Transfer of Motor Learning Engages Specific Neural Substrates During Motor Memory Consolidation Dependent on the Practice Structure

This article was downloaded by: [USC University of Southern California], [Carolee Winstein]On: 18 February 2012, At: 13:55Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,37-41 Mortimer Street, London W1T 3JH, UK

Journal of Motor BehaviorPublication details, including instructions for authors and subscription information:http://www.tandfonline.com/loi/vjmb20

Transfer of Motor Learning Engages Specific NeuralSubstrates During Motor Memory ConsolidationDependent on the Practice StructureShailesh S. Kantak a b , Katherine J. Sullivan a , Beth E. Fisher a b , Barbara J. Knowlton c &Carolee J. Winstein a da Motor Behavior and Neurorehabilitation Laboratory, Division of Biokinesiology and PhysicalTherapy at the Herman Ostrow School of Dentistry, University of Southern California, LosAngelesb Neuroplasticity and Imaging Laboratory, Division of Biokinesiology and Physical Therapy atthe Herman Ostrow School of Dentistry, University of Southern California, Los Angelesc Department of Psychology, University of California, Los Angelesd Department of Neurology, Keck School of Medicine, University of Southern California, LosAngeles

Available online: 08 Dec 2011

To cite this article: Shailesh S. Kantak, Katherine J. Sullivan, Beth E. Fisher, Barbara J. Knowlton & Carolee J. Winstein(2011): Transfer of Motor Learning Engages Specific Neural Substrates During Motor Memory Consolidation Dependent on thePractice Structure, Journal of Motor Behavior, 43:6, 499-507

To link to this article: http://dx.doi.org/10.1080/00222895.2011.632657

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions

This article may be used for research, teaching, and private study purposes. Any substantial or systematicreproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form toanyone is expressly forbidden.

The publisher does not give any warranty express or implied or make any representation that the contentswill be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses shouldbe independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims,proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly inconnection with or arising out of the use of this material.

Journal of Motor Behavior, Vol. 43, No. 6, 2011Copyright C© Taylor & Francis Group, LLC

RESEARCH ARTICLE

Transfer of Motor Learning Engages Specific Neural SubstratesDuring Motor Memory Consolidation Dependent on the PracticeStructureShailesh S. Kantak1,2, Katherine J. Sullivan1, Beth E. Fisher1,2, Barbara J. Knowlton3, Carolee J. Winstein1,4

1Motor Behavior and Neurorehabilitation Laboratory, Division of Biokinesiology and Physical Therapy at the Herman OstrowSchool of Dentistry, University of Southern California, Los Angeles. 2Neuroplasticity and Imaging Laboratory, Division ofBiokinesiology and Physical Therapy at the Herman Ostrow School of Dentistry, University of Southern California, LosAngeles. 3Department of Psychology, University of California, Los Angeles. 4Department of Neurology, Keck School ofMedicine, University of Southern California, Los Angeles.

ABSTRACT. The authors investigated how brain activity duringmotor-memory consolidation contributes to transfer of learning tonovel versions of a motor skill following distinct practice struc-tures. They used 1 Hz repetitive Transcranial Magnetic Stimulation(rTMS) immediately after constant or variable practice of an armmovement skill to interfere with primary motor cortex (M1) ordorsolateral prefrontal cortex (DLPFC). The effect of interferencewas assessed through skill performance on two transfer targets: onewithin and one outside the range of practiced movement parametersfor the variable practice group. For the control (no rTMS) group,variable practice benefited delayed transfer performance more thanconstant practice. The rTMS effect on delayed transfer performancediffered for the two transfer targets. For the within-range target,rTMS interference had no significant affect on the delayed transferafter either practice structure. However, for the outside-range tar-get, rTMS interference to DLPFC but not M1 attenuated delayedtransfer benefit following variable practice. Additionally, for theoutside-range target, rTMS interference to M1 but not DLPFC at-tenuated delayed transfer following constant practice. This suggeststhat variable practice may promote reliance on DLPFC for memoryconsolidation associated with outside-range transfer of learning,whereas constant practice may promote reliance on M1 for consol-idation and long-term transfer.

Keywords: consolidation, transcranial magnetic stimulation, trans-fer of learning

Motor memory consolidation is a postpractice time-dependent process that results in a more stable motor

memory over time. Interference to motor memory consol-idation during the immediate postacquisition period oftendegrades learning of motor skills, suggesting the critical roleof consolidation processes in motor learning (Luft, Buitrago,Ringer, Dichgans, & Schulz, 2004; Muellbacher et al., 2002;Shadmehr & Brashers-Krug, 1997). Recently, we demon-strated that the neural substrates for motor memory consoli-dation depend on the structure of practice (Kantak, Sullivan,Fisher, Knowlton, & Winstein, 2010). Interference to primarymotor cortex (M1) affected retention of the practiced skillfollowing constant practice but not variable practice. In con-trast, interference to dorsolateral prefrontal cortex (DLPFC)affected retention of the skill when the practice structure wasvariable but not when the practice structure was constant.

