In 2000 and 2004, protein scientist Douglas Axe published papers in the Journal of Molecular Biology suggesting that amino acid sequences that yield stably folding proteins may be as uncommon as 1 in 10⁷⁴ sequences. Figuring prominently in Stephen Meyer’s arguments in Darwin’s Doubt, Axe’s results are significant for the debate over Darwinian evolution and intelligent design. That is because they suggest that random mutation and natural selection would be unable to find the rare amino acid sequences that yield functional proteins. Instead, his results indicate that functional proteins are extremely rich in complex and specified information, a property that in our experience points to design.

Axe’s research has survived many attempts to refute it (for a couple of examples, see here and here). However we have been receiving inquiries about a new paper in Proceedings of the National Academy of Sciences USA by Jeffrey Skolnick and Mu Gao, “Interplay of physics and evolution in the likely origin of protein biochemical function.” Critics claim the paper shows that proteins can easily bind small molecules, and therefore that it’s not at all unlikely chance mutations could easily find amino acid sequences that generate functional proteins.

The first few paragraphs in a Science Daily article explained what the paper is about:

A new study of both computer-created and natural proteins suggests that the number of unique pockets — sites where small molecule pharmaceutical compounds can bind to proteins — is surprisingly small, meaning drug side effects may be impossible to avoid. The study also found that the fundamental biochemical processes needed for life could have been enabled by the simple physics of protein folding.

Studying a set of artificial proteins and comparing them to natural proteins, researchers at the Georgia Institute of Technology have concluded that there may be no more than about 500 unique protein pocket configurations that serve as binding sites for small molecule ligands. Therefore, the likelihood that a molecule intended for one protein target will also bind with an unintended target is significant, said Jeffrey Skolnick, a professor in the School of Biology at Georgia Tech.

Of course their conclusion was immediately controversial. As one science news article about the paper reported, “other researchers working on protein structure believe there are many more than 500 unique sites among the human proteome.”

But there’s more to say that is of relevance.

Theoretical, Not Experimental
In the original Star Wars movie, Han Solo famously said “Good against remotes is one thing. Good against the living? That’s something else.” He was talking about winning a fight with laser blasters — contrasting an artificial setting with real life — but the principle applies to molecular biology as well.

Skolnick and Gao studied “a set of artificial proteins” — meaning their research was computational and theoretical, rather than experimental. Essentially, they developed a way of classifying protein pockets, which they’ve applied to a set of make-believe proteins. As my biologist friends are quick to remind me, the best way to make a convincing claim to have discovered a biological fact is to go to the lab, and do solid laboratory work to back it up.

Thus, any claims that Skolnick and Gao have refuted a core argument for ID are based upon citing non-experimental, non-laboratory work, which is theoretical in nature. Is this sufficient to refute Axe’s empirically based research? Definitely not. Even if Skolnick and Gao’s conclusions are granted (for the sake of argument), they would not impinge upon Axe’s conclusions.

Assume a Functional Protein Fold
Skolnick and Gao’s basic argument is that functional proteins are very common in sequence space — close enough to evolve from one to another through random mutation. It is well recognized that sequence space is populated with islands of functionality where some proteins are somewhat clustered (say, for example, all proteins that share a particular fold), but those islands have steep cliffs that make it very difficult to move from one island to the next through Darwinian processes. We know that these islands of functionality are isolated because of the experimental research of Douglas Axe.

Axe studied the likelihood of an amino acid sequence generating a stably folded protein. But Skolnick and Gao studied the behavior of proteins that already are capable of producing stable folds. So even if their non-experimental work turns out to be correct, it’s not on-point for refuting Axe. They have not demonstrated that random mutation, even coupled with selection, can generate new stable, functional protein folds — which Stephen Meyer in Darwin’s Doubt calls “the smallest unit of structural innovation in the history of life.” This means that if Darwinian evolution is going to explain the origin of new proteins, it’s crucial to explain the origin of new protein folds, though Skolnick and Gao don’t address that point.

