More Recent Comments

Friday, June 28, 2019

Reactionary fringe meets mutation-biased adaptation. 2. Some objections addressed.

This is the third in a series of guest posts by Arlin Stoltzfus on the role of mutation as a dispositional factor in evolution.



Reactionary fringe meets mutation-biased adaptation. 2. Some objections addressed.
by Arlin Stoltzfus

In the previous post Part 1, we reviewed evidence from 8 analyses suggesting that modest several-fold biases in mutation may impose modest several-fold biases on the spectrum of changes involved in adaptation, including some legendary cases of natural adaptation.

Reactionary fringe meets mutation-biased adaptation
Introduction
1. The empirical case
2. Some objections addressed
3. The causes and consequences of biases in the introduction process
4. What makes this new?
5. Beyond the "Synthesis" debate
    -Thinking about theories
    -Modern Synthesis of 1959
    -How history is distorted
    -Taking neo-Darwinism
      seriously

    -Synthesis apologetics
6. What "limits" adaptation?
7. Going forward
Is the evidence strong enough already to conclude in favor of a bold new idea? The authors of the hatchet piece at TREE believe that nothing has been shown, arguing that the proposed effect is theoretically unlikely and is probably due to selection.

The focus of this post is on alternative hypotheses (theoretical arguments will be addressed later). For the sake of brevity, I will address just 2 of the many spurious objections offered by these authors in their quest to exemplify the Dunning-Kruger effect. For instance, they write "we stress that parallel genetic change underlying phenotypic convergence is not sufficient evidence for mutation bias being important in causing such convergence."

This is an inversion of the argument, common in the parallelism literature (see Bailey, et al. 2015), that the recurrence of exactly the same change is by itself evidence of selection.

In fact, the case for mutation-biased adaptation does not depend on such weak inferences. In the 8 analyses we reviewed, no change is designated as adaptive solely based on a pattern of recurrence. Instead, each mutational path has either (1) a genetic association with fitness or resistance, or (2) an experimentally verified molecular effect consistent with the adaptive story. Once adaptive changes have been identified, statistical tests are applied to detect an excess of changes of the mutationally favored class.

As another example, TREE's hatchet piece refers to selection as an independent force of adaptation, then attacks the strawman theory of mutation bias as an independent force of adaptation. To ensure that the reader is deceived about mutation-biased adaptation, and ill disposed toward this line of research, this strawman is repeated 5 times on the first page (figure).


Both arguments illustrate how reactionary minds fail to grasp new ideas, and see only perversions or inversions of cherished old ideas.

Now, let us set aside strawman arguments, to focus on genuine alternatives.

For instance, the authors suggest that transitions could be favored "owing to selection on genomic base composition," citing work on GC content. This hypothesis can not work. If the effect of selection is to conserve GC content, this can not explain a bias toward transitions, because the universe of GC-conserving mutations has a transition:transversion ratio of 0. Likewise, if the effect of selection is to change GC content, this can not explain the observed degree of bias, because the universe of GC-changing amino acid replacement mutations has roughly a 1:1 transition:transversion ratio, not large enough to explain results of Payne, et al. (2019) or Stoltzfus and McCandlish (2017).

A more plausible alternative raised by the authors, following Stoltzfus and Norris (2016), is that the observed evolutionary bias could be caused by a bias in protein-level fitness effects that happens to align with the mutation bias, e.g., they suggest that "selectively beneficial transitions and selectively beneficial transversions could also have different distributions of fitness effects."

Let us consider, for the 8 analyses addressed previously, the hypothesis that observed evolutionary biases are not due to mutation bias at all, but to a cryptic fitness bias that happens to align with the mutation bias.

First, in the studies by MacLean, et al. (2010), Sackman, et al. (2017) and Liu, et al. (2019), the authors measure fitness (or resistance). The data from MacLean, et al. (2010) reveal no correlation of mutation rate with fitness (figure).


