Inflammatory gene regulatory networks in amnion cells following cytokine stimulation: translational systems approach to modeling human parturition.

Abstract

A majority of the studies examining the molecular regulation of human labor have been conducted using single gene approaches. While the technology to produce multi-dimensional datasets is readily available, the means for facile analysis of such data are limited. The objective of this study was to develop a systems approach to infer regulatory mechanisms governing global gene expression in cytokine-challenged cells in vitro, and to apply these methods to predict gene regulatory networks (GRNs) in intrauterine tissues during term parturition. To this end, microarray analysis was applied to human amnion mesenchymal cells (AMCs) stimulated with interleukin-1β, and differentially expressed transcripts were subjected to hierarchical clustering, temporal expression profiling, and motif enrichment analysis, from which a GRN was constructed. These methods were then applied to fetal membrane specimens collected in the absence or presence of spontaneous term labor. Analysis of cytokine-responsive genes in AMCs revealed a sterile immune response signature, with promoters enriched in response elements for several inflammation-associated transcription factors. In comparison to the fetal membrane dataset, there were 34 genes commonly upregulated, many of which were part of an acute inflammation gene expression signature. Binding motifs for nuclear factor-κB were prominent in the gene interaction and regulatory networks for both datasets; however, we found little evidence to support the utilization of pathogen-associated molecular pattern (PAMP) signaling. The tissue specimens were also enriched for transcripts governed by hypoxia-inducible factor. The approach presented here provides an uncomplicated means to infer global relationships among gene clusters involved in cellular responses to labor-associated signals.

Full text

Page 1

Inflammatory Gene Regulatory Networks in Amnion Cells
Following Cytokine Stimulation: Translational Systems
Approach to Modeling Human Parturition
Ruth Li1., William E. Ackerman, IV1., Taryn L. Summerfield1, Lianbo Yu3, Parul Gulati3, Jie Zhang4,
Kun Huang4, Roberto Romero5,6, Douglas A. Kniss1,2*
1Division of Maternal-Fetal Medicine and Laboratory of Perinatal Research, Department of Obstetrics and Gynecology, The Ohio State University, Columbus, Ohio, United
States of America, 2Department of Biomedical Engineering, The Ohio State University, Columbus, Ohio, United States of America, 3Center for Biostatistics, The Ohio State
University, Columbus, Ohio, United States of America, 4Department of Biomedical Informatics, The Ohio State University, Columbus, Ohio, United States of America,
5Perinatology Research Branch, Intramural Division, Eunice Kennedy Shriver National Institute of Child Health and Human Development, National Institutes of Health,
Department of Health and Human Services, Bethesda, Maryland, United States of America, 6Hutzel Women’s Hospital, Detroit, Michigan, United States of America
Abstract
A majority of the studies examining the molecular regulation of human labor have been conducted using single gene
approaches. While the technology to produce multi-dimensional datasets is readily available, the means for facile analysis of
such data are limited. The objective of this study was to develop a systems approach to infer regulatory mechanisms
governing global gene expression in cytokine-challenged cells in vitro, and to apply these methods to predict gene
regulatory networks (GRNs) in intrauterine tissues during term parturition. To this end, microarray analysis was applied to
human amnion mesenchymal cells (AMCs) stimulated with interleukin-1b, and differentially expressed transcripts were
subjected to hierarchical clustering, temporal expression profiling, and motif enrichment analysis, from which a GRN was