Although retention of a skill requires the learner to recallthe practiced skill, transfer to a novel version of a task re-quires an ability to generalize what was learned to a new

parameter or environment. It is well established that whenan individual is required to learn multiple tasks or multiplevariants of a task, if practice is organized using a variablepractice structure, transfer performance on a novel versionof the task is better than if a constant practice structure isused (Guadagnoli & Lee, 2004; Hall & Magill, 1995; Lee &Simon, 2003; Schmidt & Bjork, 1992; Shea & Kohl, 1990,1991; Shea, Lai, Wright, Immink, & Black, 2001; Sherwood& Lee, 2003; Wulf & Lee, 1993). Practice of a task in thecontext of other task versions, such as when using a variablepractice structure, allows the learner to compare and contrastthe various task versions and in so doing process the task-related information more deeply than when the same task isrepeated trial after trial. This in-depth processing facilitateslearning of abstract rules that define the relationship betweenthe goal and action parameters required to achieve that goal.Such abstract knowledge is known to facilitate retention ofthe practiced skill and transfer to a novel skill. Specifically,transfer to a novel skill or novel version of the skill reliesmore on this abstract knowledge such that the learner caneffectively modify the action parameters required to achievethe new goal.

The purpose of this study was to extend our previous in-vestigation to explore how activity in M1 and DLPFC dur-ing motor memory consolidation specifically contributes totransfer of learning to novel versions of the motor skill fol-lowing motor practice under variable or constant conditions.We systematically manipulated the variability of motor taskpractice to differentially affect transfer of motor learning.We used 1-Hz repetitive Transcranial Magnetic Stimulation(rTMS) immediately following constant or variable motorpractice to interfere with M1 or DLPFC. The effect of rTMSinterference on transfer of learning was assessed throughmotor performance in a transfer test administered 1 day afterpractice. Based on our previous findings (Kantak, Sullivan,et al., 2010), we hypothesized that postpractice interferenceto M1 and DLPFC would attenuate transfer performance to a

Correspondence address: Shailesh S. Kantak, Division of Bioki-nesiology and Physical Therapy, Herman Ostrow School of Den-tistry, University of Southern California, 1540 E. Alcazar St., CHP155, Los Angeles, CA 90089, USA. e-mail: [email protected]

499

Dow

nloa

ded

by [

USC

Uni

vers

ity o

f So

uthe

rn C

alif

orni

a], [

Car

olee

Win

stei

n] a

t 13:

55 1

8 Fe

brua

ry 2

012

S. S. Kantak, K. J. Sullivan, B. E. Fisher, B. J. Knowlton, & C. J. Winstein

novel skill following constant and variable practice, respec-tively.

Method

Participants

Participants were 59 neurologically healthy volunteers (25men, 34 women; M age = 26.12 years, SD = 7.1 years; agerange = 23–34 years) who gave written informed consentto participate in the experimental protocol approved by theInstitutional Review board of the University of Southern Cal-ifornia. Participants were excluded from the study if they didnot meet the safety criteria for rTMS or Magnetic Reso-nance Imaging (MRI). None of the participants reported anyadverse effects with TMS or MRI during or after the exper-iment. The participants and entire experimental procedureswere the same as those reported in Kantak, Sullivan, et al.(2010). Here we present the results of transfer test perfor-mance that was tested in the protocol.

Instrumentation and Motor Task

Participants practiced to move a light eight lever with theirdominant forearm to replicate as much as possible a target tra-jectory presented on a computer. The target trajectories hada similar two-peaked movement structure and absolute tem-poral duration requirements (800 ms). The peak amplitudecould be manipulated in a way that when the target trajec-tory peak amplitude was 60◦ (60◦), the participant would berequired to move the lever in two forward-backward rever-sals of 60◦ from the start position. A linear potentiometer

recorded lever-position information with movements of thelever away from the body (forward) reflecting upward move-ments on the computer monitor and movements toward thebody (backward) reflecting downward movements on thecomputer monitor. Signals from the potentiometer were con-verted to digital signal by an A/D board of a Compaq 466vcomputer and sampled at 1000 Hz to provide feedback (FB)on the computer monitor. A custom template software pro-gram (Weekly, 2004) was used for manipulation of the targettrajectory (position-time trace) and the interval duration aswell as data storage for offline analysis of each trial.

Experimental Design

Participants were randomly assigned to one of six ex-perimental groups resulting from three rTMS sites (none,M1, DLPFC) by two practice conditions (constant, vari-able): constant practice, no rTMS (CP); variable practice,no rTMS (VP); constant practice, M1 interference (CP-M1);variable practice, M1 interference (VP-M1); constant prac-tice, DLPFC interference (CP-DLP); and variable practice,DLPFC interference (VP-DLP; Figure 1).

