HomeResearchHow Do Neurons Compute Output from Their Inputs?

Our Scientists

How Do Neurons Compute Output from Their Inputs?

Research Summary

Nelson Spruston wants to understand how the properties of individual neurons contribute to the computational performance of neural circuits controlling cognition and behavior. His lab has focused on how the structurally and molecularly elaborate dendritic trees integrate thousands of synaptic inputs to generate action potential firing in the axon, and how the axon itself performs surprisingly sophisticated integrative functions.

Neurons are the basic computational workhorses of neural circuits, but the nature of the computations these brain cells perform is poorly understood. The key to understanding this crucial aspect of the nervous system lies in determining how each neuron integrates the information it receives via synaptic inputs from its network partners. My lab uses a combination of electrophysiology, microscopy, and computer modeling to tackle this problem in the hippocampus. Neurons in this evolutionarily ancient brain structure process information about the animal's sensory environment and its movements to generate a continuous cognitive map of the animal's state. In humans and other mammals, the hippocampus is known to be a critical structure for the formation and storage of new memories, so understanding how hippocampal neurons compute is an essential step toward understanding how our brains make us who we are.

The majority of synaptic input to most neurons arrives via the dendrites—elaborately branching processes with sophisticated molecular composition. In pyramidal neurons—the principal cell type in the hippocampus—tens of thousands of synapses decorate the dendrites of a single pyramidal neuron. The challenge is to understand how these inputs are integrated to generate output in the form of action potential firing in the axon.

Movie 1: In a computational model, synaptic integration in two dendritic branches leads to the generation of dendritic spikes in each branch. When both branches are activated strongly enough to generate dendritic spikes, an action potential is generated in the axon. Subsequently, the action potential backpropagates throughout the dendritic tree. The backpropagating action potential does not propagate into the activated branches, however, because the generation of a dendritic spike in response to synaptic input caused inactivation of sodium channels in the activated branches. The axonal action potential is generated despite the fact that the spikes generated in the narrow, oblique dendrites do not propagate (i.e., fail) as they spread into the wider, main apical dendrite. Nevertheless, the dendritically generated spikes are important, both because they increase the charge entering the cell (thus increasing the probability of action potential generation) and because they contribute to synaptic plasticity. When only a single branch is activated by synaptic input (not shown), the dendritic spike fails to propagate forward through branch points and does not deliver enough charge to the axon to generate an action potential.

Movie generated by William Kath. From Katz, Y., Menon, V., Nicholson, D.A., Geinisman, Y., Kath, W.L., and Spruston, N. 2009. Neuron 63:171–177. © 2009, with permission from Elsevier.

The membranes of dendrites are excitable, so they can amplify the effects of synaptic inputs. My lab has used patch-clamp recordings from dendrites as small as 1 µm in diameter to demonstrate this directly. We have shown that a variety of voltage-gated channels, combined with dendritic branching patterns, determine how synaptic inputs are integrated. For example, we have shown that voltage-gated sodium channels and voltage-gated calcium channels contribute to the generation of different kinds of spikes in dendrites, as well as to the backpropagation of axonally generated action potentials through the dendritic tree. We have also shown that the forward propagation of dendritic spikes (i.e., toward the soma) and the backpropagation of action potentials in the other direction (i.e., into the dendrites) are differentially affected by branch points in the dendritic tree (see Movie). Understanding the intricacies of dendritic integration is important because these processes determine the computations performed by neurons, which ultimately contribute to the neural control of perception, cognition, and behavior.

Experimental work in my lab has focused on a detailed analysis of hippocampal neurons using patch-clamp recordings, including dendritic recordings, in hippocampal slices. We have also used both light and electron microscopy to perform detailed anatomical analyses. To understand how the functional and structural properties we uncover interact to produce a functional dendritic tree, we use computational modeling. This approach has allowed us not only to consider the functional implications of experimental observations but also to make experimentally testable predictions, which has led us to new discoveries.

In addition to computing, neurons also store information, providing neural circuits like the hippocampus with the ability to mediate learning and memory. My lab has made important contributions to understanding these processes in hippocampal pyramidal neurons. First, we have shown that both dendritic spikes and backpropagating action potentials make important contributions to synaptic plasticity. These events are the results of synaptic integration. Thus, there is a close relationship between dendritic structure, voltage-gated channels in dendrites, synaptic integration, and synaptic plasticity. In addition, however, my lab has shown that the intrinsic excitability of hippocampal neurons is also plastic. This plasticity results from a combination of synaptic integration and activation of modulatory neurotransmitter receptors. These modulatory receptors also have a well-established role in the regulation of synaptic plasticity, and they are known to be affected in neurological disorders such as Alzheimer's disease. Our findings suggest that neuromodulation regulates both synaptic strength and the way that synaptic inputs are integrated in the dendritic tree. Modulation of both processes works together to achieve information storage in individual neurons.

In addition to studying the excitatory pyramidal neurons, we have also studied inhibitory interneurons. Our work in this area led to the surprising discovery that some hippocampal and neocortical interneurons can also integrate information in their axons. These neurons are able to integrate action potential firing in the axon for tens of seconds to minutes. The outcome of this integration is a sudden onset of persistent firing that itself lasts tens of seconds to minutes. All of this occurs without the direct involvement of dendrites, thus identifying a previously unknown integrative function for axons. Even more surprisingly, however, firing in the axon of one neuron could lead to persistent firing in the axon of a neighboring neuron, indicating that the axons of these neurons can communicate with each other. The mechanisms by which axons may form autonomously communicating networks are not known, but we continue to study this issue in my lab.

We have also begun to study the diversity of hippocampal pyramidal neurons. The nervous system is made up of a richly diverse set of neurons that integrate and store information in a variety of ways. Differences in the structure of dendritic trees are complemented by differences in molecular composition and synaptic properties, indicating that different types of neurons process information in unique ways. The goal of my research is to determine the principles by which the cellular and molecular properties of neurons contribute to their computational properties and thus achieve sophisticated cognitive functions such as learning and memory. We seek to identify not only principles that apply to all neurons but also specializations that allow different kinds of neurons to perform different computations, thus endowing the nervous system with the broad ability to process and store information in a variety of ways.

This work has been supported in part by grants from the National Institute of Neurological Disorders and Stroke.

As of January 23, 2012

Scientist Profile

Janelia Group Leader
Janelia Research Campus