An ultrastructural map of a spinal sensorimotor circuit reveals the potential of astroglia modulation

SUMMARY

Information flow through circuits is dictated by the precise connectivity of neurons and glia. However, how glial-synaptic interactions are arranged within a single circuit to impact information flow remains understudied. Here, we use the local spinal sensorimotor circuit in zebrafish as a model to understand how neurons and astroglia are connected in a vertebrate circuit. With semi-automated cellular reconstructions and automated synaptic mapping of all cells, we identify the precise synaptic connections and the astroglia, including bona fide astrocytes, that physically associate with those synapses. Tripartite synapses are equally distributed across the circuit, and singular astroglia interact multiple times within a single circuit and across circuits. Using this ultrastructural map, we model synaptic thresholding and glial modulation on information flow. We validate this mapping by imaging neurons and glia with GCaMP6s. This work provides a foundational map that offers insights in how glia could influence information flow in complex neural networks.

In brief

Using cell reconstructions and synapse mapping in zebrafish, Koh and Avalos Arceo et. al. reveal a vertebrate local spinal sensorimotor circuit map, revealing how neurons and glia are structurally positioned in a circuit. This resource provides insight into how glia and synaptic thresholding could modulate information flow through complex neural networks.

Graphical Abstract

INTRODUCTION

The precise organization of neuronal circuits is critical for the nervous system. These circuits are established during early embryonic stages. Connectomes of invertebrates like C. elegans and recently Drosophila have contributed immensely to the understanding of the development and information flow in the nervous system.^1–7 While it is well known that precise connectivity is important, the connectome of even the simplest vertebrate circuits is not complete.

Adding a layer of complexity to circuits is the presence of glia. For example, astrocytes in the mammalian brain are predicted to interact with 100,000 synapses.⁸ We know that synapses can be tripartite, involving pre- and post-synaptic neurons and astroglia.^9,10 It is also clear that astroglia can modulate synaptic function in multiple ways, including structural support and neurotransmitter modulation. Astrocytic processes that interact with synapses can also exhibit rapid Ca²⁺ microdomains in response to neuronal activity.^10–13 Despite these reports, previous circuit maps have rarely, if ever, included astrocytes. Without an understanding of how neurons and glia are connected in a vertebrate circuit, it is challenging to develop a complete model of how circuits are constructed or how information could flow and be modulated in a vertebrate circuit. This knowledge is foundational and is a necessary resource to understand more advanced topics about circuit development and functionality in vertebrates.

As a model for a simple vertebrate circuit, we focused on the local spinal sensorimotor circuit.^14–16 This sensorimotor circuit is thought to transfer information sequentially from sensory neurons to interneurons to motor neurons, similar to the spinal reflex arc.^15,17 The sensory information from the periphery, however, must be integrated and processed within this circuit before being relayed to the motor neurons. Identifying the complete connectome of a vertebrate sensorimotor circuit is an important step to understand how circuit structure underlies function and supports precise and appropriate motor output in the spinal cord.

We used a serial electron microscopy dataset to reconstruct the glial and neuronal organization of a vertebrate sensorimotor circuit using semi-automated cell reconstructions and automated synapse identification. This work revealed that DRG neurons connect with multiple interneuron populations in the vertebrate spinal cord. The level of complexity in the circuits expands significantly at the interneuron population, with interneurons exhibiting hundreds of synapses. Eventually, interneurons connect with motor neurons. Adding a layer of potential modulation, we demonstrate that astroglial processes are equally distributed across synapses in the circuit. Finally, we validate this connectome resource with functional imaging of GCaMP6s. These findings provide a foundational resource of a vertebrate circuit and demonstrates the potential for complex and multi-layered modulation that includes glia and neurons. Such modulation is critical to ensure animals respond precisely to sensory input.

RESULTS

Reconstruction of neurons and synaptic connections within the sensorimotor circuit

To understand the organization of glia within a simple vertebrate circuit, we used the zebrafish sensorimotor spinal circuit as a model.^18–22 We first reconstructed the dorsal root ganglion (DRG) neurons in zebrafish from a serial block-face electron microscopy dataset of a 6 days post fertilization (dpf) spinal cord.²³ This region spanned 74 × 74 × 207 μm³ (voxel size 9 × 9 × 21 nm³) (Figure 1A), which has been used previously to skeletonize motor neurons.²³ We employed semi-automated reconstruction approaches to reconstruct neurons in the dataset (Figures 1B and 1C).²⁴ While the semi-automated approach was efficient for reconstructions of spinal neurons, a subset of neurons prone to inaccurate segmentation-mergers were manually reconstructed. To determine the exact connectome of the circuit, we also used computational approaches to map all of the synapses in the spinal cord section with automated synaptic mapping.²⁴ This approach previously yielded precision-recall scores between 0.91 and 0.98 on synaptic vesicle clouds and 0.88–0.98 on synaptic clefts on the zebrafish brain dataset at 14 × 14 × 25 nm³. To address potential false positives, each potential synapse in the circuit was manually verified to have both synaptic vesicles and synaptic clefts. Only synapses with both vesicles and a detectable cleft were included for the circuit reconstruction. We additionally checked for false negatives by comparison to ground truth and detected a recall score of 0.95 for automatic identification of synaptic clouds in our dataset. Using the reconstructions and synaptic locations, we then systematically reconstructed the vertebrate local spinal cord sensorimotor circuit.

Figure 1. Reconstruction of neurons and synaptic connections within the sensorimotor circuit.

(A) Electron microscopy (EM) image of the 6 dpf zebrafish spinal tissue that was used to reconstruct the local spinal sensorimotor circuit. Green arrowheads denote the location of the DRG. Scale bar: 70 μm.

(B) EM images of a DRG neuron (pseudocolored green) at its cell soma (Nu, nucleus), at the PNS/CNS interface, and at its bifurcation location. Scale bars: 4.5 μm (left) and 2 μm (middle and right).

(C) Reconstructions of all DRG neurons shown in (A), demonstrating their precise morphology.

(D) Reconstruction of the DRG neurons on both sides of the spinal cord overlaid with their synaptic sites. Blue dots denote pre-synaptic sites, and pink dots mark post-synaptic sites.

(E) Quantification of the number of post-synaptic partners of DRG neurons in the tissue section. Note that CoPA synaptic connections are over-represented.

(F) Reconstructions and EM images of synaptic sites showing that a simple circuit of a DRG, interneuron, and motor neuron could relay information both ipsilaterally and contralaterally. Second-level connections indicate a that single interneuron connects the DRG with a motor neuron, whereas the third level requires two interneuron connections. The schematic shows outline of the spinal cord and central canal (gray). Scale bars: 1 μm (second-level ipsilateral and contralateral flow and third-level ipsilateral flow) and 1.5 μm (third-level contralateral flow).

(G) Reconstructions and EM images of synaptic sites at distinct locations of the neurons. Scale bar: 1.5 μm.

(H) Quantification of the distribution of synaptic locations at each level of the spinal sensorimotor circuit, showing that most synaptic connections occur between axons and somata at all levels (DRG>IN - 1 Axon-Axon:3 Axon-Soma:33 Axon-Dendrite, IN > IN - 110 Axon-Axon:10 Axon-Soma:229 Axon-Dendrite, IN > MN - 67 Axon-Axon:113 Axon-Soma:456 Axon-Dendrite).

