This is the second in a series of guest posts by Arlin Stoltzfus on the role of Mutation as a dispositional factor in evolution. The first post was: Reactionary fringe meets mutation-biased adaptation: Introduction.
by Arlin Stoltzfus
Reactionary fringe meets mutation-biased adaptation
Introduction
1. The empirical case
2. Some objections addressed
As noted in the intro to this series, the appearance of an opinion piece on how mutation bias affects adaptation in Trends in Ecology and Evolution (TREE) appears to be a milestone for understanding this interesting new topic. The authors misrepresent a position on dual causation stated consistently in the literature for 20 years, ignore or misinterpret important new empirical work, and blithely repeat a self-serving view of history that my colleagues and I spent years debunking with careful scholarship.
In other words, as a colleague once said, "This is what victory looks like-- everyone butchering your ideas."
But it gets better. Scientists often invoke the cliche that new truths are first ridiculed as impossible, then opposed as unlikely, then claimed as traditional. QuoteInvestigator has a piece on this "stages of truth" meme, with the following from Dr. J. Marion Sims, 1868:
For it is ever so with any great truth. It must first be opposed, then ridiculed, after a while accepted, and then comes the time to prove that it is not new, and that the credit of it belongs to some one else.The idea that the course of evolution reflects internal biases in variation-- or, stated differently, that the generation of variation plays a dispositional role in evolution-- has been (1) ridiculed as an appeal to "the vagueness of inherent tendencies, vital urges, or cosmic goals, without known mechanism" (Simpson, 1967), and (2) ruled out by the famous "opposing pressures" argument from Fisher and Haldane that we will address later.
So, it was exciting that the TREE authors, who represent the reactionary fringe of evolutionary biology-- dedicating to shifting the "Synthesis" goal-posts to maintain the illusion that nothing is new--, not only butcher the idea of mutation-biased adaptation, try to minimize its importance, and misrepresent the evidence: they also want to claim it! Yes, they want to appropriate this unoriginal, unimportant, unsubstantiated idea for the legacy of Ronald Fisher, the closeted mutationist!
Seriously, you could not make this stuff up!
To dig out from under this mess will take some time. Let's begin by reviewing the evidence that the changes that occur during adaptation are enriched for mutationally likely changes.
The case for mutation-biased adaptation
First, consider 4 analyses of experimental evolution.
In the first stage of the compound study by Sackman, et al. (2017), Rokyta, et al. (2005) measured fitness for the beneficial changes-- 9 of them-- implicated in 20 replicate episodes of adaptation of phiX174, aka ID11. Because the fittest variant was found only once, yet the 4th most-fit was found 6 times, the authors explored a model of mutational effects, including transition bias and the multiplicity of mutational paths to an alternative amino acid (e.g., an ATG Met codon can mutate to Ile in 3 different ways, but an Ile codon such as ATC has only one mutational path to Met). An origin-fixation model with mutation bias fit the data better than the mutational landscape model of Orr, which considers only fixation probability. Sackman, et al. (2017) carried out this same 20-replicate protocol with 3 closely related phages: the combined results how a strong bias favoring transitions, 29:5 for paths, 74:6 for events (figure below).
Note the terminology. A "path" specifies the starting and ending genotype, e.g., "gene F, site 3665, C → T", and an event is an occurrence of change along a path-- in this case, a replicate culture in which a phage genome changes. Events along the same paths are parallel events.
MacLean, et al. (2010) repeatedly evolved Rifampicin-resistant Pseudomonas aeruginosa, with results showing a significant correlation between the chance of evolving and the measured mutation rate for 11 changes in rpoB (center panel below). All of these changes are nucleotide substitutions with mutation rates that differ due to unexamined context effects. The lack of correlation in the left panel is not too surprising, given that the calculated probability of fixation ranges only from 0.47 to 0.84, because s is so large (meanwhile, the mutation rate varies 50-fold).
Couce, et al. (2015) evolved cefotaxime resistance repeatedly in 2 different Escherichia coli mutator strains with distinctly different mutation spectra, resulting in two distinct distributions of changes among resistant strains, each with a strong correspondence to the respective parental mutation spectrum.
