Cohen, Netta et al. (2011) International Worm Meeting "A computational model of associative learning and chemotaxis in the nematode worm C. elegans."

Cohen, Netta, Appleby, Peter

[International Worm Meeting, 2011]

The challenge of understanding how the effectively hard wired circuitry in C. elegans generates complex behaviors is particularly interesting in the context of circuits that serve multiple functions and that undergo plasticity or learning. One such circuit is the relatively well characterized chemotaxis circuit in the head. In chemotaxis C. elegans will move up or down a chemical gradient dependent on whether the chemical acts as an attractant or repellent. It does this by a combination of two navigational strategies: (i) gradually steering left or right until the worm points up or down the gradient and (ii) modulating the probability of pirouettes and choosing the final orientation of the worm after the pirouette has finished. A large body of work has shown that the chemotaxis response is dynamic and that the degree of influence a particular chemical has on navigation can be changed, or even reversed depending on experience. Changes are reversible, specific to the chemical in question, and can be generated by classical conditioning experiments. All of these are hallmarks of associative learning, a sophisticated process that requires integration of multiple signals to produce a coordinated change in a behavioral response. We focus on the worm's chemotaxis and its ability to learn associations between salt (NaCl) concentrations and food. We draw upon existing experimental data from a variety of sources including electrophysiological and anatomical data to construct a simplified NaCl chemotaxis circuit in C. elegans. The circuit includes the two key NaCl sensory neurons ASEL and ASER, a simplified integration unit and a motor interface. We also define a set of experimentally observed behaviors we wish to reproduce including gentle turning, modulation of reversals and pirouette frequency, control of final orientation following a pirouette, and associative learning. In particular, we are interested in the alteration in behavioral response to NaCl that arises due to the pairing of high concentrations of NaCl with food or starvation. We present a computational model of NaCl chemotaxis and learning and show that model worms, placed in a simulated environment, exhibit qualitatively realistic chemotaxis behavior and adaptation. We further demonstrate that our model is robust and tolerant to noise. Our proposed chemotaxis circuit leads to a number of distinct predictions that could be used to test the model experimentally, including specific consequences from our postulated locus of associative learning and predicted consequences from ablations of ASEL and ASER.

Netta Cohen et al. (2005) International Worm Meeting "C. elegans forward locomotion control: A computational and simulation modeling approach"

Netta Cohen, John A Bryden

[International Worm Meeting, 2005]

We study a minimal theoretical model of the neural control for C. elegans forward locomotion. We begin with a review of the evidence about the neural circuitry, neuronal function and muscle behaviour including molecular level studies as well as studies of wild-type and mutant behaviour. Based on the available evidence, and in particular on the sparseness (and location) of synaptic connections, we construct plausible model of the neuronal circuitry. Surprisingly, we have found no evidence for central pattern generator control of forward locomotion. In fact our model circuit lacks any synaptic inhibition. Instead, our model relies on a chain of oscillating segments that use local feedback from stretch receptors placed along VB and DB motor neurons, to sustain and propagate oscillations. A computer optimisation approach was used to seek working parameters for model circuits. Simulations of optimised model circuits were found to be capable of generating undulations consistent with earlier mechanical models of the worm. Our simulations indicate that the model circuit would give rise to stable undulations that are robust to random perturbations. In addition, based on this model circuit, we make a variety of predictions about the role of command neurons, the location of stretch receptors along motoneurons, and the role of the head and tail in generating undulations.

Voegtlin, Thomas et al. (2011) International Worm Meeting "An Open-Source Neuromechanical Model of C. Elegans Locomotion."

Cohen, Netta, Voegtlin, Thomas

[International Worm Meeting, 2011]

