Age-related decline in cortical inhibitory tone strengthens motor memory

Ageing disrupts the finely tuned excitation/inhibition balance (E:I) across cortex, driven by a natural decline in inhibitory tone (γ-amino butyric acid, GABA). This causes functional decrements. However, in young adults, experimentally lowering GABA in sensorimotor cortex enhances adaptation memory. Therefore, using a cross-sectional design, here we tested the hypothesis that as sensorimotor cortical GABA declines naturally with age, adaptation memory would increase, and the former would explain the latter. Results confirmed this prediction. To probe causality, we used brain stimulation to further lower sensorimotor cortical GABA during adaptation. Across individuals, how stimulation changed memory depended on sensorimotor cortical E:I. In those with low E:I, stimulation increased memory; in those with high E:I stimulation reduced memory. Thus, we identify a form of motor memory that improves naturally with age, depends causally on sensorimotor cortex neurochemistry, and may be a potent target for motor skill preservation strategies in healthy ageing and neurore-habilitation.

Short-term retention Long-term retention

10-minutes 24-hours
Short-term retention (10-minutes) Long-term retention worn. During right-shifting prism exposure (E1-6), visual feedback enabled participants to correct their rightward pointing errors across trials. Consequent leftward after-effects were measured in intervening blocks without visual feedback throughout adaptation (AE1-6). After-effect retention was measured post-adaptation after a short (10 minutes) and long (24 hours) interval. There was significant retention at both time points. b. Age had no effect on the after-effect magnitude acquired by the end of adaptation (block AE6), nor on short-term retention (10 minutes). The key finding was that older adults showed significantly greater long-term retention (24-hours). Full statistics are in Tables S3 & S4. minutes of blindfolded rest there was significant short-term retention (mean error: −4.61 • , s.e.m.: 1 0.41 • , t (1434) = −11.36, p < 0.001; one sample t-test of mean retention: t (31) = −11.18, p <  Table S3 -model 5 4). In our previous work with young adults long-term retention was not significant 48 . The AE was 6 stable at both time points, indicated by no change in error across trials (main effect of Trial: both 7 p > 0.38). 8 Our hypothesis was that AE retention would increase with age. Fig. 1b plots the results. Age 9 had no effect on the AE magnitude acquired by the end of prism exposure (Block AE6), nor on 10 short-term retention (both p > 0.35; Fig. 1b; Table S4 -models 1 & 2). However, older age was 11 associated with greater long-term retention (Age × AE 24hrs : t (1432) = −2.24, p = 0.025, η 2 p = GABA and Glutamix ("Glx"= Glutamate + Glutamine, since these two metabolites cannot be dis-1 tinguished reliably at 3 Tesla). As expected, in both regions, age was associated with significant 2 grey matter atrophy (both p < 0.002), which could indirectly lower neurochemical concentration  Table S5 -model 3). 13 In the anatomical control region (occipital cortex), there was a qualitatively similar pattern 14 of age-related inhibitory decline, consistent with previous reports 28,29 . However this was not 15 statistically significant, likely reflecting the impact of quality controls that reduced the size of models 4-6).  Table S5.
1 Lower motor cortical inhibitory tone is associated with greater long-term retention. Based 2 on our previous work 48 , we hypothesized that lower motor cortical inhibitory tone would be asso-3 ciated with greater retention. Results confirmed this prediction (Fig. 3). Across individuals, higher 4 sensorimotor cortex E:I was associated with a larger prism AE at retention 24-hours after adapta- Plot shows relationships between brain chemistry and the magnitude of prism after-effect retained 24 hours after adaptation. Negative values on the y-axis indicate retention. a. Sensorimotor cortex ("M1") Across individuals, lower GABA was associated with greater retention. There was no relationship with Glx (Glutamate + Glutamine). b. Occipital cortex ("V1") There was no relationship between GABA or Glx and 24-hour retention. For each voxel and neurotransmitter, relationships control for the fraction of grey matter and white matter, and the other neurotransmitter. Absolute concentrations are expressed in arbitrary units. Full statistics details are in Table   S6.
1 Retention increases with age as a function of motor cortical inhibitory decline. Our key pre-2 diction was that as M1 GABA concentration declines with age, adaptation memory would increase, 3 and the former would explain the latter. We used mediation analysis to formally test this hypoth-4 esis. Mediation analysis is well suited to a situation in which the independent variable (Age) may 5 not directly influence the dependent variable (Long-term retention), but is instead hypothesized 6 to do so indirectly via its influence on candidate mediators (M1 E:I, GABA, Glx). The extent to 7 which the relationship between the independent and dependent variable is influenced by a medi-8 ator is termed the indirect effect. We tested the significance of indirect effects using a bootstrap 9 estimation approach with 10,000 samples (see Methods).
10 Figure 4 shows that, as hypothesized, the effect of age on long-term retention was mediated   (Table S8).   S1).
17 Figure 5 shows the results for the group average. Stimulation had no effect on short-term and GABA (GABA × a-tDCS: t (1415) = −1.73, p = 0.042, one-tail, η 2 p = 0.16) moderated the 4 stimulation effect, each in opposite directions (Table S9 -model 6). Across individuals, stimula-5 tion increased retention in those with higher GABA and/or lower Glx, and impaired retention in 6 those with lower GABA and/or higher Glx. This result was unchanged when controlling for av-7 erage movement speed during prism exposure (Table S10). It was also anatomically specific: V1 8 neurochemistry did not moderate the effect of stimulation on retention (Table S11).

