Children exhibit a developmental advantage in the offline processing of a learned motor sequence

Changes in specific behaviors across the lifespan are frequently reported as an inverted-U trajectory. That is, young adults exhibit optimal performance, children are conceptualized as developing systems progressing towards this ideal state, and older adulthood is characterized by performance decrements. However, not all behaviors follow this trajectory, as there are instances in which children outperform young adults. Here, we acquired data from 7–35 and >55 year-old participants and assessed potential developmental advantages in motor sequence learning and memory consolidation. Results revealed no credible evidence for differences in initial learning dynamics among age groups, but 7- to 12-year-old children exhibited smaller sequence-specific learning relative to adolescents, young adults and older adults. Interestingly, children demonstrated the greatest performance gains across the 5 h and 24 h offline periods, reflecting enhanced motor memory consolidation. These results suggest that children exhibit an advantage in the offline processing of recently learned motor sequences.


Supplementary Note 1: Participant characteristics, sleep and vigilance (Experiments 1 and 2)
Group means for the measures of both Experiments 1 and 2 are provided in Tables 1 and 2 in the main text and depicted in Supplementary Figures S1 and S2 below.Results from the corresponding statistical analyses can be found in Supplementary Tables 1 and 2 below.In brief, significant age group differences were observed for gender distribution, morningness/eveningness preference, sleep duration of the night prior to participation and subjective levels of sleepiness at the time of testing for both experiments.Additionally, the time of testing and objective levels of alertness were different among age groups in Experiment 1 and 2, respectively.These age group differences can largely be considered reflective of the convenience sample in the current research as well as known lifespan differences in a subset of these measures.We elaborate on these differences in the subsequent paragraphs.
The adult groups showed a gender distribution skewed towards more female than male participants.This group difference can largely be linked to our convenience sample.The potential impact of these differences in gender distribution across our age groups on our primary measures (i.e., learning magnitude, micro-online gains, and micro-and macro-offline gains) was assessed.
Results from the corresponding gender by age group interactions revealed no significant effects (all p > 0.08).
Older adults indicated a greater preference for mornings than the other age groups.This is in line with previous studies indicating that circadian preferences shift towards the morning with increasing age in adulthood 1,2 .
In Experiment 1, participants were instructed to complete the experiment between 9 am and 7 pm to avoid testing early in the morning or late at night when performance is more likely to be impacted by circadian influences.Nonetheless, young adults completed the experimental session later in the day as compared to the other age groups, potentially due to work and/or family responsibilities during the day.In Experiment 2, participants were given specific instructions regarding the timing of the sessions, decreasing the flexibility in the time windows for task completion and thus there were no differences among age groups.
Self-reported sleep duration was longer in children in both experiments and in adolescents in Experiment 2 as compared to the adult groups.Furthermore, older adults reported a shorter sleep duration in comparison to the other age groups in Experiment 1.These findings are consistent with a known decrease in total sleep time with age 3,4 .There were no differences between the two nights in Experiment 2.
Sleep quality during the night prior to the experimental sessions was comparable among age groups and experimental nights in both experiments.Based on previous literature, one would expect older adults to report decreased sleep quality 5 .However, as Vitiello et al. 6 found a mismatch between self-reported and objectively measured sleep quality, it is possible that our older adults overestimated the quality of their sleep.
Lastly, and unexpectedly, older adults self-reported a higher alertness than the other age groups in both experiments, with a group by session interaction in Experiment 2. We speculate that, and analogous to sleep quality, older adults underestimated their levels of sleepiness at the time of testing.This is supported by the fact that the older adults were not different from the adolescents and young adults in their performance on the PVTan objective measure of vigilance -in Experiment 2. Children were slower on the PVT as compared to the other age groups, which is in line with previous research 7 .

