Twin studies of brain, cognition, and behavior

Twin studies of heritability for neurobiological traits

For many of the papers in this volume, and most of the studies reviewed within those papers, the empirical evidence presented and analyzed is primarily about the heritability of, and environmental influences on, individual phenotypes. These data and analyses are an important first step in any study of the causes of individual differences, whether for quantitative traits or for clinical disorders.

As we have learned in the era of genome-wide studies of variation (e.g. genome wide association studies (GWAS)), twin and family study heritability estimates provide an important benchmark against which to evaluate emerging knowledge from genomic research. Without prior twin and family studies, there would likely have been no appreciation of the `missing heritability’ in common variant GWAS results and we might have concluded, prematurely and incorrectly, that heritability was usually very low for complex traits and most familial similarity was attributable to shared family environments. As it is, twin studies have provided us with higher, probably upper bound, heritability estimates and so motivated the pursuit of contributions to heritable variation from rare variants, copy number variation, non-additivity, and other sources. It has recently been demonstrated that the twin and family study estimates of the heritability of adult height can be fully recovered using whole genome sequence data that take proper account of the impact of low frequency and rare variants, especially those in low linkage disequilibrium with common SNP variants used in traditional GWAS (Wainschtein et al, 2019). Thus, the papers in this special issue that provide basic information about the heritability of a range of traits and conditions do provide important benchmark information. These papers include Abramson et al on Emotional and cognitive empathy, Valero et al on Sleep quality and duration, Picardi et al on Attachment, and Twito and Knafo-Noam on Values.

Abramson et al review and meta-analyze twin studies of the genetic and environmental origins of emotional and cognitive empathy: ‘The meta-analysis discovered that emotional empathy is more heritable than cognitive empathy, while cognitive empathy as examined by performance tests is also influenced by shared-environmental effects. In addition, a non-significant trend found for both emotional and cognitive empathy indicated that heritability of empathy might increase with age, a trend that calls for further longitudinal research. These findings set the ground for further psychological and neurophysiological research on the origins of individual differences in empathy, and several directions for how these findings may promote new research questions were proposed.’ Interesting though this research is, it is still largely in the first phase of establishing the heritability parameters, and not advancing to explore mechanisms, multivariate associations, and developmental progressions. The distinction between cognitive and emotional empathy is helpful descriptively, but unfortunately, there is not a bivariate analysis that would allow an examination of genetic and environmental generality and specificity and, further, perhaps help us understand the source of the higher heritability of emotional, as compared to cognitive, empathy. Beyond this, multivariate analyses would be especially useful to provide insights into the association of individual differences in empathy with endophenotypes and with other behaviors and personality dimensions.

Valero et al conclude that ‘Approximately one third of the variation of sleep quality and sleep duration scores, in general population samples, is explained by genetic factors. Moreover, there is a substantial heterogeneity across studies. Therefore, further research is needed to identify the variables underlying these differences.’ Here again the emphasis is simply on heritability of the traits, sleep duration and sleep quantity. While important as a first step, much more interesting are: a) how and why these phenotypes are related to other phenotypes such as drug use (e.g. Winiger et al, 2019), affective disorders, or chronic pain; b) what are the genetic and brain pathways that may be important and will need both genomic (GWAS; e.g. Lane et al, 2017) and twin studies that can interrelate brain imaging and sleep phenotypes, preferably developmentally, such as the Adolescent Brain Cognitive Development (ABCD) study (Iacono et al, 2018) in the US.

Picardi et al report that studies of attachment security in young twin children point to little genetic influence but considerable shared environment, presumably a result of sharing the same parents, while twin studies of adult attachment styles consistently find moderate heritabilities and negligible shared environmental influences as with other behavioral traits. The candidate gene and candidate gene x environment literature is predictably inconsistent, and the GWAS literature to date is inadequate (without replication). As the authors correctly conclude: ‘In order to keep pace with genomic research in other fields, attachment research would require larger samples and new approaches.’

