Meta-analysis
Peripheral cytokine and chemokine
alterations in depression: a meta-analysis of
82 studies
K
€
ohler CA, Freitas TH, Maes M, de Andrade NQ, Liu CS, Fernandes
BS, Stubbs B, Solmi M, Veronese N, Herrmann N, Raison CL, Miller
BJ, Lanct
^
ot KL, Carvalho AF. Peripheral cytokine and chemokine
alterations in depression: a meta-analysis of 82 studies.
Objective: To conduct a systematic review and meta-analysis of studies
that measured cytokine and chemokine levels in individuals with major
depressive disorder (MDD) compared to healthy controls (HCs).
Method: The PubMed/MEDLINE, EMBASE, and PsycINFO
databases were searched up until May 30, 2016. Effect sizes were
estimated with random-effects models.
Result: Eighty-two studies comprising 3212 participants with MDD
and 2798 HCs met inclusion criteria. Peripheral levels of interleukin-6
(IL-6), tumor necrosis factor (TNF)-alpha, IL-10, the soluble IL-2
receptor, C-C chemokine ligand 2, IL-13, IL-18, IL-12, the IL-1
receptor antagonist, and the soluble TNF receptor 2 were elevated in
patients with MDD compared to HCs, whereas interferon-gamma
levels were lower in MDD (Hedge’s g = 0.477, P = 0.043). Levels of
IL-1b, IL-2, IL-4, IL-8, the soluble IL-6 receptor (sIL-6R), IL-5, CCL-
3, IL-17, and transforming growth factor-beta 1 were not significantly
altered in individuals with MDD compared to HCs. Heterogeneity was
large (I
2
: 51.6–97.7%), and sources of heterogeneity were explored (e.g.,
age, smoking status, and body mass index).
Conclusion: Our results further characterize a cytokine/chemokin e
profile associated with MDD. Future studies are warranted to further
elucidate sources of heterogeneity, as well as biosignature cytokines
secreted by other immune cells.
C. A. K
€
ohler
1,
*,
T. H. Freitas
1,
*,
M. Maes
2,3,4,5,6
,
N. Q. de Andrade
1
,
C. S. Liu
7,8
,
B. S. Fernandes
2,9
,
B. Stubbs
10,11
,
M. Solmi
12,13
,
N. Veronese
12,14
,
N. Herrmann
8,15
,
C. L. Raison
16,17
,
B. J. Miller
18
,
K. L. Lanct
^
ot
7,8,15
,
A. F. Carvalho
1
1
Translational Psychiatry Research Group and
Department of Clinical Medicine, Faculty of Medicine,
Federal University of Cear
a, Fortaleza, CE, Brazil,
2
Deakin
University, IMPACT Strategic Research Centre, School of
Medicine, Geelong, Australia,
3
Department of
Psychiatry, Faculty of Medicine, Chulalongkorn
University, Bangkok, Thailand,
4
Department of
Psychiatry, Faculty of Medicine, State University of
Londrina, Londrina, PR, Brazil,
5
Department of Psychiatry,
Medical University Plovdiv, Plovdiv, Bulgaria,
6
Revitalis,
Waalre, The Netherlands,
7
Department of Pharmacology
and Toxicology, University of Toronto, Toronto, ON,
Canada,
8
Neuropsychopharmacology Research Group,
Hurvitz Brain Sciences Program Sunnybrook Research
Institute, Toronto, ON, Canada,
9
Laboratory of Calcium
Binding Proteins in the Central Nervous System,
Department of Biochemistry, Federal University of Rio
Grande do Sul, Porto Alegre, RS, Brazil,
10
Physiotherapy
Department, South London and Maudsley NHS
Foundation Trust, Denmark Hill, London, UK,
11
Health
Service and Population Research Department, Institute
of Psychiatry, Psychology and Neuroscience (IoPPN),
King’s College London, London, UK,
12
Department of
Neurosciences, University of Padova, Padova, Italy,
13
Institute of Clinical Research and Education in
Medicine (IREM), Padova, Italy,
14
Department of
Medicine, DIMED, Geriatrics Section, University of
Padova, Padova, Italy,
15
Department of Psychiatry,
University of Toronto, Toronto, ON, Canada,
16
Department of Human Development and Family
Studies, School of Human Ecology, University of
Wisconsin-Madison, Madison, WI, USA,
17
Department
of Psychiatry, School of Medicine and Public Health,
University of Wisconsin-Madison, Madison, WI, USA
and
18
Department of Psychiatry & Health Behavior,
Augusta University, Augusta, GA, USA
373
Acta Psychiatr Scand 2017: 135: 373–387 © 2017 John Wiley & Sons A/S. Published by John Wiley & Sons Ltd
All rights reserved
DOI: 10.1111/acps.12698
ACTA PSYCHIATRICA SCANDINAVICA
Key words: cytokines; chemokines; inflammation
meta-analysis; depression
Andr
e F. Carvalho, Department of Clinical Medicine,
Faculty of Medicine, Federal University of Cear
a, Rua
Prof. Costa Mendes, 1608, 4° andar, 60430-040,
Fortaleza, CE, Brazil.
