The auditory cortex mediates the perceptual effects of acoustic temporal expectation

Transcription

1 A r t i c l e s co m p u tt i o n n d s y st e m s The uditory cortex medites the perceptul effects of coustic temporl expecttion Sntigo Jrmillo & Anthony M Zdor 211 Nture Americ, Inc. All rights reserved. When events occur t predictle instnts, nticiption improves performnce. Knowledge of event timing modultes motor circuits nd therey improves response speed. By contrst, the neuronl mechnisms tht underlie chnges in sensory perception resulting from expecttion re not well understood. We developed ehviorl procedure for rts in which we mnipulted expecttions out sound timing. Vlid expecttions improved oth the speed nd the ccurcy of the sujects performnce, indicting not only improved motor prepredness ut lso enhnced perception. Single-neuron recordings in primry uditory cortex showed enhnced representtion of sounds during periods of heightened expecttion. Furthermore, we found tht ctivity in uditory cortex ws cuslly linked to the performnce of the tsk nd tht chnges in the neuronl representtion of sounds predicted performnce on tril-y-tril sis. Our results indicte tht chnges in neuronl representtion s erly s primry sensory cortex medite the perceptul dvntge conferred y temporl expecttion. Attention to moments in time is powerful cognitive mechnism for exploiting temporl structure in ehviors such s hunting moving prey or plying music in n ensemle. Anticiption of n event cn influence the speed of ehviorl response s well s our perception of the event 1, ut it is not known how these improvements in perception rise from chnges in neuronl ctivity. However, it is cler tht orgnisms cn tke dvntge of regulrities in the environment to form expecttions nd predictions tht cn e used to enhnce performnce. The uditory cortex is sensitive to mny forms of coustic regulrity 2. For exmple, corticl neurons respond more strongly to rrely presented sounds emedded within regulr sequence, even when these sounds re not required to perform tsk 3. In ddition, chnges in expecttion out the frequency of tsk-relevnt sound modulte the ctivity of single neurons in the uditory cortex 4. However, lthough temporl structure is centrl to the orgniztion of sounds, our understnding of the effects of temporl expecttion in the uditory system is lrgely limited to studies in humns 1,5, in whom it is difficult to study the underlying neuronl mechnisms. In prticulr, we know little out whether nd how uditory temporl expecttion chnges neuronl representtions in the uditory cortex. Improved performnce from temporl expecttion could rise from enhnced motor prepredness ut could lso e due to perceptul enhncement. We hve developed ehviorl procedure in rts to study the neuronl mechnisms tht underlie the perceptul consequences of temporl expecttion. Here we chrcterize the effect of expecttion out the timing of sounds on oth the speed nd ccurcy of performnce in n niml model. This procedure llows us to ssess chnges in perception due to expecttion, to estlish cusl link etween neuronl ctivity nd perception nd to quntify chnges in sensory representtion tht result from temporl expecttion. We show tht temporl expecttion modultes ctivity in primry sensory cortex, nd tht the influence of expecttion on neuronl ctivity in sensory cortex is directly relted to the oserved improvements in performnce. Our results suggest tht improvements in uditory perception tht result from vlid temporl expecttion rise from chnges in sensory representtions s erly s the first stges of corticl processing. RESULTS We crried out three sets of experiments designed to evlute the effects of temporl expecttion on sensory res tht medite the perception of sounds. First, we performed ehviorl experiments in which we quntified the effect of temporl expecttion on the perception of sounds y mesuring the rection time nd ccurcy of rts performing detection nd discrimintion tsk. Second, we used reversile lesions to test the role of the uditory cortex in this tsk. Third, we used electrophysiologicl methods to mesure the ctivity of single neurons in the primry uditory cortex of nimls performing the temporl expecttion tsk. These mesurements llowed us to chrcterize the modultion of neuronl responses y expecttion, nd to ssess the reltionship etween the modultion of neurl ctivity nd chnges in ehvior. Vlid temporl expecttion improved performnce We developed n uditory two-lterntive choice procedure, modeled fter relted visul procedure in humns 6, in which we could study the effect of temporl expecttion on performnce in rts (Fig. 1). Sujects were rewrded for correctly discriminting the crrier frequency of frequency-modulted trget sound immersed in puretone distrctors (Fig. 1). To perform this tsk the suject hd oth to detect the trget nd to discriminte its frequency. The prmeters of this tsk cn e mnipulted to mke either the detection or the discrimintion component, or oth, ritrrily difficult. Cold Spring Hror Lortory, Cold Spring Hror, New York, USA. Correspondence should e ddressed to A.M.Z. (zdor@cshl.edu). Received 11 My; ccepted 12 Octoer; pulished online 19 Decemer 21; doi:1.138/nn VOLUME 14 NUMBER 2 FEBRUARY 211 nture NEUROSCIENCE

2 r t i c l e s 211 Nture Americ, Inc. All rights reserved. Strt tril Stimulus 31 khz Rewrd In the experiments descried here, the frequency discrimintion component ws designed to e esy, so tht errors were minly due to filures of detection. We mnipulted detection difficulty y vrying trget modultion depth (TMD), so tht n unmodulted trget (TMD = ) ws indistinguishle from pure tone. On ech tril, we presented the trget either erly (3 or 45 ms fter sound onset) or lte (1,35 or 1,5 ms fter sound onset). We mnipulted temporl expecttion y presenting trils in locks of t lest 15 trils: expecterly locks consisted of 85% trils with erly trgets nd 15% with lte trgets, wheres expect-lte locks consisted of 85% lte nd 15% erly trgets (Fig. 1c). Performnce depended on the suject s expecttion out the moment of occurrence of the trget. Vlid expecttion led to etter performnce, s ssessed y either rection time (Fig. 2,) or ccurcy (Fig. 2c,d; P =.78, pired Wilcoxon signed-rnk test). Even for the esiest difficulty tested, where the frction of correctly detected trgets ws well ove 9%, sujects responded more quickly to erly trgets when these were expected (Fig. 2,). Rection time for difficult trils, which is gretly influenced y the correct detection of trget, ws even more sensitive to the sujects