Behavioural motifs connect related mutant worms

Short behaviours can be automatically identified from videos of crawling worms and used to discover genetic relationships between mutants. Automated methods for understanding changes in complex traits like behaviour are critical if we want to take full advantage of resources like the Million Mutation Project or large-scale gene knockout projects.

HFSP Long-Term Fellow Andre Brown and colleagues
authored on Fri, 15 February 2013

The goal of reverse genetics is to prevent a gene from performing its normal function and to determine what effect this has on an organism. The technology for perturbing genes in this way has advanced rapidly, in principle giving us an entry point to study the function of every gene. The problem is that we often cannot go any further because loss of the gene does not lead to an effect on an organism that is observable by the eye, even to trained experts. This is even true for the nematode worm C. elegans, which is arguably the best-described animal on the planet—we can trace every cell division that leads from a single-celled embryo to an adult and we know the wiring diagram of all its neurons.  The challenge then is to come up with better ways of measuring and understanding animal phenotypes (a phenotype is just the set of traits displayed by an organism).  Here, ‘better’ means more sensitive as well as faster with the ultimate goal of operating at a speed commensurate with the production of new mutants.

Figure: Each worm posture can be represented as a point in a shape space. In this representation, the series of postures a worm adopts while crawling traces out a line (the colour represents the position in the fourth dimension of the shape space). Each 'bird's nest' in the image represents 15 minutes of crawling behavior for a single worm. Some mutants worms cannot crawl smoothly or spend more time paused and their shape trajectories can be more irregular or show spots. By quantitatively comparing worm shapes over time, it is possible to relate mutants to each other and make hypotheses for new genetic relationships.

The phenotype that we are particularly interested in is behaviour, in this case how worms crawl around on an agar plate.  We built relatively inexpensive tracking microscopes that can be operated in parallel to collect a large data set of crawling worms that covers more than 300 different kinds of mutants.  We’ve recorded more than 12 000 individuals for 15 minutes each, more than 3000 hours of video in total.  To deal with this volume of data, it’s very helpful to develop analyses that require minimal human intervention.  In the past, researchers have focused on extracting worm images from video and measuring as many parameters as possible about worm motion and shape like velocity, body curvature, and length as well as the frequency of known behaviours like reversals or sharp turns.  This approach can be very effective, but we don’t know if the chosen parameters are optimal for comparing behaviours, and if we focus on detecting known behaviours we might be missing previously unknown ones.  Instead, we wanted to detect repeated behaviours with minimal bias and to let the data speak for themselves.

Inspired by a method developed in the data mining community for finding repeated patterns in time series, we extracted short segments of crawling that we called behavioural motifs.  Some of the motifs represent subtle behaviours that might be difficult to recognize by eye while others are quite irregular but nonetheless almost perfectly repeated at different times.  This supports the idea that we can detect previously undescribed behaviours automatically, but we also wanted to use the behavioural motifs for quantitatively comparing mutants to each other.  To do that, we collected motifs of different lengths from all the different mutant groups and combined them into a dictionary of behaviours.  Using the dictionary, a fingerprint can be generated for any individual simply by seeing how closely they reproduce each of the motifs in the dictionary.  Differences between fingerprints then serve as a phenotypic similarity measure that can be used to cluster mutants into groups.  We found that the groups derived in this way recapitulated known genetic relationships between mutants and so we reasoned that if two mutants cluster together, there’s a better chance that they are involved in a related pathway.  For mutants that lack genes with an unknown function, we can therefore make new hypotheses for genetic relationships.  In other words, just by looking carefully at how worms crawl we can, in some cases, learn something new about gene function.


A dictionary of behavioral motifs reveals clusters of genes affecting Caenorhabditis elegans locomotion. André E. X. Brown, Eviatar I. Yemini, Laura J. Grundy, Tadas Jucikas, and William R. Schafer, Proc. Nat. Acad. Sci. USA (2013) 110:791-796

Pubmed link

PNAS link

Million Mutation Project link