McCandlish and Stoltzfus (2017) gathered a large set of published cases of laboratory parallel adaptation due to recurrent amino acid replacements (389 events on 63 paths), and found that these data exhibit a substantial excess of transitions, relative to the null expectation for no mutation bias (a 1:2 ratio). Note that this study integrates data from MacLean, et al. (2010) and Rokyta, et al. (2005), but (1) this is a minority of the data (22 %), and (2) the test for transition bias is an independent result from the correlation shown earlier in data from MacLean, et al. (2010).
Next, consider 4 analyses of natural evolution. McCandlish and Stoltzfus (2017), in the same paper just mentioned, gathered data on natural parallelisms from 10 different study systems (231 events on 55 paths), and showed the same kind of transition bias. The summary table below shows that they draw from some famous adaptive stories in molecular evolution, including spectral tuning, resistance to cardiac glycosides (e.g., bird vs. monarch vs. milkweed), foregut fermentation, and echolocation.
The table above represents natural cases from Stoltzfus and McCandlish (2017). Cases 1, 4, 8 and 10 represent recent local adaptation of sub-populations (total 11:10 paths, 69:48 events), while the others represent species divergence (17:17 paths, 63:51 events).
Payne, et al. (2019) examined effects of transition bias in two different curated databases of causative mutations in antibiotic-resistant isolates of Mycobacterium tuberculosis, finding an excess of transition mutations. For instance, they take advantage of the unusual case of Met-to-Ile replacements, which can take place by 1 transition (ATG to ATA) or 2 different transversions (ATG to ATT or ATC). Instead of this 1:2 ratio of possibilities, they see a ratio of 88:49 (Basel dataset) or 96:39 (Manson dataset), roughly 4-fold above null expectations. Because all the replacements are the same type, the bias can not be due to selection preferring some replacements over others.
Storz, et al. (2019) examined changes in hemoglobins using 35 phylogenetically independent comparisons of low- and high-altitude bird populations. They identified adaptive changes in 20 comparisons, implicating 10 different paths and 22 events. (See Table below: Asterisks indicate CpG mutations.) The observation of 6 paths and 10 events associated with CpG mutations was about 6-fold over the null expectation, a statistically significant excess.
Finally, in a completely different type of study, Liu, et al. (2019) explore the emergence of imatinib resistance in leukemia patients, combining clinical data on the frequency of various resistant mutants (of the BCR-ABL oncogene) across 4 continents over 17 years, with laboratory characterization of engineered mutants. In a model for clinical frequency based on drug resistance (measured) and mutation biases (inferred from comparative data), they found that both factors were important, but the mutational factor was more important.
I mention Liu, et al. (2019) to draw attention to a fascinating, biomedically important study. However, I won't include it in future discussions about biological significance, because typically we do not include resistant tumor outgrowths in the category of natural adaptation (and I don't want to confuse people or invite distracting criticisms).
To summarize, various recent studies suggest that the changes that occur during adaptation are enriched for the kinds of changes that are mutationally likely. The spectrum of adaptive changes shows modest biases with the same orientation and magnitude as modest mutation biases (either known or suspected) that range in magnitude from a few-fold (transition bias) to 10-fold (CpG bias) to as large as 50-fold (the range of mutation rates measured by MacLean, et al., 2010).
Going further
Considered on their own, these results are perhaps uninteresting. Mutation biases are secretly influencing the details underlying adaptation. Perhaps this was not expected on classical grounds, but why should we care?
These results are important because they demonstrate a principle not previously accepted: modest quantitative biases in the generation of variation may impose predictable biases on evolution, without a requirement for absolute constraints, neutral evolution or high mutation rates, contradicting the classic logic of the opposing-pressures argument.
If this new principle is general, it would apply to other kinds of mutational biases, as well as to other types of biases, e.g., developmental biases induced by the structure of a genotype-phenotype map. That is, these results provide proof-of-principle for ideas long discussed in evo-devo. As argued by Stoltzfus (2019), the same results add plausibility to key ideas in the self-organization literature, regarding what Cowperthwaite and Ancel (2007) call "the large-scale patterns of mutational connectivity within genotype spaces."
Before exploring these implications in future posts, we need to take a more critical look at the evidence. The authors at TREE claim that nothing has been shown, and that the results are more likely to be due to selection. What is the status of this alternative hypothesis?
