I know that, in discussing Michael Behe’s book Darwin Devolves, we have been over this ground many times, but I am going to ask you to bear with me. I want to focus once again on the polar bear story, because it lays bare the motivations of some critics and how far they will go to avoid acknowledging a reasonable point. The whole thing is barely bearable.

There’s a bear in my lab, dear Liza, dear Liza

The core difficulty for some scientists who read Behe’s book is also the key idea at its heart. This is the idea: much of evolutionary change at the organismal (phenotypic) level happens as a result of losing function at the genetic level. This genetic loss can go by many names — “loss of function,” degrading of function, damaging, or deleting function entirely. This loss of function is due to mutation to the gene that disrupts or changes the coding sequence of the gene, and thus disrupts or otherwise interferes with the normal functioning of that gene product. It can also happen by a change in gene regulation that eliminates or greatly reduces gene expression. I am being a bit pedantic here. Why? The concept of getting an adaptive function from genetic breakdown is a counterintuitive thing. How can the loss of a genetic function lead to a new phenotype — a new behavior or appearance or ability — for an organism? That’s the key to understanding Behe’s new book. Loss of function at the genetic level often translates to change at the biochemical level, meaning proteins are changed or missing, or some other cellular function is missing or degraded or misregulated, which can have a surprising, even antithetical effect on phenotype — antithetical, that is, when the underlying biochemistry is not well understood.

To Spin, Not to Hear

I am also being pedantic because some people want not to hear this. They want to spin this any way they can to avoid hearing it. Behe’s thesis goes in the wrong direction. Evolution cannot build complexity if it is leaking functionality faster than it builds. It is also in the nature of things that it is much easier to break than to build. Therefore they do not want to concede that so much of evolution proceeds by damaging mutations.  

Just a note here before I forget: Nowhere does Behe say this is the only way evolution proceeds. He just says loss of function mutation is the cause of many adaptive if not most adaptive changes.

Then catch it, dear Henry, dear Henry

Some accuse Behe of not being up to date on all the marvelous things that evolution does accomplish by positive change, and the new things it does create. Sometimes rearrangements produce new proteins or cause expression in new times and places, and such changes can affect phenotype. But a large portion of mutations that cause evolutionary change are in fact loss of function, or degrading mutations. Behe discusses a number of classic examples in his book, but the one that seems to have caught the most attention from Behe’s critics is APOB and the polar bear mutations.  

A Straight Line

The reason for that, it seems to me, is that the Liu et al., authors of the polar bear study, “Population Genomics Reveal Recent Speciation and Rapid Evolutionary Adaptation in Polar Bears,” draw a straight line from phenotype (the polar bear’s adaptation to a high fat diet) to genotype (the genes proposed  to be responsible for that adaptation) to specific amino acid changes in those genes that may be responsible for that adaptation. They use comparisons between polar bears, closely related brown bears, and panda bears to identify candidate genes where specific mutations have occurred, becoming common in the polar bear population but not the others. These mutations, the authors say, have been selected because they confer some advantage on the polar bear in its arctic environment. They are adaptations, meaning they help the polar bear to survive. Notice, in this statement the authors have gone from genotype straight to adaptation. Notice also, dear reader, that these adaptations occurred under natural conditions, without directed selection or experimental intervention. It wasn’t a carefully designed lab experiment where the end result could be expected to favor whatever grew fastest on a single food source. This was evolution under real conditions, not lab conditions.

The mutations were primary; they were what identified the candidate genes as likely to be important in polar bear adaptation. So the authors used a program called PolyPhen-2 to analyze those mutations, since no one has done laboratory studies on polar bears. (No surprise. I am wondering where all the genomes they did sequence came from!) PolyPhen-2 compares the mutant amino acid’s likely structural and biochemical characteristics  to those of other species, and to those of other mutants. Then they ran the mutants through two programs. Both had been trained  by supervised machine learning, but on different data sets. HumDiv was trained on sequences of known human disease-causing genetic variants along with sequences shared between humans and closely related species, and assumed to be non-damaging. The second, HumVar, includes disease-causing variants plus any common mutations that cause a change in amino acid in their protein, in other words, any nonsynonymous mutations.

With what shall I catch it, dear Liza, dear Liza?

What exactly does PolyPhen-2 examine? Here is more detail, for those who want it.

According their website:

PolyPhen-2 is an automatic tool for prediction of possible impact of an amino acid substitution on the structure and function of a human protein. This prediction is based on a number of features comprising the sequence, phylogenetic and structural information characterizing the substitution.

