Abstract

Sequence learning relies on formation of unconscious transitional and conscious ordinal memories. The influence of practice type on formation of these memories that compose skill and systems level neural substrates is not known. Here, we studied learning of transitional and ordinal memories in participants trained on motor sequences while scanned using fMRI. Practice structure was varied or grouped (mixing or grouping sequences during training, respectively). Memory was assessed 30 min and 1 week later. Varied practice improved transitional memory and enhanced coupling of the dorsal premotor cortex with thalamus, cerebellum, and lingual and cingulate regions and greater transitional memory correlated with this coupling. Thus, varied practice improves unconscious transitional memories in proportion to coupling within a cortico-subcortical network linked to premotor cortex. This result indicates that practice structure influences unconscious transitional memory formation and identifies underlying systems level mechanisms.

INTRODUCTION

When humans learn a new skill, different memory types are formed. Transitional memories are based on the frequencies of movement-to-movement transitions (i.e., pressing “2” after “3” in “4-3-2-1”), and ordinal memories are based on the ordinal position of each single movement within a sequence (i.e., pressing “2” in the third ordinal position in “4-3-2-1”; Song & Cohen, 2014a, 2014b; Conway & Christiansen, 2001; Terrace & McGoningle, 1994; Cohen, Ivry, & Keele, 1990). As a real-life example, a pianist will be better at performing a new song in G major because of the similar transitions between keypresses among all G major songs (transitional memory) but will also show improvements as he or she learns the specific order of the notes in a particular G major song (ordinal memory).

Transitional memories are not consciously recalled, whereas ordinal memories are consciously recalled (Song & Cohen, 2014b). Ordinal memories can be formed in pigeons, monkeys, and humans (Conway & Christiansen, 2001; Terrace & McGoningle, 1994). On the other hand, transitional memories are absent in pigeons, limited to pairs of adjacent events in monkeys, and uniquely hierarchical in humans forming across several events and crucial to skill acquisition (Conway & Christiansen, 2001; Terrace & McGoningle, 1994).

Different memory types make unique contributions to daily life and are differentially affected by brain lesions (Cohen & Squire, 1980). Furthermore, practice structure is important for optimizing skill learning across the human lifespan in healthy individuals and in patients with brain lesions (Song, Sharma, Buch, & Cohen, 2012; Lin et al., 2011; Kantak, Sullivan, Fisher, Knowlton, & Winstein, 2010; Tanaka, Honda, Hanakawa, & Cohen, 2010; Brady, 2004, 2008; Cross, Schmitt, & Grafton, 2007; Schneider, Healy, & Bourne, 2002; Shea, Lai, Wright, Immink, & Black, 2001; Hanlon, 1996; Hall & Magill, 1995; Porretta & O'Brien, 1991; Shea & Morgan, 1979). Varying the practice structure of training tasks rather than practicing them in grouped sets results in superior overall skill learning, and this phenomenon is termed the “contextual interference effect” (Shea & Morgan, 1979) or the “spacing effect” (Ebbinghaus, 1885).

One influential theory proposes that varied practice benefits long-term learning because of the greater effort spent in constant reupdating within working memory of the parameters for the upcoming task (Cross et al., 2007; Immink & Wright, 1998; Lee & Magill, 1983). In line with this account, prefrontal brain regions involved in strategy and planning have been linked to the varied practice benefit for these explicit tasks (Kantak et al., 2010). However, a recent study showing the benefits of varied practice to long-term learning of more implicit tasks (Song et al., 2012) that are largely working-memory independent (Song & Cohen, 2014c; Song, Marks, Howard, & Howard, 2009; Song, Howard, & Howard, 2007) questions the universality of this account. Unlike conscious (explicit) forms of learning, implicit learning may often rely on multiple distributed neural networks that detect patterns and regularities (Song et al., 2012; Keele, Ivry, Mayr, Hazeltine, & Heuer, 2003). Given the dissociations between transitional and ordinal memories in conscious recall (Song & Cohen, 2014b), understanding how varied practice behaviorally and neurally affects these two types of memories would be useful in elucidating additional or alternative accounts for varied practice benefits. However, the influences of practice structure on formation of transitional and ordinal memories, crucial for skill, and on the underlying neural substrates for these memory components are not known.

METHODS

Participants

Twenty-four healthy volunteers (11 men, 13 women; aged 19–41 years, 28 ± 6 years) participated in the study after signing informed consent to the experimental procedure, which was approved by the National Institute of Neurological Disorders and Stroke institutional review board. All participants were right handed and naive to the task and had a normal neurological examination as assessed by a credentialed physician. One participant was removed for excessive motion (>3 mm) leaving 23 participants in the analyses. Four additional participants were enrolled but either failed to follow directions for the task or failed to complete the total number of study visits.

Behavioral Task and Stimuli

Participants were trained on two different 12-item patterns of keypresses. Keypresses were guided by targets appearing in one of four locations on a screen that corresponded to one of the four fingers of the right hand (Figure 1A). The four locations on a screen occupied approximately 10° of the visual field from left to right (Song & Cohen, 2014a). Participants made keypress responses on a button box (LU440; Cedrus, San Pedro, CA) for the entire study.

Figure 1. 

(A) Participants were trained on two different 12-item patterns of keypresses. Keypresses were guided by targets appearing in one of four locations on a screen that corresponded to one of the four fingers of the right hand. (B) During training inside the scanner, participants completed 56 task blocks (24 targets each). No-pattern blocks (white) included keypresses randomly ordered. During varied practice, keypresses were organized in pattern blocks (green, repeating 12-item sequences) intermixed with no-pattern blocks. During grouped practice, pattern blocks (blue) included repetition of a different 12-item sequence (no intermixed no-pattern blocks). Presentation of block types was counterbalanced across participants. RT data for early (first half) and late (second half) for each block type inside the scanner are shown here. Accuracy data can be found in Supplemental Figure 1b. Note that benefits of varied practice are typically not found during training time itself but only after completion of training (Song et al., 2012; Brady, 2008; Shea & Morgan, 1979). Note also that ISIs that lead to variable response-to-stimulus intervals can impair expression of learning but not learning itself (Willingham et al., 1997). Short and constant response-to-stimulus intervals were used in test blocks after training to provide more accurate behavioral measures of skill (Willingham et al., 1997). (C) Performance was tested 30 min and 1 week after training as shown (Song et al., 2012). (Top) For our main behavioral end points of transitional memory and ordinal memory at 30 min and 1 week, we employed a subtractive method that compared average RTs across triplet types. Please refer to Song and Cohen (2014a) for extensive details on this and other methods of calculating transitional and ordinal learning. This method relies on identifying triplets in no-pattern blocks that matched triplets found in pattern blocks (examples encircled in red) and triplets in no-pattern blocks that did not match triplets found in pattern blocks (example encircled in black). Because of the fact that the two different 12-item patterns had no triplets in common with each other, matching triplets for both the grouped and varied patterns were a different set of triplets. Raw RT and accuracy for the various triplet types can be found in Supplemental Figure 1a. (Bottom) For transitional memory, we compared RTs between triplets in no-pattern blocks that matched or did not match triplets in pattern blocks. For ordinal memory, we compared RTs between matching triplets in pattern and no-pattern blocks. Note the larger RT benefit of varied (green) over grouped (blue) practice on transitional memory (left), absent in ordinal memory. This significant benefit of practice structure on transitional memory was apparent at 30 min up to 1 week posttraining.

