In these two large cohort studies, no consistent evidence emerged for additive effects of candidate genes putatively involved in attachment security and disorganization. Thus, the ‘usual suspects’ (Ebstein, Israel, Chew, Zhong, & Knafo, 2010
) in the dopamine, serotonin, and oxytocin systems were not related to attachment quality. Furthermore, proposed risk models for DRD2, DRD4, 5-HTT, and OXTR failed to provide unequivocal results. No effects were found in either study for insecure or disorganized attachment in carriers of the DRD2 minor-T(A1)-allele, DRD4 7-repeat, and A-allele of OXTR. 5-HTT short-allele carriers proved more securely attached in Generation R, but this finding was not replicated in the SECCYD. Previous studies by Gervai and her team (e.g., Lakatos et al., 2000
), Spangler and his colleagues (e.g., Spangler et al., 2009
), and by Barry, Kochanska, and Philibert (2008)
reported genetic main effects and/or interactive effects of genotype and parental sensitive-responsiveness on attachment, but their samples were about four times smaller than each of the current samples. The lack of replication in the two largest attachment samples to date leads us to the conclusion that these earlier studies presented intriguing but insufficiently supported hypotheses.
That said, a co-dominant effect of the COMT Val/Met proved replicable across the studies (a small combined effect of d
= 0.22). In carriers of the Val/Met genotype, disorganization scores were higher compared to both Val/Val and Met/Met carriers, a disadvantage also referred to as negative heterosis (Comings & MacMurray, 2000
). Co-dominant effects for COMT Val/Met have been reported for neurobehavioral functioning (Gosso et al., 2008
; Wahlstrom, et al., 2010
) and schizophrenia (for a meta-analysis, see Costas et al., 2010
). However, these studies showed evidence of positive
heterosis. Molecular heterosis is thought to be biologically plausible. Several studies (e.g. Tunbridge, Harrison, & Weinberger, 2006
) suggest that there is an inverted U-shape with opposing gene expression occurring in heterozygotes compared to the homozygotes. Furthermore, the range of expression of gene products could be greater in heterozygotes, providing a broader window for plasticity or response to stress (Comings & MacMurray, 2000
Evidence from this inquiry might suggest the latter. COMT val/met carriers may be more susceptible to environmental influences, which in turn may increase risk for attachment disorganization provided the small effect identified is not a product of Type 1 error. Of course, the increased susceptibility to the environment might also result in effective gene × environment interactions which we did not find for this genotype. For attachment disorganization we did not assess the most promising candidate environment, i.e. frightening or frightened parenting (Madigan et al., 2006
). An additional explanation might be the involvement of COMT Val158Met in regulation of emotional arousal (Drabant et al., 2006
), which is considered central to disorganized attachment. Disorganized infants inability to regulate stress and emotions in arousing situations is striking, and their dysregulation is an early predictor of later psychopathology (Fearon, Bakermans-Kranenburg, Van IJzendoorn, Lapsley, & Roisman, 2010
; Sroufe, et al., 2005
). As this is the first study that reveals a replicated co-dominant effect of COMT on attachment, further studies are needed that investigate the effects of the COMT val/met genotype in combination with challenging environments, and assess outcomes related to the child's plasticity in emotion regulation.
Genetic pathways are frequently indirect and subject to numerous biological and environmental influences (Ebstein, et al., 2010
; Kendler, 2005
). Several previous attachment GxE studies have suggested that genetic effects may be contingent upon gene-environment co-action (Gervai, et al., 2007
; Spangler, et al., 2009
; Van IJzendoorn & Bakermans-Kranenburg, 2006
; see also Rutter, 2006
). Nevertheless, we did not find GxE interactions that were replicable across the two samples. Previously reported associations for genes involved in attachment (DRD4, 5-HTT) could not be replicated in the two cohorts. The contrast with previous findings might indicate the importance of large samples to test for reliable GxE effects, particularly in case of a phenotype that cannot be assessed without some error.