In their model of effects in drug-resistant tumors, Liu, et al. (2019) find that the mutational factor (estimated mutation rate) explains more variance than the fitness-related factor (measured drug resistance). Results of one-step adaptation from Sackman, et al. (2017) are shown in the figure (left: transitions are in light gray, transversions are in dark gray; upper scale is selection coefficient, lower scale is number of evolved lineages out of 20). Here the mean selection coefficients for transitions and transversions are 0.37 (CI 0.053) and 0.40 (CI 0.18), respectively, i.e., transversions are insignificantly better (data from their Table 1).

Next, consider the experimental study by Couce, et al. (2015) shown in the figure below (courtesy of Alex Couce). Among resistant mutants in PBP3, the resistant mutT isolates (blue) overwhelmingly have the kind of mutations favored by mutT (left box), and the resistant mutH isolates (red) overwhelmingly have the kind of mutations favored by mutH (center box; other types of mutations are in the right box, which includes most of the black isolates indicating a wild-type parent).


The only way to explain this as a fitness effect would be to argue that (1) the mutT and mutH genotypes have widespread, strong, and utterly distinct epistatic effects on the fitness landscape for PBP3, i.e., each mut genotype induces a distinct set of beneficial alleles, and (2) the corresponding mutations for those alleles just happen to be (overwhelmingly) the same type of mutation favored by the mutator.  This is wildly implausible because it implies that the blue-red segregation of columns in the figure above is accidental.

What about the meta-analyses of transition-transversion bias? Could there be a fitness advantage of transitions that explains this effect?

Stoltzfus and Norris (2016) analyzed data on 544 transitions and 695 transversions with experimentally measured fitness effects. Comparing various binary predictors, they considered the chance that a nominally conservative mutation is more fit than a nominally radical one, aka the AUC, which ranges from 0 to 1, with a null expectation of 0.5. Transition-transversion class is a weak predictor (AUC = 0.53, figure), out-performed by most biochemical factors, all 200 of which are out-performed by a conservative-radical distinction based on Tang's U (AUC = 0.64), an empirical measure of relative fixation probability computed from a large set of sequence alignments. Yet, the conservative-radical distinction from Tang's U corresponds to a mere 2.7-fold fixation bias in evolution. Using this relationship, Stoltzfus and Norris (2016) estimate that the transition:transversion distinction corresponds to a 1.3-fold fixation bias, with a confidence interval from 1.0 (no effect) to 1.6.

But these results use the entire distribution of mutations, including the worst ones that (in nature) would be removed by selection. Therefore, Stoltzfus and Norris (2016) truncated the data to see if a stronger benefit would emerge among benign mutations. Instead of getting stronger, the effect disappeared (their Fig. 1).

Next, Stoltzfus and Norris (2016) set aside the above data, and looked at an independent set of data from 4 studies of laboratory adaptation implicating 111 beneficial mutations with measured fitness effects. In the table below, the AUC value in the penultimate column is the chance that a transition is ranked higher than a randomly chosen transversion: the values are all < 0.5. That is, beneficial transitions rank slightly lower than beneficial transversions. The later study by Sackman, et al. (2017) (above) represents a 5th independent case in which beneficial transitions rank slightly lower than beneficial transversions.

Thus, available data, reflecting multiple lines of evidence, indicate that transitions simply do not have a fitness advantage that could explain a several-fold effect on amino acid changes in evolution.

Finally, note that Payne, et al. (2019) report evolutionary biases that cannot be explained by protein-level selection, including transition bias in non-coding changes, and the excess of Met-to-Ile transitions over Met-to-Ile transversions (which are twice as likely without mutation bias).

To summarize, in our evaluation of the cryptic-fitness-difference hypothesis, we find that: in 3 cases, the fitness effects were measured, with results that do not support the hypothesis; in 3 cases (counting 2 meta-analyses in Stoltzfus and McCandlish, 2017), the evidence indicates that the mutationally favored class (transitions) does not have a sufficient fitness advantage; in 1 case, the hypothesis is wildly implausible (Couce, et al., 2015); and in 1 remaining case, Storz, et al. (2019) invoke a mutational effect without any clear justification for assuming an absence of differential fitness effects.

