Frontiers in Neural Circuits最新文献

Bow-tie architecture (BTA) is widely observed in biological neural systems, yet the underlying mechanism driving its spontaneous emergence remains unclear. In this study, we identify a novel formation mechanism by training multi-layer neural networks under biologically inspired non-negative connectivity constraints across diverse classification tasks. We show that non-negative weights reshape network dynamics by amplifying back-propagated error signals and suppressing hidden-layer activity, leading to the self-organization of BTA without pre-defined architecture. To our knowledge, this is the first demonstration that non-negativity alone can induce BTA formation. The resulting architecture confers distinct functional advantages, including lower wiring cost, robustness to scaling, and task generalizability, highlighting both its computational efficiency and biological relevance. Our findings offer a mechanistic account of BTA emergence and bridge biological structure with artificial learning principles.

Synaptic plasticity underlies adaptive learning in neural systems, offering a biologically plausible framework for reward-driven learning. However, a question remains: how can plasticity rules achieve robustness and effectiveness comparable to error backpropagation? In this study, we introduce Reward-Optimized Stochastic Release Plasticity (RSRP), a learning framework where synaptic release is modeled as a parameterized distribution. Utilizing natural gradient estimation, we derive a synaptic plasticity learning rule that effectively adapts to maximize reward signals. Our approach achieves competitive performance and demonstrates stability in reinforcement learning, comparable to Proximal Policy Optimization (PPO), while attaining accuracy comparable with error backpropagation in digit classification. Additionally, we identify reward regularization as a key stabilizing mechanism and validate our method in biologically plausible networks. Our findings suggest that RSRP offers a robust and effective plasticity learning rule, especially in a discontinuous reinforcement learning paradigm, with potential implications for both artificial intelligence and experimental neuroscience.

The coordination of various brain regions achieves both volitional and forced motor control, but the role of the primary motor cortex in proficient running motor control remains unclear. This study trained mice to run at high performance (>10,000 rotations per day or >2,700 rotations per hour) using a running wheel, and then assessed the effects of the removal of bilateral cortical areas including the primary motor cortex on volitional (self-initiated) and forced (externally driven) running locomotion. The control sham-operated group revealed a quick recovery of volitional running, reaching half of the maximum daily rotation in 3.9 ± 2.6 days (n = 10). In contrast, the cortical injury group took a significantly longer period (7.0 ± 3.3 days, n = 15, p < 0.05) to reach half of the maximum volitional daily rotation, but recovered to preoperative levels in about two weeks. Furthermore, even 3 days after surgery to remove cortical regions, the running time on a treadmill moving at 35.3 cm/s, which is difficult for naïve mice to run on, was not significantly different from that in the sham-operated group. These results suggest that the intact primary motor cortex is not necessarily required to execute trained fast-running locomotion, but rather contributes to the spontaneity of running in mice.

Bimanual movements consist of simultaneous and nonsimultaneous movements. The neural mechanisms of unimanual and nonsimultaneous bimanual movements have been explored in rodent studies through electrophysiological recordings and calcium imaging techniques. However, the neural bases of simultaneous bimanual movements remain poorly understood because of a lack of effective training procedures for such movements in head-fixed rodents. To address this issue, we developed a task in which mice simultaneously pull right and left levers with their forelimbs in a head-fixed condition. Here, we conducted sessions with the link plate in which both levers were mechanically linked to help mice learn the importance of simultaneous bimanual movements. These sessions with the link plate enabled the mice to maintain high success rates even during independent sessions, where the right and left levers could move independently. In these independent sessions, mice were not required to pull both levers at the same time, but rather simply to hold levers simultaneously for a specific period. The mice that experienced sessions with the link plate showed a significantly higher ratio of simultaneous (i.e., lag < 20 ms) than nonsimultaneous lever pulls. In contrast, mice without experience in sessions with the link plate showed no significant increase in simultaneous over nonsimultaneous pulls. This study demonstrates the efficacy of our new task in facilitating repetitive simultaneous forelimb movements in rodents and provides a basis for understanding the neural mechanisms underlying bimanual movements.