In total, 13 DRG neurons were connected into the sensorimotor circuit (Figures 1B and 1C). The DRG neurons were symmetrically bilateral and extended the length of the tissue section (Figures S1A–S1E). The DRGs were both pre-synaptic and post-synaptic to neurons in the spinal cord and did not display any obvious clustering of synapses (Figures 1D and S1E–S1I). While synapses are known to occur on secondary branches and were identified elsewhere (Figures S1J–S1L), we did not detect synaptic sites at secondary branches of DRG neurons. Instead, we found that DRG pre-synaptic sites occurred on the main axonal shaft and thereby represented en passant synapses (Figures 1D and S1M). Volume measurements of these synaptic axonal sites revealed local swellings in which synaptic sites were 44% wider than the surrounding axonal shaft (Figures S1N–S1Q). To identify the specific interneuron subtypes with which the DRG neurons connect, we cross-referenced our reconstructions with previous morphological descriptions of interneurons in the zebrafish spinal cord (Figures S2A and S1B).^19,25–29 Connections between DRG and CoPA interneurons were over-represented in the circuits (Figures 1E and S2C). Reconstructions of the connected interneurons revealed that en passant DRG synapses could form on cell somata and dendritic processes of the interneurons (Figures 1G and 1H). 196 unique interneurons were connected in these circuits (Figure S2A). The DRG-to-interneuron connection maps showed that information could flow both ipsilaterally and contralaterally through distinct interneurons (Figure 1F). On average, the interneurons exhibited 217.75 ± 27.63 synaptic connections. These diverse interneurons eventually synapsed on 133 motor neurons (Figures 1F and 1G). Varying subtypes of motor neurons were represented in the map within 3 synaptic connections of the DRG axons (Figures 1F and S2D).^19,23,30,31 The majority of synapses in the circuit occurred between axons and dendrites, but we also detected axon/soma and axon/axon synapses at different levels of the circuit (Figures 1G and 1H).

Reconstruction of astroglia in the spinal sensorimotor circuit

To then understand the potential role of glia in a circuit, we extended the map to include glia interactions that could impact information flow through the circuit (Figure 2). In particular, we considered that synapses in the circuit could be tripartite, with pre- and post-synaptic neuronal partners in close interaction with astroglial processes.^10,32 The zebrafish spinal cord at 6 dpf has astrocytes that are positioned in the synapse-rich region where we mapped the sensorimotor circuit.^33,34 To add this additional glial layer, we identified whether each synapse in the circuit was contacted by an astroglial process. We defined tripartite synapses as those where an astroglia process was directly adjacent to the synaptic cleft region (Figure 2A). Reconstructions of these astroglial cells demonstrated a robust bushy morphology reminiscent of mammalian and Drosophila astrocytes and consistent with morphological characterization from light microscopy (Figures 2A and 2B).^10,35–37 Astroglial processes that interacted with synapses were also present in a subset of astroglia with a radial morphology that extended to the edge of the spinal cord but also had projections that were in the synapse-rich region of the spinal cord (Figures 2A and 2B).^38,39 While there were astroglial interactions with neuronal processes in non-cleft regions, we did not include such interactions in our analysis. We identified that 23.9% of synaptic connections (293 of 1,224 total) were tripartite across 3 levels of the sensorimotor circuit.

Figure 2. Reconstruction of astroglia in the spinal sensorimotor circuit.

(A) EM images and cellular reconstructions of synaptic sites within the sensorimotor circuit that do and do not contain astroglia. Note that astroglia can have a radial morphology or a protoplasmic morphology. Pseudocolors correspond with the neuronal subtype. Astroglia are shown in green. Scale bar: 750 nm.

(B) Confocal images of astrocyte types in comparison to the EM cellular reconstructions, showing distinct astrocyte morphotypes in the spinal cord. The schematic shows an outline of spinal cord and central canal (gray). Scale bar: 10 μm.

(C) Quantification of the percentage of synapses that are tripartite vs. bipartite between DRG neurons and interneurons (DRG > IN - 8 tripartite:29 bipartite), between interneurons in the sensorimotor circuit (IN > IN - 89 tripartite:260 bipartite), and between interneurons and motor neuron in the sensorimotor circuit (IN > MN - 162 tripartite:474 bipartite).

(D) EM images and cellular reconstructions of DRG/IN, IN/IN, and IN/MN synapses, showing that tripartite synapses can be at each circuit level. Scale bar: 1 μm.

(E–G) Quantification of the percentage of astroglia subtypes at each level of the sensorimotor circuit (cumulative - 96 protoplasmic:58 glial limitans, DRG>IN - 0 protoplasmic:7 glial limitans, IN > IN - 27 protoplasmic:19 glial limitans, IN > MN - 69 protoplasmic:32 glial limitans) (E), number of astroglia subtype seen within individual sensorimotor circuits (F), and number of tripartite synapses per astroglia subtype in the local spinal sensorimotor circuit (G).

(H) Number of tripartite synapses that are predicted to be excitatory vs. inhibitory.

(I and J) Quantification of the axonal area of bipartite and tripartite synapses for pre-synaptic neurons (I) and post-synaptic neurons (J).

(K) Distribution of tripartite synapses locations at each level of the circuit (DRG>IN - 0 axon-axon:0 axon-soma:8 axon-dendrite, IN > IN - 32 axon-axon:1 axonsoma:56 axon-dendrite, IN > MN - 14 axon-axon:33 axon-soma:115 axon-dendrite). t test was used in (F)–(J). All data are represented as mean ± SEM. Chi-square (E), T-test (F–J)

Although we know that synapses are tripartite, whether specific synaptic connections within a circuit are tripartite is poorly understood. To answer this for the local sensorimotor circuit, we subclassified the astrocyte-associated synapses between sensory-interneuron, interneuron-interneuron, and motor-interneuron (Figures 2C and 2D). These results revealed that tripartite synapses were equally distributed across all levels of the circuit (Figures 2C and 2D). Both protoplasmic and glial limitans astroglia subtypes interacted with synapses, although sensory neuron-to-interneuron synapses were exclusively contacted by glial limitans morphotypes (Figure 2E). It is well known that astrocytes have extensive morphology, but it is unclear how many astrocytes interact within a given local circuit. To investigate this for the sensorimotor circuit, we asked how many astrocytes were present in each single DRG neuron’s circuit and whether single astrocytes form multiple tripartite synapses within the circuit. These results revealed that 20.1 ± 2.5 astrocytes are present within each circuit and that a given astrocyte forms 4.06 ± 0.5 tripartite synapses within the circuits (Figures 2F and 2G). These results support the idea that multiple astrocytes could control synaptic transmission through the sensorimotor circuit or formation of it. To identify whether tripartite synapses were positioned in specific areas of a single circuit, we subcategorized the tripartite synapses in various ways according to their neuronal partners. We first asked whether tripartite synapses were biased toward pre-synaptic partners that would be expected to be excitatory vs. inhibitory (Figure 2H). Since zebrafish synapses in electron microscopy (EM) preparations cannot be distinguished by morphological features, similar to Drosophila synapses, we utilized previous morphological analysis and neurotransmitter expression analysis to classify each neuron (Figure S2E).^19,25–29 We assumed Dale’s law: each neuron has only one neurotransmitter profile.^40,41 Tripartite synapses were present at 46% of predicted excitatory vs. 54% of inhibitory neurons, suggesting equal distribution across neuron types. We also compared the axonal areas between bipartite and tripartite synapses and found no significant differences for both pre- and post-synaptic areas, suggesting similar synapse morphology (Figures 2I and 2J). Finally, we determined that tripartite synapses were located at synaptic connections between axon-dendrite, axon-soma, or dendrite-dendrite in proportions similar to the overall synaptic distribution in the circuit (Figure 2K). Together, these results reveal the precise and extensive network of glial interactions within a single circuit that could impact the processing of sensory information.

