Disrupted glial-mediated synaptic refinement in Fragile X syndrome
Fragile X syndrome (FXS), the most common inherited cause of intellectual disability and autism, results from the loss of the RNA-binding protein fragile X mental retardation protein (FMRP). FMRP is a translational regulator and is highly expressed in glial cells, where its role in neural circuit development remains poorly defined. Here, it was observed that Fmr1 knockout mice exhibit reduced synapse size and accelerated eye-specific segregation. To examine which cell-types participate in this process, a multi-omic framework was applied to FXS model mice at postnatal day 7, a critical window for synaptic remodeling in the retinogeniculate pathway, an established model system utilized to study synaptic pruning. Single-cell transcriptomics revealed coordinated alterations in microglia, astrocytes, and neurons in genes linked to synaptic pruning. Computational modeling further demonstrated enhanced astrocyte-to-microglia signaling, particularly through Ephrin A (EphA)- and semaphorin-mediated pathways, while lipidomic profiling revealed reductions in EphA-associated lipid species required for lipid raft stability and receptor localization. Consistent with these observations, a glial engulfment assay indicated that FXS microglia and astroglia over-engulf synaptic material in the lateral geniculate nucleus, supporting the transcriptomic profile. Together, these findings identify impaired glial-driven synaptic refinement as an early mechanistic feature of FXS pathogenesis, highlighting the genes involved in this process as potential therapeutic targets during circuit development.

(Comment on this)

Modulation of feature attention by reward prediction error explains value learning behavior
Adaptive behavior requires learning the value of environmental features while selectively attending to those most likely to yield reward. Reward prediction errors (RPEs) drive value learning and learned values guide attention, yet the computational function linking RPEs to attentional modulation remains unspecified. Here, we developed a reinforcement learning model with a perceptual front-end to investigate how value and RPE signals modulate attentional gain during learning. We compared five candidate RPE-attention transfer functions, each combined with either single- or multi-focus attention, against behavioral data from two adult male rhesus macaques performing a color-value learning task with shifting reward contingencies. Monkeys exhibited rapid initial learning followed by sub-optimal asymptotic accuracy. Overall, single-focus architectures consistently outperformed multi-focus counterparts on matching monkey errors, indicating that macaques collapse the value distribution into a winner-take-all attentional focus. Furthermore, the ``Switch'' model, in which attention targets the highest-valued feature but transiently inverts following negative RPEs, produced the fastest exploration dynamics following target switches and, together with the Absolute Value model, yielded decision confidence trajectories that positively correlated with empirical reaction times. In support of this, single-neuron correlation analyses revealed that 27-42% of neurons in prefrontal cortex, frontal eye fields, and lateral intraparietal area encoded previous-trial RPE at the time of next trial onset. In total, we conclude that capacity constrained attention that inverts its focus after negative RPE best explains value learning dynamics. These results provide a normative account for why biological learners sacrifice asymptotic precision for rapid adaptation in volatile environments.

(Comment on this)

Dissecting the molecular triggers of early and late long-term potentiation
The brain stores information by changing the strength of its synapses, a process that has at least two phases: Late long-term potentiation (L-LTP) is thought to result from the consolidation of early LTP (E-LTP), just as long-term memory requires the prior establishment of short-term memory. Recently, inhibitory avoidance experiments under CaMKII inhibition have challenged this notion, demonstrating long-term fear memory without measurable short-term memory. Here we use optogenetic activation and inhibition of CaMKII during induction of spike-timing-dependent potentiation (tLTP) to dissect the signaling pathways. While CaMKII activation in CA1 neurons was sufficient to induce E-LTP, growth of the postsynaptic density and spine neck expansion, we found that CaMKII-induced LTP does not give rise to L-LTP. Conversely, inhibition of CaMKII during tLTP induction prevented E-LTP, but FOS and L-LTP were still expressed, driven by CaMKK and PKM{zeta}. Thus, both long-term memory and L-LTP form in the absence of CaMKII activation.