expecttion (Supplementry Figs. 1 nd 2). We focused our nlysis on trils Frequency c P trget Erly Expect-erly lock Expect-lte lock 6.5 khz 6 1,2 1,8 Time (ms) Figure 1 Tsk nd mnipultion of temporl expecttion. () Rts initited ech tril y nose poke into the center port of the opernt chmer 12,36. After vrile (25 35 ms) silent period, stimulus consisting of frequency-modulted trget in trin of pure tone distrctors ws presented. Rts were required to sty in the port until the trget ws presented. The center frequency of the trget (6.5 khz or 31 khz) indicted the side port where wter rewrd would e delivered on ech tril (left or right, respectively). () The stimulus consisted of sequence of 1-ms pure tones (5 4 khz) seprted y 5 ms, which ws presented for s long s the niml styed in the port. The frequency-modulted trget ws presented in plce of one of the tones in ech tril. (c) Temporl expecttion ws mnipulted y chnging the rtio of trils with erly or lte trgets within ech lock of 15 2 trils. Figure 2 Vlid temporl expecttion improved performnce. () Behviorl responses were fster on trils with expected trgets. Exmple of the rection time distriution from one rt t the esiest difficulty tested (TMD 25%) for erly trgets tht were expected (lue) or unexpected (green). Rection time ws defined s the time etween the onset of trget nd the moment when the rt left the center port. () Medin rection time for ech rt (dots) on the esiest difficulty tested, nd verge cross ll eight rts for erly trgets tht were expected (lue) or unexpected (green). (c) Behviorl responses were more ccurte on trils with expected trgets. Exmple for one rt of the percentge of correct trils s function of difficulty, vried here y the modultion depth of the trget (TMD). Error rs correspond to the 95% confidence intervls on estimtes. The dshed line corresponds to 75% performnce used for clculting thresholds in d. (d) Modultion depth needed to chieve 75% correct trils for ech rt (dots) nd verge cross ll eight rts (colored rs). **P <.1. Lte with erly trgets ecuse, in the sence of n erly trget, ll lte trgets cn ecome eqully expected (Supplementry Fig. 3). The effect of expecttion on ccurcy lso vried with difficulty. For the esiest stimuli, performnce depended only slightly on expecttion, ut for intermedite difficulties, temporl expecttion improved performnce y s much s 15% (Fig. 2c,d). Using modified stimulus, we verified tht sujects were relying on the detection of the trget sound nd not dopting lterntive strtegies for improving their performnce (Supplementry Fig. 4). These oservtions demonstrte the influence of temporl expecttion not only on the speed of ehviorl responses ut lso on the perception of trget sounds. Inctivtion of uditory cortex decresed performnce Hving estlished n uditory procedure for studying the mechnisms of temporl expecttion, we next ssessed whether the uditory cortex ws essentil for performing this tsk. Previous studies hve shown tht nimls cn perform vrious uditory tsks even fter ilterl ltion of uditory cortex 7,8, presumly y mens of prllel pthwys etween sucorticl uditory res nd motor regions. To minimize the possiility of corticl reorgniztion following permnent lesion 9,1, we inctivted the uditory cortex reversily y pplying the GABA A receptor gonist muscimol ilterlly (Fig. 3). Inctivtion of the uditory cortex y muscimol impirs performnce on vrious uditory discrimintion tsks without ffecting performnce on olfctory discrimintion tsks 11. Muscimol inctivtion impired the performnce of ll rts (Fig. 3) compred with interleved control sessions in which sline ws used (P = , pired Wilcoxon signed-rnk test), lthough there ws considerle vriility etween rts in the mgnitude of the effect. For the esiest difficulty tested, verge rection times incresed significntly from 196 ± 18 ms (men ± s.e.m.) in sline sessions to 317 ± 76 ms in muscimol sessions. In some rts, the effect of inctivtion ws strong enough to reduce performnce to chnce levels t ll tested stimulus difficulties (Supplementry Fig. 5), even though these sujects continued to perform hundreds of trils per session s in control sessions (Supplementry Fig. 6). This reduction in performnce supports the hypothesis tht the uditory cortex hs n essentil role in this tsk. Frction of trils c Correct trils (%).2.1 Expected Unexpected Rection time (ms) 1 Expected 9 Unexpected TMD (%) Rection time (ms) d TMD for 75% correct % 8% 4% 2% 1% Exp. Exp. ** ** Unexp. Unexp. nture NEUROSCIENCE VOLUME 14 NUMBER 2 FEBRUARY

3 r t i c l e s 211 Nture Americ, Inc. All rights reserved. AC Muscimol Skull AC Temporl expecttion modulted neuronl ctivity We hypothesized tht the enhnced performnce conferred y vlid temporl expecttion would e correlted with chnges in the neurl representtion of coustic stimuli in uditory cortex. To test this hypothesis, we used tetrodes to record responses from single neurons in the primry uditory cortex of three rts performing the temporl expecttion tsk t n intermedite difficulty. To ensure controlled delivery of uditory stimuli to the unrestrined rt regrdless of the position of its hed 12, sounds were delivered through erphones during recording sessions (see Online Methods). We focused our initil nlysis on the neuronl responses to the tones tht preceded the trget, to understnd the effects of trget nticiption. To minimize the effects of coustic context (the recent history of sounds 13,14 ), we fixed the frequency of the first three tones for ll trils in given session. We compred responses elicited y these tones etween two expecttion conditions: expect-erly locks, in which the suject expected the trget to occur erly in the trin of sounds, nd expect-lte locks, in which the trget ws expected to occur lte. Responses to tones tht immeditely preceded n erly trget in expect-erly locks were enhnced compred to responses to identicl Correct trils (%) Sline Muscimol TMD (%) Figure 3 Inctivtion of uditory cortex decresed performnce. () Bilterl reversile inctivtion of uditory cortex (AC) ws performed y pplying the GABA A receptor gonist muscimol to the surfce of the exposed dur mter. Crniotomies were protected y implnted wells. Drker- nd lighter-colored regions indicte primry nd secondry uditory cortices, respectively 37. () Performnce on expected erly trgets s function of difficulty on interleved inctivtion (gry) nd control (lck) sessions. The plot shows men ± s.e.m. for five rts. tones in expect-lte locks (Fig. 4). This modultion occurred in neurons with vrious types of response dynmics: onset, sustined or offset (Fig. 4 nd Supplementry Fig. 7). The popultion of responsive neurons showed n enhncement in evoked ctivity (P = , pired Wilcoxon signed-rnk test), s indicted y the distriution of modultion indices (Fig. 4). The low sound-evoked firing rtes of uditory corticl neurons tht re typiclly oserved in the sence of creful neuron-y-neuron stimulus optimiztion tends to oscure smll chnges in firing rte 15,16. Even so, this enhncement reched significnce (P <.5, Wilcoxon rnk-sum test) in lmost one-third (14/44) of the neurons. Only single neuron showed significnt chnge in the opposite direction (Fig. 4). A qurter of the responsive neurons (11/44) showed significnt increse of t lest 5% in evoked response (P <.5, Wilcoxon rnk-sum test); no neuron hd decrese in response of this mgnitude. On verge, there ws no significnt chnge in spontneous firing rte etween conditions for these cells (Supplementry Fig. 8). Evoked locl field potentils (LFPs) lso showed stronger responses to stimuli tht immeditely preceded the expected trget (Fig. 4c,d). The modultion of evoked responses to sounds occurring round the expected ppernce of trget indictes tht informtion out current expecttions cn enhnce the representtion of stimuli in erly stges of corticl processing. We then nlyzed the effect on responses to the trget itself. A similr nlysis of the 26 neurons tht fulfilled our criteri for responsiveness did not show systemtic effect of temporl expecttion on responses to the trget stimulus (Supplementry Fig. 9). At the end of the section elow we provide n explntion for this oservtion. Modultion ws specific to driven ctivity To understnd how the modultion y temporl expecttion depended on the frequency preference of ech neuron, we performed new set of experiments in which we fixed the frequency of only the first two of the three tones tht preceded the trget nd llowed the frequency of the third tone to vry rndomly cross trils. This modified stimulus llowed more ccurte estimtion of the frequency tuning of ech neuron (Supplementry Fig. 1). From the dditionl recording sessions under these stimulus conditions, 58 cells fulfilled our selection criteri for responsiveness. As ove, there ws no significnt chnge in spontneous firing rte etween conditions for these cells (Supplementry Fig. 8). Figure 4 Temporl expecttion modulted neuronl ctivity in the uditory cortex. () Responses of single neuron to the sme sequence of tones under two temporl expecttion conditions: expecting n erly (lue) or lte (red) trget. Expected erly trgets ppered fter 45 ms, wheres expected lte trgets (not visile here) were presented fter 1,5 ms. Trils re ligned to the onset of the first tone (gry verticl line) for the spike rster (top) nd the peristimulus time histogrm for ech condition (ottom). The session included more thn two locks of trils, ut ll expect-erly or expect-lte locks re grouped together here for illustrtion. The frequency nd durtion of ech tone re indicted y the gry oxes. The gretest difference in evoked ctivity is seen for the tone tht immeditely preceded the erly trget. The stimuli presented fter 45 ms re not the sme on ech tril; verge responses for these stimuli re shown s dshed lines. () Modultion index of 44 responsive cells recorded during sessions in which ll tones tht preceded the erly trgets hd fixed frequencies. A positive modultion index indictes stronger response on expect-erly trils. Cells with sttisticlly significnt modultion (P <.5, Wilcoxon rnk-sum test) re shown in lck. The gry tringle indictes the men of the modultion index. The white tringle shows the modultion index for the exmple in. (c) Evoked locl field potentil (men ± s.e.m.) t one recording site. The onset of the erly trget is indicted y the lue tringle. (d) Modultion of locl field potentils. Colors s in. Expect erly Expect lte Firing rte (spikes per s) c Averge LFP (µv) khz 8.4 khz 24 khz Trget 31 khz 8.4 khz 24 khz Rndom Expect erly Expect lte Time from sound onset (ms) 24 khz Expect erly 5 Expect lte Time from sound onset (ms) (LFP E LFP L )/(LFP E +LFP L ) Numer of sites Numer of cells d Modultion index (R E R L )/(R E +R L ) 248 VOLUME 14 NUMBER 2 FEBRUARY 211 nture neuroscience

4 r t i c l e s 211 Nture Americ, Inc. All rights reserved. Evoked response (spikes per s) Expect erly Expect lte Frequency (khz) *** * * 1. 4 A priori, the modultion might depend on the reltionship etween the neuron s preferred frequency nd the frequency of either the stimulus or the trget. To test the first possiility, we nlyzed the chnge in firing with temporl expecttion s function of the difference etween the stimulus frequency nd the preferred frequency. On verge, the difference in firing rtes etween conditions for non-preferred stimuli were smller thn those for the preferred frequency of ech cell (Fig. 5), confirming the importnce of the reltionship etween the neuron s preferred frequency nd the stimulus frequency. To test the second possiility, we ssessed whether the mgnitude of the modultion ws lrger for neurons whose preferred frequency ws close to one of the trgets. On verge, we found no such reltionship (Supplementry Fig. 11). Furthermore, within the frequency resolution of our stimuli, temporl expecttion hd no effect on the frequency preference of cells (Supplementry Fig. 12). The dependence of the effect of temporl expecttion on the stimulus chrcteristics, together with the lck of modultion when no stimulus ws presented, re consistent with specific chnge in the responsiveness of cells tht represent the stimulus, rther thn non-specific increse in firing rte. In light of these findings, we re-nlyzed the results of the first set of experiments to ssess whether we could ccount for the lck of modultion of responses to the trgets. We hypothesized tht this pprent lck of modultion rose from mismtch etween the frequency of the trget sounds nd the preferred frequency of the recorded cells. We otined corse estimte of the neuron s preferred frequency y quntifying its response to tones tht preceded lte trgets. We found tht the mgnitude of the temporl expecttion effect ws correlted (ρ =.6, P =.1, t-test on trnsformed correltion) with the size of mismtch etween the preferred nd trget frequencies (Supplementry Fig. 13). Thus, neurons did show n increse in evoked response for expected trgets ut only when the trget mtched the neuron s preferred frequency. Becuse in mny cses the trget nd preferred frequency filed to mtch, the overll modultion of the trget ws smll nd filed