Ep 97: Mutation bias in evolution: climbing Mount Probable (with Arlin Stoltzfus)
What is mutation bias? How does it affect evolution?
In this episode, we talk with Arlin Stoltzfus, a research fellow at the University of Maryland’s Institute for Bioscience and Biotechnology Research. Arlin studies mutation bias – the idea that some types of mutations occur more often than others – and how these patterns can influence the evolutionary trajectories of populations. In the chat, we contrast this mutation-centric approach to evolution with more standard views in which selection does most of the creative heavy lifting. We center the talk around Arlin’s 2021 book - Mutation, Randomness, and Evolution, which offer a new conception of variation as a difference-maker in evolution. Looking forward, Arlin argues that a better understanding of mutation will make it easier to predict the origins and outcomes of different cancers and the evolution of infectious diseases and crop pests.
📚🐛 Big Biology Bookshelf
Cover photo: Keating Shahmehri
-
Art Woods 0:00
Hey, Big Biology listeners, that gentle reminder that we're a nonprofit and that we rely on donations from listeners to keep the show going.
Cameron Ghalambor 0:07
If you'd like to support the show, please consider making a one time donation at Big biology.org or a recurring monthly donation@patreon.com slash big bio.
Art Woods 0:17
Even small donations help us out a lot.
Cameron Ghalambor 0:27
Hey, Art, you're driving the bus on this introduction, right?
Art Woods 0:30
Uhh Cam. I'm sitting on my couch right now. No buses within my field of view. I'm not riding in one and I'm certainly not driving one.
Cameron Ghalambor 0:39
Okay, sorry. I didn't mean to confuse you, Art, I was using a metaphor.
Art Woods 0:42
Ah, of course. And folks, this little staged interaction gets at the nub of the beauty and pain of metaphors.
Cameron Ghalambor 0:50
Okay, let's talk for a moment about positive uses of metaphors. They provide shorthand ways of encapsulating complicated ideas and they build bridges between useful ideas in one area and another. Plus many thinkers over the past century have argued that metaphors are indispensable to science.
Art Woods 1:07
For example, take Theodore Brown's 2003 book called Making Truth: Metaphor in Science. In it, he writes that, "Metaphor plays a central role in the development of a scientific subject from its very beginnings through to its full development as a mature body of knowledge and understanding. It figures in the scientists' initial creative impulses in interpretation of experimental data, in formulations of scientific explanations, and in communication between scientists and between scientists and the rest of the world."
Cameron Ghalambor 1:37
Examples of influential scientific metaphors include Maxwell's demon, Schrodinger's cat, the Greenhouse Effect and phylogenetic trees. These last two are so familiar that most biologists don't even think of them as metaphors anymore.
Art Woods 1:52
But metaphors can also mislead in small and large ways. Indeed, Thomas Kuhn argued that pervasive metaphors can become so foundational to theories that they can become hard to revise or even to overturn.
Cameron Ghalambor 2:05
And going back to Brown's book, he relates an interesting anecdote about the creativity that can be unleashed when we’re released from binding metaphors. In the late 1980s, Brown headed up an effort to establish a multidisciplinary research institute at the University of Illinois, Urbana-Champaign.
Art Woods 2:21
Brown and some colleagues brought together scientists from a broad range of backgrounds including physics, chemistry, physiology, cell biology, engineering, philosophy and some other areas to work on cross-disciplinary problems. And it worked, beyond their wildest dreams.
Cameron Ghalambor 2:37
Brown writes that, "We learned that putting people together in well-designed surroundings with institutional encouragement, made for a bit of magic."
Art Woods 2:45
But what is that magic? Brown writes that sharing different metaphorical representations of a problem can open up new kinds of creative thinking.
Cameron Ghalambor 2:53
Okay, so what that's basically saying is that metaphors within each field become creative straitjackets, and that progress can emerge from collisions among metaphors from different fields.
Art Woods 3:05
Our non-metaphorical guest today is Arlin Stoltzfus. He's a fellow at the Institute for Bioscience and Biotechnology in Rockville, Maryland. Arlin has written extensively about the processes of mutation and how those processes can influence the trajectory of molecular evolution.
Cameron Ghalambor 3:21
We chat with Arlin about the concept of mutation bias, which interestingly raised his hackles, because it doesn't seem to fit well into several of the major metaphors that biologists use to describe how mutations generate genetic variation and contribute to evolutionary change.
Art Woods 3:36
One of these dominant metaphors invokes pottery. It describes selection as the potter and variation as the clay which the potter shapes into different forms. A beautiful metaphor in some sense, but notice how it assigns agency and power to selection, while relegating variation to a passive role.
Cameron Ghalambor 3:54
Arlin argues that variation as passive clay is a misleading metaphor. In several recent papers and in a 2021 book called Mutation, Randomness and Evolution, he develops the idea and reviews the evidence that patterns of variation emerge from multiple kinds of bias in how mutations arise. In some lineages, particular kinds of mutations are more likely than others to occur. And these biases influence the likelihood of getting particular evolutionary outcomes.
Art Woods 4:24
In other words, that there is structure in the potter's clay and that structure can have surprisingly strong effects on evolutionary trajectories. And to be clear right up front here, Arlin isn't claiming that directed mutation is evolutionarily important. Directed mutation is a controversial idea that organisms can generate beneficial mutations when needed.
Cameron Ghalambor 4:44
As an example, consider transition transversion biases. To understand this recall that DNA nucleotides come in two chemical classes A and G are purines whereas T and C are pyrimidines means a transition occurs when a DNA letter mutates to another within a chemical class like from A to G.
Art Woods 5:04
By contrast, transversion occurs when a DNA letter mutates into another class like the purine A mutating into the pyrimidine C. Interestingly, the molecular machinery that synthesizes DNA and repairs mistakes biases this process such that transitions typically are much more likely than transversions.
Cameron Ghalambor 5:23
Such a bias can affect which mutations are most likely to go to fixation in populations.
Art Woods 5:28
But not so much because they're selected more because they're so common in a population because of the inherent chemistry that makes them, which gives some agency back to the clay.
Cameron Ghalambor 5:39
To describe the effects on evolutionary trajectories, Arlin modifies another well known metaphor in evolutionary biology, the idea that populations gradually ascend adaptive peaks, in the same way that climbers ascend mountain ranges. We won't give it all away here except to say that Arlin's modified metaphor involves climbers that make systematically biased choices about whether to go left or right.
Art Woods 6:04
Other topics include how much mutation bias affects organismal phenotypes, the meaning of origin fixation models, and whether we can leverage our knowledge about mutation bias to predict how diseases and pests will evolve in response to our countermeasures.
Cameron Ghalambor 6:19
I'm Cameron Ghalambor
Art Woods 6:20
And I'm Art Woods
Cameron Ghalambor 6:21
And you're listening to Big Biology.
Art Woods 6:35
Arlin Stoltzfus thanks for joining us on Big Biology.