constructed. These methods were then applied to fetal membrane specimens collected in the absence or presence of
spontaneous term labor. Analysis of cytokine-responsive genes in AMCs revealed a sterile immune response signature, with
promoters enriched in response elements for several inflammation-associated transcription factors. In comparison to the
fetal membrane dataset, there were 34 genes commonly upregulated, many of which were part of an acute inflammation
gene expression signature. Binding motifs for nuclear factor-kB were prominent in the gene interaction and regulatory
networks for both datasets; however, we found little evidence to support the utilization of pathogen-associated molecular
pattern (PAMP) signaling. The tissue specimens were also enriched for transcripts governed by hypoxia-inducible factor. The
approach presented here provides an uncomplicated means to infer global relationships among gene clusters involved in
cellular responses to labor-associated signals.
Citation: Li R, Ackerman WE IV, Summerfield TL, Yu L, Gulati P, et al. (2011) Inflammatory Gene Regulatory Networks in Amnion Cells Following Cytokine
Stimulation: Translational Systems Approach to Modeling Human Parturition. PLoS ONE 6(6): e20560. doi:10.1371/journal.pone.0020560
Editor: Peter Csermely, Semmelweis University, Hungary
Received January 21, 2011; Accepted May 5, 2011; Published June 2, 2011
Copyright: ? 2011 Li et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted
use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This work was supported in part by subcontract WSU07043 under National Institute of Child Health and Human Development Contract N01-HD-2-3342
and Perinatology Research Branch, Division of Intramural Research, Eunice Kennedy Shriver National Institute of Child Health and Human Development, NIH,
DHHS. Additional support was provided by NIH K08 HD049628 (WEA) and The Ohio State University Perinatal Research and Development Fund. The funders had
no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing Interests: The authors have declared that no competing interests exist.
* E-mail: kniss.1@osu.edu
. These authors contributed equally to this work.
Introduction
The last three decades have seen prodigious growth in our
knowledge of the basic principles of parturition in women. Indeed,
we now understand that human labor in both the term and
preterm settings is fundamentally an inflammatory process typified
by the sudden, robust expression of pro-inflammatory cytokines
(IL-1, IL-6, TNF-a) and chemokines (IL-8/CXCL-8, MCP-1/
CCL-2) that elicit a panoply of downstream biological conse-
quences culminating in expulsion of a viable offspring [1–3].
Matrix metalloproteinases (MMPs) are released locally by
infiltrating leukocytes (neutrophils and monocytes/macrophages)
that modify the extracellular milieu within the cervix and at the
maternal-fetal interface [4]. Furthermore, cytokines secreted into
the intrauterine microenvironment through the induction of
prostaglandin synthase-2 (PTGS2) provoke the precipitous pro-
duction of prostaglandins (PGE2and PGF2a) that then stimulate
myometrial contractility and cervical changes that must take place
in order to deliver the fetus [2,5–7]. A collection of coordinately
regulated genes (contraction-associated proteins, CAPS), including
the oxytocin receptor (OXTR), connexin-43 (CX-43), the
prostaglandin F receptor (FP), and PTGS2 (also known as
cyclooxygenase-2, COX-2) are dramatically increased during the
early stages of labor [8].
Some of the molecular mechanisms that regulate these genes
have been investigated, and we now understand that the
inflammatory response during parturition is controlled, at least
in part, by the master transcription factor nuclear factor-kappaB
(NF-kB) [9,10]. For example, we and others have shown that
cytokines elicit the up-regulation of (PTGS2) for the synthesis of
PLoS ONE | www.plosone.org1 June 2011 | Volume 6 | Issue 6 | e20560