E-mail: andrefc7@terra.com.br or
andrefc7@hotmail.com
*Contributed equally as first authors.
Accepted for publication December 29, 2016
Summations
•
Evidence indicates that peripheral immune activation may be involved in the pathophysiology of
major depressive disorder.
•
Herein, we conducted an updated meta-analytic review of 82 studies that measured cytokines and/or
chemokines in individuals with major depressive disorder and healthy controls.
•
Levels of IL-6, TNF-a, 10, the soluble IL-2 receptor, C-C chemokine ligand 2, IL-13, IL-18, IL-12,
the IL-1 receptor antagonist, and the soluble TNF receptor 2 were elevated, whereas interferon-c
levels were reduced in individuals with major depressive disorder compared to controls. These results
add in the characterization of a putative cytokine/chemokine profile for major depressive disorder.
Considerations
•
A large degree of heterogeneity was evident in this literature. Potential sources of heterogeneity were
not consistently reported across included studies.
•
Methodological quality has varied across included studies.
Introduction
In the past two decades, an increasing body of evi-
dence indicates that aberrations in immune-inflam-
matory pathways and activation of cell-mediated
immunity represent important pathophysiological
pathways for the development of major depressive
disorder (MDD) (1, 2). In addition, converging
experimental and clinical research points that
reciprocal neuroimmune interactions may con-
tribute to the neurobiology of MDD (3, 4). A low-
grade inflammatory response characterized by
increased numbers of granulocytes and monocytes
(1), as well as the elevated levels of acute phase
reactants (e.g., C-reactive protein and hap-
toglobin) (5, 6), inflammatory cytokines (7), and
possibly chemokines (8), has been demonstrated in
groups of individuals with MDD compared to
healthy controls (HCs). These peripheral immune
abnormalities may influence brain function through
several mechanisms. For example, evidence indicates
that cytokines may cross the blood-brain barrier,
while certain cytokines (e.g., IL-1b) may convey sig-
nals to the brain via afferent nerves like the vagus (2,
4). The pathophysiological role of proinflammatory
cytokines in MDD is further supported by preclini-
cal research indicating that proinflammatory cytoki-
nes may promote depressive-like behaviours,
whereas TNF-a and IL-6 receptor knockout mice
exhibit resilience to stress-induced depressive-like
behaviours (9–11). In addition, a recent meta-analy-
sis estimates that ~25% of patients with chronic hep-
atitis C develop depression after treatment with the
proinflammatory cytokine interferon-a (IFN-a)
(12). The common denominator among these find-
ings is that peripheral immune dysregulation may
represent an important pathway for inducing func-
tional and structural brain changes that underpin
the pathophysiology of MDD. Perhaps as a conse-
quence of this, peripheral inflammatory mediators
have emerged as promising candidate biomarkers
for MDD (13), although evidence of bias may limit
inferences derived from the literature on peripheral
biomarkers for MDD (14).
