key: cord-0316904-msznwu1r
authors: Rakowska, Martyna; Lazari, Alberto; Cercignani, Mara; Bagrowska, Paulina; Johansen-Berg, Heidi; Lewis, Penelope A.
title: Distributed and gradual microstructure changes track the emergence of behavioural benefit from memory reactivation
date: 2022-04-29
journal: bioRxiv
DOI: 10.1101/2022.04.28.489844
sha: 77ab46989d4612a656f0fd25b0468b3c563ff569
doc_id: 316904
cord_uid: msznwu1r

Memory traces develop gradually and link to neural plasticity. Memory reactivation during sleep is crucial for consolidation, but its precise impact on plasticity and contribution to long-term memory storage remains unclear. We used multimodal diffusion-weighted imaging to track the location and timescale of microstructural changes following Targeted Memory Reactivation (TMR) of a motor task. This showed continuous microstructure plasticity in precuneus across 10 days post-TMR, paralleling the gradual development of behavioural benefit. Both early (0 - 24 h post-TMR) and late (24 h - 10 days post-TMR) microstructural changes in striatum and sensorimotor cortex were associated with the emergence of behavioural effects of TMR at day 20. Furthermore, the baseline microstructural architecture of sensorimotor cortex predicted TMR susceptibility. These findings demonstrate that repeated reactivation of memory traces during sleep engenders microstructural plasticity which continues days after the stimulation night and is associated with the emergence of memory benefits at the behavioural level.

The ability to change and adapt in response to internal or external stimuli by reorganising neural 21 connections, structure, and function is a fundamental property of the brain 1 . The remarkably plastic 22 nature of the nervous system forms the basis of learning and memory 2 4 . Learning-associated neural 23 processes occurring during sleep have been receiving increasing attention in recent times. The active 24 systems consolidation model suggests that newly encoded memories are reactivated during non-rapid 25 eye movement (NREM) sleep, and that this enables their re-coding from a temporary store to a more 26 permanent location 5 7 . However, while the process of memory consolidation is gradual and occurs 27 over long timescales 8 , it is unclear whether reactivation of memories during sleep leads to plasticity 28 in the neural substrate over time. Moreover, there is increasing evidence for distributed long-term 29 cortical storage of memories 9 11 , but it is unclear whether replay-driven consolidation of memories is 30 associated with plasticity at different cortical sites. Likewise, how such plasticity could lead to long-31 term memory storage in humans has not been studied sufficiently 12 .

Here we set out to track the location and timescale of microstructural changes underlying long-term 34 effects of memory reactivation during sleep using Targeted Memory Reactivation (TMR). TMR involves 35 associating learning items with sensory cues during wake and then covertly re-presenting them during 36 sleep (e.g., 13, 14 ). This is thought to trigger reactivation of the cue-associated memory representation 37 which leads to a better recall of the cued items compared to those that were not cued during the night 38 (i.e., uncued) 15 19 . In recent years, TMR has become a valuable tool to study the mechanisms of sleep-39 dependent memory processes. It has allowed us to establish a causal link between memory 40 reactivation and consolidation 20,21 , and to identify brain regions that are functionally involved in such 41 relationship 13,18,22 24 . Our prior studies further demonstrated that the behavioural effects of TMR 71 to decrease in precuneus in response to a series of repeated learning-retrieval epochs during wake 30 , 72 which could be regarded of as a proxy of memory reactivation during sleep 33 . We thus hypothesised 73 that TMR during sleep would also lead to rapid MD-driven plasticity within precuneus, thereby 74 supporting the functional activation of this structure in association with TMR that we observed in our 75 previous work 24 . However, we expected the motor-related regions to undergo long-term 76 microstructural changes, thereby reflecting their slowly evolving reorganisation 34 38 , as well as long-77 term functional engagement and volumetric increase in response to TMR 24 . In the current study, we 78 also used Restricted Water Fraction (Fr), as modelled by the CHARMED framework 32 . Fr is thought to 79 be more sensitive to microstructural changes than MD, especially in long-term assessment 28 . We thus 80 combined MD with Fr in a multimodal analysis protocol to uncover common microstructural patterns 81 across the two MRI markers and their relationship with TMR benefits across time. Finally, despite the 82 importance of tissue microstructure for memory formation 30 , it is unclear whether baseline tissue 83 microstructure is predictive of memory encoding capacity. To clarify this, we tested whether baseline 84 brain characteristics can determine TMR susceptibility, thus adding to our current understanding of 

