bioRxiv Subject Collection: Neuroscience's Journal

Audiomotor prediction errors drive speech adaptation even in the absence of overt movement
Observed outcomes of our movements sometimes differ from our expectations. These sensory prediction errors recalibrate the brain's internal models for motor control, reflected in alterations to subsequent movements that counteract these errors (motor adaptation). While leading theories suggest that all forms of motor adaptation are driven by learning from sensory prediction errors, dominant models of speech adaptation argue that adaptation results from integrating time-advanced copies of corrective feedback commands into feedforward motor programs. Here, we tested these competing theories of speech adaptation by inducing planned, but not executed, speech. Human speakers (male and female) were prompted to speak a word and, on a subset of trials, were rapidly cued to withhold the prompted speech. On standard trials, speakers were exposed to real-time playback of their own speech with an auditory perturbation of the first formant to induce single-trial speech adaptation. Speakers experienced a similar sensory error on movement cancelation trials, hearing a perturbation applied to a recording of their speech from a previous trial at the time they would have spoken. Speakers adapted to auditory prediction errors in both contexts, altering the spectral content of spoken vowels to counteract formant perturbations even when no actual movement coincided with the perturbed feedback. These results build upon recent findings in reaching, and suggest that prediction errors, rather than corrective motor commands, drive adaptation in speech.

Memorize multiple phase pattern attractors in nonlinear coupled oscillators dynamic via heterogeneous connectivity strength
Whole brain neural oscillation activities exhibit multiple wave patterns and seem to be supported by the common circuit network structure. We proposed a Hebbian-like Kuramoto model based entirely on heterogeneous connectivity strength rather than phase delay, which encodes the multiple phase patterns as attractors. We systematically investigated how the model dynamic landscape influenced by attractors and their corresponding eigenvalues, as well as how to control the stability of equilibrium points and the occurrence of high dimension bifurcations. This framework enables us to reproduce the dominant wave activity components in human brain functional MRI signal, and provides a canonical model for the multi body physical system spatio-temporal pattern attractor dynamics.

Shared and Unique Neural Codes for Biological Motion Perception in Humans and Macaque Monkeys
Throughout evolution, living organisms have honed the ability to swiftly recognize biological motion (BM) across species. However, how the brain processes within- and cross-species BM, and the evolutionary progression of these processes, remain unclear. To investigate these questions, we examined brain activity in the lateral temporal areas of humans and monkeys as they passively observed upright and inverted human and macaque BM stimuli. In humans, the middle temporal area (hMT+) responded to both human and macaque BM stimuli, while the right posterior superior temporal sulcus (pSTS) exhibited selective responses to human BM stimuli. This selectivity was evidenced by an increased feedforward connection from hMT+ to pSTS during the processing of human BM stimuli. In monkeys, the MT region processed BM stimuli from both species, but no subregion in the STS anterior to MT was specific to conspecific BM stimuli. A comparison of these findings suggests that upstream brain regions (i.e., MT) may retain homologous functions across species, while downstream brain regions (i.e., STS) may have undergone differentiation and specialization throughout evolution. These results provide insights into the commonalities and differences in the specialized visual pathway engaged in processing within- and cross-species BMs, as well as their functional divergence during evolution.

Distinct bandwidths of orexin neuron activity independently encode body movement and metabolic state
Hypothalamic orexin/hypocretin neurons (HONs) are key orchestrators of metabolism and adaptive behaviors. HON activity varies rapidly across behavioral and metabolic states: active during periods of high arousal, and silent during quiescence. However, it is unknown whether HON activity tracks specific behaviors, or a general aspect of all behaviors, and whether this depends on metabolic state. After recording HON population dynamics in awake behaving mice, we performed a spectral analysis of HON encoding of spontaneous body movements and behaviors recorded via video-tracking. Multivariate analyses of distinct HON activity bandwidths revealed that movements were encoded at a higher frequency bandwidth than blood glucose, and imposing distinct metabolic states did not interfere with HON movement encoding. HON population activity tracked the total amount of body movement across multiple classified behaviors with a high degree of precision. At key projection targets, orexin/hypocretin peptide outputs correlated with self-initiated movement in a projection-specific manner, indicating functional heterogeneity in HON outputs. Finally, we found that body movement was not encoded to the same extent in other key neural clusters related to arousal or energy. Collectively, these data reveal a general principle behind HON activation in behaving animals, that is orthogonal to their reactions to metabolic shifts. This not only enhances our understanding of the fundamental roles of HONs in brain-body coordination, but also suggests novel non-invasive methods for monitoring HON population activity using video-tracked biometrics, which are pivotal for future investigations into how movement and energy demands are encoded by the brain.