The experiment took place over two consecutive days. OnDay 1, each participant practiced the motor task for a to-tal of 120 trials during the acquisition phase. The structureof the practice was different for the constant practice andvariable practice groups. The constant practice groups (CP,CP-M1, and CP-DLP) practiced 120 trials of only one target:60◦ peak amplitude (criterion task). In the variable practicegroups (VP, VP-M1, and VP-DLP), the participants prac-ticed the target of 60◦ peak amplitude for 60 trials and 20

FIGURE 1. Experimental design: Participants practiced the task on Day 1 either under a constant practice condition or variablepractice condition. Immediately following practice, they were tested on an immediate transfer test (IT). Participants were retestedon a delayed transfer test (DT) 24 hr later. Participants from each practice condition (constant and variable) were randomized toa control (no rTMS) group (CP, VP), an M1-interference group (CP-M1, VP-M1), and a DLPFC-interference group (CP-DLP,VP-DLP). The M1-interference groups received 1Hz rTMS over M1 after IT. The DLPFC-interference groups received 1-Hz rTMSover DLPFC to interfere with consolidation processes immediately following IT. The black dots represent the time points for motorcortex excitability testing.

500 Journal of Motor Behavior

Dow

nloa

ded

by [

USC

Uni

vers

ity o

f So

uthe

rn C

alif

orni

a], [

Car

olee

Win

stei

n] a

t 13:

55 1

8 Fe

brua

ry 2

012

Memory Consolidation of Motor Transfer

FIGURE 2. (A) Targets of 30◦, 45◦, 60◦, and 75◦, each witha different absolute amplitude specifications (30◦, 45◦, 60◦,and 75◦, respectively), but a similar movement structure andsame absolute time requirements (800 ms). The constantpractice groups practiced 120 trials of target A3, whereasthe variable practice group practiced 60 trials of target 60◦

and 20 trials of targets 30◦, 45◦, and 75◦ each. The order ofpresentation of the four targets was randomized in the vari-able practice groups. (B) Example of the feedback display.Participant movement trajectory (thin line) superimposed onthe target trajectory (thick line). The root mean square error(RMSE) was displayed along with the trajectories after eachtrial.

additional trials of targets of peak amplitude 30◦, 45◦, and75◦ each (Figure 2A). The presentation order of the fourtarget amplitudes was pseudorandomized in the VP group.Immediately following practice, participants were tested onan immediate transfer (IT) test that comprised an 8-trial blockof target trajectories that were not presented during the ac-quisition phase. The transfer target trajectories had the samedouble-peaked movement structure and absolute timing re-quirement as the acquisition trajectories, but two differenttarget amplitude specifications: 50◦ and 80◦. The M1 inter-ference groups (CP-M1 and VP-M1) received 1-Hz rTMSover the biceps representation of M1 after IT (Figure 1). TheDLPFC interference groups (CP-DLP and VP-DLP) received1-Hz rTMS over the DLPFC similarly, immediately after IT

(Figure 1). On Day 2 (24 ± 3 hr later), participants wereretested on a delayed transfer (DT) test, which was similarto immediate transfer (IT) test in its composition. Beforeeach IT and DT, the participants were tested for retention oftarget 60◦. This retention data has been presented elsewhere(Kantak, Sullivan, et al., 2010). Post hoc, we recruited 12additional participants and randomized them into one of twogroups that received rTMS 4 hr postpractice: CP M1-4 hrand VP DLP-4 hr. The CP M1-4 hr group practiced the skillunder constant conditions and had 1-Hz rTMS applied overM1 4 hr after practice. The VP DLP-4 hr group practiced thetask under variable conditions and rTMS was applied overDLPFC 4 hr after practice. The effect of delayed rTMS wascompared with the immediate postpractice rTMS to investi-gate the temporal specificity of the rTMS effects.

Experimental Procedure

During practice, participants sat comfortably in front of thecomputer monitor with their testing forearm along the arm ofthe lever and their hand grasping the lever handle. For eachtrial, a target trajectory was displayed on the computer moni-tor at the beginning of each trial for a 2000-ms duration, afterwhich the trajectory disappeared from the screen. After 1000ms, a “go” signal was displayed at which point the partici-pant was instructed to move the lever in a manner to replicatethe target trajectory as closely as possible. To replicate thetrajectory, the participant had to perform a coordinated fore-arm movement that consisted of two elbow extension-flexionreversal movements of specified amplitude in the horizontalplane. After a 2000-ms delay following movement, postre-sponse feedback was displayed on the computer screen for a5000-ms duration. Feedback consisted of (a) an overall nu-meric error score (root mean square error [RMSE]) and (b)a graphic representation of the participant’s response, time-locked to onset and superimposed on the target movementpattern. Figure 2B shows an example of an individual trialand postresponse FB display with RMSE and superimposedtrajectory.

