Normal Aging Affects the Short-Term Temporal Stability of Implicit, But Not Explicit, Motor Learning following Visuomotor Adaptation

Synthesis

This manuscript provides new insights into age-related changes on short-term retention of motor memories. This might have implications for neurological disorders. The authors present results about hand motor learning task to visuomotor rotation of hand trajectory in younger and older individuals. They focus on temporal stability of explicit and implicit learning during repeated tasks (test and learning task alternated with different delays from 0 to 90 s). In general, the experiment is well conducted, and report is well written, but would benefit from condensation and clarification.

Major concerns

The authors study explicit and implicit learning, but they do not explain these concepts or theories in the Introduction at all. Please explain these concepts and explain more clearly how and why your tasks represented/teste these learning mechanisms?

The implicit motor learning task appears as kinesthetic/proprioceptive task/learning. In this task, there is no visual cue during the movement, and thus it reflects more a somatomotor integration learning/adaptation than visuomotor integration. This should be emphasized for the current task(s) and results in Introduction and Discussion. Interestingly, both the proprioceptive perception and cortical processing of the proprioception are impaired with aging, that could partly explain your result of slower learning/processing (e.g. selection time) of the currently used tasks. See:

- Goble et al. 2009. Proprioceptive sensibility in the elderly: Degeneration, functional consequences and plastic-adaptive processes. Neurosci. Biobehav. Rev. 33, 271-278., DOI: 10.1016/j.neubiorev.2008.08.012

- Piitulainen et al. 2018. Cortical Proprioceptive Processing Is Altered by Aging. Front Aging Neurosci. 10:147. DOI: 10.3389/fnagi.2018.00147.

Selection time is very long, there is almost a short-term retention period between each trial when the selection time is 10sec or more. It is very different for some of the older adults who need up to (on average) 20s (15s during the testing phase) to select movement direction. I believe that this is an underestimation as the authors excluded trials with selection time larger than 30s. Yet, these excluded trials will influence the adaptation process because they make the inter-trial time longer for these specific subjects. This will have an influence on total adaptation rate. The fact that the selection time is much larger in older adults is a major problem as these individuals have then an overall longer inter-trial duration, which has an influence on the amount of adaptation. It is also directly influence the estimation of the delay period as this can be up to 200% larger than instructed (30s instead of 10s for selection time of 10s and delay of 10s). The authors should perform their analyses with the actual elapsed time during the delay period rather than the instructed/imposed delay time (in Fig.6, replace delay time with actual elapsed time). Note that the elapsed time could be different for the explicit retention (from end of previous trial to time of selection) and for the implicit retention (from end of previous trial to time of movement start?). One way to proceed who be to use the actual average elapsed time for each age group but it would be even better if there was a way to take into account inter-individual differences in elapsed time as this varies a lot between the participants, especially for the older adults (if not possible on the graph, it should at least be added as co-variate in the statistical analysis).

Statistics of overall adaptation do not fit with what one sees on the figure. Statistics as performed currently are very dependent on the selection of time windows (middle and late). It would be better to perform an unbiased analysis by taking the mean over all the learning period (trials 1-60).

Tiny differences in baseline should be taken into account in the ANOVA as a covariate. The data from the blank trials during baseline are never reported. How accurate were people during baseline in aiming to the selected point? These values should be described and should be taken into account as a co-variate in the analysis (not simply substracting).

There is no statistical results that support the main conclusion of the paper (decay in implicit is different for young and old). Yet, when both implicit and explicit are used in the LMM (line 472), there is a main effect of age group, learning type and delay period and only an interaction between delay time and learning group. In other words, on this basis, one could not conclude that the difference in retention is specific to the implicit component as there is no interaction between learning type, delay period and age group. This statistical analysis does not match with the conclusions of the authors. The rest of this paragraph is describing that there was no difference in explicit retention between age groups but there was one in implicit retention. Yet, only an interaction can tell that this is a true difference. This is the problem reported by Nieuwenhuis et al. 2011. I believe that the absence of significant triple interaction is due to the huge variability in retention values for the group of older adults.

The task has been well validated in younger adults. On most graphs, the authors provide individual datapoints, but there should be more of them. It is also important that they measure the explicit and implicit components of motor adaptation.

Results: Figure 3. Were there any significant differences here? The inter-individual variation is huge, and thus the sample size of the current data is not sufficient. With bigger sample size many of the ns. Results would be significant. Based on the text, the only difference was for older group in middle training period between implicit and explicit training. Please indicate the significant results clearly to the Figures.

The learning task is exhaustively repeated throughout the manuscript. It should be explained as clearly as possible in the Methods section, and then only briefly reminded in the later sections. It would be good to include more details also in the Figure 1 about the task (number of repetitions etc.). This work would also benefit from illustrative Video in the Supplements about the task if possible. Now it took some time to understand the protocol from the text and Figure 1.

The Results are exhausting and could be greatly condensed to improve clarity. I would omit all discussion from the Results (e.g. L321¬-329) and move it to the Discussion only.

There is very little data on the different measures linked to retention (Fig.6). Selection time are provided on Fig.7 but the retention is primarily based on the difference between the last three trials (with visual feedback) before the break and the first trials after the break. Yet, there is currently absolutely no information on these three trial averages and this is required as they have a large influence on the retention parameters. The authors should provide 7 scatterplot graphs (one graph per delay value, on each graph: one point per participant, color-coded for age group) with average adaptation in the three movements (average across the 4 trials for that delay) on the x-axis vs. the average amount of adaptation on the retention trial without visual feedback (average across the 4 trials for that delay). The goal is to present the data from the adaptation trials during the testing phase and from the retention trials for each individual participant and each delay. This is currently missing in the paper.