9
Given the finding in Experiment 1 of an association between age and M1 E:I (Fig. 2), we 10 tested whether age could be substituted as a simpler, easy to measure, proxy for neurochemistry.

11
By contrast with M1 E:I, age alone did not moderate the effect of stimulation on retention (see  adults showed that the enhancing effect of M1 a-tDCS on motor skill memory was influenced 2 by a common polymorphism in the gene that codes for brain derived neurotrophic factor (BDNF 3 val66met 59 ). Here we tested for a similar effect on adaptation memory. BDNF polymorphism 4 (obtained in 24 participants in Experiment 2) did not significantly moderate the effect of M1 a- This study tested the hypothesis that healthy older adults would exhibit stronger adaptation mem-8 ory, owing to age-related M1 GABA decline. The results confirmed this prediction. In a cross-9 sectional study of healthy men (aged 49-81), older adults had higher long-term retention ( Fig. 1) 10 and lower M1 GABA (Fig. 2). A mediation analysis showed that the latter explained the former 11 (Figs. 3 & 4). When M1 neurochemistry was accounted for, there was no longer a relationship 12 between age and memory, consistent with full mediation. The findings were specific: anatomi-13 cally (M1 not V1), neurochemically (GABA not Glx) and functionally (long-term not short-term induced memory change fitted an inverted U-shape, suggesting there is an optimum range of M1 1 E:I within which adaptation memory is maximal. The results were specific to E:I within M1 2 (no effect for V1). Whereas GABA loss in older age has typically been associated with func-3 tional decline 23, 26, 31, 32 , the present results reveal a specific domain of motor function that instead 4 becomes naturally enhanced. Enhanced adaptation memory may help compensate for impaired 5 motor skill learning in older age 60 .

6
This study extends our previous behavioural findings that M1 a-tDCS during adaptation en- results indicate that adaptation memory strength, both intrinsic and experimentally induced, de-10 pends causally and inversely on M1 cortical inhibitory tone. Previously we provided prelim-11 inary evidence for this hypothesis from an experiment in a small sample (n = 10) of young 12 healthy adults. We inferred a likely causal link between M1 GABA and memory strength be-13 cause stimulation-induced changes in both were correlated. Here, in a larger sample of older 14 adults (a more relevant age for stroke translation) we provide more robust causal evidence for this 15 hypothesis from mediation and moderation analyses of correlational and interventional data.