Supplementary Note 2: Assessment of learning dynamics with non-normalized performance measures (Experiments 1 and 2)
Results presented in the main text were based on the normalized performance outcomes, adjusted for baseline performance on the pre-learning random run.For completeness, exploratory analyses were also performed on the non-normalized performance outcomes.
Non-normalized performance measures on the SRT task of Experiments 1 and 2 are depicted in Supplementary Figures S3 and S4 and results from the corresponding statistical analyses are provided in Supplementary Tables 3 and 4, respectively.In brief, absolute accuracy remained stable across blocks of practice for all task runs of both experiments, as evidenced by no main effects of block.Furthermore, similar changes across task blocks were found among groups (i.e., no groups x block interaction effects), but the overall performance levels were significantly different among age groups (i.e., group main effect).Specifically, pairwise follow-up comparisons indicated that for certain task runs of the two experiments, children were less accurate than adolescents (i.e., all task runs of Experiment 1, except for the post-learning random), young adults (all runs of Experiment 1 plus pre-learning random and test of Experiment 2), and older adults (all task runs of both experiments except for the post-learning random of Experiment 2).
Absolute response times (RTs) on the pre-learning random revealed a group effect but no block or group x block interaction effect in both experiments.During the training run of both experiments, RTs significantly decreased across practice blocks, as shown by a main effect of block, and the overall RT was significantly different among age groups.Furthermore, in Experiment 1, no group x block interaction effect was found and thus the decrease in RT across the training blocks was similar among age groups.In Experiment 2, however, results revealed a marginally significant group x block interaction, indicating that the change in performance across training blocks tended to be different among groups.During the post-learning test run of both experiments, there was no significant block effect, indicating a performance plateau was reached.This plateau was reached by all groups (i.e., no group x block interaction effect), but the age groups reached significantly different performance levels (i.e., group main effect).Similarly, the post-learning random run showed no block or group x block interaction effects but did reveal significant overall performance differences among groups.Pairwise follow-up comparisons for all the main effects of group indicated that the children were significantly slower than the other age groups for all task runs.Furthermore, the adolescents were slower than the young adults for all task runs of Experiment 2 except for the post-learning test.And lastly, the older adults were slower than the young adults for certain task runs of Experiment 1 (i.e., pre-learning random and post-learning test) and all task runs of Experiment 2.
In summary, and as expected, children were less accurate and slower on the task as compared to the other age groups.Results corresponding to these changes in non-normalized performance metrics across blocks of practice were largely consistent with those in the main text on normalized data.

Supplementary Note 3: Assessment of learning magnitude and micro-learning with non-normalized data (Experiment 1)
The dependent measures of learning-magnitude (assessing sequence-specific learning) as well as micro-offline and -online performance changes were computed with normalized RT data in the main text.Here, we present results from the same analyses but with these performance indices computed with non-normalized response time (RT) data.

Learning Magnitude
The difference in non-normalized RT between the post-learning test (averaged across the 4 test blocks) and the post-learning random (average across the 4 random blocks) was divided by the non-normalized RT in the post-learning random run (averaged across the 4 blocks).A one-way ANOVA revealed a significant group effect (F(3,129) = 5.440, p = 0.002, ƞ 2 = 0.115; see Supplementary Figure S5).Follow-up comparisons revealed that a significantly smaller sequencespecific learning magnitude was observed in children as compared to adolescents (p = 0.014, G = 0.714) and young adults (p = 0.001, G = 0.970).This finding is consistent with the results based on the normalized data presented in the main text.
In summary, the calculation of the micro-online and -offline gains based on the non-normalized data demonstrates that children exhibited smaller and larger micro-online and -offline performance gains, respectively compared to the other 3 age groups.

Supplementary Note 4: Assessment of macro-offline performance changes with non-normalized data (Experiment 2)
Macro-offline performance changes in the main text were based on inter-session differences in normalized performance outcomes, adjusted for the baseline performance on the pre-learning random run.Here, we conduct exploratory analyses on macro-offline performance changes computed on non-normalized response times (RTs).

Random and sequence-specific macro-offline performance gains
Macro-offline gains in the random condition were assessed with the same computation as described in the main text but using the non-normalized RTs (Supplementary Figure S8A).
Sequence-specific macro-offline gains were assessed with the same computation as described in the main text but using with non-normalized RTs (Supplementary Figure S8B).Significant main effects for both offline period (F(1,103) = 32.333,p < 0.001, partial ƞ 2 = 0.239) and group (F(3,103) = 7.523, p < 0.001, partial ƞ 2 = 0.180) were revealed, but there was no offline period x group interaction effect (F(3,103) = 2.079, p = 0.108, partial ƞ 2 = 0.057).Post-hoc pairwise comparisons on group differences across the two offline intervals indicated that children exhibited larger sequencespecific macro-offline gains as compared to young (p = 0.026, G = 0.627) and older adults (p < 0.001, G = 0.949).Additionally, adolescents showed significantly larger sequence-specific offline gains in comparison to older adults (p = 0.003, G = 1.360).These results suggest that the offline changes in children likely involved the strengthening of their sequential memory.These findings are consistent with the results based on the normalized RTs presented in Supplementary Note 5 below.

