It is now 22 years since I wrote my first open-peer commentary for Behavioral and Brain Sciences (Crusio Reference Crusio1990). How much has changed since. And how much has stayed the same. In that commentary, I argued that heritability estimates had only a very limited value, namely, that of predicting the response to selection. Being young and bold (or perhaps young and reckless), I asserted that “these estimates have only a very limited value for researchers investigating animal behavior and are without purpose in human-behavior research.” Significantly older now and perhaps even more reckless but certainly more stubborn, I still maintain that heritability estimates have only a very limited value and I still don't see much use for them in human research.
Perhaps it would be good to recall the definition of heritability: the proportion of the phenotypical variance that is attributable to genetic variation in a given population in a given environment or set of environments. (Depending on whether or not only additive-genetic variation is taken into account, we may distinguish between heritability in the broad sense or in the narrow sense, but that distinction is not important for this discussion.) The crucial words here are “in a given” population or environment. If our population is an inbred mouse strain, heritability will be zero for every measurable phenotype, as there is no genetic variation present. In a population with limited genetic variability, heritability may be relatively small, while a genetically more variable population would show a higher heritability even if it was living in exactly the same environment. Similarly, if two populations had identical levels of genetic variability, but were exposed to different environments, then they would almost certainly also show different heritabilities. In short, in human populations, where we cannot experimentally control either environments or levels of genetic variation, heritability estimates have no inherent theoretical value (as opposed to practical uses, such as determining the number of subjects needed to localize genes). As far as I can see, the only questions of theoretical import would be whether heritability is significantly different from 0 (i.e., genetic effects on the phenotype of interest exist) and from 1 (i.e., significant effects of the environment exist). The often-heard argument that higher heritabilities would mean that a trait (or disorder) will be less responsive to (therapeutic) interventions is obviously flawed.
It should perhaps be noted that the foregoing holds even in the idealized simplified situation where there is no interaction or covariation between genotype and environment. Taking into account also those effects, as well as the molecular processes so skillfully described by Charney in the current target article, really should be the final nail in heritability's coffin. It is true that the effects and processes that Charney describes can be included in quantitative-genetic models: Whatever can be measured can also be modeled. But given the foregoing, such an exercise would seem to be a waste of time and effort. In short, my only beef with Charney's criticism of heritability is the same as the one I raised in my commentary on Wahlsten's (Reference Wahlsten1990) target article, namely, that they both criticize the concept of heritability on technical points, whereas even if those could be solved, there are still much more fundamental inherent problems with the use of this construct in human-behavior genetics. That Charney nevertheless apparently felt that he had to pay the attention that he did to such a useless notion is an apt testimony of the amount of reification that the concept of heritability has undergone in the field of human quantitative behavior genetics.
Despite the foregoing, I would like to bring up one more point about heritability estimates. Animal studies are designed to “boost” heritability: All animals are raised as much as possible in uniform environments in an attempt to reduce the amount of phenotypical variance attributable to variation induced by environmental variations. The end result should be that genetic causes would become relatively more important, leading to higher heritabilities, compared to uncontrolled situations such as encountered in human studies. Yet, studies in human-behavior genetics routinely report heritabilities over 50%, whereas studies in animal genetics generally yield much more modest values, a point that has always puzzled me. The explanation for this paradox may lie in the different processes described by Charney. First of all, although their effects may result in both over- and underestimates of the heritability of a character, the general tendency seems to be to inflate these values. Second, thinking about the way animal studies are set up (e.g., comparisons of inbred strains of mice, or a species always giving birth to multiple offspring), it would appear that there are less confounding effects in these studies and, hence, a lower possibility of arriving at an overestimate.
A final, more minor, point that I would like to make concerns phenotypic plasticity. The situation here is actually even more complex than Charney describes: There is experimental evidence that the amount of phenotypical plasticity that a certain organism is capable of is itself under genetic control (Crusio Reference Crusio2006), an idea that has been around for a considerable time (Hyde Reference Hyde1973).
Summarizing, Charney raises many important points with far-reaching implications for research in human-behavior genetics and offers explanations for a number of its more puzzling findings, from heritability estimates that often are much higher than those found in animal studies to the baffling difficulties encountered when attempting to identify genes for most complex phenotypes in this field, be they psychiatric pathologies or behaviors in the normal range.
