-1
$\begingroup$

In articles like this one, I often read that several "genes variants are associated to a given trait". It is often added: "genetic factors explain (say) 20% of the trait variance."

The way I understand this is the following: researchers regress the trait upon the genome, find correlations between some variants and the trait and give some measure of the fit (or explained variance) like the $R^2$. Is my understanding correct?

If my understanding is correct, I am uncomfortable with the interpretation that the trait is partly genetic. Indeed, it could be due to rearing (that is, from the parents, yet not from the genes). I am aware of the practical difficulty (or impossibility) of finding a trait that we know for sure is from rearing and not genetic, but one can use models (described below) to study my concern. The basic model feature a random trait and the advanced one a trait transmitted by rearing.

The basic model runs as follows: define an arbitrary trait as the indicator of a random subgroup of the (global human) population. Would a study really fail to detect a genetic causality in this model? How common is it to find some gene variants associated to this arbitrary trait, and explaining (say) 20% of the trait variance? More formally, what is the distribution of the variance (apparently) explained, in function of the size of the arbitrary subgroup? I am sure there is a paper (or even a literature) about this: I'd like a reference to get the main insights.

Now, let's turn to the more advanced model, for which I am also seeking for references (I am sure it exists as well but have no clue how to look for it).

The model simulates genomes of the whole population among successive generations. Some trait appear at a given generation (call it t=0, although it is not the eldest generation modeled) among random individuals. The trait is not uniformly distributed at t=0, but has more chances to be found in individuals "close to the spatial location where it appear" (you can think of the trait as "listening to techno", and the location as Detroit). Suppose the trait is not genetic, in the sense that no gene variant influences the trait occurrence in one individual. Instead, the trait is transmitted through rearing environment: e.g. an individual has the trait with probability P if one of their parents has it, and with probability p<P otherwise (we could refine the assumption and say it could also be transmitted by acquaintances, or that it has more chances to be transmitted if both parents have it, but I think such refinements are not needed, and we could perhaps even simplify further and take p=0, P=1). Then, after T generations of breeding (realistically modeled), some biologists try to assess whether the trait is genetic. They will surely find some genetic correlations as (i) the trait has originally appeared in a specific location where people were relatively close genetically, (ii) the trait is transmitted by the parents, like genes, and (iii) there are many many genes, so that the probability is high that those who got the trait at t=0 share some gene variants. Now, let me recall that the trait is not genetic: for example, we could take the babies from their biological parents at t=T and have them raised by randomly drawn couples, it would be the adoptive children of techno listeners who would share the trait, and the (apparent) genetic link would be lost.

Hence my question: how do biologists know whether a trait is genetic or transmitted by rearing, despite the two seeming hard to distinguish in the advanced model? When I read a study claiming that a given trait is genetic, do they employ subtle statistical method to really prove so, or do they mean by that "genetic or reared", in the sense that the trait could fall in the case of the advanced model above? Does "explain 20% of the variance" mean that it is (probably) partly genetic?

$\endgroup$
10
  • 4
    $\begingroup$ Is biology.stackexchange.com/a/42280/27148 helpful at all? $\endgroup$
    – Bryan Krause
    Aug 31 at 1:14
  • 4
    $\begingroup$ I really don't understand this question. How can a trait be genetic OR hereditary? Genetic IS hereditary. $\endgroup$
    – jamesqf
    Aug 31 at 15:53
  • 1
    $\begingroup$ Does this answer your question? Why is a heritability coefficient not an index of how "genetic" something is? $\endgroup$ Aug 31 at 20:44
  • 3
    $\begingroup$ @bixiou: I may be wrong, but my understanding is that genetic == hereditary. You seem to be asking about genetic versus environmental, which could include prenatal influences such as chemical exposure (fetal alcohol syndrome comes to mind, or being taught something by parents &c. Acquisition of language would be an example of the latter: humans have the genetic ability to learn language, but what particular language(s) they learn is environmental. A child of Chinese parents adopted as an infant by an English-speaking family will grow up to speak English, not Chinese. $\endgroup$
    – jamesqf
    Aug 31 at 23:37
  • 1
    $\begingroup$ Thank you all, in particular @BryanKrause. I learned the word "rearing" thanks to you :-) Indeed, I had a wrong definition in mind for heredity. Your edits were fine, I just changed "influenced by rearing" into "transmitted by rearing". $\endgroup$
    – bixiou
    Sep 1 at 22:47
3
$\begingroup$

