Ancestral beta-lactamase enzyme

So I spent a bit of time reading through the currently pinned conversation on Doug Axe’s work with beta-lactamase. It seems the consensus opposing Gauger is that beta-lactamase enzymatic activity has a much lower probability barrier than the modern beta-lactamase enzyme. So I was wondering, do critics of Gauger and Axe mean that the modern beta-lactamase enzyme evolved from an ancient simpler beta-lactamase enzyme similar to the one discussed there based on an antibody? If so, what is your proposed evolutionary pathway for this evolution?

1 Like

I that is not what we mean. I don’t think any one thinks beta-lactamase evolved from a mammalian antibody. The phage-antibody system, instead, is an artificial system designed to sample sequence space in an undirected way (without selection), directly testing their hypothesis.

The point is, rather, that beta-lactamase activity is not nearly as rare in sequence space as Doug Axe’s (and Ann Gauger’s) analysis purports to demonstrate. This seems to directly demonstrate false both their conclusions about rarity of enzyme activity in sequence space and the experimental methods and theoretical analysis they used to arrive at this conclusion.


Not this critic.


Their problem is that they find an enzyme atop an adaptive peak and assume 1) that the peak has no flanks, only the peak itself and 2) that there are no other peaks elsewhere in the fitness space.


Yes, that part is correct. The general idea here is that the function, the capacity to break down beta-lactams, is much more frequent than the particular implementation of it we find in the specific enzyme(TEM-1 Beta-lactamase) analysed by Axe.

No definitely not from antibodies. It might be the case that the superfamily of all beta-lactamases originated in that way(from a simpler protein, not from antibodies), but that simply isn’t known at this time. Afaicg the members of the superfamily are so diverged in sequence that phylogenies are inferred on the basis of structures instead, and only within particular classes (such as class A) is it possible to do sequence-based phylogenies.

Incidentally, evidence indicates that TEM-1, a class A beta-lactamase, shares common ancestry with the rest of the superfamily of beta-lactamases. See:
Hall BG, Barlow M. Evolution of the serine beta-lactamases: past, present and future. Drug Resist Updat . 2004;7(2):111-123. doi:10.1016/j.drup.2004.02.003 (pdf here)

1 Like

Off the top of my head, I would suggest that serine- and metallo- beta lactamases probably do not share a single common ancestral protein.

Yes they are not included as part of the serine beta-lactamase superfamily.

In other words, two totally unrelated sequences and folds that catalyze the same reaction.

In this light, I don’t see why ID proponents have such a problem with the catalytic antibodies.


Hey of course someone has done this work. Two articles on ancestor reconstruction of ancient beta-lactamases:


Axe studied a version of the enzyme that was weakly functional and at the very edge of stability. This is not at a fitness peak but rather at the base of one fitness peak.

Do you propose that ancient enzymatic function exists on a different fitness peak than the modern beta-lactamase? If so, how is the ancient enzyme relevant to the evolution of the modern one?

Beta-lactamase activity was used in the study to detect the presence or absence of a stable fold. The probability of the fold in the sequence space was the purpose of the study, not enzymatic activity. This should be obvious since Axe did not mutate the active site.

So both of these studies are done on proteins with the right fold. Oddly, one touts high stability as key while the other touts high “conformational flexibility”.

Are you suggesting that these reconstructed proteins have the same likilihood of forming as the ab enzymes being discussed? I don’t see how. Both these enzymes have the modern fold. That suggests to me that Axe’s study has more relevance to their probability than the ab screens. Has anyone suggested a mutational pathway between the ancestral and the modern which can be tested?

@BenKissling, I disagree. If Axe was interested in folds, then he would have assayed, not just beta-lactamase activity, but also peptidase activity. At the very least. This is because the fold he was studying is known to be associated with more than just beta-lactamase activity.


No. I propose that the Texas sharpshooter fallacy is not a valid form of reasoning. You can’t determine the probability of an enzyme with a given function evolving by considering only the sequences that currently exist.


Yes, structural stability with respect to temperature tolerance.

No, this is actually not correct. While there is some relationship between protein fitness and protein temperature stability(the degree of stability contributes something to protein fitness), you can’t directly transform one into the other.
Another factor is protein activity(how much does it’s rate of substrate turnover translate into organismal fitness?), and another still is it’s folding efficiency(out of all of them that are synthesized, how many of these misfoldso much they become nonfunctional?).

None of these factors or relationships are simple, and one measure can’t be translated directly into the other.

The fitness landscape metaphor kind of breaks down for promiscous enzymes capable of catalyzing multiple different chemical reactions. There aren’t really “hills” or “peaks” in that classic sense. The landscape is hyperdimensional. It basically grows by length and amino acid alphabet size. Every position in a protein has 19 dimensions attached to it, one for each possible replacement. And each of those 20 total variants for each position has some fitness-score associated with it that depends on what the organism needs.

