Author response: A corticostriatal deficit promotes temporal distortion of automatic action in ageing Article Swipe

MAKE IMPACT

Miriam Matamales , Zala Skrbis , Matthew R. Bailey , Peter D. Balsam , Bernard W. Balleine , Jürgen Götz , Jesus Bertran‐Gonzalez · YOU? · · 2017 · Open Access · · DOI: https://doi.org/10.7554/elife.29908.030

Article Figures and data Abstract Introduction Results Discussion Materials and methods References Decision letter Author response Article and author information Metrics Abstract The acquisition of motor skills involves implementing action sequences that increase task efficiency while reducing cognitive loads. This learning capacity depends on specific cortico-basal ganglia circuits that are affected by normal ageing. Here, combining a series of novel behavioural tasks with extensive neuronal mapping and targeted cell manipulations in mice, we explored how ageing of cortico-basal ganglia networks alters the microstructure of action throughout sequence learning. We found that, after extended training, aged mice produced shorter actions and displayed squeezed automatic behaviours characterised by ultrafast oligomeric action chunks that correlated with deficient reorganisation of corticostriatal activity. Chemogenetic disruption of a striatal subcircuit in young mice reproduced age-related within-sequence features, and the introduction of an action-related feedback cue temporarily restored normal sequence structure in aged mice. Our results reveal static properties of aged cortico-basal ganglia networks that introduce temporal limits to action automaticity, something that can compromise procedural learning in ageing. https://doi.org/10.7554/eLife.29908.001 Introduction Learning of new skills permits optimal interactions with the environment while reducing cognitive costs, a fundamental adaptation contributing to behavioural autonomy and automaticity in many species. Motor skills are generally implemented through sequence learning, which allows elementary action units to be integrated into behavioural streams (Lashley, 1951), reducing latency and increasing speed of action (Sternberg et al., 1978). Similar to memory, movement sequences are thought to be organised into 'motor chunks', a cognitive-motor strategy proposed to depend on cortico-basal ganglia circuitry (Graybiel, 1998; Wymbs et al., 2012) that allows expression of whole behavioural programs as a single response (Abrahamse et al., 2013). Although initial learning capabilities are often preserved (Voelcker-Rehage, 2008; Matamales et al., 2016a), motor skill capacity is reduced in older adults, especially in tasks involving fine motor control. Indeed, some studies report that aged humans have a near zero capacity to form action chunks in newly acquired sequences (Shea et al., 2006; Verwey, 2010), whereas others find evidence of shortened motor chunks (Bo et al., 2009). Sequence learning appears, therefore, to be impaired in ageing, although the acquisition of new skills may still be possible, perhaps through the recruitment of compensatory strategies (Cabeza et al., 2002; Park and Reuter-Lorenz, 2009). Nevertheless, despite increasing knowledge of the neural bases of action automatisation (Jin and Costa, 2015), the precise neuronal mechanism by which advanced ageing affects the acquisition of skills remains unknown. To address this issue, we first designed behavioural tasks to determine which elements of action sequence learning were impaired by ageing. We then identified neuronal activation changes in the corticostriatal network that correlated with age-related learning defects, and reproduced some of these behavioural features through circuit-specific manipulations in young transgenic mice. Finally, we explored approaches to restore the ability of aged mice to perform action sequences through the introduction of an action-related feedback cue. Our findings suggest that the ageing brain may engage particular neuronal strategies to automatise action during skill learning, and that environmental support in the form of action-related feedback can be used to correct this behaviour. Results Aged mice display shorter patterns of action We first compared the ability of young (2 months old) and aged (20–22 months old) mice to develop action sequences. We focused on homogeneous sequences involving the concatenation of single responses because this procedure is particularly susceptible to chunking (Garcia-Colera and Semjen, 1987; Jin and Costa, 2015). After mild food restriction, we trained animals using an instrumental procedure in which the number of lever presses (LP) required to obtain food outcomes increased as training progressed on a random ratio schedule of reinforcement (from constant reinforcement, CRF, to random ratio 20, RR20; Figure 1A). The formation of action sequences was promoted by introducing a sequence trigger (ST), in which caching of 5 or 7 contiguous LP was required to access the current RR program (see Figure 1—figure supplement 1A). We used a novel approach to monitor the formation of sequences by parsing behavioural data into three elements: initiation (first LP, or LP occurring after a magazine check), execution (LP occurring after a LP) and termination (magazine check occurring after LP) (Figure 1B). Throughout training, both young and aged mice