An international team of researchers have combined ideas from biology, computer science and mathematics to explain why evolution seems to favor symmetry.
We spoke with Iain Johnston, a professor at the University of Bergen, and one of the authors behind the article Symmetry and simplicity spontaneously emerge from the algorithmic nature of evolution, published in PNAS, about his findings and conclusions.
What are the main findings of the paper?
“The main finding is a theoretical explanation of why we see symmetric and simple structures so often in biology. We look around biology, on all sorts of different scales from sunflowers and starfish to the little molecular machines that work inside our cells all the time. We often see very symmetric structures or, if not perfectly symmetric, then simple modular ones where a single subunit or a small set of subunits are repeated over and over again.”
“We don’t invoke natural selection. We say that evolution deals fundamentally with information, and from this perspective, we argue that simple and symmetric structures should be favored in an evolutionary sense.”
“You might say that traditional thinking would say evolution is driven by natural selection, so there must be some selective advantage to have this symmetric structure or this simple structure. But we suggest an alternative to that. We don’t invoke natural selection. We say that evolution deals fundamentally with information, and from this perspective, we argue that simple and symmetric structures should be favored in an evolutionary sense.”
“So the technical result is that simple and symmetric structures need less genetically-encoded information to encode them. Such structures may require “assembly rules” involving fewer types of subunits and less complicated interactions, meaning that they’re just easier to discover through the random process that is evolution. It takes fewer mutations, and fewer evolutionary innovations, to discover the rules needed to build a simple or symmetric structure. So before there’s any level of natural selection acting for any particular specific organism, we expect these simple and symmetric structures to be favored just because they’re easier to come by in an evolutionary space.”
Was the result surprising?
“The picture of evolution as constantly increasing complexity it is hugely controversial. One could argue, for example, that bacteria are the most successful things on the planet, and although they have all sorts of interesting metabolic complexity, physiologically they are far less complex than us. Of course, we’re tempted to take an anthropocentric point of view and view ourselves as the pinnacle of evolution with our complex brains – there are lots of counterarguments to that as well. Several biologists have gone explicitly against this line of thinking and said evolution doesn’t necessarily increase in complexity, it just does a good job of discovering and working with what’s good at the task at hand.”
“We were surprised by the generality of it as well. In our models, we threw different selective pressures at our model system to say “do this” or “do that”, subject yourself to these criteria, and we always saw the simplest structures that fulfill those criteria popping out.”
“To start with, we were kind of dominated by the selective mode of thinking, where you see biological structures because they fulfill a particular purpose. But our picture instead suggests that evolution is also shaped by the intrinsic properties of the algorithms, the rule sets, necessary to produce structures, before any particular selection to fulfill this complex task or live in that complex environment. We expect simple algorithms to “arrive” first in an evolutionary search. We were surprised by the generality of it as well. In our models, we threw different selective pressures at our model system to say “do this” or “do that”, subject yourself to these criteria, and we always saw the simplest structures that fulfill those criteria popping out.”
“We’re not arguing that you’ll always get the simplest possible structures in biology, and there is, of course, a clear role for natural selection in shaping structure. But amongst the set of solutions that evolution could discover to fulfill a certain criterion or survive in a particular environment, we argue that we will see the simplest subset of those possible options emerge.”
What is your background and how did you end up in Norway?
“I’m a physicist by training and I’ve moved in joint biological and mathematical directions. I worked in various universities around the UK as an early career researcher and then as a lecturer. I moved to Norway for both personal and professional reasons. It’s a wonderful and beautiful country, especially with a young family; and Brexit raised several uncertainties around EU funding which I wanted to avoid.”
How does being a physicist play into the paper that you wrote?
“Towards the end of my physics training as an undergraduate, I was lucky enough to get a project with Ard Louis, my collaborator on the PNAS paper. The project used thermodynamic simulation, a fairly standard tool in physics, to explore the self-assembly of virus capsids. These are symmetric structures, quite beautiful if you ignore what the viruses do. These beautiful structures self-assemble with high fidelity inside our cells to make symmetrical football-like icosahedra, and we were interested in the ways that they do that, and what physical rules must somehow be encoded for their structure and subunits to facilitate this self-assembly.”
“These beautiful structures self-assemble with high fidelity inside our cells to make symmetrical football-like icosahedra, and we were interested in the ways that they do that, and what physical rules must somehow be encoded for their structure and subunits to facilitate this self-assembly.”