(Comment on this)

Precursor Cells in the Parenchyma Act in Concert with Ventricular Neural Progenitors to Facilitate Dramatic Astrocyte Turnover and Recovery Following Natural Neuronal Death
Songbirds exhibit remarkable seasonal neuroplasticity, with song control nuclei undergoing seasonal cycles of extreme and rapid neuronal death and regeneration. While adult neurogenesis in these systems is well-characterized, the dynamics and functional significance of astrocytic turnover remain unknown. Here, we examined the fate of neural progenitor cell progeny born during seasonally-induced reactive proliferation and identified a rapid astrocytic turnover event in HVC following seasonal neuronal loss. Using lineage-specific and proliferation labeling, we characterized a previously undescribed SOX2-positive neural progenitor-like population within the avian parenchyma beyond the canonical ventricular zone niche. These parenchymal astrocyte precursor cells (pAPCs) proliferate at quantifiable, steady levels under homeostatic conditions, yet as a proliferative cell pool dramatically expand following non-injury induced neuronal death. Beyond their proliferative potential, pAPCs demonstrate capabilities suggestive of self-renewal and generation of astrocytes and neurons. The coordinated response of canonical neural progenitor cells and the newly-described pAPCs generates new astrocytes that persist throughout re-establishment of homeostasis, all of which together likely facilitate subsequent circuit regrowth. These findings reveal extensive astrocyte plasticity in the adult avian telencephalon and establish a foundation for understanding how astrocytes and their precursors - both within and beyond their canonical niches - contribute to neural circuit remodeling and behavioral maintenance.

(Comment on this)

Memory reactivation during sleep promotes structure abstraction
We readily detect structure in our environments, which in turn guides future learning. Disentangling this structure from the superficial features of a specific learning environment provides an especially strong basis for future generalization, but it remains unclear when and how this kind of abstraction occurs. Memory reactivation during sleep has been hypothesized to support such abstraction, but this has yet to be directly tested. Here we examined this hypothesis by teaching participants novel categories in which patterns of feature covariation were governed by different graph structures. Participants then learned a new category, defined by entirely different features, whose structure was either congruent or incongruent with a previously learned category. If structural knowledge is abstracted away from superficial features, it should facilitate transfer when structures are congruent. In Experiment 1, when two categories were learned in immediate succession, participants showed no transfer benefit, suggesting that structure understanding remained tied to the original features. In Experiment 2, we tested whether offline processing promotes abstraction. Participants either remained awake between learning phases spaced 3 hours apart, or took a nap during which a previously learned category was reactivated using targeted memory reactivation (TMR). Transfer benefits emerged only when the reactivated and target categories shared the same structure, and these benefits increased with the number of cues presented during slow-wave sleep. These findings provide the first direct evidence that memory reactivation during sleep promotes the abstraction of structure, enabling knowledge to transfer across learning episodes with no overlap in features.

(Comment on this)

Designer indicators for two-photon recording of subthreshold voltage dynamics
Subthreshold voltage dynamics are critical for neuronal information integration, yet they remain understudied in vivo due to the limitations of current tools. While genetically encoded voltage indicators (GEVIs) offer a promising alternative, their application for deep-tissue recording using two-photon (2P) microscopy--a preferred method for deep-tissue recording--has been hindered by insufficient sensitivity for detecting millivolt-scale subthreshold signals. Here, we refined our multiparametric two-photon high-throughput screening platform to develop two novel GEVIs, JEDI3sub and JEDI3hyp, tailored explicitly for subthreshold voltage detection. Through fast 2P optical recording in awake, behaving mice, we demonstrated the superior sensitivity of JEDI3 indicators compared to JEDI-2P. We also showed that JEDI3sub can track population-level subthreshold optical tuning, while JEDI3hyp reliably captured subthreshold dynamics associated with sharp-wave ripple oscillations in hippocampal PV interneurons. Finally, JEDI3hyp facilitated extended imaging of brain-state-dependent, millivolt-scale subthreshold voltage changes across deep-layer somas, fine dendritic structures, and diverse cell types. By addressing the critical gap in 2P optical recording of subthreshold voltage dynamics, JEDI3 indicators open new avenues for studying neural information processing and its alterations in health and disease.

