Examining the association between adult attachment style and cortisol responses to acute stress.

The quality of social relationships may contribute to variations in biological stress responses, thereby affecting health risk. The association between an important indicator of social relationships, adult attachment style, and cortisol has been relatively unexplored. The present study examined adult romantic attachment style and cortisol responses to acute laboratory stress. Salivary cortisol was measured in response to two behavioural tasks, a colour/word interference task and mirror tracing task, in 498 healthy men and women from the Heart Scan study, a subsample of the Whitehall II cohort. Participants were classified as secure, fearful, preoccupied or dismissive on the basis of responses to the Relationship Questionnaire. Cortisol output was lowest in the fearful group, followed by the preoccupied group, with both secure and dismissive groups having higher levels. The results from this study tentatively support the proposition that attachment style is a factor in determining the manifestation of HPA dysregulation.

Introduction

Attachment theory proposes that a biologically based system of behaviours regulates proximity between an infant and caregiver with the goal of increasing the infant's survival in the face of external threats (Bowlby, 1969). These interactions between infant and care-giver are internalised forming enduring cognitive schemas, or “internal working models”, of expectations of care that remain into adulthood (Bowlby, 1969; Hazan and Shaver, 1987; Bartholomew and Horowitz, 1991). Expectations and responses to interpersonal situations learned in these early relationships provide a template for relatively stable and enduring patterns of interpersonal behaviour which are known as the adult attachment style (Ainsworth et al., 1978; Hazan and Shaver, 1987). It is believed that the adult attachment style remains linked to the psychological and biological systems that regulate threat (stressor) appraisal, response and recovery (Bowlby, 1969). To date the main focus of adult attachment research has been on subjective accounts of distress, particularly symptom reporting (Ciechanowski et al., 2002a,b; Kidd and Sheffield, 2005; Wearden et al., 2005; Armitage and Harris, 2006). Little is known about physiological responses to the activation of the attachment system. This paper aims to extend previous work by examining adult attachment using the four style classification model and cortisol responses to acute stress in healthy older men and women.

There are two traditions for measuring adult attachment, the first examines retrospective accounts of parent–child relationships and is assessed using semi structured interview developed by Main and colleagues (1985). In the second, the peer/romantic partner tradition, authors have developed questionnaires to measure how adults with different attachment histories perceive and behave in close relationships (Hazan and Shaver, 1987). These self report measures of adult attachment assign individuals to categories of attachment style or measure the degree to which various dimensions of attachment style are present. Dimensional measures of attachment focus on attachment anxiety (expectation of separation, abandonment, or insufficient love; a preoccupation with the availability and responsiveness of others; and hyper-activation of attachment behaviour) and avoidance (devaluation of the importance of close relationships, avoidance of intimacy and dependency, self reliance, and relative deactivation of attachment behaviour) (Brennan et al., 1998). Although continuous methods of assessing attachment style have been shown to provide a more accurate picture than that available using the four discrete categories (Brennan et al., 1998), categories continue to be used in contemporary research for their ease of administration and interpretation (Maunder and Hunter, 2009), particularly so in large scale studies in which attachment is measured along with many other variables. This is the case in the Whitehall II prospective epidemiological study from which these data were drawn (Bartley et al., 2007). Moreover there is no consensus as to whether attachment phenomena are inherently categorical or dimensional (Ravitz et al., 2010).

Bartholomew and Horowitz (1991) measure reconciles categorical and dimensional models by defining categories that correspond to a combination of extreme positions on the dimensions of attachment anxiety and avoidance. They developed a two-dimensional model comprising of view of self and view of others. Scores on these dimensions produce four possible attachment prototypes. Those classed as secure (low anxiety, low avoidance) hold a positive view of self and others because of the consistent responsive care they received. They are comfortable relying on others, and are easily comforted. Preoccupied individuals (high anxiety, low avoidance) hold a negative view of themselves, but a positive view of others due to inconsistent caregiving. This style is characterised by emotional dependency on others, negative affect, being hyper-vigilant to any potential threats, and having low self esteem. Dismissive persons (low anxiety, high avoidance) have a positive view of self, where they view themselves as resilient and not needing others, but a negative view of others due to early unresponsive care. Although they are uncomfortable being close to others, they have a positive view of themselves. This strategy leads to the denial of attachment needs, avoidance of closeness, intimacy, dependence in close relationships, and self reliance and independence. Finally, fearful individuals (high anxiety, high avoidance) have a negative view of both themselves and others. Akin to preoccupied styles they seek social contact, but in this case are inhibited by fear of rejection. This leads to a behaviour style of approach and avoidance in inter-personal interactions in adult life. In common with preoccupied styles they experience high levels of negative affect and poor self esteem.

