Article (PDF Available)

Gene networks associated with conditional fear in mice identified using a systems genetics approach

Department of Molecular and Medical Pharmacology, David Geffen School of Medicine, University of California, Los Angeles, CA 90095, USA.
BMC Systems Biology (Impact Factor: 2.44). 03/2011; 5(1):43. DOI: 10.1186/1752-0509-5-43
Source: PubMed


Our understanding of the genetic basis of learning and memory remains shrouded in mystery. To explore the genetic networks governing the biology of conditional fear, we used a systems genetics approach to analyze a hybrid mouse diversity panel (HMDP) with high mapping resolution.
A total of 27 behavioral quantitative trait loci were mapped with a false discovery rate of 5%. By integrating fear phenotypes, transcript profiling data from hippocampus and striatum and also genotype information, two gene co-expression networks correlated with context-dependent immobility were identified. We prioritized the key markers and genes in these pathways using intramodular connectivity measures and structural equation modeling. Highly connected genes in the context fear modules included Psmd6, Ube2a and Usp33, suggesting an important role for ubiquitination in learning and memory. In addition, we surveyed the architecture of brain transcript regulation and demonstrated preservation of gene co-expression modules in hippocampus and striatum, while also highlighting important differences. Rps15a, Kif3a, Stard7, 6330503K22RIK, and Plvap were among the individual genes whose transcript abundance were strongly associated with fear phenotypes.
Application of our multi-faceted mapping strategy permits an increasingly detailed characterization of the genetic networks underlying behavior.

Full-text (PDF)

Available from: Christopher Park
SNP genotype
Gene expression
Behavioral phenoypes
Gene networks associated with conditional fear
in mice identified using a systems genetics
Park et al.
Park et al. BMC Systems Biology 2011, 5:43 (16 March 2011)
Page 1
Gene networks associated with conditional fear
in mice identified using a systems genetics
Christopher C Park
, Greg D Gale
, Simone de Jong
, Anatole Ghazalpour
, Brian J Bennett
, Charles R Farber
Peter Langfelder
, Andy Lin
, Arshad H Khan
, Eleazar Eskin
, Steve Horvath
, Aldons J Lusis
, Roel A Ophoff
Desmond J Smith
Background: Our understanding of the genetic basis of learning and memory remains shrouded in mystery. To
explore the gen etic networks governing the biology of conditional fear, we used a systems genetics approach to
analyze a hybrid mouse diversity panel (HMDP) with high mapping resolution.
Results: A total of 27 behavioral quantitative trait loci were mapped with a false discovery rate of 5%. By
integrating fear phenotypes, transcript profiling da ta from hippocampus and striatum and also genotype
information, two gene co-expression networks correlated with context-dependent immobility were identified. We
prioritized the key markers and genes in these pathways using intramodular connectivity measures and structural
equation modeling. Highly con nected genes in the context fear modules included Psmd6, Ube2 a and Usp33,
suggesting an important role for ubiquitination in learning and memory. In addition, we surveyed the architecture
of brain transcript regulation and demonstrated preservation of gene co-expression modules in hippocampus and
striatum, while also highlighting important differences. Rps15a, Kif3a, Stard7, 6330503K22RIK, and Plvap were among
the individual genes whose transcript abundance were strongly associated with fear phenotypes.
Conclusion: Application of our multi-faceted mapping strategy permits an increasingly detailed characterization of
the genetic networks underlying behavior.
Advances in both genetic and behavioral t echniques are
providing unprecedented opportunities for dissecting
the gene networks governing behavior. Through a vari-
ety of approaches, promising candidate genes have been
identified for a wide collection of clinically relevant
traits such as anxiety, conditional fear and spatial mem-
ory [1-3]. Intercrosses and backcrosses have been widely
used to identify behavior quantitative trait loci (QTLs)
in mice, but suffer from poor mapping resolutio n. More
rec ently, the use of outbred mice has allowed fine map-
ping of a range of biological [3] and expression traits
[4,5]. However, outbred mice are a fleeting resource and
must be regenotyped and re-phenotyped for each study.
In spite of many successes, the rece nt wave of gen-
ome-wide association studies paints an increasingly
complex picture of genes underlying beha vioral traits.
The genetic architecture of most behaviors is widely dis-
tributed, with collections of independent loci making
relatively small contributions to overall trait variability
[6,7]. The largely undefined and likely complex contri-
bution of environmental factors to both the etiology and
maintenance of behavior represents another formidable
obstacle to reliable QTL mapping.
Recent work has achieved superior resolution using
panels of inbred mouse lines [8]. Power can be further
improved by incorporating recombinant inbred (RI)
strains formed by crossing cla ssical inbred strains f ol-
lowed by repeated sibl ing mating. One such resource
is the hybrid mouse diversity panel (HMDP) which
* Correspondence:
Contributed equally
Department of Molecular and Medical Pharmacology, David Geffen School
of Medicine, University of California, Los Angeles, CA 90095, USA
Full list of author information is available at the end of the article
Park et al. BMC Systems Biology 2011, 5:43
© 2011 Park et al; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons
Attribution License (, which permits unrestricted use, distr ibut ion, and reproduction in
any medium, provided the original work is properly cited.
Page 2
combines inbred and RI lines to create a panel of 100
strains with great resolution and statistical power [9].
The HMDP consists of 29 classical inbred strains sup-
plemented with 71 RI strains derived from C57BL/6J
crossed with either DBA/2J, A/J or C3H/HeJ. In addi-
tion to enhanced resolution, there are o ther significant
advantages to using the HMDP for genetic mappin g.
Each strain has been genotyped extensively [10], and
multiple individuals can be phenotyped for the same
trait, reducing measurement variability. Furthermore,
the panel is a renewable resource, since each strain can
be propagate d indefin itely [11]. Phenot ype data c an be
pooled and shared in an ongoing fash ion, while the
effects of environmental variables are easily studied.
To leverage these emerging resources, we employed
an integrative systems approach to explore the genetics
of conditional fear. Figure 1 illustrates the sources of
data we collect and how we investigate relationships to
identify genetic pathways implicated in the predisposi-
tion to fear. Mice were phenotyped on a fear condition-
ing assay, and the quantitative data comb ined with
single nucleotide polymorphism (SNP) genotypes to
map behavioral quantitative trait loci (QTLs). We cor-
rected for the confounding effects of relatedness and
population structure between strains using efficient
mixed model association (EMMA) [12]. By combining
genome-wide expression QTL (eQTL) maps for hippo-
campus and striatum, weighted gene correlation net-
work analysis (WGCNA) [13,14], and structural
equ ation modeli ng, we identified single genes and path-
ways with relationships to fear-driven behavioral
To identify regions of the genome associated with fear-
related behavior, mice from the HMDP were subjected
to a fear conditioning procedure and characterized on
48 unique behavioral phenotypes drawn from different
test phases. Using these phenotypes as quantitative
traits, we performed a genome-wide assoc iation study
(GWAS) to identify loci associated wit h each of the
behavioral traits.
SNP genotypes
ene expression
Figure 1 A systems biology approach to dissecting fear biology. Data from behavioral phenotype analysis was integrated with SNP
genotypes to map behavioral QTLs. Behavioral phenotypes were also compared to gene co-expression modules created from hippocampus and
striatum microarray datasets. Gene expression data and SNP genotypes were used together to map expression QTLs. All three datasets were
merged to prioritize mapped genes using Network Edge Orienting. This approach identifies gene networks associated with behavioral
Park et al. BMC Systems Biology 2011, 5:43
Page 3 of 16
Page 3
Cued and context fear phenotyping
Mice were tested for cued and contextual fear acquired
through a Pavlovian conditioning procedure. Such fear
memories manifest across a variety of behavioral dimen-
sions and can be collectively quantified thr ough the use
of automated tracking and analysis [15].
Immobility (freezing) is a classical measure of fear
triggered by an environmental threat. This species-speci-
fic defense response can be reliably acquired in a single
conditioning trial, making it a widely used model for
fear expression an d lea rning and me mory. We a lso
monitored other measures of fear including velocity,
thigmotaxis (wall-preference),pathshape,andhabitua-
tion. The fear conditioning assay is depicted schemati-
cally in Figure 2A. On d ay one, a mouse is placed in a
cage where an auditory conditional stimulus (CS) tone
is played for fifteen seconds followed by a brief foot
shock. Training consisted of three tone-shock pairings.