Figure 1. 

(A) Participants were trained on two different 12-item patterns of keypresses. Keypresses were guided by targets appearing in one of four locations on a screen that corresponded to one of the four fingers of the right hand. (B) During training inside the scanner, participants completed 56 task blocks (24 targets each). No-pattern blocks (white) included keypresses randomly ordered. During varied practice, keypresses were organized in pattern blocks (green, repeating 12-item sequences) intermixed with no-pattern blocks. During grouped practice, pattern blocks (blue) included repetition of a different 12-item sequence (no intermixed no-pattern blocks). Presentation of block types was counterbalanced across participants. RT data for early (first half) and late (second half) for each block type inside the scanner are shown here. Accuracy data can be found in Supplemental Figure 1b. Note that benefits of varied practice are typically not found during training time itself but only after completion of training (Song et al., 2012; Brady, 2008; Shea & Morgan, 1979). Note also that ISIs that lead to variable response-to-stimulus intervals can impair expression of learning but not learning itself (Willingham et al., 1997). Short and constant response-to-stimulus intervals were used in test blocks after training to provide more accurate behavioral measures of skill (Willingham et al., 1997). (C) Performance was tested 30 min and 1 week after training as shown (Song et al., 2012). (Top) For our main behavioral end points of transitional memory and ordinal memory at 30 min and 1 week, we employed a subtractive method that compared average RTs across triplet types. Please refer to Song and Cohen (2014a) for extensive details on this and other methods of calculating transitional and ordinal learning. This method relies on identifying triplets in no-pattern blocks that matched triplets found in pattern blocks (examples encircled in red) and triplets in no-pattern blocks that did not match triplets found in pattern blocks (example encircled in black). Because of the fact that the two different 12-item patterns had no triplets in common with each other, matching triplets for both the grouped and varied patterns were a different set of triplets. Raw RT and accuracy for the various triplet types can be found in Supplemental Figure 1a. (Bottom) For transitional memory, we compared RTs between triplets in no-pattern blocks that matched or did not match triplets in pattern blocks. For ordinal memory, we compared RTs between matching triplets in pattern and no-pattern blocks. Note the larger RT benefit of varied (green) over grouped (blue) practice on transitional memory (left), absent in ordinal memory. This significant benefit of practice structure on transitional memory was apparent at 30 min up to 1 week posttraining.

During training inside the scanner, participants completed 56 task blocks in total. Each block contained 24 targets separated by an ISI of 800 msec (block length = 19.2 sec). There were three types of blocks (Figure 1B). For no-pattern blocks, the targets followed a pseudorandom order (randomly ordered with repetitions then removed). In varied practice pattern blocks, keypresses followed a repeating 12-item sequence, and these blocks were intermixed with no-pattern blocks. In grouped practice pattern blocks, keypresses followed a different 12-item sequence, and these blocks were presented one after another in a group. Specifically, the following two different 12-item patterns were used: “241432134231” or “313242141324.” These were structured in this way such that there was no overlap between any three keypresses in a row (triplets) between the two sequences. In this way, there was no overlap from second-order conditionals (keypress n predicted by keypress n−2) upward (for details on learning for the different orders of conditional probability in transitioning, see Song & Cohen, 2014a).

The 56 task blocks were spread across four runs each lasting 420 sec (run length = 7 min; Figure 1B). The four runs were as follows: (1) All 14 blocks in a run were no-pattern blocks, (2) seven blocks of varied practice pattern blocks were varied with seven blocks of no-pattern blocks beginning with a pattern block, (3) all 14 blocks in a run were grouped practice pattern blocks, and (4) same as (2) but beginning with a no-pattern block. The four runs were counterbalanced across participants.

After participants were taken out of the scanner, performance was tested 30 min and 1 week after training (Figure 1B and C). Each block during tests contained 96 targets separated by a response-to-stimulus interval of 120 msec. Hence, the next target appeared 120 msec after participants made the correct keypress response. Unlike ISIs that lead to variable response-to-stimulus intervals, short and constant response-to-stimulus intervals will provide more accurate measures of skill (Willingham, Greenberg, & Thomas, 1997). During test, varied practice and grouped practice pattern blocks were flanked by no-pattern blocks (Figure 1B). Five seconds of rest separated each block. After each block, participants were given feedback on accuracy and were asked to respond as fast as possible while maintaining perfect accuracy. At 1 week, participants were also tested for intermanual transfer in the left hand in the egocentric and allocentric coordinate frames (Perez, Tanaka, Wise, Willingham, & Cohen, 2008; see Supplemental Figure S1).

Imaging Data Acquisition

The experiment was performed on a 3.0-T GE Excite scanner using an eight-channel coil (GE Medical Systems, Milwaukee, WI). BOLD data were obtained with a gradient-echo EPI sequence (repetition time [TR] = 2400 msec, echo time [TE] = 30 msec, flip angle = 90°, 40 axial contiguous varied slices per volume, 3.5-mm slice thickness, field of view [FOV] = 24 × 24 cm2, 64 × 64 acquisition matrix). The four runs all began with 18 sec of fixation, followed by fourteen 19.2-sec task blocks that were separated by 9.6 sec of fixation, and ended with 8.4 sec of fixation.