Twito and Knafo-Noam ‘reviewed all published twin studies on human values, classified as representing four higher order values across two bipolar dimensions: self-transcendence versus self-enhancement and openness to change versus conservation. Across most studies, and most values, monozygotic twins correlated more strongly than dizygotic twins, indicating genetic contribution to values. Significant heritability estimates ranged from 24.5 to 85.7%. The effects of the environment shared by family members were generally weaker. Finally, there was a contribution of the non-shared environment for all values.’ ‘The genetic contribution to values suggests a biological infrastructure to the preference of certain values over others. There has been surprisingly little research on the topic.’ This immediately raises questions about endophenotypes, brain mechanisms, personality and temperament, and genetic pathways. What are the implications for theories of values and socialization? Very similar questions arise in the study of political ideology and social attitudes that have, in the past, been assumed to be solely environmentally driven but for which the twin study literature suggests in fact moderate or high heritability (Hatemi et al, 2014).

Twin studies of relationships between phenotypes

Beyond estimating the importance of heritable and environmental contributions to variation or liability to disease, the more interesting and often more clinically or theoretically significant questions are about relationships between phenotypes. The relationships we are interested in might be between brain structure/function and behavior/disorder, but also between behavior (e.g. eating) and physical outcomes (e.g. obesity), or between earlier developmental characteristics (e.g. cognitive ability and personality traits) and a later outcome (e.g. educational attainment). To these aims, data and analyses should be structured from the outset to address multivariate and developmental genetic questions. The need for this approach has been appreciated for at least forty years, and has become the central theme of modern twin study methodology since Neale and Cardon (1992) and the inception of the annual workshops on Statistical Genetic Methods for Human Complex Traits that have been taught in one form or another since 1987 (see https://www.colorado.edu/ibg/workshop for the most recent incarnation). Within this special issue, several papers have taken these questions as their primary focus. These include Khan et al on Pain and mood disorders (depression and anxiety), Malanchini et al on Cognitive ability and educational attainment, Tistarelli et al on ADHD and its comorbidities, and Silventoinen and Konttinen on Obesity and eating behavior.

Kahn et al present an important study of co-occurrence of pain with depression and/or anxiety finding both genetic and environmental overlap, but a larger role for genetics. The paper highlights an important phenotype, pain, and makes a good case for a greater focus on pain in data collection for twin studies. This would have implications for, among other things, the opioid crisis and other treatment approaches for pain management. ‘In this systematic review, the covariation of pain with depression and anxiety was found to have common underlying genetic and environmental factors. Perhaps the most important implication for public health policies is that the relationship between pain and anxiety or depression is unlikely to be causal (i.e. pain determines depression/anxiety or vice-versa). Rather, this covariation appears to stem from shared elements of risk with genetic factors having a preponderant role.’

Malanchini et al discuss the evidence for largely genetic correlation between cognitive ability and education attainment, from both twin studies and GWAS. Importantly, although there is a large degree of overlap, it is not complete, and there is increasing evidence of non-cognitive factors being associated with educational attainment. The pattern of evidence, including from longitudinal studies, suggests that factors like self-control and openness to experience, not traditionally thought of as components of general cognitive ability, also independently predict academic achievement, and that their associations are again largely genetic in origin. In discussing together both the twin study research accumulated over many decades and the more recent GWAS research, they underscore their complementary contributions, which will no doubt become even more important in the future.

Tistarelli et al review the nature and nurture of ADHD and its comorbidities; ‘by comparing monozygotic (MZ) and dizygotic (DZ) ‘cross-twin/cross-trait’ correlations (i.e. between one trait or disease in a twin and another trait or disease in the co-twin), it is possible to assess if and to what extent an observed association is due to shared genetic and environmental aetiologies.’ After reviewing the ADHD twin study literature, it is evident that ADHD is related to many other conditions and characteristics --- reading difficulties, conduct problems, specific structural and functional brain features, risk of exposure to abuse and neglect, Autism Spectrum Disorder, depression and anxiety, substance use and abuse, and even asthma, to name a few --- and that much of this relationship is genetic in origin. Many, but not all, of these comorbidities reflect the association between ADHD and externalizing disorders generally and, in fact, symptoms of ADHD are often included in defining a latent behavioral disinhibition or externalizing spectrum factor. However the authors also go out of their way to note that not all associations of ADHD are with ‘bad’ characteristics and there are positive associations with entrepreneurship for example. Nevertheless, the overwhelming conclusion from their review is that the genetics of ADHD cannot be considered independently of a wide range of comorbid character and cognitive phenotypes.