16
Two manipulations, one natural and one experimental, indicate that adaptation memory de-17 pends causally on M1 E:I -ageing (Fig. 4) and brain stimulation (Fig. 6). Collectively, they show 18 that M1 E:I determines retention. On average, lower inhibitory tone is associated with stronger an optimal level of E:I at which retention is maximal. This optimum differs across individuals.
1 Increasing E:I via stimulation moves individuals with naturally low E:I towards their maximum, 2 whereas individuals with naturally high E:I exceed that maximum and retention becomes impaired.  Rather than enhance retention, stimulation impaired it in individuals with intrinsically high 16 M1 E:I (Fig. 6a). An alternative interpretation of this result (to that offered above, Fig. 6b ceiling to become breached (Fig. S5). The relative (non-exclusive) contributions of these two 4 potential mechanisms to impaired retention is a question for future work. Nonetheless, under 5 either scenario, the data provide causal evidence that there is an optimal range of M1 E:I, that 6 varies across individuals, within which retention is maximal.

7
In our previous work on neglect 48 , we speculated that M1 a-tDCS might antagonize (rather 8 than enhance) prism adaptation therapy in some patients. The present data support this since 9 here stimulation disrupted retention in healthy older individuals with naturally higher M1 E:I.

10
Stroke disrupts E:I across distributed cortical networks, and how this interacts with age-related 11 dysregulation is likely to vary by region and time. How the neurochemical constraints identified 12 here apply in stroke populations remains to be tested. The present normative dataset could help 13 guide interpretation of future stroke data. Serendipitously, in our earlier work, we established proof 14 of concept via experiments in young adults 48 . These showed that M1 a-tDCS during adaptation 15 specifically enhanced retention. If instead we had started by testing older healthy controls, we are 16 unlikely to have ever progressed to testing prism therapy + M1 a-tDCS in neglect, since (as in Fig.   17 5) we would not have found evidence that stimulation enhances retention. What reconciles our 18 previous and present results, across the combined evidence from younger and older adults, is the While the results provide a conceptual replication and extension of our previous finding that 1 intrinsic differences in neurochemistry explain inter-individual variability in stimulation response, 2 a critical caveat is that we sampled only men. This choice was informed by the fact that GABA 3 levels change across the menstrual cycle in women and hence adaptation memory, E:I, and stim-4 ulation responsivity are also likely to fluctuate accordingly. Given hormonal changes across the 5 lifespan, womens M1 inhibitory tone may have a different or more variable age-related trajectory 6 than men. Hence, to rule out these hormonal sources of variance, which would require much 7 larger samples, we did not recruit women. In so doing, we follow a long tradition in biomedi-8 cal research, the limitations and adverse consequences of which for women are significant 67 . Of 9 note, in our previous work, all three patient cases (by coincidence) were men. This may matter for 10 the therapeutic effects observed. By excluding gender-related heterogeneity, we could identify an 11 important mediator (M1 E:I) of variation in response to stimulation-induced functional plasticity, 12 at least in men. Dissecting out intrinsic biological factors in this way helps to causally explain 13 inter-individual differences and to dispel scepticism that this variability somehow renders brain 14 stimulation (and tDCS in particular) suspect as a neuroscience tool 68 . We also assessed genotypic 15 variation. Unlike previous work on motor skill retention 59 , we did not find that the BDNF val66met 16 genetic polymorphism influenced the effect of M1 a-tDCS on retention of adaptation, at least not 17 in this sample of older adult men (Fig. S4).

18
Our previous behavioural findings causally implicated sensorimotor cortex in retention be-19 cause, on average, M1 a-tDCS during adaptation led to stronger memories. Yet the spatially dif-20 fuse electric field induced by tDCS complicates functional localization. The current work provides 21 complementary evidence at the level of individuals that differences in adaptation memory relate 1 to differences in M1 neurochemistry. This need not be interpreted as evidence that memories are 2 formed and/or stored locally and/or exclusively in sensorimotor cortex. Of note, we measured 3 brain chemistry only in M1 and V1. Adaptation memory, like most functions, is likely to be dis-4 tributed,implemented through parieto-premotor-cerebellar circuit interactions. Yet we targeted M1 5 owing to evidence that it has a causal role in the early consolidation of motor learning 69-73 . These 6 data strengthen this evidence base in the case of adaptation. We interpret the data to indicate that 7 M1 is a privileged node in the distributed cortical circuitry that implements the early formation of 8 adaptation memory. That is, the strength of that memory trace can be changed during its formation 9 by tonic disinhibition of M1 (via a-tDCS), and the impact on individuals memory is quantitatively 10 and causally related to their local E:I balance within M1. This local neurochemical measure has 11 been shown to correlate with sensorimotor network resting state functional connectivity 74, 75 . Thus,