We sought to examine whether repeated reactivation of a motor memory trace could engender 105 microstructural plasticity in the brain. We tested whether behavioural cueing benefit at different 106 sessions is associated with changes in brain microstructure within predefined regions of interest (ROIs) 107 over set periods of time. Thus, individual modality maps collected at different sessions were 108 subtracted from each other, yielding measures of early (S1-S2) and late (S2-S3) microstructural 109 plasticity. The resultant difference maps were used as inputs in the multimodal analysis, set out to 110 uncover common trends in MD and Fr change measures. Cueing benefit at S2, S3 and S4 were entered 111 as regressors, and separate contrasts were run for cueing benefit at each session. Given the 112 correlations between cueing benefits at different sessions ( Fig.1b-c) , sessions of no interest in any 113 given contrast were covaried out to test if the effects are specific to the cueing benefit tested.

We hypothesised that TMR during sleep would lead to rapid plasticity within precuneus. Consistent 115 with this, the analysis revealed a positive relationship between early plasticity in left precuneus (-6, -116 60, 18) and cueing benefit at S3 when controlling for the behavioural effects at S2 and S4 (p = 0.041; Table S1A , Fig.S3a ), such that greater cueing benefit was associated with greater reductions 118 in MD and greater increases in Fr. In addition, late plasticity in bilateral precuneus (4, -58, 16) was 119 associated with cueing benefit at S4 when controlling for the behavioural effects at S2 and S3 (p = 120 0.027; Fig.2d Table S1D , Fig.S3d ). To confirm that the results were specific to S4, we also tested the 142 relationship between microstructural plasticity and cueing benefit at S2 and S3. In line with our 143 expectations, there was no relationship between motor network plasticity and cueing benefit at S2 or 144 S3 (p > 0.05).

Together, these results provide evidence for gradual plasticity in the microstructure of precuneus, 147 striatum and sensorimotor cortex that underpins long-term behavioural effects of TMR. Interestingly, 148 both the early and late plasticity results for motor ROIs seem to be specifically related to S4, the time inter-individual variability in brain microstructure could confer susceptibility to the manipulation. For 165 this, we were not interested in the TMR effect at any particular session, but rather in the common 166 variance of the cueing benefit shared across the post-stimulation sessions (Fig.1b-c) . Thus, we 7 performed a Principal Component Analysis (PCA) on the cueing benefit at S2, S3 and S4 in order to 168 obtain a single measure which we will call TMR susceptibility . We used baseline (S1) maps of MD and

Fr as inputs, with TMR susceptibility entered as a regressor. This showed a relationship between 170 baseline microstructure in right precentral and postcentral gyrus (58, -6, 20) and TMR susceptibility (p 171 = 0.041; Fig.4a -c, Table S2 ; Fig.S4 

Indeed, TMR and memory reactivation per se share a lot of parallels with memory retrieval. However, 211 the long-term time scale of our current results as well as the microstructural changes that we report 212 suggest that the role of precuneus may extend beyond retrieval only. We believe that precuneus could 213 42 , building up representations of the retrieved information before they are 214 transferred to a more permanent store. Indeed, precuneus has already been shown to undergo rapid, uncued sequence that we observe 20 days post-TMR. Thus, we argue that (targeted) memory 223 reactivation during sleep has a powerful impact on both memory processing and brain plasticity, and 224 that its effects extend beyond the initial night of sleep.

Our findings support the suggestion that the long-term cueing benefit of TMR is mediated by early 

The results of the current study demonstrate that DW-MRI can provide a valuable tool to investigate 261 behaviourally relevant changes in brain microstructure. Furthermore, the multimodal approach that 262 we adopted here revealed a common pattern across two diffusion markers: MD and Fr. This not only 263 makes our findings more robust but also provides insights into the biological changes that could unable to complete the study, missing either one (n = 1) or two (n = 5) sessions. One additional 340 participant could not physically attend S3; they performed the SRTT online, but their MRI data could 341 not be collected and therefore the sample size for the MRI analyses of S3 had to be decreased by one.

Finally, Fr maps collected from three additional participants failed a visual quality check after pre-343 processing and were thus excluded from the Fr analysis (n = 3). Hence, the final sample size for MD 344 analysis was n = 23 for S1, n = 23 for S2 and n = 19 for S3, and the sample size for Fr analysis was n = 345 13 20 for S1, n = 20 for S2 and n = 16 for S3. A flowchart of participants included and excluded from the 346 different analyses is presented in Fig.S1 .