Silventoinen and Konttinen present and review evidence for a hypothesis that ‘obesity is a neuro–behavioral disease having a strong genetic background mediated largely by eating behavior and being sensitive to macroenvironment’. The evidence for a moderate or even high heritability comes from both twin and family studies (e.g. Maes, Neale, and Eaves, 1997, for a classic analysis) and, more recently, from GWAS studies which, even in genome-wide complex trait analysis (GCTA) or related analyses, report somewhat lower heritabilities. This is not unexpected for all the usual reasons related to common versus rare variation etc, but it is also possible that gene by age interaction contributes to the lower apparent heritability in GWAS that do not take this interaction into account. Evidence for gene by age interaction comes from the new genetic variation emerging after early childhood that Silventoinen and Konttinen highlight, but also from similar evidence relating to young adulthood and middle age (Fabsitz, Carmelli, and Hewitt, 1992). Moreover, heritabilities are, of course, expected to be different in different environments, and Silventoinen and Konttinen suggest that the evidence points to greater heritabilities in more obesogenic environments; these could be characterized as also more ‘permissive’ for obesogenic behavior. But the GWAS evidence is nevertheless striking: there are robust findings like the influence of the FTO gene, and clear evidence of polygenicity --- over 900 SNPs identified and, although these explain only about 6% of variation, the GCTA estimates of heritability from sequence data are as high as 40% (Wainschtein et al, 2019). In terms of mechanisms, the BMI ‘hits’ are enriched in brain, justifying a focus on behavior. Genetics also influences appetite, diet to some extent, and how individuals respond to an increasingly obesogenic environment in developed or developing economies. Silventoinen and Konttinen review the evidence from twin, family, and GWAS studies, showing that relevant behaviors are indeed heritable, although the estimates are less robust than for BMI.

However, the critical questions are about the relationship between BMI and eating behavior and sensitivity to the macroenvironment. Here the relevant data are less numerous or complete, and the results to date have been suggestive but not definitive: ‘the common risk variants for obesity can partly exert their effects through appetite-related eating behavior traits. However, […] these findings are mainly based on cross-sectional data and especially results related to restrained eating need to be replicated. It should also be noted that the magnitude of both indirect (i.e. mediated) and direct associations have been small.’ ‘More rigorous prospective study designs controlling the well-known biases of measuring food intake would be necessary to prove this part of the hypothesis or to show that other behavioral mechanisms are also important when explaining the effect of genes on BMI.’

Co-twin control studies of environmental influences and epigenetics

Perhaps the most common methodological theme across this special issue is the use of the co-twin control design to control for genetic and shared family environmental factors while testing for the impact of environmental factors affecting individuals within a twin pair. This represents one of the many ways in which the study of twins is of value even in the genomic era. The conceptual clarity and uniquely powerful test for an environmental or epigenetic influence, controlling automatically for a wide range of observed and unobserved confounding genetic, biological, and environmental variables, is deservedly becoming much more appreciated by the scientific community. Papers in this special issue that make use of the co-twin control quasi-experimental design include Besteher et al on Brain structure and cognition in schizophrenia, Buscarinu et al on Molecular mechanisms in multiple sclerosis, Delvecchio et al on brain structure and affective disorders, Palma-Gudiel et al on Depression and epigenetics, Squarcina et al on Genetic and environmental determinants of brain morphology and function in the early lifespan, Iso-Markku et al on the association of physical activity with cerebral and cognitive outcomes, and Qihua Tan on the opportunities more generally for studying epigenetics and behavior.

The review by Besteher et al addresses twin studies of biomarkers for schizophrenia, brain volumes in schizophrenia, and progressive brain changes in schizophrenia, and concludes that each of these is heritable. Furthermore, they draw attention to the substantial evidence of overlap between variation in cognitive function and liability to schizophrenia. These findings are based on published papers reporting what are generally quite small numbers of twin pairs ascertained as best as could be done. The problem of finding pairs of twin subjects when the prevalence of schizophrenics who are also twins is somewhere around .01 * .02 = .0002, or 2 in ten thousand, is clearly considerable. One good use of the available twins is through the co-twin control design, not to address genetics but to address environmental difference controlling for genetics. Phenotypic differences, e.g. in brain volume, are observed between the members of a pair discordant for schizophrenia. Such approaches are needed to bridge the gap between endophenotypes and disorders, and offer complementary information to that provided by GWAS.