Arlin Stoltzfus 6:37
Thank you. It's great to be here.
Art Woods 6:39
We're looking forward to talking about a bunch of aspects of your work on evolution. We're going to focus primarily on ideas related to mutation and mutation bias and the roles that that mutation may play in shaping the evolutionary process. And we intend to do that in the context of your really great 2021 book called "Mutation, Randomness and Evolution," which is published by Oxford, and a recent 2022 preprint and EcoEvoRxiv that was co-led by Alejandro Cano and Brian Gitschlag. So yeah, Cam, do you want to start us off with the first question?
Cameron Ghalambor 7:14
Yeah, sure. So let's just dive in and start talking about the place of mutation in evolutionary biology. I'll start off by maybe trying to articulate maybe an older or traditional, but maybe common view among evolutionary biologists, which is that the process of mutation is recognized as sort of the ultimate source of variation, but the specifics of the mutation process may not be particularly relevant for understanding evolution. And maybe that sort of irrelevance arises from the way that we think about or conceive how mutations and variation are generated in populations. And then how selection acts on that variation. So like a population genetic perspective would be like this mutation-selection balance.
Arlin Stoltzfus 8:05
Exactly. I mean, that is a very traditional view that the details of mutation are irrelevant to how evolution turns out. And, and it goes back to Darwin, he had this analogy with a builder who builds something from stones that fall from a cliff, and they fall from a cliff according to natural laws. But the built structure doesn't have anything to do with that Darwin says. I like to explain it with the with a sandcastle, you know, so, imagine a sandcastle that's made by a person or is made out of a mold, it has a particular shape. And if we want to explain the shape, it's based on the mold and the sand grains just provide substance. If you focus in with a microscope, and you look at an individual sand grain, each one is different. Each one has a history that reflects laws that govern, you know, the emergence of sand grains, but none of that matters if you're going to explain the shape of the sandcastle, because the sand is replaceable. And it's just like Darwin's metaphor, except I think it's a little bit better one. But Darwin said he believed in a law-governed universe, he believed that there were laws of variation, but he said that those laws bear no relation to the structures built by natural selection, just because it was kind of happening at a different level. So yeah, that's absolutely the traditional view. And I would call that the raw materials view: variation just supplies the raw materials that selection shapes into adaptations.
Cameron Ghalambor 9:22
So is there something wrong with that view?
Arlin Stoltzfus 9:23
Yeah. So I would say that, first of all, it's just kind of a folk theory. Right? I would say that, if you want to get into formal theory, then the replacement for that, you know, you sort of graduate from that to evolutionary quantitative genetics, you know, Lande Arnold 1983, where you have a theory that tells you what the relationship is between standing variation and the course that's taken by evolution. And it's very much like raw materials. The intuitions are very similar, but they're not exactly the same, because once you get into multivariate evolutionary quantitative genetics, then you the direction and rate of evolution depends jointly on selection, and the G matrix of variances and covariances. So it's not just passive raw material anymore.
Art Woods 10:10
So like what you're saying is that that raw material has a structure to it that's really important, that can shape the response to selection. Right?
Arlin Stoltzfus 10:17
Right. So that's the world of sort of quantitative traits. But I think that you actually need a different kind of theory, separate from the raw material. So set the raw materials theory aside, there's a different theory, which, depending on how you think about, it sort of goes back a long way, and that's something like constraints that says that variation is acting in a manner that's prior to selection, setting out some possibilities. So this was part of early Evo Devo, thinking, for instance. So I think there's Per Ahlburg that said in his 1980 article, that selection may determine the winner of the game, but development non-randomly defines the players. And there have been, you know, multiple attempts to establish that kind of idea. And I would say that, just like evolutionary quantitative genetics is the sort of the formalized theory that you graduate to from raw materials, I think that the formalized version of that theory is biases in the introduction process. That tells you how to think about it in a more rigorous way and provides a quantitative predictive framework.
Art Woods 11:18
I was going to try to link that idea that you just stated back to this idea of sand grains and sandcastles. And it sounds like what you just said there is that there is something about the origin of the sand grains and the sort of flavor that they have that does affect the outcomes of the sandcastle. Is that a fair way of saying what you just said?
Arlin Stoltzfus 11:36
Um, yes, and no, because if you're still in the realm of sand grains, if you're still in the realm of quantitative characters, where you're literally assuming infinitesimal effects, there's only so much influence that biases and variation can have, and in the classical evolutionary quantitative genetics framework mutation is actually in the background a little bit and you're talking about standing variation, that's the main focus. But if the sand grains get too big, and they're big chunks, right, now you need to have a theory for what's the character and timing of appearance of those chunks, alright? And that moves you into this other realm.
Arlin Stoltzfus 12:10
But I want to mention another theory that I think Cam just briefly touched on when he talked about mutation selection balance. So early on in classical population genetics, Haldane and Fisher thought about this idea of an internalist theory, where mutation would be important or variation would be important. And they interpreted it as a theory of evolution by mutation pressure, which means the transformation of the population by many events of mutation, right, sort of in parallel. And what they worked out was that it's not a very good theory, because you need to have high mutation rates unopposed by selection. No, it does. So the question that they were trying to answer originally is, why do we have diseases right? If selection is acting against these diseases, like Hemophilia is one that Haldane wrote about repeatedly, why don't they just disappear? And the answer is, the pressure of recurrent mutation pushes up the frequency and selection pushes down the frequency. And in a sufficiently large population, it reaches a balance point. And the balance point is closer to zero than to one because mutation is a much weaker force.
Art Woods 13:10
These are negative alleles. Yeah,
Arlin Stoltzfus 13:12
Yeah these are deleterious alleles. Yeah. So Haldane and Fisher generalized from that to say, well, we don't think mutation will be an important source of direction in evolution, because it would need to have high mutation rates. But the problem was, they weren't thinking about the introduction process. So that theory of evolution by mutation pressure is not really very important for understanding ongoing evolution, it helps to understand things like genetic diseases.
Art Woods 13:38
Let me state another distinction that might be relevant here, and that is, you mentioned this phrase standing genetic variation, so this is this idea that in populations, there's just some pre existing background amount of variation that is there, and that can be selected on under the right circumstances. And that's as opposed to sort of thinking about the ongoing introduction of new variation and the effects of that on ongoing evolution. So is that a distinction you're trying to make? Or is it more about the kinds of processes that lead to the flavor of the standing genetic variation?