The next day, the mouse returned to the same chamber
and contextual fe ar is indexed through a collection of
behavioral endpoints including immobility. On the third
day, the mouse is placed in a novel chamber and given
a series of CS prese ntations with no foot shoc k. Cued
fear is quantified across the same behavioral endpoints
used to assess contextual fear.
Variability in freezing across the panel is s hown in
Figure 2B. Further testing d etails for each of the beha-
vioral phenotypes (labeled from B1 to B48) are provided
in Additional file 1 (Supplementary methods and T able
S1). A cluster dendrogram depicting the similarity
between the quantitative behavioral phenotypes across
the HMDP is sh own in Addi tional file 1 Figure S1. Sur-
prisingly, context and cue immobility measures clustered
closely together although they index different types of
Mapping of conditional fear QTLs
We mapped loci for behavioral phenotypes using EMMA
and 101,629 SNPs ([12], METHODS). Across 48 mea-
sured behavioral phenotypes, QTL analysis revealed 27
Day 1: Fear conditioning
Day 2: Context fear test
Day 3: Cued fear test
% immobility
% immobility
% immobility
B12 post training immobility mean
B44 context immobility mean
B25 cue immobility mean
Figure 2 Fe ar conditioning in the HMDP. A) Behavioral procedure for cued fear conditioning. Mice were s ubjected to a three-phase
procedure. On day 1, mice received 3 auditory conditional stimuli (CS) co-terminating with 0.75 mA foot shock. On day 2, mice were returned
to the conditioning chamber for an 8 minute extinction test. On day 3, mice were placed in a novel chamber and given a series of 10 CS
presentations (inter-trial interval 1 minute). Green horizontal lines show time periods when fear endpoints were measured. B) Behavioral
distributions for selected endpoints across HMDP, corresponding to panels in A. Percent immobility calculated for three separate test phases.
Park et al. BMC Systems Biology 2011, 5:43
Page 4 of 16
Page 4
loci with a P value < 4.48 × 10
, corresponding to a gen-
ome-wide false discovery rate (FDR) of 5%. This threshold
value is comparable to that from another study using the
same panel [9], which employed permutation testing to
calculate a genome-wide significance threshold of P = 4.1
or a family-wise error rate of 0.05. QTL plots for
the entire battery of behavioral endpoints are in Additional
file 1 Figure S2. The significant loci and corresponding
closest genes are summarized in Table 1.
We mapped a highly significant QTL on chromosome
7 for cued immobility (P =4.40×10
). There are two
peak markers for this locus, located ~102 kb apart and
residing in different linkage disequilibrium blocks (Addi-
tional file 1 Figure S3). One peak marker is located
within the Tyrosinase (Tyr)gene.SincetheHMDPis
composed of inbred mouse strains, a number are homo-
zygous for a recessive mut ation in Tyr leading to a n
albino coat color (26 of 94 strains phenotyped).
One study looked directly at the effects of Tyr on cue
dependent freezing behavior [16] using both B6 mice with
amutantTyr alle le and an AJ congenic strain wit h the
wildtype B6 allele substituted for the albino Tyr allele. Tyr
had only a small influence on fear learning with only
minor (if any) learning deficits due to reduced visual acuity
[17-19] and was one of likely many alleles influencing this
phenotype. Interestingly, the second peak has the same P
value as the first and lies in the glutamate receptor gene
metabotropic 5 (Grm5), which is involved in glutamatergic
neurotransmission. Homozygous null mice for Grm5 have
been shown to have reduced hippocampal long term
potentiation (LTP) [20] and impaired spatial learning [21].
These mice also have a behavioral phenotype associated
with a rodent model of schizophrenia [22]. Polymorphism
at this locus may contribute to a variance in motor activity
as a conditioned response to a tone.
eQTL mapping in hippocampus and striatum
Using gene expression measures of 25,697 transcripts as
quantitative traits from tissue fro m both the hippocam-
pus (98 strains, n = 1) and striatum (96 strains, n =1),
Table 1 Behavioral QTLs with FDR < 0.05
Quantitative Behavioral Phenotype Chromosome Base Position Nearest gene P value
B3 pre training thigmotaxis mean distance to point 9 61,060,175 Tle3 1.14 × 10
B6 post training velocity mean 15 5,887,595 Dab2 2.92 × 10
B11 pre training immobility mean 2 6,186,281 Echdc3 1.77 × 10
B11 pre training immobility mean 7 126,370,751 Gpr139 1.31 × 10
B12 post training immobility mean 8 68,297,006 March1 4.41 × 10
B24 precue immobility mean 7 94,641,553 Tyr 5.58 × 10
B24 precue immobility mean 7 94,744,373 Grm5 5.58 × 10
B24 precue immobility mean 7 107,177,259 Chrdl2 5.14 × 10
B25 cue immobility mean 3 103,364,188 Syt6 1.56 × 10
B25 cue immobility mean 3 130,123,970 Col25a1 3.44 × 10
B25 cue immobility mean 4 6,678,672 Tox 2.58 × 10
B25 cue immobility mean 7 94,641,553 Tyr 4.40 × 10
B25 cue immobility mean 7 94,744,373 Grm5 4.40 × 10
B25 cue immobility mean 7 104,540,350 Alg8 7.06 × 10
B25 cue immobility mean 15 37,521,578 Ncald 1.76 × 10
B25 cue immobility mean 19 26,658,546 Smarca2 3.80 × 10
B27 precue mobility mean 7 94,641,553 Tyr 1.37 × 10
B27 precue mobility mean 7 94,744,373 Grm5 1.37 × 10
B30 precue thigmotaxis mean distance to point 1 163,397,742 Tnfsf18 3.17 × 10
B31 cue thigmotaxis mean distance to point 11 48,065,799 Gnb2l1 1.24 × 10
B33 precue thigmotaxis mean 2 151,612,920 Psmf1 3.36 × 10
B33 precue thigmotaxis mean 11 52,523,068 Fstl4 2.20 × 10
B33 precue thigmotaxis mean 13 72,750,827 D430050G20 3.73 × 10
B38 context thigmotaxis mean distance to point 1 172,955,973 Fcgr4 1.22 × 10
B38 context thigmotaxis mean distance to point 8 53,062,087 Aga 3.62 × 10
B38 context thigmotaxis mean distance to point 9 61,070,635 Tle3 2.16 × 10
B42 context meander mean 2 129,472,283 Sirpa 3.65 × 10
B44 context immobility mean 2 128,198,673 Gm14005 3.32 × 10
B44 context immobility mean 6 71,209,634 Smyd1 5.22 × 10
B47 context mobility extinction 11 70,800,475 Dhx33 4.27 × 10
Park et al. BMC Systems Biology 2011, 5:43
Page 5 of 16
Page 5
we mapped expression quantitative trait loci (eQTLs)
and the ir corresponding expression SNPs (eSN Ps) using
EMMA ([12], see METHODS). For each tissue, we cal-
culated an independent genome-wide significance
threshold corresponding to a false discovery rate (FDR
or Q value) < 5% [23]. In hippocampus, this threshold
was P <9.21×10
while in striatum the corresponding
threshold w as P <1.19×10
. We separated the eSNPs
from each tissue into two separate categories: markers
within 2 Mb of the probe start position (termed cis or
local) and markers mor e than 2 Mb away (termed trans
or distant).
In hippocampus, we mapped 2,128 cis eQTLs, while in
striatum we mapped 2,528. There was strong overlap in
the cis eQTLs of the two tissues with 1,641 in common
= 11,831, df =1,P <10
) indicating that tran-
scription regulation due to polymorphism is strongly
preserved between tissues. Interes tingly, t he set o f cis
eQTLs unique to hippocampus was enriched in genes
from the gene ontology (GO) category [24] involved in
the positive regulation of behavior (Q =1.8×10
The top 100 cis eQTLs in each tissue along with loca-
tions of their corresponding peak markers and mini-
mum P values are provided in Additional file 1 (Tables
S2 and S3).
sequence of the transcripts interrogated by the microar-
ray might produce spurious false positive cis eQTLs due
to a change in binding avidity. To investigate this possi-
bility, we downloaded a list of 8,265,759 known SNPs
from the Perlegen SNP Database http://mouse.cs.ucla.
edu/mousehapmap and searched for each of these SNPs
in the 25,697 pr obes on the Illumina microarray. Of the
SNPs in this list, 3,841 probes contained at le ast one
SNP. In the hippocampus, we observed 535 eQTLs with
SNPs while 317 were expected proportionally (c
= 22.0,
df =1,P <2.7×10
). The striatum also showed slight
enrichment with 602 cis eQTLs exhibiting SNPs in
probes with 372 expected (c
=3.0,df =1,P =0.08).
