Genetic, environmental and gender influences on attachment disorder behaviours

Background Despite current interest in attachment disorder,there is concern about its discrimination from other disordersand an unproven assumption of an environmental aetiology.

Aims To test whether behaviours suggestive of attachment disorder are distinct from other childhood behavioural and emotionalproblems and are solely environmentally determined.

Method In a community sample of 13 472 twins, we carried outfactor analysis of questionnaire items encompassing behavioursindicative of attachment disorder, conduct problems, hyperactivityand emotional difficulties. We used behavioural genetic model-fittinganalysis to explore the contribution of genes and environment.

Results Factor analysis showed clear discrimination between behaviours suggestive of attachment disorder, conduct problems,hyperactivity and emotional problems. Behavioural geneticsanalysis suggested a strong genetic influence to attachmentdisorder behaviour, with males showing higher heritability.

Conclusions Behaviours suggestive of attachment disorder canbe differentiated from common childhood emotional and behaviouralproblems and appear to be strongly genetically influenced,particularly in boys.

INTRODUCTION

TOP

INTRODUCTION

There have been recent attempts to codify behaviours associatedwith early neglect and institutionalisation (Chisolm et al, 1995;Zeanah et al, 2004) into a psychiatric category. Both DSM–IVand ICD–10 describe reactive attachment disorder, withtwo subtypes encompassing inhibited and disinhibited behaviour (World Health Organization, 1992; American Psychiatric Association, 1994).Questions remain about the nosology of the syndrome beyondage 5 years (American Academy of Child and Adolescent Psychiatry, 2005),therefore we simply refer to ‘attachment disorder behaviours’.We seek to extend the extant literature by testing two hypotheses:first, that the two subtypes are distinct from one anotherand from other common behavioural and emotional problems inyoung children, and second that these behavioural patternsare environmentally mediated. We capitalise on a twin study,a design that provides particular leverage in testing environmentalhypotheses.

METHOD

TOP

METHOD

Participants
The sampling frame, described in detail elsewhere (Trouton et al, 2002),was 13 940 twin pairs from the 1994 and 1995 birth cohorts ofthe Twins Early Development Study, tested as they reached theireighth birthday. Informed written consent was obtained fromall participants. Questionnaires were sent to the parents of13 940 twin pairs aged 7–9 years, and questionnairesfor 6771 pairs (48.6%) were returned. Thirty-five pairs oftwins were excluded because of missing data, leaving a finalsample of 13 472 twins of average age 7.9 years. There wereminor differences between the twin pairs enrolled in the studyinitially and those for whom questionnaires were returned,in ethnicity (87.5% White mothers in first wave v. 93.9% incurrent wave), mothers with A-level as highest qualification(11.6% v. 14.1%) and mothers who were working (39.6% v. 43.5%).

Measures
We previously used a questionnaire for reactive attachment disordersin clinical and general population samples aged 5–16years (Minnis et al, 2002). It was a checklist of attachmentdisorder behaviours of both the inhibited and disinhibitedtypes, as described in ICD–10 (World Health Organization, 1992).During pilot work, items were added at the suggestion of parentsand clinicians, the wording of other items was modified (Minnis et al, 2002) and items were removed that failed to discriminate betweenchildren from the general population and children living infoster care (Millward et al, 2006). The resulting questionnaireused in the present study, the Relationship Problems Questionnaire(RPQ; see online appendix), is an 18-item parent-report questionnairewith an internal consistency (Cronbach’s ${alpha}$ ) of 0.85 inthis data-set. It has four possible responses (‘exactlylike my child’, ‘like my child’, ‘a bit like my child’ and ‘not at all like my child’),scored 3, 2, 1 and 0 respectively (maximum possible score 54).

