Modeling somatic and dendritic spike mediated plasticity at the single neuron and network level

Jacopo Bono, Claudia ClopathNature Communications 2017 · DOI

Introduction

For decades, computational neuroscience has relied heavily on the "point neuron" abstraction—treating a neuron as a single dimensionless point that sums inputs and fires an output. In these models, learning is usually governed by Spike-Timing-Dependent Plasticity (STDP), where a synapse strengthens only if a presynaptic input is closely followed by a postsynaptic action potential (the cell's main output spike). However, real neurons are sprawling, tree-like structures. Their branches, or dendrites, are not just passive cables; they are active processors capable of generating their own local electrical events, such as NMDA spikes. This research investigates how these local dendritic events interact with global somatic spikes to shape learning and solve the "stability-plasticity dilemma"—the challenge of learning new information without immediately overwriting old memories.

Biophysical Modeling of Dendritic Integration

Researchers Jacopo Bono and Claudia Clopath employed multi-compartmental biophysical models of cortical pyramidal neurons (specifically Layer 5 and Layer 2/3) using the Brian2 simulator. Unlike standard neural network models, these simulations included detailed ion channel dynamics (sodium, potassium, and calcium) distributed across the dendrites.

Figure 1: Biophysical model of a pyramidal neuron showing location-dependent spike thresholds and integration gradients. (a) Neuron morphology. (b) Voltage responses to synaptic inputs with varying temporal offsets. (c-d) Spatial distribution and histogram of thresholds for NMDA and somatic spikes.

They implemented a voltage-dependent plasticity rule, which allows synaptic changes to occur based on local membrane potential rather than just the timing of the final output spike. To scale their findings to the network level, they developed a "reduced" compartmental model that captured the essential non-linearities of the full biophysical version while remaining computationally efficient enough to simulate associative learning in groups of neurons.

The Plasticity Gradient and Decoupled Learning

A key insight of the study is the "plasticity gradient." STDP is most effective near the cell body (soma), where the backpropagating action potential is strongest. As you move further out into the dendrites, this signal weakens, making traditional STDP difficult. However, distal regions compensate by being much more sensitive to local NMDA spikes.

This leads to a decoupling of learning from output. In point neurons, a cell must fire to learn. In dendritic models, distal synapses can undergo Long-Term Potentiation (dLTP) via local spikes even if the neuron as a whole remains silent. This decouples the "learning" signal from the "output" signal. Furthermore, distal inputs can act as a "teacher" or "gate": a local dendritic spike provides enough sustained depolarization to allow other, weaker proximal inputs to reach the threshold for STDP.

Memory Protection and Network Stability

The model predicts that synapses clustered on the same dendritic branch are far more likely to form stable, bidirectional connections. Re-analysis of experimental data from Markram et al. (1997) confirmed this prediction: bidirectional connections were found significantly further from the soma ($106.9 \pm 33.7\ \mu m$) than unidirectional ones ($89.6 \pm 28.6\ \mu m$).

In associative learning simulations, networks with active dendrites retained memories significantly longer than point-neuron networks. Because dLTP doesn't require a global somatic spike, it is less susceptible to interference. The dendritic spikes effectively "shielded" the learned weights from being degraded by the noise of subsequent, unrelated neuronal firing. This provides a biophysical rationale for why biological brains are more adept at lifelong learning than current artificial neural networks.

Limitations and Future Work

The authors acknowledge several simplifications, such as the focus on basal dendrites and the omission of the apical tuft's complex calcium dynamics. Additionally, the model omitted inhibitory synapses, which likely play a massive role in shunting or controlling local spikes in a real brain. Future work could explore how these synaptic clusters form in the first place through structural plasticity—the physical growing and pruning of connections.

Summary updated

Abstract

Synaptic plasticity is thought to be the principal neuronal mechanism underlying learning. Models of plastic networks typically combine point neurons with spike-timing-dependent plasticity (STDP) as the learning rule. However, a point neuron does not capture the local non-linear processing of synaptic inputs allowed for by dendrites. Furthermore, experimental evidence suggests that STDP is not the only learning rule available to neurons. By implementing biophysically realistic neuron models, we study how dendrites enable multiple synaptic plasticity mechanisms to coexist in a single cell. In these models, we compare the conditions for STDP and for synaptic strengthening by local dendritic spikes. We also explore how the connectivity between two cells is affected by these plasticity rules and by different synaptic distributions. Finally, we show that how memory retention during associative learning can be prolonged in networks of neurons by including dendrites.Synaptic plasticity is the neuronal mechanism underlying learning. Here the authors construct biophysical models of pyramidal neurons that reproduce observed plasticity gradients along the dendrite and show that dendritic spike dependent LTP which is predominant in distal sections can prolong memory retention.
BibTeX
@inproceedings{bono2017modeling,
  author = {Jacopo Bono and Claudia Clopath},
  title = {{Modeling somatic and dendritic spike mediated plasticity at the single neuron and network level}},
  booktitle = {Nature Communications},
  year = {2017},
  doi = {10.1038/s41467-017-00740-z},
}
Figure