User talk:Pietrobiroli

GxE interactions
You added sources to Gene–environment interaction supporting the claim that "virtually all traits show gene–environment interaction." My reading of this literature indicates that this is not the case, at least as far as human behavior genetics is concerned. Neven Sesardic, in his book Making Sense of Heritability, has collected comments (p. 66ff) on GxE from a number of researchers showing that the evidence for GxE is mixed at the very least. Some quotes:


 * nonadditive interactions rarely account for a significant portion of variance. (Plomin et al. 1988: 228–229)


 * there is very little empirical support for [the] existence [of genotype–environment interactions] in the behavioral domain. (McGue 1989: 507)


 * Interactions are known to occur, but they are far from invariable occurrences. Under many, if not most, circumstances, effects are additive and, moreover, people tend to show broadly similar responses to the same environmental stimuli. (Rutter quoted in Plomin et al. 1988: 228)


 * There is no conspiracy against interaction: If an interactive model could be shown to fit the data better than the traditional model, researchers would be quick to use it. (Plomin 1990: 144)


 * [The studies of interactions in nonhuman species have established that] although genetic control of sensitivity to the environment is widespread, the contribution of G × E to the overall population variance is typically smaller than the main effects of G and E even in controlled experiments using extreme environments. (Rutter & Silberg 2002: 465)


 * One aspect of genetic influence that environmental researchers seem to support is that of genotype–environment interaction. I agree that intuitively it seems as if these interactions must exist; however, genotype–environment interactions are non-existent in human literature. (Thompson 1996: 181)


 * The most plausible explanation . . . for the dearth of genotype–environment interactions in human behavior-genetics research on intelligence . . . may be . . . that genotype–environment interaction may not be important for individual differences in intelligence in most populations, given the ranges of intelligence, genotypes, and environments represented in these populations. (Waldman 1997: 558, the word order slightly changed, for ease of citation)

What do your sources exactly say about GxE? How do you reconcile them with the above? One problem is that some researchers think that the mere existence of norms of reaction indicates that there's GxE, even when the norms for different genotypes are parallel, i.e., even if different genotypes are equally sensitive to changes in environment.--Victor Chmara (talk) 22:04, 14 November 2012 (UTC)

Reply
Maybe it is too much to say that "virtually all" traits of human behavior show gene-environment interaction. However my reading of the recent literature is that interactions between gene and environment have been found to be systematically important, although potentially with a smaller contribution to the overall variance than the gene or environment taken alone. After the seminal paper of Caspi et al (2002, 2003), many others have found significant interactions. Some citations:


 * A gene for aggression lands you in prison if you are from the ghetto, but in the board room if you are a manor born (Dalton Conley, 2011)


 * Flamingos everywhere are famous for their beautiful pink color. [But] if flamingos do not have access to their usual diet [of shrimp and plankton] for any reason, they are white, not pink. Their color is entirely dependent on the environmental influence of diet. On the other hand, the flamingo's ability to turn pink with diet is entirely dependent on their genes. You could feed seagulls forever on the same diet and they would never turn pink. It would make no sense to say the flamingo's color was 50% due to genes and 50% due to diet. The color is due to the joint action of genes and environment (Micheal Rutter, 2006)


 * The nature versus nurture distinction, although traditional, is obsolete. Gene and environment cannot be meaningfully parsed by traditional linear models that assign unique variances to each component (James Heckman, 2007)

Pietrobiroli (talk) 22:36, 27 November 2012 (UTC)


 * Yes, there is evidence for GxE effects for some human behavioral disorders (although replication studies have only sporadically been able to confirm these findings). However, there is no such evidence for most human phenotypes (such as IQ, which is by far the most studied variable in behavioral genetics). Heckman refers to Caspi's research on behavioral disorders, not any other evidence relevant for human traits. Heckman's claim about "traditional linear models" is also misleading, because tests for non-linear or GxE effects have been around for a long time -- the most powerful method for this was introduced in Fulker and Jinks's classic 1970 paper. Additive models are preferred simply because they fit the data better, and it's absurd to claim that they are obsolete in absence of corroborating data. Even if GxE is present, it will prevail over the main effects of G and E only if it is very strong, and, as pointed out by Rutter and Silberg in the quote above, GxE is usually smaller than G and E even in experimental nonhuman studies using extreme environments.