Mapping eye, arm, and reward information in frontal motor cortices using electrocorticography in non-human primates
Goal-directed reaches give rise to dynamic neural activity across the brain as we move our eyes and arms, and process outcomes. High spatiotemporal resolution mapping of multiple cortical areas will improve our understanding of how these neural computations are spatially and temporally distributed across the brain. In this study, we used micro-electrocorticography ({nu}ECoG) recordings in two male monkeys performing visually guided reaches to map information related to eye movements, arm movements, and receiving rewards over a 1.37 cm2 area of frontal motor cortices (primary motor cortex, premotor cortex, frontal eye field, and dorsolateral pre-frontal cortex). Time-frequency and decoding analyses revealed that eye and arm movement information shifts across brain regions during a reach, likely reflecting shifts from planning to execution. We then used phase-based analyses to reveal potential overlaps of eye and arm information. We found that arm movement decoding performance was impacted by task-irrelevant eye movements, consistent with the presence of intermixed eye and arm information across much of motor cortices. Phase-based analyses also identified reward-related activity primarily around the principal sulcus in the pre-frontal cortex as well as near the arcuate sulcus in the premotor cortex. Our results demonstrate {nu}ECoG's strengths for functional mapping and provide further detail on the spatial distribution of eye, arm, and reward information processing distributed across frontal cortices during reaching. These insights advance our understanding of the overlapping neural computations underlying coordinated movements and reveal opportunities to leverage these signals to enhance future brain-computer interfaces.

Behavior- and circuit-specific cortico-striatal decoupling during the early development of Parkinson's disease-like syndrome
Despite that cortico-striatal decoupling has been widely reported in individuals diagnosed with Parkinson's Disease (PD), its onset, circuit specificity and underlying mechanism remain largely unclear. To investigate these questions, dual fiber photometry is established for the first time to evaluate cortico-striatal coupling during varied motor behaviors, whose cell-type resolution was provided by the usage of Cre transgenic mouse lines. Contralateral turning, digging and licking show distinct coupling patterns, among which digging induces the strongest coupling. Striatal D1R-expressed medium spiny neurons (dMSNs) and D2R-expressed MSNs (iMSNs) similarly contribute to the cortical-striatal coupling during turning and licking but not digging, with much tighter coupling between the dMSNs and the M1 cortex. In PD-like mouse model induced via intra-striatal injection of synthetic mouse wildtype -synuclein preformed fibril (PFF), digging-associated cortical-striatal decoupling emerges as early as 1-month post induction (Mpi), which becomes significant since 2 Mpi and correlates with later-onset behavioral deficit. Notably, impaired dMSNs but not iMSNs mediate this decoupling, which can be rescued by activation of D1 but not D2 receptor. Mechanistically, we found an inverted U-shape decline in striatal dopamine level along the disease development in PFF-injected mice. Supplement with L-DOPA alleviates the decoupling and motor deficit, suggesting that early dopamine deficiency directly contributes to the cortical-striatal decoupling and the associated motor deficit. These findings provide new insights into the temporal profile and mechanisms underlying the PD-associated cortico-striatal decoupling, which has been implicated as functional biomarker for early diagnosis of PD.

Single-cell DNA methylation analysis tool Amethyst reveals distinct noncanonical methylation patterns in human glial cells
Single-cell sequencing technologies have revolutionized biomedical research by enabling deconvolution of cell type-specific properties in highly heterogeneous tissue. While robust tools have been developed to handle bioinformatic challenges posed by single-cell RNA and ATAC data, options for emergent modalities such as methylation are much more limited, impeding the utility of results. Here we present Amethyst, a comprehensive R package for atlas-scale single-cell methylation sequencing data analysis. Amethyst begins with base-level methylation calls and expedites batch integration, doublet detection, dimensionality reduction, clustering, cell type annotation, differentially methylated region calling, and interpretation of results, facilitating rapid data interaction in a local environment. We introduce the workflow using published single-cell methylation human peripheral blood mononuclear cell (PBMC) and human cortex data. We further leverage Amethyst on an atlas-scale brain dataset to discover a noncanonical methylation pattern in human astrocytes and oligodendrocytes, challenging prior assumptions that this form of methylation is only biologically relevant to neurons in the brain.