significantly escalated their overall performance (LP rate) as the ratio requirement to obtain an outcome increased (mixed ANOVA using factors of training and age F(3.5,48.9) = 116.852, p<0.001). A non-significant training x age interaction (F(3.5,48.9) = 1.153, p=0.341) suggested that overall instrumental performance was equal in young and aged mice (Figure 1—figure supplement 1B). However, half way through this training, aged mice produced a larger number of action sequences per minute, whereas young mice stabilised, or diminished, their sequence rates (Figure 1C). This was supported by a significant effect of training (F(16,224)= 28.885, p<0.001) and a significant training x age interaction (F(16,224) = 3.716, p<0.001). Further, a comparison of action sequences between days 7 and 17 of training revealed that both young and aged mice were able to extend the number of execution elements in their sequences throughout training (F(1,14) = 158.282, p<0.001) (Figure 1D). This increase was of a different magnitude (as suggested by a significant day x age interaction (F(1,14) = 40.03, p<0.001)), indicating that aged mice extended their sequences to a lesser extent than young mice. Separate analysis confirmed, however, that, although at a different level, both young (F(1,7) = 108.562, p<0.001) and aged (F(1,7) = 106.664, p<0.001) groups showed significant increases in sequence length (Figure 1D). In contrast, analysis of sequence duration showed a significant day x age interaction (F(1,14) = 8.368, p<0.05) with significant increases of sequence duration in young (F(1,7) = 18.376, p<0.01) but not aged (F(1,7) = 1.745, p=0.228) mice (Figure 1E). Importantly, aged mice showed evidence of learning about the ST applied from day 10, as they displayed a marked increase in the number of sequences that broke through the ST and therefore frequently accessed the RR schedule (Figure 1—figure supplement 1C). In contrast, young mice kept these numbers low, suggesting that they comfortably incorporated the ST requirement (F(3.065,42.908) = 7.757, p<0.001; training x age interaction F(3.065,42.908) = 6.027, p<0.01) (Figure 1—figure supplement 1C). Event-time plots recorded in both groups on day seven revealed no sequence structure in their instrumental performance, despite all rewards being obtained by all of the mice (Figure 1—figure supplement 1D). In contrast, analysis on day 17 revealed that both young and aged mice developed clearly defined action sequences over time (with identifiable initiation, execution and termination elements), although these sequences were more frequent and much shorter in the aged group (Figure 1F). Figure 1 with 2 supplements see all Download asset Open asset Evidence of sequence learning in aged mice. [Figure 1—source data 1] (A) Experimental design of instrumental conditioning under an increasing random ratio (RR) schedule of reinforcement, from constant reinforcement (CRF) to RR20. Sequence triggering (ST) programs were introduced on day 10. (B) Initiation, execution and termination elements were identified in each sequence according to lever press (LP) and magazine entry interspersing. (C) Number of action sequences per minute (sequence rate) displayed throughout instrumental learning [StatsReport1]. (D, E) Average length (D) and duration (E) of LP sequences produced by each young and aged mouse on days 7 and 17 of training. Orange traces represent individual mice. Asterisks denote significant day x age interaction (red) and simple effects (black, see text). N.S., not significant [StatsReport2] [StatsReport3]. (F) Event-time diagrams showing linearly organised LP sequences produced by all young and aged mice on training day 17. Shaded segments (gray) are amplified LP sequences (seconds). See Figure 1—figure supplement 1D for pre-sequence diagrams. https://doi.org/10.7554/eLife.29908.002 Figure 1—source data 1 Source data for Figures 1 and 3. https://doi.org/10.7554/eLife.29908.005 Download elife-29908-fig1-data1-v2.xlsx Figure 1—source data 2 Source data for Figure 1. https://doi.org/10.7554/eLife.29908.006 Download elife-29908-fig1-data2-v2.xlsx Figure 1—source data 3 Source data for Figure 1. https://doi.org/10.7554/eLife.29908.007 Download elife-29908-fig1-data3-v2.xlsx We next ruled out that this effect was due to the inability of aged mice to hear when pellets were delivered (Figure 1—figure supplement 2). First, we confirmed that the delivery of pellets in the operant conditioning boxes generated a clear vibration signal, which was more salient than the sound of the dispenser engagement and/or pellet dropping (Figure 1—figure supplement 2A). We then exposed a group of young and aged mice to six days of magazine training, where 20 pellets were randomly delivered in ~30 min sessions (Figure 1—figure supplement 2B–C). We found that all mice reduced their reaction time between pellet delivery and first magazine check (first check intervals) as training progressed (mixed ANOVA with factors training and age F(5,70) = 25.822, p<0.001), indicating that they learned to detect when pellets were delivered. A non-significant training x age interaction (F(5,70) = 0.360, p=0.874) confirmed that this reduction was similar in both groups (Figure 1—figure supplement 2B and D). This similar performance was not due to a higher overall number of