| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
ORIGINAL ARTICLES |
From the Centers for Behavioral and Preventive Medicine (J.M.M., R.N., J.F.T.), Brown Medical School and The Miriam Hospital, Providence, Rhode Island; and SRI International (G.E.S., D.C.), Menlo Park, California.
Address reprint requests to: Jeanne McCaffery, PhD, Centers for Preventive and Behavioral Medicine, Coro Building, Suite 5000, One Hoppin Street, Providence, RI 02903. Email: Jeanne_McCaffery{at}brown.edu
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
METHOD: Twin structural equation modeling was employed to estimate genetic and environmental contributions to the covariation of depressive symptoms, as indexed by the Centers for Epidemiological Studies–Depression Scale, and common variance among blood pressure, body mass index, waist-to-hip ratio, and serum triglycerides and glucose among 87 monozygotic and 86 dizygotic male twin pairs who participated in the NHLBI twin study.
RESULTS: Depressive symptoms were associated with individual components of the metabolic syndrome and common variance among the risk factors. Twin structural equation modeling indicated that the associations were attributable to environmental (nongenetic) factors.
CONCLUSIONS: These results support the hypothesis that depressive symptoms may increase risk for a pattern of physiological risk consistent with the metabolic syndrome.
Key Words: depression, • metabolic syndrome, • twins, • cardiovascular disease, • diabetes, • cholesterol.
Abbreviations: a2 = additive genetic variance;; BMI = body mass index;; c2 = shared environmental variance;; CES-D = Centers for Epidemiological Studies–Depression Scale;; d2 = dominant genetic variance;; DBP = diastolic blood pressure;; DZ = dizygotic;; e2 = nonshared environmental variance;; GLUC = glucose;; HDL = high-density lipoprotein cholesterol;; MZ = monozygotic;; NHLBI = National Heart, Lung, and Blood Institute;; SBP = systolic blood pressure;; MAP = mean arterial pressure;; TRIG = triglycerides;; WHR = waist-to-hip ratio.
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
One possible mechanism through which depression may influence subsequent cardiac events is an accumulation of cardiovascular risk factors associated with the metabolic syndrome. The metabolic syndrome has been defined as a clustering of elevated glucose (GLUC), blood pressure (SBP, DBP), triglycerides (TRIG), waist-to-hip ratio (WHR) and body mass index (BMI), and decreased high-density lipoprotein (HDL) cholesterol and is thought to develop from impaired insulin-mediated glucose uptake, primarily in skeletal muscles (12). Previous studies of depression and characteristics of the metabolic syndrome suggest that clinically depressed patients exhibit elevated insulin and glucose responses to glucose tolerance tests (13–15) and lower fasting HDL cholesterol concentrations (16), compared with nondepressed controls. Indeed, one study suggested that the elevated glucose and insulin responses to oral glucose challenge observed among clinically depressed patients may be improved after treatment of depression with tricyclic antidepressant medication (17). Depressive symptoms, as indexed by questionnaire, also seem to correlate with insulin responses to an oral glucose challenge (18).
Twin studies have characterized genetic and environmental contributions to the covariation among risk factors associated with the metabolic syndrome. For example, SBP, BMI, TRIG, HDL, and insulin resistance shared common genetic variance in a sample of middle-aged twins (19). In addition, the covariance of hypertension, obesity, and diabetes seemed attributable to common genetic and nonshared environmental effects in a second sample of middle-aged twins (20). Lastly, genetic and, to a lesser extent, environmental variance also contributed to covariation among SBP, DBP, BMI, TRIG, and total serum cholesterol in young adult twins (21).
Genetic and environmental contributions to depressive symptoms differ by whether clinical diagnosis (eg, DSM-III-R) or questionnaire (eg, Centers for Epidemiological Studies–Depression Scale (CES-D)) measures are used. Clinical depression tends to be moderately heritable with an additional contribution of individual (unique) environmental factors (22). However, individual (unique) environmental variance tends to account for the majority of individual differences in self-report measures of depressive symptoms, such as the CES-D (23–25). The goals of this investigation were: 1) to further explore the nature of the association between depressive symptoms and the metabolic syndrome by examining the covariation of depressive symptoms with common variance among physiologic risk factors for cardiovascular disease; and 2) to estimate genetic and environmental contributions to any observed associations among adult male twins (ages 59–69) enrolled in the National Heart, Lung, and Blood Institute (NHLBI) twin study.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
This analysis is based on data collected from four study centers during the third examination (1986–1987) (27). Of the 514 twin pairs who participated at Time 1, at least one twin of 94 pairs was known to have died and at least one twin from 152 pairs was lost to follow-up, leaving a total of 268 complete twin pairs who participated at Time 3. Of the 268 pairs, 54 were excluded due to self-report of diabetes by at least one twin and 41 pairs were excluded due to missing lipid or glucose data for at least one twin. Thus, 173 twin pairs, 87 monozygotic (MZ) and 86 dizygotic (DZ), with complete metabolic and anthropometric data and who did not report diabetes were included in the present study. For analyses incorporating the CES-D, the sample with complete data included 76 MZ and 73 DZ twin pairs due to missing CES-D data for at least one twin from 24 pairs.