“Molecular simulation was quite a useful tool to look at that, and I worked with Ard doing some simulations on this virus assembly and looking at what challenged it and what facilitated it. Then we got interested in self-assembly a bit more generally in biology, the physical process where subunits come together to make things that are more than the sum of their parts. And that was really what paved the way for this project looking at how such structures might evolve the ability to self-assemble.”
What are the implications of this in terms of our understanding of evolution?
“The most general implication is that we don’t necessarily need natural selection to explain a particular aspect of biological form. If you see a symmetric structure, you don’t need necessarily to invoke the idea that this must have evolved because it has a particular selective pressure governing it.”
“We’re arguing that this symmetry, this simplicity, is the default. So in terms of the evolution of form, you don’t always need natural selection to explain the sort of beautiful symmetry that we see throughout life.”
An example of a highly symmetric evolved structure – a light-harvesting complex from a bacterium. Image is produced by Iain Johnston using the Protein Data Bank; structure 1nkz
What are the future directions of research in this area?
“We worked a lot with protein quaternary structures – these are the machines that proteins form in the cell to catalyze chemical reactions and perform all sorts of physical processes. We also worked with some RNA modeling and some biological networks, but it’s mainly sub-cellular structures that we’ve been looking at. A big question is, how much does this theory generalize to larger structures, such as tissue, organ, or full organismal forms? Can we use the same arguments to say – we expect symmetry to emerge naturally at the molecular level – do we also expect it to emerge naturally at the organismal level?”
“Can we explain our bilateral symmetry and the rotational symmetry of other interesting organisms, flowers for example, through this kind of algorithmic picture of evolution? We have some preliminary modeling studies using very coarse-grained organismal models to suggest that, yes, actually, if you view these things as sets of rules, algorithms, that are evolving to produce certain forms, again, simplicity tends to be favored.”
How does the research challenge the orthodoxy in evolutionary thinking?
“Sometimes, the more you pick at an orthodoxy in biology, the more you realize there isn’t really an orthodoxy. People will tend to occupy quite different places on the spectrum of accepted knowledge. If it is orthodoxy to say that every form must have a selected function, then we are explicitly challenging that. We’re saying that there is a class of forms that emerge before any degree of selection is invoked, just from the nature of evolution, dealing with algorithms.”
“We’re trying to take a different perspective and ask, regardless of the particular encoding details of this system, what can we say generally about the rules involved in the algorithms that produce these structures? That’s how we are working in a slightly unusual way perhaps.”
“The algorithmic picture of evolution itself is something that is not contradictory with a molecular perspective of evolution, but it’s one that not many people are looking into at the moment. Many research projects focus on the specific molecular details of DNA, for example. A lower proportion of projects think about the broader “principles” that those A, T, Cs and Gs are encoding, and that’s what we were trying to do. We’re trying to take a different perspective and ask, regardless of the particular encoding details of this system, what can we say generally about the rules involved in the algorithms that produce these structures? That’s how we are working in a slightly unusual way perhaps.”
So what are the limitations or potential biases in the research?
“There are several specific biases that are a necessary part of doing this research. In our work with protein quaternary structure, we make use of an amazing resource called the Protein Data Bank, the PDB. That contains a set of protein structures that scientists have solved, often through x-ray crystallography or other methods. But there is some inset bias in that database because we’re necessarily looking at structures that are easier to solve, rather than ones that are hard to solve. And that relates to particular classes of proteins, being over-sampled in that database and that could skew the numbers in some of our biological examples. I’d be very surprised if it skewed the overall trends that we find, but there’s no denying that some of our illustrative examples are built on a sub-sampling of biology.”
“The same goes for the modeling paradigm more generally. We’re using computational models to explore what we think is a fundamental principle. Biology itself is very, very complex and there’s no way a given computational model can address all of the subtle details that go into the evolution of a particular molecule. So it’s a necessarily coarse-grained perspective. But in a sense that is taken into account by how we work. We’re interested in the general rules and we’d like to drill down into specific instances that are of particular interest. The length scale of our biological examples is another limitation. We focus on the molecular level, but left it quite open on how much this could translate up through different length scales to the organ, the tissue, and the organismal scale.”
“One other open question is that we’re making an argument based on the spread of things that could happen, the arrangement of algorithms in some evolutionary space. We’ve included some simple models of different selective pressures in our simulations, but the general interplay between the spread of algorithms in space and particular selective pressures for individual cases is still very much open for exploration.”