rTMS Interference Procedure

We applied 1-Hz rTMS at 110% of resting motor thresh-old (MT) with a Magstim 70 mm figure-eight coil attachedto a Magstim Rapid2 magnetic stimulator. Immediately afterIT, participants in the CP-M1 and VP-M1 groups receiveda 1-Hz train of rTMS for 10 min (600 pulses) at 110% MTintensity over the hotspot of biceps brachii in contralateralM1. We have previously demonstrated that rTMS transientlydownregulated the motor cortical excitability, as evidencedby a reduction in the size of the MEP amplitude post-rTMS(Kantak, Fisher, Sullivan, & Winstein, 2010). This transientdownregulation of cortical excitability has been used as a vir-tual lesion to interfere with time-dependent neurocognitiveprocesses such as consolidation. In the CP-DLP and VP-DLP groups, 1-Hz rTMS at 110% MT intensity for 10 min(600 pulses) was applied over the contralateral DLPFC.

2011, Vol. 43, No. 6 501

Dow

nloa

ded

by [

USC

Uni

vers

ity o

f So

uthe

rn C

alif

orni

a], [

Car

olee

Win

stei

n] a

t 13:

55 1

8 Fe

brua

ry 2

012

S. S. Kantak, K. J. Sullivan, B. E. Fisher, B. J. Knowlton, & C. J. Winstein

The coil was positioned on the scalp overlying the DLPFCusing a structural MRI-based, stereotaxic neuronavigationsystem, Brainsight Frameless (Rogue Research, Montreal,Quebec, Canada). Details of the neuronavigation procedureshave been previously described in Kantak, Sullivan et al.(2010).

Dependent Measures

Performance accuracy (motor behavior) was assessed foracquisition and transfer tests (IT and DT). The dependentmeasure for accuracy included RMSE, which is the aver-age difference between the goal movement trajectory and theparticipant’s response, calculated over the participant’s totalmovement time (Schmidt & Lee, 2004). RMSE was calcu-lated for each trial. For the acquisition phase, only the trialson the criterion task (60◦) from the variable practice groupand corresponding trials in the constant practice group wereincluded in behavioral analysis. Thus, for each participantwe included the performance on the 60 trials of target 60◦ forthe acquisition phase analysis. For each participant, RMSEdata for these 60 trials were averaged into 12 five-trial blocksfor the acquisition phase. For each of the transfer tests (ITand DT), average RMSE for the four trials was calculatedseparately for the 50◦ and 80◦ targets.

Data Analysis

For motor skill practice data during the acquisition phase,we used a 2 Practice Condition (constant, variable) × 3rTMS Site (none, M1, DLPFC) × 12 Block (1–12 acqui-sition blocks) analysis of variance (ANOVA) with repeatedmeasures on the last factor. Because we were specificallyinterested in understanding the effects on postpractice motormemory consolidation, we contrasted the DT performancewith the IT performance to infer offline performance changesduring motor memory consolidation. For the transfer tests,we used a 2 Practice Condition (constant, variable) × 3 rTMSSite (none, M1, DLPFC) × 2 Transfer Test (IT, DT) ANOVAwith repeated measures on the last factor separately for the50◦ and 80◦ target blocks. Also, we ran a separate 2 PracticeCondition (constant, variable) × 3 rTMS Site (none, M1,DLPFC) ANOVA for each IT and DT performance of 50◦

and 80◦ trial blocks. For all statistical tests, the significancelevel was set at p < .05. SPSS version 15.0 statistical softwarewas used for all statistical analyses.

Results

Acquisition Phase

Practice improved motor performance in all the groupsas evidenced by a reduction in the global error (RMSE) inperforming the target-matching task (practice effect), F(1,58) = 351.66, p < .001. There was no statistically significantdifference between the groups across the entire acquisitionphase (Figure 3), F(2, 53) = 0.458, p = .635. To furtherconfirm that the experimental groups were not different at

the beginning of acquisition, we used a separate 2 PracticeCondition × 3 TMS Site ANOVA on the first practice block.At the first practice block, there was no significant differencein the skill performance on the criterion task (60◦) among thesix experimental groups (Figure 3), F(5, 58) = 1.324, p =.268. More importantly, all groups had similar RMSE by theend of practice, after which rTMS was applied to interferewith consolidation processes.

Transfer to 50◦ Target

Offline memory stabilization was determined by measur-ing the performance change from the IT to the delayed trans-fer DT for each group. A 2 Practice Condition (constant,variable) × 3 rTMS Site (none, M1, DLPFC) × 2 TransferTest (IT, DT) ANOVA with repeated measures on the lastfactor for transfer to the 50◦ target yielded a significant maineffect of practice (Figure 4), F(1, 53) = 7.201, p = .01. Simi-lar to previously reported findings, variable practice structurebenefited offline memory stabilization for transfer comparedwith constant practice structure. There were no significantinteraction effects or effect of rTMS site. Further analysisrevealed that the participants in the variable practice groupsperformed with significantly lower error on the DT comparedwith those in the constant practice groups, F(1, 58) = 6.351,p = .015.

