In this study, we investigated the genomic factors contributing to the co-occurrence and differentiation of behavioural and emotional problems in early life. Our findings revealed systematic genomic signal in both differentiation and total problems. Overall, associations of genetic liabilities to psychiatric and neurodevelopmental conditions with differentiation were at least as strong as for total problems. Furthermore, genomic structural equation modelling indicated that, while the “p-factor” was associated with somewhat higher total levels of behavioural and emotional problems, genetic liability to neurodevelopmental conditions was specifically and strongly associated with differentiation toward behavioural problems. In line with Grotzinger et al. [5], the p-factor accounted for heterogeneous genomic signal, suggesting limited informativeness of a genomic “p” in explaining associations with childhood emotional and behavioural problems. Trio model results indicated that genetic effects were primarily direct, consistent with previous studies [33]. PGS effects on traits associated with behavioural and emotional conditions were mostly unmediated or mediated via total problems, whereas PGS associations with later diagnoses were primarily unmediated. These results enhance our understanding of the genetic underpinnings of different neurodevelopmental processes related to mood and behavioural problems from early in life. They also underscore the value of looking not only at generalised liability such as that which is typically captured by the p-factor, but also domain-level sources of variability for both gene discovery and the investigation of aetiological mechanisms.
We identified genetic correlations between a propensity to develop more behavioural relative to emotional problems in early life with ADHD, autism spectrum conditions, alcohol dependence, and depression. In recent work by Grotzinger and colleagues, these conditions loaded on a shared neurodevelopmental factor at the genomic level [5]. Here, one implication could be that genetic liability to conditions underpinned by neurodevelopmental processes may be associated with differentiation toward behavioural problems in early childhood, which was also supported by our modelling results. However, liability to ADHD could also be the driving factor behind these associations. In the multivariate PGS analyses, where each score is adjusted for all the others, liability to ADHD was the predominant predictor of phenotypic differentiation toward behavioural problems - though it is important to note that such results reflect the size and power of the underlying GWAS as well as differences between the various polygenic liabilities. In genomic SEM model comparisons, we found that the pattern of genetic associations was better explained by the neurodevelopmental factor than the condition-specific genetic components (including ADHD) for differentiation, but not for total problems. This suggests that liability to the conditions loading on the neurodevelopmental factor - rather than ADHD specifically - was the driver of differentiation toward behavioural problems.
An intriguing finding was the notably larger ADHD PGS effect for differentiation than for total problems. Previous studies have found similar or stronger associations between liability to ADHD and a general “p-factor” than specific sub-factors [18, 19, 21, 65]. This apparent discrepancy is most likely due to differences in modelling strategy, or the specific measures included (e.g., the behavioural problem scale included multiple questions related to inattention). Also, children with a high burden of generalised genetic risk may display a broad range of problems from early in life, whereas those who predominantly display behavioural problems may be more likely to have specifically elevated liability to ADHD. Alternatively, genetic liability to ADHD might be associated with increasing behavioural problems across early childhood, and more general aspects of mental health problems later in development [18, 19, 21, 65].
Leveraging a large sample of genotyped parent-offspring trios, we found little consistent evidence of indirect genetic effects on differentiation or total problems with effect sizes of meaningful magnitudes. There seemed to be a small indirect effect of maternal genetic liability to autism spectrum conditions on offspring total problems. These results may be because mothers' liability to autism affects their children’s behaviour. Alternatively, it may reflect how mothers with high liability to autism perceive and report on their children’s behaviour. It is important to note that, although statistically significant, this effect was small. First, one implication is that biases from population phenomena may not necessarily substantially inflate genetic associations with psychiatric traits (supported by converging evidence from multiple lines of research [21, 33, 66]). Only the major depression PGS association with total problems was attenuated in the trio model (as compared to the child-only model), which aligns with recent within-sibship GWAS findings [66]. Second, another implication for future studies is that any indirect effects of specific psychiatric PGS on behavioural outcomes may be small in magnitude [33], suggesting that careful interpretation of their potential practical relevance is warranted. This applies more generally to observational associations between parental psychiatric traits and offspring emotional and behavioural outcomes [67, 68], often assumed to be caused by parenting. If causal parental effects of the magnitude often postulated as explanations for these observational associations existed, we would have expected to see evidence of them as indirect genetic effects here.
Overall, mediation analyses showed the importance of total problem development in early childhood, and to some extent differentiation, as intermediate pathways between polygenic liability and traits associated with behavioural and emotional conditions in middle childhood. Associations between genetic liability and later diagnoses (which had low prevalences in this population-based sample) were mostly direct. An exception was genetic liability to ADHD, which showed sizeable associations with different behavioural and emotional conditions, via differentiation and total problems.
Overall, genetic liability to neurodevelopmental conditions was the most important contributor to early-life behavioural and emotional problems, and their link to later behavioural and emotional conditions. Recent evidence suggests that a distinguishing factor between liability to child and adult mental health problems is the key role of neurodevelopmental processes in childhood, relevant to broad aspects of mental health and not just neurodevelopmental conditions such as autism and ADHD [69]. Future GWAS in children would help to delineate these processes further, as most current GWAS samples consist of adults. Focusing on specific neurodevelopmental processes could increase our understanding of the aetiology of mental health conditions and pave the way for effective prevention and treatment.
Our findings warrant further investigation and replication in other samples [70] and ancestries [71]. Future studies could help elucidate the developmental co-occurrence and differentiation of behavioural and emotional conditions in diverse populations, employ Mendelian Randomisation [72] to identify putatively causal exposures, and examine gene-environment interplay. This would contribute to a more comprehensive understanding of differentiation in childhood and beyond.
Limitations
There are some limitations to our study. First, the estimates may be affected by measurement (un)reliability of the differentiation scores. To address this, we modelled the outcomes using a latent growth process, which partitions out measurement error. Potential limitations of the approach to model differentiation have been addressed in more detail elsewhere [25]. Second, a limitation of all PGS analyses is that the size of the GWAS for each trait influences their predictive power. Therefore, PGS for traits with larger GWAS are more likely to have detectable associations with our phenotypic outcomes. This must be accounted for when comparing the different PGS associations between the 11 psychiatric and neurodevelopmental conditions. We mitigated this issue by conducting multivariate GWAS of the latent growth factors of differentiation and total problems and modelling the overlap with the 11 conditions at the genomic level - via genetic correlations and path estimates. These estimates are much less variable with GWAS power than those derived from PGS-based analyses. Conducting GWAS of longitudinal phenotypes is a relatively novel and promising approach [73], which may increase power relative to cross-sectional GWAS [74]. Finally, our results could be affected by selective attrition. The presence of behavioural problems or ADHD in children has been identified as predictors of attrition in similar cohorts [75], which would attenuate any links with our predictors. We have previously reported limited attrition based on the behavioural and emotional scales in this sample [25]. Here, in part because the slope factor would be most affected by selective attrition, our main focus of interpretation is on the intercept factor.