The adaptation behavior during the testing phase should be described and illustrated for both age groups. To account for potential differences in adaptation before retention, I would like the authors to perform their analyses on relative retention parameters (adapt-retained/adapt) rather than on the absolute difference as they did here.

Influence of baseline: the angle decay over time (ADoT) is computing by subtracting the baseline-corrected adaptation measured during retention trials (R-B) from the average of non-baseline-corrected adaptation trials before the delay (A). By doing so, the researchers are essentially adding the baseline correction to the angle decay observed during the delay period (ADoT = (A-R) + B).

The values presented on Lines 419-420 seem inconsistent with the values presented on Fig.6. For the young group, there seems to be a small increase in adaptation during the 0s break (from 43.47{degree sign} to 44.03{degree sign}). Yet, the average of total adaptation for the 0s delay on the left panel of Fig.6 is below zero while it should be above it. How is the value on line 433 (-0.27{degree sign}) is obtained?

Minor comments

Conclusions are missing from the end of the Discussion, but are more or less mentioned in the 1st paragraph of Discussion. I would move the primary Conclusions to a specific Conclusions section in the end of the Discussion.

Please add titles to the panels of Figures in addition to panel labels. It would be much easier to understand what is presented in each panel.

It would be more correct to call ‘elderly’ group as ‘older’ group

How does Fig.5 contribute to our understanding of possible differences in short-term retention between young and old adults. Furthermore, I don’t agree that this is consistent with Haith et al. (lines 389-391) given that their low and high PT was in the order of a few 100ms and not several seconds. In addition, it does not make sense that the slope was steeper for older adults early in the adaptation period.

Vandevoorde and Orban de Xivry 2019 reported no difference in short-term retention between young and old participants over a one-minute break for total adaptation and for implicit adaptation only (isolated by design)

Why does the effect of group reported on line 429 does not match the effect of group on line 476? The model is different but should not the statistical results be the same because total adaptation = exp.

Author Response

Dear Dr. Palva:

We have revised our manuscript titled “Normal aging reduces the short-term temporal stability of implicit, but not explicit, motor learning following visuomotor adaptation “ (eN-TNWR-0527-20X). We thank the Reviewers for their very careful and insightful comments. Our manuscript is much improved based on their suggestions. We have addressed each of the raised concerns marked in red below. Specifically, in the revised manuscript we now use the actual elapsed time in our quantification of learning decay and statistical analysis, include the suggested analysis directly comparing the learning on probe trials to the previous re-training trials, and have significantly reduced the redundancy in the Methods and Results. Changes are marked in red throughout the redline version of the revised manuscript. Please note that the lines where the revised text can be found corresponds to the redline document. We believe that we have fully addressed the Reviewers’ concerns, and as a result, we now have a much improved manuscript that we respectfully resubmit for your consideration.

Sincerely,

Guneet Bindra, Rylee Brower, Ryan North, Weiwei Zhou and Wilsaan M. Joiner

Synthesis Statement for Author (Required):

This manuscript provides new insights into age-related changes on short-term retention of motor memories. This might have implications for neurological disorders. The authors present results about hand motor learning task to visuomotor rotation of hand trajectory in younger and older individuals. They focus on temporal stability of explicit and implicit learning during repeated tasks (test and learning task alternated with different delays from 0 to 90 s). In general, the experiment is well conducted, and report is well written, but would benefit from condensation and clarification.

We thank the Reviewers for the careful and insightful review of our manuscript. We have addressed these concerns below.

Major concerns

1. The authors study explicit and implicit learning, but they do not explain these concepts or theories in the Introduction at all. Please explain these concepts and explain more clearly how and why your tasks represented/teste these learning mechanisms?

In the Introduction we have now added a brief explanation of implicit and explicit learning where we describe the current understandings of these learning components. In addition, we connect these learning mechanisms to the visuomotor rotation adaptation paradigm we implemented, in order to demonstrate how the task accounts for both learning components (Lines 58-70):

"One notable task involves visuomotor rotation (VMR), in which individuals adapt to a visual feedback perturbation (i.e., a rotation of the feedback trajectory, Krakauer et al. 1999; Redding et al. 2005). This adaptation has been shown to involve concurrent explicit and implicit learning mechanisms (Taylor et al. 2014). Implicit learning mechanisms utilize sensory-prediction errors to update the internal forward model, likely mediated by the cerebellum on a subconscious level (Synofik et al. 2008; Tseng et al. 2007). Additionally, explicit processes also contribute to the adaptation to the rotation of the feedback through conscious efforts, minimizing the experienced target errors (Bond et al. 2015; Mazzoni and Krakauer 2006). Thus, the VMR paradigm provides a method for assessing both explicit and implicit components of motor adaptation; the explicit component of learning is determined through an aiming report made by the subject, while the implicit learning is based on subtracting this planned movement direction from the actual movement trajectory (Taylor et al. 2014; Morehead et al. 2015).”