Neuronal and astroglia organization in single sensorimotor circuits

Now that the glial and neuronal organization was mapped, we compiled it to understand the potential information flow. We represented the circuit with reconstructions at different levels, a heatmap of connected neurons, and a Sankey plot that showed information flow based on synaptic connections (Figures 3A–3C). Level 1 connections were interactions directly between DRG neurons and their partners (Figures 3A and 3C). The second level represents the interconnectivity of interneuron connection (Figures 3A–3C). Finally, the third level exhibited the extensive connections that ended at motor neurons (Figures 3A–3C). Even looking at a single DRG neuron’s circuit reconstructions demonstrated the potential for complicated flow of information. In the representative DRG neuron, the neuron is connected to two excitatory neurons, CoPA and CiA, and an inhibitory neuron, CoSA (Figures 3A and 3C). Examining the second level connections identifies that information from a single DRG neuron can travel to a motor neuron on the ipsilateral side through the CiA neuron (Figure 3A). It also revealed that the CoSA neuron could relay that information to the contralateral side of the spinal cord (Figure 3A). The second level of connections demonstrated a network of interneuron-to-interneuron connections that increased the complexity of the circuit (Figure 3A). Finally, the third level of connections demonstrated that information could travel both ipsilaterally or contralaterally to eventually impact motor neurons (Figure 3A). For even a single DRG neuron, the circuit is complex, connected through 155 synapses (Figures 3A–3D). For example, the motor neuron on the ipsilateral side can be excited by an interneuron. The activity of the motor neuron on the contralateral side can be changed as well, either by predicted excitatory interneurons that are commissural or through two inhibitory neurons in succession (Figures 3C and 3D). The number of times a cell is synaptically connected in a circuit could indicate its influence in that circuit. To probe this concept and determine how many cells and synapses per cell are present within a single DRG neuron’s circuit, we aggregated those numbers into a heatmap (Figure 3B). The heatmap demonstrates that, while many cells are only connected in the circuit through one synapse, others make multiple synaptic connections (Figure 3B). Aggregating the classes of neurons in a Sankey plot provided an understanding of how neuronal information could flow through the local circuit. For example, the local circuit relays information directly to motor neurons within two synaptic connections but also expands significantly through the connections of CiA interneurons (Figures 3C and S3A). Last, we mapped the predicted excitatory vs. inhibitory neurons that are present in a single DRG neuron’s circuit (Figure 3D).^19,25–29 This demonstrated that excitatory information could flow quickly to motor neurons and further refined the understanding of how information flows within the local spinal circuit.

Figure 3. Organization of neurons and astroglia in single sensorimotor circuits.

(A) Cellular reconstructions of neurons in the local spinal sensorimotor circuit from a single DRG neuron’s circuit. The schematic of the neurons is color coded based on the excitatory (green), inhibitory (red), or mixed (blue) identity of the neurons, DRG neurons (yellow), and motor neurons (pink). The schematic shows an outline of the spinal cord and central canal (gray).

(B) Heatmap showing the number of synapses for each cell in a single DRG neuron’s circuit. Each row represents an individual neuron of the bracketed identity.

(C) Schematic showing potential information flow, demonstrating each of the classes of neurons and how that could be impacted when incorporating excitatory or inhibitory neurons. The size of the bar represents the number of synapses that connect the classes of neurons.

(D) Schematic of a single DRG neuron’s local circuit, demonstrating the predicted excitatory (blue), inhibitory (magenta), and mixed connections (blue/magenta).

(E) Cellular reconstructions of neurons and astroglia in the circuit depicted in (A). All astroglia (sky blue) that are associated with the synapses in that circuit are included.

(F) Schematic showing tripartite (purple) vs. bipartite synapses (orange) within a single DRG neuron’s circuit that is depicted in (A).

(G) Heatmap of the number of synapses with which each astroglia interacts in the DRG neuron circuit depicted in (A). Each row represents an individual astroglia of the bracketed identity.

To determine how astrocytes could contribute to the complexity of the circuit, we added the astrocytic interactions onto the circuit of a single DRG neuron (Figures 3E and S3B). These results revealed where astrocytes could modulate the circuit. In just the 3 levels, astrocytic processes contacted multiple synapses. In particular, we revealed that astrocytes are in structural proximity to modulate the first level of synaptic connections between the DRG neurons and the interneurons (Figure 3E). To represent the control of astrocytes on a single DRG neuron circuit, we collapsed the synapses onto a single map (Figure 3F). These results revealed a network of synapses that would be expected to be excitatory and inhibitory neurons (Figure 3D). The collapsed circuit demonstrates a precise map of tripartite synapses vs. bipartite synapses (Figure 3F). It is possible that astrocytic modulation could be controlled by a collection of astroglia or by a single highly connected astroglia. To explore this, we created a heatmap of how many times each astrocyte participated in a tripartite synapse within a single DRG neuron circuit (Figure 3G). Although a portion of astroglia only form one tripartite synapse in the circuit, at least 3 astroglia had the potential to modulate multiple synapses within the circuit (Figure 3G). Incorporating astrocytes into the map of the sensorimotor circuit demonstrates a complex circuit that could be modulated by both neurons and glia at multiple layers.

Spinal sensorimotor circuits are interconnected

Spinal somatosensory responses have spatial acuity, but interneurons also extend through multiple hemisegments, suggesting that spatial acuity might be in part coded by interconnected circuits. To next explore the extent of interconnectivity, we compiled the maps of all DRG neurons in the tissue segment (Figure 4). To understand the scale of information flow at each level, we created a Sankey chart where the number of synapses was collapsed at each level (Figures 4A and S4A). These results demonstrated that the synaptic connections expanded significantly at the interneuron level (Figure 4A). This potential information flow funneled into the motor neurons. Layering glial cells onto this information flow map demonstrated that glia had potential to modulate the circuit at all phases equally (Figure 4A). To next understand the directional flow of potential information, we generated a map where individual synapses of DRG neurons are mapped onto interneuron and motor neuronal subtypes (Figure 4B). These results demonstrated DRG neurons synapse onto most interneuron subtypes (Figure 4B). Even within the 3 segments of the spinal cord that we mapped, there was massive interconnection between DRG circuits. For example, synapses within the integrated sensorimotor circuits existed between CiA, CoSA, CoPA, CoBL, CiD, MCoD, UcoD, and COLA interneurons and motor neurons (Figure 4B).

Figure 4. Spinal sensorimotor circuits are interconnected.

(A) Sankey plot of the DRG, interneuron (IN), and motor neuron (MN) connections aggregated from all DRGs in the tissue section to show where information can expand in the circuit. The height of the bar represents the number of synaptic connections at that layer of the circuit.

(B) Heatmap of every neuron that is connected within the local spinal sensorimotor circuit. Colors denote the number of synapses on each given neuron within each circuit. Columns are organized to demonstrate each DRG neuron’s circuit. Classes of neurons are clustered together. The outflux column shows the cumulative amount of post-synaptic connections a cell has in all circuits.

(C) Sankey plot of DRG R1B and L3B, with the lines representing information flow colored based on whether they are unique to R1B (orange) or L3B (blue) or shared (purple) to demonstrate how interconnected each DRG neuron’s circuit is.

(D) Heatmap of every astroglia that is connected within the local spinal sensorimotor circuit. Colors denote the number of synapses with which each astroglia interacts in each single DRG neuron’s circuit.

(E) Quantification of the neurons in each DRG neuron’s circuit if information flow was controlled by the number of synapses required to induce neuronal activity. Note that, as the threshold of synapses increases, the number of neurons in a circuit decreases.

(F) Quantification of the neurons in each DRG neuron’s circuit if flow of information was controlled by astroglia.

(G) Sankey plot of potential information flow if the number of synaptic connections on a given neuron controls the flow. Represented are the interconnected circuits of R1B and L3B depicted in (C). ≥1 and ≥2 demonstrate the drastic change in information flow when synapse thresholding is applied. Also depicted is the information flow if astroglia could modulate the circuits of R1B and L3B. The astroglia control plot assumes that synaptic thresholding is ≥ 1.