Bowlby (1969) believed that adult attachment style remains linked to the psychological and biological systems that regulate threat appraisal, stress response, and recovery from stress. Research has shown that there are differences between attachment styles on threat appraisal during both attachment and non-attachment related stressors (Mikulincer and Florian, 1998; Schmidt et al., 2002). Individuals high in attachment anxiety are hyper-reactive to threats, tend to report greater levels of perceived stress, and also are much more likely to ruminate over the event (Shaver and Mikulincer, 2007). Those high in attachment avoidance are more likely to employ defensive regulation mechanisms, such as repression, to allow them to control unpleasant emotionally stressful situations (Shaver and Mikulincer, 2007). Although past work suggests that attachment insecurity could be an important factor in regulating the perception of threat, less is known regarding any association between attachment style and biological response to stress.

Stressful situations are thought to activate the attachment system (Bowlby, 1969; Mikulincer et al., 2003), and physiological systems are an important mechanism for the expression of stress responses. It has been argued that the hypothalamic–pituitary–adrenal (HPA) axis will be particularly sensitive to attachment processes because it specifically responds in situations that evoke social-evaluative threat (Blascovich and Tomaka, 1996; Dickerson and Kemeny, 2004; Denson et al., 2009), is sensitive to interpersonal situations (Diamond, 2001; Kirschbaum et al., 1995), and shows individual variation in response (Gerra et al., 2001). Stimulation of the HPA axis leads to the release of adrenocorticotropin hormone (ACTH) by the anterior pituitary, which results in the adrenal cortex releasing cortisol into the bloodstream. Although activation of these systems in response to stress is functional, repeated activation of the stress system can ultimately compromise functioning (Taylor, 2010).

There is an extensive literature that supports the existence of a relationship between early care-giving experiences and dysregulation of the HPA axis in animals and humans, whereby adverse early life experience stimulates upregulation or down-regulation of adult cortisol stress responses resulting in dysregulation of stress responsivity (Heim et al., 2000; Luecken and Lemery, 2004; Carpenter et al., 2009; Lupien et al., 2009). Much of the research to date has examined these processes in populations that have experienced trauma or abuse (Repetti et al., 2002; Taylor, 2010). However there is mounting evidence that lower level stressful early life experiences can also compromise the HPA axis pathways such as non nurturant behaviour, neglect, non-responsive parenting, and maternal withdrawal (Taylor, 2010). In a review Chorpita and Barlow (1998) reported that families characterised by low levels of warmth, and high/low restrictions, and controlling parenting had children with a hyper-reactive cortisol response to stress. It is widely believed that over time this chronic response gives way to hypocortisolism as the HPA axis loses its resilience (McEwen, 1998). Thus, the effects of early environment on HPA stress response may occur across a spectrum of early environment quality and involve multiple parameters of the HPA axis, as well as alterations in functioning over time. Not surprisingly, attachment insecurity has been associated with adverse early life experience (Quirin et al., 2008). It has been suggested that human attachment relationships may function to modulate physiological reactions to distress in both children (Gunnar et al., 1996) and adults (Carpenter and Kirkpatrick, 1996; Maunder et al., 2006; Pierrehumbert et al., 2009).