2. The implicit motor learning task appears as kinesthetic/proprioceptive task/learning. In this task, there is no visual cue during the movement, and thus it reflects more a somatomotor integration learning/adaptation than visuomotor integration. This should be emphasized for the current task(s) and results in Introduction and Discussion. Interestingly, both the proprioceptive perception and cortical processing of the proprioception are impaired with aging, that could partly explain your result of slower learning/processing (e.g. selection time) of the currently used tasks. See:

- Goble et al. 2009. Proprioceptive sensibility in the elderly: Degeneration, functional consequences and plastic-adaptive processes. Neurosci. Biobehav. Rev. 33, 271 278., DOI: 10.1016/j.neubiorev.2008.08.012

- Piitulainen et al. 2018. Cortical Proprioceptive Processing Is Altered by Aging. Front

Aging Neurosci. 10:147. DOI: 10.3389/fnagi.2018.00147.

We thank the Reviews for this comment. However, as noted in the Methods and Materials, there is a visual cue (the target), full visual feedback of the cursor on the training and re -training trials, and additional endpoint feedback at the end of the movement on these respective trials. If we

have misinterpreted the comment, we will gladly address this concern. That said, in the revised manuscript we now include these age-related deficits in proprioceptive perception and the cortical processing of this information in the Discussion. Specifically, we include this information when discussing how these issues that accompany natural aging must be taken into account when distinguishing motor learning and retention deficits associated with normal aging to those caused by brain disorders (Lines 670-680):

"Our experimental paradigm may provide a basis for clinical assessments of implicit and explicit learning and memory deficits. For example, it is well known that normal aging contributes to neurochemical and physiological alterations within the brain, including cortical thinning and decreased hippocampal volume (Nyberg et al. 2012), leading to reduced working and episodic memory (Park et al. 2002; Rönnlund et al. 2005; Malouin et al. 2010; Sauzéon et al. 2016), pair association (Light et al. 2004; Old and Naveh-Benjamin 2008), and decision-making (Cai et al.

2014). There is also evidence that proprioceptive perception and associated processing declines with age (Adamo et al. 2007; Goble et al. 2009; Piitulainen et al. 2018), which likely contribute to deficits in error-based learning and decision tasks. Treatment of neurodegenerative diseases requires a thorough understanding of how these age-related changes in sensory processing and memory stability are distinct from pathological deficits.”

3. Selection time is very long, there is almost a short-term retention period between each trial when the selection time is 10sec or more. It is very different for some of the older adults who need up to (on average) 20s (15s during the testing phase) to select movement direction. I believe that this is an underestimation as the authors excluded trials with selection time larger than 30s. Yet, these excluded trials will influence the adaptation process because they make the inter-trial time longer for these specific subjects. This will have an influence on total adaptation rate. The fact that the selection time is much larger in older adults is a major problem as these individuals have then an overall longer inter-trial duration, which has an influence on the amount of adaptation. It is also directly influence the estimation of the delay period as this can be up to 200% larger than instructed (30s instead of 10s for selection time of 10s and delay of 10s). The authors should perform their analyses with the actua l elapsed time during the delay period rather than the instructed/imposed delay time (in Fig.6, replace delay time with actual elapsed time). Note that the elapsed time could be different for the explicit retention (from end of previous trial to time of selection) and for the implicit retention (from end of previous trial to time of movement start?). One way to proceed who be to use the actual average elapsed time for each age group but it would be even better if there was a way to take into account inter-individual differences in elapsed time as this varies a lot between the participants, especially for the older adults (if not possible on the graph, it should at least be added as co-variate in the statistical analysis).

We agree with the Reviewers and thank them for these very helpful suggestions. We also apologize for the oversight in not clearly presenting the analysis of the selection time ; as pointed out, the selection time should be included in the change in adaptation with the imposed delay and in the statistical analysis.

On the first point, the reason we excluded the trials with a selection time larger than 30s is that all of these trials were at beginning of the training session when subjects first experienced the visuomotor rotation and were initially becoming familiar with how to make the selection using the knob (2.1% of training trials with selection time large than 30s occurred for the elderly group and

0.6% for the young group. All of these trials were in the first half of the training phase: trials 1-30). Based on our statistical analysis, selection times larger than 30s were outliers (greater than 3 times of the standard deviation of the average selection time for each respective group). Subjects

in both groups did not make selections with larger than 30s for trials in the second half of the training session (trials 31-60) and the entire testing session. Thus, excluding these trials did not affect the main findings of the study. We have clarified these points in the revised manuscript (Lines 234-243):

"Trials that had very slow/fast movement speeds (peak speed < 0.22 m/s or > 0.5 m/s) and long selection time (selection time > 30 seconds) were excluded (∼3% of total trials for the young group and ∼4.5% of total trials for the older group). The reason we excluded the trials with a selection time larger than 30s is that all these trials were at beginning of the training session when subjects first experienced the visuomotor rotation and were becoming familiar with rotating the knob and making an aiming selection. There were 2.1% of training trials with a selection time larger than

30s for the older group and 0.6% for the young group. All of these trials were in the first half of the training phase; none of the subjects made a selection with a time larger than 30s for trials in the second half of the training session and the entire testing phase.”

On the second point, we agree with the Reviewers that using the actual elapsed time (the imposed delay and the selection time) is a much better way to characterize the decay for both the explicit and implicit learning. In the revised manuscript we have described this in the Methods (Lines 206-

213):

"Due to the variance in the selection time and movement duration across subjects, we used the actual elapsed time to define the delay for explicit and implicit learning instead of the imposed 7 delay periods in the experiments to better capture the influence of the delay for each learning component. Thus, the delay for the explicit learning component was defined as the time interval from the end of previous trial to the time when the selection was made. The delay for the implicit learning component was defined as the time interval from the end of the previous trial to the movement onset. (Note the actual elapsed time was always slightly longer for the implicit learning component.)"

