Association between the serotonin transporter promoter polymorphism (5-HTTLPR) and adult unresolved attachment

Abstract

Research on antecedents of organized attachment has focused on the quality of caregiving received during childhood. In recent years, research has begun to examine the influence of genetic factors on quality of infant attachment. However, no published studies report on the association between specific genetic factors and adult attachment. This study examined the link between the 5-HTTLPR promoter polymorphism of the serotonin transporter gene and adult unresolved attachment assessed with the Adult Attachment Interview. Genetic material and information on attachment-related loss or trauma were available for 86 participants. Multivariate regression analyses showed an association between the short 5-HTTLPR allele and increased risk for unresolved attachment. Temperament traits and psychological symptoms did not affect the association between 5-HTTLPR and unresolved attachment. The authors hypothesize that the increased susceptibility to unresolved attachment among carriers of the short allele of 5-HTTLPR is consistent with the role of serotonin in modulation of frontal–amygdala circuitry. The findings challenge current thinking by demonstrating significant genetic influences on a phenomenon previously thought to be largely environmentally driven.

Introduction

Research analyzing childhood sibling similarities on attachment largely supports common environmental influences on the organization of attachment (Bakermans-Kranenburg, Van IJzendoorn, Bokhorst, & Schuengel, 2004; Bokhorst et al., 2003; O’Connor & Croft, 2001; Van IJzendoorn et al., 2000). For disorganized infant attachment, common environmental influences have not been demonstrated (Bakermans-Kranenburg et al., 2004; Bokhorst et al., 2003; O’Connor & Croft, 2001; Van IJzendoorn et al., 2000). Recent analyses of adult siblings produced similar findings, with substantial similarities between siblings on organized representations of attachment but not disorganized attachment (Caspers, Yucuis, Troutman, Arndt, & Langbehn, 2007; Constantino et al., 2006). The lack of common environmental influences on disorganized attachment suggests genetic variability as a potential source of influence on adult disorganized attachment. Therefore, this study examines the serotonin transporter promoter polymorphism (5-HTTLPR) as a potential candidate gene in the susceptibility to disorganized attachment in adulthood.

The examination of genetic contributions to disorganized attachment is not without precedence. The long variant of the dopamine D4 receptor (DRD4) gene has been shown to significantly predict infant disorganized attachment (Lakatos et al., 2000), although not all studies have been consistent (Bakermans-Kranenburg & Van IJzendoorn, 2004). The DRD4 polymorphism has also been found to moderate the intergenerational transmission of disorganized attachment whereby maternal unresolved attachment predicts disorganized infant attachment only among carriers of the 7-repeat DRD4 polymorphism (Van IJzendoorn & Bakermans-Kranenburg, 2006). A recent study introduces further complexity to understanding the interplay between genes and environment in attachment (Gervai et al., 2007). Disruptive maternal affective communication has been proposed as a mechanism for intergenerational transmission of disorganized attachment (see Lyons-Ruth, Bronfman, & Parsons, 1999; Madigan, Moran, Schuengel, Pederson, & Otten, 2007). Gervai et al. (2007) examined the interaction between DRD4 gene polymorphisms and disruptive maternal affective communication in the prediction of infant disorganized attachment. Higher rates of infant disorganized attachment were found when the 7-repeat DRD4 allele was absent.

The above findings on associations between specific genes and infant disorganized attachment demonstrate the complexity of identifying candidate genes and highlight the necessity for theory-driven hypotheses bridging genetic and attachment research. Determination of adult disorganized attachment, herein referred to as unresolved attachment, relies on expert examination of discourse patterns elicited during recollection of experiences of loss and trauma (Main & Goldwyn, 1998). Shifts in discourse and reasoning patterns may represent lapses in consciousness, undue influence of overwhelming emotions, or inappropriate interference of memories surrounding the event such that speech is no longer actively being monitored (Hesse & Main, 2000; Hesse & Van IJzendoorn, 1999). For example, an individual might speak about a deceased loved one as though the person was still alive (i.e., indicating a lapse in reasoning). Another individual might speak about a traumatic event in such detail that it suggests a loss of awareness of the immediate purpose and context of the discourse (i.e., an indication of lapse in thought). This affective modulation of language suggests that the cognitive processes of adults with disorganized attachment may be influenced by alteration in the neurobiological processes governing emotions (Phillips, Drevets, Rauch, & Lane, 2003). Therefore, we constructed our hypotheses combining evidence about genetic variability influencing susceptibility to environmental stressors and on the neural structures subserving emotional response (Main, 1999).

Serotonin (5-HT) is a major neurotransmitter involved in emotion regulation (Ressler & Nemeroff, 2000). 5-HT removal from thesynaptic cleft is largely achieved through the activity of 5-HT transporter. The amount of 5-HT transporter is influenced by the 5′ promoter region regulating the transcription of the 5-HTT gene. The promoter contains a polymorphic region with a variable number of tandem repeats (5-HTTLPR), with the short allele responsible for less efficient production of the 5-HT transporter (Collier et al., 1996). The discovery of the 5-HTTLPR promoter polymorphism has led to important developments on predisposing factors for psychopathology (Canli & Lesch, 2007; Ebstein, 2006; Lesch et al., 1996). For example, profound alterations in the functioning of the 5-HT system (documented as lower cerebrospinal fluid 5-HT metabolite) have been shown in animal and human carriers of the short allele in response to chronic or acute stressful experiences (Bennett et al., 2002; Williams et al., 2003). Greater severity of mood disorder and behavioral phenotypes indicative of psychopathology are reported among carriers of the short allele of 5-HTTLPR who experienced stressful life events (Caspi et al., 2003; Fox et al., 2005; Kaufman et al., 2004; Kendler, Kuhn, Vittum, Prescott, & Riley, 2005; Wilhelm et al., 2006), although not all findings are consistent. Finally, Suomi and colleagues(Champoux et al., 2002; Suomi, 1999, 2003, 2006) have shown a significantly greater impact of maternal deprivation among infant rhesus monkeys who are carriers of the risk allele.