Circadian gene Bmal1 in the lateral habenula regulates alcohol drinking behavior in a sex specific manner
The circadian clock regulates most aspects of mammalian physiology and behaviour, including alcohol drinking behaviour. Disrupted circadian clock function via deletion of clock genes along the mesolimbic dopamine (DA) pathway has been linked to altered patterns of alcohol drinking behaviour. The lateral habenula (LHb), an epithalamic structure that house a semi-autonomous circadian clock, is a negative regulator of the mesolimbic DA system and of alcohol consumption. To study the role of the LHb circadian clock in alcohol consumption, we knocked out the core clock gene, Bmal1 specifically in the LHb and examined the impact on various alcohol drinking paradigms and affective behaviours in male and female mice. Our findings demonstrate that Bmal1 deletion in the LHb leads to sex-specific alterations in alcohol consumption. Male knockout mice exhibited increased voluntary alcohol intake, enhanced consumption of a bitter alcohol solution, and elevated alcohol binge drinking compared to controls. Conversely, female knockouts showed a marginal decrease in voluntary intake, significantly reduced consumption of an aversive alcohol solution, and lower post abstinence, relapse-like drinking. These results indicate that Bmal1 in the LHb exerts a repressive effect on alcohol intake in males, while it facilitates intake under certain aversive conditions in females. Interestingly, Bmal1 deletion did not significantly affect anxiety-like or depressive-like behaviours, suggesting that the habenular clock's role in alcohol consumption is independent of affective state. These findings mark Bmal1 in the LHb as a novel sexually dimorphic regulator of alcohol consumption in mice. Potential mechanisms involving the circadian modulation of DA and serotonin signaling pathways by the LHb clock is discussed.

Age-dependent cortical overconnectivity revers under anesthesia in Shank3 mice
Increasing evidence points to brain network dysfunction as a central neurobiological basis for autism spectrum disorders (ASDs). Consequently, studies on Functional Connectivity (FC) have become pivotal for understanding autism-related large-scale network dynamics. While autism is a neurodevelopmental disorder, existing FC studies in mouse models largely focus on adult subjects under anesthesia. Given that FC can be significantly influenced by brain state, the differential impact of anesthesia on cortical functional networks in autistic subjects remains unexplored. To fill this gap, we conducted a longitudinal evaluation of FC across three distinct brain states in the Shank3b mouse model of autism. We utilized wide-field calcium imaging to monitor cortical activity in Shank3b+/- and Shank3b+/+ mice from late development through adulthood, and isoflurane anesthesia to manipulate the brain state. Our findings reveal that network hyperconnectivity, initially evident in the barrel-field cortices during the juvenile stage, progressively expands to encompass the entire dorsal cortex in adult Shank3b+/- mice. Notably, the severity of FC imbalances is highly dependent on the brain state: network alterations are more pronounced in the awake state and shift towards hypoconnectivity under anesthesia. These results underscore the crucial role of anesthesia in detecting autism-related FC alterations and identify a significant network of early cortical dysfunction associated with autism. This network represents a potential target for non-invasive translational treatments.

A Biologically Inspired Attention Model for Neural Signal Analysis
Understanding how the brain represents sensory information and triggers behavioural responses is a fundamental goal in neuroscience. Recent advances in neuronal recording techniques aim to progress towards this milestone, yet the resulting high dimensional responses are challenging to interpret and link to relevant variables. In this work, we introduce SPARKS, a model capable of generating low dimensional latent representations of high dimensional neural recordings. SPARKS adapts the self-attention mechanism of large language models to extract information from the timing of single spikes and the sequence in which neurons fire using Hebbian learning. Trained with a criterion inspired by predictive coding to enforce temporal coherence, our model produces interpretable embeddings that are robust across animals and sessions. Behavioural recordings can be used to inform the latent representations of the neural data, and we demonstrate state-of-the-art predictive capabilities across diverse electrophysiology and calcium imaging datasets from the motor, visual and entorhinal cortices. We also show how SPARKS can be applied to large neuronal networks by revealing the temporal evolution of visual information encoding across the hierarchy of the visual cortex. Overall, by integrating biological mechanisms into a machine learning model, we provide a powerful tool to study large-scale network dynamics. SPARKS' capacity to generalize across animals and behavioural states suggests it is capable of estimating the internal latent generative model of the world in animals, paving the way towards a foundation model for neuroscience.