Recently, five studies have been published that examine adult attachment and cortisol response in healthy populations. Four of the studies measure adult romantic attachment using the Experiences in Relationships Questionnaire (Brennan et al., 1998). This is a continuous measure of attachment anxiety and avoidance. Powers et al. (2006) examined 124 heterosexual dating couples’ patterns of cortisol reactivity and recovery in response to a relationship conflict. They found that avoidance in women and a combination of attachment anxiety and avoidance in men were associated with greater stress reactions than was secure attachment. Similarly, Ditzen et al. (2008) examined the role of social support on psychological and cortisol response to stress using the Trier Social Stress Test (TSST) on 63 healthy men. The results suggest an interaction effect for secure attachment and social support for reducing self reported anxiety, but attachment did not directly moderate cortisol response to acute stress. Quirin et al. (2008) examined both the cortisol awakening response and cortisol responses to an acute stressor, which consisted of a cognitive task with unpredictable uncontrollable noise on 48 healthy women. They found a significant positive association between attachment anxiety and cortisol for the acute phase of the study, but a significant negative relationship between attachment anxiety and cortisol awakening response. Rifkin-Graboi (2008) examined attachment and cortisol during an interpersonal challenge based around attachment topics such as death, separation and abandonment, followed by a series of cognitive challenges in 46 healthy males using the Adult Attachment Interview (Main et al., 1985). She found that dismissing individuals had greater cortisol responsivity during the interpersonal challenges but not the cognitive tasks. Cortisol awakening responses were also measured but no associations were reported. Most recently Smeets (2010) investigated the association between attachment, vagal and cortisol response to the TSST task in a sample of 68 healthy young men. No differences were found on any outcome measures.

However, due to the considerable variation between studies, direct comparison is difficult. Part of this variation is due to the type of task employed to evoke a stress response. Some tasks are attachment specific threats such as conflict with a partner (Powers et al., 2006). Others include both attachment and non-attachment related tasks, such as the paced auditory serial addition task (PASAT) (Quirin et al., 2008), and some are non-attachment related (Ditzen et al., 2008; Smeets, 2010). These different approaches to measuring stress reflect a current debate among researchers as to whether attachment is a state or trait characteristic. On one hand attachment behaviours are not always on display and are activated by stressful events, with larger effects being found for attachment specific threats (Powers et al., 2006). On the other hand, there is a trait like consistency to the patterns of behaviour that is evident in these situations, as demonstrated in subjective reports of distress (Ravitz et al., 2010). This suggests that attachment may influence stress responses on a more global level, and not just be limited to attachment specific threats such as inter-personal conflict. Although the physiological research so far seems to favour the state approach (Powers et al., 2006; Rifkin-Graboi, 2008), research is still at a very early stage and further work is needed to assess the processes associated with attachment and physiological activation. Moreover, the research to date has focused on the physiological responses of young adults. However, early life research suggests that stress response can change over time as a result of chronic activation of the stress systems (Repetti et al., 2002; Taylor, 2010). To date these assertions regarding the nature of attachment, be it state or trait activated, have not been examined looking at cortisol response in an older population.

Therefore our hypotheses are as follows: first, those individuals high in attachment anxiety (preoccupied and fearful) will report greater levels of perceived stress in comparison to those low in anxiety (secure and dismissive) during a non-attachment related acute stress task. As findings from previous research are quite diverse it was not possible to develop specific hypotheses regarding cortisol reactivity during an acute stressor. Therefore we have adopted a more general approach. Our second hypothesis is that there will be a difference in cortisol response between attachment styles during a non-attachment related task in an older population.

Method

Participants

These analyses were carried out on a subsample of participants in the Heart Scan study, a psychophysiological investigation of the relationship between socioeconomic and psychosocial factors, physiological stress responsivity, and subclinical coronary artery disease (Hamer et al., 2010; Steptoe et al., 2010). The Heart Scan study was conducted with healthy middle-aged men and women who were members of the Whitehall II epidemiological cohort (Marmot et al., 1991). It involved 543 participants aged 53–76 years, 498 of whom (271 men, 227 women) completed the attachment measure (age range 45–68 years). All procedures were carried out with the written consent of the participants. Inclusion criteria included white European origin, no history or objective signs of coronary heart disease (CHD), hypertension, or inflammatory disease, no history of mental illness, or any medication that might affect cortisol levels, including hormone treatment. Recruitment was stratified by grade of employment in order to investigate socioeconomic status (SES).