These data support the role of the 5-HT system in modulating emotional response to environmental stressors. Progress on the neurobiology of arguably one of the most overwhelming of human emotions (i.e., fear) allows the development of specific hypotheses on the genetic mechanisms underlying unresolved–disorganized attachment. Theamygdaloid complex is a central station for processing emotionally charged stimuli, especially frightening socially relevant stimuli (Adolphs, Baron-Cohen, & Tranel, 2002; Phelps & LeDoux, 2005). Application of neuroimaging to the study of attachment in humans has highlighted the role of the amygdala during positive and negative attachment experiences (Buchheim et al. 2006; Leibenluft, Gobbini, Harison, & Haxby, 2004; Lemche et al., 2006). Recent studies in humans have helped to show how the association between the 5-HTTLPR short allele and stress-related emotional reactivity may be displayed at the neural level. Carriers of the short allele (s/s or s/l) of 5-HTTLPR exhibit greater amygdala activity in response to salient frightening stimuli compared with individuals homozygous for the long allele (l/l; Hariri et al., 2002, 2005). Similar findings are reported in four separate samples of healthy participants (Bertolino et al., 2005).

The present original research aims to determine the extent to which the short variant of the 5-HTTLPR allele predicts unresolved adult attachment. On the basis of the influences of the 5-HTTLPR polymorphism on amygdala reactivity, we predicted that individuals who are carriers of the short 5-HTTLPR allele would show greater shifts in discourse and reasoning patterns as a result of poor emotion regulation and would thus demonstrate higher rates of unresolved attachment. On the basis of the above-summarized literature, we further hypothesized that carriers of the short 5-HTTLPR allele would only display differential coherence of discourse during discussions of loss or trauma. In addition, short allele carriers were not predicted to be differentially exposed to loss or trauma experiences. Studies have typically converged on dominanceof the short 5 -HTTLPR allele (Ebstein, 2006; Canli & Lesch, 2007). We chose to directly test the assumption of dominance because some studies have shown “dose-related” effects of the short 5-HTTLPR variant in addition to dominance effects (Caspi et al., 2003; Kaufman et al., 2004; Wilhelm et al., 2006). Because alteration in 5-HT metabolism has been posited to affect personality and mood disorders (Phillips et al., 2003; Ressler & Nemeroff, 2000), we examined the association between 5-HTTLPR genotype and unresolved attachment with mood and personality measures.

Methods

Participants

Participants for this study were enrolled as part of a large adoption study consisting of adoptees separated from their biological parents at birth. The average age at adoption was 2.42 months (SD = 6.39 months) with 71% of the adoptees placed with the adoptive parents before 1 month of age and 82% before 3 months of age. Adoptees were originally selected for participation on the basis of the psychiatric diagnoses of their biological parents. Adoption agency and institutional (e.g., hospital and prison) records were reviewed by board-certified psychiatrists to determine the diagnoses of the biological parents (e.g., alcoholism, antisocial behaviors; for review of the methods, see Yates, Cadoret, & Troughton, 1999). Adoptees were classified as a proband when a diagnosis was present in either biological parent or as a comparison when no diagnosis was present in either biological parent. Interviewers were naive to the psychiatric history of the biological parents of all the participants. Adoptive families were predominantly upper (20%) and middle class (76%). Average adoptee household income was $40,000 to $49,999 per year. Participants were predominantly White, non-Hispanic (n = 81; 91%), with the remainder of the participants African American, non-Hispanic (n = 4; 4.5%); African American, Hispanic (n = 1; 1%); Caucasian, Hispanic (n = 1; 1%); or mixed race (n = 2; 2%).

Procedure

All procedures were approved by the Carver College of Medicine, University of Iowa Internal Review Board. The Adult Attachment Interview (AAI; Main& Goldwyn, 1998) was administered, transcribed, and coded for 217 individuals between the years of 2000 and 2004.

Age of participants at the time of interview ranged between 28 and 62 years (M= 40 years, SD = 7.63), and 52% were women. The interviews were anonymously assigned and coded blindly by raters who were trained to be reliable to the coding standards of the laboratory of Mary Main and Eric Hesse (Rebecca Yucuis and Kristin M. Caspers were trained by D. Jacobvitz, Austin, TX, 2000; Jeanne Frederickson and Beth Troutman were trained by J. Sroufe, Minneapolis, MN, 1999 and 2001). Roughly 50% of all interviews were rated by two coders. If there was disagreement between coders and consensus could not be reached, a third rater was selected. Interrater agreementwas 93% for the unresolved –not unresolved classification (κ= .71, p < .001). The intraclass correlation, computed with exact agreement methods, for unresolved loss or trauma was .76 and for coherence of transcript was .77. The frequency distribution of unresolved attachment differed significantly from expected rates, χ²(1, N= 86) = 13.35, p = .001, signifying that in our sample unresolved attachment was overrepresented (Van IJzendoorn & Bakermans-Kranenburg, 2008).

The molecular genetic component of the present study began in June of 2001, at which time newly recruited participants were asked to provide buccalswabs for genetic analysis. A total of 111 participants with coded AAIs were asked to provide consent for genetic analysis, and 89 agreed. The difference in the distributionof unresolved versus not unresolved attachment between individuals who provided consent and those who did not was not statistically significant, χ²(1, N= 111) = 0.21, p = .65. Men and women provided cheek swabs at equal rates, χ²(1, N= 111) = 3.11, p = .08. The present study is based on data of only adoptees who provided genetic data and had completed the AAI (N= 89). Among participants who provided cheek swabs, 86 reported experiencing loss or trauma and 3 did not. Because reporting a loss or trauma is a necessary condition for exhibiting characteristics of unresolved attachment on the AAI, data analysis was restricted to those individuals who reported experiencing loss or trauma (final N= 86). Race (e.g., Caucasian, non-Hispanic vs. other) was not associated with unresolved attachment, χ²(1, N= 86) = 0.00, p = .98, or 5-HTTLPR genotype,χ²(1, N = 86) = 0.06, p = .97.

Measures

Adult attachment

Adult attachment representations were derived using the AAI (Main & Goldwyn, 1998). Interviews were audiotaped, transcribed verbatim, and scored with the standard AAI classification system (Main & Goldwyn, 1998). Participants are asked to provide five adjectives for their childhood relationship with their mother and father. Participants are then asked to provide experiential support for the descriptors (e.g., a detailed recount of personal events). Questions about parental responses during episodes of emotional upset, illness, and injury are also probed. Participants are asked about experiences of loss or trauma. On the AAI, loss is defined as deaths of individuals who were important to the interviewee (occurring at any point during their lifetime). Trauma is defined as maltreatment by parents (occurring during childhood) or overwhelmingly frightening experiences (occurring at any point during their lifetime). Finally, the individual is asked to describe the degree to which his or her current feelings differ from past feelings toward his or her parents.

