Cowen 2012

16
doi:10.1152/jn.01012.2011 107:2393-2407, 2012. First published 8 February 2012; J Neurophysiol Stephen L. Cowen, Glen A. Davis and Douglas A. Nitz effort and reward to their associated action sequences Anterior cingulate neurons in the rat map anticipated You might find this additional info useful... 55 articles, 24 of which can be accessed free at: This article cites http://jn.physiology.org/content/107/9/2393.full.html#ref-list-1 including high resolution figures, can be found at: Updated information and services http://jn.physiology.org/content/107/9/2393.full.html can be found at: Journal of Neurophysiology about Additional material and information http://www.the-aps.org/publications/jn This information is current as of May 11, 2012. American Physiological Society. ISSN: 0022-3077, ESSN: 1522-1598. Visit our website at http://www.the-aps.org/. (monthly) by the American Physiological Society, 9650 Rockville Pike, Bethesda MD 20814-3991. Copyright © 2012 by the publishes original articles on the function of the nervous system. It is published 12 times a year Journal of Neurophysiology on May 11, 2012 jn.physiology.org Downloaded from

description

Effort and Reward in the Anterior Cingulate Cortex

Transcript of Cowen 2012

Page 1: Cowen 2012

doi:10.1152/jn.01012.2011 107:2393-2407, 2012. First published 8 February 2012;J NeurophysiolStephen L. Cowen, Glen A. Davis and Douglas A. Nitzeffort and reward to their associated action sequencesAnterior cingulate neurons in the rat map anticipated

You might find this additional info useful...

55 articles, 24 of which can be accessed free at:This article cites http://jn.physiology.org/content/107/9/2393.full.html#ref-list-1

including high resolution figures, can be found at:Updated information and services http://jn.physiology.org/content/107/9/2393.full.html

can be found at:Journal of Neurophysiologyabout Additional material and information http://www.the-aps.org/publications/jn

This information is current as of May 11, 2012. 

American Physiological Society. ISSN: 0022-3077, ESSN: 1522-1598. Visit our website at http://www.the-aps.org/.(monthly) by the American Physiological Society, 9650 Rockville Pike, Bethesda MD 20814-3991. Copyright © 2012 by the

publishes original articles on the function of the nervous system. It is published 12 times a yearJournal of Neurophysiology

on May 11, 2012

jn.physiology.orgD

ownloaded from

Page 2: Cowen 2012

Anterior cingulate neurons in the rat map anticipated effort and rewardto their associated action sequences

Stephen L. Cowen,1 Glen A. Davis,1 and Douglas A. Nitz2

1The Neurosciences Institute and 2the University of California, San Diego, California

Submitted 4 November 2011; accepted in final form 4 February 2012

Cowen SL, Davis GA, Nitz DA. Anterior cingulate neurons in therat map anticipated effort and reward to their associated actionsequences. J Neurophysiol 107: 2393–2407, 2012. First publishedFebruary 8, 2012; doi:10.1152/jn.01012.2011.—Goal-directed behav-iors require the consideration and expenditure of physical effort. Theanterior cingulate cortex (ACC) appears to play an important role inevaluating effort and reward and in organizing goal-directed actions.Despite agreement regarding the involvement of the ACC in theseprocesses, the way in which effort-, reward-, and motor-relatedinformation is registered by networks of ACC neurons is poorlyunderstood. To contrast ACC responses to effort, reward, and motorbehaviors, we trained rats on a reversal task in which the selectedpaths on a track determined the level of effort or reward. Effort waspresented in the form of an obstacle that was climbed to obtainreward. We used single-unit recordings to identify neural correlates ofeffort- and reward-guided behaviors. During periods of outcomeanticipation, 52% of recorded ACC neurons responded to the specificroute taken to the reward while 21% responded prospectively to effortand 12% responded prospectively to reward. In addition, effort- andreward-selective neurons typically responded to the route, suggestingthat these cells integrated motor-related activity with expectations offuture outcomes. Furthermore, the activity of ACC neurons did notdiscriminate between choice and forced trials or respond to a moregeneralized measure of outcome value. Nearly all neural responses toeffort and reward occurred after path selection and were restricted todiscrete temporal/spatial stages of the task. Together, these findingssupport a role for the ACC in integrating route-specific actions, effort,and reward in the service of sustaining discrete movements through aneffortful series of goal-directed actions.

medial prefrontal cortex; decision making; reversal learning; ensem-ble recording

DURING DECISION MAKING, the effort required to obtain an objec-tive must often be compared with anticipated rewards. Theanticipation of effort also influences behaviors that follow adecision, as goal-directed actions must often be maintaineddespite increasing energetic demands. Mounting evidence sug-gests that the anterior cingulate cortex (ACC) plays an impor-tant role in effort- and reward-guided behaviors (Amiez et al.2006; Floresco and Ghods-Sharifi 2007; Hauber and Sommer2009; Pratt and Mizumori 2001; Schweimer et al. 2005; Shimaand Tanji 1998; Walton et al. 2003). This role is supportedfurther by results from recent physiological studies that haveidentified ACC neurons that respond prospectively to botheffort and reward (Hillman and Bilkey 2010; Kennerley et al.2009). It is also clear that the role of the ACC extends beyondoutcome anticipation and evaluation, as considerable behav-ioral and physiological evidence indicates a role of the ACC in

motor control, action selection, and feedback control (reviewedin Hikosaka and Isoda 2010; Rushworth et al. 2004). Themultiple roles of the ACC in effort- and reward-driven evalu-ations and in motor control suggest that networks of ACCneurons may be involved in the prediction of anticipated effortlevels in order to guide action selection prior to a decision.Alternatively, ACC neurons may become involved after actionselection by maintaining specific goal-directed behaviors in theface of experienced or anticipated effort.

Addressing such questions requires an appreciation of thediverse response properties of ACC neurons to reward, move-ment, uncertainty, and sensory cues (Amiez et al. 2006; Baeget al. 2001; Cowen and McNaughton 2007; Euston and Mc-Naughton 2006; Jung et al. 1998; Kargo et al. 2007; Lapish etal. 2008; McCoy and Platt 2005; Narayanan and Laubach2006; Pratt and Mizumori 2001; Procyk and Joseph 2001;Shima and Tanji 1998). Although such heterogeneity suggestsa role in the association of multimodal stimuli with actions, italso complicates the analysis of neural function. For example,neurons responding to motor activity that is correlated with anoutcome would appear to be outcome-selective. To overcomesuch issues and to determine the extent to which ACC neuronsrespond to expected physical effort and reward, we developeda novel automated task that requires animals to adapt tomultiple path-reward and path-effort contingency reversals. Inthe task, the high-effort condition required animals to climb anobstacle to receive a food reward. This condition is widelyused to assess effort-related behaviors in rodents (Salamone etal. 1994; Walton et al. 2003). Frequent contingency reversalsenabled the isolation of path-, effort-, and reward-selectiveresponses of ACC neurons.

Analysis of neural responses revealed that ACC neuronswere most selective for the path to the outcome and to antic-ipated and experienced effort. Importantly, most effort-selec-tive neurons responded to conjunctions between the position ofthe animal and anticipated effort. Perhaps as a consequence ofthe sensitivity of neurons to movement, these responses werespatially and temporally limited, indicating that the populationof action- and effort-signaling neurons changed through thecourse of a trial. Finally, action, effort, and reward responsesoccurred after animals initiated movements toward the left orright path, suggesting a more prominent role for ACC involve-ment in maintaining ongoing goal-directed actions.

MATERIALS AND METHODS

Animals and Surgical Procedures

Neurophysiological recordings were acquired from five maleSprague-Dawley rats (9–14 mo old). Rats were housed individually

Address for reprint requests and other correspondence: S. L. Cowen,Neurosciences Inst., 10640 John Jay Hopkins Dr., San Diego, CA 92121(e-mail: [email protected]).

J Neurophysiol 107: 2393–2407, 2012.First published February 8, 2012; doi:10.1152/jn.01012.2011.

23930022-3077/12 Copyright © 2012 the American Physiological Societywww.jn.org

on May 11, 2012

jn.physiology.orgD

ownloaded from

Page 3: Cowen 2012

and maintained on a 12:12-h light-dark cycle. Recordings took placeduring the dark (waking) phase of the cycle. Surgery was conductedaccording to National Institutes of Health guidelines for rodents andInstitutional Animal Care and Use Committee (IACUC)-approvedprotocols. The rats were implanted with arrays of moveable micro-drives while under isoflurane anesthesia. Each microdrive held two orthree stereotrodes (McNaughton et al. 1983). Each stereotrode con-sisted of two polyimide-coated nichrome wires (24-�m diameter,electroplated with gold, impedance: 150–500 k�; California FineWire, Grover Beach, CA). A skull screw over the cerebellum servedas an indifferent reference for local field potential (LFP) recording.

A single craniotomy was drilled in the region above the right ACCand OFC (ACC: 0.8 mm mediolateral, �3.2 mm anteroposterior tobregma; OFC: 2.4 mm mediolateral, �3.6 mm anteroposterior tobregma). An insufficient number of electrodes reached the OFC, andso data from the OFC were not used in the present study. A secondcraniotomy was drilled over the right hippocampus (2.0 mm medio-lateral, �3.0 mm anteroposterior to bregma). After the microdrivewas cemented in place, a pair of LFP probes (stainless steel, coateddiameter of 0.0045 in., 300-k� impedance; Medwire, Mt. Vernon,NY) were lowered into the hippocampal craniotomy to depths of 2.7mm (fissure) and 2.1 mm (CA1 principal cell layer). LFP recordingswere not analyzed in this study. Rats were given a single injection ofbuprenorphine after surgery. All implanted animals received oralantibiotics (Cefa-Drops) on a daily basis.

