In the present study, we aimed to conduct polygenic interaction (sG × E) analyses for adolescent social anxiety symptom development across 3 successive years that involved genetic variation in the oxytocin system and both constraining and facilitative aspects of parenting, that is, parental psychological control and autonomy support. To create polygenic components, we applied a novel analytical approach, that is, Principal Covariates Regression (De Jong and Kiers
1992; Vervloet et al.
2015), to 223 SNPs linked to the oxytocin system in a large sample of adolescents. Using this analytical technique, we identified one polygenic component—consisting of small contributions of many SNPs across multiple genes—that was strongly positively associated with adolescent social anxiety symptoms. Importantly, significant sG × E interactions suggested that higher scores on this polygenic oxytocin component were associated with higher initial levels of adolescent social anxiety in a context of higher parental psychological control or lower parental autonomy support. In contrast, lower scores on this polygenic oxytocin component were consistently associated with lower initial levels of social anxiety regardless of the parenting context. Collectively, these findings suggest that genetic variability in the oxytocin system interacts with both psychologically controlling and autonomy supporting parenting to predict adolescent social anxiety symptoms in line with the diathesis-stress model (Pluess
2015).
Implications for Research on Social Anxiety Symptom Development in Adolescence
The present study expands substantially on current understanding of both genetic and environmental correlates of social anxiety symptom development in adolescence and their interactions. Regarding genetics, we identified five oxytocin-related polygenic components using Principal Covariates Regression. The oxytocin system has only recently been put forward as particularly relevant to symptoms of social anxiety (Gottschalk and Domschke
2016; Neumann and Slattery
2016) and has received some attention in earlier research on the genetic origins of social anxiety (e.g., Notzon et al.
2015; Thompson et al.
2011). Four of the identified polygenic components mainly represented non-random associations of alleles within a gene (i.e., LD structure) and thereby reflect information concerning population stratification (see Online Supplementary Material
1). Importantly, however, one polygenic component showed a strong positive association with adolescent social anxiety symptoms; a “multi-gene” risk component with very small contributions from multiple SNPs located in several genes across several chromosomes. This finding is in line with the prevailing notion of genetic vulnerability to psychopathology derived from GWAS studies (Visscher et al.
2012), which stresses the importance of polygenic approaches that consider many SNPs across different genes, as many genetic variants each appear to contribute only a small amount of risk.
While research on youths’ problem behavior and oxytocin-related genes other than
OXTR is limited, the contribution of both the
PRLR and
OXTR genes to this polygenic component is in line with earlier findings that these genes are linked to restricted affiliative behaviors in autism (Yrigollen et al.
2008). Furthermore,
GABRA6, the other gene that contributed to this polygenic component, has been associated with stress responsiveness, in social situations in particular, in past research (Frigerio et al.
2009; Uhart et al.
2004). As several genes were found to contribute to adolescent social anxiety symptoms, the oxytocin-related genetic basis of social anxiety seems complex. Oxytocin genes are assumed to confer some of their effects through intermediate physiological processes that have an impact on social interactions and social approach behavior, by affecting individuals’ responses to social stress (Bethlehem et al.
2014). However, being a statistical entity representing the combined genetic risk across multiple genetic markers like all polygenic scores, it is unknown through what biological (and potential psychological and social) mechanisms the captured genetic variability in oxytocin genes exactly affect a particular phenotype such as social anxiety. Also, additional research on the exact part of the genetic variability that is captured by the polygenic components uncovered in this study is clearly indicated. Furthermore, it should always be kept in mind that all genes and neurotransmitter systems interact as a whole in the human body and brain. We know that oxytocin is only part of a complex network of neurotransmitters and by focusing on some oxytocin-related genes, we look at a small part of these complex neurobiological circuits and from a specific angle only.
Regarding environmental influences, the current study revealed, in line with previous studies and theoretical models on the development of social anxiety symptoms (McLeod et al.
2007; Spence and Rapee
2016; Wong and Rapee
2015,
2016), that both constraining and facilitative aspects of the parental environment were associated with adolescent social anxiety symptoms. Some models suggest that when parents are highly psychologically controlling this may negatively affect adolescents’ feelings of self-efficacy and thereby be associated with elevated levels of anxiety, whereas parental encouragement of autonomy may positively affect adolescents’ feelings of self-efficacy and thereby be associated with lowered levels of anxiety (Chorpita and Barlow
1998; see also McLeod et al.