Transcripts were scored using 9-point scales. A coherence score was also assigned on the basis of overall narrative consistency. In addition to the coherence score, three primary classifications representing organized attachment were derived: dismissing, autonomous, and preoccupied (see Main, 2000). When significant, albeit brief, lapses in discourse were observed during descriptions of loss or trauma, participants were classified as having unresolved attachment. Examples of speech patterns indicative of unresolved attachment included (a) change to the present tense when describing the dead person and/or indication that the dead person was still playing an active role in the participant’s life, (b) excessive detail surrounding the event of death or trauma, (c) identifying the self as causing the death of the loved one, (d) viewing themselves as deserving of abuse, and/or (e) reporting extreme reactions to experiences of loss or trauma.

All but 3 participants with available genetic data reported at least one experience of loss. Although there are several possible reasons why the majority of the individuals in this sample reported experiencing at least one major loss, the most plausible explanation is that the chance of experiencing the death of a loved one increases with age. The average age of individuals in this sample was 40 years at the time of the AAI interview, and their adoptive parents’ median age was 30 years at the time of the adoption. Indicators of unresolved speech were evaluated independently for each loss.

Trauma was only scored if the experience met specific criteria (e.g., hitting that is inappropriate or induces pain, leaving bodily marks, overwhelmingly frightening parental rage directed toward the child or in the presence of the child) and sufficient information was available to determine the experience was abusive or overwhelmingly frightening. Sixteen transcripts met these criteria. An unresolved scale score (1–9) was independently assigned to each passage of loss (UL) and trauma (UTr). An overall unresolved scale score (UO) was determined from the highest rating across all loss and trauma. Participants were classified with unresolved attachment when the overall score for unresolved loss or trauma was 6 or above. Transcripts with borderline unresolved scale scores (i.e., 5) were reviewed by at least two coders, and the final classification of unresolved or not unresolved was determined through conference.

Temperament traits

The Schedule for Nonadaptive and Adaptive Personality (Clark, 1995)was used to assess temperament traits. Participants indicated whether each of 375 items accurately described them. We used the following three primary temperament scales: Negative Temperament (28 items), Positive Temperament (27 items), and Disinhibition (35 items). We calculated T scores using published norms (Clark, 1995). Cronbach’s alphas within each of the three scales were good and ranged from .82 to .92.

Mood disorder symptoms

The Brief Symptom Instrument (Derogatis, 1996) is a short form of the Symptom Checklist-90-Revised and assesses dimensions of psychological health. Participants rated on a 5-point Likert scale (0 = not at all, 1 = a little bit, 2 = moderately, 3 = quite a bit, 4 = extremely) the degree to which they experienced symptoms of depression, anxiety, and interpersonal sensitivity in the previous 7 days. The Brief Symptom Instrument was administered following administration of the AAI as a measure of concurrent mood disturbance. We derived T scores for symptoms of depression from published, gender-specific adult nonpatient norms. Cronbach’s alphas within each of the scales were adequate (α= .79 to .89).

5-HTTLPR

Buccal swabs were obtained using Cytotech brushes (Medical Packaging Corp, Camarillo, TX). These swabs were assigned a study code, stripped of all other participant identifiers, and stored in a refrigerator at 4 °C until analyzed. DNA from these swabs was prepared using a QIAmp DNA minikit (Qiagen, Inc., Valencia, CA). Genotyping of the 5-HTTLPR locus was carried out using the primers F- GGCGTTGCCGCYCYGAATGC and R-GAGGGACTGAGCTGGACAACCAC (Persico et al., 2000). Vent polymerase was used according to manufacturer’s suggestion (New England Biolabs, Ipswich, MA) and 100 μmol/L 7-deaza guanosine triphosphate (Boehringer Mannheim, Indianapolis, IN) was added to aid amplification through this GC-rich region. Cycling parameters were as follows: 98 °C × 15 s, 68 °C × 15 s, and 72 °C × 45 s, with a 7-min final extension at 72 °C. Approximately 3 μL of each of the above polymerase chain reaction products were denatured, then loaded on a standard 6% polyacrylamide sequencing gel and electrophoresed for 2 to 3 hr. The gels were exposed to standard X-ray film and the visualized polymerase chain reaction products sized by comparison to an internal sequencing ladder. Gels were read independently and naively with respect to participant behavioral outcome or genetic background. Frequencies of the 5-HTTLPR alleles (s and l) and genotypes (s/s, s/l, and l/l) are shown in Table 1.

Table 1.

Frequency of 5-HTTLPR Alleles and Genotypes

Measure	Allele		Genotype			Total
	s	l	s/s	s/l	l/l
Frequency	72	100	21	30	35	86
Percentage	42	58	24	35	41	100

Genotypic frequencies were in Hardy–Weinberg equilibrium for the s/s, s/l, and l/l genotypes, χ²(2, N= 86) = 3.56, p = .17.

Statistical Analyses

Univariate associations between study variables were tested. Nonparametric tests (e.g., Kruskal–Wallis, Mann–Whitney U, Spearman rho correlations) were used for nonnormally distributed variables. The association of unresolved attachment with 5-HTTLPR genotype was tested using the dichotomous classification (i.e., not unresolved, unresolved) and the 9-point scales to avoid spurious effects (Eaves, 2006). A secondary set of analyses explored individual characteristics (temperament, psychological symptoms) that may be associated with 5-HTTLPR genotype or influence its association with unresolved attachment. Finally, we used regression analyses to examine the influence of control variables (including sex, age, age at adoption, and biological parent diagnosis) on the association between 5-HTTLPR genotype and unresolved attachment. For the logistic regression analyses, 5-HTTLPR was entered as three-level categorical variable (s/s, s/l, or l/l), with the l/l genotype designated as the indicator variable. For the linear regression analyses, we tested the association between unresolved attachment scale scores and 5-HTTLPR genotype using two dummy coded vectors. The first vector, hereafter referred to as homozygous short, coded the s/s genotype as 1 and the s/l or l/l genotypes as 0. The second vector, hereafter referred to as heterozygous short, coded the s/l genotype as 1 and the s/s or l/l genotypes as 0. The two 5-HTTLPR dummy variables were entered simultaneously in the regression model. For the logistic and linear regression analyses, we compared the s/s and s/l 5-HTTLPR genotypes using a second dummy variable (s/s = 0 and s/l = 1). Finally, we tested dominance of the short 5-HTTLPR allele by constructing a third dummy variable where s/s or s/l 5-HTTLPR genotypes were coded 1 and the l/l 5-HTTLPR genotype was coded 0. The final dominance contrast was tested using both logistic and linear regression.

Results

Sample Characteristics

Descriptivestatistic sfor study variables are presented in Table 2.

Table 2.

