone) tht directed rief lterl hed movement (to the left or the right; Supplementry Video 1). On 3% of trils, this Go tone ws followed y cue (white noise), instructing tht the rt should sty in the centrl port (Fig. 1,c). For oth Go trils nd trils, we rewrded correct performnce y delivery of sugr pellet. As typiclly oserved for stop-signl tsks, rection times for Filed trils corresponded to the fster portion of the rection-time distriution in Go trils (Fig. 1d). This is consistent with rce models: when the go process hppens more quickly, stop process is less likely to suppress ehvior. For our first set of recordings (experiment 1), four well-trined sujects received tetrode implnts tht simultneously trgeted sensorimotor stritum, STN, glous pllidus (corresponding to glous pllidus prs extern in primtes) nd SNr 1. We isolted spikes from individul neurons during tsk performnce, from ech rin region (for ntomicl loctions, see Supplementry Fig. 1). A chllenge when studying ehviorl inhiition is to disentngle neurl ctivity specificlly linked to stopping, rther thn going. To do this, we followed ltency-mtching procedure,3 (Online Methods), which exploits the similrity in rection times, nd thus presumly the go process, etween Filed trils nd trils. We compred the firing rte of ech neuron etween these tril types, nd etween Correct trils nd trils. We then ssessed the frction of ech neuronl 1 Deprtment of Psychology, University of Michign, Ann Aror, Michign, USA. Deprtment of Neurology, University of Michign, Ann Aror, Michign, USA. 3 Centre Ntionl de l Recherche Scientifique, Institut des Mldies Neurodégénértives, Université Victor-Seglen, Bordeux, Frnce. Correspondence should e ddressed to J.D.B. (jderke@umich.edu). Received Mrch; ccepted 31 My; pulished online 14 July 13; doi:1.138/nn VOLUME 16 NUMBER 8 AUGUST 13 nture NEUROSCIENCE

2 Figure 1 Tsk events nd ehvior. () Simplified scheme of neurl circuitry under investigtion during the stop-signl tsk. Projections from stritum (STR) nd STN converge on SNr, which provides tonic inhiition of motor output. GP, glous pllidus. () Configurtion of the opernt chmer with five nose ports on one side nd food port on the opposite side. Entry into ny port is detected y reks of photodiode em (red dshed lines). (c) Tsk events in Go trils nd trils re shown in sequence from left to right. Thick rs indicte occurrence of sensory cues ( udio nd house light ) nd rt position in center nd side ports. Rection time (RT) is mesured etween onset of Go cue nd onset of movement (tht is, nose out of the center port). Movement time (MT) is the time it tkes the rt to go from the center port to the side port. In trils, the stop-signl dely (SSD) is the time etween onset of Go cue nd onset of cue. (d) Rection time distriutions for experiment 1 (rts 1 13, top to ottom). Correct Go trils re shown in lue, nd Filed trils in purple. Note tht Filed trils hve similr rection times to the fster prt of the distriution for Go trils. npg 13 Nture Americ, Inc. All rights reserved. popultion tht exhiited significnt (P <.; shuffle test) differences t ech moment in time (Fig. nd Supplementry Fig. ). Stritl neurons showed little or no fst popultion-level response to the signl. In contrst, oth STN nd SNr contined significnt proportion of neurons with rpid responses to the signl (P <., inomil test; Fig.,). For STN, this proportion ws similr for Correct nd Filed trils (see Supplementry Fig. for comprison etween these tril types) nd thus resemled sensory -like popultion response to the cue. However, for SNr, the cue evoked fst ctivity chnge only in Correct nd not Figure Distinct processing of the cue cross sl gngli components. () Frction of neurons whose firing rte significntly differed etween the tril types under comprison, for indicted rin res during contrlterl trils. To screen for stop-relted ctivity, we compred Correct trils with trils (top), nd Filed trils with trils (ottom). For exmple, movement-relted ctivity ws very similr on nd Filed trils, so it does not show up in this comprison. Activity is ligned to onset of the cue (or for Go trils, the point t which the cue would hve een presented hd it een tril). Upwrd nd downwrd rs denote the frction of units tht fired more on nd Go trils, respectively. Filled rs indicte times when this frction significntly exceeded chnce level (inomil test; P <. with ple rs uncorrected nd drk rs corrected for multiple comprisons; horizontl gry lines mrk these respective significnce thresholds. In this ltency-mtched nlysis, the proportions of STN neurons with fst cue responses were similr in Correct nd Filed trils (for the two STN ins just fter the onset of cue; P =.17 nd.1, shuffle test). By contrst, these proportions were significntly different for SNr (for the two filled red SNr ins, P =.8 nd., shuffle test). () Exmples of individul neuron ctivity in STN nd SNr during Go trils in the four relevnt tril types s indicted (oth ipsilterl nd contrlterl movements re shown). The STN unit showed fst, trnsient increse in ctivity fter the cue in oth Correct nd Frction of units Tril Tril in Filed trils (Fig. ). Thus, lthough the ctivity in STN ws consistent with sensory response, ctivity in SNr insted reflected the ehviorl outcome on ech tril. STR (n = 3) STN (n = 1) SNr (n = 146) GP (n = 14) Time from cue (s) STN (single unit) Correct SNr (single unit) Correct Time from cue (s) Filed trils. On Correct trils, the SNr unit lso exhiited fst increse in firing, nd no movement-linked puse. By contrst, on Filed trils, the SNr unit simply showed movement-linked decrese in firing rte nd no response to the cue, very similr to trils. Filed 1 Correct Filed Filed nture NEUROSCIENCE VOLUME 16 NUMBER 8 AUGUST

