Depressive symptoms are known to be prevalent in prepubescent children (e.g., Costello, Mustillo, Erkanli, Keeler, & Angold, Reference Costello, Mustillo, Erkanli, Keeler and Angold2003) and can affect both social and academic development (Edelsohn, Ialongo, Werthamer-Larsson, Crockett, & Kellam, Reference Edelsohn, Ialongo, Werthamer-Larsson, Crockett and Kellam1992) and long-term mental health (Dunn & Goodyer, Reference Dunn and Goodyer2006). In order to develop preventative interventions to attenuate these negative outcomes, it is important to understand the mechanisms by which symptoms of depression first develop. Models aimed at explaining why some children are more prone to developing depressive symptoms than others have underscored the role of both nurture and nature. Some of the most consistent findings regarding children's depressive symptoms relate to the role of stress, with stressful life events implicated in the onset of symptoms, but chronic difficulties and adversities as significant contributors too (Eley & Stevenson, Reference Eley and Stevenson2000). Among the set of chronic difficulties, interpersonal problems such as poor parent–child relationships and negative peer relationships have been highlighted (Rudolph, Hammen, & Burge, Reference Rudolph, Hammen and Burge1997). Alongside environmental influences, data from family and twin studies also suggest the role of some genetic contributions to depressive symptoms in children, although studies vary over how large this contribution is (Lau, Rijsdijk, Gregory, McGuffin, & Eley, Reference Lau, Rijsdijk, Gregory, McGuffin and Eley2007; Rice, Harold, & Thapar, Reference Rice, Harold and Thapar2002). There is also some suggestion that genetic and environmental factors correlate and interact (Rice, Harold, Shelton, & Thapar, Reference Rice, Harold, Shelton and Thapar2006; Wilkinson, Trzaskowski, Haworth, & Eley, Reference Wilkinson, Trzaskowski, Haworth and Eley2013). However, what remains unknown is how genetic and environmental risks on symptoms are expressed.
One possibility that we have explored in previous studies is that genetic and environmental risks for children's depressive symptoms are expressed as biases in the way in which children process information about emotional events and activities (Eley et al., Reference Eley, Gregory, Lau, McGuffin, Napolitano and Rijsdijk2008; Gregory et al., Reference Gregory, Rijsdijk, Lau, Napolitano, McGuffin and Eley2007; Lau, Belli, Gregory, Napolitano, & Eley, Reference Lau, Belli, Gregory, Napolitano and Eley2012; Lau, Rijsdijk, et al., Reference Lau, Gregory, Goldwin, Pine and Eley2007). Similar to depressed adults (Kovacs & Beck, Reference Kovacs and Beck1978), children with mood symptoms have been reported to show biases in automatic forms of information processing such as in the interpretation and attribution of the causes of ambiguous, positive, and negative events (Dineen & Hadwin, Reference Dineen and Hadwin2004; Gladstone & Kaslow, Reference Gladstone and Kaslow1995). Analogous biases have also been shown in cognitions, which can reflect the cognitive products of maladaptive information processing such as in their perceptions and expectations of daily events and activities, and of other people (Gregory et al., Reference Gregory, Rijsdijk, Lau, Napolitano, McGuffin and Eley2007; Rudolph et al., Reference Rudolph, Hammen and Burge1997). It is plausible that this collection of “cognitive biases” is shaped by more distal sources of genetic and environmental influence. There is also some work showing that negative patterns of information processing and negative cognitions can be acquired through exposure to (Murray, Woolgar, Cooper, & Hipwell, Reference Murray, Woolgar, Cooper and Hipwell2001) and learning from negative social environments (Field, Reference Field2006; Haddad, Lissek, Pine, & Lau, Reference Haddad, Lissek, Pine and Lau2011). In contrast, there is currently little work investigating the extent to which these maladaptive information-processing styles and cognitions reflect markers of inherited risks on depressive symptoms in children. Addressing whether cognitive factors can reflect genetic vulnerability is important, not only for understanding genetic risk mechanisms and associated pathways, but also for dissecting depressive phenotypes in children into more genetically homogeneous subgroups that can inform the search for genes in molecular genetic studies.