Buscarinu et al bring twin studies to bear on the potential molecular mechanisms of multiple sclerosis (MS). As for many diseases, MZ twins are substantially more likely to be concordant for MS than are DZ twins, confirming a heritable risk, but the number of discordant MZ twins greatly outnumbers those concordant and almost all DZ pairs are discordant. These observations are consistent with a moderately heritable multifactorial disorder of low prevalence. They also provide ample opportunity to use the discordant twin design to explore environmental etiological factors and non-genetic biomarkers. This approach has been used to study the roles of sun exposure, vitamin D, adipocytokines in cerebrospinal fluid, regulation of gene expression, and gut microbiota. Of these, probably the last mentioned has the most empirical support through an experimental extension of transplanting gut microbiota into a transgenic mouse model of spontaneous brain autoimmunity: ‘microbiota coming from MS twins induced a significantly higher incidence of encephalomyelitis compared to microbiota derived from healthy co-twins’ (Berer et al, 2017). Broadly, this paper underscores the fact that the utility of twins extends beyond heritability estimation to providing experimental material for exploration of environmental risks and endophenotypes, controlling for genetics and family background. The promising leads from a study of gut microbiota in MZ twins discordant for MS may be the most interesting to emerge from such studies to date.

Delvecchio et al review evidence from studies of twins concordant or discordant for bipolar disorder (BD) or major depressive disorder (MDD). Their general conclusion is that ‘the results showed a complex interplay between gene and environment in affective disorders, the evidence seem to underline that both genetic and environmental risk factors have an impact on brain areas and vulnerability to MDD and BD. However, the precise mechanism of action and the interaction between these factors still needs to be unveiled. Therefore, future larger studies on concordant or discordant twins should be encouraged, because this population provides a unique opportunity to probe separately genetic and environmental markers of disease vulnerability.’ Thus it seems that, unfortunately, the empirical data are insufficient for clear conclusions to be drawn. The recurring theme, and this is not a criticism of the worthy attempts to gather the appropriate data, is that the studies to date are too small and too haphazard in methodology to yield robust conclusions.

Palma-Gudiel et al set out ‘to systematically assess all research papers assessing DNA methylation correlates of depression by means of MZ twin studies, and [to] analyze the advantages and disadvantages of each of the methodologies reviewed.’ In doing so, they present a promising use of discordant MZ twins to explore epigenetic associations with disease or disorder. In this case the focus was on depression, but it could similarly be applied to other diseases or disorders. The key advantages over other approaches are the removal of genetic and environmental confounding as potential causes of association and the simultaneously increased precision afforded by pairing of highly correlated individuals (for phenotypes unrelated to the discordance). Limitations include the inability to determine cause and effect, at least in the cross-sectional studies that are currently available. At this stage in the development of these methods, it is not surprising the results are not compelling, but there is clearly great promise for future studies of epigenetics as well as genetics.

Squarcina et al review twin MRI studies on genetic and environmental determinants of brain morphology and function in the early lifespan. Structural brain measures --- cortical thickness, surface area, white matter, gray matter --- show substantial heritability in most studies from a young age (neonatal) onwards. There are some patterns of difference across regions, but it is hard to assess how important these differences are. A consistent trend for those structural measures, in line with behavioral development, is increasing heritability from childhood to adulthood, although other brain characteristics, such as those captured by Diffusion Tensor Imaging (DTI), may show decreasing heritability across development.

Twin studies of functional MRI, using both classical twin study and discordant MZ twin designs lead to the general conclusion that: ‘Overall, these studies provided information on the determinants of BOLD neurobiology, confirming the key role of genetics but also suggesting selective environmental influences on the brain functional circuits underlying cognition.’ ‘The overall evidence from MRI studies on twins suggests the presence of a core genetic component affecting global brain structure from early infancy to adulthood, which is accompanied by differential profiles of genetic expression, progressively interacting with environment, that drive structural and functional brain regional specialization. Environment also plays a key role in brain development by inducing epigenetic changes that can deviate the developmental trajectories of MZ twins.’