Page 2

PGE2and PGF2athrough the activation of heterodimeric (p65/
p50) NF-kB and its inhibitor IkB-a in a variety of in vitro culture
preparations of intrauterine cell types [11–15]. Furthermore, the
genes encoding many cytokines (e.g., IL-1b and TNF-a),
chemokines (e.g., IL-8) and MMPs (e.g., MMP-2, MMP-9) are
themselves driven from NF-kB-binding promoters [16,17]. Most
of this information has been obtained by studying one or a few
biomolecules in any given body of work.
Experimental reductionism has yielded many important and
interesting insights into how a given gene (or gene product) might
participate in mammalian parturition. Indeed, today we know far
more about the underlying cellular and molecular participants that
mediate uterine contraction and the biochemical events of cervical
dilatation and effacement in humans than we did just a few years
ago. The endocrine signaling of parturition that takes place via the
hypothalamic-pituitary-adrenal-placental axis in the sheep [18],
and the role of the corpus luteum in sustaining rodent pregnancy
and the consequences of its regression on birth [8] are now
understood in significant detail. Yet, collectively, this information
has not substantively improved our ability to prevent, diagnose or
arrest preterm labor in women to any significant degree. Nor do
we yet fully appreciate the molecular nuances that distinguish term
and preterm labor and birth.
A reductionist approach permits the intricacies of a given
biochemical process to be unraveled, but the inherent quest for
simplicity precludes an assessment of potential complex interac-
tions that may exist between several molecular pathways.
Increasingly, networks of interacting pathways are evaluated by
examining a myriad of individual genes expressed under a given
set of experimental or clinical contexts. In this regard, Romero
and colleagues recently reported that mRNAs isolated from cells of
the fetal membranes in women in spontaneous parturition at term
manifest a characteristic ‘‘inflammatory gene signature’’ when
compared to a cohort of women prior to the onset of active clinical
signs of labor [19]. A similar pattern of gene expression was
unveiled by this group in the uterine cervix in association with
labor [20]. These studies offer the intriguing opportunity to
investigate the complex molecular interactions that occur as the
uterus and cervix transition from quiescence to active labor.
However, we know clearly that the onset of labor is not a binary
switch that is suddenly unleashed; rather, it is a series of subtle
biochemical and physiological epochs that arise in the last several
weeks of gestation [21–23]. Therefore, the inability to assess this
protracted time-course in women due to ethical realities renders this
transcriptomic approach in women scientifically very challenging.
Thus,inthe presentwork,wehaveassembledareductioniststrategy
using monolayer cultures of human amnion mesenchymal cells as a
means to model the inflammatory gene expression signature
previously reported by Haddad et al. [19]. Using a systems
approach, we have examined the complex interactions that occur
between networks of genes expressed preferentially following
cytokine challenge in amnion cells. In addition, we have used
computational methods to investigate the control of many of these
genes by transcription factor binding motif analysis. The results
from thissimple cell culture preparation when compared to the data
reported by Romero and colleagues offer a novel translational
strategy for a more thorough molecular appreciation of human
parturition in the term and preterm settings.
Methods
Cell Culture
Primary cultures of human amnion mesenchymal cells were
prepared from amnion membranes obtained prior to labor at
term, as previously described [12]. Cells were cultured in DMEM
(25 mM glucose) supplemented with 10% fetal bovine serum,
2 mM L-glutamine, 1 mM sodium pyruvate, 50 mg/ml of
gentamicin sulfate, and 0.5 mg/ml of amphotericin B at 37uC in
a 5% CO2atmosphere. Cell culture media, supplements and sera
were purchased from Invitrogen (Carlsbad, CA).
Treatment and RNA Extraction
Recombinant human interleukin-1b (IL-1b) was obtained from
R&D Systems (Minneapolis, MN). Early passage cells at
confluence were treated with IL-1b (10 ng/ml) for 1 or 8 h,
whereas control cells were treated with medium only. Total RNA
from control and experimental groups were extracted using Trizol
reagent (Invitrogen, Carlsbad, CA) and the RNeasy Mini Kit
(Qiagen, Valencia, CA) as previously described [12,24]. Experi-
ments were repeated three times for a total of nine samples used
for microarray analysis. Three biological replicates were used to
ensure that any results were not biased by day-to-day variation in
RNA extraction.
Microarray Analysis and Data Processing
Quantification and quality of total RNA samples were evaluated
with the 2100 Bioanalyzer (Agilent Technologies, Foster City,
CA). Three sets of RNA samples including vehicle control, 1 h and
8 h post IL-1b treatment were hybridized to GeneChip Human
Genome U133A 2.0 Array (Affymetrix, Inc., Santa Clara, CA). All
arrays were processed using the GeneChip System (Affymetrix,
Inc., Santa Clara, CA) at the OSU Comprehensive Cancer Center
Microarray Shared Resource facility following the manufacturer’s
protocol. The R statistics package version 2.7.1 [25] was employed
for microarray data analysis. First, background correction and
quartile normalization were performed to remove technical bias,
and gene expression levels were summarized over probe set using
the Robust Multichip Average (RMA) method [26]. These data
have been deposited in NCBI’s Gene Expression Omnibus [27]
and are accessible through GEO Series accession number
GSE26315 (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc
=GSE26315). A filtering method based on percentage (7 out of 9
chips) of arrays below noise cutoff (log2scale of 6 expression level)
was applied to filter out probes whose expressions are not
detectable. Analysis of variance (ANOVA) modeling was per-
formed and t-test was used to detect differentially expressed genes.
To improve estimates of variability and statistical tests for
differential expression, variance shrinkage methods were employed
for this study [28]. The significance level was adjusted by
controlling the expected number of false positives for all
comparisons [29]. After these statistical analyses, significant genes
were filtered from the three sets of RNA samples using the criteria
of a 62.0 fold change and a p-value of #0.05 for comparisons
among the vehicle, 1 h and 8 h time points. A final list of genes
was generated by eliminating duplicate probe sets and averaging
the expression values of genes with multiple probe sets. The final
lists of genes were subjected to subsequent clustering, time series,
and transcription factor binding motif enrichment analyses as
described in their respective sections below. For functional
pathway and network analysis, the fold change (62.0) and
p-value (#0.05) cut off, and probeset averaging for genes were
completed directly with the Ingenuity Pathway Software. A
heatmap representing the gene lists was generated with hierarchi-
cal clustering (Euclidean distance and complete linkage) using the
MultiExperiment Viewer (MeV) software, part of the TM4
microarray software suite [30]. Short Time-series Expression
Miner (STEM) [31] was used to identify temporal expression
profiles, ranked by the number of genes fitted to each profile.
A Systems Approach to Modeling Human Parturition
PLoS ONE | www.plosone.org2June 2011 | Volume 6 | Issue 6 | e20560

