Professional Documents
Culture Documents
Gene Expression Profiling in Sepsis: Timing, Tissue, and Translational Considerations
Gene Expression Profiling in Sepsis: Timing, Tissue, and Translational Considerations
Sepsis is a complex inflammatory response to infection. Sepsis, one of the most prevalent diseases in the ICU, is
Microarray-based gene expression studies of sepsis a clinical syndrome that is characterized by a systemic
have illuminated the complex pathogen recognition inflammatory response to infection, typically bacterial in
and inflammatory signaling pathways that characterize origin. It is defined as documented or suspected infection in
sepsis. More recently, gene expression profiling has the setting of a subset of four findings that describe the
been used to identify diagnostic and prognostic gene systemic inflammatory response syndrome (SIRS) [4], and
signatures, as well as novel therapeutic targets. Studies it can progress rapidly, resulting in organ failure (severe
in pediatric cohorts suggest that transcriptionally dis- sepsis) or impaired tissue perfusion (septic shock). Sepsis
tinct subclasses might account for some of the hetero- syndromes are both common and dangerous, with the
geneity seen in sepsis. Time series analyses have incidence increasing in both adults and children, and
pointed to rapid and dynamic shifts in transcription mortality rates as high as 10–50% depending on age and
patterns associated with various phases of sepsis. disease severity [5,6].
These findings highlight current challenges in sepsis The genetic influences on the pathogenesis of acute
knowledge translation, including the need to adapt conditions such as sepsis are often under-appreciated,
complex and time-consuming whole-genome methods but they are striking. In adoptee studies, death from
for use in the intensive care unit environment, where infection has been shown to be fivefold more heritable than
rapid diagnosis and treatment are essential. death from cancer [7]. The innate immune response that
accounts for the physiologic derangements of bacterial
Sepsis and its genomic influences infection is associated with altered expression of more
Since its introduction in the late 1990s, microarray-based than 3,700 genes [8], making gene expression analysis a
gene expression profiling has had a significant impact on potentially useful tool for discovery-oriented studies of the
the field of medicine. In cancer biology, molecular subtypes pathogenesis of sepsis and severe infection. Published
of diseases have been identified [1], as well as transcrip- findings based on this research paradigm are increasing;
tional signatures predicting clinical outcome [2] and furthermore, expression data are accumulating in publicly
response to specific therapies [3]. Useful biomarkers have available repositories, and a few active clinical trials
been found that in some cases can obviate the need for include a gene expression component (Figure 1).
genome-wide approaches, enabling the translation of gene
expression research into clinical practice. Nonetheless, the Goals and challenges of transcriptome research in
impact of genome science remains far from pervasive, sepsis
especially in the intensive care unit (ICU), where diseases Gene expression analysis of sepsis is distinguished from
evolve rapidly, resulting in systemic illness, organ failure, analyses of cancer and chronic diseases in a number of ways,
and high mortality. both conceptual and pragmatic (Figure 2). First, sepsis
investigators must make decisions about which tissues to
Corresponding author: Wong, H.R. ([email protected]). sample and at what time points, which in the absence of
Keywords: sepsis; septic shock; gene expression; microarrays; genomics; bioinfor- additional clinical data or a priori hypotheses, can be arbi-
matics. trary in nature. Tissue from the source of infection can be
1471-4914/$ – see front matter difficult to sample directly because biopsies are seldom
ß 2014 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.molmed.2014.01.006
practical in the critically ill. As an easily accessible compart-
ment of the immune system, whole blood and its various
leukocyte fractions have therefore been the source tissue of
interest in most gene expression studies of sepsis. The
204 Trends in Molecular Medicine, April 2014, Vol. 20, No. 4
Review Trends in Molecular Medicine April 2014, Vol. 20, No. 4
300
200
Key:
Clinical trials
2
4
Counts
6
8
Microarray samples
100
Figure 1. Trends in gene expression profiling of sepsis. The bars represent the number of PubMed citations per year for the search term ‘gene expression AND sepsis’. The
trend line shows the number of individual microarray assays added each year to a publicly available repository of gene expression data (ArrayExpress). The sizes of the
circles at the bottom of the plot reflect the number of clinical trials initiated in each year, as identified by a trials registry (ClinicalTrials.gov, search terms ‘gene expression
AND sepsis OR septic shock OR severe sepsis’).