As compared with the full sample at Time 1, the 173 twin pairs eligible at Time 3 were on average younger (46.7 vs. 47.9 years old, p < .01) and had lower SBP (124.8 vs. 127.1, p < .01) and DBP (79.9 vs. 81.6, p < .05). These twin pairs were comparable to the full sample in BMI (25.4 vs. 25.4, p = .87) and years of education (13.2 vs. 13.1, p = .54).
Measures
Measures were taken on one occasion and twins were encouraged to participate on the same day. Height and weight were used to calculate BMI (weight/height in kg/cm2), and WHR was defined as the ratio of waist circumference at the level of the umbilicus to the hip circumference at the maximum width. Blood pressure was measured by mercury sphygmomanometer on the right arm and was averaged across two seated measurements taken at a 1-minute interval. Blood draws were taken after overnight fast and consumption of a 50-g load of glucose. Thus, glucose concentrations followed a glucose tolerance challenge but the lipids were not fasting concentrations. Lastly, participants also completed the CES-D, a 20-item scale assessing depressive symptoms (28).
Data Analysis
The associations between individual metabolic risk factors and depressive symptoms were first characterized by correlation. Next, to determine whether common variance among metabolic risk factors could be reduced to component factors, phenotypic factor analysis was employed. Lastly, twin structural equation modeling was used to partition the observed total variation and covariation between MZ and DZ twins in terms of latent causes due to genetic and environmental effects for the derived metabolic factors and depressive symptoms. This method allows the characterization of genetic and environmental contributions to each of the risk factors individually and the estimation of genetic and environmental contributions to any covariation between depression and common variance among the metabolic risk factors (29).
In twin modeling, it is assumed that rearing environment for the behaviors under study was similar for MZ and DZ twin pairs. Empirical tests of the equal environment assumption by zygosity suggest that it holds for many traits (30). Given this equal environments assumption, it is possible to estimate genetic variance from the difference in the degree of similarity between MZ and DZ twins. If the degree of similarity between MZ and DZ twins is proportional to their degree of genetic relatedness – MZ twins are more similar genetically than DZ twins by a factor of 2 – genetic variance is modeled as additive. Additive genetic variance has been conceptualized as the combined effect of several genetic loci that are roughly equipotent with regard to the phenotype of interest. This additive genetic variance is also reflected in traditional heritability estimates (a2 or h2). To the extent that the degree of similarity of MZ twins exceeds that of DZ twins by a factor greater than 2, results are suggestive of genetic dominance (d2). Dominance refers to an interaction of alleles at one locus and can be accommodated using twin structural equations.
Although one of the strengths of twin modeling is the ability to estimate genetic effects, the twin design is also one of the best methods to isolate environmental variance in a phenotype within a population. For example, to the extent that similarities among DZ twins are comparable to those observed for MZ twins, environments shared by the twins (c2; shared environment) are likely to play a role. In addition, as MZ twins generally share all of their DNA, any differences between MZ twins must be nongenetic, or environmental, in origin. This environmental variance, which includes error of measurement, is labeled nonshared environment (e2) because it does not contribute to similarities among MZ and DZ twins. Indeed, it contributes to similar differences in both. It should be noted that shared environment cannot be concurrently modeled with a dominance term, as these effects are correlated among MZ and DZ twins reared together; thus, these terms will not be included in the same model. In the models presented in this paper, it is also assumed that assortative mating, genotype-environment covariance, genotype x environment interaction, dominance, and epistasis do not substantially influence the variables under study (29).