Study Variables for Total Sample

Variable	N	M (SD) or %	Skewness(SE)	Kurtosis (SE)
Age	86	36.89 (7.03)	0.91 (0.26)	0.63 (0.51)
No. of loss-trauma events	86	3.23 (1.39)	0.28 (0.26)	−0.29 (0.51)
AAI unresolved attachment (present)	25	29
AAI scale scores	86
Overall unresolved		3.58 (2.15)	0.21 (0.26)	−1.28 (0.51)
Unresolved loss	84	3.45 (2.17)	0.30 (0.26)	−1.28 (0.52)
Unresolved trauma	16	3.22 (2.10)	0.50 (0.56)	−1.04 (1.09)
Coherence of transcript	86	4.76 (1.68)	−0.16 (0.26)	−1.15 (0.51)
Sex
Male	39	45
Female	47	55
Biological parent diagnosis
Control	51	59
Proband	35	41
Age at adoption
Less then 1 month	59	70
1 to 3 months	10	12
3 to 6 months	8	10
Greater than 6 months	7	8
SNAP
Negative Temperament	82	44.19 (11.48)	0.81 (0.27)	−0.43 (0.53)
Positive Temperament	82	47.95 (10.60)	−0.72 (0.27)	−0.43(0.53)
Disinhibited Temperament	82	42.39 (8.48)	0.65 (0.27)	1.20 (0.53)
BSI
Interpersonal sensitivity	81	50.63 (10.17)	0.93 (0.27)	−0.19 (0.53)
Depression	81	52.73 (10.17)	0.66 (0.27)	−0.57 (0.53)
Anxiety	81	49.04 (10.47)	0.82 (0.27)	−0.25 (0.53)
Hostility	81	51.32 (10.02)	0.35 (0.27)	−0.49 (0.53)

Note: Biological parent diagnosis includes alcohol problems and/or antisocial behaviors. AAI = Adult Attachment Interview; SNAP = Schedule for Nonadaptive and Adaptive Personality: BSI = Brief Symptom Instrument.

The types of losses ranged from loss of a parent (n = 24; 28%), grandparent (n = 63; 73%), other family member (e.g., aunt, uncle, cousin; n = 27; 31%), and/or a nonfamily member (e.g., friend, coworker; n = 27; 31%). The average ages at the time of each type of loss were as follows: parent (M = 31.63 years, SD = 8.48; minimum, maximum = 20, 50), grandparent (M = 14.89 years, SD = 8.49; minimum, maximum = 2, 39), other family member (M = 20.26 years, SD = 9.26; minimum, maximum = 6, 36), and nonfamily member (M = 19.41 years, SD = 9.36; minimum, maximum = 8, 44). There were no participants reporting loss of a parent in childhood. Types of traumas, as defined by the AAI manual, consisted primarily of excessive physical punishment (e.g., leaving welts or bruises; n = 14; 16%). Four participants were rated for the following overwhelmingly frightening events with no corresponding parental abuse: burglary (n = 1), car accident (n = 2), and combat (n = 1). The average age at the time of trauma was 10.17 years (SD = 5.30). Participants reported an average of 3.24 losses (SD = 1.40; minimum, maximum = 1, 7; n = 86) and an average of 1.28 traumatic events (SD = 0.54; minimum, maximum = 1, 3; n = 16).

Univariate Associations Between Participant Characteristics and Unresolved Attachment and 5-HTTLPR Genotype

Means and standard deviations arepresented for continuous variables, and cell frequencies (percentages) are presented for nominal variables. For comparisons involving unresolved attachment, t tests were used for normally distributed continuous variables, Mann–Whitney U comparisons for variables not normally distributed, and chi-square analyses for nominal data. For tests of 5-HTTLPR genotype associations, we used overall F tests for normally distributed variables, Kruskal–Wallis tests of association for variables not normally distributed, and chi -square tests for nominal data. Finally, nonparametric tests were used for comparisons of the 9-point unresolved scale scores and included Mann–Whitney U, Kruskal–Wallis, and Spearman correlations.

AAI unresolved attachment classification

Descriptives for study variables by unresolved and not unresolved attachment are presented in Table 3.

Table 3.

Study Variables by Unresolved Attachment Classification

Variable	Not unresolved		Unresolved		Test statistic
	N	M (SD) or %	N	M (SD) or %
Age	61	37.85 (6.11)	25	40.72 (8.69)	r(84) = −1.74
No. of loss-trauma events	61	3.08 (1.33)	25	3.60 (1.47)	r(84) = −1.59
AAI coherence or transcript	61	5.21 (1.65)	25	3.66(1.21)	r(84) = 4.24***
Sex					χ2(1) = 4.28*
Male	32	52	7	28
Female	29	48	18	72
Biological parent diagnosis					χ2(1)= 0.78
Control	38	62	13	52
Proband	23	38	12	48
Age at adoption					χ2(3) = 3.55
Less then 1 month	44	73	15	63
1 to 3 months	5	8	5	21
3 to 6 months	5	8	3	13
Greater than 6 months	6	10	1	4
SNAP
Negative Temperament	58	43.62 (12.04)	24	45.82 (10.68)	z = −1.25
Positive Temperament	58	47.97 (10.27)	24	47.68(11.84)	z = −0.27
Disinhibited Temperament	58	41.65 (8.87)	24	44.54 (7.67)	z = −1.63
BSI
Interpersonal sensitivity	57	49.58 (92.00)	24	53.17(10.66)	z = −1.27
Depression	57	51.74 (10.39)	24	55.63 (9.60)	z = −1.59
Anxiety	57	48.39 (10.35)	24	50.50(11.17)	z = −0.65
Hostility	57	50.31 (10.18)	24	52.42 (9.89)	z = −0.72

Note. Biological parent diagnosis includes alcohol problems and/or antisocial behaviors. The z statistic is from the Mann-Whiney U test. AAI = Adult Attachment Interview: SNAP = Schedule for Nonadaptive and Adaptive Personality: BSI = Brief Symptom Instrument.

^*p <.05.

^***p <.001.

Overall coherence of transcript was lower for unresolved attachment (Cohen’s d = 1.07). Women were nearly three times more likely to be classified as unresolved than were men. No other associations between study variables, including temperament traits and psychological symptoms, and unresolved attachment reached significance (see Table 3).

5-HTTLPR genotype

Descriptive statistics for the study variables by 5-HTTLPR genotypes are presented in Table 4. None of the associations were found to be significant. Furthermore, there were no significant effects of 5-HTTLPR genotype on any of the temperament traits or psychological symptoms (see Table 4). The absence of significant associations suggests that these characteristicsdo not account for the association between 5-HTTLPR genotype and unresolved attachment.