Arlin Stoltzfus 14:15
It depends on how you think about the standing variation. But yeah, that's definitely a distinction that I'm getting at. And if you remember, in the book, there was this distinction between the buffet and the sushi conveyor. So the original modern synthesis view was that there's this buffet it's all full of all sorts of different variations that sort of cover all the possibilities selection can do anything at once out of this abundance of the gene pool, and it's actually maintained. Dobzhansky, his birthday was yesterday, you know, this was his big conceptual contribution to the modern synthesis . Natural populations maintain abundant variation because of recessive genes, and balancing selection and heterosis. And then they mix it all together by sexual mixes and recombination and crossing over. And this is where all the diversity comes from to make evolution. And you can just based on the standing variation, I mean, there are plenty of experiments that were done in the early half 20th century that showed that you could just take a population, and you could select, and you could move the population very quickly, in such a short period of time that it didn't really need new mutations. Okay, so the other view is like the sushi conveyor view, where there's opportunities coming along at different times, you don't choose what it is that's coming up, you don't choose when it's coming up, you just choose yes or no, am I going to take this, right?
Art Woods 15:38
Yeah you grab the smoked salmon as it goes by.
Arlin Stoltzfus 15:40
That's right. And the difference is that then the dynamics with which things appear on the sushi conveyor matter a lot. Whereas in the buffet view, there's just all this abundance, selection just chooses whatever it wants. And formally, if all of the alleles relevant to the outcome of evolution are present in an initial population, at some frequencies, where they're not going to get lost, then basically the most fit one wins, you know, it has a probability of fixation of one, and all the others have a probability of fixation at zero. But if you're at the other extreme, where they're coming along rarely, then the probability fixation is a graduated function of their fitness benefit.
Cameron Ghalambor 16:14
I find that perspective really interesting because I hadn't heard the sushi conveyor belt analogy before, but I really liked that. My kind of impression has been that, you know, the importance of mutation and standing genetic variation has really been this like kind of back and forth in terms of kind of the importance and, you know, I think about the molecular revolution, and Hubby and Lewinton finding all this like protein variation, and then Kimura coming along with the neutral theory of molecular evolution, and that really dominating the way we thought about standing genetic variation for a very long time. But then also this view, that adaptation was really driven by a mutation process of a beneficial mutation shows up in the population and then gets swept to fixation. And then there were these studies like I don't know if you're familiar with this study by Felicity Jones on stickleback that showed repeated parallel adaptation to freshwater from standing variation, but the allele responsible for that occurred in the population at very, very low frequency, and it was probably maladaptive or deleterious. And it seemed like then the pendulum swung again towards thinking more about standing variation again. So where do things stand, sort of, from your perspective now, given all the molecular tools that are available, and the ability to sequence genomes and look at the timing and origin of where mutations arise, and how selection acts on those?
Arlin Stoltzfus 17:48
Yeah, so the molecular revolution was really weird going back and looking at it, neutrality kind of sucked all the oxygen out of the room, and people were not thinking about other things. Because I would say, looking back at that era, that the biochemists who started comparing protein sequences in the late 50s, and then through the 60s, they pretty much immediately began to talk like mutation as they began to talk the way Thomas Hunt Morgan or or RC Punnett had talked about evolution as a succession of mutations that are accepted, because they're beneficial. But they had this strong dose of neutralism in there. They said that mutations are constantly coming on, and they're either accepted or rejected based on fitness effects, but oftentimes, they're just sort of tolerated. And they started to say something, a new thing, which is that it was the least important parts of molecules that changed the most. And the kind of evolution that they were describing this sort of Markov chain of mutation fixation events, you know, that wasn't the classical view. The classical view was adaptation is this multi-threaded process, there are many allelic effects at different loci that are brought together by recombination and selection, and that's how you get adaptation. So they were talking about things in a quite different way, and in that 1969, King and Jukes' co-proposal of the neutral theory, they said, "We need new rules to understand evolution." That was a pretty serious challenge to the conventional view, but what sort of happened, I think, was that most organismal biologists just didn't really care, because it's just these molecules. And there was kind of a truce that developed and eventually, you know, Zuckerkandl and Kimura signed on to this truce, which said that, "Okay, we get molecular evolution, but it's this invisible thing that happens down where you can't see it. And what's really important, for the organismal biologists is selection, and that happens on phenotypes, right? And that's just a separate realm down there." And that's the way things were in the 70s and 80s, except for this debate about neutral evolution among the population geneticists who were focused on understanding variation. For the people who were looking at divergence, like long term divergence comparing molecules, they already went over to this mutationist view of evolution. Ok so I want to kind of get back to the thread and some of the things that Cam was saying reminded me a little bit the issues that you were bringing up, Cam.
Cameron Ghalambor 20:04
Well, given the molecular tools that we have now, the juxtaposition of the two perspectives, sort of that adaptation occurs when there's a new mutation that gets swept to fixation by selection, versus maybe the more of the buffet view that you were talking about that there is a lot of standing, genetic variation, even more than we previously appreciated, and even the existence of variants that may be very rare and even deleterious, but still persist in populations that under the right conditions, can still be swept to fixation.
Arlin Stoltzfus 20:43
So I wouldn't agree that there's sort of a pendulum swinging back and forth between people thinking about evolution from new mutations and people thinking about evolution for standing variation. I feel like the structure of thinking is really based on assuming evolution from standing variation, and it hasn't really been questioned. It's a different theory from origin fixation models that treat evolution as this two-step process of mutational introduction, and then fixation by selection or drift. And if you look at the history of those models, the history of those models is interesting. So first of all, they didn't emerge mathematically until 1969, in the midst of the molecular revolution. And then for like 20 or 30 years, they were mostly used by the molecular evolutionists, who were looking at long term divergence, right? And they were making all these sequence models. And they began to be used in lots of things, even in phylogenetics. That's something that became popular in the 90s, you know, with H. Allen Orr, with the mutational landscape model, which I mean, it dates back a little bit earlier to Gillespie and the 80s. But it didn't really become a thing, you know, until the 90s. So I would say that among theoreticians, who are making models, the recognition that there's an introduction process, and you need to have it to get the dynamics right, didn't really happen until the end of the 20th century. And I think we're still working through the implications of this, and it hasn't really been integrated into evolutionary thinking. So an example of that would be the whole Evo Devo thing. So what happened in the 1980s was developmental biologists said, "Oh, we think development shapes evolution by setting out possibilities prior to selection." And the response from authorities was "No, you're confused. Development is not an evolutionary cause." A theory that formalizes that intuition didn't exist in population genetics at the time. So Cam and I may be looking at things in different ways, because I'm thinking about the development of formal theory, and this is very much delayed. Whereas it could be that, you know, people working on experimental systems have been thinking about evolution from new mutations, you know, without having this kind of theory.
Art Woods 22:46
I think this is actually a good time to segue to maybe a new idea. And this sort of builds on this metaphor of the sushi conveyor belt and thinking about origin fixation models. And I'd say, to me, and I think you've argued in the, in your book and some of the papers, this opens the door to thinking more seriously about how the processes of mutation itself can influence the trajectories of the evolutionary process. And this is something that you and others call mutation bias. So maybe let's just dwell on that idea for a second, what is mutation bias? And how can it affect evolutionary trajectories?