A meta-analysis that included 24 studies provided
evidence that peripheral levels of tumor necrosis fac-
tor-alpha (TNF-a) and interleukin-6 (IL-6) are sig-
nificantly elevated in individuals with MDD
compared to healthy controls (HCs) (7). However,
374
K
€
ohler et al.
between-study heterogeneity for these estimates was
high (7). A more recent cumulative meta-analysis
confirmed that peripheral levels of IL-6 are elevated
in individuals with MDD and HCs, whereas no con-
sistentevidenceofchangesinTNF-a and IL-1b in
patients with MDD compared to controls was found
(5). This significant degree of heterogeneity could be
explained by a number of factors including but not
limited to the following: differences in assay methods
across laboratories, medication status, and potential
confounders (e.g., body mass index and smoking).
Furthermore, it has been increasingly recognized that
the phenotypic heterogeneity of MDD may contribute
to discrepant findings. For example, melancholic
depression is associated with elevated HPA axis activ-
ity (15, 16), whereas individuals with atypical depres-
sion appear to have higher levels of proinflammatory
markers (16). In addition, each individual cytokine/
chemokine may have different functions relevant to
the pathophysiology of MDD. For example, some
cytokines/chemokines are predominantly proinflam-
matory, whereas others are mainly anti-inflammatory,
and some of these immune mediators have been
increasingly implicated in neuroplasticity mechanisms
(17, 18). Therefore, the characterization of peripheral
levels of a wider array of cytokines and chemokines
may be of particular relevance to this field.
Aims of the study
Since the publication of these previous meta-ana-
lyses (5, 7, 19, 20), additional studies have been con-
ducted e xamining a wider range of immune
biomarkers. Therefore, the aims of this large, collab-
orative meta-analysis were to investigate differences
in peripheral levels of a wider range of cytokines and
chemokines among individuals with major depres-
sive disorder compared to healthy controls and to
explore potential sources of heterogeneity across
studies. We anticipated that the large number of
new studies would allow for a more precise charac-
terization of the role of cytokines and chemokines as
peripheral biomarkers for major depressive disorder.
Methods
This study comprised a between-group meta-analy-
sis of studies that compared cytokine or chemokine
levels between adults with MDD and healthy con-
trols. We complied with the Preferred Reported
Items for Systematic Reviews and Meta-analysis
(PRISMA) statement (21). The literature search,
title/abstract screening, final decision on eligibility
after full-text review, and data extraction were inde-
pendently performed by two investigators. An apri-
ori defined yet unpublished protocol was followed.
Search strategy
A systematic search was conducted in the PubMed/
MEDLINE, EMBASE, and PsycINFO databases
from inception up until May 30, 2016. The detailed
search strings used in this review are presented in
the supporting information that accompanies the
online version of this article. This search strategy
was augmented through tracking the citation of
included articles in Google Scholar (22).
Study selection
We included original references published in any
language. Eligible studies had to measure periph-
eral cytokine or chemokine levels in adult subjects
(age ≥ 18 years old) that met either DSM (23) or
ICD (24) criteria for MDD, and a comparison
group of healthy controls (HCs). The following
exclusion criteria were adopted: (i) studies that
reported that participants had medical and/or psy-
chiatric comorbidities were excluded (except cur-
rent smoking); (ii) studies that included pregnant
women or women in the postpartum period; (iii)
case reports or case series (N < 10); (iv) studies
that assayed the mediators in specimens/tissues
other than blood; and (v) studies in animals. The
authors of meeting abstracts that met inclusion cri-
teria were contacted by e-mail to provide data for
analysis (no additional data were provided).
Data extraction
For each cytokine/chemokine, we extracted the
means, variance estimates [standard deviation
(SD), standard error of the mean (SEM), or 95%
confidence interval (CI)], and sample sizes for both
MDD and HC groups. From studies that pre-
sented only results of the comparison of the MDD
and HC groups, we extracted the appropriate mea-
sure (z-score or t-score). In studies that provided
median IQR or median range, we estimated
the mean SD following a standard method (25).
We also extracted the following data whenever
available: (i) first author, (ii) publication year, (iii)
gender distribution of study sample (% females),
(iv) mean age and BMI, (v) mean illness duration
(years), (vi) treatment status (drug-free during
assessment and/or treatment-na
€
ıve), (vii) percent-
age of the sample with atypical and/or melancholic
depression, (viii) measurement of depressive symp-
toms, and (ix) % of current smokers.
