(C) PLOS One
This story was originally published by PLOS One and is unaltered.
 . . . . . . . . . .



A framework to determine active neurons and networks within the mouse brain reveals how brain activity changes over the course of the day [1]

['Guanhua Sun', 'Department Of Mathematics', 'University Of Michigan', 'Ann Arbor', 'Michigan', 'United States', 'Tomoyuki Mano', 'Department Of Systems Pharmacology', 'Graduate School Of Medicine', 'The University Of Tokyo']

Date: 2025-11 

The mouse brain’s activity changes drastically over a day despite being generated from the same neurons and physical connectivity. To better understand this, we develop an experimental-computational pipeline to determine which neurons and networks are most active at different times of the day. We genetically mark active neurons of freely behaving mice at four times of the day with a c-Fos activity-dependent TRAP2 system. Neurons are imaged and digitized in 3D, and their molecular properties are inferred from the latest brain spatial transcriptomic dataset. We then develop a new computational method to analyze the network formed by the identified active neurons. Applying this pipeline, we observe region and layer-specific activation of neurons in the cortex, especially activation of layer five neurons at the end of the dark (wake) period. We also observe a shift in the balance of excitatory (glutamatergic) neurons versus inhibitory (GABAergic) neurons across the whole brain, especially in the thalamus. Moreover, as the dark (wake) period progresses, the network formed by the active neurons becomes less modular, and the hubs switch from subcortical regions, such as the posterior hypothalamic nucleus, to cortical regions in the default mode network. Taken together, we present a pipeline to understand which neurons and networks may be most activated in the mouse brain during an experimental protocol, and use this pipeline to understand how brain activity changes over the course of a day.

Competing interests: I have read the journal’s policy and the authors of this manuscript have the following competing interests: H.R.U. is a co-inventor on patent applications covering the CUBIC reagents and a co-founder of CUBICStars Inc.

Funding: We acknowledge the following funding: 1) HFSP (Human Frontier Science Program) RGP0019/2018 to DBF, SB and HU, 2) NSF DMS (National Science Foundation Division of Mathematical Sciences) 2052499 to DBF, and 3) ARO MURI (Multidisciplinary University Research Initiatives) W911NF-22-1-0223 through 5/29/25 to DBF, and 4) Velux Stiftung 1812 to KK. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Data Availability: The data used in the analysis for the paper is shared under the following repository: https://figshare.com/s/5c518ccaedc4dd5baeaa . Some raw files are too large to upload and compressed versions are provided. The code used to analyze these data is also uploaded.

Copyright: © 2025 Sun et al. This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Next, we introduced a novel approach that integrates the neuronal information with the mesoscopic structural connectivity data from the Allen Brain Connectivity Atlas. This approach is based on the assumption that any effective communication between two brain regions requires neurons in both regions to be active and structurally connected. We refer to this derived network as the "active connectivity." We then used established tools from network science [ 26 , 27 ] to analyze the structure of the active connectivity across different time windows, hypothesizing that network hubs, regions with high centrality, would shift over time and play key roles in mediating inter-regional communication. Our analysis revealed notable structural differences in active connectivity across time points, with hub regions at each time window corresponding to brain areas that are known to regulate vigilance and cognitive performance.

To achieve this, we developed a novel experimental and computational framework to identify the most active neurons during different times of the day. We use a Targeted Recombination in Active Populations (TRAP2) system [ 23 ] induced by c-Fos expression to label the active neurons in the mouse brain at four different time windows of the day: the beginning (ZT0-4) and the ending (ZT8-12) hours of the resting period of mice, as well as the beginning (ZT12-16) and ending (ZT20-24) hours of the active period of mice. Subsequently, we combine tissue clearing technologies (CUBIC [ 24 ]) with light sheet microscopy to attain images with active neurons marked by fluorescent signals. We then applied machine learning techniques to segment individual neurons and align the anatomy to the Allen Brain Atlas Common Coordinate Framework (ABA-CCFv3) [ 25 ]. This enables the usage of established spatial transcriptomics data [ 6 ] that is used to identify the molecular properties of each active neuron we identified.

