Skip to Content

Evolution in Action – Literally

David Liu’s group at Harvard has been working on a technique for a few years now called PACE, and I keep meaning to write about it. A new paper in Nature gives me the opportunity. The acronym stands for “Phage-assisted continuous evolution”, and it’s as neat an example of directed evolution as I’ve ever seen. The idea is that you use bacteriophage as your vehicle for evolving proteins, because of their extremely short generation time. These infect E. coli bacteria, and the two of them are modified in this system so that you can use selection pressure to get to a protein with specific binding properties. It’s been used to look at protease inhibitor resistance and DNA-binding specificity, and now it’s put to work for a protein-protein interaction.

Here’s how that works: the M13 bacteriophage have a specific surface protein, pIII, that they need for their progeny to be infectious. In this case, they’ve had that gene stripped out of their sequence, but it’s been added instead to the E. coli cells that they’re going to be infecting. That gene is set up, on an accessory plasmid (AP) downstream of a promoter that’s tied to the protein-protein interaction that you’re trying to evolve – so only when some protein can bind to its target up there will pIII get produced. That protein is what the phage are going to have to evolve to keep themselves reproducing – they infect the bacteria with a selection plasmid (SP), and if that plasmid DNA can generate a protein that starts the pIII production machinery, then they can get enough pIII for their progeny (being produced inside the infected bacterium) to be infectious in turn. The bacteria also have a mutagenesis plasmid (MP), arabinose-driven, that bumps up the rate of error-prone DNA copying, to keep a mutagenic foot down on the evolutionary gas pedal.

And here’s the final trick: the bacteria are in a “lagoon” vessel, and are being continuously dripped in (and the contents of the lagoon are being similarly slowly drained out. You tweak these flow rates to fit the rate of phage mutations, so that they’re in a constant fitness race. If they can evolve proteins to get their pIII protein and go on for another generation faster than they’re being diluted out of the lagoon, they’ll keep going. Otherwise, out they go. You have to play around with these, of course, so that you’re giving them a fair shot in their c. 20-minute lifespan: run things through too quickly, and the phage won’t have a chance, but if you do it too slowly they have no incentive to better themselves.

As you might well expect, you have to tweak some other things besides a stopcock opening to really get this to work well:

To maximize the sensitivity of the bacterial two-hybrid, we extensively optimized parameters including (1) transcriptional activation and DNA-binding domains, (2) protein expression level, (3) interaction binding affinity, (4) DNA-binding domain multivalency state, (5) reporter gene ribosome-binding site, (6) operator–promoter distance, (7) RNA polymerase–promoter affinity, and (8) DNA-binding domain–bait linker length.

Another thing that they’ve found is that you want to let the phage/bacteria system just roll along and drift for a while before you start applying pressure, which I would guess is a way of getting a lot of potential starting points built in. What happens once you do bring the hammer down is that the phage population drops for several hours and then recovers, and that presumably reflects the majority of the phage present at the beginning finding themselves with no path forward. As with any of these hook-the-microorganisms-to-the-wheel schemes, one of the key things you have to look out for is that they don’t have an easy way to just completely jump out of the stall and go back to a wild-type existence. Bacteriophage live a reasonably precarious lifestyle as it is, so making them jump through hoops for you at the same time takes some doing. But when you get this sort of thing right, you’ve harnessed an extraordinary optimization engine – in theory, you just walk away for a few days and come back to find an optimized protein waiting for you.

What the team is trying to evolve here are mutant forms of Bt proteins, the bacterial pesticides that are used extensively in agriculture. This family of proteins binds to specific receptors in insect midgut cells, but as you’d expect, the insects themselves are evolving under the pressure of Bt use and are becoming resistant. So Liu and co-workers set up a system to evolve a Bt protein (Cry1Ac) that can target a receptor from the cabbage looper caterpillar, which is already becoming resistant in the wild. 528 hours of the lagoon system, under varying mutagenesis conditions, flow rates, etc. did the trick. At least 25 mutations were seen in the protein at the end of the process, with strong biases towards several key changes. Interestingly, it appears that there were some recombination events along the way, which must have happened when more than one phage infected the same luckless bacterium, and these seemed to be especially useful in the evolutionary trajectory.

Synthesizing consensus mutation Cry1Ac proteins using the data from the evolutionary run gave variants that bind to the insect receptor down in the tens of nanomolar range, which is pretty impressive considering that it started out hardly binding at all. When they tried them out on real cabbage loopers, they found that indeed, the new proteins were hundreds of times more potent on the resistant organisms, and were several times more potent even on the susceptible strains. They went on to try these proteins against eleven other insect pests, and found that they were as potent or more so than wild-type protein across the board. So you can indeed evolve insecticidal proteins that will bypass the current resistance mechanisms, and produce what appear to be very useful new variants. And when the insects find a way around those, well, you can presumably turn around and use the fast-forward button to evolve around those as well.