All twin structural equation models were based on standardized variance/covariance matrices and estimated using the Mx program (31). This package provides maximum likelihood estimates of model parameters and a 
2 goodness-of-fit test to assess overall model fit. The significance of individual parameters was determined by comparing the fit of models omitting the parameters of interest with the fit of a full model.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
|
Two measures of degree of contact differed between MZ and DZ twins: frequency of contact and degree of relationship closeness (t(464) > 5.40, p value < .01). To address whether this difference between twin types affected the phenotypes of interest, linear regression was used to determine whether frequency of contact or relationship closeness was associated with absolute within-pair differences in depressive symptoms, the metabolic risk factors, or the factor derived from the metabolic risk factors, independent of zygosity. Within-pair differences in degree of closeness were not associated with any of the phenotypes (ß values < 0.15, p values > .08). Within-pair differences in the degree of contact were also not associated with the phenotypes (ß values < 0.17, p values > .10) with the exception of TRIG and WHR (ß values = 0.18, 0.20; p values = .03, .01; respectively). This suggests that genetic variance in these measures may be somewhat inflated due to environmental differences between MZ and DZ twins.
Phenotypic Correlations and Factor Analysis
Phenotypic correlations derived from twin structural equation modeling are listed in Table 2. Overall, small-to-moderate correlations were observed between the metabolic variables, with the exceptions of SBP, DBP, and MAP (which were highly intercorrelated), and GLUC, which did not correlate significantly with DBP, MAP, and WHR. CES-D scores also showed small but significant correlations with MAP, BMI, WHR, and TRIG and tended to correlate with SBP and DBP (p values < .10). Due to the strong association between the blood pressure measures (r = 0.54, p < .01), only MAP [(SBP + (DBP*2))/3] was included in further analyses.
|
Univariate Twin Modeling
Twin-pair correlations and univariate twin structural equation modeling parameters are presented in Table 3. For each of the physiologic variables and the metabolic factor derived from MAP, BMI, WHR, TRIG, and GLUC, the correlations among MZ twins were moderate to strong in magnitude and exceeded the correlation among DZ twins, suggesting genetic effects. For SBP and WHR, the correlations among MZ twins were less than twice the correlations among DZ twins, suggesting a small contribution of environments shared by twins. Lastly, each of the correlations among MZ twins was substantially less than perfect (or 1.0), indicating that environmental factors that make twins differ (nonshared environment; including error of measurement) also influence the physiologic variables and underlying metabolic factor. For CES-D scores, MZ correlations were small in magnitude, although they were approximately twice the correlation of DZ twins. These results suggest that variability in CES-D scores was predominantly influenced by nonshared environment with, perhaps, a small genetic contribution.
|
For DBP, BMI, GLUC, TRIG, the metabolic factor, and the CES-D, the correlation among MZ twins exceeded the correlation among DZ twins, suggesting the possibility of genetic dominance or epistasis. To test this possibility, a model including a dominance term (a2+d2+e2) was compared with a model including only additive genetic and nonshared environmental effects (a2+e2). Only in the case of GLUC did the dominance model provide superior fit to that of the model including additive genetic and nonshared environmental effects alone (a2+d2+e2 vs. a2+e2: 
2(1) = 4.45; p < .05).
Overall, model fits for the univariate modeling tended to be good, reflecting models that closely represent the observed data. Only the models for GLUC and WHR did not provide adequate fit to the data, likely due to small variance differences between twins labeled 1 and 2.
Bivariate Twin Modeling
Bivariate analysis of genetic and environmental contributions to the covariation of CES-D scores and the metabolic factor are presented in Figure 1. The phenotypic correlation between depressive symptoms and the metabolic factor was small in magnitude (r = 0.16, p < .01), accounting for 4% of the variance in each of the variables. Nonshared environmental effects accounted for the majority of common variance between CES-D scores and the metabolic factor, with a smaller but nonsignificant contribution of genetic effects. Indeed, a model in which genetic and shared environmental effects on covariation were fixed to 0 fit as well as the full model (2(2) = 0.40, NS), suggesting that genetic and shared environmental effects do not contribute to the association between depressive symptoms and the metabolic factor in this sample. Similar to the univariate results, variance in the CES-D that is independent of the association with the metabolic factor is primarily nonshared environmental in origin, whereas variance in the metabolic factor not common with CES-D is attributable to genetic and nonshared environmental effects.
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
Depressive Symptoms and Metabolic Risk
In phenotypic analyses, depressive symptoms, as measured by the CES-D, showed small but significant correlations with MAP, BMI, WHR, TRIG, and GLUC and also tended to correlate with SBP and DBP. Previous studies have generally shown an association between clinically diagnosed depression and metabolic risk, as indexed by elevated glucose and insulin responses to a glucose tolerance test (13–15). For example, in one study that administered a 5-hour glucose tolerance test to depressed patients and healthy controls (13), depressed patients demonstrated significantly higher basal glucose levels and greater cumulative glucose and insulin responses after the glucose tolerance test than control subjects. Taken together with this type of result, the present study provides further evidence of an association between depressive symptoms and the metabolic syndrome using a factor analytic approach to metabolic risk.
