Insecure attachment during infancy predicts greater amygdala volumes in early adulthood

Introduction

Animal research has established that early stress exposure is a key influence on neural development in regions that are central to affective responding and stress reactivity (Huang & Lin, 2006; Meaney, 2001; Ono et al., 2008; Raineki, Cortes, Belnoue, & Sullivan, 2012). In particular, rodent studies have demonstrated that maternal separation or poor care may result in decreased neurogenesis and glucocorticoid receptor density in the hippocampus (Huot, Plotsky, Lenox, & McNamara, 2002; Ladd, Huot, & Plotsky, 2000), with associated hippocampal volumetric reductions (Meaney, 2001). Early stress exposure has also been linked to altered amygdala morphology and function in animal studies (Caldji et al., 1998); disrupted maternal care in rodents has been associated with precocious myelination (Ono et al., 2008) and enhanced amygdala activation in response to stress (Huang & Lin, 2006; Raineki et al., 2012). These observations are of considerable interest, given the central role that these regions play in emotional processing and responding. The amygdala is a major component of the neural circuitry of emotion, contributing to detection and evaluation of emotionally salient stimuli, as well as the expression of emotional responses such as fear (Whalen & Phelps, 2010). By contrast, the hippocampus exerts strong regulatory control over the hypothalamic–pituitary–adrenal (HPA) axis stress response system, ultimately serving to modulate cortisol output via hippocampal glucocorticoid receptors (Dedovic, Duchesne, Andrews, Engert, & Pruessner, 2009). Although the findings from nonhuman primate studies concerning longer term effects of maternal separation on brain development are more complicated, such observations from the animal literature have stimulated investigation of whether human neural development may be similarly sensitive to the quality of care received in early life (Gunnar & Quevedo, 2007).

Early experience has been examined in relation to human hippocampal and amygdala morphology across several contexts, with some consistent observations emerging. Adults with PTSD who experienced maltreatment in childhood show smaller hippocampi than their nonmaltreated counterparts, although an equivalent effect has not been found in children (McCrory, De Brito, & Viding, 2010). In addition, in a study of normative variations in maternal care, higher levels of maternal support in early childhood were found to be strongly predictive of larger hippocampal volumes in school-age children (Luby et al., 2012). With regard to the amygdala, overt childhood maltreatment has not been linked to morphology, at least as evidenced by the gross alterations indexed by volumetric measurement (McCrory et al., 2010). However, enlarged amygdala volumes have been observed in two samples of orphanage-reared children, where environments are characterized by neglect (Mehta et al., 2009; Tottenham et al., 2010). In addition, Lupien and colleagues found amygdala volumes to be increased in the context of chronic maternal depressive symptoms, an association that was hypothesized to be a consequence of the withdrawn parenting that typically occurs in this context (Lupien et al., 2011). Direct observation of aspects of the parent–child relationship is required to confirm that variations in the early nurturing environment are linked to gross alternations in amygdala morphology.

The attachment bond between an infant and their primary caregiver is considered to be a biologically primed, fundamental aspect of the early caregiving environment (Ainsworth, Blehar, Waters, & Wall, 1978; Bowlby, 1979). Securely attached infants experience interactions with their primary caregivers that are broadly sensitive and responsive to the infant's emotional needs, whereas insecurely attached infants are more often subject to inconsistent, withdrawn or negative/rejecting caregiver responses (Ainsworth et al., 1978; de Wolff & van IJzendoorn, 1997). Correspondingly, different patterns of infant attachment behaviours are displayed. Infants with secure attachment patterns tend to seek proximity when anxious and find contacting effective in reducing their distress. Insecure attachment patterns, where infants may avoid contact, become angry and inconsolable or disorganized in their attachment behaviours, are thought to represent suboptimal behavioural strategies for responding to threat which leave insecurely attached infants more vulnerable to stress than their secure counterparts (Cassidy & Shaver, 2008; Fox & Hane, 2008). Attachment is thus a potentially important mechanism via which the quality of care may influence the developing child's experience of and responding to stress. Consistent with this, several studies have indicated that children who show insecure attachment patterns respond with heightened cortisol responses to separation and to other stressors relative to secure children (Ahnert, Gunnar, Lamb, & Barthel, 2004; Hertsgaard, Gunnar, Erickson, & Nachmias, 1995; Spangler & Grossmann, 1993).