findings from gene expression profiling of blood cells might pathophysiology of acute illness. Samples collected repeat-
not accurately reflect expression patterns from immune edly over the course of an illness episode should therefore
cells resident in other tissues, such as alveolar macrophages ideally be analyzed together rather than in isolation in an
or splenic lymphocytes, although the significance of this attempt to describe illness trajectory and differentiate
potential discordance remains uncertain [9,10]. responses to treatment. Unlike with trauma, the precise
Second, sepsis is a dynamic process within a relatively onset of sepsis can be difficult to pinpoint accurately, intro-
narrow time period. Thus, whereas genomic changes can ducing further complexity when comparing time course gene
occur in tumors over weeks or months, large-scale tran- expression profiles from different patients with sepsis.
scriptional shifts in leukocytes have been shown to occur Third, whereas gold standard diagnostic labels can be
within just a few hours of an inflammatory stimulus [8,11]. arrived at for most tumors on the basis of anatomic and
In the setting of blunt trauma, a condition with considerable molecular pathology findings, the diagnosis of sepsis is
inflammatory features that is often complicated by sepsis, predominantly a clinical one. Moreover, the criteria on
the leukocyte transcriptome is substantially altered to upre- which the diagnosis is based lack specificity, with more than
gulate inflammatory and pathogen recognition pathways 40% of cases having negative bacterial cultures [13]. The
in the days and weeks after injury [12]. These investiga- absence of reliable classification complicates statistical ana-
tions into the functional trajectory of cellular processes lyses of gene expression data that use supervised methods to
constitute a unique method by which to model the dynamic detect differences in expression between groups.
205
Review Trends in Molecular Medicine April 2014, Vol. 20, No. 4
(A)
Sepsis episode
T1 T2 T3 T4 ... Tn
(B)
WB PMNs PBMCs
Gene expression levels
(C)
A B
Gene list
ALOX5
IL8
MMP9
IFN
MAPK1
…
Clinical phenotypes
(D)
CCL3
Barcode Sequence Count
CCL3
AATGCGTACAG 132
HSPA1B HSPA1B
IL8
TGCAGCCGAG 1,290
IL8
GCCCGTAGCA 32,199
ELA2 ELA2
Figure 2. Overview of gene expression profiling in sepsis. (A) Because sepsis syndromes are characterized by rapid shifts in gene expression over hours and days, blood
samples can be collected for analysis at various time points. Multiple samples taken over the course of resuscitation, stabilization, and convalescence can be used to
generate time series of gene expression. (B) After samples are collected, RNA can be extracted either directly from whole blood or from different leukocyte fractions. RNA
transcript levels are derived from gene expression microarrays. (C) Bioinformatics pathways can be used to compare gene expression profiles between two or more groups
of patients (e.g., sepsis and non-infectious systemic inflammatory response syndrome), resulting in a list of differentially expressed genes, and their associated pathways.
Unsupervised machine learning methods, including partitional clustering algorithms, can be used to identify previously unrecognized sepsis subclasses. Expression data
from multiple time points can be analyzed together to generate expression trajectories, which may differ between patients. The interpretation of differences in gene
expression is facilitated through comparison with clinical phenotypes derived from patient data collected from electronic medical records or patient registries, or in the
context of a clinical trial. (D) Unlike with diseases managed in the outpatient setting, the treatment of sepsis relies on diagnostic testing that can rapidly return easily
interpreted results. High-dimensional gene expression data must therefore be ‘downsized’ to more easily derived and understood signals. Strategies include using serum
biomarker assays to develop patient classifiers, generating gene expression mosaics that visually represent complex expression signals, and deploying sophisticated
multiplex assays that measure a limited number of transcripts using molecular barcoding technology. Abbreviations: PBMCs, peripheral blood mononuclear cells; PMNs,
polymorphonuclear neutrophils; WB, whole blood.
