Introduction
There is consistent evidence that genes influence individual variation in the personality trait of neuroticism, as well as internalizing disorders such as anxiety and depression, and somatization syndromes (most commonly presenting as prolonged fatigue or chronic pain). In adolescents, these measures are moderately heritable [i.e. ~30–60% (Gillespie et al. Reference Gillespie, Evans, Wright and Martin2004; Rettew et al. Reference Rettew, Vink, Willemsen, Doyle, Hudziak and Boomsma2006; Lamb et al. Reference Lamb, Middeldorp, van Beijsterveldt, Bartels, van der Aa, Polderman and Boomsma2010; Bartels et al. Reference Bartels, van de Aa, van Beijsterveldt, Middeldorp and Boomsma2011)], consistent with that for adults (e.g. Kendler et al. Reference Kendler, Gardner, Gatz and Pederson2007; Vassend et al. Reference Vassend, Roysamb and Nielsen2011). Genetic covariation has been shown between neuroticism and anxiety and/or depression (Boomsma et al. Reference Boomsma, Beem, van den Berg, Dolan, Koopmans, Vink, de Geus and Slagboom2000; Hettema et al. Reference Hettema, Prescott and Kendler2004, Reference Hettema, Neale, Myers, Prescott and Kendler2006; Kendler et al. Reference Kendler, Gardner, Gatz and Pederson2007), neuroticism and somatic health (Vassend et al. Reference Vassend, Roysamb and Nielsen2011), anxiety and depression (Gillespie et al. Reference Gillespie, Zhu, Heath, Hickie and Martin2000; Kendler et al. Reference Kendler, Gardner, Gatz and Pederson2007) and anxiety, depression and somatic syndromes (Hickie et al. Reference Hickie, Kirk and Martin1999b; Gillespie et al. Reference Gillespie, Zhu, Heath, Hickie and Martin2000; Kato et al. Reference Kato, Sullivan, Evengard and Pedersen2009). However, it remains unclear if the genetic overlap between anxiety and depression symptoms and common somatic complaints such as prolonged fatigue and pain is due largely to their relationship with neuroticism.
Extensive co-morbidity between depression and somatoform symptoms is frequently reported (e.g. Goldberg, Reference Goldberg1996; Vaccarino et al. Reference Vaccarino, Sills, Evans and Kalali2009). Previous work, particularly in health care settings, has indicated the extent to which the two syndromes can be distinguished cross-sectionally and, more importantly, longitudinally (Hickie et al. Reference Hickie, Hadzi-Pavlovic and Ricci1997, Reference Hickie, Koschera, Hadzi-Pavlovic, Bennett and Lloyd1999c; Gillespie et al. Reference Gillespie, Kirk, Heath, Martin and Hickie1999; van der Linden et al. Reference van der Linden, Chalder, Hickie, Koschera, Sham and Wessely1999). Both syndromes appear to have relatively early ages of onset, with some subjects largely only ever developing one or other form of illness. Our previous genetic analyses (Hickie et al. Reference Hickie, Kirk and Martin1999b; Gillespie et al. Reference Gillespie, Zhu, Heath, Hickie and Martin2000) also showed that, while significant genetic risk is shared between measures of psychological and somatic distress, there were also independent genetic and environmental risk factors influencing somatic health. This has also been shown in a large Swedish study exploring somatic syndromes in women (Kato et al. Reference Kato, Sullivan, Evengard and Pedersen2009). A variety of different psychosocial and neurobiological paths have been proposed to explain both the common and distinctive aspects of the two syndromes (Rief et al. Reference Rief, Hennings, Riemer and Euteneuer2010).
Boomsma et al. (Reference Boomsma, Beem, van den Berg, Dolan, Koopmans, Vink, de Geus and Slagboom2000) found that genetic covariance between measures of anxiety, depression, somatic anxiety and neuroticism could be attributed to a common genetic source in adolescents. However, studies in young adults suggest that genetic covariation between anxiety and depression measures and somatic health may exceed that due to the relationship with neuroticism. Gillespie et al. (Reference Gillespie, Zhu, Heath, Hickie and Martin2000) showed that 67% of the genetic variance in somatic distress was due to sources that also influence measures of depression and phobic anxiety, whereas Vassend et al. (Reference Vassend, Roysamb and Nielsen2011) showed recently that only 35–48% of the genetic variance in somatic health appears due to a source influencing neuroticism.