To understand the hierarchical interaction between different brain regions, researchers have identified key hubs and networks in the mouse brain by studying its structural and functional connectivity [ 9 , 10 ]. For primates in general, three core neurocognitive networks have been suggested to play key roles in the regulation of primate behavior: the salience, default mode, and central executive networks (SN, DMN and CEN) [ 11 ]. These networks have more recently been identified in the rodent brain [ 12 – 14 ] and are known to vary in a time-dependent manner [ 15 – 18 ]. While functional connectivity studies have provided valuable insights into the short-term, state-dependent interaction of different brain regions, they fall short in identifying specific neurons or circuits linked to particular behaviors or brain functions. To tackle this limitation, neuronal tagging systems based on genes transcribed shortly after neuronal activation, namely immediate early genes [ 19 ], have been developed [ 20 , 21 ]. Some studies have used such tagging systems to reveal regional variation in active neurons [ 22 ], but no study to date has provided a basic framework that can study how specific neurons or networks are active at a certain time window during an experimental protocol at the whole-brain level.

The physical connectivity of the mouse brain has been mapped with increasing precision, ranging from mesoscopic, regional connectivity [ 1 ], to its voxelized extension [ 2 ], and further to layer-specific and cell-type-specific connectivity in the thalamocortical system [ 3 ]. Most notably, recent efforts have led to the complete reconstruction of a block of the visual cortex, including the wiring diagram of neurons within [ 4 ]. While the structural connectivity of the brain and the transcriptomic profiles of its constituent neurons [ 5 , 6 ] are generally considered stable, both the strength of these connections and the activity of individual neurons can vary significantly throughout the day due to sleep and circadian regulation, environmental conditions, and homeostatic needs [ 7 , 8 ]. Consequently, the neural networks governing brain-wide behaviors across different times of the day emerge from the interplay between structural connectivity and the dynamic activation of interconnected neuronal populations.