Last, whole-genome approaches to sepsis research pathologic specimens, targeted genotyping, or other com-
must include strategies for ‘downsizing’ the methods used, plex and time-consuming assays, these techniques are
from high-dimensional, resource- and time-intensive gene impractical in the management of sepsis. Useful assays
expression assays to rapid, cost-effective diagnostic tests must reflect the rapidly evolving, dynamic nature of sepsis,
that can be deployed at the point of care. Whereas findings the need for quick information in the acute resuscitative
from gene expression studies in cancer can translate to phases of this condition, and the necessity that individual
clinical practice via immunohistochemical staining of samples be analyzed at any time of day or night, with short
206
Review Trends in Molecular Medicine April 2014, Vol. 20, No. 4
turnaround time, and without waiting to be batched with Unlike with tightly controlled experiments in healthy
other specimens. subjects, clinical studies of gene expression in sepsis must
confront substantially more uncertainty, heterogeneity,
Experimental designs and complexity in the inflammatory manifestations of
As a primarily immunological phenomenon, sepsis is often infection.
studied by examining leukocytes, including ex vivo immu- Expression patterns are modulated by a number of
nostimulation experiments. Patients with sepsis syndromes factors, such as age, gender, and ethnicity, the presence
manifest various dynamic shifts in leukocyte populations, of comorbidities, the timing of inflammatory stimulus, and
often transitioning between states of leukocytosis and leu- the patient’s state of immune activation at the time of
kodepletion, and exhibiting differences in the relative abun- infection. This is reflected in the results of clinical studies
dance of granulocytes, lymphocytes, and specialized subsets of gene expression in adults with sepsis syndromes. A
thereof. Each of these cellular subtypes exhibits a distinct systematic review of a dozen such studies suggests nearly
gene expression pattern tailored to the specialized function universal upregulation of the pathogen recognition and
of the respective cell type [14], with expression profiles from signal transduction pathways identified in controlled
whole blood representing a weighted sum of these patterns. endotoxin experiments, but it provides a far more mixed
Because different cellular compartments of the blood per- picture when it comes to the pro- and anti-inflammatory
form different immunological functions in response to infec- pathways mediated by genes that govern lymphocyte
tion, leukocyte gene expression in sepsis is cell-type specific. differentiation, antigen presentation, and cytokine
The set of genes that distinguishes sepsis from non-infec- expression [20]. Disagreement between studies in this
tious inflammation in the neutrophils of the innate immune regard might reflect differences in patient population, in
system demonstrates little overlap with the similarly that trauma patients who are otherwise healthy might be
focused set of genes identified from the lymphocytes of predisposed to immunostimulation following injury,
the adaptive immune system [15,16], whereas patients with primary sepsis exhibit a greater
The use of whole blood to derive gene expression data in degree of immunosuppression [18,21].
sepsis has the pragmatic advantage of straightforward Gene expression studies have been used to identify novel
sample collection, minimal preprocessing, and limited therapeutic targets in sepsis on the basis of molecular
induction of expression artifacts related to isolation of pathways that are differentially expressed between cases
leukocyte subsets [17]. Most clinical studies in both adults and controls, or between survivors and non-survivors. Com-
and children have used either whole blood or peripheral plementing findings from basic science research, animal
blood mononuclear cells (PBMCs). In the case of whole studies, and clinical trials in humans [22], gene expression
blood studies, statistical methods have been used to studies have highlighted the importance of zinc homeostasis
account for the relative abundance of each leukocyte sub- in immune functioning, particularly among children with
type in the sample and to attribute gene expression find- sepsis syndromes [23]. Children admitted to the ICU with
ings to specific populations of cells [18]. septic shock have been shown to exhibit diminished expres-
sion of numerous genes that influence or rely on zinc home-
Insights into molecular mechanisms of sepsis ostasis. Evidence suggests that this pattern is more evident
Initial studies of gene expression in sepsis were largely in a certain subset of patients and is associated with poor
exploratory in nature, aiming to describe the complex outcomes [24]. Clinical studies of zinc supplementation have
immunologic and inflammatory pathways that character- demonstrated salutary effects on the incidence and severity
ize this condition. Early insights came from studies of of certain infections in both children and the elderly [25–27],
healthy volunteers exposed to bacterial endotoxin but in both adult and pediatric ICU patients, larger rando-
[8,11,19], which revealed significant changes in the tran- mized trials of mixed micronutrient supplementation that
scription of more than 3,700 genes as soon as 2 hours after included zinc showed no significant effects on the incidence
endotoxin exposure. Early after endotoxin exposure, of infection [28,29].