The temporal retention of the explicit and implicit learning in now in terms of actual elapsed time in Figure 7A and B (see below). (Lines 431-436):

"Thus, the overall temporal retention for both groups decayed significantly as the actual delay period increased (actual delay at 0s: 2.47s {plus minus} 0.19s, 5s: 8.01s {plus minus} 0.23s, 10s: 13.19s {plus minus} 0.27s, 20s:

23.11s {plus minus} 0.30s, 30s: 33.32s {plus minus} 0.24s, 50s: 53.71s {plus minus} 0.33s, 90s: 93.07s {plus minus} 0.65s for the young group;

0s: 4.01s {plus minus} 0.52s, 5s: 10.55s {plus minus} 0.81s, 10s: 16.14s {plus minus} 0.75s, 20s: 26.74s {plus minus} 0.88s, 30s: 36.81s {plus minus}

0.86s, 50s: 55.16s {plus minus} 0.1.19s, 90s: 96.91s {plus minus} 0.96s for the older group).”

We also modified the corresponding statistical analysis (taking into account the actual elapsed time as a covariate, Lines 247-250):

"To examine the potential difference in the temporal retention of the learning between the two groups, we fitted the absolute angular decay and the relative decay percentage (see above) using a LMM with fixed effects of group and learning type, random effect of subjects, and actual elapsed time as a covariate.”

This greatly improved our conclusions on the role the different learning mechanisms played during the decay of adaptation (Lines 461-469):

"Significance was found for all three fixed effects (F(1, 488.16) = 49.35, P < 0.0001 for actual elapsed time, F (1, 484.68) = 156.10, P = 0.0018 for learning type and F (1, 75.87) = 10.52, P =

0.0018 for group) and interaction among the three fixed effects (F(1, 484.89) = 9.62, P = 0.002). Post hoc tests showed that for both groups, implicit learning decay was significantly larger than the explicit learning decay over time (tratio(484) = 3.02, P = 0.0026 for the young group; tratio(486)

= 7.19, P < 0.0001 for the older group). However, there was no significant difference in the temporal angular decay of explicit aiming between the two groups (tratio(486) = 0.18, P > 0.85) and temporal angular decay of the implicit learning for the older group was significantly larger than the decay of the young group (tratio (484) = -4.17, P < 0.0001).”

Lines 486-493:

"The same LMM with random effect of subjects was performed to examine the effects of actual elapsed time, learning type, and age group on the relative temporal decay of the explicit and implicit learning. Significance was found for the interaction among the three fixed effects (F(1,

484.89) = 4.80, P = 0.029). Post hoc tests revealed the same results that relative retention of the explicit learning were not significantly different between the two groups (tratio(486) = 0.36, P > 0.71) and the relative temporal decay of the implicit learning for the older group was significantly larger than the decay of the young group (tratio (484) = -2.69, P = 0.0074).”

4. Statistics of overall adaptation do not fit with what one sees on the figure. Statistics as performed currently are very dependent on the selection of time windows (middle and late). It would be better to perform an unbiased analysis by taking the mean over all the learning period (trials 1-60).

We agree with the Reviewers that the statistics on the adaptation levels during training are somewhat dependent on the selection of the windows (placement and size). However, it is important to note that our goal was to show the progression of the different learning components and the selection time for both groups at different points during the training. We chose the periods (early, middle, and late) based on the number of trials and made sure the window sizes were consistent for all comparisons/analyses. Unfortunately, taking the mean over the entire learning period would obscure the ability to observe any progressive change in performance. We therefore maintained the selection time windows, but now provide more context for these periods (Lines

161-166):

"We defined three periods during training in order to compare learning between groups (early training: the first 5 training trials (1-5), middle training: the middle 5 trials of training (28-32) and late training: the last 5 training trials (56-60)). Note that these periods were used to show the progression in behavior throughout training and were based on the total number of training trials. Also, the results are similar for different window sizes (3 or 7 trials).”

As suggested we also report the mean values for the different learning components and overall learning (Lines 285-292):

"We first evaluated the mean adaptation over the training period for both groups. Throughout the entire training period, the subjects in the young group had a significantly higher explicit learning level (17.73o {plus minus} 3.40o for the young group and 11.56o {plus minus} 3.6o for the older group; two-tailed unpaired t-test, P > 0.022, d = 0.76) and a similar implicit learning level compared to the subjects in the older group (16.35o {plus minus} 3.43o for the young group and 18.65o {plus minus} 3.64o for the older group; two-tailed unpaired t-test, P = 0.26, d = 0.37), leading to significantly higher overall adaptation for the young group (34.07o {plus minus} 4.55o for the young group and 30.20o {plus minus} 4.68o for the older group; two-tailed unpaired t-test, P > 0.023, d = 0.75).”

5. Tiny differences in baseline should be taken into account in the ANOVA as a covariate. The data from the blank trials during baseline are never reported. How accurate were people during baseline in aiming to the selected point? These values should be described and should be taken into account as a co-variate in the analysis (not simply substracting).