 * The flamingo example highlights a definitional problem that is not discussed in the Wikipedia article. The interaction between diet and plumage color is analogous to the fact that a farmer can increase his crop yields by using fertilizers. However, the latter is not usually thought of as an example of GxE. As I wrote above, some researchers think that the mere existence of reaction norms entails GxE, while others argue that there's GxE only if the shapes of the reaction norms of different genotypes are dissimilar (and even a difference in reaction norms can be regarded as simply a scale artifact). In any case, the practical implications of these two types of GxE are vastly different. These issues should be sorted out in the article. The claim about ubiquitous GxE is also highly dubious in light of evidence.--Victor Chmara (talk) 09:39, 28 November 2012 (UTC)

The research of Turkheimer and colleagues shows evidence of interaction between socioeconomic status and genotype also in the case of IQ. (I believe that we can safely regard socioeconomic status as an environment). Qi and Cho  focus on obesity. Therefore, I don't think that the evidence of GxE is relegated only to human behavioral disorders. As said above, it might be overreaching to say that virtually all traits show GxE, but extensive research has been focusing on various human traits, finding evidence of GxE. It might not be an important driver of the variance decomposition in population genetics, but it should not be dismissed as non-existent only because simple linear additive models fit the data better.

Regarding the flamingo example, I believe it is different from the example of fertilizers: if you do not use fertilizers, the crop still produces a certain yield, albeit lower. This is a good example of norm of reaction, where variation in the environment is mapped into variation in phenotype. On the other side, if you don't feed shrimps to the flamingo, it doesn't turn out as a paler pink, but it turns out white. There is not a scale or a variation of environment, but a sudden change.

Finally, thank you for pointing out the concept of norm of reaction, which I didn't know. I found it very useful and agree with the idea that a GxE can be represented as a non-isomorphic shift in the shape of reaction norm due to different genotypes.

Pietrobiroli (talk) 18:10, 2 December 2012 (UTC)


 * Turkheimer's study is another example of the definitional or terminological confusion. What he found was that the environmental factors that are relevant for IQ are not uniformly distributed across social classes. (Note that some other studies have failed to replicate this finding, e.g. .) This is again not a GxE effect in the sense that Caspi et al. have used the term in the case of behavioral disorders. Caspi et al. found that exposure to the same stressful (or bad in other way) environment causes behavioral disorders in individuals with certain genotypes, while those with different genotypes do not develop any disorders in that environment. Furthermore, in a good, stress-free environment individuals with either kinds of genotypes do not develop these disorders. This is sometimes called statistical GxE, and it's clearly different from, say, the flamingo example. If some flamingos raised without shrimp and plankton nevertheless turned pink, while some others (with different genes) with the usual diet turned white, then we'd have statistical GxE. (I don't think it matters if plumage color is a categorical or continuous variable, but it's probably continuous anyway, i.e., the shade of pink depends on diet composition.)


 * What I'm concerned is that if we say "GxE universally happens", then what do we actually mean? I agree that there are usually better and worse environments for phenotypes, but it is definitely not true that the effect of the same environment is usually different for different genotypes, which is what statistical GxE is about. Sesardic, in his book mentioned above, puts it this way:


 * A widespread conviction that heritability claims are devoid of almost any interesting explanatory content is often based on an argument that genes and environments interact, and that for this reason their causal contributions to phenotype cannot be separated and measured independently. An immediate problem with this argument is that there are two very different meanings of “interaction”: commonsense and statistical. According to the commonsense notion (interactionc), to say that two causes A and B interact means that neither can produce the effect without the presence of the other. To use a standard example, striking a match and the presence of oxygen interact to produce fire. According to the statistical notion (interactions), however, to say that two variables A and B interact means that a change in one variable does not always have an effect of the same magnitude: its effect varies, depending on the value of the other variable. For instance, the very same life event, such as parental divorce, may affect children with different personality characteristics quite differently.


 * What this means is that if there is "commonsense GxE", genetic and environmental effects act separately (G+E), and the estimation of independent genetic and environmental components of variance is just about having an adequate sample, i.e., heritability estimation is, in principle, unproblematic. If, on the other hand, there is (strong) "statistical GxE", then the environmental and genetic effects cannot be separated, and heritability cannot be estimated, i.e., phenotypic variance is due to GxE, not G+E. Using reaction norms, commonsense GxE looks like this, while (very strong) statistical GxE looks like this.--Victor Chmara (talk) 22:28, 2 December 2012 (UTC)