Neurophysiological Recordings

Neural data were recorded once implanted animals could completefour blocks of trials within a single session (�1 mo after surgery).Once recordings began, all electrodes were lowered �40 �m aftereach successful recording session. Analysis was restricted to ACCneurons recorded at depths between 1.0 and 2.5 mm. Stereotrodeswere connected to a 54-channel unity-gain headstage (Neuralynx,Bozeman, MT). A multiwire cable connected the headstage to digi-tally programmable amplifiers. Spike signals were amplified (1,000–5,000 times), band-pass filtered (600–6,000 Hz), and transmitted tothe data acquisition system (Neuralynx). Recorded spike events were

initially and automatically sorted off-line on the basis of peak height,waveform energy, and principal components (KlustaKwik, author:K. D. Harris, Rutgers-Newark). These results were refined manuallywith custom-written software (MClust, author: A. D. Redish, Univer-sity of Minnesota; Waveform Cutter, author: S.L. Cowen). No attemptwas made to match cells from one daily session to the next. Conse-quently, the number of reported cells does not account for duplicatecells across consecutive days.

Histology

After completion of the study, recording sites were marked withdirect current stimulation (15 �A for 15 s) at least 2 days beforetranscardial perfusion. Perfusion was performed once animals weredeeply anesthetized with isoflurane. Prepared coronal sections wereNissl stained and visually inspected to determine electrode placement.Two representative sections are presented in Fig. 1E. Depth recordsindicated that the median depth for all recorded cells was 1,710 �m.

Behavioral Apparatus

All behavior was performed on a rectangular running track (10 cmwide) with a central choice region (Fig. 1A). Trials started whenanimals stepped on a trigger pad located on one end of the track. Thestart of a trial prompted the opening of one or two servo-controlleddoors located in the choice region of the track. On any given trial,either one door (forced trials) or two doors (choice trials) would open.A 5-cm wall was mounted on the inside edge of the section of trackcontaining the trial-start trigger. This wall obstructed the view of thedoors until the animal turned the corner that preceded the choiceregion. The doors led to two paths that eventually converged. Thedelivered amount of reward or effort (a climbable obstacle) wascontingent upon the path choice. Entry into the selected path wasregistered when animals stepped on a touchpad located just inside thedoorway. Stepping on this pad triggered the delayed (400 ms) deliveryof reward and/or the automatic rotation (via a servomotor) of a20-cm-tall triangular wire mesh obstacle (see Walton et al. 2003) ontothe track. Reward consisted of one or five drops of liquid Ensure

Fig. 1. Behavioral apparatus and adaptive behavior.A: layout of the track. Rats ran clockwise and selected apath when they reached the central choice region. Thisregion contained either 1 or 2 open doorways (forced orchoice trials). The chosen path determined either theamount of food or the presence or absence of a 20-cmobstacle (effort). B: organization of behavioral sessions.During each block of trials animals discriminated betweenthe level of effort (blue triangles) or reward (red squares).Contingencies were switched once animals chose the pathleading to the optimal response (low effort or high reward)on at least 8 consecutive choice trials. Block order waschanged on each session. C: example of choice behaviorfrom 1 session. Red and green squares represent choicetrials in which the rat chose the left (top) or right (bottom)path. The color of the square indicates correct (green) orincorrect (red) choices. Forced trials were randomly inter-spersed throughout the session (P � 0.5), and these trialsare omitted from the plot for visual clarity. Arrows indi-cate changes in outcome contingencies. Background colorindicates the type of block. D: reversal performance. The boxand whisker plot indicates the median number of incorrectchoice trials that occurred before rats made 5 consecutivecorrect responses (whiskers � 1.5 IQR). Animals adaptedmore quickly to effort reversals (fewer error trials) than toreward reversals (n � 58, *P � 0.00006, Wilcoxon rank sumtest). E: 2 representative histological sections with corre-sponding diagrams from Paxinos and Watson, The Rat Brainin Stereotaxic Coordinates, Copyright Elsevier, 1998. Ar-rows indicate estimated electrode position.

2394 ANTERIOR CINGULATE RESPONSES TO ACTION AND EFFORT

J Neurophysiol • doi:10.1152/jn.01012.2011 • www.jn.org

on May 11, 2012

jn.physiology.orgD

ownloaded from

Page 4: Cowen 2012

delivered through a solenoid valve. The entire apparatus was con-trolled by a microcontroller (ZX-1280, ZBasic, Elba) using customsoftware. The microcontroller also monitored performance and deter-mined the timing of path-outcome contingency reversals. Duringrecording sessions, the position of the animal was tracked from avideo camera mounted 2.5 m above the track. Position was sampled at30 frames/s.

Behavioral Protocol

Pretraining. Animals were shaped to perform the target behaviorover the course of �7 wk of daily training (2 h/day, 5 days/wk).Animals were first trained to run clockwise around a shortenedversion of the track (Fig. 1A) for food reward (1–2 wk). Successfulperformance was followed with reward discrimination training thatrequired learning which path led to food reward. This period oftraining ended once animals adapted to three path-reward contingencyreversals within a session. Effort discrimination followed rewarddiscrimination training and also concluded once animals adapted tothree path-effort reversals within a session. The final stage combinedeffort and reward discrimination blocks within a session. Criterionperformance was reached once animals completed two blocks ofeffort and reward discrimination trials within a single session (�140trials and 3 reversals). The details of this behavior are provided in thenext section.

Tested behavior. The goal of the behavior was to facilitate thedissociation between neural responses to actions, anticipated costs,and anticipated rewards. As a result, trials involving effort discrimi-nation and reward discrimination were performed in separate blocks.The levels of effort and reward were varied on separate blocks inorder to allow the clear identification of neural selectivity to amountsof anticipated effort or reward. This design separates this study fromother studies that explicitly investigate cost-benefit decision making,as these studies necessarily require nonindependent combinations ofeffort and reward (e.g., high effort/high reward vs. low effort/lowreward; see Salamone et al. 1994; Walton et al. 2002).

In the present behavior, rats were required to run laps around arectangular track and select a path at a “choice region” (see Fig. 1A).The choice region presented animals with either one (forced trials) ortwo (choice trials) open doorways according to a pseudorandomschedule (P � 0.5). The principal function of choice trials was todetermine whether animals learned the association between the pathand the outcome and thus whether the animals had a specific expec-tation for high or low effort and reward. Forced trials were used fortwo reasons. First, they ensured adequate sampling of neural datarelated to the nonpreferred path and outcome, as animals wererequired to sample the nonpreferred path. Second, forced trials short-ened the time for animals to adapt to a reversal, presumably becauseanimals were able to intermittently sample the outcome associatedwith the nonpreferred arm. Reducing the time between reversals wasimportant given the limited number of trials that animals wouldperform before becoming sated.

Each doorway led to an alternate route along the track, with theselected route determining either the presence or absence of a 20-cmtriangular obstacle (Salamone et al. 1994) or the amount of reward (1or �5 drops of liquid Ensure). The obstacle was swiveled onto themaze automatically through a servomotor mounted on its base. Theobstacle was retracted before the start of each trial. The obstacle andfeeders were only triggered 400 ms after animals left the doorwayregion so that outcome-related auditory or visual cues would notinfluence choice behavior.

Trials were divided into four blocks (see Fig. 1B), and within eachblock there was a single optimal choice (e.g., choose the high-rewardpath or avoid the high-effort path). The contingency between eachpath and the level of effort or reward was reversed when animalsselected the optimal path on eight consecutive choice trials. Allimplanted animals successfully completed at least 150 trials within

each session and adapted to at least 3 reversals (see Fig. 1C for anexample).

Analysis

All statistical tests were performed with tools from the MATLABStatistics Toolbox (version 2008b) and R (version 2.11). Shufflecontrols were also used to determine whether results differed fromchance. These controls were created by randomly permuting trialorder with respect to trial category (e.g., high or low reward). Allanalysis was performed with a Windows 7 computer (Intel Core i3).

Categorization of Neurons

A major objective of the present experiment was to determinewhether ACC neurons respond to and integrate information aboutactions, effort, and reward. Addressing these issues involved classi-fying neurons according to their selectivity for the chosen path (left orright), the level of effort (high or low), or the level of reward (high orlow). The first step of the classification procedure involved subdivid-ing behavioral trials into binary groups (e.g., high or low effort).Trials were further restricted to trials containing runs of at least fiveconsecutive correct responses (e.g., choosing the high- over thelow-reward path). This was done to better ensure that animals were“expecting” a high or low amount of effort or reward. In addition,analyses, unless otherwise stated, were restricted to forced trials. Thisrestriction ensured equal sampling of trials in the high and low effortand reward conditions and better ensured that responses were tooutcome expectancy and not to competing contributions related toprechoice deliberation.

Neuron classification was performed by comparing trial-by-trialspiking activity under each trial condition (e.g., left or right path).Comparisons were performed by standard statistical tests, with thespecific test depending on the measure of neural activity that wasanalyzed (e.g., spike counts or ratemaps). Consequently, the statisticaltests used (e.g., ANOVA or Wilcoxon rank sum test) are described inthe appropriate RESULTS sections. A neuron was considered to have aselective response to the chosen feature if the statistical comparisonwas significant at P � 0.05.