We agree with the Reviewers that the baseline should be evaluated and reported for both groups. This information was provided in the initial submission under Baseline Performance. We observed no difference in reach angle between the groups, and the learning components were not significantly different from zero (Lines 264-279):

"All subjects practiced selecting the aiming direction and reaching to the targets during the baseline block with 20% blank no-visual feedback trials (a total of 8 blank movements). During the baseline trials, all participants in both groups moved the cursor directly to the target, and the reach angles were not significantly different from each other (young: 0.45o {plus minus}

0.23o, older: 0.47o {plus minus} 0.29o; two-tailed unpaired t-test, P = 0.97, d = 0.012) and not significantly

different from 0o (two-tailed one sample t-test, P = 0.16, d = 0.37 for the young group and P =

0.12, d = 0.44 for the older group). Participants demonstrated no explicit aiming or implicit learning during this block, as expected (explicit: young: -0.029o {plus minus} 0.046o, older: -0.031o {plus minus} 0.10o; two-tailed one sample t-test, P = 0.53, d = 0.14 for the young group and P = 0.75, d = 0.073 for the older group; implicit: young: 0.48o {plus minus} 0.24o, older: 0.50o {plus minus} 0.32o; two-tailed one sample t-test, P = 0.06, d = 0.45 for the young group and P = 0.13, d = 0.36 for the older group). Additionally, there were no significant differences between the explicit and implicit components of adaptation in both groups (An analysis using LMM with fixed effects of group and learning type and random effect of subjects: F(1, 74) = 0.0012, P = 0.97 for group, F(1, 74) = 0.35, P = 0.12 for learning type and F(1, 74) = 0.0021, P = 0.96 for the interaction between the two fixed effects).”

When we included the differences in baseline as a covariate in the analysis we saw no difference in the statistical results. Thus, due to a lack of any differences in baseline we decided not to include these measures as a covariate in the revised manuscript.

6. There is no statistical results that support the main conclusion of the paper (decay in implicit is different for young and old). Yet, when both implicit and explicit are used in the LMM (line

472), there is a main effect of age group, learning type and delay period and only an interaction between delay time and learning group. In other words, on this basis, one could not conclude that the difference in retention is specific to the implicit component as there is no interaction between learning type, delay period and age group. This statistical analysis does not match with the conclusions of the authors. The rest of this paragraph is describing that there was no difference in explicit retention between age groups but there was one in implicit retention. Yet, only an interaction can tell that this is a true difference. This is the problem reported by Nieuwenhuis et al. 2011. I believe that the absence of significant triple interaction is due to the huge variability in retention values for the group of older adults.

We agree with the Reviewers that some of reported statistics in the initial submission were problematic for the exact reasons outlined above. Based on the suggestion above (see comment

3), in the revised manuscript we now perform the analyses with the actual elapsed time instead of the imposed delay periods. We took the inter-individual differences in actual elapsed time into

consideration and set elapsed time as a covariate in our linear mixed -effect models. We also

assessed the relative retention level of the learning (compared to the average adaptation at the end of preceding re-training trials, see comment 12 below). These new and modified analyses have provided much stronger evidence to support our main conclusions.

7. The task has been well validated in younger adults. On most graphs, the authors provide individual datapoints, but there should be more of them. It is also important that they measure the explicit and implicit components of motor adaptation.

For all relevant figures (Figures 3, 4 and 8) we verified that all individual data points are present in order to show the individual differences along with the average statistics. Note that there may be some substantial overlap of individual data points, obscuring an accurate account. If we have misinterpreted this comment and/or there remains an issue with the revised figures, we will gladly fix any remaining concerns.

8. Results: Figure 3. Were there any significant differences here? The inter-individual variation is huge, and thus the sample size of the current data is not sufficient. With bigger sample size many of the ns. Results would be significant. Based on the text, the only difference was for older group in middle training period between implicit and explicit training. Please indicate the significant results clearly to the Figures.

We agree with the Reviewers. In the revised manuscript significance is now reported in all figures. We apologize that this information was missing in the initial submission. Also, we should note that the focus of our study was to examine and quantify the temporal decay of these different components of learning (along with the selection time), which the current sample size is sufficient. Although there is variability during the training phase, in our analysis of the selection time and learning decay (explicit, implicit and overall) we normalize within each subject which strengthens our overall analysis. We are especially thankful for the analysis suggestions (e.g., taking into account the actual elapsed time and examining the relative retention parameters) that greatly strengthen our conclusions.

9. The learning task is exhaustively repeated throughout the manuscript. It should be explained as clearly as possible in the Methods section, and then only briefly reminded in the later sections. It would be good to include more details also in the Figure 1 about the task (number of repetitions etc.). This work would also benefit from illustrative Video in the Supplements about the task if possible. Now it took some time to understand the protocol from the text and Figure 1.