Only a handful of studies have explored the genetics of maladaptive information-processing style and its association with depressive symptoms in children. One study showed that negative interpretational style (the tendency to draw negative interpretations of ambiguous words or scenarios) was moderately heritable (30%) and that genetic influences on this processing style overlapped with genetic influences on depressive symptoms, a genetic correlation of 0.65 (Eley et al., Reference Eley, Gregory, Lau, McGuffin, Napolitano and Rijsdijk2008). However, in another study of the same sample, this time measuring attributional style (the tendency to attribute positive and negative events to internal or stable, global or specific, and stable or unstable causes), genetic influences were minimal (Lau et al., Reference Lau, Belli, Gregory, Napolitano and Eley2012). Instead, across two time points (at age 8 and 10 years), a pattern of shared and nonshared environmental influences contributed to negative attributions. Shared environmental influences are family-wide environmental influences that contribute toward the phenotypic similarity of family members growing up in the same environment, while nonshared environmental influences are individual-specific environmental influences that contribute toward differences among family members. Data analysis also showed that the same shared environmental influences that contributed to negative attributions also tended to contribute to depressive symptoms (via a shared latent psychometric factor), suggesting that in this period of middle to late childhood, attributional style may reflect family-wide environmental risks, rather than inherited risks on depressive symptoms. One study examined the genetics of negative interpersonal cognitions and links with depressive symptoms in 8-year-old children (Gregory et al., Reference Gregory, Rijsdijk, Lau, Napolitano, McGuffin and Eley2007). Negative interpersonal cognitions may reflect the products of biased information processing. This study found that shared environmental rather than genetic influences shaped both negative and positive perceptions about the self and others, as well as negative expectations of both peers and parents. Moreover, like the findings on attributional style, these dysfunctional interpersonal cognitions held shared and nonshared environmental influences in common with depressive symptoms, again suggesting that these negative cognitions are likely to reflect environmental experiences of children rather than inherited risks for symptoms at age 8 years.
The paucity of studies investigating whether maladaptive information-processing styles and cognitions reflect genetic or environmental risks on children's depressive symptoms makes it difficult to draw firm conclusions across studies, particularly given mixed findings across these different processing styles and cognitions, with some studies reporting modest heritability of cognitive biases and others suggesting that biases reflect environmental experiences. Adding additional complexity is that in adolescence, some information-processing styles, attributional style in particular, have been found to be genetically influenced and to share genetic overlap with depressive symptoms (Lau & Eley, Reference Lau and Eley2008; Lau, Rijsdijk, & Eley, Reference Lau, Rijsdijk and Eley2006; Zavos, Rijsdijk, Gregory, & Eley, Reference Zavos, Rijsdijk, Gregory and Eley2010). These age differences in the findings may be because attributional style does not fully mature (i.e., is not stable, or does not show traitlike qualities or act as a diathesis–stress factor) until adolescence (Cole & Turner, Reference Cole and Turner1993; Turner & Cole, Reference Turner and Cole1994). As such, any mediation of genetic risk on depressive symptoms may be contingent on the developmental maturity of the specific process under assessment. In comparison, as interpretational style has been found to operate as a trait-vulnerability factor for emotional symptoms in early childhood (Pass, Arteche, Cooper, Creswell, & Murray, Reference Pass, Arteche, Cooper, Creswell and Murray2012; although see Haller, Cohen Kadosh, & Lau, Reference Haller, Cohen Kadosh and Lau2013, for the developmental timecourse of interpretational style), it may be more likely to reflect inherited risk for depressive symptoms from a young age, consistent with the findings from genetic designs.
It is as yet unclear when negative interpersonal cognitions mature as stable, vulnerability factors for depressive symptoms, and if this has implications for whether these reflect inherited versus environmental risks. The present study addresses these questions. While the previous study (Gregory et al., Reference Gregory, Rijsdijk, Lau, Napolitano, McGuffin and Eley2007) analyzed data from Wave 1 of the Emotions, Cognitions, Heredity and Outcomes (ECHO) twin study (when children were aged 8 years), here we combined those data with data collected when children were aged 10 years. Using the two waves of data together, we examine the extent to which these interpersonal cognitions are “mature.” We explore this using two indices of developmental maturity: whether these factors showed stability across two time points and whether these operate as trait-vulnerability factors by predicting depressive symptoms within and across time (from ages 8 to 10 years). Next, we explore the extent to which these negative interpersonal cognitions reflect inherited and/or environmental risks for children's depressive symptoms across 8 and 10 years.