Measure of Selectivity

The selectivity index (SI) was derived from the area under thecurve (AUC) measure of category separation as determined from thereceiver operator characteristic curve (Green and Swets 1989).The AUC measures how easy it is to perform binary classification ona data set and is related to the Wilcoxon rank test. AUC was used toshow the degree to which spiking activity (or head position) wasdifferent under two conditions (e.g., left or right path). The classicAUC measure assumes that the user knows a priori which distributionis the leftmost distribution. The leftmost distribution could be deter-mined by measuring the mean or median; however, such a methodcould yield different results depending on the measure of centrality.Consequently, we modified the AUC measure to be independent ofassumptions of centrality. The following equation describes thismeasure:

SI � �AUC � 0.5� � �AUCshuffle � 0.5�where AUCshuffle is the AUC value computed after randomly shufflinggroup membership for each trial (e.g., level of effort) relative to thespike counts. The SI value is zero when two distributions cannot bediscriminated.

Identifying Time of Earliest Selective Response

To assess whether ACC neurons were involved in planning actions,it was necessary to establish whether selective neural responses

2395ANTERIOR CINGULATE RESPONSES TO ACTION AND EFFORT

J Neurophysiol • doi:10.1152/jn.01012.2011 • www.jn.org

on May 11, 2012

jn.physiology.orgD

ownloaded from

Page 5: Cowen 2012

associated with the chosen path preceded or followed movementstoward that path. It was therefore important to estimate the point intime at which either neural activity or head/body movements firstdiscriminated between the two paths. Two methods were used toestimate this time point. The first involved using repeated t-tests onthe event-aligned measures of selectivity (averaged across experimen-tal sessions). The onset of the selective response was defined as thetime in which four consecutive time bins (100 ms) exhibited a meanselectivity value that was significantly greater than zero. The secondmethod applied change-point analysis (Chow 1960) to identify struc-tural changes within the measure of selectivity for the path. Change-point analysis was implemented with the strucchange library (Kimchiand Laubach 2009; Zeileis et al. 2002) for the R statistical program-ming environment.

RESULTS

Neural recordings (n � 380 neurons) were acquired fromfive rats as the animals performed an effort- and reward-guideddecision-making behavior (see MATERIALS AND METHODS). Theanalysis of behavioral performance on the task is presented inthe next section, and the analysis of neural responses duringthis behavior follows.

Effort Avoidance Strongly Impacts Choice Behavior

Assessment of decision-making performance provided ameasure of the rate at which animals learned associationsbetween the two paths and the associated levels of effort andreward. Performance was evaluated by measuring the numberof incorrect choice trials that occurred before the animalcompleted five consecutive correct responses (Fig. 1D). Ap-plying this criterion to the reward and effort blocks revealedthat animals required roughly half as many trials to adapt toreversals in effort level relative to reward (P � 0.00006, n �58 sessions, Wilcoxon rank sum test). One hypothetical sourcefor the difference in behavioral performance between the effortand reward conditions is that the obstacle and/or the movementof the obstacle into the path of the animal was a more salientcue compared with cues associated with reward delivery (e.g.,the audible click of the feeder). The increased salience ofeffort-associated cues may have therefore facilitated adaptationto changes in effort contingencies. A second observation wasthat, despite months of training, performance peaked at sub-optimal levels, with all animals experiencing at least two errors(and typically many more) before reversing choice behavior.This was surprising, as a single error was always 100%predictive of a contingency reversal. The suboptimal reversalperformance suggests that the animals were integrating errorsover trials instead of adopting a cue- or rule-driven strategy. Asimilar effect has been reported in primates (Kennerley et al.2006).

ACC Neurons Discriminate Between Paths, AnticipatedEffort, and Anticipated Reward

The first objective of the experiment was to determine theextent to which ACC neurons integrate information aboutactions, effort, and reward. A second objective was to deter-mine the stage or stages of the task at which these responsesoccur. Addressing these issues involved the categorizing ofindividual neurons according to the extent to which neuralactivity discriminated between the two paths or the levels ofeffort and reward. The categorization procedure is presented in

MATERIALS AND METHODS. Briefly, trials were organized bycondition (e.g., high- vs. low-reward trials), and neural activity(defined below) in each condition was compared to determinewhether the neuron exhibited a selective response to the fea-ture. Examples of the response of six neurons that wereselective to the path, effort, and reward are presented in Fig. 2.In these examples, neurons were categorized as selective ifthere was a significant difference (ANOVA, Pthreshold � 0.05)in the vectors of spike counts, measured during the 600-msinterval that preceded entry into the doorway, in each trialcondition (e.g., high or low reward). These examples illustratethe rich variation among neurons in their response to thelocation of the animal, the amount of effort, and the amount ofreward.

To evaluate selectivity across the population of ACC neu-rons (n � 380) and across all locations on the track, thetrial-by-trial ratemaps (bin size � 3 cm) were analyzed. Figure 3summarizes responses across the population and presents thepercentage of neurons whose firing activity discriminated path,effort, and reward (P � 0.05, Wilcoxon rank sum test) at eachlocation on the track. The analysis indicates a strong and earlyselective response to the path taken as animals approached andentered the choice zone, with the majority of neurons becom-ing selective after animals entered the choice region. Subsets ofneurons also became selective for the anticipated presentationof the obstacle or to reward once animals entered the choiceregion. These anticipatory responses were not due to visual/auditory cues associated with the outcome, as the obstacle andfeeder were not triggered until 400 ms after the rats left thedoorway. In addition, nearly half of recorded neurons becameselective for the level of effort (presence or absence of theobstacle) when rats reached the location of the obstacle. Se-lectivity at this location may be due to an ACC response tosensorimotor feedback associated with climbing the obstacle. Itis conceivable that such sensitivity could contribute to effortassessment, as integration of the sensorimotor signal wouldproduce an estimate of the effort. Alternatively, the responsescould be due to a motor control signal involved in organizingthe act of climbing the barrier.

Given behavioral evidence for the involvement of the rodentdorsomedial prefrontal cortex in working memory (Ragozzinoand Kesner 2001), it was conceivable that individual ACCneurons may maintain selective responses for previously ex-perienced actions or outcomes after animals leave the rewardzone. Contrary to this suggestion, ACC neurons were notselective to the preceding action or outcome during the returnjourney (orange section in Fig. 3). The lack of a response,however, could be due to the fact that the task used in thisstudy did not explicitly require animals to maintain a retro-spective memory of the previous action or outcome in order tomaintain correct performance.

Neural Responses to Anticipated Effort and the Two PathsExceed Responses to Anticipated Reward

To better assess the significance of the above observations,a measure of neural selectivity for the left or right path, thelevel of effort, and the amount of reward was computed. Thisselectivity index (SI) was derived from the AUC measure ofclass separation (see MATERIALS AND METHODS). The index wascalculated from distributions of perievent spike counts identi-

2396 ANTERIOR CINGULATE RESPONSES TO ACTION AND EFFORT

J Neurophysiol • doi:10.1152/jn.01012.2011 • www.jn.org

on May 11, 2012

jn.physiology.orgD

ownloaded from

Page 6: Cowen 2012

fied in a 600-ms window aligned on the time that animalsreached prespecified locations along the track (inset in Fig. 4Dillustrates these locations). The selectivity of neurons for thepaths, effort, and reward during the approach to the doorway is

indicated in Fig. 4B. During the approach, ACC neuronsresponded selectively to the path, effort, and reward, with thestrongest response being to the path (Fig. 4B; P � 0.05,ANOVA with Tukey-Kramer correction). Furthermore, selec-

Fig. 2. Single-unit responses to path, effort, and reward.A: schematic of the track with colors indicating task-relevantlocations. These colors correspond to the colors indicated onthe x-axis of the linearized version of the maze. B: example ofa 2-dimensional (2D) ratemap from a single anterior cingulatecortex (ACC) neuron with activity averaged across all trials.C–E: mean (�SE) linearized ratemap responses of 6 exampleneurons that were classified as being selective to the left andright path (C), to the level of effort (D), and to the amount ofreward (E). Colors on the x-axis correspond to the color-codedsections of the track (A).

Fig. 3. Percentage of ACC neurons from the population ofrecorded cells (n � 380) that were selective for the path, thelevel of effort, and the amount of reward at each location on thetrack. Cell counts were determined by identifying neurons withsignificant differences in activity (P � 0.05, Wilcoxon ranksum test) for trials under each condition (e.g., high- vs. low-reward trials). Neural activity in each trial was quantified as thelinearized ratemap (3-cm bin size). Colors on the x-axis indicatelinearized position on the 2D track (inset). Red dashed lineindicates chance. Trials were restricted to forced trials thatfollowed a series of correct choices to better ensure that animalshad a specific outcome expectancy on each trial and to ensurethat an approximately equal number of trials were in eachcondition. A large portion of neurons were selective for thetraversed path as animals approached the choice region. Asmaller but significant proportion of neurons were also selectivefor effort and reward in the choice region. Gray dashed lineindicates the selectivity index for the left or right path (righty-axis) determined from the tracked movements of the animal.

2397ANTERIOR CINGULATE RESPONSES TO ACTION AND EFFORT

J Neurophysiol • doi:10.1152/jn.01012.2011 • www.jn.org

on May 11, 2012

jn.physiology.orgD

ownloaded from

Page 7: Cowen 2012

tivity for the level of effort was significantly greater than forreward (P � 0.05). Similar results were observed from theanalysis of cell counts, with 196 neurons (52%) respondingselectively to the choice path, 80 (21%) responding selectivelyto effort level, and 44 (12%) responding to the amount ofreward (Fig. 4C).

The analysis of selectivity was applied to six additionaltrigger-points on the track (Fig. 4D). This analysis revealedthat at the approach trigger (Ap), some ACC neurons werealready selective for the left or right path but not to effort orreward (P � 0.05, 2-tailed t-test with Bonferroni correction).ACC neurons did become selective for effort and reward whenanimals were near the doorway, and, in the case of effort, thisselectivity persisted until animals left the reward zone. Selec-tivity for reward was also observed at these locations, with thenotable exception of the “effort” zone (Ef). Reward selectivitywas significantly lower than selectivity for effort at all of thesepoints (P � 0.05, paired 2-tailed t-test with Bonferroni correc-tion). As suggested from Fig. 3, selectivity for the path, effort,or reward did not persist during the return journey (location Rtin Fig. 4D; P � 0.05, 2-tailed t-test with Bonferroni correc-tion). To summarize, the pattern of selectivity for the path,effort, and reward suggests that ACC neurons are notablyselective for ongoing actions and anticipated levels of effort.