Adolescence is the peak age of onset for all of the major adult psychiatric phenotypes, with the emergence of depressive disorders post-puberty being of greatest significance (Merikangas et al. Reference Merikangas, He, Burstein, Swanson, Avenevoli, Cui, Benjet, Georgiades and Swendsen2010). Of prime importance in clinical psychiatry is the identification of the earliest phenotypes that emerge during this key developmental period and the extent to which they predict transition to the major anxiety, mood and psychotic disorders in adult life. Such work lies at the heart of current international efforts to promote earlier intervention (Hetrick et al. Reference Hetrick, Parker, Hickie, Purcell, Yung and McGorry2008; Hamilton et al. Reference Hamilton, Naismith, Scott, Purcell and Hickie2011; McGorry et al. Reference McGorry, Purcell, Goldstone and Amminger2011) or ‘pre-emptive’ psychiatry (McGorry, Reference McGorry2011). Characterization of key genetic or environmental risks (and the extent to which they are shared or unique) during the same period has the capacity to inform both the type and timing of more relevant preventive and early intervention strategies. Other modelling of risks to depression based on twin studies in the teenage and early adult years indicate the extent to which there are likely to be both multiple relevant time points and changing patterns of both genetic and environmental risk (Kendler et al. Reference Kendler, Gardner and Lichtenstein2008).
In the current study, we explore these relationships in a primarily adolescent population sample comprising twins from the Brisbane Longitudinal Twin Study (Wright & Martin, Reference Wright and Martin2004). Measures of anxiety/depression and somatic distress, as well as a measure of overall mental health and well-being were assessed by self-report using the Somatic and Psychological Health Report (SPHERE) questionnaire (Hickie et al. Reference Hickie, Davenport, Hadzi-Pavlovic, Koschera, Naismith, Scott and Wilhelm2001a). The instrument was developed specifically to explore these types of relationships in those with common forms of psychological distress, but particularly those with affective syndromes. The subscales measure somatic and psychological symptoms independently (van der Linden et al. Reference van der Linden, Chalder, Hickie, Koschera, Sham and Wessely1999; Wijeratne et al. Reference Wijeratne, Hickie and Davenport2006) and characterize symptomatology (mood and behavioural features) as continuous dimensional traits, so are advantageous for genetic modelling in a population sample (i.e. twin studies have greater power to resolve sources of familial resemblance when using continuous compared with binary or ordinal data; Neale et al. Reference Neale, Eaves and Kendler1994).
Given the variations in prevalence and age of onset of depressive disorders between boys and girls in the post-pubertal period, we tested for potential differences in SPHERE scores by gender and explored differences across adolescence to young adulthood (12–25 years). Using a trivariate twin design, we then investigated the relationship of neuroticism, which captures trait-based anxiety present from early childhood and is likely to be indicative of genetic risk factors (Kotov et al. Reference Kotov, Gamez, Schmidt and Watson2010), to the co-morbidity found between anxiety/depression and somatic distress.
Method
Sample
The sample comprised 2459 adolescents and young adult twins [1168 complete pairs, 35.4% monozygotic (MZ), 53% female], mean age 15.5±2.9, range 12.0–25.6 years. Participants are typical of the South East Queensland adolescent and young adult population on a range of traits and had taken part in one or more studies (Fig. 1). Written, informed consent was obtained from all participants and a parent or guardian for those aged <18 years. The study was approved by the Human Research Ethics Committee at the Queensland Institute of Medical Research.
Fig. 1. Venn diagram showing study participation numbers for Somatic and Psychological Health Report (SPHERE) collection across three studies (Wright & Martin, Reference Wright and Martin2004), with added components for participants in studies 1A and 2 only (54 participants) and studies 1B and 3 only (four participants). Data were collected once only for 1281 participants [52% of total sample (n=2459)], twice for 523 (21%), three times for 526 (21%), four times for 129 (5%). A measure of neuroticism was available for 84% of the sample. For approximately one-third of the sample a single assessment of neuroticism from the NEO was available, a further third were assessed on the Junior Eysenck Personality Questionnaire (JEPQ) (one to two occasions) and the remaining third were assessed on both the NEO (single occasion) and JEPQ (one to three occasions; JEPQ and NEO were collected at the same time point for 393 individuals, r=0.71). The summed scores for each neuroticism measure were standardized (Z-scores: mean=0±1) and averaged to produce a composite measure. For participants with multiple SPHERE measures, a mean measure (and mean age) was used in analyses. Studies 1 and 2 are ongoing in-person studies of melanocytic naevi (moles) at age 12 and 14 years (studies 1A and 1B) and cognition at 16 (Study 2). Study 3 was a mail and phone study of health and well-being targeting adolescent and young adult twins. Exclusion criteria for the cognition study at 16 were parental report of head injury, neurological or psychiatric illness, substance abuse/dependence or current use of psychoactive medication in either twin. Availability was the only criterion for the other studies. We determined zygosity from DNA using a commercial kit (AmpFlSTR Profiler Plus Amplification Kit; ABI, USA) and this was later confirmed for >80% of the sample genotyped on a genome-wide single nucleotide polymorphism (SNP) genotyping platform [610K Illumina; Illumina Inc., USA (Medland et al. Reference Medland, Nyholt, Painter, McEvoy, McRae, Zhu, Gordon, Ferreira, Wright, Henders, Campbell, Duffy, Hansell, Macgregor, Slutske, Heath, Montgomery and Martin2009)].