In other words, as the website further explains, they look at known information of the sequence in its context, whether it is globular, membrane or lipid associated, metal binding, associated with an active site, modified by carbohydrate, coiled, a signal peptide, etc.:

A substitution may occur at a specific site, e.g., active or binding, or in a non-globular, e.g., trans-membrane, region. PolyPhen-2 tries to identify a query protein as an entry in the human proteins subset of UniProtKB/Swiss-Prot database and use the feature table (FT) section of the corresponding entry. PolyPhen-2 checks if the amino acid replacement occurs at a site which is annotated as:

• DISULFID, CROSSLNK bond or
• BINDING, ACT_SITE, LIPID, METAL, SITE, MOD_RES, CARBOHYD, NON_STD site

At this step PolyPhen-2 memorizes all positions which are annotated in the query protein as BINDING, ACT_SITE, LIPID, and METAL. At a later stage if the search for a homologous protein with known 3D structure is successful, it is checked whether the substitution site is in spatial contact with these critical for protein function residues.

PolyPhen-2 also checks if the substitution site is located in the region annotated as:

• TRANSMEM, INTRAMEM, COMPBIAS, REPEAT, COILED, SIGNAL, PROPEP

Then they look at structural information, either from known structures or homologous structures: solvent accessibility, side chain volumes and dihedral angles, and whether there are spatial conflicts. All of these analyses allow PolyPhen-2 to detect potential damage to the   protein’s structure or function. They also compare the sequence being studied (in this case, the polar bear sequence) to either the HumVar or HumDiv libraries to determine if the sequence’s mutations (polar bear mutations) are possibly or probably damaging to the protein’s function.

A Wild Claim

A few comments here. Some have made the wild claim that PolyPhen-2 does not measure the likelihood of biochemical disruption but rather it measures adaptive function. Wow. As the PolyPhen-2 literature shows, this is completely wrong: PolyPhen-2 is designed to detect disruption and does a pretty good job of it. In a study where it was tested against positive and negative controls, PolyPhen-2 had an accuracy of .72, a sensitivity of .8 and a specificity of .7.

The one thing PolyPhen-2 cannot do is predict which amino acid changes might be adaptive. At the protein level, the disruption is going to be … disrupting. Damaging, by all the criteria PolyPhen-2 measures.  The program is cueing off of disease-causing mutations, not beneficial ones. It looks for biochemical and structural problems. Its function is to identify SNPs that are likely to have a damaging effect on structure or chemistry or both.

Darwin Devolves

In his book, Behe used the polar bear analysis as a first demonstration of his idea that a majority of evolutionary adaptations are actually the result of a damaging mutation. A damaged gene makes a damaged protein that in turn through some process makes a new adaptation. So, to focus in on one example, the polar bear has a gene called APOB that is involved in lipid metabolism. APOB is the gene for apolipoprotein B (apoB), the primary lipid binding protein for LDL and other compounds. Its biochemistry is not simple — it has several forms, to be discussed in another post. Having high levels of LDL cholesterol in the blood is associated with heart disease in humans. Polar bears have a very high fat diet and high levels of cholesterol in the blood. Yet they do fine. It would make sense for there to be some adaptive mutations in this gene; nine candidate mutations are present in the polar bear population in higher than normal rates (one out of two bears have them). Five of those nine mutations are in the amino terminal globular region where lipids are bound by the protein. The authors of the polar bear study, Liu et al., chose to put those five APOB mutations through PolyPhen-2 to analyze their predicted effects.

The authors of the paper scored the APOB mutation results as damaging if either the HumDiv or HumVar programs scored them as possibly damaging. These specific results were discussed previously at Evolution News, but I have again pulled their results on APOB straight from their table reporting the data for all the polar bear genes they analyzed.

With a pole and a noose, dear Henry, dear Henry

Please notice. APOB was identified as probably adaptive, because in the population one out of two bears had these mutations. The logic is that because mutation is random, the only reason so many polar bears would have these five mutations was because they were beneficial and gave an advantage to any bear that bore them. In his book Behe points out that these “adaptive” APOB mutations are all predicted by PolyPhen-2 to be damaging to some degree. (And the authors of the polar bear paper, Liu et al., say the same thing in the paper. They agree that PolyPhen-2 identifies them as damaging, though as we will see they still hold to the idea that they are somehow adaptive.) Behe’s thesis in Darwin Devolves is that most adaptive mutations are adaptive because they degrade function. And here we have a case where an adaptation is tied to a gene with specific mutations that are probably damaging to the protein’s function. 

Mitigating the “Damage”

So how do the critics respond? They try to mitigate the “damage” by saying that the program PolyPhen-2 cannot predict damage but only change in function, and thus that to claim that these polar bear mutations are damaging is more than the evidence can “bear.” They may be potentially damaging, but that more work needs to be done to fully demonstrate they are.

This is true as far as it goes. To definitively prove they are damaging would require more work — empirical studies to be precise. But the computer studies conducted thus far give us good reasons to suspect the mutations were damaging. The ironic thing is I have seen comments online saying that programs like PolyPhen-2 are used because they are often right. On the other hand, some have said PolyPhen-2 is trained on human data and polar bears are different. (Their APOB proteins are 79 percent identical and 88 percent similar according to BLAST, so not that different. I am willing to bet the structural alignment is excellent.) PolyPhen-2 reportedly has a good track record, so it seems unlikely PolyPhen-2 would get it wrong five out of five times.