Laboratory stress procedures

Each participant underwent a psychophysiological stress testing session in the laboratory. Participants were tested individually in either morning or afternoon sessions in a light and temperature controlled laboratory. Participants were asked to refrain from alcohol and exercise in the evening before or during the day of testing, and not to drink caffeinated drinks, or smoke at least 2 h prior to testing. After a 30 min rest period, a baseline saliva sample was taken. Two non-attachment related behavioural tasks were administered in a random order. The first task was a mirror tracing task, where participants had to trace a mirror image of a star using a metal stylus. Participants were told that the average person completed the circuit five times in the time allowed and they were asked to prioritise accuracy over speed throughout both tasks. The second task was a computerised version of the stroop test, a colour interference task, where a colour word (e.g. blue) is presented on screen in an incongruous colour. Participants were asked to press a computer key that corresponded to the location of the correct colour of the word presented. Each task lasted 5 min, and task order was randomised across participants. Saliva samples were collected immediately after the tasks were completed, and then at 20, 45, and 75 min intervals after the end of tasks for the assessment of salivary cortisol. The samples were collected using Salivettes (Sarsted, Leicester, UK), which were then stored at −30 °C until analysis. Levels of cortisol were assessed using a time resolved immunoassay with fluorescence detection, at the University of Dresden. The intra and inter-assay coefficients of variation were less than 8%. Participants were asked to rate their level of stress at baseline, immediately after each task, and during recovery on a seven point Likert scale ranging from 1 (no stress) to 7 (feeling very stressed). Scores from both stress tasks were aggregated to produce one average stress task score.

Questionnaires

The Relationship Questionnaire (RQ) is a single item measure made up of four short paragraphs, each describing a prototypical attachment pattern as it applies in close adult peer relationships (Bartholomew and Horowitz, 1991). Participants are asked to rate their degree of correspondence to each prototype out of 100 where 100 meant that “the statement describes me exactly.” These ratings provide a profile of an individual's attachment feelings and behaviour. Participants were asked to provide different ratings for each paragraph. The highest of the four attachment prototype ratings can be used to classify participants into an attachment category of secure, fearful, preoccupied or dismissive. The statements are:

It is easy for me to become emotionally close to others. I am comfortable depending on them and having them depend on me. I don’t worry about being alone or having others not accept me. (Secure)

I am uncomfortable getting close to others. I want emotionally close relationships, but I find it difficult to trust others completely, or to depend on them. I worry that I will be hurt if I allow myself to become too close to others. (Fearful)

I want to be completely emotionally intimate with others, but I often find that others are reluctant to get as close as I would like. I am uncomfortable being without close relationships, but I sometimes worry that others don’t value me as much as I value them. (Preoccupied)

I am comfortable without close emotional relationships. It is very important to me to feel independent and self-sufficient, and I prefer not to depend on others or have others depend on me. (Dismissive)

Each person was classified into their attachment style on the basis of their highest score for each paragraph. Participants were excluded if two or more paragraphs had the same rating out of 100. This measure was administered in phase five (1997–1999) of the Whitehall II study (Bartley et al., 2007).

Statistical analysis

The participant characteristics for each attachment style were compared using Chi-square tests for categorical data and univariate analysis of variance for continuous data. The pattern of cortisol over the laboratory session was analysed using individual values, and also by computing cortisol area under the curve with respect to ground (AUCground) for the complete session, using the procedures described by Pruessner et al. (2003). Task scores were aggregated to increase reliability. Cortisol values were skewed and so were log transformed for analysis; geometric means are presented in the manuscript for the sake of comprehensibility and comparison with other studies. The relationships between attachment style and subjective stress and cortisol responses were analyzed using repeated measures analysis of variance with trial as the within-person and attachment style as the between-person factor. The Greenhouse–Geisser correction of degrees of freedom (df) was applied when the sphericity assumption was violated, but unadjusted dfs are presented in Section 3. Additionally, cortisol AUCground was analysed using analysis of covariance. Age, gender, BMI, smoking status, and employment grade as covariates for all the analyses. These factors were included since they are known to influence responses (Hamer et al., 2010; Steptoe et al., 2003). Since half the sessions were held in the morning and half in the afternoon, time of day was included as an additional covariate. Significant difference between means were analysed using Tukey's least significant difference test.