12
M1 E:I may be an informative measure because it also serves as a proxy readout of sensorimotor 13 network strength. Extra-synaptic GABAergic tone, that measured by magnetic resonance spec-14 troscopy, has also been linked to oscillatory markers of inter-regional neuronal communication 76, 77 . 15 Hence, this local M1 readout may also indirectly index inter-individual differences in propensity 16 for inter-areal communication strength, of functional relevance during adaptation. Thus, we con-17 clude that M1 is a sensitive node at which to both measure and manipulate adaptation memory 18 formation.

19
Previous work has generally found adaptation to be preserved or somewhat impaired in older 20 adults 35-46 . Instead, our work reveals that a specific sub-domain of adaptation -long-term mem-21 ory -is naturally enhanced in older adults and provides a neurochemical explanation of this phe-1 nomenon. Whether these findings are specific to reach adaptation, or prisms, or may also generalize 2 to other effectors and forms of adaptation remains to be investigated. These results identify a domain of upper limb motor functional plasticity that is naturally 18 increased in healthy ageing. Is this likely to be beneficial? That will vary with context. Adaptation perturbation 84 . For long-lasting, slowly evolving perturbations, such as gradual muscle stiffening 1 with increasing age, adapting to that and maintaining it over time would help to offset these dele-2 terious effects. Conversely, in volatile environments that require agents to quickly learn and forget 3 new visuomotor transformations (e.g., playing basketball on a windy day), slow forgetting would 4 be maladaptive. Hence, whether reduced M1 inhibitory tone and stronger retention is adaptive 5 or maladaptive depends on the context and the task. Higher GABAergic tone (in physiologically 6 younger adults) may enable more selective functional release of inhibition during learning 30 and 7 thus promote more selective retention (e.g. of those motor memories likely to be useful in future).

8
Hence, stronger adaptation memory in older age is best conceived of as a "paradoxical functional 9 facilitation" 85 -an isolated domain of upregulated function that may have benefits, but which is 10 a side effect of a deleterious process (age-related GABA loss) that primarily causes functional 11 decline 23, 26, 31, 32 .

12
Nonetheless, our findings provide grounds for optimism about healthy motor ageing. The 13 usual narrative is one of decline and loss. Here instead we identify a natural age-related functional 14 gain. Maybe we cannot "teach an old dog new tricks", but we can instead focus effort on adapting 15 and retaining existing skills, promoted by natural neurochemical changes that may contribute to 16 maintaining motor function for longer.

13
In Experiment 1, the sample size (n = 32) was chosen for comparability with previous 14 MRS studies that successfully identified associations between behaviour and age-related GABA 15 change 23,26,28,29,86 . Sample sizes in these previous studies ranged from 11 to 37. In Experiment touchscreen. There were two lateral targets situated 10 cm to the left and right of the central target.

12
The distance between participants' eyes and the central target was 57 cm.

13
During PA participants alternated between two types of task block: Visual feedback was prevented on each trial by the LCD shutter turning opaque at reach onset, thus 2 occluding vision of the target, reach and endpoint error. This enabled the leftward after-effect to 3 be measured without participants de-adapting in response to visual error feedback.

4
In both experiments, each PA session measured pointing accuracy during: baseline, adapta-

13
The purpose of washout was to maximise the likelihood of observing a stimulation-related reten-14 tion benefit at 24-hours. We reasoned that, if memory formation was strengthened by stimulation 15 during adaptation, then washout should negatively interfere with long-term retention in the sham 16 condition, but not in the anodal condition.