In addition, the following anatomical/connectome scans were obtained: a high-resolution T1-weighted anatomical (magnetization-prepared rapid gradient echo, TE/TR = 2672/6256 msec, 198 slices per volume, 1-mm thickness, FOV = 24 × 24 cm2, 256 × 256 acquisition matrix), diffusion-weighted images (TE/TR = 50/12709.2 msec, 60 slices per volume, 2.5-mm thickness, b value of 1100 s/mm2 in 50 directions, and three volumes at a b value of 0 s/mm2, for 53 brain volumes, array spatial sensitivity encoding technique acceleration factor = 2; two identical diffusion series were collected), a structural T2-weighted anatomical (T2-weighted fast spin echo, TE/TR = 122.304/8333.33 msec, 1.5-mm slice thickness, FOV = 24 × 24 cm2, 512 × 512 acquisition matrix), and pretask and posttask resting state (EPI: TR = 2400 msec, TE = 30 msec, flip angle = 90°, 40 axial contiguous varied slices per volume, 3.5-mm slice thickness, FOV = 24 × 24 cm2, 64 × 64 acquisition matrix, 116 consecutive whole-brain volumes). These scans will be the focus of a separate report.

Data Analysis and Statistics: Behavioral

Performance was tested 30 min and 1 week after training (Song et al., 2012). Skill was parsed into transitional and ordinal memories using a subtractive method previously described (Song & Cohen, 2014a) by chunking every keypress with the two prior keypresses to create overlapping chunks of three items (“triplets”). Transitions in random blocks by chance bear semblance to transitions in pattern blocks, and this method requires the identification of trials after transitions (from n−2) that bear semblance and those after transitions that do not and then simple subtraction of RTs between the two. Hence, for transitional memory, we compared RTs between triplets in no-pattern blocks that matched or did not match triplets in pattern blocks. For ordinal memory, triplets that matched between no-pattern blocks and pattern blocks contained the same transitions (from n−2), but those in pattern blocks contained additional ordinal information. Hence, for ordinal memory, we compared RTs between matching triplets in pattern and no-pattern blocks (Figure 1C). The average RT measures for triplets used to make these calculations are shown in Supplemental Figure S1.

Data were analyzed with repeated-measures ANOVAs (ANOVARMs) with within-subject factors of Type (transition vs. ordinal), Delay (30 min vs. week), and Practice (grouped vs. varied). ANOVAs were followed by post hoc t tests. Before the ANOVAs, sphericity was confirmed with Mauchly's test. If the test for sphericity failed, we applied a Greenhouse–Geisser correction, which is reflected as a correction to the degrees of freedom. In figures, data are shown as group means ± standard error, and results were considered significant at p < .05. All statistical analyses were performed in SPSS (SPSS, Inc., Chicago, IL).

Data Analysis and Statistics: Imaging

Before statistical analyses, image preprocessing was conducted using the AFNI software package (Cox, 1996). The first six EPI volumes were removed from each run, and large transients in the remaining volumes were removed through interpolation. EPI volumes were then slice-time corrected, coregistered to the anatomical scan, resampled to 2-mm isotropic voxels, smoothed with an isometric 6-mm FWHM Gaussian kernel, normalized to the mean signal intensity in each voxel to reflect percent signal, and transformed into the standardized Talairach and Tournoux (1988) volume for the purposes of group analyses. To remove nuisance physiological and nonphysiological artifacts from the EPI data without the independent physiological measures, we applied the basic model (Gotts, Saad, et al., 2013; Saad et al., 2013) that is a reduced version of the full ANATICOR model (Jo et al., 2013). The steps were as follows: The anatomical scan was segmented into tissue compartments using Freesurfer (Fischl, 2012). Ventricle and white matter masks were created, eroding the white matter masks to prevent partial volume effects with gray matter. Masks were then applied to the volume-registered EPI data before smoothing to yield pure nuisance time series for the ventricles as well as local estimates of the signal in white matter that were averaged within a 20-mm-radius sphere. All nuisance time series were detrended with fourth-order polynomials before least-squares model fitting to each voxel's time series. Nuisance variables for each voxel included an average ventricle time series, a local average white-matter time series, and six parameter estimates for head motion. The predicted time course of these nuisance variables was then subtracted from the voxel time series. Band-pass filtering was not applied as physiological variation will not typically be well removed by the popular band-pass filtering step as the problematic frequencies (∼0.3 Hz for respiration cycles and ∼0.9–1 Hz for cardiac cycles) have already been aliased to frequencies below the Nyquist frequency (Gotts, Saad, et al., 2013).

For analysis of BOLD activation across task conditions, task regressors for each condition (early and late training for varied and grouped practice) were modeled as box-car functions convolved with the hemodynamic response. Specifically, we included task regressors for pattern blocks for early and late training for varied and grouped practice. These task regressors modeled the same number of pattern blocks that were of equal length (14 blocks each, each block contained 24 keypresses and two repetitions of the 12-unit pattern). Task regressors for random blocks were also included, but random blocks were not considered further in any of the analyses. Multisubject analysis was based on a random-effect general linear model. Global brain signal was not included among the nuisance variables for reasons detailed in Saad et al. (2012). These resulting maps were subject to a 2 × 2 Training (early vs. late) × Practice (varied vs. grouped) ANOVARM (AFNI 3dANOVA3). The early (Blocks 1–7) versus late (Blocks 8–14) comparison was used here as prior fMRI studies of practice structure find neural changes from early to late in training (Cross et al., 2007). To control for family-wise Type I error, a cluster threshold adjustment method was used, based on Monte Carlo simulations (AFNI AlphaSim), and thus, a minimum cluster size of 154 voxels at p < .005 was used to correct for multiple comparisons (p < .05 corrected).

For analyses of average functional connectivity over the whole brain across task conditions, we first calculated the average residuals time-series correlation of each voxel with every other voxel in the whole brain for each participant in each task condition (Gotts, Jo, et al., 2013; Saad et al., 2013; Gotts, Simmons, et al., 2012). Specifically, the residual time series for pattern task blocks offset by three TRs were concatenated for early and late training for varied and grouped practice, and average functional connectivity in each brain voxel represented the average residual time-series correlation of that voxel with every other voxel in the whole brain for each participant for each condition (AFNI 3dTcorrMap, Figure 2A). Note again that only pattern blocks were included in this analysis and that each condition contained the same number of pattern blocks that were of equal length (14 blocks each, each block contained 24 keypresses and two repetitions of the 12-unit pattern). All Pearson's r values were then transformed using Fisher's z to yield normally distributed values that were submitted to a 2 × 2 Training × Practice ANOVARM (AFNI 3dLME; Chen, Saad, Britton, Pine, & Cox, 2013), which included the average global correlation value for each participant as a nuisance regressor (Gotts, Saad, et al., 2013; Saad et al., 2013).

