Poster session 2 (1:00-2:30pm, Day 2)

Background/Aims:
A growing body of research is concerned with modelling magneto- or electroencephalography (MEEG) responses to speech. This usually focuses on predicting the low-frequency time-domain portion of the signal in the delta (1 − 4 Hz) and theta (4 − 8 Hz) ranges from different stimulus features. Here, we instead focus on oscillatory power and aim to predict it with a stimulus-computable computational model.

Methods:
We re-analyse an MEG dataset of passive story listening. We obtain spatio-spectral response filters from a canonical correlation analysis (CCA) that maps MEG sensor power time courses onto the time-lagged envelope of the speech stimulus. We then predict the projections of the MEG responses through the spatio-spectral filters with features of a recurrent network that predicts the future of the speech envelope at multiple forecast horizons by parameterising it as two-component Gaussian mixture distributions.

Results:
The CCA weights suggest generators in the beta frequency range (13 − 30 Hz) in bilateral auditory cortices, which we confirm with source localisation. We then find that network outputs related to the variance but also the mean of the upcoming sound energy predict the MEG response component better than acoustic baseline models.

Discussion/conclusions/implications
Our model is only a suggestion of a computational process that could be underlying the generation of the biological beta power signal, and more work is needed to disambiguate the contributions from mean and variance outputs of our network. Based on its success in outperforming acoustic baseline models, we however conclude that our approach of ignoring textual linguistic annotations and instead considering a deceptively simplistic optimisation objective is promising in order to arrive at a stimulus-computable model of the listening brain.

Background/Aims: 
In real-life scenarios, individuals frequently engage in tasks that involve searching for one of various items stored in memory. This intertwined cognitive operation, referred to as hybrid search, is essential for activities like driving following reference landmarks. While behavioural aspects of hybrid search have received extensive attention, in this study our focus is on delving into the underlying neural mechanisms.

Methods: 
We recorded concurrent magnetoencephalography (MEG) and eye tracking recordings while participants engaged in a visual and memory search task involving items embedded within naturalistic photographs.

Results: 
By applying a deconvolution analysis approach to disentangle brain activity from eye movements artifacts, we found a robust marker that could differentiate between targets and distractors. We also found strong decrease in frequency power in the alpha and beta bands, which was associated with increased memory load.

Discussion/conclusions/implications: 
This study casts a spotlight on the pivotal role of alpha oscillations in both memory retention and visual search processes. Beyond this foundational insight, our approach offers a gateway to investigating neural processes in contexts mirroring real-world situations, such as the cognitive demands associated with driving.

Background/Aims: 
Many tasks require the skilled interaction of both hands, such as eating with a knife and fork or keyboard typing. However, our understanding of the behavioural and neurophysiological mechanisms underpinning bimanual motor learning is still sparse. Here, we aimed to address this by characterising learning-related changes of different levels of bimanual interaction.

Methods: 
To explore early bimanual motor learning, we designed a bimanual motor learning task. In the task, a force grip device held in each hand (controlling x- and y-axis separately) was used to move a cursor along a path of streets at different angles (0°, 22.5°, 45°, 67.5°, and 90°). Each street corresponded to specific force ratios between hands, which resulted in different levels of hand interaction, i.e., unimanual (Uni, i.e., 0°, 90°), bimanual with equal force (Bieq, 45°), and bimanual with unequal force (Biuneq 22.5°, 67.5°). 42 healthy participants performed the task for 100 trials, whereby each trial comprises six streets.

Results: 
On the behavioural level, the three conditions differed in their movement time and error. Further, we found that the novel task-induced improvements in movement time and error, with no trade-off between movement time and error, and with distinct patterns for the three levels of bimanual interaction. On the neural level, beta event-related desynchronisation and synchronisation show different spatial topographies for the unimanual conditions. Moreover, we found that learning-related changes are manifold and are explored in detail using the amplitude-envelope variant of the Hidden Markov Model.

Discussion/conclusions/implications:
Overall, this complex bimanual motor task allows us to characterise bimanual motor learning with different levels of bimanual interaction. This should pave the way for future neuroimaging studies to further investigate the underlying mechanism of bimanual motor learning.

Aims:
To understand prediction-for-perception, studies should address where, when, and how the brain predicts the stimulus features that change behavior. Typical multivariate classifiers are trained to contrast the bottom-up patterns of neural activity between two stimulus categories. These classifiers then quantify predictions as reactivations of the category contrast but cannot capture the features reactivated for each category. We addressed these predicted features.

Method:
In a prediction experiment, we used Dali’s ambiguous painting–Slave Market with the Disappearing Bust of Voltaire which contains two possible perceptions (Nuns vs. the bust of Voltaire). In each of 3150 trials, 10 participants were cued to each perception, followed by deciding whether the stimulus was “Volaire” or “Nuns.” Stimuli comprised random samples (Gabor) of the features leading to each perception. Using each participant’s reconstructed MEG on 8196 sources, we (1) decoded the category-contrast prediction (Voltaire vs. Nuns) using classifiers trained to discriminate Nuns vs. Voltaire images in a localizer run prior to the experiment; (2) decoded the category-feature predictions (separately for Voltaire and Nuns predicted features) with classifiers trained to discriminate High (> 70%) from Low (< 30%) proportions of Voltaire and Nuns features in uncued trials; (3) studied how trial-by-trial reactivation of category-feature representations in the brain changes subsequent behavior.

Result:
We show that top-down reactivations of the category-features (vs. category-contrast) are more precisely localized (i.e. lateralize to left occipital cortex for LSF Voltaire and right LOC for HSF Nuns), with per-trial reactivation strength that more strongly biases participant’s perceptual behavior across a wider range of stimulus evidence.

Discussion:
Our novel approach can therefore track the reactivation of category-specific visual contents to better relate mechanisms of prediction to behavior.

Background/Aims:
Apathy is a pervasive symptom across a spectrum of neurological conditions. There are currently no approved treatments, despite negative links with caregiver burden, quality of life and survival. In this ongoing project, we propose and test a new model of apathy focused on a reduction in prior precision of action outcomes. We explore this new framework using psychophysical analysis of performance and expectations in a goal directed task, in combination with MEG and Bayesian modelling, to identify the neural mechanisms underpinning apathy.

Methods:
N=20 participants with no history of psychological or neurodegenerative condition completed a goal directed task in the MEG scanner. Our primary task involves participants landing a virtual ball on a target. In a subset of trials, the ball disappears, and participants estimate the end position. From performance and estimation errors, we can infer the priors on outcomes. Participants with more precise prior beliefs are systematically biased towards the target in their estimates, and their prior precision is inversely proportional to trait apathy (Hezemans et al, 2020) as measured by the Apathy Motivation Index.

Hypotheses:
Our primary (confirmatory) hypothesis is that there is a negative correlation between trait apathy and prior precision. Our co-primary (exploratory) hypothesis is that including prior precision as an empirical prior for physiological response can explain activity in the prefrontal and motor cortex as measured by MEG. We will test this formally using dynamic causal modelling, employing Parametric Empirical Bayes to compare models with and without prior precision. Additional exploratory hypotheses will be specified in an online preregistration.

Conclusion:
In summary, the aim of this ongoing project is to test the neural underpinnings of apathy in healthy adults, with the goal of laying groundwork for new experimental medicine studies in people with neurological disorders.

Background:
Simulations have shown that MEG is capable of localising brain signals with laminar precision, given high enough levels of data quality, and these levels are achievable with high precision, head-cast MEG. Previous laminar source reconstruction efforts used two surfaces, representing deep and superficial layers. One of these approaches used a sliding time window model comparison for temporally resolved laminar inference, limiting its applicability across experimental conditions, and limiting inference to whether activity is strongest in deep or superficial layers. Here we aimed to provide a more complete picture of neural dynamics across cortical layers.

Methods:
A depth electrode-like source space was created by generating 11 equidistant layers between the white matter and pial surfaces, with vertices matched across layers. We then applied the Empirical Bayesian Beamformer to visual and motor ERFs from high precision, head-cast MEG data. This yielded, for each pial surface vertex, 11 source time series at different laminar depths. We then applied two analyses to these series: a current source density (CSD) transform, revealing temporally dynamic current sources and sinks, and a relative power analysis, thought to be a marker of the boundary between deep and superficial layers.

Results:
The boundaries of current source and sink patterns tightly corresponded to the estimated thickness of each cortical layer as estimated from the BigBrain histological atlas. Moreover, alpha/beta power was strongest in deep layers and gamma in superficial layers, and the crossover point of relative alpha/beta and gamma power was correlated with the estimated depth of layer IV from the BigBrain atlas.