Measures
SPHERE
Three measures were obtained from the 34-item SPHERE questionnaire (Hickie et al. Reference Hickie, Davenport, Hadzi-Pavlovic, Koschera, Naismith, Scott and Wilhelm2001a, Reference Hickie, Davenport, Scott, Hadzi-Pavlovic, Naismith and Koscherab). Participants indicated if they had been troubled by symptoms over the past few weeks, making one of three response choices: sometimes/never (coded as zero); often; most of the time (each coded as 1). Items were summed to obtain scores for SPHERE-34 (all 34 items), PSYCH-14 (14 items tapping anxiety/depression), SOMA-10 (10 items, non-overlapping with PSYCH-14, tapping somatic distress). Internal consistency was good (Cronbach's α=0.89 for SPHERE-34, 0.84 for PSYCH-14, 0.70 for SOMA-10). Items missing for 50 individuals (mean=1.3±0.8 items, ranging 1–6 items, <0.001% of the dataset) were imputed in prelis 2.30 (Scientific Software International, UK) based on sex, age and remaining items (i.e. 28–33 items).
SPHERE measures were collected on at least two occasions for just under half the sample (i.e. 48%; see Fig. 1). To increase test measurement reliability for this first analysis, we averaged the data collected on multiple occasions to give a single measure, together with an ‘average age’ at assessment. This decision was further supported by the subtleness and generally linear nature of change with age and the relative stability of the data over time (in a subsample of 91 individuals tested twice within 2–6 months, r=0.43 for SPHERE-34, 0.48 for PSYCH-14, 0.64 for SOMA-10).
Neuroticism was obtained from either 20 items (scored as yes=1, no=0) from the Junior Eysenck Personality Questionnaire (JEPQ; Eysenck, Reference Eysenck1972; Eysenck & Eysenck, Reference Eysenck and Eysenck1975) and/or 12 items (using a 5-point Likert scale and scored 0–4) from the NEO Five-Factor Inventory (NEO-FFI) of the NEO-PI-R (Costa & McCrae, Reference Costa and McCrae1992), with more recent assessments using revised versions (NEO-FFI-R, NEO-FFI-3: McCrae & Costa, Reference McCrae and Costa2004, Reference McCrae and Costa2010). Participants with missing items were excluded and neuroticism was included only when SPHERE had been collected on the same occasion (i.e. 2065 individuals, 84% of the SPHERE sample).
Statistical analyses
Distributions for each of the SPHERE measures were normalized by converting to a proportional scale before transformation into arcsin values (Freeman & Tukey, Reference Freeman and Tukey1950) (e.g. Birley et al. Reference Birley, Gillespie, Heath, Sullivan, Boomsma and Martin2006; Wray et al. Reference Wray, Birley, Sullivan, Visscher and Martin2007), with outliers (four to eight individuals) winsorized to s.d.±3.3. Neuroticism was normally distributed with no outliers. In addition, a single family was found to be outlying for the three SPHERE measures, using the %P option in Mx (Neale et al. Reference Neale, Boker, Xie and Maes2003), which provides a likelihood statistic for each family conditional on the genetic model, and this family was excluded from further analyses.
Modelling, which uses all data points regardless of missingness, was performed in Mx using a full information maximum likelihood estimator. The fit of constrained models was compared with the full model by examining the difference in the −2 log likelihood, which is distributed as a χ2 for given degrees of freedom. We first assessed homogeneity of sampling by examining the means and variances for birth order and zygosity effects as described in McGregor et al. (Reference McGregor, Pfitzner, Zhu, Grace, Eldridge, Pearson, Mayne, Aitken, Green and Martin1999), as well as the effects of sex, age and sex×age. Those with significant effects were retained as covariates. We also tested whether the twin correlations for both MZ and dizygotic (DZ) boys and girls could be set equal. If not, this is suggestive of magnitude differences in genetic and/or environmental estimates for boys and girls. Similarly, if correlations for opposite-sex DZ pairs are significantly lower than those of same-sex DZ pairs, this indicates different sources of influence between boys and girls.
At the univariate level and using the five zygosity groups (i.e. MZ females, MZ males, DZ females, DZ males and opposite-sex pairs), we decomposed the variance of each variable into additive genetic (A), common environmental (C) and unique environmental (E) sources of variance. We tested for sex limitation effects relating to the source of genetic influence by setting the correlation between additive genetic sources of influence on opposite-sex pairs to 0.5 and comparing the fit of this model with that of the fully saturated model in which the correlation was free to vary. Magnitude effects were examined by setting A, C, and E influences to be equal for boys and girls and comparing model fit with the fully saturated model, which allowed these estimates to vary.