A neuromechanical model of locomotion in C. elegans was recently proposed by Jordan H. Boyle [1]. One of the main results is that both swimming and crawling can be generated by a single neural circuit, reflexively modulated by the environment. This supports the known experimental results showing that different forms of C. elegans forward locomotion (e.g., swimming and crawling) can be described by a modulation of a single biomechanical gait [2]. The modelling result illustrates the importance and the potential of neuromechanical simulations for the analysis of the worm's behaviour. In order to continue this work, and to make it usable by a broader audience, we have developed a similar neuromechanical model of the worm using CLONES. CLONES (Closed Loop Neural Simulation) is an open source framework for neuromechanical simulations. CLONES implements a communication interface between a neural simulator, called BRIAN [3], and a physics engine for biomedical applications, called SOFA [4]. BRIAN and SOFA are open-source simulators that are easy to use and provide high performance. Our implementation of the worm's locomotion reproduces the neural model described in [1]. However, there are two key differences between the original physical model and our implementation. Firstly, Boyle's model considers that the body of the worm has zero mass (a low Reynolds number approximation). In contrast, the SOFA simulator allows us to integrate equations with mass and inertia. Secondly, the original model uses rigid rods of fixed length orthogonal to the body axis (approximating the incompressibility of the body due to high internal pressure). In SOFA rigid rods are modeled as springs of very high stiffness. The physical system simulated in SOFA is described using a XML syntax. The neural network model interpreted by BRIAN is written in Python, using MATLAB-like syntax. Thus, the model is completely interpreted, and it is possible to visualize/interact with the simulation during runtime. Physical environments containing obstacles or chemical concentration gradients can be defined easily. References 1. Boyle JH: C. elegans locomotion: an integrated approach. PhD thesis, university of Leeds, 2009 2. Berri S, Boyle JH, Tassieri M, Hope IA and Cohen N, Forward locomotion of the nematode C. elegans is achieved through modulation of a single gait HFSP J 3:186, 2009; 3. Goodman DF, Brette R: Brian: a simulator for spiking neural networks in Python. Front Neuroinform 2:5, 2008 4. Allard J, Cotin S, Faure F, Bensoussan PJ, Poyer F, Duriez C, Delingette H, Grisoni L: SOFA - an Open Source Framework for Medical Simulation. Medicine Meets Virtual Reality (MMVR'15), pp. 13-18, 2007.

Sanders, Tom et al. (2015) International Worm Meeting "Nonlinear sensory integration in C. elegans: a computational model."

Cohen, Netta, Ghosh, D.D., Sanders, Tom, Nitabach, M.N.

[International Worm Meeting, 2015]

Animal survival depends on a combination of often conflicting demands such as foraging and evading of dangers. To navigate effectively in such unknown and changing conditions, animals must continuously integrate over a variety of sensory cues, and adapt their decision making strategy in a context dependent manner. Here, we examine the neural control of a sensory integration task in the nematode C. elegans. The task involves an ASH-triggered aversive response to high osmolarity fructose and an AWA-triggered attractive response to diacetyl [1]. In the assay, worms are placed in the center of a ring of fructose; two drops of diacetyl are located outside the ring. We present a computational model, consisting of point worms, situated in a virtual arena that closely mimics this experimental assay, and endowed with a sensory motor pathway of two sensory neurons, a neural integration pathway and two motor programs (pirouettes and steering). A monoamine (PDF-2 and tyramine) modulation circuit involving RIM and ASH is overlaid on the synaptic circuit, in line with molecular data [1]. Model parameters were constrained by behavioral data for wild type and mutant animals for a range of stimulus concentrations. Based on our simulation results, we reject a null hypothesis of a linear sensory integration mechanism in RIM and present results that are consistent with the data for a sensory "coincidence detector" like process in RIM.[1] Ghosh, D.D., Sanders, T., Hong, S., Chase, D.L., Cohen, N., Koelle, M.R., and Nitabach, M.N. "Neuroendocrine reinforcement of a dynamic multisensory decision." International C. elegans meeting.

Brittin, Christopher et al. (2021) International Worm Meeting "Nerve ring reconstructions reveal principles of brain organization across larval development"

Brittin, Christopher, Cohen, Netta

[International Worm Meeting, 2021]