Neurons were less responsive to anticipated levels of reward,and the lack of selectivity to the path, effort, or reward duringthe return journey suggests that ACC neurons do not maintainany retrospective information about the preceding action oroutcome after reward consumption.

Neural Responses to Choice Path Occur After Path Selection

The high degree of selectivity of ACC neurons for the left orright path suggested that the neurons may be involved inplanning actions or action sequences, a role that has beenattributed to the region (Rushworth et al. 2004). However,subsequent analysis of movements and neural activity indi-cated that selectivity for the path occurred only after animalsinitiated movements toward the chosen doorway (Fig. 5). Thisanalysis was performed by evaluating the spiking activity andhead position as animals approached the doorways. If ACCneurons are involved in planning actions, then neural selectiv-ity for the path would be expected to precede even slightmovements toward one of the two doorways. This predictionwas investigated by comparing the measures of selectivitygenerated from movement and neural data to determine the firstpoint in time at which a significant selective response wasobserved. Two methods were used to estimate these onset

Fig. 4. Analysis of selectivity. A: spike rastergrams and perievent responses of 3 path-, effort-, and reward-selective neurons with responses aligned at the timeanimals entered the doorway on the track. The 3 neurons were selective for either the selected path, effort, or reward. B: mean selectivity (�SE) of all ACCneurons (n � 380) at the doorways (Dw). The selectivity index (see MATERIALS AND METHODS; 0 indicates no selectivity) was calculated from the distributionsof perievent spike counts (600-ms window) triggered at the moment animals crossed the Dw trigger. Only forced trials following correct performance wereincluded in this analysis. Average selectivity differed for path, effort, and reward (P � 0.05, ANOVA, Tukey-Kramer correction). C: number of path-, effort-,and reward-selective neurons at point Dw. Dashed line indicates chance (P � 0.05, ANOVA). D: mean (�SE) selectivity of the population of ACC neurons (n �380 neurons) at 7 locations along the track indicated in inset. St, trial start; Ap, approach; Dw, doorways; Cv, convergence; Ef, effort; Rw, reward; Rt, return.Selectivity for the path differed from chance at all locations except St and Rt (P � 0.05, 2-way t-test, Bonferroni correction). In contrast, neurons did not becomeselective for effort and reward until point Dw. The selectivity for effort was greater than for reward at the Dw, Cv, Ef, and Rw locations (P � 0.05, paired t-test,Bonferroni correction). Inset: schematic of the track. Dashed lines mark virtual triggers used to align perievent responses.

2398 ANTERIOR CINGULATE RESPONSES TO ACTION AND EFFORT

J Neurophysiol • doi:10.1152/jn.01012.2011 • www.jn.org

on May 11, 2012

jn.physiology.orgD

ownloaded from

Page 8: Cowen 2012

times. The first approach involved creating a perievent measureof selectivity aligned on the time that the rats passed over theapproach trigger (Ap in Fig. 4D). These aligned responses wereaveraged across all path-selective neurons (n � 196 neurons)or movement recordings (n � 54 sessions) and evaluated ateach point in time (100-ms bins) with a t-test (Pthreshold � 0.05,ANOVA). To correct for multiple tests, an event was onlyconsidered significant if P � 0.05 on at least four consecutivetime bins. Figure 5A presents the results of this perieventanalysis. The analysis revealed that neural selectivity for thepath lagged 200 ms behind selectivity derived from positiondata, suggesting that ACC neurons did not signal plannedactions or routes.

The preceding method did not explicitly test for the point atwhich the SI fundamentally changed its state (e.g., changingfrom nonselective to selective), nor does it perform a within-session comparison between the neural and position measures.Consequently, change-point analysis, a technique developedfor the identification of state changes in continuous data (Chow1960; Kimchi and Laubach 2009), was used to identify theonset of the first selective response within each session. Meth-ods developed by Bai and Perron (2003) were used to identifysignificant change points in the time course of the SI (seeMATERIALS AND METHODS). Neural or movement data that did notyield a significant change point (P � 0.05) were omitted fromthe analysis. The differences in onset time between movementand neural responses were averaged across comparisons (Fig.5B). In agreement with results from the previous analysis, themean time of the change point identified from the movementdata was less than (preceded) the change point identified for theneural data (P � 0.05, 2-tailed t-test), with the mean differencebeing 172 ms. The results of both of the analyses of selectivitysupport the conclusion that neural activity associated withmovement toward the targeted path occurs shortly after theanimals begin physically moving toward the path. As a result,ACC activity in this behavior may not be related to actionplanning, but may instead be involved in postplanning pro-cesses such as action maintenance or action monitoring.

Firing Response of Over 50% of Effort- andReward-Selective Neurons Also Discriminates BetweenLeft and Right Paths

An outstanding question is whether subpopulations of ACCneurons respond exclusively to the path, effort, or reward, orwhether individual neurons integrate information about theseparameters. To examine this, a two-way ANOVA was per-formed for each ACC neuron during effort trial blocks (factor1: left/right path, factor 2: high/low effort; P � 0.05), and aseparate two-way ANOVA was performed for reward trialblocks (factor 1: left/right path, factor 2: high/low reward). TheANOVA was calculated from spike counts measured duringthe 600-ms window preceding door entry. The analysis re-vealed that 51% (41/80) of effort-selective neurons and 56%(19/34) of reward-selective neurons were also path-selectiveafter controlling for outcome sensitivity (Fig. 6). Furthermore,subsets of effort (25%, 20/80)- and reward (26%, 9/34)-selec-tive neurons expressed a significant interaction with the path.There was also very little overlap between the populations ofeffort- and reward-selective neurons, as only 13 neurons (3%of ACC neurons) exhibited a main effect of effort and reward(Fig. 6C). In summary, the considerable overlap between path-and effort-selective neurons suggests that individual ACCneurons are involved in associating actions with effort. Theminimal overlap between effort- and reward-selective cells, incontrast, suggests that such associations do not extend to effortand reward.

Mean Firing Rate Increases During Anticipation of Effort

A specialization of the ACC for processing effort could beexpressed in a number of ways. For example, ACC neuronscould exhibit a global bias in firing activity during trials inwhich the rats anticipated the obstacle. Alternatively, individ-ual ACC neurons could respond heterogeneously, with roughlyequal numbers of neurons responding with increased activity tohigh or low effort. In this scenario, the overall populationwould not exhibit a global bias for a particular outcome. To

Fig. 5. Neurons responded to the choice path after animals initiated movement toward the path. A: the selectivity of movement (black, n � 58 sessions) or neuralactivity (gray, n � 192 neurons) for the 2 paths on the track (selectivity index) was analyzed to determine whether movements preceded path-related neu-ral activity. The selectivity index is aligned on the time that animals crossed point Ap on the track (see Fig. 4D). Similar results were also obtained by averagingacross sessions and determining the earliest time at which spiking or movement-related activity reached significance (P � 0.05, ANOVA). Results from thisanalysis are presented as vertical dashed lines that indicate the earliest times at which movement (�350 ms) and neural activity (�150 ms) discriminated betweenthe paths. Onset times were defined as the time at which 4 consecutive bins (bin size � 100 ms) reached significance (P � 0.05, 2-tailed t-test). B: a secondanalysis of selectivity onset was performed with change-point analysis (see MATERIALS AND METHODS). Within each experimental session the time of the changepoint identified from the position data was subtracted from the time point determined from the neural data. These data were averaged across all neurons exhibitinga significant change point. The change point estimated from head movement preceded the change point estimated from neural activity (n � 73, P � 0.05, 2-tailedt-test; error bars � SE), with the average difference being 173 ms.

2399ANTERIOR CINGULATE RESPONSES TO ACTION AND EFFORT

J Neurophysiol • doi:10.1152/jn.01012.2011 • www.jn.org

on May 11, 2012

jn.physiology.orgD

ownloaded from

Page 9: Cowen 2012

investigate these two possibilities, the average firing rate of alleffort-selective neurons was measured separately for high- andlow-effort trials. In addition, differences in firing responsewere measured for the amount of reward and the path taken atthe choice region (Fig. 7, A–C). Neurons were categorized asselective for effort, reward, or the path according to theprocedures described in MATERIALS AND METHODS. These proce-dures were applied to the trial-by-trial spiking activities ob-served at three locations along the track: the doorway (Dw), the

convergence point (Cv), and the effort zone (Ef). The conver-gence point and effort zone were also included in this analysis,as the question applied to all locations on the track thatpreceded outcome delivery (presumably periods where the ratanticipated a specific outcome). Although no differences inmean firing rate were observed for path- and reward-selectiveneurons, neurons responsive to effort increased firing rate inthe high-effort condition at the doorway and convergence point(P � 0.05, balanced ANOVA, Tukey-Kramer correction). The

Fig. 6. Populations of effort- and reward-selective cells alsorespond to path. A: a 2-way ANOVA was performed for eachACC neuron (factor 1: left/right path, factor 2: high/low effort).The ANOVA was calculated from spike counts measured dur-ing the 600-ms window preceding door entry (Dw). Black barsindicate the number of neurons with a significant main effect ofeffort (Pthreshold � 0.05), gray bars indicate the portion of theseneurons that also exhibited a main effect of path (Pthreshold �0.05), and white bars indicate neurons with an interaction effect.Percentages indicate percentage of all recorded neurons (n �380). Dashed lines indicate chance as computed by shufflingtrial categories prior to the ANOVA. B: same as A except for thepopulation of reward-selective neurons. C: overlap between thepopulations of effort- and reward-selective cells (black bars inA and B). A total of 13 neurons responded to differences in botheffort and reward level.