magazine checks in the aged group, since both young and aged mice showed a similar increase of magazine approach responses throughout this training (F(5,70) = 25.822, p<0.001; training x age interaction F(5,70) = 0.360, p=0.874) (Figure 1—figure supplement 2C). Overall, our results showed that aged mice produced a higher number of action sequences, and in contrast to previous studies in humans (Shea et al., 2006; Verwey, 2010), these sequences could be extended (in terms of action number) but not temporally sustained. We thus explored whether sustained actions in aged mice were subject to temporal constraints, something that could explain the formation of shorter sequences of action. To address this, we used a Lever Hold (LH) instrumental procedure (Figure 2), a variant of the variable interval hold task (Bailey et al., 2015) in which the delivery of the reward depended on pseudorandom increments of LH time during training (Figure 2A and B). In this task, both young and aged groups displayed parallel escalations of LH times up to ~0.5 s. However, young mice increased their LH times throughout subsequent stages of training, whereas aged mice maintained their maximum performance around 0.5 s holds (Figure 2C). Statistical analysis supported this observation: mixed ANOVA (factors: training and age) showed that the time of lever hold significantly varied as instrumental training progressed in both age groups (F(16,176) = 73.471, p<0.001), although there was a significant training x age interaction, indicating that the age of the animals significantly affected their performance (F(16,176) = 16.796, p<0.001). Importantly, simple effects analyses confirmed that despite these differences, both groups independently elevated their performance throughout training (Young: F(16,112) = 69.387, p<0.001; Aged: F(16,64) = 51.33, p<0.001). Event scatter plots at different LH requirements revealed that, although both groups displayed a large number of very short presses, only young mice managed to produce sustained presses that approached, or exceeded, each LH requirement (Figure 2D). Event-time diagrams showed how all aged mice displayed much shorter holds than younger controls from the start of the session (Figure 2—figure supplement 1). Figure 2 with 1 supplement see all Download asset Open asset Aged mice are unable to temporally sustain their actions. [Figure 2—source data 1] [StatsReport4] (A) Mice were trained to press and hold the lever down for increasing periods to obtain rewards. (B) Experimental design of instrumental conditioning under an increasing pseudorandom lever hold (LH) schedule of reinforcement. Average LH times are indicated in seconds. (C) Acquisition of LH instrumental behaviour across training. Data are mean ±SEM (7 and 8 mice per group). Asterisk denotes significant training x age interaction. (D) Representation of each individual LH performed by all young and all aged mice on the last day of LH0.3, LH0.8, LH2, LH3.2 and LH5 training. LH times are presented in a Log10 scale (x axis), and sequences are distributed randomly in the y axis. https://doi.org/10.7554/eLife.29908.008 Figure 2—source data 1 Source data for Figure 2. https://doi.org/10.7554/eLife.29908.010 Download elife-29908-fig2-data1-v2.xlsx As such, these results provide evidence of action sequence formation in aged mice while also suggesting that these sequences could be influenced by temporal constraints, in line with previous results in humans (Bo et al., 2009). Sequence learning leads to rapid response bursts and aberrant chunking in aged mice We next investigated the forms of automaticity generated in both age groups during LP-based sequence learning. First, we analysed in our sequence timestamp dataset whether aged mice showed higher overall sequence speed. We found that average speed on the last day of training was higher in aged mice, although the effect fell below significance (t(14) = −2.104, p=0.054) (Figure 3A). However, subsequent analysis of the serial response times based on inter-press-intervals (IPIs) recorded on the same day revealed that aged mice produced a much higher proportion of very short IPIs (Figure 3B, Figure 3—figure supplement 1A), suggesting that they were reaching a higher sequence speed than young mice. Due to a resolution limit in the timestamp data (Figure 3—figure supplement 1B), we sought to establish the true maximum sequence speed by means of acoustic data, where the fastest bursts of LP in each mouse were identified by extracting the acoustic fingerprint of lever depression (Figure 3C, see Materials and methods). We found a strikingly high speed of responding in aged mice, in some instances displaying spates of LP with over 13 presses per second, much faster than the fastest bursts recorded in young mice (Figure 3C, Video 1 and Video 2). These speeds are similar to fully automatised scratch reflex movements in mice (Inagaki et al., 2003). Quantification analysis of fast sequences confirmed that average speed within fast bursts (in number of presses/sec) was significantly higher in aged than young mice (t(7.417) = −4.201, p<0.01, Figure 3D), whereas within-burst IPIs were significantly shorter (t(14) = 2.724, p<0.05, Figure 3E). Interestingly, despite their speed of responding, aged mice were less efficient at earning rewards (t(14) = 