The study consisted of four sessions (Fig.5a) , all scheduled for the same time in the evening (~ 8 pm).

Upon arrival for the first session (S1), participants completed Pittsburgh Sleep Quality Index (PSQI) 68 350 to examine their sleep quality over the past month. S1 consisted of a motor sequence learning task 

Session 2 (S2), session 3 (S3) and session 4 (S4) took place 23-26 h, 10-14 days, and 16-21 days after 365 S1, respectively. During S2 and S3 DW-MRI data were acquired as before, followed by an SRTT re-test.

Here, the first half of the SRTT blocks (24 sequence blocks + 4 random blocks) was performed in the 367 3T scanner and the second half (24 sequence blocks + 4 random blocks) in the 0T scanner. Note that 368 the order of scans (3T vs 0T) was flipped from S1 to S2 and S3 for the microstructural assessment to 

For offline data collection, the SRTT (S1-S3) was back projected onto a projection screen situated at 

The SRTT was used to induce and measure motor sequence learning. It was adapted from 17 and 394 implemented exactly as described in 24 . Briefly, participants learned two 12-item sequences of 395 auditorily and visually cued key presses. The task was to respond to the stimuli as quickly and interleaved pseudo-randomly with no more than two blocks of the same sequence in a row.

Participants were aware that there were two sequences but were not asked to learn them explicitly.

Block order and sequence replayed were counterbalanced across participants.

During each run of the SRTT, sequence blocks A and B were followed by 4 random blocks, except for 421 the first half of S1 (to avoid interrupted learning). Random blocks were indica 422 centrally on the screen and contained pseudo-randomised sequences, the same visual stimuli, and 423 tones matching sequence A for half of them (Rand_A) and sequence B for the other half (Rand_B).

Blocks Rand_A and Rand_B were interleaved, and the random sequences contained within them 425 followed three constraints: (1) each cue was represented equally within a string of 12 items, (2) two 426 consecutive trials could not contain the same cue, (3) random sequence did not share a string of more 427 than four items with either sequence A or B. 

Signals were recorded using BrainVision Recorder software (Brain Products GmbH).

Tones associated with one of the learned sequences (A or B, counterbalanced across participants) 441 were replayed to the participants during N2 and N3, as assessed with standard AASM criteria 70 .

Volume was adjusted for each participant to make sure that the sounds did not wake them up. One 443 repetition of a sequence was followed by a 20 s break, with the inter-trial interval jittered between 444 2500 and 3500 ms. Upon arousal or leaving the relevant sleep stage, replay was paused immediately 445 and resumed only when stable N2/N3 was apparent. TMR was performed for as long as a minimum 446 threshold of ~1000 trials in N3 was reached. On average, 1552.91 ± 215.00 sounds were delivered.

The protocol was executed using MATLAB 2016b and Cogent 2000. 

(restricted) space 32 , thereby providing a more sensitive method to look at the microstructural changes 512 than DTI 28 . Fr is one of the outputs from the CHARMED framework. In grey matter, Fr changes are 513 thought to reflect remodelling of dendrites and glia, and were observed both short-term (2 h) and 514 long-term (1 week) following a spatial navigation task 28 .

Co-registration, spatial normalisation and smoothing of the MD and Fr maps were performed in 517 SPM12, running under MATLAB 2015a. First, we co-registered the pre-processed diffusion images with 518 structural images using a rigid body model. The co-registration output was then spatially 519 normalised to MNI space. This step involved resampling to 2 mm voxel with B-spline interpolation and 520 utilised T1 deformation fields generated during fMRI analysis of the same participants 24 . That way, 521 the resulting diffusion images were in the same space as the fMRI and T1w data. Finally, the 522 normalised data was smoothed with an 8 mm FWHM Gaussian kernel.

All behavioural tests conducted were two-tailed, and both positive and negative contrasts were 525 performed for the MRI analyses. MRI results were voxel-level corrected for multiple comparisons by 526 family wise error (FWE) correction for the whole-brain grey matter (GM) and for the pre-defined 527 anatomical regions of interest (ROI, see section 2.7.4.6), with the significance threshold set at pFWE < 528 0.05. To obtain a whole-brain GM mask, the SPM12 tissue probability map of GM was thresholded at 529 50% probability 78 . 

