Learning mutation bias


I have been reading about mutational bias primarily through the work of Arlin Stoltzfus and it’s been a bit difficult to decipher so far. For some reason I cannot find a source that provides a good explanation of how the different biases work and the relevant research. If anyone can recommend a source (a review article, a book, a web site, etc.), I would appreciate it!

In this post though, I will describe what I have taken away from my readings, however, and show you what progress I have made while I haven’t been posting. Basically, mutations ain’t random!

The gist is that mutation matters. According to Stoltzfus, the thinkers behind the Modern Synthesis largely disregarded mutation as an evolutionary “force” or “pressure” – the mutation rate was simply too low to make a dent in allele frequencies like selection, drift, and migration did. Instead, the pool of variants produced by mutation was portrayed as nearly infinite – the trait needed for an adaptation was always at the ready. (This may be a bit extreme but it’s the general sentiment.) (If you want more information on this history, be sure to check out Arlin Stoltzfus’ blog series, The Curious Disconnect, at molevol.org or Sandwalk.)

Now, however, Stoltzfus argues that mutation bias is powerful enough to overpower selection (which I will talk about in a later post). Stoltzfus readily admits that mutational bias is nothing new but the concept hasn’t become popular outside of molecular biology and it has yet to enter our verbal understanding of evolution – something of which I think is crucial to the theory as I discussed before.

So what is mutation bias? The Modern Synthesis held that mutation was random and I don’t blame them – this was even before the discovery of the structure of DNA, but we now know that some mutations are more likely than others. A quick example (among many others) is transition:transversion bias.

A and G are purines, C and T are pyrimidines. A mutation from one purine to another (or pyrimidine to another pyrimidine) is called a transition, whereas a mutation from a purine to a pyrimidine (or vice versa) is called a transversion. Although I do not understand the biochemistry behind all of this, it makes intuitive sense that a transition is more likely than a transversion. In fact, Stoltzfus has calculated the transition:transversion bias to be around 3. This is particularly striking when you take into account that there are twice as many possible transversions than transitions (8 to 4).

Such a strong bias must have some evolutionary effect. My stumbling block was how did this bias matter? Mutation biases such as CpG mutability, CG/AT ratios, etc. certainly control (in some sense) the evolution of DNA sequence of a gene, but could we know how mutation bias had effected the gene’s evolution historically or predict its future? Natural selection “just-so stories” are easy to generate and imagine in our heads – could we do the same for mutation bias?

Well… yes, I think. It wasn’t until I read Stoltzfus’ 1999 paper, On the possibility of constructive neutral evolution, that mutation bias’ effect on the direction of evolution made sense. The paper uses several examples but I will just briefly focus on one.

Stoltzfus proposes that “duplicate gene retention” can be explained by non-adaptive (in this case, mutation) evolution as opposed to selection. When a gene or genome is duplicated, a significant chunk of the duplicates are actually retained (25-50% in bony fish (Stoltzfus 1999). Why? The selection hypothesis claims that because there are two copies of the same gene, one gene is allowed to diverge and evolve a novel function while the other gene retains and expresses the same phenotype and that this is adaptive. However, the hypothesis doesn’t explain how that duplicated gene managed to stay in the first place.

Stolzfus thinks that once the duplicated gene creates an “excess capacity” (2 working genes vs 1), the two genes will actually suffer mutations that reduce productivity because the other gene can pick up the slack. Productivity reduction will occur because “reducing” mutations are more likely to occur than productivity enhancement mutations. Think entropy here – there are more “faulty” combinations than “perfect” combinations. (If I misused entropy in the preceding sentence, please let me know.)

Stoltzfus also claims that these production reductions will stabilize the gene pair and can help explain the retained gene duplicates. There is some empirical evidence for his claims so don’t think this is merely a just-so story as I presented here. I will write a more in-depth analyses later on once I feel comfortable with my understanding – right now I’m just explaining the gist.

Because productivity-reduction mutations are more likely than productivity-enhancement mutations, mutation bias gives evolution a direction. Because mutation bias is probabilistic the direction isn’t determined – it just leans one way as opposed to another. As Stoltzfus opens his 2009 paper, mutation bias can be analogized to climbing a mountain with a slight leftward lean – you’re still ascending but just a little to the left.