Arlin Stoltzfus 22:55
So once you've decided that maybe mutation is not irrelevant anymore, now it becomes important to have a theory for that, right? So that you like, again, if evolution is coming in chunks, you want to have a theory, a predictive theory for what those chunks look like, and how often they come along. So if you were to take any organism and just measure a specific mutation rate, and keep doing that, you would immediately find that they're patterns, not everything happens at the same rate. This would be true for any kind of mutation, if you were looking at deletions, or insertions or whatever, but nucleotide substitutions are the most familiar mutations, just TCAG, you know, you change from one to the other. And one of the patterns you'd notice immediately is that transitions, T to C, or the reverse of that, and A to G or the reverse of that those happen more often than transversions in most organisms.
Art Woods 24:14
Then like a T to a G, for example.
Arlin Stoltzfus 24:16
Right, exactly. So that's one kind of bias. But if you kept looking, you'd see other things like in a lot of organisms, there's a bias toward more AT.
Art Woods 24:25
And would you say a bias toward AT, you mean that in genomes, there's more letters A and T than there are G and C generally?
Arlin Stoltzfus 24:30
Uh no I'm talking about mutation rates. So if you look at mutation rates, then the form of an AT bias could be it could mean that G's and C's are just more mutable, and so they change more often, or it could be that there's a direction going from GC to At. But that's the overall direction of the biases toward more At. If you just let it work, that's what would happen.
Art Woods 24:30
Got it. And does that result in greater fractions of A and T in genomes, that bias?
Arlin Stoltzfus 24:31
Well, mutation bias is one of the factors that has to be considered when you're looking at things like codon usage and genome composition. There are other factors. I mean, a selection is obviously involved, like if you're talking about codon usage, and also in organisms that are diploid, bias gene conversion can be important as well.
Art Woods 25:16
So that all makes sense at some level in a kind of abstract way. Maybe could you give us an example of how a transition transversion bias could end up, you know, having an effect on genes or the genome?
Arlin Stoltzfus 25:28
Yeah, so this might be the time to talk about climbing mount probable. And just to talk about, in general, how do you think about this role? Yeah, so it's just a metaphor to explain how to think about biases in the introduction process, and to think about this differently from raw materials. So imagine mountain climbing as a metaphor for evolution. But in order to make it mechanistic, you can't like look at the peak and find your way to the peak, you just have to climb in the dark or the climber has to be blind or it has to be moved by a completely local algorithm. And we'll make it a two step algorithm where first the climber reaches out with a hand or foot to sample different handholds. And then in the second step, with some probability, the climber is going to commit and shift its weight, alright? So if we put a bias on the acceptance step, the second step, so that there's a greater chance of accepting handholds and footholds that are higher up, then the climber is going to climb. But we could also impose a lateral bias at the proposal step, we could say that our climbing robot has longer or more active limbs on the left side than on the right side, so it's more likely to sample handholds. So then the joint probability of proposal and acceptance is going to be higher on the left. Now it's still climbing, but it's going to climb up and to the left. On a rough landscape, and most landscapes are rough, you always end up on some local peak. And so the local peak will tend to be up and to the left. If you're on a totally smooth landscape, eventually, you'll get to the summit, you know, you spiral out around and get to the summit, and it doesn't matter, but your pathway will either be to the right or to the left, depending on the biases. So there are theoretical results that tend to confirm what this metaphor suggests, but let me just mention some concrete examples. So just considering one step adaptation, simplest cases, you've just got two steps, we can go up and to the left or up and to the right. And what the theory says is that, even if going to the left isn't a selectively favored, it can happen a lot more often if it's just more mutationally likely. So an example of that from for instance, Justin Pritchard lab a few years ago, they were looking at resistance to Imatinib, which is an anti-leukemia drug. And there's two different resistant variants at the same side of a particular protein. And when they measured the resistance in the laboratory, the one that's not as clinically frequent, that was actually better, alright? And it turned out that it's because of the one that's more frequent arises at a higher mutation rate. There's a exactly analogous case from HIV where there's a particular site in the reverse transcriptase that responds to an anti reverse transcriptase drug. And again, the mutation that is more frequent clinically, is not the one that's more resistant, alright? They're both resistant, but this one is slightly favored by an A to G mutation bias.
Arlin Stoltzfus 25:29
It just happens more frequently.
Arlin Stoltzfus 25:51
Yeah. So that's the implication for one step evolution, just if a change is more mutationally likely, you have to weigh both the fitness benefit and the degree of mutational likelihood.
Art Woods 28:27
Yeah, I guess this is partly about sort of long term trajectories and sort of thinking again, about your metaphor of climbing mount probable, if the landscape is rough, and populations always get stuck on suboptimal peaks, then couldn't you say that mutation bias doesn't really matter, because either way, the population is going to get stuck on some local peak. The mutation bias may influence, you know, which of those local peaks it is, but does it really change the overall course of the sort of macroevolutionary trajectory of the populations?
Arlin Stoltzfus 29:02
Well, it depends on what you're trying to predict. So like, as a first approximation, if you only care about fitness, and you don't care about movement in the directions of phenotype space, then yeah, maybe that's okay, right? But when you're actually looking at evolution, you don't know that, you know that the phenotype has changed. You don't necessarily know whether that's a fitness increase.
Art Woods 29:24
Yeah. Okay.
Cameron Ghalambor 29:25
So Arlin, can I follow up on that? So I really liked the metaphor, it's both a function of the landscape itself. So how rough it is, for example, and also that the length of the arm, for example, could vary between the left side and the right side of the climber. But it also strikes me that like, especially with regard to the length of the arm, that also has the possibility of evolving, as maybe does the landscape in terms of the mutational inputs. And so like, you had mentioned that when you get to genes that are highly expressed and potentially really important that selection becomes really important at that stage. But it also strikes me that those would be the genes that maybe play the biggest role in adaptation. And so like, I think there's this pattern that there's fewer mutations or a lower mutation rate in important sort of coding genes than there are, say, in noncoding regions of the genome. So I'm trying to reconcile kind of the evolution of the mutation rate itself and like the mechanisms that are at play that are available for selection to sort of act at that level, in terms of either putting the brakes on that bias or promoting it, you know, when it might be beneficial.
Arlin Stoltzfus 30:49
So this gets us into something that I think of as a different area. So I think you were referring to the Grey Monroe's paper that came out early, early last year showing this pattern in Arabidopsis. So what they showed is that if you do a detailed analysis of mutation in Arabidopsis, what you find is that the rate of mutation is lower in non-gene regions than in genes. And it's lower in the apparently less important genes than in the more important genes. And they argued that this was adaptive. And I actually suspect that that's a solid result for Arabidopsis, but it's not really known for other species. It's an open question whether that pattern would be found more broadly. I also think it's an open question whether that pattern is adaptive. The pattern can be explained by certain features that are found more often in genes than in non-genes. I can't off the top of my head, remember what they are, but they're things like the chromatin, that organized chromatin or something like that. If it is something that evolved by natural selection, it would have happened by this theory of what's called amelioration. So in order to improve the mutation rate, you can, it's possible theoretically, you know, the numbers add up in terms of population genetic modeling to reduce the mutation rate by reducing the deleterious load, if you can satisfy a particular inequality that has to do with how much you're reducing the deleterious load. The system's not understood in sufficient detail for them to really make that argument about whether it's plausible for that to happen.
