On 30th – 31st May 2024 I attended the third European Social Science Genetics Network Conference hosted at Erasmus University in Rotterdam, funded by the Erasmus Research Institute of Management, the European Research Council, the ESRC Research Centre on Micro-Social Change (MiSoC) and Erasmus School of Economics. As an interdisciplinary PhD student with a background in biology and psychiatric genetics, I was interested to hear from researchers at the forefront of my field, and to learn about the use of genetics in social science more broadly.
Key concepts: heritability and prediction of outcomes
A fundamental notion in sociogenomics is that our life chances, be they socioeconomic or health-related, are partly “heritable”, which means that when we look at the population as a whole, some of the variation in these outcomes can be ascribed to genetic factors. The amount of heritability is usually estimated either by family-based studies (which look at the similarity between individuals of varying degrees of relatedness), or by genome-wide association studies (GWAS) which use direct measures of participants’ DNA. Importantly, heritability is not fixed – it can vary in different contexts. In her keynote, Professor Dorret Boomsma highlighted that the heritability of educational attainment varies across different birth cohorts, and by sex.
Another key question in sociogenomics is how best to predict outcomes. Oftentimes, predictors take the form of a polygenic index (PGI; a.k.a. polygenic score) which can be thought of as a measure of underlying genetic propensity towards an outcome, based on GWAS. In health, the predictive power of PGIs is lower than expected based on other ways of estimating predictive power, for example the amount of concordance between identical twins. Professor Boomsma highlighted that predictive power and heritability are different things, such that identical twin concordance is a function of both heritability and prevalence, for example, schizophrenia is around 80% heritable, but its prevalence of 1% leads to a identical twin concordance of around 33%.
It is important to note that heritability and polygenic prediction of behavioural outcomes are complex, controversial topics which warrant ongoing discussion and scrutiny.
The impact of policy
Several presenters were interested in the effects of policy on a range of outcomes, and how these interact with genetic predictors. Alicia García-Sierra and Aysu Okbay found that educational reforms (in the UK and Sweden, respectively) increased school participation but had little to no effect on later socioeconomic outcomes. PGIs and childhood socioeconomic position (SEP) were independently predictive of outcomes, and PGIs exhibited significant interactions with the reforms. Aysu concluded that, while schooling reforms do increase equality, they may not be “the great equaliser” many hoped they would be. Focussing on health, Emil Sorensen found evidence that the UK’s Clean Air Act of 1956 reduced levels of black smoke, improving birth weight by 60g and adult height by 1cm. Changes in birth weight were driven by individuals with higher PGIs for birth weight.
Intergenerational transmission
Many studies approach the timeless question of the extent to which parents impact their children’s life chances a) via genetic endowment and b) by shaping their environment. Using both an adoption study and a multiple-children-of-twins study, Arno van Hootegem demonstrated that both play a role in shaping children’s outcomes, but different study designs produce different effect size estimates. Using MoBa, Alexandra Havdahl found a negative association between mothers’ educational attainment and helping children with school work, possibly due to differing amounts of help required, or spare time available, between families. Using Lifelines, Sjoerd van Alten found that wealth demonstrated the greatest “genetic nurture” (i.e. the apparent effect of parental genotype on offspring’s outcomes), followed by educational attainment, with income the least affected. Professor Boomsma highlighted genetically informed research which points to body mass index (BMI) as a target for reducing healthcare costs. On this topic, Liam Wright examined the association between parents’ BMI and children’s health outcomes. When measured directly, parents’ BMI had a positive association with children’s BMI, but using PGIs as proxies gave a less robust association for mothers and no association for fathers.
Biomarkers: beyond genetics
Professor Boomsma spoke about the potential utility of DNA methylation (DNAm) biomarkers in social science. Her team have found that DNAm scores add power to predictive models when combined with PGIs, but this was only observed in adults, not children, potentially due to differing source tissue. Furthermore, several of the conference posters used DNAm-derived ageing biomarkers as outcomes to examine how stressors “get under the skin”, and Professor Boomsma found that maternal PGIs were associated with DNAm in their offspring.
Genetic predictors: what’s “under the hood”?
Throughout the conference, several presenters indicated that behavioural PGIs can be noisy and imperfect. To uncover the complex mechanisms captured within PGIs, Engin Keser and Evelina Akimova used genomic structural equation modelling (genomic SEM) and instrumental variable GWAS (IV-GWAS), respectively. For each of several neuropsychiatric conditions, Engin identified both a unique genetic component, and a common genetic component which was shared across all included disorders. Evelina’s two-step process allowed her to separate the genetics of occupational status into associations which are mediated by educational attainment and those which are not, identifying a new locus on chromosome 11. Both found that the popular method “GWAS-by-subtraction” yielded comparable results.
Gene-environment correlation
When gene-environment correlation (rGE) is overlooked, it can lead to spurious gene-environment interaction (GxE) effects. Using a novel method designed to model rGE and GxE, Margherita Malanchini found that 14 psychiatric PGIs together explained around 2.7% of the variance in youth mental health in TEDS, while the environment explained up to 9%. Weak rGE was widespread, and GxE played a very minimal role. Using the Swedish Twin Registry, Rafael Ahlskog found that passive rGE (indexed by parents’ educational attainment PGIs) on neighbourhood characteristics was strongest during adolescence, whereas evocative or active rGE (one’s own PGI) was strongest between the ages of 65-80. Using genetic-relationship matrix structural equation modelling (GRM-SEM) to understand the origins of childhood skills in ALSPAC, Beate St Pourcain found that cognitive and social skills each had one genomic and one non-genomic (residual) factor, and that rGE existed between the corresponding genomic and residual factors for each skill.
Neurodevelopmental conditions
Daniel Malawsky and Jitse Amelink both found that the genetic variants associated with neurodevelopmental outcomes were enriched for genes expressed during the prenatal period. Daniel found that many “single-gene” neurodevelopmental conditions are also influenced by common genetic variants. In the form of PGIs, common variants explain around 9% of the variation in IQ while rare exonic variants explain around 2%. Looking at language networks in UK Biobank, Jitse found three genetic loci associated with brain lateralization (the extent to which activity occurs in the right or left hemisphere), one of which was also associated with dyslexia. Using TEDS, Yujing Lin found that teacher ratings of child behaviours support a dyadic model of autism (social and non-social difficulties) while parent ratings favour the triadic model (wherein the social component is divided into verbal and non-verbal domains).
The future of sociogenomics
Professor Boomsma summarised the current state of health research by saying that family history is still the best predictor of health outcomes, but emphasised the benefits of PGIs in healthcare for people who don’t know their family histories. She emphasised the need to increase diversity, not only of participants but also of the outcomes studied, and noted the largely untapped potential of studying half siblings who have been raised apart. Several presenters argued for better scientific communication, rejecting the terms “direct” and “indirect” genetic effects, and emphasising the role of vertical pleiotropy (i.e. when genetic associations are driven by mediating factors upstream of the outcome). As a biologist, I found it heartening to see so many efforts to unpick the mechanisms captured by behavioural PGIs, but I would like to see more cross-disciplinary conversation on this topic.
Anna Dearman is a PhD student funded by the ESRC and BBSRC via the Soc-B Centre for Doctoral Training, based at the Institute for Social and Economic Research at the University of Essex.