Although probe SNPs did increase the number o f
observed cis eQTLs, the proportion was <15%, suggest-
ing that >85% of cis eQTLs do not have evidence of
being artifacts du e to poly morphism. Of c ourse, other
naturally occurring polymorphisms likely exist that are
not contained in the Perlegen SNP database and could
also lead to false positive associations.
In the hippocampus, we mapped 481,099 trans eSNPs
striatum, we mapped trans 619,418 eSNPs regulating a
total of 15,348 unique probes. Using a counting algorithm
(METHODS), we estimated these numbers corresponded
to a total of 19,876 trans eQTLs in the hippocampus and
60,150 trans eQTLs in the striatum. Genome-wide
probe/marker plots for each significant eSNP are
provided in the Supplementary materials (Additional file
1 Figures S4 and S5). Selected cis and trans eQTLs from
each tissue are shown in Figure 3A - 3D.
Comparison of our data with a recent eQTL survey in
the hippocampus using heterogeneous stock mice [25]
showed significant preservation of cis eQTLs (c
1,171, df =1,P =1.1×10
), while trans eQTLs did
not show significant overlap. This discrepancy could b e
due to weaker effect sizes for tran s eQTLs in general
compared to cis or due to differing thresholds for signif-
icance. Previous studies also found that trans eQTLs
replicated less frequently than cis [26,27]. A recent
study of liver using the HMDP [9] found 2,691 cis
eQTLs and 3,174 probes with at least one trans eQTL
with P < 4.1 × 10
. We detected similar numbers of cis
eQTLs but more trans loc i, even though the same sig-
nificance threshold was employed for both types of
eQTL. This discrepancy suggests differences in the regu-
latory networks of hepatic versus neural tissue a nd may
reflect greater transcriptional complexity in the brain.
To survey whether trans gene regulation in hippocam-
pus was similar to that found in the striatum, we com-
pared the probes regula ted by each marker across the
mined if a probe was regulated by each marker in the
hippocampus or not (surpassing a glo bal FDR of 5%)
and regulated by the same marker i n the striatum or
not. There was a significant overlap in t he genes regu-
lated by each marker across the tissues (FishersExact
Test, df =1,medianomnibus-log
(Q) = 4.1), suggest-
ing strong similarities in the regulatory networks of the
two tissues. A genome-wide plot of the -log
degree of overlap between tissues for genes regulate d by
each marker between tissues is shown in Figure 3E.
Some markers clearly show better preservation of regu-
lated probes than others. For instance a SNP on chro-
mosome 7 at 104 .063430 Mb reg ulates 33 uni que genes
in the hippocampus and 36 genes in the striatum, with
29 of the genes in common. These hubs may have
strong control of expression across different tissues.
Despite the significant overlap, differences in regulation
are likely important in delineating the cellular disparity
between hippocampus and striatum.
Weighted gene correlation network analysis (WGCNA)
We looked at the large scale organization of gene co-
expression networks in the hippocampus and striatum
microarray datasets. Weighted gene co-expression net-
work analysis is a data reduction method that groups
genes into modules in an unsupervised manner based
on self-organizing properties of com plex systems. These
co-expression networks are based on topological overlap
between genes while considering the correlation
genes have with each other and the degree of shared
Park et al. BMC Systems Biology 2011, 5:43
Page 6 of 16
Page 6
Cathepsin C (Ctsc)
chr7: 88.185676 Mbp
Usher syndrome 2A homolog (Ush2a)
chr1: 190.781342 Mbp
Leucine rich repeat containing 40 (Lrrc40)
chr3: 157.731200 Mbp
Similar to Ube2j2 protein (LOC545056)
chr14: 54.964400 Mbp
Figure 3 Examples of cis and trans eQTLs in hippocampus and striatum. A) Hippocampus cis eQTL. B) Striatum cis eQTL. C) Hippocampus
trans eQTL. D) Striatum trans eQTL. Red horizontal line represents genome wide significance threshold of FDR < 5% for each tissue. Blue vertical
line represents gene position. E) Degree of overlap between tissues for probes regulated by each marker between tissues at FDR < 5%.
Significance shown as - log
Park et al. BMC Systems Biology 2011, 5:43
Page 7 of 16
Page 7
connections w ithin the network. This method has been
used in several recent systems genetics studies to reveal
functional gene networks [28,29].
We identified 30 modules in hippocampus containing
39 to 8,445 genes and 25 modules in the striatum con-
taining 34 to 14,582 genes (Additional file 1 Table S4).
The largest module in each tissue is the grey module
which is reserved for genes that do not separate into any
other modules (noise genes). The hippocampus expres-
sion d ata organized into five more modules than the
striatum. This finding could reflect a greater cellular het-
erogeneity of the hippocampus compared to the striatum,
as module construction can tease apart patterns of differ-
ential expression in mixtures of cell types [30]. There
were other differences in co-expression networks between
the two tissues. Fo r instance the sienna3 module in the
hippocampus was not preserved in striatum. This module
was significantly enriched in neuropeptide hormone activ-
ity (Q = 6.25 × 10
) and oxygen binding (Q = 3.68 × 10
) indicating that these molecular classes may play impor-
tant roles in hippocampal function.
To eva luate the degr ee of module conservation across
the hippocampus and striatum, w e calculated Z scores
for preservation of each module using the hippocampus
as a reference. The Zsummary statistic encapsulates evi-
dence that a network module is preserved between a
reference and a test network based on aspects of within-
module network density and connectivity patterns [31].
Lower Z .summary.pres scores imply module differences
while larger ones indicate preservation. Figure 4 demon-
stratesthatmostgeneco-expre ssion modules s howed
some degree of preservation across hippocampus and
striatum, with larger modules showing better preserva-
tion than smaller ones.
The gene expression properties of each of these mod-
ules can be conden sed into module eigeng enes (MEs)
which represent the first principal component of each
module [32,33]. By correlating these MEs to behavioral
with relationships with aspects of conditional fear. Fig-
ure 5 shows the correlation of each ME in the hippo-
campus with the behavioral phenotypes of cued and
context immobility (B25 and B44). We focused on hip-
pocampus, as this tissue has been previously implicated
in learning, memory, and fear [34].
The context immobility phenotype (B44) showed the
strongest correlations with two MEs in the hippocam-
pus: brown (r = -0.43, P = 0.002, Q = 0.07) and darkgrey
(r = 0.4, P =0.005,Q = 0.08). We focus on these two
mod ules for further analysis and annotate them context
fear module 1 (CF1) and context fear module 2 (CF2)
respectively. Notably, no MEs showed significant corre-
lations with cued immobility (B25) even though cue and
context immobility phenotypes clustered together (Addi-
tional file 1 Figure S1). This observ ation is consi stent
with the biology of cued immobility which relies on the
amygdala but is hippocampal dependent [35].
We looked for functional enrichment of specific gene
ontologies (GO) in the two selected context fear mod-
ules using the program GOEAST, which provides an
FDR corrected Q value[36]scoreforenrichmentin
each category. The most highly represented ontologies
are shown in Additional file 1 Tables S5 and S6. Ge nes
in the intracellular portion of the cell were enriched in
both modules (CF1: Q =1.54×10
,CF2:Q =2.3
), as were those involved in the mitochondrion
(CF1: Q =4.38×10
, Q =2.1×10
). By contrast,
classes of genes involved in metabolic processes and
gene expression were specific to CF1. Genes involved in
protein t argeting and the rough endoplasmic reticulum
were prominent in CF2 but not in CF1. Results of corre-
lations between MEs and all quantified behavioral traits
for the hippocampus and striatum are provided in Addi-
tional file 1 (Figures S6 and S7).
Genes within each module are prioritized according to
their intramodular connectivity (the sum of connection
strengths with other genes within the network). Those
with a high degree of c onnectivity are considered hubs
and can be viewed as important players in molecular
pathways. T here was a high correlation between the
intramodular connectivity measures of each gene across
the hippocampus and striatum (r = 0.53, P < 2.2 × 10
indicating strong similarities in the transcriptional net-
works of these neural tissues.