Group level analyses of DW-MRI data was performed in FSL (FMRIB's Software Library, 546 http://www.fmrib.ox.ac.uk/fsl) 81 . To examine the relationship between brain characteristics and our 547 variables of interest we performed non-parametric combination (NPC) for joint interference analysis 548 (Fig.5b) , as described before 82,83 . Specifically, NPC was performed over MD and Fr maps to uncover 549 common trends related to non-myelin GM microstructure 26,28 .

The analysis was performed through Permutation Analysis of Linear Models (PALM) in FSL 84 

The resultant images were entered into a one-sample t-test with cueing benefit at S2, S3 and S4 added 569 as regressors. We run separate contrasts for each session, whereby the session of interest was 570 specified as the main regressor whereas the remaining sessions were treated as the covariates of no 571 interest (nuisance covariates). This ensured that the results were specific to the session analysed.

Additionally, the nuisance covariates also included sex and age to control for the differences between 573 males and females, as well as the effect of age. Baseline reaction time (i.e., average reaction time on 574 the random blocks performed during S1) and baseline learning capabilities (i.e., difference between 575 the average of the last 4 blocks and the first 4 blocks performed during S1) were also specified as the 576 variables of no interest to ensure that the results were independent of baseline SRTT performance.

To determine whether individual differences in baseline brain characteristics can predict susceptibility 579 to the manipulation we tested the relationship between baseline (S1) GM microstructure and the 

To determine individual contribution of each microstructural modality to the multimodal results, we 589 performed unimodal analyses of individual modalities, in FSL and through non-parametric, 590 permutation-based voxel-wise comparisons using the randomise function 86 . Results were derived 591 from 5000 permutations. Correction for multiple comparisons was carried out by FWE correction both 592 for the whole brain GM and for the pre-defined ROIs, as for the multimodal analysis. Multiple 

The Impact of Studying Brain Plasticity

Plasticity in gray and white: Neuroimaging 635 changes in brain structure during learning

Changing Rules of Neuroplasticity: Implications for Learning and Recovery

Common molecular mechanisms in explicit and implicit 640 memory

System consolidation of memory during sleep

The organization of recent and remote memories

Memory engrams: Recalling the past and imagining the future

The Neurobiology of Consolidations, Or, How Stable is the Engram?

Finding the engram

Post-learning micro-and macro-structural neuroplasticity changes 654 with time and sleep

Odor Cues During Slow-Wave Sleep Prompt 656

Reactivating Them During Sleep. Science (80-. )

Gobel, 660 during sleep influences skill learning

Strengthening Procedural Memories by Reactivation in 662

Reactivation during Slow-Wave Sleep Promotes Explicit Knowledge of a Motor Sequence

Long term 670 effects of cueing procedural memory reactivation during NREM sleep

Memory Reactivation Reoccurs in a Slow-Oscillatory Rhythm during NREM Sleep

Identification of memory reactivation during sleep by EEG classification

Memory stabilization with targeted reactivation during human slow-678 wave sleep

Odor-evoked category 680 reactivation in human ventromedial prefrontal cortex during sleep promotes memory 681 consolidation

Cueing motor memory reactivation during NREM sleep engenders learning-684 related changes in precuneus and sensorimotor structures. bioRxiv (2022)

Principles of Diffusion Tensor Imaging and Its Applications to Basic 687

Learning in the Fast Lane: New Insights into Neuroplasticity

Matter Plasticity in the Fornix

Micro-structural assessment of short term plasticity 693 dynamics

following motor sequence learning revealed by diffusion magnetic resonance imaging

Fast track to the neocortex: A memory engram in the posterior parietal cortex

The sensitivity of diffusion MRI to microstructural properties and 700 experimental factors

New modeling and experimental 702 framework to characterize hindered and restricted water diffusion in brain white matter

Rehearsal initiates systems 705 memory consolidation

Functional MRI evidence for adult motor cortex plasticity during motor skill 707 learning

Cortical Synaptogenesis and Motor Map Reorganization Occur during Late

Skill Representation in the Primary Motor Cortex After 711

Long-Term Practice

Sensorimotor Cortex Maintain Procedural Long-Term Memory Storage. iScience 5

Emergence of a Stable Cortical Map for Neuroprosthetic 716

Promoting memory consolidation during sleep: 718 A meta-analysis of targeted memory reactivation

