Behav Genet

DOI 10.1007/s10519-011-9506-x

ORIGINAL RESEARCH

Two Sources of Genetic Liability to Depression:

Interpreting the Relationship Between Stress Sensitivity
and Depression Under a Multifactorial Polygenic Model
Xinying Li Matt McGue Irving I. Gottesman

Received: 13 April 2011 / Accepted: 12 September 2011

Springer Science+Business Media, LLC 2011

Abstract Psychopathology theories, clinical observations, and research all point to multiple sources of liability
to depression. This article uses a longitudinal twin-study
design to characterize the contribution of two geneticallyinfluenced sources of depression risk: the first corresponding to stress sensitivity and the second representing
risk that is independent of stress sensitivity. The sample
consisted of 606 pairs of same-sex adolescent twins
recruited from Beijing, China. Mean (SD) age at intake
(Wave1) and follow-up (Wave2) was 13.2 (2.6) and 15.1
(2.6) years, respectively. A Reaction Level index was
developed to reflect individual differences in stress sensitivity. Biometric models were fit to examine the genetic
influence on the variance of and covariance between stress
sensitivity and depressive symptoms. Results showed that
both Reaction Level and depressive symptoms were moderately heritable. The genetic correlation between depressive symptoms and Reaction Level was estimated to be
.884. Genetic contributions to Reaction Level accounted
for 37.5% of the total variance of depressive symptoms.
Another set of genetic factors, which did not contribute to

Edited by Deborah Finkel.

X. Li
Key Laboratory of Mental Health, Institute of Psychology,
Chinese Academy of Sciences, Beijing, China
M. McGue
I. I. Gottesman
Department of Psychology, University of Minnesota,
Minneapolis, MN, USA
I. I. Gottesman (&)
Department of Psychiatry, University of Minnesota Medical
School, 2450 Riverside Ave, Minneapolis, MN 55454-1495,
USA
e-mail: gotte003@umn.edu

Reaction Level, accounted for 10.5% of the total variance

of depressive symptoms. We interpret our results within the
context of a multifactorial polygenic model, whereby
depression risk is due to the combined contribution of
multiple genetic and environmental factors.
Keywords China
Twins
Stress sensitivity

Endophenotype
Depressive symptoms
Adolescence

After Caspi et al. (2003) reported an interaction effect of

life stress and 5-HTTLPR genotype on risk of depression,
the importance of G 9 E interactions in the etiology of
depression has been widely accepted, albeit with a few
exceptions (Eaves 2006). The resulting prominence G 9 E
research has gained within the field raises a practical
question concerning future genetically-informed research
on depression: Must all future genetic research on
depression be based on a G 9 E model or is there more to
be learned about the genetic contributions to depression
risk? In the current article we make the case that there is
still much to be learned about the nature of the genetic
influence on depression using traditional twin-study
designs. Specifically, we seek to determine how depression
and stress sensitivity are genetically related.
The multifactorial polygenic model (Gottesman and
Shields 1967, 1982) provides a conceptual foundation for
our research. In this model, which was originally put forward
for schizophrenia, Gottesman and Shields (1967, 1982)
proposed three sources of liability: a general environmental
liability, a specific genetic one and a general genetic one.
The specific genetic liability conveys a specific risk for the
development of schizophrenia, whereas the general genetic
liability would be a common background contributor to
many psychiatric disorders. The two genetic sources of

123

Behav Genet

liability combine in some fashion to yield a combined

genetic liability. The latter would be offset by genetic and
environmental assets, which together with epigenetic
(Haque et al. 2009) and stochastic factors (Woolf 1997),
would yield a systems-based appreciation of total risk.
In extending this framework to depression, Fowles
(1992) claimed that reactivity to stressful life events
(SLEs) (i.e., stress sensitivity) represents a general genetic
liability for multiple psychiatric disorders, including
depression. Consistent with this notion, stress sensitivity
has been implicated in various disorders, such as affective
disorders and drug addiction (Becker et al. 2007; Tellegen
et al. 2003). Regarding the etiology of depression, similar
notions have been proposed by others (Kendler et al. 1995;
McGuffin et al. 1988). In the present study, we applied the
multifactorial polygenic model to depression, and explicitly proposed two major sources of genetic liability. The
first one, corresponding to the general genetic liability in
the multifactorial polygenic model, involves a genetically
influenced sensitivity to stress. The second one, corresponding to the specific genetic liability in the multifactorial polygenic model, has its effect regardless of stress
sensitivity and the presence/absence of SLEs.
Although stress sensitivity has been frequently discussed in previous studies due to its important role in the
etiology of psychopathology, it still lacks a clear operational definition. Each study has usually focused on one
aspect of reactivity to stressors, such as negative affect
reactivity (Wichers et al. 2009), hormonal reactivity (Bale
2006), or depressive reactivity (e.g., Caspi et al. 2003). In
the present study, we have followed Fowles (1992) and
assessed stress sensitivity as an individuals depressive
reaction to SLEs.
Clues for the two sources of genetic liability have long
been noticed in both clinical practice and scientific
research. The first line of evidence, coming from research
on recurrent depression, has suggested that stress sensitivity and the genetic loading of depression could function
in somewhat independent ways. For example, although the
association between SLEs and the risk of onset of a
depressive episode decreases as the number of recurrent
episodes increases (Burcusa and Iacono 2007), this phenomenon was weak to absent among individuals with high
genetic background (e.g., monozygotic twin, co-twin
affected) (Kendler et al. 2001). The second line of evidence
has shown that depression is genetically influenced even in
the absence of SLEs, implying the existence of a genetic
liability that is independent of SLEs. For example, in a
population-based twin study, Kendler et al. (1995) found
that in the absence of SLEs, the risk of onset of major
depression for individuals with a high genetic background
(monozygotic twin, co-twin affected) was significantly
higher than for individuals with a low genetic background