Figure 2. 

(A) Average functional connectivity in each brain voxel represented the average residual time-series correlation (Pearson's r) of that voxel with every other voxel in the whole brain (Gotts, Jo, et al., 2013; Saad et al., 2013; Gotts, Simmons, et al., 2012). (B) Only a single cluster of voxels that contained the left PMd and extended to the primary motor cortex (M1) showed a significant Training (early vs. late) × Practice (varied by grouped) interaction on average functional connectivity. This cluster showed an increase in average functional connectivity with the rest of the brain with varied practice over training time and, by late training, greater average functional connectivity in varied compared with grouped practice (inset).

Figure 2. 

(A) Average functional connectivity in each brain voxel represented the average residual time-series correlation (Pearson's r) of that voxel with every other voxel in the whole brain (Gotts, Jo, et al., 2013; Saad et al., 2013; Gotts, Simmons, et al., 2012). (B) Only a single cluster of voxels that contained the left PMd and extended to the primary motor cortex (M1) showed a significant Training (early vs. late) × Practice (varied by grouped) interaction on average functional connectivity. This cluster showed an increase in average functional connectivity with the rest of the brain with varied practice over training time and, by late training, greater average functional connectivity in varied compared with grouped practice (inset).

To localize the network of regions driving the change in whole-brain average functional connectivity, spheres of 6-mm radius centered on peak voxels of local maxima for any identified brain regions were used as seeds for seed-based correlation (Gotts, Jo, et al., 2013; Saad et al., 2013; Gotts, Simmons, et al., 2012; Figure 3A). These resulting maps were submitted to a 2 × 2 Training × Practice ANOVARM (AFNI 3dLME). To control for family-wise Type I error, a cluster threshold adjustment method along with Bonferroni correction for the number of seeds (2) was used based on Monte Carlo simulations (AFNI AlphaSim), and thus, a minimum cluster size of 172 voxels at p < .005 was used to correct for multiple comparisons (p < .05/2 corrected).

Figure 3. 

(A) Seven brain regions showed a significant Training × Practice interaction in seed-based correlation with the left PMd: L/R Cingulate, L/R Lingual gyrus, L. Thalamus, L/R Anterior cingulate, R. Cerebellum, L/R Cingulate/Medial Frontal, L. Postcentral gyrus. (B) Correlation matrices for all ROI pairs were constructed using residual time series for the peak voxels (correlation matrices found in Supplemental Figure S2). A 2 × 2 Training × Practice ANOVARM on correlation matrices revealed a significant Training × Practice interaction in coupling between left PMd with the left thalamus, right cerebellum, bilateral lingual region, and the three bilateral cingulate regions as well as between M1 and a medial frontal/cingulate region (F(1, 22) ≥ 12.8, p ≤ .05 with Holm–Bonferroni correction, identified by dark red, left). Post hoc t tests showed decreased coupling with grouped practice and increased coupling with varied practice over training time (|t(22)| > 3.7, p < .05 with Holm–Bonferroni correction, identified by dark red for positive values and dark blue for negative values, right). Note that one-sample t tests comparing correlation matrices to zero demonstrate correlation values that were either null or positive in valence (Supplementary Figure S2), suggesting that findings should be interpreted as coupling turning on or off rather than anticorrelation.

Figure 3. 

(A) Seven brain regions showed a significant Training × Practice interaction in seed-based correlation with the left PMd: L/R Cingulate, L/R Lingual gyrus, L. Thalamus, L/R Anterior cingulate, R. Cerebellum, L/R Cingulate/Medial Frontal, L. Postcentral gyrus. (B) Correlation matrices for all ROI pairs were constructed using residual time series for the peak voxels (correlation matrices found in Supplemental Figure S2). A 2 × 2 Training × Practice ANOVARM on correlation matrices revealed a significant Training × Practice interaction in coupling between left PMd with the left thalamus, right cerebellum, bilateral lingual region, and the three bilateral cingulate regions as well as between M1 and a medial frontal/cingulate region (F(1, 22) ≥ 12.8, p ≤ .05 with Holm–Bonferroni correction, identified by dark red, left). Post hoc t tests showed decreased coupling with grouped practice and increased coupling with varied practice over training time (|t(22)| > 3.7, p < .05 with Holm–Bonferroni correction, identified by dark red for positive values and dark blue for negative values, right). Note that one-sample t tests comparing correlation matrices to zero demonstrate correlation values that were either null or positive in valence (Supplementary Figure S2), suggesting that findings should be interpreted as coupling turning on or off rather than anticorrelation.

To further characterize these interactions, we constructed correlation matrices by taking residual time series for peak voxels (spheres of 6-mm radius centered on peak voxels) and calculated the time-series correlation (coupling) between all ROI pairs (Gotts, Jo, et al., 2013; Gotts, Simmons, et al., 2012; Supplemental Figure S2). These correlation matrices were submitted to a 2 × 2 Training × Practice ANOVARM (Figure 3B, left) as well as post hoc paired t tests (Figure 3B, right) and one-sample t tests (Supplemental Figure S2; MATLAB; Mathworks, Inc., Natick, MA). To control for family-wise Type I error, Holm–Bonferroni correction for multiple comparisons was applied.

RESULTS

We trained 24 participants who learned motor sequences (two different 12-item patterns of keypresses) practiced with either a grouped or varied practice structure in a within-subject design, while they were scanned using fMRI (Figure 1A and B). The two training patterns included uniquely different ordinal and transitional information. Performance was tested 30 min and 1 week after training (Figure 1C, Supplemental Figure 1A).

Measures of transitional and ordinal memory were submitted to a 2 × 2 × 2 Memory type (transitional vs. ordinal) × Practice (grouped vs. varied) × Delay (30 min vs. 1 week) ANOVARM that revealed significant main effects of Type (F(1, 22) = 12.8, p < .002) and Practice (F(1, 22) = 7.6, p < .012). These findings indicate different learning for memory types as well as an overall effect of practice structure. A 2 × 2 Practice × Delay ANOVARM on transitional memory revealed a significant main effect of Practice (F(1, 22) = 21.2, p < .0001). Post hocs showed a significant benefit of varied practice on transitional memory at 30 min (t(22) = 4.3, p < .0001) and at 1 week (t(22) = 3.8, p < .001; Figure 1C, bottom left). A 2 × 2 Practice × Delay ANOVARM on ordinal memory did not reveal any significant main effects or interactions (Figure 1C, bottom right). Hence, varied practice significantly improved transitional but not ordinal memory.