Discussion:
Multilaminar source localisation with high precision MEG is possible and extremely promising. Contrary to sparse sampling of intracranial electrodes, laminar MEG has the potential to non-invasively and globally test hypotheses about cortical layer dynamics.

Background:
Analysis of neural activity in various frequency bands is ubiquitous in systems and cognitive neuroscience. Recent analytical breakthroughs and theoretical developments rely on phase maintenance of oscillatory signals or a clean separation in power between aperiodic and periodic activity, without considering whether or not such assumptions are met. Lagged coherence, the coherence between a signal and itself at increasing temporal delays, has been proposed as a way to quantify the rhythmicity, or periodicity, of a signal. However, current lagged coherence algorithms suffer from poor spectral accuracy and resolution, aliasing effects that become more pronounced at higher frequencies, and conflation with amplitude covariation, especially in frequency ranges in which the signal power is low.

Methods:
We introduce a continuous lagged coherence metric, lagged Hilbert coherence, that addresses these shortcomings by using multiplication in the frequency domain for precise bandpass filtering, instantaneous analytic signals via the Hilbert transform, and thresholding using the amplitude covariation of surrogate data generated by an autoregressive model.

Results:
We show that this version of lagged coherence yields vastly higher spectral accuracy and resolution that previous versions, and demonstrate how it can be used to 1) examine the relationship between mean frequency-specific rhythmicity and response time, 2) improve parameterization of the aperiodic and periodic components of power spectral densities, and 3) detect the occurrence of transient oscillatory bursts.

Implications:
Lagged Hilbert coherence thus offers a significant toolset advancement for complex neurophysiological spectral analysis.

Background:
Forming predictions based on statistical stimulus regularities is essential for adaptive behaviour. Such regularities pertain not only to stimulus contents (“what”) but also their timing (“when”), and both can interactively modulate sensory processing. In speech streams, predictions can be formed at multiple hierarchical levels, both in terms of contents (e.g. single syllables vs. words) and timing (e.g., faster vs. slower time scales). It is unknown if the brain integrates these predictions in a hierarchically specific way (e.g., faster “when” predictions selectively modulating “what” predictions of single syllables), and if prediction integration at different hierarchical levels relies on dissociable neural correlates.

Methods:
We manipulated “what” and “when” predictions at two levels – single syllables and disyllabic artificial words – while neural activity was recorded using magnetoencephalography (MEG) in healthy volunteers (N=22). We analysed event-related fields evoked by syllable and/or word deviants, focusing on their modulation by “when” predictability. We used source reconstruction and dynamic causal modelling to explain the observed effects in terms of the underlying effective connectivity.

Results:
“When” predictions modulated “what” mismatch responses in a hierarchically specific way. However, these modulations were shared across hierarchical levels in terms of the spatiotemporal distribution of MEG signals. Effective connectivity analysis showed that the integration of “what” and “when” predictions selectively increased connectivity at relatively late processing stages, between the superior temporal gyrus and the fronto-parietal network.

Discussion:
These results suggest that the brain integrates different predictions with a high degree of mutual congruence, but in a shared and distributed cortical network. This contrasts with recent studies indicating separable networks for different levels of hierarchical speech processing.

Background:
Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disorder of the motor neuronal system spanning cortex to muscle. Biomarkers are needed to serve as outcome measures in therapeutic trials. We examined corticomuscular coherence (CMC) via MEG, as a more holistic motor system biomarker of disrupted neural dynamics in ALS.

Methods:
In an ongoing study, data from 25 ALS patients and 30 healthy age-matched controls (HC) were analysed. Participants completed 120 bilateral and 60 unilateral trials of a gripper task during MEG. CMC was calculated for each trial type, and group comparison via cluster-based permutation testing.

Results:
During bilateral contraction, ALS patients showed significantly reduced beta-band CMC in the motor cortex. A reduction of CMC between cortex and contralateral muscle was also evident when considering its topographical distribution. In both groups, coherence was localised to the same contralateral motor channels, but was considerably weaker in ALS. No significant differences were found in grip strength, ipsilateral CMC, mean beta power, interhemispheric coherence, or right-hand task CMC. However, the ALS left-hand task showed a significantly reduced contralateral beta-band CMC, with localised weaker coherence in the motor channels. This was independent of handedness, site and side of symptom onset or grip strength.

Discussion:
Independently of grip strength, a localised CMC reduction is associated with ALS, potentially with unexpected pathological lateralisation. Beta-band CMC was easily acquired in physically disabled individuals and is an attractive potential biomarker for the assessment of therapeutic potential in drug screening studies.

Background/Aims:
Remifentanil and midazolam are well-known sedative drugs used in anaesthesia. The former is a short-acting opioid analgesic drug, whereas the latter is part of the family of benzodiazepines. Auditory mismatch negativity (MMN) is the component of the event-related potential (ERP) induced by deviant auditory stimuli occurring in a sequence of regular stimuli. The principal theory suggests that MMN reflects auditory sensory memory and is an index of pre-attentive information processing. Here, for the first time, we measured the MMN using magnetoencephalography (MEG) while remifentanil or midazolam was administered to participants to reveal drug specific effects on network connectivity within the brain.

Methods:
Participants were 18 fit and healthy male adults between 18-43 years old. A passive auditory oddball task was used to elicit MMN during two MEG sessions. In each session, participants were presented with standard and deviant tones, before and after mild sedation using one of the two sedative drugs. Sensor and source level connectivity analysis will be conducted, to investigate differences in functional connectivity during the MMN task in contrast to the two drugs, as well as before and after sedation. Given the different mechanisms of action of the two drugs, we anticipate they will have differential modulatory profiles on the task-relevant networks involved in the MMN. Benzodiazepines have been shown to reduce the MMN oddball response, whilst evidence for remifentanil is mixed, with some evidence for an increased response.

Discussion/implications:
Knowing more about how the two sedative drugs can influence processing of auditory stimuli during the MMN task could offer important clinical insights and a better understanding of whether MMN can be used as a marker for sedation level, under different sedative agents, via their modulation of task-based functional connectivity.

Alpha oscillations have been demonstrated to link to attention in many attention-demanding tasks (e.g. Jensen & Mazaheri, 2010). However, the majority of studies have been conducted in neurotypical adults. Alternatively, children with attention difficulties, including attention-deficit / hyperactivity disorder (ADHD), are often studied during the resting-state and/or with EEG with limited spatial resolution. The overall goal of this study is to uncover the neural sources relating to control of attention in children (age 8-11), those both with and without ADHD. By using MEG, this study aims to localise sources relevant in an attention-demanding working memory task (e.g. Lenartowicz et al., 2014) as well as neural connectivity during the same task, and link this to ‘standard’ measures used in clinical ADHD research, such as resting-state connectivity, theta-beta ratio, performance on a clinically valid Continuous Performance Task (CPT) and questionnaire scales. This poster will present preliminary data. Future work within this project will aim to link functional connectivity (from MEG) to structural connectivity (DTI) and assess test-retest reliability.

A common assumption in Cognitive Neuroscience is that brain rhythms vary with task-specific cognitive demands and reflect the neural support of the cognitive operation performed. Power modulations of alpha oscillations are consistently related to working memory (WM). However, there is an inconsistency in the literature regarding the direction of this association between al-pha power and WM load: some findings suggest an increase while others suggest a decrease.
To shed light on this topic, a pilot study (N=10) was conducted using the Sternberg and N-back task. The study aimed to explore whether different gaze patterns during these tasks could predict variations in alpha power. The preliminary findings revealed that the relationship between alpha power and WM load varied depending on the variability of participants’ oculomotor actions.
Specifically, when participants exhibited higher variability in their eye movements, there was a stronger decrease in alpha power with increasing WM load and lower variability was associated with a weaker decrease. These preliminary results suggest that variations in alpha power with WM load, and power modulations more generally, may primarily serve to support oculomotor control, rather than solely reflecting the cognitive demands of the task.
Furthermore, the study identified several other noteworthy preliminary conclusions. There was a close relationship between gaze patterns and alpha power modulation during WM, indicating that eye movements and alpha oscillations are intricately linked. Higher WM load was associat-ed with a stronger reduction in alpha power and increased gaze variability. The modulation of posterior alpha power with WM load displayed significant inter-individual variability.
These findings will be used as the basis for a larger study (N=90) which will provide valuable insights into the role of oculomotor actions in the dynamics of alpha power during working memory tasks.