123

(monozygotic twin, co-twin unaffected). This result was

replicated in another study (Kendler et al. 2001).
Despite the supportive evidence, some studies have not
replicated the existence of two sources of genetic liability.
For example, McGuffin et al. (1988) investigated individuals from 83 families ascertained through clinically
depressed probands. As expected, they found a higher
prevalence of depression among the first-degree relatives
than among a community sample, suggesting familial/
genetic aggregation of depression. However, the relatives
of probands whose onset of depression followed exposure
to SLEs were no more sensitive to recent SLEs than the
relatives of probands whose onset of depression had no
obvious preceding SLE. Thus, they did not observe familial
aggregation for sensitivity to stress.
Taken together, although psychopathology theories and
some research results suggest the existence of two sources
of genetic liability to depression, empirical evidence is
inconsistent and does not lead to definitive conclusions. In
our opinion, two limitations of previous research should be
addressed. First, a methodology that allows for the separate
investigation of depression risk mediated through stress
reactivity and independent of stress reactivity is needed.
Second, a methodology that allows for the comparison of
individuals who experience different amounts of SLEs is
needed. Efforts were made in the present study to compensate for the two limitations by adopting modern statistical methods and introducing a novel variable, Reaction
Level, as an index of individual differences in stress sensitivity (see Methods for details).
We propose to test the following hypotheses: (1) Individual differences in stress sensitivity are heritable. (2) The
genetic contributors to stress sensitivity, which corresponds
to the general genetic liability to depression within the
multifactorial polygenic framework, serve as part of
genetic component of depressive symptoms. (3) There will
also be a specific genetic liability to depression, which is
independent of stress sensitivity.

Methods
Sample
This study is based on data from an ongoing longitudinal
study of adolescent behavioral and emotional disorders
conducted in Beijing, China. Participants were recruited
from elementary and high schools that were randomly
selected from all 18 counties or districts in the Beijing
municipality. For the present study, we focused on 606
pairs of same-sex twins who joined the study in 2007
(Wave1), and consented to participate in the follow-up in
2009 (Wave2). Among them, 331 pairs were female twins

Behav Genet

and 275 pairs were male twins. According to the results of

a DNA-based zygosity determination method (Chen et al.
2010) whose predictive accuracy is close to 100%, 454
pairs were monozygotic (MZ) twins (238 female, 216
male), and 152 pairs were dizygotic (DZ) twins (93 female,
59 male). Only seven pairs dropped out of the study at
Wave2, giving a follow-up participation rate of 99%. Mean
(SD) age at intake and follow-up were 13.2 (2.6) and 15.1
(2.6) years, respectively. To assess the representativeness
of this twin sample, the socioeconomic status of the twin
families was compared to that of 10,509 Beijing families
from a large population-based sample (Chen et al. 2009), in
which the same measurement of family income and
parental education level were used as in this twin sample.
Although more parents in the twin sample received college
education than parents in the population-based sample
(25.2% vs. 17.5% for fathers, 26.9% vs. 16.7% for mothers),
there was no significant difference between the twin families
and families from the population-based sample in terms of
family income. Thus, the representativeness of this twin
sample appears acceptable.
Consents were obtained from the twins themselves and
their parents. Arrangements were made for the twins to
participate in their classrooms, after school hours. After
describing the purposes of the study and explaining the
procedures, trained research staff distributed the questionnaires to the participants and instructed them to complete
the questionnaire independently. Research staff were there
to answer any questions that the students might have about
the questionnaire. After completion, twins were asked to
provide their saliva samples using the Oragene DNA selfcollection kit (Genotek Inc.). Questionnaires for parents
were taken home by the twins and mailed back to our
laboratory. 79.9% of the parents completed and returned
the questionnaires. There was no significant difference
between the twin families whose parents returned questionnaires and those who did not in terms of family income.
Confidentiality and anonymity of the participants responses to the questionnaires were ensured. Institutional
Review Board approved the research protocols.
Measures
Participants depressive symptoms were measured using the
Childrens Depression Inventory (CDI, Kovacs 1992),
which was developed for children and adolescents. The
reliability and validity of CDI have been well established
(Smucker et al. 1986; Timbremont et al. 2004). Its Chinese
version has been widely used in China. Participants rated
depressive symptoms that they had experienced over the
past 2 weeks on a 3-point scale. This measure consists of 27
items dealing with negative feelings, suicidal ideation,
biased cognition, or somatic symptoms. The answers of 13