Finally, we examined the covariation between neuroticism, PSYCH-14 and SOMA-10 in a multivariate sex-limitation model and in a model collapsed over sex. Cholesky decomposition and independent and common pathway modelling approaches (Neale & Cardon, Reference Neale and Cardon1992) were examined, which provide A, C, and E variance/covariance matrices from which genetic, common environmental and unshared environmental correlations can be calculated. As Cholesky decomposition is the standard general approach to decomposing variance into genetic and environmental sources, we used this model to test the significance of A and C influences. Akaike's Information Criterion (AIC) was examined to compare model fit between Cholesky, independent and common pathway models.
Results
Means, standard deviations and ranges, as well as sex and age effects for all measures, are shown in Table 1. We found no evidence of birth order [Δχ2 ranged 0.5–7.1, df=4 (i.e. Δχ2 4)] or zygosity effects (Δχ2 6 ranged 2.8–4.9), but significant sex and sex×age effects were found. Differences were subtle. Girls scored higher for SPHERE-34 (8.6 v. 8.2), PSYCH-14 (3.8 v. 3.4) and neuroticism (NEO: 23.2 v. 21.0; JEPQ: 10.1 v. 9.3). Post-hoc analyses showed that, for SPHERE-34 and SOMA-10, scores increased with age for girls (Δχ2 1=17.1 and 24.7 respectively), but decreased for boys (Δχ2 1=5.1 and 4.5 respectively). For PSYCH-14, scores increased with age for girls (Δχ2 1=19.0), but did not change significantly for boys (Δχ2 1=2.9). These subtle effects of sex and age are shown in a cross-sectional format in Fig. 2, with the sample divided into four age categories (based on the mean age, and mean score, for individuals with multiple measures).
Fig. 2. Summed scores for Somatic and Psychological Health Report (SPHERE-34), PSYCH-14 and SOMA-10, meaned separately for sex and for four age categories.
Table 1. Sample demographics: mean, s.d., range, sex and age effects and twin correlations (with 95% CI) for SPHERE-34, PSYCH-14, SOMA-10 and neuroticism
CI, Confidence intervals; SPHERE, Somatic and Psychological Health Report; MZ, monozygotic; DZ, dizygotic; F, female; M, male; OS, opposite sex.
a Means for neuroticism are based on the combined NEO/Junior Eysenck Personality Questionnaire (JEPQ) measure. NEO raw scores=22.4±7.2 (2, 45), JEPQ raw scores=9.4±4.5 (0, 20).
b β weights for sex and age effects are based on standardized covariates and dependent variables. Effects for neuroticism are for a combined NEO/JEPQ measure, which is described in the method.
c For all variables, MZM=MZF and DZM=DZF (Δχ2 2 ranged 0.4–2 .3), suggesting no magnitude differences in genetic influence based on sex. However, sex limitation modelling suggests that, for PSYCH-14, genetic estimates are larger in boys than girls (Δχ2 1=4.8, p=0.03). Also, different genetic and/or environmental sources influencing boys and girls are indicated for SPHERE-34 (i.e. DZO <DZ same-sex, Δχ2 1=4.0, p=0.04) and PSYCH-14 (Δχ2 1=4.0, p=0.05) and are suggestive for SOMA-10 (Δχ2 1=3.0, p=0.08), but not neuroticism (Δχ2 1=0.2, p=0.67). Sex limitation modelling further assessed genetic sources, but did not find significant differences between boys and girls (Δχ2 1 ranges 0.0–2 .0, p ranges 0.16–0 .99).
d Neuroticism was available for 84% of the SPHERE sample, for which pair numbers are listed here.
* p<0.05, ** p<0.01, *** p<0.00001.
Neuroticism was strongly correlated with PSYCH-14 (Table 2) and to a lesser, but still substantial extent, with SOMA-10 (r=0.58 with SPHERE-34). While there was no overlap in items for the SPHERE subscales, they nevertheless showed a strong phenotypic correlation.
Table 2. Model fitting results (best-fitting model shown in bold), plus the additive genetic (A) and unshared environmental (E) estimates (shown as percentages of total variance with 95% CI) and phenotypic, genetic and unshared environmental correlations derived from trivariate AE Cholesky analyses of neuroticism, PSYCH-14 and SOMA-10
CI, Confidence intervals; AIC, Akaike's Information Criterion.
a In a Cholesky decomposition, each of the trivariate variance/covariance matrices is decomposed into the product of a lower triangular matrix and its transpose. This decomposition generates a first factor that influences all variables, a second factor independent of the first that influences the second and third variables and a third factor independent of the first and second that influences only the third variable. Independent pathway models allow for one or more genetic or environmental common factors to be specified, with any remaining variance showing as specific influences. Common pathway models specify genetic and environmental influence on a latent variable that loads onto each phenotype, with remaining variance showing as specific influences.