Transfer to 80◦ Target

A comparison of the changed performance on 80◦ targetfrom the IT to DT across all groups indicated how rTMSinterference to the neural substrates following constant andvariable practice affected the offline stabilization. Becausethere was a significant effect of practice structure at the ITfor the 80◦ target, F(1, 58) = 20.462, p < .001, we used theperformance at the IT as a covariate for a 2 Practice Condi-tion (constant, variable) × 3 rTMS Site (none, M1, DLPFC)× 2 Transfer Test (IT, DT) ANOVA with repeated measureson the last factor. A significant main effect of practice, F(1,52) = 8.22, p = .006, indicated that offline stabilization wasenhanced by variable practice compared to constant practice.A significant interaction of rTMS site and practice structure,F(2, 52) = 6.371, p = .003, suggested that the postprac-tice rTMS over M1 or DLPFC affected offline stabilizationdifferently for the two practice conditions. Further analysis(ANOVA) on the DT for the 80◦ target yielded a significantinteraction between the rTMS site and practice condition.In those participants who practiced under constant practiceconditions, postpractice rTMS interference over DLPFC didnot affect transfer to the 80◦ target compared with the no-rTMS control CP group (Figure 5, RMSE: CP-DLP = CP),F(1, 18) = 0.008, p = .930. In contrast, when rTMS was ap-plied over M1 following constant practice, delayed transferto 80◦ was significantly attenuated (Figure 5; RMSE: CP-M1> CP), F(1, 19) = 4.495, p = .048. In contrast, postpracticerTMS over M1 following variable practice did not signifi-cantly affect delayed transfer performance on the 80◦ target

502 Journal of Motor Behavior

Dow

nloa

ded

by [

USC

Uni

vers

ity o

f So

uthe

rn C

alif

orni

a], [

Car

olee

Win

stei

n] a

t 13:

55 1

8 Fe

brua

ry 2

012

Memory Consolidation of Motor Transfer

FIGURE 3. Practice performance: participants in the control group (open circles), M1 interference group (black filled circles), andDLPFC interference group (gray filled circles) during the practice phase. The inverted triangles are the temporal control groups (VPDLP-4 hr, and CP M1-4 hr). The delayed interference groups were recruited post hoc to investigate if the interference effects ofrTMS in the VP-DLP and CP-M1 groups were temporally specific to the immediate postpractice period. The left side of the figurerepresents the participants who practiced under variable practice conditions; the right side of the figure represents the participantswho practiced under constant practice conditions.

compared with the no-rTMS control VP group (Figure 5;RMSE: VP-M1 rTMS = VP), F(1, 20) = 0.015, p = .904.In contrast, when rTMS was applied over DLPFC followingvariable practice, delayed transfer performance was signifi-cantly attenuated (Figure 5; RMSE: VP-DLP rTMS > VP),F(1, 20) = 4.487, p = .048. Thus, a double dissociationwas observed 1 day after practice between practice condi-tion structure (variable vs. constant) and the neural locus ofthe rTMS interference for transfer performance on the 80◦

target.

Temporal Specificity of the rTMS Interference Effects

It could be argued that the interference effects of rTMSon delayed transfer of the 80◦ target could be attributed toresidual effects of rTMS on the following day’s transfer test,and not to interference to consolidation. To explore this, wecompared the performance change from IT to DT in theVP DLP-4-hr group with that in the VP-DLP group andwith that of the CP M1-4 hr to CP-M1 group. rTMS appliedover M1 immediately after constant practice significantly

impaired DT performance on the 80◦ target compared withwhen rTMS was applied over M1 after 4 hr (Figure 5), F(1,13) = 6.395, p = .025. DT performance on the 80◦ targetwas also impaired when rTMS was applied immediately aftervariable practice compared with when rTMS applied overDLPFC 4 hr postpractice. However, this difference did notreach statistical significance (Figure 5), F(1, 12) = 1.856, p =.198. DT performance yielded a high effect size (Cohen’s d =0.8), suggesting that the lack of statistical difference may belikely attributed to the smaller sample size.