3 npg 13 Nture Americ, Inc. All rights reserved. Figure 3 cues increse firing in STN efore SNr. () Firing rte time courses for the neuronl supopultions tht distinguish from Go trils (in contrlterl trils; Fig. nd Online Methods). Colored lines show the men (± s.e.m.) z-score of the firing rte cross units (rt rekdown: 11 nd 16 STN units from rts 1 nd 11, respectively; 14, nd SNr units from rts 1, 1 nd 13, respectively; see Supplementry Tle 1). Horizontl colored rs t the top of ech pnel indicte times with significntly different versus Go firing rtes (shuffle test, P <., corrected for multiple comprisons). Verticl gry rs show SSRTs for the corresponding recording sessions. () Comprison of cue response ltencies for the sme STN (lck) nd SNr (green) units (top, Correct trils; ottom, Filed trils). To id comprison, selines re shifted so tht lowest ctivity is in ll cses t zero. Verticl dshed lines indicte ltency of pek response (STN, 1 ms; SNr, 36 ms); red line mrks shortest SSRT. Insets, distriutions of single unit response ltencies. (c) Pek cue response mplitudes for individul neurons in Correct versus Filed trils (top, STN; ottom SNr; gry lines ± s.e.m.). P vlues were derived from shuffle tests. Our glous pllidus recordings did not yield such unmiguous results. Although the initil screen indicted tht some neurons my selectively respond for Correct s (Fig. ), the direct comprison did not confirm selective glous pllidus response in Correct rther thn in Filed trils (Supplementry Fig. ). We therefore focused on STN nd SNr next. We exmined the time course of ctivity in these stop-relted STN nd SNr neurons. STN neurons responded to the onset of the cue with trnsiently incresed firing (Fig. 3) tht in some cses took the form of just single, precisely timed extr spike (Fig. ). These STN increses hd consistently very low ltencies (pek response ~1 ms; Fig. 3; see ref. 4 for similrly low STN ltencies) tht were not different etween Correct nd Filed trils (P =.41, pired t-test; Fig. 3). The mgnitude of the pek STN response exhiited no consistent preference for Correct versus Filed trils (Fig. 3c). SNr neurons lso incresed firing in response to the cue (Fig. 3) Firing rte (z score) Firing rte (z score) Correct STN (n = 7) Filed.... Time from cue (s) SNr (n = 18) Correct Filed.... Time from cue (s) Numer of units Firing rte (z score) Numer of units Firing rte (z score) STN: 1 ms SNr: 36 ms ut with longer ltency (pek response t ~36 ms; Fig. 3; P =.4 compring STN to SNr ltencies, one-sided Kolmogorov-Smirnov test) nd preferentilly on Correct trils (Fig. 3c). We oserved this ltency difference even when we restricted the nlysis to units recorded in the sme session (n = 1 pirs; STN cells preceded SNr cells y n verge of 13.6 ms, P =.41, shuffle test). All SNr neurons tht responded to the cue on Correct trils did so efore the stop-signl rection time (SSRT; Fig. 3), stndrd, inferred ehviorl mesure for how quick process must e to influence stopping performnce 3,4,7. Thus, SNr ctivity not only distinguished etween Correct nd Filed trils, it did so quickly enough to ffect the tril outcome. Most of the SNr units with fst responses to the cue (1/18) lso mrkedly decresed their ctivity eginning just efore movement (Supplementry Fig. 3). This suggests tht the cue my not lter SNr ctivity glolly, ut rther hve selective influence over Correct (z score) 1 1 Time from cue (ms) Correct Filed STN: 1 ms SNr: 1 1 Time from cue (ms) c Correct (z score) 1 STN: P =.61 1 Filed (z score) SNr: P =.1 Filed (z score) Figure 4 An SNr hotspot for cue responses. () Exmple of silicon proe recording from SNr. Tips of the eight proe shnks were coted in DiO (green) for histologicl visuliztion. One tip is visile here (the others were more nterior nd posterior). Dshed line mrks SNr oundry. () Reconstructed loctions of SNr single units from ll nine rts, on SNr coronl tls oundries 1 (Online Methods). Neurons showing significnt (P <.; shuffle test) differences etween Correct nd Filed trils ( 1 ms fter cue in either ipsilterl or contrlterl trils) re shown in red; others re in cyn. Numers indicte pproximte nterior to posterior coordinte reltive to regm. (c) Functionl mp otined y stcking tls sections. Note the dorsolterl cluster of outcome-dependent SNr units (1, 11 nd 3 units from rts 11, 1 nd 18, respectively; Supplementry Tle 1). This cluster ws oserved when either ipsi- or contrlterl movements hd to e stopped nd lso in ltencymtched control comprisons (Supplementry Fig. 4 d). (d) Representtive unit from the hotspot (from rt 1, mrked y white x in ) showing similr ctivity ptterns to SNr units from experiment 1. µm c x d Rte (Hz) Tril Rte (Hz) Tril Correct Filed.. Time from cue (s) 11 VOLUME 16 NUMBER 8 AUGUST 13 nture neuroscience

4 Frction of units.. Stritum (n = ) Contrlterl Ipsilterl Firing rte (z score) 3 1 Correct Filed Stritum (n = 74) c Firing rte (z score) 3 1 Correct Filed.... Time from Go cue (s) Time from movement (s).. Time from Go cue (s).. Time from movement (s).... Time from cue (s) Time from cue (s) Figure Vrile timing of stritl go process criticlly determines whether stopping is successful. () Frctions of stritl units distinguishing etween contr- nd ipsilterl movements, t ech time point during Go trils. Lyout is s in Figure. On the right, lck solid rs efore the red dshed line indicte significnt (P <., shuffle test) coding of movement direction efore the onset of movement. The 74 units tht contriuted to these rs were considered potentil contriutors to go process (rt rekdown:, 3, 11 nd 3 units from rts 1 13, respectively; Supplementry Tle 1). () Men (± s.e.m.) firing rte z score for these 74 stritl units. Cyn r t the top indictes times with significntly different firing rtes on Fst versus trils, lue r indictes the sme for versus Correct trils, nd purple r for versus Filed trils (shuffle test, P <., corrected for multiple comprisons). (c) Activity of the sme stritl units ligned to the cue. Note the different time scle thn in. Formt is s in Figure 3. npg 13 Nture Americ, Inc. All rights reserved. cells nd suregions involved in controlling the movement tht needs to e inhiited. We therefore recorded from second set of sujects (experiment ) using oth high-density silicon proes (in three rts, Fig. 4) nd more tetrodes (in two rts) to trget wide rnge of SNr loctions. Comining ll SNr results together reveled cler hotspot of SNr cells tht distinguished Correct trils from Filed trils, riefly fter the cue (Fig. 4 nd Supplementry Fig. 4 