The neural circuits of the striatum (caudate and putamen) constitute a crucial component of the extrapyramidal motor system, and dysfunction in these circuits is correlated with significant neurological disorders including Parkinson's disease and Huntington's disease. Many previous studies in rodents revealed the neural connections of the rostral and intermediate parts of the striatum, but relatively fewer studies focused on the caudal striatum, which likely contains both the tail of caudate (CaT) and caudal putamen (PuC). In this study, we investigate the gene markers for the CaT and PuC and brain-wide afferent and efferent projections of the caudal striatum in mice using both anterograde and retrograde neural tracing methods. Some genes such as prodynorphin, otoferlin, and Wolfram syndrome 1 homolog are strongly expressed in CaT and PuC while some others such as neurotensin are almost exclusively expressed in CaT. The major afferent projections of the CaT originate from the substantia nigra (SN), ventral tegmental area, basolateral amygdala, parafascicular nucleus, and visual, somatosensory, auditory and parietal association cortices. The PuC receives its main inputs from the posterior intralaminar nucleus, ventroposterior medial nucleus (VPM), medial geniculate nucleus, and entorhinal, motor and auditory cortices. Both CaT and PuC neurons (including dopamine receptor 1 expressing ones) project in a rough topographical manner to the external and internal divisions of globus pallidus (GP) and SN. However, dopamine receptor 2 expressing neurons in nearly all striatal regions (including CaT and PuC) exclusively target the external GP. In conclusion, the present study has identified the mouse equivalent of the primate CaT and revealed detailed brain-wide connections of the CaT and PuC in rodent. These findings would offer new insights into the functional correlation and disease-related neural circuits related to the caudal striatum.

An accurate diagnostic method using biological indicators is critically needed for bipolar disorder (BD) and major depressive disorder (MDD). The excitatory glutamate α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor (AMPAR) is a crucial regulator of synaptic function, and its dysregulation may play a central role in the pathophysiology of psychiatric disorders. Our recently developed positron emission tomography (PET) tracer, [¹¹C]K-2, enables the quantitative visualization of AMPAR distribution and is considered useful for characterizing synaptic phenotypes in patients with psychiatric disorders. This study aimed to develop a machine learning-based method to differentiate bipolar disorder from major depressive disorder using AMPAR density. Sixteen patients with BD and 27 patients with MDD, all in depressive episodes, underwent PET scans with [¹¹C]K-2 and structural magnetic resonance imaging. AMPAR density was estimated using the standardized uptake value ratio from 30 to 50 min after tracer injection, normalized to whole brain radioactivity. A partial least squares model was trained to predict diagnoses based on AMPAR density, and its performance was evaluated using a leave-one-pair-out cross-validation. Significant differences in AMPAR density were observed in the parietal lobe, cerebellum, and frontal lobe, notably the dorsolateral prefrontal cortex between patients with BD and patients with MDD during a depressive episode. The model achieved an area under the curve of 0.80, sensitivity of 75.0%, and specificity of 77.8%. These findings suggest that AMPAR density measured with [¹¹C]K-2 can effectively distinguish BD from MDD and may aid diagnosis, especially in patients with ambiguous symptoms or incomplete clinical presentation.

N-methyl-D-aspartate receptor (NMDAR) antagonists, including ketamine, phencyclidine (PCP), and dizocilpine (MK-801), are an important class of drugs that can produce antidepressant, hallucinogenic, dissociative, psychotomimetic, and anesthetic effects in humans and animal models. To understand the effects of NMDAR antagonists on the brain, it is essential to map their actions at cellular resolution. We quantified c-Fos expressing cells in the mouse telencephalon after systemic injection of the potent NMDAR antagonist MK-801 and found a 10-fold higher density of c-Fos in the medial entorhinal cortex (MEC) compared to other regions of the telencephalon. c-Fos density was high in layer 3 of the dorsal MEC but low in other parts of the MEC. Since previous studies have shown that parvalbumin (PV) staining shows a strong dorsal-ventral gradient in the MEC, we investigated the spatial correlation between c-Fos and PV staining. We classified PV neurons based on their level of immunoreactivity and found that high and medium PV neurons were positively correlated with c-Fos density, while low PV neurons were negatively correlated. To understand the temporal correlation of c-Fos and PV staining, we examined their expression patterns after MK-801 injections during postnatal development. PV expression emerged on postnatal day 12, preceding c-Fos expression, which emerged on postnatal day 16. Our results suggest that local circuits comprising specific subtypes of inhibitory and excitatory neurons are critical for generating a sustained neuronal response to NMDAR antagonists. Furthermore, a high density of PV neuron input may be a prerequisite for the induction of c-Fos expression observed in MEC principal neurons. This study contributes to our understanding of how the brain responds to NMDAR antagonists in the developing and adult brain and reveals cell types in the dorsal MEC that are highly sensitive to this class of drugs.

