We developed an unsupervised clustering method that can robustly identify different swim bout categories based on a combination of kinematic similarity and behavioral usage (Marques et al., 2018; Marques and Orger, 2019). We found that larvae show a small set of preferred swim patterns, which systematically tile the space of bout kinematics. While some patterns are used broadly across many conditions, others are used only in very specific contexts. By analyzing the sensory stimuli preceding different bout types in freely swimming fish in social or feeding contexts, we could identify the natural stimuli that trigger different bout types. This quantitative analysis of behavior provides a crucial framework for subsequent investigations of neural circuit function. While this analysis depended on characterization of high-speed kinematics based on high resolution recordings, we have shown, using machine learning methods (Jouary et al., 2024; megabouts.ai) that high-speed kinematics can be reliably inferred even from low resolution data, enabling high-throughput applications in the areas of drug screening or disease phenotyping. Using a approach based on control theory, we extending this framework to allow for more a more continuous representation of swimming kinematics, and infer the underlying dynamics and control architecture (Mullen et al. 2024)
We used the swim bout framework to investigate the effects of experience on social avoidance in zebrafish larvae (Groneberg et al., 2020). Week old larvae avoid each other at short distances, but, for larvae reared in isolation, this distance is increased. We found that isolation reared larvae use different bout types in response to social stimuli that are sensed by the lateral line. This work establishes a powerful paradigm to investigate the effect of experience on the development of social circuits.
A major challenge in neuroscience is to relate activity in large populations of neurons to behavioral features which are also high-dimensional. In collaboration with Christian Machens’ group (Feierstein et al., 2023), we applied dimensionality reduction methods to identify behavioral features represented in population activity in the hindbrain. Surprisingly we found that just two behavioural features are sufficient to explain most of this activity, and that these features define functional clusters of neurons that are distributed in a stereotyped spatial organization across individuals.
To interpret whole-brain activity maps, and propose realistic circuit models, it is critical to know not just where active neurons are, but also their morphology, projection patterns and neurotransmitter phenotypes. We have generated transgenic lines targeting different neurotransmitter classes and cell-types and passed each through an imaging pipeline to record the responses of labeled neurons to a suite of behaviorally relevant stimuli. One line identified in this screen labeled a previously uncharacterized cluster of inhibitory direction selective (DS) neurons in the rostral thalamus (see Figure). Thalamic DS neurons mostly have binocular receptive fields and show a striking topographic map of direction preference, in contrast with pretectal DS neurons. Since the anatomy and function of the thalamus in larval fish is not well understood, we undertook to more systematically map visual and motor responses throughout this region, together with in situ expression patterns of markers for subdivisions of the mammalian thalamus.
Climbing fibre projections from the inferior olive (IO) constitute one of the two major excitatory inputs to cerebellar Purkinje Cells. We set out to investigate the functional organization of this pathway in zebrafish (Felix, Markov et al., 2024). Based on stochastic single neuron labeling, inferior olive neurons can be divided into anatomical classes that differ in their dendritic morphology and axonal projection fields. We mapped the responses of IO neurons to whole-field motion stimuli, using two-photon imaging. Neurons sensitive to rotational vs. translational motion and different directions of motion, were localized to distinct regions of the olive. To understand how this organization shaped the projections to the cerebellar cortex, we used volumetric light-sheet imaging of the axon terminals and IO cell bodies simultaneously, to map the point-to-point topography of the projection.