3.101, p<0.01), indicating that although their action bursts often broke through the sequence trigger (see Figure 1—figure supplement 1C), they rarely completed the RR requirements (Figure 3F). Figure 3 with 2 supplements see all Download asset Open asset Aged mice produce ultrafast sequences and aberrant chunking. [Figure 1—source data 1] [Figure 3—source data 3] (A) Average sequence speed calculated from timestamp data after 20 min of continuous instrumental performance on training day 18 [StatsReport33]. (B) Frequency distribution of the number of inter-press-intervals (IPIs) recorded in all young and aged mice at increasing intervals on training day 18. (C) Real IPIs from the fastest sequence recorded in a young and an aged mouse on day 18 (Fast sequence one in Video 1 and Video 2). Real lever press times were identified by aligning the sound waveform (top) and acoustic spectrogram (bottom) generated by lever press activity during the session. Identified presses are marked with a blue dashed line. (D, E) Average speed (B) and IPIs (C) of the five fastest LP sequences displayed by young and aged mice during the 20 min session on day 18 extracted from acoustic data [StatsReport5] [StatsReport6]. (F) Efficiency of LP instrumental behaviour (rewards earned per minute) in young and aged mice on day 18 [StatsReport7]. Asterisks (B–D) denote significant effects (see text). (G) Scatter plots of each within-sequence inter-press-interval (IPIs, blue) and sequence boundary interval (orange) produced by one young and one aged mouse (Young #7 and Aged #5) at different stages of training. Kernel density curves of each interval type are plotted at the bottom. Grey shades delimitate the within-sequence chunking space. (H, I) Average within-sequence IPIs (G) and sequence boundary (H) intervals. Insets are corresponding standard deviations. Data are mean ±SEM. Asterisk denotes significant training x age interaction [StatsReport27] [StatsReport28] [StatsReport31] [StatsReport32]. (J, K) Proportional scatter plots and kernel density curves showing intervals of the within-sequence (I) and sequence boundary (J) elements produced by all young and all aged mice during the first 600 s of training day 18. Insets are the total element counts for this period [StatsReport29] [StatsReport30]. Data in A, D-F, J and K are mean +SEM. p, p-value; asterisk, significant effect; N.S., not significant. https://doi.org/10.7554/eLife.29908.011 Figure 3—source data 1 Source data for Figure 3. https://doi.org/10.7554/eLife.29908.014 Download elife-29908-fig3-data1-v2.xlsx Video 1 Download asset This video cannot be played in place because your browser does support HTML5 video. You may still download the video for offline viewing. Download as MPEG-4 Download as WebM Download as Ogg Representative performance of a young mouse during the last day of sequence training. Two fast action sequences are captured (indicated in red font, top left corner). Fast action sequence one is represented in Figure 3A. Subject Young#5; active lever: right; training day 18. https://doi.org/10.7554/eLife.29908.015 Video 2 Download asset This video cannot be played in place because your browser does support HTML5 video. You may still download the video for offline viewing. Download as MPEG-4 Download as WebM Download as Ogg Representative performance of an aged mouse during the last day of sequence training. Two fast action sequences are captured (indicated in red font, top right corner). Fast action sequence one is represented in Figure 3A. Subject Aged#4; active lever: left; training day 18. https://doi.org/10.7554/eLife.29908.016 We next sought to analyse whether different chunking patterns emerged in young and aged mice during sequence learning by studying the temporal relationships between consecutive action elements throughout training (Figure 3G–K). We categorised the action repertoire into two classes of interval: the within-sequence interval (IPIs that reside within each sequence) and the sequence boundary interval (a combination of press-check and check-press intervals that lay in-between sequences) (see methods). The analysis of the distribution of these intervals at different stages of training provided a visual readout of the chunking that emerged during sequence development (Figure 3G). In both young and aged mice, within-sequence IPIs (Figure 3G, blue) clustered around a similar interval space (from ~0.1 to ~1 s intervals) by day 13, suggesting that both groups developed chunking of within-sequence (i.e. execution) elements. In support of this, we found that both young and aged mice similarly reduced the within-sequence IPIs throughout training (Figure 3H), as reflected by a significant effect of training (two-way mixed ANOVA with factors training and age: F(1.955,27.377) = 18.606, p<0.001) but no training x age interaction (F(1.955,27.377) = 0.562, p=0.573). Similarly, the variability of these IPIs was equally reduced in both groups (Figure 3H, inset), again showing a significant effect of training (F(2.352,32.904) = 6.545, p=0.003) with no training x age interaction (F(2.352,32.904) = 1.785, p=0.179). On the other hand, the development of sequence boundary intervals throughout training (Figure 3G, orange) was very different in young and aged mice. Whereas young mice displayed longer boundary intervals that