2 Results

2.1 Framework to identify, digitize, and analyze active neurons and networks in the mouse brain Our framework to identify, digitize, and analyze the active neurons and networks in the mouse brain is divided into three modules: Experiment and image acquisition. To effectively capture and visualize active neurons, we first crossed mice carrying the TRAP2 system (aka Fos2A-iCreERT2; see Methods) with the Ai14 mice (Ai14(RCL-tdT)-D; also see Methods), thereby establishing a line conditionally expressing tdTomato in c-fos expressing cells (Fig 1a, upper panel). More specifically, as spontaneous neuronal activity triggers the expression of the immediate early gene c-fos, it allows for the c-Fos-driven expression of a Cre protein which is active only in the presence of hydroxytamoxifen (4-OHT). Meanwhile, the same neuron also expresses the fluorescent marker tdTomato in a Cre-dependent manner, thereby finally tagging neurons that are active simultaneously with 4-OHT activity (Fig 1a, bottom panel). Employing this mouse line to examine how active brain connectivity is modified to underlie behavior throughout the 24-hour day, we designed our experiment to tag neurons that are active specifically at circadian windows with known opposing behaviors, such as rest, arousal, and response to the presence of light (Fig 1b). We thus TRAPped neurons at the beginning (ZT0-4) or close to the end (ZT8-12) of their resting period (light period), as well as at the beginning (ZT12-16) or near the end (ZT20-24) of their active period (dark period). After the experiment, all mouse brains underwent extraction, clearing according to the CUBIC tissue clearing pipeline [24], and image acquisition with mesoSPIM (mesoscale selective plane illumination microscopy; [28]) (Fig 1b, bottom right). PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 1. Overall framework to identify, digitize, and analyze the active neurons or networks in the mouse brain. (a) Schematic of the fosTRAP2 x Ai14 system to capture active neurons. Upper panel: mice expressing the fosTRAP2 system were crossed with mice endogenously expressing Cre-dependent tdTomato (Ai14 line). Bottom panel: Neuronal activity induces expression of fos, which promotes expression of CreERT2. In the presence of 4-OHT, CreERT2 is active, thereby cleaving the stop codon from the tdTomato cassette and enabling its transcription. The result is tdTomato expression only in neurons that were active when the 4-OHT was present. (b) Experimental design for data acquisition at different circadian times. Trapping of neurons was induced by intraperitoneal administration of 4-OHT to fosTRAP2 x Ai14 mice during either the light (ZT0 or 8) or the dark period (ZT12 or 20), as depicted by the red arrows. A week later, mouse brains were harvested and underwent the CUBIC-HV whole brain clearing pipeline, including nucleic staining, prior to image acquisition with mesoSPIM and preprocessing with BigStitcher (FIJI package). (c) Sample fluorescent images of one mouse brain from different view angles (Left: horizontal, right top: coronal, and right bottom: sagittal). d) 3D digital reconstruction of the neurons after segmenting them from the fluorescent images and registering them to the Allen Brain Atlas Common Coordinate Framework. e) Matching neurons in our dataset to the established spatial transcriptomic dataset to identify neuron types. f) Construction and analysis of active connectivity at different time windows. https://doi.org/10.1371/journal.pbio.3003472.g001 Neuron segmentation and reconstruction. After obtaining the fluorescent images (see Methods), we used the machine learning algorithm used in [29] to segment all the c-fos expressing neurons based on the fluorescent signals from the stacked 2D images and reconstruct their positions in 3D space (Fig 1c and 1d). Furthermore, we transform the images, as well as the segmented neurons, based on a SyN algorithm (Methods) to the Allen Brain Atlas—Common Coordinate Framework (ABA-CCF; [25]). Aligning the reconstructed neuronal data to ABA-CCFv3 enables us to use other data that is also aligned to the same framework, including anatomical information of the mouse brain, spatial transcriptomics, and connectivity datasets that are essential for our analysis later. Cellomics and network analysis. After reconstructing and aligning our dataset to the Allen Brain Atlas Common Coordinate Framework (ABA-CCF), we developed two complementary computational methods to analyze brain-wide neuronal activity by integrating spatial transcriptomic data and mesoscopic structural connectivity. The first method aligns the spatial distribution of active neurons in our dataset with a reference spatial transcriptomic dataset based on proximity, allowing us to infer the molecular identities of active neurons and examine cell-type-specific activity across different time windows (Fig 1e). The second method constructs an “active connectivity” matrix by integrating regional information of active neurons with the mesoscopic structural connectivity from the Allen Brain Atlas [1]. For each brain, we compute this active connectivity and use established tools from network science to analyze the structural changes of active connectivity across the four periods (Fig 1f). We now present the results derived from our framework.

2.2 c-Fos expressing neurons show variation of neuronal activity across different times of the day After obtaining the 3D fluorescent images, segmenting the neurons and registering them to the ABA-CCF, we obtain the digital 3D reconstruction of neurons of all brains (S1 Fig). With careful observation of all the collected data, we exclude one brain that can not be properly aligned to the Allen Brain Atlas. Moreover, the olfactory area, medulla, and cerebellum of some brains are not intact during the experimental procedure, and therefore excluded from the later analysis. We pick one brain (ZT0-4a) here as an example to show the spatial distribution of the neurons based on their regional information (S2a-c Figs). Horizontal visualizations of 8 brains, two for each time window, are presented in Fig 2a. We see a variation in the total number of neurons that are identified, with most occuring at ZT20-24 and ZT 0-4 (Fig 2b). To avoid potential discrepancies between different images based on the experimental procedures, we calculated the ratio of the number of neurons in each region to the total number of neurons in a brain to investigate its regional composition (Figs 2c and S2). Among the different major brain regions, we found that both the proportion and absolute number of active neurons in the isocortex are significantly higher at ZT0-4, a period with a high amount of sleep compared to ZT12-16, compared to a period when the animals are waking from the rest period (Fig 2c). To our surprise, the isocortex is the only major subregion that shows a significant change in the number of active neurons. PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 2. Variation of active neurons in the whole brain across different time windows. (a) Visualizations of neurons in major brain regions for each time window (2 brains for each time point). The full visualizations of all brains and all brain regions (including the cerebellum, olfactory areas, and the medulla) are presented in S2 Fig. (b) The total number of active neurons in the brain at different time windows during the day. (c) Ratios of different major brain regions for different time windows (ratio is calculated by dividing the number of neurons in each region by the total number of neurons). The ratio of the isocortex is significantly higher at ZT0-4 compared to ZT12-16 (p-value = 0.004), which is further z-scored in the panel inside. The data underlying this figure can be found in S1 Data and at https://figshare.com/s/5c518ccaedc4dd5baeaa. https://doi.org/10.1371/journal.pbio.3003472.g002