Results

Demographics

Participant characteristics are detailed in Table 1. 236 participants were classed as having a predominantly secure attachment style, 68 fearful, 31 preoccupied and finally 163 were classed as having a dismissive attachment style. There were no significant differences between attachments styles on gender, age, employment grade, or smoking behaviour (p > 0.05). Although all groups would be classed as being over-weight using BMI categorisation, there was a significant difference on total BMI (kg), with those classed as preoccupied having higher BMI values (F (3, 492) = 3.77, p = 0.04) than those with a secure, fearful or dismissive attachment style.

Table 1

Characteristics of participants by attachment style.

	Secure	Fearful	Preoccupied	Dismissive
Age Mean (SD)	54.0(5.68)	52.0(4.61)	52.0(4.43)	55.0(5.80)
BMI (kg) Mean (SD)	25.34(3.73)	25.52(3.55)	27.39(4.93)	25.36(3.73)
Gender
Men N	129(26%)	37(7%)	21(4%)	83(18%)
Women N	107(21%)	31(6%)	10(2%)	80(16%)
Employment grade
Higher N	95(20%)	22(4%)	12(2%)	64(13%)
Intermediate N	90(18%)	26(5%)	16(3%)	66(13%)
Lower N	51(10%)	20(4%)	3(1%)	33(7%)
Current smokers N	19(4%)	6(1%)	2(0.4%)	10(2%)

(%) are in relation to the whole sample.

Attachment style and perceptions of stress

The analysis of subjective stress showed a main effect of trial (F (4, 1952) = 623.5, p < 0.001) and for attachment style (F (3, 488) = 2.72, p = 0.04), but no interaction between the two. Stress ratings increased in response to tasks, returning to baseline in the post-task recovery phase (see Fig. 1). Overall, the average stress rating was lower for participants in the secure group (mean 1.91 ± (SD) 0.66, adjusted for covariates) than in the fearful (2.11 ± 0.72, p = 0.02) or preoccupied (2.19 ± 0.65, p = 0.03) groups, while not differing significantly from the dismissive style group (2.02 ± 0.61, p = 0.09). But subjective responses to tasks did not differ. No differences were reported between attachment styles on ratings of task difficulty (F (3, 494) = 1.75, p = 0.16), involvement in the task (F (3, 494) = 1.14, p = 0.33), perceived performance (F (3, 494) = 1.11, p = 0.34), or control during the tasks (F (3, 494) = 1.26, p = 0.29).

Fig. 1

Mean perceived stress ratings (range 1–7) in the four attachment style groups for baseline, the tasks, and 20, 45, and 75 min following tasks.

Attachment style and cortisol stress responses

Repeated measures analysis of variance across the laboratory session confirmed a significant effect of trial for cortisol (F (4, 1976) = 46.61, p = 0.001, η2 = 0.04). It can be seen from Fig. 2 that the tasks induced increases in cortisol, with a subsequent decline over the recovery period. In addition, there was a main effect of attachment style, (F (3, 494) = 3.59, p = 0.01), but no interaction between attachment style and trial after adjustments for age, gender, grade of employment, smoking, BMI and time of testing. Cortisol output was lowest in the fearful followed by the preoccupied group, with the secure and dismissive style groups having higher levels.

Fig. 2

Cortisol response to acute stress (nmol/l) in the four attachment style groups for baseline, immediately post tasks, and 20, 45, and 75 min following tasks.

The association between attachment style and cortisol was further examined using the cortisol AUCground measure. The geometric mean cortisol AUCground in the fearful group was 409.9 nmol/l, significantly lower than in the secure (473.0 nmol/l) and dismissive (492.3 nmol/l, both p < 0.05), after adjustment for age, gender, grade of employment, smoking status, BMI and time of laboratory testing. The preoccupied group had a cortisol AUCground averaging 449.9 nmol/l that did not differ from other groups.