It is now 22 years since I wrote my first open-peer commentary for Behavioral and Brain Sciences (Crusio Reference Crusio1990). How much has changed since. And how much has stayed the same. In that commentary, I argued that heritability estimates had only a very limited value, namely, that of predicting the response to selection. Being young and bold (or perhaps young and reckless), I asserted that “these estimates have only a very limited value for researchers investigating animal behavior and are without purpose in human-behavior research.” Significantly older now and perhaps even more reckless but certainly more stubborn, I still maintain that heritability estimates have only a very limited value and I still don't see much use for them in human research.
Perhaps it would be good to recall the definition of heritability: the proportion of the phenotypical variance that is attributable to genetic variation in a given population in a given environment or set of environments. (Depending on whether or not only additive-genetic variation is taken into account, we may distinguish between heritability in the broad sense or in the narrow sense, but that distinction is not important for this discussion.) The crucial words here are “in a given” population or environment. If our population is an inbred mouse strain, heritability will be zero for every measurable phenotype, as there is no genetic variation present. In a population with limited genetic variability, heritability may be relatively small, while a genetically more variable population would show a higher heritability even if it was living in exactly the same environment. Similarly, if two populations had identical levels of genetic variability, but were exposed to different environments, then they would almost certainly also show different heritabilities. In short, in human populations, where we cannot experimentally control either environments or levels of genetic variation, heritability estimates have no inherent theoretical value (as opposed to practical uses, such as determining the number of subjects needed to localize genes). As far as I can see, the only questions of theoretical import would be whether heritability is significantly different from 0 (i.e., genetic effects on the phenotype of interest exist) and from 1 (i.e., significant effects of the environment exist). The often-heard argument that higher heritabilities would mean that a trait (or disorder) will be less responsive to (therapeutic) interventions is obviously flawed.
It should perhaps be noted that the foregoing holds even in the idealized simplified situation where there is no interaction or covariation between genotype and environment. Taking into account also those effects, as well as the molecular processes so skillfully described by Charney in the current target article, really should be the final nail in heritability's coffin. It is true that the effects and processes that Charney describes can be included in quantitative-genetic models: Whatever can be measured can also be modeled. But given the foregoing, such an exercise would seem to be a waste of time and effort. In short, my only beef with Charney's criticism of heritability is the same as the one I raised in my commentary on Wahlsten's (Reference Wahlsten1990) target article, namely, that they both criticize the concept of heritability on technical points, whereas even if those could be solved, there are still much more fundamental inherent problems with the use of this construct in human-behavior genetics. That Charney nevertheless apparently felt that he had to pay the attention that he did to such a useless notion is an apt testimony of the amount of reification that the concept of heritability has undergone in the field of human quantitative behavior genetics.
Despite the foregoing, I would like to bring up one more point about heritability estimates. Animal studies are designed to “boost” heritability: All animals are raised as much as possible in uniform environments in an attempt to reduce the amount of phenotypical variance attributable to variation induced by environmental variations. The end result should be that genetic causes would become relatively more important, leading to higher heritabilities, compared to uncontrolled situations such as encountered in human studies. Yet, studies in human-behavior genetics routinely report heritabilities over 50%, whereas studies in animal genetics generally yield much more modest values, a point that has always puzzled me. The explanation for this paradox may lie in the different processes described by Charney. First of all, although their effects may result in both over- and underestimates of the heritability of a character, the general tendency seems to be to inflate these values. Second, thinking about the way animal studies are set up (e.g., comparisons of inbred strains of mice, or a species always giving birth to multiple offspring), it would appear that there are less confounding effects in these studies and, hence, a lower possibility of arriving at an overestimate.
A final, more minor, point that I would like to make concerns phenotypic plasticity. The situation here is actually even more complex than Charney describes: There is experimental evidence that the amount of phenotypical plasticity that a certain organism is capable of is itself under genetic control (Crusio Reference Crusio2006), an idea that has been around for a considerable time (Hyde Reference Hyde1973).
Summarizing, Charney raises many important points with far-reaching implications for research in human-behavior genetics and offers explanations for a number of its more puzzling findings, from heritability estimates that often are much higher than those found in animal studies to the baffling difficulties encountered when attempting to identify genes for most complex phenotypes in this field, be they psychiatric pathologies or behaviors in the normal range.