Our study also suggests that the association between depressive symptoms and metabolic risk extends to depressive symptoms in a nonclinical range. In this study, the mean CES-D score was 6 and the score at the 75th percentile of this sample was 8, indicating that the majority of participants scored in the low range of depressive symptoms. It has been argued that the CES-D may not be a pure measure of depression and that it may reflect general distress and negative affect (33). In addition, a cutoff of CES-D 
16 is commonly considered suggestive of a major depressive episode (4). Thus, it is likely that most of the variability in CES-D scores in this sample likely reflects distress/negative affect and not symptoms that would meet criteria for a major depressive episode.
It is conceivable that the etiology of clinical depression may differ from that of CES-D scores. For example, it is possible that liability to major depressive episodes reflects a biologic or genetic predisposition, whereas distress/negative affect is influenced more by daily environmental challenges, such as daily hassles or life events, particularly in the low range of symptomology observed in this sample. Such a hypothesis is consistent with genetic contributions to clinical depression (22) whereas more nonshared environmental variability has been associated with CES-D scores (23–25). It is possible that differences in the proportion of genetic and environmental contribution between clinical depression and CES-D scores may reflect measurement error or reliability differences across the two measures. However, to the extent that different etiologic effects influence CES-D scores and clinical depression, it would be expected that the causes of covariation between each of these variables and metabolic risk may also differ.
Factor Structure of Metabolic Risk Factors
Factor analysis of common variance across MAP, BMI, WHR, TRIG, and GLUC produced a single underlying factor that correlated positively with depressive symptoms. BMI and WHR loaded most highly on the factor (0.72 and 0.65, respectively) with small-to-moderate loadings for TRIG (0.58), MAP (0.40), and GLUC (0.20).
It is somewhat difficult to compare our factor analytic results with previous studies, as factor analysis is dependent on the interrelationships of variables included in each study. Previous investigations have included between 5 and 17 variables in factor analyses of metabolic risk (19, 21, 34–39). These studies generally derive between one and four latent factors, with studies including more variables tending to uncover more factors (38). Our factor analytic results are most consistent with investigations including between 5 or 6 variables, each deriving one latent factor (19, 21, 40). Measures across these studies included blood pressure, WHR, BMI, HDL, TRIG, GLUC, insulin, and GLUC and insulin responses to a glucose tolerance test.
Of note, although results for HDL cholesterol were not presented due to variance differences between MZ and DZ twins, phenotypic factor analysis including HDL with the other metabolic variables yielded one factor, consistent with the factor structure presented earlier. In contrast, if SBP and DBP were included in analyses in place of MAP, two factors resulted, likely due to the high correlation of SBP and DBP. Factor structures were consistent with a metabolic factor resembling the metabolic factor presented earlier and a second blood pressure factor with substantial loadings for SBP and DBP only. Interestingly, both factors showed comparable associations with CES-D scores and twin analysis suggested that associations between depressive symptoms and both the metabolic and blood pressure factors were accounted for by environmental, as compared with genetic, effects.
Twin Modeling
The twin design permitted another extension of previous studies: a partitioning of genetic and environmental contributions to covariance. Results from this type of analysis can guide future research to focus on specific genetic and/or environmental variables that contribute to common variance across the CES-D and metabolic risk. In this study, the metabolic risk factors and the latent factor underlying these variables were either significantly heritable or tended to be heritable. These results are consistent with previous studies (19, 21) and suggest that genetic factors influence individual differences in each of these variables and common variance across these variables. In contrast, CES-D scores were primarily environmental in origin, suggesting a different etiology for depressive symptoms in this study.
Twin analyses partition environmental variance into shared environment and nonshared environment. Shared environment is defined as nongenetic factors that increase similarity among twins and is typically thought to reflect early rearing experiences such as parenting. Nonshared environment is defined as environmental factors that reduce similarity among twins, including measurement error. In this study, nonshared environmental factors influenced CES-D scores and the association between CES-D scores and common variance among metabolic features, suggesting that environmental factors differing between co-twins (perhaps daily hassles or life events) account for this variation and covariation.