to rech significnce. This oservtion might ccount for the lck of modultion of responses to trget sounds descried ove. In ddition to the lock-y-lock comprisons presented ove, we tested whether there ws n increse in evoked response s the expected time of the presenttion of trget pproched within tril. In this cse, we nlyzed the neurl responses elicited y the long trin of rndom tones tht preceded lte trget. This llowed us to estimte the frequency tuning of ech cell t vrious moments within tril, lthough on ny given tril the specific rndom sequence of frequencies cn exert strong effect 13,14. On verge, the response to tones t the preferred frequency of ech neuron incresed s Normlized response.5 Expect erly Expect lte Freq. reltive to preferred (octves) Figure 5 Modultion of neuronl ctivity ws specific to driven ctivity. () Frequency tuning of single cell, estimted from responses to the third tone in ech expecttion condition (Supplementry Fig. 1). Ech point represents the men ± s.e.m. firing rte for ech tone frequency. *P <.5, ***P <.1, Wilcoxon rnk-sum test. () Averge frequency tuning of 58 cells recorded with third tone of rndom frequency. Individul tuning curves were ligned ccording to the preferred frequency of ech cell nd normlized efore verging. Ech point indictes the men ± s.e.m. cross neurons. the expected time of the trget pproched (Fig. 6). Note tht this effect is in the direction opposite to the rte-dependent suppression generlly oserved when presenting trin of sounds of equl relevnce 12,17,18. Neuronl ctivity in uditory cortex predicted performnce The cusl role of the uditory cortex in our tsk, together with the modultion of ctivity ssocited with chnges in the suject s cognitive stte, suggested correltion, on tril-y-tril sis, etween neuronl ctivity nd the ehviorl improvements resulting from temporl expecttion. Becuse of the inry-choice nture of our tsk, sustntil proportion of rewrded trils might hve een correct guesses fter missed trget detections. As consequence, n nlysis sed solely on rewrded versus not-rewrded trils did not yield significnt differences. To improve our estimtes of correct detections, we used rection times s surrogte mesure of the niml s internl stte. Slow ehviorl responses were ssocited with missed detections (Supplementry Fig. 2). We clculted the verge neuronl response to stimuli tht preceded erly trgets on trils when the suject rected quickly, nd compred this with the verge from trils with slow rections. Neuronl responses in the uditory cortex to the sme stimulus were stronger on trils in which the sujects responded quickly (Fig. 7). This difference in neuronl response ws consistent cross the popultion of LFPs (P < 1 5, pired Wilcoxon signed-rnk test, Fig. 7) nd single cells (P =.18, pired Wilcoxon signed-rnk test, Fig. 7c). An nlysis of the correltion etween LFP mgnitude nd rection time supports these results. Aout one-fifth of recording sites showed significnt negtive correltion etween the LFP mgnitude nd rection time when we included ll trils with erly trgets (P <.5, t-test on trnsformed correltion). In ddition, the verge correltion cross ll sites (ρ LFP,RT =.3 ±.1, men ± s.e.m.) ws significntly different from zero (P =.1, Wilcoxon signed-rnk test) in the direction tht indicted fster rection times on trils with stronger evoked LFPs. These oservtions show tht, even for identicl stimuli, the suject s ehviorl response is relted to neuronl ctivity in sensory cortex. DISCUSSION We trined rts to perform n uditory tsk in which expecttions out the timing of sounds were mnipulted. We quntified the influence of temporl expecttion on the rts performnce of the tsk, s well s on the ctivity of single neurons in uditory cortex. In ddition, we Evoked response (spikes per s) ms 1,5 ms Frequency (khz) 4 Normlized resp. to PF * * * ** ** ** 1, Time from trget (ms) Figure 6 Neuronl response incresed s the expected moment of the trget pproched. () Frequency tuning of cell in Figure 5 s the time of the lte trget pproches. For ech time slot preceding the lte trget, the estimted tuning curve is plotted in different color. The color r shows the time of ech time slot with respect to the trget onset. () Neuronl response to ech cell s preferred frequency (PF) s the time of the trget pproches. Responses were normlized with respect to the response to the fourth tone ( 1,5 ms from lte trget onset). Ech point corresponds to the medin cross cells, with error rs proportionl to the medin solute devition. Only responses to the preferred frequency for ech cell re shown, nd only cells with responses ove spontneous firing for ech time slot were included (n = 2). *P <.5, **P <.1 with respect to response to fourth tone; pired Wilcoxon signed-rnk test. nture NEUROSCIENCE VOLUME 14 NUMBER 2 FEBRUARY

5 r t i c l e s Figure 7 Neuronl ctivity in uditory cortex ws correlted with ehviorl performnce. () Averge evoked LFP from single electrode for trils with expected erly trgets grouped ccording to rection time. Averges re tken over those trils with the 2% fstest (lue solid) or the 2% slowest (lue dshed) rection times. Evoked LFP for trils with lte trgets is shown in red for comprison. The light-colored nds surrounding ech trce indicte the s.e.m. cross trils. The stimulus (third tone with fixed frequency cross trils) is indicted Averge LFP (µv) khz Fst (expect-erly) Slow (expect-erly) Expect-lte trils Time from sound onset (ms) y the gry r, nd the onset time of n erly trget is represented y the lue tringle. () Evoked LFPs were lrger on trils with fster ehviorl responses. Difference in evoked LFP mgnitude etween trils with the fstest nd slowest ehviorl responses. The difference is quntified y modultion index etween the verge response on fst (LFP F ) nd slow (LFP S ) trils for ech recording site (n = 59). Sites with significnt difference (P <.5, permuttion test) re shown in lck. The gry tringle indictes the men modultion index. (c) Evoked spiking ctivity ws higher on trils with fster ehviorl responses. Difference in evoked ctivity on single cells etween trils with the fstest nd slowest ehviorl responses. The difference is quntified y modultion index etween the verge response on fst (R F ) nd slow (R S ) trils for ech cell (n = 44). Cells with significnt difference (P <.5, Wilcoxon rnk-sum test) re shown in lck. The gry tringle indictes the men modultion index. Only sessions in which the third tone hd the sme frequency on ll trils were included in this nlysis. Numer of sites Modultion index (LFP F LFP S )/(LFP F +LFP S ) cnumer of cells Modultion index (R F R S )/(R F +R S ) 211 Nture Americ, Inc. All rights reserved. evluted the reltionship etween neuronl ctivity in the uditory cortex nd ehvior y mesuring the effects of reversile inctivtion, nd y correlting evoked ctivity with performnce. We found tht: (i) vlid temporl expecttion led to improvements in oth the speed nd ccurcy of responses on n uditory detection nd discrimintion tsk; (ii) the uditory cortex hd cusl role in the performnce of this tsk; (iii) neuronl ctivity in the primry uditory cortex ws correlted with the ehviorl performnce of the sujects on tril-y-tril sis; nd (iv) temporl expecttion enhnced the neuronl response of cells in the primry uditory cortex. Our results suggest tht temporl expecttion improves uditory perception y modulting the response properties of single cells t erly corticl stges of processing. Temporl expecttions in ction nd perception Optiml processing of signls requires exploiting the sttisticl structure of such signls. This is key principle in the design of rtificil systems 19,2. In nture, evolution fvors the survivl of sensory processing systems whose properties re mtched to the sttisticl structure of their environment 21. A good exmple is the dpttion of retinl gnglion cells to the intensity sttistics of visul stimuli so s to mke full use of the cells dynmic rnge 22. Temporl structure on different timescles is n importnt form of regulrity in coustic signls 2. For instnce, neurons in the uditory cortex respond more strongly to rrely presented stimuli ( oddlls ) emedded in regulr sequence 3, ehvior tht cn e understood y positing tht these neurons hve dpted to the recent sttistics of their inputs nd tht they preferentilly represent sounds tht re outliers with respect to these recent sttistics. In n oddll experiment the sttistics re simple nd cn e estimted over minutes. On longer time-scles, speech processing requires us to mke predictions on the sis of the sttisticl structure of lnguge lerned over months or yers. In principle, temporl regulrities could e exploited to improve performnce in t lest two wys. First, they might e used to prime the motor system to respond quickly, leding to decrese in rection time ut not to n improvement in response ccurcy. Alterntively, they could e used to enhnce sensory processing t pproprite moments, improving the ccurcy nd lso possily the speed of responses. Although severl studies hve shown improved speed consistent with enhnced motor prepredness 6,23, improved ccurcy is typiclly oserved only in perceptully demnding tsks 1,24. In our study, regulrities in trget timing occurred over locks of either short-dely or long-dely trils. Rts exploited these regulrities to improve oth speed (Fig. 2,) nd ccurcy (Fig. 2c,d nd Supplementry Fig. 4). Thus we conclude tht in our study vlid temporl expecttion led to improved sound processing, which might lso hve een ssocited with incresed motor prepredness. Neuronl mechnisms of temporl expecttion We hypothesized tht the oserved improvements in perception cused y temporl expecttion resulted from chnges in the sensory representtion of sounds. Becuse rin res in the uditory processing strem my e recruited to differing degrees depending on the ehvior under study 7,8,1, we first ssessed whether the uditory cortex ws essentil for this tsk. Inctivtion mrkedly reduced performnce (Fig. 3), confirming role for the uditory cortex. Indeed, the impirment ws so lrge tht it prevented us from evluting the effect of inctivtion on improvements in performnce due to temporl expecttion. The results of our inctivtion experiments, together with potentil functionl projections into uditory cortex from rin regions tht provide expecttion or timing signls 25 27, suggested tht the uditory cortex would e promising re in which to look for neurl correltes of temporl expecttion. Single-unit tetrode recordings from the primry uditory cortex of ehving rts showed cler enhncement of ctivity evoked y stimuli close to tsk-relevnt moment (Fig. 4). This oservtion shows tht temporl expecttion not only influences lte stges of processing 28,29 ut lso cn enhnce the representtion of stimuli s erly s primry sensory cortex. These chnges in neuronl response re not limited to the representtion of rewrds 3 ut occur even for non-trget sounds tht occur ner the expected moment of relevnt stimulus. Temporl nticiption might e expected to rise the overll level of neuronl ctivity in non-specific wy. However, we found no chnge in spontneous firing rte, indicting tht the oserved chnges were specific to driven ctivity, s often seen with visul ttention 31,32. Furthermore, the difference in firing rte etween the expected nd unexpected conditions ws gretest ner the neuron s preferred frequency (Fig. 5), indicting tht the ctivity of neurons involved in the representtion ws preferentilly enhnced. Our results re rodly consistent with multiplictive increse in response ove seline 31,32, ut ecuse of our reltively corse frequency smpling, we cnnot ssess whether the gin ws constnt cross frequencies. These oservtions suggest tht the top-down circuit nd synptic mechnisms tht underlie temporl expecttion might overlp with those tht medite other ttentionl phenomen. The modultion of ctivity descried here should not e interpreted s evidence tht the uditory cortex helps to keep the time of 25 VOLUME 14 NUMBER 2 FEBRUARY 211 nture neuroscience

6 r t i c l e s 211 Nture Americ, Inc. All rights reserved. relevnt events t the scle of seconds, s is oserved, for exmple, in the prietl cortex 33. Insted, we interpret our results s showing tht the uditory cortex chnges its representtion of sounds ccording to timing informtion it receives from other res. In ddition to the strong modultion of responses to tones tht preceded the tsk-relevnt trget, our dt lso suggest n enhncement, leit sutler, of responses evoked y the trget sound itself (Supplementry Fig. 13). The fct tht this enhncement ws less pronounced thn tht of the responses to the preceding tones ws proly consequence of the detils of our experimentl design. The smll numer of unexpected trgets of prticulr frequency, in ddition to the vrile durtion of the trget sound (ecuse sounds stopped once the suject withdrew from the center port), ffected the reliility of estimtes of responses to trget sounds. Nevertheless, responses to oth trgets nd non-trgets were consistent with selective increse in response strength for stimuli close to the expected moment of the trget. Cusl role