d). This hotspot corresponds remrkly well to the SNr sensorimotor core suregion tht hs een descried in ntomicl studies, locted dorsolterlly nd extended long the rostrl-cudl xis. This suregion projects to specific prts of the superior colliculus involved in orienting movements,6, so the signl influences ctivity in n SNr suregion tht is likely criticl for exerting fst ehviorl control 7. The distinct ltencies of STN nd SNr cue responses re consistent with stop informtion eing conveyed long the STN-SNr pthwy. Yet the selectivity of this trnsmission to Correct trils suggests some form of gting mechnism. In other words, given tht the glutmtergic STN cells spike on Filed trils, why re SNr neurons not responsive to this input? The nswer my lie in the movementrelted firing-rte decreses of SNr neurons (Supplementry Figs. 3 nd 4e,f). Such SNr firing puses re well known from studies of eye nd lim movements 7,8, nd re thought to fcilitte ction through disinhiition of superior colliculus nd other structures more directly linked to motor output 16,9. SNr puses re driven y incresed firing of the GABAergic stritl direct pthwy neurons 17,3, plusile prticipnts in go process. To ssess how stritl neurons my contriute to movement preprtion nd initition, we looked for units tht distinguish movement direction efore onset of movement. We found n rupt increse in contrlterl coding strting ~14 ms efore movements (Fig. ; see Supplementry Fig. for nlyses of other rin regions). When we compred the ctivity of these directionselective stritl neurons (74 cells) etween different tril types, we oserved rpid ccelertion of firing rte just efore the onset of movement 31 (Fig. ) tht followed the sme trjectory for, nd Filed trils. Aligned on the erlier Go cue, this stritl ctivity remined very similr etween nd Filed trils ut distinct to nd Correct trils (Fig. ). These results fit well with simple rce model, in which vriility in the timing of stritl-sed go process determines the outcome on trils. On Filed trils, movement-relted stritl ctivity hs lredy egun to increse y onset of cue (Fig. c nd Supplementry Fig. 6). This effect ws prticulrly pronounced when exmining individul presumed stritl projection neurons (Supplementry Fig. 6). Therefore, the lck of SNr responses to the cue on Filed trils my e due to the erly rrivl of stritl GABAergic input, shunting wy the effects of glutmtergic inputs from the STN. To confirm the viility of this ide, we studied gting of the cue responses in simple integrte-nd-fire model of n SNr neuron. This neuron received excittory pulses, mimicking STN sensory responses to cues, nd (s in prior sl gngli models 3,33 ) this excittory input ws influenced y GABAergic inhiition 34,3 (Online Methods). For GABAergic input, we used the verge stritl popultion ctivity during initition of movement (Fig. ) to pproximte rel input ptterns. We djusted synptic strengths of inhiitory nd excittory inputs to provide good qulittive mtch with the cue-evoked increses nd movement-relted decreses in SNr firing. A criticl prmeter in the model is the reltive timing of excittion nd inhiition. We defined s the intervl etween the cue onset cue nd the point in the stritl output t which movements egn on Go trils. If the cue egn long efore initition of movement (Fig. 6; = ms), stritl inhiition ws low t tht time nd the cue evoked full response in the SNr cell. In contrst, if the cue occurred only riefly efore initition of movement (Fig. 6; = ms), high level of stritl inhiition suppressed the SNr cue response. A systemtic vrition of in the ehviorlly relevnt rnge yielded gting curve tht quntified the model response to the cue (Fig. 6,c). The gting phenomenon required strong divisive inhiition, for exmple, through shunting inhiition, rther thn simple summtion of inhiitory nd excittory conductnces (Fig. 6c). We then used the ehviorl dt of ech rt to estimte the ctul distriution of for oth Correct nd Filed trils (Supplementry Fig. 7). Then we could use these distriutions to clculte model firing rtes for these tril types (Fig. 6d). Just s for rel rt SNr cells, the model SNr cell selectively responded to the cue in Correct trils ut not in Filed trils. We conclude tht the integrtion of distinct excittory nd inhiitory synptic inputs y individul SNr neurons provides strightforwrd, mechnistic ccount of how go nd stop processes cn rce in the rin. nture NEUROSCIENCE VOLUME 16 NUMBER 8 AUGUST

5 npg 13 Nture Americ, Inc. All rights reserved. Figure 6 Modeling sensorimotor gting in SNr neurons. () Model responses for two illustrtive vlues of, the intervl etween onset of cue nd onset of movement. Red nd green lines indicte STN nd stritl (STR) inputs to the SNr model, nd lue line shows the output firing rte of the model SNr cell. Note the cler SNr response to the cue with = ms ut not with = ms. () SNr model responses to the cue over rnge of. For smll, strong shunting inhiition from stritum prevents STN-evoked spiking to the cue (white rrow). (c) Comprison etween model output with nd without the cue, mesured in the ms fter STN input reches SNr (left). Enlrged view of the gry re (right), for the rnge of where gting of cue occurs. Red line is the sme s on the left; other lines show the effects of different levels of shunting inhiition (low vlues of the reference current J correspond to strong divisive inhiition; Online Methods). In ll cses the lines indicte the difference etween model SNr firing rte with nd without the cue. Note tht without shunting inhiition the model does not gte the cue s oserved in the experimentl SNr dt. (d) Model SNr output (colored lines) exhiits response to the cue in Correct trils (top) ut not in Filed trils (ottom). Filed nd Correct trils in the model re sed on rt rection time dt (Supplementry Fig. 7). Blck histogrm shows one exmple rt SNr cell for qulittive comprison to the model. Note tht in the model, incresed firing to the cue response ws lwys followed y movement-relted decrese s, for simplicity, we did not incorporte our oservtion tht stritl output is susequently suppressed on Correct trils (Fig.,c). DISCUSSION Rce models