Arlin Stoltzfus 30:57
The reason I think of this as quite separate is because, well, in terms of the theory in terms of theoretical apparatus, the theory for amelioration is a different kind of theory than, like, the theory of biases in the introduction process. And the other thing is, I wouldn't want people to get confused when I talk about mutation bias devolution or the effect of biases in the introduction process. This depends on some kind of tendency for the mutations to be biased toward things that are more fit, because they're not. Like in the example that I gave you, specifically, the model is the one that's more mutational favorable is not the one that's more beneficial, right? So the theory doesn't depend on there being any kind of adaptive,
Art Woods 33:18
Or like directed even
Arlin Stoltzfus 33:20
Right exactly.
Art Woods 33:21
There's no directed mutation?
Arlin Stoltzfus 33:22
Well, the model doesn't depend on it. Now, if you asked me the question, because I wrote this book about mutation, randomness and evolution, I had to address the issue of specialized mutation systems, even though it's kind of a side branch, relative to the rest of the book. Because there's absolutely no doubt that there are specialized mutation systems. I mean, yeast mating type switching is basically a deterministic mutation, you know, that happens in the germination of spores. There are these shuffling systems in like Trypanosomes, Cerevisiae
Art Woods 33:54
CRISPR.
Arlin Stoltzfus 33:55
CRISPR is another category. Yeah, there's like five or six different in the book, I go through six different categories of these systems that are, they're elaborate, you know, and in some cases, one case that I mention in the book, it's been demonstrated experimentally that this shuffling system helps a pathogen evade the host immune response, because if you knock out the shuffling, that reduces the survival of this bug, but only in immunocompetent mice. If they're immunocompromised mice, then they don't need the shuffling, and that's just, you know, a very direct experimental result there that establishes the importance of this shuffling system for immune evasion. So specialized mutation systems exist, but they're not what drives the theory of mutation biased evolution.
Cameron Ghalambor 34:42
So I'm gonna add a bit of a follow up but maybe it's really more of kind of a stepping back. In this literature, I see a lot of terms that are sometimes I'm not quite sure on, on how to differentiate between them. So for example, a mutation hotspot versus mutation bias versus fixation bias. And I think we've maybe used fixation bias and mutation bias in our conversation so far, but I'm not sure that we actually formally defined how they're different. And I think I have an idea. But I'd like to hear your sort of how you differentiate like, between the two, when you're looking empirically a data.
Arlin Stoltzfus 35:28
Yeah, the literature is confusing. And the term like a mutation hotspot or mutation bias, those terms may be used, where they're actually referring to an evolutionary hotspot. So from my perspective, as someone who is really interested in the role of mutation in evolution, I really want to strictly distinguish between origination and fixation as two different steps in the evolutionary process. So sometimes out there, people will say that mutation bias affects the fixation probability, and I wouldn't never say that, because that's confusing the two steps. So if you really want to keep these separate, you need to have a language for origination, a language fixation and a language for the joint process. Let's say the frequency of evolutionary changes happening, that's the whole process is jointly dependent on biases in origination and biases in fixation. That's the way to say, that is a way to speak clearly about these things. But there's a lot of confusion in the literature about this.
Arlin Stoltzfus 36:24
So let me get back to your first question. So a mutation bias is a predictable asymmetry in mutation rates. And I specifically mean mutation rates and not evolutionary rates. When we talk about the mutation spectrum, it's just a broad way of referring to a whole set of possibilities. And again, for me, mutation spectrum is a spectrum in the rates of mutation, the process of mutation itself. Whereas if you go out there in the clinical literature, people will talk about a spectrum of mutations for a particular kind of cancer. And they mean what are the clinically manifested allelic types that they see, they're not actually talking about the process of mutation, they're talking about that joint process of mutation and establishment. I often as a researcher, who depends on other because I'm a computational person. I depend on data that's already published, you know, and reading lots of papers. And I often get stuck, because I can't tell what authors are talking about, I can't tell whether they actually have knowledge of mutation preferences, or whether they're talking about evolutionary preferences or clinical preferences. So yeah.
Cameron Ghalambor 37:27
But if there's a fixation bias, is that strictly? Well, I guess it's, it reflects, in part, the input of mutations into the process, but it's primarily also then being driven by selection. Is that fair?
Arlin Stoltzfus 37:43
Yeah. So I would say that to you, if you're using the word fixation bias, you should be talking about selection.
Cameron Ghalambor 37:48
Okay. Yeah.
Arlin Stoltzfus 37:49
But an evolutionary bias could be due to the fixation bias or the origination bias.
Cameron Ghalambor 37:54
I think it'd be helpful maybe to take a step back a little bit, because I think this idea of directed mutation is a fairly controversial, maybe loaded topic within evolutionary biology and instantly raises red flags, because it sort of invokes this maybe Lamarckian kind of view where the organism is faced with some kind of challenge, and then creates and generates mutations to help it adapt to this kind of new challenge that it faces. So we could talk probably for hours about that, but-
Arlin Stoltzfus 38:37
Yeah, this is an issue that makes people's heads explode. Yeah. And it is addressed in the book. And okay, so here's what I would say about this. There's a particular way of looking at independence, the idea that nutrition and fitness are independent of each other. Okay, so let me back up just a little bit. There's this randomness doctrine in the evolutionary literature. People say mutation is random, or they refer to mutation as being random or associated with the concept of randomness. And if you go through and you look at those statements, they're just all over the place. They're a mess. And I tried to sort out a bunch of that in the first chapters of the book. If you get into real detail, and you try to make sense of the randomness doctrine, it goes in two directions. One of them is irrelevance, we've already talked about that. That's just a restatement of the Neo-Darwinian dichotomy of selection as the potter and variation as the clay. Variation supplies raw material that selection shapes into adaptation. And randomness as irrelevance is just kind of a way of sort of pushing that down to make it seem like it's a bottom-up principle, when it's really not.
Arlin Stoltzfus 39:39
The other interpretation of the randomness doctrine is that it sounds more mechanistic, it sounds like there's something about mutation that makes it independent of fitness, alright? And a lot of those ideas don't work very well either, and I give some examples in the book about how there are these obvious cases where you're going to see correlations between mutation and fitness just because of the way the biology works. There was there's that example about lateral transfer, and there's that example about nucleotide precursor pools.
