| The accumulation and storage of information over time, temporal integration, is key to numerous behaviors. In the oculomotor system, the integration of eye-velocity signals to eye-position commands is achieved by a hindbrain cell group called the velocity-to-position neural integrator (VPNI). Here the mechanisms underlying temporal integration were explored in the larval zebrafish VPNI using optical and computational approaches.;We used calcium imaging during oculomotor behaviors and single-cell electroporation to map VPNI activity onto a hindbrain scaffold consisting of alternating glutamatergic and glycinergic stripes, linking VPNI coding properties with structure and genotype. Three distinct classes of VPNI cells were identified. Cells from the glutamatergic alx class had ipsilateral projections to motoneurons, and gave local collaterals within the VPNI consistent with integration through recurrent excitation. Cells from a second glutamatergic dbxlb class had contralateral projections to motoneurons, consistent with a role in vergence behaviors. A third putative GABAergic dbxlb class also had contralateral projections to motoneurons; these neurons collateralized onto contralateral VPNI cells, consistent with a role in coordinating activity between functionally-opposing populations. Together, these results clarify and augment the canonical positive excitatory feedback hypothesis of integration.;To gain insight into the role of feedback interactions supporting integration, provided through connectivity and dendritic processing, we built a conductance-based multi-compartment network model in which dendritic compartments containing plateau potentials are sequentially recruited with increasing network inputs. In addition to serving as a framework for experiments, the model showed how microscopic components can effect macroscopic changes in the network, simultaneously highlighting the challenge and utility of modeling recurrent networks with realistic components.;We also directly imaged the dendrites of integrator neurons to distinguish between models for temporal integration proposing different dynamics along the dendritic arbor. The presence of discrete hotspots of activity throughout the dendritic tree constrains the possible mechanisms to those proposed by the dendritic plateau recruitment model and the simple recurrent feedback model. The distance- and branch order-dependent correlation of the hotspots to somatic activity further suggests that VPNI neurons utilize more than one layer of processing.;Together, these results provide insights into potential network and cellular mechanisms underlying temporal integration. |
