The effect of birth-weight with genetic susceptibility on depressive symptoms in childhood and adolescence

A total of 1,581 families, a sub-sample from a systematically ascertained, population-based register of all twin births between 1980 and 1991, the Cardiff Study of All-Wales and North-west England Twins (CASTANET) were invited to participate. Twins on this register were identified through child health databases and the characteristics of the sample have been described in detail elsewhere [38, 47] but are representative of the local population in terms of social class and ethnicity. The percentages of families from social class groups I, II, III, IV and V were 11.4%, 29.3%, 30.1%, 4.9% and 1.4%, respectively, with the remainder classified as at-home or unemployed. These figures are in keeping with those expected from a population from Greater Manchester [36]. The present wave of data collection took place in 2000 including twins from the register aged less than 17 years (range 8–17) and in full-time education born in Greater Manchester. The mean age of participating twins was 11.17 (standard error=0.068). Completed questionnaires were received from 1,023 families, giving an overall response rate of 65% (1,023/1,581). Data on birth-weight, antenatal and perinatal factors and social class had been collected 3 years previously. There were 934/1,545 twin pairs who took part at both assessment points (60%). (Note: There were 934 pairs who took part in both waves of the study but missing computed scores for some variables meant that they were not included in regression analyses). There were no significant differences between those individuals who had complete data available and those who did not take part at Time 2 in terms of birth weight (t=0.545, SE=0.15, P=0.586) or those who did not take part at time 1 in terms of depressive symptoms (t=0.730, SE=0.06, P=0.465). In addition, information on emotional symptoms as assessed by parental report on the Rutter questionnaire (40) at Time 1 did not reveal any significant differences between those with complete data and those who did not participate at Time 2 (t=1.114, SE=0.06, P=0.465). Multi-centre ethical approval was obtained. The study was described to participants and informed consent was obtained following return of the completed questionnaire package. Complete data on birth weight, gestation and depressive symptoms were available for 370 monozygotic (MZ) and 495 dizygotic (DZ) pairs. Zygosity was assigned according to parent responses to the twin similarity questionnaire, which has been found to be over 90% accurate in distinguishing MZ and DZ twins. An algorithm based on previous work was used [8, 47].

Mothers of the twins reported information on antenatal and obstetric factors using a questionnaire adapted from Lewis & Murray [27], which included the twins’ birth-weights, number of weeks of gestation at birth, maternal age at birth and maternal smoking during pregnancy (yes/no). Mother reports of birth weight and gestation have been found to be very accurate in comparison to antenatal records with correlations of over 0.90 on average [16, 51]. Social class information was also collected and classified according to Standard Occupational Classification [31]. Depression in the children and mothers was assessed 3 years later in the year 2000. Child depressive symptoms were assessed using the Mood and Feelings Questionnaire (MFQ) [11]. The MFQ is a 34-item scale based on DSM-III-R symptoms of depression that is a reliable quantitative measure of depression and a useful screening questionnaire for clinical depression in children in the community [11, 48] as well as in clinical populations [57]. Parent reports of children’s depression symptoms were used as child reports are only reliably obtained for those over the age of 11 years. The mean depression score for the sample was 9.15 (range 0–60). Boys’ scores (mean=8.63; range 0–43) were lower than those of girls (mean=9.59; range=0–60) although this difference was not significant (t=−1.45, SE=0.05, P=0.15). Given that there was no significant gender difference in mean symptom counts for the sample, and no evidence for gender differences in the magnitude of the genetic influence on parent-rated depressive symptoms in a previous analysis of this sample [38], the primary analyses were conducted using the combined sample rather than separately for each gender. Parents also completed the depression sub-scale of the Hospital Anxiety and Depression Scale (HADS) [59] about their own depression symptoms in the past week. The mean depression score for parents was 4.46 (range 0–21). The majority of analyses were based on total symptom scores given that there are strong arguments for regarding depression as a continuous dimension of risk rather than as a categorical disorder (affected/unaffected) both in adults and in children. Firstly, studies of adults and children show that depressive symptoms behave as a continuum with no evidence of increased morbidity once clinical diagnostic criteria are met [23, 51]. For example, Pickles & colleagues [33] found that number of depressive symptoms but not functional impairment predicted future depressive episodes. Second, sub-clinical depressive symptoms are associated with impairment [2, 26]—one of the main criteria for fulfilment of disorder. Third, sub-clinical symptoms are associated with service use and significantly increase the risk for depressive disorder in adulthood [21, 34]. Moreover, it appears that a continuous measure of depression is more indicative of genetic risk in children and adolescents. Several studies have found that the aetiology of very high levels of depressive symptoms is less influenced by genetic factors than depressive symptoms across the normal range [13, 14, 38]. Thus, at least in children, symptoms of depression may give a better indicator of genetic risk than categorical diagnoses. Nonetheless, because of the suggestion of aetiological heterogeneity, depressive symptoms were analysed both as a continuous and a categorical measure in this report. Birth weight was examined both as a dimensional measure and as a category since both birth weight across the continuum and LBW have been reported to be associated with depression in adults [7, 15, 32].

