Optimal designs, rugged fitness landscapes and the Texas sharpshooter fallacy

Conversation
1

Very often, when ID proponents argue that the appearance of a specific new complex functional biological structure is out of reach of the RV + NS process on the basis of its improbability, evolutionists resort to the Texas sharpshooter fallacy to defeat the argument, meaning that for them the improbability is delusional for they posit that many other alternatives biological structures potentially exist that could have performed the same function. However, although the Texas sharpshooter argument may seem convincing for undefined situations, there is one situation where it is powerless, it is when complex optimal solutions are associated with rugged fitness landscapes such as vertebrate limbs. As Miller said in the piece below, the only plausible explanation for the perfection of design observed in vertebrate limbs is that a mind engineered them, for only a mind can choose highly optimized solutions out of a sea of possibilities.

2

You are taking something that you see as perfect, and then making that the target. That’s where “Texas sharpshooter” arises. You are supposed to decide on a target first, not afterward.

I’ll remind you that natural selection can get to “highly optimized solutions”.

5 Likes
3

I avoid reading the DI’s misinformation, and I wish everyone did the same but I felt some professional obligation to check this out since I like to think about fitness landscapes and evolutionary explorations, and because I once wrote a series (at BioLogos) about tetrapod limbs. Besides, it’s always possible that a DI person will write something interesting. Right?

This piece starts as a nice (if simplistic) explanation of the basic concept of a fitness landscape. In a bit of bitter irony, the author uses some of Bjørn Østman’s work (animations of populations on fitness landscapes) — Bjørn used to write at Panda’s Thumb and ran the Carnival of Evolution. The overview is mostly fine; the biggest weakness IMO is the failure to understand that actual fitness landscapes are essentially always multidimensional.

The author then works on suggesting, falsely, that rugged fitness landscapes are “greatly limiting” for “evolutionary change.” (The truth is that this is an open question of strong current interest.) He uses selective citation to suggest that this is some kind of big problem in the field, and even claims that the “problem” is worse at the molecular level. He either has not read, or has elected to omit, recent very interesting work about how rugged fitness landscapes are navigated. A widely discussed paper from Andreas Wagner’s group, published in Science last year, asks the very question that our DI agent wants us to consider, then… oops… answers it. The paper is “A rugged yet easily navigable fitness landscape.”

Here’s the Editor’s summary and structured abstract:

Editor’s summary
How many mutations does it take to move from one genetic fitness peak to another in a fitness landscape? Papkou et al. performed mutagenesis to survey the combinatorial genotypic space of nine nucleotides encoding three successive amino acids in a protein targeted by antibiotics in Escherichia coli. The authors found that most genotypes had low fitness, but that traveling between high fitness peaks required surprisingly few mutations. This work represents an exhaustive examination of more than 260,000 genotypes, surveying a nearly complete network of mutational paths to answer a long-standing question. —Corinne Simonti