Early experiences of stress that occur in association with attachment insecurity may, in principle, serve to influence neural development in sensitive regions including the hippocampus and amygdala (Gunnar & Quevedo, 2007; Lupien, McEwen, Gunnar, & Heim, 2009). Little direct evidence on this point is available. However, previous work by our group supports the possibility of altered neural outcomes in relation to infant attachment security; in a functional magnetic resonance imaging (fMRI) study, we found evidence of less effective prefrontal neural engagement during attempts to upregulate positive emotions in young adults who were insecurely versus securely attached infants (Moutsiana et al., 2014). Here, in the same longitudinally studied sample, we investigated whether insecure versus secure early mother–infant attachments predict hippocampal and/or amygdala volumes in adulthood, as a preliminary test of the hypothesis that key aspects of the caregiving environment that are captured by attachment status may influence morphological neurodevelopment. Our sample, originally recruited shortly following birth, comprised individuals whose mothers were suffering from postnatal depression (PND; n = 28) together with control participants whose mothers did not have PND (n = 31). Attachment status was assessed at 18 months via direct observation using the Strange Situation, the gold standard in the field; and structural MRI scans were completed at age 22 years. We hypothesized that insecure versus secure attachments would be associated with smaller hippocampal volumes, and/or larger amygdalae.

Several covariates were considered. First, whole brain volume, hemisphere/laterality and gender were taken account of in all analyses, as each of these typically explains important variance in neural structures (Brain Development Cooperative Group, 2011; Free et al., 1995). We also considered maternal depression and participant's own history of depression and anxiety as potential covariates for analyses. Previous research has linked the presence of youth or adult depression (Caetano et al., 2007; Campbell, Marriott, Nahmias, & MacQueen, 2004; MacMaster et al., 2008; Videbech & Ravnkilde, 2004), and also family history of depression (Baare et al., 2010; Chen, Hamilton, & Gotlib, 2010; Rao, Chen, et al., 2010) with reduced hippocampal volumes, although not always consistently (Luby et al., 2012; Lupien et al., 2011; Pannekoek et al., 2014; Rosso et al., 2005). Similarly, there have been observations of enlarged amygdalae in association with familial risk of depression (Lupien et al., 2011; Romanczuk-Seiferth et al., 2014), and with individual symptoms of anxiety (Baur, Hanggi, & Jancke, 2012; MacMillan et al., 2003; Qin et al., 2014; Tottenham et al., 2010), although again there have been some mixed findings (Munn et al., 2007). Given that maternal PND was prevalent in our sample and that significant rates of anxiety and depressive disorders were also identified in our participants by adolescence (Halligan, Murray, Martins, & Cooper, 2007), these factors were considered to be important possible confounds.

Methods

Procedures were approved by the University of Reading and the National Health Service Research Ethics Committees. Participants provided written informed consent prior to taking part.

Results

Sample characteristics are presented in Table 1, by attachment status; age was highly constrained (range 21–23 years), and rate of psychotropic medication use was extremely low. As previously described for this sample, nonsecure attachments were almost exclusively of the avoidant subtype (n = 21) versus ambivalent (n = 1) or disorganized (n = 2); and the PND versus control group contained a higher proportion of participants who were insecurely attached as infants (Murray, Halligan, Adams, Patterson, & Goodyer, 2006). Similarly, there was a trend for total months of maternal depression during the course of the study to be higher for insecure versus secure participants.