Data were analysed using STATA [45] and SPSS [46]. We adopted two approaches for testing the effect of birth-weight for gestation and genetic/familial risk. First, genetic regression analysis [22, 56] was used to test for genetic effects and for interaction of genetic risk and birth-weight. We made use of the twin design that relies on the observation that identical (monozygotic; MZ) twins share all of their genes in common (coefficient of relationship=1), while on average, non-identical (dizygotic; DZ twins) share only half of their genes in common (coefficient of relationship=0.5). Genetic regression analysis is undertaken using standard forced entry linear regression analysis with one twin’s depression symptom scores being the dependent variable. The twin’s depression score was predicted using the formula T=B₀+B₁ C+B₂ R+B₃ I+B₄ M+B₅ G where T=the twin depression score, B₀=the regression constant, C=the co-twin depression score, R=a coefficient of genetic relatedness (0.5 for DZ twins; 1 for MZ twins), I is the interaction term (C*R*birth-weight corrected for gestation), M=the main effect of birth-weight (with the length of gestation regressed out) and G=twin gender. One twin’s depression score was entered as the dependent variable, with co-twin depression, genetic relatedness and the interaction term as independent variables. Regression analyses were carried out using both continuous and categorical measures of birth-weight corrected for gestational age. Low birth-weight for gestation was classified as ≤10th centile since standard published cut points were not available for birth weights of twins.

Next, linear regression analysis was used to test for a main effect of birth-weight on depressive symptoms. The survey commands in STATA were used to take account of clustering within twin pairs by likening the twin data to a two-stage cluster design with the twin pairs as the primary sampling unit. A robust variance matrix calculation is utilised that relaxes the assumption of independence within groups. Birth-weight was corrected for number of weeks gestation at birth by regressing out the effect of length of gestation and using the unstandardized residuals for further analysis. As boys tended to be heavier at birth than girls, and the association between birth-weight and depressive symptoms may differ by gender. The effect of sex was controlled for by including sex as a predictor variable in the regression model. A number of factors have been reported to be associated with LBW [12]. Information on three potential influences was available—maternal smoking during pregnancy, social class and maternal age at birth. Social class was not associated with birth weight for gestation (b=−4.20, SE=3.32, P=0.206) or depressive symptoms (b=0.147, SE=0.107, P=0.169) in the present sample. Maternal pre-natal smoking (yes/no) was inversely associated birth weight for gestation (b=−141.62, SE=18.32, P=0.001) and positively associated with depressive symptoms in the offspring (b=2.16, SE=0.618, P=0.001). Maternal age at birth of child was negatively associated with child depressive symptoms (b=−.143, SE=0.053, P=0.007) and positively associated with twin birth weight (b=6.68, SE=1.65, P=0.001). These three variables were entered as predictor variables in regression models in addition to child gender.

The second approach to test the effects of birth-weight and genetic/familial risk was to use standard regression analysis but this time defining (1) genetic and (2) familial risk according to cut points on the depression questionnaires for (1) the co-twin and (2) the mother. Regression analysis was used to estimate the unit increase in depressive symptoms per unit decrease in birth weight for those individuals at genetic risk or family risk of depression and those not at risk. Both familial risk and genetic risk for depression was used for this analysis because to estimate the unit increase in symptoms by genetic risk meant restricting the analysis to MZ twins and thus a substantial reduction in sample size. Using familial risk allowed results to be replicated using the much larger full sample. Furthermore, using mother’s depression to index familial risk rather than the DZ twins allowed the entire sample to be double entered i.e. each twin to be entered as an individual in the analysis. This is because maternal depressive symptoms is a variable that is obligatory shared for each member of a twin pair. Appropriate correction for the non-independence of twins was undertaken using the survey commands in STATA as described above. For genetic risk, birth-weight and gestation were entered as independent variables to predict MZ twin depressive symptoms for MZ twins individuals without an affected co-twin (MZ co-twin depressive symptoms <90th centile) and those with an affected co-twin (MZ co-twin depressive symptoms ≥90th centile). This analysis was then repeated for the whole sample using parent depression symptoms as the cut point (parental depression score ≥90th centile) as a measure of familial risk. Dummy coded groups [9] were then analysed to assess whether there was an interaction between genetic or familial risk for depression and birth weight in predicting early depressive symptoms. These dummy variables allow the difference between the beta coefficients for the low-risk and high-risk groups to be tested.