hve een centrl to theories of ction suppression for decdes, yet cler evidence tht they ctully descrie neurl processes hs een elusive. Here we demonstrted tht ctivity in two key sl gngli pthwys for ction control closely resemles rce etween go nd stop processes. Individul SNr neurons exhiited oth movement-relted puses in firing (driven y stritum) nd rpid increses in firing rte fter cues (driven y STN), nd the reltive timing of these influences corresponded to whether stopping ws successful. These SNr cells re locted in specific dorsolterl suregion, tht projects to colliculr intermedite lyers importnt for the control of orienting movements,6. Furthermore, the evidence we found for shunting inhiition of STN inputs y stritl inputs egins to revel how mechnisms operting in single cells cn contriute to sensorimotor gting. Neurons in STN nd SNr with fst cue responses lso incresed spiking with the Go cue tht instructed contrlterl movement (Supplementry Figs. 3 nd 8). Thus, the STN-SNr pthwy does not solely convey signls tht instruct stopping ut lso other tskrelevnt cues. The effect of oth Go nd cues ws to trnsiently increse firing of popultion of SNr neurons tht decrese firing with the onset of movement. We found tht trils in which STN nd SNr responded more strongly to the Go cue hd longer rection times (Supplementry Fig. 9), consistent with role for STN-SNr trnsmission in delying ehviorl output rther thn cusing outright stopping (see elow). A rpid, utomtic inhiition of ehviorl responses y tsk-relevnt cues my help prevent responses tht re impulsive or premture (tht is, when preprtion for movement is incomplete) nd lso explin why even cues tht instruct sujects not to stop ut insted continue s plnned result in longer rection times 36. c Conductnce (.u.) Conductnce (.u.) STR STN SNr Go Go = ms = ms Time from cue (ms) With Without Difference 4 (ms) Rte difference (Hz) 3 1 Shunting inhiition: None J =. na J = 1. na J = 1. na 3 (ms) (ms) Rte (Hz) 3 3 Time from cue (ms) d 1 Our results contriute informtion to the ongoing dete out whether certin rin res contriute to ction inhiition versus cue-evoked reorienting of ttention 36 38, indicting tht these functions re not necessrily distinct. The very low, fixed ltency of cue-evoked ctivity in STN is informtive in severl wys. First, rce models often incorporte vrile timing of oth go nd stop processes, yet we found tht the STN response to the cue occurred t the sme time in Correct nd Filed trils. Thus, if these responses re prt of stop process, performnce vriility rises directly from the vrile timing of the go process (corresponding to vrile rection times), t lest in this version of the stop-signl tsk. This result is consistent with recent simultions of sl gngli networks during inhiitory control 39 : STN provides the sme fst signl to puse ction, whether stopping is ctully successful or not. Second, incresed STN spiking within just 1 ms of the onset of cue constrins the sophistiction of prior informtion processing, nd which fferents cn drive this response. Recent studies in humns hve emphsized the role of frontl corticl inputs to STN in ction suppression 1,14, ut it is not cler tht cue informtion cn e pssed quickly enough through cortex to cuse this fst STN spiking. Our implementtion of the stop-signl tsk encourged very quick responses nd my hve incresed the importnce of sucorticl pthwys tht support sensory processing nd fst orienting-like movements. In prticulr, mny neurons in the thlmic intrlminr complex (centromedin nd prfsciculr nuclei, CM-Pf) nd pedunculopontine nucleus hve short-ltency responses to slient uditory stimuli 4,41 nd project to rnge of sl gngli trgets, including the STN 4,43. CM-Pf projections to stritum re importnt for ehviorl switching nd lerning fter unexpected cues Correct Filed Time from cue (ms) 11 VOLUME 16 NUMBER 8 AUGUST 13 nture neuroscience

6 npg 13 Nture Americ, Inc. All rights reserved. We hypothesize tht the STN responses we oserved re one component of roder interrupt system, medited y CM-Pf nd/or pedunculopontine nucleus, tht coordintes response to slient cues cross multiple timescles using multiple pthwys (Supplementry Fig. 1). In this scheme, very fst yet trnsient excittion of STN nd SNr serves to dely ctions tht re close to execution, similr to previous descriptions of STN uying time during decision-mking 47. However, the STN-driven increse in SNr firing is highly trnsient; in our simultion it delyed, ut did not fully cncel, the stritum-driven decrese in firing tht releses movements (Fig. 6e). We lso found evidence tht second, slower mechnism my ct in the stritum to help shut down the go process. Movement-relted stritl ctivity ruptly decresed in Correct trils, compred to trils (Fig.,c nd Supplementry Fig. 6), nd similr suppression of contrlterl-coding stritl ctivity hs een oserved in n ntisccde tsk 48. Such suppression of go process is key feture of interctive rce models of stop-signl performnce 4. Yet stritum-sed processing y itself is unlikely to ccount for stopsignl performnce, s the reduction in stritl output ws not consistently efore the SSRT (Fig. c). It thus ppers tht complementry mechnisms llow ction suppression to e oth fst (vi STN) nd selective (vi stritum) 19,39,48. Future studies will investigte how the stritl go process is suppressed in Correct trils. Direct pthwy neurons cn e inhiited in mny wys, nd some (non-exclusive) possiilities include the influences of indirect pthwy cells 39,48,49 nd cholinergic interneurons 4,46. We used sic stop-signl tsk, designed to investigte rective spects of ehviorl inhiition (responding to cue). This tsk does not ssess ll the complexities of ehviorl inhiition, such s proctive components (tht is, prepredness to stop). It hs een proposed tht proctive inhiition involves yet nother sl gngli circuit, the indirect pthwy from stritum through glous pllidus 19. In follow-up studies, we pln to investigte whether systemticlly vrying prepredness to stop revels cler role for glous pllidus tht ws not pprent here. Finlly, we hd previously reported 1 (using the experiment 1 dt) tht slient tsk