Fig. 7. Analysis of firing rate. A–C: mean (�SE) firingresponse of ACC neurons to the selected path and thelevel of effort and reward. Neurons were first categorizedaccording to their selectivity for the effort, reward, orpath by applying a t-test to the firing activity measured ateither the doorway (Dw), the convergence point (Cv), orthe effort zone (Ef) (threshold: P � 0.05). No differenceswere observed for the path- or reward-selective neurons(A and C); however, significant differences were ob-served for effort-selective neurons (B) at points Dw, Cv,and Ef (*P � 0.05, balanced ANOVA, Tukey-Kramercorrection). D: firing rate of all ACC neurons was sig-nificantly higher during the approach to the doorway(Ap) than at all other regions of the track (balancedANOVA, *P � 0.05, Tukey-Kramer correction).E: neuronal variability (Fano factor) decreased at pointAp, but only relative to points Cv, Ef, and Rw (ANOVA,*P � 0.05, Tukey-Kramer correction). F: mean (�SE)running speed at each location on the track (n � 58sessions).

2400 ANTERIOR CINGULATE RESPONSES TO ACTION AND EFFORT

J Neurophysiol • doi:10.1152/jn.01012.2011 • www.jn.org

on May 11, 2012

jn.physiology.orgD

ownloaded from

Page 10: Cowen 2012

magnitude of the firing rate difference in all of these conditionswas relatively small (�0.3 Hz). Consequently, these resultssuggest a weak bias in global firing rate during the anticipationof effort.

Changes in Global Firing Activity and Neuronal Variability

The ACC has been implicated in many phases of adaptivebehavior such as conflict monitoring, action selection, andoutcome evaluation (reviewed in Ridderinkhof et al. 2004). Itwas therefore conceivable that overall ACC activity wouldvary as a function of the phase of the behavior and/or thelocation of the animal on the track. The analysis of the meanfiring activity of all ACC neurons revealed a global peak inactivity during the approach toward the choice zone (Fig. 7D;P � 0.05, ANOVA with Tukey-Kramer correction). The in-creased firing activity did not correspond to a clear peak ortrough in the overall running speed of the animal (Fig. 7F).

The variability of the neural response, independent of meanrate, has been shown to correlate with attentional demands(Mitchell et al. 2007), reward magnitude expectancy (Kargo etal. 2007), and the time of stimulus onset (Churchland et al.2010). Although the causes of changes in neuronal variabilityare unknown, decreased variability may result from the reduc-tion of internal noise generated within individual cells (Kargoand Nitz 2004; Mitchell et al. 2007) or as a consequence ofsynaptic mechanisms associated with a neuron becoming in-corporated into a larger functional group. Accordingly, weinvestigated trial-by-trial variability as a function of the stageof the tasks (Fig. 7E). Variability was quantified as the Fanofactor (ratio of the variance of the spike count to the mean).The Fano factor was calculated for each ACC neuron (n �380). The Fano factor was lowest at the approach zone;however, the size of this effect was small and only significantlydifferent from values observed near the effort zone (points Cvand Ef on the track; P � 0.05, ANOVA with Tukey-Kramercorrection).

It is notable that a Fano factor � 1 was observed at alllocations along the track. This result contrasts with observa-tions from studies that report sub-Poisson responses in subcor-tical structures (Kara et al. 2000) and V1 (Gur et al. 1997). Incontrast, our results are in agreement with results from studiesthat cover a range of cortical regions in which Fano factors �1 have been reported (Churchland et al. 2010; Hussar andPasternak 2010; Mitchell et al. 2007). The super-Poisson re-sponse observed in the ACC and reported in other corticalregions could be due to factors such as a higher sensitivity ofneurons in these regions to higher-order and difficult to mea-sure information (e.g., internal states, level of attention) or to asensitivity to unmeasured sensorimotor variables (Gur et al.1997). The observation of super-Poisson and sub-Poisson re-sponses also suggests that efforts to model neuronal activitymay be improved if potential sources of violations of thePoisson model are considered.

Path-Selective Neurons Respond to Running Speed, butEffort- and Reward-Selective Neurons Do Not

The variations in running speed observed during differentstages of the behavior (Fig. 7F) prompted an analysis ofrunning speed as a function of path, effort, and reward. Thisanalysis revealed differences in running speed for each condi-

tion (Fig. 8, A–C). The largest observed difference in runningspeed at the choice region (Dw) was observed for high and loweffort, with running speed being slightly slower in the high-effort condition (P � 0.05, n � 58 sessions, balanced ANOVAwith Tukey-Kramer correction).

The above result suggested that some neuronal variabilityobserved for the effort condition may be explained bysensitivity to running speed or its correlates (e.g., motiva-tion level, fatigue). To investigate this, the vector of trial-by-trial measures of running speed was correlated withtrial-by-trial spike counts for each neuron. Neurons with asignificant Pearson’s correlation with running speed (P �0.05) during the approach to the doorway (Dw) were iden-tified as speed-selective. The overlap between the popula-tions of speed-selective neurons and neurons classified asselective for the path, effort, or reward is presented in Fig.8D. This analysis revealed that nearly half (46%, 90/196) ofpath-selective neurons were also selective to running speed;however, less than 16% (13/80) of effort- and 20% (9/44) ofreward-selective neurons were speed-selective, and thesevalues approached values computed from trial-shuffled con-trols. Similarly, the analysis of explained variance indicatedthat a small (�10%) but significant amount of variance inthe population of path-selective neurons could be explainedby variations in running speed (Fig. 8E). No significantamount of variance was explained for the populations ofeffort- and reward-selective neurons (Fig. 8E).

ACC Neurons Responded to Either Effort or Reward but Notto Outcome Value

The observation that few ACC neurons responded to botheffort and reward (Fig. 6C) suggested that individual ACCneurons respond to specific outcomes rather than to a general-ized estimate of outcome value. To investigate this questionexplicitly, we determined whether ACC neurons expressed atendency to respond to valued outcomes in both the effort andreward discrimination blocks. For example, a neuron respond-ing to positive value should fire more robustly during thelow-effort condition during effort discrimination and alsorespond more during the high-reward condition during re-ward discrimination. Accordingly, the difference betweenthe mean firing response of each neuron to the high- andlow-value alternative was calculated for the effort andreward discrimination trials. This measure was computedduring the 600-ms window that preceded door entry. Thedifferences in firing activity (high value � low value) areplotted in Fig. 9. Neurons expressing a generalized responseto value across outcome types should respond in a similardirection to either low- or high-valued outcomes in botheffort and reward discrimination blocks. As a result, pointsshould cluster in the lower left or upper right quadrants.Instead, results from the analysis indicated that more pointsclustered in the upper left and lower right quadrants both forthe entire population of ACC neurons and for the subpopu-lations of effort- and reward-selective cells, and the corre-lation between the level of value in both conditions wasnegative (rall ACC � �0.21, P � 0.00006; reffort � �0.41,P � 0.0002; rreward � �0.20, P � 0.26).

2401ANTERIOR CINGULATE RESPONSES TO ACTION AND EFFORT

J Neurophysiol • doi:10.1152/jn.01012.2011 • www.jn.org

on May 11, 2012

jn.physiology.orgD

ownloaded from

Page 11: Cowen 2012

Few ACC Neurons Discriminate Between Forced andChoice Trials

Given the involvement of the ACC in decision making, itwas conceivable that spiking activity would differ betweenforced and choice trials. Such an analysis, however, shouldbegin with the caveat that the effort and reward discriminationtask used in this study does not challenge the animals to

reevaluate options once the optimal arm is identified. In con-trast, tasks such as probability matching (Sugrue et al. 2004) orgame-theoretic behaviors (Barraclough et al. 2004) requireconsistent reevaluation of available options.

To determine whether ACC neurons responded differently tochoice and forced trials, the populations of path (n � 196)-,effort (n � 80)-, and reward (n � 44)-selective neurons

Fig. 8. Modulation by running speed. A–C: running speed was measured at 7 locations on the track and by experimental condition (path, effort, and reward level).Significant differences in running speed were observed under each condition (*P � 0.05, n � 58 sessions, paired signed-rank test, Bonferroni correction for multiplecomparisons). The largest differences were observed under the effort condition, with rats slowing down on high-effort trials (B). D and E: neurons identified as selectivefor the chosen path, level of effort, or level of reward (see Fig. 4) were compared with neurons sensitive to running speed. Sensitivity to running speed was calculatedby determining the correlation between trial-by-trial measures of speed with trial-by-trial spike counts. The explained variance (R2) and significance of this correlationwere used as measures of the strength and significance of the influence of running speed. This analysis was restricted to the approach to the doorway (Dw). D: numberof path-, effort-, and reward-selective neurons (black bars) that were also selective for running speed (gray bars). A trial-shuffled control is included for comparison (whitebars). Right y-axis indicates the percentage from the population of recorded neurons (n � 380). E: amount of explained variance (R2) between running speed and neuralactivity was computed for the subsets of path-, effort-, and reward-selective neurons. To determine whether values were above chance, an R2 computed from randomlypermuting trial order was subtracted from the unshuffled measure. Values differed significantly from zero for the path-selective neurons (*Ppath � 0.0000004, pairedt-test), but not for the reward- or effort-selective cells (Peffort � 0.08, Preward � 0.26, paired t-test).