Background:
Amyotrophic lateral sclerosis (ALS) is a fatal adult-onset neurodegenerative disorder of the motor system characterised by progressive muscle weakness. It involves widespread cerebral extra-motor as well as motor pathology, including cortical hyperexcitability with paired-pulse transcranial magnetic stimulation. Biomarkers are needed to provide trial outcome measures that are more sensitive than disability or survival, against which to screen potentially therapeutic drugs.

Methods:
Ten minutes of resting state MEG were recorded in ALS (n=36) and healthy controls (n=51), followed by a structural MRI scan for co-registration. Extracted metrics from 52 regions and 6 frequency bands (δ, θ, α, β, γ, high-γ) included static power, amplitude envelope correlation (connectivity), 1/f exponent and Higuchi fractal dimension (complexity), which were entered into a permutations-based general linear model with correction for multiple comparisons.

Results:
The ALS group showed lower cortical sensorimotor β and higher high-γ power. Greater disability was associated with increased δ, θ and high-γ global connectivity, increased fractal dimension and a lower 1/f exponent. Increases in temporal connectivity were driven by intra-hemispheric hyperconnectivity, whereas frontal and occipital connectivity increases were driven by global hyperconnectivity.

Discussion:
Resting state MEG identified key elements of a cortical neurophysiological signature for ALS. The combined findings of reduced β power, increased γ power and increased complexity metrics are compatible with the existing hypothesis that the loss of inhibitory GABAergic interneurons is a key feature of pathogenesis. Increased connectivity may represent compensatory responses to a failing motor system. MEG has potential to provide early sub-clinical biomarkers of therapeutic benefit in ALS.

Background:
Investigating natural vision in settings where participants can saccades is becoming increasingly important. Here, we investigate the categorization of visual objects during natural viewing, characterized by frequent eye movements. Human saccades occur every ~250 ms, leaving only ~150 ms for foveal processing and planning the next saccade. The extent to which a parafoveal object is processed before we saccade to it remains unknown. Serial processing models posit that parafoveal processing occurs only when the foveated object has been processed, while parallel processing proposes that foveal and parafoveal objects are processed simultaneously. These two mechanisms predict varying degrees of processing of parafoveal previewing before saccade onset. Using Multivariate Pattern Analysis (MVPA), we investigate in which detail a parafoveal object is processed.

Methods:
We used a free-viewing paradigm with naturalistic equidistant images, from different categories (animal, food, object) there were displayed in greyscale or in colors. After a mask, the images were presented once more, except one that has been replaced and had to be identified. 36 participants performed the task while eye movements and brain activity were acquired respectively with an eye tracker and magnetoencephalography (MEG). MVPA was conducted to classify feature (greyscale vs color) and semantic (category) characteristics of objects in the fovea and parafovea.

Results:
Preliminary results suggest that feature and semantic information can be extracted at the fovea during natural vision. We are currently exploring the decoding of parafoveal information. The time courses and the source-level analysis of the classification results will allow the reconstruction of the processing of foveal and parafoveal objects along the visual hierarchy.

Conclusion:
In sum, our study will provide a stronger neuroscientific understanding of human’s ability to explore visual scenes efficiently in daily life.

Background/Aims:
Psychedelic drugs have shown great promise for treating a number of psychiatric conditions, such as depression and anxiety, but the underlying brain mechanisms that subserve their therapeutic potential are still unknown. In this study, we apply a biophysical Hopf bifurcation model of MEG oscillations (Cabral et al., 2022) to shed light on the network properties that drive previously-observed changes in neural synchronization on psychedelics (Muthukumaraswamy et al., 2013; Carhart-Harris et al., 2016).

Methods:
We analysed an existing MEG dataset in which the psychedelic LSD was administered to 17 healthy participants (Carhart-Harris et al., 2016). In particular, we determined the values of the free parameters (global coupling strength and mean conduction delay) that optimised the fit between the Hopf bifurcation model and the data, for both the placebo and LSD conditions. We also computed the metastable oscillatory modes (MOMs) – clusters of strongly synchronised regions – at this optimal point in parameter space and measured the differences in summary metrics, such as duration, size, and fractional occupancy, between the placebo and LSD conditions.

Results:
We detected significant differences in MOM properties between conditions. That is, LSD strongly reduced the duration, size, and fractional occupancy of MOMs in all the measured frequency bands (theta, alpha, beta, gamma). These changes could be explained by changes in global model parameters, namely an increase in the mean conduction delay and coupling strength.

Discussion:
The results not only align with but also explain previous findings of broadband desynchronisation in the acute psychedelic state, as observed in MEG. They suggest that psychedelics may exert their characteristic effects on human brain activity by slowing down connectivity and diminishing the influence of local synchronisation.

Background:
Autism spectrum disorders (ASD) are neurodevelopmental conditions characterized by difficulties in social communication and interaction, restricted interests, and repetitive behaviours. Sensory and motor difficulties are also common in the majority of children with ASD, affecting more than 80% of individuals. However, our understanding of the neural mechanisms associated with sensorimotor processing in children with ASD is limited.

Methods:
We recruited 18 children diagnosed with ASD (mean age = 6.00 years, SD = 0.59, 5 females, 13 males) and 19 typically developing (TD) children matched based on age and IQ (mean age = 5.71 years, SD = 0.46, 4 females, 15 males). We designed a child-friendly video game-like motor task, where participants had to press a button in response to a visual target while we recorded their brain activity using a child-customized MEG system.

Results:
We observed significant gamma power increases at 70 to 90 Hz and 0 to 100 ms period following the button response onset in the primary motor cortex (M1) and gamma power increases at 50 to 60 Hz and 150 to 450 ms period following the visual target onset in the bilateral cuneus in both TD and ASD groups. We identified statistically significant differences in motor-related gamma power in the right M1 (t = 2.412, p = 0.021), but not in the left M1, but not in visual gamma power within the bilateral cuneus between the two groups. Furthermore, we conducted correlation analyses to investigate the relationship between visual and motor gamma power increases. Within the TD group, we discovered significant negative correlations between visual and motor gamma power specifically within the left hemisphere (ρ = -0.553, p = 0.014). However, such correlations were not observed within the ASD group.

Discussion:
These findings might provide compelling evidence for distinct neural mechanisms underlying varied patterns of visuomotor processing in individuals with ASD.

The post-movement beta rebound, as measured with MEG, has neuroscientific and clinical importance. This is the most widely studied example of a “post-task response” (PTR), i.e., a response that occurs in between periods of task and rest. We recently reported PTRs in MEG following cessation of working memory processes, using an n-back task. These responses occurred across the cortex in theta, alpha and beta bands, scaled with working memory load, and left lateral visual alpha PTR correlated with reaction times.
This study aims to determine whether PTRs following higher cognitive processes are driven by the same underlying phenomenon as the post-movement beta rebound. This was addressed by performing a burst analysis using a hidden Markov model (HMM), a technique shown to be effective in characterising bursts. Using a univariate HMM, we compare PTRs in the n-back dataset with a visuomotor dataset. Using k-means clustering of states across 78 AAL regions, we identify a PTR state in both tasks from the average HMM state probability timecourse. Binary HMM timecourses were visually compared to single-trial time-frequency responses to verify that PTRs are driven by transient bursts. Bursts were characterised in terms of burst duration and spectral content, comparing region-wise variation of the PTR state across tasks. Results show that both burst duration and spectral content in the PTR state show remarkable similarity across tasks, with alpha and beta content of bursts across brain regions correlating strongly between tasks (R^2 = 0.89, R^2 = 0.53, for alpha and beta, respectively), as well as burst durations (R^2 = 0.56). Burst durations had a mean of 310 ms and 320 ms for n-back and visuomotor tasks.
These results suggest that PTRs induced by different cognitive processes may be driven by the same underlying neural phenomenon, which, based on previous evidence, may serve a self-stabilising inhibitory function to bring active networks back to rest following task cessation.

Background:
Aging is a significant risk factor for many neuropsychiatric disorders. To properly assess the impact of these conditions against controls, a comprehensive understanding of healthy aging is necessary. While the effects of healthy aging have been widely investigated in the resting-state networks (RSNs) of EEG and MEG, a formal comparison of these modalities in capturing such effects remains lacking.

Methods:
In this study, we qualitatively compared M/EEG-driven static and dynamic brain network features to characterise how each modality represents age-related neural differences. We used openly available EEG LEMON and MEG CamCAN datasets to compute power spectra, power spatial maps, and functional connectivity (FC) of the whole-brain RSNs from 86 young (20-35 years) and 29 old (55-80 years) participants.