Parvalbumin expression identifies subicular principal cells with high projection specificity
The calcium-binding protein parvalbumin is an established marker for a subset of cortical inhibitory interneurons with similar biophysical features and connectivity. However, parvalbumin is also expressed in a small population of excitatory cells in layer 5 of the neocortex with specific sub-cortical projection targets. Parvalbumin may thus also in principal cells identify particular subclasses with distinct connectivity and function. Here we investigated whether parvalbumin is expressed in excitatory neurons of the hippocampal formation and if so, whether it delineated neurons with specific features. We report parvalbumin-expressing glutamatergic cells in the distal subiculum, which - based on location, connectivity and gene expression - separated into two subclasses: neurons in deep layers, which specifically project to the antero-ventral thalamus and neurons in superficial layers, which project to the mamillary bodies. Contrary to most adjacent pyramidal cells parvalbumin-positive neurons were non-bursting and displayed straight apical dendrites devoid of oblique dendrites. Functionally, the projections diverged from classical driver/modulator subdivisions. Parvalbumin expression thus marks two sub-types of subicular projection neurons with high target specificity and unique functional features.

The relationship between sleep and cognitive performance on tests of pattern separation
Study objectives: Sleep disturbances are considered both a risk factor and symptom of dementia. The present research aimed to identify cognitive tests that are sensitive to sleep-dependent cognition, focusing on cognitive tests designed to evaluate the earliest cognitive changes in dementia. Methods: In Experiment 1, we recruited younger (n=89) and older (n=40) adults and remotely monitored their sleep patterns for 7 consecutive days using actigraphy and sleep diaries. On day 7, participants completed a battery of cognitive tests, which included the Prodromal Alzheimer's and MCI battery from the Cambridge Neuropsychological Test Automated Battery (CANTAB) and the Mnemonic Similarity Task (MST). In Experiment 2, we examined the effects of a night of total sleep deprivation on cognitive performance in young adults. We observed a sleep deprived group (n=16) overnight in the lab. The rested control group (n=32) slept normally at home. Results: In Experiment 1, there was a significant relationship between performance on MST and total sleep time in the older adults. There were no relationships between cognitive performance and sleep quantity or quality in the younger adults. In the older adults, Montreal Cognitive Assessment (MoCA) scores were correlated with performance on the CANTAB and MST. In Experiment 2, the young adults who underwent 24-hour sleep deprivation performed worse than the rested participants on the MST task. Conclusion: Performance on cognitive tests designed to assess pattern separation are sensitive to sleep patterns, particularly in older adults and should be evaluated for potential use as clinical trial outcome measures for sleep-promoting treatments.

The brain simulates actions and their consequences during REM sleep
Vivid dreams mostly occur during a phase of sleep called REM. During REM sleep, the brain's internal representation of direction keeps shifting like that of an awake animal moving through its environment. What causes these shifts, given the immobility of the sleeping animal? Here we show that the superior colliculus of the mouse, a motor command center involved in orienting movements, issues motor commands during REM sleep, e.g. turn left, that are similar to those issued in the awake behaving animal. Strikingly, these motor commands, despite not being executed, shift the internal representation of direction as if the animal had turned. Thus, during REM sleep, the brain simulates actions by issuing motor commands that, while not executed, have consequences as if they had been. This study suggests that the sleeping brain, while disengaged from the external world, uses its internal model of the world to simulate interactions with it.

Network models incorporating chloride dynamics predict optimal strategies for terminating status epilepticus
Seizures that continue for beyond five minutes are classified as status epilepticus (SE) and constitute a medical emergency. Benzodiazepines, the current first-line treatment, attempt to terminate SE by increasing the conductance of chloride-permeable type-A GABA receptors (GABAARs). Despite their widespread use, benzodiazepines are ineffective in over a third of cases. Previous research in animal models has demonstrated that changes in intraneuronal chloride homeostasis and GABAAR physiology may underlie the development of benzodiazepine resistance in SE. However, there remains a need to understand the effect of these changes at a network level to improve translation into the clinical domain. Therefore, informed by data from human EEG recordings of SE and experimental brain slice recordings, we used a large spiking neural network model that incorporates chloride dynamics to investigate and address the phenomenon of benzodiazepine resistance in SE. We found that the GABAAR reversal potential (EGABA) sets SE-like bursting and determines the response to GABAAR conductance modulation, with benzodiazepines being anti-seizure at low EGABA and ineffective or pro-seizure at high EGABA. The SE-like activity and EGABA depended on a non-linear relationship between the strength of Cl- extrusion and GABAAR conductance, but not on the initial EGABA of neurons. Independently controlling Cl- extrusion in the pyramidal and interneuronal cell populations revealed the critical role of pyramidal cell Cl- extrusion in determining the severity of SE activity and the response to simulated benzodiazepine application. Finally, we demonstrate the models utility for considering improved therapeutic approaches for terminating SE in the clinic.