items were arranged in the opposite direction to the other 14

items. Scores of these 13 items were reversed prior to data
analysis, so that higher scores indicate more depressive
symptoms. There is no validity scale within this measure.
Thus, we trained our research staff to pay attention to the
participants behavior while they were filling out the questionnaires. When cleaning the entire data set, cases whose
response seemed odd were excluded according to specific
criteria. Cronbachs a of the self-reported CDI raw score was
.82. Parents also rated their twin childrens depressive
symptoms using the same questionnaire. The inter-rater
reliability was .38, which is consistent with the finding that
parents are relatively poor reporters of childrens depressive
symptoms (Angold et al. 1987). Hence, we did not include
the parent-reported data in the following data analysis.
SLEs that occurred during the interval between Wave1
and Wave2 were assessed with a modified version of the
Life Events Checklist (Johnson and McCutcheon 1980).
The original items regarding positive life events were
excluded because negative life events are more likely than
positive life events to be related to psychological problems
(Kessler 1997). Finally, 22 independent SLEs, which are
largely independent of the childrens behavior (e.g., death
of a family member, severe family economic difficulty),
and 17 dependent SLEs that may be dependent on the
childrens behavior (e.g., trouble with teacher, failing in
competition) were included. We assessed and analyzed
both dependent and independent SLEs, because they are
both able to influence childrens depressive symptoms.
Each item in the checklist was scored 1 if the target adolescent indicated a specific event had occurred and 0 if the
event had not occurred. Both parents and children completed this questionnaire. The inter-rater reliability was .35.
Thus, we did not include the parent-reported data in the
following data analysis.
Statistical analyses
Before testing our main hypotheses, we introduced an
index, namely Reaction Level, as a measure of individual differences in sensitivity to SLEs that was independent of the level of SLEs experienced. We constructed
this index in multiple steps. First, we tested whether there
was a linear association between SLEs and change in
depressive symptoms between Wave1 and Wave2. If there
was a linear association (In fact, this was the case.), we
then divided participants according to the number of SLEs
they had experienced between the two waves. Finally, to
quantify individual differences in stress sensitivity and
facilitate direct comparison between individuals in different SLE groups, each individuals CDIdiff score was
transformed (CDIdiff = CDIwave2 - CDIwave1) into z-score
based on his/her SLE group mean and SD. That is, we

123

Behav Genet

computed z-scores separately within each SLE group. This

z-score is consequently uncorrelated with number of SLEs
and is what we called Reaction Level. This idea is
consistent with the concepts of reaction range (Gottesman
1974) and reaction surface (Turkheimer et al. 1995) that
recognize that individuals differ in how they respond to the
same environmental event.
A 0 Reaction Level means that the child had an
average change in depression symptoms for those having
the same number of SLEs as she or he had. A positive
Reaction Level means that this child was more sensitive
than average, i.e., more easily depressed when responding
to the same number of SLEs as others. Alternatively, a
negative Reaction Level means that this child was less
sensitive than average, i.e., less easily depressed when
responding to same amounts of SLEs as others.
Biometric modeling was utilized to test our main
hypotheses (Neale et al. 1999). First, univariate models
were fit to estimate the relative contributions of genes and
environment to the variance of the depressive symptoms
(CDIwave1 and CDIwave2), Reaction Level and SLE level.
Total variance was parsed into additive genetic (A), shared
environmental (C), and non-shared environmental
(E) components. Estimates of the A, C and E components
were derived by comparing the correlation among MZ twins
(who share 100% of their genes) to that among DZ twins
(who share on average 50% of their genes). Greater MZ
relative to DZ correlation is attributed to A (rMZ [ rDZ).
Within-pair correlation not due to genetic factors is attributed to C (rDZ [ 1/2rMZ). Non-shared environmental
influences (E) are estimated from the within-pair differences, which also include measurement error. Sex differences in total variance or the relative contribution of A, C
and E to total variance were estimated by using sex-limitation models (Neale and Maes 1996). A significant change
in model fit (v2) following constraining parameters for male
and female to be equal indicates a significant sex difference.
All variables were age-regressed prior to univariate analysis. Our first hypothesis is supported if a significant contribution of A to Reaction Level was observed.
Next, a bivariate Cholesky decomposition model (Neale
and Maes 1996) was fit to examine our second hypothesis
that the genetic contributors to stress sensitivity serve as
part of the genetic component of depressive symptoms. As
shown in Fig. 1, this model estimates the A, C and E
components contributing to the covariance between Reaction Level and depressive symptoms (i.e., Ag, Cg and Eg in
Fig. 1) as well as the A, C and E components contributing
to depressive symptoms specifically (i.e., As, Cs and Es in
Fig. 1). In order to control for baseline level of depression,
Wave2 CDI scores were regressed on Wave1 CDI scores
and the resulting residuals used as the depression phenotype in the bivariate model. The order for how Reaction

123

Fig. 1 Bivariate Cholesky model of Reaction Level and depressive

symptoms for one member of a twin pair. Depressive symptoms in
this model are Wave2 CDI scores with Wave1 CDI score being
controlled

Level and depressive symptoms were arranged in the

Cholesky model is consistent with the hypothesized relationship between predictor and consequence. The parameters a2 and a3 indicate the two sources of genetic influence
on depressive symptoms, with the former representing the
general genetic liability and the latter representing the
specific genetic liability. Total genetic variance of
depressive symptoms is calculated as: a22 ? a32. The proportions by which the general and specific genetic components explain the total genetic variance of depressive
symptoms are calculated as: a22/(a22 ? a32) and a32/
(a22 ? a32), respectively. Similar to the genetic influence,
the parameters c2 and c3 are two possible sources of shared
environmental influence on depressive symptoms, and the
parameters e2 and e3 are two possible sources of nonshared environmental influences on depressive symptoms,
although the environmental influences were not our major
interest in the present study. Given that the Reaction Level
is not correlated with SLE level, the main effect of SLE on
depressive symptoms was added to the model by estimating the mean of depression as l ? bmM, where M is the
number of SLEs and bm is the regression coefficient.
To further confirm that the specific genetic component
(Ag) works regardless of the SLE level, we incorporated
gene-environment interaction terms (Purcell 2002) into the
bivariate model to examine the moderation effect of SLE
on the paths from As ? depressive symptoms, and from
Ag ? depressive symptoms (Fig. 2). The two genetic
paths were re-expressed as linear functions of the SLE
level: a ? baM, where a is the main genetic effects and
the ba is the regression coefficient marking the extent to
which the genetic effects vary as a function of the SLE
level. Similarly, paths from shared and non-shared environment factors to depressive symptoms were re-expressed
as: c ? bcM and e ? beM, respectively. If As has its effect
on depression regardless of SLE level, then, a significant
bas (regression coefficient in the path from As ? depressive symptoms) should not be expected.

