Yeast de résistance, part deux
In May, I began another Open Science experiment. I shared a link to an open research proposal with the hope of soliciting peer review before ever laying a finger on a pipet. What’s the project? To validate an improved method for rapidly and thoroughly identifying yeast mutants that are resistant to overdose by psychoactive drugs, like the colonies growing on drug-laden plates above. What’s up with ODing yeast cells, you ask? (Skip ahead if you’ve already heard my evolutionary pharmacology spiel). Well, I spent the last five years in academia demonstrating that even a unicellular eukaryote sheds light on the “dark matter” of psychopharmacology. Selecting for overdose resistance is a validated warm-up exercise, part of a summer of experimental prototyping as I seek out partners and patrons for a rare disease Moon Shot.
Scitweeps and regulars at PerlsteinLab offered constructive criticism on my proposal from all corners of the Internet, which makes me think open proposals could be something other scientist do, whether they are independent or not. As I’ll elaborate below, an old friend from my grad school days who has experience with DNA sequencing analysis posted specific experimental suggestions in a DISQUS thread on my lab website. A former colleague from my postdoc days thoughtfully weighed in with a Facebook comment based on her own lab’s experiences. And strangers from unrelated fields kindly retweeted and occasionally chimed in on Twitter:
I’m putting a brief proposal for my first project as independent scientist up for open review. Feedback encouraged! http://t.co/erOXAP6j3K
— Ethan O. Perlstein (@eperlste) May 8, 2013
The consensus is that my proposal would work, but I’d have to be mindful of potential challenges. Last week, I finally secured summer lab space at the Molecular Sciences Institute (MSI), and this week (today, in fact!) my first batch of supplies arrives at the lab. Here I’ll update my original open proposal based on all the terrific feedback I received.
This proposal really goes back to the Fall of 2007. The goal was to examine scads of psych drugs; we started with the SSRI antidepressant Zoloft. Whole-genome sequencing, even for the Lilliputian 12 million basepair yeast genome, wasn’t a cost-effective option back then. Instead we had to use what are now antiquated yeast whole-genome tiling arrays from Affymetrix, which cost $150 a pop. My very first technician, Meredith Rainey, and I started out with a batch of 15 Zoloft-resistant mutants, which became the crux of our 2010 Genetics paper…
Now it’s Summer 2013. The cost of reading DNA has dropped 1-million fold in the last six years, according to sequencing Svengali George Church. The current cost of a single lane of a next-gen sequencer, e.g., Illumina HiSeq 2500, yielding 40Gb of usable data, or 30X genome coverage of 100 yeast strains, is now equal to the cost of microarray-based genotyping the original 15 Zoloft-resistant mutants in 2007. This stunning price drop allows for scaling up genetic discovery. To see what I mean, consider this representative drug-resistance plate:
Before, we would pick individual colonies, do basic genetic tests to convince ourselves that a single nuclear-encoded mutation is responsible for the observed phenotype of drug resistance, and then identify the exact DNA spelling changes, which are often just a single base pair alteration. But now I’m not concerned with isolating and studying individual mutants since I don’t have a proper lab to drill down into resistance mechanisms. Instead, I want to know the entire distribution of resistance-conferring mutations for a given psych drug, a process that involves pooling all colonies from a single plate, like the one above.
As pointed out by several commentators, the challenge of identifying mutations in a pool, where some alleles are present at high frequency (plump colonies) while others are present at low frequencies (diminutive colones), is similar to the challenge of identifying all or as many of the cancer-driving mutations present in a complex tumor biopsy, or all or as many of the driver mutations present at a particular generation/time point during an in vitro evolution experiment. So the very first revision based on open feedback: I’ll aim for 100X, instead of the typical 30X, coverage per pooled sample. That should abet detection of rare mutations present in only a few colonies, especially the waifish ones.
Graham Ruby, my old grad school friend and currently a postdoc in Joe DeRisi’s lab at UCSF, exemplifies the kind of substantive comment I had hoped to receive at the outset. He too was concerned with the sensitivity of pooling to pick up all but the most common mutations on a plate. So he proposed that I sector drug-selection plates into slices, and then pool all the colonies from each slice as one (composite) sample. His concern was the tiniest resistant mutant colonies would not be present in sufficient quantify to be detected if I pooled all the colonies from a plate. However, because I’ll be increasing the average genome coverage to 100X from 30X, I think sectoring may be overkill. My compromise: bisect the above drug-selection plate in two halves using an imaginary divider. Pool all the colonies from the left half of the plate into one tube, and then either randomly or based on criteria like size and shape select several colonies from the right half of the plate as individual samples. Examining mutants as pools vs individuals from the same plate is not only a good control, but also a way to empirically determine the sensitivity of the pooling approach.
Maitreya Dunham, who was a fellow Lewis-Sigler Fellow with me at Princeton and is now an assistant professor at UW, made an astute observation. The geneticists in the audience will appreciate this one. All of the drug-resistance selections we’ve done in the past have been with haploid yeast cells, and that’s the plan for these new experiments, too. Maitreya recommended we try drug-resistance selections with diploid yeast cells, on the logic that dominant resistance-conferring mutations may be more illuminating than recessive-conferring mutations. Naturally, both dominant and recessive mutations arise in haploid populations, but only dominant mutations arise in diploid populations (unless two independent recessive mutations strike the same gene, whose probability is the square of an already low spontaneous mutation rate). So as a control and for kicks, I will also select for drug-resistant diploid mutants.