Methods
Sample
Data come from monozygotic (MZ) and dizygotic (DZ) twins taking part in the ECHO study (Eley, Gregory, Clark, & Ehlers, Reference Eley, Gregory, Clark and Ehlers2007). ECHO twins were selected from the Twins Early Development Study (Trouton, Spinath, & Plomin, Reference Trouton, Spinath and Plomin2002), a larger ongoing longitudinal sample of twins born in England and Wales during 1994–1996. The ECHO Wave 1 sample comprised 300 8-year-old twin pairs (mean age = 8.47 years, SD = 0.18): 247 pairs were selected from the Twins Early Development Study on the basis of high parent-reported anxiety at age 7 (those scoring in the top 15% for anxiety); 53 control twin pairs not scoring in the top 15% of anxiety scores were also chosen to ensure coverage of the full range of scores on test measures. After the testing session at Wave 1, data from 11 families were considered unusable because of autistic spectrum disorders, severe receptive language impairments, or persistent attention problems in at least one of the twins, so only 289 families were contacted at Wave 2 approximately 2 years later (Lau, Gregory, Goldwin, Pine, & Eley, Reference Lau, Gregory, Goldwin, Pine and Eley2007). Of these, 250 families (87%) agreed to participate. Mean age of twins at Wave 2 was 10.09 years (SD = 0.26). At Wave 1, the sample was predominantly (80%) White, and all parents of children in the sample were in employment and had remained in education until 18 years of age.
Data collection for ECHO took place at the Institute of Psychiatry in London, with a small number of families visited in their homes. Written informed consent was obtained from parents of all twins. Current analyses utilized Waves 1 and 2 data. Full data for all questionnaires across waves were available for 245 twin pairs. Of these, there were 34 male MZ pairs, 48 female MZ pairs, 25 male DZ pairs, 42 female DZ pairs, and 96 opposite-sex DZ pairs. Because the ECHO sample is subject to selection biases (oversampling symptomatic children) and response biases (individuals with mothers reporting higher levels of emotional symptoms and who experienced greater negative life events were less likely to participate at Wave 2), a weighting factor was constructed for use in subsequent analyses. This multiplied together the ratio of the probability of selection of high symptom families to that of nonsymptom control families, with the inverse of the predicted probability of families remaining at Wave 2 using significant predictors. By incorporating weights into analyses, parameter estimates are adjusted to reflect less weight being assigned to individuals from categories overrepresented (and greater weight to those underrepresented) by the sampling process.
Measures
Depressive symptoms
These were measured using the Children's Depression Inventory (Kovacs, Reference Kovacs1985), a 27-item self-report questionnaire adapted from the Beck Depression Inventory for use in children and adolescents. Individual items consist of three statements about the frequency with which a depressive symptom has occurred over the past 2 weeks (e.g., 0 = “I am sad once in a while,” 1 = “I am often sad,” and 2 = “I am sad all the time”). Total scores range between 0 and 54, with higher scores indicating higher levels of depression. One item concerning thoughts about suicide was removed from the questionnaire for ethical reasons (it was deemed inappropriate for 8-year-old children). The internal consistency (α) of the remaining items was 0.82 at 8 years and 0.80 at 10 years. The measure has demonstrated good discriminant and convergent validity in 6- to 16-year-olds (Hodges, Reference Hodges1990).
Peer perceptions
This was measured using the Perception of Peers and Self Questionnaire (Rudolph, Hammen, & Burge, Reference Rudolph, Hammen and Burge1995), which assesses children's perceptions of themselves and others in social contexts. It comprises 30 items, 15 of which concern beliefs about others (e.g., “Other kids are pretty helpful when you need them” and “Other kids can sometimes be pretty mean”) and 15 of which concern beliefs about the self (e.g., “I am a lot of fun to be with” and “It's a waste of other kids' time to be friends with me”). Children rate how much they agree with each item on 4-point Likert-like scales (1 = not at all, 4 = very much). Some items are positive (“Other kids are pretty easy to get along with”); others are negative (“Other kids are really out to get you”). Positive items are reverse-coded such that a higher total score indicates more negative perceptions. Rudolph et al. (Reference Rudolph, Hammen and Burge1995) found test–retest reliabilities of r = .69, p < .0001 for both other and self subscales over a 1-month period, and r = .55, p < .005 and r = .60, p < .002 for these same subscales over a 5-month period.