Concluding thoughts


In recent years, systematic data have begun to accumulate on molecular changes implicated in phenotypic adaptation. The pattern emerging from these data is that the molecular changes implicated in adaptation are enriched for the kinds of changes that are favored by mutation, and this enrichment cannot be explained by a cryptic fitness bias that happens to align with the mutation bias.

We could treat this merely as a pattern, as a new and useful empirical generalization.

But there is much more to the story. Mutation-biased adaptation was predicted under a theory that contrasts sharply with classical thinking, which holds that internal tendencies of variation cannot cause evolutionary trends or biases, because mutation rates are too small: in order for mutation biases to be important, mutation rates must be very large, or the opposing pressure of selection must be absent, i.e., effects of biases in ordinary mutations will be limited to neutral evolution.

Yampolsky and Stoltzfus (2001) argued that this view, which derives from the mutation-selection balance model of Fisher and Haldane, assumes that evolution can be treated as a short-term process of shifting the frequencies of pre-existing alleles, without considering the (potentially biased) introduction of new alleles. Using a simple model, they showed that the efficacy of biases in introduction does not require absolute constraints, neutral evolution, or high mutation rates. They argued that this conclusion applies to developmental biases as well as mutation biases.

Thus, it is time to understand this theory, what it implies, and why it differs from classical thinking-- the topic of the next post in the series.


Bailey SF, Blanquart F, Bataillon T, Kassen R. (2017). What drives parallel evolution?: How population size and mutational variation contribute to repeated evolution. Bioessays 39:1-9.[doi.org/10.1002/bies.201600176]

Couce A., RodrÃ-guez-Rojas A., and Blázquez J. (2015) Bypass of genetic constraints during mutator evolution to antibiotic resistance. Proc. Biol. Sci. Apr 7;282(1804):20142698 [doi: 10.1098/rspb.2014.2698]

Liu, C., Leighow, S., Inam, H., Zhao, B., and Pritchard, J.R. (2019) Exploiting the 'survival of the likeliest' to enable evolution-guided drug design. bioRxiv 557645; [doi: 10.1101/557645

MacLean R.C., Perron G.G., and Gardner A. (2010) Diminishing returns from beneficial mutations and pervasive epistasis shape the fitness landscape for rifampicin resistance in Pseudomonas aeruginosa. Genetics 186: 1345-1354. [doi: 10.1534/genetics.110.123083]

Payne J.L., Menardo F., Trauner A., Borrell S., Gygli S.M., Loiseau C., et al. (2019). Transition bias influences the evolution of antibiotic resistance in Mycobacterium tuberculosis. PLoS Biol 17(5): e3000265. [doi: 10.1371/journal.pbio.3000265]

Sackman, A.M., McGee, L.W., Morrison, A.J., Pierce, J., Anisman, J., Hamilton, H., Sanderbeck, S., Newman, C., and Rokyta, D.R. (2017) Mutation-Driven Parallel Evolution during Viral Adaptation. Mol. Biol. Evol. 34:3243-3253. [doi: 10.1093/molbev/msx257]

Stoltzfus, A. and McCandlish, D.M. (2017) Mutational Biases Influence Parallel Adaptation. Molecular Biology and Evolution 34:2163–2172, [doi: 10.1093/molbev/msx180]

Stoltzfus A, Norris RW. (2016). On the Causes of Evolutionary Transition:Transversion Bias. Mol Biol Evol 33:595-602. [doi.org/10.1093/molbev/msv274]

Storz J.F., Natarajan C., Signore A.V., Witt C.C., McCandlish D.M. and Stoltzfus A. (2019) The role of mutation bias in adaptive molecular evolution: insights from convergent changes in protein function. Phil. Trans. R. Soc. B [doi: 10.1098/rstb.2018.0238]

No comments :

Post a Comment