were distant from the chunking space, aged mice tended to accumulate them in a narrow band near the chunking space (Figure 3G). Importantly, this band appeared earlier in training (day 10) in all aged animals, before within-sequence intervals chunked (Figure 3G, Figure 3—figure supplement 2A and B). This was evidenced in the quantitative analysis, where the young group showed an increase in sequence boundary intervals from days 10 to 17, whereas the aged group kept these intervals consistently low (Figure 3I), as indicated by a significant effect of training (F(3.884,54.375) = 8.005, p<0.001) and a significant training x age interaction (F(3.884,54.375) = 4.729, p=0.003). Although no drop in variability for this interval was found from day 10 to 17 (F(7,98) = 1.428, p=0.203), aged mice displayed lower variability than young mice in this period (F(1,14) = 8.855, p=0.01) (Figure 3I, inset), suggesting that the temporal relationships of boundary elements in aged mice had already been established in day 10 (see Figure 3—figure supplement 2). In order to obtain proportional measures of chunking in both groups, we compared the within-sequence and sequence boundary intervals produced by all young and all aged mice during the first 600 s of continuous action sequence behaviour (Figure we found in aged mice a larger proportion of very short within-sequence IPIs (Figure although the overall number of within-sequence presses was equal across groups (t(14) = (Figure The sequence boundary intervals produced during this period were also shorter in aged mice (Figure and the larger number of boundary elements in this group = p=0.003) (Figure reflected the higher number of action sequences produced by aged mice per minute (see Figure 1C). we found clearly different action automatisation within-sequence and sequence boundary elements in young and aged mice. young mice appeared to the elements within the sequences, aged mice displayed ultrafast chunking of within-sequence and showed evidence of chunking in sequence boundary elements. these results suggest that chunking of the sequence elements in aged mice could behaviour into ultrafast in automatic action in ageing. Sequence features with development of corticostriatal activity in aged mice The development of action sequences and is to of activity in cortico-basal ganglia circuits et al., and 2010), a of corticostriatal in et al., et al., Here, we the of activation of the corticostriatal network at different in young and aged mice action sequences for 20 min (Figure We identified over across and striatal in mice using a approach based on the and mapping of (Figure and a that is used to neuronal activity throughout the brain et al., We first the extent of activation in and striatal in young and aged mice (Figure ANOVA (factors: revealed no overall effect of age although there was a significant age x interaction = effect analyses age) showed that this interaction was by in the 2 = p<0.05) and the of the = p<0.01), as as in their of the = p<0.05) et al., and We next used this extensive dataset to the corticostriatal that were to action sequence We a by the corticostriatal activity based on activation in different and striatal in each mouse with the different features of sequence We found that sequence duration and speed of action displayed on the last day of training could only be by activation in and whereas activation in all other not significantly with of sequence execution (Figure We then investigated the engagement of different across the corticostriatal network in each age group by a analysis in which the mean of activation in each was correlated with all other across animals (Figure mice showed very low of with no significant relationships between and striatal throughout the extent of the perhaps due to in neuronal variability and/or et al., and (Figure left In contrast, aged mice showed that were significant in different of the corticostriatal especially the but not (Figure right In order to obtain information on this we compared the distribution of in the corticostriatal network by mapping the of individual in each of from young and aged mice et al., (Figure Our data in aged mice revealed large within the and of the that were of neuronal activation (Figure Overall, these results suggested an age-related in the corticostriatal network during action sequence execution that in and the two in motor chunking et al., Figure Download asset Open asset sequence features with activation of corticostriatal networks in aged mice. [Figure data 1] (A) analysis of neuronal activation in different 2 and striatal after 20 min of instrumental performance (day (B) and of each were through quantitative (C) displaying the of counts in different and striatal of young and aged mice. maximum and mean of the data are represented individual counts for each mice per group). Asterisks denote significant age x interaction (red) and simple effects (black, see (D) performance the counts of in the different and striatal from to and the action sequence features duration and as as overall instrumental performance number of lever displayed by each young and aged mouse during the last day of training. The corresponding the significance of each and diagrams are for each (E) analysis of the activity recorded in the different from to in all young