Finally, to understand how individual neurons were represented in the integrated map, we generated heatmaps where each neuron in the collective circuits was colored based on how frequently it was represented within the circuits (Figure 4B). The heatmaps revealed that individual DRG circuits contain both unique synaptic connections for that circuit and shared synaptic connections with other DRG circuits (Figure 4B). An example of this is DRG neurons R1B and L3B, which are on opposite sides of the spinal cord but share a common CoBL neuron (Figures 4B, 4C, and S4B). While there is convergence onto single neurons, the DRG neuron circuits also display circuit divergence. This is prominent with CiA and CiD neurons, which are typically exclusive to a single DRG neuron’s circuit (Figure 4B). This concept is highlighted by again contrasting DRG neurons R1B and L3B, with L3B connecting to 73 unique connections between neurons compared to R1B (Figure 4C). To determine the potential of glia to modulate this interconnectivity, we asked how glia are positioned within these interconnected circuits by performing a similar analysis for glia. This showed distinct astroglia that had potential to modulate multiple synapses within a single circuit and between distinct local circuits (Figure 4D). Interestingly, protoplasmic astrocytes were more interconnected than glial limitans astroglia (Figure S4C). Together, these results demonstrate a complex sensorimotor circuit that has both potential for convergent and divergent information flow through spinal neurons that could be modulated by glia.

With the massive interconnectivity of the neurons in local circuits, it seemed possible that the simple connectivity of neurons could not provide acuity in sensory responses. To explore how sensory acuity could be coded in the spinal cord, we considered the hypothesis that the number of synaptic connections for a given neuron could refine the number of neurons involved in a circuit.^5–7,42 To test this idea, we determined how many neurons were in each circuit if the threshold for firing a neuron was based on the number of synaptic inputs. While the circuit was complex if an input threshold of one synapse was required, connecting 117 ± 12.2 neurons, thresholding the system to different levels of synaptic abundance reduced the total number of neurons in the circuit (2 synapses: 46.6 ± 6.6 [39.8% ± 7.0%], 3 synapses: 21.8 ± 3.9 [18.6% ± 3.9%] (Figure 4E). Astroglia have been proposed to provide inhibitory modulation, which could function to reduce the influence of neurons in a circuit.^43,44 To determine how glia could function in this thresholding, we created similar calculations by subtracting the number of synaptic connections that contained glia. Thresholding with glia reduced the circuits on average from 117 ± 12.2 to 60.3 ± 6.9 neurons without any additional synaptic thresholding (Figure 4F), demonstrating the potential control that glia could have on circuit complexity and significantly reducing the scale of thresholding required to simplify a circuit.

While the number of neurons in a circuit provides an understanding of circuit complexity, it does not incorporate the directional flow of information through circuit levels. To determine how the information flow could be altered from thresholding of synaptic number, we generated flow charts of every neuron in two interconnected DRG circuits (R1B and L3B) (Figure 4G). Theoretically, circuit information could flow through 217 connections between neurons if only 1 synapse is required (Figure 4G). To determine how thresholding impacted direction flow, we generated flow charts with thresholding at 2 synapses, which caused the sharpest decline in circuit complexity (Figure 4E). Thresholding the synapses to >2 downstream of the first-level DRG to interneuron synapses reduced the number of neurons that would mediate information flow to a single UCoD interneuron (Figure 4G). This interneuron is then connected to 2 interneurons and a single motor neuron.

To again determine how glia could modulate this information flow, either positively or negatively, we generated a map that labeled tripartite connections and any downstream information flow (Figure 4G). The extent of this could be visualized by color coding the Sankey plot so that synaptic connections with astrocytes and downstream information flow from those are grayed, with information flow without astrocyte control labeled in green (Figure 4G). Although the complexity of the circuit is not reduced as much as synapse thresholding, the presumed inhibitory astroglia control does have the potential to simplify the circuit compared to no synapse thresholding (Figure 4G). To test how astrocyte control could change the interconnectivity (in either an excitatory or inhibitory way), we assumed that information flow could be altered by astrocytes. This control would significantly alter the interconnectivity of R1B and L3B DRG neurons’ circuits, shifting the “shared” portion of the circuit by a tripartite synapse located between DRG R1B and an interneuron (Figure S4D). These results revealed an impressive potential for astroglia to modulate information flow within and between two circuits. The potential of glial modulation, assuming it is negatively regulating circuit activity, is similar to increasing the synaptic threshold, highlighting the potential of two cooperative mechanisms to modulate circuit information flow.

GCaMP6s reveals interconnected circuits

The interconnected nature of the connectome predicted that, if DRG neurons were simultaneously excited, then a large portion of spinal neurons and glia would also be excited or active. To test this, we induced the activity of DRG neurons using cold water immersion and then imaged GCaMP6s as a proxy for neuronal and glial activity.^22,45,46 This sensory assay causes a shivering response by 3 dpf that is dependent on intact afferent axons that connect the DRG neurons with the spinal cord.^20,22,46 This imaging was performed with Tg(neurod:GAL4); Tg(UAS:GCaMP6s) animals at 5 dpf, which express GCaMP6s nearly pan-neuronally in the spinal cord and DRG.⁴⁵ We utilized the neurod promoter because it still provided sparse labeling that allowed us to identify single neuron cell bodies in the packed spinal cord.⁴⁵ As a control to test that the addition of water itself does not cause DRG neuronal activity, we first exposed each animal to 23°C water before 4°C water (Figure S4E).⁴⁶ These results confirmed that GCaMP6s fluorescence changes only after administration of 4°C water. To better define the dynamics of this, we then captured high-speed confocal z stacks that spanned 40 μm of the spinal cord every 0.971 s before and after submersion in 4°C water (Figure 5A). The dynamics of GCaMP6s does not allow us to detect the order of activity but does indicate whether two neurons are likely activated within a similar period of time and within the fluorescence dynamics of GCaMP6s. For each movie, we calculated the Z score of the GCaMP6s intensity over all time points and defined an active neuron as having a Z score greater than 1.0.^45,46

Figure 5. GCaMP6s imaging reveals interconnected circuits.

(A) Representative confocal images and quantification of DRG and spinal neurons when uninjured Tg(neurod:GAL4); Tg(UAS:GCaMP6s) animals are exposed to 4°C water. Quantifications of neuronal GCaMP transients are represented as Z scores on the corresponding line graphs, where each line represents one cell. The color of these lines indicates their response to cold water exposure, signifying when the Z score was >1.0. DRG neurons with a rapid change in Z score >1.0 are colored red or navy blue if delayed (Z score only >1.0 one or more frames after initial cold water submersion). Spinal neurons with rapid changes are colored green, blue if delayed by 1 s, light blue if delayed by 2 s, pink if delayed by more than 2 s, and neurons that do not change (Z score never >1.0) are black. Note that exposure to 4°C water causes DRG and spinal neurons to rapidly increase GCaMP6s intensity. Scale bar: 10 μm.

(B) Sankey plot showing the flow of information from DRG neurons to interneurons and then to motor neurons in the volumetric EM that contains 3 hemisegments. The width of the bar represents the number of synapses for that given class of neuron. Colors are randomly assigned.

(C) Representative confocal images and quantifications of DRG and spinal neurons of Tg(neurod:GAL4); Tg(UAS:GCaMP6s) animals 3 h after undergoing serial axotomy. Note that DRG neurons still show a Z score >1.0 after 4°C but fewer spinal neurons show rapid Z score changes >1.0. Scale bar: 10 μm.