Page 3

Transcription Factor Binding Motif Enrichment Analysis
NCBI reference sequence mRNA accession numbers for each
unique, significantly expressed probeset from our microarray
analysis, as well as the published full-thickness fetal membrane
data [19], were curated by conversion from Entrez Gene IDs and/
or Human Genome Organisation (HUGO) gene symbols using
g:Profiler (http://biit.cs.ut.ee/gprofiler/) [32]. Multiple probesets
mapping to identical genes were consolidated based on unique
gene symbol. Genes with multiple reference mRNA sequences
were resolved by using the accession number for the longest or
most frequently appearing transcript. The RefSeq mRNA
accession numbers were then subjected to transcription factor
binding motif analysis using the web-based software Pscan (http://
159.149.109.9/pscan/) [33]. The JASPAR [34] database of
transcription binding factor sequences was utilized for the analysis
on position 2950 to +50 of the 59 upstream promoters. The range
of 2950 to +50 was selected out of the available range options in
Pscan to best cover the range of interest of 21000 to +50 bp.
Hypothetical proteins, pseudogenes, and expressed sequence tags
lacking reference sequences were not included in the transcription
factor binding motif analysis. Similar conditions and exclusions
were used to perform promoter scanning analysis with the dataset
of Haddad et al. [19]. MATLAB 2009 (MathWorks, Inc., Natick,
MA) was used to generate a colormap showing the frequency of
transcription factor motif occurrences, while Graphviz Editor
version 2.26.3 [35] was used to generate a gene regulatory network
map based on results of the transcription factor binding motif
analysis.
Functional and Pathway Analyses
Network pathway analyses were generated using Ingenuity
Pathways Analysis (version 8.5, Ingenuity Systems, www.ingenuity.
com). A full list of the normalized, filtered data set from the 9
microarray chips were uploaded into Ingenuity for analysis. For
the Haddad et al. dataset, the microarray data taken from the
published data were uploaded into Ingenuity for analysis. The
genes from each dataset were set as molecules of interest which
interact with other molecules in the Ingenuity Knowledge Base
(identified as ‘‘Network Eligible molecules’’). For network
generation, the molecules from the normalized, filtered microarray
dataset were each mapped to their corresponding object in
Ingenuity’s Knowledge Base. A fold change cutoff of $2 (log2-fold
change cutoff of 1.0) was set to identify molecules whose
expression was significantly differentially regulated. Multiple
probesets that map to the same gene were averaged (preliminary
analyses of individual probesets compared to multiple probesets
representing the same gene demonstrated the validity of this
approach). Network eligible molecules were mapped onto a global
molecular network developed from information contained in
Ingenuity’s Knowledge Base. Networks of Network Eligible
Molecules were then algorithmically generated based on logical
connectivity. In the graphical representation of networks,
molecules are represented as nodes (filled shapes), and the
biological relationship between two nodes is represented as an
edge (line). All edges are supported by at least 1 reference from the
literature, from a textbook, or from canonical information stored
in the Ingenuity Pathways Knowledge Base. Human, mouse, and
rat orthologs of a gene are stored as separate objects in the
Ingenuity Pathways Knowledge Base, but are represented as a
single node in the network. The intensity of the node color
indicates the degree of up- or down-regulation (red and green,
respectively). Nodes are displayed using various shapes that
represent the functional class of the gene product. Canonical
pathways analysis identified the pathways from the Ingenuity
Pathways Analysis library of canonical pathways that were most
significant to the data set. The significance of the association
between the data set and the canonical pathway was measured in 2
ways: (1) A ratio of the number of molecules from the data set that
map to the pathway divided by the total number of molecules that
map to the canonical pathway is displayed; and (2) the Benjamini-
Hochberg method used to focus on the most significant biological
functions associated with the dataset by adjusting calculated p-
values that determine the probability that the association between
the genes in the dataset and the canonical pathway is explained by
chance alone.
Biological functional analysis identified the biological functions
and/or diseases that were most significant to the dataset.
Molecules from the dataset that met the fold change cutoff of 2
and were associated with biological functions and/or diseases in
Ingenuity’s Knowledge Base were considered for the analysis.
Benjamini Hochberg method was used to calculate a p-value
determining the probability that each biological function and/or
disease assigned to that data set is due to chance alone.
Quantitative Real Time PCR Analysis
Reverse transcription was carried out with 2 mg of total RNA
using oligo primers and Superscript III reverse transcriptase
(Invitrogen, Carlsbad, CA). TaqMan Universal PCR Master Mix
and TaqMan Gene Expression Assays (Applied Biosystems, Foster
City, CA) were used for quantitative real time PCR (qRT-PCR)
withb-actin(ACTB) as a control. The followingAppliedBiosystems
(ABI) primer/probe sets were used: Nfkbia, Hs00153283_m1; Rela,
Hs00153294_m1; Il1b, Hs01555413_m1; Il6, Hs00985641_m1;
Il8, Hs00174103_m1; Cxcl2, Hs00601975_m1; Ptgs2, Hs00-
153133_m1; and Actb, 4333762F. Quantitative RT-PCR was
performed on a 7300 Real-Time PCR System (Applied Biosystems,
Foster City, CA) using recommended cycling conditions and
associated SDS software, and subsequent analysis with the
comparative CTmethod [36] was conducted using Microsoft Excel.
Plots were constructed and statistical analysis was completed using
GraphPad Prism software (La Jolla, CA). Statistical analyses were
performed using the Kruskal-Wallis statistical test with post-hoc
testing using Dunn’s multiple comparison test when appropriate,
after it was determined that gene amplification data exhibited a
non-Gaussian distribution.
Results
Temporal transcription profiling of cytokine-stimulated
amnion mesenchymal cells in vitro
Interleukins-1a and -1b (IL-1a and IL-1b) are encoded by
separate genes, but share nearly identical biological activities. The
pro-inflammatory response to either is elicited upon activation of a
common interleukin-1 receptor (IL-1R), a heterodimeric complex
comprised of the IL1R1 and Interleukin-1 receptor accessory
protein (IL1RAP) gene products [37,38]. The constituent IL-1
receptor proteins are themselves members of the IL-1R/Toll-like
receptor (TLR) superfamily [37,38]. As a result of conserved
intracellular domains among these receptors, the signaling
cascades elicited by IL-1a or IL-1b are generally similar to those
activated by TLR ligation by viral and bacterial pathogens, with
some exceptions [38,39]. Therefore, the genes upregulated in
response to IL-1b have broad implications in the context of innate
host defense and, importantly, can mimic the initial response to
bacterial pathogens in a microbe-free environment [40].
In the current experiments amnion mesenchymal cells (AMCs)
were chosen as a relevant in vitro model system in which to study
the inflammatory component of labor. We have previously
A Systems Approach to Modeling Human Parturition
PLoS ONE | www.plosone.org3 June 2011 | Volume 6 | Issue 6 | e20560

