Friday, July 29, 2011

Mendelian Inheritance: Basic Genetics or Basic Mistake? Part V.

This series of posts has been about the illusion of simple Mendelian inheritance, that has been an enormously powerful tool in understanding how genes are inherited, working through carefully chosen experimental situations in which traits were so closely tied to specific alleles (variant states of a single gene) that it seemed as if the trait itself were being inherited.  But it's only genes that are inherited (except for the goop that's in the fertilized egg or other cell that starts a new organism on its way).

We've been saying that the effect of  alleles on a trait varies with the alleles' contexts.  We can call this variation a 'spectrum' of effects, just as a rainbow is a spectrum of color: most variants have very small average effects, but a few, usually rare ones, can have such large effects that it seems that whenever you inherit one of those alleles you're nearly sure to have the trait.  Just as in the classical pattern of occurrence in families represented in this figure that we've grabbed from the web--almost.  We used the original in earlier posts in this series, because it is that which is in all the textbooks.  But this is more like reality and illustrates a main point.  The Aa's in the middle are not exactly as 'dominant' as the true-red Aa individuals.  And the light orange aa in the grandchild could easily be classified as unaffected, or affected....depending on what theory you were trying to confirm:

But we've said that the more clear-cut (or argued to be clear-cut) version is what one sees for a selection of variants in a selection of genes, and it grabbed scientific attention because it fit our expectations of  how inheritance works, based on Mendel's results with carefully chosen traits in peas, and similar trait-selection for the following century.  Even  then candid acknowledgment will be typically made that there is variation in the actual traits that people in the same family have: not all the red dots are equally solid or red, as in the figure.  And we know that as a rule, since most effects of individual variants are very small, and many variants contribute to most traits, even the basic idea of Mendelian inheritance makes little sense as a rigorous theory of inheritance--and we should realize that, accept that, even if we wish things were simpler.  The real world doesn't have to fulfill our wishes!

In fact, fine classical papers from around 1960 showed that the appearance--the illusion--of Mendelism can arise in another way unrelated to the extreme, usually rare ends of the allelic effects distributions.  If a presence/absence trait, like hyptertension vs normotension, stroke vs non-stroke, cancer vs non-cancer, arises when a threshold is exceeded on some underlying quantitative trait (e.g., if your blood pressure rises above some agreed-on cutoff level for calling it hypertension), and if many different genes contribute to it, the trait can occur in families in a way that appears clearly 'Mendelian', that is, as if only a single allele were responsible.

We know that even with complex inheritance, children will resemble their parents, and we know the average extent to which that should happen.  Generally, you're half-way between your parents' trait levels, such as stature.  But the usual idea is that this doesn't apply to discrete (yes/no) traits that follow Mendel's rules: you're either this or that, but not a half-way blend of your parents for such traits.  As Darwin would say, the traits do not 'blend'.

Nonetheless, as was shown in the '60s, the probability of a qualitative trait (yes/no, like hypertension of diabetes) can be similar enough between close relatives that it appears to follow Mendel's rules.
This is because if you have a combination of genetic variants, across your genome, that makes your blood pressure high, your children will inherit half of those (on average) plus whatever similar risk effects your spouse may have, and for a generation or two it can have a net result of around half of children of an affected parent also being affected with hypertension, which is what you'd expect if there were just one Hypertension gene with two states, normal and hypertensive.  The illusion can arise under a broad set of circumstances, and can fuel hopes of simple situations----or hopes that GWAS will, after all, really work.

This means that, in addition to all the other things discussed in this series, and not even considering shared environmental effects which can often be by far the most important, multi-gene causation can reinforce the ideas of someone assuming that Mendelian inheritance of the trait is true.   

We can relate all of this to GWAS findings in another way as well.  If a given allele is common enough in cases for its effects to be found and reach statistical significance, relative to its frequency in controls, then it has a chance to be detected in a GWAS study.  But this depends on its penetrance, that is, on the strength of its effects, on their own, on a trait measure like stature or blood pressure, or on the presence of a disease.  Highly penetrant alleles will be found more frequently in cases than controls because they have a higher chance of 'causing' the trait.  The greater the effect, the more likely if you have the allele you have the disease.

This is a kind of 'dominance', because the allele is being detected against the other allele at that gene in individuals, plus whatever other relevant variants they have in their genome. That typically only a few genes are identified in this way shows how relatively rare real dominance is--how far it is from being the baseline, basic nature of inheritance!  In fact, most variants that contribute contribute so little--have so little 'dominance' in this context--that we simply cannot detect their individual effects (or those effects are not enough to generate a statistically 'significant' association with the trait).

Put another way, these various considerations show how, if we assume a theory, we can make the data fit the theory and also assume we understand the data.  But if that theory is wrong or very inaccurate, it can lead us far Mendelian theory has indeed been doing for more than 150 years.


Corneel said...

That typically only a few genes are identified in this way shows how relatively rare real dominance is--how far it is from being the baseline, basic nature of inheritance!

I liked reading your posts, but at this point I disagree. GWA studies are relatively unsuccesful, because they often aim to find alleles associated with a certain disease, which are rare and typically recessive. Thus, it is in fact dominance that makes it hard to detect these alleles. Although I broadly agree that dominance is not a fixed property of an allele, I think that it can be quite stable across different genomic backgrounds. Dominance explains why we have a genetic load of recessive deleterious alleles, and why we see inbreeding depression as soon as the homozygosity increases.

Apart from that, I think the point that genes are not traits is worth making. A long time ago, the botanist Wilhelm Johannsen introduced the perfectly fine terms "genotype" and "phenotype" to make this distinction clear. It probably won't harm to remind people of that.

Ken Weiss said...

I think the difference between our views is largely semantic. I would say alleles have a spectrum of effects, with most being very small to a few being large, and most being more but a few being less context-specific. The spectrum is largely quantitative rather than qualitative, and differs in detail for each trait and species.

Mendelian dominance should not be treated as a qualitative ground state of nature as it is in most textbooks, which I believe inculcates students into confusing inheritance of genes and the nature of traits, which I believe leads to a lot of poor biology. And I think quantitative genetics should not be treated as a separate field, as it typically used to be and in some senses still is.

I think most variation at most genes has little phenotypic effect (lethals being excepted though even those can be context-dependent). Even traditional dominance is sometimes, if not usually at least somewhat context dependent.

I think this explains the under-performance of GWA, because that was based on false expectations. Actually, I think the idea of inbreeding and recessiveness is also oversold. Most isolates are not heavily burdened with recessive diseases overall, even though some arise that are much rarer in their parent population--and we treat them as inherently recessive, which may overlook context-dependency in that population at a genome background level. That would in my view also apply to inbreeding depression, which is genomwide but doesn't necessarily mean specific loci have classical dominance.

In that sense, inbreeding depression is about statistical dominance, in the terminology we used in our post. I think the loss of variation genomewide by inbreeding can rightly account for the depression phenomenon without needing classical single-locus, single-allele inherent dominance, which was the main point of our posts (at least that was our intention).

Genetic load is a different subject that is interesting and in a way the reverse of its sense in the classical population genetics of decades ago, when the term arose. It's a very important but I think unsolved issue. We may want to do a post or two on that, come to think of it.

Holly Dunsworth said...

Do you think that we should we be talking in terms of penetrance (along a spectrum of more or less) rather than using the terms dominant and recessive?