The approach established here enables targeting of a Bt-resistant pest through the evolution of high-affinity Bt toxin variants that bind a specific target insect protein. In principle, this strategy should be applicable to target a variety of insect pests. While the evolution of insect resistance to an evolved Bt toxin is a likely possibility, this work has the potential to provide access to many new Bt toxins that, individually or in combination, may manage resistance and extend the effectiveness of this important approach to pest control. We also envision that this system may be used to explore potential resistance mechanisms by evolving the receptor in the presence of a Bt toxin, analogous to the recent use of PACE to identify protease inhibitor drug resistance mechanisms. Finally, we note that the ability of protein-binding PACE to rapidly evolve novel protein–protein interactions may prove useful in the discovery or improvement of protein therapeutics.

Excellent stuff. The machinery of protein synthesis and heredity is so powerful; it’s no wonder that people have been working so long to harness it.

16 comments on “Evolution in Action – Literally”

  1. exGlaxoid says:

    Can they use the system to select new antibiotics without resistance to current drugs? (see May 13th story) Seems like it could be ultimately used for that as well. Maybe it will take a little more work to start out. But well worth the trouble if it works.

    1. A Cheme says:

      It seems like to do that you’d have to integrate a second selection pressure; not only would you want to kill the target organism, but you’d have to integrate a human cell (or maybe just mammalian, or even just eukaryotic) feedback mechanism into the process as well. After all, it’s not killing bacteria alone that’s hard, it’s not killing the person as well.

      I would suspect that adding this second selection point would do a lot more than double the complexity of this sort of thing.

  2. Argon says:

    ‘lagoon’ vessel = chemostat?

    Oh, I see. The set-up is a two-stage chemostat: One to continuously produce E. coli hosts and another to feed the E. coli containing medium into the phage mix. Good thing they didn’t run the experiment too long or they might have had genetic drift issues with their E. coli host culture.

    Chemostats make great Evolution-o-trons.

    1. SP says:

      No, they feed in new E. coli and flush out old ones before they have a chance to replicate. The residence time of any bacterial host has to be less than bacterial replication time or you do start to get non-desired survival mechanism (e.g. mutations in promoter for pIII). You get what you select for.

      1. Mark Thorson says:

        These people must have some serious gadget-builders on staff.

        1. David Liu says:

          SP is exactly correct. And we don’t have any formal staff (gadget builders or otherwise); just some really talented and hard-working graduate students and postdocs!

          Thanks for your interest in our work!

          Best,
          David

  3. gippgig says:

    It should be noted that M13 is an unusual virus; it doesn’t kill its host cell. Instead, mature viruses are secreted thru the cell membrane as the host continues to grow (more slowly).

  4. Jay Bradner says:

    A stunner from the Liu lab. I had the pleasure of hearing David present this story earlier today…minutes after the post above. Genome editor-in-chief, crop savior, genius multi-tasker. In all seriousness the Lagoon is proving to be a serious proving ground for pre-clinical drug assessment, enzyme evolution and next-generation biotechnology NOS (all the while invoking gene control at its roots). It is a sparing few who invent new technology and stay so decisively in a space after published proof-of-concept to innovate at depth and breadth. Bravo, David.

  5. Anonymous Researcher snaw says:

    Nifty research! There may also be applications in the chemical industry of this approach: chemical companies are increasingly interested these days in doing on the metric ton scale what we pharma/biotech people do on the milligram scale because they are trying to move away from current petroleum-heat-and-catalyst approaches. Oil may be cheap now, but they don’t expect it to stay cheap forever. Also, they expect eventually they will have to pay a carbon tax or something called a different name for political reasons that amounts to a carbon tax. So I know of more than one chemical company investing in microbial strain engineering these days.

    1. Paul D. says:

      As I mentioned in a comment in the previous incarnation of this blog, the greentech startup Siluria has also been using M13 phage display, in their case to make industrial catalysts. The phages are used as templates for biomineralization and are destroyed by heating, so it’s not the same sort of lab scale evolution. Doing tens of thousands of trials they’ve come up with what look like practical catalysts for oxidative coupling of methane + oxygen to ethane + ethylene + water, which has long been a Holy Grail of the petrochemical industry.

      http://siluria.com/Technology/Oxidative_Coupling_of_Methane

      (I have no stake in this company or technology.)

  6. anon2231 says:

    love it. Now I need to read the actual paper.

  7. Li Zhi says:

    Can weaponization be far behind?

    1. matt says:

      If you are an insect targeted by Bt, the weapon is already here. But the little buggers will fight back.

  8. Poul-Henning Kamp says:

    I’m vaguely reminded about Douglas Adams joke about evolution (“… but quite incapable of drinking coffee”) here.

    How do they find out if undesirable changes also happened to that protein ?

    Do they have to start all over again and treat it as a new protein or is it safe to assume some properties are invariant under this kind of “random” changes ?

    1. Slurpy says:

      I was thinking the same thing – won’t someone have to vet each new Bt variant as safe before incorporating it into the next generation of GM crops?

  9. Morten G says:

    Urgh, couldn’t they at least split the Bt into multiple subunits so it takes a little longer for the gene to migrate into weeds and disrupt the ecosystem?

Comments are closed.