Discussion

Previous research suggests that insecure attachment is associated with increased perceptions of stress (Ditzen et al., 2008), increased reactivity (Powers et al., 2006; Quirin et al., 2008; Rifkin-Graboi, 2008), and delayed recovery in young adults (Powers et al., 2006). The aims of the present study were twofold: first, to investigate the relationship between attachment styles and perceptions of stress in response to acute behavioural challenge: second, to examine the association between adult attachment styles on cortisol response to an acute stressor in an older population.

Adult attachment style and subjective stress reports

Previous research has shown that preoccupied (high anxiety/low avoidance) and fearfully (high anxiety/high avoidance) attached individuals consistently report more subjective distress than either dismissing or secure groups during both attachment specific threat (Maunder et al., 2006; Ditzen et al., 2008) and non-attachment provoking situations (Kidd and Sheffield, 2005). Correspondingly, those who had a fearful or preoccupied attachment style reported significantly greater levels of perceived stress compared to those with a secure attachment style in the current study. However, it is notable that effects were present across the complete session including baseline and recovery, and not in response to tasks themselves. This may be explained, in part, by the behavioural strategies developed over time to maximise closeness to the attachment figure and aid in the regulation of affect.

The normal function of attachment is to regulate distress (Shaver and Mikulincer, 2007). As insecurely attached individuals are not able to regulate their emotions internally a series of strategies have developed to reduce or manage any distress experienced (Shaver and Mikulincer, 2007). Bartholomew and Horowitz (1991) describe preoccupied individuals as overly dependent on others and in constant need of attention, whereas those who are fearfully attached seek social contact but are inhibited by fear of rejection. This leads to an approach and avoidance style in interpersonal interaction in adult life. It may be that fearful and preoccupied attachment styles, both associated with high levels of attachment anxiety, utilise hyper vigilant strategies and over report so that their distress is acknowledged, and to ensure care will be provided (Ciechanowski et al., 2002a,b, 2003; Mikulincer et al., 2003). In contrast those with a dismissing style, (low anxiety, high avoidance), report less distress than both fearful and preoccupied styles. Again, this is consistent with the research literature that suggests that those who are high in avoidance attachment disassociate themselves from situations that may threaten their autonomy (Mikulincer, 1998).

Adult attachment style and physiological response to acute stress

Although a significant main effect was found for attachment style on cortisol response there was no interaction with trial. These findings are similar to the results obtained for perceived stress, whereby differences between attachment styles on cortisol response are present across the complete session, and not in response to the tasks themselves. Further analysis examining the AUC revealed that the differences between attachment styles were on total cortisol output over the study period. Significant differences were reported between those with a secure and fearful attachment style, and those with a dismissive and fearful attachment style. No differences were reported between secure, preoccupied or dismissive attachment styles on this outcome. Individuals with a fearful attachment style had a blunted cortisol response over the entire trial period. These results are somewhat in contrast to previous work that has found either no association between adult attachment and cortisol response (Ditzen et al., 2008; Smeets, 2010), or cortisol hyper-responsivity to an acute stress task in healthy populations (Powers et al., 2006; Rifkin-Graboi, 2008). For example, Powers et al. (2006) found that men with both high attachment anxiety and avoidance (fearful) and women with high levels of attachment avoidance displayed greater cortisol reactivity during a conflict negotiation task. Rifkin-Graboi (2008) found dismissing attachment was associated with increased cortisol reactivity in a sample of college males. Somewhat in line with our findings, Quirin et al. (2008) reported an association between attachment anxiety and a decrease in cortisol response to awakening.

Our results also indicate an inverse relationship between subjective and objective stress measures for fearfully attached individuals. To our knowledge this association between perceived stress and neuroendocrine response to a stressor has not been described before. Previously, Ditzen et al. (2008) examined perceived and objective stress response to an acute stressor, and although they found that those who were high in anxious attachment reported higher levels of stress as in the case of our study, there was no relationship to cortisol response for the attachment dimensions. So ours is the first study to report that fearfully attached individuals report higher levels of perceived stress but have the lowest cortisol response during an acute stress trial in an older population.