Supplementary Note 5: Random and sequence-specific macro-offline performance changes with normalized performance measures (Experiment 2)
Figure S11A displays the random macro-offline gains across both offline intervals per age group and as a function of age.Significant main effects for both offline period (F(1,103) = 21.409,p < 0.001, partial ƞ 2 = 0.172) and group (F(3,103) = 4.097, p = 0.009, partial ƞ 2 = 0.107) were revealed, but there was no offline period x group interaction effect (F(3,103) = 0.601, p = 0.616, partial ƞ 2 = 0.017).Post-hoc pairwise comparisons on group differences across the two offline intervals indicated that older adults exhibit lower random macro-offline gains as compared to children (p = 0.023, G = 0.700) and adolescents (p = 0.012, G = 1.041).Importantly, the children did not differ from any of the other age groups.
Figure S11B displays the sequence-specific macro-offline gains across both offline intervals per age group and as a function of age.Significant main effects for both offline period (F(1,103) = 32.558,p < 0.001, partial ƞ 2 = 0.240) and group (F(3,103) = 7.210, p < 0.001, partial ƞ 2 = 0.174) were revealed, but there was no offline period x group interaction effect (F(3,103) = 1.737, p = 0.164, partial ƞ 2 = 0.048).Post-hoc pairwise comparisons on group differences across the two offline intervals indicated that children exhibited larger sequence-specific macro-offline gains as compared to young (p = 0.034, G = 0.609) and older adults (p < 0.001, G = 0.923).Additionally, adolescents showed significantly larger sequence-specific offline gains in comparison to older adults (p = 0.003, G = 1.340).(d).Details of the corresponding statistical results are presented in the main text.In brief, there was a significant relationship between micro-and macro-offline gains collapsed across the 4 age groups (Panel a).Between group comparisons revealed that this relationship was significantly different between young adults and each of the other 3 age groups.Note, however, that these between-group differences were no longer significant if the young adult with the large negative 5hr offline gains was excluded from analyses.n = 27 in each of the 4 age groups.

Figure S3 Figure
Figure S3

Figure S4 Figure
Figure S4

Figure S6 Figure
Figure S6

Figure S7 Figure
Figure S7

Figure S9 Figure
Figure S9

Figure S10 Figure
Figure S10

Table 2 | Participant characteristics in Experiment 2.
9 and subjective and objective vigilance measures of the four age groups.Gender reports chi square statistics and the Cramer's V effect size, whereas all other variables list F-values and eta-squared effect sizes.Significant values are marked with an asterisk.Df = degrees of freedom; ƞ 2 = eta squared; SSS = Stanford Sleepiness Scale8.PVT = psychomotor vigilance task9.Corresponding group means are provided in Table2of the main text and depicted in FigureS2as part of this Supplementary Information.FigureS2also indicates significant pairwise comparisons that were conducted as follow-ups to significant contrasts shown above.

Table 3 | Absolute (non-normalized) task performance in Experiment 1.
Results from statistical analyses of non-normalized response time (RT) and accuracy measures for all task runs of Experiment 1. Significant values are marked with an asterisk.Df = degrees of freedom; part ƞ 2 = partial eta squared; B x G = block x group interaction.

Table 4 | Absolute (non-normalized) task performance of session 1 in Experiment 2.
Results from statistical analyses of non-normalized response time (RT) and accuracy measures for all task runs of session 1 of Experiment 2. Significant values are marked with an asterisk.Df = degrees of freedom;part ƞ 2 = partial eta squared; B x G = block x group interaction.

Table 5 | Statistical results on initial learning dynamics with normalized performance measures in Experiment 2. RT Accuracy
Results from statistical analyses of normalized response time (RT) and accuracy measures for all task runs of session 1 in Experiment 2. Significant values are marked with an asterisk.Df = degrees of freedom; part ƞ 2 = partial eta squared; B x G = block x group interaction.Figure4in the main text depicts the learning dynamics of these normalized measures.