Fig. 9. Effort- and reward-selective neurons respond to the type of outcome but not to outcome value. If ACC neurons respond to the expected value of outcomes,regardless of outcome type, then firing responses to positive outcomes should be similar in the effort and reward conditions (e.g., respond to both low effort andhigh reward). Responses to low-value outcomes should also generalize across conditions. To examine this issue, the difference in firing activity to anticipatedlow or high effort was compared to differences between high and low reward. These 2 values were computed for each neuron and plotted. Analysis was restric-ted to the 600-ms window preceding door entry. A response to value predicts that neurons will cluster in the top right or bottom left quadrants of the plot andregression analysis should indicate a positive slope. A: responses of all ACC neurons. A negative regression slope was observed (P � 0.00006, r � �0.21). Barsat bottom right of each plot represent the relative number of points in the top left and bottom right quadrants (left bar) relative to the points in the bottom leftand top right quadrants (right bar). B and C: slope was negative for the populations of effort (B)- and reward (C)-selective neurons; however, the regression wassignificant for the population of effort-selective cells (P � 0.0002, r � �0.41) but not for the reward-selective neurons (P � 0.26, r � �0.20).

2402 ANTERIOR CINGULATE RESPONSES TO ACTION AND EFFORT

J Neurophysiol • doi:10.1152/jn.01012.2011 • www.jn.org

on May 11, 2012

jn.physiology.orgD

ownloaded from

Page 12: Cowen 2012

identified at the doorway (Fig. 4C) were analyzed for theircapacity to discriminate between trials in which either one ortwo doors were open. Trials were restricted to those in whichanimals ultimately entered the low-effort and high-rewardpaths. This restriction was necessary as animals infrequentlyselected the nonoptimal paths (e.g., high effort or low reward)during choice trials. Results from the analysis indicated thatfew path (12 of 196 neurons)-, effort (9 of 80 neurons)-, andreward (4 of 44 neurons)-selective neurons discriminated be-tween forced and choice trials (P � 0.05, ANOVA).

Subset of Neurons Responding to Effort or Reward ChangesThrough Course of a Trial

Our analyses demonstrated that the population of ACCneurons maintained selectivity for effort or reward from thepoint at which animals approached the doorway to the point atwhich they reached the reward (Figs. 3, 4D, 10A). This obser-vation suggests that individual ACC neurons maintain a per-sistent trace of the anticipated outcome through the course of atrial. Alternatively, individual ACC neurons could remainselective for only brief periods or across short stretches of thetrack. In the latter case, different subpopulations of cells wouldregister anticipated effort and reward for different track posi-tions. Such a result would be more consistent with neuronsbeing sensitive to particular actions or action sequences.

To differentiate between these two forms of activity, wedetermined the extent to which outcome-selective neuronsmaintained their selective responses through the duration of atrial (Fig. 10B). The analysis indicated that �40% of effort-selective neurons maintained selectivity through the course ofthe trial (dashed line in Fig. 10B). The value fell to 15% forreward-selective neurons. Similar results were obtained whenthe analysis was performed with the original population ofselective neurons identified at subsequent positions along thetrack (Fig. 10C). These observations suggest that most ACCneurons do not maintain a consistent or sustained selectiveresponse for outcomes. Instead, different populations of cells

become selective for constrained portions of the task. Thissuggestion was supported further by results from the analysisof population firing activity at each location along the track(Fig. 11; see Fujisawa et al. 2008). In this analysis, a statevector matrix was created composed of the mean activity ofeach neuron (n � 380) at each location on the track (n � 150,bin size � �3 cm). The similarity between state vectors wascompared by computing all possible pairwise correlations(Pearson) between states (corrcoef, MATLAB). These matriceswere created for each major trial condition (left/right turn,high/low reward, high/low effort) and then averaged to pro-duce the state vector correlation matrix presented in Fig. 11B.The observed strength of the correlations along the diagonal ofthe matrix and the relative weakness of correlations off thediagonal suggest that population activity in the ACC at eachlocation is relatively unique. This also suggests that populationstates in the ACC continually change through the course of atrial (Fujisawa et al. 2008; Lapish et al. 2008). A variation tothis pattern appears to occur during the return journey thatfollows reward consumption. During this section of the track,state vectors remained similar for the majority of the journey.To conclude, aside from the interesting exception of the returnjourney, results from the preceding analyses suggest that thepopulation of ACC neurons responds to each phase of the taskwith a unique pattern of activity.

DISCUSSION

The ACC plays an important role in adaptive behavior andeffort- and reward-guided decision making (Ragozzino andRozman 2007; Rushworth et al. 2003; Walton et al. 2007).How the activities of ACC neurons support such behaviors isunknown; however, connectivity between the ACC and motorand limbic structures (Gabbott et al. 2005) suggests that theACC integrates information about actions and their anticipatedoutcomes. To identify ACC responses to actions, effort, andreward, we varied all of these conditions in a reversal behaviorso that specific sensitivities of ACC neurons to each of these

w w w

www w w w

Fig. 10. Populations of outcome-selective neurons changeduring the course of a trial. A: number of neurons determinedto be selective (n � 380, P � 0.05 ANOVA) for effort andreward at 4 successive locations along the track. Dashed lineindicates chance. B: % of cells determined to be selective forthe outcome during the approach to the doorway (from “D” inA) that also remained selective at subsequent locations alongthe track. The % of neurons that retained their selectivityduring subsequent locations fell to �40% for the effort con-dition even though the overall number of effort-selective cellsdid not decrease. The effect for reward-selective neurons wasalso large, as �15% of reward-selective neurons maintainedtheir selectivity through the trial. C: same analysis as B exceptbased on the initial population of selective cells at the 3remaining locations along the track (indicated by underlinedlabel).

2403ANTERIOR CINGULATE RESPONSES TO ACTION AND EFFORT

J Neurophysiol • doi:10.1152/jn.01012.2011 • www.jn.org

on May 11, 2012

jn.physiology.orgD

ownloaded from

Page 13: Cowen 2012

factors could be assessed. Populations of ACC neurons werefound to be notably sensitive to movement along the left andright paths and to the anticipated presence or absence of aclimbable obstacle that stood between the animal and reward.Given mounting evidence linking obstacle avoidance to ACCfunction and effort-guided decision making (Floresco and Ghods-Sharifi 2007; Rudebeck et al. 2006; Schweimer and Hauber2006; Walton et al. 2003), we interpret the neural sensitivity tothe obstacle as a likely response to anticipated physical effort.Even so, it must also be acknowledged that such discriminativeresponses could be influenced by other affective or visceralreactions to the barrier such as the anticipated discomfortassociated with climbing the obstacle.

In contrast to the sensitivity of neurons to the path and anti-cipated effort, few neurons were sensitive to anticipated re-ward. In addition, neural selectivity for the path, effort, andreward only occurred after animals initiated movements towardthe chosen path, a result suggesting ACC involvement inmaintaining and potentially modulating ongoing goal-directedactions according to the magnitude of expected outcomes. Thisrole is supported by the observation that half of effort- andreward-selective neurons also discriminated between the paths

after controlling for any sensitivity for the specific outcome(through the use of path-outcome reversals). Furthermore,ACC neurons did not exhibit any significant retrospectiveresponse for past choices or outcomes.

ACC Neurons Associate Actions with Anticipated Effort

A specialization for processing effort in the ACC is sug-gested by the fact that neural activity was more selective foreffort relative to reward (Fig. 4) and differences in firing ratebetween low and high effort were larger than differencesobserved for reward (Fig. 7). Furthermore, over half of effort-selective neurons discriminated between the two paths (Fig. 6),suggesting that separate populations of neurons may be in-volved in associating actions with their effortful consequencesor for processing effort alone.

ACC Involvement in Ongoing but Not Planned Behavior

Neural activity in the ACC was found to be selective formovements along the left and right paths and for the associatedlevel of effort and reward. However, selectivity for thesefeatures only appeared after animals began moving toward oneof the open doorways. That is, path-selective ACC activity wasnot prospective in the sense of predicting an upcoming decisionor action plan. However, effort- and reward-related activitywas prospective in that it emerged well before the time andplace of effort and reward delivery.

The lack of path-predicting ACC activity suggests that, atleast for this task, ACC activity is not related to the action-planning component of decision making. This finding is con-sistent with prior experiments in which little evidence forprospective responses for actions and trajectories was found fora probabilistic decision-making task (Sul et al. 2010) and for aworking memory task (Euston and McNaughton 2006). Incontrast, prospective responses for future goals or actions wereobserved during an odor-cued choice behavior (Fujisawa et al.2008) and an alternation task (Baeg et al. 2003). Differencesbetween these studies could potentially be explained by uniquefeatures of the behavioral tasks or by the fact that most studiesinclude neurons from multiple medial prefrontal structures.Indeed, of the publications considered above, only one (see Sulet al. 2010) examined recordings specific to the ACC, and thisstudy did not find strong evidence for action selection.