The reason I want to discuss mutation bias in general is because I think non-adaptive evolution isn’t given enough credit and I find the work of scientists like Michael Lynch and Arlin Stoltzfus fascinating. I don’t know if these proposed models really apply to how evolution actually works or not; I just know that the models seem to make sense and do provide interesting alternatives to selection-based models. The topic is worth investigating and I intensely believe that we should always question the foundational principles of the arguably most successful scientific theory so far produced. We won’t know we’re wrong until we look.

If anyone has any suggestions on what to read regarding mutation bias (or non-adaptive evolution, in general), please let me know in the comments!

Also if you want to read some more on “constructive neutral evolution,” be sure to check out Psi Wavefunction’s commentary at Skeptic Wonder (Part I and Part II)!
Stoltzfus, A. (1999). On the Possibility of Constructive Neutral Evolution Journal of Molecular Evolution, 49 (2), 169-181 DOI: 10.1007/PL00006540

Stoltzfus A, & Yampolsky LY (2009). Climbing mount probable: mutation as a cause of nonrandomness in evolution. The Journal of heredity, 100 (5), 637-47 PMID: 19625453

17 thoughts on “Learning mutation bias

  1. Here’s a couple classic references for biased mutation (or directed mutation), which may be slightly different than what you’re discussing above:

    Cairns, J., J. Overbaugh, and S. Miller. 1988. The origin of mutants. Nature 335 142-145.

    Cairns, J. and P.L. Foster. 1991. Adaptive reversion of a
    frameshift mutation in Escherichia coli. Genetics 128: 695-701.

    These got a ton of coverage at the time (you could also read some of what Frank Stahl wrote about this debate).

    The idea of “random mutation” that developed as part of the synthesis was about a different sort of randomness than transition/transversion rate differences. I think those involved would have readily agreed that there would be things like this (rate differences, mutational hotspots, etc.) The contention was (and is) that mutation is random with respect to the utility of the mutation. So populations that could really use a mutation to get by are no more likely to get it than those that don’t need it at all.


    • Thank you for the references; I’ll be sure to check those out soon.

      What you say makes sense and clarifies the topic a bit for me (in regards to how mutation is viewed as random). There is a second major part to Stoltzfus’ argument that I didn’t want to discuss until I understood mutation bias a bit more. As far as I understand it, Stoltzfus argues for a revision (or scratching altogether) the Modern Synthesis because he thinks there is a bias in the introduction of alleles. He views the Synthesis as seeing evolution as a fixation process shifting allele frequencies, whereas he views evolution as an origin-fixation process. The Synthesis left out the first part.

      I don’t think Stoltzfus’ argument disagrees with the sentiment of “mutation is random with respect to the utility of the mutation.” I will read more of Stoltzfus’ papers to see how he views mutations as non-random – I believe he sees non-randomness in the introduction of alleles but not necessarily in their utility.

      I apologize for how terribly vague this is and as I have said, I am still trying to understand his argument myself. I will definitely keep your comment in mind though.


  2. Kele–

    Hi. Thanks for writing about these topics. I’d be happy to try to explain anything that isn’t clear about my publications. But I think you are doing very well so far without any help from me.

    You write “For some reason I cannot find a source that provides a good explanation of how the different biases work and the relevant research.”

    I would love to start a dialog or thread on this– ideally a structured dialog that could be used later for didactic purposes. When I talk about these subjects, I spend a lot of time fending off misinterpretations from folks who see mutational explanations as threats and are eager to mischaracterize and dismiss them. So it would be helpful for me to work with folks who are interested in understanding (rather than misunderstanding). Maybe we could come up with some ways to make the concepts clear and unmistakable.



    • Upon quickly rereading through my post, I noticed some statements which slightly misconstrue what your point is, like selection and mutation being directly opposed to each other.

      With that small bit of evidence in hand, I think a dialog of some kind is a fantastic idea for many reasons. What do you have in mind?

      By the way, would you recommend reading JL King’s “The Role of Mutation in Evolution”? Is it still relevant?



  3. I can summarize what’s relevant about King’s paper. Mainly its relevant for historical reasons, to remind us how much the new “molecular” view diverged from the view that Mayr, et al worked so hard to establish.