(Comment on this)

Dynamic thermodynamic-informational entropic relationship (TIER) models of selective vulnerability to neurodegeneration
BACKGROUND: Neurodegenerative diseases share selective vulnerability patterns suggesting common physical mechanisms. We apply unified mechanics theory to neural systems, predicting that brain regions accumulate structural damage proportional to computational workload. METHODS: We simulated a hierarchical neural network implementing relationships between mechanical work (W = F * D), proportional thermodynamic entropy accumulation, and structural failure thresholds. Neural architectures at three hierarchical levels employed Hebbian learning across 2000 simulation sets, tracking entropy generation and dynamic stability. A coupled "siphon" model simulated cortical and subcortical support populations under constant cognitive demand. RESULTS: Heteromodal integration nodes consistently exhibited elevated work, accelerated entropy accumulation, and dynamic instability across architectures. Support systems reached 50% population loss before cortical systems despite lower absolute work, demonstrating accelerated compensatory failure. DISCUSSION: These thermodynamic-informational entropic relationships (TIER) reveal physical mechanisms underlying selective vulnerability across neurodegenerative diseases, reframing neurodegeneration as inevitable consequence of evolutionary trade-offs optimizing cognition over longevity.

(Comment on this)

A region-delineated snRNA-seq atlas of mouse spinal cord across lifespan resolves the interaction of normative aging programs with SOD1-G93A ALS
Aging is the strongest risk factor for amyotrophic lateral sclerosis (ALS), yet how normative aging programs intersect with disease mechanisms remain unclear. Here we generated a lifespan-resolved, cell type- and region-specific single-nucleus RNA-sequencing atlas of the mouse spinal cord spanning embryonic development through advanced age in WT mice and end-stage disease in the SOD1-G93A ALS model. This resource enabled systematic comparison of physiological aging trajectories with disease-associated transcriptional changes across spinal cord cell types and rostrocaudal regions. We found that SOD1-G93A transcript and protein states differed markedly across spinal regions during disease onset and progression, and these molecular patterns paralleled the relative resilience of cervical regions and the heightened vulnerability of lumbar regions to degeneration in this transgenic mouse model. Prior to disease onset, we identified reduced ubiquitin expression that primed region-specific disruption of proteostasis in the SOD1-G93A spinal cord. Despite these disease-associated changes, aging-related transcriptional programs were largely preserved across most cell types, arguing against a global acceleration of aging in ALS. Instead, microglia emerged as a key exception, exhibiting accelerated and rewired aging- and disease-associated gene expression modules regulated by MITF and NRF2. Together, these findings provide an anatomically, cellularly, and temporally resolved framework for understanding how aging programs interact with disease-specific pathways to shape regional dysfunction and neurodegeneration in ALS.

(Comment on this)

Whole-organism spatial transcriptomics at single-cell resolution in C. elegans
Spatial transcriptomics has advanced our understanding of tissue organization by mapping gene expression in its native context yet applying these techniques to whole organisms remains a significant challenge. Caenorhabditis elegans is well-suited to whole organism-level analysis because its compact size, transparency, reproducible anatomy, and genetic tractability make it possible to link molecular and cellular changes to circuit function and behavior within the same animal. However, current transcriptomic approaches in C. elegans are often limited by spatial resolution or multiplexing capacity, making it challenging to profile multiple gene expression patterns across intact worms while preserving spatial context. Here, we present a single-molecule fluorescence in situ hybridization workflow that enables multiplex imaging with single-cell resolution across the entire worm. This approach allows sequential imaging of one gene per fluorescent channel using two channels across 20 hybridization rounds, enabling the profiling of up to 40 genes while preserving spatial context. We further provide a curated marker-gene panel for reproducible neuron identification, which, together with probabilistic assignment of transcripts to segmented nuclei, enables quantitative measurements of gene expression levels. We used this method to identify up to 86 neuronal classes and reveal sex- and neuron-specific expression patterns at single-cell resolution. Together, these results establish a scalable framework for the spatial analysis of gene expression and cell identity in intact C. elegans.