The gene mitogen-activated protein kinase 1
(Map2k1) was one of the most highly connected genes
in CF1 and has been previously implicated in long-term
synaptic plasticity and memory [37]. The gene protea-
some (prosome, macropain) 26 S subunit, non-ATPase,
6(Psmd6) acted as another hub in CF1, while in CF2,
the genes ubiquitin-conjugating enzyme E2A (Ube2a),
nuclear factor I/B (Nfib), and ubiquitin specific pepti-
dase 33 (Usp33) had the strongest intramodular connec-
tivity and served as hubs for this module. These results
suggest a role for targeted protein degradation in path-
ways associated with context dependent fear, consistent
with a recent s tudy that showed that synaptic protein
degradation through polyubiquitination underlies the
destabilization of retrieved fear m emory [38]. Other co-
expressed genes identified in these modules may also
play critical roles in the molecular mech anisms go vern-
ing learning and memory. Complete details for the gene
co-expression network analysis for each tissue and the
corresponding measures of intr amodular connectivity
for each gene can be found i n Supplementary materials
(Additional file 2).
Park et al. BMC Systems Biology 2011, 5:43
Page 8 of 16
Page 8
MEs as quantitative traits
Each module eigengene can be considered a quantitative
trait, allowing for mappi ng of SNPs associated wi th var-
iation in groups of co-expressed genes. This strategy
revea ls lo ci th at perturb the expression of gene modules
with hopes of uncovering key drivers for traits of phy-
siological relevance [39]. Mapping results that sur vive a
Bonferroni correction for all 101,629 markers are sum-
marized in Table 2. Lo ci regulating six MEs in the hip-
pocampus were mapped, of which four were preserved
in the striatum and two were specific to hippocampus.
The first hippocampal specific locus regulated the dar-
kolivegreen module and mapped to a SNP on chromo-
some 7 within t he intron for the gene TEA domain
family member 1 (Tead1),ageneknowntobeasso-
ciated with transcription factor complexes. This module
was enriched in the cellular component flotillin complex
(Q =4.910
) and the molecular function calmodu-
lin-dependent protein kinase activity (Q =4.77×10
The second hippocampal specific locus regulated the
white module and mapped to a SNP on chromosome 1
at 173.121821 Mb. This module consisted of genes
involved in the positive regulation of the acute inflam-
matory response to antigenic stimulus (Q = 4.54 × 10
Themodulewiththestrongest association to physio-
logically relevant GO categories that also possessed reg-
ulatory loci for both tissues was the yellowgreen module
in the hippo campus (saddlebrow n in striatum). This
mod ule was enriched in antigen processing and presen-
tation (Q =1.61x10
) and MHC protein complex (Q
). This module may play a role in synaptic
remodeling, as neuronal MHC class I molecules were
recently found to regulate synapses in the central ner-
vous system in response to activity [40]. Interestingly,
Figure 4 Gene co-expression module preservation across hippocampus and striatum. Modules were constructed separately for each tissue
and preservation assessed by Zsummary score using hippocampus modules as reference set. Larger modules tended to be better preserved
across tissues.
Park et al. BMC Systems Biology 2011, 5:43
Page 9 of 16
Page 9
0.073 / (0.6) 0.089 / (0.5)
-0.029 / (0.8) 0.28 / (0.06)
0.17 / (0.2) 0.0032 / (1)
0.081 / (0.6) -0.12 / (0.4)
0.17 / (0.2) 0.0061 / (1)
0.029 / (0.8) 0.36 / (0.01)
0.13 / (0.4) 0.4 / (0.005)
-0.093 / (0.5) 0.15 / (0.3)
0.088 / (0.5) -0.13 / (0.4)
-0.027 / (0.9) -0.27 / (0.06)
-0.15 / (0.3) 0.11 / (0.5)
-0.16 / (0.3) -0.046 / (0.8)
0.19 / (0.2) 0.11 / (0.4)
0.17 / (0.3) -0.09 / (0.5)
0.23 / (0.1) 0.06 / (0.7)
0.061 / (0.7) -0.34 / (0.02)
-0.079 / (0.6) -0.26 / (0.07)
0.066 / (0.7) -0.23 / (0.1)
-0.11 / (0.5) -0.26 / (0.07)
-0.07 / (0.6) -0.17 / (0.2)
0.14 / (0.3) -0.32 / (0.03)
-0.033 / (0.8) -0.43 / (0.002)
0.0096 / (0.9) -0.37 / (0.01)
0.0068 / (1) -0.16 / (0.3)
-0.12 / (0.4) 0.18 / (0.2)
-0.15 / (0.3) -0.023 / (0.9)
0.025 / (0.9) -0.076 / (0.6)
0.1 / (0.5) 0.29 / (0.04)
0.13 / (0.4) 0.11 / (0.5)
0.041 / (0.8) -0.097 / (0.5)
B25 cue immobility mean
B44 context immobility mean
Figure 5 Correlation of module eigengenes with cued and context immobility phenot ypes in the hippocampus.Columnsrepresent
cued and context immobility phenotypes and the rows represent MEs. Correlations between MEs and phenotype represented by colors ranging
from red (high positive correlation) to green (high negative correlation). Correlation coefficient shown for each comparison with corresponding P
value in parentheses. Two highlighted modules shown in boldface.
Table 2 Loci regulating module eigengenes and significance
Hippocampus module Striatum module Chromosome Base Position Hippocampus P valve Striatum P valve
darkmagenta paleturquoise 17 24,843,527 9.38 × 10
1.75 × 10
yellowgreen saddlebrown 17 33,901,252 2.31 × 10
3.34 × 10
skyblue3 skyblue 8 125,688,170 1.93 × 10
3.52 × 10
Orange steelblue 14 50,200,200 1.87 × 10
9.56 × 10
darkolivegreen - 7 108,611,544 2.28 × 10
White - 1 173,121,821 1.18 × 10
Park et al. BMC Systems Biology 2011, 5:43
Page 10 of 16
Page 10
the regulatory locus for this m odule was identical for
hippocampus and st riatum. A potential candidate for
this locus was flotillin 1 (Flot1),agenewithacis eQTL
in both hippocampus and striatum ~24 kb away from
this peak mark er. This gene product has been found to
accumulate in tangle-bearing neurons of Alzheimer s
disease [41] and may play a role in learning. In addition,
the f lotillin complex featured in the darkoliv egreen
module regulated by a hippocampal locus (above).
Other genes in these identified modules should be
examined as potential players in the molecular pathways
for fear conditioning.
Network edge orienting: prioritizing directed trait
To look for relatio nships between genetic variation, dif-
ferences in gene expression, and behavioral phenotypes,
we employed the Network Edge Orienting (NEO) [42]
algorithm. Using SNP markers as causal anchors, NEO
assigns directionality to trait networks and provides a
way to prioritize genes with expression profiles that are
coincident with quantitative behavioral phenotypes (Fig-
ure 6A).
We perfor med a NEO single ma rker analysis on mar-
kers with an FDR < 10% in the behavioral QTL map-
ping. The software uses structural equation modeling to
fit five models: causal, reactive, independent, and two
confounded models. NEO compares the best fitting
model relative to the next best fitting model, yielding a
likel ihood ratio, LEO.NB. AtoB, for each significant
SNP for each of the behavioral endpoints. Values greater
than 0.3 for this score indicate that the causal model fits
the input data twice a s well as the next best model; a
score of 1 indicates a ten-fold better fit. The measure
RMSEA.AtoB is an index of model fit, with values <
0.05 representing a good fit.
Figure 6B shows the results of NEO analysis in the
hippocampus. The results indicate that two SNP mar-
kers located on chromosome 7 regulate the expression
of two nea rby genes on chromosome 7 (630503K22RIK
and Rps15a) which in turn influence the immobility of
the animals before training (B11: Pre training immobility
Genetic variation at a SNP on chromosome 11 at
51.279205 was also shown to influence the expression of
the nearby gene kinesin-like protein 3A (Kif3a)which
then contributed to variation in thigmotaxis ( B33: Pre
cue thigm otaxis mean). Kif3a is a kinesin gene involved
in moving axon cargo [43] and has been implicated in
amyotrophic lateral sclerosis, a disease involving degen-
eration of motor neurons [44].