Arlin Stoltzfus 40:08
Now you can get beyond all those by this idea of conditional independence by saying that mutation and selection are independent of each other conditional on the common conditions that affect them both. And that's a good definition in the sense that we don't know any biological systems that contradict that. In order to contradict that, you would have to have a mutation system, where the mutation develops through intermediates, and during that process, it somehow senses what the incipient fitness effects of those intermediates would be. And then that feeds back and it either inhibits the things that look like they're going to be deleterious or promotes the things that look like they're going to be favorable. That's clearly the kind of thing people mean, when they talk about directed mutation. In the adaptive mutations of the directed mutations, controversy that was sparked in 1988 from that Karen's paper, that is what I would say, what that controversy is about. Are there are these real time responsive systems? And there's no convincing evidence that those real time responsive systems exist. Okay. But I just said a minute ago, they're specialized mutation systems. Well, they're special, they do these elaborate things like I mean, the paradigm example, the exemplar that I use in the book is somatic recombination, the development of antibody genes by shuffling together three different parts, you know, and there's like four thousand, different combinations. And then there's junctional, sliding and there's hypermutation. And that's obviously an elaborate evolved system, but it doesn't have real time responsiveness. So my point is that you can have all kinds of strange and interesting things happening with mutation that are not directed mutation, right? They don't involve real time responses, but they're still specialized, they still help the bacterium, in one of the cases that I cited, to escape the immune response of the host.
Art Woods 42:05
Yeah that's a really nice distinction.
Cameron Ghalambor 42:06
I was just curious about these specialized systems, not knowing a lot about them. But it strikes me that those types of systems would be particularly useful in situations like, say, the immune response or in situations where the organism or the immune system has to deal with a lot of variation that may be unpredictable. And so having these kinds of specialized systems that are generating variation, even if it's relatively blind to the specific kind of challenge that might be coming, they would be strongly favored in a system. Is that a fair way of characterizing those kinds of systems?
Arlin Stoltzfus 42:51
Yeah, I tried to develop some thinking about that in the book. And that the thing that takes a while to wrap your head around, is that it's hard to design a mutation system that responds to new challenges, to genuinely new challenges, right? Comes up with new things. But that's not really how these systems work. They're switching between two known things, or they're evolving away from the previous thing. So there's many examples in nature of away from adaptation, for instance, tetrodotoxin, or something like that, you know, it affects a particular sodium channel, there's a binding pocket. And so to evolve to tetrodotoxin resistance, what you have to do is disrupt that pocket in some way. And so there's a particular loop that I'm thinking of where there's multiple changes at exactly the same site that all confer resistance, and it's because it's evolution away from something. And that's what's happening when you're evading an immune response. So that Tripanosan that I talked about, it coats its whole body with this big surface protein. So that's the major antigen. And then that is what's changed. And what it's doing is it's just evolving away from whatever it had before.
Art Woods 44:00
Yeah. It just needs to be different.
Art Woods 44:02
Yes, yes. So that's an elaborate system for evolving away from what the previous state was. It's not sensing and responding to some new condition that it's never seen before.
Cameron Ghalambor 44:12
Yeah, so maybe a better analogy is like systems that are in sort of evolutionary arms races.
Arlin Stoltzfus 44:19
It does appear, yeah, that there's some association between these, because a lot of those systems that I- so there's diversity generating retroelements is one of them. There's the CRISPR CAS system, there's phase variation. Those three are all involved in bacteriophage competition with each other. And we know this, because there's a lot of viruses that have anti CRISPR proteins, right? So they must be fighting with each other over CRISPR. And in the book, I kind of give a hand waving argument that, you know, there's a global population of bacteria that's just enormous, okay. It's been estimated like 10 to the 32, I think, something like that. And they're turning over rapidly, and a lot of bacterial mortality is due to phages, so there's just an enormous amount of death. Life, birth and death happening and birth and death is what fuels adaptation. So there's a huge amount of fuel there. And maybe that's why, I speculate, that we see such elaborate systems that appear to be enhancing evolution in this interaction of bacteria and phages. But you see a little bit of the same thing in the interaction between pathogens of mammals and their hosts, which have that because of the adaptive immune system.
Art Woods 45:34
Speaking of population sizes, I think that leads to another interesting question. And I want to just sort of reframe this question of mutation bias in terms of maybe a population genetics question, which is, how much does the importance of mutation bias depend on the overall rate of mutation, on the population size, and on the strength of selection? And maybe that's giant question, and we can just pick out some parts of that.
Arlin Stoltzfus 45:58
Yeah, I mean, that's a giant question. The first thing I would say is that if you're really, really interested in this, there's a paper from last spring from Alejandro Cano was the first author in PNAS, where there's a simple gene model, and we're just looking at what's the effect of the mutation spectrum. And the effect of the mutation spectrum is strongest when mutation supply is low, when new mutations are entering the population rarely. And then when mutation supply becomes very high, as I mentioned previously, if all of the relevant alleles are present, or all the allelic types, if you're looking at mutation biases, are present, then the best one wins, and there's no room for mutation biases. It's also important to have like, if you want to see the effects of mutation bias, you won't really see it statistically, unless there's a diverse mutational target, unless there's many ways. If there's only like two ways to get this antibiotic resistance, then there's not many degrees of freedom for mutation bias to influence what happens. So the one thing I would say is, yeah, there's a bunch of population genetic work that's been done on this. And mutation supply is very important. The diversity mutational target is very important.
Arlin Stoltzfus 47:04
In the big picture, however, what I would say is that there's always something in evolution that could be happening based on rare mutations, even if there's something else that's happening based on standing variation. So in any mammal population, you've got all these short tandem repeat loci, right? And they're so variable, because the mutation rate is like 10 to the minus three per generation for changes in the repeat length of short, tandem repeats. That's why they're so useful in forensics, right? Because we're so likely to differ from each other. And to the extent that those are affecting things like gene regulation, there's just like this automatic supply of variability that's present all the time, because the mutation rates are so so high. But at the same time, you know, there's always been an intuition, like a minority view in evolutionary biology, that there are rare important things, right, rare innovations that play an important role in evolution. So I would say, there's always something that can happen, that's dependent on rare changes. Does that satisfy?
Art Woods 48:06
Yeah, I mean, I'm satisfied in the sense that I think it's a giant question, and it feels like it would need, you know, hours of conversation to get through, but maybe let's just leave that there for now.
Arlin Stoltzfus 48:16
I mean this is a big issue, it's an open issue, it's gonna take, you know, generations.
Art Woods 48:30
So Arlin, in this conversation, we've touched a couple of times on processes and variation and biases occurring at the molecular level, we've connected that in a few ways to variation and biases at sort of higher phenotypic levels, but I just wanted to get your broad perspective on how much does mutation bias affect bias in the sort of organismal level traits that end up coming to fixation? Or that, you know, that end up evolving in different lineages? And I guess I'm asking this because it feels like, you know, still, there's this possibility that somebody could, you know, listen to what you're saying, read your book and say, you know, mostly what he's talking about is bias at the molecular level. And that's still, in some way, kind of partitioned off from this other kind of trait evolution that I, you know, that we care about.