Table 4.

Study Variables by 5HTTLPR Genotype

Variable	5 HTTLPR genotype						Test statistic
	s/s		s/l		l/l
	N	M (SD) or %	N	M (SD) or %	N	M (SD) or %
Age	21	39.05 (6.44)	30	39.50 (7.40)	35	37.77 (7.05)	F(2, 83) = 0.52
No. of -trauma events	21	3.48(1.72)	30	3.11(1.15)	35	3.14(4.38)	F(2, 83) = 0.43
AAI coherence of transcript	21	4.92(1.65)	30	4.25 (1–78)	35	5.10(1.55)	F(2, 83) = 2.42
Sex							χ2(2) = 0.54
Male	10	48	12	40	17	49
Female	11	52	18	60	18	51
Biological parent diagnosis							χ2(2) = 2.11
Control	11	52	16	53	24	69
Proband	10	48	14	47	11	31
Age at adoption							χ2(6) = 5.95
Less than 1 month	12	60	20	67	27	81
1 to 3 months	3	15	6	19	1	3
3 to 6 months	3	15	2	~7	3	8
Greater than 6 months	2	10	2	~7	3	8
SNAP
Negative Temperament	20	45.24 (12.54)	28	44.31 (11.44)	34	43.65 (11.56)	K-W: χ2(2) = 0.27
Positive Temperament	20	49.25 (10.48)	28	46.76 (11.80)	34	48.00 (10.03)	K-W: χ2(2) = 0.59
Disinhibited Temperament	20	45.57 (10.71)	28	41.79(7.72)	34	41.26 (7.67)	K-W: χ2(2) = 3.02
BSI
Interpersonal sensitivity	20	50.65 (11.09)	28	51.96 (10.18)	33	49.52 (9.88)	K-W: χ2(2) = 0.57
Depression	20	55.25 (11.35)	28	52.64 (9.75)	33	51.67 (10.06)	K-W: χ2(2) = 1.25
Anxiety	20	51.90 (11.76)	28	47.89 (10.01)	33	48.21 (10.28)	K-W: χ2(2) = 1.58
Hostility	20	52.35 (10.05)	28	50.18 (9.89)	33	51.58(10.43)	K-W: χ2;(2) = 0.67

Note. Biological parent diagnosis includes alcohol problems and/or antisocial behaviors. AAI = Adult Attachment Interview: SNAP = Schedule for Nonadaptive and Adaptive Personality: K-W = Kruskal-Wallis test: BSI = Brief Symptom Instrument.

AAI unresolved attachment scale scores

Further examination of the data revealed that 92% (23/25) of unresolved attachment cases were classified as such based on examination of speech during description ofa loss. Therefore, we conducted two sets of univariate analyses involving the unresolved scale scores. First, we analyzed the 9-point scale score representing the maximum score assigned to any loss and/or trauma (UO). Second, we analyzed the unresolved 9-point scale score specific to descriptions of loss (UL). Similar analyses were not conducted for the unresolved scale score specific to trauma (UTr) because of insufficient sample size.

Descriptive statistics for associations between the study variables and unresolved attachment scores are presented in Table 5.

Table 5.

Study Variables by Unresolved Attachment Scale Scores

Variable	Unresolved scale score
	UO			UL
	N	M (SD)	Test statistic	N	M (SD)	Test statistic
Age	86		r = .15	84		r= .21
No. of loss-trauma events	86		r = .24*	84		r = .32**
AAI coherence of transcript	86		r = −.28**	84		r = −.24*
Sex			z = −1.92			z = −1.80
Male	39	3. 09 (1.89)		38	2.97 (1.92)
Female	47	3.99 (2.28)		46	3.85 (2.30)
Biological parent diagnosis			z = −0.05			z = −0.04
Control	51	3.57 (2.14)		51	3.44 (2.11)
Proband	35	3.60 (2.20)		33	3.47 (2.29)
Age at adoption			K-W: χ2(3) = 2.22			K-W: χ2(3) = 2.42
Less then 1 month	59	3.52 (2.20)		58	3.34 (2.20)
1 to 3 months	10	4.30 (2.43)		10	4.30 (2.43)
3 to 6 months	8	3.81 (1.69)		7	3.79 (1.82)
Greater than 6 months	7	2.86 (1.68)		7	2.86 (1.68)
SNAP
Negative Temperament	82		r=.17	80		r = .21
Positive Temperament	82		r = −.08	80		r = −.05
Disinhibited Temperament	82		r= .19	80		r = .16
BSI
Depression	81		r= .21	79		r = .24*
Interpersonal sensitivity	81		r =.21	79		r = .24*
Anxiety	81		r =.11	79		r = .12
Hostility	81		r =.15	79		r = .12

Note. The rs are Spearman rho correlations, and the; z statistics are from Mann-Whitney U tests. U_o = unresolved loss or trauma seals score; U_L. = unresolved loss scale score; AAI = Adult Attachment Interview; K-W = Kraskal-Wallis: SNAP = Schedule for Nonadaptive and Adaptive Personality; BSI = Brief Symptom Instrument.

^*p <.05.

^**p <.01.

Sex, biological parent diagnosis, age at adoption, and temperament traits were not significantly associated with unresolved scale scores. Higher UL scores were associated with a higher number of reported losses or traumas, lower overall coherence of transcript, and higher symptoms of depression and interpersonal sensitivity. Temperament traits or psychological symptoms were not found to be associated with the UO scale scores.

In summary, examination ofthe study variables by unresolved attachment and 5 -HTTLPR showed significantly higher unresolved attachment among women and among individuals reporting a greater number of loss–trauma events. Higher symptoms of depression and interpersonal sensitivity were found with higher UL scores. 5-HTTLPR genotype was not associated with any study variables including overall coherence of transcript.

Univariate Associations Between 5-HTTLPR and Unresolved Attachment Associations

Data showing univariate associations between unresolved attachment and 5-HTTLPR genotype are shown in Table 6.

Table 6.