Results:
Our findings indicate that MEG outperforms EEG in revealing static and dynamic differences between age groups. While our analysis of static power spectra unveiled comparable frequency ranges with age effects in MEG and EEG, only MEG demonstrated spatially localised age effects in source space. Furthermore, when examining dynamic network features in source space, MEG exhibited a greater number of network states with between-group power and FC differences compared to EEG. Nonetheless, our results do not suggest dismissing EEG, as it identified spectral and spatial age effects that do not overlap with those of MEG, implying the potential presence of distinct but complementary information within each modality.

Conclusion:
Our study, therefore, proposes that the distinction between EEG and MEG should be carefully considered when interpreting the results of aging studies while recognising the complementary potentials of these modalities. Future studies combining the two will be instrumental in identifying how aging influences changes in healthy and diseased brains, leading to a more concrete picture of neuropsychiatric disorders associated with aging.

Background:
Multiple sclerosis (MS), a neurodegenerative disease characterized by inhibitory and excitatory synaptic loss, imposes working memory (WM) impairment. The 1/f slope has recently been hypothesized to provide an accessible marker of E/I ratio. In this study, besides the well-known oscillatory alpha suppression during WM tasks, we explored the task-induced variations in non-oscillatory component particularly the 1/f slope (indicating the steepness of the 1/f power-law component). We also hypothesized that healthy subjects (HC) have a higher level of inhibition (steeper 1/f slope) after distractor stimuli compared to MS patients.

Methods:
MEG data were recorded from 38 HC and 60 MS patients during a visual-verbal n-Back task which includes 0, 1 and 2-back conditions. Data were preprocessed using the OSL library and source reconstructed using an LCMV beamformer and then parceled into 42 parcels. We used the FOOOF algorithm to estimate the 1/f exponent and correct the power spectra by subtracting the non-oscillatory component. We used non-parametric statistics to compare the 1/f slope and periodic alpha power over the whole brain and then re-done at the parcel level before analysing the spatial structure.

Results:
Besides the alpha suppression, as expected we observed a steeper 1/f slope after both target and distractor stimuli onset. In line with our hypothesis, the 1/f slope was steeper after the distractor stimuli in HC as compared to MS patients, suggesting a higher level of inhibition of distractor stimuli in HC subjects which is important for optimal WM performance in all three conditions (p(0-back)=0.043, p(1-back)=0.02, p(2-back)=0.043). We also observed a significant correlation between 1/f slope changes and visuospatial working memory performance measured by the Brief Visuospatial Memory Test.

Conclusion:
Our results suggest that the 1/f slope variations may serve as a potential biomarker to monitor the WM performance during visuospatial WM task.

Dyspraxia/Developmental coordination disorder (DCD) is characterised by an impairment in the acquisition and performance of motor skills. End-state comfort research suggests the motor deficit is related to planning of movement sequences; but it is unclear what component of planning is affected. Neurophysiology and behavioural findings in controls show that movements in a sequence are pre-ordered in parallel, and the strength of pre-ordering, known as competitive queuing (CQ), predicts the accuracy of performance. This is accompanied by movement-related beta desynchronisation (MRBD) during planning and execution. This study aimed to build on our behavioural results, which found a reduced CQ of movements during planning, to examine the neural mechanisms of sequence planning in adults with DCD. Participants who took part in the preceding behavioural learning study were reinvited to perform finger sequences from memory in the Magnetoencephalography (MEG) scanner after a refresher of the delayed sequence production task. We used multivariate linear discriminant analysis of whole-head MEG activity patterns associated with the execution of each finger press to quantify the relative pattern probability of each press position during planning. Based on the behavioural data it is expected that the DCD group will show a weaker neural CQ gradient, i.e. reduced pre-ordering of press-related patterns during planning. Additionally, we hypothesise that the DCD group will have a higher beta power during baseline and a less pronounced MRBD compared to controls, particularly during planning, based on results in motor-impaired stroke and Parkinson’s disease patients. Critically, the study aims to tease apart effects related to performance from group, by looking at performance-matched trial analysis. This will promote understanding of the neural basis of disorderly motor planning in DCD and prepare interventions that target the neural organisation of memory-guided sequential movements.

Due to the high sensitivity of the sensors, magnetoencephalography (MEG) data are susceptible to noise, which can severely corrupt the data quality. Consequently, quality control (QC) of such data is an important step for valid and reproducible science (Niso et al., 2022). However, the visual detection and annotation of artifacts in MEG data requires expertise, is a tedious and time extensive task and is hardly standardized. Since quality control is commonly done in an idiosyncratic fashion it might also be subject to individual biases. Despite the minimization of human biases, standardization of QC routines will additionally enable comparisons across datasets and acquisition sites. Hence, an automated and standardized approach to QC is desirable for the quality assessment of in-house and shared datasets. Therefore, we developed a software tool for automated and standardized quality control of MEG recordings: MEGqc. It is inspired by a software for quality control in the domain of fMRI, called mriqc (Esteban et al., 2017). MEGqc strives to support researchers to standardize and speed up their quality control workflow and is designed to be easy and intuitive to use, e.g. only minimal user input (path to the dataset) is required. Therefore, the tool is tailored to the established BIDS standard (Gorgolewski et al., 2016; Niso et al., 2018). Among other metrics we detect noise frequencies in the Power Spectral Density and calculate their relative power, calculate several metrics to describe the ‘noisiness’ of channels and/or epochs, e.g. STD or peak-to-peak amplitudes, and quantify EOG and ECG related noise averaged over all channels and on a per-channel basis. MEGqc generates BIDS compliant html reports for interactive visualization of the data quality metrics and moreover provides machine interoperable JSON outputs, which allow for the integration into automated workflows. MEGqc is open source, can be found on Github, and its documentation is hosted on readthedocs.

Background/Aim:
Recent studies support the central role of synaptic loss leading to excitation/inhibition (E/I) imbalance, which prompts subclinical epileptiform activity as the primary initiator of Alzheimer’s disease (AD), and consequently neural network dysfunction. The classic brain criticality hypothesis posits that neuronal systems operate in the vicinity of continuous phase transition, regulated by E/I balance. This provides brain with optimal dynamic range, which is essential to healthy cognition and behaviour. However, due to positive feedback ―a slow parameter in addition to E/I― neurons show bistable activity, demonstrating discontinuous phase transition. It is suggested that moderate and elevated degree of bistability in ongoing neuronal oscillations were predictive of cognitive performance in healthy adults and neuropathology in epilepsy and geriatric subjects, respectively. Taken that since there is growing evidence of presence of subclinical epileptiform activity in patients of AD that can hasten cognitive decline, we aim to characterize such events with bistability analysis and provide mechanistic understanding of disease progression.

Methods:
We analysed resting-state MEG data recorded from 85 preclinical (SCD: Subjective Cognitive Decline), 142 prodromal (MCI: Mild Cognitive Impairment), 14 AD patients, and 116 healthy controls (HC). MNE estimated sources were collapsed into 400 parcels with a fidelity-optimized operator. Parcel broadband data were then filtered with 32 wavelets within a range of 2−90 Hz and bistability (BiS) indices were estimated.

Results:
We found aberrant BiS for SCD, MCI and AD compared to HC over the spectrum of 7–40 Hz. Importantly, BiS differentiated early disease stages and had frequency specific between-cohort differences.

Conclusion:
The results suggest BiS already alters at the early stages of AD and progressively alters with disease progression, and potentially be utilized as a biomarker for AD diagnosis and prognosis.

Background & Aims
Inter-areal coupling of neuronal oscillations is essential for regulation of neuronal processing and communication. Oscillatory activity in the brain is affected by neuromodulatory systems which exhibit regional specificity in afferent connections, receptor distributions, and neurotransmitter reuptake regulation. Here, we set out to investigate how frequency-specific coupling of neuronal oscillations covaries with the spatial distributions of NT receptors and transporters in the human cortex.

Methods
We computed phase-and amplitude-coupling in source-reconstructed human magnetoencephalography (MEG) data from 67 healthy subjects in frequencies from 1 to 96 Hz and estimated the covariance of local coupling strength with the density of 19 NT receptors and transporters across 200 cortical parcels. We further used principal component analysis (PCA) to identify common structures shared between these receptor and transporter density maps.