Type

peer-review

Language

en

Landing Page

https://doi.org/10.7554/elife.29908.030

OA Status

gold

Related Works

10

OpenAlex ID

https://openalex.org/W2982850725

Raw OpenAlex JSON

OpenAlex ID

https://openalex.org/W2982850725

Canonical identifier for this work in OpenAlex

DOI

https://doi.org/10.7554/elife.29908.030

Digital Object Identifier

Title

Author response: A corticostriatal deficit promotes temporal distortion of automatic action in ageing

Work title

Type

peer-review

OpenAlex work type

Language

en

Primary language

Publication year

2017

Year of publication

Publication date

2017-10-04

Full publication date if available

Authors

Miriam Matamales, Zala Skrbis, Matthew R. Bailey, Peter D. Balsam, Bernard W. Balleine, Jürgen Götz, Jesus Bertran‐Gonzalez

List of authors in order

Landing page

https://doi.org/10.7554/elife.29908.030

Publisher landing page

Open access

Yes

Whether a free full text is available

OA status

gold

Open access status per OpenAlex

OA URL

https://doi.org/10.7554/elife.29908.030

Direct OA link when available

Concepts

Automaticity, Neuroscience, Basal ganglia, Cognition, Action (physics), Psychology, Sequence learning, Serial reaction time, Action selection, Procedural memory, Cognitive psychology, Computer science, Cognitive science, Physics, Central nervous system, Perception, Quantum mechanics

Top concepts (fields/topics) attached by OpenAlex

Cited by

0

Total citation count in OpenAlex

Related works (count)

10

Other works algorithmically related by OpenAlex

Full payload