Methodological quality of included studies
We devised a score to estimate the methodological
quality of each study based on the following
375
Cytokines and chemokines in depression
parameters: (i) study sample ≥ 50 participants
(1 = Yes; 0 = No); (ii) Did the study control results
for potential confounders (e.g., age, BMI, gender,
race)? (1 = Y; 0 = No); (iii) Were participants with
MDD and HCs age- and gender-matched? (1 = Y;
0 = No); (iv) Was the time of sample collection
specified? (e.g., morning vs. evening) (1 = Y;
0 = No); (v) Were participants with MDD free of
antidepressant drugs during sample collection?
(1 = Y; 0 = No); and (vi) Reporting of either the
manufacturer of the test or its parameters (detection
limit and coefficient of variation) (1 = Y; 0 = No).
Thus, the score may vary from 0 to 6, with higher
scores indicating better methodological quality.
Statistical analysis
Because studies used different measurement meth-
ods, we estimated a standardized mean difference
and 95% CI (Hedges’s g) for each immune media-
tor, which provides an unbiased effect size (ES)
adjusted for small sample sizes (26). We assessed the
heterogeneity across studies using the Cochran Q
test, which provides a weighted sum of the squares
of the deviations of individual study ES estimates
from the overall estimate. In addition, heterogeneity
across studies was quantified with the I
2
statistic,
which indicates the percentage of total variation
across several studies due to heterogeneity and
which is considered high when ≥50% (27). We antic-
ipated a high degree of heterogeneity. Therefore, we
pooled ES using a random-effects model according
to the DerSimonian and Laird method (28). Meta-
analyses were conducted only for immune mediators
with at least three individual datasets.
Studies with statistically non-significant (i.e.,
negative) results are less likely to be published than
studies with significant results (14, 29). To assess
publication bias, we inspected a funnel plot graph
for asymmetry and calculated the Egger’s regres-
sion test for funnel plot asymmetry (30). Evidence
of small-study effects (indicative of publication
bias) was considered when the P-value of the
Egger’s test was <0.1, and the ES of the largest
study was more conservative or changed direction
when compared with the overall ES estimate (fun-
nel plots of ES estimates in which evidence of pub-
lication bias was observed are illustrated in Figs
S10–S14) (14). The trim-and-fill procedure was
used to estimate the ES adjusting for publication
bias (31), while the fail-safe N (i.e., the file drawer
statistic) was used to determine how many addi-
tional studies would be necessary to turn a signifi-
cant ES non-significant (32).
We explored potential sources of heterogeneity
across studies for each ES estimate, using either
subgroup (if there were at least three studies in each
subgroup) or random-effects meta-regression analy-
ses. Meta-regression analyses were conducted only
when at least 10 studies provided moderator; this
decision was made aprioribecause with fewer data-
sets, this analytic tool may provide spurious results
(33). The following variables were considered in
meta-regression analyses: sample size, mean age of
MDD group, mean age of the HC group, differences
in mean age (MDD group minus HC group), mean
body mass index (BMI) of MDD group, mean BMI
of the HC group, differences in mean BMI (MDD
group minus HC group), % of females in the MDD
group, % of females in the HC group, difference in
% of females (MDD group minus HC group), % of
current smokers, latitude of the country where the
study was executed, depression severity (expressed
as a percentage of the cutoff for severe depression in
the rating scale), methodological quality of each
included study, and mean illness duration in years.
Studies were weighted in such a way that investiga-
tions with more precise parameters (indicated by
sample size and 95% CI) had more influence in
meta-regression analyses (34). For statistically sig-
nificant ES estimates, we performed sensitivity anal-
ysesinwhichweexcludedeachstudyfromanalyses
to verify whether a single study turned results non-
significant or otherwise changed the direction of the
ES. In addition, cumulative meta-analysis was per-
formed for significant ES with at least 10 datasets.
All analyses were conducted in Stata MP soft-
ware version 14.0 (Stata Corp, College Station,
TX, USA) using the metan package. Statistical sig-
nificance was considered at an alpha level of 0.05.