(Comment on this)

Neurophysiological correlates of processing Agreement and Tense in Arabic
Most syntactic approaches converge on the fact that Tense and Agreement are two different functional categories, although there is less agreement on their exact representation and relative hierarchical order. Cross-linguistic agrammatic data seems to support the difference between Tense and Agreement, with patterns of dissociation reported from agrammatism between them, in which Tense is generally more impaired than Agreement. To examine whether there is evidence for such a dissociation of tense and agreement processing in neurotypical individuals, the present study employed Event-Related brain Potentials (ERPs) to study the real-time comprehension of Modern Standard Arabic sentences. Critical stimulus sentences were of the form Temporal Adverb-Subject-Verb-PP, in which the intransitive verb was in either the past or future tense, and was preceded by a singular or plural subject and an adverb indicating past or future tense. The subject nouns were all human and either masculine or feminine. The verbs either agreed with the subject noun or presented a person, number or gender agreement violation. They also either agreed or showed a mismatch with the temporal frame of the adverb, the latter being a tense violation. Results at the verb showed that both tense and agreement violations yielded a biphasic N400-P600 effect. We discuss these results in light of previous ERP findings and conclude that despite the putative configurational differences between Tense and Agreement, the processing of the two categories in Arabic may deploy the same underlying cognitive mechanisms.

(Comment on this)

Kinematics-based assessment of reaching and grasping movements in LRN ablated animals identifies a role for the LRN in endpoint stabilization and reach timing.
The lateral reticular nucleus (LRN) is thought to contribute to skilled forelimb control but its specific contributions to reaching and grasping remain unclear. In this paper, we examine skilled reaching in intact adult female Long-Evans rats after bilateral LRN ablation through single-pellet reaching tasks. Tasks were analyzed using sensitive quantitative kinematic analyses and qualitative behavioral scoring. Overall, limb transport was largely preserved after ablation, with results appearing in temporally restricted differences. The clearest deficits emerged in pellet-directed endpoint control. LRN-ablated animals showed broad variability in endpoint covariance, endpoint spread, and increased trial-to-trial variability, indicating that the movement became less precise and less consistent. These effects were more consistent than any single spatial difference seen, suggesting that ablation of the LRN impairs movement refinement rather than inducing a simple directional bias, although the paw height during the reach was significantly effected. Reach duration also changed, but this temporal difference emerged later and was less prominent. Our results suggest that the LRN acts as an important contributor to endpoint stabilization and reach timing during skilled forelimb behavior.

(Comment on this)

Topographic structure and function of locus coeruleus norepinephrine neurons
Norepinephrine (NE) is released throughout most of the central nervous system by neurons in the locus coeruleus (LC). We found a relationship between the morphologies, gene expression, and activity of LC-NE neurons in mice. Axonal projections of individual neurons were extensive but largely confined to subsets of brain regions. Axonal projections and graded gene expression correlated with locations of cell bodies in LC. In a behavioral task requiring ongoing learning from actions, neurons in dorsal LC projecting to the cerebral cortex were excited when mice made a different choice from the previous one and by reward prediction errors, a signal driving learning. Background activity of neurons in ventral LC was higher when mice ignored stimuli indicating potential reward availability. These observations reveal a topographically organized structure and function of a neurotransmitter system and show that it contains learning signals for flexible behavior.