b Total A [heritability (h2)] for Somatic and Psychological Health Report-34, determined from univariate analyses,=49%.
c Additive genetic factor A1 for neuroticism (52%) includes specific genetic variance (As1=23%). Remaining variance for neuroticism (Ac1=29%) represents genetic influences common to all measures. Similarly, the unshared environmental factor E1 for neuroticism (48%) includes specific environmental variance (Es1=33%) and variance due to environmental factors that are common to all measures (Ec1=15%). In the same way, additive genetic factor A2 (18%) includes specific genetic variance (As2=7%) and genetic variance in common with SOMA-10 (Ac2=11%). A similar breakdown can be shown for the unshared environmental factor E2 (41%; Es2=36, Ec2=5). Variance specific to neuroticism and PSYCH-14 was identified by changing the order of the variables and running the model with either neuroticism or PSYCH-14 specified last. This does not change model fit. A3 and E3 in the model above represent variance specific to SOMA-10.
Genetic modelling
The MZ and DZ twin correlations (Table 1) indicate additive genetic (A), common environmental (C) and unique environmental (E) influences on both SPHERE-34 and PSYCH-14 (i.e. the DZr>0.5×MZr) (Martin et al. Reference Martin, Jardine, Andrews and Heath1988), whereas for SOMA-10 and neuroticism, A and E influences are indicated (i.e. the DZr≈0.5×MZr). There was some suggestion of magnitude differences in the heritability for boys and girls for PSYCH-14 (i.e. reduced A and increased C suggested for girls compared to boys, as indicated by DZ correlations being similar to MZs for girls, but less than MZs for boys) and preliminary univariate sex limitation modelling showed that PSYCH-14 was significantly less heritable for girls compared to boys (p=0.03). In addition, there was also some indication that different genetic or environmental factors may influence individual differences in boys and girls for both SPHERE-34 and PSYCH-14 (i.e. opposite-sex correlations significantly lower than same-sex DZ correlations), but this finding was not supported by sex limitation modelling, which showed that genetic sources of influence did not differ significantly for boys and girls.
Covariation between neuroticism, PSYCH-14, and SOMA-10, was initially examined in a sex-limited ACE Cholesky model. While the genetic correlations suggested neuroticism was more strongly related to anxiety/depression than somatic distress in girls [0.89 (95% confidence intervals (CI) 0.32–1.0) v. 0.51 (−0.01 to 1.0)] than boys [0.76 (0.17–1.0) v. 0.67 (0.002–1.0)], the CIs were wide and both genetic and environmental pathways could be set equal for boys and girls (Δχ2 1 ranged 0.0–3.0). Thus, we collapsed over sex to examine the significance of A and C contributions in a Cholesky model (Table 2). Small C components, which accounted for 3% of total variance for neuroticism, 11% for PSYCH-14 and 7% for SOMA-10 could be dropped from the model without worsening fit (Δχ2 6=2.3, p=0.89), while A components were essential to maintain fit (Δχ2 6=41.5, p=2.3×10−7). In examining models containing only A and E components, both the Cholesky and independent pathway models containing either one or two common A and E factors provided a good fit to the data (Table 2) with the AE Cholesky having the lowest AIC and thus designated as the best-fitting model. We then estimated variance specific to neuroticism and PSYCH-14 by changing the order of the variables and re-running the model with each of these variables specified last. Note that this does not change model fit.
Table 2 shows estimates for the best fitting model with 95% CI (Fig. 3 shows the pathway model). A genetic factor (A1) accounts for all of the genetic variance in neuroticism, which includes 23% specific variance for neuroticism and 29% that is in common with the variance for PSYCH-14 and SOMA-10, plus 22% of the variance in PSYCH-14 and 16% in SOMA-10. A second genetic factor (A2), independent of A1, accounts for a further common source of genetic variance between PSYCH-14 (i.e. 11% of the variance: 18% minus 7% specific genetic variance) and SOMA-10 (17%). The genetic factor (A3) accounts for specific genetic variance for SOMA-10 (10%). These common sources of genetic influence are reflected in the genetic correlations (as the smaller environmental overlap is reflected in the unshared environmental correlations; see Table 2).
Fig. 3. Parameter estimates for the trivariate AE Cholesky model showing covariation between neuroticism, PSYCH-14 and SOMA-10. The model includes additive genetic (A1, A2, A3) and unshared environmental (E1, E2, E3) sources. Estimates are standardized such that when squared they indicate the percentage of variance accounted for. The factors A1 and E1 account for all of the variance for neuroticism [i.e. they include specific genetic (23%) and environmental (33%) variance for neuroticism], while the factors A2 and E2 are independent of neuroticism [Note: A2 and E2 include specific genetic (7%) and environmental (36%) variance for PSYCH-14]. Heritability (h2) is shown for each variable.
