A fundamental question in neuroscience is how neural networks generate behavior. The first challenge in answering this question is to identify neurons and circuits controlling the behavior of interest. To identify such key neuronal subtypes, current approaches perturb one neuron subtype at a time and measure the consequences. Considering the large number of neuronal subtypes in most model organisms, implementing unbiased search becomes difficult. Further, unique genetic tools to perturb to only one neural subtype at a time do not always exist, making this search for essential neuron subtypes even more challenging.

In a recent study (PDF), we developed an experimental framework to identify essential neurons for a given behavior with far fewer perturbation experiments than traditional methods by exploiting non-specific genetic markers and promoters. In order to make this possible, we took advantage of a method called compressed sensing from statistics. This method provided an alternative strategy for rapidly inferring the contributions of different neuron subtypes to a behavior.

For a qualitative understanding of compressed sensing, consider a high school puzzle. You have a pile of 64 coins. 63 coins each have an identical weight, and the remaining one is heavier. How do you find the heaviest coin with the minimal number of measurements? The most tedious way would be to measure the coins one at a time (quite like finding they key subtypes of neurons, looking for them one at a time). The way to find the heavy coin quickly is to split the pile of coins into two piles of 32…weight them against each other to find the heavier pile. Take that pile, split it into two equal piles again and repeat the process. This way, you can find the heaviest coin in 5 (or log (64) measurements). Now consider a much harder problem that was thought to very difficult before the advent of compressed sensing. You have 64 coins again. Most of them (you don’t know how many) have the same weight. The remaining (could be one, two, three….the only condition is that these remaining coins are much less than 64) each have a weight different from the rest (some of the remaining coins could be heavier, others lighter etc).We ask you to find the coins with different weights from the majority using the smallest number of possible measurements. Turns out, thanks to compressed sensing, that you can find these coins again with a number of proportional to log(64), and you do not have to be anywhere as smart as the smart high school student when determining how to break the coins up into piles.

Using this framework of compressive sensing, we identified the interneuron subtypes regulating the speed of locomotion of the nematode *Caenorhabditis elgans*. We first designed an optogenetic screen. We expressed arhaerhodopsin protein, which is responsive to green light and inhibits the activity of neurons under green light, into multiple predetermined groups of neurons (using different known non-specific promoters in the literature). While inhibiting these neurons, we monitored the behavior of the animal. From a small number of such measurements, we identified three key interneuron subtypes controlling the speed of locomotion.

In order to validate the results, we developed a tracking microscope for imaging neural activity and optogenetic manipulation of targeted neurons on freely moving animals. By tracking a neuron with one-micron precision in x, y and z, we stabilized the image of the worm under the field of view for accurate imaging and targeted illumination of specific neurons. With this level of stabilization, we could use an order of magnitude lower light level than previously possible. We could thus image the activity of neurons in freely moving animals for more than an hour without any effects phototoxicity. By imaging and manipulating the activity of neurons we discovered that the three interneuron subtypes identified by compressive sensing control different aspects of the speed regulation in worms. One interneuron subtype acts as a switch, determining whether the animal moves or not. Another acts as a rectifier to determine whether the animal moves forward or not, and the third modulates the speed of locomotion continuously over slower time scales.

In summary, we showed that a compressed sensing-based framework that exploits non-specific genetic tools, in conjunction with the microscope enables rapid and comprehensive understanding of the neural circuits that drive the behavior of *C. elegans*. Similar experimental methods based on compressed sensing have the potential to discover the key nodes that control a phenotype in complex biological networks, including the nervous systems of higher organisms as well as gene regulatory networks.