Key sample characteristics, namely maternal PND status, total months of maternal depression, and participant's own history of depression or anxiety (present/absent), level of education and socioeconomic status, were screened as potential covariates by examining their relationship with volumetric outcomes. Multiple linear regression analyses were conducted, examining total hippocampal/amygdala volumes in relation to each potential confound, controlling for TIV and gender. Results are presented in Table 2. For hippocampal volume, significant positive associations were observed for maternal total months of depression and the presence of a lifetime history of depressive disorder in the participant, but there were no other significant effects. There was no evidence that amygdala volume was related to any of the covariates. Based on these observations, maternal total study months of depression and participant history of depression were controlled for in analyses of hippocampal volume. Gender and TIV were included as covariates for all analyses.

Infant attachment at 18 months

Infant attachment status was investigated as a predictor of hippocampal and amygdala volumes. First, we completed a 2 × 2 ANCOVA examining hippocampal volume in relation to infant attachment status (secure/insecure) and laterality (left vs. right hemisphere), with TIV, gender, participant history of depression and maternal total months of depression included as covariates. Results are depicted in Figure 1. We found no significant effects of attachment on hippocampal volume, either as a main effect (F_1,52 = 0.12, p = .73, partial η² = .002) or in interaction with hemisphere (F_1,52 = 0.04, p = .85, partial η² = .001).

Second, an equivalent analysis was completed for amygdala volume. Results indicated a significant main effect of infant attachment (F_1,55 = 4.39, p = .041, partial η² = .074). As can be seen in Figure 1, in line with hypotheses, greater amygdala volumes were found in adults who had been insecurely attached infants relative to their securely attached counterparts. There was no significant attachment x laterality interaction (F_1,55 = 2.54, p = .12, partial η² = .044). Given that the presence/absence of maternal PND was a key defining feature of the sample which was strongly related to attachment, we repeated analyses including PND status as a second independent variable. The attachment main effect was retained (F_1,53 = 4.66, p = 0.035, partial η² = .081) and no significant effect of maternal PND emerged (F_1,53 = 0.41, p = 0.53, partial η² = .008). Importantly, the maternal PND by attachment status interaction term was also not significant and the estimated effect size was extremely small (F_1,53 < 0.001, p = .99, partial η² < .001). Thus, we had no reason to believe that associations between attachment status and amygdala volume were particularly present in one or other of the original recruitment groups.

Discussion

In our sample of young adults who had been studied from birth, we found that the quality of the early mother–child relationship was a significant predictor of amygdala volume in adulthood. Specifically, participants who had insecure attachments at 18 months of age had larger amygdala volumes at 22 years than those who had been securely attached. By contrast, we did not find the quality of the early attachment relationship to be a predictor of hippocampal volumes.

Our finding that insecure attachment in infancy predicts amygdala volume in adulthood is consistent with previous reports of enlarged amygdalae, but no changes in hippocampi, in children who experienced early institutional deprivation (Mehta et al., 2009; Tottenham et al., 2010). The majority of insecure attachments (87%) in the current sample were Type A-Avoidant, which is typical of dyads in which mothers are relatively unresponsive to their infant (Ainsworth et al., 1978). As such, our observations complement the conclusions from the postinstitutional adoption studies, by suggesting that normative variations in the extent to which the infant experiences an involved and responsive caregiver may also predict amygdala morphology. With regard to the underlying mechanisms, in principle differences in amygdala volumes may arise as a consequence of more frequent or prolonged activation of the developing child's stress response systems. Not only is attachment conceptualized as both a consequence of less optimal parenting and a behavioural mechanism for regulating resultant stress, but several studies have also found direct evidence that insecurely attached infants show greater stress reactivity, as indicated by cortisol response to separation and other stressors (Fearon, Groh, Bakermans-Kranenburg, van IJzendoorn, & Roisman, 2014; Fox & Hane, 2008). Research from the animal literature demonstrates that early stress can significantly influence amygdala development (Lupien et al., 2009). Based on the current observations, the possibility that similar influences operate in human development warrants further investigation. Nevertheless, other mechanisms should also be considered, particularly given the correlational nature of our study. Greater amygdala volumes may be related to persistent environmental or behavioural effects occurring in association with insecure infant attachment status, may be a contributor to versus a consequence of attachment status, or may reflect underlying, shared genetic influences.