Page 4

reported that basal release of PGE2in AMCs is 5-fold higher than
induced release of PGE2from amnion epithelial cells [12]. Our
laboratory previously used this model to examine the induction of
selected gene targets, such as prostaglandin E synthase (Ptges, also
known as microsomal prostaglandin E synthase-1, mPGES-1) and
prostaglandin-endoperoxide synthase 2 (Ptgs2), in response to
IL-1b [12]. In addition, AMCs are a non-transformed cell model,
which eliminates from consideration the strong affect of oncogenes
and other factors that might influence the global transcriptional
response [41,42]. It was also reported that the epithelial-to-
mesenchymal cell ratio was 4.3 to 1 at preterm (15 cases at 23 – 36
weeks) and 7.8 to 1 at term (27 cases) [43], suggesting a higher
population of mesenchymal cells in the setting of preterm labor.
To expand upon the prior published results [12], we analyzed
the expression of 14,500 genes in AMCs in three biological
replicates at 1 and 8 h following IL-1b challenge (10 ng/ml) using
transcriptional profiling. After background correction, normaliza-
tion, and filtering of genes with low expression, a 2-fold or greater
increase (p,0.05) in transcripts was observed for 137 probe sets
(representing 108 unique genes) in AMCs after 1 h of IL-1b
stimulation, while 153 probe sets (representing 125 unique genes)
were up-regulated by 2-fold or greater after 8 h of cytokine
exposure, as summarized in the volcano plots shown in Figure1A.
Compared with the 1 h time point, 48 probe sets (representing 38
unique genes) were significantly (p,0.05) upregulated at 8 h, while
48 probe sets (representing 39 unique genes) were downregulated.
The full list of 190 unique probe sets with their average log2
expression values at each time point is presented in Table S1.
The normalized AMC microarray data have been deposited in
NCBI’s Gene Expression Omnibus [27] and are accessible
through GEO Series accession number GSE26315 (http://www.
ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE26315).
The 190 unique genes with statistical significance (p,0.05), $2
fold-changes in at least one of the pairs of time point comparisons
were subjected to hierarchical clustering (Figure 1B). There were
three major groups into which the genes clustered, each with
subgroups bearing differing temporal expression profiles. Genes
that exhibited the most dramatic levels of induction following
IL-1b challenge included those encoding several cytokines (Il1a,
Il6, Tnf, and Csf2) and chemokines (Cxcl1, Cxcl2, Cxcl3, Ccl2, and
Il8), the adhesion molecule Vcam1, transcription regulatory
proteins (Irf1, TnfaipP3, Nfkbia, and Zc3h12a), the prostaglandin
synthesizing enzyme Ptgs2, a stress-activated protein kinase
(Nuak2), and an apoptotic suppressor protein that interacts with
the TNFR2-TRAF signaling complex (Birc3). The basal levels of
expression (represented as the log2in Figure 1B) varied widely,
with some genes exhibiting low levels of basal expression
(including Bcl2a1, Birc3, Ccl5, Csf2, Cxcl3, Icam4, Tnf, and Vcam1),
and others exhibiting relatively high levels of basal expression (e.g.,
Cebpd, Cxcl6, Ier3, Il8, Nfkbia, Plau, and Rgs2).
Due to dramatic differences in the absolute expression
intensities, hierarchical cluster analysis per se was insufficient to
group co-expressed (and presumably, co-regulated) genes based
upon relative levels of induction. To evaluate subgroups of genes
with similar temporal expression patterns, the 190 unique genes
selected above were subjected to clustering using the Short Time-
series Expression Miner (STEM) software [31]. Of the 80% of
genes that were mapped to temporal expression profiles, five major
patterns were discerned (Figure 1C): genes that were induced at
1 h for which expression remained elevated (Group A); genes that
were induced at 1 h then gradually repressed (Group B); genes for
which expression steadily increased throughout treatment (Group
C); genes exhibiting delayed induction (Group D); and genes that
were rapidly induced, then repressed quickly (Group E). A