It's purely and simply that there is no single answer - as in the linked paper, there's no "gay gene", there is a group of identified genes that contribute, but not all the variance in the population seen can be attributed to those genes. i.e. you can have some or none of those genes and still be gay, or indeed all the genes and not be gay.

This could mean that there are other genes still to be identified as playing a role, or it could mean that environment, epigenetic factors (e.g. methylation, some of these seem to be hereditary too!), expression levels etc. play roles in this trait, but as it is multi-factorial, we don't know the answer(s) yet. It may well be that we will never know, or that there are more than 1 answers to the trait.

$\endgroup$
5
  • $\begingroup$ I am sorry but this does not answer the question nor does it add information with respect to it. $\endgroup$
    – bixiou
    Aug 31 at 7:42
  • 2
    $\begingroup$ @bixiou - yes it does - if you want me to spell it out explicitly - your "models" make some assumptions that are simply incorrect. There are heritable traits of which we know no genetic component - e.g RogerVadim's handedness in comment above, and this is partly because we don't fully know what all of our gene products do and how they are regulated. Have a look at the functional interactions of p53 at stringdb - keep clicking the "more" button. We've known and heavily studied this protein for ~50 years and are still finding interactions... $\endgroup$
    – bob1
    Aug 31 at 9:39
  • 1
    $\begingroup$ @bixiou ...there are about 25,000 genes in the genome, each would need to be studied to the same level and we would need to be able to work out the undoubtably multi-factorial models to explain some of these traits. We don't have the technology yet, though we are just starting to scratch the surface. We are also finding new ways in which we regulate genes and gene products - M6A (6-methyladenosine) in RNA is a good example of this, known for 30+ years, only now being identified as to what it does. $\endgroup$
    – bob1
    Aug 31 at 9:44
  • $\begingroup$ That's interesting that we are not able to study all genes at the same level, it shows that my understanding (as I have defined it at the beginning of the post) is wrong. I'd like to know how they do though, because knowing that I am wrong does tell me the correct answer. $\endgroup$
    – bixiou
    Aug 31 at 11:05
  • $\begingroup$ @bixiou why they aren't studied has many answers, partly funding, partly man-power, partly reagents, partly we haven't looked for that interaction (yet). p53 is heavily studied because it is a cancer gene, one of the more common changes, it also happens to be a cell cycle regulator, so it is easy to get money and thence man-power to study it. A lot (almost all) of the interactions at a protein level are determined empirically and 1 by 1, so you can see that it takes a long time and a lot of work to get it done. $\endgroup$
    – bob1
    Aug 31 at 20:54
2
$\begingroup$

General remark
One does not necessarily need to use a subtle statistical method, but one does need good understanding of the experimental design and statistical analysis in order to draw reliable conclusions from the data (or know when not to draw such conclusions). It is for a good reason that statistics is a field of its own (just like biology) and there exist a dedicated stack exchange community (by far more active than biology). Pubmed is also full of articles explaining why this or that approach needs to be carefully - just try to search for spirious correlations and see how many articles come up.

Correlations and non-correlations
Closer to the point: the model in the OP assumes that certain trait can be a consequence of the location (or other non-genetic factor, not necessarily hereditary) and the genotype. The co-occurence of certain genotypes and locations confounds the problem. Moreover, it is possible that this co-occurence actually leads to real correlations between these two factors.

One thing to look for is the appropriate sampling procedures, especially the sample size. As an extreme example, let us consider preference for wearing warm clothes in winter - is it a function of location (Moscow vs. Miami) or a trait coupled with Y-chromosome? We could do the analysis of variance, proposed in the OP and easily prove that there is no correlation with the presence of Y-chromosome... unless most of the individuals sampled in Moscow were men and most of those sampled in Miami were women, in which case we may erroneously attribute the preference to warm clothing to possessing Y-chromosome.