17
Transcranial direct current stimulation. In Experiment 2, tDCS was delivered by a battery Pulses and Analysis). The unsuppressed water signal acquired from the volume of interest was 14 used to remove eddy current effects and to reconstruct the phased array spectra 99 . Single scan 15 spectra were corrected for frequency and phase variations induced by subject motion before sum-16 mation. Glutamix (Glx) was used in the current study due to the inability to distinguish between 17 glutamate and glutamine using a 3T MRI scanner. To avoid biasing the sample towards high con- in the VOIs were calculated using FMRIBs automated segmentation tool 108 . They are reported 20 together with MRS data quality metrics in Table S2.

21
Across individuals, the total creatine (tCr) concentration estimate was negatively correlated 1 with age in the M1 voxel (r (22) = −0.46, p = 0.04) although not in the V1 voxel (r (18) = 2 −0.06, p = 0.81; Fig. S2b). Owing to this confound with age, tCr could not be used as a valid 3 internal reference for metabolite estimates. Hence, throughout this work, we used absolute con-4 centration estimates for GABA and Glx, rather than expressing the data as ratios of tCr.     sciencedirect.com/science/article/pii/S1076633218304008.        metabolite concentration estimates would be associated with high %CRLB, in a way that truly re-2 flects high estimation uncertainty. In this scenario, it would be valid to mistrust the data based on 3 a high %CRLB. However, because of the relative nature of the %CRLB, this metric also strongly 4 depends on its denominator, i.e. the estimated metabolite concentration. Hence, for two samples 5 acquired with equivalent levels of noise, the sample with a lower metabolite concentration will 6 have a higher %CRLB. In that scenario, rejecting the dataset based on interpreting its high %CRLB 7 as an indicator of high estimation uncertainty would be invalid.