Social expectations
The Children's Expectations of Social Behaviour Questionnaire (Rudolph et al., Reference Rudolph, Hammen and Burge1995) provided an index of children's expectations about the prospective behavior of their mothers and peers. Children are read 30 descriptions of hypothetical interpersonal situations (15 featuring their mother and 15 featuring peers) and instructed to choose the most likely outcome from three alternatives for each. The three alternative responses include a positive or accepting behavior, an indifferent behavior, and a negative, hostile, or rejecting behavior. One item is “You see some kids playing a game during break one day so you go over and ask if you can play with them. What do you think they might say?”; the three responses are “They might tell me to join in and make room for me” (positive), “They might just act like I wasn't even there and keep playing” (indifferent), and “They might say mean things about me and tell me to go away” (negative). Responses are rated 0 (positive), 1 (indifferent), or 2 (negative). Summing across items, higher scores indicate more negative expectations of social partners' behaviors. The Children's Expectations of Social Behaviour Questionnaire shows high test–retest reliability in 7- to 12-year-olds over a 5-month period: r = .82 (mother items), r = .68 (peer items).
Analyses
Phenotypic analyses
Confirmatory factor analyses were first used to replicate whether the same four interpersonal cognitive factors identified at Wave 1 (i.e., at age 8) also characterized children at Wave 2 (i.e., at age 10). These were conducted first for half of the sample (where only one twin from each twin pair was selected randomly) and then conducted in the second half, such that the second half reflected an internal replication of the first set of results. Correlational analyses were performed next between Waves 1 and 2 variables to identify the following: (a) significant concurrent phenotypic associations between interpersonal cognitive factors and depressive symptoms at Wave 2, (b) the stability of each interpersonal cognitive factor from Wave 1 to Wave 2, (c) significant phenotypic associations between Wave 1 depressive symptoms and Wave 2 interpersonal cognitive factors, and (d) significant phenotypic associations between Wave 2 depressive symptoms and Wave 1 interpersonal cognitive factors. Again, these were conducted for each half of the sample separately.
Regression analyses were then conducted to investigate the extent to which Wave 2 depressive symptoms were predicted by each Wave 1 interpersonal cognitive factors over and beyond effects of Wave 1 depressive symptoms. These analyses were conducted in the whole sample to ensure maximal power, but while controlling for nonindependence of cases by clustering observations within families.
Genetic analyses
We estimated genetic and environmental effects on Wave 2 interpersonal cognitive factors. Of note, genetic and environmental influences on Waves 1 and 2 depressive symptoms and Wave 1 interpersonal cognitive bias measures have already been reported elsewhere (Gregory et al., Reference Gregory, Rijsdijk, Lau, Napolitano, McGuffin and Eley2007). Estimates of genetic and environmental influences are made by comparing within-pair similarity (twin correlations) among MZ twins, who share 100% of their genetic makeup, and DZ twins, who share on average 50% of segregating genes. Higher MZ compared to DZ resemblance is attributed to increased genetic similarity among MZ twins and used to estimate genetic (a 2) influences that are additive. Within-pair similarity not due to genetic factors is assigned as shared environmental variance (c 2) contributing toward resemblance among individuals in the same family. Finally, nonshared environmental influences (e 2) create differences among individuals from the same family and are estimated from within-pair differences between MZ twins (i.e., the one minus MZ twin correlations), although this term also includes measurement error.
Next, we used multivariate model-fitting analyses to confirm and refine these estimates on each measure, and also to examine the extent to which genetic, shared, and nonshared environmental influences between interpersonal cognitive factors and depressive symptoms overlapped within and across time. Only three of the four Wave 1 interpersonal cognitive factors significantly predicted depressive symptoms at Wave 2 in phenotypic analyses. Thus, three independent pathway models (Figure 1) were run to investigate the extent to which a common genetic and/or environmental factor explained phenotypic overlap between the interpersonal cognitive factor and depressive symptoms within and across time, and the extent to which each measure was explained by specific genetic and/or environmental factors. A single model including all interpersonal cognitive factors as well as depression was not run, owing to power constraints due to the sample size. Parameters in multivariate models are estimated from cross-twin cross-measure covariance matrices. As can be seen from Figure 1, two sets of genetic and environmental influences are assumed: “common” genetic and environmental factors (AC, CC, EC) contribute to all measured variables, while “specific” genetic and environmental influences account for unique variance on each variable at each time point (ASD1, ASD2, ASI1, ASI2, CSD1, CSD2, CSI1, CSI2, ESD1, ESD2, ESI1, and ESI2). Total genetic, shared, and nonshared environmental effects on each individual measure can be obtained by summing all specific genetic and environmental paths to that measure with common genetic and environmental influences.