Univariate Associations Between 5-HTTLPR Genotype and Unresolved Attachment

AAI measure	5-HTTLPR genotype						Test statistic
	s/s		s/l		l/l
	N	M (SD) or %	N	M (SD) or %	N	M(SD) or %
Unresolved classification							χ2:(2) = 9.07*
Not Unresolved	13	21	17	28	31	51
Unresolved	8	32	13	52	4	16
AAI scale scores
Overall Unresolved	21	3.81 (2.05)	30	4.09 (2.43)	35	3.01 (1.85)	K-W: χ2(2) = 4.08
Unresolved loss	21	3.81 (2.05)	30	4.09 (2.43)	33	2.64 (1.75)	K-W: χ2;(2) = 7.33*
Unresolved trauma	3	3.33 (2.52)	4	3.13 (2.39)	9	3.22 (2.12)	k-w: χ2(2) = 0.02
Coherence of transcript	21	4.92 (1.65)	30	4.25(1.78)	35	5.10 (1.55)	K-W: χ2(2) = 4.00

Note. AAI = Adult Attachment Interview; K-W = Kruskal-Wallis.

^*p < .05.

The overall chi-square statistic was significant for the association between unresolved versus not unresolved attachment and 5-HTTLPR genotype. A significant overall association between 5-HTTLPR genotype was also found for the UL scale scores (see Table 6). In contrast, the association between 5-HTTLPR genotype and UO scale scores did not reach statistical significance. Despite the nonsignificant overall effect of 5-HTTLPR genotype and the UO scale score, we present effect sizes for both scale scores. For UO, medium effect sizes were observed for the homozygous short (Cohen’s d = 0.41) and heterozygous short (Cohen’s d = 0.50) genotypes. In comparison, effect sizes were medium to large for the UL scale scores (Cohen’s d = 0.61 and 0.68, respectively).

In summary, univariate analyses suggest that carriers of the short variant of5 -HTTLPR show higher rates of unresolved attachment. Multivariate analyses presented below test associations between 5-HTTLPR and unresolved attachment after adjusting for significant control variables. These analyses also directly test dominance of the short 5-HTTLPR allele.

Multivariate Analysis of Unresolved Attachment Classification and 5-HTTLPR

Unresolved attachment classification

Logistic regression tested multivariate associations between 5-HTTLPR genotype and the unresolved attachment classification (see below and see Table 7).

Table 7.

Multivariate logistic Regression Predicting Unresolved Attachment Classification

Source of variation	B	Wald	df	P	OR	95% CI
Model 1: Control variables
Current age	0.07	2.43	1	.12	1.07	0.98, 1.16
Sex	−1.17	3.48	1	.06	0.31	0.09. 1.06
Biological parent diagnosis	−0.85	1.93	1	.16	0.43	0.13, 1.42
Age adopted		1.77	3	.62
1 to 3 months	1.03	1.73	1	.19	2.81	0.60, 13.10
3 to 6 months	0.49	0.23	1	.63	1.63	0.22, 11.98
Greater than 6 months	−21.03	0.00	1	1.00	0.00	0.00
Depression	0.04	0.83	1	.36	1.04	0.96, 1.12
Interpersonal sensitivity	0.04	0.85	1	.36	1.04	0.96, 1.12
Model 2: 5-HTTLPR genotype
Sex	−1.21	4.57	1	.03	0–30	0.10,0.91
Trauma reported	1.17	2.78	1	.10	3–21	0.82, 12.61
5-HTTLPR contrastsa		8.79	2	.01
s/s genotype	1.85	6.05	1	.02	6.34	1.46, 27.64
s/l genotype	2.01	8.15	1	<.01	7.43.	1.88,29.41
Model 3: 5-HTTLPR dominance effect
Sex	−1.21	4.64	1	.03	0.30	0.10,0.90
Trauma reported	−1.16	2.77	1	.10	0.31	0.08, 1.23
5-HTTLPR dominanceb	1.94	8.73	1	<.01	6.97	1.92,25.24

Note. N = 86. Biological parent diagnosis includes alcohol problems and/or antisocial behaviors. OR = odds ratio; CI = confidence interval.

^aThe contrast option available in the logistic regression command was used, The l/l 5-HTTLPR genotype was designated as the reference group.

^bDummy coding was used to construct the dominatice contrast: s/s or s/l = 1 and l/l = 0.

We constructed a dichotomous trauma variable (0 = no trauma, 1 = trauma reported) because the experience of trauma was associated with degree of unresolved attachment. We present findings for three unnested models. In Model 1, we determined which control variables significantly predicted unresolved attachment. In Model 2, we retained the significant control variables from Model 1 and added the presence of reported trauma and 5-HTTLPR genotype. Model 3 presents the dominance model of 5-HTLLPR genotype.

The interaction between sex and 5-HTTLPR genotype did not reach statistical significance, Wald(2, N = 86) = 4.63, p = .10. Therefore, we present findings for the three main effects models (see Table 7). None of the control variables were significant in Model 1. Sex approached significance and was therefore included in Model 2 because of significant univariate associations. Women were more likely to be classified as unresolved (Model 2). The p values associated with the homozygous and heterozygous 5-HTTLPR contrasts were both significant (see Model 2, Table 7). The chi-square for the log-likelihood ratios associated with Model 2 was also significant (−2 log likelihood = 86.91), χ2(4, N = 86) = 16.77, p < .01. The dummy variable comparing the homozygous or heterozygous genotypes (−B = 0.18), Wald(1, N = 86) = 0.09, p = .77, odds ratio = 0.84, 95% confidence interval = 0.26, 2.70, and the log-likelihood ratio for the model (−2 log likelihood = 66.44),χ2(3, N = 86) = 2.67, p = .45, did not reach statistical significance. The test for dominance of the short 5-HTTLPR allele with the s/s or s/l genotypes (coded as 1) and the l/l 5-HTTLPR genotype (coded 0) are presented in Table 7, see Model 3, Table 7). The contrast and the overall model (−2 log likelihood = 86.97),χ2(3, N = 86) = 16.70, p < .001, were significant (see Model 3, Table 7). Predicted probabilities estimated from Model 3 were .03 for men reporting no trauma and who had the homozygous long 5-HTTLPR genotype, compared with .75 for female carriers of the short 5-HTTLPR allele who reported trauma.

Overall scale scores for unresolved loss and/or trauma

Examination of the 9-point unresolved scale scores revealed a bimodal distribution. For this reason, we constructed the following dummy variable from the overall unresolved scale score: Unresolved scale scores of 1 were coded as 0 and scale scores above 1 were coded 1. The significance of the dummy variable is not of theoretical importance and allows interpretation of significant associations with the scale scores as predicting degree of unresolved attachment given lack of resolution. To minimize the importance of the inflated significance of the model attributable to the categorical variable, we also present overall model fit attributable solely to the 5-HTTLPR genotypes. We tested the three models described above using linear regression with contrasts. Because of asymmetrical distributions, the unresolved scale scores were ranked prior to being submitted to the regression analysis to approximate a nonparametric test (Conover, 1980, 1999; Conover & Iman, 1981).