To gain insight into the neural substrates underlying these memory gains, BOLD activation was assessed across conditions. A 2 × 2 Training (early vs. late) × Practice (varied vs. grouped) ANOVARM on BOLD activity revealed no significant effects or interactions. Next, average functional connectivity was assessed over the whole brain across conditions (Gotts, Jo, et al., 2013; Saad et al., 2013; Gotts, Simmons, et al., 2012; Figure 2A). A 2 × 2 Training × Practice ANOVARM identified only one cluster with a significant Training × Practice interaction on average functional connectivity: the left dorsal premotor cortex (PMd) extending to the primary motor cortex (M1; Figure 2B). Post hocs showed increasing average functional connectivity in this cluster over training time with varied practice (t(22) = 2.6, p < .02) but not with grouped practice. Whole-brain average functional connectivity in late training with varied practice was greater in this cluster than with grouped practice (t(22) = 3.7, p < .002; Figure 2B, inset).

To localize the network of regions driving the change in whole-brain average functional connectivity, we used the two local maxima (PMd and M1) for seed-based correlations (Gotts, Jo, et al., 2013; Saad et al., 2013; Gotts, Simmons, et al., 2012; Figure 3A). A 2 × 2 Training × Practice ANOVARM revealed seven regions with a significant Training × Practice interaction with PMd (Figure 3A, Table 1) but none with M1.

Table 1. 

Regions Showing Differential Changes in Coupling with Practice Structure

Talairach Coordinates of ROI “Peak”Volume (Vox/mm3)
Average functional connectivity 
L. dorsal premotor (PMd, BA 6) 33 46 215/1720 
L. primary motor (M1, BA 4; second local maxima) 29 31 46 
 
Seed-based correlation with PMd seed 
L./R. cingulate cortex (BA 31) −7 27 38 985/7880 
L./R. lingual gyrus (BA 18) −7 57 520/4160 
L. thalamus 15 19 −4 264/2112 
L./R. anterior cingulate (BA 32) −7 −25 16 255/2040 
R. cerebellum (Lobule VII) −37 71 −36 251/2008 
L./R. cingulate/medial frontal −1 −21 34 232/1856 
L. postcentral gyrus (BA 13) 47 18 173/1384 
Talairach Coordinates of ROI “Peak”Volume (Vox/mm3)
Average functional connectivity 
L. dorsal premotor (PMd, BA 6) 33 46 215/1720 
L. primary motor (M1, BA 4; second local maxima) 29 31 46 
 
Seed-based correlation with PMd seed 
L./R. cingulate cortex (BA 31) −7 27 38 985/7880 
L./R. lingual gyrus (BA 18) −7 57 520/4160 
L. thalamus 15 19 −4 264/2112 
L./R. anterior cingulate (BA 32) −7 −25 16 255/2040 
R. cerebellum (Lobule VII) −37 71 −36 251/2008 
L./R. cingulate/medial frontal −1 −21 34 232/1856 
L. postcentral gyrus (BA 13) 47 18 173/1384 

L. = left; R. = right.

To further characterize these interactions, we constructed correlation matrices by taking residual time series for peak voxels and calculated the time-series correlation (coupling) between all ROI pairs (Figure 3B). A 2 × 2 Training × Practice ANOVARM revealed a significant Training × Practice interaction (F(1, 22) ≥ 12.8 for p ≤ .05, Holm–Bonferroni corrected; Figure 3B, left) on functional connectivity in seven ROI pairs: the left thalamus, right cerebellum, lingual region, and three cingulate regions differentially changed coupling with the left PMd over training time. Post hoc t tests identified decreasing coupling with the left PMd with grouped practice and increasing coupling with varied practice over training time (Figure 3B, right). Taken together, these findings indicate an increase in coupling within a specific cortico-subcortical network linked to the premotor cortex with varied but not grouped practice.

As practice structure benefits to transitional memory were apparent 30 min after training and retained to 1 week (Figure 1C), we correlated 30-min measures of transitional memory to average coupling in the network. There was a significant correlation between average strength of coupling (Figure 4) and transitional memory with varied practice (r(21) = .43, p < .04) but not with grouped practice (p > .2). Hence, greater transitional memory correlated with coupling in this premotor network only with varied practice. Visual presentation of the correlation between transitional and ordinal memory with varied and grouped practice with average and pairwise coupling in all regions found in the PMd network can be found in Supplemental Figure 3.

Figure 4. 

The average strength of coupling (Average R(z′) in late training for ROI pairs that showed a significant Training × Practice interaction at p < .05 with Holm–Bonferonni correction) was related to posttraining transitional memory. A significant correlation was found only with varied practice (r(21) = .43, p < .04). Correlation between behavioral measures of transitional memory and coupling in specific ROI pairs in both grouped and varied practice can be found in Supplemental Figure S3.

Figure 4. 

The average strength of coupling (Average R(z′) in late training for ROI pairs that showed a significant Training × Practice interaction at p < .05 with Holm–Bonferonni correction) was related to posttraining transitional memory. A significant correlation was found only with varied practice (r(21) = .43, p < .04). Correlation between behavioral measures of transitional memory and coupling in specific ROI pairs in both grouped and varied practice can be found in Supplemental Figure S3.

DISCUSSION

Here, we report for the first time that varied practice resulted in superior transitional memory and increased coupling of a specific left dorsal premotor cortico-subcortical network and that greater transitional memory acquired with varied practice correlated with greater coupling.

Practice structure is important for skill learning across the human lifespan in healthy individuals and in patients with brain lesions (Song et al., 2012; Kantak et al., 2010; Tanaka et al., 2010; Lin et al., 2009; Brady, 2004, 2008; Cross et al., 2007; Schneider et al., 2002; Shea et al., 2001; Hanlon, 1996; Hall & Magill, 1995; Porretta & O'Brien, 1991; Shea & Morgan, 1979). Although such benefits are often thought to be because of greater conscious rehearsal in working memory with varied practice (Cross et al., 2007), in a prior study, we found that conscious processes were not necessarily involved in varied practice benefits (Song et al., 2012). Here, we confirm that varied practice can benefit transitional memory, a component of motor skill that is not consciously recalled (Song & Cohen, 2014b; Figure 1C). Transitional memories can also be unconsciously formed (Song & Cohen, 2014a; Remillard, 2010; Jiménez, 2008; Remillard & Clark, 2001) and are uniquely hierarchical in humans (Conway & Christiansen, 2001; Terrace & McGoningle, 1994). In the example as stated above, varied practice may help a pianist form memories that help in performance for all songs in a G major key rather than for the specific order of notes in a song. Transitional memories contribute not only to motor skill learning but also to human language learning as well (Schneider et al., 2002; Conway & Christiansen, 2001; Terrace & McGoningle, 1994).