Independent factor models investigating the extent to which each interpersonal cognitive factor shares common genetic and environmental influences (AC, CC, and EC) with depressive symptoms within and across time and the extent to which specific genetic and environmental influences are important (ASD1, ASD2, ASI1, ASI2, CSD1, CSD2, CSI1, CSI2, ESD1, ESD2, ESI1, and ESI2).
Raw data (rather than twin variance–covariance matrices) were modeled to maximize the sample. To assess model fit, that is, how well-observed values compare to expected values, the –2 log likelihood statistic of the data is produced. Although the –2 log likelihood does not represent an overall measure of model fit, relative measures of fit, such as chi-square (χ2), can be obtained by subtracting the difference in the log-likelihood statistic of a tested model with that of a saturated model containing the same number of measured variables. A saturated model estimates the maximum number of parameters to describe variances, covariances, and means of all measured variables from raw data. The lower χ2 relative to the degrees of freedom (i.e., a nonsignificant χ2) generally indicates a better fit of the model to the data. Another consideration of model fitting, parsimony, is to identify the best fitting model with the fewest parameters. The Akaike information criterion (AIC), which is calculated as χ2 – 2 df, is an index of both goodness of fit and parsimony. When comparing AIC values across models, more negative values indicate better fit relative to the number of parameters estimated. Models were fit to age-regressed and where appropriate log-transformed scores to minimize mean age effects and to correct for positive skew. Mean or variance differences between males and females identified by descriptive analyses were incorporated in genetic models.
Results
Interpersonal cognitive factors at Wave 2 (at age 10)
Confirmatory factor analyses conducted in each half of the Wave 2 twin sample showed that the factor structure at Wave 2 (i.e., at age 10) closely replicated that at Wave 1 (at age 8), with the same four factors emerging across waves and with similar items loading onto each (see online only Supplementary Table S.1). The four factors were Absence of Positive Peer/Self-Perceptions, Negative Peer/Self-Perceptions, Negative Expectations of Mother, and Negative Expectations of Peers. Factor scores were generated by summing the scores of the items loading onto each factor. Each factor comprised 15 summed items and showed good reliability with Cronbach α values of 0.85, 0.76, 0.78, and 0.67, respectively. Means and standard deviations of the four factors, together with depressive symptom measures at Wave 2, are reported in Table 1. Model-fitting analyses used to investigate mean sex differences on each measure revealed that, at Wave 2, depressive symptoms and interpersonal perceptions and expectations were comparable across boys and girls. To assess whether depressive symptoms and these factor scores changed much at the mean level, Wave 1 variables are also included in Table 1. As can be seen, mean levels were comparable across waves.
Means (standard deviations) for Wave 2 measures and correlations between variables at Wave 2 and cross-wave
Note: The correlations in parentheses reflect internal replication in the other half of the sample. Correlations in italic do not reach statistical significance at p < .05.
Four sets of correlations (for each half of the sample) are presented in Table 1. The first set shows that all Wave 2 interpersonal cognitive bias measures correlated significantly with Wave 2 depressive symptoms, suggesting cross-sectional associations between cognitions and symptoms. The second set shows significant cross-time correlations for all interpersonal cognitive bias measures from Wave 1 to Wave 2. The third set of correlations show that Wave 1 depressive symptoms (mostly) significantly correlated with all Wave 2 interpersonal cognitive bias measures. The last set of correlations show that Wave 1 interpersonal cognitive bias measures significantly correlated with Wave 2 depressive symptoms.
