Aim 1: Define computational rules of dendritic integration
We showed that dendrites in medial entorhinal cortex neurons exhibit supralinear input–output transformations (Schmidt-Hieber et al. 2017), demonstrating that dendrites do not simply sum inputs but perform nonlinear computations that define the functional signature of neurons. This supported the proposal’s goal of defining dendritic computational rules by providing empirical evidence of active dendritic integration in a functionally relevant system (grid cells).
We also demonstrated that Purkinje cell dendrites exhibit nonlinear filtering properties that can be modulated by noise, affecting their excitability and output (Buchin et al. 2016). This suggested that dendrites follow complex computational rules beyond simple integration, reinforcing the idea that dendritic branches act as independent computational units.
Using detailed modelling of presynaptic input and postsynaptic integration, we showed (Goetz et al. 2021), that as few as 1% of the strongest synaptic inputs to a pyramidal neuron can determine the tuning of its action potential output in vivo. These strong but sparse synaptic inputs exert their influence on the output by triggering dendritic Na+ and NMDA spikes.
Aim 2: Map sensory-driven dendritic computations in vivo
We showed that cortical feedback selectively engages dendritic nonlinearities, supporting a role for dendritic computation in processing of specific visual stimuli (Fisek et al. 2023). This paper provided causal evidence for active dendrites contributing to sensory processing in vivo, aligning with the goal of mapping how sensory patterns activate dendritic computations.
We developed a closed-loop optogenetic system for manipulating and recording neuronal activity in awake animals, allowing targeted stimulation of individual dendritic branches (Zhang et al. 2018). This paper provided a crucial methodological approach for mapping how sensory input is processed at the dendritic level.
Aim 3: Understand how circuit activity engages dendritic computation
We showed that dendritic signals triggered by climbing fiber inputs to Purkinje cells encode predictive and reactive reward signals, suggesting that Purkinje cell dendrites play a role in encoding distinct behavioural states (Kostadinov et al. 2019).
We provided evidence that specific inputs are mapped onto distinct dendritic regions, supporting the idea that dendrites, not entire neurons, function as computational subunits. Specifically, we demonstrated that feedback projections selectively engage dendrites in the visual cortex, suggesting that circuit-level activity is structured to interact with dendritic computation (Fisek et al. 2023). This paper supported the proposal’s goal of understanding how synaptic input is spatially and functionally organized onto dendrites.
Aim 4: Establish causal links between dendritic computation and behaviour
We enabled precise, real-time manipulation of activity in awake animals, providing a new tool for researchers to test how dendritic computations influence behaviour (Zhang et al. 2018). We developed and disseminated a toolkit for experimentally manipulating dendritic computations and test their causal role in sensory processing and behaviour.
Conclusion
Our research provided direct experimental evidence and methodological tools to investigate the proposal’s hypothesis that individual dendrites, rather than the whole neuron, are the fundamental units of computation in the brain.