Behav Genet

Results
Creation of Reaction Level

Fig. 2 Bivariate Cholesky model incorporating interactions between

SLE and latent genetic and environmental factors on depressive
symptoms

As shown in Fig. 3, a linear association between SLEs and

change in CDI scores was observed (r = .13, p \ .001).
The more SLEs one had experienced, the more depressed
this child would be (indicated by greater increase in his/her
CDI score from Wave1 to Wave2). This linear association
between SLEs and depression has also been reported in
previous studies (e.g., Ge et al. 2009). Meanwhile, individual differences in depression were clearly observed
within each SLE group: Change in CDI scores varied
among children who had experienced similar amounts of
SLEs. Based on these findings, we divided the participants
responses into 7 SLE groups and created a Reaction Level
index for each participant.
Descriptive analyses

Fig. 3 The association between the number of SLEs and depressive

symptoms. Participants were divided into seven groups according to
the number of SLEs they had experienced. Y-axis represents the
change in their CDI scores during the time interval between the two
waves (CDIdiff = CDIwave2 - CDIwave1)

Model-fitting analyses were performed with the Mx

software (Neale 1997). Model fit was evaluated by comparing -2 9 loglikelihood (-2LL), Akaikes Information
Criterion (AIC), Bayesian Information Criterion (BIC), the
sample-size adjusted BIC, and the Deviance Information
Criterion (DIC) between nested models. Lower values
indicate better fit. All variables were age and sex-regressed
prior to bivariate analysis to correct for potential biases.
Skewed variables were log-transformed to better approximate normality.

Table 1 provides descriptive statistics for depressive

symptoms (CDIwave1 and CDIwave2), SLEs and Reaction
Level. Male and female adolescent twins reported similar
levels of depressive symptoms in both Wave1 (37.32 for
males and 36.97 for females, t = .91, p = .36), and Wave2
(39.45 for males and 39.28 for females, t = .42, p = .68).
They also experienced similar amounts of SLEs during the
time interval between the two waves (2.74 for males and
2.72 for females, t = .14, p = .89). The participants
reported more depressive symptoms in Wave2 than in
Wave1 (39.35 v.s. 37.08, t = -11.30, p \ .001). The mean
increase in CDI scores did not differ between the two sexes
(t = -.38, p = .70). There was also no sex difference in
Reaction Level (t = -.07, p = .94).
Wave2 depression was significantly correlated with
Wave1 depression, SLE level and Reaction Level (r = .48,
.35 and .52, respectively, ps \ .001). Multiple linear
regression further revealed that each of these variables
contributed independently to the prediction of Wave2
depression (B = 1.00, .34 and 6.80, SE = .004, .008, .023
for Wave1 depression, SLE level and Reaction Level

Table 1 Descriptive statistics of depressive symptoms in Wave1 and Wave2, stressful life events occurring between the two waves, and the
Reaction Level
MZ

DZ

Male
Mean

(N = 432)
SD

Female
Mean

(N = 476)
SD

Male
Mean

(N = 118)
SD

Female
Mean

(N = 186)
SD

CDI score in Wave1

37.37

6.40

36.48

6.33

37.13

6.28

38.16

6.95

CDI score in Wave2

39.14

6.90

38.86

6.36

40.52

7.94

40.27

7.10

Stressful life events

2.68

2.73

2.70

2.38

2.96

3.13

2.83

2.20

Reaction Level

-.06

.99

-.01

.95

.13

.96

-.01

.99

123

Behav Genet
Table 2 Twin correlation and univariate A, C and E estimates (95% confidence intervals) and model fit statistics
rMZ

rDZ

Fit statistics
v2

df

AIC

CDI score in Wave1

.48

.30

.42 (.12, .60)

.06 (.00, .33)

.52 (.45, .60)

3.28

.51

-4.72

CDI score in Wave2

.57

.31

.58 (.36, .64)

.00 (.00, .20)

.42 (.36, .49)

5.76

.22

-2.24

Reaction Level

.42

.20

.39 (.10, .48)

.00 (.00, .26)

.61 (.52, .71)

5.02

.29

-2.98

Stressful life events

.49

.37

.33 (.04, .56)

.18 (.00, .44)

.49 (.43, .57)

1.73

.78

-6.27

respectively, ps \ .001; for the full regression model,

R2 = .90, F(3) = 4.31, p \ .001). The results of regression
analysis reinforced the rationale used to set up the bivariate
Cholesky model (Fig. 1).
Univariate models
Fitting the sex-limitation model did not reveal any significant sex difference regarding either the total phenotypic
variance or the relative contribution of A, C and E components to total variance for depressive symptoms (Wave1
and Wave2) and Reaction Level. For the SLEs, however,
constraining the total variance to be equal between the two
sexes led to a significant decrease in model fit (Dv2(1) =
18.40, p \ .001), indicating greater variance for males than
for females. Because the relative contributions of A, C, and
E in males and females are however similar, we report the
common parameter estimates for male and female twins in
Table 2.
As shown in Table 2, both the Wave1 and Wave2
depressive symptoms were moderately heritable. Interestingly, the heritability of depressive symptoms increased by
16% during the time interval from Wave1 to Wave2,

implying that adolescence might be a critical period in the

etiology of depression. Depressive symptoms also exhibited substantial non-shared environmental effects. The
contribution of shared environment to depressive symptoms, however, was minimal and non-significant. As
expected, a moderate and significant heritability of Reaction Level was observed, supporting our first hypothesis.
The remaining variance in Reaction Level could mainly be
attributed to the non-shared environmental effect. SLEs
also exhibited a moderate heritability (cf. McGuffin et al.
1988), a minor shared environmental effect and a substantial non-shared environmental effect. The non-shared
environment effect also included error.
Bivariate models
Since the shared environmental effects to both depressive
symptoms and Reaction Level were minimal and non-significant, the Cg and Cs components were excluded from the
bivariate models. As shown in the upper part of Table 3, all
path coefficients in the bivariate model were significant.
Closely relevant to the purpose of this study, the a2 and a3
were significant, indicating substantial effects from the

