E squared – The Evolution of Evolution
For example, imagine trying to “evolve” a software program to alphabetize lists of words. Say you have a program that does the job tolerably well – it works, but it’s slow and inefficient. To simulate mutation, we change one letter of the program at a time and we ask “better or worse?” If the program now works better, we keep the change; otherwise we keep the original. If we do this long enough, can we make a better and better computer program?
It may not surprise you to hear that for all standard computer languages, this procedure won’t work at all. So we might try to enhance the workability of the model by simulating sex. Imagine breaking apart and recombining pieces from a number of very similar programs, all of which can alphabetize a word list. But this is not a practical way to create a better computer program, either. Even if this whole evolutionary process is realized in software that runs at many gigaflops, the program could go on for many times the life of the Universe without ever creating a better algorithm.*
Computer languages do not constitute an evolvable system. Living systems, on the other hand, have evolvability. How lucky for us!
We might not be satisfied attributing the evolvability of life to “luck”. Perhaps at the dawn of life, a lot of proto-living systems began in many different forms, but it was only a few that happened to be evolvable, and those are the ones that survived. In other words, evolvability evolved. But the truth is larger than this and far stranger. The evolution of evolvability has been an ongoing process, interwoven with the “normal” evolution of fitness, and continuing all through the history of life.
We know this because there are traits that are obviously highly-evolved, but they offer no selective advantage whatsoever, in the traditional sense of survival and reproduction – their only advantages are in the long-range prospects for adaptive change. How did evolvability traits manage to evolve, without ever offering a selective advantage to the individual carrying that trait, but only to its great, great grandchildren?
The genome is organized like a bureacracy, with command-and-control genes at the top and implementation genes underneath. In the 1990s, it was discovered that a single gene could be inserted into a fruit fly’s DNA that would cause the ectopic appearance of an entire eye or a wing or a leg on a part of the body where it does not belong. The term invented for this was hox genes and they perform a function similar to calling a subroutine in a computer algorithm, or a homeowner hiring a contractor to work on his house, or a general issuing an order down the chain of command.
How did the genome come to be organized hierarchically? This feature offers no advantage in fitness for the individual. It does, however, contribute to the rate of increase of fitness over evolutionary time.
The advantage of such a system is not that it makes it easier for the body to construct an eye or a leg – it doesn’t. The advantage is that it permits evolutionary experimentation. Using HOX genes, the placement of limbs or organs can be optimized in an evolutionary trial-and-error process. Without having to re-invent the eye or the kidney each time, different body parts can be moved around to create “endless forms most beautiful and wonderful” that Darwin described. As a way to design any particular organ for one animal, it is a very inefficient way to go; but as a system that can flexibly experiment with legs or wings or eyes or kidneys, hox genes are a brilliant invention. Did I say “invention”? Of course, they’re not an invention at all – merely a product of evolution. But this is a kind of evolution that expands on the traditional “survival of the fittest”. The idea that ‘evolution = blind variation + natural selection’ has become untenable.
How does evolution manage to give the impression of being “smart”? There is a chicken-and-egg problem here. You need an evolvable system to get started with evolution. You need a highly evolvable system in order to select for evolvability. So evolvability is a property that is needed in order to create itself. Think “bootstrapping”.
Besides hierarchical organization of the genome, there are additional ways in which life is optimized for evolution. The most obvious and prominent is sex, to which we’ll return presently (gives me something to look forward to). Some places in the DNA are thousands of times more likely to mutate than others, and these hot spots always correspond to opportunities for experimentation. Meanwhile, genes that control the core metabolism common to all life are tucked away safely beyond the reach of mutation
Genes are not coded into the DNA as contiguous segments**, but are spread out over smaller units (“transposable elements“) that have to be cut and spliced together to make each single protein. This is a complex and inefficient process, adding time and energy and potential for errors. The benefit is that this system promotes evolvability, because functional segments of protein can be cut and spliced in new ways to try out new possibilities without having to evolve them from scratch.
The maintenance of diversity is a major ingredient in evolvability, and it is predominantly appropriate and useful diversity that persists. How does this come about?
Darwin and the Sources of Variability
Through Darwin’s career, the missing piece in his theory, the mystery that he recognized but never resolved was the maintenance of diversity. Natural selection cannot work in a uniform population. It requires diversity as a kind of raw material, which it “consumes” as the less-fit are selected out.
Variability is governed by many unknown laws, more especially by that of correlation of growth. Something may be attributed to the direct action of the conditions of life. Something must be attributed to use and disuse. The final result is thus rendered infinitely complex…These facts seem to be very perplexing, for they seem to show that this kind of variability is independent of the conditions of life. (Origin of Species, First Edition, 1859)
A partial response to this mystery came with Mendel’s understanding of genetics and the mechanism of sexual inheritance. But it remains true in the 21st Century that when we estimate the rate at which selection collapses diversity and the rate at which useful new diversity is generated by mutation and recombination, we cannot escape concluding that the gain in diversity ought to fail by many orders of magnitude to keep up with its loss. 150 years after Darwin, we still fail to account for the maintenance of diversity in nature.
