Combined membrane potential imaging and connectome of behavioral circuits in an annelid worm

In most animal phyla, nerve cells form highly speciﬁc connections with each other 1,2 . The resulting intricate networks determine what activity patterns a nervous system can sustain, and hence what behaviors an animal can exhibit 3,4 . Accordingly, understanding the relationship between connectivity and activity is a major goal of neuroscience. However, despite major recent advances 5,6 , no current technology can comprehensively record activity in large nervous systems such as the human or even the mouse brain, nor can connectivity be reconstructed at synaptic level in such systems 7 . Doing both at once in the same specimen remains a distant goal. However, smaller nervous systems offer exciting opportunities today, and here we report for the ﬁrst time obtaining joint functional and anatomical data from a major functional unit of the nervous system of the medicinal leech, by combining voltage-sensitive dye (VSD) 8 imaging with serial blockface electron microscopy (SBEM) 9 . We simultaneously recorded from the majority of the neurons in a segmental ganglion during several motor behaviors with a VSD sufﬁciently sen-sitive to record subthreshold neuronal activity 10,11 . We then anatomically imaged the entire ganglion with SBEM 12 . As a proof of concept, we have manually traced a


Main
Understanding the relationship between connectivity and activity in the nervous system remains one of the great intellectual challenges in science and it requires the integration of functional and anatomical approaches.In the age of large-scale "omics" projects, one would ideally like to record from every single neuron in a nervous system during all of the behaviors the animal can execute, and then reconstruct the anatomical connections between those neurons.Although still monumental, neither of these endeavors is out of reach anymore: Activity imaging using calcium dyes has advanced to the point where simultaneous recordings from the vast majority of individual neurons in smaller species can be accomplished, for instance in larval zebrafish 5 .This technique has even been applied to behaving animals 13 .Likewise, anatomical imaging using electron microscopy has advanced to the point that brains as large as that of the fruit fly Drosophila melanogaster can be imaged-and substantial fractions of their circuitry reconstructed-at a synaptic resolution 6,14,15,16 .Even small mammalian brains like that of larval zebrafish are yielding to this approach 17 .
Critically, however, robustly linking function to connectivity requires a combined anatomical and functional assessment within the same animal.After all, even the simplest nervous systems exhibit some variability in their connectomes 18 (and larger ones exponentially more).Yet, each of the studies mentioned above addressed either function (by recording neuronal activity) or connectivity (by reconstructing anatomy) in isolation.Here we performed for the first time a combined assessment by first recording with a voltage-sensitive dye (VSD) 19 from a segmental ganglion of the medicinal leech Hirudo verbana while its nervous system expressed several behaviors and then imaging the same individual ganglion with serial blockface electron microscopy (SBEM; Fig. 1).A leech segmental ganglion is a good stand-in for a whole nervous system, because its neurons capture the entire pathway from sensory input through self-generated interneuronal rhythms to motor output.We chose VSD imaging over calcium imaging for this pioneering study, because VSDs can detect both action potentials and subthreshold excitatory and inhibitory potentials.Likewise, we chose SBEM over serial-section transmission EM 20 because SBEM can reliably process large numbers of slices with much lower risk of sectioning artifacts.Lastly, we chose the leech for this study because it robustly expresses several behaviors even in reduced preparations 21 , its neurons are uncommonly accessible to physiological recording, and its cell bodies are relatively large and thus yield strong VSD signals 22 .The nervous system of the leech comprises cephalic ganglia, a tail ganglion, and 21 nearly identical segmental ganglia connected by a ventral nerve cord 23,24 .We focused on the segmental ganglia, because those are largely responsible for processing sensory information and generating muscle activity 21 .
author/funder.All rights reserved.No reuse allowed without permission.

X-ray imaging
Serial blockface electron microscopy 22 × 10¹² voxel traceable volume Figure 1: Approach.Several fictive behaviors were induced in the isolated nervous system of a medicinal leech while one segmental ganglion was imaged using a VSD.After fixation and resin embedding, the ganglion was x-ray-imaged to verify that the geometry of somata was preserved.Finally, the neuropil was imaged at nanometer resolution with SBEM.