2.4 Identifying neuron types with spatial transcriptomic data Next, we leverage a spatial transcriptomic dataset to identify the properties of the identified active neurons. This dataset(Zhuang-ABA1) provides the exact location with respect to the Allen Brain Atlas CCF, as well as specific molecular information, including the neurotransmitter it releases, of more than 9 million cells across the whole mouse brain [6]. The underlying technique used to generate this dataset, multiplexed error-robust fluorescence in situ hybridization (MERFISH) [34], is not known to be significantly influenced by circadian variation, making it a suitable molecular reference for comparison with our dataset. Because Zhuang-ABA1 has a much higher density of neurons, our dataset can be well immersed with it (Fig 5a). We know the exact location and region of each neuron in both datasets, and thus we can calculate the distance between any neuron in our dataset and the Zhuang-ABA1 data. For each neuron in our dataset, we compute its distance to all neurons within the same anatomical region in the Zhuang-ABA1 data, and assign the closest neighbor within that region as its “twin” (Fig 5b). This approach assumes that neurons of the same type are likely to reside in close spatial proximity. We run this matching algorithm only through the same region, eliminating discrepancies that can happen near the region boundary. As an example, we demonstrate this neuron-matching procedure in the primary visual cortex on the ZT8-12D brain (Fig 5c). To evaluate the quality of these matches, we also examine the distribution of distances between matched neuron pairs. As shown in Fig 5d, nearly all neurons in our dataset find a match within 100 , with the majority of matches occurring within 40 . PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 5. Matching active neurons to the spatial transcriptomic dataset. (a) Spatial distribution of the neurons in Zhuang-ABA1 (grey), our ZT8C brain (colored based on region) and merged together in the same coordinate framework. (b) Schematics of finding the twin neuron in the spatial transcriptomic dataset (Zhuang-ABA1) for each active neuron in our data based on spatial proximity. (c) A demonstration of the neuron matching process with the right primary visual cortex of the ZT8C brain. Green points are our data, grey points are Zhuang-ABA1 and yellow points are the twin neurons after matching. (d) The distribution of distances between the twin neurons in Zhunag-ABA1 and the neuron in our dataset for the ZT8C brain. https://doi.org/10.1371/journal.pbio.3003472.g005 After assigning each active neuron in our dataset a “twin” from the reference transcriptomic dataset(Zhuang-ABA1), we assume that the matched neuron shares the same molecular identity as its twin. To validate the accuracy of our neuron-matching approach, we selected two anatomically and molecularly distinct regions: the reticular nucleus (RT), a thalamic region composed almost entirely of GABAergic inhibitory neurons [35], and the ventral tegmental area (VTA), a dopaminergic hub in the hypothalamus [36]. We applied our matching algorithm to the ZT0–4a brain and examined the inferred molecular profiles in these regions (S5a and b Figs). As expected, the method correctly identified a large proportion of GABAergic neurons in the RT, as well as dopaminergic neurons in the VTA (S5c and d Figs). Furthermore, the resulting proportions of GABAergic and dopaminergic neurons in these regions were consistent with values reported in the literature and closely aligned with the distributions observed in the Zhuang-ABA1 dataset, further supporting the validity of our spatial matching framework.