    In the earlier paper that co-proposed the neutral theory, King and his co-author Thomas Jukes (1969) argued that the observed correlation between a) frequencies of amino acids in proteins and b) numbers of codons assigned to an amino acid in the genetic code was evidence for neutral evolution. In a non-neutral view, they said, “one particular amino acid will be optimal at a given site in a given organism, and it matters little whether there are six possible codons (as there are for serine) or only one (as there is for methionine).”

    However, in his 1971 article, King recanted and said that this was not evidence for neutral evolution, because the same pattern could be seen under a lucky mutant view of adaptation in which “the adaptive mutations which are fixed are those which arise first and are not lost”. If so, then:

    “Suppose that, at a given time, there are several possible amino acid substitutions that might improve a protein, and among these are changes to serine and to methionine; serine, with its six codons, has roughly six times the probability of becoming fixed in evolution. Once this has occurred, the mutation to methionine may no longer be advantageous. Ultimately, the amino acid frequency composition of proteins will reflect the frequencies at which the various amino acids arise by mutation, which in turn depends on the genetic code and DNA nucleotide frequency composition.”

    Thus King said that the correlation was not evidence against “adaptation”, but only evidence against a deterministic model in which selection picks out the optimal genotype from the gene pool. Maynard Smith (1975) picked up on the significance of this, repeating King’s argument and concluding:

    If we accept the selectionist view that most substitutions are selective, we cannot at the same time assume that there is a unique deterministic course for evolution. Instead, we must assume that there are alternative ways in which a protein can evolve, the actual path taken depending on chance events. This seems to be the minimum concession the selectionists will have to make to the neutralists; they may have to concede much more.”

    What’s shocking about this whole episode is how this remarkable “concession” was never acknowledged. What King and Maynard Smith are accepting is a stochastic “lucky mutant” view of evolution, as distinct from the deterministic view in which selection picks the best codon from all the alternative codons in the “gene pool”. King made this distinction very clear in his paper, though he did not refer back to the mutationists.

    The alternative, in 1971, was that, if a gene has codon X at position Y, this is because codon X is the best possible codon for position Y. That is what being a neo-Darwinist meant in 1971. No one actually adopts that view today, although its often hard to tell because the language has not changed. “Selectionists” today write triumphantly as though they had won the debate over molecular evolution. Nothing could be further from the truth.


    • Were Jukes and King predicting codon bias in their 1968 paper?

      So is the view adopted by King and Maynard Smith basically what Rokyta et al. were testing in their adaptive walk experiment? That the most beneficial mutation isn’t necessarily the one that gets fixed?

      This seems like an example of where many biologists accept these ideas but it has yet to enter the verbal conception of evolution. It is weird (but cool) how old these ideas really are. Does someone like Dawkins not agree with these ideas?



    • Arlin: What’s shocking about this whole episode is how this remarkable “concession” was never acknowledged.

      What, exactly, should have been acknowledged and who, exactly, should have acknowledged it? As Kele points out, beginning from roughly this point in time (1975), most theorists came to agree not only that the course of evolution is not deterministic, but also that most substitutions are neutral

      Kele’s readers who have not yet discovered The Curious Disconnect are probably puzzled regarding what is at stake in this dispute.


  4. Perplexed wrote:

    “What, exactly, should have been acknowledged and who, exactly, should have acknowledged it?”

    I don’t know– I’m not interested in what folks “should” do. I said it was remarkable that such a critical change in thinking could happen, yet not be widely known.

    But I would say that, in fact, a profound change in thinking did NOT happen, but only a muddled compromise to the effect that evolution is “not deterministic”, which barely scratches the surface.

    Let’s review. Then we can ask if this is “puzzling”.

    According to King, one way of thinking about why we observe some pattern or feature goes this: there is some set of possible codons at a site, and selection chooses the best one based on the encoded amino acid, regardless of whether it can be encoded by 1, 2, 3, 4 or 6 codons. King says that this way of thinking has a justification according to neo-Darwinian assumptions (the “gene pool” and so on), and indeed, he says its still valid when applied to “phenotypes”.

    The alternative, according to King, is that change just keeps on happening and happening at a rate depending on the mutation rate AND the chance of acceptance, so that the steady-state distribution of amino acids will reflect codon number due to an effect of mutation (more codons in a set, more mutational pathways to that set), even when all the changes represent *selective* (not random) fixations.