The Strengths and Difficulties Questionnaire (SDQ; Goodman, 1997)is a 25-item screening instrument for common child mental healthproblems which has been well validated against other screeninginstruments such as the Child Behavior Checklist (Goodman & Scott, 1999)and against psychiatric diagnosis (Goodman & Scott, 1999; Goodman et al, 2003). It has sub-scales for emotional problems,conduct problems, hyperactivity, problems with peer relationshipsand prosocial behaviour. It has three possible responses (‘nottrue’, ‘somewhat true’ and ‘definitelytrue’), scored 0, 1 or 2.

Parents completed questions described in detail elsewhere (Asbury et al, 2003) on the use of discipline, including reasoning, spanking andignoring misbehaviour, which gave composite scores for ‘parentalwarmth’, ‘negativity’ and ‘harsh parenting’.For harshness, items were rated on a six-point scale from ‘Irarely or never do this’ to ‘I usually do this’.One-year test–retest reliability was 0.52. Warmth andnegativity were rated on a five-point scale from ‘definitelytrue’ to ‘definitely untrue’, and the 1-yeartest–retest reliability was 0.50. A measure of generalcognitive functioning was derived from verbal and non-verbalcognitive ability tests adapted for telephone administration (Petrill et al, 2002).

Phenotypic factor analysis
Principal components analysis was used to explore the underlyingstructure of the RPQ. The optimal number of factors was identifiedusing a scree plot (Cattell, 1966). Both orthogonal and obliquemethods of rotation were tried and all gave similar results.Varimax rotation is the analysis presented here. The twin designwas exploited as an opportunity to repeat the factor analysisand see if similar results were produced on both occasions.Data were analysed separately for each member of a twin pairand correlations were calculated between each of the factorloadings. In order to explore whether it is possible to discriminatebetween attachment disorder behaviours and other mental health problems, the factor analysis included RPQ items plus SDQ itemsfor emotional problems, conduct problems and hyperactivity.

RPQ scores and parenting
We explored the association between RPQ scores and parentalwarmth/harsh parenting using linear regression analysis controllingfor age, gender, paternal social class (Office of PopulationCensuses and Surveys, 1995) and the child’s cognitiveability.

Quantitative genetic analyses
The hypothesis that there is a genetic component to attachmentdisorder behaviours was tested first by comparing intraclasscorrelations between RPQ scores for monozygotic twins withthose for dizygotic twins, and then by model fitting.

Each genetic factor influencing human behaviour is presumedto contribute only a small amount, and may have an additiveeffect with other genetic factors. Dominance effects may alsobe important – dominance is the extent to which the effectsof alleles at a locus fail to ‘add up’ to producegenotypic values. If the effect of a locus involves dominance, there are effects of a combination of alleles at that particularlocus. Additive and dominant genetic effects are defined soas to be independent of one another (Plomin et al, 2001).Environmental factors can be shared, i.e. can be influences that make children growing up in the same family similar, ornon-shared, which refers to all other environmental factors (Plomin & Daniels, 1987).

Intraclass correlational analyses
The fact that monozygotic twins share all of their genetic materialwhereas dizygotic twins share only about 50% can be used toestimate the genetic and environmental influences on attachmentdisorder behaviours. If shared environmental influences werepredominant, twin correlations would be large and similar formonozygotic and dizygotic twins. If non-shared environmental influences were predominant, twin correlations would be smallbut similar for both types of twins. If, however, genetic influenceswere significant, monozygotic twin correlations would exceeddizygotic twin correlations.

Model-fitting analyses
Maximum likelihood model-fitting analyses estimate the contributionsof additive genetic (A), shared environmental (C), dominance (D) and non-shared environmental effects (E). A model incorporatingadditive genetic, shared and non-shared environmental effects (ACE model) was considered first. An ADE model, considering dominance effects instead of shared environmental effects,was then fitted and the two compared. A chi-squared goodness-of-fittest was applied to each model (Neale et al, 2003).

Analyses were carried out using both the total RPQ score andthe sub-scales generated in the factor analysis and were doneseparately for male and female twins. The twins were double-enteredso each child appears as twin one and as twin two to help toeliminate any bias due to birth order, and 95% confidence intervalswere adjusted accordingly. Behavioural genetic modelling wasdone using Mx (Neale et al, 2003) and all other analyses usedthe Statistical Package for the Social Sciences, Version 11.The covariance matrices were used to input the data into Mxand only same-sex twin pairs were included in the analysis.