complete list of the genes mapping to these profiles is provided
in Table S2.
The genes that fit the Group A expression profile were
significantly enriched for gene ontology (GO) terms such as
‘‘leukocyte chemotaxis/migration’’ and ‘‘inflammatory response’’.
Among these genes were a number of highly-upregulated
cytokines and chemokines (Ccl2, Ccl20, Csf1, Cxcl1, Cxcl2, Cxcl3,
Cxcr7, Il1a, Il1b, Il6, and Il8), as well as genes involved in NF-kB
activation and its regulation (Nfkb1, Nfkbie, Tnfaip2, Tnfaip3,
Tnfaip8, Tnfrsf9, and Traf1). The Group B profile was enriched
for the GO term ‘‘regulation of gene expression’’, and included
several genes encoding transcriptional regulators (Ets1, Fosl1, Jun,
Maff, Nfkbia, Nr4a2, Phlda1, Rel, Twist1, and Zfp36l1). Numerous
transcription factors (Bhlhb2, Egr1, Egr2, Egr3, Fos, Fosb, Ier2, Junb,
and Klf6) were also present among the genes of Group E, for which
the GO term ‘‘regulation of RNA metabolic processes’’ was
overrepresented. The Group C profile was enriched for GO terms
including ‘‘cell adhesion’’ and ‘‘integral to plasma membrane’’,
and encoded proteins involved in cell adhesion (Icam4, Cldn1),
integral membrane receptor proteins (Dram1, Nrp2, Olr1, Rarres1),
and membrane transport proteins (Slc12a7, Slc2a6, Pdpn). For
Group D genes, the most highly-ranked GO term was ‘‘response
to virus’’; interestingly, several interferon (IFN)-inducible genes
were present in this profile (such as myxovirus resistance genes
[Mx1, Mx2], the gene encoding guanylate binding protein (Gbp1),
the single-stranded RNA exonuclease Isg20, a retinoic acid-
inducible gene I [Rig-I]-like receptor involved in the host detection
of double-stranded RNA viruses [Ifih1], and genes of unknown
function [Ifi35, Ifi44, Ifit3]), in addition to some cytokines/
chemokines exhibiting delayed induction (Ccl5, Cxcl5, Cxcl6, and
Tnfsf18). Insofar as the expression of IFN genes (Ifna isoforms,
Ifnb1, Ifng, and Ifnw1) themselves remained very low throughout
the course of treatment (average log2expression values between
2.3 and 4.1, data not shown), it was unexpected that IFN-regulated
genes would be upregulated in response to the inciting stimulus of
IL-1b [44,45]. Moreover, it was interesting to note that in addition
to the Mx pathway, components of other IFN-induced antiviral
proteins (e.g., OAS3, of 29-59-oligoadenylate synthetase system)
also had statistically significant induction, but fell short of the
chosen fold-change threshold [46–48].
To verify the results of the microarray analysis, we next
examined the expression of seven transcripts (Rela, Nfkbia, Cxcl2,
Il1b, Il6, Il8, and Ptgs2) by qRT-PCR using total RNA collected
from replicate experiments using IL-1b treatment conditions
identical to those used for the transcription profiling studies. This
select mRNA subset was chosen for the frequent high expression of
these genes and their cognate proteins in in vitro models of
inflammation-mediated preterm labor (including the model used
in this work) and in vivo in biological fluids or tissue specimens
collected from women in preterm labor. Consistent with the
STEM profile analysis, we found that Nfkbia exhibited maximal
induction following 1 h of stimulation, with decreased expression
thereafter (Figure 2A). Of four genes that mapped to the Group A
STEM profile (i.e., genes with elevated expression at both 1 and
8 h), only one (Cxcl2) followed the predicted temporal course when
analyzed using qRT-PCR (Figure 2B); the remainder (Il1b, Il6,
Il8, and Ptgs2) exhibited modest upregulation at 1 h, followed by
markedly elevated expression at 8 h (Figure 2, C–F). These
results suggest that, as has been noted in prior studies, microarray
analysis tends to underestimate true fold changes in transcript
expression [49]. In keeping with this observation, we found that
Rela expression was moderately increased at 8 h following cytokine
challenge by qRT-PCR (Figure 2G), but no change was detected
by transcriptional profiling.
A Systems Approach to Modeling Human Parturition
PLoS ONE | www.plosone.org4 June 2011 | Volume 6 | Issue 6 | e20560