Results
Local strengths of large-scale phase and amplitude coupling strongly covaried with receptor and transporter densities across brain areas in a frequency-specific manner. Specifically, we found that dopaminergic, GABA, NMDA, muscarinic, and most serotonergic densities were positively correlated with local strength of both phase- and amplitude coupling in delta and gamma bands, and negatively in high-alpha and beta bands. In contrast, in theta and low-alpha bands, phase and amplitude node strengths showed more distinct coviance with receptor densities. PCA revealed several distinct anatomical patterns underlying the distribution of receptor and transporter densities, which also covaried with coupling strength.

Implications
Our findings indicate that oscillatory activity and coupling between neuronal oscillations are likely influenced by fundamental neuroarchitectionical principles underlying the distribution of NT receptors and transporters.

Objective
Proper name anomia is a common language deficit observed in people with dementia (PWD), impacting their ability to recall and retrieve the names of familiar people. We investigated the neural changes associated with a 6-week, app-based, proper-name anomia therapy in PWD using MEG.

Methods
14 PWD with proper name anomia were recruited. Following baseline assessment, patients underwent a structured 6-week proper-name anomia therapy program using a novel app called Gotcha! Participants were trained to name 6-10 familiar faces (usually their relatives and close friends), using confrontation naming and audio cueing methods. MEG recordings were obtained at two time points: pre- and post-therapy while PWD were presented with pictures of the familiar or untrained famous faces, which they named aloud. MEG data were analysed in SPM. We measured source localised gamma-band (30-80 Hz) power 0-1000 ms after the onset of a face. We ran a 2×2 factorial analysis (familiar/famous; pre-/post-therapy) on our source images using a repeated-measures ANOVA to look for changes in power across conditions. The behavioural data was analysed using a repeated-measures ANOVA with named faces during free-naming as the dependent variable.

Results
Behavioural data analysis revealed that a significant effect of training (post>pre), F(1,14)=8.79, p=0.01. For the MEG analysis, we identified a large cluster situated in the left ventral temporal lobe (MNI: -50 -28 -26, F=9.19, p=0.004, k=813) where gamma reduction was associated with training (pre>post) of familiar faces, but not (untrained) famous faces.

Discussion
This is the first study to demonstrate that the left ventral temporal lobe supports re-learning for familiar face-name associations in PWD. App-based proper name anomia retraining appears to be an effective therapy for PWD. Initial MEG findings suggest that therapy effects are manifest in areas associated with face-naming.

Abstract: Chronic cluster headaches (CCH) are debilitating conditions often accompanied by mood disorders such as depression. Deep Brain Stimulation (DBS) of the Ventral Tegmental Area (VTA) is an experimental treatment for this group of disorders and provides a unique opportunity to study VTA activity and connectivity using simultaneous MEG and intracranial recordings. In the present study we used a reinforcement learning task to explore VTA responses to reward and loss and their possible correlation with depressive symptoms.

Methods: Fourteen patients with VTA-DBS electrodes for CCH treatment were scanned using MEG. A paradigm was designed, incorporating three outcome types (win, loss, neutral) associated with unique fractal images. Hierarchical behavioural modelling code was developed to capture the dynamics of decision-making, prediction error, and learning. We examined VTA evoked responses to the presentation of trial outcome.

Results: A clear response to the outcome event was detected peaking around 200ms. The early part of this response was correlated with reward prediction error derived from behavioural modelling.

Conclusions and future work: The investigation of reward prediction errors in the VTA is critical, as it could reveal potential therapeutic DBS targets for depression and mood disorders comorbid with CCH. Understanding the underlying neural mechanisms is vital in advancing treatments for both cluster headaches and general mood disorders like depression. We will next look for correlation between the magnitude or VTA response to reward and depression severity. We would expect the correlation to be negative in line with the suggestion that depression is associated with decreased reward sensitivity. We will also look for MEG sources that could be upstream or downstream relative to VTA in the reward processing sequence.

Predicting an individual’s cognitive traits or clinical condition using brain signals is a central goal in modern neuroscience. This is commonly done using either structural aspects, or aggregated measures of brain activity that average over time. But these approaches are missing the unique ways in which brain activity unfolds over time. The reason why these dynamic patterns are not usually taken into account is that they have to be described by complex, high-dimensional models; and it is unclear how best to use information from these models for a prediction. We here propose an approach that describes dynamic frequency-specific power- and phase-coupling patterns using a Hidden Markov model (HMM) and combines it with the Fisher kernel, which can be used to predict individual traits. The Fisher kernel is constructed from the HMM in a mathematically principled manner, thereby preserving the structure of the underlying HMM. In this way, the unique, individual signatures of brain dynamics can be explicitly leveraged for prediction. The model is generally applicable to neuromaging and non-invasive electrophysiological data, including MEG. In summary, our approach makes it possible to leverage information about an individual’s brain dynamics for prediction in cognitive neuroscience and personalised medicine.

Background/Aims:
Beyond structural and static connectivity brain measures, the way brain activity dynamically unfolds can add important information when investigating individual cognitive traits. One approach to leverage this information is to infer on generative models of brain network dynamics and, from the model fits, extract features that can predict individual traits. However, there can be two potential sources of variation in these predictions. First, in certain cases, the run-to-run variability (e.g., due to different initialisations); and second, the variability induced by the choice of the model hyperparameters that determine the complexity of the model.

Methods:
To improve prediction accuracy, we propose an approach that leverages the useful aspects of this variability —in the sense of carrying complementary information rather than being the mere result of statistical noise— by combining predictions from different models and runs. Specifically, we use stacking to combine predictions from multiple models of brain dynamics to generate predictions that are accurate and robust across multiple cognitive traits.

Results:
We demonstrate the approach by describing dynamic frequency-specific power- and phase-coupling patterns using the Hidden Markov Model, and show that the use of stacking can significantly improve the accuracy and robustness of subject-specific phenotype predictions.

Discussion/conclusions/implications:
Looking forward, stacking predictions opens up avenues for integrating a wider variety of data or models that may improve predictions further, for example combining predictions generated using data from different brain imaging modalities, as well as static functional connectivity and structural information. Our model is broadly applicable to neuroimaging and non-invasive electrophysiological data, including MEG. In summary, our approach leverages multiple perspectives of brain dynamics to improve prediction in cognitive neuroscience.

Introduction:
Understanding human cognition requires mapping brain activity’s spatio-temporal structure. Time-varying functional connectivity (FC) methods are popular for describing statistical coupling changes in the brain. However, some existing methods suffer from unstable estimations across data runs. To address this, we propose two robust approaches based on Hidden Markov Models (HMMs): Best-Ranked HMM (BR-HMM) and Hierarchical Cluster HMM (HC-HMM).

Methods:
We applied the BR-HMM and HC-HMM to fMRI (100 subjects; 4800 timepoints per subject) and MEG (10 subjects; ≈72500 timepoints per subject) datasets. The HMM characterized concatenated timeseries as state activations, with each state representing a pattern of FC. For the MEG dataset, HMM was applied to power (band-limited to 8-12 Hz). BR-HMM involved running the model multiple times and selecting the best-ranked run based on free energy values. HC-HMM applied hierarchical clustering on state timeseries from multiple runs.

Results:
Both BR-HMM and HC-HMM showed significantly higher similarity scores (≈[0.85-0.95]) for fMRI and MEG datasets compared to individual runs (≈0.6), indicating improved stability. The BR-HMM approach required more runs for fMRI, while HC-HMM achieved high stability with fewer runs. For MEG, both approaches were comparable. However, when the lowest free-energy runs exhibited variations within the same HMM decomposition, the simpler BR-HMM approach was effective in capturing the dynamics, while the added complexity of HC-HMM may affect its performance.

Conclusions:
Despite HMM’s stochastic nature, our proposed approaches reliably captured momentary changes in FC for fMRI and MEG. BR-HMM offers higher stability scores but can be computationally costly. HC-HMM provides a more affordable solution. These methods yield stable and reliable estimates of time-varying FC, facilitating our understanding of neural processes and cognitive functions.

Unsupervised, data-driven methods are commonly used in neuroscience to automatically decompose data into interpretable patterns. These patterns differ from one another depending on the assumptions of the models. How these assumptions affect specific data decompositions in practice, however, is often unclear, which hinders model applicability and interpretability. For instance, the hidden Markov model (HMM) automatically detects characteristic, recurring activity patterns (so-called states) from time series data. States are defined by a certain probability distribution, whose state-specific parameters are estimated from the data. But what specific features, from all of those that the data contain, do the states capture? That depends on the choice of probability distribution and on other model hyperparameters. Using both synthetic and real MEG data, we aim to better characterize the behavior of two HMM types that can be applied to electrophysiological data. Specifically, we study which differences in data features (such as frequency, amplitude, or signal-to-noise ratio) are more salient to the models and therefore more likely to drive the state decomposition. Overall, we aim at providing guidance for the appropriate use of this type of analysis on electrophysiological data and an informed interpretation of its results given the characteristics of the data and the purpose of the analysis.

