Dendritic nonlinearities are tuned for efficient spike-based computations in cortical circuits

Balázs Ujfalussy, Judit K. Makara, Tiago Branco, Máté LengyeleLife 2015 · DOI

TL;DR

Cortical neurons use nonlinear processing in their dendrites to efficiently integrate incoming signals, a process that is surprisingly well-explained by a theory linking dendritic nonlinearities to the statistical properties of presynaptic activity. This work bridges the gap between cellular and systems-level neuroscience, suggesting a fundamental computational principle for how neural circuits perform analog computations with spiking neurons.

What Problem Does This Paper Address?

The brain is incredibly efficient at processing complex information, and a key part of this is how individual neurons, particularly their dendrites, handle the thousands of inputs they receive. These inputs are not simply summed up; dendrites exhibit complex, nonlinear behaviors that can lead to either sublinear or superlinear integration of signals. While these dendritic nonlinearities have been studied at the single-neuron level, it's been unclear how they contribute to the overall computation performed by neural circuits. This paper aims to bridge this gap by developing a theoretical framework to explain how dendritic nonlinearities are tuned to the statistical properties of incoming signals, and how this tuning enables efficient computation in spiking neural circuits.

Methods

The authors developed a computational model based on Bayesian inference to predict the "optimal" way a neuron should integrate its inputs to perform a specific computation, given the statistical properties of those inputs. This theory formalizes the idea that the optimal integration strategy depends on the statistics of presynaptic activity, including firing rates, membrane potential fluctuations, and correlations between neurons. They then compared the predictions of this theory to experimental data obtained from two-photon glutamate uncaging in cortical pyramidal neurons. This technique allows for precise stimulation of dendritic inputs and recording of the resulting subthreshold responses. They also explored simplified neuron models (linear, somatic nonlinear, random dendritic nonlinear, clustered dendritic nonlinear) to understand the computational advantage of dendritic nonlinearities.

Datasets

The study utilized in vivo data to characterize presynaptic population activity statistics. Specifically, they used data on:

  • Membrane potential fluctuations of layer 2/3 pyramidal cell-pairs in the barrel cortex during quiet wakefulness.
  • Membrane potential fluctuations and multineuron spiking patterns of hippocampal pyramidal cells during sharp wave activity.

These datasets were used to fit the parameters of their statistical model of presynaptic activity, which then informed the predictions of the optimal dendritic response.

Key Insights and Ideas

  • Dendritic nonlinearities are not arbitrary: The paper proposes that dendritic nonlinearities are not just a biophysical quirk but are actively tuned to match the statistical properties of the incoming synaptic activity.
  • Correlation is key: The degree and type of correlation between presynaptic neurons fundamentally dictate whether linear or nonlinear integration is optimal. Independent inputs are best integrated linearly, while correlated inputs demand nonlinear processing.
  • A unified principle: The theory suggests that dendritic nonlinearities, particularly those mediated by NMDA receptors, are crucial for approximating optimal input integration, effectively bridging the gap between the "digital" nature of spikes and the "analog" computations performed by circuits.
  • Clustering matters: The theory predicts that correlated inputs should cluster their synapses on dendrites, allowing for efficient engagement of nonlinear integration mechanisms.

Key Results

  • The authors' theory accurately predicted the responses of two types of cortical pyramidal cells (neocortical and hippocampal) to patterned stimulation, demonstrating a strong match between experimental data and theoretical predictions when presynaptic statistics were correctly accounted for.
  • When presynaptic neurons were statistically independent, linear integration was optimal. However, with correlated presynaptic activity, nonlinear integration became essential, with the specific form of nonlinearity (sublinear or supralinear) depending on the nature of the correlations.
  • A simple nonlinear dendritic model with clustered connectivity achieved near-optimal performance in approximating the true signal, significantly outperforming linear or purely somatic nonlinear models.
  • Pharmacological blockade of NMDA receptors disrupted the precise match between dendritic integration and presynaptic statistics, highlighting the critical role of NMDA receptor-mediated nonlinearities.

Limitations and Future Work

  • Simplification of computation: The authors assumed a linear computational function for the postsynaptic neuron when fitting experimental data, acknowledging that real neurons perform more complex nonlinear computations. They suggest that future work could explore these more complex functions.
  • Tractability of the optimal response: The full mathematical implementation of the optimal response is computationally complex, necessitating approximations and simplified models for practical analysis.
  • Inhibitory neurons: The study focused on excitatory inputs to pyramidal neurons. The authors suggest that their theory can be extended to include inhibitory neurons, which may play a crucial role in supporting computations.
  • Synaptic clustering: While the theory predicts synaptic clustering for correlated inputs, experimental confirmation remains an active area of research. Future work could further investigate this prediction.
  • Adaptation mechanisms: The paper proposes several mechanisms for how dendritic nonlinearities might adapt to changing presynaptic statistics over different timescales, suggesting avenues for future experimental and theoretical investigation.

Summary updated

Abstract

Cortical neurons integrate thousands of synaptic inputs in their dendrites in highly nonlinear ways. It is unknown how these dendritic nonlinearities in individual cells contribute to computations at the level of neural circuits. Here, we show that dendritic nonlinearities are critical for the efficient integration of synaptic inputs in circuits performing analog computations with spiking neurons. We developed a theory that formalizes how a neuron's dendritic nonlinearity that is optimal for integrating synaptic inputs depends on the statistics of its presynaptic activity patterns. Based on their in vivo preynaptic population statistics (firing rates, membrane potential fluctuations, and correlations due to ensemble dynamics), our theory accurately predicted the responses of two different types of cortical pyramidal cells to patterned stimulation by two-photon glutamate uncaging. These results reveal a new computational principle underlying dendritic integration in cortical neurons by suggesting a functional link between cellular and systems--level properties of cortical circuits.
BibTeX
@inproceedings{ujfalussy2015dendritic,
  author = {Balázs Ujfalussy and Judit K. Makara and Tiago Branco and Máté Lengyel},
  title = {{Dendritic nonlinearities are tuned for efficient spike-based computations in cortical circuits}},
  booktitle = {eLife},
  year = {2015},
  doi = {10.7554/elife.10056},
}
Figure