How will you follow up on your findings?
“There were a handful of authors on that paper and I think we’re all taking new avenues of research. Ard Louis, one of the co-authors, is doing lots of work, including an extension to more organismal scale models and using algorithmic models of development to explore if we can generalize this to an organismal form.”
“So the number of mistakes I make when I copy my genome is a function of the enzymes that do that copying, and they themselves are a part of my phenotype, so they can themselves evolve. And so, how should I evolve my evolutionary process in that sense? Can I evolve the process to be optimal?”
“I’ve been looking at how we can use the ideas from the paper to make claims about evolution itself. So if we want an evolutionary process itself to be optimal, and good at discovering new forms, while also good at retaining useful information that it’s discovered in the past, what properties should the evolutionary process itself have? I’m quite interested in this because evolution is driven by mutation, but mutation itself is evolvable. So the number of mistakes I make when I copy my genome is a function of the enzymes that do that copying, and they themselves are a part of my phenotype, so they can themselves evolve. And so, how should I evolve my evolutionary process in that sense? Can I evolve the process to be optimal?”
“My other collaborators are taking different perspectives in exploring this link from algorithmic information theory a bit more deeply, looking at what quantitative predictions can be made using this branch of computer science, about the algorithmic picture of evolution that we’re trying to paint.”
How does this research relate to the origin of life and the emergence of complexity?
“Well, I have to preface this entire answer by saying this is all highly speculative but something that I think it’s quite interesting. We would at the very, very early, almost protocellular level of life, have expected simple structures to have dominated. Evolution – whatever it looked like then – wouldn’t yet have had a chance to explore much of its available algorithmic space, which would itself be very small. So we’d see structures that, if they involved interactions between subunits at all, would likely be very simple single subunits repeated over and over, perhaps almost like a modern virus capsid.”
“Then as evolution proceeded, our theory might suggest that we proceeded exploring outwards in the space of algorithms, typically finding the simplest possible structural solution to a given problem. So if our new form of life had to colonize a new niche, it would probably adopt the simplest possible set of molecular structures that allowed it to do so.”
“It’s extremely speculative, but I’d love to make stronger connections between the exploration of algorithmic space and the observed diversity of structures in the fossil record.”
“It’s perhaps a picture of complexity evolving just as much as it has to at each stage. We normally favor simple structures. I’m quite interested in the Cambrian explosion of forms and their subsequent disappearance. Here’s something like a “just so story”. We start with very simple structures. As life is starting to profilerate, it can colonize different niches, and it’s got a lot of resources available. Perhaps it can now explore weirder structures, as the number of individuals exploring our space of algorithms is increasing, and no-one’s facing much survival pressure. So life is just increasing and exploring the algorithmic space a lot more. Then after building all these weird and wonderful forms, after expanding into that space, resources start to be limited and then some pressures kick in. It starts to be those things which are simple, requiring fewer resources to build, requiring less high-fidelity maintenance of information, that survive. And so we lose the weird and wacky forms and just retain on those which are more simple and more symmetric and perhaps more evolvable. It’s extremely speculative, but I’d love to make stronger connections between the exploration of algorithmic space and the observed diversity of structures in the fossil record.”
Image produced by Iain Johnston using custom code and R libraries ggplot2 and ggtree.
What are any potential applications of your research in medicine or biotechnology?
“This is more of a basic biology paper and project, but there are some implications in the rational design of biological structures. Quite a lot of our early self-assembly work was at least tangentially motivated by nanotechnology. So how can we learn the interactions necessary to build or self-assemble a given mini-robot or similar out of subunits? The evolutionary algorithms that we’re using in this biological model can also be used to discover simple sets of those interaction rules that you’d need to self-assembly a structure in an engineering sense.”
“So how can we learn the interactions necessary to build or self-assemble a given mini-robot or similar out of subunits?”
“We’re saying it would make engineering sense to build a given structure based on as few repeated subunits as possible, not just for efficiency of encoding, but our project implies that they are actually more robust and they could be more evolvable as well. If we proceed to the case where we want to somehow create structures that can themselves evolve, by making a link with algorithmic information theory we could potentially say, this structure is more or less resistant to mutations, it can absorb a given amount of mutational damage before getting compromised, and so on. Although we’re fundamentally thinking about the basic biological process of evolution in this research – how can we borrow ideas from information theory to explain the beautiful emergence of structures across life?”
Iain Johnston. Photo: Øystein Rygg Haanæs/UiB