We understand the Reviewers’ comment. Our intent was to be as thorough as possible and provide information on the task if the reader only focused on the Results or Discussion sections of the manuscript. That said, throughout the Results and Discussion in the revised manuscript we have significantly modified and removed sections that explained the task. Also, in the Figure 1 description, we have added more information regarding the organization of the task, number of trials in each block, and the overall design of the experimental apparatus (Lines 908-929):

"Figure 1. Experimental design and task structure. (A) The experimental setup consisted of a LCD monitor that displayed targets and visual feedback of the hand, a digitizing tablet, and a PC that controlled the paradigm. The monitor was 10 inches above the digitizing tablet. The midline of the subject was aligned with the center of the tablet and monitor. (B) Depiction of the explicit aim selection which consisted of a circle centered at the start location (gray filled circle). The movement goal is represented by the red filled circle, and a yellow filled circle represents the selection marker. This circle appeared 10 cm from the starting position. Subjects turned a knob at the beginning of each trial to move the yellow marker around the circle and indicate their explic it aiming direction. (C) The sequence of training and testing the subjects. After 20 familiarization trials, the subjects underwent 40 baseline trials where the cursor followed the hand position. This was followed by 60 training trials, during which a +45{degree sign} (clockwise) or -45{degree sign} (counterclockwise)

visuomotor rotation of the cursor movement feedback was applied. During the testing phase, each subject experienced a delay period (0, 5, 10, 20 30, 50, and 90 seconds). This was followed by a single probe trial, and then 5-7 re-training trials before the next delay period was tested. This pattern was repeated for a total of 4 probe movements for each delay duration. (D) Depiction of hand and cursor movements during baseline, and early and late training. During baseli ne the cursor and hand paths were aligned. When the visuomotor rotation was applied, there was a rotational offset between the cursor and hand path. Late in training subjects adjusted the hand path to guide the cursor to the target. (E) A representation of the explicit and implicit components of overall adaptation. Explicit aiming was measured as the angle difference between the aiming direction and the target, while implicit adaptation was defined as the angle difference between reach direction and aiming direction.”

Although we do not provide a video of the task, we hope that the improved description of the task setup and experimental paradigm have addressed any ambiguity noted in the initial submission.

10. The Results are exhausting and could be greatly condensed to improve clarity. I would omit all discussion from the Results (e.g. L321¬-329) and move it to the Discussion only.

Based on this comment we have tried to streamline the Results section. First, we omitted sentences and phrases within the Results section that were repetitive, including sections that explained the task. We condensed the section by re-structuring and editing some of the sentences for clarity. We also moved certain parts of the Results, such as the discussion of Figure 5 (now lines 642-667), to the Discussion. Finally, we have removed the summary of the mean values and corresponding standard error ranges for Figures 7 and 8. This information can be found in the statistics summary. On the last point (mean values and standard error), if the Reviewers and editor feel that this is necessary information we will gladly include them again in the manuscript, but we thought these details may be excessive in light of the comment above.

11. There is very little data on the different measures linked to retention (Fig.6). Selection time are provided on Fig.7 but the retention is primarily based on the difference between the last three trials (with visual feedback) before the break and the first trials after the break. Yet, there is currently absolutely no information on these three trial averages and this is required as they have a large influence on the retention parameters. The authors should provide 7 scatterplot graphs (one graph per delay value, on each graph: one point per participant, color-coded for age group) with average adaptation in the three movements (average across the 4 trials for that delay) on the x-axis vs. the average amount of adaptation on the retention trial without visual feedback (average across the 4 trials for that delay). The goal is to present the data from the adaptation trials during the testing phase and from the retention trials for each individual participant and each delay. This is currently missing in the paper.

We agree with the Reviewers and now show this supporting data. A new figure (Figure 6) was added following this suggestion to illustrate the relationship between the adaptation level in the last 3 retraining trials and retention on the delay probe trials for explicit aiming, implicit learning and overall learning:

With a corresponding figure legend (Lines 968-976):

Figure 6. Retention of learning with respect to previous re-training trials. The amount of learning (aiming and reaching angle) on the probe trial is plotted against the amount of learning on the previous re-training trials for (A) explicit aiming, (B) implicit learning and (C) overall learning for the seven different delays (each respective column). The individual data for each subject are represented by the green circles (young subjects) and purple squares (older subjects). The group mean in each panel is depicted by the larger filled symbols for each respective group, and the shaded ellipses represent the 95% confidence interva ls. Horizontal and vertical dashed lines show the maximum VMR perturbation (45o) and the diagonal dashed line is the unity line.

The interpretation of this analysis is now included in the Results section (Lines 403-414):

"We first assessed retention of explicit aiming, and implicit and overall learning by comparing the aiming angles and reaching angles on the non-visual feedback probe trials to the average aiming and reaching angles over the last three re-training trials (see Materials and Methods). W e plotted the former as a function of the latter for each individual probe trial in Figure 6. For the explicit learning (Figure 6A), the majority of the data points for both groups were on the diagonal unity lines for the different delay periods, indicating no obvious decay over time (note the mean values in each panel are on the unity lines). For the implicit learning and overall learning (Figure 6B and C), most data points started to fall below the unity line as the delay period increased. Also, the data points for the older group (purple squares) deviated further from the diagonal unity lines than those for the young group (green circles), suggesting greater decay with time (again, the mean values in these panel are below the unity line, with the data for the older group below the young).”

12. The adaptation behavior during the testing phase should be described and illustrated for both age groups. To account for potential differences in adaptation before retention, I would like the authors to perform their analyses on relative retention parameters (adapt-retained/adapt) rather than on the absolute difference as they did here.

We agree and thank the Reviewers for these suggestions. In the revised manuscript we have now described these analyses in the Methods (Lines 198-205):

"For each time delay, retention of learning was quantified by comparing the angular deviation on the no-visual feedback probe trials to the average angular deviation at the end of the preceding re-training trials (the last three trials of the 5-7 re-training sequence). The absolute retention (difference in angular deviation between the probe trial and the average on the re-training trials) and relative retention (percentage of the adaptation retained: (probe trial angle - average angle on the re-training trials) / average angle on the re-training trials) were used to examine the decay of the two learning components and overall learning over the delay periods.”