Evolvability and Sex
The vast majority of species shares genes between consenting adults, mixing and matching in a never-ending quest for new combinations. Bacteria are promiscuous, floating their genes out into the environment in the form of “plasmids“, and constantly pick up new genes, without regard to their origin. Single-celled protists swap genes through a process of “conjugation”, actually merging and re-shuffling their genetic identities. This is sex without reproduction, in which two individuals come together and scramble their genomes. The two individuals that emerge from the process are re-shuffled combinations of the two original cells.
Almost all multi-celled organisms include some kind of sexual reproduction. And yet, sex is not at all adaptive in the traditional sense. For individual fitness, sex is a disaster. If we cloned ourselves instead of requiring male + female to reproduce, we could be (at least) twice as fit. The most efficient way to reproduce is simple cloning, and if the most successful individuals reproduced (rapidly!) via cloning, the entire population would, within a few generations, consist in copies of this one type alone. The advantage of sex comprises only a contribution to evolvability.
By chance, I was witness to the dawn of evolvability theory.
In 1980 I was a grad student and teaching assistant, working for physics Prof David Layzer of the Harvard Astrophysical Observatory. Layzer is a broadly-cultured man, a musician and a scholar of many sciences. The course that I taught with him that year was offered to non-science majors, tying together ideas about the behavior of collections of similar objects, from molecules in a gas to animals in a population to galaxies in the Cosmos. That same year, Layzer wrote a paper entitled Genetic Variation and Progressive Evolution, which he succeeded in getting published in the high-profile journal, American Naturalist. Layzer was writing for biologists, while thinking like a physicist. Suppose there were a gene, he mused, that offered no fitness advantage whatever, but which promoted the gradual increase in fitness of offspring and offsprings’ offspring over evolutionary time. Could such a gene be selected in a Darwinian process. “Yes”, was what Layzer concluded, and proferred a mathematical proof.
Layzer’s paper and the ideas within it were roundly ignored, both because he was ahead of his time and because Layzer didn’t speak the language of biologists. It was not until sixteen years later that a Yale biologist and an AI expert from Hawaii paired up to describe the same ideas in language that a biologist might appreciate. They were not aware of Layzer’s precedent, and arrived at their ideas completely independently. This seminal paper of Gunter Wagner and Lee Altenberg put evolvability on the map, and sparked a revolution in evolutionary thinking. Well, perhaps I overstate the situation; though the paper has been widely cited and the issue recognized, these ideas have yet to affect the foundations of evolutionary theory in a way that logically must follow.
There can be no doubt that without evolvability adaptations, evolvability could never have evolved. In other words, evolvability promotes itself in a positive feedback loop, or bootstrapping process. The further evolution of evolvability progresses, the more rapid is further progress in evolvability [sic].
This idea gives us greater respect for evolution, the foundation and basis for life. Evolution is not a simple process that is bound to happen, beginning whenever some chemical happens to catalyze its own synthesis and proceeding inexorably onward and upward from there. Evolution as we know it has required this further action of exponentially increasing its own effectiveness, a process that modern evolutionary science can barely describe, let alone understand.
Evolvability and Group Selection
Most evolutionary biologists strain at the gnat of ‘group selection’ but they swallow whole the camel of evolvability. What I mean by this is that multi-level selection theory (MLS) is well-grounded in traditional evolutionary theory, and requires only a modest theoretical step beyond kin selection. For historic and cultural reasons going back to the 1960s, many evolutionary biologists categorically dismiss the body of MLS research, insisting that the “selfish gene” is a one-size-fits-all explanation for all evolutionary processes.
Evolvability, in contrast, is an irriducibly radical concept. It requires group selection on a vast scale that dwarfs MLS accounts. Evolution of evolvability is a story of how evolution came to be smart, or at least give the illusion of being smart.
A simple yet controversial idea from MLS is that local geography ties together fate of a local animal community, which can be described as having a collective fitness, and which experiences Darwinian selection as a unit. But evolution of evolvability (E2) goes far beyond this, requiring that selection work on entire lineages that last over many generations required for significant evolution to take place. Somehow, during all that time, the fittest individuals don’t manage to crowd out those that are collectively good evolvers, though much less fit (by the traditional definition)
Evolvability and Aging
You’ll have to wait until next week.
* There’s a science devoted to evolving computer programs in this way, and it is calledgenetic algorithms. The process can work when the rules are carefully defined to make sure that pieces of different programs must fit together in a way that makes logical sense.
**in higher life, but not bacteria
This post previously appered in Josh’s blog here: http://joshmitteldorf.scienceblog.com/2013/07/16/e-squared-the-evolution-of-evolution/