History's largest genetic twin study looked at 14.5 million pairs of twins like Marian and Vivan Brown. (Image: Photographed by Christopher Michel/Wikimedia)The largest meta-study of twins in history—which examined 14.5 million sets of twins — found that, generally, genes and environment play an equal role in human development.

“If we pool all data, the balance between nature and nurture is near perfect: across all traits the heritability is 49 percent, and environmental influences account for 51 percent,” senior author Danielle Posthuma, Ph.D., Neuroscience Campus Amsterdam Complex Trait Genetics department head, told Bioscience Technology.

University of Edinburgh geneticist and statistician William Hill, OBE, FRS, FRSE, was uninvolved in the study. He told Bioscience Technology: “This is an important paper, reporting an amazingly large amount of work, doing what seems a sensible analysis, and arriving at useful results.”

University of Oregon biologist Patrick Phillips, Ph.D., also uninvolved in the study, told Bioscience Technology he was “completely blown away” by the size of it, and that it was “fascinating, if it doesn’t really tell us about gene function per se.”

But, he said, “It does tell us something about the genetic basis of these traits. And by and large it is consistent with genome-wide association studies (GWAS). It seems that most complex states are caused by many, many genes with fairly small effects. And when that is true, you expect that most overall effects would be caused by adding up those genes together, the general pattern people have been discovering over the past five or six years.”

Meanwhile, however, another scientist believes far more work needs to be done.

“The study reports a meta-analysis based on the summary-statistics from a diverse set of traits and populations,” Swedish University of Agricultural Sciences computational geneticist Orjan Carlborg told Bioscience Technology. Carlborg was also uninvolved with the study. “It is, however, unclear how to interpret these meta-population results genetically and biologically, as these are computed across diverse populations and traits without accounting for the genetic heterogeneity.”

The study

Posthuma’s team set out to review every published genetic twin study done over 50 years. Published in Nature Genetics, it is the largest meta-analysis in the field, including data on 17,804 traits, measured in 14,558,903 twin pairs, published between 1958 and 2012 in 2,748 papers.

Beyond the overall 50/50 genes/environment find, smaller interesting links were unearthed. The team found that, as thought, bipolar disorder  is 68 percent heritable and schizophrenia is 77 percent heritable, while social values are only 30 percent heritable.

Read More: Potentially Reversible Driver of Aging Found

Key take-home points, Posthuma told Bioscience Technology, are, first, that it is the largest meta-analysis of twin-heritability studies, “with the most reliable estimates of twin correlations, and heritability for hundreds of different traits.” Second: “There was not one trait where the weighted heritability was estimated at zero. Three: For the majority—2/3—of traits we found the monozygous [MZ, or identical] twin correlation was exactly twice the dizygotic [DZ, or fraternal] twin correlation. This suggests that for the majority of traits, all genetic variance is additive.”

Senior author Danielle Posthuma, Ph.D.Among the study’s surprises, said Posthuma: “The field is heavily biased towards investigating a small set of traits that fall under the psychiatric and metabolic domain. We found hundreds of studies investigating the heritability of alcohol abuse, yet very few studies investigating disease like Parkinson’s.”

Also, she said: “The finding that 2/3 of traits follow a simple additive genetic model was surprising. I would have expected, given 'missing heritability,’ to find more evidence for non-additive genetic influences.”

Critical next steps, she said, are twin studies of traits not heavily investigated, and mapping studies consulting the team’s MATCH database “to determine whether a simple additive genetic model, or a more complex non-additive genetic model, is most likely for traits of interest.”


The study matters, Posthuma told Bioscience Technology, as “insight into observed variation in human traits is important in medicine, psychology, social sciences, and evolutionary biology, and has gained new relevance with the ability to map genes for human traits. Individual differences in human traits have been studied for more than a century, yet the causes of human trait variation remain uncertain and controversial. Our study is a thorough analysis of the evidence provided by virtually all published human twin studies of genetic variation underlying individual differences, and of the importance of non-additive genetic variation.”

She noted her study found that, “across all traits, there is strong evidence that genetic variation underlies a substantial proportion of individual differences in the population. We show that the relative influence of genes and environment is not randomly distributed across traits, but clusters in functional domains. We also show that reported estimates of variance from model fitting are likely to be an underestimation of true trait heritability.”

In the new study, for a huge majority of traits, “the observed pattern of twin correlations is consistent with a simple and parsimonious underlying model, where twin resemblance is solely due to additive genetic variation, and not consistent with a substantial influence of either shared environment, or non-additive genetic variation. This is an important observation and relevant for the continuing debate about the contribution of gene-gene interactions to trait variation. What we show is that, across hundreds of traits, there just isn’t any evidence of a substantial proportion of individual differences being due to non-additive genetic variation.”

It should interest professionals and the public alike, she said, as it “provides the most extensive meta-analysis of heritability studies to date, across many disciplines of scientific research, including medicine and the social sciences. The conclusions we draw are general and should appeal to researchers working across multiple disciplines. The results will guide future gene-mapping efforts that rely on assumptions concerning the relative influence of genes and environment, as well as concerning the presence or absence of non-additive genetic variance. Given the comprehensiveness of our study, the sheer volume of data collected and results obtained, and the creation of a user-friendly website with all information, we expect this work to become the standard reference for heritability estimates of a wide variety of complex traits, going across the biological and social sciences.”

She said the website is easy to navigate, allowing meta-analysis on twin correlations, and estimates of genetic and environmental influences on hundreds of traits that can be visualized across sex, age, and population.


Carlborg said the authors "have made an impressive effort to collect twin-correlations and heritabilities from earlier published studies and make them easily accessible for others. This is a commendable effort that will make it easier for others to get a good overview of earlier estimates of quantitative genetic population statistics on the human trait or disease that they are interested in.”