It is clear what has gone wrong in the example above:

  1. the experimental design was not balanced
  2. statistical analysis was not corrected for this lack of balance

One can thus expect improvements along these two axes: by designing experiments that allow disentangling undesired correlations and by employing the approrpiate methods of analysis. Let me however add a few more remarks:

  • not all correlations can be disentangled - sometimes creating appropriate design is difficult or even impossible. This is especiallyw hen we are talking about complex genetic traits.
  • there may be genuine correlation between the traits - e.g., the trait of interest and settling at a certain location may be both functions of genotype.
  • In regard to the correlations that may arise after several generations, as the OP suggests, it is worth keeping in mind that such correlations require evolutionary timescales - they are a real issue when comparing Native Americans and Chinese, but less of a problem when comparing populations in New Yorkers and Detroit.

Suggested Reading
I suggest starting with the Wikipedia article on the Experiment design or an equivalent chapter in a biostatistics textbook. Statistics community Cross Validated is rather welcoming to biostatistics questions. Finally, there are many good statistics and biostatistics textbooks - the obstacle is usually not the availability of materials, but the level of math and abstraction.

$\endgroup$
9
  • $\begingroup$ The reason why I am asking on a biology forum is because 1. geneticists have for sure thought about that (and master the relevant statistics), 2. I am interested in what geneticists do, what statistical methods they use, whether and how they are able to distinguish heredity from genetics. I am sorry but your answer does not really help answering the questions I raised. $\endgroup$
    – bixiou
    Aug 31 at 11:04
  • 2
    $\begingroup$ @bixiou you asked a very general question - one cannot describe all possible statistical methods used in genetics in a short answer. Try to ask a more specific question, and also give an idea on how much background you have in genetics and statistics. Alternatively, you may consult books, such as springer.com/gp/book/9780387953892 $\endgroup$ Aug 31 at 11:21
  • 1
    $\begingroup$ @bixiou three people tried to answer your question, but it seems that none of us understands what you are really asking. You know better what your question really is, so try reformulating it, and adding some references, so that we can have a better idea of what you mean. $\endgroup$ Aug 31 at 19:53
  • 1
    $\begingroup$ @bixiou - that's environmental, language is a fine example. The reason for our confusion is that often biologists have a very strong understanding of genetics and heredity and use those terms to mean specific things... hence my answer talking about factors that we don't know about yet. $\endgroup$
    – bob1
    Sep 1 at 21:10
  • 1
    $\begingroup$ @bixiou Re: "when an article claims that a trait is (partly) genetic (because some share of the variance is explained by genetics), should we interpret this literally?" Sometimes yes, sometimes no. As Roger points out, It depends on how well the experiment and statistical analysis were designed and executed. If you don't feel you can evaluate this yourself, you might want to stick to review articles and textbooks, where someone has already done at least some evaluation and vetting. $\endgroup$
    – Armand
    Sep 1 at 21:32
1
$\begingroup$

Some papers that should help and provide further info in their refs:

An oldy but goody: "Genetic dissection of complex traits: guidelines for interpreting and reporting linkage results" (1995) Eric Lander & Leonid Kruglyak

Here's a discussion of techniques involved, with an emphasis on linkage analysis but discussing association studies as well: "Genetic linkage analysis in the age of whole-genome sequencing" (2015) Jurg Ott, Jing Wang, and Suzanne M. Leal

A more recent example: Höglund, J., Rafati, N., Rask-Andersen, M. et al. Improved power and precision with whole genome sequencing data in genome-wide association studies of inflammatory biomarkers. Sci Rep 9, 16844 (2019).

$\endgroup$
1
  • $\begingroup$ Thank you @Armand! These references should answer the question. I'll probably accept this answer once I have read them. $\endgroup$
    – bixiou
    Sep 2 at 8:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.