Arlin Stoltzfus 49:18
Yeah, you could say that. And some people are going to say that, you know, until they're forced to believe otherwise. Right. So, generally-
Art Woods 49:26
Until they listen to this podcast
Arlin Stoltzfus 49:27
In general, that's a big question. And we don't know the answer. So let me suggest some ways to think about it. So first of all, one of the reasons that I'm focused on the molecular level is I just feel like we can prove stuff here, especially because there's experimental work, right? I mean, the principle of selection, the idea that selection is a powerful, powerful force, you know, that was established theoretically. And then it was established with experiments. It's very difficult to show selection in nature. And in the same way we feel that the effect of biases in the introduction process has been demonstrated very clearly experimentally. And then we also see it in some retrospective analysis of adaptation. But you're right, those are mostly, we're talking about changes that can be traced to the molecular level where it's literally a macro-mutation, right? So if you have something like a change that confers resistance to tetrodotoxin, with just one change, or confers resistance to an antibiotic, that is a macro-mutation, it's a large effect mutation. So to some extent, we're talking about that. And so people will think well, how does this relate to the visible morphologies and behaviors of charismatic megafauna that evolutionary biologists have been concerned with for 150 years? And again, the answer is that, I'm not sure, but let me tell you how to think about it.
Arlin Stoltzfus 50:43
The way to think about it is to approach this with a genotype-phenotype map. So imagine a grid, you know, highly abstractly, there's a grid, right? And that's all the genotypes and things that are close to each other on the grid are close genotypes that can be reached by mutation easily. And then there's some blobs on the grid, and those are how phenotypes map to genotypes. So one of the things that we know is that there's usually many, many genotypes that can encode a phenotype. And they're often close together, although they're not always close together. So the network of genotypes that map to a particular phenotype may be disconnected, there may be different disconnected parts in genotype space. So the principles that we know about from theoretical modeling are there's two big ones. So one of them is about the adjacency of phenotypes in genotype space. So think about those blobs on the grid, some of those blobs will have a lot of adjoining surface area, right, there'll be abutting each other, they'll be next to each other. And the more space they have next to each other, the more possibilities to mutate from one phenotype to the other phenotype. And the other big principle is findability. So if you have a phenotype that has lots and lots of genotypes, it's more spread out in genotype space, there's essentially more mutational arrows pointed at it from other regions of genotype space, and that makes it more findable by an evolutionary process that proceeds by mutations.
Arlin Stoltzfus 52:09
So an example where we can show this process in a model, and it also corresponds with what we see, would be for RNA folds. RNA sequences, you know, you can just imagine any sequence in a computer, and you can fold it up with an algorithm and it has a shape, okay, that's its phenotype. And if you do this a gazillion times, and you take those different shapes and cluster them into families, there are some fold families that are much more common in sequence space. And if you do evolutionary modeling, you can show that those families or those folds are more accessible, they're more findable, evolution is more likely to end up on those folds. And you can also show that there's a very, very remarkable correspondence between the RNA folds that are most common in nature, alright, that are found encoded in genomes, and the folds that are most common in sequence space. So that was all very, very abstract, alright? Now, the reason that we can't do that with animal body plans is that we can't sort of make all genotypes of animals and just like produce them like, but with the RNA sequences, you can just fold them up in a computer and get a fold. The thing that you can do with a more complex animal system is to just explore the neighboring genotype space, and to try to map that out, just from one from one local point. And you can also try to do some modeling about what larger regions of the space would be.
Art Woods 53:40
So you're thinking of things like digit loss in salamanders, that kind of thing? I mean, that's a body plan issue. But you know, losing this digit versus that digit is an adjacent way of thinking about that, right?
Arlin Stoltzfus 53:52
Absolutely. Absolutely. And there's this classic paper from Evo Devo, which is the Alberch and Gale paper, maybe that's the one that you're thinking of where- so this actually, that's actually great example of what you can do, so let's talk a little bit about that study. So they were talking about was it salamanders and frogs?
Art Woods 54:08
Yeah, I think so.
Arlin Stoltzfus 54:09
So salamanders and frogs, they looked at these patterns of digits. And you can lose entire digits, you know, or you can lose just a phalange like, just one of those little bones. And they looked at patterns of loss in evolution in salamanders and frogs, and they saw there were different patterns, you know, you can lose a digit axially or distally, right? On one side of the hand or on the other side of the hand. And that was one of the differences that they saw in the preference for axial or for distal loss. And there were also some preferences about which digits tended to lose phalanges. And then what they were interested in was whether that might be a developed, there might be some kind of developmental bias. So what they did was they took developing frogs and they treated them with colchicine, which is a microtubule inhibitor. And it just kind of messes up development. So why did they do that? Well, the idea is that there's a dynamic developmental system, and if you disrupt it, it's going to show certain preferences. And they were thinking of this disruption by colchicine as an analog of mutation. So they could do this very quantitatively, just lots and lots and lots of frogs, or lots and lots of salamanders, and look at what happens to the digits. And what they found was that there was a correspondence between the pattern of digit and phalange loss due to colchicine that had kind of developmental preferences and an evolutionary pattern of loss. And what was great, what was so nice about that argument is that distal versus axial loss seems very arbitrary, so it's difficult to make up a selective explanation for why that would happen. So if you accept their premise that, you know, colchicine is like is a kind of perturbation, like genetic perturbation, they were looking at the the local accessible possibilities, right, and assaying them in that way, and then showing that those preferences map to evolutionary preferences.
Art Woods 56:03
Just to fold this back into your RNA example, it sounds like you're saying that, you know, the RNA folding example, there's a relatively simple and straightforward genotype to phenotype map. In this sort of digit loss example, it's not genetic, but it's sort of an analogue of a genetic process that's a fairly straightforward, quote unquote, genotype to phenotype map, but that we can't do it for like entire animal body plans, because that G to P map is just way too complicated. And so to arrive at the idea that there might be biases in these really complicated G to P maps, it's a question of extrapolating from these simpler examples.
Arlin Stoltzfus 56:40
Yeah, something like that. So before I go on, I just want to call out the authors of that RNA study, so Kamaludin Dingle, is the first author and he's working in Ard Louis' lab, just great, great work that they've done. So to get to your question, I hope I don't offend people too much. But I think that there are things that you can prove at the molecular level that are just never going to be proved for animal body plans, for like the visible morphologies and behaviors of charismatic megafauna. That's what people have been talking about for 150 years, but they're just so inaccessible to experiment. So-
Art Woods 57:15
I mean, I care about these organismal traits, but I'm not offended by that.
Arlin Stoltzfus 57:19
Ok it could be the best that we can do. See, if we can establish using the examples that are accessible to us that there's this principle of the effect of biases in the introduction process, then you're more confident to apply that to something where you don't really have the leverage to prove it. So, you know, going back to selection, it's you can show in the laboratory and in the field that selection is very powerful with experiments, it's a different set of techniques to do that when you're looking retrospectively at evolution.