The Sex × 5-HTTLPR Genotype interactions were not significant for the UO or UL scale scores, ΔF(2, 72) = 1.03, p = .36, ΔRadj2= .02, and ΔF(2, 70) = 0.95, p = .39, ΔRadj2= .03, respectively. The final models are presented in Table 8.

Table 8.

Multivariate Linear Regressions Predicting Ranked Unresolved Scale Scores

Source of variation	Ranked unresolved scale scores
	UO			UL
	β	t	P	β	t	P
Model 1: Control variables
Sex	.19	1.67	.10	.20	1.77	.08
Current age	.26	2.17	.03	.33	2.66	.01
Biological parent diagnosis	.07	0.66	.51	.05	0.47	.64
Age adopted contrasts
1 to 3 months	.05	0.48	.64	.06	0.57	.57
3 to 6 months	−.02	−0.18	.86	−.06	−0.51	.61
Greater than 6 months	−.31	−2.45	.02	−.29	−2.37	.02
Depression	.13	0.84	.41	.16	1.03	.31
Interpersonal sensitivity	.19	1.21	.23	.18	1.14	.26
Model 2: 5-HTTLPR genotype
Sex	.15	2.42	.02	.14	2.18	.03
Current age	06	0.86	.39	.11	1.67	.10
Trauma reported	−.03	−0.50	.62	−.11	−1.66	.10
Dummy unresolved presenta	.79	12.44	.001	.78	11.79	.001
5-HTTLPR contrasts
Homozygous short	.07	1.02	.31	.14	2.03	.05
Heterozygous short	.18	2.70	.01	.26	3.71	.001
Model 3: 5-HTTLPR dominance effect
Sex	.16	2.51	.01	.14	2.28	.03
Current age	.06	0.92	.36	.11	1.73	.09
Trauma reported	−.03	−0.42	.67	−.10	−1.58	.12
Dummy unresolved presenta	.78	12.30	.001	.75	11.67	.001
5-HTTLPR dominancec	.14	2.30	.02	.22	3.47	.001

Note. Biological parent diagnosis includes alcohol problems and/or antisocial behaviors, U_O = scale score for unresolved loss or trauma (N = 86): U_L = scale score for unresolved loss (N = 84).

^aUnresolved scale score recoded into dummy variable (scale score of l =0; scale scores > l = 1).

^bDummy coding was used to construct the 5-HTTLPR contrasts; two vectors were created: the homozygous short vector (s/s = l, s/l = 0, l/l = 0) and the heterozygous short vector (s/s = 0, s/l = l, l/l = 0).

^cDummy coding was used to construct the dominance contrast: s/s or s/l = l and l/l = 0.

The overall F statistic for Model 1 did not reach statistical significance for the UO scale scores, F(8, 70) = 1.99, p = .08, Radj2 = .07 The overall F test for Model 2, which included sex, current age, presence of trauma, the dummy unresolved scale score variable, and 5-HTTLPR genotype, was significant, F(6, 79) = 32.08, p < .001, Radj2 = .69. The estimated increase in R2 due to 5- HTTLPR genotype in Model 2 was also significant, ΔF(2, 79) = 3.66, p = .03, ΔRadj2 = .03, with significance for the heterozygous short contrast. The comparison of the homozygous short and heterozygous short 5-HTTLPR genotypes did not reach statistical significance (β = 0.11), t = 1.34, p = .19. Finally, the dominance contrast, where s/s or s/l 5-HTTLPR genotypes were coded 1 and the l/l genotype was coded 0, was significant in Model 3, ΔF(1, 80) = 5.31, p < .05, ΔRadj2 = .02.

The overall control model (Model 1) was significant for the UL scale score, F(8, 68) = 2.71, p = .02, Radj2= .15 (see Table 8). Women and older participants received higher unresolved loss scale scores. The model testing the significance of 5-HTTLPR genotype after controlling for covariates was significant, F(6, 77) = 31.74, p = .001, Radj2 = .69, as was the increase in R2 specific to 5-HTTLPR genotype, ΔF(2, 77) = 6.97, p < .01, ΔRadj2 = .05. The comparison of the homozygous short and heterozygous short 5-HTTLPR genotypes did not reach statistical significance (β = 0.11), t = 1.35, p = .19. The overall F test for the dominance model was significant, ΔF(5, 78) = 37.38, p < .001, ΔRadj2 = .69, with significant and unique contributions of the short 5-HTTLPR allele to UL, ΔF(1, 78) = 12.07, p = .001, ΔRadj2 = .05.

In summary, multivariate analyses tested the association between 5-HTTLPR genotype and unresolved attachment after controlling for significant control variables. Biological parent diagnosis and adoptee age at time of adoption did not reach statistical significance. Significant control variables included sex and current age (for scale scores). Unresolved attachment was greater among women and older participants. There was not a statistically significant interaction between sex and 5-HTTLPR, suggesting the association with unresolved attachment was invariant. Finally, support for dominance of the short variant of 5-HTTLPR in unresolved attachment was found.

Discussion

Incorporating specific genetic markers into models predicting psychosocial outcomes has enriched our understanding of the complex interplay between genes and environment. The present study examined the association between 5-HTTLPR genotype and unresolved attachment. We found a strong dominant effect of the short allele of 5-HTTLPR on unresolved loss. The magnitude of the effect was equivalent to that found for anxiety-related traits (~3–9%; Lesch et al., 1996), with the short variant of the 5-HTTLPR allele predicting increased risk of unresolved attachment. The effect of 5-HTTLPR genotype was found for both the dichotomous classification and the continuous scale score for unresolved loss. This suggests that our findings are not a statistical artifact (Eaves, 2006). In addition, significant associations between temperament traits or psychological symptoms and 5-HTTLPR genotype were not statistically significant, signifying that these characteristics may not act as mediators. Finally, the attachment–5-HTTLPR genotype association was found for speech related to loss but not overall coherence.