So what could be a highly fit enzyme with respect to one particular reaction, could be much lower for another. A landscape like this is better pictured as a sort of network, with each particular sequence represented as a node, connecting to hundreds, or thousands, of other nodes. The “shape” of the landscape depends entirely on the number and types of activities you look at, and has to be be cast in the context of the needs of the organism.


Those two are not in contradiction. A protein can be both highly stable, in that it takes a high temperature to make it lose structural integrity such that it becomes nonfunctional, while also being able to perform it’s function while undergoing some degree of structural change. And increased ability to accommodate other substrates with different shapes and sizes, yet retain a similar overall structure.

No. I don’t know what the likelihood of forming any of them is. That obviously depends a lot on what their ancestral states are. That’s the problem with this kind of ad-hoc probability calculations of de novo emergence is.

Forming the sequence AAAAAG by random mutation of AAAAAA is much more likely (here there just needs to be one “correct” mutation), than forming it from GCCTTT. Thus the prior probability of it’s de novo emergence is meaningless to calculate unless you know what sequence gave rise to it. You are committing the Texas Sharpshooter fallacy.

And it might not even have emerged in that kind of spontaneous fashion. The ultimate ancestor could be a sort of fusion protein derived from fragments of other extinct proteins, or homologoues that have long since lost all trace of detectable sequence similarity. Or it could, in principle, have grown from a much smaller peptide that had an entirely different function. We don’t know, and hence these sorts of calculations are meaningless mental masturbation.

Not really. If Axe really wanted to know the probability of this sort of protein fold’s de novo emergence, he should first of all have at least attempted to reconstruct the most ancestral version he could reliably infer, as that sequence would be closer in time, sequence, function, and structure, to the one that actually did originate and subsequently give rise to the superfamily.

And then he shouldn’t have somehow tried to alter it to make it less tolerant to mutation by deliberately making a highly temperature sensitive variant of it, as that directly affects the so-called “pass rate” he uses to calculate how many sequences are able to fold and function. Obviously a much more stable protein is able to tolerate many more substitutions before it loses integrity and stops functioning. If it takes more mutations to destabilize it, it will have a higher pass rate. So Axe is stacking the deck against it with a temperature sensitive variant.

It is noteworthy that even the ancestor of class A is highly thermostable, and quite promiscuous. This suggests there are many, many more possible functional versions than what Axe infers from his highly sensitive variant, that was only tested indirectly for one function. Since these enzymes are capable of acting on a range of substrates, low-activity versions for one might still be useful, high-fitness variants for another. He can’t just pretend this reality is irrelevant.

That’s pretty much what they do in the 2nd reference:
You can see the data in the supplementary information.

They reconstruct four consecutive ancestral nodes in the phylogenetic tree of class A enzymes(see figure 2) and show each of these four “steps” in this evolutionary history are functional (and that they are functional on a range of different substrates) despite being different from their extant TEM-1 descendant by 93 out of 262 total amino acid positions.

Key question: If phylogenies are just fantasies and this evolutionary history did not actually occur, and if proteins are generally not all that tolerant to mutation, why is it possible to use a phylogenetic algorithm to infer protein sequences that are that different from their modern counterparts, which are still highly functional, and higher temperature stability?


Scissors are rigid with conformational flexibility at the same time. The two properties at the same time are critical to its operation.


Taking the whole system into account, abzymes can almost certainly be reproducibly formed under the experimental conditions of the paper. It did not require sampling that much of sequence space at all.

I would say also that beta lactamase activity (through not necessarily the precise beta lactamase enzyme under discussion) can almost certainly evolve in the right environment, likely on observable time scales. There is a clear set of steps towards evolving the enzyme activity from penicillin binding proteins.

See this paper for an overview:


Yes absolutely. Which highlights the uselessness of focusing on a particular protein fold as opposed to the function performed by it. If the function of beta-lactam hydrolysis, that is the breaking down of the antibiotic molecule, is ultimately the barrier that stands between future extinction and evolving successful resistance, then it is the probability of evolving that function that matters. Not whether it is achieved by some particular protein structure.

1 Like

If Axe is trying to figure out the probability of a beta-lactamase enzyme evolving then the probability of a specific protein fold is not relevant. You need to be looking for activity, not a specific protein fold.

The very fact that we can find beta-lactamases that don’t have that specific fold tells us that Axe’s study can’t be used to determine the probability of such an enzyme evolving. If your math says it should take 10^60 attempts to get a beta-lactamase and it only takes 10^9 in the real world then the math is wrong, not the real world.