Then I have seen comments saying that the results are untrustworthy. I must say that the people making the last kind of comment probably had Behe’s name attached to that word  untrustworthy.  To toss out a whole methodology that has been used successfully by many, just because you don’t like the results of one study as interpreted by one man, is simply scurrilous. And as Behe himself has said, if a tool is reliable until Behe relies on it, the problem is not with the tool or with Behe. 

You’ve got to be joking, dear Liza, dear Liza

A Critic Equivocates

Yet another critic has said that PolyPhen-2 cannot predict damage but only change in function. This is an equivocation. Anything new, including damage, results in a change in function. However, the program can only infer whether mutated amino acids are neutral or damaging in a given position, based on whether they are likely to be disrupting to their protein’s structure or chemistry, or not. It cannot predict benefit or improvement. In addition, the databases being used were based on known disease-causing mutations. Disruption, damage, or “benign” as they called it (meaning neutral, harmless), not a beneficial change in function, were the only outcomes. “Change in function” thus obscures and diverts from the point, which I think was the point of the “change.”

The only way to know if these mutations enhance or improve the biochemical function of APOB is to test them. And that requires some really scary experiments (picture a tranquilized polar bear female being implanted with an engineered embryo), or developing another animal model. 

With sedatives and gloves, dear Henry, dear Henry

How do the authors explain their results? Liu et al. say:

…We suggest that the shift to a diet consisting predominantly of fatty acids in polar bears induced adaptive changes in APOB, which enabled the species to cope with high fatty acid intake by contributing to the effective clearance of cholesterol from the blood. [Emphasis added.]

So they are proposing adaptive changes to APOB itself that enables “effective clearance” of cholesterol from the blood. No one disagrees with this. But do they conclude this is the result of constructive mutations that enhance the function of APOB? Don’t forget that they previously had said:

Cholesterol levels in blood plasma of polar bears are extreme (e.g., Ormbostad, 2012); in humans, elevated cholesterol levels are a major risk factor for the development of cardiovascular disease (Cannon et al., 2010). It remains an enigma how polar bears are able to deal with such lifelong elevated levels of cholesterol.

So maybe they aren’t so sure after all about how exactly polar bears reduce cholesterol. More will be said about this in a forthcoming post. 

The Best Explanation

Then how to account for the apparent phenotypic adaptive effect of APOB mutations? Mike Behe gave the best explanation I’ve seen, in a post a few weeks ago, which the critics seem to have missed. (Do you remember how I said we needed an alternate animal model to polar bears?)

In 1995 researchers knocked out (destroyed) one of the copies of the APOB gene in a mouse model — the same gene as has been selected in polar bears. Although APOB is itself involved in the larger process of the transport of cholesterol, mice missing one copy of the APOB gene actually had lower plasma cholesterol levels than mice with two copies. (Mice missing both copies died before birth.) What’s more, the researchers noted that “When fed a diet rich in fat and cholesterol, heterozygous mice were protected from diet-induced hypercholesterolemia.”

The researchers admitted they did not know how it all came together — how that effect on the complex cholesterol-transport system resulted from breaking the gene. Nonetheless, there is no ambiguity about the mouse results. Simply by lowering the amount/activity of APOB mice were protected from the effects of a high-fat diet….

Thus there is no good reason to speculate about possible new activities of the coded protein in the polar bear. Rather, the simplest hypothesis is that the mutations in the polar bear lineage that were judged by computer analysis to likely be damaging did indeed blunt the activity of the APOB protein in that species. That molecular loss gave rise to a happy, higher-level phenotypic result — an increased tolerance of polar bears for their high fat diet.

Not Radical, Just Counterintuitive

Look, it’s really not so hard. Behe is proposing a hypothesis: that mutations judged to be damaging by PolyPhen-2 really were damaging. It’s not radical. It’s just counterintuitive to think that damage leads to a benefit. However, as Behe reports in his book, a lot of evolutionary change happens as a result of genetic loss or damage. The result can be adaptive, in part because of the complexity of biochemical and physiological interactions of fatty acid metabolism. There are many pieces to the fatty acid metabolism puzzle, and tugging on the one called APOB can apparently cause a helpful readjustment to the bear’s ability to cope with high fat.

Is there an easier way dear Liza, dear Liza?

Of course, in an ideal world where polar bears were as docile as my poodle, we could have genetically engineered versions of APOB inserted into their genomes and then screen them for expression of the modified gene, and any effect on phenotype. Or we could develop cell culture assays as an approximation. Or use mouse models, and take the results seriously. But in vivo studies in the organism in question are the gold standard. BUT…until that happens, scientists use the evidence and arguments available to them. As did Liu et al. As does Behe. The next step in genuine  scientific debate ought to be to test the hypothesis about the five mutations. Not to disparage, or denigrate, or deny. That accomplishes nothing, reveals nothing, and does not increase understanding. The work is begging to be done. My hope is someone will, or has already begun.

Photo credit: Eva Blue via Unsplash.