In his new book, Meyer titles Chapter 11 “Assume a Gene” because evolutionary biologists often try to explain the origin of new genes by assuming the existence of pre-existing, functional genes. In a similar way, Skolnick and Gao assume a set of folding proteins. Thus their work can tell us nothing about Axe’s conclusions on how functional protein folds originated in the first place.

Troubles in Traversing the Island
Now a critic might claim instead that in the new paper, Skolnick and Gao provide a test of the co-option model, showing that an already-functional protein could easily undergo mutations to acquire a finely tuned, highly active binding function. In other words, such a critic might argue that once you land on an island of functionality (e.g., a functional protein fold), Skolnick and Gao show that you can easily evolve from one protein to another so long as you remain on the “island.” Even this claim, however, is contradicted by experimental results from some of Axe’s other work.

In the past, we’ve discussed a 2011 paper by Axe and Ann Gauger, “The Evolutionary Accessibility of New Enzymes Functions: A Case Study from the Biotin Pathway,” which tried to experimentally convert one enzyme (Kbl₂) into another very similar enzyme, thought to be closely related (BioF₂). After trying multiple combinations of different mutations, they found:

We infer from the mutants examined that successful functional conversion would in this case require seven or more nucleotide substitutions.

This presents a serious problem for Darwinian evolution since a 2010 paper by Axe found that a feature which would require more than six neutral mutations before providing an advantage could not arise in the entire history of the earth. Axe and Gauger (2011) thus concluded:

[E]volutionary innovations requiring that many changes would be extraordinarily rare, becoming probable only on timescales much longer than the age of life on earth. Considering that Kbl₂ and BioF₂ are judged to be close homologs by the usual similarity measures, this result and others like it challenge the conventional practice of inferring from similarity alone that transitions to new functions occurred by Darwinian evolution.

Let’s return now to Skolnick and Gao’s 2013 paper. Even if their paper is correct, and even if it is true that functional binding pockets are very common in sequence space, they have not demonstrated that transitioning from one enzyme function to another can be accomplished by a stepwise evolutionary pathway — even when the enzymes in question are similar. Indeed, as we just saw, Axe and Gauger (2011) found that proteins can be very close to one another in sequence and structure, but moving from one function to another might still require more simultaneous mutations than could arise through a Darwinian evolutionary process over the entire history of the earth. Given that Axe (2010) found that only a small number of neutral and/or deleterious mutations can arise before one outstrips the probabilistic resources available to Darwinian selection, it seems that the evolutionary bridgability of these binding pockets has not been demonstrated.

In other words, Skolnick and Gao’s paper doesn’t even try to show you can hop from one island of functionality to the next (a major problem raised by Doug Axe’s research), but it also doesn’t show you can easily traverse an island — i.e., move from different protein to protein within those islands (assuming you can find an island in the first place).

Binding a Ligand is Quite Different from Binding an Enzyme
Another weakness in the applicability of Skolnick and Gao is that their paper merely sought to test binding affinities for ligands rather than other types of common protein functions, such as enzyme catalysis. Ligand binding is much simpler, as it only involves interacting with a very small molecule, whereas enzyme catalysis requires interacting with much larger molecules with much more complex shapes, and would be far more difficult to evolve. Thus Skolnick and Gao only address perhaps the simplest case of protein action upon a substrate.

More Artificial Tests: Protein Promiscuity
There’s a final, more fundamental problem facing Skolnick and Gao’s paper. They claim to support the “inherent functionality model” of proteins, where irregularities in the surfaces of proteins readily create pockets that can bind lots of other small molecules. In their view, protein may “promiscuously” bind with many different small molecules, and selection can then fine-tune those interactions that provide a functional advantage. But does this theory play out in the real world?