    We need a third view to cover mainstream thinking in molecular evolution, which presents mutational explanations as “neutralist” hypotheses, assuming implicitly that mutational effects only happen under the condition of neutral evolution. This view is consistent with the idea that evolution is “not deterministic”. Its consistent with the Modern Synthesis view of mutation as a “weak force” unable to “oppose” selection. But conceptually its a muddle, neither fish nor fowl.

    Is that puzzling?



  5. Thanks for the responses.

    Kele asks: “So is the view adopted by King and Maynard Smith basically what Rokyta et al. were testing…? That the most beneficial mutation isn’t necessarily the one that gets fixed?”

    Yes and no. Clearly I’m not getting my message through. If you’ll pardon me for saying it, my frustration is the sense that folks keep taking what I’m saying, which is rich in implications, and throwing out the interesting bits, and transforming it into (what seem to me to be) rather lame and impotent expressions of “not”: evolution is “not deterministic”.; the most beneficial mutation is “not necessarily the one that gets fixed”; evolution does “not” look like mainstream neo-Darwinism ca. 1959.

    This gives me a depressing “end of science” (John Horgan) feeling that evolutionary thinking has reached its permanent endpoint, with panglossian neo-Darwinism as the reference point for every discussion, so that whatever comes out of my mouth, the message that enters peoples’ ears is limited to “not neo-Darwinism”. It reminds me of the time I presented a very elegant constructive-neutral-evolution model of RNA pan-editing, including some precise quantitative predictions that seemed to be working out right, and an editing expert in the audience raised her hand after the talk and said “So, you’re saying its all just random?” She could have taken the model back to her lab and tested some implications, but the take-home message for her was that I didn’t agree that pan-editing is God’s special gift to trypanosomes.

    What King said, and what I’m saying here, is far richer and full of implications than just saying “not deterministic”. I’m saying that the beneficial mutation that gets fixed tends to be the one favored by mutation. I’m saying that there is a conception of causation that makes sense of these effects, based on recognizing the introduction process, and distinct from the classic “forces” view. I’m saying that evolution *does* look like mutationism, with mutation providing iniative, discontinuity, creativity and direction.

    Kele writes “This seems like an example of where many biologists accept these ideas but it has yet to enter the verbal conception of evolution. It is weird (but cool) how old these ideas really are. Does someone like Dawkins not agree with these ideas?”

    I think “accept” is a strong word in this context, but I’m with you on “weird”.


    • I think I understand your frustration. Evolution not being deterministic, etc. are trivially true. A lot of the problem is that when raised under the Modern Synthesis, it is hard to think outside that paradigm.

      “I’m saying that the beneficial mutation that gets fixed tends to be the one favored by mutation.”

      By “beneficial” do you mean generally beneficial, but not necessarily the most beneficial? That is, in Rokyta et al., the mutations most frequently fixed are ranked 2, 3, 4 (also favored by mutation) – not 1 but not 7 or 8 either.

      I apologize if I am failing to understand this specific point. This is how I understand it generally:

      Not only is there a bias in the introduction of mutations (transition:tranversion rate, 4-fold degeneracy, etc.) but this bias also influences which mutations become fixed. This happens because the gene pool is not infinite – a mutation has to occur first and thus its introduction biases the proceeding direction of evolution.

      Right? Not right?


      • Right. Thanks.

        The Rokyta, et al study is a great example. Let’s apply the three modes of thinking that I outlined above.

        Classic. All single-nucleotide variants of the phage are present in the ‘gene pool’ at non-zero frequencies. “Evolution” is the smooth (continuous) and largely deterministic process of shifting these frequencies to a new optimum. Parallel evolution is evidence of selection, which repeatedly picks out the best alleles.

        Mutationist. Evolutionary change happens by the introduction of a new mutation and its fixation by selection or drift. Parallel evolution happens because the mutation-fixation probabilities are non-uniform, due to mutational effects and fitness effects.

        Muddled 1. Mutation and selection are opposing pressures. Selection usually wins, but mutation can bias evolution if there is no selection.

        Muddled 2. Selection drives adaptation, but mutation imposes “constraints”.

        The classic view fails to explain the results of Rokyta, while the mutationist view succeeds. Muddled view #1, which is common among molecular evolutionists, fails. Muddled view #2 succeeds superficially, but only because (in muddled view #2) “constraint” is a catch-all that can explain anything.