Table 3 Parameter estimates with 95% confidence intervals for the bivariate models
Model

Path

Intercept
Parameter

Coefficient for moderator effect

Estimates

Parameter

Estimates

Bivariate model

Ag ? Reaction Level

a1

.61 (.54, .69)

Without moderation effect of SLEs

Ag ? depressive symptoms

a2

3.48 (3.01, 3.94)

As ? depressive symptoms

a3

1.84 (1.63, 2.05)

Eg ? Reaction Level

e1

.72 (.67, .77)

Eg ? depressive symptoms

e2

3.68 (3.39, 4.00)

Es ? depressive symptoms

e3

1.81 (1.68, 1.96)

Bivariate model

Ag ? Reaction Level

a1

.61 (.53, .68)

With moderation effect of SLEs

Ag ? depressive symptoms

a2

3.61 (3.15, 4.06)

Bag

-.11 (-.22, .01)

As ? depressive symptoms

a3

1.85 (1.64, 2.06)

Bas

.06 (-.05, .17)

Eg ? Reaction Level

e1

.73 (.68, .78)

Eg ? depressive symptoms

e2

Es ? depressive symptoms

e3

3.60 (3.32, 3.91)

Beg

.18 (.09, .27)

1.61 (1.48, 1.76)

Bes

.15 (.08, .23)

Upper panel presents the results of fitting the bivariate Cholesky model without moderation effect of SLEs. Lower panel presents the results
of fitting the bivariate Cholesky model with moderation effect of SLEs

123

Behav Genet
Table 4 Fit statistics for bivariate models
Model

-2LL

df

AIC

BIC

BICadj

DIC

Full bivariate model without moderation effect of SLEs

7636.15

1978

3680.15

-2442.02

697.56

-624.36

No general genetic component (Ag)

7773.82

1983

3807.82

-2389.01

758.50

-566.76

No specific genetic component (As)

7750.57

1983

3784.57

-2400.49

747.03

-578.23
-183.18

No general non-shared environmental component (Eg)

8540.96

1983

4574.96

-2005.44

1142.08

No specific non-shared environmental component (Es)

8806.28

1983

4840.28

-1872.78

1274.73

-50.52

Full bivariate model with moderation effect of SLEs

7636.15

1978

3680.15

-2442.02

697.56

-624.36

Without genetic moderators (bag & bas)

7639.72

1980

3679.72

-2446.57

696.18

-627.07

Without non-shared environmental moderators (beg & bes)

7667.45

1980

3717.45

-2427.70

715.05

-608.20

hypothesized general genetic source (Ag) and specific

genetic source (As) to depression. Results of fit statistics of
this model and its nested models are shown in Table 4.
Fixing either a2 or a3 to 0 leads to poorer model fits
(Table 4). Thus, neither genetic component (Ag and As)
could be excluded from the model. Similarly, excluding
either environmental component (Eg and Es) led to poorer
model fits.
Further analysis showed that the genetic correlation
between depressive symptoms and Reaction Level is .884
(95% CI: .843.916). The hypothesized general genetic
source (Ag) contributed 37.5% (95% CI: 29.445.1%) to
the total variance of depressive symptoms, while the
hypothesized specific genetic source (As) contributed
10.5% (95% CI: 8.113.0%) to the total variance of
depressive symptoms.
As shown in the lower part of Table 3, both the general
genetic source (Ag) and the specific genetic source (As) did
not interact with the SLE level. On the contrary, the general non-shared environmental component (Eg) and the
specific non-shared environmental component (Es) interacted with SLE level. To determine whether the absence of
significant moderation effect of SLE on genetic pathways
was a consequence of insufficient statistical power, we
tested all competing models from the models with one
moderator (any one out of the four) to multiple moderators.
As shown in Table 4, the best fitting model was a model
that contained the two non-shared environmental moderators (beg and bes), but did not contain the two genetic
moderators (bag and bas). Hence, we successfully showed
evidence for the two sources of genetic liability to
depression: one corresponding to genetic contributors of
stress sensitivity; and another which is specific, that is,
statistically independent of stress sensitivity.

Discussion
In the present study, we successfully decomposed the
genetic contribution to depressive symptoms into two

parts: One part that had its effects on depression in terms of

genetically-influenced individual differences in stress sensitivity, and another part that worked independently of
stress sensitivity. These findings are consistent with the
multifactorial polygenic model whereby the genetic liability to psychopathology represents the combined contributions of both general and specific factors. It is
noteworthy that depressive symptoms, which were
measured by the CDI in the present study, refers to a
symptomatic spectrum detectable in the general population. It is not conceptually identical to Major Depressive
Disorders or other forms of mood disorders which are
measured by clinical diagnostic criteria, such as DSM-IV.
However, since many scholars have argued that normal and
abnormal trait variation can be treated within a single,
unified structural framework, and that abnormal traits can
be modeled as extremes of normal trait variation (Markon
et al. 2005; OConnor and Dyce 2001), results of this study
will contribute to our understanding of genetic and environmental liability to clinical depression. In fact, evidence
has shown that depressive symptoms as measured by
CDI are predictive of clinical depressive disorder. Moreover, the CDI total score differentiates depressive disorder
from anxiety disorder and disruptive behavior disorder
(Timbremont et al. 2004).
To the best of our knowledge, this is the first adolescent
twin study on the heritability of stress sensitivity.
According to previous literature, stress sensitivity acts as a
trait that plays an important role in the etiology of
depression. For example, heightened stress sensitivity
could be a psychological scar which influences the recurrence of depression (Burcusa and Iacono 2007). Albeit
important, since most studies investigating the genetic
basis of stress sensitivity have used animal models (e.g.,
Bartolomucci et al. 2010) or studied specific gene loci in
terms of G 9 E interaction (e.g., Caspi et al. 2003), little is
known about the extent to which stress sensitivity is
attributable to genetic influences. Thus, the present study
added to the literature by showing that stress sensitivity is
moderately heritable.