2.5 Cell-type-specific variation of active neurons at different times of the day By applying this method to our dataset, we were able to determine the neurotransmitter profile of all active neurons, enabling a more refined analysis of temporal and regional variation in neurotransmitter composition across the brain. We began by examining the distribution of different neuron types across the entire brain. Notably, we observed a significant shift in the balance between excitatory (glutamatergic) and inhibitory (GABAergic) neurons between ZT12–16 and ZT20–24. The proportion of excitatory neurons was significantly higher during ZT20–24 compared to ZT12–16, whereas the opposite trend was observed for inhibitory neurons (Fig 6a). This shift was also apparent in visualizations in which neurons were color-coded according to neurotransmitter type (Figs 6d and S6). PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 6. Cell-type-specific variation of active neurons across the brain. (a) The ratio of neuronal populations that release different neurotransmitters in the whole brain. (b and c) The ratio of neuronal population of different neurotransmitters in the isocortex and thalamus (TH). (d) Visualizations of neurons based on their neurotransmitter information. We chose two brains for each time segment, which correspond to the brains shown in Fig 2a. The data underlying this figure can be found in S2 Data. https://doi.org/10.1371/journal.pbio.3003472.g006 To further explore the source of this excitatory/inhibitory (E/I) balance shift, we assessed the E/I ratio across major brain regions. Interestingly, the cortical E/I balance remained relatively stable throughout the day (Fig 6b), suggesting that other brain regions—most notably the thalamus—were primarily responsible for the overall change in E/I ratio (Fig 6c). While the cortex as a whole did not show major fluctuations, several cortical subregions did display significant changes in E/I balance, including various regions in the visual cortex (VISp, VISa, VISrl), the ventral anterior cingulate cortex (ACAv), the ventral auditory cortex (AUDv) and the frontal pole (S7 Fig). At the same time, the thalamus exhibited a pronounced reduction in the proportion of active excitatory neurons during ZT20–24, alongside a marked increase in active inhibitory neurons. This regional shift may reflect an increased demand for sleep-related processes late in the day, as inhibitory populations such as those in the reticular nucleus become more active. This result may also be influenced by the relatively uniform neuronal composition observed in regions such as the sensory thalamic nuclei. In contrast, areas like cortical layers 2/3 exhibit a more heterogeneous and intermixed distribution of neuronal subtypes, which may reduce the accuracy of the method.

2.6 Construction of active connectivity Previous observations revealed that active neurons in certain regions, such as the anterior cingulate and anterior insular areas, both components of the salience network, exhibit similar activity patterns. This prompted us to ask whether network-level information could be extracted from our dataset following region-specific analyses of active neurons. To explore this, we leveraged the mesoscopic projectome data from the Allen Brain Atlas, which quantifies the strength of structural connectivity between different brain regions based on viral tracing of EGFP-expressing neurons [1]. To conceptualize communication from one brain region to another, consider the scenario where signals are sent from region A to region B, which has a structural connection density denoted as . If is very low, indicating a lack of structural connectivity, then no signal transmission is expected. However, even when substantial structural connections exist, if region A lacks active neurons, no signal can be initiated. Similarly, if region B lacks active neurons, incoming signals may not be propagated further effectively. Based on this reasoning, we define the “active” connection strength from active neurons in region A to those in region B as the product of three factors: the average structural connection density , the number of active neurons in region A, and the number of active neurons in region B (Fig 7a). PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 7. Construction of the active connectivity. The active connectivity measures the strength of communication between active neurons in different regions (a). By using the normalized connection density between different regions and the regional information of active neurons (b), we attain different active connectivity matrices for different time windows (c). https://doi.org/10.1371/journal.pbio.3003472.g007 Building on this, we compute a directed, weighted connectivity matrix—termed the active connectivity — by multiplying both the rows and columns of the mesoscopic projectome with a vector representing the number of active neurons of different regions (Fig 7b and Appendix). Each entry in this matrix represents the active connection strength from the active neurons of a source region to those of a target region. Using this method, we computed the active connectivity for each brain at different time windows (Fig 7c) and applied standard network analysis techniques to investigate potential differences in network structure across various time windows (Appendix).

[END]
---
[1] Url: https://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.3003472

Published and (C) by PLOS One
Content appears here under this condition or license: Creative Commons - Attribution BY 4.0.

via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/plosone/