View larger version (13K):
[in this window]
[in a new window]
[as a PowerPoint slide]
Fig. 1 Distribution of Relationship Problems Questionnaire scores.

RESULTS

TOP

RESULTS

Distribution of RPQ scores
The RPQ scores in this population appeared to be continuouslydistributed, although strongly skewed (Fig. 1).

Factor analysis
For the RPQ factor analysis, a scree plot suggested a three-factor solution. Rotated factor 1 had six questions that loaded highlyon it: ‘unpredictable friendliness’, ‘runsaway when approached’, ‘false affection’,‘has no conscience’, ‘aggressive to self’and ‘looks frozen with fear’. This first rotatedfactor explained 30% of the variance. Loading highly onto thesecond factor were ‘gets too physically close’,‘too cuddly’, ‘too friendly with strangers’and ‘asks personal questions’; this factor explained10% of the variance. Four questions loaded highly on the last factor: ‘afraid of new situations’, ‘actsyounger than age’, ‘often unhappy’ and ‘veryclingy’; this factor explained 7% of the variance. Theremaining four questions loaded most highly onto the thirdfactor, but their loadings were fairly equally spread acrossall three factors. These were removed, one at a time, to seewhat effect removing them had on the remaining factor loadings.Removing them improved discrimination between the three factorsand the final model used 14 questions, each of which loadedclearly and highly onto one of the three factors.

When the factor analysis was repeated for the two members ofeach twin pair, the correlations between the factor loadingsfor each of the three factor pairs were 0.998, 0.998 and 0.992respectively, each with P $≤$ 0.001. The first factor appeared toindex behaviours indicative of the inhibited form of an attachmentdisorder. The second factor indexed behaviours that reflectthe disinhibited form of attachment disorder. The third factorsuggested behaviours typical of behaviourally inhibited temperament(Muris et al, 2005), which may not be directly linked to attachmentdisorder. To avoid confusion with inhibited attachment disorderbehaviours, we refer to this factor as the ‘temperamentfactor’.

Items from the conduct problems, hyperactivity and emotionalproblems (anxiety and depression) scales of the SDQ were thenincluded in the factor analysis along with the 14 remainingRPQ items. The three RPQ sub-scales were still clearly distinctfrom one another and from the SDQ sub-scales (Table 1), withthe exception that the SDQ item ‘nervous/clingy’loaded with the RPQ temperament subscale rather than with theSDQ emotional problems sub-scale. The RPQ item ‘oftenunhappy’ did not load with any particular factor.

View this table:
[in this window]
[in a new window]

Table 1 Rotated factor loadings for 14 of the Relationship Problems Questionnaire items and 14 Strengths and Difficulties Questionnaire items, from a six-factor solution using the principal components extraction method (varimax rotation)

RPQ scores and parenting variables
There were significant associations between both the inhibitedand disinhibited subscales and harsh parenting and parentalnegativity, and significant negative associations between bothsub-scales and parental positivity (Table 2) after controllingfor age, gender and cognitive ability, which partially confounded these relationships (not social class, which did not act asa confounder).

View this table:
[in this window]
[in a new window]

Table 2 Linear regression analysis of the association between harsh parenting, parental negativity and positivity, and the Relationship Problems Questionnaire¹ inhibited and disinhibited sub-scales.

Developing sub-scales of the RPQ
Sub-scales of the RPQ were developed from the results of thefactor analysis for use in behavioural genetics analyses (see Table 1). The ‘inhibited’ sub-scale included thesix questions that loaded highly onto factor 1. The ‘disinhibited’sub-scale comprised the four questions that loaded highly ontofactor 2. The inhibited and disinhibited sub-scale scores areonly modestly correlated (0.443) with each other and with theSDQ sub-scales (0.176–0.318). The behavioural geneticsanalyses were performed both for the whole 18-item RPQ and forthe sub-scales.