In addition, the analysis of the relative retention was added in the Results (Lines 470-493):

"To better understand the temporal decay for the two learning components, we also measured the relative retention compared to the end of the preceding re-training trials (Figure 7C and D). The relative retention of the overall learning and explicit learning for both groups follow the same trends as the absolute angular retention shown in Figure 7A and B. Both groups had similar relative retention of the explicit learning, and the relative overall learning decay for the older group was significantly larger than that for the young group. For actual elapsed time up to 100s, the average decay of the overall learning for the older group (15%) was about twice of the decay for the young group (8%). One interesting result we observed was that the relative decay of th e implicit learning for both groups was about twice that of the relative decay of their overall learning. This difference is due to dividing the absolute angular decay by the implicit learning levels at the end of the re-training trials which accounted for about 50% of the overall learning (i.e., both explicit and implicit learning contributed about 50% to the overall learning for both groups at the end of training, Figure 2 and 3). Thus, the relative temporal decay of the implicit learning was actually greater than the relative decay of the overall learning for both groups due to the stability of the explicit aiming (For the young group, time constant: 16.76 {plus minus} 23.44s for the implicit decay and

17.38 {plus minus} 23.84s for the overall decay; for the older group, slope: -0.26 {plus minus} 0.095 for the implicit decay and -0.14 {plus minus} 0.052 for the overall decay). The same LMM with random effect of subjects was performed to examine the effects of actual elapsed time, learning type, and age group on the relative temporal decay of the explicit and implicit learning. Significance was found for the interaction among the three fixed effects (F(1, 484.89) = 4.80, P = 0.029). Post hoc tests revealed the same results that relative retention of the explicit learning were not significantly different between the two groups (tratio(486) = 0.36, P > 0.71) and the relative temporal decay of the implicit learning for the older group was significantly larger than the decay of the young group (tratio (484)

= -2.69, P = 0.0074). “

Finally, the corresponding results are now shown in Figure 7C and D.

With a corresponding figure legend (Lines 978-993):

"Figure 7. Temporal decay curves. The decay of implicit, explicit and overall learning as a function of the actual elapsed time is shown for (A and C) young and (B and D) older subjects. The actual elapsed time is the time interval from the end of previous trial to the time when the selection was made for the explicit learning component, and the time interval from the end of previous trial to the movement onset for the implicit learning component. Note the actual elapsed time was always slightly longer for the implicit learning component. The absolute retention (angular deviation) and relative retention (percentage of the adaptation at the end of preceding re-training trials) are shown on the top and bottom row, respectively. The mean value for each respective measure (overall learning: light blue symbols, explicit aiming: red symbols, implicit learning: dark blue symbols) on the probe trial is plotted for the actual elapsed time for each respective delay period (0, 5, 10, 20, 30, 50, and 90 seconds). Each panel includes the best fit function (Panels A and C: exponential for young subjects, Panels B and D: linear for older subjects) for the decay of overall adaptation and implicit learning. Explicit learning for each group was best described by a linear function. Vertical error bars represent SEM in the respective measures of decay, and horizontal error bars represent SEM in the actual elapsed time.”

The new results based on the relative retention have indeed strengthen our conclusions and we thank the Reviewers again for this suggestion.

13. Influence of baseline: the angle decay over time (ADoT) is computing by subtracting the baseline-corrected adaptation measured during retention trials (R-B) from the average of non- baseline-corrected adaptation trials before the delay (A). By doing so, the researchers are essentially adding the baseline correction to the angle decay observed during the delay period (ADoT = (A-R) + B).

This was a mistake in our description of the analysis and we apologize for the confusion. The angle decay over time was computed by subtracting the non-baseline-corrected adaptation measured during retention probe trials from the average of the non-baseline-corrected re- adaptation trials before the delay. We have corrected the mistake and made this point clear in the revised manuscript (Lines 200-206):

"The absolute retention (difference in angular deviation between the probe trial and the average on the re-training trials) and relative retention (percentage of the adaptation retained: (probe trial angle - average angle on the re-training trials) / average angle on the re-training trials) were used to examine the decay of the two learning components and overall learning over the delay periods. These measures of retention were a direct comparison before and after the delay; there was no need for a baseline correction.”

14. The values presented on Lines 419-420 seem inconsistent with the values presented on Fig.6.

For the young group, there seems to be a small increase in adaptation during the 0s break

(from 43.47{degree sign} to 44.03{degree sign}). Yet, the average of total adaptation for the

0s delay on the left panel of Fig.6 is below zero while it should be above it. How is the value

on line 433 (-0.27{degree sign}) is obtained?

We agree with the Reviewers and apologize for the confusion. Although these values have been removed from the Results section in order to reduce its length, this was a miscalculation on our part. We have corrected this in the revised manuscript (Lines 418-422):

"For angular decay (Figure 7A), at the delay of 0s, the amount of compensation for the VMR was not significantly different from the amount of overall recalibration during re -training for both groups (re-training: 44.94o {plus minus} 0.66o and 0s: 44.97o {plus minus} 0.81o, two-tailed paired t-test P > 0.94, d =

0.0065 for the young group; re-training: 42.29o {plus minus} 0.91o and 0s: 41.47o {plus minus} 1.22o, two-tailed paired t- test P > 0.48, d = 0.17 for the older group).”

