Demolishing the modern synthesis

Back in July of 1999 I posted a “demolition” of the modern synthesis. This led to a lively discussion with various attempts at refuting the demolition and arguments against the attempt. There was no clear resolution and in the end the discussion moved to other matters. I thought it would be of interest to represent the discussion.

This article is organized in several parts. Part I consists of the original demolition. I have taken the liberty on making a few minor corrections, and of inserted some marked, indented paragraphs containing clarifying material. I also have numbered the paragraphs so as to simplify making references. Part II contains arguments against the demolition by various persons. Part III contains counter arguments. Part IV contains an analysis of the demolition, the arguments, and the counter arguments. Finally part V contains my comments on the Pimple Nosed Antarctic Anteater.

Part I: The demolition

(1) There is a fundamental problem with the modern synthesis. Basically, it is broken. The gist of the matter is that there aren’t enough genes.

NOTE: In this article I am talking about the “modern synthesis” put together by people like Fisher, Mayr, and Dobzhansky that formed the basis for genetics well into the late twentieth century. In the past few decades the entire subject has been altered beyond recognition. However the text books are only slowly catching up.

(2) The conflict which the synthesis purportedly resolved was a dispute between the naturalists and the population geneticists as to the nature of heredity as to whether it was soft or hard, i.e., whether variation was continuous or discrete. The compromise was effected by the observation that soft inheritance (continuous variation) is emulated by hard inheritance (discrete variation) if a trait is the cumulative result of a number of genes.

(3) Continuous variation (or emulation thereof) is necessary for the Darwinian model of natural selection which supposes a process of accumulating small advantages. It is, moreover, that which is observed for many traits.

(4) The catch is that there aren’t enough genes. The human genome has approximately 70,000 genes. If genes are to determine traits quasi-continuously it will take 10-20 genes to control one trait which means that the number of traits controlled by the genome is on the order of 5,000 traits OR LESS.

NOTE: The actual number of genes is about 33,000. This makes the effective number of traits even less. However the effective number of genes is greater because of various reading tricks.

NOTE: This was a point of confusion in the discussion. Despite the disclaimer, some assumed that I was arguing for a direct gene to trait mapping. I was not. However the model used in the modern synthesis does implicitly assume presume such a mapping.

NOTE: This is oversimplified in that changes in the base pairs that dictate the folding are relevant. None-the-less the effective number of bits in a gene is far less than the nominal number.

Part II – Arguments against the demolition

The arguments of PZ Myers:

1. . The combinatorial argument. N genes don’t code for just N traits, they can code for 2^N traits. That is, one gene allows for 2 possible cell states, 2 allow for 4 states, 3 allow for 8, etc.
2. . The regulatory argument. What’s critical in defining a cell is its regulatory state — and any one gene may have a large number of regulatory sites. (OK, it’s a variant of #1…)
3. . The development argument. Genes aren’t adequate to specify an organism– there are also significant influences from the cytoplasm and the environment, and each cell has an independent history that influences gene expression.

   What it amounts to is that your argument was a rather more subtle and cleverer- than-usual variation of the creationist demand that we show “fin genes” that get turned into “arm genes”. You were making an unrealistic assumption that there is some kind of simple one-to-one mapping of genes to discrete morphological traits, and there isn’t one. There is no “arm gene”. Similarly, for Bubba, there is no “7,351st Purkinje cell from the midline on the left side of the cerebellum gene”.

There isn’t a need for continuity of genetic determination. The key is of course additive advantages and not continous variation. More to the point is that potential advantages need to be available for selection. That is rather broad and deep with a suspect smell. A lack of determinancy of phenotype from genotype slows selection and at some point leaves drift as the dominant factor. But I don’t see that continuity is required.

Let’s try to start on a real example. What’s involved in the “fight or flight” response. We get (partial list) a release of adrenaline, binding of adrenaline to a receptor, G-protein activation, G-protein inactivation, scavanging of adrenaline and various cascades of 2ndary messages.

Focusing on the receptor which bind adrenaline, there may indeed be only a handfull of amino acids which form the actual binding pocket but the whole rest of the molecule can participate to provide nearly continous variation to binding affinities.

The G-protein which is activated only has a few amino acids which participate in the converison of ATP to cyclic ATP but the rate of that conversion is subltly affected by the rest of the protein. Elsewhere that protein has a delayed fuse that will hydrolyze a GTP to a GDP and then turn off the production of cAMP. That clock can also be tweaked by subtle effects elsewhere which can be independent or interdependent to the rate of cAMP formation. Likewise we have multiple routes to varition in the scavanging of Adrenaline or its release concentration, the concentration of the receptors themselves, their ability to reprime for the next response and on and on.

All of these and more contribute to the difference between acting like a sloth which slowly turns around to see who just ate its hindquarters and a shrew that runs and hide from the sound of a butterfly landing on a bush 10 feet away.

Part III – Counter arguments to the objections

…It is given that there is a messy map from the genotype to the phenotype (and even that the phenotype is a function of the environment as well.) The key is the number of degrees of freedom, i.e., the dimensionality of the two spaces. If there are N dimensions in gene space they can determine at most N dimensions in trait space. This is not changed by the messiness of the mapping….

We are not, repeat not, repeat NOT, talking about a one-to- one mapping of genes to traits. We are talking about the number of variables (on the gene side) and the number of function values (on the trait side). The variables are independent of each other. The function values are not, in general, independent of each other. To give a simple example:

Suppose we have two variables, x and y, and three functions f, g, and h which are given by:

           f = x+y
           g = x - 2*y
           h = 2*x - y
   

We have three functions and only two variables; the dimensionality of the function space is apparently three (three different functions) but in actuality is only two because the three functions are not independent. That is, we can express h as a combination of f and g. The situation is general. If we have N independent variables and M functions of them (M>N) we can have at most N independent functions; the remaining M-N ones can be expressed as combinations of N of them.