for uditory cortex We hve descried the influence of temporl expecttion on ehvior s well s on neuronl signls relted to sensory stimuli, ut these results lone do not link neuronl ctivity to improvements in performnce. Our nlysis of correltions etween response mgnitude nd performnce (Fig. 7) closes this gp. Our oservtions indicte tht the rt s response is relted to neurl ctivity in sensory cortex even for identicl stimuli, s hs een seen in rin res tht re cuslly relted to perceptul decisions 34. This correltion is consistent with the ide tht signls in uditory cortex re used y the suject to detect the occurrence of trget, nd tht chnges in these signls cn generte improvements in perception. We hve developed n niml model in which to study the effects of temporl expecttion on sound-driven ehviors. This procedure llowed us to study the neuronl mechnisms tht underlie expecttion, eyond wht is currently possile in humn studies 1,5,35. Tking dvntge of this niml model, we hve dvnced our understnding of the effects of temporl expecttion in the following directions: first, we showed tht temporl expecttion produced chnges in the representtion of sensory stimuli, in ddition to the oserved modultion of motor circuits. Second, we showed tht these effects on sensory representtion occurred s erly s primry sensory cortex, nd relted this modultion to the physiologicl fetures of single cells. Lst, we provided evidence for cusl link etween neuronl ctivity nd ehvior, nd relted improvements in performnce to chnges in neuronl ctivity. Methods Methods nd ny ssocited references re ville in the online version of the pper t Note: Supplementry informtion is ville on the Nture Neuroscience wesite. Acknowledgments We thnk G.H. Otzu, A.E. Bker, J.M. Aolfi nd B.J. Burch for ssistnce with preliminry studies. This reserch ws supported y postdoctorl fellowship from the Swrtz Foundtion nd y grnts from the Swrtz Foundtion, the US Ntionl Institutes of Helth nd the Mrie Roertson Fund. AUTHOR CONTRIBUTIONS S.J. nd A.M.Z. designed the project nd wrote the mnuscript. S.J. collected the dt nd performed the nlyses. COMPETING FINANCIAL INTERESTS The uthors declre no competing finncil interests. Pulished online t Reprints nd permissions informtion is ville online t reprintsndpermissions/. 1. Nore, A., Corre, A. & Coull, J. The hzrds of time. Curr. Opin. Neuroiol. 17, (27). 2. Winkler, I., Denhm, S.L. & Nelken, I. Modeling the uditory scene: predictive regulrity representtions nd perceptul ojects. Trends Cogn. Sci. 13, (29). 3. Ulnovsky, N., Ls, L. & Nelken, I. Processing of low-proility sounds y corticl neurons. Nt. Neurosci. 6, (23). 4. Fritz, J., Shmm, S., Elhilli, M. & Klein, D. Rpid tsk-relted plsticity of spectrotemporl receptive fields in primry uditory cortex. Nt. Neurosci. 6, (23). 5. Lnge, K., Rosler, F. & Roder, B. Erly processing stges re modulted when uditory stimuli re presented t n ttended moment in time: n event-relted potentil study. Psychophysiology 4, (23). 6. Nore, A.C. Orienting ttention to instnts in time. Neuropsychologi 39, (21). 7. Ohl, F.W., Wetzel, W., Wgner, T., Rech, A. & Scheich, H. Bilterl ltion of uditory cortex in mongolin geril ffects discrimintion of frequency modulted tones ut not of pure tones. Lern. Mem. 6, (1999). 8. Syk, J., Rylko, N., Mzelová, J. & Drug, R. Gp detection threshold in the rt efore nd fter uditory cortex ltion. Her. Res. 172, (22). 9. Newsome, W.T. & Pré, E. A selective impirment of motion perception following lesions of the middle temporl visul re (MT). J. Neurosci. 8, (1988). 1. Tlwr, S.K., Musil, P.G. & Gerstein, G.L. Role of mmmlin uditory cortex in the perception of elementry sound properties. J. Neurophysiol. 85, (21). 11. Ti, L. & Zdor, A. Neurl mechnisms of selective uditory ttention in rts (Disserttion). Nture Precedings < (28). 12. Otzu, G.H., Ti, L., Yng, Y. & Zdor, A.M. Engging in n uditory tsk suppresses responses in uditory cortex. Nt. Neurosci. 12, (29). 13. Br-Yosef, O., Rotmn, Y. & Nelken, I. Responses of neurons in ct primry uditory cortex to ird chirps: effects of temporl nd spectrl context. J. Neurosci. 22, (22). 14. Asri, H. & Zdor, A.M. Long-lsting context dependence constrins neurl encoding models in rodent uditory cortex. J. Neurophysiol. 12, (29). 15. Hromádk, T., DeWeese, M.R. & Zdor, A.M. Sprse representtion of sounds in the unnesthetized uditory cortex. PLoS Biol. 6, e16 (28). 16. Wng, X., Lu, T., Snider, R.K. & Ling, L. Sustined firing in uditory cortex evoked y preferred stimuli. Nture 435, (25). 17. Brosch, M. & Schreiner, C.E. Time course of forwrd msking tuning curves in ct primry uditory cortex. J. Neurophysiol. 77, (1997). 18. Wehr, M. & Zdor, A.M. Synptic mechnisms of forwrd suppression in rt uditory cortex. Neuron 47, (25). 19. Shnnon, C.E. A mthemticl theory of communiction. Bell Syst. Tech. J. 27, , (1948). 2. Hykin, S. Adptive Filter Theory 4th edn. (Prentice Hll, 21). 21. Rieke, F., Wrlnd, D., de Ruyter Vn Steveninck, R. & Bilek, W. Spikes: Exploring the Neurl Code (MIT Press, 1996). 22. Smirnkis, S.M., Berry, M.J., Wrlnd, D.K., Bilek, W. & Meister, M. Adpttion of retinl processing to imge contrst nd sptil scle. Nture 386, (1997). 23. Niemi, P. & Näätänen, R. Foreperiod nd simple rection time. Psychol. Bull. 89, (1981). 24. Corre, Á., Lupiáñez, J. & Tudel, P. Attentionl preprtion sed on temporl expectncy modultes processing t the perceptul level. Psychon. Bull. Rev. 12, (25). 25. Kilgrd, M.P. & Merzenich, M.M. Corticl mp reorgniztion enled y nucleus slis ctivity. Science 279, (1998). 26. Tremly, L. & Schultz, W. Reltive rewrd preference in primte oritofrontl cortex. Nture 398, (1999). 27. Ro, S.M., Myer, A.R. & Hrrington, D.L. The evolution of rin ctivtion during temporl processing. Nt. Neurosci. 4, (21). 28. Ghose, G.M. & Munsell, J.H.R. Attentionl modultion in visul cortex depends on tsk timing. Nture 419, (22). 29. Anderson, B. & Sheinerg, D.L. Effects of temporl context nd temporl expectncy on neurl ctivity in inferior temporl cortex. Neuropsychologi 46, (28). 3. Shuler, M.G. & Ber, M.F. Rewrd timing in the primry visul cortex. Science 311, (26). 31. Treue, S. & Trujillo, J.C.M. Feture-sed ttention influences motion processing gin in mcque visul cortex. Nture 399, (1999). 32. McAdms, C.J. & Munsell, J.H. Effects of ttention on orienttion-tuning functions of single neurons in mcque corticl re v4. J. Neurosci. 19, (1999). 33. Jnssen, P. & Shdlen, M.N. A representtion of the hzrd rte of elpsed time in mcque re LIP. Nt. Neurosci. 8, (25). 34. Britten, K.H., Newsome, W.T., Shdlen, M.N., Celerini, S. & Movshon, J.A. A reltionship etween ehviorl choice nd the visul responses of neurons in mcque MT. Vis. Neurosci. 13, 87 1 (1996). 35. Corre, Á., Lupiáñez, J., Mdrid, E. & Tudel, P. Temporl ttention enhnces erly visul processing: review nd new evidence from event-relted potentils. Brin Res. 176, (26). 36. Uchid, N. & Minen, Z.F. Speed nd ccurcy of olfctory discrimintion in the rt. Nt. Neurosci. 6, (23). 37. Pxinos, G. & Wtson, C. The Rt Brin in Stereotxic Coordintes 5th edn. (Acdemic, 25). nture NEUROSCIENCE VOLUME 14 NUMBER 2 FEBRUARY