cues cuse rpid reset of et oscilltory phse throughout the sl gngli, whether or not the cues ctully direct ehvior on given tril. By contrst, cue-induced increses in et power only occur for cues tht re used, for exmple, fter the cue on Correct ut not Filed trils. This distinction corresponds closely to the difference etween STN nd SNr descried here: events tht cused rupt increses in STN firing lso produced restet of et phse, wheres those events tht dditionlly incresed SNr firing susequently produced increses in et power. Furthermore, there is evidence tht other oscilltory frequencies such s delt nd thet cn influence the prmeters of ehviorl control, such s decision thresholds 47. An importnt direction for future investigtion will e to determine the mechnistic reltionships etween rpid firing rte chnges nd ltered dynmic sttes in sl gngli circuitry. Methods Methods nd ny ssocited references re ville in the online version of the pper. Note: Supplementry informtion is ville in the online version of the pper. Acknowledgments We thnk V. Stuphorn, D. Weissmn, A. Aron, G. Morris, M. Churchlnd, M. Bevn nd D. Meyer for their helpful comments. J. Pettione nd A. Cse provided vlule ssistnce. This work ws supported y Deutsche Forschungsgemeinschft grnt SCHM 74/1-1, the US Ntionl Institute on Drug Ause, Ntionl Institute on Neurologicl Disorders nd Stroke, nd the University of Michign. AUTHOR CONTRIBUTIONS J.D.B. designed nd oversw the project. D.K.L. helped develop the ehviorl tsk. D.K.L., N.M. nd F.C. performed electrophysiologicl experiments. R.S. developed nd performed the dt nlyses nd computtionl modeling. R.S. nd J.D.B. wrote the mnuscript. COMPETING FINANCIAL INTERESTS The uthors declre no competing finncil interests. Reprints nd permissions informtion is ville online t reprints/index.html. 1. Aron, A.R. & Poldrck, R.A. Corticl nd sucorticl contriutions to signl response inhiition: role of the suthlmic nucleus. J. Neurosci. 6, (6).. Frnk, M.J. Hold your horses: dynmic computtionl role for the suthlmic nucleus in decision mking. Neurl Netw. 19, (6). 3. Logn, G.D., Cown, W.B. & Dvis, K.A. On the ility to inhiit simple nd choice rection time responses: model nd method. J. Exp. Psychol. Hum. Percept. Perform. 1, (1984). 4. Boucher, L., Plmeri, T.J., Logn, G.D. & Schll, J.D. Inhiitory control in mind nd rin: n interctive rce model of countermnding sccdes. Psychol. Rev. 114, (7).. Roins, T.W., Gilln, C.M., Smith, D.G., de Wit, S. & Ersche, K.D. Neurocognitive endophenotypes of impulsivity nd compulsivity: towrds dimensionl psychitry. Trends Cogn. Sci. 16, (1). 6. Egle, D.M. et l. -signl rection-time tsk performnce: role of prefrontl cortex nd suthlmic nucleus. Cere. Cortex 18, (8). 7. Hnes, D.P. & Schll, J.D. Neurl control of voluntry movement initition. Science 74, (1996). 8. Osmn, A., Kornlum, S. & Meyer, D.E. The point-of-no-return in choice rectiontime controlled nd llistic stges of response preprtion. J. Exp. Psychol. Hum. Percept. Perform. 1, 43 8 (1986). 9. Hikosk, O. & Wurtz, R.H. Effects on eye-movements of GABA gonist nd ntgonist injected into monkey superior colliculus. Brin Res. 7, (1983). 1. Bunez, C. et l. Effects of STN lesions on simple vs choice rection time tsks in the rt: preserved motor rediness, ut impired response selection. Eur. J. Neurosci. 13, (1). 11. Bergmn, H., Wichmnn, T. & DeLong, M.R. Reversl of experimentl prkinsonism y lesions of the suthlmic nucleus. Science 49, (199). 1. Bllnger, B. et l. Stimultion of the suthlmic nucleus nd impulsivity: relese your horses. Ann. Neurol. 66, (9). 13. Mjid, D.S., Ci, W., George, J.S., Verruggen, F. & Aron, A.R. Trnscrnil mgnetic stimultion revels dissocile mechnisms for glol versus selective corticomotor suppression underlying the stopping of ction. Cere. Cortex, (1). 14. Jhfri, S. et l. Effective connectivity revels importnt roles for oth the hyperdirect (fronto-suthlmic) nd the indirect (fronto-stritl-pllidl) frontosl gngli pthwys during response inhiition. J. Neurosci. 31, (11). 1. Frnk, M.J., Smnt, J., Moustf, A.A. & Shermn, S.J. Hold your horses: impulsivity, deep rin stimultion, nd mediction in prkinsonism. Science 318, (7). 16. Hikosk, O. & Wurtz, R.H. Visul nd oculomotor functions of monkey sustnti nigr prs reticult. I. Reltion of visul nd uditory responses to sccdes. J. Neurophysiol. 49, (1983). 17. Krvitz, A.V. et l. Regultion of prkinsonin motor ehviours y optogenetic control of sl gngli circuitry. Nture 466, 6 66 (1). 18. Bevn, M.D., Bolm, J.P. & Crossmn, A.R. Convergent synptic input from the neostritum nd the suthlmus onto identified nigrothlmic neurons in the rt. Eur. J. Neurosci. 6, (1994). 19. Aron, A.R. From rective to proctive nd selective control: developing richer model for stopping inpproprite responses. Biol. Psychitry 69, e e68 (11).. Gge, G.J., Stoetzner, C.R., Wiltschko, A.B. & Berke, J.D. Selective ctivtion of stritl fst-spiking interneurons during choice execution. Neuron 67, (1). 1. Leventhl, D.L. et l. Bsl gngli et oscilltions ccompny cue utiliztion. Neuron 73, 3 36 (1).. Stuphorn, V., Brown, J.W. & Schll, J.D. Role of supplementry eye field in sccde initition: executive, not direct, control. J. Neurophysiol. 13, (1). 3. Brown, J.W., Hnes, D.P., Schll, J.D. & Stuphorn, V. Reltion of frontl eye field ctivity to sccde initition during countermnding tsk. Exp. Brin Res. 19, (8). nture NEUROSCIENCE VOLUME 16 NUMBER 8 AUGUST