123

Behav Genet

Stress sensitivity could be a promising candidate as an

endophenotype for clinical depression (cf. Farmer 2004).
The term endophenotype was first described as an
internal phenotype for psychopathology (Gottesman and
Shields 1972, 1973). The main aim of an endophenotype
strategy is to reduce the complexity of phenotypes and to
fill the gap in the pathway between symptoms and putative
genes by introducing variables that are measurable constructs in the pathway. The identification and validation of
endophenotypes is crucial for this strategy. Although stress
sensitivity has been proposed as a candidate endophenotype for depression (Hasler et al. 2004), little effort has
been made to test its validity. In this study, we found that
the Reaction Level fulfilled at least two criteria out of six
for being an endophenotype (Chan and Gottesman 2008;
Gottesman and Gould 2003); that is, it is heritable and
associated with symptoms (depressive symptoms in this
study). Furthermore, the observed genetic correlation
between Reaction Level and depressive symptoms also
supports that stress sensitivity could be an endophenotype
for depression.
The participants of the current study were in adolescence, a transitional period within which dramatic changes
occur in biological domain as well as in psychosocial
domain. Both biological factors (such as puberty) and
psychosocial factors (such as life stressors) could account
for the significant rise in rates of depressive symptoms for
adolescents (Ge et al. 2001). Findings of this study highlight the importance of both biological factors and psychosocial factors in the development of depression. First,
robust genetic influence appears during adolescence.
Consistent with several other studies (Kendler et al. 2008;
Thapar and McGuffin 1994), the heritability of depressive
symptoms observed in our adolescent twin sample is much
higher than the heritability observed in adult twins or twins
younger than 1011 years (McGue and Christensen 1997,
2003; Thapar and McGuffin 1994). Furthermore, the heritability of depressive symptoms increased by 16% from
Wave1 to Wave2. All the evidence implies that dramatic
changes may occur in the underlying mechanisms of
depression during the transitional period from childhood to
adulthood. Future research is needed to answer why genetic
factors exhibit more impact in adolescence than in any
other period of ones life.
It is worth mentioning that some adolescent twin studies
(e.g., Silberg et al. 1999) report lower heritability of
depression than we reported in our study and that others
have reported in some similar studies (Kendler et al. 2008;
Thapar and McGuffin 1994). Difference in measurement
methodology may account for this inconsistency. Studies
using clinical interview techniques (Silberg et al. 1999)
usually get relatively low estimates, whereas studies using
self-reported questionnaires usually get relatively higher

123

estimates (Kendler et al. 2008; Thapar and McGuffin

1994).
Second, the finding that the effects of non-shared environmental factors on depressive symptoms were moderated
by SLE level reminds us of the importance of E 9 E
interactions in the development of depression. Adolescence
is such a life stage during which one often faces a confluence of diverse life stressors. Multiple environmental
risk factors may have their adverse effects in an interactive
as well as additive way. Thus, simultaneous exposure to
multiple environmental risk factors would constitute an
extraordinarily level of risk. However, in contrast to G 9 E
interactions and G 9 G interactions, the E 9 E interactions, or cascading effects, seem to be underemphasized in
recent studies. They warrant more attention in future
research.
Different types of SLEs (e.g., chronic v.s. transient ones,
dependent v.s. independent ones) may have different
effects on adolescents. In addition to the SLEs that
occurred between Wave1 and Wave2 (and therefore may
be transient), we also measured major SLEs that occurred
before Wave1 (such as parental divorce) as well as chronic
stressors. Regression analysis showed that Wave2 depression was significantly predicted by recent SLEs (B = .31,
p \ .001), but not by SLEs prior to Wave1 (B = - .01,
p = .33) when controlling for Wave1 depressive symptoms
and Reaction Level. Thus, we focused on more recent
SLEs in the present study. Comparison was also made
among different types of SLEs. Some events, such as
being blamed and spanked by parents quite often were
more potent than others in predicting depressive symptoms.
Nevertheless, the depressogenic effects of different SLEs
appeared to cumulate, such that the number, rather than just
the occurrence, of SLEs was associated with depressive
symptoms. Thus, we did not differentiate the effects of
different SLEs in this study. Consistent with previous
studies (e.g., Kendler et al. 1999), we found that dependent
SLEs were more strongly associated with depressive
symptoms than independent SLEs (r = .41 for dependent
SLEs, r = .18 for independent ones, both ps \ .001). Since
the independent SLEs were associated with depression
symptoms, however, we included them along with the
dependent SLEs in our SLE scale.
The gender difference in depression, whereby females
are more vulnerable than males, is a well established
finding in western samples. In contrast, this finding is not
consistently found in Chinese samples (e.g., Lee et al.
2009). Some studies (e.g., Sun et al. 2010) even observe a
higher prevalence of depression in boys than in girls.
Researchers have speculated that more life stressors and/or
less coping strategy in boys might be able to account for
the absence of strong evidence for gender difference in
Chinese samples. More evidence is needed before one can