Variation at a SNP on chromosome 2 resulted in a
change in expression of the gene START domain-con-
taining 7 (Stard7) which then influenced immobility
induced by a novel context (B44 Context immobility).
The genes 6330503K22R IK and Kif3a a lso appear as
strong candidates for fear related behavior in the NEO
analysis for the striatum (Additional file 1 Figure S8),
underscoring the similarity of transcriptional regulation
in the two tissues.
Fear conditioning provides anopportunitytosurveya
range of clinically relevant processes including short and
long-term memory, co ntext general ization, and memo ry
extinction, making it an efficient tool with which to
probe the genetics of f ear dependent behavior. To map
fear related QTLs, we subjected a population of inbred
mouse strains to a stan dard fear conditi oning procedure
and follow-up memory t ests. We then combined beha-
vioral phenotype data with SNP genotypes and tissue
specific gene expression to search for candidate genes
and related networks associated with f ear phenotypes.
Across 48 behavioral endpoints, we mapped a total of
27 QTLs, highlighting the complexity of behavioral reg-
ulation and showcasing the value of HMDP for mapping
fear loci.
The inbred strains of the HMDP were not randomly
selected, but were, in fact, carefully chosen to avoid,
insofar as possible, high correlation of non-linked gen-
ome segments. Nevertheless, there are some shared seg-
ments across the genome due to bottlenecks in the
breeding and the history of the strains. EMMA endea-
vors to correct for these artifacts in the association ana-
lysis. However, some caution should be applied to the
interpretation of the mapping results, since bias may
remain which cannot be overcome by the analysis of the
The strongest behavioral QTL in our investigation was
for the phenotype cue immobility and had two peak
markers on chromosome 7. These markers were located
in the adjacent genes Tyr and Grm5 and had identical P
values of 4.4 × 10
, yet there were recombination
breakpoints betwee n them . Many HM DP strai ns have
mutations in Tyr and are albino, resulting in possibly
learning and memory deficits due to decreased visual
acuity. However, a study that examined this allele speci-
fically showed that it plays only a minor role in cue
immobility and that additional loci are likely to influ-
ence fear conditioning [16]. Grm5 is an attractive candi-
date gene for this locus, s ince it has previou sly been
shown to be involved in hippocampal LTP.
We surveyed the architecture of transcriptional regula-
tion across two brain regions. We found a smaller num-
ber of cis and trans eQTLs in the hippocampus than in
the striatum. This diminution may be caused by signal
dilution due to the heterogeneous cellular nature of the
hippocampus. However we found that the cis and trans
Park et al. BMC Systems Biology 2011, 5:43
Page 11 of 16
Page 11
eQTLs in the two tissues overlapped significantly, indi-
cating that DNA polymorphism has a robust effect in
modulating gene expression across tissues.
By simplifying the gene expression data int o modules,
we identified groups of genes that are related to fear
related behavior. Two such modules in the hippocampus
(CF1 and CF2) showed strong correlations with context-
dependent fear mea sures, allowing identification of n et-
works of genes whose co-expression co-varied with fear
phenotypes across the HMDP. We assig ned priorities to
genes within each module based on t heir level of intra-
modular connectivity and mapped loci responsible for
regulating MEs in b oth hippocampus and striatum.
Cued and context immobility were phenotypically simi-
lar as they clustered together in the behavioral dendro-
show strong correlations with cued fear, confirming sug-
gesting that the two different types of fear are expressed
through different neural and/or molecular pathways.
A hub gene in CF1 (Psmd6) and two of the most
highly connected genes in CF2 (Ube2a and Usp33)have
been shown to play roles in ubiquitination. Interestingly,
others have shown that ubiquitin-mediated proteolysis is
involved in initiating long-term stable memory, as both
specific removal o f s pecific inhibitory proteins and gene
induction are likely to be critical players in fear condi-
tioning [45]. Other components in these modules may
be implicated by association in these gen etic pathways
and provide attractive targets for further investigation.
Structural equation modeling allowed us to identify
single markers that influenced the expression of single
genes which in turn influence fear related phenotypes.
We identified fiv e genes with causal relationships for
fear-related phenotypes in the hippocampus and stria-
tum including 6330503K22RIK, Rps15a, Kif3a, Stard7,
and Plvap.
In summary, looking at expression patterns in genes and
groups of genes in various neural tissues has helped to
elucidate the complex molecular networks contributing
to fear dep endent behavior. While the current approach
6330503K22RIK Rps15a Stard7Kif3a
Behavioral QTL
Gene Expression QTL
2.65 0.5851.73.03
RefSeq Gene
Quantitative Behavioral
B33: Pre cue
thigmotaxis mean
B11: Pre training
immobility mean
B44: Context
immobility mean
B11: Pre training
immobility mean
Probe Position
SNP Marker Position
Figure 6 Selected causative genes in the hippocampus found using network edge orienting. A) Model fitted by NEO software implicates
a marker (DNA) as causal for a phenotypic trait through expression of a gene (RNA) B) Thresholds for FDR < 5% shown as red horizontal lines.
Vertical black lines indicate the start position of the gene. 6330503K22RIK, Rps15a, Kif3a, and Stard7 are genes with local markers that perturb
gene expression levels, which in turn contribute to fear phenotypes.
Park et al. BMC Systems Biology 2011, 5:43
Page 12 of 16
Page 12
yielded several potential loci and candidate genes, a ddi-
tional inbred strains would provide increased power for
more comprehensive mapping. Next generation seq uen-
cing technologies and proteomics should afford
even deeper views of genetic polymorphism and expres-
sion as we continue to refine gene networks of fear
Mouse population
Male mice from the M ouse Diversity Panel (HMDP)
were used for all behavioral analyses. This panel of mice
consists of 100 inbred strains comprised of 29 classical
inbred strains paired with three sets of RI strains
selected for diversity [9]. All mice (n = 700) were
obtained through Jackson Laboratory at approximately
55 days old then housed for a 14-day acclimation period
prior to testing. Mice were housed in groups (3-4 per
cage) under a 12hr/12hr day/night cycle with ad lib
access to food and water. All behavioral testing was con-
duc ted during the day portion of the cycle, between the
hours of 10 AM and 4 PM. Protocols conformed to
NIH Care and Use Guidelines and were approved
through the UCLA Animal Research Committee. Mice
were housed in their covere d home cages and placed in
an adjacent holding room. Auditory background stimu-
lus in the form of white noise (80db) was delivered
through overhead speakers. Previous unpublishe d obser-
vation showed no evidence of orienting response, or any
behavioral responses to stimulus presentation while in
the holding room [15].
Fear Conditioning
All HMDP strains were exposed to a fear c onditioning
procedure followed by two independent memory tests.
Parameters and procedures were identical to those pre-
viously described [15]. On each test day, mice were
wheeled to a holding room for a 30 min acclimation
period prior to test ing. Each mouse was tested individu-
ally and then transferred to a holding cage. On day 1,
mice were placed in a 25 cm × 20 cm conditioning
chamber with grid floors and white plexiglass. Following
a 3 minute exploration period, mice received three audi-
tory conditional stimu li (CS; 2000Hz, 15 seconds, 80
dB) co-terminating with footshock unconditional stimu-
lus (US; 0.75 mA, 1 second), delivered with an inter-trial
interval (ITI) of 1 minute. Mice were removed 2 min-
utes following the final US. On day 2, contextual fear
was assessed. Mice were then returned to the condition-
ing chamber under conditions identical to day 1.
Neither the CS nor US was presented during an 8 min-
ute test. On day 3, cued f ear was assessed following a
contextual shift. Mice w ere placed in a novel, rectangu-
lar activity chamber (50 cm × 25 cm), given a 3 minute
exploration period followed by a series of ten CS pre-
sentations (ITI 1 min), then removed from the chamber
1 minute following the final CS. No US were presented
during this test. This apparatus was cleaned with 70%
ethanol between tests.
Behavioral Data Analysis
Behavior was recorded digital ly from a camera mounted
above each test chamber, then digitized at 15 frames per
second with the EthoVision Pro tracking system (Noldus
Information Technology). For each mouse a total of 48
unique endpoints were quantified automatically with
EthoVision software (Additional file 1 Table S1). Vary-
ing numbers of biological replicates were ob tained for
each strain (ranging from n =3ton = 16, mean = 7.3) .
