Gene Expression

Some of the most fascinating theoretical evolutionary biology that I’ve run into emerges out of David’s Haig’s work on genetic conflict. You’ve probably stumbled into it somewhere, whether via popularizers like Matt Ridley, or other researchers like Robert Trivers and Sarah Blaffer Hrdy. Haig is a biologist who extrapolates from the familiar axioms of evolutionary genetics and hacks his way through the jungle of derivations and wanders into a world turned upside down. Though his work in the area of mother-offspring genetic conflict in utero is probably what is in widest circulation in the public mind space, Haig’s refinment of “kinship theory” of genomic imprinting is a far more general paradigm, and nearly as accessible.

What is genomic imprinting? Consider a diploid organism. It has two parents (e.g., it is part of a sexual species) of two sexes, male and female. On each genetic locus the offspring of the two parents receives a copy of a given gene. Convential Mendelianism offers a few outcomes of the pairing of the two genes. The two variants, alleles, could be functionally identical (both “wild type”). Or, one could be wild type and another could be a mutant. In some cases a mutant will be a loss of function, and the wild type might “mask” its effect, so the mutant will be recessive to the dominant wild type. In some cases the mutant could actually be dominant, for example, it produces a wide ranging protein factor which gums up the wild type copy and prevents its expression (this is a negative gain of function). Sometimes each copy offers its own “dosage” to the phenotype, so that if you have one wild type copy you have half expression. And so on. But, note, I haven’t mentioned at this point whether a copy is maternal or paternal, it doesn’t matter. Unless you are speaking of the sex chromosomes, it is irrelevant if the allele comes from the mother or father.

In genomic imprinting it does. Consider igf2, an insulin growth promoter, in general the paternal copy tends to express, while the maternal copy is silent. In other words, if you receive a defective paternal copy you are shit out of luck in regards to the function of igf2 because the other copy isn’t talking. Additionally, in most mammals there is another gene, igf2r, which acts in reverse of igf2 as a growth inhibitor, and, it is active in the maternal copy and silent in the paternal copy. This is all strange, no?

How does this happen? Well, it occurs epigenetically, and methylation is usually involved. I’m not going to say more about the molecular genetics because that is an enormous field and it would draw us away from the “big picture.”

Instead, let’s ask the evolutionary question: why does this happen?

Haig forwards what he terms kinship theory, though in the past it has been called “genetic conflict theory.” But first, let’s present two alternatives:

1) The evolvability hypothesis
2) The ovarian time bomb hypothesis

The evolvability hypothesis is pretty simple, the silencing of one uniparental lineage results in the masking of its existence from natural selection. This allows that lineage (for example, igf2 maternal and igf2r paternal) to build up novel mutations, which might eventually be “called up” to generate new adaptations. Haig dismisses this theory on a variety of grounds. First, this seems a rather peculiar way to generate new variation, especially considering that turning the lineage neutral will probably simply create loss of function, not the synergy and novelty the theorists propose. Additionally, organisms already have a way to build up mutations: recessive alleles. In a random mating population recessive alleles are only expressed phenotypically, that is, seen by selection, to the square of their frequency. So, if a recessive allele is floating at 1%, only .01% of individuals in a random mating population will exhibit it, so 99% of copies of the allele are invisible to negative selection. Finally, this model is just too general, if imprinting emerged to mask alleles then it would be more ubiquitous. The reality is that diploid organisms are diploid for a reason, it forestalls loss of heterozygosity and increases fitness. But the reality is only a few hundred loci have been discovered in humans that exhibit genomic imprinting, it is simply too rare a genetic phenomenon to increase evolvability!

Next, the ovarian time bomb hypothesis posits that paternal expression and maternal suppression is crucial in preventing premature and unfertilized oocyte development. The logic behind this is clear: to prevent maladaptive triggering of development prior to fertilization, simply make the paternal copy a necessary precondition of the developmental pathway. But the problem with this hypothesis the inverse of that of the evolvability hypothesis: it is too specific. Why are loci which have no relationship to oocyte development imprinted? The researchers propose ‘bystander effects.’ Why are some maternal loci active and paternal silent when the point is that maternal loci are silent and paternal active so that development is triggered after fertilization? Also, imprinting is found in taxa which lack invasive placenta. Finally, imprinting is persistent in somatic tissue where premature oocyte development is irrelevant. Imprinting results in highly risky monoallelic expression. If you want to know why this is highly risky, read about Retinoblastoma.