Behav Genet

draw any conclusion about gender differences in depression among Chinese children and adolescents.
Limitations should be kept in mind when interpreting
the results of the present study. First, the Reaction Level is
not a precise measure of stress sensitivity. However, since
the definition and nature of stress sensitivity are far from
clear, it should be meaningful to introduce such an index to
roughly reflect individual differences in stress sensitivity.
Alternative measures do exist, such as the hormonal cortisol reaction to stressful stimuli in the laboratory. But
those measures usually focus on one specific aspect of the
stress response, and results obtained in laboratory studies
may not be generalized to real life. On the contrary, the
Reaction Level could be considered as an overall consequence of all mechanisms or processes related to stress
response in real life. Second, instead of a causal relationship, the observed correlation between Reaction Level and
depressive symptoms might be a consequence of other
mechanisms, such as poverty or family conflict. However,
the longitudinal design of the present study and a further
regression analysis, which revealed a significant correlation
between Reaction Level and Wave2 depressive symptoms
after controlling for family income and the quality of
parents marriage life, enhance the possibility that the
Reaction Level leads to depression. Third, the Reaction
Level was derived from the difference between Wave1 and
Wave2 depression (within level of SLEs), while the outcome variable in the biometric model is the residual Wave2
depression predicted by Wave1 depression, there could be
potential confounding in the interpretation of the results.
But since the effect of SLE level on depression is strong,
the confounding may be minor. Fourth, although the
residual genetic effects in the bivariate model were interpreted as specific genetic liability in this study, it is possible that these genetic effects are shared by other
disorders, just not via stress reactivity.
In summary, multiple sources of genetic and environmental factors combined to convey risk for indicators to
depression. Investigating the genes associated with stress
sensitivity could be a useful strategy for identifying
depression-related genes. Practical implications for schools
and mental health counselors include developing strategies
to reduce adolescents stress sensitivity, and paying more
attention to children who may have a genetic disposition
toward high stress sensitivity, and who are exposed to
multiple SLEs and other adverse environments.
Acknowledgments This study was funded by the Knowledge
Innovation Program of the Chinese Academy of Sciences (KSCX2EW-J-8) and the National Natural Science Foundation of China
(31170993). The first author was a postdoctoral guest worker in the
Department of Psychology and the Institute of Child Development at
the University of Minnesota while working on this project. We thank
all the personnel involved in the sample recruitment and data

collection. We dedicate this article to the memory of our late and

beloved colleague Dr. Xiaojia Ge for his great contribution in
founding the Beijing Adolescent Twin Project.
Conflict of interest The authors do not have any financial relationship with the organization that sponsored the research. We have
full control of all primary data and we agree to allow the journal to
review our data if requested.

References
Angold A, Weissman MM, John K, Merikangas KR, Prusoff BA,
Wickramaratne P et al (1987) Parent and child reports of
depressive symptoms in children at low and high risk of
depression. J Child Psychol Psychiatry 28(6):901915
Bale TL (2006) Stress sensitivity and the development of affective
disorders. Horm Behav 50(4):529533
Bartolomucci A, Carola V, Pascucci T, Puglisi-Allegra S, Cabib S,
Lesch KP et al (2010) Increased vulnerability to psychosocial
stress in heterozygous serotonin transporter knockout mice. Dis
Model Mech 3(78):459470
Becker JB, Monteggia LM, Perrot-Sinal TS, Romeo RD, Taylor JR,
Yehuda R et al (2007) Stress and disease: is being female a
predisposing factor? J Neurosci 27(44):1185111855
Burcusa SL, Iacono WG (2007) Risk for recurrence in depression.
Clin Psychol Rev 27(8):959985
Caspi A, Sugden K, Moffitt TE, Taylor A, Craig IW, Harrington H
et al (2003) Influence of life stress on depression: moderation by
a polymorphism in the 5-HTT gene. Science 301(5631):386389
Chan RC, Gottesman II (2008) Neurological soft signs as candidate
endophenotypes for schizophrenia: a shooting star or a Northern
star? Neurosci Biobehav Rev 32(5):957971
Chen Z, Guo F, Yang X, Li X, Duan Q, Zhang J et al (2009)
Emotional and behavioral effects of romantic relationships in
Chinese adolescents. J Youth Adolesc 38(10):12821293
Chen J, Li X, Chen Z, Yang X, Zhang J, Duan Q et al (2010)
Optimization of zygosity determination by questionnaire and
DNA genotyping in Chinese adolescent twins. Twin Res Hum
Genet 13(2):194200
Eaves LJ (2006) Genotype 9 environment interaction in psychopathology: fact or artifact? Twin Res Hum Genet 9(1):18
Farmer A (2004) Bad luck and bad genes in depression. In: DiLalla
LF (ed) Behavior genetics principlesperspectives in development, personality, and psychopathology. American Psychological Association, Washington, pp 107121
Fowles DC (1992) Schizophrenia: diathesis-stress revisited. Annu
Rev Psychol 43:303336
Ge X, Conger RD, Elder GH Jr (2001) Pubertal transition, stressful
life events, and the emergence of gender differences in
adolescent depressive symptoms. Dev Psychol 37(3):404417
Ge X, Natsuaki MN, Neiderhiser JM, Reiss D (2009) The longitudinal
effects of stressful life events on adolescent depression are
buffered by parent-child closeness. Dev Psychopathol
21(2):621635
Gottesman II (1974) Developmental genetics and ontogenetic psychology: overdue detente and propositions from a matchmaker.
In: Pick AD (ed) Minnesota symposia on child psychology, vol 8.
University of Minnesota Press, Minneapolis, pp 5580
Gottesman II, Gould TD (2003) The endophenotype concept in
psychiatry: etymology and strategic intentions. Am J Psychiatry
160(4):636645
Gottesman II, Shields J (1967) A polygenic theory of schizophrenia.
Proc Natl Acad Sci USA 58(1):199205