      • I think those are reasonable portrayals of what people think. I totally agree with your assessment of Muddled 2. I have recently taken interest in the idea of “constraints” and parallel evolution and as far as I can tell, constraints as normally conceived are so trivially true to not even be worth mentioning. I am thinking of pursuing that topic for the paper we have to write in developmental biology though.

        What do you think of the idea of the large amount of standing variation found in wild populations? I have heard this idea before and Mayr reiterated it. I know you’re not a fan of the “gene pool” – how do you reconcile these ideas? Do you think the standing variation idea is without merit?


  6. “Selection drives adaptation, but mutation imposes “constraints”

    Funny, I tend to think of it pretty much the exact opposite: selection is a [probabilistically-acting] constraint, mutation, drift and whatever else ‘drive’ change. ‘Drive’ in the sense of a blind drunk driver aimlessly wandering about with no intention of ever actually getting anywhere. Sometimes he drives off a cliff. Selection takes care of the aftermath. Sometimes it doesn’t, and the mangled wreck somehow survives and becomes some lineage of large multicellular eukaryotes, or a kinetoplastid mitochondrial genome.

    This is why I can’t fight the Intelligent Design crowd – the very premise that any of it is intelligent or designed makes me choke up laughing…



  7. Kele to Arlin: What do you think of the idea of the large amount of standing variation found in wild populations? I have heard this idea before and Mayr reiterated it. I know you’re not a fan of the “gene pool” – how do you reconcile these ideas? Do you think the standing variation idea is without merit?

    You may be confusing two different notions of “standing variation” – two notions that are different in their statistical profile and in their theoretical implications.

    The first notion is “widespread polymorphism”. This is the fact, empirically demonstrated by Lewontin and Hubby, that most loci are polymorphic – that is, there are multiple alleles, each measurably common (>5% of population) – in a typical wild population.

    The implication of widespread polymorphism is that neutral mutation must be common – that Kimura must be roughly correct – since that degree of polymorphism cannot be maintained by balancing selection.

    The second notion is the “gene pool” idea that Arlin talks about. This notion claims that most possible alleles at most loci exist in the population, even though they may be very rare. And if they don’t exist in the current generation, they will surely arise anew by mutation within a generation or two. This notion is empirically false – obviously so if by ‘locus’ we mean protein-coding gene.

    The theoretical significance of the gene pool in this sense would be that if it existed, time-to-fixation would not include a significant amount of time waiting for the allele to appear.


  8. The “gene pool” idea was used to rationalize a way of thinking that is inadequate to encompass evolution. 70 years ago, this was a hand-waving argument. Today, we have massive amounts of data on long-term divergence, and these results show that evolution cannot be understood as a deterministic process of shifting frequencies of alleles already in the population, with no dynamic dependence on mutation rates.

    Nevertheless, its important to understand that the “shifting gene frequencies” view of evolution has a basis in reality.

    The reality is that if you take two fly populations, and select for fat flies in one population, and skinny flies in another, you likely will succeed in getting major differences in weight between the fat line and the skinny line, and this will turn out to be due to “shifting gene frequencies” without new mutations. Though this process is not entirely deterministic, it is reproducible in the sense that the same kinds of shifts tend to happen recurrently. For a recent example, see the paper by Burke, et al. that just came out in Nature, regarding selection for rapid development in fruit flies:


    The question is, and always has been, how relevant this kind of experiment is to evolution over the long-term.



  9. Only saw this interesting post. You might find a paper we recently published in PLoS Genetics interesting.
    Nucleotide content in bacteria is extremely variable (some bacterial genomes have over 75% GC content while others have less than 25%). For a long time it has been widely assumed that this is due to differences in mutational biases. In fact, in most papers you read about codon usage or nucleotide content you can find the words nucleotide content, and mutational biases used as synonyms. This would imply that in AT-rich bacteria mutations from GC to AT occur more frequently, while the opposite is true for GC-rich bacteria. In our PLoS Genetics paper, we show that this does not seem to be the case. We focus on bacterial lineages that have been recently evolving under severely relaxed selection, and using large quantities of genomic sequence data demonstrate that mutation appears to be universally AT-biased, even for lineages that are very GC-rich. Our results indicate that mutational biases might be less variable than people have been thinking, and that selection, or a selection like process such as biased gene conversion are is likely involved in determining bacterial nucleotide content.
    The full paper can be found here:


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s