Voltage-dye imaging of behavior
Each ganglion consists of about 400 neurons 25 with cell bodies arranged in a spherical monolayer around a central neuropil.In the neuropil, neurons communicate through chemical and electrical synapses located along extensively branched neurites 26,27,12 .Unlike in mammals, leech neuronal processes cannot be subdivided into axons and dendrites.
We expressed (fictive) swimming, crawling, and local bending behavior in the isolated nervous system of a single leech following the same protocol used for a previous extensive study of these behaviors using VSD imaging in a larger group of animals 11 .As in the previous study, one segmental ganglion in the chain was prepared for VSD imaging and we recorded from both the ventral and dorsal aspects simultaneously with a double-sided fluorescence microscope (Fig. 2a).
We were able to record from 250 neurons simultaneously, similar to our previous results.Fictive swimming was induced by electrical stimulation of a posterior segment, which resulted in characteristic rhythmic activity in dorsal motor neurons and many other neurons on both sides of the ganglion (Fig. 2b).Coherence analysis confirmed that the rhythms of the various neurons were indeed related (Fig. 2c, d).In a similar manner, we induced fictive local bending (Fig. S1) and crawling (Fig. S2).
We established a mapping between the neurons seen in the VSD images and the canonical maps of the ganglion 28 based on geometry and on the involvement of the neurons in the various behaviors.

Electron microscopy
At the end of the (fictive) behavior experiment, the ganglion was fixated and embedded in a resin.
We then re-imaged the ganglion using x-ray tomography and verified that the cell bodies seen in the VSD images could still be identified (Fig. S3a).The x-ray image stack was also used to trace neuronal processes from the somata to the edge of the neuropil (Fig. S3b).This obviated the need to capture the somata in the subsequent electron microscopy, and instead allowed us to restrict the EM effort largely to the neuropil.
Electron microscopy itself took nearly 7 months of imaging, during which we acquired 78,803 images from 9604 slices, totaling 22.8 terapixels.We periodically paused the acquisition to clean the knife or to adjust the imaging area so as to include the entirety of the neuropil but not too much additional space.

Tracing a motor neuron and all its synaptic inputs
With current technology, automated tracing of the fine processes in our EM image set was not possible.Therefore we decided to manually trace one motor neuron of particular interest and all of its presynaptic inputs.The motor neuron we chose was Dorsal Excitor motor neuron DE-3 R , a common output of all of the behaviors included in our functional data set.The combined path length of the entire arborization of DE-3 R was 6,109 µm (Fig. 3a, b, c).
In addition to tracing the neuron, we marked all of its postsynaptic sites and then traced each of its presynaptic partners to their somata.Several visually distinct types of synapses were author/funder.All rights reserved.No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the .https://doi.org/10.1101/2020.03.09.984013 doi: bioRxiv preprint c 100 μm found, among which most prominently: (1) bright terminals with large dark vesicles (Fig. 3d) and (2) darker terminals with smaller vesicles that occurred mainly in fiber endings and varicosities (Fig. 3e).The small vesicles are barely resolved in our data set and appear merely as fields of granules.We used TEM on thin slices of a second ganglion to confirm our interpretation of these granules as vesicles (Fig. S4).No attempt has been made as of yet to interpret the anatomically distinct types of synapses as physiological classes.
We identified 531 synapses onto DE-3 R .Of these, 44 were formed by cells with somata in neighboring ganglia, which were not included in our EM volume.Of the rest, 387 could be traced to their somata with a high degree of confidence.To avoid false positives, we only considered presynaptic neurons that formed at least two synapses onto DE-3 R .There were 51 of those.Of those, 27 could be confidently matched to cell bodies seen in the VSD record, and of those, 14 could be confidently matched to specific identified neurons on the canonical map (Fig. 4a).

Linking form to function: Synaptic clustering
Spatial clustering of synapses on the dendrites of a neuron has been proposed as a key organizational principle of circuits that integrate multiple sources of information 29,30,31 .Since motor neuron DE-3 R must integrate input from its presynaptic partners to accurately generate multiple behaviors, we therefore assessed the spatial distribution of synapses on the neurites of DE-3 R .Cluster analysis (Fig. 4b) revealed a multitude of synaptic clusters.Most clusters contained synapses from 6 author/funder.All rights reserved.No reuse allowed without permission.
A key strength of our dataset is that it contains both functional and anatomical information.
Accordingly, we were able to assess the functional relevance of these spatially defined synaptic clusters by investigating whether neurons that contribute synapses to a specific cluster share commonalities in their activity during behaviors.An ANOVA-style procedure was used that yielded an F-ratio indicating whether coherence values of neurons were more similar within spatial clusters than between clusters.The procedure was applied to each of the trials separately and for a range of clustering parameters (see Methods).In all but one trial, parameter ranges could be identified for which spatial clusters were found to correspond to functional groupings (Fig. 4c and Fig. S5).We used a least-squares fit approach to find the location in parameter space of the strongest correspondence (Fig. S6 and Methods).In the two swim trials, the peaks were located at e C = 61 ± 2 µm and 65 ± 2 µm respectively; in the two crawl trials at 14 ± 5 µm and 18 ± 3 µm.In summary, the clustering parameters that led to the strongest correspondence in swim trials were quite different compared to crawl trials, but consistent within a trial type.(Results for local bend trials were not consistent; Fig. S5).