Animal nervous system organization relies on features at every scale, from nano-level localization of synapses, through neuronal morphologies, to the high-level stereotyped connections between different regions of the brain. Previously, we developed methodologies for determining and characterizing multi-scale brain features from reconstructed serial sectioned electron micrographs (EM) of the L4 and adult nerve rings1. To obtain a development timeline of the nerve ring, we now integrate our analysis with results that we obtain from new EM datasets for the L1, L2, L3 and adult2. We find that membrane contacts between neurites are well described by a conserved core, embedded in a sea of variable contacts. We present a parsimonious model that consistently predicts that about 28-33% of conserved membrane contacts are actively targeted for synapse formation with high precision (≈93%), while the significant variability across datasets is accounted for by a non-negligible basal synaptic contact rate (≈20-30%), across all these developmental stages. Thus, while the numbers of membrane and synaptic contacts increases with age, our model predicts that the tendency to make synaptic contacts remains relatively constant. By clustering membrane contacts on each of the animals, we found a nerve ring organization of five spatial neighborhoods that supports a similarly modular information processing synaptic circuit. Extending this analysis over development, we find that the structural and synaptic modularity of the nerve ring is robust across all developmental stages, indicating that the macro-structure of the nerve ring is rooted in embryonic development (see accompanying abstract for a postulated mapping between this structure of the nerve ring and collective cell behaviors in the embryo3). We further present a developing brain map (a complete single cell resolution synaptic map of the C. elegans nerve ring) and use it to highlight key features of the post-embryonic development of the nerve ring. Our network analysis of the brain map points to a combination of individuality and robustness of brain organization that likely scale to larger nervous systems. 1. Brittin et al., (2021) Nature. https://doi.org/10.1038/s41586-021-03284-x 2. Witvliet et al. (2020), biorxiv. https://doi.org/10.1101/2020.04.30.066209 3. Brittin et al. (2021). Multicellular rosettes organize neuropil formation. [C. elegans conference abstract]

Sanders, Tom et al. (2013) International Worm Meeting "The computational role of the head navigation circuit in C. elegans: Exploring the interneuron layer in silico."

Cohen, Netta, Sanders, Tom

[International Worm Meeting, 2013]

In the nematode C. elegans the head circuit drives the worm's navigation. Interestingly, computational models of C. elegans navigation tend to focus on the sensory layer and motor outputs, completely neglecting the complex interneuron circuit midstream. What then are the computational roles of the ~30 interneurons in this circuit? Here, we propose a role in sensory integration. We created computational models of the worm situated in virtual environments mimicking an integration assay. In this assay animals are presented with an attractive odorant (diacetyl) on the opposite side of an aversive barrier (copper). Animals have to integrate both signals and either cross the barrier to reach the attractant or stay on their side of the barrier. Behavior is measured by the number of worms crossing the barrier as a function of the barrier and attractant concentrations. We tested the ability of different models, with increasing biological detail and complexity, to perform effective sensory integration. Our final model is capable of reproducing the behavior of wild type worms and known mutants, for separate stimuli and combinations thereof. Interestingly, highly abstract models could already show complex integration through nonlinearities in their motor system alone. The model sheds light on the computational role of specific head interneurons, and makes a number of concrete predictions, including a new stimulus encoding strategy in sensory neurons and a previously undescribed motor program.

Holbrook, Robert et al. (2015) International Worm Meeting "TRAMS: A system for three-dimensional tracking of behaviour."

Holbrook, Robert, Cohen, Netta

[International Worm Meeting, 2015]

Under natural conditions C. elegans moves through three-dimensional environments, yet in the lab the majority of locomotion research has focused on movement on two-dimensional surfaces. This raises the question of whether a significant amount of the worm's movement, locomotion and behaviour has gone unexplored and unexplained. To address this we designed and built a tri-axial microscope system (TRAMS). Computer vision is used to track the position, orientation and configuration of the worm body in three-dimensions. This is achieved by placing a worm in a glass cube containing fluid. Three cameras were placed facing three adjacent sides, illuminated with infrared backlighting. Standard photogrammetry was then used to solve the camera geometry and reconstruct known points within the volume. Contrast-informed shape carving was used for the images in each camera to create a voxel for each worm, which was then skeletonised. Preliminary three-dimensional kinematic data of wild-type and mutant strains of C. elegans are presented.

Brittin, Christopher et al. (2015) International Worm Meeting "Spatial aggregation of neurons contributes to network structure in the nematode C. elegans."

Cook, Steven, Cohen, Netta, Brittin, Christopher

[International Worm Meeting, 2015]

Previous studies of real-world networks have suggested that networks motifs, defined as connectivity patterns that are significantly over-represented when compared to random networks with the same degree distribution, may arise due to evolutionary design principles and serve as computational units. For neural networks and other spatially embedded systems, where network connections form between physically proximate nodes, this approach carries a risk of overstating the statistical significance of connectivity patterns. Previous studies have attributed a number of network motifs to the C. elegans neuronal network and suggested that the worm's nervous system may be constructed from these computational modules (Milo et al., 2002; Reigl et al. 2004). However, other groups have conjectured that the high frequency of observed connectivity patterns may simply be a consequence of the organization and localized connectivity of the neuropile (White et al., 1983; Artzy-Randrup et al. 2004). To test these two hypotheses, we measured the spatial aggregation of neurons in the nematode C. elegans and used the data to construct a statistical model with a spatially constrained null-hypothesis. We found that a number of motifs, including the 3-node feed forward loop, are no longer over-represented in our spatially constrained model. Thus, the observed network structure in the C. elegans nervous system may simply be the consequence of how the neuropile is organized.