Discussion

Previously, Kantak, Sullivan et al. (2010) showed thatpractice structure affects the neural substrates that implementmotor memory consolidation for retention of the practicedskill. In the present article, we present additional data regard-ing the effects of rTMS interference to DLPFC or M1 duringmotor memory consolidation following either constant orvariable practice on the transfer of learning to novel move-ment parameters (targets 50◦ and 80◦), both not practiced

2011, Vol. 43, No. 6 503

Dow

nloa

ded

by [

USC

Uni

vers

ity o

f So

uthe

rn C

alif

orni

a], [

Car

olee

Win

stei

n] a

t 13:

55 1

8 Fe

brua

ry 2

012

S. S. Kantak, K. J. Sullivan, B. E. Fisher, B. J. Knowlton, & C. J. Winstein

FIGURE 4. Immediate and delayed transfer performance (RMSE) for the 50◦ target of the participants in the control groups,M1 interference groups, DLPFC interference groups, and delayed interference (VP DLP-4 hr, CP M1-4 hr) groups. The delayedinterference groups were recruited post hoc to investigate if the interference effects of rTMS in VP-DLPFC and CP-M1 groups weretemporally specific to the immediate postpractice period. The left panel represents the data of the variable practice groups and theright panel shows the data of the constant practice groups. Each data point represents a mean of four trials; error bars represent thestandard error of the mean.

by either groups. During acquisition, there was no beneficialeffect of constant practice. This was contrary to the expecta-tion that error would be higher in the variable practice groupscompared with the constant practice groups. However, simi-lar to previous reports, our data demonstrated robust delayedtransfer benefits of variable practice structure compared withconstant practice structure, a well-documented finding thathighlights the learning–performance distinction. For the 50◦

target, rTMS interference to M1 or DLPFC did not signifi-cantly affect the delayed transfer performance following vari-able or constant practice structure. For the 80◦ target, rTMSinterference to DLPFC but not M1 affected the delayed trans-fer performance following variable practice compared withthe no-rTMS VP group. Also, for the 80◦ target, rTMS in-terference to M1 but not DLPFC affected delayed transferperformance following constant practice compared with theno-rTMS CP group. Thus, rTMS interference significantlyaffected the transfer of learning to the 80D target but notthe 50◦ target. We speculate that rTMS interference to M1or DLPFC did not significantly affect the delayed transferperformance of the 50◦ target because the participants weretested for retention of the criterion target (amplitude 60◦) justbefore they were tested on the 50◦ transfer target. It is likelythat the small difference of 10◦ allowed for relearning within

the testing phase and possibly occluded the effects of rTMS.Evidence exists to suggest that knowledge of the abstract re-lationship between the goal and action parameters acquiredfollowing variable practice allows for wider generalizationto novel skill parameters, even to those outside the rangeof practice movements. Our findings support this predictionsuch that when tested on a transfer target that was furtheraway from the mean of the practiced targets (i.e., 80◦), therTMS interference effects were robustly evident. One poten-tial limitation is that the transfer tasks used in the study maynot sufficiently different from the practiced task. It is likelythat with more different transfer task parameters, the differ-ences would have been more evident. Nevertheless, similarto our retention findings (Kantak, Sullivan et al., 2010), weobserved a double dissociation between the rTMS site andpractice structure for transfer of learning to the 80◦ target.

The double dissociation between the neural substrate andpractice structure offers insights into the nature of mem-ory processing induced by each practice structure. Higherreliance on DLPFC for motor memory consolidation follow-ing the variable compared with constant practice conditionssupports the behavioral theory of active cognitive processingduring variable practice (Cross, Schmitt, & Grafton, 2007).Participants in the variable practice group were required to

504 Journal of Motor Behavior

Dow

nloa

ded

by [

USC

Uni

vers

ity o

f So

uthe

rn C

alif

orni

a], [

Car

olee

Win

stei

n] a

t 13:

55 1

8 Fe

brua

ry 2

012

Memory Consolidation of Motor Transfer

FIGURE 5. Immediate and delayed transfer performance (RMSE) for the 80◦ target of the participants in the control groups,M1 interference groups, DLPFC interference groups, and delayed interference (VP DLP-4 hr, CP M1-4 hr) groups. The delayedinterference groups were recruited post hoc to investigate if the interference effects of rTMS in the VP-DLP and CP-M1 groups weretemporally specific to the immediate postpractice period. The left panel represents the data of the variable practice groups and theright panel shows the data of the constant practice groups. Each data point represents a mean of four trials; error bars represent thestandard error of the mean.

select and execute a different movement plan at each trial,as prompted by the target display. Behavioral literature sug-gests that variable practice, due to interspersing of tasks, mayforce the learner to forget an action plan from the workingmemory for an executed task because he or she has to per-form a new task on the next trial. So, when the initial taskversion is presented again, it provides an opportunity for re-trieval or reconstruction of the action plan during practice.These higher order cognitive processes of action selection,planning (Pochon et al., 2001; Rowe, Toni, Josephs, Frack-owiak, & Passingham, 2000), and processing of informationfrom working memory (Fuster, 2000) during variable prac-tice may be particularly suited for the unique processing ca-pabilities of the DLPFC. Our results suggest that, even afterpractice ends, DLPFC is actively engaged in the consolida-tion process, and its role is necessary for transfer of motorlearning.