Regression analysis to identify significant early predictors of Wave 2 depressive symptoms
In separate models, three of the four interpersonal cognitive factors at Wave 1 (Absence of Positive Peer/Self-Perceptions, Negative Expectations of Peers, and Negative Expectations of Mothers) significantly predicted variance in depressive symptoms at Wave 2, over and beyond Wave 1 depressive symptoms in regression analyses (Table 2). Negative peer/self-perceptions predicted depressive symptoms at Wave 2 at trend-level significance.
Regression analysis investigating the effects of Wave 1 interpersonal cognitive bias measures on depressive symptoms at Wave 2 after controlling for depressive symptoms at Wave 1
MZ and DZ twin correlations
MZ and DZ correlations for each Wave 1 and 2 measure are presented in Table 3. In Wave 1, depressive symptoms reflected a combination of modest genetic influences, shared environmental influences, and largely nonshared environmental influences. As can be seen, the pattern of twin correlations for Wave 2 depressive symptoms support mainly shared environmental and nonshared environmental influences. For the Absence of Positive Peer/Self-Perceptions factor, both Wave 1 and Wave 2 twin correlations suggest modest heritability but again shared environmental and nonshared environmental contributions. Wave 2 negative peer/self-perceptions and negative expectations of peers support modest heritability, but Wave 1 twin data on these variables suggest mainly shared and nonshared environmental influences. For negative expectations of mother, MZ and DZ twin correlations suggested mainly shared and nonshared environmental effects.
Monozygotic (MZ) and dizygotic (DZ) twin correlations of Wave 1 and 2 variables
a These data were reported in our previous study (Gregory et al., Reference Gregory, Rijsdijk, Lau, Napolitano, McGuffin and Eley2007).
Genetic analysis to identify genetic influences on interpersonal cognitive factors and the genetic overlap with depressive symptoms across time
Parameter estimates from three independent pathway models are presented in Table 4. All models fit well, as indexed by the root mean square error of approximation (Steiger, Reference Steiger1990; Steiger & Lind, Reference Steiger and Lind1980) and the AIC (Akaike, Reference Akaike1974, Reference Akaike1987). All root mean square error of approximation values were ≤0.03, and all AIC values were <–29 (see Table 4 for full fit statistics). Estimates of heritability and shared and nonshared environmental effects can be calculated by summing the appropriate common and specific factors together. Across all three models, adding up the estimates for common and specific genetic, shared, and nonshared environmental effects for Wave 1 depressive symptoms, estimates for total genetic influence varies from 0 to 0.33 and total shared environmental effects from 0.07 to 0.26. These estimates are similar to the ones reported in our previous paper. Summing the common and specific genetic estimates for Wave 2 depressive symptoms yields similar total estimates that vary between 0.04 and 0.24 (falling within the 95% confidence interval estimates made in any one model), while total shared environmental effects are estimated between 0.16 and 0.29. At first glance, these estimates appear to be discrepant with the MZ and DZ twin correlations, which suggest no genetic influence (Table 3). However, multivariate models also draw on cross-twin cross-measure MZ and DZ correlations (see Supplementary Table S.2), and these supply additional information that contribute to the heritability estimates of each individual measure. For example, it can be seen that correlations between Wave 2 depressive symptoms for Twin 1 and Wave 1 depressive symptoms for Twin 2, for MZ and DZ, are 0.33 and 0.25, respectively. Because the MZ cross-twin cross-measure correlation is larger than the corresponding DZ correlation, this suggests that some genetic influence is likely to contribute toward the relationship between these two measures (i.e., depressive symptoms at Wave 1 and Wave 2). Similarly, the MZ and DZ cross-twin cross-measure correlations for Wave 2 depressive symptoms and absence of positive peer/self-perceptions are 0.23 and 0.13. Again, the larger MZ correlation suggests the role of genes in explaining the phenotypic correlation between these measures. These additional pieces of information thus indirectly implicate genetic influences to each measure alone, explaining the discrepancy in the genetic estimates derived from multivariate models and the univariate MZ and DZ twin correlations.
Squared parameter estimates from independent pathway models of interpersonal cognitive factors and depressive symptom scores over time
Note: The 95% confidence intervals are in parentheses and significant estimates at this level are in bold. –2LL, –2 Log likelihood; AIC, Akaike information criterion; RMSEA, root mean square error of approximation.