In contrast to the strong genetic association among the three measures, common environmental effects (i.e. the common influence subsumed in factors E1 and E2) are generally much less, with specific environmental effects accounting for a substantial amount of the variance (i.e. 33% for neuroticism, 36% for PSYCH-14, 47% for SOMA-10). As can be seen in Table 2, E1, after adjusting for the specific environmental variation for neuroticism, accounted for only 15% of the variance for neuroticism and 4% for SOMA-10, but 19% for PSYCH-14 (i.e. variance accounted by A1 and E1 are approximately the same for PSYCH-14). A second independent factor (E2) accounts for further common environmental influences for PSYCH-14 (5%) and SOMA-10 (6%), which again is approximately half that accounted for by A2.
Discussion
In the current study we sought to identify the role of neuroticism, a known risk factor for common mental disorders (Kotov et al. Reference Kotov, Gamez, Schmidt and Watson2010), as the major shared risk that may best explain the co-morbidity between anxiety/depression and somatic distress. We targeted adolescents and young adults as they pass through this key stage of onset of both of these syndrome sets (Merikangas et al. Reference Merikangas, He, Burstein, Swanson, Avenevoli, Cui, Benjet, Georgiades and Swendsen2010). All measures from the SPHERE questionnaire were moderately heritable (40–49%), consistent with similar measures (Lamb et al. Reference Lamb, Middeldorp, van Beijsterveldt, Bartels, van der Aa, Polderman and Boomsma2010; Bartels et al. Reference Bartels, van de Aa, van Beijsterveldt, Middeldorp and Boomsma2011; Vassend et al. Reference Vassend, Roysamb and Nielsen2011) and similar to neuroticism (52%). Genetic sources accounted for most of the covariation between neuroticism, anxiety/depression and somatic symptoms. However, genetic overlap between anxiety/depression and somatic symptoms was not solely due to their relationship with neuroticism. This represents the first genetic study to examine the role of neuroticism in the covariation between anxiety/depressive and somatic symptoms in adolescence.
A strongly influential common factor was identified, accounting for just over half of the genetic variance in anxiety/depression and neuroticism and approximately one-third of the genetic variance in somatic distress. This factor may reflect a susceptibility to psychological distress (i.e. an increased likelihood of responding to situations with fear, sadness, embarrassment, anger, guilt and disgust), which is core to the neuroticism domain, and neuroticism-related characteristics, such as a proneness to irrational ideas, poor impulse control and poor stress management (McCrae & Costa, Reference McCrae and Costa2010). Further, neuroticism is related to a self- or body-focused disposition (Pennebaker & Watson, Reference Pennebaker, Watson, Kirmayer and Robbins1991), which is known to correlate with level of somatic symptom reporting (Robbins & Kirmayer, Reference Robbins, Kirmayer, Kirmayer and Robbins1991) and to be an associated feature of mood and anxiety disorders (APA, 2000). Cognitive processes may also contribute as difficulty in discriminating emotional feelings and bodily sensations or in expressing emotions have been related to neuroticism and depressed mood (Parker et al. Reference Parker, Taylor and Bagby1989) and may also influence self-report of somatic symptoms (Kirmayer et al. Reference Kirmayer, Robbins and Paris1994).
At a neurobiological level, there are likely to be a variety of paths that are relevant to the expression of both depression and somatic symptoms. For example, serotonergic activity appears to influence both depression and somatization, although with differing relevance to each (Rief et al. Reference Rief, Pilger, Ihle, Verlerk, Scharpe and Maes2004) and has also been linked to neuroticism (Frokjaer et al. Reference Frokjaer, Vinberg, Erritzoe, Baare, Holst, Mortensen, Arfan, Madsen, Jernigan, Kessing and Knudsen2010). Disturbed circadian function (Hickie & Rogers, Reference Hickie and Rogers2011) appears to be a major risk factor to onset of both mood disorders and related somatic syndromes (particularly prolonged fatigue) and treatments targeting melatonin secretion may well provide novel treatment strategies. We have investigated extensively the relationships between exposure to infective agents and onset of both somatic and affective syndromes (Hickie et al. Reference Hickie, Davenport, Wakefield, Vollmer-Conna, Cameron, Vernon, Reeves and Lloyd2006). Disturbed immune function may also be a factor and one that we have previously investigated in adult twin samples (Hickie et al. Reference Hickie, Bennett, Lloyd, Heath and Martin1999a). For example, cytokine activity can influence pain perception and induce symptoms such as fatigue, depressed mood and altered cognition (Vollmer-Conna et al. Reference Vollmer-Conna, Fazou, Cameron, Li, Brennan, Luck, Davenport, Wakefield, Hickie and Lloyd2004; Dimsdale & Dantzer, Reference Dimsdale and Dantzer2011), although concentrations of immune parameters can differ between patients with depression and those with somatization (Rief et al. Reference Rief, Hennings, Riemer and Euteneuer2010). Further, levels of pro-inflammatory cytokines have been associated with neuroticism (Sutin et al. Reference Sutin, Terracciano, Deiana, Naitza, Ferrucci, Uda, Schlessinger and Costa2010).