Understanding the neuronal mechanisms supporting consciousness is a fundamental question in neuroscience. Several competing theories have been proposed. To accelerate research, the predictions of these theories should be tested together under a common framework. This is the aim of COGITATE, an adversarial collaboration testing predictions from Global Neuronal Workspace (GNW) and Integrated Information Theory (IIT).
Here we tested two predictions made by the two theories regarding activation and inter-areal communication using MEG. Participants were presented with visual stimuli that were undoubtedly consciously perceived. GNW predicted a phasic activation in prefrontal cortex at both stimulus onset and offset, while IIT predicted content-specific sustained activation in posterior cortex during stimulus presentation. Additionally, GNW predicted stronger synchronization between prefrontal and category-selective areas in the “ignition” time window, whereas IIT predicted sustained synchronization between early visual cortex and category-selective areas.
The results indicated the presence of the predicted sustained alpha activity in posterior cortex. Furthermore, we observed the predicted late phasic ignition in prefrontal cortex at stimulus offset in the alpha band. However, this result was not supported by control analyses. Concerning phase-synchronization, neither the frequency band nor the temporal patterns of connectivity were consistent with the predictions of either theory.
By integrating our MEG results with other neuroscientific techniques (fMRI, intracranial EEG) and testing additional theoretical predictions (e.g., decoding of conscious content), we will get more conclusive evidence supporting or refuting the two theories and to clarify how consciousness arises in the human brain.

Understanding visual perception involves gaining knowledge of its temporal dynamics, stability and dependency on the participant’s cognitive state and context. However, in many cognitive electrophysiological studies participants are typically measured during a single experimental session, which restricts the investigation of processing variations over time scales longer than a few hours. We acquired and analysed a unique MEG dataset containing recordings from one adult subject. The dataset comprises data from 11 scanning days over the course of over 5 months in which the subject was instructed to mentally perform (with no behavioural output, and with exactly equal stimulus input) either a memory or visual task, thus enabling investigation of the influence of both cognitive context and temporal variations on visual processing. Temporal generalisation matrices were obtained using linear decoders. Context-related processing differences were explored through the training of a decoder to discriminate between memory and visual tasks. Additionally, decoders were trained for each session to discriminate between animate and inanimate trials. Their performance was evaluated by testing them on the other sessions, enabling investigation into how decoders trained on one task perform on the other. To investigate how processing of visual stimuli varies over time we compared decoding generalisation across sessions as a function of how far away in time are such sessions. Evidence of context-dependent differences in processing visual stimuli was found, but decoding was shown to be very robust across days or even months. In conclusion, our findings highlight the main differences in visual stimulus processing, with cognitive state (task) playing a significant role, while time appears to have little to no influence. However, validation across multiple participants is necessary to corroborate these results.

Background/Aims:
22q11 deletion syndrome (22q11ds) is a genetic condition caused by a deletion along chromosome 22. Common cognitive and psychiatric symptoms include learning disabilities, attention deficit hyperactivity disorder (ADHD), autism, and adulthood schizophrenia. MEG studies of children with 22q11ds have demonstrated altered static power and functional connectivity patterns which reflect those seen in adults with schizophrenia and correlate with cognitive scores. The dynamic sub-second activation of discrete functional networks (microstates) have also widely been shown to be altered in people with schizophrenia and are known to correlate with different domains of cognition, yet cortical microstates have not been studied in 22q11ds. Here we apply a recently published approach to source-space MEG cortical microstate analysis to uncover how dynamic cortical networks are altered in 22q11ds.

Methods:
Resting-state MEG and cognitive test scores were collected from 10-18 year-old participants with 22q11ds (N=35) and sibling controls (N=25). MEG cortical microstate analysis was performed and microstate statistics calculated using the +microstate toolbox.

Results:
Microstates were identified reflecting resting-state networks and previously published MEG cortical microstates in normative controls. A range of statistics including activation patterns, frequency content, and transitioning statistics differed between participants with 22q11ds and controls, and correlated with cognitive test scores.

Discussion:
Our results suggest that the activation and transitioning of dynamic cortical networks are altered in children with 22q11ds and are correlated with their cognitive and psychiatric symptoms. Microstate analysis uncovered altered brain dynamics in 22q11ds that were complementary to static connectivity and power analysis, suggesting the use of MEG cortical microstates may give novel insight into psychiatric and neurological disorders.

Predicting subject traits from data is crucial for understanding cognitive processes, brain disorders, diseases and normal ageing. In this work, we propose a mathematically principled method to predict subject traits from M/EEG spectrograms. The idea is to interpret a spectrogram as a probability distribution and apply the Kernel Mean Embedding of distributions, a powerful kernel-based approach that takes probability distributions as inputs. Focusing on both accuracy and robustness, we demonstrate its use and improvement for age estimation over a baseline method like ridge regression by leveraging the HarMNqEEG dataset—a multinational compilation of EEG recordings—, which we assessed using leave-one-country-out cross-validation. Our method shows key insights, such as identifying brain regions with larger effects on ageing and intriguing gender-related patterns. We observed that for both genders, the frontotemporal region presents a slightly higher ageing impact than the other regions. In general, men exhibit a more pronounced effect than women. These results are consistent with previous studies based on MRI, indicating that the more pronounced ageing effects are observed in the healthy brain for men than for women. Remarkably, our approach can be used for broader cross-modal applications, including MEG data. This study advances M/EEG-based age prediction and underscores the versatility and efficacy of our proposed method.

An important aspect of the brain processes underlying language and cognition is the integration and transformation of information across multiple brain systems. Thus, a detailed characterisation of brain connectivity is key. In order to characterize brain connectivity most accurately, connectivity methods should make use of the full multivariate and multidimensional information available from neuroimaging data. This should include a characterization of transformations between patterns of activation across brain regions, and their dependence on stimulus features, task and context.
Here, we describe novel methods developments to estimate the multidimensional relationships between patterns of brain activity from different brain regions. In particular, we will highlight their potential to estimate the voxel-to-voxel transformations between these patterns. This opens up opportunities to characterise these transformation with metrics such as sparsity, divergence, convergence, etc. We will specifically focus on methods that are suitable for event-related experimental designs. A few recent studies employed ridge regression to estimate linear transformation matrices. In fMRI data from an object recognition experiment this revealed that transformations between early visual cortex and inferior temporal areas are relatively sparse. In dynamic EEG/MEG data, this approach supported a central role for bilateral ATLs with a wider semantic brain network. The latter results have been confirmed using a nonlinear extension of this method, indicating that linear methods provide an efficient approximation of multidimensional brain connectivity. A multivariate as well as multidimensional extension of this method has also recently been proposed.
We propose methods for analysing pattern transformations in language research in more detail. We illustrate this on simplified examples from the neuroscience of word recognition.

Subthalamic deep brain stimulation (STN-DBS) is an effective therapy for alleviating motor symptoms in people with Parkinson’s disease (PwP), although some may not receive optimal clinical benefits. One potential mechanism of STN-DBS involves antidromic activation of the hyperdirect pathway (HDP), thus suppressing cortical beta synchrony to improve motor function, albeit the precise mechanisms underlying optimal DBS parameters are not well understood.
To address this, 20 PwP with STN-DBS completed a 2 Hz monopolar stimulation of the left STN during MEG. MEG data were imaged in the time-frequency domain using MNE. Peak vertex time series data were extracted to interrogate the directional specificity and magnitude of DBS current on evoked and induced cortical responses and accelerometer metrics of finger tapping using linear mixed-effects models and mediation analyses.
We observed increases in evoked responses (HDP ~3-10ms) and synchronization of beta oscillatory power (14-30Hz, 10-100ms) following DBS pulse onset in the primary sensorimotor cortex (SM1), supplementary motor area (SMA) and middle frontal gyrus (MFG) ipsilateral to the site of stimulation. DBS parameters significantly modulated neural and behavioral outcomes, with clinically-effective contacts eliciting significant increases in HDP responses, reductions in induced SM1 beta power and better movement profiles compared to suboptimal contacts, often regardless of the magnitude of current applied. Finally, HDP-related improvements in motor function were mediated by the degree of SM1 beta suppression in a setting-dependent manner.
Together, these data suggest that DBS-evoked brain-behavior dynamics are influenced by the level of beta power in key hubs of the basal ganglia-cortical loop, and this effect is exacerbated by the clinical efficacy of DBS parameters. Such data may be useful for characterizing DBS programming strategies to optimize motor symptom improvement in the future.