2007, for a brief discussion of different models). In our study, these associations between parenting and social anxiety were specifically found in relation to mean-level differences between adolescents, rather than changes in social anxiety symptoms over time. Importantly, parental psychological control and parental autonomy support—indicators of constraining and facilitative parenting, respectively, and assessed through a latent longitudinal multi-informant approach—only showed a modest negative correlation with one another across informants. This fits with conceptualizations of psychological control and autonomy support as independent parenting subdimensions rather than opposite ends of a common underlying continuum (Silk et al.
2003).
Regarding interactions between the polygenic components and the parental environment (sG × E), our findings suggested that the “multi-gene” polygenic oxytocin component interacted with both environmental factors to affect adolescent social anxiety symptoms. In line with the diathesis-stress framework, the combination of high genetic load and less adequate parenting (i.e., high psychological control or low autonomy support) represented a “dual risk” and was associated with the highest levels of adolescent social anxiety symptoms (Pluess
2015). Furthermore, our findings suggested that more adequate parenting, characterized by low constraining or high facilitative parenting, did not seem to protect adolescents’ much concerning their social anxiety symptom development in the presence of high genetic load. Low polygenic load, by contrast, was consistently associated with lower adolescent social anxiety symptoms regardless of the parenting context. Adolescents with low genetic risk thus appeared to be relatively unaffected by both positive and negative aspects of parenting, even though higher parental psychological control and lower parental autonomy support in itself were, as expected, significantly associated with adolescent social anxiety symptoms in our study.
So, whereas genetic risk and less adequate parenting appeared to be independently associated with adolescent social anxiety symptoms, the combined effect of these two types of risk factors should also be considered in sG × E analyses. Our polygenic sG × E findings are in line with (a) the developmental psychopathology framework (Cicchetti and Rogosch
2002), which recognizes that the development of all forms of psychopathology is affected by multiple risk as well as protective factors at multiple levels of analysis that interact with one another in complex ways over time, (b) current models of the development of social anxiety that include such interactions as a key element (e.g., Wong and Rapee
2015), and (c) earlier work on G × E interactions involving social anxiety that relied on a monogenic approach (e.g., Notzon et al.
2015; Reinelt et al.
2014; Thompson et al.
2011).
Strengths, Limitations, and Directions for Future Research
The present study has several important strengths. First, by using polygenic components in our interaction analyses we implemented the latest recommendations in the research field, which involves a gradual shift from a focus on a single polymorphism in a single gene (i.e., candidate gene studies) to polygenic approaches (Belsky and Israel
2014; Dick et al.
2015). Second, by applying Principal Covariates Regression to genetic data, which is an established analytical approach in the field of psychology, we introduced researchers to a novel and versatile analytical tool for future polygenic sG × E studies. Third, by focusing on the relatively unexplored oxytocin system, which may have particular relevance for social anxiety symptoms compared to other psychopathological symptoms (Gottschalk and Domschke
2016; Neumann and Slattery
2016), we expanded on traditional approaches that rely on genetic variants in the dopamine or serotonin systems. Fourth, by taking a longitudinal approach, we took an important step in genetic interaction research in general, because most research so far has been cross-sectional in nature. Fifth and finally, by using a multi-informant latent index of parenting across 3 successive years as environmental factor in our sG × E analyses, we increased the assessment quality of the environmental factor (Wong et al.
2003). Furthermore, we considered both constraining and facilitative aspects of parenting in our interaction analyses. Whereas past research has predominantly focused on processes of risk, focusing on both positive and negative dimensions of parenting improves our understanding of processes of both resilience and risk, which are central processes in the developmental psychopathology perspective (Cicchetti and Rogosch
2002).
Still, our study should be considered in the light of some limitations, which may provide directions for future research. First, the usefulness of Principal Covariates Regression when creating polygenic components for use in sG × E studies is in need of replication in independent samples. Although several auxiliary analyses supported the validity, robustness, and specificity of our findings (see Online Supplementary Material
1), such analyses, valuable as they are, do not preclude the need for independent replications. Importantly, there is a growing need for replication and validation in the field of molecular genetics of complex traits in general (Dick et al.
