Dendritic branch structure compartmentalizes calcium-dependent plasticity signals and supports the acquisition of complex tuning functions

Abstract

Life as an organism depends on flexible and intelligent neural activity for perceiving the sensory world, constructing models of the environment, and generating adaptive behaviors. It is evident that the collective activity of all the neurons in the brain accomplishes this goal, even though each individual neuron only has access to the activity of its synaptic partners. Somehow, developmental mechanisms and activity-dependent learning rules must lead to the emergence of adaptive neural by shaping neural connectivity throughout the brain. In this thesis, we investigated the mechanisms and computational functions of synaptic, activity-dependent learning rules.

First, we investigated the calcium signals underlying spike-timing dependent plasticity (STDP). STDP has been experimentally observed across brain regions and species, both in vitro and in vivo, and represents the best-established example of Hebb’s postulate. Numerous studies investigating the patterns of activity that evoke STDP and the cellular mechanisms that implement STDP have pointed towards control of calcium signaling as the key locus for activation of STDP-dependent potentiation or depression. However, most studies have focused on global cellular properties of the calcium signals involved in STDP and have not explored intracellular variety in how these signals are regulated in different compartments of dendritic trees.

We investigated intracellular variation in activity-dependent calcium signals by performing whole-cell recording, two-photon calcium imaging, and two-photon glutamate uncaging. We found that back-propagating action potential (bAP) evoked calcium influx, which is critical for the signaling cascade that evokes STDP-dependent depression, is selectively reduced in a subset of dendritic branches in cortical layer 2/3 (L2/3) pyramidal cells of the somatosensory cortex in mice. Interestingly, bAP-dependent amplification of synaptically-evoked calcium influx, which is critical for STDP-dependent potentiation, is retained in these dendritic branches, indicating an intracellular mismatch in the ratio of STDP-dependent depression and potentiation. We show that bAPs successfully propagate to all branches and that all branches contain voltage-gated calcium channels (VGCCs) that can be opened by sufficient depolarization. However, using a combination of dendritic voltage imaging and biophysical compartmental modeling, we demonstrate that dendrites with a reduction in bAP-evoked calcium influx have lower amplitude action potentials due to their elaborate dendritic branch patterns, which explains the reduction in calcium influx. Finally, we used conductance models to explain why reductions in bAP amplitude would lead to a selective reduction in bAP-evoked calcium influx through VGCCs while sparing amplification of synaptic calcium influx through NMDA-type glutamate receptors. Together, our experimental work suggests that the ratio of STDP-dependent depression to potentiation varies within individual neurons.

STDP has multiple adaptive properties. Computational work has shown that simple implementations of STDP rules leads to acquisition of input selective tuning functions and homeostatic regulation of firing properties. Although these properties are attractive as models for cellular learning, these models are limited in their ability to explain the diversity and complexity of experimentally observed receptive fields in vivo. To explore how intracellular variation in the ratio of depression to potentiation affects the tuning functions that emerge from STDP, we adapted previous models to reflect the properties of neurons we observed in slice preparations. We demonstrate that the ratio of depression to potentiation scales with the average pairwise correlation between strongly weighted synaptic inputs. This finding suggests that complex tuning functions, which require synaptic input from weakly correlated inputs, depends on reductions in synaptic depression at distal synapses. We then used a simplified model of visual cortical processing to demonstrate that the observed distribution of dendritic spine tuning in vivo only emerges when distal inputs have a reduction in synaptic depression, just as we predict from our experimental data. This work demonstrates how a combination of simple computational rules and complex biological parameterization can support a richer and more sophisticated understanding of cellular learning in the brain.