OPM-MEG offers sensitive measurement of brain activity in a wearable system, enabling naturalistic movement, lifespan compliance and comfortable scanning environments. Systems have recently been set up in research centres globally, facilitating the opportunity for large collaborative datasets. However, variability between OPM-MEG systems in different laboratories must first be characterised to achieve data harmonization. The present study compares gamma oscillations, a crucial aspect of neural communication and cognition, across OPM-MEG systems in Nottingham, UK (Notts) and SickKids hospital, Canada (SK). 52 healthy adults (24 male, average age 27) were recruited, 26 from each site. Participants passively viewed a visual stimulus (a circularly oscillating grating). Participants wore size-matched rigid 3D printed helmets with 58 triaxial OPM sensors (Notts) or 40 dual-axis sensors (SK). 3D structure scans were taken; one of the head-shape to warp age-matched template MRIs, and one wearing the helmet to coregister sensors to template anatomy. Data were epoched and filtered to the gamma band. Beamformer analysis was conducted on both datasets, generating pseudo-t statistical maps by contrasting task and rest periods. Results revealed increased stimulus induced gamma power, returning to baseline at rest. Average peak frequencies of gamma modulation were measured at 54 Hz in Notts and 56 Hz in SK. The relative change in signal amplitude in the 50 – 60 Hz band was 35% in Notts and 24% in SK. When equivalent channel count was implemented, Notts had peak frequency at 55 Hz and 28% change. This cross-site investigation of human visual gamma oscillations with OPM-MEG provides valuable insights into the reliability of measurement across systems, enabling future collaborative data collection across OPM-MEG sites.

The perception of polyphonic music is an effortless and enjoyable task that is made possible by complex neural mechanisms processing multiple musical streams. Previous studies demonstrated that the human brain learns the statistical regularity of a melody and actively attempts to anticipate upcoming notes. The perception and enjoyment of music has been suggested to arise from the combination of the input melody and our expectations, in line with the predictive processing framework. However, there remains considerable uncertainty on how polyphonic music is processed. Here, we tested the hypothesis that our brains can build predictions for multiple melodic streams simultaneously, rather than for just the most salient stream. To this end, we carried out a novel electroencephalography (EEG) experiment involving the natural listening of diaphonic classical music i.e., involving two simultaneous melodic streams. Temporal response functions (TRF) were used to estimate the neural encoding of melodic expectations for the diaphonic streams. Behavioural measures and a control condition were included to determine the most salient stream, allowing us to measure the impact of salience on melodic processing. Further analyses were carried out to discern the acoustic and melodic features driving salience in music.

Recent research indicates a correlation between beta band (13-30Hz) modulation and top-down influence, where heightened beta power suggests heightened inhibition. We investigated the effects of attention on beta amplitude using Optically Pumped Magnetometer (OPM) based Magnetoencephalography (MEG). MEG data were collected from 17 healthy participants undertaking a spatially selective tactile attention paradigm (Bauer et al., 2006). Tactile stimulation of the index fingers was achieved using ‘braille’ stimulators. Participants focused on one hand following a visual cue and identified target patterns presented in subsequent pattern sequences via button presses. 80 trials were conducted, interspersed with rest intervals. Data acquisition was conducted using a 192-channel OPM-MEG system. For each participant, sensor space time-frequency spectra were calculated, and source space analysis was performed using a linearly constrained minimum variance beamformer to project data onto 78 cortical regions. Task-induced SNR was calculated for each region, and beta envelope analysis compared attended and unattended stimuli (collapsed across hands and participants, excluding response trials). Sensor level results show the expected beta band modulation across stimuli segments. Beamformer analysis identified the postcentral gyrus as having the highest SNR during stimulation, other regions included the inferior and superior parietal lobes, integral for multimodal sensory processing. Further investigation of the beta envelope revealed significantly higher beta band amplitude for non-attended vs. attended stimuli (p=4.01e-5 and 6.97e-5 for left and right motor cortex; Wilcoxon sign rank test), implying beta band modulation is linked to inhibitory mechanisms preventing further action on non-attended stimuli. Our findings support the proposition that beta oscillations serve as markers of top-down inhibition in primary sensory cortices and are susceptible to attentional modulation.

Metacognition represents evaluation of our own knowledge and behaviours. Deficits or alterations in metacognition have been implicated in a range of cognitive functions and mental health issues, but the neural mechanisms of metacognition in naturalistic tasks remain poorly understood. Metacognitions are typically studied in 2 alternative forced choice (2-AFC) tasks, with simple artificial stimuli varying on a few discrete levels of a one-dimensional feature. To explore metacognition with more ecological validity and reveal the specific visual contents used by individuals, we consider rich sampling of a naturalistic ambiguous image that affords different perceptions in a 3-AFC task.
We apply Bubbles sampling to a naturalistic ambiguous image that affords two perceptions in a 3-AFC task—i.e. perceiving the nuns, the bust of Voltaire, or neither. Bubbles selectively reveal parts of the image randomly on each trial, resulting in a high-dimensional stimulus evidence space while we concurrently recorded MEG. Each participant also rated their confidence in each response. To explore the neural mechanisms of metacognition, we use mutual information (MI) to quantify the trial-by-trial dependence between confidence ratings and MEG responses.
We reveal the critical information for category judgment and confidence and find two distinct neural responses related to confidence: a slow response, visible in the evoked response after 500ms, and an alpha/beta desynchronisation effect between 800-1500ms. Source localisation results show these responses originate from different locations: slow response starts from parietal and then move to occipital sources, alpha desynchronisation mainly in occipital.
Our design enables us to explore metacognition with more ecological validity. Analysis results show that rather than reflecting the same evidence processing, these two neural representations of confidence may reflect different stages of metacognitive evaluation.

Background:
Acquired demyelinating syndromes (ADS) disrupt brain connectivity, especially in paediatrics. Autoantibodies to myelin oligodendrocyte glycoprotein (MOG-ab) have been newly identified in proportion of children with ADS with a significantly altered phenotype. We demonstrated MOG-ab +ve ADS to have more favourable cognitive outcomes, compared to seronegative counterparts.

Aim:
To establish whether the less severe phenotype in MOG-ab +ve ADS can be explained through differences in neurophysiological responses which have been identified as ‘atypical’ in ADS.

Methods:
To date, n=17 children with ADS have been recruited (n=7 tested positive for MOG-ab) and n=12 healthy controls (HCs). Post-movement beta (14–30 Hz) rebound (PMBR), a motor response following movement cessation, was assessed with a child-adapted visuomotor task. We performed source localisation of greatest difference in beta-band oscillations during the active contrasted against baseline and extracted virtual electrode time-series. Time-frequency spectrograms of beta-band oscillatory power were then used to estimate PMBR neurophysiology (e.g. peak power, latency of peak power). Participants completed structural MRI and cognitive assessment.
Results:
Preliminary analyses (n=7 MOG –ve ADS, n=3 MOG +ve ADS) showed no differences between participants with and without MOG-ab, in behaviour or neurophysiology. We hypothesise, in the larger sample, increased latency in both patient groups compared to controls, which will be lesser in the MOG-ab +ve children, in line with more favourable cognitive outcomes. Exploratory analyses will investigate white matter integrity of corticospinal motor tracts as a potential structural correlate.

Implication:
This research will tease apart the neurophysiological basis of spared cognitive functioning seen in MOG-ab +ve patients, compared to seronegative counterparts. This may therefore identify therapeutic targets to help protect cognition in seronegative ADS.