7 npg 13 Nture Americ, Inc. All rights reserved. 4. Cheruel, F., Dormont, J.F. & Frin, D. Activity of neurons of the suthlmic nucleus in reltion to motor performnce in the ct. Exp. Brin Res. 18, 6 (1996).. Deniu, J.M., Milly, P., Murice, N. & Chrpier, S. The prs reticult of the sustnti nigr: window to sl gngli output. Prog. Brin Res. 16, (7). 6. Pre, M. & Hnes, D.P. Controlled movement processing: superior colliculus ctivity ssocited with countermnded sccdes. J. Neurosci. 3, (3). 7. Hndel, A. & Glimcher, P.W. Quntittive nlysis of sustnti nigr prs reticult ctivity during visully guided sccde tsk. J. Neurophysiol. 8, (1999). 8. Schultz, W. Activity of prs reticult neurons of monkey sustnti-nigr in reltion to motor, sensory, nd complex events. J. Neurophysiol., (1986). 9. Bsso, M.A. & Sommer, M.A. Exploring the role of the sustnti nigr prs reticult in eye movements. Neuroscience 198, 1 (11). 3. Alin, R.L., Young, A.B. & Penney, J.B. The functionl-ntomy of sl gngli disorders. Trends Neurosci. 1, (1989). 31. Lo, C.C. & Wng, X.J. Cortico-sl gngli circuit mechnism for decision threshold in rection time tsks. Nt. Neurosci. 9, (6). 3. Humphries, M.D. & Gurney, K.N. A pulsed neurl network model of ursting in the sl gngli. Neurl Netw. 14, (1). 33. Humphries, M.D., Stewrt, R.D. & Gurney, K.N. A physiologiclly plusile model of ction selection nd oscilltory ctivity in the sl gngli. J. Neurosci. 6, (6). 34. Blomfield, S. Arithmeticl opertions performed y nerve cells. Brin Res. 69, (1974). 3. Segev, I. Dendritic processing. in The Hndook of Brin Theory nd Neurl Networks. (ed. Ari, M.A.) 8 89 (MIT Press, 1998). 36. Shrp, D.J. et l. Distinct frontl systems for response inhiition, ttentionl cpture, nd error processing. Proc. Ntl. Acd. Sci. USA 17, (1). 37. Shulmn, G.L. et l. Interction of stimulus-driven reorienting nd expecttion in ventrl nd dorsl frontoprietl nd sl gngli-corticl networks. J. Neurosci. 9, (9). 38. Levy, B.J. & Wgner, A.D. Cognitive control nd right ventrolterl prefrontl cortex: reflexive reorienting, motor inhiition, nd ction updting. Ann. NY Acd. Sci. 14, 4 6 (11). 39. Wiecki, T.V. & Frnk, M.J. A computtionl model of inhiitory control in frontl cortex nd sl gngli. Psychol. Rev. 1, 39 3 (13). 4. Mtsumoto, N., Minmimoto, T., Gryiel, A.M. & Kimur, M. Neurons in the thlmic CM-Pf complex supply stritl neurons with informtion out ehviorlly significnt sensory events. J. Neurophysiol. 8, (1). 41. Pn, W.X. & Hylnd, B.I. Pedunculopontine tegmentl nucleus controls conditioned responses of midrin dopmine neurons in ehving rts. J. Neurosci., (). 4. Deschenes, M., Bourss, J., Don, V.D. & Prent, A. A single-cell study of the xonl projections rising from the posterior intrlminr thlmic nuclei in the rt. Eur. J. Neurosci. 8, (1996). 43. Kit, T. & Kit, H. Cholinergic nd non-cholinergic mesopontine tegmentl neurons projecting to the suthlmic nucleus in the rt. Eur. J. Neurosci. 33, (11). 44. Kimur, M., Minmimoto, T., Mtsumoto, N. & Hori, Y. Monitoring nd switching of cortico-sl gngli loop functions y the thlmo-stritl system. Neurosci. Res. 48, 3 36 (4). 4. McHffie, J.G., Stnford, T.R., Stein, B.E., Coizet, W. & Redgrve, P. Sucorticl loops through the sl gngli. Trends Neurosci. 8, (). 46. Ding, J.B., Guzmn, J.N., Peterson, J.D., Golderg, J.A. & Surmeier, D.J. Thlmic gting of corticostritl signling y cholinergic interneurons. Neuron 67, (1). 47. Cvngh, J.F. et l. Suthlmic nucleus stimultion reverses mediofrontl influence over decision threshold. Nt. Neurosci. 14, (11). 48. Wtne, M. & Munoz, D.P. Neurl correltes of conflict resolution etween utomtic nd volitionl ctions y sl gngli. Eur. J. Neurosci. 3, (9). 49. Cui, G. et l. Concurrent ctivtion of stritl direct nd indirect pthwys during ction initition. Nture 494, 38 4 (13).. Zndelt, B.B. & Vink, M. On the role of the stritum in response inhiition. PLoS ONE, e13848 (1). 1. Pxinos, G. & Wtson, C. The Rt Brin in Stereotxic Coordintes th edn. (Elsevier Acdemic Press, ). 114 VOLUME 16 NUMBER 8 AUGUST 13 nture neuroscience

8 npg 13 Nture Americ, Inc. All rights reserved. ONLINE METHODS Experimentl procedures. Behviorl electrophysiology methods hve een previously descried in detil,1,. All niml experiments were pproved y the University of Michign Committee for the Use nd Cre of Animls. Sujects were dult mle Long-Evns rts, housed on 1:1 reverse light:drk cycle nd tested during the drk phse. Rts were housed in groups of 3 4 with moderte environmentl enrichment (toys, vriety of edding, 9 cm 39 cm cm cges) during presurgicl trining, then were singly housed fter surgery. The opernt chmer hd five nose-poke holes on one wll nd food dispenser on the opposite wll. At the strt of ech tril, one of the three more-centrl holes (chosen rndomly) ws illuminted, indicting tht the rt should poke nd hold its nose in tht port. After vrile hold dely ( 1, ms), tone (Go cue; 6 db, ms) instructed the rt to move promptly into the djcent hole either to the left (1-kHz tone) or right (4-kHz tone); correct choices triggered immedite delivery of sugr pellet rewrd (signled y n udile click of the food dispenser). To encourge rts to respond quickly, on Go trils rts hd to leve the initil port within limited hold period, nd then poke the djcent hole within movement hold (Supplementry Tle 1). On trils (3%), the Go cue ws followed fter short dely (the stop-signl dely, SSD) y cue (white noise urst, 6 db, 1 ms). If the rt moved efore the SSD, the cue ws not plyed nd the tril ws treted s Go tril. To successfully complete tril, the rt hd to mintin its nose in the initil port until the limited hold period would hve expired on Go tril. At tht point, the udile click of the food dispenser signled rewrd delivery. Errors of ny type produced time-out (house light on for 8 s). Otherwise, the next tril ws initited fter the rt otined its rewrd. The computer-controlled sequence of trils ws rndomized nd experimenters were linded to the tril sequence. However, to further discourge the rt from dopting holding strtegy, the rt hd to perform correct Go tril efore tril could occur. Other rndomiztion or linding procedures were not performed in this study. After chieving stle tsk performnce (typiclly ~ 3 months of trining, >7% correct choices on Go trils) rts in experiment 1 received implnts contining 1 individully drivle tetrodes trgeting sl gngli structures (stritum (STR), glous pllidus, STN nd SNr), nd recording sessions egn ~1 week lter. SSD ws held constnt in ech recording session to fcilitte electrophysiologicl nlyses ut ws djusted etween stop-signl sessions so tht Correct nd Filed trils were pproximtely equl in numer. On lternte dys, rts performed the stop-signl tsk nd go/no-go tsk, which ws identicl in most respects to the stop-signl tsk ut with n SSD of zero (on No-Go trils, the white noise ws plyed