Population stratification, sufficient power and accurate assessment of the phenotype are crucial methodological aspects (Ebstein, 2006
; Ioannidis, 2007
; Little et al., 2009
). High-quality GxE studies with careful measurement of the environment and the outcome variables are essential, as well as explicit hypotheses about how a specific
gene and a specific
environmental condition interact to predict a specific outcome (Bakermans-Kranenburg & Van IJzendoorn, 2010
). Here the study populations were selected for Caucasian ethnicity, securing an ethnically homogenous sample that might restrict the generalizability of the results but also make them more robust. Although only small single-gene effects were anticipated (Plomin & Davis, 2009
), power was sufficient to detect rather small effects. Furthermore, the phenotype was assessed carefully, as the SSP is the gold standard for assessing attachment quality. Finally, direct replications were possible by using the two largest attachment cohorts with molecular genetic data to date.
Nevertheless, the absence of a replicable G × E effect in explaining variation in attachment security and disorganization may be related to the assessment of the outcome or the candidate environments in the current studies. The assessments of attachment and sensitivity in the SECCYD sample were based on gold standard procedures in this field of inquiry, and they showed the expected co-variation, with an effect size equal to the combined effect size of a series of earlier, smaller studies (NICHD, 2005
; De Wolff & Van IJzendoorn, 1997
). The unexpected association between sensitivity and attachment disorganization found in one of the analyses of the SECCYD data should be taken as a spurious and non-replicated outcome.
In the Generation R study a slightly modified Strange Situation Procedure was used, with pre-separation and separation episodes shortened by one minute each. This modified procedure however was stressful enough to yield the expected distribution of secure and insecure attachments. Moreover, in a previous report on the Generation R study we showed that infant attachment quality was related to cortisol stress reactivity as assessed before and after the SSP, with resistant infants showing the largest increase in cortisol excretion after the SSP and disorganized infants displaying a more flattened diurnal slope than non-disorganized infants (Luijk et al., 2010
), indicating the validity of the procedure. However, in the Generation R sample no significant association between maternal sensitivity and attachment security was found. The lack of association runs counter to meta-analytic evidence on the relation between parental sensitivity and infant attachment security, not only in correlational studies (see De Wolff and Van IJzendoorn, 1997
, though it should be noted that effect sizes were found to be significantly smaller in larger samples) but also in experimental intervention studies (Bakermans-Kranenburg et al., 2003
). We note that the assessment of sensitivity in Generation R was less than optimal as it took place during a rather brief session with simultaneous psychophysiological assessments, and this may have decreased the association between observed sensitivity and infant attachment security.
In terms of predicting attachment, sensitivity to positive signals of the infant in settings in which the parents can fully concentrate on their child might not be the optimal way of measuring this complex construct. Parent-infant interactions in situations with competing demands (Pederson et al., 1990
) might entail more ecological validity, and parental responses to infants' negative or distress signals may be more powerful in shaping attachment (Cassidy, 2008
; Goldberg et al., 1999
; Thompson, 1997
). In both studies the sensitivity assessments did not include these more challenging components of parenting. For attachment disorganization the most important determinant has been found to be frightening or atypical parenting behaviors (Lyons-Ruth & Jacobvitz, 2008; Madigan et al., 2006
). In the current studies this type of parenting has not been assessed. Furthermore, other risk factors in the infants' environment that may lead to attachment disorganization have not been assessed either, such as parental psychopathology (e.g., bipolar depression) or family violence (Cyr et al., 2010
). In samples with more variety in clinical symptoms or in risk environments and with parenting assessments in more challenging settings replicable gene × environment effects might be revealed.
Genetic contributions to attachment may operate in ways not tested in this study. For example, epistatic effects could play a role (e.g., Pezawas et al., 2008
). Before evaluating these gene-gene interactions, more knowledge is needed about functionality and specific pathways of targeted genes. Genome-wide analyses (GWAS) and pathway analyses might uncover genetic associations beyond the usual suspects. Also, effects of deletions or multiplications of larger DNA segments—copy number variations (CNVs)—are known to affect protein expression and gene function. These CNVs might act as vulnerability factors for neurodevelopmental phenotypes (Merikangas, Corvin, & Gallagher, 2009
). Furthermore, epigenetic processes merit consideration, as these can modify gene expression and neural function without changing nucleotide sequence (Van IJzendoorn, Caspers, Bakermans-Kranenburg, Beach, & Philibert, 2010
; Zhang & Meaney, 2010