pathogen recognition cell surface receptors, including One particular family of zinc-related proteins has been
those from the Toll-like receptor (TLR) family, are up- shown to be consistently overexpressed in sepsis and septic
regulated, along with various proinflammatory cytokines shock. The matrix metalloproteinases (MMPs) are a series
and chemokines, such as tumor necrosis factor (TNF), of proteases that degrade extracellular matrix, inflamma-
interleukin-1a (IL-1a), IL-1b, CXC chemokine ligand 1 tory cytokines, and other proteins, thereby mediating a
(CXCL1), CXCL2, monocyte chemotactic protein 1 (MCP- variety of immunologic and neoplastic processes [30,31].
1), and IL-8 [8,11]. These changes are accompanied by the Gene expression studies, along with confirmatory serum
activation of signal transduction pathways including the assays, have shown that MMP-8 and MMP-9 are upregu-
nuclear factor-kb (NF-kb), mitogen-activated protein lated in injury and sepsis, correlating in some cases with
kinase (MAPK), Janus kinase (JAK), and signaling trans- disease severity [11,32–34]. Consistent with these clinical
ducer and activator of transcription (STAT) pathways. In observations, MMP-8-null mice and wild-type mice treated
parallel, signaling to restrain the immune response is with a pharmacological inhibitor of MMP-8 have a survival
increased, both by the upregulation of suppressor of cyto- advantage when subjected to a model of sepsis [32].
kine signaling genes [e.g., suppressor of cytokine signaling
3 (SOCS3)] and the downregulation of cytokine expression Influence of demographic factors
itself. Expression patterns return to baseline within Although patient age, gender, and ethnicity have been
24 hours of endotoxin exposure. accounted for in traditional clinical and basic science
207
Review Trends in Molecular Medicine April 2014, Vol. 20, No. 4
research in sepsis, the importance of these demographic including Gram-positive bacteria, Gram-negative bacteria,
factors in gene expression studies has yet to be fully and fungi, can be clinically indistinguishable, and both the
explored. Nonetheless, there is reason to believe that yield and lag time of microbial cultures limit their use in
demographic features exert considerable effects on gene practice [4,13]. Because the molecular pathways under-
expression patterns in sepsis. Ethnic background is known pinning the cellular immune response to these various
to be a strong determinant of gene expression in general types of infection have distinctive features, gene expres-
[35], and there is some evidence to suggest its influence on sion profiling has been investigated as a means to identify
sepsis in particular. In one small study examining gene culprit organisms in patients with sepsis.
expression patterns among critically ill patients with ven- Early studies in animal models and ex vivo models of
tilator-associated pneumonia (VAP), far more genes varied human cell types suggest that regardless of the invading
between ethnic groups (Caucasian or African-American) pathogen, a core group of co-expressed genes is upregu-
than between groups sorted by age or gender [36]. Epide- lated in the face of infection, constituting a so-called ‘com-
miological studies suggest differences in sepsis incidence mon host response’ to sepsis [43,44]. This common
and outcomes among patients with different ethnic back- response, which is expressed in a variety of cell types,
grounds, although the extent to which these findings includes the activation of inflammatory mediators and
reflect genomic differences is not known [37]. signal transduction pathways, as well as negative feedback
Concurrent with evolving immune function, both early pathways and apoptotic pathways that put infected cells in
in development and later in life, different genetic responses a state of ‘high alert’, whereby programmed cell death can
to sepsis are seen among various patient age groups. In one be initiated in the event of progressive infection [43].