7 211 Nture Americ, Inc. All rights reserved. ONLINE METHODS Animl sujects. Animl procedures were pproved y the Cold Spring Hror Lortory Animl Cre nd Use Committee nd crried out in ccordnce with Ntionl Institutes of Helth stndrds. Fourteen dult mle Long Evns rts (Tconic Frms) were used for the experiments: 8 for ehviorl nlysis (t lest 3 sessions per niml), 5 for inctivtion experiments (1 sessions per niml), nd 3 implnted with tetrode microdrives for recordings (2 of these included in the ehviorl nlysis). Rts hd free ccess to food, ut wter ws restricted to ehviorl sessions. Free wter ws provided on dys with no experimentl sessions. Experiments were conducted in single-wlled soundooths (Industril Acoustics Compny). Behviorl tsk. The tsk consisted of n uditory two-lterntive choice procedure for freely moving rts. Rts initited ech tril y poking their nose into the center port of three-port opernt chmer. After silent dely of rndom durtion (25 35 ms, uniformly distriuted), sequence of 1-ms pure tones seprted y 5 ms ws presented. The frequency of the tones, from 5 khz to 4 khz, ws chosen rndomly for ech time slot nd tril, except on recording sessions s descried in the text. Sound intensity ws set t 7 db-spl. The niml ws required to sty in the port until frequency-modulted trget sound (Supplementry Eqution 1) ws presented in plce of one of the distrctor tones. The modultion frequency ws set to 15 Hz. The crrier frequency of the trget indicted to the niml which of the two side ports would provide 24 µl of wter rewrd on ech tril. For trget crrier frequency of 6.5 khz, rewrd ws ville only t the left port. For trget t 31 khz, rewrd ws provided t the right port. Sounds stopped once the suject left the center port. We presented mximum of 14 sounds (2.1 s) on tril. Nothing ws presented fter this period even if the niml styed in the center port, ut this rrely occurred (<2% of the time). Behviorl nlysis included only vlid trils in which the niml styed in the center port until the time of the trget onset. Animls were punished with 4 s time-out (during which nothing ws presented) if they withdrew from the center port fter the sequence of tones strted ut efore the onset of the trget. These events were ignored nd the tril ws re-strted. An equivlent time-out ws pplied whenever suject tried to collect rewrd on the wrong side port fter the presenttion of the trget. These were considered error trils. The frequency discrimintion component of the tsk ws designed to e very esy for the sujects, nd it ws used s wy to verify correct trget detection. The chllenging component of the tsk ws the detection of the frequency-modulted sound immersed in sequence of tones, which we could mke ritrrily difficult y reducing its TMD. Detection difficulty ws rndomized for ech tril, except on recording sessions where it ws set to n intermedite level resulting in n verge percentge of correct trils etween 7% nd 8%. Temporl expecttion ws mnipulted in locks of trils. On n initil lock, 85% of the trils contined the trget fter 3 or 45 ms (uniformly distriuted) from the first tone onset. In the remining 15% of trils, the trget ws presented fter 1,35 or 1,5 ms. In following lock the rtios were swpped. All four possile trget onset times were used in ech session, except during recordings when only 45 nd 1,5 ms were used. The lock type switched utomticlly every 15 or 2 trils until the session ended. No explicit cue ws given to the niml t the end of lock. Stimulus delivery. During trining nd inctivtion sessions, uditory stimuli were delivered through generic electromgnetic dynmic spekers clirted using pressure-field microphone (Brüel & Kjær) to produce 7 db-spl in the rnge of 5 4 khz t the position of the suject. During electrophysiologicl recording sessions, uditory stimuli were delivered through erphones (Etymotic Reserch) ttched to the hed of the niml nd clirted in similr wy. Wveforms were creted in softwre t smpling rte of 2, smples per second nd delivered to spekers or erphones through Lynx L22 sound crd (Lynx Studio Technology). We pplied rise nd fll liner envelopes of 2 ms to ll sounds. Surgery. Animls were nesthetized with n intrperitonel injection of mixture of ketmine (6 mg kg 1 ) nd medetomidine (.5 mg kg 1 ). Wounds were infiltrted with lidocine. For inctivtion experiments, plstic wells were implnted to protect ilterl crniotomies over uditory cortex (3 6 mm posterior to Bregm). The dur mter ws left intct. For recording experiments, ech rt ws surgiclly implnted with custom-mde microdrive in left uditory cortex contining 6 14 independently djustle tetrodes. Electrodes were implnted etween 3.5 nd 6 mm posterior to Bregm nd 6.5 mm left from the midline. Animls were llowed to recover for severl dys efore resuming wter restriction nd strting recording or inctivtion sessions. Before recording sessions, erphones were fixed to plstic rings ttched to the microdrive during surgery. Inctivtion of uditory cortex. We used the GABA A receptor gonist muscimol (Sigm-Aldrich) to reversily inctivte uditory cortex ilterlly. After lightly nesthetizing the rt (2% (vol/vol) isoflurne), we plced gelfom soked in 12.5 µl of either muscimol (3.44 µg µl 1 ) or sline (.9% (wt/vol) NCl) on the exposed dur mter on ech side. The rt ws given 3 min to recover from nesthesi efore strting the tsk. Dt from t lest five inctivtion nd five sline-control interleved sessions were collected for ech rt. Neurl recordings. Ech tetrode consisted of four polyimide-coted nichrome wires (Knthl Plm Cost; wire dimeter 12.7 µm) twisted together nd goldplted to n impednce of.3.4 MΩ t 1 khz. Electricl signls pssed through unity-gin hed stge (Neurlynx) connected to custom-uilt tetrode microdrive efore reching the cquisition system. Signls were recorded using the NSpike multichnnel cquisition system (L. Frnk nd J. McArthur). Tetrode depths were djusted etween recording sessions to smple n independent