Uchiyama et al. address the role cultural variation may play in influencing estimates of the quantitative genetic parameter of heritability, h 2. They argue, rightly in our view, that cultural differences among human groups can contribute an important component of environmental variation in human behaviour. Indeed, given the preponderance of social learning in human populations, it is right to consider how this source of variance may influence our estimates of heritability, by shaping the relationship between genetic and non-genetic sources of variation in behaviour and other traits.
Where we depart from Uchiyama et al.'s view is in terms of their emphasis on the generalisability of heritability estimates. Heritability estimates are always context specific (Falconer & Mackay, Reference Falconer and Mackay1996; Visscher, Hill, & Wray, Reference Visscher, Hill and Wray2008). Even if one is able to identify, and then control for, components of environmental variance (as those studying quantitative genetic parameters in the wild often do: Charmantier, Garant, and Kruuk, Reference Charmantier, Garant and Kruuk2014), one is still left with a context-specific estimate. Importantly, this context specificity is not just in terms of the role of gene-by-environment (G × E) interactions – emphasised in the context of culture by Uchiyama et al. – but also in terms of all the components of variance that go into the heritability calculation, including other sources of environmental variance, and population genetic parameters such as the frequencies of causal alleles segregating in the population. This means that it is a mistake, and a misunderstanding, to expect a generalisable estimate of heritability for any given trait, an aim stated more than once by Uchiyama et al.
The interpretation of heritability is crucial. Uchiyama et al. provide many reasons for not over-interpreting heritability, but these are at odds with imagining that there could be such a thing as a “true” heritability. Heritability only speaks to the sources of variance expressed by a trait in a given sample. In terms of environmental sources of variation, if these are large then additive genetic effects may be swamped, but it does not necessarily mean that they are absent. Behavioural traits are often thought to be contaminated with large sources of environmental variation for example (for discussion see Dochtermann, Schwab, Anderson Berdal, Dalos, & Royauté, Reference Dochtermann, Schwab, Anderson Berdal, Dalos and Royauté2019; Stirling, Réale, & Roff, Reference Stirling, Réale and Roff2002). Similarly, if environmental sources of variation are small, then heritabilities may be high, but again this tells us rather little about the additive genetic variance itself. It is worse than that though, as additive and residual sources of variance may be non-independent (Hansen, Pélabon, & Houle, Reference Hansen, Pélabon and Houle2011; Houle, Reference Houle1992). One alternative is the coefficient of additive genetic variation, CVA, which may be more comparable across contexts (Houle, Reference Houle1992).
As such, differences in heritability can be because of both differences in environmental and additive genetic components (bundling away non-additive effects for brevity), but of course the same is true for similarities: Heritabilities may be alike, but for different causal reasons. Uchiyama et al. talk about techniques such as polygenic scores in terms of unpicking the quantitative genetic basis of traits, but as they note recent research has shown that the causal variants identified by polygenic score methods do not replicate well across populations (including for intensively studied human populations and traits such as height, which are highly heritable across populations: Mathieson, Reference Mathieson2021). This means that even if there was a “true” heritability, repeatable across populations (which there isn't), and we could unpick the cultural influences, it would not necessarily mean that we were looking at the same underlying genetics. And if we are not looking at the same underlying genetics, then what is the purpose of trying to generalise heritability estimates? It is after all uncontroversial that most traits exhibit heritable variation (Lynch & Walsh, Reference Lynch and Walsh1998; but see Blows & Hoffmann, Reference Blows and Hoffmann2005 for complications).
Heritability is a useful statistic, particularly coming into its own in comparative studies across traits and organisms (Mousseau & Roff, Reference Mousseau and Roff1987; Weigensberg & Roff, Reference Weigensberg and Roff1996). Hundreds of studies across animals have told us that morphological, life-history, and behavioural traits typically vary in their h 2 estimates, going from higher to lower respectively. Moreover, within a species, variation in heritabilities with age, for example, can give us hypotheses about (a) how selection acts at different ages, or (b) how developmental processes, and the genes and environments they influence and call upon, change over the lifetime (Wilson, Kruuk, & Coltman, Reference Wilson, Kruuk and Coltman2005). But there is no generalisable, canonical h 2 waiting to be discovered. Developmental processes, via the moment-to-moment interactions of organisms in their environments, do not call on genes in such a way that could generate such a canonical measure; put simply, the whole genome is not scrutinised moment-to-moment by an organism, in its environment. Instead, if one wants such an over-arching genetic perspective, then the molecular basis of traits of interests needs to be considered more directly, one that embraces changes in gene expression, within- and across-tissues, across time, as the organism lives its way through its social, cultural, and other environments.
Why the emphasis on h 2? Given our clarification that a given h 2 estimate says rather little without understanding the underlying sources of variance, and indeed given much of Uchiyama et al.'s discussion, why view h 2 as potentially generalisable at all? We are not sure. Heritability is a fundamentally flawed way of arguing that some traits are more or less “genetic” in origin than others. All phenotypes, including culturally inherited behaviours and artefacts, have a genetic component to them, because the bodies and brains that produce those phenotypes are built by genes living in environments. Heritability does not speak to that aspect of the genetic basis of traits though, it only speaks to the variance in those traits. So, what are we trying to generalise?
For humans, culture is in the environmental mix in terms of sources of phenotypic variance. As such, controlling for cultural exposure may help reveal patterns in variation in h 2 that can lead to interesting hypotheses and further tests. But to over-emphasise heritability, and to imagine that it can be meaningfully generalised, is to misunderstand what it can, and cannot, tell us about the evolution of humans and other organisms.