(D and E) Quantifications for the percentage of DRG neuron (D) and spinal neuron (E) responses in the uninjured (n = 6) and 3 h post-injury (hpi) (n = 7) animals. Fisher’s exact test was performed to compare the percentages of rapidly active and delayed activity in DRG neurons (D, p = 0.5546, Fisher’s exact) and spinal neurons (E, p < 0.0001, Fisher’s exact) between the uninjured vs. injured animals.

(F) Schematic of the serial axotomy paradigm, where 8 consecutive DRG central-projecting nerves (DRGs 4–11) were disconnected from the spinal cord.

The results revealed that, in laterally mounted 5 dpf animals, 66.67% ± 10.85% of individual DRG neurons fired in the first capture immediately following 4°C water exposure (Figure 5D). These experiments also revealed that 71.53% ± 6.739% of detectable spinal neurons fired simultaneously with DRG neurons (Figures 5A and 5E). Imaging demonstrated that these spinal neurons were responding in both the dorsal and ventral locations of the spinal cord (Figures 5A and S4E), consistent with the locations of interneurons and motor neuron cell bodies. These results were predicted by our serial electron microscopy mapping of the circuits, which demonstrated a vast network of interneurons that were connected to the DRG neurons in the section (Figures 5B and S4A). While most spinal neurons fired synchronously with DRG neurons, 27.33% ± 6.393% of spinal neurons were not rapidly activated with the DRG neurons and fired 1 or more seconds after the DRG (Figures 5A and 5E), consistent with our ultrastructural reconstruction.

To test whether this spinal neuron activity was propagated from the DRG neurons, we disconnected 8 DRG neurons from the spinal cord circuit via serial axotomy to the centrally projecting axons of DRG neurons in spinal segments 4–11 (right side) at 5 dpf (Figure 5F)^22,47,48 and then repeated the cold water assay 3 h post injury (3 hpi) (Figure 5C). We previously identified that these nerves do not regenerate.²² When DRG neurons are disconnected, 37.31% ± 10.73% of spinal neurons were rapidly responsive to 4°C exposure, and 54.62% ± 10.15% had delayed activity (Figures 5C and 5E), demonstrating a significant decrease in the proportion of active spinal neurons, consistent with the idea that spinal neuron activity is dependent on DRG information flow. Despite altered spinal neuron activity, axotomy did not alter the rapid GCaMP6s activity of DRG neurons (Figure 5D).

GCaMP6s demonstrates astroglial response after stimulus

Our circuit reconstruction of the glia indicated that they had the potential to be integrated into the somatosensory circuit. To test this possibility, we generated a new transgenic animal, Tg(gfap:GCaMP6s-CAAX), that expressed GCaMP6s in astroglia using the regulatory regions of gfap.³⁹ We then repeated the cold water immersion assay and recorded the activity of astroglia in the spinal circuit. GCaMP6s transients were not detected during 23°C exposure but were robustly present after exposure to 4°C water (Figures S5A–S5C and 6A). The robustness of the GCaMP6s signal prevented quantification of single astroglial cells, so we instead quantified the overall area (Figure 6B) and fluorescence intensity (Figures 6C and 6D) of GCaMP6s fluorescence changes above threshold as an astroglial population response. These results demonstrate an immediate increase in GCaMP6s intensity that continued to rise at sequential time points (Figures 6B–6D). To determine whether this robust response could be predicted by the EM analysis, we segmented all astroglia located at tripartite synapses in the reconstructed DRG sensorimotor circuits (Figure 6E). The GCaMP6s response versus the EM prediction were qualitatively comparable.

Figure 6. GCaMP6s imaging demonstrates astroglial response after stimulus.

(A) Representative confocal images of Tg(gfap:GCaMP6s-CAAX) animals before and after cold water submersion. Scale bar: 20 μm.

(B–D) Quantifications of the glial GCaMP6s fluorescence intensity above threshold in Tg(gfap:GCaMP6s-CAAX) animals, displayed as changes in area above threshold (B), as well as in fluorescent intensity above threshold (Z score) in a heatmap (C) and line graph (D) before and after 4°C water exposure.

(E) Cellular EM reconstructions of all astroglia that interact with synapses in the aggregate DRG neuron’s circuits.

(F) Representative confocal images and quantifications of gfap:GCaMP6s transients in uninjured and 3 h post serial axotomized Tg(gfap:GCaMP6s-CAAX);Tg(ngn1:GFP) animals before and after cold water submersion. Scale bar: 20 μm.

(G and H) Quantification of changes in area (G) and fluorescence intensity (H) of glial GCaMP transients above threshold in uninjured animals (n = 5).

(I and J) Quantifications of GCaMP transients before and after cold water submersion regarding changes in area (I) and fluorescence intensity (J) above threshold in 3 hpi animals (n = 6).

(K) Quantification of ΔF/F₀ in the uninjured (n = 5) and injured (n = 6) animals, comparing fluorescent integrated densities before and after 4°C water submersion (Mann-Whitney test, p = 0.0079). Data are represented as mean ± SEM.

(L–N) Representative confocal images (L) and quantification of fluorescence intensity of single glial cells (created by injecting gfap:GCaMP6s-CAAX) before and after 4° C water submersion. Scale bar: 10 μm. Single gfap:GCaMP6s⁺ cell changes in fluorescence intensity are represented as Z scores in a line graph (M) and heatmap (N).

If astroglia were responding to the DRG circuit, disruption of that circuit would be expected to reduce astroglial GCaMP6 dynamics. To test this, we performed serial axotomy on 8 DRG central projections in Tg(gfap:GCaMP6s-CAAX); Tg(ngn1:GFP) animals (Figure 5F) and then repeated the 4°C submersion assay. This quantification of GCaMP6s dynamics revealed a reduction in the astroglial response to 4°C compared to uninjured controls, consistent with the idea that the astroglial GCaMP6s dynamics are dependent on DRG stimulus (Figures 6F–6K). Finally, to determine how individual astroglia were impacted by 4°C submersion, we mosaically labeled them by injecting a gfap:GCaMP6s-CAAX construct and repeated the submersion assay. While this experiment was low throughput and prevented robust analysis of individual glial responses, we confirmed that individual GCaMP6s transients could be detected in both protoplasmic and radial glial morphotypes (Figures 6L–6N). Together, the functional analysis is consistent with the interconnected nature of individual DRG neuron circuits and the potential for glia within them.

DISCUSSION

The sensoriomotor circuit is critical for animals to respond to their environment. Here, we mapped the vertebrate sensorimotor circuit at the ultrastructural level. In summary, our work provides a foundational resource of the map of a vertebrate local spinal sensorimotor circuit and introduces a potential layer of modulation that could occur at tripartite synapses. We show that astrocytic processes localize equally to all synapses in simple circuits and that a single astrocyte could regulate multiple and distinct synaptic connections within a single circuit. This foundational knowledge can be used for future studies that investigate other circuits that underlie important animal behaviors.

Until recent advances in serial EM and automated cellular reconstructions, circuit maps were typically reserved for model systems with a relatively smaller number of neurons.¹ For example, the C. elegans ultrastructural connectome was established in 1986.¹ Even with an organism with 302 neurons, it is clear that the flow of information is complicated. This is particularly the case when animals are exposed to multiple sensory signals, which requires the integration of sensory information. This is exemplified by the neurons in C. elegans that respond to harsh vs. light touch.^49–55 Both subtypes of neurons synapse onto interneurons in the ventral nerve cord of C. elegans, which transmit that information to motor neurons that move muscle. ¹ While both subtypes of harsh- and light-touch neurons are activated during a harsh-touch response, the animal responds in a stereotypical way by curling up.^52,54–57 Light touch to the animal, however, engages a distinct backward or forward movement.⁵⁷ Thus, while synapses occur on the same interneuron,^1,54,56 the information of harsh touch is somehow coded to ignore the light-touch neuron circuit. How this works in vertebrates is unclear. However, this lack of understanding is not surprising given our incomplete understanding of a vertebrate sensorimotor circuit map.