The absence of positive peer/self-perceptions measures support heritability at Wave 1 and Wave 2 (with total genetic estimates of 0.25 and 0.13) and nonshared environmental influences, again at both waves (with total estimates of 0.73 and 0.87). A similar pattern arose for negative peer/self-perceptions measures, with total genetic estimates of 0.26 and 0.32 at Wave 1 and Wave 2, respectively, and total nonshared environmental influences of 0.74 and 0.64. Total shared environmental influences were minimal for both of these interpersonal cognitive bias measures. Negative expectations of mother showed no genetic effects at Wave 1 or Wave 2 (total estimates of 0 and 0), but rather supported modest shared (total estimates of 0.08 and 0.18) and large nonshared environmental (total estimates of 0.92 and 0.82) effects. Finally, for negative expectations of peers, genetic influences appear to play a modest role at Wave 2 (total estimate of 0.32), but not at Wave 1 (total estimate of 0.02), where shared environmental influences are relatively more important (total estimate of 0.30). Total nonshared environmental effects are estimated at 0.68 and 0.62 for this variable at Waves 1 and 20 respectively.
In terms of common and specific genetic and environmental influences, significant common genetic effects were found between Absence of Positive Peer/Self-Perceptions and depressive symptoms within and across time, with some support also for common nonshared environmental influences. Specific nonshared environmental influences contributed to each variable too. Similarly, Negative Peer/Self-Perceptions also shared common genetic influences with depressive symptoms within and across time, which largely accounted for common variance between these variables. Specific nonshared environmental influences were also important. Negative expectations of peers and negative expectations of mother showed a more distinct profile of effects, with no support for common genetic influences across measures. Instead, common variance across measures seemed to be explained by common shared and nonshared environmental influences. Again, specific nonshared environmental influences were generally apparent on each measure.
Discussion
In the present study, we first sought to examine whether interpersonal cognitive bias factors previously linked to internalizing symptoms in children were maturing as stable, traitlike vulnerability factors to predict depressive symptoms within and across time. Cross-time correlations between waves showed that all four interpersonal cognitive factors were moderately stable across time. Mean levels of these four factors were also comparable across waves. Although all four factors at Wave 1 correlated significantly with Wave 2 depressive symptoms, only three (Absence of Positive Peer/Self-Perceptions, Negative Expectations of Peers, and Negative Expectations of Mother) significantly predicted depressive symptoms across time (in separate regression models), over and beyond their associations with Wave 1 depressive symptoms. Then we investigated the genetic and environmental contributions to these interpersonal cognitive factors and specifically whether they shared common genetic and environmental variance with depressive symptoms. Positive peer/self-perceptions and negative peer/self-perceptions were heritable and shared genetic risks with depressive symptoms. In contrast, negative expectations of peer and negative expectations of mother generally reflected overlapping shared and nonshared environmental influences on depressive symptoms.
Drawing on the sparse data in this area, we had speculated that maladaptive information-processing styles and cognitions might only come to mediate inherited risks once they are developmentally mature. Such a trend seems to characterize attributional style, which has been shown to only operate as a diathesis–stress factor for depressive symptoms in adolescence and not in childhood (Cole & Turner, Reference Cole and Turner1993; Turner & Cole, Reference Turner and Cole1994), and to only reflect genetic risks from adolescence onward (Lau & Eley, Reference Lau and Eley2008; Lau et al., Reference Lau, Rijsdijk and Eley2006) and not in childhood (Lau et al., Reference Lau, Belli, Gregory, Napolitano and Eley2012). Similarly, interpretational style, which may act as an early familial precursor for social anxiety symptoms, present in at-risk infants (Pass et al., Reference Pass, Arteche, Cooper, Creswell and Murray2012), also shows moderate heritability in childhood (Eley et al., Reference Eley, Gregory, Clark and Ehlers2007). In comparison, negative cognitions such as perceptions and expectations of daily events and activities and of other people show minimal genetic influences in childhood and, according to our previous study, reflect recent environmental experiences instead (Gregory et al., Reference Gregory, Rijsdijk, Lau, Napolitano, McGuffin and Eley2007). It may then be that these negative cognitions have not yet begun to develop traitlike qualities in middle childhood. By including an additional wave of data 2 years later, we were able to show across ages 8 and 10 years that preadolescent children's positive and negative perceptions of the self and other people, and negative expectations of peers and their mother, were showing moderate stability. With exception to negative self-/other perceptions, these variables also showed a temporal relationship with depressive symptoms across this period of middle to late childhood. These data are consistent with other studies explaining why preadolescent children with internalizing symptoms might attract and respond differently to negative stressors such as peer rejection (Caldwell, Rudolph, Troop-Gordon, & Kim, Reference Caldwell, Rudolph, Troop-Gordon and Kim2004; Rudolph et al., Reference Rudolph, Hammen and Burge1997).