Varied practice, characterized by an increase in the number of shifts in action plans and better learning relative to grouped practice (Shea & Morgan, 1979), enhanced coupling of the left PMd with the thalamus, right cerebellum, and lingual and cingulate regions. This finding is consistent with our understanding of the role of the cingulate and premotor cortices. Cingulate regions are hubs connected to almost every other brain area (van den Heuvel, Kahn, Goni, & Sporns, 2012; van den Heuvel & Sporns, 2011), integrate the history of actions and their outcomes (Kennerley, Walton, Behrens, Buckley, & Rushworth, 2006), and are required for encoding shifts in expected actions and outcomes across time (Behrens, Woolrich, Walton, & Rushworth, 2007; Kennerley et al., 2006). The PMd, on the other hand, can form representations for multiple action plans such as different motor sequences and select between them (Chouinard & Paus, 2006; Cisek & Kalaska, 2005; Brasted & Wise, 2004; Picard & Strick, 2001; Murray, Bussey, & Wise, 2000).

Coupling in this network including PMd and cingulate regions as well as thalamus and cerebellum correlated with greater transitional memory. Movement transitions do have specific representations in both primary motor and dorsal premotor cortices (Wiestler & Diedrichsen, 2013). The PMd, in particular, contains Layer 4 granular cells that allow for thalamic input and form powerful information processing loops between the neocortex and phylogenetically older areas such as the basal ganglia and the cerebellum that project via the thalamus back to premotor regions (Murray, Wise, & Rhodes, 2011; Houk & Wise, 1995; Dum & Strick, 1991). These loops confer unique reinforcement learning abilities as required to encode movement transitions (Murray et al., 2011; Houk & Wise, 1995). Recent functional imaging studies have found concordance between increased network functional connectivity and enhanced efficiency via increased synchrony between regions in networks (Gotts, Chow, & Martin, 2012). This theoretical background is in agreement with the current study finding of a correlation between transitional memory and increased synchrony in premotor loops. Furthermore, prior imaging studies in sequence learning under various learning conditions (Cross et al., 2007; Grafton, Hazeltine, & Ivry, 1998) identify many of these same regions identified in the current study, with comprehensive review suggesting the involvement of multiple networks in sequence learning (Keele et al., 2003). This and other studies such as this one that parse sequence learning into differing transitional and ordinal memories may provide detail into how these multiple networks uniquely contribute to sequencing skill.

In summary, we report that varied practice improves unconscious transitional memories that compose skill in proportion to coupling within a cortico-subcortical network linked to the premotor cortex. Because skills are amalgamations of both unconscious and conscious memories, our results document an unconscious behavioral systems-level mechanism that may underlie benefits of practice structure on skill. The neural synchrony account proposed here offers an additional or alternative account for varied practice benefits, whereas prior accounts focused on conscious effort spent in constant reupdating within working memory of the parameters for the upcoming task (Cross et al., 2007; Immink & Wright, 1998; Lee & Magill, 1983). It remains to be determined if practice structure protocols could improve these memory types in the setting of rehabilitation after brain lesions (O'Shea, Johansen-Berg, Trief, Gobel, & Rushworth, 2007; Johansen-Berg et al., 2002) as was the case here in healthy participants.

Acknowledgments

The authors thank Richard Reynolds and Gang Chen of the NIH AFNI Core Staff and Marco Sandrini for helpful comments on the manuscript. This work was supported by the Intramural Research Program of the National Institute of Neurological Disorders and Stroke at the National Institutes of Health and utilized the high-performance computational capabilities of the Biowulf Linux cluster at the National Institutes of Health (http://biowulf.nih.gov).

Reprint requests should be sent to Sunbin Song or Leonardo G. Cohen, Human Cortical Physiology and Neurorehabilitation Section, National Institute of Neurological Disorders and Stroke (NINDS), National Institutes of Health, Bethesda, MD 20892, or via e-mail: songss@ninds.nih.gov, cohenl@ninds.nih.gov.