This level of processing and integration may be expected to be more complex in vertebrates. The sensory neurons in the vertebrate spine are located in the DRG.^16,58–60 The DRG in each somite contains a collection of neurons that each respond to distinct sensory modalities with spatial acuity.^58,61–63 The sensorimotor circuit we mapped contributes to unraveling how this segregation of different sensory inputs can be accomplished. For example, the mapping demonstrated that convergent and divergent circuit connections of individual DRG neurons could amplify or interrupt information flow. We also observed a network of interconnected interneurons where a single DRG could excite two interneurons, one of which would go on to inhibit an interneuron that was excited by another DRG. Our connectomes of each DRG neuron showed variability in the connectivity among DRG neurons to their downstream neuron partners. This variation could be attributed to the developmental state of the DRG neurons, as each DRG circuit is going through varying stages of synaptogenesis and refinement. The variation could also be an attribute of the neuron modality, which, beyond responses to 4°C, is not well defined in larval zebrafish. The higher connectivity of different DRG neurons would likely impact the information processing, with higher-connected neurons requiring more information processing than less connected neurons. The complexity of the circuit could also impact processing speed, with less connected neurons providing a quicker route to elicit immediate motor response. The ultrastructural map of these connections is only the first step in understanding how the animal distinguishes an environment that simultaneously activates multiple modalities.

Adding to the complexity of the circuit is the potential role of astrocytes in vertebrates. Studies from invertebrates have already supported the idea that the potential for glia to modulate circuits is evolutionarily conserved.^35,36,64 In C. elegans, glial cells control the hyperactivity of dopamine neurons.⁶⁵ In Drosophila and zebrafish, astrocytes occupy synapse-rich regions and modulate circuits that drive simple behaviors.^13,66–69 We also know that astrocytes in vertebrates can interact extensively with synapses.¹⁰ In mice, it is hypothesized that a given astrocyte can interact with 100,000 synapses.^8,70 Even more synaptic connections per astrocyte are predicted in humans. We also know that astrocytes provide structural support for neurons and can modulate synaptic function through neurotransmitter regulation.^71–73 In particular, astrocytes express transporters that can remove neurotransmitter from the synapse cleft. They also are hypothesized to participate by releasing gliotransmitters.^71,74,75 However, the extent to which a single circuit can be modulated by astrocytes needs more investigation. Our work shows the limits of the potential for astrocytic modulation at the synaptic level in a vertebrate sensorimotor circuit. In our map, we revealed that astrocytic processes contact synapses equally throughout the circuit. There are several reasons why such a distribution would be advantageous. It is possible that synaptic control at each area of the circuit ensures that there are multiple levels that can be controlled, including initial relay of sensory information, integration of information from interneurons, and finally firing of motor neurons to drive muscle movement. The equal distribution of tripartite synapses might also reflect the development of astrocytes and synapses, which is still occurring at 6 dpf, when the EM dataset was collected. These concepts are beyond the scope of this report but warrant future exploration.

The connectome of the nervous system is a critical and foundational resource if we want to further investigate the development and functionality of circuits. This work provides a key exploration of the connectivity of the vertebrate sensorimotor circuit and reveals the precise synaptic locations where glial cells could modulate the circuit.

Limitations

The circuit map includes only chemical synapses and therefore neglects the important contribution of electrical synapses in information flow. The circuit map is generated from a 6 dpf zebrafish and thus likely represents a circuit that will change as the animal ages. Further, the full complex connectivity of local spinal sensorimotor circuits is likely to expand significantly beyond what is reported here because synaptic connections outside of the 3 hemisegments are not included. While the number of times a neuron is synaptically connected could impact the circuit, a more accurate model would include other parameters, like synaptic strength. Similarly, the represented circuit is likely to expand in animals with a larger number of complex neurons. The number of tripartite synapses is likely underestimated because of current approaches that limit the segmentation of small astrocytic processes. Further, this report does not consider the potential astrocytic modulation of information flow beyond direct contact with synaptic cleft regions or that astroglia can positively impact neuronal excitability, both of which are reported in the literature and can be the focus of future investigation.^13,76,77

RESOURCE AVAILABILITY

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Cody J. Smith (csmith67@nd.edu).

Materials availability

All unique reagents created from this study are available upon request from the lead contact if not already present in a public repository.

Data and code availability

• All data collected for this study are included in the figures and supplemental information.

• Custom code was written for Python and deposited at Zenodo, publicly available at https://doi.org/10.5281/zenodo.17475612 as of the date of publication. Custom code was used to generate Figures 1E, 1H, 2C, 2E–2I, 3B, 3C, 3G, 4A–4G, S1B–S1D, S1I–S1L, S2C, S3A, S4A, S4C, and S4D. Data generated from the custom code was imported into Prism or SankeyMATIC for figure generation.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

STAR★METHODS

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Experimental procedures followed the NIH guide for the care and use of laboratory animals. The Animal studies were approved by University of Notre Dame Institutional Animal Care and Use Committee (IACUC) (protocol 19–08-5464), which is guided by the United States Department of Agriculture, the Animal Welfare Act (USA) and the Assessment and Accreditation of Laboratory Animal Care International.

Animal Specimens. Danio rerio (zebrafish) were utilized in this study. The stable strains used for this study were: AB, Tg(UAS: GCaMP6s)⁷⁸, Tg(neurod:GAL4+myl7:GFP)⁷⁹, Tg(gfap:GCaMP6s-CAAX) (generated for this study), Tg(ngn1:GFP)⁸⁰, Tg(sox10:mRFP).⁸¹ All embryos were produced through pairwise matings and grown in 28°C in constant darkness. At 24 hpf, zebrafish were exposed to PTU (0.0003%) to reduce pigmentation for intravital imaging. Age of animals was determined by hour postfertilization and stages of development.⁸⁵

METHOD DETAILS

Generation of transgenics

Tg(gfap:GcaMP6s-CAAX) was created by injecting tol2 RNA and psSL08 (gfap:GcaMP6s-CAAX-pA) into one cell animals. The plasmid was generated using gateway plasmids (p5e-gfap, pMe-GcaMP6s-CAAX, p3e-polyA, p394). Injected animals were grown to adults and then screened for GCAMP6s-CAAX fluorescence. Once founders were identified, animals were outcrossed to generate a stable Tg(gfap:GcaMP6s-CAAX) transgenic.

Mosaic injection

gfap:GcaMP6s-CAAX (psSL08) was injected at the one cell stage with Tol2 mRNA into AB animals. Animals were grown to 5 dpf and screened for successful expression via confocal microscopy. Animals were subjected to cold water submersion assay as noted below.

Serial electron microscopy

The serial electron microscopy dataset used in this study was previously published in Svara et al. 2018.²³ The images were collected with serial block-face electron microscopy from a 6 day post fertilization zebrafish larva. The images were collected from the middle of the spinal cord, near the anal pore. The volume is 74 × 74 × 207 μm³, with a voxel size of 9 × 9 × 21 nm³. All the EM images for this study were managed using https://knossos.app²⁴

Semi-automated cellular reconstructions

Cellular reconstructions in this study were completed using Knossos based on Svara et al. 2022.²⁴ Semiautomated reconstructions were completed by clicking on the cell of interest in Knossos. This revealed the morphology of the cell in the reconstruction window. To complete the cellular reconstructions, each endpoint was confirmed. If the axon continued in the section, segments of neurons were joined using the agglomeration function in Knossos. Cells were reconstructed until all endpoints were confirmed. For a subset of reconstructions, improper mergers of the agglomeration prevented the semiautomated approach. These neurons were reconstructed using the skeletonization function in Knossos. The center of the axon was marked throughout the section. The skeletonization reconstruction included nodes every 10 z-positions in the section. All skeletonized neurons were confirmed with two authors. Glial cells were reconstructed with the semi-automated approach.

