Many times in experiments, there will be a few pieces of data that don't make sense. Confronted with an odd data point or two, most researchers will think, `Huh. That's weird,' and move on. David McCrea from the University of Manitoba, on the other hand, has found a way to make his weird data points work for him.
He works on the neural circuitry for walking in cats, called the central pattern generator (CPG), which is contained in the spinal cord. Using the right sort of stimulation, the cat spinal cord (or in fact almost any vertebrate spinal cord) will `walk' even with the muscles paralyzed and with no connection to the brain. Activity in nerves running to the limb muscles looks more or less like activity during walking: flexor muscle nerves generally alternate with extensor muscle nerves, and within that alternation,muscle activity is ordered more or less appropriately. But there is often a strange effect: sometimes certain nerves `forget' to turn on or off. The rest of the pattern may stop as one nerve fires anomalously or the rhythm may keep on going, roughly as usual. And sometimes the CPG keeps time throughout the`deletion', but sometimes it doesn't.
Most other researchers had shrugged off deletions as just another weird effect to be avoided, but McCrea realized that they might tell him something about the underlying structure of the CPG. So he teamed up with Ilya Rybak of Drexel University College of Medicine to produce a mathematical model of the CPG that could occasionally `forget' to turn on one muscle, but not lose its rhythm.
They decided that they'd have to make a CPG model with two levels: a higher-level `rhythm generator' to produce the basic timing and a lower-level`pattern generator' to give the characteristic ordering of muscle activity. Using a standard Hodgkin-Huxley neural model, which models the characteristics of electrically excitable cells like neurons, they made a simulation of a pair of generic flexors and extensors, with one excitatory neuron at each level for the flexor and one per level for the extensor, a total of four main neurons. They then added in four inhibitory neurons, so that the flexors would inhibit the extensors and vice versa. This crossed inhibition, combined with a slow self-inactivation built into each neuron, results in rhythmic alternation; first the flexor neurons start firing, inhibiting the extensors,but they slowly deactivate themselves until the extensor neurons can come on,inhibiting the flexor, and so forth.
Deletions, they thought, might come from temporary fluctuations in the excitability of the CPG neurons. Increasing the excitability of an extensor neuron in either level, for example, keeps the extensors on, swamping the usual alternation with the flexors. But exciting the extensor neuron in the rhythm generator level resets the rhythm, while exciting the same neuron in the pattern generating level doesn't: the CPG keeps time, even though the flexors don't fire.
McCrea and Rybak's model can replicate some of the subtleties of deletions. For example, in the cat data, they would sometimes see the flexors turn on,firing continuously for a while, with little or no activity in the extensors. Other times, the flexors might go silent while the extensors kept up their rhythm as usual. In the model, exciting the pattern generating flexor neuron creates the first effect, while inhibiting it results in the second.
The two-level model is still a hypothesis – no one knows whether there really are `rhythm generator' and `pattern generator' neurons in the spinal cord – but it predicts differences in the behaviour of the two classes of neurons, particularly during deletions. Now the goal is too see whether they can find spinal neurons that match their predictions.