ACC Neurons Responded to Specific Outcomes but Not toOutcome Value

An important question regarding ACC function is whetherACC neurons respond to specific outcomes (e.g., anticipatedeffort or reward) or to outcome value. Value-driven responsescould be useful as they would indicate the utility of optionsregardless of outcome type. Contrary to this suggestion, wefound little evidence for value-driven responses in the ACC inour task. For example, only 13 of 380 neurons responded toboth effort and reward (Fig. 6C), and ACC neurons did notexpress value-driven responses across effort and reward con-ditions (Fig. 9). The absence of an observed value-drivenresponse complements and contrasts with observations fromthe few published physiological studies of cost-benefit andeffort-guided decision making. In agreement with our obser-vations, Kennerley and Wallis (2009) describe populations ofneurons in the primate ACC that responded to sensorimotor

Fig. 11. Neuronal activity states respond to localized regions of the track.A: schematic of the track with colors corresponding to locations indicated onthe axes of B. B: state vector correlation matrix for the population of recordedneurons (n � 380) at all locations on the track. This matrix was generated byfirst determining the mean firing rate of each neuron at each location on thetrack (bin size � �3 cm). This produced a neuron (380 neurons) by position(150 positions) matrix. The similarity between all of the state vectors wasmeasured with the Pearson’s correlation coefficient and stored in a position-by-position correlation coefficient matrix. These matrices were created foreach major trial condition (left/right turn, high/low reward, high/low effort).These individual matrices were then averaged to produce the state correlationmatrix presented in B. The pattern of correlation indicates that populationstates tended to remain correlated with nearby states at nearby locations anduncorrelated with other sections of the track. This suggests that populationactivity in the ACC changes dynamically through the course of the trial. Anexception to this pattern occurs during the return journey to the start trigger(orange region on the track). During this section of the track, state vectorsremained similar for the majority of the journey.

2404 ANTERIOR CINGULATE RESPONSES TO ACTION AND EFFORT

J Neurophysiol • doi:10.1152/jn.01012.2011 • www.jn.org

on May 11, 2012

jn.physiology.orgD

ownloaded from

Page 14: Cowen 2012

activity, reward, and effort. In contrast to our observations,however, the majority of effort-selective neurons also re-sponded to reward. In addition, the authors presented evidencethat neurons responded to the overall value of an outcomeregardless of its form, a result that is also supported by a recenthuman imaging study (Croxson et al. 2009). In contrast, wefound little evidence for value-specific responses and observedthat individual neurons in the rat ACC appear to be morespecialized for processing actions and effort.

Functional variations within the ACC may contribute to thedifferences observed between primates and rats. For example,Kennerley and Wallis (2009) observed that many more neuronsresponded to conjunctions between effort and reward in theanterior relative to the posterior ACC. Differences have alsobeen observed within the rodent ACC. For example, the sen-sitivity of ACC neurons to sensorimotor features varies alongthe dorsoventral axis (Cowen and McNaughton 2007). Indeed,the median depth of the electrodes (1,710 �m) indicated thatmost recordings were in the dorsal ACC, a region that may bemore sensitive to specific sensorimotor features associated withoutcomes.

Populations of Effort- and Reward-Selective NeuronsChange Through Course of a Trial

In the present work, the form of anticipatory effort- andreward-related activity was found to be dynamic in nature aslargely different populations of neurons exhibited effort- orreward-related activity at different locations along the track(Fig. 10). This result sets anticipatory activity in the ACC apartfrom such activity observed in other regions of the brain duringreward anticipation. For example, dopamine neurons can fire asa group and in brief bursts in response to expected reward(Fiorillo et al. 2003), while striatal neurons exhibit persistent,ramping activity profiles to reward anticipation (Apicella et al.1992). The result from the present study parallels recentfindings obtained from the posterior parietal cortex of mon-keys, where, across time, object-centered activity was given bychanging populations of neurons (Crowe et al. 2010), and inthe rat, where transiently active and changing populations ofmedial prefrontal neurons were found to discriminate betweentrajectories and locations on a T-maze (Fujisawa et al. 2008)and to changes in task epoch or decision-making errors (Lapishet al. 2008). The dynamic character of path-, effort-, andreward-related activity yields, as a consequence, a distinctpattern of ensemble activity within ACC for each path positionand its associated action. Below, we consider how this featureof ACC activity may provide clues as to the overall function ofthe ACC.

Proposed Role for ACC Activity in Action Maintenance inthe Face of Effortful Outcomes

A significant role for the ACC in supporting ongoing asopposed to prospective motor activity is suggested by theobserved predominance of path-selective responses and thelack of prospective activity for the path. In this context,ACC activity reflecting anticipated effort and reward can beinterpreted as supporting motor behavior carried out subse-quent to an initial decision. That is, effort-selective ACCactivity could be critical to carrying out a behavioral planthat demands greater effort, in the short term, so that an

animal may reach a desired outcome. This interpretation isconsistent with the behavioral consequences of ACC inac-tivation on cost-benefit behaviors that require animals tochoose between high-effort/high-reward and low-effort/low-reward alternatives (Floresco and Ghods-Sharifi 2007; Hau-ber and Sommer 2009; Schweimer et al. 2005; Walton et al.2003). The most common observation from these studies isthat lesioned animals avoid the previously preferred high-effort/high-reward option in favor of a low-effort/low-re-ward alternative. A reasonable interpretation of the role ofthe ACC in this behavior is for evaluating future outcomesin order to bias decision making toward the high-reward/high-effort alternative. In consideration of our observationsand of the behavioral consequences of ACC inactivation, wepropose that ACC inactivation reduces the capacity ofanimals to sustain the motivation required to maintain theset of actions associated with a costly alternative. Withoutsuch a capacity, animals lacking an ACC may initially selectand even move toward the high-effort/high-reward alterna-tive but abandon this plan once the specific sequence ofactions leading to the high-effort outcome is initiated (e.g.,when animals reach the base of the barrier). As an alterna-tive explanation, it is conceivable that the tasks used incost-benefit behaviors may place additional evaluative de-mands on the ACC relative to the effort- and reward-guidedbehavior used in the present experiment. The additionaldemands may change ACC responses, resulting in strongerprospective activity for effort and reward and perhaps alsotriggering value-driven responses in the region. This activitymay be required for the evaluation of conflicting outcomes.The possibility that the ACC responds differently duringeffort discrimination as opposed to effort-guided cost-ben-efit evaluations has been suggested by Hillman and Bilkey(2010).

The connection between the ACC and affective and mo-tivational centers such as the amygdala and the nucleusaccumbens supports ACC involvement in connecting motoractivity with outcome expectation (Conde et al. 1995; Gab-bott et al. 2005; Sesack et al. 1989). ACC responses to theanticipation of effort may trigger activity in efferent struc-tures that, in turn, generates the motivational responserequired to sustain the motor plan required to overcome thephysical obstacle. The rich responses of ACC neurons tomovements, paths, and task phase suggest that any suchprocessing must be intimately tied with the ongoing motorbehavior and sensorimotor feedback, and therefore goesbeyond the calculation of value. Integrating these ideas, wepropose that movement- and effort-related activity in theACC triggers a persistent motivational signal that drives theperformance of goal-directed movements, especially whenthose movements are associated with anticipated physicaleffort. Future studies designed to evaluate the behavioraland physiological responses of animals to effort-associatedcues will be required to directly evaluate this proposal.

ACKNOWLEDGMENTS

The authors give special thanks to Gerald Edelman, Einar Gall, NathanInsel, Jason Fleischer, Fred Jones, Paul Greenberg, John Iversen, and JosephGally for comments and critiques.

The views, opinions, and/or findings contained in this article are those of theauthors and should not be interpreted as representing the official views or

2405ANTERIOR CINGULATE RESPONSES TO ACTION AND EFFORT

J Neurophysiol • doi:10.1152/jn.01012.2011 • www.jn.org

on May 11, 2012

jn.physiology.orgD

ownloaded from

Page 15: Cowen 2012

policies, either expressed or implied, of the Defense Advanced ResearchProjects Agency (DARPA) or the Department of Defense. Approved for PublicRelease, Distribution Unlimited.

GRANTS

This work was supported by grants to the Neurosciences Research Foun-dation from The G. Harold and Leila Y. Mathers Charitable Foundation andfrom DARPA through Office of Naval Research (ONR) Grant N00014-08-1-0728. S. L. Cowen was the Blasker-Rose-Miah Fellow of The San DiegoFoundation.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

Author contributions: S.L.C. and D.A.N. conception and design of research;S.L.C. and G.A.D. performed experiments; S.L.C. and G.A.D. analyzeddata; S.L.C. and D.A.N. interpreted results of experiments; S.L.C. preparedfigures; S.L.C. drafted manuscript; S.L.C., G.A.D., and D.A.N. edited andrevised manuscript; S.L.C., G.A.D., and D.A.N. approved final version ofmanuscript.

REFERENCES

Amiez C, Joseph JP, Procyk E. Reward encoding in the monkey anteriorcingulate cortex. Cereb Cortex 16: 1040–1055, 2006.

Apicella P, Scarnati E, Ljungberg T, Schultz W. Neuronal activity inmonkey striatum related to the expectation of predictable environmentalevents. J Neurophysiol 68: 945–960, 1992.

Baeg EH, Kim YB, Jang J, Kim HT, Mook-Jung I, Jung MW. Fast spikingand regular spiking neural correlates of fear conditioning in the medialprefrontal cortex of the rat. Cereb Cortex 11: 441–451, 2001.

Baeg EH, Kim YB, Huh K, Mook-Jung I, Kim HT, Jung MW. Dynamicsof population code for working memory in the prefrontal cortex. Neuron 40:177–188, 2003.

Bai J, Perron P. Computation and analysis of multiple structural changemodels. J Appl Econometrics 18: 1–22, 2003.

Barraclough DJ, Conroy ML, Lee D. Prefrontal cortex and decision makingin a mixed-strategy game. Nat Neurosci 7: 404–410, 2004.

Chow GC. Tests of equality between sets of coefficients in two linearregressions. Econometrica 28: 591–605, 1960.

Churchland MM, Yu BM, Cunningham JP, Sugrue LP, Cohen MR,Corrado GS, Newsome WT, Clark AM, Hosseini P, Scott BB, BradleyDC, Smith MA, Kohn A, Movshon JA, Armstrong KM, Moore T,Chang SW, Snyder LH, Lisberger SG, Priebe NJ, Finn IM, Ferster D,Ryu SI, Santhanam G, Sahani M, Shenoy KV. Stimulus onset quenchesneural variability: a widespread cortical phenomenon. Nat Neurosci 13:369–378, 2010.