Automated synaptic identification

Synapses were identified using the procedures outlined in Svara et al. 2022 and reconstructed using the 3dEMtrace service provided by ariadne.ai ag (https://ariadne.ai).²⁴ Synapses were defined as: 1. Two adjacent cellular surfaces were parallel and juxtaposed. 2: Vesicle clouds were present in the axon in close proximity to the membrane and opposite the adjacent cell. 3: A thick and dark labeling, although faint at times, of the postsynaptic membrane was present.²⁴ Two parts of the synapse were labeled throughout the dataset. Characteristics 1 and 3 were marked by pink and synaptic clouds defined by characteristic 2 were labeled in blue. We defined mature synapses as requiring both pink and blue markings and thereby all three characteristics. While synapses without all their characteristics were initially mapped for sensory neurons, the remainder of the mapping only included synapses with all three characteristics.

Neuron classification

Neurons were classified based on their morphology and previous literature that defined different subtypes of cells in the spinal cord. A decision tree was created based on the previous literature and each cell was classified by the newly created decision tree. The excitatory vs. inhibitory classification was based on cross-referencing the neuron morphologies with neurotransmitter classification in published literature.^19,25–29 CoSA neurons were excluded from analysis with excitatory/inhibitory because they are characterized in the literature as both.

Glia classification

Astroglia subtypes were defined by the morphology of the cell. Cells with radial projections that extended to the edge of the spinal cord and made up a significant portion of the glial limitans were characterized as glial limitans astroglia. Protoplasmic astroglia exhibited morphology with numerous small processes that extended into the neuropil. Astroglia that exhibit large mergers were not included in the subclassification. We define large mergers as those with multiple cell bodies within a single cellular reconstruction. We could not identify the morphology of each individual cell. Such mergers prevented synaptic contacts to be attributed to a single cell and thus were removed from analyses that sought to quantify the characteristics of different astroglia (Figure S3C). This represented 24.42% of astroglia that were present at synapses. For visualization in Figure 3E, improper mergers between astroglia and neurons were removed via Photoshop. For transparency’s sake, unedited and edited versions are represented in Figure S3B.

Tripartite synapses

At every defined synapse (per the criteria above), all membranes directly in contact with both pre and post synaptic membranes were clicked to display the cell reconstruction and identify it as astroglia. Any “tertiary” membrane that revealed an astrocyte through semi-automated reconstructions was characterized as a tripartite synapse.

In vivo imaging

Animals were anesthetized with veterinary grade 3-aminobenzoic acid ester during mounting and then were bathed in PTU during imaging.^{21,22,45,47,79} A glass-bottom 35 mm Petri dish was used to image the animals. Each animal was mounted on the glass-bottom dish with 0.8% low melt agarose. All confocal images were collected on a custom-built (3i) spinning disk confocal microscope. The microscope contains: Zeiss Axio Observer Z1 Advanced Mariana Microscope, X-cite 120LED White Light LED System, filter cubes for GFP and mRFP, a motorized X,Y stage, piezo Z stage, 20X Air (0.50 NA), 63X (1.15NA), 40X (1.1NA) objectives, CSU-W1 T2 Spinning Disk Confocal Head (50 μm) with 1X camera adapter, and an iXon3 1 K × 1 K EMCCD camera or Prime 95B back illuminated CMOS camera, dichroic mirrors for 446, 515, 561, 405, 488, 561,640 excitation, laser stack with 405 nm, 445 nm, 488 nm, 561 nm and 637 nm. Z-stacks were collected with a step size of 1 μm covering 40 μm of the spinal cord of the animal. High speed imaging was captured using a 4D rapid imaging module in Slidebook that utilizes free run captures. Exposure was set at 10 ms to reduce the interval of imaging. Brightness and contrast were only adjusted for generation of the figures in this study.

Temperature submersion and GCaMP6s assays

Animals were exposed to PTU from 1 to 5 dpf to restrict skin pigmentation. All imaging was performed on the spinning disk confocal inverted microscope as noted in microscopy section. Tg(gfap:GCaMP6s-CAAX) larvae were raised until 5 dpf for analysis of astroglia. The Tg(gfap:GCaMP6s-CAAX) animals were mounted laterally, and the glial limitans/spinal cord boundary was positioned 2–4 μm from the bottom of a 40 μm z stack (2 μm step-size) in order to fully capture the glia of the spinal cord.

Tg(neurod:GAL4+myl7-GFP);(UAS:GCaMP6s) animals were raised and imaged at 5 dpf to image neuronal activity. For the neuronal activity experiments, both laterally and dorsally mounted Tg(neurod:GAL4+myl7-GFP);(UAS:GCaMP6s)+ animals were used. In either mounting position, the appearance of the DRG was positioned 5–7 μm from the bottom of a 40 μm z stack (2 μm step-size).

In the injury context, Tg(neurod:gal4+myl7-GFP);(uas:GCaMP6s);Tg(sox10:mRFP) and Tg(ngn1:GFP); Tg(gfap:GCaMP6s-CAAX) animals were raised to 5 dpf at which point the centrally-projecting axons at spinal segment 4–11 were injured as previously described^22,47; 3 h post injury, these animals underwent the cold water immersion assay during calcium imaging.^20,22,45

For the water immersion assay, animals were anesthetized and individually mounted in 0.8% low melting agarose in 4-well glass bottom dishes, wherein the agar was thin enough that the body of the animal created a bump in the agar. After the agar solidified, fresh, room temperature egg water was added to the wells and the animals were given 20 min to recover from the anesthetic. Egg water was then removed from the well and a 24-time point timelapse was started. At a step size of 2 μm, z-range of 40 μm, and exposure time set to 10 ms, each capture of the timelapse took exactly 0.971 s. After time point 12, the water is carefully aspirated from the well without disturbing the dish/the xy position. At time point 15 - just before timepoint 16, 4°C egg water was added to the well, where it remained for the rest of the 24 timepoint timelapse movie.

Sankey plots

Cells were given individual cell identity numbers in Knossos based on the segmentation ID of the neuronal cell body. A custom code was written to generate sankey plot information. In brief, each the segmentation IDs of each synaptic connection in the circuit was generated. For example, segmentationID1 [2] segmentationID2 would represent cell#1 is connected through 2 synapses to cell#2. This process was completed for every connection in the circuit. Once completed for a single DRG neuron’s circuit, the code was repeated on each circuit. To represent the interconnectivity of the circuits, synaptic connections between DRG R1B and L3B were combined. Sankey plot data was then imported into sankeymatic and color coded according to the figure panel.

QUANTIFICATIONS AND STATISTICAL ANALYSIS

Quantification of cell measurements

The area of the axon was measured using the line ROI tool in ImageJ. The total area of the ROI was first calculated in pixels and then converted to micrometers. The swelling at the synapse was measured where the synaptic cloud are centered. Before and after the swellings were defined as directly adjacent locations from the synaptic cloud regions. Length of DRG axons was estimated by taking multiple points within the axons and summing the Euclidean distance between each of these points, with respect to the scalebar given in Knossos.