analysis combining results from five separate studies, Whereas targeted experiments suggest that isolated path-
researchers identified differences in gene expression ways coordinate the immune response to Gram-positive
among pediatric patients with septic shock. Full-term and Gram-negative infections, microarray experiments
neonates (28 days) were shown to have gene expression suggest considerable overlap between these types of infec-
patterns distinct from those of infants (1 month to 1 year), tion. Both TLR2 (which is associated with the transcrip-
toddlers (2–5 years), and school-aged children (6 years) tional response to Gram-positive infections) and TLR4
when assessed within 24 hours of ICU admission [38]. (which binds lipopolysaccharide from Gram-negative bac-
Importantly, and in contrast to older children and adults, teria) initiate signals that culminate in the common host
these results suggest that neonates not only failed to response [43]. This result is reflected in clinical studies of
mount a robust inflammatory response but also demon- gene expression in sepsis, which for the most part have
strated downregulation of antigen presentation and NF-kb shown few differences in expression patterns between
pathway genes, and an overall decrement in immune patients infected with different types of bacteria [36,45].
response to infection. Reduced expression of triggering In the pediatric population, gene expression data have
receptor expressed on myeloid cells 1 (TREM1) pathway- been used to distinguish bacterial infections from viral
related genes was also seen, suggesting that neonates infections such as influenza A [46]. Not surprisingly,
might have limited capacity to amplify immune signals patients with influenza were identifiable on the basis of
related to pathogen recognition and might be less respon- increased expression of interferon (IFN) pathways. Up to
sive to novel therapies targeting this pathway in septic one-third of patients with bacterial sepsis also demon-
shock [39,40]. This study included a relatively small num- strated upregulation of IFN genes, suggesting the possi-
ber of neonates (n = 17), and the findings would be bol- bility of a preceding or concurrent viral infection in these
stered by validation in a dedicated prospective study. cases. Differences were also found between patients with
At the other end of the age spectrum, there are fewer Gram-positive infections and those with Gram-negative
whole-genome data about the effects of ageing on immune infections. Beyond its focus on pediatric rather than adult
functioning in sepsis. In one mouse study examining indi- patients, this study differed from those that found no
vidual cytokine levels using a cecal ligation and puncture differentially expressed genes between Gram-positive
(CLP) model of sepsis, older mice showed more pronounced and Gram-negative sepsis. Instead of using neutrophils,
local and systemic inflammatory responses compared to the investigators generated gene expression profiles from
younger mice with similar survival rates [41]. Although PBMCs, which, because of their pleomorphism, might be
gender is known to influence gene expression patterns in expected to better reflect pathogen differences. These find-
other conditions such as ischemic heart diseases (albeit ings highlight the importance of considering tissue type
modestly) [42], sex-specific differences in gene expression when designing and interpreting gene expression studies
in sepsis remain largely unexplored. Complex interactions in sepsis.
between demographic factors are also likely to influence In addition to determining what type of infection has
gene expression in sepsis. triggered a sepsis response, it might be equally important
to determine the nature of the response mounted by an
Gene expression and sepsis subtypes individual patient. The existence of heretofore unrecog-
One of the most clinically relevant questions following the nized sepsis subtypes is suggested by the heterogeneity
diagnosis of sepsis is that of which invading pathogen is of physiologic and molecular phenotypes encompassed
responsible for the acute infection. Proper knowledge of the under the non-specific clinical definition of sepsis. This
inciting cause is useful in selecting appropriate antimicro- inadvertent case mixing in clinical studies leads to indis-
bial agents and identifying uncontrolled sources of infec- tinguishable survival curves, overlapping histograms of
tion. Sepsis arising from different types of organisms, measured outcomes, and a deficit of actionable evidence.