Haig contends that his kinship theory does not suffer from the faults of over generality and over specificity. It can explain the finite number of loci which are imprinted, and it can explain the preponderance of these loci in terms of the functional-adaptive reason for their sex specific expression and silencing patterns.

To understand kinship theory, you need to start with kin selection. Consider the following coefficients of relatedness:

You:You = 1
You:Monozygotic Twin = 1
You:Sibling = 1/2
You:Parent = 1/2
You:Half-sibling = 1/4
You:Uncle/Aunt = 1/4
You:Grandparent = 1/4
You:First cousin = 1/8

So far so good, you understand the general logic here, your genetic relationship drops off by 1/2 each generation or line of relationship between you and the individual in question (neglect inbreeding). Now, let’s focus on half-siblings. Using Haig’s model it matters whether you share a mother or a father with your half-sibling. Though the averaged coefficient of relatedness is 1/4, if you share a mother than your maternal alleles have a coefficient of relatedness of 1/2 and your paternal alleles one of 0. If you share a father the situation is inverted. For illustrative purposes let us assume you share the same mother (among most mammals mothers raise offspring, so you would have a chance to socialize with half-siblings in this case). In a Hamiltonian context inclusive fitness is defined like so: do action if cost is less than the benefit X coefficient of relatedness (C < B X r). This is the origin of J.B.S. Haldane's famous quip about being willing to sacrifice his life for an x number of relatives in proportion to his genetic relationship (the more distant the set of relatives the more of them needed to balance out his sacrifice). In any, if you have a maternal half-sibling C < B X r differs whether it is a paternal or maternal allele because r is either 1/2 or 0! Your maternal alleles would benefit from inclusive fitness because they are likely to be present in your half-sibling (1/2 likelihood at any given maternal focal point on the genome), but your paternal allele would get nothing genetically (remember, we are ignoring inbreeding).

And so this sets up the scenario for conflict. Remember igf2? It is expressed in paternal copies, and, in most mammals it is counterbalanced by a repressor, igf2r which is only expressed in maternal copies. When igf2 is expressed it induces growth of the offspring, and while igf2r counteracts this. If you break igf2 in the paternal copy than the offspring will be diminutive, while if you break igf2r you may see higher basal concentratiosn of IGF in the offspring and a larger size. What does this have to do with genetic conflict? A large fetus is risky for the mother, and a large offspring is calorically more demanding. If the father is a random variable from mating to mating than paternal copies view the mother as dispensible and wish to maximize the survival of the offspring. In contrast, maternal copies benefit from a long reproductive lifespan for the mother and so it will attempt to restrain excessive fetal and offspring demands on the mother which might reduce her long term reproductive fitness, ergo, the probability of producing future maternal copies of the alleles in question. This explains the limited set of imprinted genes: they are implicated in mother-offspring relationships in regards to growth, development and behavior. It can also explain the expression and silencing in regards to why they are biased between the father and the mother.

There are many more details, and I will get to them in the coming week as I continue my review of Haig et. al.’s work. But note the big picture here, the intial assumptions are spare and minimal within the context of Mendelian genetic principles, and yet from this one can derive a rich set of inferences and predictions. Of course, some may this “well, this is theory, aren’t we getting ahead of ourselves?” True, as I suggest above, many of the epigenetic processes and mechanisms which control and modulate imprinting are poorly understood. But, Darwin’s theory of evolution based on natural selection was presented 40 years before the proper analytic model of genetical inheritance was understood (Mendelianism). And, note that I was another 50 years that the genuine replicative substrate, DNA, upon which Mendelian abstractions depended was understood to be the root of the issue. Theory without data is irrelevant, but data without theory is blind.


Wilkins et. al., What Good is Genomic Imprinting: the function of parent-specific gene expression, Nat Rev Genet. 2003 May;4(5):359-68

Burt et. al., Genes in Conflict, 2005