Page 5

Collectively, these results confirm that signaling through the
IL-1 receptor is biased toward an innate immune response in
cultured AMCs. While the inflammatory response became
heterogeneous over time due to upregulation of several auto-
crine/paracrine factors and their receptors, we observed that none
of the TLRs surveyed by the microarray platform (Tlr1 through
Tlr8) exhibited pronounced levels of expression (average log2
expression values between 2.6 and 5.0, data not shown), and
therefore were less likely to have contributed substantively to the
observed transcriptional response, even under non-sterile condi-
tions. Nevertheless, the IL-1b-induced gene expression changes in
AMCs bore some striking similarities to the transcriptional
Figure 1. Amnion mesenchymal cell (AMC) microarray data. (A) Volcano plot showing the criteria (62.0 no-log fold change, ,0.05 p-value)
for the filtered (blue points) and unfiltered (black points) AMC microarray data. (B) Hierarchical clustering heatmap of discriminant genes from
analysis of AMC microarray data (a complete list of the discriminant genes is found in Table S1). (C) Top significant profiles from temporal expression
analysis of the AMC data; the bottom graph shows the average fold change for each profile. Details of the genes that mapped to each temporal
profile are found in Table S2.
doi:10.1371/journal.pone.0020560.g001
A Systems Approach to Modeling Human Parturition
PLoS ONE | www.plosone.org5June 2011 | Volume 6 | Issue 6 | e20560