REFERENCES

REFERENCES
Behrens
,
T. E.
,
Woolrich
,
M. W.
,
Walton
,
M. E.
, &
Rushworth
,
M. F.
(
2007
).
Learning the value of information in an uncertain world
.
Nature Neuroscience
,
10
,
1214
1221
.
Brady
,
F.
(
2004
).
Contextual interference: A meta-analytic study
.
Perceptual Motor Skills
,
99
,
116
126
.
Brady
,
F.
(
2008
).
The contextual interference effect and sport skills
.
Perceptual Motor Skills
,
106
,
461
472
.
Brasted
,
P. J.
, &
Wise
,
S. P.
(
2004
).
Comparison of learning-related neuronal activity in the dorsal premotor cortex and striatum
.
European Journal of Neuroscience
,
19
,
721
740
.
Chen
,
G.
,
Saad
,
Z. S.
,
Britton
,
J. C.
,
Pine
,
D. S.
, &
Cox
,
R. W.
(
2013
).
Linear mixed-effects modeling approach to fMRI group analysis
.
Neuroimage
,
73
,
176
190
.
Chouinard
,
P. A.
, &
Paus
,
T.
(
2006
).
The primary motor and premotor areas of the human cerebral cortex
.
The Neuroscientist
,
12
,
143
152
.
Cisek
,
P.
, &
Kalaska
,
J. F.
(
2005
).
Neural correlates of reaching decisions in dorsal premotor cortex: Specification of multiple direction choices and final selection of action
.
Neuron
,
45
,
801
814
.
Cohen
,
A.
,
Ivry
,
R. I.
, &
Keele
,
S. W.
(
1990
).
Attention and structure in sequence learning
.
Journal of Experimental Psychology: Learning, Memory, and Cognition
,
16
,
17
.
Cohen
,
N. J.
, &
Squire
,
L. R.
(
1980
).
Preserved learning and retention of pattern—Analyzing skill in amnesia: Dissociation of knowing how and knowing that
.
Science
,
210
,
207
210
.
Conway
,
C. M.
, &
Christiansen
,
M. H.
(
2001
).
Sequential learning in non-human primates
.
Trends in Cognitive Science
,
5
,
539
546
.
Cox
,
R. W.
(
1996
).
AFNI: Software for analysis and visualization of functional magnetic resonance neuroimages
.
Computers and Biomedical Research, an International Journal
,
29
,
162
173
.
Cross
,
E. S.
,
Schmitt
,
P. J.
, &
Grafton
,
S. T.
(
2007
).
Neural substrates of contextual interference during motor learning support a model of active preparation
.
Journal of Cognitive Neuroscience
,
19
,
1854
1871
.
Dum
,
R. P.
, &
Strick
,
P. L.
(
1991
).
The origin of corticospinal projections from the premotor areas in the frontal lobe
.
Journal of Neuroscience
,
11
,
667
689
.
Ebbinghaus
,
H.
(
1885
).
Memory: A contribution to experimental psychology
.
New York
:
Dover
.
Fischl
,
B.
(
2012
).
FreeSurfer
.
NeuroImage
,
62
,
774
781
.
Gotts
,
S. J.
,
Chow
,
C. C.
, &
Martin
,
A.
(
2012
).
Repetition priming and repetition suppression: A case for enhanced efficiency through neural synchronization
.
Cognitive Neuroscience
,
3
,
227
237
.
Gotts
,
S. J.
,
Jo
,
H. J.
,
Wallace
,
G. L.
,
Saad
,
Z. S.
,
Cox
,
R. W.
, &
Martin
,
A.
(
2013
).
Two distinct forms of functional lateralization in the human brain
.
Proceedings of the National Academy of Sciences, U.S.A.
,
110
,
3235
3244
.
Gotts
,
S. J.
,
Saad
,
Z. S.
,
Jo
,
H. J.
,
Wallace
,
G. L.
,
Cox
,
R. W.
, &
Martin
,
A.
(
2013
).
The perils of global signal regression for group comparisons: A case study of autism spectrum disorders
.
Frontiers in Human Neuroscience
,
7
,
356
.
Gotts
,
S. J.
,
Simmons
,
W. K.
,
Milbury
,
L. A.
,
Wallace
,
G. L.
,
Cox
,
R. W.
, &
Martin
,
A.
(
2012
).
Fractionation of social brain circuits in autism spectrum disorders
.
Brain
,
135
,
2711
2725
.
Grafton
,
S. T.
,
Hazeltine
,
E.
, &
Ivry
,
R. B.
(
1998
).
Abstract and effector-specific representations of motor sequences identified with PET
.
The Journal of Neuroscience
,
18
,
9420
9428
.
Hall
,
K. G.
, &
Magill
,
R. A.
(
1995
).
Variability of practice and contextual interference in motor skill learning
.
Journal of Motor Behavior
,
27
,
299
309
.
Hanlon
,
R. E.
(
1996
).
Motor learning following unilateral stroke
.
Archives of Physical Medicine and Rehabilitation
,
77
,
811
815
.
Houk
,
J. C.
, &
Wise
,
S. P.
(
1995
).
Distributed modular architectures linking basal ganglia, cerebellum, and cerebral cortex: Their role in planning and controlling action
.
Cerebral Cortex
,
5
,
95
110
.
Immink
,
M.
, &
Wright
,
D. L.
(
1998
).
Contextual interference: A response planning account
.
Quarterly Journal of Experimental Psychology
,
51A
,
735
754
.
Jiménez
,
L.
(
2008
).
Taking patterns for chunks: Is there any evidence of chunk learning in continuous serial reaction-time tasks?
Psychological Research
,
72
,
387
396
.
Jo
,
J. H.
,
Gotts
,
S. J.
,
Reynolds
,
R. C.
,
Bandettini
,
P. A.
,
Martin
,
A.
,
Cox
,
R. W.
, et al
(
2013
).
Effective preprocessing procedures virtually eliminate distance-dependent motion artifacts in resting state fMRI
.
Journal of Applied Mathematics
.
Article 935154
.
Johansen-Berg
,
H.
,
Rushworth
,
M. F.
,
Bogdanovic
,
M. D.
,
Kischka
,
U.
,
Wimalaratna
,
S.
, &
Matthews
,
P. M.
(
2002
).
The role of ipsilateral premotor cortex in hand movement after stroke
.
Proceedings of the National Academy of Sciences, U.S.A.
,
99
,
14518
14523
.
Kantak
,
S. S.
,
Sullivan
,
K. J.
,
Fisher
,
B. E.
,
Knowlton
,
B. J.
, &
Winstein
,
C. J.
(
2010
).
Neural substrates of motor memory consolidation depend on practice structure
.
Nature Neuroscience
,
13
,
923
925
.
Keele
,
S. W.
,
Ivry
,
R.
,
Mayr
,
U.
,
Hazeltine
,
E.
, &
Heuer
,
H.
(
2003
).
The cognitive and neural architecture of sequence representation
.
Psychological Review
,
110
,
316
339
.
Kennerley
,
S. W.
,
Walton
,
M. E.
,
Behrens
,
T. E.
,
Buckley
,
M. J.
, &
Rushworth
,
M. F.
(
2006
).
Optimal decision making and the anterior cingulate cortex
.
Nature Neuroscience
,
9
,
940
947
.
Lee
,
T. D.
, &
Magill
,
R. A.
(
1983
).
