Despite the fact that many diseases likely result from mutations in multiple genes, there is very little in vivo data on the genetic interactions that underlie these phenotypes in complex organisms. This is largely due to the difficulty in genetically manipulating animals compared to yeast or bacteria. RNAi, however, is making large-scale genetic screens in animals feasible.

Because RNAi can reduce the expression of a specific gene, it can be used to generate pseudo-knockout animals with the same phenotype as a genetic knockout even though they still possess the gene of interest. The trick is to be able to efficiently deliver small interfering RNA to the animal and to quickly screen for phenotypes—two criteria that Caenorhabditis elegans fulfills beautifully.

While in Julie Ahringer's laboratory, Andrew Fraser became an expert at using RNAi in C. elegans screens. In work just published he has used these techniques to systematically test 65,000 pairs of genes for genetic interactions. You might expect such a large scale experiment to require the development of a complex and sophisticated automated approach, but you would be wrong.

In fact, the major methodological change was the use of a 96-well liquid format for feeding small interfering RNA–containing bacteria to C. elegans instead of agar plates. And their readout? Well, they looked at the worms. As Fraser puts it, "I always tend to go low-tech before I go high-tech. I want to make sure that if we do these screens it gives us biologically informative data before I build a very clever way to do it."

In vivo screens are much more difficult to automate than screens in cell lines. "The problem in vivo is that until you've had a look at the data you're not quite sure what you should be looking for. You can miss lots of the weird phenotypes because you didn't tell your machine how to count them—teaching a machine to find the unexpected is hard. Those are things that you rarely miss by eye," says Fraser.

Not only can manual examination spot phenotypes that might be missed by automated phenotyping, bizarrely it can also be faster because the human eye spots wild-type phenotypes with barely a glance while a computer would have to acquire an image and analyze it. Fraser remarks, "The manual method is really amazingly quick for that, and there is nothing we have automated that even gets close."

This low-tech method allowed Fraser and colleagues to identify 349 interactions out of 65,000 tested, which means that pair-wise combinations of mutations are many times more likely to result in nonviable phenotypes than single mutations. And this is only looking at the highest-confidence hits.

To try to identify weaker interactions, Fraser is now building an automated image analysis system. He explains, "Now at least we have test sets so we know how well the screen works, what it gives us, and that it is worth doing." Fraser is quick to add, however, that even if they get the automated method working well they will likely do a very rapid manual screen as a first pass and then do all the repeats in a quantitative way using the automated screening.

So before you embark on your next high-throughput screening experiment, it might be worth considering whether it might be good to do an old-fashioned manual screen before investing time and effort in something that does not work out quite as planned.

1. Lehner, B. et al. Systematic mapping of genetic interactions in Caenorhabditis elegans identifies common modifiers of diverse signaling pathways. Nat. Genetics 38, 896–903 (2006). | ChemPort |