Simulated neural circuits can generate similar output despite variability in the intrinsic properties of individual neurons and the connections among them (synaptic properties). But few studies have attempted to actually measure just how much a particular neural circuit varies between individuals in an animal population. Jean-Marc Goaillard, Adam Taylor, David Schulz and Eve Marder took on this challenge in a recent paper published in Nature Neuroscience.
The researchers work with a small neural network in the crab nervous system: it consists primarily of a pair of coupled, rhythmically active cells (pacemakers), which make inhibitory synapses onto two types of follower neurons. The follower neurons are inhibited during periods of pacemaker activity, but then rebound and become active (with characteristic delays) as the pacemaker falls silent. This ends up producing a rhythm with three distinct activity bursts within each cycle.
First, the team measured animal-to-animal variability in the output of their network and found that within an animal, the speed of the rhythm was constant, but between animals, rhythm speed varied 2- to 4-fold. They also found that, in every animal, each burst of activity occupied the same percentage of a cycle, regardless of the speed of the rhythm in that particular animal. It was as if, from animal to animal, the same burst pattern was simply stretched or squished horizontally so that its duration varied but occupied the same fraction of the cycle in every individual. This suggests that some features of the network must remain constant even at different cycle speeds.
To examine the underlying causes of population variability, Goaillard and colleagues measured parameters thought to play important roles in determining the phases of network activity. First, they measured the relative strengths of the synapses from the two pacemaker cells onto one follower cell. They found that the conductances (strengths) of both synapses varied widely between crab individuals; however, the conductances of the two synapses were always negatively correlated with each other such that the total synaptic conductance of the two synapses was held approximately constant. Since follower neuron activity can also be modulated by neuropeptides, the team measured a modulator evoked intrinsic current and found that the modulator current was highly variable, but positively correlated with the strength of one of the pacemaker–follower synapses and the firing properties in the follower. In a final tour de force, the team plucked out each pacemaker and follower cell and measured expression levels of a whole battery of ion channels known to affect neuronal firing properties. Again, there were large variations from animal to animal, but they also found strong correlations between expression levels and various circuit parameters.
The work of Goaillard and colleagues is important because it provides experimental evidence for the idea that neural circuits can find more than one way to produce functional output that is ‘good enough’ for the role it has to satisfy. This work also establishes a new standard for detailed circuit analysis. The authors' multi-dimensional analysis techniques reveal a complex network of correlations between intrinsic properties, synaptic properties and the network output. Lists of correlations may seem like dull accounting at first but, taken as a whole, they provide important insights into how multiple parameters within a neural network are co-regulated and which parameters are functionally critical. These kinds of insights will be essential if we want to understand how circuits with variable underlying components organize themselves.