Understanding how vagus nerve stimulation (VNS) modulates cortical information processing is essential to developing sustainable, adaptive artificial intelligence inspired by biological systems. This study presents the first evidence that VNS alters the representation of auditory information in a manner that is both layer- and frequency band-specific within the rat auditory cortex. Using a microelectrode array, we meticulously mapped the band-specific power and phase-locking value of sustained activities in layers 2/3, 4, and 5/6, of the rat auditory cortex. We used sparse logistic regression to decode the test frequency from these neural characteristics and compared the decoding accuracy before and after applying VNS. Our results showed that VNS impairs high-gamma band representation in deeper layers (layers 5/6), enhances theta band representation in those layers, and slightly improves high-gamma representation in superficial layers (layers 2/3 and 4), demonstrating the layer-specific and frequency band-specific effect of VNS. These findings suggest that VNS modulates the balance between feed-forward and feed-back pathways in the auditory cortex, providing novel insights into the mechanisms of neuromodulation and its potential applications in brain-inspired computing and therapeutic interventions.

This study explores the efficacy of Bayesian estimation in modeling the orientation and direction selectivity of neurons in the primary visual cortex (V1). Unlike traditional methods such as least squares, Bayesian estimation adeptly handles the probabilistic nature of neuronal responses, offering robust analysis even with limited data and weak selectivity. Through the analysis of both simulated and experimental data, we demonstrate that Bayesian estimation not only accurately fits the neuronal tuning curves but also effectively captures parameter certainty or uncertainty of both strongly and weakly selective neurons. Our results affirm the complex interdependencies among response parameters and highlight the variability in neuronal behavior under varied stimulus conditions. Our findings provide guidance as to how many response samples are necessary for Bayesian parameter estimation to achieve reliable fitting, making it particularly suitable for studies with constraints on data availability.

Like other brain circuits, the brainstem respiratory network is continually modulated by neurotransmitters that activate slow metabotropic receptors. In many cases, activation of these receptors only subtly modulates the respiratory motor pattern. However, activation of some receptor types evokes the arrest of the respiratory motor pattern as can occur following the activation of μ-opioid receptors. We propose that the varied effects of neuromodulation on the respiratory motor pattern depend on the pattern of neuromodulator receptor expression and their influence on the excitability of their post-synaptic targets. Because a comprehensive characterization of these cellular properties across the respiratory network remains challenging, we test our hypothesis by combining computational modeling with ensemble electrophysiologic recording in the pre-Bötzinger complex (pre-BötC) using high-density multi-electrode arrays (MEA). Our computational model encapsulates the hypothesis that neuromodulatory transmission is organized asymmetrically across the respiratory network to promote rhythm and pattern generation. To test this hypothesis, we increased the strength of subsets of neuromodulatory connections in the model and used selective agonists in situ while monitoring pre-BötC ensemble activities. The in silico simulations of increasing slow inhibition were consistent with experiments examining the effect of systemic administration of the 5HT1aR agonist 8-OH-DPAT. Similarly, the effects of increasing slow excitation in the model were experimentally confirmed in pre-BötC ensemble activities before and after systemic administration of the μ-opioid receptor agonist fentanyl. We conclude that asymmetric neuromodulation can contribute to respiratory rhythm and pattern generation and accounts for its varied effects on breathing.