(Comment on this)

Electrophysiological properties of mesodiencephalic junction neurons projecting to the inferior olive
Neocortex has the largest volume, while cerebellar cortex houses most neurons of the mammalian brain, underscoring paramount functions for interactions between them. Neocortex projects to the cerebellum via both the mossy fibre and climbing fibre system; whereas the mossy fibres find their origin in the pons, the climbing fibres are relayed via the subnuclei of the inferior olive (IO) that receive their cortical input from the mesodiencephalic junction (MDJ). Since climbing fibres regulate cerebellar plasticity in a timing-dependent manner, the MDJ-IO pathway probably contributes to both sensorimotor and cognitive learning. However, the physiological properties of IO-projecting MDJ neurons remain largely unknown. Here, we made targeted whole-cell recordings in acute brain slices of the murine MDJ, separating IO- from non-IO-projecting neurons following retrograde tracing. We show that IO-projecting neurons are spontaneously active and are capable of high frequency-bursts up to 350 Hz during current injections. Action potentials of IO-projecting MDJ neurons depolarize and re-polarize quickly, and hyperpolarizing inputs consistently trigger rebound action potentials upon stimulus offset. Instead, non-IO-projecting neurons in the same MDJ region are hardly spontaneously active with little rebound activation. Even so, IO-projecting MDJ neurons can also transform excitatory and inhibitory inputs into a proportional output, indicating that they can also employ rate coding. Moreover, neocortical inputs to IO-projecting MDJ neurons can directly be integrated with their inputs from the cerebellar nuclei. Our results highlight how neocortical inputs to the MDJ can be transformed into diverse IO firing patterns and thereby contribute to the required specifics of cerebellar learning.

(Comment on this)

The Temporal Constraints of the Cerebellum's Timekeeping
The cerebellum plays a central role in generating temporal predictions from past sensory regularities, yet the temporal boundaries of this predictive capacity remain unclear. Using magnetoencephalography (MEG), we investigated somatosensory and cerebellar responses to omissions within rhythmic somatosensory stimulation trains across six inter-stimulus intervals (ISIs) ranging from 0.5 to 5.5 seconds. We hypothesised that cerebellar prediction signals would follow a logistic decay pattern, remaining robust at short ISIs before declining beyond a 2-4-second temporal threshold. As a first step, we validated the omission paradigm by confirming the expected SI and SII response pattern to stimulations and the preservation of the SII response to omissions. Cerebellar source reconstruction revealed consistent beta band (14-30 Hz) responses to omissions peaking at 40-50 ms post-omission in right lobule VI, replicating previous findings. Critically, cerebellar activation was compatible with a logistic decay pattern with increasing ISI, with the inflexion point estimated within the hypothesised 2-4-second window, though precise localisation of this threshold warrants further investigation. Together, these findings establish empirical boundaries for cerebellar temporal prediction, suggesting that the cerebellum operates as a precise but duration-limited internal clock with implications for understanding the brain's timing mechanisms and their functional consequences for perception.

(Comment on this)

Rare variants alter mitochondrial lipid homeostasis and neuronal excitability in PD patient-derived dopaminergic neurons
Parkinson's disease (PD) exhibits substantial genetic heterogeneity, yet how combinations of rare variants converge on disease-relevant cellular mechanisms remains unclear. Here, we generated human induced pluripotent stem cell-derived dopaminergic neurons from PD patients carrying rare variants in recently implicated genes and performed integrated electrophysiological, proteomic, lipidomic, and genetic analyses. Patient-derived neurons showed reduced membrane capacitance and altered action potential firing, indicating impaired intrinsic excitability and synaptic dysfunction, with marked variability across genetic backgrounds. Multi-omics profiling revealed dysregulation of mitochondrial function, glycolysis, and oxidative phosphorylation, accompanied by extensive lipid remodeling, including increased fatty acids, acylcarnitines, and sphingolipids, and reduced gangliosides. These alterations were more pronounced in neurons harboring specific variant combinations in KIF21B, SLC6A3, HMOX2, TMEM175, and AIMP2. Integrative analyses uncovered coordinated protein-lipid changes linking mitochondrial dysfunction and membrane homeostasis. Notably, Calpastatin and CXCR4 were consistently dysregulated across PD neurons. Genetic association analyses in independent cohorts identified PD-associated variants in genes encoding dysregulated proteins, supporting the functional relevance of these pathways. Overall, our results define convergent and variant-specific mechanisms underlying PD and highlight candidate biomarkers and therapeutic targets.