As has been stated elsewhere, the current state-of-the-art requires the authors to patch together information from multiple different studies in a way that is far from satisfactory. Hopefully this situation will be superseded by large scale purposefully designed longitudinal studies such as the ABCD study currently underway in the US (Iacono et al, 2018).

Iso-Markku et al review twin studies of the association of physical activity (PA) with cerebral and cognitive outcomes. While almost all analyses of the phenotypic association find that PA and cognition and/or dementia are inversely associated, ‘none of the twin studies has so far been able to show a significant association between PA and cognition or dementia in a co-twin control design controlling for genetic factors and shared environment […], except for Gatz et al. (2006), who were able to find a significant association between midlife PA and dementia other than AD type in a co-twin control design of 55 MZ twin pairs. However, the result was not significant for dementia in general or AD.’ This is a challenging conclusion, seemingly negating a very clear message that ‘physical activity is a good thing, even for your brain’. These kinds of results are counter intuitive and deserve close scrutiny. One suspects that the important effects are in the extremes and that these extremes might not be well represented in discordant twin pairs from a general population sample. Whether this kind of explanation might also apply to PA and cognition is an open question. Or, of course, it may just be that most of the observed phenotypic associations are in fact the result of genetic pleiotropy. It should also be considered that available co-twin control studies tend to have small numbers of pairs discordant for both PA and dementia or cognitive impairment, and may thus have been underpowered to detect possible associations.

Lastly, among the collection of papers emphasizing the use of the co-twin control study, Qihua Tan presents an overview of how to use twins to assess epigenetic changes related to disease, especially utilizing MZ twins discordant for disease. The author and his colleagues have published empirical studies utilizing the co-twin control study, e.g. on rheumatoid arthritis, as well as theoretical and methodological papers, including on the power of the co-twin control design. This work provides a good argument for the largely untapped potential for clinically significant studies using this approach.

Paying attention to sex and zygosity groups in twin studies

Finally, one paper in this special issue that does not fall easily into these three classifications (heritability analyses, bi- or multi-variate analyses, and co-twin control studies), is that by Ahrenfeldt et al examining the Twin Testosterone Transfer hypothesis. This hypothesis suggests that twins in utero are affected by hormones from the embryonic co-twin such that twins from opposite-sex pairs will show resulting differences from their corresponding sex twins from same-sex pairs (and, presumably, from singletons). To summarize their findings, no consistent evidence to support the hypothesis was found for physiological, morphological, reproductive traits, health related traits, or behavioral traits other than some evidence for cognitive traits, although this was inconsistent across studies. The authors note that there was no direct assessment of hormone exposure, with differential exposure being inferred from animal model studies. Despite these largely negative findings, the review highlights the potential information available in mean phenotypes for twins from different sex/zygosity groups that is largely ignored in the twin study literature.

Summary and final word

The papers in this special issue underscore the various ways in which the study of twins can provide important information about the etiology of individual differences in brain, cognition, and behavior, and of vulnerability to diseases and disorders involving these domains. The benchmark information provided by heritability analyses has proven its worth in the genome era, leading to the highly productive scientific questions about ‘missing heritability’. Equally, the focus on environmental variation and epigenetics afforded by both classical twin analyses and co-twin control studies is providing clinically significant information and promises much more once the value of this approach is fully appreciated. Bi- and multi-variate genetic studies of twins are providing an understanding of the role of genetic pleiotropy as well as potential causal relationships between brain and behavior, and developmental genetic longitudinal studies of twins are informing us about gene by age interactions which are often ignored by other approaches. And just when many were prepared to throw out family studies, including those of twins, altogether, there has been a renewed appreciation of the need for within-family tests of polygenic prediction, most efficiently conducted with dizygotic twin pairs, to understand newly recognized limitations and complexities of genotype-phenotype associations (Selzam et al, 2019; Young et al, 2019). In these and other ways, twin studies continue to illuminate neuroscience and biobehavioral research.