Fetal magnetocardiography (fMCG) has emerged as a non-invasive method of electrophysiological imaging of the fetal heart. Currently, cryogenic systems are one-size-fits-all, and use unnatural scanning positions, making them generally inappropriate for fetal imaging. Optically pumped magnetometers (OPMs) allow for room temperature, on-skin measurements of the fetal heart in a flexible array of sensors that can conform to the individual mother’s abdominal geometry. The aims of this work are to show the ability of OPMs to detect fMCG, and extract clinically useful information: signal morphology, fetal heart rate variation (FHRV) and movement detection. Fetal biomagnetic activity data were collected by QuSpin OPMs (16-35 sensors). Six subjects were scanned, between 32-35 weeks gestational age (GA), with one subject additionally scanned at 19 and 28 GA, providing 22 datasets. Sensors were secured to subjects via plaster of Paris ‘bump casts’. 10 minute recordings were taken from subjects sitting down. Movement felt by the mother was recorded. Independent component analysis was used to separate the fetal and maternal cardiac signals and an R-peaks extraction algorithm applied. The quality of the recordings were quantified by peak amplitudes and signal-to-noise ratio (SNR). R-peaks were used to perform FHRV analysis. OPMs were able to successfully detect the fMCG in all cases with GA > 32 weeks. The quality of the fMCG signals are sufficient for clinical applications. Movement was felt by mother in 14/22 datasets, and correlations between perceived movement and changes in FHRV observed. In the early GA scans, the fetal fMCG is undetectable at 19 weeks GA, and detectable at 28 weeks GA, showing the emergence of a sufficiently strong fetal heart signal between 19 and 28 weeks GA. OPMs can successfully detect the magnetic field signals originating from the fetal heart, and hold significant potential for clinical applications whilst being adaptable to fit a range of subjects.

Alzheimer’s disease affects neurophysiology by loss of neurones, synapses and neurotransmitters. A mechanistic understanding of the human disease will facilitate new treatments. Recent developments in biophysically-informed dynamic causal models enable inferences around laminar and cell-specific disease effects from human non-invasive imaging. Based on pre-clinical models and effects of cholinesterase inhibitors, we hypothesised that Alzheimer’s disease would affect superficial pyramidal cell gain and extrinsic connectivity in hierarchical cognitive networks.
Magnetoencephalography was recorded during a mismatch negativity (MMN) task from healthy adults (n=15) and people with symptomatic Alzheimer’s disease or mild cognitive impairment (n=47, amyloid-biomarker positive) at baseline and 16 months. Fifteen people from the patient group had repeat magnetoencephalography at two weeks. We inverted MMN responses to dynamic causal models. Second-level parametric empirical Bayes of the dynamic causal models examined the effect of group and progression (baseline vs follow-up) on pyramidal cell self-inhibition and extrinsic connectivity.
Sensor data confirmed the effect of disease and progression (patients vs controls, T=-1.80, p=0.04; patient baseline vs follow-up, T=-1.82, p=0.03) and reliability of the mismatch negativity amplitude (ICC=0.94, p<.001). Parametric empirical Bayes revealed that there is strong evidence (posterior probability>95%) in Alzheimer’s disease of reduced connectivity between pyramidal cells and reduced superficial pyramidal cell gain which changed further during follow up.
Dynamic causal models confirmed that reduced pyramidal cell gain and connectivity can explain the observed physiological effect of Alzheimer’s disease. This approach to non-invasive magnetoencephalography data may be used for experimental medicine studies of candidate treatments and bridge clinical to preclinical models of drug efficacy.

Background:
Essential tremor (ET) manifests in pathological tremors that vary with factors such as stress and motor demands. This study investigates the brain circuits governing these variations in a cohort of patients and healthy controls, to identify oscillatory biomarkers associated with endogenous tremor suppression that have the potential to be leveraged by brain stimulation therapies.

Methods:
The first cohort included 12 ET patients and 12 age-matched controls (using 128 channel EEG) and a second cohort including 4 controls/3 ET patients (using OPM/MEG). Brain signals, electromyography (EMG), and kinematics were recorded during a cued, whole limb reaching task using high-density neuroimaging, employing a 2×2 design (high vs low uncertainty cues; small vs large targets). Dynamic Imaging of Coherent Sources (DICS) was used to localize sources modulated by motor demands as well as sources in the brain that were synchronized to tremor activity.

Results:
In controls and patients, both response times and reach duration were modulated by cue uncertainty and target size (ANOVA (21) P < 0.001). Tremor amplitude was suppressed by ~10% when reaching for small targets (T-test (12), P < 0.05). Beta band (14-30 Hz) movement related synchronization was localized to the supplementary motor cortex (SMA). For small targets, beta resynchronization was significantly slowed. Increased motor precision elicited increases in gamma power that was most clear in OPM recordings (permutation test (5), P< 0.01) and localized to the posterior parietal cortex.

Discussion/Conclusions/Implications:
Recordings of brain activity with OPMs during large scale, whole limb movement reflects a novel achievement and are well validated against EEG data recorded using the same task. This work shows that activity across physiological beta/gamma bands are responsive to changes in motor demands during whole limb reaching and begins to untie how they co-modulate with tremor in pathologies such as ET.

Background/Aims: The Hurst exponent is a simple autocorrelation measure describing self-similarity or long- range dependence in timeseries data. In other words, it evaluates whether the observed deviations as part of a dynamical system are substantially outside the normal state of balanced criticality. It is increasingly being applied in neuroscience, in the context of scale- free brain networks, and has shown promise as a method for localising epileptogenic zones in patients awaiting neurosurgery. Changes in brain architecture resulting from brain damage, as well as brain diseases, are known to interfere with the dynamics of neuronal activity; this is often seen in the form of abnormal slow waves or other deviations from expected patterns of oscillatory activity. In this study, we aim to explore the potential of the Hurst exponent to quantify deviations from healthy brain activity in MEG data from patients who have suffered a mild traumatic brain injury (mTBI) and characterise its natural variability in adult controls. Methods: Resting state MEG data from the current study were used to produce a broad-band beamformer source model, reconstructing the timeseries for every voxel in the brain. The Hurst exponent was computed for each voxel to produce a volumetric map, which enables the quantification of regional and localised deviations in neurological dynamics for both control and patient groups. Results: We present group comparisons to determine whether the Hurst exponent can dissociate healthy brain activity in controls from the disturbed activity and abnormal neurological features seen in data from mTBI patients. Implications: Since these alterations in brain dynamics are typically associated with significant behavioural and quality-of-life changes, finding methods to quantify them and distinguish from other diagnoses with overlapping symptoms (e.g. depression) has clinical value. Using algorithms like the Hurst exponent could thus be incorporated in future diagnostic tools.

Background: In recent years, machines powered by deep learning have achieved near-human levels of performance in speech recognition. The fields of artificial intelligence and cognitive neuroscience have finally reached a similar level of performance, despite their huge differences in implementation, and so deep learning models can, in principle, serve as candidates for mechanistic models of the human auditory system. Methods: Utilizing high-performance automatic speech recognition systems, and advanced non-invasive human neuroimaging technology such as magnetoencephalography and multivariate pattern-information analysis, the current study aimed to relate machine-learned representations of speech to recorded human brain representations of the same speech. Results: In one direction, we found a quasi-hierarchical functional organization in human auditory cortex qualitatively matched with the hidden layers of deep artificial neural networks (DNN) trained as part of an automatic speech recognizer. In the reverse direction, we modified the hidden layer organization of the artificial neural network based on neural activation patterns in human brains. The result was a substantial improvement in word recognition accuracy and learned speech representations. Discussion: We have demonstrated that artificial and brain neural networks can be mutually informative in the domain of speech recognition. The neurocomputational function of superior temporal gyrus regions is akin to later layers of the DNN, computing complex auditory features such as articulation and phonemic information. On the other hand, “reverse-engineering” human learning systems implemented in brain tissue in such a bidirectional fashion provides a complementary approach in developing and refining DNN learning algorithms.

State-space models are widely employed across various research disciplines to study unobserved dynamics. Conventional estimation techniques, such as Kalman filtering and expectation maximisation, offer valuable insights but incur high computational costs in large-scale analyses. Sparse inverse covariance estimators can mitigate these costs, but at the expense of a trade-off between enforced sparsity and increased estimation bias, necessitating careful assessment in low signal-to-noise ratio (SNR) situations. To address these challenges, we propose a three-fold solution: 1) Introducing multiple penalised state-space (MPSS) models that leverage data-driven regularisation; 2) Developing novel algorithms derived from backpropagation, state-space gradient descent, and alternating least squares to solve MPSS models; 3) Presenting a K-fold cross-validation extension for evaluating regularisation parameters. We validate this MPSS regularisation framework through lower and more complex simulations under varying SNR conditions, including a large-scale synthetic MEG data analysis. In addition, we apply MPSS models to concurrently solve brain source localisation and functional connectivity problems for real event-related MEG/EEG data, encompassing thousands of sources on the cortical surface. The proposed methodology overcomes the limitations of existing approaches, such as constraints to small-scale and region-of-interest analyses. Thus, it may enable a more accurate and detailed exploration of cognitive brain functions.