If these interpersonal cognitive variables are beginning to show traitlike characteristics as risk factors for children's depressive symptoms, do they now also begin to reflect inherited risks? Our data provide tentative support for this hypothesis but only for the absence of positive self-/other perceptions. This interpersonal cognitive bias factor alone was beginning to show some traitlike qualities: moderate stability, moderate temporal precedence, and genetic effects that were shared with depressive symptoms within and across time. In contrast, negative peer/self-perceptions showed genetic effects, but their capacity to predict depressive symptoms across time was mediated through their concurrent effects on depressive symptoms. Although showing the capacity to predict depressive symptoms, negative expectations of peers and negative expectation of mother did not show heritability or shared genetic effects with depressive symptoms. Together these data may suggest that these interpersonal cognitive biases are beginning to stabilize and may therefore only reflect genetic risks at later stages of development beyond childhood. Nonetheless, it is interesting to note that the absence of positive perceptions may emerge first as an inherited precursor of children's depressive symptoms.
Why might maturation of particular cognitive traits elicit genetic contributions? Maturation is a gradual process involving protracted neural development but also experience-dependent pruning of these functions (Nelson, Lau, & Jarcho, in press). As these developmental neural changes emerge, it is possible that the opportunity for inherited biological differences to manifest on cognition increases. Several well-known indices of cognitive ability and functioning reflect increasing heritability with age (Haworth et al., Reference Haworth, Wright, Luciano, Martin, de Geus and van Beijsterveldt2010). While these hypotheses on how genetic effects on cognitive processing wax and wane across development to differentially shape depressive symptoms are intriguing, more systematic investigations of cognitive vulnerability factors, their stability and heritability, will have to be conducted.
These data and conclusions are subject to a number of limitations. First, all measures here are self-report. Although there are strengths of using self-reported data particularly on internalizing symptoms and cognitions concerning validity, these findings could be further strengthened with reports from peers or teachers. The exclusive reliance on self-reports may also have artificially increased associations between variables within and across time. Second, the sample size was relatively small and associated power was low, leading to wide confidence intervals that overlapped with zero among many of the parameter estimates of genetic models. Third, the cross-time correlations reported here may not be strong enough to conclude that cognitive biases are stable. The degree of the stability coefficients was moderate, indicating some degree of stability but also change. Furthermore, because only two waves of data were used in this study, it is difficult to conclude that these cognitive biases have stabilized at this stage of development. Fourth, the usual caveats associated with twin analyses, such as violations of the equal environments assumption, assortative mating, and differences between twin and nontwin individuals, may collectively act to alter estimated parameters.
Notwithstanding these limitations, our data provide some interesting findings on the role of negative cognitions, or rather the absence of positive self-/other perceptions, as reflecting early inherited markers of risk for children's depressive symptoms. What are the translational implications of these findings? Many psychological interventions, the frontline treatment for paediatric mood and anxiety conditions, target maladaptive information processing and cognitions. Although there is some debate over whether this is an age-appropriate strategy for children given their level of cognitive maturity (Grave & Blisset, Reference Grave and Blissett2004), the present results suggest that some biased cognitions relating to interpersonal situations are beginning to emerge in this age range, showing moderate stability and a capacity to predict depressive symptoms across time. Nonetheless, because the degree of stability coefficients was only moderate, this suggests that these biases are also amenable to change. Targeting these cognitions in this age range may therefore be important, as part of interventions for early-emerging mood symptoms. Novel interventions that target biases in information processing have so far aimed to modify patterns of attention-vigilance for threats and interpretative styles through training, with some success (Bar-Haim, Reference Bar-Haim2010; Lau, Reference Lau2013). Such training programs could be extended to challenge more concrete products of biased cognitive processing too, such as perceptions and expectations.
Supplementary Materials
The supplementary materials for this article can be found online at http://journals.cambridge.org/dpp.