These measures were designed to charac terize multiple
dimensions of defensive behavior. The methodology and
rationale behind these measures has been discussed pre-
viously [15].
Mea n performance for each endpoint was d etermined
by either collapsing across the entire test session for
context fear endpoints or across specific test phases for
fear conditioning (pre-US, post-US) and cued fear test
(pre-CS, CS) endpoints. T he pre-US period consisted of
the 3 minutes prior to the initial CS presentation, while
the post-US period encompassed the 4.25 minute inter-
val between the first US presentati on and removal from
the chamber. Likewise, the pre-CS period spann ed the 3
minutes pr ior to CS presenta tion, and t he CS period
covered the 12.5 minute period between the first CS
presentation and removal from the chamber. Measures
reflecting rate c hanges were quantified by analyzing
time course data within individual test phases.
For the context test, endpoint rate changes were calcu-
epoch to the final 2 mi nute epoch. For mul ti-phas e tests
(training, cued fear test), rate changes were calculated as
suppression ratios based on mean values from the relevant
test phases (pre/(pre+post)). Strain means were calculated
and served as the behavioral phenotypes for downstream
analysis. Velocity is the mean rate of movement in any
given interval (e.g. cm/s), while mobility is the time spent
mobile, expressed as a percentage of total time.
Genotype analysis
The classical inbred and RI strains were genotyped pre-
viously [9] by the Broad Institute (classical) and the Well-
come Trust Center for Human G enetics (RI). The
imputed from the Wellcome Trust genotypes. Only SNPs
with a minor allele frequency greater than or equal to
10% were used in the analysis to minimize false positives
due to small samp le size. All genome coordinates are
based on NCBI build 35 (mm7) of the mouse genome.
Park et al. BMC Systems Biology 2011, 5:43
Page 13 of 16
Page 13
Behavioral QTL mapping
Using the collected behavioral phenotypes, we
performed a genome-wide associati on test using the
software package EMMA (Efficient Mixed-Model Asso-
ciation) [12]. This program calculates P values which
quantify the degree of association between each probe-
marker pair while correcting for confounding effects of
population structure and genetic relatedness between
strains in the panel. We used a genome-wide Q value
threshold of 5% [23] which corresponds to a P value
of 4.1 × 10
. To count the number of significant QTL,
markers were found in adja cent bins, markers w ere
combined and counted as a single QTL.
Tissue harvesting
Brains were removed from each animal after euthanasia.
Hippocampus and striatum were dissected out and flash
frozen in liquid nitrogen. RNA was extracted from each
sample using the Qiagen RNeasy kit.
Microarray data collection
Gene expression levels were quantified using Illumina
Mouse-Ref 8 v2.0 Expression BeadChip microarrays.
option in the software package BeadStud io (Illumina)
[46]. The microarray data are available at the Gene
Expression Omnibus (GEO) http://www.ncbi.nlm.nih.
gov/geo/ under accession number GSE26500.
Expression quantitative trait loci (eQTL) mapping
Using the marker genotype information from the
HMDP and RNA expression data from hippocampus
and striatum, we performed a genome-wide associati on
test for each of the 25,697 probes (genes) on the micro-
array compared to each of the 101,629 SNP markers
using the software package EMMA. Markers w ithin 2
Mb of the probe position for each gene were considered
cis (local), while those greater than 2 Mb from the
probe position were considered trans (distant). Genome-
wide significance thresholds were determined by calcu-
lating the P valuecorrespondingtoaBenjaminiand
Hochberg corrected FDR of 5% [23]. To count the num-
ber of significant trans loci, we divided the genome into
bins of 2 Mb in width and counted whether or not a
marker that s urpassed an FDR of 5% was observed in
the bin or not. If ad jacent bins contained at least one
significant marker, the bins were combined together and
counted as a single locus.
Gene ontology enrichment analysis
Groups of identified genes were checked for enrichment
in gene ontology categories using the package GOEAST
[24]. Significance wa s reported as Q values (P value cor-
rected false discovery rates [36]).
Identification of gene co-expression modules associated
with behavioral phenotypes
We used the R package WGCNA [47] to create gene co-
expression modules. The inp ut data consisted of gene
expression data from the hippocampus (n = 94) and the
striatum (n = 94). This program created modules or clus-
ters of highl y correlated genes in ea ch tissue separately.
For each of the modules, the program produced a mod-
ule eigengene (ME) which enabled us to find relation-
ships of modules with behavioral phenotypes.
Module preservation
We used the modulePreservation function from the
WGCNA library to calculate module preservation statis-
tics [31]. The Zsummary is derived fr om seven underly-
ing statistics t hat measure preservation of various
aspects of within-module network density and connec-
tivity patterns. The underlying preservation statistics are
based on permutation tests and their values represent
evidence that a module is signif icantly better preserved
between the referenc e and test netwo rks than a ran-
domly sampled group of genes of the same size. A
Zsummary < 2 indicates no evidence of module preser-
vation, 2 < Zsummary < 10 indicates weak to moderate
module preservation, and Zsummary > 10 indicates
strong preservation.
Network edge orienting
Markers surpassing a FDR threshold of 10% in the beha-
vioral QTL analysis along with gene expression data for
hippocampus and striatum were used as input to the
Network Edge Orienting (NEO) software package in R
[42]. We selected marker, gene, and phenotype combi-
nations that yielded a LEO, NB.AtoB score > 0.3 a nd
RMSEA.AtoB score < 0.05 for further analysis.
Additional material
Additional file 1: Supplementary Methods, Tables and Figures. The
Supplementary Methods describe further analyses of fear phenotypes in
the HMDP and gene regulation hotspots from the eQTL mapping.
Supplementary Tables are Table S1, Classification of quantified behavioral
phenotypes; Table S2, Top 100 cis eQTLs in hippocampus; Table S3, Top
100 cis eQTLs in striatum; Table S4, Gene co-expression modules; Table
S5, Functional classification for genes in context fear module 1; Table S6,
Functional classificat ion for genes in context fear module 2.
Supplementary Figures are Figure S1, Cluster dendrogram by behavioral
phenotype across HMDP; Figure S2, Mapped locus for cue immobility on
chromosome 7; Figure S3, QTL plots for 48 tested behavioral phenotypes
after EMMA correction for population structure; Figure S4, Hippocampus
eQTLs; Figure S5, Striatum eQTLs; Figure S6, Hippocampus module-trait
correlations; Figure S7, Striatum module-trait correlations; Figure S8,
Striatum NEO results.
Park et al. BMC Systems Biology 2011, 5:43
Page 14 of 16
Page 14
Additional file 2: Gene connectivity and module information. Table
provides details of gene co-expression network analyses for each tissue
and corresponding measures of intramodular connectivity for each gene.
We acknowledge funding from the National Institutes of Health RO1
Author details
Department of Molecular and Medical Pharmacology, David Geffen School
of Medicine, University of California, Los Angeles, CA 90095, USA.
Department of Medical Genetics and Rudolf Magnus Institute of
Neuroscience, UMC Utrecht, 3584 CG, Utrecht, The Netherlands.
Department of Medicine - Cardiology, David Geffen School of Medicine,
University of California, Los Angeles, CA 90095, USA.
Department of Human
Genetics, David Geffen School of Medicine, University of California, Los
Angeles, CA 90095, USA.
Department of Computer Science, University of
California, Los Angeles, CA 90095, USA.
University of California, Los Angeles,
Center for Neurobehavioral Genetics, David Geffen School of Medicine, CA
90095, USA.
Center for Public Health Genomics, School of Medicine,
University of Virginia, VA 22908, USA.
Authors contributions
CCP participated in the analysis of the expression and behavior data and
drafting of the manuscript. GDG participated in the analysis of the behavior
data and drafting of the manuscript. SdJ, AG, PL, AL, EE and SH participated
in the statistical analyses. BB, CRF and AK participated in sample collection
and analysis. AJL, RAO, EE, and SH participated in the design of the study.
DJS conceived of the study, and participated in its design and coordination.
All authors read and approved the final manuscript.
Received: 27 September 2010 Accepted: 16 March 2011
Published: 16 March 2011
1. Barrett GL, Reid CA, Tsafoulis C, Zhu W, Williams DA, Paolini AG, Trieu J,
Murphy M: Enhanced spatial memory and hippocampal long-term
potentiation in p75 neurotrophin receptor knockout mice. Hippocampus
2010, 20:145-152.