Parent-rated MZ mean depression symptoms were lower than those of DZ twins but this was not significant (MZ mean=8.97, standard deviation=9.49; DZ mean=9.14, standard deviation=8.92; t=0.980, df=967, P=0.327). There were no significant differences on the ante-natal and labour complications assessed according to zygosity (admission to hospital for high blood pressure χ²=0.328, P=0.609, vaginal bleeding χ²=0.088, P=0.800, labour less than 3 h χ²=1.035, P=0.345, emergency Caesarean, χ²=2.270, P=0.151, maternal smoking during pregnancy, χ²=0.192, P=0.697, forceps or ventouse delivery? χ²=1.398, P=0.237).

Descriptive statistics for the birth-weight and weeks gestation of twins are shown in Table 1. It can be seen that second-born twins were lighter at birth than first-born twins. MZ twins were lighter than DZ twins although they also tended to be born earlier in pregnancy. The average birth-weight for the entire sample was 2469.81 g (standard deviation=553.38) and the average length of gestation was 36.36 weeks (standard deviation=2.87).

Linear regression showed that birth-weight for gestation significantly predicted parent-rated child depression symptoms when controlling for child gender (b=−0.001, SE=0.001, P=0.003, 95% CI −0.003, −0.0006) (n=1,740). This remained significant when child gender, maternal age at birth, pre-natal smoking and social class were also included in the regression model (b=−0.001, SE=0.001, P=0.015, 95% CI=−0.003, −0.0003). This illustrates that children who were small for gestation had higher depressive symptoms in childhood and adolescence.

The results of genetic regression analyses with depression as a continuous variable are shown in Table 2. It can be seen from Table 2 that birth-weight for gestation, genetic risk and maternal smoking during pregnancy showed main effects on depressive symptoms. When birth-weight for gestation was included as a continuous variable there was no birth-weight by genetic risk interaction effect although there was a slight trend (β=−0.048, P=0.096) (n=861). Data were also analysed using LBW as a categorical variable. Significant main effects of birth-weight for gestation, genetic risk, pre-natal smoking and a significant genetic risk by LBW interaction were found (β=0.394, P=0.001) (n=861). This relationship is illustrated graphically in Fig. 1 and shows a stronger relationship between genetic risk (C*R) and depressive symtpoms for those individuals who were small for gestational age at birth (≤10th centile). This suggests that it is LBW for gestational age that interacts with genetic risk for depression when depression is assessed as a continuous variable.

When twins were classified at genetic risk of depression if they had an affected MZ co-twin, the coefficients derived showed that for those twins at genetic risk for depression (N=32), for each 1 kg decrease in birth weight corrected for gestation, gender, social class, maternal age at birth and pre-natal smoking there was an increase of 2 depressive symptoms. As birth weight was measured in grams, this was calculated by the regression coefficent multiplied by 1000 (−0.002×1,000=−2). For those identical twins (N=338) not at genetic risk, the increase in depressive symptoms for each 1 kg decrease in birth weight was 1 symptom (−0.001×1,000=−1). Thus, there was a two fold increase in the number of depressive symptoms for each 1 kg decrease in birth-weight for twins at genetic risk (2 vs. 1 symptom) when controlling for covariates. Similarly, for each decrease in birth-weight of 1 kg there was an increase of 7 depressive symptoms for those at familial risk of depression (N=196) (−0.007×1,000=−7) and an increase of 1 depressive symptoms for those not at familial risk of depression (N=1478) (−0.001×1,000=−1) representing a seven-fold increase in number of symptoms.

Regression with dummy coded variables showed that the effect of birth-weight on depressive symptoms was not significantly different for those at high genetic risk of depression compared to those not at genetic risk of depression (for interaction term, β=0.009, P=0.853). It should be noted that regression with dummy coded variables for genetic risk of depression required restriction to MZ twins only and thus a substantial reduction in sample size. There was, however, a significant interaction according to familial risk (β=−0.086, t=−2.381, P=0.017).