Sanders, Tom et al. (2013) International Worm Meeting "A circuit for decision making in C. elegans: a computational approach."

Jansen, Gert, Sanders, Tom, Cohen, Netta

[International Worm Meeting, 2013]

The head navigation circuit in the nerve ring of C. elegans performs a wide variety of computations with a limited number of neurons. Indeed, overlapping neurons have been implicated in navigation, sensory integration and decision making. Here, we create a computational model capable of reproducing a wide variety of behaviors, as observed in various assays performed on wild type, mutant and other defective animals. To this end, we developed a modular computational framework, in which neuronal, circuit and motor output functions and properties, as well as the assay specification can all be easily modified. In our model, individual worms are represented by position, direction pairs controlled by a reduced nervous system consisting of several sensory neuron pairs, an interneuron layer and an abstract motor system. We focus on a decision making assay, in which animals are placed on a dish with quadrants containing a particular concentration of NaCl or no NaCl. Before the assay, the animals are washed for 15 minutes with a low salt buffer to test their naive response, or a buffer containing 100 mM NaCl to test the response of animals that have associated NaCl with the absence of food (called gustatory plasticity). We show that the model is capable of reproducing the behavior of naive and conditioned animals in decision making. Model analysis suggests specific neuronal roles and circuit mechanisms. Specifically, our model distinguishes the roles of ASEL and ASER in decision making and suggests that ASH sensory neurons are recruited into the functional circuit, switching the preferred decision of the animals. We are currently testing the performance of our model in other navigation assays, for example a NaCl chemotaxis assay where animals are exposed to a shallow gradient of NaCl and a copper-diacetyl integration assay. The ability to validate the model by testing it in a variety of assays, and under a variety of conditions is a first step towards building more complete models of ever richer behaviors in C. elegans.

Berri, Stefano et al. (2009) International Worm Meeting "Agar groove masks locomotion related phenotypes."

Berri, Stefano, Boyle, Jordan H., Cohen, Netta, Tassieri, Manlio, Hope, Ian A.

[International Worm Meeting, 2009]

When C. elegans crawls on agar, it carves a groove in the surface and leaves a track behind it. It has been postulated that this groove is integral to the mechanism of crawling, allowing the sinusoidal trajectory of the head to dictate the shape on the rest of the body. Indeed, a stiff groove would result in a strong asymmetry in the environment''s resistance to motion in the normal (sideways) and longitudinal (along the body) directions, constraining the movement of the body. However, our recent findings show that the groove is not essential to generate a crawling shape in the wild type worm, as it can produce the typical crawling waveform even on a flat non-deformable surface. These results are consistent with the fact that swimming and crawling actually represent a single behavior[1], produced by a single neural control system. Any mechanism that fails without a groove could not account for locomotion in water. We next asked whether the physical forces the groove applies could have greater significance when the worm is defective. We recorded various locomotion-related mutants moving in a range of media with increasing visco-elasticity. We found that the movement of unc-8 (e49) and vab-7 (e1562) mutants on agar is very similar to that of wild type worms. When observed in water, however, it is clear that they are highly defective in the posterior half of their body which usually fails to show dorso-ventral oscillations at the same frequency of the head. When observed in media with increasing stiffness, the phenotype gradually disappears as the properties of the environment progressively mask the defect in locomotion.We conclude that while the groove has minimal importance to wild type locomotion, it can have a strong masking effect when the worm is defective. We suggest that, when screening for locomotion related phenotypes, water may be a more appropriate medium than agar. A screen for mutants defective in swimming may actually be a screen for mutants affected in their single locomotion gait, but for which the phenotype is masked by the groove on the agar surface. [1] Berri S, Boyle JH, Tassieri M, Hope IA and Cohen N. 2009. "Forward locomotion of the nematode C. elegans is achieved through modulation of a single gait." HFSP Journal, In press.