(Comment on this)

Sex-specific plasticity mechanisms mediating fear extinction
There is strong evidence that synaptic plasticity is a critical cellular mechanism underlying learning and memory. Although the forms of synaptic plasticity used by different circuits and cell types vary, a widespread presumption is that the male and female brain has evolved to use the same form of plasticity within the same circuits during performance on the same task. Here, we used complimentary approaches to determine how activity in the mouse frontal cortex supports the extinction of associative memories in a context-dependent manner. While in vivo recordings show that both male and female mice have similar cue-relevant activity patterns and ensemble dynamics in excitatory neurons from the infralimbic cortex (IL) during learning, activity in amygdala-projecting IL neurons was indispensable for extinction memories only in male mice. Likewise, male but not female mice showed evidence for the recruitment of IL by structural remodeling and clustering of dendritic spines on these neurons, and extinction memory impairments were evident only in male mice after projection-specific IL deletion of the glutamate receptor subunit GRIN2B. This work provides strong evidence that synaptic plasticity mechanisms employed during learning and critical for memory retrieval differ between males and females, which undercuts the utility of one-size-fits all therapeutic approaches for mental health conditions in which memory is disrupted.

(Comment on this)

Upregulating action semantics with neuromodulation and gesture observation to facilitate verb retrieval in aphasia
Although many individuals with chronic aphasia respond to language therapy, there remains a need for adjunctive interventions that can enhance treatment response. Approaches targeting multiple modalities, such as gesture cueing, and neuromodulation techniques, such as transcranial magnetic stimulation, have shown promise for supporting language recovery. The present pilot study investigated whether enhancing activation of the action semantic network could facilitate verb production in individuals with chronic aphasia. Participants were recruited as a convenience sample and completed a within-subject design in which all individuals received each condition. Two non-linguistic methods of activating the action semantic network were evaluated: (1) pantomimed gesture cues to prime action concepts and (2) intermittent theta-burst stimulation to the left posterior middle temporal gyrus (pMTG), an intact action-semantic network node in our participants. We examined individual and combined effects of gesture priming and stimulation to test whether a combined approach would yield additive or interactive benefits. Using a Bayesian generalized linear mixed-effects model, we observed a moderate interaction between gesture priming and stimulation site. Contrary to predictions, combining gesture priming with pMTG stimulation did not produce additional benefits over either intervention alone. Instead, pMTG stimulation attenuated the priming advantage observed under vertex stimulation, and gesture priming attenuated the advantage observed with pMTG stimulation alone. Posterior estimates provided substantial preliminary evidence for this interaction in our pilot sample size. These findings suggest that combined activation of the action semantic network through gesture and neuromodulation approaches may not benefit verb retrieval above and beyond each approach alone.

(Comment on this)

Cognitive Vergence and Pupil Response During Oddball Task are Associated With Alzheimers Disease Cerebrospinal Fluid Neurodegenerative Biomarkers
Background: Alzheimers disease (AD) can be diagnosed using cerebrospinal fluid (CSF) biomarkers reflecting amyloid and tau pathology. However, it provides no information about functional network status. We aimed to determine whether CSF biomarkers (AB 42, p-Tau, t-Tau, and AB 42/p-Tau ratio) are associated with altered stimulus differentiation in vergence and pupil responses during an oddball task, and to evaluate oculomotor metrics as predictors of CSF core AD biomarkers in patients at mild cognitive impairment (MCI) stage. Methods: Thirty-eight participants with abnormal CSF core AD biomarkers at MCI stage completed a visual oddball task while oculomotor responses were recorded. Linear mixed- effects models examined condition x biomarker interactions, controlling for sex, age, and MMSE. Temporal and magnitude features were tested as predictors using linear regression. Results: Higher p-Tau levels were negatively associated with target-distractor differentiation in cognitive vergence (B; = -0.035, p < 0.001) and pupil responses (B = -0.060, p < 0.001). Higher AB42 and AB 42/p-Tau showed positive associations with vergence differentiation but opposite effects on pupil responses. Oculomotor features predicted p-Tau levels (R2= 0.20-0.21). Conclusion: Oculomotor differentiation metrics capture functional signatures of tau-related network dysfunction, positioning them as accessible biomarkers complementing CSF measures for detecting network disruption at MCI stage.