We also identified a second common genetic factor, independent of neuroticism and specific to anxiety/depression and somatic distress. It is unclear what this genetic factor may represent, but given the complex psychological and biological processes involved as we discuss above, it is plausible that there may be some genetic effects in common with anxiety/depression and somatic distress that are independent of neuroticism. Interestingly, while this second genetic factor has a strong influence on somatic distress, accounting for more than one-third of the genetic variance, it has less influence on anxiety/depression, where it accounts for approximately one-quarter of the genetic variance. As the complex physiology underlying somatic symptoms is increasingly recognized as immune related (Dimsdale & Dantzer, Reference Dimsdale and Dantzer2011), this factor may reflect cytokine activity specific to these traits.
Our results extend those found in adult samples and are consistent with reported relationships between neuroticism and generalized anxiety disorder (Hettema et al. Reference Hettema, Neale, Myers, Prescott and Kendler2006; Kendler et al. Reference Kendler, Gardner, Gatz and Pederson2007), major depression (Hettema et al. Reference Hettema, Neale, Myers, Prescott and Kendler2006; Kendler et al. Reference Kendler, Gardner, Gatz and Pederson2007) and somatic health (Vassend et al. Reference Vassend, Roysamb and Nielsen2011). The identification of a second genetic factor is in contrast with Boomsma et al. (Reference Boomsma, Beem, van den Berg, Dolan, Koopmans, Vink, de Geus and Slagboom2000), where a single common factor was found to account for all genetic covariation between measures of depression, anxiety, somatic anxiety and neuroticism. Notably, we found an independent pathway model allowing only a single common genetic source to have only a slightly worse fit than our best-fitting Cholsky model; thus, both models are worthy of consideration. However, in contrast to the independent pathway model, all elements of the Cholesky model were significant, adding to confidence in this model, which had the best AIC fit.
In contrast to the genetic influences, unshared environmental influences were largely specific to each variable, (r g ranges 0.61–0.87 while r e ranges 0.25–0.56). These reflect environmental risk factors unique to the individual and to the trait and at least a moderate degree of trait-specific measurement error [based on reliability reports for similar symptom scales (e.g. Vallejo et al. Reference Vallejo, Jordan, Diaz, Comeche and Ortega2007) and our own estimates of stability over 2–6 months]. Even so, the finding of small overlapping unshared environmental factors suggests that some environmental risk factors are relevant to all measures or are independent of neuroticism and common to anxiety/depression and somatic distress. For example, exposure to stressors can promote pro-inflammatory processes (Raison et al. Reference Raison, Capuron and Miller2006), which could potentially influence multiple related traits. However, common unshared environmental factors may also include correlated measurement error, including state effects.
In addition to the finding of a major role of genetic factors on the relationship between neuroticism and anxiety/depressive and somatic symptoms in adolescence, the subtle differences for girls and boys across this period of adolescence are worth noting. In girls, we found that, for each of the SPHERE measures, the number of symptoms increased slightly over time (from 12 to 25 years), whereas in boys the symptom scores either decreased or showed no significant change. Although not all studies are in agreement (Bartels et al. Reference Bartels, van de Aa, van Beijsterveldt, Middeldorp and Boomsma2011), similar patterns have been reported previously for depression in adolescents (Sund et al. Reference Sund, Larsson and Wichstrom2001; Angold et al. Reference Angold, Erkanli, Silberg, Eaves and Costello2002; Lamb et al. Reference Lamb, Middeldorp, van Beijsterveldt, Bartels, van der Aa, Polderman and Boomsma2010). For neuroticism, while girls had higher scores than boys, as has been found previously (McCrae et al. Reference McCrae, Costa, Terracciano, Parker, Mills, de Fruyt and Mervielde2002), we detected no change in scores across adolescence in either girls or boys. This is consistent with what has been shown in adulthood (Terracciano et al. Reference Terracciano, Costa and McCrae2006), while varying age effects have been reported for adolescents. McCrae et al. (Reference McCrae, Costa, Terracciano, Parker, Mills, de Fruyt and Mervielde2002) found no age affects in American high school students, but found small increases for girls with age in a replication sample of Flemish adolescents and, interestingly, in a sample of gifted students, perhaps suggesting a cognitive component to the increases found. It is possible that increases over age in girls compared to boys, as found for our anxiety/depressive and somatic symptoms, may also reflect differing cognitive styles.