You have
Access
Uchiyama et al. address the role cultural variation may play in influencing estimates of the quantitative genetic parameter of heritability, h 2. They argue, rightly in our view, that cultural differences among human groups can contribute an important component of environmental variation in human behaviour. Indeed, given the preponderance of social learning in human populations, it is right to consider how this source of variance may influence our estimates of heritability, by shaping the relationship between genetic and non-genetic sources of variation in behaviour and other traits.
Where we depart from Uchiyama et al.'s view is in terms of their emphasis on the generalisability of heritability estimates. Heritability estimates are always context specific (Falconer & Mackay, Reference Falconer and Mackay1996; Visscher, Hill, & Wray, Reference Visscher, Hill and Wray2008). Even if one is able to identify, and then control for, components of environmental variance (as those studying quantitative genetic parameters in the wild often do: Charmantier, Garant, and Kruuk, Reference Charmantier, Garant and Kruuk2014), one is still left with a context-specific estimate. Importantly, this context specificity is not just in terms of the role of gene-by-environment (G × E) interactions – emphasised in the context of culture by Uchiyama et al. – but also in terms of all the components of variance that go into the heritability calculation, including other sources of environmental variance, and population genetic parameters such as the frequencies of causal alleles segregating in the population. This means that it is a mistake, and a misunderstanding, to expect a generalisable estimate of heritability for any given trait, an aim stated more than once by Uchiyama et al.
The interpretation of heritability is crucial. Uchiyama et al. provide many reasons for not over-interpreting heritability, but these are at odds with imagining that there could be such a thing as a “true” heritability. Heritability only speaks to the sources of variance expressed by a trait in a given sample. In terms of environmental sources of variation, if these are large then additive genetic effects may be swamped, but it does not necessarily mean that they are absent. Behavioural traits are often thought to be contaminated with large sources of environmental variation for example (for discussion see Dochtermann, Schwab, Anderson Berdal, Dalos, & Royauté, Reference Dochtermann, Schwab, Anderson Berdal, Dalos and Royauté2019; Stirling, Réale, & Roff, Reference Stirling, Réale and Roff2002). Similarly, if environmental sources of variation are small, then heritabilities may be high, but again this tells us rather little about the additive genetic variance itself. It is worse than that though, as additive and residual sources of variance may be non-independent (Hansen, Pélabon, & Houle, Reference Hansen, Pélabon and Houle2011; Houle, Reference Houle1992). One alternative is the coefficient of additive genetic variation, CVA, which may be more comparable across contexts (Houle, Reference Houle1992).
As such, differences in heritability can be because of both differences in environmental and additive genetic components (bundling away non-additive effects for brevity), but of course the same is true for similarities: Heritabilities may be alike, but for different causal reasons. Uchiyama et al. talk about techniques such as polygenic scores in terms of unpicking the quantitative genetic basis of traits, but as they note recent research has shown that the causal variants identified by polygenic score methods do not replicate well across populations (including for intensively studied human populations and traits such as height, which are highly heritable across populations: Mathieson, Reference Mathieson2021). This means that even if there was a “true” heritability, repeatable across populations (which there isn't), and we could unpick the cultural influences, it would not necessarily mean that we were looking at the same underlying genetics. And if we are not looking at the same underlying genetics, then what is the purpose of trying to generalise heritability estimates? It is after all uncontroversial that most traits exhibit heritable variation (Lynch & Walsh, Reference Lynch and Walsh1998; but see Blows & Hoffmann, Reference Blows and Hoffmann2005 for complications).
Heritability is a useful statistic, particularly coming into its own in comparative studies across traits and organisms (Mousseau & Roff, Reference Mousseau and Roff1987; Weigensberg & Roff, Reference Weigensberg and Roff1996). Hundreds of studies across animals have told us that morphological, life-history, and behavioural traits typically vary in their h 2 estimates, going from higher to lower respectively. Moreover, within a species, variation in heritabilities with age, for example, can give us hypotheses about (a) how selection acts at different ages, or (b) how developmental processes, and the genes and environments they influence and call upon, change over the lifetime (Wilson, Kruuk, & Coltman, Reference Wilson, Kruuk and Coltman2005). But there is no generalisable, canonical h 2 waiting to be discovered. Developmental processes, via the moment-to-moment interactions of organisms in their environments, do not call on genes in such a way that could generate such a canonical measure; put simply, the whole genome is not scrutinised moment-to-moment by an organism, in its environment. Instead, if one wants such an over-arching genetic perspective, then the molecular basis of traits of interests needs to be considered more directly, one that embraces changes in gene expression, within- and across-tissues, across time, as the organism lives its way through its social, cultural, and other environments.
Why the emphasis on h 2? Given our clarification that a given h 2 estimate says rather little without understanding the underlying sources of variance, and indeed given much of Uchiyama et al.'s discussion, why view h 2 as potentially generalisable at all? We are not sure. Heritability is a fundamentally flawed way of arguing that some traits are more or less “genetic” in origin than others. All phenotypes, including culturally inherited behaviours and artefacts, have a genetic component to them, because the bodies and brains that produce those phenotypes are built by genes living in environments. Heritability does not speak to that aspect of the genetic basis of traits though, it only speaks to the variance in those traits. So, what are we trying to generalise?
For humans, culture is in the environmental mix in terms of sources of phenotypic variance. As such, controlling for cultural exposure may help reveal patterns in variation in h 2 that can lead to interesting hypotheses and further tests. But to over-emphasise heritability, and to imagine that it can be meaningfully generalised, is to misunderstand what it can, and cannot, tell us about the evolution of humans and other organisms.
Conflict of interest
None.