Contingency awareness shapes neural responses in fear conditioning
Contingency awareness refers to an observer's ability to identify the association between a conditioned (CS) and an unconditioned stimulus (US). A widely held belief in human fear conditioning is that this form of associative learning may occur independently of contingency awareness. To test this hypothesis, in this preregistered study (https://osf.io/vywq7), we recorded electroencephalography (EEG) during a task, where participants were presented with compounds of words (from two semantic categories) and tactile stimulation, followed by either a neutral sound (US-) or a loud noise (US+). Based on interviews, participants were divided into an aware (N=50) and an unaware (N=31) group. Only the aware group showed signs of learning, as expressed in larger stimulus-preceding negativity developing before US+ and a stronger theta response to vibrations predicting US+. The aware group also showed stronger alpha and beta suppression around the vibrations and a weaker theta response to US+, possibly indicating heightened attention to the cue and the violation/confirmation of expectation. Personality tests showed that elevated anxiety, neuroticism, higher intolerance of uncertainty, or harm avoidance are not predictive to the acquisition of contingency awareness. Our findings support the notion that fear conditioning, as reflected in cortical measures, cannot occur without contingency awareness.

The ageing stopping network: Regional and network changes in the IFG, preSMA, and STN across the adult lifespan
Response inhibition, the cancellation of planned movement, is essential for everyday motor control. Extensive fMRI and brain stimulation research provides evidence for the crucial role of a number of cortical and subcortical regions in response inhibition, including the subthalamic nucleus (STN), pre-supplementary motor area (preSMA), and the inferior frontal gyrus (IFG). Current models assume that these regions operate as a network, with action cancellation originating in the cortical areas and then executed rapidly via the subcortex. Response inhibition slows in older age, a change that has been attributed to deterioration or changes in the connectivity and integrity of this network. However, previous research has mainly used whole-brain approaches when investigating changes in structural connectivity across the lifespan, or have used simpler measures to investigate structural ageing. Here, we used high-resolution quantitative and diffusion MRI to extensively examine the anatomical changes that occur in this network across the lifespan. We found substantial changes in iron concentration in these tracts, increases in the apparent diffusion coefficient, and some evidence for demyelination. Conversely, we found very little evidence for age-related anatomical changes in the regions themselves. We propose that some of the functional changes observed in these regions in older adult populations (e.g., increased BOLD recruitment) are a reflection of alterations to the connectivity between the regions, rather than localised regional change.

EEG source imaging technique to investigate sleep oscillations for simultaneous EEG-fMRI
Simultaneous recording of electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) is a widely used non-invasive neuroimaging technique in sleep studies. However, EEG data are strongly influenced by two types of MRI-related artefacts: gradient artefacts (GA) and ballistocardiogram artefacts (BCG). If artefacts correction is suboptimal, the BCG obscures the EEG signals below 20Hz and could make it difficult to investigate sleep oscillations, especially sleep spindles, sleep specific oscillations occurring within 11-16Hz frequency band. We previously demonstrated the utility of beamforming spatial filtering in correcting MRI-related artefacts on EEG. Here, we investigated the use of beamforming spatial filtering for improving the detection of EEG oscillations during sleep, in particular, 1) to accurately estimate single-event spindle EEG power changes, and 2) to demonstrate the potential improvement of fMRI general linear model (GLM) analysis when involving such additional EEG information. We analysed EEG-fMRI data acquired during a recovery nap after sleep deprivation in 20 young healthy participants (12 females, 8 males, age=21.3+/-2.5 years). Based on spindle events (onset and duration) detected by trained sleep scorers on BCG corrected EEG signals through a conventional average artefact subtraction (AAS) method, we compared four different EEG processing steps: non-BCG corrected; AAS BCG corrected; beamforming BCG corrected; beamforming+AAS BCG corrected. These processing steps consist of non-BCG corrected and AAS BCG corrected considered either at the sensor level or at the source-level (after beamformer localization) to evaluate the impact of the BCG artefact on the detection of spindle activity. Then we further investigated four different fMRI GLM approaches using 1) the spindle onset and duration (GLM1), 2) spindle onset, duration, and parametric modulation of single-spindle power change from the Cz electrode of the AAS BCG corrected data (GLM2), 3) spindle onset, duration, and parametric modulation of single-spindle power change from the beamforming+AAS BCG corrected (GLM3) and 4) spindle onset, duration, and parametric modulation of single-spindle power change from the beamforming BCG corrected data (GLM4). We found that the beamforming approach did not only attenuate the BCG artefacts, but also recovered sleep spindle activity occurring during NREM sleep. Furthermore, this beamforming approach allowed us to accurately estimate single-event power change of each spindle in the source space when compared to the channel level analysis, and therefore to further improve the specificity of fMRI GLM analysis, better localizing the recruited brain regions during spindles. Our findings show the benefit of applying beamforming source imaging technique to EEG-fMRI acquired during sleep. We demonstrate that this approach would be beneficial especially for long EEG-fMRI data acquisitions (i.e., sleep, resting-state), when the BCG correction becomes problematic due to inherent dynamic changes of heart rates. Our findings extend previous work regarding the application of the source imaging to the sleep EEG-fMRI. Combining with this advanced methodology and analysis, sleep EEG-fMRI will help us better understand the functional roles of human sleep.