insted of Go cue). During tsk performnce, wide-nd (1 9, Hz) rin signls were recorded continuously t 31. khz. Tetrodes were usully moved y t lest 8 µm etween two stop-signl sessions. In some cses (for exmple, if the numer of trils ws low in one recording session), tetrodes were not moved etween sessions, nd we only included the etter session in the nlysis. Individul neurons were isolted offline using wvelet-sed filtering followed y stndrd mnul spike-sorting, nd clssifiction into different presumed cell types 3. No sttisticl methods were used to predetermine smple sizes. In experiment we exmined whether stop responses re loclized to specific suregion of SNr. To fcilitte systemtic functionl mpping, three rts received 8-shnk, 64-chnnel silicon proes (Neuronexus Inc.) in the SNr. Silicon proe shnks were coted in the lipophilic dye DiO efore implnttion nd not moved fter initil surgery. For these fixed-loction silicon proes we only included dt from single session per rt, to void including duplicte cells. Two dditionl rts received similr tetrode implnts s in experiment 1. Loctions of SNr single units were reconstructed on the pulished SNr coronl tls oundries 1. Dt nlysis. All nlyses were performed using custom Mtl routines. Dt distriution ws not formlly tested for normlity, ut we insted mostly used sttisticl methods tht re roust for non-norml-distriuted dt (Kolmogorov- Smirnov nd shuffling tests). SSRTs were estimted for ech session individully y the integrtion method 4, s follows. First, we determined the percentge of Filed trils f. Then we clculted the stopping time s the f-th percentile of the distriution of Go tril rection times. The SSRT is then the stopping time minus the SSD 4. This SSRT vlue ws lso used to seprte Go trils in the sme session into nd. To exmine the ctivity of ech neuron we used 4-ms time ins (sliding in steps of ms) to otin spike-count distriutions for different tril types, ner key tsk events. A neuron hd to exceed 3-Hz firing rte in t lest one time in to e included in susequent nlyses. To est isolte ctivity ssocited with cues, we compred tril types for which the ctivity ssocited with movement preprtion is most similr. Tht is, we compred Filed trils with trils, nd Correct trils with trils,3 (ltency mtching). For trils, neurl ctivity ws ligned to the onset of the cue, nd for Go trils, we used the time t which the signl would hve occurred, tht is, Go cue onset + SSD. For few Filed trils, rection times were very long (Fig. 1d). We interpreted these s trils for which the initil stopping ws ctully successful ut susequent holding on for rewrd ws not (oth spike nd LFP mesures were consistent with this interprettion; dt not shown). We therefore excluded Filed trils with rection times > ms from ll nlyses. To compre whether spike rtes were different etween two tril types, we used shuffle test for ech time in. We shuffled the tril type lels 1, times, nd for ech shuffle we compred the mens of the two resulting spike count distriutions. To otin P vlue, we determined the frction of shuffles in which the difference etween the shuffled mens ws lrger (or smller) thn the difference etween the two ctully oserved mens. We used P vlue of. to determine significnt coding of tril types. It follows tht % of popultion of rndomly ctive units should, on verge, e clssified s coding (ll flse positives). A inomil test ws then used to determine whether the empiriclly mesured frction of coding units ws significntly higher thn tht expected y chnce. We corrected for multiple testing with respect to the overll timewindow round tsk events (for exmple, for Fig., ech single test ws done for 4-ms time window, yielding /4 independent tests round the tsk events). This correction is overly conservtive for some key time points of interest, s we hypothesized priori tht firing-rtes would chnge shortly fter cue onset. Therefore we lso (in Fig. ) indicted times when the P vlue ws elow. without djusting for multiple comprisons. The nlyses shown in Figure 3 include only STN nd SNr units identified s stop-relted. For STN, we included units tht contriuted to the significnt responses in either Correct or Filed trils (filled red nd mgent rs for STN in Fig. ). As the overll popultion of SNr units did not rech significnce in Filed trils, we included only those individul SNr units tht contriuted to the significnt cue response in Correct trils (filled red rs for SNr in Fig. ). The firing rte time course of ech unit ws then estimted y verging spike counts over trils of the sme type (for exmple, Filed right trils) with sliding ms window in steps of ms round importnt tsk events (for exmple, the cue; Fig. 4), nd then smoothing with three-point verge. To compre men firing rte time courses etween tril types, ctivity of ech unit ws first trnsformed to z scores using the men nd s.d. of session-wide firing rte estimtes (otined from 1-s-wide windows). To compre the mgnitude of the cue response etween Correct nd Filed trils (Fig. 3c), we used pek firing rtes in the rnge 1 7 ms fter onset of cue. To more precisely identify the times t which STN nd SNr neurons responded to the cue (Fig. 3), we used non-overlpping 3-ms time windows smoothed with threepoint verge for the firing rte estimtion. The ltency ws tken s the time of pek firing within the rnge 1 7 ms fter onset of cue. Units were mrked in red (in Fig. 4 nd Supplementry Fig. 4 d) if the corresponding tril types were significntly different in t lest two out three ins (ech in ws 4 ms wide, centered t 4, 6 nd 8 ms fter the stop cue). For the dt shown in Figure 4, we compred tril types seprtely for ipsilterl nd contrlterl conditions nd mrked units red if in one or oth cses the ctivity differences were significnt. Similrly, in Supplementry Figure, unit ws counted in the histogrm if it exhiited significnt difference in ctivity etween Filed nd Correct trils for ipsilterl or contrlterl trils. Computtionl modeling. To descrie the gting of the cue response, we implemented n integrte-nd-fire model of single SNr cell. Chnges in the memrne potentil V t time t were given y 46 dv t m = V( t) + E m r m I( t) dt where τ m is the memrne time constnt, E m the memrne resting potentil, nd r m the memrne input resistnce. I is sum of currents: I (t) = I (t) + I STR (t) + I STN (t). The current I ws used to mimic the high spontneous ctivity oserved in SNr cells: I (t) = g (V (t) E ) (see elow for prmeter settings) 3,. Synptic doi:1.138/nn.346 nture NEUROSCIENCE