Quantification of synaptic and glia organization

Synapses and glia at synapses were counted within each DRG neuron’s circuit. To keep the data consistent across DRG circuits, all the data was combined into a larger database. 3 lists were made using a custom Python code – an object list, a synapse list and a tripartite glia list. Each element of the object list had 2 components – the Knossos ID of the object, and the cell type of the object. Each element of the synapse list had 4 components – the coordinate of the synapse, the synaptic type, the pre-synaptic reference number, and the post-synaptic reference number (pointing to the position of the objects within the object list). Each element of the tripartite glia list had 3 components – the coordinate of the glia, the Knossos ID of the glia, and the synaptic reference number (pointing to the position of the synapse within the synapse list). These lists were populated with data from manually counting the synapses, first the object list, then the synapse and tripartite glia list together. Duplicate synapses across circuits were removed by getting the Euclidean distance between known coordinates and setting a threshold of 35 voxels, which we determined a distance that ensured unique synapses could be identified.

Quantification of synapse types

Synapse types were identified into 4 subtypes by where the synapse occurred with respect to the pre-synaptic and post-synaptic neurons: A-A for axon to axon, A-D for axon to dendrite, A-S for axon to soma, and the rare A-G for axon to glia. Axons, dendrites and cell somas were defined by specific criteria that define that area of the neuron. Cell bodies were defined by the location of the nucleus. Neuronal branches were defined as dendrites if they extended multiple projections from the cell soma and branched extensively into the neuropil. Dendrites typically did not contain synaptic clouds/vesicles. Axons were defined as singular processes that extended from the cell soma and contained pre-synaptic material.

Quantification of tripartite types

Glial modulators were identified by morphology as mentioned above (under Tripartite Synapses), and the number of unique glia and the number of times each one appeared in a given circuit were calculated.

Quantification of GCaMP6s co-activity

Maximum z-projections were made of each capture. GCaMP6s was analyzed by exporting the 16-bit timelapse movies into ImageJ software (Fiji). A single rectangle was drawn around any apparent fluorescent/cellular landmarks that could be distinguished throughout the timelapse. Using the Template Matching plugin, frames of each timelapse were aligned according to the visible landmarks throughout the timelapse to correct for any movements from the animal and/or additions of water.

For the glial calcium imaging, the scale of the file is set according to the microscope’s specifications, in which the distance in pixels is scaled to make a 1.0 pixel aspect ratio and known distance of 1.00 μm. The threshold plugin was run and Otsu settings were used to make the imaging file binary. The low-end threshold is set so that almost no ROIs are visible in timepoints preceding 4°C water exposure, when no calcium activity would be expected. The high end of the threshold was set at its highest setting. The Analyze Particles feature was then used with the size was set to 3.00-Infinity (microns ^2). It was then confirmed that no spots were identified during room temperature frames unless a random activation event occurred. This process was completed for each timelapse, however, the low-end threshold values varied in each animal. From each of the 24 timepoints, the average and standard deviation was used to calculate the Z score of the integrated density (the product of area and mean fluorescent value) in each frame. A Z score of 2.0 or more was set as a significant activation event for slower (5-s interval) acquisition movies. For shorter (0.971-s) intervals Z score of 1.0 was set as a significant activation because more high fluorescence intensity timepoints are incorporated into normalization parameters. These z-scores were plotted into a stacked xy line graph or heatmap. The total area above threshold was also used to display the global activation of the glial limitans.

For the neuronal calcium imaging, the first capture immediately after the cold water immersion was used to set ROIs. ImageJ’s ROI Manager feature was utilized and all distinguishable (regardless of visible brightness) GCaMP⁺ DRG and spinal neurons were free-hand traced and added to the ROI manager. Once all cells were traced in 4°C immersion timepoint, any other neurons not previously traced that fired in later time points were also traced. The multi-measure tool was used to obtain the integrated density for each traced cell. The integrated density measurements for each cell at each timepoint was used to obtain the average and standard deviation in order to calculate z-scores. A Z score of 1.0 or more was marked as a neuronal firing event. Neuronal GCaMP6s dynamics were classified into 3 categories. If GCaMP6s changes underwent a Z score change of >1.0 in the 1st frame after 4°C submersion was classified as a rapid firing event. If a neuron did not have a Z score of 1.0 or more until at or after the 2^nd frame of the 4°C submersion, that was classified as a delayed firing event. If the cell never underwent a Z score change of 1.0 or more then that was classified as no firing event. z-scores for all neurons were plotted in a stacked xy line graph for each animal. The percent of rapid, delayed and no firing event spinal neurons was collected for each animal for comparisons. A Fisher’s exact test was used to compare the percentage of rapidly responding versus delayed firing spinal neurons between groups. Note, in the injured GCaMP+ animals, only DRG 6–9 were analyzed to ensure that all spinal neurons analyzed in this assay were flanked by at least two DRG that had previously been injured.

Statistical analysis

Prism was used for all statistical analysis. Sample sizes were based on previous publications but were not predetermined by statistical methods. All statistical tests were performed with biological replicates and not technical replicates. Some analysis excluded any data points associated with neuronal branches without cell soma in the EM section. Heatmaps of astrocyte synaptic interactions excluded astrocytes that were merged. Data extracted from EM were from a single healthy animal. For GCaMP6s experiments, healthy animals were randomly selected. GCaMP6s experiments were repeated at least twice.

Software

Slidebook, Prism, ImageJ, Adobe Illustrator, and Knossos were used to acquire, analyze, and compile figures. Sankey plots were generated with https://sankeymatic.com/build/.

Supplementary Material

SUPPLEMENTAL INFORMATION

Supplemental information can be found online at https://doi.org/10.1016/j.celrep.2025.116761.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Experimental models: Organisms/strains
Danio rerio: AB	N/A	N/A
Danio rerio: Tg(UAS:GCaMP6s)	Thiele et al.78	ZFIN: ZDB-ALT-140811-3
Danio rerio: Tg(neurod:GAL4+myl7:GFP)	Nichols and Smith79	ZFIN: ZDB-ALT-191209-8
Danio rerio: Tg(gfap:GCaMP6s-CAAX)	This paper	N/A
Danio rerio: Tg(ngn1:GFP)	Prendergast et al.80	ZFIN: ZDB-ALT-090806-1
Danio rerio: Tg(sox10:mRFP)	Kucenas et al.81	ZFIN: ZDB-ALT-080321-3
Chemicals, peptides, and recombinant proteins
PTU (1-phenyl-2-thiourea)	Fisher Scientific	Lot: A0393290 CAS: 103-85-5
P396: tol2 RNA	Kwan et al.82	p396: https://tol2kitkwan.genetics.utah.edu/index.php/Full_Clone_List
psSL08 (gfap:GcaMP6s-CAAX-pA)	This paper	-
p5e-gfap	Don et al.83	Addgene: 75024
pMe-GcaMP6s-CAAX	Hughes and Appel84	-
p3e-polyA	Kwan et al.82	p302:https://tol2kitkwan.genetics.utah.edu/index.php/Full_Clone_List
p394	Kwan et al.82	p394:https://tol2kitkwan.genetics.utah.edu/index.php/Full_Clone_List
Software and algorithms
Knossos	N/A	https://knossos.app/
Python (3.13.7)	N/A	https://www.python.org/
Adobe Photoshop	N/A	https://www.adobe.com/products/photoshop
Adobe Illustrator	N/A	https://www.adobe.com/products/illustrator.html
ImageJ	N/A	https://imagej.net/ij/
Prism	N/A	https://www.graphpad.com/features
SankeyMatic	N/A	https://sankeymatic.com/
Microsoft Word	N/A	https://www.microsoft.com/en-us/microsoft-365/word
Others
Custom code written for data analysis	This paper	https://doi.org/10.5281/zenodo.17475612

Highlights.

• Detailed map of local spinal sensorimotor circuits

• Astroglial synaptic contacts are equally distributed in the local sensorimotor circuits

• Glial modulation and synaptic thresholding could reduce circuit complexity

• GCaMP imaging confirms circuit activity and astroglia respond robustly to stimuli