208
Review Trends in Molecular Medicine April 2014, Vol. 20, No. 4
The discovery of sepsis subtypes has thus been identified biomarkers that distinguish patient subsets of interest.
by some as a key goal in sepsis research [47,48]. Typically, this involves examining genes with the highest
This problem is inherently one of unsupervised machine fold-change difference in expression between patients with
learning, in which patients are grouped according to simi- sepsis and those with non-infectious SIRS. In one pediatric
larities across multiple dimensions of gene expression data study, gene expression profiling was used to identify class-
rather than clinical labels assigned by investigators. Var- predictor genes distinguishing SIRS with negative bacter-
ious cluster identification algorithms have been used for ial cultures from sepsis with positive bacterial cultures
this purpose. In one such study, pediatric patients with [51]. The most predictive gene was Epstein–Barr virus-
septic shock were partitioned into distinct gene expression induced gene 3 (EBI3), which encodes a subunit of IL-27;
clusters on the basis of the expression levels of genes that high levels of IL-27 in the serum had high specificity and
differentiated septic patients from a group of non-septic positive predictive value (>90%) for the diagnosis of sepsis.
controls [24]. Hierarchical clustering was used with an a However, a comparable study in adults showed that serum
priori decision to designate the second-order branch points IL-27 levels were relatively less specific for sepsis, and IL-
as distinct clusters. This approach resulted in 3 subclasses 27 was outperformed as a biomarker by procalcitonin,
of septic shock (subclasses A, B, and C), with nearly 7,000 which is already used in clinical practice [52].
genes differentially expressed between them. Clinical phe- Biomarkers identified by gene expression profiling have
notyping of the subclasses after clustering showed signifi- also been used to stratify patients with sepsis according to
cant differences in important clinical outcomes, with mortality risk. The IL8 gene was shown to be upregulated
patients in subclass A having more severe manifestations in nonsurvivors of pediatric septic shock, with elevated
of sepsis, a greater degree of organ injury, and higher serum levels of IL-8 significantly increased as compared to
mortality. Patients in this subclass also tended to be survivors and controls [23]. Again this biomarker was
younger, with a median age of 3.6 months, compared to shown to have less predictive value in adults with sepsis
4.3 years for subclass B and 2.0 years for subclass C. From [53]. Although there were methodological differences
a genomic standpoint, subclass A was characterized by a between the pediatric and adult studies in the case of both
relative downregulation of adaptive immune pathways IL-27 and IL-8, these findings again underscore the influ-
and glucocorticoid receptor signaling. In a separate, multi- ence of age in immune functioning, and the importance of
center validation study, 82 patients from an independent considering age in the design and analysis of gene expres-
cohort were grouped according to their level of expression sion studies in sepsis.
of the top 100 class-defining genes identified in the first In other microarray studies, CX3C chemokine receptor 1
study. Patients from subclass A again showed a greater (CX3CR1; fractalkine receptor) was shown to be upregu-
severity of illness, a trend towards higher mortality, and lated in sepsis survivors compared to non-survivors [54],
younger median age [49]. and serum levels of its ligand, CX3CL1, were found to be
Although no prospective studies in adults have been increased in patients with sepsis, as compared to healthy
dedicated to sepsis subtype discovery, a post hoc analysis of controls [55]. CC chemokine ligand 4 (CCL4; also known as
gene expression profiles generated in the course of other MIP-1b) has also been identified as a potential biomarker
investigations suggests their existence. Using separate on the basis of differential gene expression and was shown
derivation and validation cohorts, patients were clustered to have a high negative predictive value for mortality in
using a partitioning around medioids (PAM) algorithm pediatric septic shock [56].