In addition, we have modified this analysis based on the Reviewers’ suggestions and made sure that any errors in these calculations have been corrected.

Minor comments

1. Conclusions are missing from the end of the Discussion, but are more or less mentioned in

the 1st paragraph of Discussion. I would move the primary Conclusions to a specific

Conclusions section in the end of the Discussion.

As suggested, in the revised manuscript we moved the first paragraph of Discussion to the end so that it now is the Conclusion (Lines 700-712):

"In this study we examined the differences in the short-term (seconds to 1.5 minutes) temporal stability of explicit and implicit motor learning following adaptation of arm reaching movements. We specifically focused on possible variations in the stability due to normal aging. In general, both young and older subjects showed similar overall learning rates of adaptation and by the end of training there was not a significant difference in the level of explicit aiming and implicit learning between the two groups. Interestingly, over a short-term delay period (from 0s to 90s), both young and older subjects showed almost complete retention of the explicit aiming achieved during adaptation. However, the temporal change for implicit learning was very different between subject groups. Consistent with previous work, young subjects had an exponential decrease in the implicit learning component. In stark contrast, older subjects had a linear decrease over the same time period. These results demonstrate that within the context of this motor adaptation task, normal

aging reduces the short-term stability of implicit motor learning across time, with very little impact

on explicit motor learning processes.”

2. Please add titles to the panels of Figures in addition to panel labels. It would be much easier to understand what is presented in each panel.

As suggested, in the revised manuscript we have added panel titles to all figures.

3. It would be more correct to call ‘elderly’ group as ‘older’ group

We agree with this suggestion. Throughout the revised manuscript we replaced the term ‘elderly’

with ‘older’.

4. How does Fig.5 contribute to our understanding of possible differences in short-term retention between young and old adults. Furthermore, I don’t agree that this is consistent with Haith et al. (lines 389-391) given that their low and high PT was in the order of a few 100ms and not several seconds. In addition, it does not make sense that the slope was steeper for older adults early in the adaptation period.

The Reviewers are correct and we apologize for the mistake. The slope for the older adults was less steep compared to the young subjects in the early adaption period. In the revised manuscript we now correct this mistake and soften our comparison to the previous work of Haith et al. and provide more context (multiple timescales of movement planning) of why this analysis gives novel insight (Lines 642-667):

"Our results presented in Figure 5 suggests that during early training the increase of the explicit aiming was related with the increase of the selection time; with longer selection time subjects in both groups had a larger explicit aim to counter the perturbation. Although demonstrated on a significantly smaller time scale (milliseconds as opposed to seconds/minute), this benefit of increased decision time is observed in previous findings (Haith et al. 2015, 2016) where subjects exhibited significantly greater error on trials with low preparation time compared with high preparation time trials. Importantly, the results suggest that the time allowed to prepare the motor output differentially influence the explicit an d implicit learning components-more time allows a greater contribution of explicit strategy. The shallower relationship in early training for older subjects suggests that they required more time to form an aiming direction that young subjects were able to formulate in much less time. In addition, all subjects required more time to select an aiming direction as the delay period increased (Figure 8). However, similar to training (Figure 4), the older subjects often required twice the amount of time compared to young subjects, suggesting greater uncertainty in forming the explicit plan. This finding is consistent with prior studies (Trewartha et al. 2014; 2016), where older subjects had a longer task reaction time during an explicit memory task. Furthermore, the increased reaction time in the older group for the memory task was found to correspond with reduced explicit memory performance, demonstrating a negative relationship between reaction time and explicit memory. Trewartha (2016) demonstrated similar findings in the context of a weight prediction task. Our results did not show a similar temporal relationship; older subjects were able to maintain aiming accuracy although the selection time increased (Figures 7 and 8). However, in both the training and testing phase, the longer selection time for older subjects may result from uncertainty in the aiming direction, rather than a general difficulty in making the selection. That is, there was no significant change in the explicit aiming at the end of training (Figure 3) or over the testing phase (Figure 7), but older subjects continued to require significantly more selection time compared to young subjects.”

5. Vandevoorde and Orban de Xivry 2019 reported no difference in short-term retention between young and old participants over a one-minute break for total adaptation and for implicit adaptation only (isolated by design)

We thank the Reviewers for this comment. In the revised manuscript we now clearly contrast the previous finding from Vandevoorde and Orban de Xivry (2019) to the current study (Lines 609-

619):

"It is important to note that these results point to an age-related deficit in the short-term temporal stability, which appears in contrast to the previous finding that older and younger subjects had similar retention of implicit and overall adaptation over a one-minute break during learning (Vandevoorde and de Xivry, 2019). However, Vandevoorde and de Xivry (2019) assessed this retention after a single time point, whereas we were interested in how this retention systematic changed over time. In addition, even in our study this single time point shows similar levels of decay between the two groups (see Figure 7 near the 50 second time point); extending the range beyond a minute (> 90 seconds) provides a full view of the difference in the decay rates. Furthermore, we assessed age-based difference in stability after asymptotic levels of learning had been achieved and maintained (through re-training), not during training as in this previous study.”

6. Why does the effect of group reported on line 429 does not match the effect of group on line

476? The model is different but should not the statistical results be the same because total

adaptation = exp.

We agree with the Reviewers and apologize for the confusion. We revised the entire analysis based on the suggestions detailed above and now use the actual elapsed time as a covariate in our model. In the revised manuscript these new results are presented in the Results section (please see above).