Proteins are composed of pieces of secondary structure that bind together to form tertiary structures with irregular surfaces. If it were really the case that these surface irregularities, bumps and crevices, had a reasonable chance of doing what enzyme active sites do, then it should be easy for the functions of deleted genes to be filled in by these accidental, promiscuous activities. But when this is tested we find functional compensation by promiscuous activities to be rare, and in no case has it been shown to result from surface irregularities outside of active sites. Furthermore, an attempt to optimize one of these promiscuous functions produced nothing remotely comparable to a natural enzyme. Axe and Gauger (2011) explain:

Although it is possible to compensate for poor functional conversion to OSBS by over-expressing the converted genes, this reduces the evolutionary plausibility in two respects. First, the over-expression itself would require particular genetic modifications, making the total complexity of the adaptation greater than the single change to the enzyme. Second, because over-expression involves a significant metabolic cost, adaptive evolution may eliminate over-expressed genes more readily than it tinkers with them. This presents a catch-22 situation for the fate of duplicate genes. If they are strongly expressed they are vulnerable to rapid elimination, but if they are weakly expressed the new function would need to appear with high proficiency in order to have a selective effect.

The promiscuity hypothesis seems to offer a way out of this by positing that small-scale innovations can originate as secondary functions in enzymes that are already highly beneficial because of their primary functions. The primary function guarantees that the gene is preserved and expressed, potentially making it a good platform for secondary functions to ‘hitchhike’ their way to selective success. The obvious difficulty, though, is that efficient performance of the primary function seems to require that hitchhiking be minimized. Indeed, an important study by Patrick et al. shows this promiscuous hitchhiking to be a limited exception rather than a rule. They used 104 auxotrophic E. coli strains, each with a single-gene knockout, and a plasmid library in which all E. coli genes are individually over-expressed to find out how many of the missing gene functions can be filled in by other genes. Functional rescue was found to be possible for 21 of the knockouts, with fifteen of these cases appearing to involve metabolic workarounds of various kinds and only six appearing to involve catalytic promiscuity. This shows that promiscuous activities do exist in modern enzymes, but it also indicates that they are rare. Furthermore, considering the high expression levels of the rescuing genes and the poor growth of the rescued strains, it is again unclear whether the activities demonstrated are of evolutionary significance. An attempt at evolutionary optimization of one of these activities fell ten-million-fold short of wild-type proficiency, suggesting that they may actually be evolutionary dead ends. (internal citations omitted)

Conclusion
Some people think Skolnick and Gao (2013) refutes the implications of ID research on the origin of proteins, but I find that claim very surprising. Why?

• (1) Skolnick and Gao don’t cite any work of ID proponents nor do they claim their paper pertains to it.
• (2) Skolnick and Gao’s work is entirely theoretical and computational, and is thus hard-pressed to impinge upon Axe’s experimental results.
• (3) Many other researchers working on protein structure disagree with their conclusion, and believe “there are many more than 500 unique sites among the human proteome.”
• (4) Skolnick and Gao only studied the behavior of an artificial set of proteins which were assumed to already be capable of folding into a stable structure, and thus their paper doesn’t address Axe’s research which investigated the likelihood of producing stably folded proteins in the first place.
• (5) If Skolnick and Gao are correct that proteins are clustered closely in sequence space, it still doesn’t show that closely related proteins can always evolve from one function to another: Axe and Gauger (2011) experimentally found that even with proteins very close in sequence and structure, moving from one function to another can require more simultaneous mutations than could arise through a Darwinian evolutionary process over the entire history of the earth.
• (6) Skolnick and Gao only studied proteins that bind small molecules like ligands. They did not study the feasibility of evolving much more complex protein shapes required for many common enzyme functions, like enzyme catalysis, involving much larger and more complex molecules.
• (7) Their paper attempts to theoretically bolster the “promiscuity hypothesis” of protein evolution, a hypothesis that has not fared well in experimental tests.

Skolnick and Gao ask, “How can one demonstrate that the ability to engage in a variety of low-level biochemical functions without selection is an intrinsic property of proteins… ?” In other words, what should convince us that the functions of natural proteins are abundantly present in random polypeptide chains? The answer is obvious: Go into the lab and impress us with what you find in a mixture of random polypeptides.

Neither they, nor other critics, have done this.