2. Peters M, Bletsch M, Catapano R, Zhang X, Tully T, Bourtchouladze R: RNA
interference in hippocampus demonstrates opposing roles for CREB and
PP1alpha in contextual and temporal long-term memory. Genes Brain
Behav 2009, 8:320-329.
3. Yalcin B, Willis-Owen SA, Fullerton J, Meesaq A, Deacon RM, Rawlins JN,
Copley RR, Morris AP, Flint J, Mott R: Genetic dissection of a behavioral
quantitative trait locus shows that Rgs2 modulates anxiety in mice. Nat
Genet 2004, 36:1197-1202.
4. Farber CR, van Nas A, Ghazalpour A, Aten JE, Doss S, Sos B, Schadt EE,
Ingram-Drake L, Davis RC, Horvath S, et al: An integrative genetics
approach to identify candidate genes regulating BMD: combining
linkage, gene expression, and association. J Bone Miner Res 2009,
5. Ghazalpour A, Doss S, Kang H, Farber C, Wen PZ, Brozell A, Castellanos R,
Eskin E, Smith DJ, Drake TA, et al: High-resolution mapping of gene
expression using association in an outbred mouse stock. PLoS Genet
2008, 4:e1000149.
6. Flint J, Corley R, DeFries JC, Fulker DW, Gray JA, Miller S, Collins AC: A
simple genetic basis for a complex psychological trait in laboratory
mice. Science 1995, 269:1432-1435.
7. Flint J: Analysis of quantitative trait loci that influence animal behavior. J
Neurobiol 2003, 54:46-77.
8. Grupe A, Germer S, Usuka J, Aud D, Belknap JK, Klein RF, Ahluwalia MK,
Higuchi R, Peltz G: In silico mapping of complex disease-related traits in
mice. Science 2001, 292:1915-1918.
9. Bennett BJ, Farber CR, Orozco L, Min Kang H, Ghazalpour A, Siemers N,
Neubauer M, Neuhaus I, Yordanova R, Guan B, et al: A high-resolution
association mapping panel for the dissection of complex traits in mice.
Genome Res 2010, 20:281-290.
10. Kirby A, Kang HM, Wade CM, Cotsapas CJ, Kostem E, Han B, Furlotte N,
Kang EY, Rivas M, Bogue MA, et al: Fine Mapping in 94 Inbred Mouse
Strains Using a High-density Haplotype Resource. Genetics 2010,
11. Broman KW: The genomes of recombinant inbred lines. Genetics 2005,
12. Kang HM, Zaitlen NA, Wade CM, Kirby A, Heckerman D, Daly MJ, Eskin E:
Efficient control of population structure in model organism association
mapping. Genetics 2008, 178:1709-1723.
13. Horvath S, Zhang B, Carlson M, Lu KV, Zhu S, Felciano RM, Laurance MF,
Zhao W, Qi S, Chen Z, et
al: Analysis of oncogenic signaling networks in
glioblastoma identifies ASPM as a molecular target. Proc Natl Acad Sci
USA 2006, 103:17402-17407.
14. Zhang B, Horvath S: A general framework for weighted gene co-
expression network analysis. Stat Appl Genet Mol Biol 2005, 4:Article17.
15. Gale GD, Yazdi RD, Khan AH, Lusis AJ, Davis RC, Smith DJ: A genome-wide
panel of congenic mice reveals widespread epistasis of behavior
quantitative trait loci. Mol Psychiatry 2009, 14:631-645.
16. Ponder CA, Huded CP, Munoz MB, Gulden FO, Gilliam TC, Palmer AA: Rapid
selection response for contextual fear conditioning in a cross between
C57BL/6J and A/J: behavioral, QTL and gene expression analysis. Behav
Genet 2008, 38:277-291.
17. Cohen RM, Kang A, Gulick C: Quantitative trait loci affecting the behavior
of A/J and CBA/J intercross mice in the elevated plus maze. Mamm
Genome 2001, 12:501-507.
18. Crawley JN, Belknap JK, Collins A, Crabbe JC, Frankel W, Henderson N,
Hitzemann RJ, Maxson SC, Miner LL, Silva AJ, et al: Behavioral phenotypes
of inbred mouse strains: implications and recommendations for
molecular studies. Psychopharmacology (Berl) 1997, 132 :107-124.
19. DeFries JC: Pleiotropic effects of albinism on open field behaviour in
mice. Nature 1969, 221:65-66.
20. Ayala JE, Chen Y, Banko JL, Sheffler DJ, Williams R, Telk AN, Watson NL,
Xiang Z, Zhang Y, Jones PJ, et al: mGluR5 positive allosteric modulators
facilitate both hippocampal LTP and LTD and enhance spatial learning.
Neuropsychopharmacology 2009, 34:2057-2071.
21. Xu J, Zhu Y, Contractor A, Heinemann SF: mGluR5 has a critical role in
inhibitory learning. J Neurosci 2009, 29:3676-3684.
22. Wang H, Westin L, Nong Y, Birnbaum S, Bendor J, Brismar H, Nestler E,
Aperia A, Flajolet M, Greengard P: Norbin is an endogenous regulator of
metabotropic glutamate receptor 5 signaling. Science 2009,
23. Benjamini Y, Hochberg Y: Controlling the false discovery rate - a practical
and powerful approach to multiple testing. J Royal Stat Soc, Series B 1995,
24. Zheng Q, Wang XJ: GOEAST: a web-based software toolkit for Gene
Ontology enrichment analysis. Nucleic Acids Res 2008, , 36 Web Server:
25. Huang GJ, Shifman S, Valdar W, Johannesson M, Yalcin B, Taylor MS,
Taylor JM, Mott R, Flint J: High resolution mapping of expression QTLs in
heterogeneous stock mice in multiple tissues. Genome Res
Peng J, Wang P, Tang H: Controlling for false positive findings of trans-
hubs in expression quantitative trait loci mapping. BMC Proc 2007,
1(Suppl 1):S157.
27. van Nas A, Ingram-Drake L, Sinsheimer JS, Wang SS, Schadt EE, Drake T,
Lusis AJ: Expression Quantitative Trait Loci: Replication, Tissue- and Sex-
Specificity in Mice. Genetics 2010, 185:1059-1068.
28. Ghazalpour A, Doss S, Zhang B, Wang S, Plaisier C, Castellanos R, Brozell A,
Schadt EE, Drake TA, Lusis AJ, et al: Integrating genetic and network analysis
to characterize genes related to mouse weight. PLoS Genet 2006, 2:e130.
29. Keller MP, Choi Y, Wang P, Davis DB, Rabaglia ME, Oler AT, Stapleton DS,
Argmann C, Schueler KL, Edwards S, et al: A gene expression network
model of type 2 diabetes links cell cycle regulation in islets with
diabetes susceptibility. Genome Res 2008, 18:706-716.
30. Oldham MC, Konopka G, Iwamoto K, Langfelder P, Kato T, Horvath S,
Geschwind DH: Functional organization of the transcriptome in human
brain. Nat Neurosci 2008, 11:1271-1282.
31. Langfelder P, Rui L, Oldham MC, Horvath S: Is my module network
preserved and reproducible? PLoS Comput Biol 2011, 7:e1001057.
32. Horvath S, Dong J: Geometric interpretation of gene coexpression
network analysis. PLoS Comput Biol 2008, 4:e1000117.
Park et al. BMC Systems Biology 2011, 5:43
Page 15 of 16
Page 15
33. Langfelder P, Horvath S: Eigengene networks for studying the
relationships between co-expression modules. BMC Syst Biol 2007, 1:54.
34. Kim JJ, Rison RA, Fanselow MS: Effects of amygdala, hippocampus, and
periaqueductal gray lesions on short- and long-term contextual fear.
Behav Neurosci 1993, 107:1093-1098.
35. Fanselow MS, LeDoux JE: Why we think plasticity underlying Pavlovian
fear conditioning occurs in the basolateral amygdala. Neuron 1999,
36. Benjamini Y, Yekutieli D: Quantitative trait Loci analysis using the false
discovery rate. Genetics 2005, 171:783-790.
37. Kelleher RJ, Govindarajan A, Tonegawa S: Translational regulatory
mechanisms in persistent forms of synaptic plasticity. Neuron 2004,
38. Lee SH, Choi JH, Lee N, Lee HR, Kim JI, Yu NK, Choi SL, Lee SH, Kim H,
Kaang BK: Synaptic protein degradation underlies destabilization of
retrieved fear memory. Science 2008, 319:1253-1256.