Structured Abstract
INTRODUCTION
The fitness landscape is a foundational concept in evolutionary biology that has also served to study complex optimization problems in multiple other disciplines. It is an analog to a physical landscape in which a location corresponds to a genotype, and the elevation at that location corresponds to the fitness of an organism with this genotype. Darwinian evolution can be viewed as an exploration of such a landscape by evolving organisms, in which the highest peaks correspond to the best-adapted organisms. When Sewall Wright coined the landscape concept in 1932, he was concerned that biological fitness landscapes may have an astronomical number of peaks, most of which may have low fitness. In such landscapes, evolving populations are likely to become trapped on low fitness peaks from which natural selection cannot help them escape. For almost 80 years after Wright’s discovery, virtually all work on landscapes remained theoretical, and even though experimental landscape studies are becoming more frequent now, we still do not know whether rugged landscapes impair adaptive evolution.
RATIONALE
To tackle this fundamental question experimentally, we created a large biological fitness landscape (>260,000 mutants) by CRISPR-Cas9 gene editing of the key Escherichia coli metabolic gene folA, which encodes dihydrofolate reductase. We mapped the fitness landscape of this enzyme by exhaustively mutating nine nucleotides at three amino acid positions that can confer resistance to the clinical antibiotic trimethoprim. We passaged sixfold replicated mutant libraries of all folA variants in an antibiotic-containing environment and used deep sequencing to obtain fitness estimates for nearly 99.7% of all sequence variants. Our nearly combinatorially complete data allowed us to determine the ruggedness of this high-dimensional landscape. We identified its fitness peaks, their basins of attraction, and evolutionarily accessible paths to these peaks. To find out whether landscape ruggedness impairs adaptive evolution, we simulated the evolutionary dynamics on this landscape under various population genetics scenarios.
RESULTS
We found that the landscape is highly rugged. It has 514 fitness peaks, most of which have low fitness. Nonetheless, the landscape has multiple properties of a smooth landscape. These include an abundance of monotonically fitness-increasing paths to high fitness peaks, large basins of attraction of these peaks, and easy reachability of these peaks by >75% of evolving populations. Furthermore, most evolving populations can access multiple high fitness peaks. All 74 high fitness peaks effectively share one enormous basin of attraction (104,496 variants). This leads to low predictability of evolution on the molecular level because each population can take multiple alternative paths that lead to different high fitness peaks. High fitness peaks remain accessible under various evolutionary dynamics on the landscape.
CONCLUSION
Our work shows that adaptive evolution on realistic high-dimensional and rugged fitness landscapes may be easier than commonly thought. Our finding calls for new and improved theory to understand the counterintuitive geometry of realistic high-dimensional fitness landscapes.

Our DI man either has not read, or omitted mention of, several other recent empirical examinations of fitness landscapes, rugged and otherwise. One really cool one is in Cell Systems earlier this year: Rugged fitness landscapes minimize promiscuity in the evolution of transcriptional repressors. Another is a 2020 eLife paper, Predictable properties of fitness landscapes induced by adaptational tradeoffs, from a group that our man quotemines in his tale, pulling a single irrelevant quote from a 2014 review article.

After stuggling through his misleading tale of the horrors of rugged fitness landscapes, he ends with the predictable and unsupported assertion that “the vast number of suboptimal local peaks in the fitness landscape precludes any possibility of an evolutionary search ever discovering the perfection of design consistently seen in vertebrates and in other taxa.” He offers no data or analysis to support this, for a very good reason. You can guess.

The DI article is typical dreck, and @Giltil has written a typically misleading and ridiculous post about it. Neither is worth your time.

8 Likes
4

Sorry but nowhere in that article does Brian Miller show how many pathways through which the vertebrate limb (or indeed any actual biological adaptation) had available to it when it evolved. The whole article is one big mere assertion (as is your claim) and nothing more.

There is no math, no mutation rates, no population sizes, no selection coefficients, no number of mutational steps, no calculations or simulations that show, given known facts, that the vertebrate limb couldn’t have evolved.

5 Likes
5

Meanwhile here’s a nice introduction to what is wrong with creationist obsessions with the so-called “Waiting time problem”:

1 Like
6

Miller’s opinion is overruled by the evidence of vertebrate limb homology that would be unnecessary under design (but necessary under evolution) and by the known examples of evolutionary algorithms that have the ability to “choose highly optimized solutions out of a sea of possibilities”.

He is demonstrably wrong.

3 Likes
7

So here’s a question. Is the fitness landscape for vertebrate limbs actually that rugged? Doesn’t the flexibility of the three-hinge design argue against that? And how do you tell if the limbs considered are truly optimal rather than just near-optimal?

6 Likes
8

That’s an interesting assertion, but why should anyone believe it?

2 Likes
9

Or that there are many alternative pathways to whatever the most optimal solution is. Like (2) here:

waiting time
waiting time478×592 4.5 KB

So even if there is just one single optimal solution (there doesn’t have to be just one way genetically to produce the single best solution to the shape of a limb, say, that’s a baseless assumption, just look at how for example the torpedo shapes of fish and aquatic mammals have distinct genetic solutions), there could still be numerous different pathways to any one of them.