2015; Halldorsdottir and Binder
2016). As sG × E studies become more complex in design, for example as they adopt a polygenic approach, focus on specific biological pathways, follow their participants over time, and use multi-informant assessments of the environment and/or the phenotype of interest, international and multidisciplinary collaboration in larger cohorts or consortia may become even more important in this field to be able to replicate and validate sG × E findings. At this moment, we have been unable to find studies with characteristics comparable to the key features of our own study to replicate and validate both the usefulness of Principal Covariates Regression for creating polygenic components and our significant sG × E findings.
Second, despite its strengths, this analysis technique has some weaknesses. SNP density within the genes, overall gene size, and—to a certain extent—the underlying LD structure are confounding factors in the type of analyses we conducted. It is no surprise, therefore, that the genes that contain the largest number of SNPs in our study (see Table
S1 in Online Supplementary Material
1) strongly help to define the polygenic oxytocin components that we found. Or, in other words: genes with a small number of SNPs are less likely to define underlying polygenic components. Furthermore, it remains important to examine the composition of the identified genetic components in-depth to check it against known genetic interdependency of the polymorphisms (i.e., LD structure and ancestry) to avoid interpretation of confounding signals. In addition, our polygenic analyses build on the assumption that a genetic sensitivity to a certain phenotype can be detected in a first step, which moderates the association between environmental factors and this phenotype in a second step. This same assumption is also at the heart of the MPS and PRS approaches that our Principal Covariates Regression approach was intended to improve upon. Alternatively, some approaches exist that directly concentrate on significant G × E interactions at the gene or polymorphism level, such as
Genome-Wide by Environment Interaction Studies (GWEIS; Dunn et al.
2016; Van Assche et al.
2017), but studies applying such approaches are still scarce.
Third, care should be taken not to overgeneralize our findings. The data were collected in a particular region of Western Europe on a relatively well-functioning community sample of adolescents with a relatively homogeneous ethnic background. It is as yet unclear whether our results can be extended to adolescents who live in other regions of the world, who have a more diverse socio-economic and ethnic background, and who are more diverse in functioning. Furthermore, interpretation of the associations and interactions observed is limited to the oxytocin system and the particular environmental variables included in our analyses. As indicated, other biological systems, most notably the serotonin and the dopamine systems (e.g., Enter et al.
2014; Gottschalk and Domschke
2016), also appear to play a role in social anxiety symptoms. However, the polygenic approach adopted in the current study, which involves a careful selection of genes within the oxytocin system and a subsequent selection of genetic variants within each of these genes, is generalizable and can easily be applied to any selection of genetic variants of interest. Similarly, other aspects of the parenting environment than the ones we examined, and parental overprotection in particular (Rubin et al.
2009; Wong and Rapee
2016), have also been linked with the etiology of social anxiety. Also, although we included adolescents’ sex, age, and living situation in our analyses as covariates of both the polygenic components and the development of adolescent social anxiety symptoms, there are many other unexamined potentially relevant covariates, also of parenting. Future research should expand on our study focusing on polymorphisms related to other neurotransmitters and other, or additional, parenting variables, as well as including other unexamined potentially relevant covariates, particularly of the parenting variable(s) under study. On a related note, we were unable to control for potential effects of pubertal status or the menstrual cycle of females on adolescent social anxiety symptom development (Reardon et al.
2009; Van Veen et al.
2009), which would be important potential covariates for future research to include.
Fourth and finally, an important challenge for the research field is to incorporate longitudinal assessments of the environmental factor included in sG × E interactions in addition to longitudinal assessments of the phenotype of interest, because we know that environmental factors including parenting are dynamic in nature and may thereby show change over time. Also, it is important to realize that all significant findings, including our own sG × E findings, are mere statistical results that are in need of an explanation in terms of their underlying mechanisms. Only if those mechanisms are understood at the molecular level and linked to increased risk for social anxiety can we further increase our understanding and interpretation of polygenic component-by-environment interactions (Halldorsdottir and Binder
2016).