What mechanisms underlie the risk to develop unresolved attachment when having the short 5-HTTLPR allele? We propose that unresolved attachment is indicative of an emotional regulatory system that has short circuited. The ventral and medial prefrontal cortexes participate in regulation of the amygdaloid complex (Blair et al., 2007; Drevets et al., 1997). These networks are influenced by the 5-HTTLPR genotype (Heinz et al., 2005; Pezawas et al., 2005) and regulate subjective appreciation of emotional experiences, thereby playing a role in the recall and inhibition of emotional memories (Grimm et al., 2006; Phan et al., 2004; Phan, Wager, Taylor, & Liberzon, 2004; Sierra-Mercado, Corcoran, Lebron-Milad, & Quirk, 2006). The presence of the 5-HTTLPR short allele may influence the interconnectivity between these brain regions (Bertolino et al., 2005), thereby increasing susceptibility to the disorganizing effects of elevated affective intensity experienced during discussions of loss (Fearon & Mansell, 2001; Hariri et al., 2005; Heinz et al., 2005; Hesse & Main, 2006). Impaired interconnectivity may result in a reduced ability to effectively regulate heightened emotion and ultimately interfere with active monitoring of speech. This interpretation is consistent with several indexes of unresolved attachment. For example, “dead–not dead” is an example of a lapse of reasoning in which a deceased person is spoken about as though the person was still alive and is indicated by shifts to present tense. Heightened affectivity associated with talking about the deceased, coupled with impaired connection of memories surrounding the events (e.g., date, time), may result in the momentary and unmonitored shift to present tense. Excessive attention to detail when describing a traumatic event might arise because of heightened emotional reactivity and a simultaneous inability to regulate orientation.

In addition to understanding potential underlying physiological mechanisms by which 5-HT regulation affects attachment, we must recognize the plausibility that multiple neurotransmitter systems and neural circuitries are involved in emotion regulation (Luciana, Collins, & Depue, 1998; Meyer-Lindenberg et al., 2005; Ressler & Nemeroff, 2000). In this context, comparison of the findings in this study with the findings on the neurobiology of attachment in childhood is warranted. Research on the genetic susceptibility for childhood disorganized attachment has focused on the dopaminergic system (i.e., DRD4 gene; Bakermans-Kranenburg & Van IJzendoorn, 2004; Gervai et al., 2005; Lakatos et al., 2000, 2002; Van IJzendoorn & Bakermans-Kranenburg, 2006). Disorganized infant attachment is viewed as a breakdown in the infant’s behavioral and attentional regulation of fear. Infants may demonstrate sequential and contradictory approach and avoidance behaviors, freezing, stereotypies, disoriented wandering, or movement without direction or finality. The amygdala and associated neural circuits are also influenced by dopamine (Phelps & LeDoux, 2005; Rosenkranz & Grace, 2001, 2003; Sesack, Carr, Omelchenko, & Pinto, 2003), as are brain regions that play a functional role in emotion regulation and nonsocial regulatory abilities (e.g., attention; Chamberlain, Muller, Robbins, & Sahakian, 2006). Therefore, exploring the role of dopaminergic regulation in infant disorganized attachment, and possibly adult unresolved attachment, appears warranted.

Finally, studies in molecular psychiatry have shown phenotypic specificity when exploring genetic susceptibility for a wide variety of behaviors(Rhee & Waldman, 2002; Roisman & Fraley, 2006). Similarly, our findings suggest that the origins of unresolved attachment in adulthood may have etiologic specificity that is experientially and possibly genetically moderated (Hughes, Turton, Hopper, McGauley, & Fonagy, 2004; Jacobvitz, Leon, & Hazen, 2006; Lyons-Ruth, Yellin, Melnick, & Atwood, 2003). Although conclusions drawn from our data are limited, we showed significant 5-HTTLPR genotype associations specific to unresolved loss. All but 2 of our participants were classified unresolved with regard to loss experiences, thus the specificity of the association with the scale scores for unresolved loss is not surprising. However, the specific influence of the 5-HTTLPR genotype is also consistent with differences in the behavioral correlates among parents classified as unresolved because of loss versus trauma (Abrams, Rifkin, & Hesse, 2006; Hesse & Main, 2006; Jacobvitz et al., 2006; Lyons-Ruth, Yellin, Melnick, & Atwood, 2005). The results of the present study may offer a basis for further hypothesis testing aimed at explaining incomplete prediction of infant disorganized attachment from parental unresolved attachment (Gervai et al., 2007; Madigan et al., 2006; Schuengel, Bakermans-Kranenburg, & Van IJzendoorn, 1999; Van IJzendoorn & Bakermans-Kranenburg, 2006; Van IJzendoorn, Schuengel, & Bakermans-Kranenburg, 1999).

Some caveats need to be addressed. Because our sample was composed of adoptees, the results of this study need to be interpreted with caution. In the present study, rates of unresolved attachment were higher than expected. The loss of a child by miscarriage or perceived loss of a child because of infertility by the adoptive parent(s) may influence parental behavior toward future children, placingan individual at greater risk for unresolved status in adulthood (Bakermans-Kranenburg, Schuengel, & Van IJzendoorn, 1999; Hughes, Turton, McGauley, & Fonagy, 2006). Furthermore, adopted individuals may be more susceptible to experiences of loss, given possible feelings of loss associated with being adopted (Borders, Penny, & Portnoy, 2000; Feeney, Passmore, & Peterson, 2007). Our study is also limited by the examination of a single genetic polymorphism. Evidence of gene–gene interactions (e.g., the DRD4 gene and the functional −521 C/T promoter polymorphism) have been reported, albeit inconsistently, which further complicates interpretation of isolated genetic effects (Bakermans-Kranenburg & Van IJzendoorn, 2004; Gervai et al., 2005; Lakatos et al., 2002; Van IJzendoorn & Bakermans-Kranenburg, 2006). Finally, it is standard among researchers using the adoption paradigm to assume that passive gene–environment correlations are absent because of the lack of biological relatedness between parent and child. Passive gene–environment correlation refers to the inheritance of genes that are also associated with characteristics of the environment. We acknowledge that passive gene–environment correlations are possible in these data. The adoptive parents and their biologically unrelated offspring could have common genotypes for a specific genetic marker (e.g., 5-HTTLPR polymorphism) that might also be correlated with environmental factors associated with unresolved attachment (e.g., frightened–frightening maternal behavior). Unfortunately, direct testing of passive gene–environment correlations is not possible with these data because of the absence of genetic information on the adoptive parents (or a comparable biologically intact comparison group). Therefore, we can onlytentatively argue that passive gene–environment correlations do not contribute to our findings.

In summary, this study provides preliminary evidence of an underlying genetic susceptibility to unresolved attachment in adulthood. The findings of the present research are intriguing, but replication is necessary and needs to be extended to samples closely matched to the characteristics of ours as well as to more generally defined populations. We hope that this study fosters research aimed at uncovering the exact mechanisms and specificity by which the 5-HTTLPR genotype regulates emotions (Fox, Hane, & Pine, 2007).