In reality the situation is of course going to be a mix of situation (1) and (2), with each distinct adaptation having numerous pathways through which it could have evolved (this is what they found in Starr et al. 2017).

Incidentally you are also mistaken to think a rugged fitness landscape means there can’t be numerous pathways to whatever an optimal solution is.

4 Likes
10

Point out that this an example of the Texas sharpshooter fallacy, because it simply is.

There’s no “Texas sharpshooter argument,” Gil.

But we know that’s not true, as evolutionary algorithms routinely do so. One example of this is evolution itself in antibody production, which happens in only 2 weeks. There’s no mind doing any choosing AFAIK.

And I’ll add that as a vertebrate myself, Many nonoptimal aspects of the design of my limbs are painfully (not a metaphor) obvious.

1 Like
11

Apply this to whale flippers. The bones structure appears to owe its arrangement to descent rather than being a clean sheet design as a hydrofoil. Even as a retrodiction, I have never seen any optimization based argument for the flipper design outside of some “common blueprint” vaguery which assumes that creation sprang from a limited palette of imagination.

Next question, how would you distinguish “perfection of design” from “good enough” adaptations to environment?

And although wings on bats and birds work well enough, but why should designing them with wings to fly require sacrificing arms? In general, design in nature looks more constrained by phylogeny than purpose driven innovations. Human technology does much better, with additions and leaps that are not dependent on prior approaches to problem solving.

And in any event, phenotypical traits of proportion and size generally exhibit existing variation within a population, which can trend by drift and selection. Different proteins are not required. Your friends vary in both proportion and size. Island species are usually different from mainland species. What is the supposed barrier that is unbridgeable by RV + NS? Give me something real.

4 Likes
12

This:

Or that there are many alternative pathways to whatever the most optimal solution is. Like (2) here:

waiting time
waiting time478×592 4.5 KB

So even if there is just one single optimal solution (there doesn’t have to be just one way genetically to produce the single best solution to the shape of a limb, say, that’s a baseless assumption, just look at how for example the torpedo shapes of fish and aquatic mammals have distinct genetic solutions), there could still be numerous different pathways to any one of them.

In reality the situation is of course going to be a mix of situation (1) and (2), with each distinct adaptation having numerous pathways through which it could have evolved (this is what they found in Starr et al. 2017).

Let’s take a familiar two case studies. Which of these is the optimal solution for flight, and why? Or, alternatively, perhaps two or all three are optimal, but only optimal for their respective situations. If so, why?
image

We can do the same with this example.

image
image1000×500 46.7 KB

The evolutionary explanation for the differences between structures that perform the same function points to historical contingency (see this paper for a good review). For example, whales (like dolphins) have this particular internal anatomy of their flippers because they are descendants of terrestrial tetrapods, specifically mammals. The limb is just modified to fit for the purpose of swimming. Penguins are also descended from terrestrial tetrapods, so they also have the familiar humerus, radius, ulna, carples, digits. However, they are from a different lineage of tetrapods, the birds. Hence why the internal anatomy of penguin flippers are that of bird wings. Sharks didn’t descend from tetrapods, which is why their internal anatomy is very different from that of the former two. The details of their anatomy are NOT necessary for the function. These are contingent on their respective ancestral history.

‘Cdesign Proponentsists’ really… and I do mean REALLY… despise historical contingency. To them it implies that the current state of affairs are “merely” due to chance or accidents (while related, contingency is actually NOT synonymous with chance, see the aforementioned review paper). They don’t like that because their view is deeply entrenched in teleology. Everything is the way it is due to a designer that orchestrates every detail with complete foresight and a specific goal in mind. This interaction between Behe and Thornton really exemplifies that. Thornton’s reaction to Behe’s comment:

Thanks for asking for my reaction to Behe’s post on our recent paper in Nature. His interpretation of our work is incorrect. He confuses “contingent” or “unlikely” with “impossible.” He ignores the key role of genetic drift in evolution. And he erroneously concludes that because the probability is low that some specific biological form will evolve, it must be impossible for ANY form to evolve. Behe contends that our findings support his argument that adaptations requiring more than one mutation cannot evolve by Darwinian processes. The many errors in Behe’s Edge of Evolution — the book in which he makes this argument — have been discussed in numerous publications. In his posts about our paper, Behe’s first error is to ignore the fact that adaptive combinations of mutations can and do evolve by pathways involving neutral intermediates. Behe says that if it takes more than one mutation to produce even a crude version of the new protein function, then selection cannot drive acquisition of the adaptive combination.

6 Likes
13

The process of somatic mutations involved in finding optimal solutions for antibodies is most likely associated with fitness landscapes that have properties of smooth landscapes.

14

From the referenced article:

The referenced paper actually does say that. I notice that the author was funded by the Biologic Institute, so it’s all pretty incestuous.

2 Likes
15

And yet no conflicts of interest declared… that seems like an academic faux pas.

16

“Have properties of” is so vague so as to be meaningless. You’re just blathering.

All measured fitness landscapes have “properties of” both smoothness and ruggedness. It isn’t either-or, it comes in degrees.

5 minutes of googling results in hundreds of papers showing that epistasis is rampant in the process of affinity maturation, which is to be expected since this is a general feature of most interactions between proteins and other organic molecules.

Here’s a few, found in less than 2 minutes, on SARS-Cov2 (Hey it’s Tyler Starr, Joseph Thornton etc. again):
Starr, T. N. et al. Deep mutational scanning of SARS-CoV-2 receptor binding domain reveals constraints on folding and ACE2 binding. Cell 182, 1295–1310.e20 (2020).

Starr, T. N. et al. Shifting mutational constraints in the SARS-CoV-2 receptor-binding domain during viral evolution. bioRxiv https://doi.org/10.1101/2022.02.24.481899 (2022).

Moulana A, Dupic T, Phillips AM, et al. Compensatory epistasis maintains ACE2 affinity in SARS-CoV-2 Omicron BA.1. Nat Commun . 2022;13(1):7011. Published 2022 Nov 16. doi:10.1038/s41467-022-34506-z

Of course, in reality the fitness landscape for antibody-antigen binding is better described as fluid/switching, or a never-ending arms-race, where any local optimum is eventually transformed into another valley when the pathogen mutates again and the immune system has to play catch-up.

2 Likes
17

I think you just made that up without doing any math. Far more likely is that the ruggedness of fitness landscapes differ by orders of magnitude for different antigens. Yet mutation+selection finds a way.

So on a similar subject, why did you write,

Why did you omit their very explicit, quantitative qualification of that, Gil?

1 Like
18

Is it incestuous when organizations devoted to naturalistic evolution fund works that are friendly to naturalistic evolution ?

19

Are there any that do so?

20

Miller’s pitch to ID readership

His article demonstrates that the similarities between vertebrate limbs is best explained not by common ancestry but by intelligent design. He explains how the general vertebrate limb layout (aka architecture or plan) is the optimal design pattern for complex motion in diverse environments. He also demonstrates that six specific limb designs in five different vertebrate taxa (i.e., groups in animal hierarchy) are the best possible designs for the animals’ environment and behaviors.

From Burgess source paper - gets through peer review (supposedly double blind)

The great versatility of the vertebrate limb pattern challenges the limb homology argument that the skeletal layouts of the whale flipper and bird wing are not what would be expected for those applications and make sense only when seen to be a consequence of evolutionary inheritance. This paper argues that the vertebrate limb pattern is so versatile that it is actually highly optimal not just for arms and legs but also for flippers and wings. All the musculoskeletal structures of flippers and wings are actually fully functional and fully explainable in terms of optimal design.

All bold mine.

It is subtle, but Miller’s review emphasis or cast is on design and ancestry as mutually exclusive, which Burgess is careful not to make explicit.

2 Likes