Conclusion
We successfully combined voltage-sensitive dye imaging with serial-blockface electron microscopy to obtain both detailed functional and anatomical information from the same individual nervous system.For this purpose we chose a segmental ganglion from the medicinal leech Hirudo verbana which contains the neuronal circuits that support motor behaviors including swimming, crawling, and local bending.After functional and anatomical imaging, we focused on the reconstruction of the main excitatory motor neuron of dorsal longitudinal muscles, DE-3 R along with its presynaptic partners 32 .
The reconstructed morphology of the motor neuron DE-3 R was in accordance with previous light microscopic studies in adult 33,27 and electron microscopy in a juvenile ganglion 12 : its primary neurite emerged from the soma laterally and traveled toward the ipsilateral roots before making a 180°turn to run laterally across the ganglion.We could positively identify the EM image of the soma of DE-3 R with its image in VSD data (Fig. S3).Our tracing revealed 531 synapses onto DE-3 R .This was slightly lower than the number previously reported for a juvenile ganglion (650) 12 .
The difference may be due to individual variability or developmental plasticity.Several of the synapses we found were with presynaptic partners that had previously been reported 34 based on paired electrophysiological recordings.We focused here on chemical synaptic connections, since SBEM does not yet allow identification of gap junctions.Work on molecular markers of electrical synapses that do not require expression of a particular connexin 35 may overcome this hurdle in the future.This is important, because electrical synapses between DE-3 and their contralateral homologs as well as some interneurons have been reported 27 .
author/funder.All rights reserved.No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the .https://doi.org/10.1101/2020.03.09.984013 doi: bioRxiv preprint Motor neuron DE-3 and its presynaptic partners form a multifunctional neuronal circuit; DE-3 being a common output for multiple motor behaviors.Combining VSD and SBEM data allowed for an integrated assessment of structure and function of this circuit: First, the detailed anatomical reconstruction at synaptic level revealed the local convergence of presynaptic cells within synaptic clusters on the neurites of the motor neuron.Second, referring back to the VSD results showed that these clusters act as functionally cohesive units.That is, anatomically defined clusters tended to comprise presynaptic partners with like activity patterns in the behaviors we studied.An attractive interpretation of our results is that the clusters are the loci where inputs from synchronized presynaptic cell assemblies are integrated 36 .The limited experimental evidence currently available suggests that even clusters of two synapses are functionally relevant in the mammalian cortex 29,37 .In agreement with earlier reports 38 , the strongest correspondence between spatial clusters and activity patterns was observed when synaptic clusters were defined by a maximum distance between nearest neighbors (d NN ) of up to 10 µm.In contrast to earlier work that relied on light microscopy and could therefore not identify presynaptic partners, our use of SBEM allowed us to base our assessment directly on individually identified synapses.
All of our data are publicly available and can form the basis of future investigations of the relationship between form and function in nervous systems.They can also serve as a large-scale ground truth for EM segmentation algorithms 39 .The combination of anatomical methods with synaptic resolution and imaging techniques that can record from the entirety of the neurons of a circuit promises an extraordinary opportunity to assess neural computations at the level of circuit dynamics.Functional maps 40 from recorded activity combined with anatomical connectomes 41,42,43 are therefore poised to become a powerful tool not only to have a better understanding of complex behaviors, but also to predict the outcomes of new manipulations 44,45,46 .