Parietal lobe contributions to episodic 720 memory retrieval

The hippocampus as the switchboard between perception and memory

Memory retrieval and the parietal cortex: A review of evidence 726 from a dual-process perspective

Behaviorally Relevant Abstract Object Identity Representation in the 728

Rapid and independent memory formation in the parietal cortex

A common reference frame for movement plans in the 732 posterior parietal cortex

Neural Correlates of Motor Memory Consolidation

Distinct contribution of the cortico-striatal and 736 cortico-cerebellar systems to motor skill learning

The Ventral Striatum in Off-Line Processing: Ensemble Reactivation during 738 Sleep and Modulation by Hippocampal Ripples

Coupling between motor cortex and striatum increases during sleep over 740 long-term skill learning

A sleep spindle framework for motor memory consolidation

Differential Corticostriatal Plasticity during Fast 744 and Slow Motor Skill Learning in Mice

Targeted Memory Reactivation during 26

Multiple memories can be simultaneously reactivated during sleep as 748 effectively as a single memory

Targeted Memory Reactivation for Consolidation in Sleep are Contingent on Memory 751 Accuracy and Direct Cue-Memory Associations

Activity-Dependent Structural and Functional 753 Plasticity of Astrocyte-Neuron Interactions

Diffusion MRI of Structural Brain 755 Plasticity Induced by a Learning and Memory Task

Rapid formation and selective stabilization of synapses for enduring motor 757 memories

Sleep promotes branch-specific formation of dendritic spines after learning

REM sleep selectively prunes and maintains new 761 synapses in development and learning

Intrinsic optical signals in the rat optic 763 nerve: Role for K+ uptake via NKCC1 and swelling of astrocytes

Diffusion, confusion and functional MRI

Motor learning induces astrocytic hypertrophy in the cerebellar cortex

The memory function of sleep

Primed to Sleep: The Dynamics of Synaptic Plasticity Across Brain 772

Enhanced Implicit Sequence Learning 774 in College-age Video Game Players and Musicians

Visual-spatial sequence learning and 776 memory in trained musicians

The Pittsburgh sleep 778 quality index: A new instrument for psychiatric practice and research

PsychoPy2: Experiments in behavior made easy

AASM Scoring Manual Version 2.2 Updates: New Chapters for Scoring Infant 783

Sleep Staging and Home Sleep Apnea Testing

Measuring compartmental <math altimg="si4.svg"> <msub> <mi>T</mi> tiltable RF coil and diffusion-<math altimg="si4.svg"> <msub> <mi>T</mi> 787 <mn>2</mn> </msub> </math>

Fast and accurate Slicewise OutLIer Detection 791 (SOLID) with informed model estimation for diffusion MRI data

Gibbs-ringing artifact removal based on local 794 subvoxel-shifts

Diffusion tractography in crossing fiber 796 regions

Can we correct for interactions 798 between subject motion and gradient-nonlinearity in diffusion MRI

Why diffusion tensor MRI does 801 well only some of the time: Variance and covariance of white matter tissue microstructure 802 attributes in the living human brain

The Impact of Lesion In-Painting and Registration Methods on Voxel-Based 804

Morphometry in Detecting Regional Cerebral Gray Matter Atrophy in Multiple Sclerosis

Fitting Linear Mixed-Effects Models Using lme4

Estimated marginal 809 means, aka least-squares means

Advances in functional and structural MR image analysis and 812 implementation as FSL

Reassessing associations between white matter and behaviour with 814 multimodal microstructural imaging

White matter structure and myelin-related gene expression 816 alterations with experience in adult rats

Non-parametric combination and related permutation tests for 818 neuroimaging

Evolving plasticity in brain and behaviour after targeted memory reactivation

Permutation 822 inference for the general linear model

Motor Memory Consolidation in Sleep Shapes More Effective Neuronal 824 Representations

Sleep-dependent motor 826 memory plasticity in the human brain

Both the Hippocampus and Striatum Are Involved in Consolidation of Motor 828

Brain plasticity related to the consolidation of motor sequence learning and 830 motor adaptation

Experience-dependent changes in cerebral activation during human REM 832 sleep

Memory processing 834 during human sleep as assessed by functional neuroimaging

An automated method for 837 neuroanatomic and cytoarchitectonic atlas-based interrogation of fMRI data sets

Elegant Graphics for Data Analysis