9 npg 13 Nture Americ, Inc. All rights reserved. input from the stritum ws summrized y I STR (t) = g STR (t) (V(t) E STR ). The time-dependent synptic conductnce g STR (t) ws modeled using the ctivity (t) of go-relted stritl cells during movement initition shown in Figure on the right, verged over slow nd trils, s: g STR ( t ) h ( t ) ( i ) = i = The scling fctor η controls the movement-relted decrese in our model SNr cell nd ws fitted to reproduce the verge time course in Supplementry Figure 4. This provided n importnt constrint on the overll strength of stritl inhiition. Although further increses in stritl input strength could lock STN inputs without the need for shunting inhiition, this chnge lso produced too erly puse in SNr firing (tht is, mismtch with the timing of stritl nd SNr t chnges oserved in vivo). The sum ( i) i = is over previous ctivity in the current tril (strting ms efore onset of movement) nd is used s simplified description of fcilittion found t stritonigrl synpses 6. In our model this fcilittion effectively led to roder pek of the stritl ctivity rmp nd therey to roder puse in SNr ctivity, s we oserved in the experimentl dt. Although rel SNr neurons puse t rnge of times reltive to the onset of movement (Supplementry Fig. 3 nd ref. 9), for our simple model we used fixed stritl input time course. We further ssumed tht STN input to SNr consisted of single spike t time t s evoked y the Go or cue (see STN exmple cell in Fig. ). Ech spike gve rise to modulted lph function yielding synptic current I STN (t) = νh(t)α(t,t s )(V(t) E STN ) with scling fctor ν nd shunting fctor h(t). The scling fctor ν controls the mplitude of the SNr response to the STN input. We choose ν to fit the mplitude of SNr responses to the Go cue. Owing to the high seline ctivity in SNr medited y I, in some trils STN inputs cn shift the spike timing rther thn increse the SNr firing rte. When the SNr seline ctivity is diminished y inhiitory input, this effect is diminished so tht STN input ecomes more likely to evoke n SNr spike (gry line in Fig. 6c on the right). The lph function ws zero for t < nd otherwise defined s ( t, ts) = exp( ( t ts)/ q) exp( ( t ts)/. q) with time constnt q. The shunting fctor h(t) tkes vlues etween zero nd one depending on the mount of inhiitory current 3,33 : I ( t) I ( t) h( t) = 1 STR J Θ 1 STR J, where Θ denotes the Heviside step function. The reference current J controls the efficcy of the shunting inhiition (Fig. 6c). A spike ws generted y the model SNr cell if V(t) reched the threshold voltge V thr = mv. After one millisecond, V(t) ws then reset to the equilirium memrne potentil E m = 8 mv. Other prmeter vlues were τ m = ms; r m = 1 MΩ; g =.9 µs; E = mv; E STR = 1 mv; E STN = mv; η = 1/1,; ν =.1; q = 1 ms; nd J = 1 na, 1. na nd na for strong, medium nd wek shunting inhiition, respectively. Further, complex effects on synptic integrtion of GABA A were neglected in our model s for simplicity we set E STR < E m. t However, we performed dditionl simultions with more relistic vlues for shunting inhiition where E m < E STR < V thr. With ppropritely rescled vlues for η nd J, the results remined the sme. Therefore, the key feture of inhiition ws the divisive vetoing effect 34,3 on excittory STN input. For the simultion without shunting inhiition h(t) ws lwys set to 1. In the model we kept the SSD constnt t 3 ms, nd ssumed tht STN input reched SNr strting 3 ms fter cue onset. The SNr response to the cue ws mesured s the firing rte within the susequent -ms time window. The intervl etween presenttion of cue nd onset of movement ws vried in 1-ms time steps, over the rnge of ms. For ech, trils with nd trils without the cue were simulted nd the verge firing rtes were then used for the results. To otin SNr output for Filed trils, we used the ehviorl dt of the rts. From the mesured rection times of Filed trils, we sutrcted the SSD for tht session. The resulting distriution of -ligned rection times (Supplementry Fig. 7) ws used s the model prmeter for Filed trils. For Correct trils, no direct rection time mesure is ville. However, using the rection time distriution on Go trils, it is possile to estimte wht the rection time would hve een if the cue hd not een presented. More formlly, for ech rt we hd n empiriclly mesured rection time distriution for Go trils F go (t) = P(RT = t), with P denoting proility nd RT eing ligned rection times in -ms-wide ins. With the corresponding proility distriution for Filed trils F FS, the hypotheticl distriution of correct stop rection times ws estimted s F CS FGo ( t) pffs ( t) ( t) = 1 p where p denotes the overll proility of filing to stop (numer of Filed trils divided y the numer of trils). Owing to noise in the ehviorl dt in few cses the estimtes of F CS were negtive, so we pplied lower ound for F CS (t) of zero nd rescled the whole proility distriution to mintin totl proility of 1. The resulting estimtes of F CS were then used s the distriution of for Correct trils.. Wiltschko, A.B., Gge, G.J. & Berke, J.D. Wvelet filtering efore spike detection preserves wveform shpe nd enhnces single-unit discrimintion. J. Neurosci. Methods 173, 34 4 (8). 3. Berke, J.D. Uncoordinted firing rte chnges of stritl fst-spiking interneurons during ehviorl tsk performnce. J. Neurosci. 8, (8). 4. Verruggen, F. & Logn, G.D. Models of response inhiition in the stop-signl nd stop-chnge prdigms. Neurosci. Bioehv. Rev. 33, (9).. Gerstner, W. & Kistler, W.M. Spiking Neuron Models (Cmridge University Press, ). 6. Connelly, W.M., Schulz, J.M., Lees, G. & Reynolds, J.N.J. Differentil short-term plsticity t convergent inhiitory synpses to the sustnti nigr prs reticult. J. Neurosci. 3, (1). nture NEUROSCIENCE doi:1.138/nn.346