Limitations
A limitation of this study, despite the large sample (n=2459), was the lack of power to explore sex differences underlying covariation between the traits. Although not significant, our results suggested neuroticism had a stronger genetic relationship with anxiety/depression than somatic distress and this was more prominent in girls than boys. Fanous et al. (Reference Fanous, Gardner, Prescott, Cancro and Kendler2002) hypothesized that neuroticism and depression may be more genetically correlated in females, but, in an adult sample, they found correlations to be higher, although not significantly, in males. At a univariate level, we did find significantly higher heritability in boys compared to girls for anxiety/depression (PSYCH-14). A recent review of childhood and adolescent anxiety and depression reports small to negligible sex differences in genetic aetiology (Franic et al. Reference Franic, Middeldorp, Dolan, Ligthart and Boomsma2010), but, interestingly, a higher heritability for boys has been found for self-rated depression in children and adolescents, although no difference was found for parental ratings of depression in the same individuals (Rice et al. Reference Rice, Harold and Thapar2002). Most recently, a higher heritability for adolescent girls for self-reported anxiety/depression and somatic complaints was found (Bartels et al. Reference Bartels, van de Aa, van Beijsterveldt, Middeldorp and Boomsma2011), a finding also reported for teacher ratings of depressive symptoms in young adolescents (Happonen et al. Reference Happonen, Pulkkinen, Kaprio, van der Meere, Viken and Rose2002) and for mother-rated separation anxiety in children and adolescents (Feigon et al. Reference Feigon, Waldman, Levy and Hay2001). Clearly, further research using large samples is required to clarify the role of sex in the genetic aetiology of childhood and adolescent anxiety and depression and the role of factors such as phenotype definition and participant age.
The use of dimensional (continuous) rather than categorical measures of anxiety/depression and somatic distress is also a potential limitation. Categorical classification is optimal when no meaningful clinical variation exists among those diagnosed positive or among those negative (Kraemer, Reference Kraemer2007). However, the range in symptom count found for anxiety/depression (0–14) and somatic distress (0–10) suggests that the full range of variability would not be captured in a select number of categories. Supporting the use of a dimensional measure, at least for depression, are studies showing a linear relationship between symptom count and impairment or disability (Ustun and Sartorius, Reference Ustun and Sartorius1995; Sakashita et al. Reference Sakashita, Slade and Andrews2007) and others indicating depression is best conceptualized as one latent continuous dimension (Ruscio & Ruscio, Reference Ruscio and Ruscio2000; Slade & Andrews, Reference Slade and Andrews2005).
A further limitation may be the conceptualization of anxiety/depression and somatic distress as distinct syndromes, given the relative lack of empirical data to support a clear separation. Somatic syndromes, together with anxiety and depression, may be considered part of a broader spectrum of internalizing disorders (Krueger et al. Reference Krueger, Chentsova-Dutton, Markon, Goldberg and Ormel2003). Nevertheless, factor analytic studies show a clear separation (Gillespie et al. Reference Gillespie, Kirk, Heath, Martin and Hickie1999; Kirk et al. Reference Kirk, Hickie and Martin1999), consistent with studies showing, for example, that not all patients with somatic disorders meet criteria for other psychological disorders (Hickie et al. Reference Hickie, Lloyd, Wakefield and Parker1990) and that patients with fatigue do not show specific response to antidepressant pharmacotherapy (Vercoulen et al. Reference Vercoulen, Swanink, Zitman, Vreden, Hoofs, Fennis, Galama, van der Meer and Bleijenberg1996).
We must also acknowledge that we are not yet sufficiently powered to detect common environmental influences. Based on our current sample size and twin correlations, we have >75% power to detect additive genetic influences in a univariate model, but only 5–37% power to identify common environmental influences (i.e. 37% for SPHERE-34, 25% for PSYCH-14, 5% for SOMA-10). Therefore, our genetic estimates from the AE model may be slightly inflated. However, negligible C estimates are consistent with the literature for self-rated depression and somatic complaints (Happonen et al. Reference Happonen, Pulkkinen, Kaprio, van der Meere, Viken and Rose2002; Rice et al. Reference Rice, Harold and Thapar2002; Bartels et al. Reference Bartels, van de Aa, van Beijsterveldt, Middeldorp and Boomsma2011).
Although our study design precluded examination of longitudinal relationships, data collection is ongoing and future analyses will address age-related changes in genetic and environmental influence. Here, we show that, in adolescents, the genetic risk factors indexed by neuroticism do not fully account for the genetic overlap found between measures of psychological and somatic health.
Acknowledgements
This work was supported by grants from the Australian Research Council (ARC), the National Health & Medical Research Council (NHMRC) and Beyond Blue, Australia.
Declaration of Interest
None.