Behavioural genetics analysis of total RPQ scores
The correlation for the 18-item RPQ items in male monozygotictwins was 0.917 (P<0.0001) and 0.599 for male dizygotictwins. This marked difference in monozygotic v. dizygotic correlationgives a clear indication of a strong genetic influence. Totest this hypothesis, an ACE model was fitted and provideda significantly good fit ( ${chi}$ ² goodness-of-fit test), whereasthe ADE model was a poor fit. Parameter estimates are shownin Table 3.

View this table:
[in this window]
[in a new window]

Table 3 Estimates from the ACE model based on the Relationship Problems Questionnaire and its ‘inhibited’ and ‘disinhibited’ sub-scales

Behavioural genetic modelling assumes multivariate normality,but these data were highly skewed to the left. Various transformationsof the data were unsuccessful in producing a normal distribution.More importantly, if a suitable transformation was achieved,this would almost certainly lead to a loss of much of the importantinformation relating to the variation. However, the ACE modelgives almost identical results to those produced by the correlationcalculations, which make no assumption about the distributionsof the data.

The same analyses were performed on the female twin data. TotalRPQ scores for female monozygotic twins are highly correlated(correlation coefficient 0.914). The female dizygotic twinswere more highly correlated (correlation coefficient 0.716)than the male dizygotic twins (0.599). The ACE model was fittedand again the ${omega}$ ² goodness-of-fit test indicated that this wasthe best fit. The ADE model again demonstrated a significantreduction in fit. Parameter estimates are shown in Table 3.The confidence intervals for additive genetic effects and sharedenvironment do not overlap when comparing males and females,indicating that they are significantly different.

Behavioural genetics analysis of inhibited and disinhibited sub-scales
It was clear that there is genetic influence on the inhibitedsub-scale scores, as the monozygotic correlation was 0.880for males and 0.846 for females, compared with dizygotic correlationsof 0.571 and 0.713 respectively for males and females. Forthe disinhibited sub-scale the monozygotic correlation was0.923 for males and 0.918 for females, compared with dizygotic correlations of 0.533 and 0.616 respectively for males andfemales. Model fitting found that the ACE models again gavethe best fit compared with the ADE model and estimates areshown in Table 3. For males, the majority of the variancein the inhibited and disinhibited sub-scales was explainedby additive genetic effects. This was also true for the disinhibited sub-scale for females, whereas for the inhibited sub-scalethe majority of the variance was explained by shared environmentaleffects. Again, the contribution of additive genetic effectsand shared environment was significantly different for malesand females.

DISCUSSION

TOP

DISCUSSION

Our findings demonstrate that behaviours suggestive of attachmentdisorder can be identified in school-age children from thegeneral population, are associated with harsh parenting andcan be discriminated from conduct problems, emotional problemsand hyperactivity. In this first twin study of these behaviours,we have shown that both genes and environment have a significantrole in their aetiology.

Genetic effects appear particularly important for boys. Interestingly, Zeanah & Fox (2004) have postulated that temperamentalfactors such as withdrawn–inhibited behaviour or impulsivitymay put a child at greater risk of attachment disorder in thecontext of maltreatment. They give the example of 20-month-old twins who were raised in the same seriously neglectful environment;the boy developed disinhibition, whereas the girl became emotionallywithdrawn and inhibited (Hinshaw-Fuselier et al, 1999). Quitewhat the gender modification of the genetic effect means isnot yet clear and requires replication, but a range of biologicalcandidates could be investigated in this context, includingstress hormones and neuropeptides.

Shared environment explained more variance in females than inmales. Although this could perhaps be accounted for by a greatersimilarity in parental behaviour with girls than with boys,it is an intriguing finding. As the shared environmental effectthat is likely to be of most importance in the aetiology ofattachment disorder is maltreatment (Zeanah & Fox, 2004),this needs to be further investigated in maltreated children.