Limitations
Limitations on the interpretation of these results should be noted. First, this study was cross-sectional in design and the direction of effects cannot be determined. It is possible that depressive symptoms affect metabolic risk; however, it is also possible that metabolic risk influences depressive symptoms. For example, obesity, one of the components of metabolic risk that is mostly strongly associated with depressive symptoms in this sample, has been shown to predict depressive symptoms in cross-sectional and longitudinal studies (42, 43). In addition, a third, as yet undetermined, variable may also account for the association between depressive symptoms and metabolic risk. One potential variable of interest in this regard is physical activity, which has been associated both with metabolic risk and depressive symptoms (44). In future studies, we plan to examine the association between depressive symptoms and metabolic risk longitudinally to further understand the direction of causation.
Second, the degree of contact among twins predicted variability in TRIG and WHR. These results suggest that the heritability estimates for TRIG and WHR were somewhat inflated due to additional contact among MZ twins as compared with DZ twins. Interestingly, this result indicates that shared environment may also contribute to these metabolic risk factors, in addition to genetic and nonshared environmental effects.
Third, approximately 20% of the veterans reported treatment with antihypertensive medications at the time of assessment. Data on lipid-lowering medications and antidepressant medications were not collected. It is possible that medication use may have altered the structure of the metabolic factor, perhaps accounting in part for the smaller loading of MAP as compared with the other metabolic risk factors.
Fourth, the sample was limited to male veterans with a mean age of 63, thus, it is unclear whether these results would generalize to females and individuals of different ages. At the age of 63, men have already entered the high-risk period for CVD and thus may be expected to show a greater extent of metabolic risk factor clustering than younger individuals. It is also possible that these men experience some disability secondary to their level of risk, potentially increasing the association between metabolic risk factors and depressive symptoms in this sample. These issues are best addressed in a longitudinal study incorporating both men and women.
Lastly, this study did not measure clinical depression and the preponderance of scores are in the nonclinical range. Thus, it remains to be determined whether the association between depressive symptoms would be strengthened in a sample with a greater prevalence of clinical depression. To the extent that a stronger association is seen in samples with a greater prevalence of clinical depression, it would be interesting to examine whether variability in the metabolic syndrome accounts, in part, for the association between clinical depression and CVD.
Received for publication December 3, 2001.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
 |
|
 |
|
  J. A. Dunbar, P. Reddy, N. Davis-Lameloise, B. Philpot, T. Laatikainen, A. Kilkkinen, S. J. Bunker, J. D. Best, E. Vartiainen, S. Kai Lo, et al. Depression: An Important Comorbidity With Metabolic Syndrome in a General Population Diabetes Care, December 1, 2008; 31(12): 2368 - 2373. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 C. V. Igna, J. Julkunen, H. Vanhanen, P. Keskivaara, and M. Verkasalo Depressive Symptoms and Serum Lipid Fractions in Middle-Aged Men: Physiologic and Health Behavior Links Psychosom Med, November 1, 2008; 70(9): 960 - 966. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
V. Vaccarino, C. McClure, B. D. Johnson, D. S. Sheps, V. Bittner, T. Rutledge, L. J. Shaw, G. Sopko, M. B. Olson, D. S. Krantz, et al. Depression, the Metabolic Syndrome and Cardiovascular Risk Psychosom Med, January 1, 2008; 70(1): 40 - 48. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. H. Boyle, R. S. Surwit, A. Georgiades, B. H. Brummett, M. J. Helms, R. B. Williams, and J. C. Barefoot Depressive Symptoms, Race, and Glucose Concentrations: The role of cortisol as mediator Diabetes Care, October 1, 2007; 30(10): 2484 - 2488. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
N. Kupper, J. Denollet, E. J. C. de Geus, D. I. Boomsma, and G. Willemsen Heritability of Type-D Personality Psychosom Med, September 1, 2007; 69(7): 675 - 681. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. A. Bermudes, P. E. Keck Jr., and J. A. Welge The Prevalence of the Metabolic Syndrome in Psychiatric Inpatients With Primary Psychotic and Mood Disorders Psychosomatics, December 1, 2006; 47(6): 491 - 497. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. E. Elwing, P. J. Lustman, H. L. Wang, and R. E. Clouse Depression, Anxiety, and Nonalcoholic Steatohepatitis Psychosom Med, July 1, 2006; 68(4): 563 - 569. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. Herva, P. Rasanen, J. Miettunen, M. Timonen, K. Laksy, J. Veijola, J. Laitinen, A. Ruokonen, and M. Joukamaa Co-occurrence of metabolic syndrome with depression and anxiety in young adults: the Northern Finland 1966 Birth Cohort Study. Psychosom Med, March 1, 2006; 68(2): 213 - 216. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