Art Woods 58:02
I want to switch now over to thinking about sort of more practical applications for all of this theory that we've been talking about. And I would say, one important goal of evolutionary theory, at least for some people in some contexts, is to make better predictions about evolutionary trajectories into the future, to predict, you know, what kinds of drug resistance are going to evolve and how rapidly and in what way? And I'm just wondering what you think, Arlin, about how your ideas about mutation bias and the importance of the introduction of mutation, how does that inform the kinds of predictions that we can make?
Arlin Stoltzfus 58:40
Yeah, well, you mentioned that paper that's on a preprint server right now, it's Alejandro Cano, is again, one of the first authors. So there we're trying to apply these ideas to, to prediction and forecasting. The answer is, we don't know, we don't know how important this is going to be in practice. But what we did notice, you know, if you look at the prediction literature, you can see that there's a hole there because everybody, in the first paragraphs of their prediction paper, they say there's some things that make evolution predictable, selection. And there's other things that make evolution unpredictable, mutation, right? So there's gap there, because biases and mutation make evolution more predictable, so how can we take that into account? Again, we're not really sure how important this would be practically, but certainly, we can look at the kinds of things that people do forecasting on or want to predict, like cancer, you know, mutational effects are all over the place in cancer studies, there. TP53, for instances, this is this very common cancer driver locus, and the top five most frequent mutation, clinical cancer mutations and TP53 or CpG, mutations, mutations at CpG hotspots. So effects like effects like that can be important. And there's, I mean, it's actually an interesting and complex story there about CpG so do you know what I'm talking about the CpG sites?
Art Woods 1:00:03
Maybe for our listeners just explain the CPG process.
Arlin Stoltzfus 1:00:06
Yeah, okay, so CpG means C phospho G, so it's just talking about two sites adjacent to each other and DNA where there's a C, and then there's a G. In mammals and birds, the cytosine tends to be methylated. It's not always methylated, but it's often methylated. And this has a particular effect on mutation. So when cytosine gets oxidatively deaminated, it turns into uracil, which doesn't belong in DNA, and an enzyme comes along and cuts it out. If it's methylated cytosine, it turns into thymine, which does belong in DNA. And so it's not recognized as well, but there's still a miss-pair. And so that's why CpG is a hotspot because of this effect of the methylation. Now, in the human genome, and in other mammalian genomes, there's a lot less CpG than you'd expect from the frequency of the two nucleotides. It's about, I think, eight fold less frequent, something like that. And here's the funny thing, so over time, if a dinucleotide has an elevated mutation rate, and it evolves, the sites evolve away over time, so that you have fewer of them, then the ones that are left are going to be the ones that are difficult to change, because they're deleterious. And so I don't think this has ever actually been proven, but I'm very sure I will predict that this is why there are so many CpG changes that are involved in monogenic diseases in humans, there's just a ton of them. And part of the reason is surely that the CpG mutations take place 10 times more frequently than other nucleotide mutations. But I think another contributing factor is just those are the bad sites, the ones that are left, because the ones that were easy to get rid of evolved away from CpG, because it has a high leading rate, right, because of the high mutation. So that's an example of, you know, something that's really fascinating about the evolution of the genome that reflects an effect of mutation bias, or at least I'm hypothesizing that that has a very practical role, I think, in disease. Obviously, for COVID, we're hearing about mutants all the time, and it would be nice to be able to predict those. And again, I don't know how important it will be, but it's definitely something that's going to improve prediction.
Cameron Ghalambor 1:02:24
So one example that I know a little bit about, but I think, I suspect, you know a lot more about is the evolution of hypoxia tolerance at high elevations like in mice and birds, and how the hemoglobin molecule evolves to have a higher affinity for oxygen under these low oxygen conditions. And I remember a paper by Jay Storz several years ago, where he kind of describes the process by which those changes occurring, and he talked about it in the context of epistasis. And I remember being really confused when I first read that because, you know, my thinking with epistasis was really more statistical, as opposed to kind of thinking about it from a molecular perspective. And I've seen that example come up also in the context of mutation bias, and the sort of predictability of how adaptation to high elevation occurs. So can you can you talk about how mutations play out in this particular system
Arlin Stoltzfus 1:03:28
Well, that's one of the natural cases where I think we have good evidence for effects of mutation bias. So one of the things that Jay's group did was to assemble this really powerful data set where they have 35 pairs of low and high altitude species. So there's a high altitude population, and there's a low altitude population, and they're related to each other. Sometimes there's subpopulations of the same species, but they're quite separate and other times they're related species. So that's like 35 matched pairs to do phylogenetic contrasts with, which is great. And in most of these, they've done this detailed biochemical work where they, you know, they actually take the birds and they get blood, and they take hemoglobin out and they make measurements on it. So one of the things that we did, we actually were lucky to collaborate with Jay's group, and looked at an effect of, again, the CpG mutation bias on these changes in hemoglobin. And I don't remember how the the numbers off the top of my head, but there were like, I think 22 different changes, where you could say credibly, that the high altitude species has a higher affinity hemoglobin, and it's due to this particular change, and ten of those were due to CpG changes where you'd only expect about two of them if the mutation rates didn't matter. So, you know, high altitude adaptation to hemoglobin, you know, that's a classic example of adaptation in evolutionary biology, and if you look in detail at this case with these with these birds, you see an influencing mutation bias
Art Woods 1:05:00
Okay well, Arlin, thanks so much for this great conversation. I really learned a lot. And this is pretty far outside of my normal wheelhouse. I really enjoyed it. We'd like to wrap up just by asking a question that we ask all of our guests, and that is whether we've covered everything that you want to cover, and if there's any last thing you'd like to say?
Arlin Stoltzfus 1:05:19
Oh gosh, we've covered so, so many things. I mean, invite me back for a second session, you know, after you get some feedback, or you have some time to think about it, because, you know, this, this work is something that takes a while to sink in.
Art Woods 1:05:34
Yeah, well, great. I really enjoyed your book and papers and just a lovely conversation. So thank you.
Cameron Ghalambor 1:05:39
Yeah, thanks a lot.
Arlin Stoltzfus 1:05:40
Yes, thanks a lot.
Cameron Ghalambor 1:05:57
Thanks for listening to this episode. If you like what you hear, let us know via Twitter, Facebook, Instagram, or leave a review wherever you get your podcast. And if you don't, well, we'd love to know that too. All feedback is good feedback
Art Woods 1:06:10
thanks to Steve Lane, who manages the website and Ruth Demree for producing the episode.
Cameron Ghalambor 1:06:14
Thanks also to interns Dayna De La Cruz and Kyle Smith for helping produce this episode. Keating Shahmehri produces our awesome cover art.
Art Woods 1:06:23
Thanks to the College of Public Health at the University of South Florida, the College of Humanities and Sciences at the University of Montana and the National Science Foundation for support.
Cameron Ghalambor 1:06:33
Music on the episode is from Podington Bear and Tieren Costello.