This also doesn’t work although the issues are subtler. The problem is that development is not heritable. Consider a parent organism creating an egg. The parent not only passes on a genotype, it also passes on an environment in which the child organism will develop. Fine, this apparently is information that is not in the child’s genotype. Consider, however, what happens when the child in turn creates an egg. It must supply the same developmental environment to its offspring. Now where does that information come from? There are two possibilities. One, which is actually the case, is that the information is encoded in the genome. (I.e., a mother inherits genes from her mother that “describe” how to create the eggs environment. There is some interesting genetics there.) The other, which is not the case, is that it has recorded somehow the information about the environment given it and sets up the same environment for its offspring. The latter possibility, if it were to occur, would be a form of direct Lamarckian inheritance.

Information in the cytoplasm which is under genetic control, even indirectly, counts as information in the genome – you need genes for it. If you are going to talk about *additional* information in the cytoplasm you are talking about *nongenetic* information.

Continuity or a reasonable approximation thereof is required. (“Was” – we are talking about the synthesis as formulated.) It was and is the biologists, the ones that actually study plants and animals, that demanded soft inheritance because that is WHAT IS OBSERVED. Yes, there are traits that exhibit hard inheritance – color of eyes, for example, and smooth vs wrinkled peas. There are a lot of traits that are effectively continuous (I presume that we all understand continuity can be approximated by additive combination). We need only mention some examples, e.g., height, the size of a finch’s beak, and the placement of jawbones in therapsids.

Now this actually is a good argument, the essence being that variations in the protein structure supply de facto continuous variation in function. There are problems, though.

For this to work (in the context of the synthesis) it is necessary that there be sequential incremental change. Thus suppose we have a molecule X which controls a rate. There will be accessible (by mutation) variants X.1, X.2, etc. all of which, for Darwinian selection to work effect small changes in rate. Of these some one (or few) are selected, say X.1. In turn new variants arise, X.1.1, X.1.2, etc. The same requirement holds here. Thus the approximation of continuity must hold not only at the base point X but also in the space that X is resident in. This is a strong requirement.

Part IV – After thoughts

I think the upshot is that the demolition is correct, sort of. PZ Myers put it this way:

Does this spell trouble for the “change in allele frequencies” mantra? Sort of, I think. Ultimately, it’s all going to come down to some messy molecular biology, but for now there is more complicated stuff going on than we can understand. Look at beak size in those finches, for instance — it’s variation that can be quantified with a few simple parameters, but all the underlying biology is a total mystery. How is shape and size of a beak specified? I doubt that there is a “beak gene” anywhere in the bird. There are genes that somehow specify growth rates in certain bones, genes that define adhesivity in migrating tissues, genes that allocate cells to certain fates. You can measure beak length, but there are a thousand sneaky changes beneath that that you don’t see at all — and who knows which one is the genuinely significant one that selection sees.

The demolition and the discussion raised the issue of information: How much heritable information is there and where is it. Myers made a point that there is heritable information in the cytoplasm, i.e., information that the mother passes directly to the egg that is not in the genome. Gene imprinting is an example; development directed by maternal hormones is another, e.g., the testerone shot that mothers give their male fetuses.

I made the counterpoint that for this information to be stable it must ultimately be derived from information in the genome. Information that is part of the cell’s dynamic processes degrades over a few generations at most.

It occurs to me that this is a good thing, that much of evolution (or at least what looks like evolution) has very little to do with changing alleles. Consider those finch beaks again. To make it simple lets suppose the mother adds a hormone to the egg that controls how big the beak will become – and that she puts in about the same amount that her mother put in her egg.

The effect is that there will be a lot of variation in beak sizes without any underlying genetic variation. Darwinian evolution can operate very quickly on that variation, much more quickly than it could on genetic variation.

What this means is that there can be mechanisms for quickly changing phenotype attributes within a few generations to respond to environmental changes.

It also occurs to me that there may be rather less information in the genome than one would suspect from observing the phenotype. After all, most animals are simply variations of the segmented tube with things sticking out.

Part V – The Pimple Nosed Antarctic Anteater

There is a catch. For population genetics to give us anything besides drift we need non-zero selection coefficients for genes. To do this what we really need is that similar phenotypes (at least as far as fitiness is concerned) have similar genotypes. Thus, suppose that purple pimple noses are particularly favorable and that genomes Abc, aBc, and abC all produce purple pimple noses. Unfortunately genomes ABc and AbC produce green pimple noses which are the favorite diet of the Antarctic Fly Trap and Antarctic Anteaters with aBC genome are, ahem, obligate homosexuals. There is a further complication. In some years the Antarctic Grasshopper swarms; it dines exclusively on purple pimple nosed Antarctic Anteaters and on the Antarctic Fly Trap. In those years green is in and purple is out. We have this problem that colors do not generally breed true.

The situation is worse with respect to the synthesis and soft inheritance. In traits with soft inheritance (if there were such a thing) the trait has a numerical measure(s), e.g., the size of the bazoonga. [This being a family group we shall not discuss what a bazoonga is or what it used for.] The offspring have bazoongas whose size varies around the mean of the size of their parent’s bazoongas. Finch beak size will do as an example. The point is that natural selection in the Darwinian formulation operates on traits with soft inheritance. The explanatory stories – the exaptations, the evolution of the angler fish’s bait – all rely on the accumulation of small favorable variations. These in turn depend upon the determination of traits being approximately additive.

If genes in general encoded trait values and there were no correlation between fitness values and alleles, i.e., to get a fit genotype you have to have the whole sequence exact, selection of genotypes would quickly break down because there would be too many different genotypes.