Dissection and voltage-sensitive dye imaging
Detailed procedures have been described before 47 .Briefly, leeches (Hirudo verbana, obtained from Niagara Leeches, Niagara Falls NY) were maintained on a 12h:12h light:dark cycle in temperaturecontrolled aquariums filled with artificial pond water.The entire nervous system of a leech was removed and pinned down on silicone (PDMS, Sylgard 184, Dow Corning, Midland, MI).The sheath surrounding one segmental ganglion (M10) was removed from both ventral and dorsal aspects to allow access with voltage-sensitive dyes.Most of the nerves that innervate the periphery were cut short, but several were kept long to allow extracellular stimulation as described before 11 .
A voltage-sensitive dye (VF2.1(OMe).H 48 provided by Evan Miller) was bath-loaded at a concentration of 800 nM in leech saline using a pair of peristaltic pumps to evenly load cell membranes on both sides of the ganglion.The preparation was placed on a custom-built dual-headed microscope which was used to image neuronal activity during fictive behaviors triggered by electrical stimulation, as in our previous work 11 .
We manually drew regions of interest (ROIs) around neuronal cell bodies and used custom software to associate those ROIs with named cells on the canonical maps of the leech ganglion 28 .
For each of the behavior trials separately, we calculated the spectral coherence between each of the neurons and DE-3 R at the frequency of the dominant peak in the power spectrum of DE-3 R for the given behavior.

Histology
After dye imaging, the preparation was reduced to just one segmental ganglion by transecting the anterior and posterior connectives.The ganglion was mounted on a slab of silicone (DPMS) with a hole cut out in the center so that the somata would not be in direct contact with the silicone.This preparation was transferred into a glass container and incubated for 72 hours at 4 °C in 2% paraformaldehyde, 2.5% glutaraldehyde in 0.15 M cacodylate buffer containing 2 mM CaCl 2 .
Subsequently, the ganglion was washed in cacodylate buffer for 10 minutes and then incubated in an aqueous solution of 2% OsO 4 and 1.5% potassium ferrocyanide.During this incubation, the sample was microwaved in a scientific microwave (Pelco 3440 MAX) three times at 800 W with a duty cycle of 40 seconds on and 40 seconds off at a controlled temperature of 35 °C and subsequently left at room temperature (RT) for 30 minutes.The sample was then washed twice in ddH 2 O and then microwaved three times at 30 °C with a duty cycle of 2 minutes on and 2 minutes off.
The sample was incubated in 0.5% thiocarbohydrazide (Electron Microscopy Sciences, Hatfield, PA).During this incubation, the sample was microwaved three times at 800 W with a duty cycle of 40 seconds on and 40 seconds off at 30 °C and subsequently left at RT for 15 minutes.
The ganglion was then washed again, followed by the same microwave incubation as described above.
Next, the sample was incubated in 2% aqueous OsO 4 , microwaved three times at 800 W with a duty cycle of 40 seconds on and 40 seconds off at 30 °C, and left for 30 minutes at RT.After another wash, the sample was left overnight in 2% uranyl acetate at 4 °C.
The next day, the sample was incubated in a lead aspartate solution at 60 °C for 25 minutes 49 .
The sample was then washed and dehydrated through a series of ethanol solutions (50%, 70%, 90%, 100%, 100%, 10 minutes each) at RT and incubated in acetone.After this, the sample was infiltrated with epoxy resin by first incubating it for one day at RT in a solution of 25% Durcupan (Sigma, St. Louis, MO) in acetone.On subsequent days, the concentration of Durcupan was increased to 50%, 75%, and finally 100%.After that, the sample was transferred to freshly prepared 100% Durcupan and incubated at 60 °C for 3 days.
author/funder.All rights reserved.No reuse allowed without permission.

Micro-CT Imaging
We used Micro-CT scanning to confirm that the above sample preparation had left the overall geometry of the ganglion intact and to trace portions of neurons outside of the neuropil.Scans were collected using the 20x objective on a Zeiss Versa 510 X-ray microscope.Epoxy-embedded ganglia were attached to the end of an aluminum rod using cyanoacrylate glue and then scanned at 80 kV, collecting 2401 projection images while rotating the specimen 360 degrees.The final pixel size was approximately 0.75 µm.Volumes were reconstructed using Zeiss Reconstructor software and visualized in custom software (GVox, see "Data availability").