Results
Study selection
Following removal of duplicates, the title/abstracts
of 4911 unique references were screened for eligibil-
ity. A total of 4432 references were excluded, while
479 full texts were retrieved and screened for eligi-
bility. Of those articles, 397 were excluded (see
Table S1 for reasons for exclusion). Finally, 82 orig-
inal studies met inclusion criteria, which provided
data from 6010 participants (3212 participants with
MDD and 2798 HCs). Figure 1 provides the
PRISMA flowchart for study selection.
Characteristics and methodological quality of included studies
Of the 82 studies included in our meta-analysis, in
43 studies (52.4%), participants with MDD and
HCs were age- and gender-matched, while 35
376
K
€
ohler et al.
studies (42.7%) adjusted results for potential con-
founders (e.g., age, gender distribution, depressive
symptom scores, or BMI). In addition, most stud-
ies ( k = 81; 98.8%) either reported the manufac-
turer of the assay or provided values of the
coefficient of variation (CV) of the test. In addi-
tion, most studies (k = 65; 79.3%) provided data
regarding medication status, whereas three (3.7%)
included only treatment-na
€
ıve (i.e., never treated
with antidepressants) participants with MDD.
Finally, most studies did not provide information
on illness duration (k = 56; 68.3%); the remaining
studies included MDD participants with illness
duration of 2.97 6.15 (mean SD) years. The
methodological quality scores of each study varied
from 1 to 6 (median: 4) (Table S2).
Studies of IL-6
Il-6 measurements were extracted from 42 studies
(1587 cases and 1183 controls). Participants with
MDD had higher concentrations compared to
HCs (g = 0.621; P < 0.001; Table 1 and Fig. 2a).
No evidence of small-study effects (which provides
an indication of publication bias) was observed.
Possible sources for the large heterogeneity
(I
2
= 64.9%) were explored using meta-regression
and subgroup analyses (Tables S3 and S4). In
meta-regression analyses, differences in gender
distribution (% females) in the MDD and HC
groups emerged as a significant moderator
(P = 0.046). Subgroup analyses showed that
heterogeneity was smaller in studies that measured
IL-6 in serum and whole blood samples compared
to plasma, while results suggest that the measure-
ment of IL-6 with ELISA is associated with higher
heterogeneity compared to other types of assay.
Of 42 studies that measured IL-6, 10 provided
adjusted differences in peripheral levels of this
cytokine to confounders (e.g., age, gender, BMI,
smoking, among other variables specific to the
study) (35–44). We re-calculated this ES consider-
ing those adjusted values. Then, the overall ES of
IL-6 was 0.543 (95% CI = 0.435‒0.651; P < 0.001).
The I
2
value was 49.8% (P < 0.001). In addition,
we performed a subgroup analysis considering
studies which did vs. did not adjust comparisons to
confounders. The overall ES for the unadjusted
studies was 0.574 (95% CI = 0.437–0.711;
P < 0.001; k = 32), with an I
2
of 52.3%
(P < 0.001). The overall ES for the adjusted stud-
ies was 0.467 (95% CI = 0.303–0.631; P < 0.001;
k = 10), with a I
2
of 38.6% (P = 0.101). Therefore,
studies that adjusted to potential confounders had
a lower degree of heterogeneity.
In sensitivity analysis, the exclusion of any indi-
vidual study from the analysis did not alter the
direction or statistical significance of the ES
gnineercS
Included
Eligibility noitacifitnedI
Studies included in
qualitative synthesis
(N = 82)
Records identified through
database searching
(N = 6104)
Additional records identified
through other sources
(N = 2)
Records after duplicates removed
(N = 4911)
Records screened
(N =4911)
Records excluded
(N = 4432)
Full-text articles assessed
for eligibility
(N = 479)
Full-text articles excluded,
with reasons
(N = 397)
Studies included in meta-
analysis
(N = 82)
Fig. 1. PRISMA flowchart of study
selection for systematic review and
meta-analysis. [Colour figure can be
viewed at wileyonlinelibrary.com]
377
Cytokines and chemokines in depression
![Fig. 1. PRISMA flowchart of study selection for systematic review and meta-analysis. [Colour figure can be viewed at wileyonlinelibrary.com]](/figures/fig-1-prisma-flowchart-of-study-selection-for-systematic-3bdpvcne.webp)