popultion of cells ech time. Tetrode loctions were confirmed histologiclly on the sis of electrolytic lesions (Supplementry Fig. 14). Although lesions were loclized to primry uditory cortex, it is possile tht some of the recorded cells come from non-primry res s we could not verify the exct loction of ech recording site. One of the three nimls hd to e lightly nesthetized (1.5% isoflurne) efore recording sessions to ese the connection of the hed stge nd erphones. In this cse, recovery period of 2 min ws given efore the eginning of the tsk. Dt nlysis. Dt were nlyzed using in-house softwre developed in MATLAB (Mthworks) nd Python ( For ll nlyses, trils fter those with unexpected trgets were ignored. The first few trils fter lock switching were not treted ny differently from susequent trils, to void setting ritrry thresholds on how mny trils constituted the expecttion updte stge. Behviorl nlysis. Ech session included in the ehviorl nlysis contined interleved trils of six different difficulties (TMD.15 24%) uniformly distriuted throughout the session. Rection time mesurements for Figure 2, included only correct trils t the esiest difficulty. At this difficulty level, nimls performed >9% correct. Rection time ws defined s the time etween the onset of trget nd the moment when the niml withdrew its nose from the center port. Erly trgets t oth 3 nd 45 ms were pooled together when estimting rection time. Averge rection time for ech niml ws clculted s the medin over trils from ll sessions. The percentge of correct trils for ech niml ws estimted using trils from ll sessions pooled together. Confidence intervls for these estimtes were clculted using the MATLAB function inofit, which computes the mximum likelihood estimte of the proility of success in inomil tril on the sis of the numer of successes oserved in set of independent trils. Performnce thresholds were estimted y fitting sigmoidl function (Supplementry Eqution 2) to the psychometric dt (percent correct versus difficulty) collected for ech niml nd finding the TMD tht would yield performnce of P = 75% correct. Single-cell nlysis. The spiking ctivity of 226 single cells ws isolted offline y comintion of n utomted expecttion mximiztion lgorithm (Klustkwik, K.D. Hrris) nd y mnully clustering spike fetures derived from the smpled wveforms using the MClust softwre pckge (A.D. Redish). Only cells with n evoked response of t lest 3 spikes per s ove spontneous firing were included in the nlysis. This resulted in 12 cells: 44 (out of 127) from sessions in which the first three tones hd fixed frequencies cross trils, nd 58 (out of 99) from sessions with only the first two tones fixed. nture NEUROSCIENCE doi:1.138/nn.2688

8 211 Nture Americ, Inc. All rights reserved. Averge spontneous firing ws evluted in the 2-ms window preceding the onset of the first tone. Neurl responses were estimted from the most responsive 25-ms window (verged cross ll trils) in the period etween the onset of the first tone nd the onset of the erly trget (45 ms lter). The mgnitude of the response ws clculted s the evoked firing rte minus the spontneous rte. The modultion index (MI) ws defined s Rexpected Runexpected MI = Rexpected + Runexpected where R is the evoked response, mesured s spikes per second or mgnitude of LFP (Fig. 4,d). This quntity is ounded etween 1 nd 1 for positive evoked chnges in ctivity, which ws the cse in our mesurements. The sttisticl significnce of the modultion of ctivity for single cells ws evluted with Wilcoxon rnk-sum test on the spike counts (over the response window) from ech expecttion condition. For the popultion, we used pired Wilcoxon signed-rnk test on the verge response for ech condition to evlute significnce. Becuse there were fewer trils ville for ech frequency when estimting the tuning of cell, response curves (to generte Fig. 5) were clculted from vrile-length window round the mximl response. The window ws formed y ll contiguous 25-ms ins tht showed n evoked response of t lest 7% of the mximum. For this figure, neuronl ctivity is presented without sutrcting spontneous firing to test for non-specific modultion. Individul tuning curves were ligned ccording to the preferred frequency of ech cell nd normlized efore verging. To evlute the chnges in response over time cross the popultion of cells (Fig. 6), we first estimted the est frequency for ech unit y pooling together ll responses to tones etween the erly nd the lte trgets. We then evluted the response to the est frequency tone t ech time slot, nd normlized it to the response to the erliest tone in this rnge (the fourth tone in the sequence, presented 1,5 ms efore the lte trget). Responses were estimted using the sme 25-ms time window with respect to ech tone onset. Only cells with responses ove spontneous firing for ech time slot were included (n = 2). Sttisticl significnce t ech time slot ws evluted with pired Wilcoxon signed-rnk test with respect to the fourth tone. Locl field potentils. The mgnitude of the evoked locl field potentil (LFP) ws clculted s the root men squre (RMS) power of the signl in the period ms fter the onset of the stimulus (Supplementry Eqution 3). This period corresponds to the onset response to the tone preceding the time of erly trgets. For comprisons of evoked LFP mgnitudes etween conditions (Fig. 4c,d nd Fig. 7), trils were first verged seprtely for ech condition. Chnges in evoked LFP mgnitudes were quntified y modultion index etween conditions, s descried ove. Sttisticl significnce for individul sites on given comprison ws evluted using permuttion tests. Correltion etween evoked responses nd ehvior. To evlute the reltionship etween neuronl ctivity nd ehviorl performnce, we compred the neuronl ctivity of trils with erly trgets grouped ccording to rection time. On ech individul ehviorl session, we sorted trils ccording to rection time nd selected two groups: one contining the 2% of trils with the fstest ehviorl responses, nd second contining the 2% of trils with the slowest ehviorl responses. We estimted verge response mgnitudes for oth LFP nd single cells, s descried erlier, nd clculted modultion index in ech cse to evlute the differences. Sttisticl significnce of the correltion etween LFP mgnitude nd rection time ws clculted using the function corrcoef from MATLAB, which trnsforms the correltion vlues to crete t-sttistic in order to estimte the proility of getting correltion s lrge s the oserved vlue y rndom chnce. doi:1.138/nn.2688 nture NEUROSCIENCE