Neurons typically form elaborate tree-like branches with two functional sites (Figure a). One site, called a dendrite, receives inputs from other neurons. After integration and processing of the received information, the message is passed on to the other end of the neuron, called an axon. The axon transmits signals to the dendrites of downstream neurons, and this is how information propagates throughout neural networks.

However, neural cells may form much simpler structures. The simplest structure is presumably a single neurite extending from the cell body (Figure b). Such nerve cells characterize simple creatures with a small and compact neural network, such as that of the worm C. elegans, which is comprised of exactly 302 neurons. Most of these 302 neurons are simply structured with a unipolar neurite extension, a feature that greatly limits the computational capacity of the worm's neural network. So how exactly does the worm manage to perform complex tasks, such as intelligently search for food or evading predators? 

Research student Rotem Roach (now Dr. Ruach) in Prof. Zaslaver’s group discovered that simple nerve cells of type (b) can perform a parallel calculation. How? The researchers showed that on the surface of the neurite branch, synaptic inputs and outputs are organized in discrete clusters (marked in figure b with dashed circles), so that each such cluster constitutes a separate computational unit. The neural calculation (receiving the information, processing it and sending it on to nearby nerve cells) is therefore carried out locally, thus bypassing the need for the information to pass through the entire cell. The fact that each neuron can perform several such operations independent of one another greatly enhances the computational capacity of small neural networks made of simply-structured neurons.

The research was published in the prestigious journal PNAS, and was performed in collaboration with Prof. Scott Emmons from Albert Einstein University in New York. Read the paper