Selection beats design, again
This week's paper is "HIV-1 proviral DNA excision using an evolved recombinase" by Indrani Sarkar and others, published in Science (vol.316, p.1912). This paper is yet another example showing that selection (natural or artificial) can outperform design.
To illustrate the point, let me start with a well-known example from plant breeding. Suppose you wanted to make broccoli, starting with its ancestor, wild kale? You could cross them, identify which genetic differences are most responsible for the large edible inflorescence, and transfer those genes to the wild kale. But what if broccoli didn't exist?
We know something about which genes are involved in flowering, branching, etc., so we could mess with those genes and see what happened. But I don't think all of the genetic engineers in the world, working together, could design a genome that would result in broccoli. Not on their first try, anyway.
And yet, we know that broccoli was developed by humans, without using any molecular methods at all. They did this growing a bunch of kale plants and planting seed only from those that were slightly more broccoli-like than the rest. They presumably selected for bigger inflorescences, without necessarily expecting that the final product would have one more than 6" across. In an old gardening book, I once saw a picture of an intermediate, called "brokale" or something like that, with several 1" inflorescences.
The authors of this week's paper faced a similar problem. HIV, the virus that causes AIDS, is getting easier to control, but it's almost impossible to get rid of. Like other retroviruses, it copies itself into the DNA of host cells. They wanted to make an enzyme that would cut the HIV provirus out of human DNA, starting with an existing enzyme that does the same thing with certain other viruses. The problem was, they didn't know which amino acids in the enzyme to change, to make it cut out HIV rather than its current target virus.
So, they did what people are increasingly doing when a problem is too tough for anyone to design a solution: they evolved a solution. Their approach to directed evolution was ingenious and fairly complex. They identified a sequence in HIV that had some similarity to that recognized by their starting enzyme. They modified this target sequence until the "ancestral" enzyme occasionally recognized it, then tested various mutant versions of the enzyme to find those that worked better. They recombined those that worked best, mutated again, and so on. All in all, it took 126 cycles of selection to get an enzyme that worked well. A less-sophisticated approach, analogous to breeding broccoli from wild kale without knowing the DNA sequence of either, might also have worked, but it would have taken much longer.
Some version of directed evolution is often the best approach, whenever you can't design an answer to a problem, but 1) you can compare several possible answers and say which is better, and 2) you can generate a bunch of new answers that are somehow similar to your best current answers. My personal favorites are evolving an enzyme made out of DNA, rather than protein or RNA (Science 286:2441) and this "story problem" I used in my Crop Ecology class:
A certain weed produces 100,000 seeds. The number of weeds in the field is the same from year to year. What is the chance that a given seed will grow into an adult plant and produce seeds of its own?
If you don't see the answer immediately, guess a number at random, then see what effect it would have on changes in the weed's population over years. If weed population would decrease, your number is too big, so decrease it by a random amount, and so on. Of course, more sophisticated versions of this algorithm will find the answer more quickly.