In terms of genes that might be involved, an X-linked genetic–environmentalinteraction has been found in the development of conduct disorder(Caspi et al, 2002) as well as a link between the dopamineD4 receptor gene (DRD4) and disorganised attachment (Lakatos et al, 2000).It is early days in molecular genetic attachment research, butour findings reinforce the notion that this might be a fruitfulfuture direction.

Methodological considerations
We used the RPQ as a screening tool in a community sample, anddo not assert that the children reported as demonstrating thesebehaviours had reactive attachment disorder; in order to definesuch disorder, detailed diagnostic examinations would be requiredand, according to DSM–IV, symptoms would have had tobe present before the age of 5 years (American Psychiatric Association, 1994).We are not certain whether mothers or fathers completed the parent-report questionnaires, which could have affected theresults. This kind of population-based research requires simpletools, and complements but does not replace more clinicallyfocused research.

Our study is limited by factors known to apply to twin studiesin general (Maccoby, 2000). For example, correlations betweentwins’ scores could be affected by reporting bias onthe part of parents. The skewness of the distribution mighthave limited the model-fitting analysis, but is unlikely tohave seriously affected the interpretation of the results,because the correlational calculations (which do not dependon a normal distribution) gave very similar results. The responserate of just less than 50% means that the sample may differ systematically from the general population in known and unknownways. We are likely to have lost to follow-up the participantswith the most significant psychosocial problems, so it is particularlyinteresting that even in a sample that was probably healthierthan the general population, behaviours suggestive of attachmentdisorder were identified.

To our knowledge, no diagnostic instrument yet exists for attachment disorder in children of this age, but one is currently beingdeveloped by our group and will include information from parents,teachers and observation of the child. Only future researchwill determine whether these behaviours do, in fact, predicta diagnosis of attachment disorder and one method would be to follow up children who had high RPQ scores with a detaileddiagnostic assessment. For a disorder in which some behaviours,such as overfriendliness, are on a continuum with normal behaviour,the lack of more detailed clinical information may increasethe likelihood of false positive responses.

Two of the items that loaded on the ‘inhibited’factor – ‘has no conscience’, and ‘falseaffection’ – are not part of the DSM or ICD classificationof inhibited reactive attachment disorder. False affectionwould perhaps be expected to load with the disinhibited factor,although recent research has suggested that clinically thetwo subtypes can be mixed (Zeanah et al, 2004). Including thesetwo items would broaden the phenotype of inhibited reactiveattachment disorder and, as there is consensus that the inhibitedphenotype is less well defined than the disinhibited phenotype(American Academy of Child and Adolescent Psychiatry, 2005),clarity about the nosological boundaries of the inhibited disorderwill be an important future research focus. In the light ofthis apparent broadening, it is reassuring that our factoranalysis suggests clear demarcation between both attachmentdisorder subtypes and other forms of child psychopathologysuch as conduct disorder.

This research design allowed us to examine attachment disorderbehaviours as they were distributed in a sample approximatingthe general population, their discrimination from behaviourssuggestive of other disorders and the possibility of geneticmediation. The significant associations between RPQ scoresand indices of harsh or negative parenting suggest we are investigating the same domain of functioning (but perhaps less extreme behaviours)that we would be investigating in a maltreated sample. A studyof maltreated or severely neglected twins might yield differentfindings regarding the balance of genetic and environmentalinfluence, but would be difficult if not impossible to construct.

Clinical implications
Attachment disorder behaviours have previously been consideredin samples of children who are known to have been maltreatedor institutionalised. These data demonstrate that attachmentdisorder behaviours are present in the general population,are associated with harsh or negative parenting behaviour andmay be mediated by both environment and genetics. The cleardemarcation, in our factor analysis, of reactive attachmentdisorder behaviours from other forms of psychopathology mayhelp clinicians develop appropriately targeted treatments forthese behaviours. Future research identifying the candidate genes and the types of environments that have a causal rolewill have a major impact on prevention and intervention strategies.

REFERENCES

TOP