Scanning electron microscopy
Ganglia were mounted onto aluminum pins using conductive silver paint.The ganglia were mounted in a vertical orientation (with the anterior connective pointing upwards).The sample was imaged with a Zeiss Gemini 300 SEM with a Gatan 2XP 3View system.The microscope was run in focal charge compensation mode using nitrogen gas (40% pressure), with an accelerating voltage of 2.5 kV, a 30 µm objective aperture, magnification of 336x, a raster size of 17100 × 17100 pixels, a 5.5 nm pixel size, a 0.5 µs dwell time, and 50 nm section thickness.Stage montaging with an overlap of 8% between tiles was used to cover the complete extent of the ganglion in any given image.The backscatter detector was a Hamamatsu diode with a 2-mm aperture.
Outside of the neuropil, neuronal processes could be traced in the micro-CT scan, which allowed us to reduce the total volume needed to be imaged with SBEM by almost a factor two (Figure S3b).Still, at the widest points of the neuropil as many as 7 × 2 tiles (119,700 × 34,200 pixels) were needed at a given z-position.
After approximately every 500 sections, the run was stopped to clear sectioning debris from the diamond knife and prevent contamination of the block-face, diode, or column.The run was also stopped when we reached significantly wider or narrower regions of the neuropil as indicated above.Ultimately, the run was subdivided into 61 subruns.There was only one instance in the run where a significant loss of tissue occurred (approximately 150 nm) following the re-approach of the knife to the tissue after an interruption for clearing debris.

Transmission electron microscopy
Image quality and specimen preservation was verified using an additional ganglion prepared as above, but imaged in ultrathin sections on a conventional transmission electron microscope (JEOL JEM-1200EX, 120 kV, 12,000x-20,000x magnification).

Image processing
Images were aligned using custom software.First, we reduced the linear resolution of the original images by a factor five.Then we split each image into 5x5 sub-tiles and calculated the optimal Because of limited resolution in our SBEM images, synaptic vesicles appear merely as gray granules (Fig. 3c,d), but fields of such granules were clearly distinct from other gray areas in the SBEM images.Comparison with digitally blurred TEM images (Fig. S4) confirmed this interpretation.
We found that granular areas were concentrated in fiber endings and varicosities.

Synaptic clustering analysis
The analysis was based on data from the 45 synaptic partners of DE-3 R for which both anatomical as well as VSD recordings were available.We defined synaptic clusters using an agglomerative hierarchical clustering algorithm with two parameters: (1) The maximum allowed distance between nearest neighbors (d NN ); (2) The maximum overall spatial extent of the cluster (e C ).
The algorithm began by treating each synapse as an individual cluster.Then, it iteratively joined the two clusters with minimum distance between their most proximal elements ("single-linkage" clustering).However, if a joint cluster would exceed the limit on overall spatial extent (e C ), its putative constituents were not joined.Aggregation stopped when no pairs of clusters were left with acceptable nearest-neighbor distance (i.e., less than d NN ) and acceptable joint spatial extent (i.e., less than e C ).All distances were measured along the neurites of DE-3 R rather than by Euclidean metric in the volume.We explored maximum nearest-neighbor distances between 5 and 25 µm, and maximum spatial extents between 10 and 100 µm.Clusters comprising only a single synapse were not considered for further analysis.
The analysis of functional significance of spatial clusters used an ANOVA-like procedure on the complex spectral coherence values of neurons within clusters relative to DE-3 R .As in ANOVA, we calculated sums of squares within and between clusters.Since coherence values are complex numbers, we used the absolute square value.In standard ANOVA, the ratio of these sums of squares (the "F-ratio") follows an F-distribution under the null hypothesis.In the complex-valued case, that is no longer true, so we calculated empirical distributions of the F-ratios by randomly shuffling the list of per-neuron coherence values 1000 times.The empirical p-value p was then defined as p = m+1 N+1 , where N = 1000 is the number of randomizations and m is the number of times the F-ratio from shuffled data exceeded the experimentally observed F-ratio.These p-values are reported in the supplemental table.We generated plots of the F-ratio as a function of the cluster parameters d NN and e C .For each trial, we first determined the value of d NN for which the largest F-ratio was obtained.Then, we fitted a Gaussian of the form to the F-ratio as a function of e C (Fig. S6).The µ-values from those fits and their uncertainties according to least-squares fitting are reported in the text.
author/funder.All rights reserved.No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the author/funder.All rights reserved.No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the author/funder.All rights reserved.No reuse allowed without permission.
The copyright holder for this preprint (which was not peer-reviewed) is the .https://doi.org/10.1101/2020.03.09.984013 doi: bioRxiv preprint    The copyright holder for this preprint (which was not peer-reviewed) is the author/funder.All rights reserved.No reuse allowed without permission.