123

Behav Genet
Gottesman II, Shields J (1972) Schizophrenia and genetics; a twin
study vantage point. Academic Press, New York
Gottesman II, Shields J (1973) Genetic theorizing and schizophrenia.
Br J Psychiatry 122:1530
Gottesman II, Shields J (1982) Schizophrenia: the epigenetic puzzle.
Cambridge University Press, New York
Haque FN, Gottesman I, Wong AH (2009) Not really identical:
epigenetic differences in monozygotic twins and implications for
twin studies in psychiatry. Am J Med Genet C 151C(2):136141
Hasler G, Drevets WC, Manji HK, Charney DS (2004) Discovering
endophenotypes for major depression. Neuropsychopharmacology 29(10):17651781
Johnson JH, McCutcheon SM (1980) Assessing life stress in older
children and adolescents: preliminary findings with the life
events checklist. In: Sarason IG, Spielberger CC (eds) Stress and
anxiety, vol 7. Hemisphere, Washington, pp 111125
Kendler KS, Kessler RC, Walters EE, MacLean C, Neale MC, Heath
AC et al (1995) Stressful life events, genetic liability, and onset
of an episode of major depression in women. Am J Psychiatry
152(6):833842
Kendler KS, Karkowski LM, Prescott CA (1999) Causal relationship
between stressful life events and the onset of major depression.
Am J Psychiatry 156(6):837841
Kendler KS, Thornton LM, Gardner CO (2001) Genetic risk, number
of previous depressive episodes, and stressful life events in
predicting onset of major depression. Am J Psychiatry 158(4):
582586
Kendler KS, Gardner CO, Lichtenstein P (2008) A developmental twin
study of symptoms of anxiety and depression: evidence for genetic
innovation and attenuation. Psychol Med 38(11):15671575
Kessler RC (1997) The effects of stressful life events on depression.
Annu Rev Psychol 48:191214
Kovacs M (1992) Childrens depression inventory manual. NultiHealth Systems, North Tonawanda
Lee S, Tsang A, Huang YQ, He YL, Liu ZR, Zhang MY et al (2009)
The epidemiology of depression in metropolitan China. Psychol
Med 39(5):735747
Markon KE, Krueger RF, Watson D (2005) Delineating the structure
of normal and abnormal personality: an integrative hierarchical
approach. J Pers Soc Psychol 88(1):139157
McGue M, Christensen K (1997) Genetic and environmental contributions to depression symptomatology: evidence from Danish twins
75 years of age and older. J Abnorm Psychol 106(3):439448
McGue M, Christensen K (2003) The heritability of depression
symptoms in elderly Danish twins: occasion-specific versus
general effects. Behav Genet 33(2):8393

123

McGuffin P, Katz R, Bebbington P (1988) The Camberwell collaborative depression study. III. Depression and adversity in the
relatives of depressed probands. Br J Psychiatry 152:775782
Neale MC (1997) Mx: statistical modelling, 4th edn. Department of
Psychiatry, Richmond
Neale MC, Maes HH (1996) Methodology for genetics studies of
twins and families, 6th edn. Kluwer, Dordrecht
Neale MC, Boker SM, Xie G, Maes HH (1999) Mx: statistical
modeling, 5th edn. Department of Psychiatry, Richmond
OConnor BP, Dyce JA (2001) Rigid and extreme: a geometric
representation of personality disorders in five-factor model
space. J Pers Soc Psychol 81(6):11191130
Purcell S (2002) Variance components models for gene-environment
interaction in twin analysis. Twin Res Hum Genet 5(6):554571
Silberg J, Pickles A, Rutter M, Hewitt J, Simonoff E, Maes H et al
(1999) The influence of genetic factors and life stress on
depression among adolescent girls. Arch Gen Psychiatry 56(3):
225232
Smucker MR, Craighead WE, Craighead LW, Green BJ (1986)
Normative and reliability data for the childrens depression
inventory. J Abnorm Child Psychol 14(1):2539
Sun Y, Tao F, Hao J, Wan Y (2010) The mediating effects of stress
and coping on depression among adolescents in China. J Child
Adolesc Psychiatr Nurs 23(3):173180
Tellegen A, Ben-Porath YS, McNulty JL, Arbisi PA, Graham JR,
Kaemmer B (2003) The MMPI-2 restructured clinical (RC)
scales: development, validation and interpretation. University of
Minnesota Press, Minneapolis
Thapar A, McGuffin P (1994) A twin study of depressive symptoms
in childhood. Br J Psychiatry 165(2):259265
Timbremont B, Braet C, Dreessen L (2004) Assessing depression in
youth: relation between the childrens depression inventory
and a structured interview. J Clin Child Adolesc Psychol 33(1):
149157
Turkheimer E, Goldsmith HH, Gottesman II (1995) Some conceptual
deficiencies in developmental behavior genetics: comment.
Hum Dev 38:143153
Wichers M, Barge-Schaapveld D, Nicolson N, Peeters F, de Vries M,
Mengelers R et al (2009) Reduced stress-sensitivity or increased
reward experience: the psychological mechanism of response
to antidepressant medication. Neuropsychopharmacology 34:
923931
Woolf CM (1997) Does the genotype for schizophrenia often remain
unexpressed because of canalization and stochastic events during
development? Psychol Med 27(3):659668