Welcome to the Genomics Forum blog


Based at The University of Edinburgh, the ESRC Genomics Policy and Research Forum is part of the ESRC Genomics Network and pioneers new ways to promote and communicate social research on the contemporary life sciences.

Monday 1 November 2010

The problem with "heritability"

by David Shenk - Genomics Forum Visiting Fellow August 2010

I want to extend my deep appreciation to the Genomics Forum for hosting me as a visiting fellow in August. In my all-too-brief stay, I interacted with some great minds, and was very much impressed with the Forum's mission to improve the understanding of genetics among the general public. My work in my recent book The Genius in All of Us nicely intersects with that mission.

At the Forum, I was very glad to spend some time with Dr. Wendy Johnson from the University of Edinburgh's Department of Psychology. Dr. Johnson has written some of the most articulate work I've come across about the meaning and limitations of so-called "heritability" studies on intelligence.
 

While these studies have been crucially important in establishing substantial genetic influences on intelligence, the precise nature of that influence is widely misunderstood. Heritability studies have often been misinterpreted as proving that much of intelligence is predetermined by the genes, which is simply not true. Dr. Johnson's paper carefully explains how high heritability can co-exist with individual malleability.

Heritability studies were set up to set up to calculate the respective effects of genes and the environment on various traits, and they were built on the assumption that gene effects and environmental effects are independent of each other. In humans, the most common method is to compare the extent of similarity in identical twins with the extent of similarity in fraternal twins. Since identical twins share close to 100 percent of their DNA and fraternal twins share, on average, 50 percent of the genes that differ among all humans (like any ordinary siblings), comparing these two groups allows for a somewhat tidy statistical calculation.

Unfortunately, while the calculation is straightforward and simple to perform, the interpretation of the result is not. This is a dangerous combination in science. There are big problems with how heritability has been interpreted by some scientists, who have often failed to take into account the independence assumption on which the calculation rests, and who have sometimes improperly inferred that high heritability implies genetic determinism of trait levels. Consequently, there are even bigger problems with how these studies have been reported in the mainstream media and interpreted by the general public.

Dr. Johnson does a marvelous job of addressing the first issue -- the misinterpretation of these studies within the scientific community -- in her recent paper, "What Do Genes Have to Do with Cognition?" She describes two gigantic problems with heritability as it commonly has been measured and interpreted:

  • A. "Heritability" ignores gene-environment interaction.
The statistical model of heritability was created to distinguish genetic influences from environmental influences, and it rests, as Patrick Bateson has said, on "the extraordinary assumption that genetic and environmental influences are independent of one another and do not interact. That assumption is clearly wrong." Johnson expands on this point, and takes us back to R.A. Fisher's original thinking behind heritability:

"Fisher's treatment was mathematically rigorous and his (rather limited) experimental work convinced him that interactions were rare and of little practical importance. Many followed him in effectively disregarding them.
Embryologists and evolutionary biologists, however, took a different view.
To them, phenotypic characteristics were the current products of ongoing series of transactions between environmental circumstances and genetic material, and individual differences in those characteristics could be attributed to genetic or environmental influences only if the developmental conditions were specified. Failures of independence of genetic and environmental influences were not only not just occasional and annoying exceptions to a general rule; they were standard operating procedure in the natural world."

Johnson suggests that there are statistical adjustments that can be made to partly make up for this discrepancy, but that they are difficult and very rarely done.

  • B. Heritability is population and time-dependent.
By definition, any heritability measure is entirely specific -- and limited -- to a precise population and time. The results cannot be applied to any other population or to any individual. Scientists routinely acknowledge this point, but many then rhetorically override it by pointing out that, as Dr.
Johnson herself acknowledges, "heritabilities of similar traits are often very similar across populations, and even across species. This can leave the impression that heritability is robust to population differences." She
continues:

"There are two reasons, however, to think this an illusion. First, the heritability of many traits is moderated by environmental circumstances within populations. For example, heritability of physical health decreased with level of income and level of perceived control over life (Johnson & Krueger, 2005a; Johnson & Krueger, 2005b) in a national US sample.
Emphasizing the importance of remembering that heritability is a ratio [of genetic to total variance], this was because genetic variance decreased with increasing income  In contrast, heritability of life satisfaction increased with level of income (Johnson & Kureger, 2006) due to decreased environmental variance with increasing income.

"The second reason to think that robustness of heritability to population differences an illusion is that heritability estimates tend to gravitate toward the moderate range of perhaps 30-50% for essentially psychometric reasons completely unrelated either to genetic or environmental population circumstances. These reasons include the presence of substantial measurement error in environmental variance and, ironically, efforts to reduce measurement error such as aggregating collections of measurement items and including measurement items that generate a range of response frequencies.
Thus similar heritabilities in different populations can arise for very different combinations of reasons involving both factors intrinsic to the genetic and environmental population circumstances of primary interest and properties of the measurement instruments used to assess the traits."

***

In the mainstream press, "heritability of a trait" is interpreted as "percentage of that trait inherited and dictated by genes." To begin with, this is a profound misunderstanding of the scientific meaning of heritability, which applies at the level of population variance only, and is irrelevant to the level of a trait in any particular individual. When combined with the fact that the use of heritability in science is beset with the problems stated above, we're left with a public that is ill-informed about overly simplistic scientific applications and interpretations. This is not to suggest that heritability studies are useless. But it does suggest that much communication about them is , at present, a significant hindrance to the popular understanding of genes and their influence on the formation of traits.
------------------------------------------------
For anyone interested in the topic, I urge you to read Dr. Johnson's entire paper, “What Do Genes Have to Do with Cognition?" which will appear next year in the book Motivation and Cognition, edited by S. Kreitler, published by Cambridge University Press.

No comments:

Post a Comment