Extended data figures and tables
The copyright holder for this preprint (which was not peer-reviewed) is the .https://doi.org/10.1101/2020.03 The copyright holder for this preprint (which was not peer-reviewed) is the

3Figure 2 :
Figure 2: Fictive swimming imaged using VSD.a. Images of the dorsal (left) and ventral (right) aspects of a leech ganglion simultaneously obtained using a double-sided microscope."R" indicates the right side of the ganglion (i.e., the animal's right when dorsal side up).b.Selected VSD traces during fictive swimming.From the dorsal surface: dorsal and ventral inhibitory and excitatory motor neurons DI-1, VI-2, DE-3, and VE-4; from the ventral surface: AP ("Anterior Pagoda") cells and the Retzius cells (neuromodulatory interneurons).Vertical scale bars: 0.2% relative fluorescence change c.Magnitude (radial axis from 0 to 1) and phase (angular coordinate) of the coherence of activity in individual neurons with the swim rhythm.Error bars indicate confidence intervals based on a multi-taper estimate.d.Coherence maps of the VSD signals of all cells on the dorsal (left) and ventral (right) surfaces of the ganglion.Colors of cell bodies indicate coherence relative to excitatory motor neuron DE-3 R .Color scale applies to all panels.

Figure 3 :
Figure 3: Electron microscopic tracing: neurites and synapses of motor neuron DE-3 R .a.The principal neurite of DE-3 R near its entrace to the neuropil (dashed yellow outline).b.Two branches of the neurite of DE-3 R (dashed outlines).c.Fully reconstructed arborization of DE-3 R with a transverse section of the ganglion.(In the perspective, anterior is up, dorsal is near.)d.A synaptic connection onto DE-3 R from an inhibitory motor neuron (DI-1 R ).Arrowheads: synapses, Pre: presynaptic terminal, v: vesiscles.e.A synapse onto DE-3 R from an interneuron (cell 24 on the canonical map 28 ).

Figure 4 :
Figure 4: Partner neurons of DE-3 R with multiple synapses.a. Full tracing of DE-3 R (thick black line, soma location marked by boldface "3") and backtracings of all synaptic partners.Partners that we could identify as known entities on the canonical ganglion map are colored (arbitrarily) and labeled in black type for confident identification, light gray type for low-confidence partners.Small gray disks indicate partner neurons that could not be cross-identified between EM and VSD imagery.b.Tracing of DE-3 R with synapses and synaptic clusters obtained with parameter values (d NN , e C ) = (5 µm, 65 µm).Synapses (small dots) are colored by the coherence between the activity of their presynaptic partner and DE-3 R during swimming (as in Fig. 2).Clusters (elliptic areas) are colored by the average coherence of their constituent presynaptic partners.c.All clustering results for swim and crawl trials.Color indicates the degree of correspondence between spatial clusters and functional grouping expressed as an F-ratio from complex ANOVA (see Methods) as a function of clustering parameters.

Figure S1 :Figure S2 :Figure S3 :
Figure S1: Fictive local bending imaged using VSD.a. Selected VSD traces during fictive local bending.From the dorsal surface: dorsal and ventral inhibitory and excitatory motor neurons DI-1, VI-2, DE-3, and VE-4; from the ventral surface: AP cells (well-known postsynaptic partners of the P cells with unknown function) and the Retzius cells.Vertical scale bars: 0.2% relative fluorescence change.b.Magnitude (radial axis from 0 to 1) and phase (angular coordinate) of the coherence of activity in individual neurons with the local bend rhythm.Error bars indicate confidence intervals based on a multi-taper estimate.c.Coherence maps of the VSD signals of all cells on the dorsal (left) and ventral (right) surfaces of the ganglion.Colors of cell bodies indicate coherence relative to DE-3 R ."R" indicates the right side of the ganglion (i.e., the animal's right when dorsal side up).Color scale applies to all panels.

Figure S4 :
Figure S4: Comparison of SEM with TEM for interpreting synapses.a.Our interpretation of a small section of our SBEM image: a process of cell DE-3 (yellow) and a presynaptic partner (purple).b.Same area without overlay.Arrowhead: synapse.Stars: Mitochondria.c.A similar area imaged with TEM (from a ganglion from another leech).d.Same area as (c), computationally blurred to simulate the lower resolving power of SEM.

Figure S5 :Figure S6 :
Figure S5: Clustering results for the local bend trials.Out of the four trials, the first three involved stimulation of the left P V cell; the final trial involved stimulation of the right P V cell.In contrast to the swim and crawl trials (4c), the clustering results for local bend trials are inconsistent.
21author/funder.All rights reserved.No reuse allowed without permission.

Table S3 :
Results of the ANOVA analysis of synaptic clusters, local bend trials.25 author/funder.All rights reserved.No reuse allowed without permission.