39. Fuller TF, Ghazalpour A, Aten JE, Drake TA, Lusis AJ, Horvath S: Weighted
gene coexpression network analysis strategies applied to mouse weight.
Mamm Genome 2007, 18:463-472.
40. Goddard CA, Butts DA, Shatz CJ: Regulation of CNS synapses by neuronal
MHC class I. Proc Natl Acad Sci USA 2007, 104:6828-6833.
41. Girardot N, Allinquant B, Langui D, Laquerriere A, Dubois B, Hauw JJ,
Duyckaerts C: Accumulation of flotillin-1 in tangle-bearing neurones of
Alzheimers disease. Neuropathol Appl Neurobiol 2003, 29:451-461.
42. Aten JE, Fuller TF, Lusis AJ, Horvath S: Using genetic markers to orient the
edges in quantitative trait networks: the NEO software. BMC Syst Biol
2008, 2:34.
43. Camargo LM, Collura V, Rain JC, Mizuguchi K, Hermjakob H, Kerrien S,
Bonnert TP, Whiting PJ, Brandon NJ: Disrupted in Schizophrenia 1
Interactome: evidence for the close connectivity of risk genes and a
potential synaptic basis for schizophrenia. Mol Psychiatry 2007, 12:74-86.
44. Pantelidou M, Zographos SE, Lederer CW, Kyriakides T, Pfaffl MW,
Santama N: Differential expression of molecular motors in the motor
cortex of sporadic ALS. Neurobiol Dis 2007, 26:577-589.
45. Chain DG, Schwartz JH, Hegde AN: Ubiquitin-mediated proteolysis in
learning and memory. Mol Neurobiol 1999, 20:125-142.
46. Kuhn K, Baker SC, Chudin E, Lieu MH, Oeser S, Bennett H, Rigault P,
Barker D, McDaniel TK, Chee MS: A novel, high-performance random array
platform for quantitative gene expression profiling.
Genome Res 2004,
47. Langfelder P, Horvath S: WGCNA: an R package for weighted correlation
network analysis. BMC Bioinformatics 2008, 9:559.
Cite this article as: Park et al.: Gene ne tworks associated with
conditional fear in mice identified using a systems genetics approach.
BMC Systems Biology 2011 5:43.
Submit your next manuscript to BioMed Central
and take full advantage of:
Convenient online submission
Thorough peer review
No space constraints or color figure charges
Immediate publication on acceptance
Inclusion in PubMed, CAS, Scopus and Google Scholar
Research which is freely available for redistribution
Submit your manuscript at
Park et al. BMC Systems Biology 2011, 5:43
Page 16 of 16
Page 16
  • Source
    • "Encouraged by this success, GWASs also have been proposed in mice using a recently developed panel of classic inbred and RI mouse strains, termed the Hybrid Mouse Diversity Panel (HMDP). Our prior studies with the HMDP have identified key associations for a variety of complex traits that are highly relevant to human diseases, which collectively illustrate the power of this approach for gene discovery in mice (Bennett et al., 2010Bennett et al., , 2012 Farber et al., 2011; Ghazalpour et al., 2011 Ghazalpour et al., , 2012 Ghazalpour et al., , 2014 Park et al., 2011; Orozco et al., 2012; Davis et al., 2013; Parks et al., 2013; Hartiala et al., 2014). In the present study, we used this mouse platform to carry out a comprehensive genetic screen for HSPC frequency in the adult BM. "
    [Show abstract] [Hide abstract] ABSTRACT: Prior efforts to identify regulators of hematopoietic stem cell physiology have relied mainly on candidate gene approaches with genetically modified mice. Here we used a genome-wide association study (GWAS) strategy with the hybrid mouse diversity panel to identify the genetic determinants of hematopoietic stem/progenitor cell (HSPC) frequency. Among 108 strains, we observed ∼120- to 300-fold variation in three HSPC populations. A GWAS analysis identified several loci that were significantly associated with HSPC frequency, including a locus on chromosome 5 harboring the homeodomain-only protein gene (Hopx). Hopx previously had been implicated in cardiac development but was not known to influence HSPC biology. Analysis of the HSPC pool in Hopx(-/-) mice demonstrated significantly reduced cell frequencies and impaired engraftment in competitive repopulation assays, thus providing functional validation of this positional candidate gene. These results demonstrate the power of GWAS in mice to identify genetic determinants of the hematopoietic system. Copyright © 2015 The Authors. Published by Elsevier Inc. All rights reserved.
    Full-text · Article · Jun 2015 · Stem Cell Reports
  • Source
    • "An extensive overview of WGCNA, including numerous tutorials, can be found at http://labs.genetics.ucla. edu/horvath/CoexpressionNetwork/Rpackages/WGCNA/ and this method has been extensively used to create coexpression networks232425262728. To begin, we filtered the array data to include 8173 probes expressed in the liver as previously described [29]. "
    [Show abstract] [Hide abstract] ABSTRACT: Background Atherosclerosis, the underlying cause of cardiovascular disease, results from both genetic and environmental factors. Methods In the current study we take a systems-based approach using weighted gene co-expression analysis to identify a candidate pathway of genes related to atherosclerosis. Bioinformatic analyses are performed to identify candidate genes and interactions and several novel genes are characterized using in-vitro studies. Results We identify 1 coexpression module associated with innominate artery atherosclerosis that is also enriched for inflammatory and macrophage gene signatures. Using a series of bioinformatics analysis, we further prioritize the genes in this pathway and identify Cd44 as a critical mediator of the atherosclerosis. We validate our predictions generated by the network analysis using Cd44 knockout mice. Conclusion These results indicate that alterations in Cd44 expression mediate inflammation through a complex transcriptional network involving a number of previously uncharacterized genes.
    Full-text · Article · Aug 2014 · BMC Medical Genomics
  • Source
    • "Finally candidate genes associated with the trait are prioritised based on network statistics like module membership and gene significance. WGCNA has been used to identify genes and gene networks associated with specific tissues, distinct biological states or diseases, and qualitative as well as quantitative phenotypes [22–24]. "
    [Show abstract] [Hide abstract] ABSTRACT: Background Salmonella enterica serovar Typhimurium is a gram-negative bacterium that can colonise the gut of humans and several species of food producing farm animals to cause enteric or septicaemic salmonellosis. While many studies have looked into the host genetic response to Salmonella infection, relatively few have used correlation of shedding traits with gene expression patterns to identify genes whose variable expression among different individuals may be associated with differences in Salmonella clearance and resistance. Here, we aimed to identify porcine genes and gene co-expression networks that differentiate distinct responses to Salmonella challenge with respect to faecal Salmonella shedding. Results Peripheral blood transcriptome profiles from 16 pigs belonging to extremes of the trait of faecal Salmonella shedding counts recorded up to 20 days post-inoculation (low shedders (LS), n = 8; persistent shedders (PS), n = 8) were generated using RNA-sequencing from samples collected just before (day 0) and two days after (day 2) Salmonella inoculation. Weighted gene co-expression network analysis (WGCNA) of day 0 samples identified four modules of co-expressed genes significantly correlated with Salmonella shedding counts upon future challenge. Two of those modules consisted largely of innate immunity related genes, many of which were significantly up-regulated at day 2 post-inoculation. The connectivity at both days and the mean gene-wise expression levels at day 0 of the genes within these modules were higher in networks constructed using LS samples alone than those using PS alone. Genes within these modules include those previously reported to be involved in Salmonella resistance such as SLC11A1 (formerly NRAMP1), TLR4, CD14 and CCR1 and those for which an association with Salmonella is novel, for example, SIGLEC5, IGSF6 and TNFSF13B. Conclusions Our analysis integrates gene co-expression network analysis, gene-trait correlations and differential expression to provide new candidate regulators of Salmonella shedding in pigs. The comparatively higher expression (also confirmed in an independent dataset) and the significantly higher connectivity of genes within the Salmonella shedding associated modules in LS compared to PS even before Salmonella challenge may be factors that contribute to the decreased faecal Salmonella shedding observed in LS following challenge. Electronic supplementary material The online version of this article (doi:10.1186/1471-2164-15-452) contains supplementary material, which is available to authorized users.
    Full-text · Article · Jun 2014 · BMC Genomics
Show more