Notes on the genetics, traits, and evolution.
Everything about us is a trait. Not everything about our traits is genetic, though. The conundrum of nature vs nurture, of genes vs environment, and the structure and meaning of genetics goes to the heart of biology. A few traits, like eye color, are simple enough. But they are the exception, by far. Body mass index is influenced by practically every gene we have. And autism has, by this point, hundreds of contributing genes. Both traits are highly heritable, in the sense that inheritance/genes are the dominant influence, vs environment (as seen in twins). But environment can easily be dominant over BMI when conditions change and starvation sets in.
A puzzle that came out of the early human genome studies was how unhelpful it was to do genome-wide association studies (GWAS) to approach some of these questions- that is, studies of what variants in the population at large contribute to particular traits, especially to serious diseases. Study after study was done, and disappointment mounted that what were found were genes with minor effects, in tangential biological processes. This was supposed to be the holy grail- the payoff for sequencing the human genome- and what came up was dud after dud.
A recent paper plows over this ground again with a new mathematical synthesis of genetic structure of human traits, genetic alleles, and selection. What it finds is sort of obvious, but there are some intriguing observations along the way. It is critical to note at the outset that evolution as Darwin understood and described, and natural selection in particular, is absolutely at the heart of this or any contemporary analysis of genetics. While Darwin's understanding was revolutionary and broad, subsequent decades of work have brought these concepts to a very concrete, operational, indeed mathematical, level.
Let's start with the concept of allele frequency, also called minor allele frequency, of MAF. Over any individual genome, there will be millions of "variants", which are coding differences from the reference genome. Which does not have any special status... it is just the genome of some guy from Buffalo. Variants (or alleles) are bases in the genome that differ from the reference. With three billion positions and four possible bases per position, that means that there are nine billion possible variants. How common is a particular variant in a population? That is its allele frequency. For GWAS and related studies, the threshold is commonly set at variants seen in the population at over one percent frequency, while minor alleles are seen under that frequency. A variant that causes some devastating disease is typically one that sprang up recently, and is heavily selected against. That is why it must have an exceedingly low allele frequency. On the other hand, a variant may have no discernable effect at all, not being selected for or against, thus just drifts along in the genome, not subject to natural selection. Such alleles may, over long periods of time, drift to higher or lower frequency by random chance.
However, the focus of GWAS association studies are variants between these extremes. These are variants that have some effect on a trait (or may be physically close to others that do, thus get "carried" along over time). At the same time, they are also common in the population, at least common enough to be discernable in an association study. Such a study needs some statistical correlation between the occurrence of the variant, and the occurrence of the trait. That means that the variant can not just appear once, but must appear many times over a large population. At the same time, a study that focuses on a trait like, say, high blood pressure, will be seeing variants that are, by definition, deleterious. That means that any allele with a large effect will be subject to strong selection, and driven out of the population. Only alleles with more modest effects will be able to survive at all, and even then, at low frequencies. So a GWAS focuses on medium-to-low prevalence variants, hunting for alleles on the loose in a large population that have typically modest effects on a given disease or other trait. Such alleles will have typically survived for tens or even hundreds of thousands of years, so they will have some complex relationship to natural selection.
In contrast to all this is the family study, which focuses on a dramatic variant that causes some terrible disease. Such studies have been remarkably productive, because they deal with extremely rare, high-effect variants, which are as a rule very informative about the genesis of that disease. Such variants, as mentioned above, would be heavily selected against, thus disappear rapidly. But mutation is always happening, so all sorts of mutations arise in large enough populations. Assuming that, as biologists, we are interested in the core ten or fewer genes that most influence a given trait or condition, the hundreds or thousands of significant, but low-effect variants that come out of GWAS are almost by definition guaranteed to be tangential and minor. It turns out that most traits are complex, in the sense of being influenced in various minor ways by hundreds of genes.
So, what is the typical genetic structure of complex traits? That is what this paper set out to answer. "Structure" in this case means ... what is the normal distribution of selective target / effect size versus frequency/prevalence in the population of variants that, in combination, add up to a complex trait. The assumption (as discussed above) is that the core armature of such traits does not have variants at all, due to strong selection, while the available variants in the population all have minor effects that in sum form the genetic variation seen in the trait in the population.
While other researchers have attempted to fit the variation distribution of complex traits to typical formulas like the normal distribution, these authors found that a natural selection-informed approach gave a clear and simple result. All traits follow the same general scaling, with only two parameters- the mutational target size of the trait (that is, the proportion of the genome capable of appearing as relevant variants), and the effect that each site has on the given trait, termed (very poorly) the site's "heritability". It is important to note that every site in the genome is equally and fully heritable. The term refers to the trait, and the size of the effect from variations of that site on that trait. Summed over all sites in the genome and all variations in the population, this heritability ultimately equates to the overall variation of the trait that is genetically caused.
 |
| A comparison of two traits, and how they might look in a genetic variation study. In blue is a trait skewed towards small effect variants, with weak selection and consequently variants with higher frequency. In red is a different trait that partakes more from stronger effect variants. On the whole, this kind of difference is not common among complex traits that arise from thousands of loci. MAF = minor allele frequency; Z-score is the score in a GWAS study indicating how statistically significant the variant's effect on the trait is. Note how lower Z-score correlates with more variants at the higher MAF frequencies. At the same time, the GWAS method overall has some skew to higher MAF frequencies, since only those provide sufficient statistical power to get any results at all. Log(s) is the strength of selection; L is the genetic target size for the whole trait, and h*2 is the heritability, or proportion of the trait effect due to the causal variant. |
The model they come up with accounts for the selective effect of trait effects (the larger the effect of the variation on traits, the lower its frequency in the population). It also accounts for the fact that variants that affect one trait often affect other traits as well, so the selective effect needs to be considered over all affected traits, most of which are probably unknown, but can be inferred. And conversely, most traits are composed of contributions from many genes and their alleles, sometimes thousands- they are genetically complex traits.
 |
| The model the researchers come up with can normalize among the huge population of variants that affect a single trait, in this case blood pressure. Left shows the effect sizes of individual variants, and right shows the scaled (normalized) version from the paper's model, showing that all these variants follow the same overall rule / logic, using the custom parameters of h*2 and L- trait heritability and mutational target size. All this is to say that the lower effect variants (skewed to left) are assumed to have higher selection coefficients. |
The researchers go on to show that various traits do look different under this analysis. Some differ mostly by target size, accounting for more or fewer variants, but having a similar spread of effect sizes over the population. Others differ by the scale of effects that each variant contributes, thus skewing toward higher or lower allele frequencies overall. Interestingly, they add an analysis of the age of these low-frequency, low-effect variants that are the grist for GWAS, finding that they are on average 137,000 years old. That compares with an average age of 600,000 years for variants that are neutral, thus would not come up in GWAS or be under selection. This is fascinating in its implications both for human population genetics in general, and for the fact that most human variation- even that under modest selection- predates the divergence between African and non-African populations.
No comments:
Post a Comment
Thank you for commenting!