M1 processing during the practice and postpractice con-solidation phases is critical for retention of a newly prac-ticed motor skill (Galea & Celnik, 2009; Galea, Vazquez,Pasricha, Orban de Xivry, & Celnik, 2010). The present re-sults advance our understanding of the role of M1 in motormemory consolidation by demonstrating that M1 process-ing is also necessary for the transfer of motor skill learning

following constant practice but not variable practice. Thisfinding is in apparent contradiction with previous reports.Multiple studies have shown that M1 is more actively en-gaged during variable practice compared with constant prac-tice (Lin et al, 2008). Further, Wymbs and Grafton (2009)observed an increase in the medial prefrontal cortex activ-ity on functional MRI during blocks of repetitive practice.Thus, the apparent contradiction between our findings andthose of others is likely attributed to the difference in thephases of skill acquisition investigated in these studies. Allof the aforementioned researchers investigated the neuralsubstrates during the within-training (during practice) phase.In contrast, we used direct TMS perturbation to investigatethe neural substrates associated with each practice structureduring the postpractice offline motor memory consolidationphase. Our results suggest that practice structure influencesthe relative role of M1 and DLPFC during motor memoryconsolidation that is critical for transfer of learning to novelmotor parameters.

Interestingly, the effects of rTMS interference to DLPFCand M1 on transfer tests were similar to those on retentionof the skill. Although transfer and retention are distinct phe-nomena, there is evidence that behavior and neural substratesfor transfer and learning (often inferred from retention)

2011, Vol. 43, No. 6 505

Dow

nloa

ded

by [

USC

Uni

vers

ity o

f So

uthe

rn C

alif

orni

a], [

Car

olee

Win

stei

n] a

t 13:

55 1

8 Fe

brua

ry 2

012

S. S. Kantak, K. J. Sullivan, B. E. Fisher, B. J. Knowlton, & C. J. Winstein

partially overlap (Seidler, 2010). In addition to the neuralstructures engaged during learning, transfer to a new sen-sorimotor adaptation condition is shown to engage neuralsubstrates such as cingulate gyrus, superior parietal lob-ule, inferior parietal lobule, and left middle occipital gyrus(Seidler & Noll, 2008). The effects of interference to these re-gions in future research may help unravel the precise neuralsubstrates engaged in the transfer of motor learning. Nev-ertheless, the present findings suggest that when practicestructure is constant, M1 processing is necessary during mo-tor memory consolidation for transfer of the learned skill.In contrast, when the practice structure is variable, process-ing within DLPFC during motor memory consolidation isnecessary for skill transfer.

Motor memory consolidation is a time- and sleep-dependent process. In the present study, we focused on anearly, time-dependent phase of consolidation that occurs im-mediately postpractice and results in a more stable motormemory representation over time. We did not directly probethe consolidation processes during sleep. All participants had6–7 hr of sleep between the end of practice and second daytesting. Therefore, we are confident that the observed resultswere not a result of differences in hours of sleep betweengroups. However, it is impossible to rule out the possibil-ity that rTMS immediately postpractice might have, in someway, affected the characteristics of motor memory, therebyaffecting its sleep-mediated consolidation process. Alterna-tively, rTMS might have modulated brain-activity duringsleep such as slow-frequency oscillations of nonrapid eyemovement sleep known to facilitate consolidation within theprefrontal cortex. The temporal control groups suggest that itis unlikely that rTMS affects the sleep-related consolidationbecause when rTMS was applied ∼4 hr postpractice it did notaffect transfer, compared with the control groups. However,further research is warranted to understand the interactionbetween time- and sleep-dependent consolidation processesand practice structures.

In summary, we showed that practice structures that lead todifferent levels of delayed transfer performance engage dis-tinct neural substrates during the postpractice consolidationphase. Our findings highlight how behavioral interventionsthat invoke different cognitive processes drive distinct brainnetworks during motor memory consolidation. This supportsthe notion of experience dependent neuroplasticity, whichsuggests that the nature of behavioral interventions (practicestructure) dictates the nature of neuroplasticity. Conversely,the results also demonstrate that time- and substrate-specificinterference to neural processes such as consolidation affectssubsequent transfer of motor learning.

REFERENCES

Cross, E. S., Schmitt, P. J., & Grafton, S. T. (2007). Neural substratesof contextual interference during motor learning support a modelof active preparation. Journal of Cognitive Neuroscience, 19,1854–1871.

Fuster, J. M. (2000). Executive frontal functions. ExperimentalBrain Research, 133, 66–70.

Galea, J. M., & Celnik, P. (2009). Brain polarization enhances theformation and retention of motor memories. Journal of Neuro-physiology, 102, 294–301.

Galea, J. M., Vazquez, A., Pasricha, N., Orban de Xivry, J. J., &Celnik, P. (2010). Dissociating the roles of the cerebellum andmotor cortex during adaptive learning: The motor cortex retainswhat the cerebellum learns. Cerebral Cortex, 21, 1761–1770.