To help explain possible reasons for this association it may be useful to draw on the developing literature relating early life adversity with reduced cortisol responsivity. Dysregulation of the HPA axis represents a potential marker of vulnerability for a variety of stress related diseases and neuropsychiatric disorders (Gervasoni et al., 2004; Miller et al., 2007). There is increasing evidence (Fries et al., 2005; Gunnar and Vazquez, 2006) that stress over the life span results in hypocortisolism, specifically low early morning levels of cortisol and blunted adrenocorticotropic-hormone (ACTH) and cortisol response to acute stressors. Clinical studies of adults who experience even low level early life stress have shown alterations in HPA axis reactivity. Heim et al. (2000) found lower cortisol concentrations to acute stress in people who experience early life adversity. Carpenter et al. (2007) reported diminished total plasma cortisol and ACTH response during a psychosocial stress test for individuals who reported moderate to severe early life stress. It is possible that the participants with a fearful attachment style experienced greater early life adversity than other groups, leading both to their adult attachment style and to attenuated cortisol secretion.

Another possible explanation for our results may be related to the age of the participants in the current work as compared with previously conducted studies. The average age for our group was 54 years compared with 23 years for other studies (Powers et al., 2006; Quirin et al., 2008; Rifkin-Graboi, 2008). Carpenter et al. (2009) and Lupien et al. (2009) have shown that age may be an important factor in determining the magnitude of the stress response. Authors such as Carpenter et al. (2009) suggest the significant interaction of exposure to an early stressor and increasing age on the HPA axis response to a standardised challenge might reflect cumulative or progressive effects of early environment across the life span. Such an interpretation would be consistent with the concept of allostatic load, or long term wear and tear on physiological systems because of chronic stress or demand on stress adaptive biological mechanisms. Lupien et al. (2009) has noted that the aging brain seems more vulnerable to hippocampal damage from prolonged psychosocial stress associated with these processes. Inconsistency in the literature may be due to the degree and chronicity of contemporary stressors in adulthood, and may contribute significantly, particularly if the trajectory from HPA axis hyper responsivity to hyporesponsivity occurs during several decades of stress exposure (McEwen, 2008). Accordingly, it may be that over time specific aspects of attachment strategies, such as chronic hyper-vigilance to perceived stress as found in fearful attachment style, may be a causative factor in blunted cortisol responses to perceived stress.

Limitations and future directions

Caution needs to be taken in interpreting the data presented here. Although cross sectional studies allow us to establish the existence of relationships, they do not allow us to state cause and effect. One possible explanation is that attachment style influences the development of the stress responses. However, other explanations are also possible; for example hyporeactivity may be an adaptive response acquired in childhood maintained into adulthood, and not a result of allostatic load (Gunnar et al., 2009). It is also important to take into account the considerably smaller sample size of the preoccupied group. Those with a preoccupied style are high in attachment anxiety and similarly to fearful individuals report greater perceived stress. It may be that blunted cortisol responses are specific to the fearful group who are high in both attachment anxiety and avoidance. Individuals who are fearfully attached face an approach-avoidance dilemma during times of perceived threat, where the attachment figure (romantic partner) represents simultaneously a haven of safety and a source of threat. A consequence of this behavioural strategy is that the attachment system remains activated, which may result in increased allostatic load over time (Pierrehumbert et al., 2009). However, it cannot be ruled out that the lack of association may be a result of the relatively small number of those with a preoccupied attachment style in the study. In addition this study was limited to white, middle aged participants and so we cannot generalise the results to other age or ethnic groups.

Type of task might also have had an impact on the results. The stroop and mirror tracing tasks were deliberately chosen for the study as they were non-attachment related stress tasks, and they have been shown to reliably produce a cortisol response during stress testing (Dickerson and Kemeny, 2004). We wanted to examine whether attachment is activated during all types of stress, or whether these responses are specific to attachment related threats. If attachment style responses are activated during a non-attachment related stressor it could be argued that attachment is a trait characteristic. Our results tentatively support the hypothesis that attachment might be a trait characteristic because significant differences were found between attachment styles during the trial period. Specifically, individuals with a fearful attachment style had the lowest cortisol response during the trial. Moreover the difference was in overall cortisol level during the study period and not in responses to tasks. On the one hand this lends support to the trait proposition as over the course of the life span fearfully attached individuals, who are constantly alert to any type of perceived threat, develop a hypo-reactive stress response as a result of their attachment system being constantly activated (Shaver and Mikulincer, 2007; Pierrehumbert et al., 2009). On the other hand, this may indicate that the task was not stressful enough, and the task did not produce a stress reaction to produce a change in cortisol levels.