He offered several “howevers.”

On the conclusion we are 50/50 nature/nurture, he was unsure how to interpret results. “Implicitly, the current analyses assume that the analyzed groups of traits are due to the same genetic architectures in all populations,” he told Bioscience Technology. “Although this assumption might be reasonable for low-level biological traits measured in a standardized way in closely related populations, it is unlikely to hold when grouping all traits in all populations to conclude a 49 percent across trait heritability. The estimate of an equal contribution by genetic and non-genetic factors in the shaping trait variability in humans is therefore likely to be very uncertain, if it has any biological relevance at all.” 

On genetics causing phenotypic differences in individuals, Carlborg said, the statistics obtained from the meta-analysis “are further of limited use for inferring the contribution by genes to individual phenotypes. More specifically, as the study is not based on data on individuals, one cannot identify the genetic mechanisms by which genes contribute to individual differences in human traits. To gain such insights, at least data on the genotypes and phenotypes of individuals are needed. The ability to guide future gene-mapping efforts that aim to comprehensively explore the causes of individual differences in human traits in a particular population is therefore limited.”

Then there is additivity. Carlborg told Bioscience Technology that statistical additivity “is not the same as biological additivity.  Statistical additivity, which is the focus in this study, quantifies how much of the total variance—or variation—in a population of twins that can be explained by their average sharing of genes. Resemblances between the twins can, however, arise via many other types of gene-action than by independent, additive contributions by individual genes. In fact, most types of non-additive biological mechanisms important for determining the phenotypes of individuals will lead to substantial statistical additive genetic variance when studied in populations. Hence, the finding here of mostly additive variance for a majority of the traits is a result of the statistical genetic analyses being biased towards inferring additive variance due to the averaging of genetic contributions across individuals and families in a population. The meta-analysis performed here is likely to be even more biased towards statistical additivity than the original studies, as it also aggregates effects across populations and traits.” 

Furthermore, non-additive genetic effects are likely needed for prediction of individual phenotypes, he said. “The additive genetic-effects model assumes that a gene variant has the same effect in all genetic contexts, and hence can be estimated without bias across heterogenous populations where it is present. However, if the gene variant is dependent on the genetic background, specific combinations of genetic variants might have considerably larger or smaller effects in different genetic backgrounds (individuals or sub-populations) than expected from their additive effects obtained by averaging across the whole population. If those kinds of non-additive effects are important, that seriously degrades one’s ability to predict individual phenotypes based on scores derived from the whole population.”

In all, he said, given the inherent bias toward statistical additivity in the study’s analysis, “it is striking that almost a third of the traits are inconsistent with the twin-resemblance being due only to additive genetic variance. This illustrates the need to not only consider alternative strategies to identify the biological mechanisms by which genes contribute to the phenotypes of individuals—and to understand how non-additive genetic mechanisms lead to additive statistical variance in populations. It also illustrates the need to develop experimental designs and analytical approaches that can also utilize the non-additive statistical genetic variance, to identify the genetic contributions that will be missed if focusing only on the additive genetic variance.”


Regarding Carlborg’s reservations over the 50/50 nature/nurture issue, Posthuma and co-author Peter Visscher responded their study offers “an average across all traits, and should be interpreted as such. Thus, it does not make sense to try to link the 50 percent nature for all traits to a specific biological mechanism. It merely means that, across all traits, the influence of nature and nurture is equal. In our study, we also analyzed all specific traits ever investigated with the classical twin study, and we provide estimates of nature vs nurture for each specific trait (see our webtool). These estimates can also be displayed for specific populations, diminishing genetic heterogeneity.”

On the criticism of the idea that genetics cause phenotypic differences in individuals, the Posthuma team responded that “the meta-analysis is based on data from more than 14 million non-independent twin pairs, and more than 28 million individuals, which were analyzed in the classical twin design. In this design we use the known genetic relationship between MZ twins and DZ twins to infer the influence of genetics on a trait, without the need to have actual genotypes. The classical twin design is not suited for identifying genetic mechanisms, and that is also not the goal of the meta-analysis. What the meta-analysis can show, however, is evidence of additive or non-additive genetic influences on a trait, by comparing the MZ and DZ twin correlations.” 

Regarding additivity, the Posthuma crew said: “It is wrong to say that our analyses of published twin studies, or those studies themselves, are somehow biased towards finding additive genetic variation. And the contrast between statistical and biological interaction is misleading. We are interested in explaining individual differences—variation--in the population. If the correlation pattern of the resemblance between relatives--here: MZ and DZ twin pairs--is such that they are linear in the proportion of genes they share by descent, then it means that the variation of average effects of causal variants (= additive genetic variation) is sufficient to explain the data. That does not exclude within or between locus interactions. It just means that deviations from additivity do not explain individual differences in the population. It implies that loci that contribute to complex trait variation can be mapped using simple additive genetic models, and that 'genetic background' is irrelevant for prediction of future phenotypes in the population.  Additive genetic variation has a precise meaning in quantitative (complex trait) genetics. 'Heterogeneity' and 'genetic background' do not.”

Responding to the idea that "non-additive genetic effects” are “needed for prediction of individual phenotypes,” Posthuma and Visscher contended that “the results of our meta-analysis suggest that this is not expected to be of importance.” All told, they said: “It is not correct that the twin design, or our meta-analysis, is biased toward statistical additivity.”

Phillips told Bioscience Technology the study didn’t offer “a particularly powerful test of epistatic [gene-on-gene] interactions." To address that, another important next step may be to start identifying "exactly which chromosome segments are shared between identical and fraternal twins," since many genes are linked together on shared chromosomes in twins, resulting in fewer gene-gene effects.

Still, Phillips concluded: "It is amazing they compiled all this data. There is a huge amount of information here."