Guadagnoli, M. A., & Lee, T. D. (2004). Challenge point: A frame-work for conceptualizing the effects of various practice condi-tions in motor learning. Journal of Motor Behavior, 36, 212–224.

Hall, K. G., & Magill, R. A. (1995). Variability of practice andcontextual interference in motor skill learning. Journal of MotorBehavior, 27, 299–309.

Kantak, S. S., Fisher, B. E., Sullivan, K. J., & Winstein, C. J. (2010).Effects of different doses of low frequency rTMS on motorcortical excitability. Journal of Neurology & Neurophysiology,1(102). doi: 10.4172/2155-9562.1000102

Kantak, S. S., Sullivan, K. J., Fisher, B. E., Knowlton, B. J., &Winstein, C. J. (2010). Neural substrates of motor memory con-solidation depend on practice structure. Nature Neuroscience, 13,923–925.

Lee, T. D., & Simon, D. (2003). Contextual interference. Informa-tion processing perspectives. In A. M. Williams & N. J. Hodges(Eds.), Skill acquisition in sport: Research, theory, and practice(pp. 29–44). London: Routledge.

Luft, A. R., Buitrago, M. M., Ringer, T., Dichgans, J., & Schulz,J. B. (2004). Motor skill learning depends on protein synthe-sis in motor cortex after training. Journal of Neuroscience, 24,6515–6520.

Muellbacher, W., Ziemann, U., Wissel, J., Dang, N., Kofler, M.,Facchini, S.,et al. (2002). Early consolidation in human primarymotor cortex. Nature, 415, 640–644.

Pochon, J. B., Levy, R., Poline, J. B., Crozier, S., Lehericy, S., Pil-lon, B.,et al. (2001). The role of dorsolateral prefrontal cortex inthe preparation of forthcoming actions: An fMRI study. CerebralCortex, 11, 260–266.

Rowe, J. B., Toni, I., Josephs, O., Frackowiak, R. S., & Passing-ham, R. E. (2000). The prefrontal cortex: response selection ormaintenance within working memory? Science, 288, 1656–1660.

Schmidt, R. A., & Bjork, R. A. (1992). New conceptualization ofpractice: Common principles in three paradigms suggest newconcepts for training. Psychological Science, 3, 207–217.

Schmidt, R. A., & Lee, T. D. (2004). Motor control and learning:A behavioral emphasis (4th ed.). Chicago, IL: Human Kinetics.

Seidler, R. D. (2010). Neural correlates of motor learning, transferof learning, and learning to learn. Exercise and Sport SciencesReviews, 38, 3–9.

Seidler, R. D., & Noll, D. C. (2008). Neuroanatomical correlates ofmotor acquisition and motor transfer. Journal of Neurophysiol-ogy, 99, 1836–1845.

Shadmehr, R., & Brashers-Krug, T. (1997). Functional stages inthe formation of human long-term motor memory. Journal ofNeuroscience, 17, 409–419.

Shea, C. H., & Kohl, R. M. (1990). Specificity and variabilityof practice. Research Quarterly for Exercise and Sport, 61,169–177.

Shea, C. H., & Kohl, R. M. (1991). Composition of practice: in-fluence on the retention of motor skills. Research Quarterly forExercise and Sport, 62, 187–195.

Shea, C. H., Lai, Q., Wright, D. L., Immink, M., & Black, C.(2001). Consistent and variable practice conditions: effects onrelative and absolute timing. Journal of Motor Behavior, 33, 139–152.

Sherwood, D. E., & Lee, T. D. (2003). Schema theory: criticalreview and implications for the role of cognition in a new theory

506 Journal of Motor Behavior

Dow

nloa

ded

by [

USC

Uni

vers

ity o

f So

uthe

rn C

alif

orni

a], [

Car

olee

Win

stei

n] a

t 13:

55 1

8 Fe

brua

ry 2

012

Memory Consolidation of Motor Transfer

of motor learning. Research Quarterly for Exercise and Sport,74, 376–382.

Weekly, A. (2004). TEMPLATE [computer software]. GardenGrove, CA: Weekly Consulting.

Wulf, G., & Lee, T. D. (1993). Contextual interference in move-ments of the same class: Differential effects on program andparameter learning. Journal of Motor Behavior, 25, 254–263.

Wymbs, N. F., & Grafton, S. T. (2009). Neural substrates of practicestructure that support future off-line learning. Journal of Neuro-physiology, 102, 2462–2476.

Submitted December 23, 2010Revised August 3, 2011

Accepted October 12, 2011

2011, Vol. 43, No. 6 507

Dow

nloa

ded

by [

USC

Uni

vers

ity o

f So

uthe

rn C

alif

orni

a], [

Car

olee

Win

stei

n] a

t 13:

55 1

8 Fe

brua

ry 2

012