(Comment on this)

Identifying Phelan-McDermid-Like Electrophysiological Subtypes in Autism Using EEG and Machine Learning
Background: Phelan McDermid syndrome (PMS), caused by SHANK3 haploinsufficiency, is a genetic form of autism spectrum disorder (ASD) that provides a genetically defined model for studying ASD-related circuit dysfunction. SHANK3 mutations disrupt synaptic organization and cortical synchrony, leading to attenuated gamma-band auditory steady-state responses (ASSRs). We investigated whether PMS-related electrophysiological signatures could be identified using machine learning and whether similar patterns are present in a subset of individuals with idiopathic ASD (iASD). Methods: EEG recorded during a 40-Hz ASSR paradigm was collected from 123 participants (42 TD aged 2-30, 56 iASD aged 3-31, 25 PMS aged 2-26). We extracted time-series, ERSP, FOOOF-derived spectral, and intertrial phase coherence (ITPC) features. XGBoost models with leave-one-out cross-validation classified PMS versus TD; the best age/sex-adjusted ITPC model was then applied to iASD participants to derive a Synchrony Atypicality Index (SAI). Unsupervised clustering of high-dimensional ITPC features was also performed. Results: ITPC-based models showed the strongest discrimination between TD and PMS participants (AUROC = 0.83). When applied to iASD participants, 35.7% exhibited elevated SAI, indicating a PMS-like gamma-band phase-locking profile. Classification of iASD versus PMS performed poorly in the full sample but improved markedly after excluding high-SAI iASD individuals, consistent with substantial heterogeneity within iASD. Unsupervised clustering of ITPC features identified PMS-enriched clusters that also captured high-SAI iASD participants. Results were consistent after controlling for age in sensitivity analyses. Conclusions: Reduced 40-Hz ITPC is a mechanistically interpretable electrophysiological signature of PMS and identifies a biologically meaningful PMS-like subgroup within iASD, supporting biomarker-guided stratification.

(Comment on this)

From sequences to schemas: low-rank recurrent dynamics underlie abstract relational representations
A hallmark of intelligent behavior is the ability to extract abstract relational structure from temporal sequences, recognizing, for instance, that seq{aab}, seq{ccd}, and seq{eef} all follow the same underlying pattern, regardless of the specific elements involved. This capacity, observed across species and sensory modalities, is thought to underlie the formation of cognitive schemas: compressed internal models that support rapid generalization to novel experiences. Yet the neural circuit mechanisms by which such abstract, identity-independent representations emerge from sequential experience remain largely unknown. Here, we investigate this question using Recurrent Neural Networks (RNNs) as mechanistic models of neural circuits, trained to classify sequences based on their latent algebraic patterns (e.g., seq{aab}, seq{aad} to seq{AAB}; seq{aba}, seq{aca} to seq{ABA}) without supervision on intermediate transitions. We demonstrate that RNNs spontaneously learn low-dimensional representations that mirror the hierarchical generative structure of the sequences, and that this abstraction is mechanistically supported by the emergence of low-rank recurrent connectivity. The leading singular component of the recurrent weight matrix integrates relational transition information, whether consecutive tokens are the same or different, across time, driving the formation of a structured, tree-like geometry in the population state space. Through singular vector ablation, we establish a causal role for this component: removing it selectively erases memory for earlier transitions while leaving local, single-step sensitivity intact. Finally, while RNNs trained on next-token prediction do not spontaneously acquire these abstract representations, transferring the low-rank scaffold learned from classification significantly accelerates learning and improves generalization, an effect specific to the abstract structure of the scaffold rather than generic statistical pretraining. These findings offer a computational account of how task demands shape recurrent connectivity to support temporal abstraction, with direct implications for understanding schema formation in biological brains.

(Comment on this)