The locus of contextual interference in motor-skill acquisition
.
Journal of Experimental Psychology: Learning, Memory, and Cognition
,
9
,
730
746
.
Lin
,
C. H.
,
Fisher
,
B. E.
,
Wu
,
A. D.
,
Ko
,
Y. A.
,
Lee
,
L. Y.
, &
Winstein
,
C. J.
(
2009
).
Neural correlate of the contextual interference effect in motor learning: A kinematic analysis
.
Journal of Motor Behavior
,
41
,
232
242
.
Lin
,
C. H.
,
Knowlton
,
B. J.
,
Chiang
,
M. C.
,
Iacoboni
,
M.
,
Udompholkul
,
P.
, &
Wu
,
A. D.
(
2011
).
Brain-behavior correlates of optimizing learning through interleaved practice
.
Neuroimage
,
56
,
1758
1772
.
Murray
,
E. A.
,
Bussey
,
T. J.
, &
Wise
,
S. P.
(
2000
).
Role of prefrontal cortex in a network for arbitrary visuomotor mapping
.
Experimental Brain Research
,
133
,
114
129
.
Murray
,
E. A.
,
Wise
,
S. P.
, &
Rhodes
,
S. E. V.
(
2011
).
What can different brains do with reward?
In
J. A.
Gottfried
(Ed.),
Neurobiology of sensation and reward frontiers in neuroscience
(pp.
61
98
).
Boca Raton, FL
:
Taylor & Francis
.
O'Shea
,
J.
,
Johansen-Berg
,
H.
,
Trief
,
D.
,
Gobel
,
S.
, &
Rushworth
,
M. F.
(
2007
).
Functionally specific reorganization in human premotor cortex
.
Neuron
,
54
,
479
490
.
Perez
,
M. A.
,
Tanaka
,
S.
,
Wise
,
S. P.
,
Willingham
,
D. T.
, &
Cohen
,
L. G.
(
2008
).
Time-specific contribution of the supplementary motor area to intermanual transfer of procedural knowledge
.
Journal of Neuroscience
,
28
,
9664
9669
.
Picard
,
N.
, &
Strick
,
P. L.
(
2001
).
Imaging the premotor areas
.
Current Opinion in Neurobiology
,
11
,
663
672
.
Porretta
,
D. L.
, &
O'Brien
,
K.
(
1991
).
The use of contextual interference trials by mildly mentally handicapped children
.
Research Quarterly for Exercise and Sport
,
62
,
240
244
.
Remillard
,
G.
(
2010
).
Implicit learning of fifth- and sixth-order sequential probabilities
.
Memory and Cognition
,
38
,
905
915
.
Remillard
,
G.
, &
Clark
,
J. M.
(
2001
).
Implicit learning of first-, second-, and third-order transition probabilities
.
Journal of Experimental Psychology: Learning, Memory, and Cognition
,
27
,
483
498
.
Saad
,
Z. S.
,
Gotts
,
S. J.
,
Murphy
,
K.
,
Chen
,
G.
,
Jo
,
H. J.
,
Martin
,
A.
, et al
(
2012
).
Trouble at rest: How correlation patterns and group differences become distorted after global signal regression
.
Brain Connectivity
,
2
,
25
32
.
Saad
,
Z. S.
,
Reynolds
,
R. C.
,
Jo
,
H. J.
,
Gotts
,
S. J.
,
Chen
,
G.
,
Martin
,
A.
, et al
(
2013
).
Correcting brain-wide correlation differences in resting-state FMRI
.
Brain Connectivity
,
3
,
339
352
.
Schneider
,
V. I.
,
Healy
,
A. F.
, &
Bourne
,
L. E.
, Jr.
(
2002
).
What is learned under difficult conditions is hard to forget: Contextual interference effects in foreign vocabulary acquisition, retention, and transfer
.
Journal of Memory and Language
,
46
,
419
440
.
Shea
,
C. H.
,
Lai
,
Q.
,
Wright
,
D. L.
,
Immink
,
M.
, &
Black
,
C.
(
2001
).
Consistent and variable practice conditions: Effects on relative and absolute timing
.
Journal of Motor Behavior
,
33
,
139
152
.
Shea
,
J. B.
, &
Morgan
,
R. L.
(
1979
).
Contextual interference effects of the acquisition, retention, and transfer of a motor skill
.
Journal of Experimental Psychology: Human Learning and Memory
,
5
,
179
187
.
Song
,
S.
, &
Cohen
,
L. G.
(
2014a
).
Practice and sleep form different aspects of skill
.
Nature Communications
,
5
,
3407
.
Song
,
S.
, &
Cohen
,
L. G.
(
2014b
).
Conscious recall of different aspects of skill memory
.
Frontiers in Behavioral Neuroscience
,
8
,
233
.
Song
,
S.
, &
Cohen
,
L. G.
(
2014c
).
Impact of conscious intent on chunking during motor learning
.
Learning and Memory
,
21
,
449
451
.
Song
,
S.
,
Howard
,
J. H.
, Jr.
, &
Howard
,
D. V.
(
2007
).
Implicit probabilistic sequence learning is independent of explicit awareness
.
Learning and Memory
,
14
,
167
176
.
Song
,
S.
,
Marks
,
B.
,
Howard
,
J. H.
, Jr.
, &
Howard
,
D. V.
(
2009
).
Evidence for parallel explicit and implicit sequence learning systems in older adults
.
Behavioural Brain Research
,
196
,
328
332
.
Song
,
S.
,
Sharma
,
N.
,
Buch
,
E.
, &
Cohen
,
L. G.
(
2012
).
White matter microstructural correlates of superior long-term skill gained implicitly under randomized practice
.
Cerebral Cortex
,
22
,
1671
1677
.
Talairach
,
J.
, &
Tournoux
,
P.
(
1988
).
Co-planar stereotaxic atlas of the human brain
.
New York
:
Thieme
.
Tanaka
,
S.
,
Honda
,
M.
,
Hanakawa
,
T.
, &
Cohen
,
L. G.
(
2010
).
Differential contribution of the supplementary motor area to stabilization of a procedural motor skill acquired through different practice schedules
.
Cerebral Cortex
,
20
,
2114
2121
.
Terrace
,
H. S.
, &
McGoningle
,
B.
(
1994
).
Memory and representation of serial order by children, monkeys, and pigeons
.
Current Directions in Psychological Science
,
3
,
180
185
.
van den Heuvel
,
M. P.
,
Kahn
,
R. S.
,
Goni
,
J.
, &
Sporns
,
O.
(
2012
).
High-cost, high-capacity backbone for global brain communication
.
Proceedings of the National Academy of Sciences, U.S.A.
,
109
,
11372
11377
.
van den Heuvel
,
M. P.
, &
Sporns
,
O.
(
2011
).
Rich-club organization of the human connectome
.
Journal of Neuroscience
,
31
,
15775
15786
.
Wiestler
,
T.
, &
Diedrichsen
,
J.
(
2013
).
Skill learning strengthens cortical representations of motor sequences
.
eLife
,
2
,
e00801
.
Willingham
,
D. B.
,
Greenberg
,
A. R.
, &
Thomas
,
R. C.
(
1997
).
Response-to-stimulus interval does not affect implicit motor sequence learning, but does affect performance
.
Memory and Cognition
,
25
,
534
542
.