8
This is relevant to the present study, in which we aimed to measure a reduction in GABA 9 concentration with older age. The %CRLB cutoff criterion introduces a potential selection bias that 10 could artificially bias the sample towards excluding participants with low GABA concentrations 11 (and correspondingly high %CRLB). Therefore, to avoid this potential methodological confound 12 we developed a data quality filtering approach that did not solely consider high %CRLB with 13 respect to an arbitrary cutoff, but also considered the concentration estimate itself when deciding 14 whether or not to reject datasets for quality control. Thus we aimed to better deal with the following 15 scenarios: 1) datasets with a high %CRLB because of a low concentration estimate, rather than an 16 excessive level of noise -such datasets should not be excluded; 2) Datasets with a low %CRLB 17 simply because the concentration estimate is high might in fact be excessively high, given the 18 metabolite concentration -such datasets should be excluded.
quality filtering. First, the following model was fitted to the "concentration estimate × %CRLB" 1 relationship: where N i represents a group noise constant and C i the concentration estimates for a metabo- prisms off). Retention of the after-effect was measured 10 minutes and 24 hours post-adaptation. c. Procedure for Experiment 2. The procedure for Experiment 2 was the same as Experiment 1, except that left M1 anodal tDCS (real/sham) was applied throughout adaptation (grey shading). Short-term retention was followed by washout, during which participants observed and b. This panel presents the association between age and total Creatine (tCr), controlling for the fraction of WM and GM, in the M1 voxel (in green) and the V1 voxel (in blue). Shading indicates 95% confidence intervals. This panel shows that the tCr estimate was negatively correlated with age in M1, but not in V1. Because of this relationship, we use absolute concentrations of GABA and Glx throughout the paper, rather than using tCr for internal referencing. c. Magnetic resonance spectroscopy voxels group overlap map. The M1 voxel was centred on the left central sulcus in 22 participants (in green, MNI coordinate z = 52).
The control V1 voxel was centred on the bilateral calcarine sulcus in 20 participants (in blue, MNI coordinate z = 2). Colour bar represents the degree of overlap. All images are displayed in radiological convention (i.e. left side of the image corresponds to the right side of the brain).
Plot shows the same data with the same conventions as in Fig. 5, separated by genotype. a.
Behaviour of individuals with val/val genotype. b Behaviour of individuals with val/met genotype.
c. Mean, distribution and individual datapoints for short-term (10-min) and long-term (24-hour) retention separated by genotype. Linear mixed models showed that BDNF val66met genotype did not significantly influence how stimulation changed short (BDNF × a-tDCS: t (2141) = 1.00, p = 0.320) or long-term retention (BDNF × a-tDCS: t (2141) = −1.61, p = 0.107).  Figure S5: Alternative account of how stimulation impairs adaptation memory in individuals with high M1 E:I. a For comparison purposes, the model proposed in Fig. 6b is reproduced here. b. The schematic offers an alternative mechanistic interpretation of the data presented in Fig. 6a. This model assumes that in a healthy brain there is a ceiling on cortical excitation which, when exceeded, triggers homeostatic mechanisms that reduce E:I, to bring it back within physiological range. In individuals with naturally high M1 E:I (near ceiling), homeostasis could overshoot, leading to an overall decrease in E:I. This mechanism could explain why stimulation impairs retention in those with high baseline M1 E:I, without requiring a reversal in the sign of the relationship between M1 E:I and retention.     Table S3: Experiment 1: Prism adaptation behaviour. All LMMs analysed the normalised pointing error as the dependent variable (i.e. trial endpoint errors minus mean baseline error). Model (1) assesses the reduction of CLP errors throughout prism exposure (blocks E1-6), while model (2) captures the development of an after-effect on OLP trials (blocks AE1-6). Models (3) and (4) assess the persistence (intercept) and stability (main effect of Trial) of the after-effect (OLP) at the 10-minutes and 24-hours retention intervals. * p<0.1; * * p<0.05; * * * p<0.01 (all two-tailed). (1) (2) (3) (4) (5)    (1), (2) and (3) examine the relationship between age and prism after-effect at the end of adaptation (block AE6), and at the 10-minutes and 24-hour retention time points, respectively. Only 24-hour retention was related to age, such that older participants showed a larger (more negative) after-effect (AE). The next three models assess the robustness of this result when controlling for the average AE at the end of adaptation (model 4), the average AE at the 10minutes retention interval (model 5), and the average movement duration on CLP trials during prism exposure (model 6). The relationship between age and long-term adaptation memory survived controlling for all three factors, confirming that it was not an artefact of older participants adapting to a greater extent on the first day or pointing more slowly. * p<0.1; * * p<0.05; * * * p<0.01 (all two-tailed).  Table S5: Experiment 1: Older participants have a higher excitation:inhibition ratio in sensorimotor cortex. The linear regressions reported in this table examine the relationship between age and metabolite concentration within the motor (labelled "M1") and occipital (labelled "V1") cortex voxels. All models controlled for the fraction of grey and white matter within the MRS voxel, and included the MRS measure as the dependent variable. Model (1) shows the predicted significant positive relationship between age and E:I ratio (Glx:GABA). Models (2) and (3) decompose this relationship into its GABA and Glx constituents respectively. They highlight that the age-related increase in E:I was mainly due to a loss of GABA-ergic inhibition. The final three models show a qualitatively similar, though not significant, pattern within the bilateral occipital cortex. * p<0.1; * * p<0.05; * * * p<0.01 (all two-tailed).    Table S7: Experiment 1: Motor cortical GABA mediates the relationship between age and long-term adaptation memory.
Models (2) and (3) show that this relationship is driven by GABA (p = 0.013) and not Glx (p = 0.27). Model (4) shows that older age is associated with greater 24-hour retention (p = 0.02). Crucially, model (5) demonstrates that the association between age and 24-hour retention is no longer significant when accounting for M1 E:I. Further, model (6) shows that the mediation is specifically driven by GABA (p = 0.004) and not Glx (p = 0.78). Overall, these regression models provide support in favour of M1 GABA mediating the relationship between age and long-term retention, which was subsequently assessed formally. The   Table S7 are unchanged: M1 GABA, but not Glx, mediates the relationship between age and 24-hour retention. * p<0.1; * * p<0.05; * * * p<0.01 (all two-tailed).