It should be noted that cortisol responses were modest in the present study, and this is a reflection of the stress task used. It may be that attachment is associated with larger cortisol responses only when the stressor is attachment related, thus supporting the state activated approach to attachment (Powers et al., 2006). However, it should be noted that participants in the study were healthy older individuals, not on any medication, and without any disease, so we would not expect to see large cortisol responses (Dickerson and Kemeny, 2004). However, this is the first study to find evidence in support of the trait approach using only a non-attachment related task. Further research should examine these associations using other non-attachment related tasks, such as the TSST, which has been found to elicit consistently larger cortisol responses (Dickerson and Kemeny, 2004).

Much criticism has been levelled at the use of attachment styles instead of dimensions in the research literature. It has been argued that the dimensional approach is more accurate as it is thought that statistical power is lost when continuous variables are converted into categories. However, a recent review suggested that there is no consensus regarding whether attachment is best described analytically as categorical or dimensional (Ravitz et al., 2010). Additionally there is debate over the number of attachment styles, with some investigators preferring the original three style model proposed by Ainsworth et al. (1978) comprising of secure, anxious and avoidant, while others advocate the four style approach where the avoidant style is split into dismissing and fearful (Bartholomew and Horowitz, 1991). Our results strongly support the latter of the two, as there were clear differences between the dismissive and fearful groups. For example, opposing responses were found between fearful and dismissive on both perceived stress and cortisol response. The majority of the work that has been undertaken has used dimensional measures of attachment, and only one of these studies found an interaction between anxious and avoidant attachment (Powers et al., 2006). Interestingly, the interaction was for high avoidance and high anxiety, the fearful group classification. This again suggests that fearful attachment in particular may have a deregulatory effect on the HPA axis. Clearly more work is needed to establish the most appropriate way of measuring adult attachment in clinical research.

The results from this study add to the existing literature in showing differences between attachment styles in endocrine responses to non-attachment related stress. They also suggest that responses are different in an older population compared with the younger populations previously studied. Future work is needed to understand the mechanisms and processes that may be occurring over the life course. One direction that may be useful to pursue is the role of the hippocampus. Prolonged cortisol secretion has been related to destruction of neurons in the hippocampus and to metabolic and immune changes (Pruessner et al., 2010). The hippocampus plays an important role in the regulation of the HPA axis. A recent study by Quirin et al. (2010) reported a link between attachment anxiety and avoidance and reduced hippocampal cell density in healthy young adults. It may be that reduced hippocampal cell density constitutes a neural underpinning of reduced stress regulation ability among insecurely attached individuals, and especially those who are fearfully attached.

The examination of the associations between attachment and stress response in clinical populations may help identify some of these processes, particularly in those who have experienced chronic stress from an early age. Pierrehumbert et al. (2009) also found an inverse relationship between subjective and objective stress reports for those with an unresolved attachment style in a sample of women who had been abused during childhood or adolescence. They postulate that chronic exposure to stress may cause a desensitization of the endocrine system, whereby hypocortisolism may be seen as an exhaustion of the stress response. Taking this into consideration the results from the current study tentatively support the proposition that attachment style is a factor in determining the manifestation of HPA axis dysregulation. Further work is needed to understand the activation of these processes, the impact on the stress systems, and the consequences for health.

Role of the funding source

Funding for this study was provided by the British Heart Foundation (RG/05/006). The British Heart Foundation had no further role in the study design; in the collection, analysis and interpretation of data; in the writing of the report; and in the decision to submit the paper for publication.

Conflict of interest

All authors declare that they have no conflicts of interest.

Contributors

Mark Hamer and Andrew Steptoe designed the study and wrote the protocol. Tara Kidd managed the literature searches and analyses. Tara Kidd and Andrew Steptoe undertook the statistical analyses, and Tara Kidd wrote the first draft of the manuscript. All authors contributed to and have approved the final manuscript.