Conde F, Maire-Lepoivre E, Audinat E, Crepel F. Afferent connections ofthe medial frontal cortex of the rat. II. Cortical and subcortical afferents. JComp Neurol 352: 567–593, 1995.

Cowen SL, McNaughton BL. Selective delay activity in the medial prefrontalcortex of the rat: contribution of sensorimotor information and contingency.J Neurophysiol 98: 303–316, 2007.

Crowe DA, Averbeck BB, Chafee MV. Rapid sequences of populationactivity patterns dynamically encode task-critical spatial information inparietal cortex. J Neurosci 30: 11640–11653, 2010.

Croxson PL, Walton ME, O’Reilly JX, Behrens TEJ, Rushworth MFS.Effort-based cost-benefit valuation and the human brain. J Neurosci 29:4531–4541, 2009.

Euston DR, McNaughton BL. Apparent encoding of sequential context in ratmedial prefrontal cortex is accounted for by behavioral variability. J Neu-rosci 26: 13143–13155, 2006.

Fiorillo CD, Tobler PN, Schultz W. Discrete coding of reward probabilityand uncertainty by dopamine neurons. Science 299: 1898–1902, 2003.

Floresco SB, Ghods-Sharifi S. Amygdala-prefrontal cortical circuitry regu-lates effort-based decision making. Cereb Cortex 17: 251–260, 2007.

Fujisawa S, Amarasingham A, Harrison MT, Buzsáki G. Behavior-depen-dent short-term assembly dynamics in the medial prefrontal cortex. NatNeurosci 11: 823–833, 2008.

Gabbott PL, Warner TA, Jays PR, Salway P, Busby SJ. Prefrontal cortexin the rat: projections to subcortical autonomic, motor, and limbic centers. JComp Neurol 492: 145–177, 2005.

Green DM, Swets JA. Signal Detection Theory and Psychophysics. Los Altos,CA: Peninsula, 1989.

Gur M, Beylin A, Snodderly DM. Response variability of neurons in primaryvisual cortex (V1) of alert monkeys. J Neurosci 17: 2914–20, 1997.

Hauber W, Sommer S. Prefrontostriatal circuitry regulates effort-relateddecision making. Cereb Cortex 19: 2240–2247, 2009.

Hikosaka O, Isoda M. Switching from automatic to controlled behavior:cortico-basal ganglia mechanisms. Trends Cogn Sci 14: 154–161, 2010.

Hillman KL, Bilkey DK. Neurons in the rat anterior cingulate cortex dynam-ically encode cost-benefit in a spatial decision-making task. J Neurosci 30:7705–7713, 2010.

Hoover WB, Vertes R. Anatomical analysis of afferent projections to themedial prefrontal cortex in the rat. Brain Struct Funct 212: 149–179, 2007.

Hussar C, Pasternak T. Trial-to-trial variability of the prefrontal neuronsreveals the nature of their engagement in a motion discrimination task. ProcNatl Acad Sci USA 107: 21842–21847, 2010.

Jung MW, Qin Y, McNaughton BL, Barnes CA. Firing characteristics ofdeep layer neurons in prefrontal cortex in rats performing spatial workingmemory tasks. Cereb Cortex 8: 437–450, 1998.

Kara P, Reinagel P, Reid RC. Low response variability in simultaneouslyrecorded retinal, thalamic, and cortical neurons. Neuron 27: 635–646, 2000.

Kargo WJ, Nitz DA. Improvements in the signal-to-noise ratio of motorcortex cells distinguish early versus late phases of motor skill learning. JNeurosci 24: 5560–5569, 2004.

Kargo WJ, Szatmary B, Nitz DA. Adaptation of prefrontal cortical firingpatterns and their fidelity to changes in action-reward contingencies. JNeurosci 27: 3548–3559, 2007.

Kennerley SW, Walton ME, Behrens TEJ, Buckley MJ, Rushworth MFS.Optimal decision making and the anterior cingulate cortex. Nat Neurosci 9:940–947, 2006.

Kennerley SW, Dahmubed AF, Lara AH, Wallis JD. Neurons in the frontallobe encode the value of multiple decision variables. J Cogn Neurosci 21:1162–1178, 2009.

Kennerley SW, Wallis JD. Evaluating choices by single neurons in the frontallobe: outcome value encoded across multiple decision variables. Eur JNeurosci 29: 2061–2073, 2009.

Kimchi EY, Laubach M. Dynamic encoding of action selection by the medialstriatum. J Neurosci 29: 3148–3159, 2009.

Lapish CC, Durstewitz D, Chandler LJ, Seamans JK. Successful choicebehavior is associated with distinct and coherent network states in anteriorcingulate cortex. Proc Natl Acad Sci USA 105: 11963–11968, 2008.

McCoy AN, Platt ML. Risk-sensitive neurons in macaque posterior cingulatecortex. Nat Neurosci 8: 1220–1227, 2005.

McNaughton BL, O’Keefe J, Barnes CA. The stereotrode: a new techniquefor simultaneous isolation of several single units in the central nervoussystem from multiple unit records. J Neurosci Methods 8: 391–397, 1983.

Mitchell JF, Sundberg KA, Reynolds JH. Differential attention-dependentresponse modulation across cell classes in macaque visual area V4. Neuron55: 131–141, 2007.

Narayanan NS, Laubach M. Top-down control of motor cortex ensembles bydorsomedial prefrontal cortex. Neuron 52: 921–931, 2006.

Paxinos G, Watson C. The Rat Brain in Stereotaxic Coordinates. New York:Elsevier, 1998.

Pratt WE, Mizumori SJ. Neurons in rat medial prefrontal cortex showanticipatory rate changes to predictable differential rewards in a spatialmemory task. Behav Brain Res 123: 165–183, 2001.

Procyk E, Joseph JP. Characterization of serial order encoding in the monkeyanterior cingulate sulcus. Eur J Neurosci 14: 1041–1046, 2001.

Ragozzino ME, Kesner RP. The role of rat dorsomedial prefrontal cortex inworking memory for egocentric responses. Neurosci Lett 308: 145–148,2001.

Ragozzino ME, Rozman S. The effect of rat anterior cingulate inactivation oncognitive flexibility. Behav Neurosci 121: 698–706, 2007.

Ridderinkhof KR, Ullsperger M, Crone EA, Nieuwenhuis S. The role of themedial frontal cortex in cognitive control. Science 306: 443–447, 2004.

Rudebeck PH, Walton ME, Smyth AN, Bannerman DM, Rushworth MFS.Separate neural pathways process different decision costs. Nat Neurosci 9:1161–1168, 2006.

Rushworth MFS, Walton ME, Kennerley SW, Bannerman DM. Actionsets and decisions in the medial frontal cortex. Trends Cogn Sci 8: 410–417,2004.

2406 ANTERIOR CINGULATE RESPONSES TO ACTION AND EFFORT

J Neurophysiol • doi:10.1152/jn.01012.2011 • www.jn.org

on May 11, 2012

jn.physiology.orgD

ownloaded from

Page 16: Cowen 2012

Rushworth MF, Hadland KA, Gaffan D, Passingham RE. The effect ofcingulate cortex lesions on task switching and working memory. J CognNeurosci 15: 338–353, 2003.

Salamone JD, Cousins MS, Bucher S. Anhedonia or anergia? Effects ofhaloperidol and nucleus accumbens dopamine depletion on instrumentalresponse selection in a T-maze cost/benefit procedure. Behav Brain Res 65:221–229, 1994.

Schweimer J, Saft S, Hauber W. Involvement of catecholamine neurotrans-mission in the rat anterior cingulate in effort-related decision making. BehavNeurosci 119: 1687–1692, 2005.

Schweimer J, Hauber W. Dopamine D1 receptors in the anterior cingulatecortex regulate effort-based decision making. Learn Mem 13: 777–782,2006.

Sesack SR, Deutch AY, Roth RH, Bunney BS. Topographical organizationof the efferent projections of the medial prefrontal cortex in the rat: ananterograde tract-tracing study with Phaseolus vulgaris leucoagglutinin. JComp Neurol 290: 213–242, 1989.

Shima K, Tanji J. Role for cingulate motor area cells in voluntary movementselection based on reward. Science 282: 1335–1338, 1998.

Sugrue LP, Corrado GS, Newsome WT. Matching behavior and the repre-sentation of value in the parietal cortex. Science 304: 1782–1787, 2004.

Sul JH, Kim H, Huh N, Lee D, Jung MW. Distinct roles of rodentorbitofrontal and medial prefrontal cortex in decision making. Neuron 66:449–460, 2010.

Walton ME, Bannerman DM, Rushworth MF. The role of rat medial frontalcortex in effort-based decision making. J Neurosci 22: 10996–11003, 2002.

Walton ME, Bannerman DM, Alterescu K, Rushworth MFS. Functionalspecialization within medial frontal cortex of the anterior cingulate forevaluating effort-related decisions. J Neurosci 23: 6475–6479, 2003.

Walton ME, Rudebeck PH, Bannerman DM, Rushworth MF. Calculatingthe cost of acting in frontal cortex. Ann NY Acad Sci 1104: 340–356, 2007.

Zeileis A, Leisch F, Hornik K, Kleiber C. strucchange: An R package fortesting for structural change in linear regression models. J Stat Software 7:1–38, 2002.

2407ANTERIOR CINGULATE RESPONSES TO ACTION AND EFFORT

J Neurophysiol • doi:10.1152/jn.01012.2011 • www.jn.org

on May 11, 2012

jn.physiology.orgD

ownloaded from