Distinct Contributions of the Dorsal and Ventral Hippocampus to Spatial Working Memory and Spatial Coding in the Prefrontal Cortex
The hippocampus (HPC) supports spatial working memory (SWM) through its interactions with the prefrontal cortex (PFC). However, it is not clear whether and how the dorsal (dHPC) and ventral (vHPC) poles of the HPC make distinct contributions to SWM and whether they differentially influence the PFC. To address this question, we optogenetically silenced the dHPC or the vHPC while simultaneously recording from the PFC of mice performing a SWM task. We found that whereas both HPC subregions were necessary during the encoding phase of the task, only the dHPC was necessary during the choice phase. Silencing of either subregion altered the spatial firing patterns of PFC neurons. However, only silencing of the vHPC affected their coding of spatial goals. These results thus reveal distinct contributions of the dorsal and ventral HPC poles to SWM and the coding of behaviorally-relevant spatial information by PFC neurons.

Assessing white matter plasticity in a randomized controlled trial of early literacy training in preschoolers
Reading is a cognitive skill that requires our brain to go through a myriad of changes during learning. While many studies have described how reading acquisition shapes childrens' brain function, less is known about the impact of reading on brain structure. Here we examined short-term causal effects of reading training on preschoolers' behavior and white matter structure. Forty-eight English-speaking preschoolers (4y10m to 6y2m) participated in a randomized controlled trial where they were randomly assigned to two training programs: the Letter training program was focused on key skills for reading (e.g., decoding and letter knowledge), while the Language training program strengthened oral language comprehension skills without exposure to text. Longitudinal behavioral data showed that only the Letter Training group increased letter knowledge and decoding skills after the two-week training. Diffusion MRI measures (FA and MD) of eighteen white matter pathways (including the left arcuate and the left inferior longitudinal fasciculus) did not reveal any statistically significant changes for either group despite high degrees of scan-rescan reliability across sessions. These findings suggest that a two-week reading training program can cause changes in preschoolers' letter knowledge and decoding abilities, without being accompanied by measurable changes in the diffusion properties of the major white matter pathways of the reading network. We conclude highlighting possible constraints (i.e., age, training onset and duration, cognitive profile) to reading-related white matter plasticity.

Instruction-tuned large language models misalign with natural language comprehension in humans
Transformer-based language models have significantly advanced our understanding of meaning representation in the human brain. Prior research utilizing smaller models like BERT and GPT-2 suggests that "next-word prediction" is a computational principle shared between machines and humans. However, recent advancements in large language models (LLMs) have highlighted the effectiveness of instruction tuning beyond next-word prediction. It remains to be tested whether instruction tuning can further align the model with language processing in the human brain. In this study, we evaluated the self-attention of base and finetuned LLMs of different sizes against human eye movement and functional magnetic resonance imaging (fMRI) activity patterns during naturalistic reading. Our results reveal that increases in model size significantly enhance the alignment between LLMs and brain activity, whereas instruction tuning does not. These findings confirm a scaling law in LLMs' brain-encoding performance and suggest that "instruction-following" may not mimic natural language comprehension in humans.