The functional implications of the current findings remain to be determined. Theoretically, attachment insecurity is held to represent a risk factor for a number of adverse outcomes, particularly in the socioemotional domain (Cassidy & Shaver, 2008). Previous observations from the same sample studied here have found mother–infant attachment patterns to be an important predictor of offspring socioemotional adjustment and responding in childhood and adolescence (Murray et al., 2006, 2011). As already described, insecure attachments are also associated with greater physiological reactivity to stress, which is likely to be directly influenced by neural activity in the amygdala (Fox & Hane, 2008; Whalen & Phelps, 2010). Previous research has indicated that greater amygdala volumes are associated with trait negative affectivity (Holmes et al., 2012), sensitivity to negative experiences (Barros-Loscertales et al., 2006; Gerritsen et al., 2012), and with the presence of elevated levels of anxiety in both children and adults (Baur et al., 2012; MacMillan et al., 2003; Qin et al., 2014), including in the aforementioned study of postinstitutionalized children (Tottenham et al., 2010). Such findings are notable, given that meta-analysis of attachment-related outcomes has indicated that avoidant attachments, in particular, are associated with elevated rates of internalizing symptoms (Groh, Roisman, van IJzendoorn, Bakermans-Kranenburg, & Fearon, 2012), and suggest that the larger amygdalae observed in the insecure group in the current study may predispose to elevated sensitivity to stress and/or symptoms of anxiety. Nonetheless, we found no association between participant lifetime history of anxiety or depressive disorder and amygdala volumes in our sample. Moreover, in an fMRI study of the same sample studied here, we also found no differences in amygdala responding during an emotion regulation task between those who were securely versus insecurely attached during infancy (Moutsiana et al., 2014). These observations are certainly not conclusive; the fMRI task utilized in our previous report focused on effortful regulation of emotion and particularly recruited prefrontal versus limbic neural regions; and we have limited power to detect associations between amygdala volumes and the presence of psychiatric disorder. However, direct evidence to show that the volumetric differences in amygdalae we observed in association with infant attachment security are of functional importance is still needed.

We failed to confirm previous reports of links between hippocampal volume and aspects of the caregiving environment, although we note that previous findings have been mixed. Thus, while one study found that maternal support in early childhood predicted larger hippocampal volumes in school-aged children (Luby et al., 2012), a second study found that higher levels of maternal warmth in childhood predicted smaller hippocampal volumes in adolescence (Rao, Betancourt et al., 2010). Such discrepancies are likely to reflect differences in the ages of the samples and the types of developmental experiences studied. It is worth noting that we did find that history of offspring and maternal depression were both related to greater hippocampal volume. However, these findings were in the opposite direction to that reported in some previous research, where smaller hippocampal volumes have been linked to a familial history of depression (Baare et al., 2010; Chen et al., 2010; Rao, Chen, et al., 2010). Given that other studies have also failed to find any association between familial risk of depressive disorder and hippocampal volumes (Luby et al., 2012; Lupien et al., 2011), or have identified larger hippocampi in association with risk status (van Erp et al., 2012; Romanczuk-Seiferth et al., 2014; Ladouceur et al., 2008), the current evidence base does not allow for clear conclusions.

The current study has some key strengths, including the longitudinal design and the rigorous observational assessment of attachment status. There are also some limitations. We have applied overall volumetric measurement, which is a gross indicator of neural morphology that potentially misses fine-grained structural differences that might be detectable in larger samples, or with more sophisticated (e.g. multimodal) or higher field strength scanning protocols. Moreover, findings are essentially correlational and causal influences cannot be inferred. Finally, there are some limitations in our ability to generalize findings; insecure attachments were almost exclusively of the avoidant type and studies of insecure ambivalent or disorganized attachment groups may yield different findings; and the sample had some unique characteristics, including the low risk profile, and the inclusion of a group with maternal PND. Nonetheless, the current analyses provide a proof of principle, demonstrating that the caregiving experiences captured by early attachment status can predict amygdala structure in adulthood. It remains for future studies to evaluate the significance of our observations and to explore the environmental and/or genetic factors that may have contributed to them.