Recovery of Verbal Working Memory Depends on Left Hemisphere White Matter Tracts
Researchers propose that the recovery of language function following stroke depends on the recruitment of perilesional regions in the left hemisphere and/or homologous regions in the right hemisphere. Many investigations of recovery focus on changes in gray matter regions, whereas relatively few examine white matter tracts and none address the role of these tracts in the recovery of verbal working memory (WM). The present study addressed these gaps, examining the role of left vs. right hemisphere tracts in the longitudinal recovery of phonological and semantic WM. For 24 individuals with left hemisphere stroke, we assessed WM performance within one week of stroke (acute timepoint) and at more than six months after stroke (chronic timepoint). To address whether recovery depends on the recruitment of left or right hemisphere tracts, we assessed whether changes in WM were related to the integrity of five white matter tracts in the left hemisphere which had been implicated previously in verbal WM and their right hemisphere analogues. Behavioral results showed significant improvement in semantic but not phonological WM from the acute to chronic timepoints. Improvements in semantic WM significantly correlated with tract integrity as measured by functional anisotropy in the left direct segment of the arcuate fasciculus, inferior fronto-occipital fasciculus and inferior longitudinal fasciculus. The results confirm the role of white matter tracts in language recovery and support the involvement of the left rather than right hemisphere in the recovery of semantic WM.

Relationships of hematocrit concentration with dementia from a multiethnic population-based study
Objective: Red blood cell (RBC) concentration impacts cerebrovascular disease, yet it is unclear whether RBC concentrations relate to dementia risk, particularly in racially/ethnically diverse cohorts. We investigated whether RBC concentrations associate with incident dementia risk in a diverse population of stroke-free individuals and explored whether cerebral small vessel disease (CSVD) mediates this relationship. Methods: A longitudinal observational analysis was performed using a population-based cohort of stroke-free, older adult participants (>50 years) from the Northern Manhattan Study (NOMAS) enrolled between 2003-2008. Participants received baseline hematocrit testing, MRI neuroimaging, and cognitive assessments at baseline and long-term follow-up. Associations of baseline hematocrit as a categorical variable (low, normal [reference], and high based on laboratory reference levels) with incident dementia were assessed using Cox models adjusting for relevant covariates. Separate analyses investigated whether MRI CSVD mediated these relationships. Results: We studied 1207 NOMAS participants (mean age 71{+/-}9 years, 60% female, 66% Hispanic). Mean hematocrit was 41.2% ({+/-}3.8) with 16% of participants developing incident dementia. Lower hematocrit associated with increased dementia risk (adjusted hazard ratio 1.81 [1.01-3.23]) after adjusting for age, sex, race/ethnicity, education, APOE status, and comorbidities. High hematocrit was not associated with dementia risk. No interactions by sex or race/ethnicity were seen and baseline CSVD did not mediate relationships between hematocrit and dementia. Conclusions: Low hematocrit associated with dementia risk in our diverse population cohort. Further work is needed to assess mechanisms behind anemia's relationship with dementia to assess whether this can serve as a trackable, preventable/treatable risk factor for dementia.

Novel tau filament folds in individuals with MAPT mutations P301L and P301T
Mutations in MAPT, the microtubule-associated protein tau gene, give rise to cases of frontotemporal dementia and parkinsonism linked to chromosome 17 (FTDP-17) with abundant filamentous tau inclusions in brain cells. Individuals with pathological MAPT variants exhibit behavioural changes, cognitive impairment and signs of parkinsonism. Missense mutations of residue P301, which are the most common MAPT mutations associated with FTDP-17, give rise to the assembly of mutant four-repeat tau into filamentous inclusions, in the absence of extracellular deposits. Here we report the cryo-EM structures of tau filaments from five individuals belonging to three unrelated families with mutation P301L and from one individual belonging to a family with mutation P301T. A novel three-lobed tau fold resembling the two-layered tau fold of Pick's disease was present in all cases with the P301L tau mutation. Two different tau folds were found in the case with mutation P301T, the less abundant of which was a variant of the three-lobed fold. The major P301T tau fold was V-shaped, with partial similarity to the four-layered tau folds of corticobasal degeneration and argyrophilic grain disease. These findings suggest that FTDP-17 with mutations in P301 should be considered distinct inherited tauopathies and that model systems with these mutations should be used with caution in the study of sporadic tauopathies.