based on the expression levels of a subset of genes identi- Although variously sensitive or specific in certain popu-
fied in the literature as being meaningful in sepsis and lations, these single-marker diagnostic strategies do not
septic shock [50]. The existence of two distinct clusters was have well-rounded performance characteristics that would
best supported by the data, and although no significant justify their broad use in clinical practice [57]. Moreover,
clinical differences were found between subtypes, there the use of individual biomarkers in isolation in many ways
were important differences in the expression of genes defeats the purpose of using high-throughput technologies
involved in inflammatory and pathogen recognition path- such as gene expression microarrays to study multifaceted,
ways, as well as key pharmacogenes involved in the meta- complex conditions such as sepsis. Because biomarkers are
bolism of drugs used commonly in sepsis. traditionally identified by knowledge-based approaches
predicated on known biological functions and pathways,
‘Downsizing’ genome-wide data for clinical use such searches tend to leave unexamined scores of other
Multidimensional gene expression platforms that sample proteins that might be useful alone or in combination, but
thousands of genes at once are ideally suited for discovery- whose biological function in sepsis has yet to be fully
oriented tasks in sepsis research. Their use in clinical elucidated [58]. One strategy to overcome this bias is to
practice, however, is limited by a number of practical con- leverage the considerable coverage of gene expression
siderations. To be useful in the evaluation and treatment of microarrays to identify candidate biomarkers that seem
patients with sepsis, assays must be widely available, rapid, to be promising on the basis of statistical rather than
repeatable, consistent, and inexpensive. As such, there is a biological features.
need to ‘downsize’ the high-dimensional data generated by In the Pediatric Sepsis Biomarker Risk Model (PERSE-
gene expression microarrays to more targeted signals that VERE) project, investigators used such an approach to
can be produced closer to the point of care. derive and validate a panel of serum biomarkers to assign
The most reductive approach to this problem is to use a mortality probability in pediatric sepsis [57–59]. Genes
gene expression microarray data to identify individual differentially expressed between sepsis survivors, sepsis
209
Review Trends in Molecular Medicine April 2014, Vol. 20, No. 4
non-survivors, and healthy controls were identified by findings from the leukocyte fractions of 167 trauma
analysis of variance (ANOVA) and further refined by post patients, researchers derived an expression signature of
hoc pairwise comparisons to identify 137 candidate bio- 63 genes that varied most between patients with uncom-
marker genes. These were cross-referenced with a list of plicated, intermediate, and complicated clinical courses
4,397 candidate genes identified using support vector following trauma [34]. To create an assay that could rea-
machine (SVM) classifiers based on sepsis mortality and sonably be used in clinical practice, leukocytes were iso-
further reduced by selecting genes whose protein products lated by means of red cell lysis in microfluidics chambers,
could be easily measured in the blood [58]. The final panel and samples were analyzed using the NanoString platform
of 12 biomarkers was subsequently used to derive a clas- to evaluate expression of the 63 signature genes. These
sifier based on classification and regression tree (CART) results were further downsized to a single expression
analysis and validated in a separate patient cohort. The metric, the difference from reference value (DFR), based
model had adequate sensitivity (91% in the derivation on a summation of expression differences for each gene,
cohort, 89% in the validation cohort), but lacked specificity from age-, gender-, and ethnicity-matched controls.
(86% in the derivation cohort, 64% in the validation Results were generated within 8 to 12 hours, showed good
cohort), and had an area under the receiver operating agreement with microarray expression values, and per-
characteristic (ROC) curve of 0.759 in the validation cohort formed better than both microarray-derived DFR and
[59]. Recently, prospective validation of an updated model conventional clinical severity of illness scores.
yielded an ROC of 0.811 [60], and an analogous model was
derived and tested in adult populations [61]. Temporality of gene expression in sepsis
The shortcomings in gene expression-derived biomarker Most clinical studies of gene expression in sepsis have been
performance are likely to reflect a complex interplay of based on the analysis of a single time point in the illness
individual genes and transcripts in sepsis, to say nothing of trajectory or on the comparison of an early time point with
the post-translational modifications and protein interac- a later one. However, genomic shifts in response to inflam-
tions that exert influence on function. Cellular signals in mation are known to occur rapidly, as seen in clinical
sepsis are dynamic, changing rapidly with inflammatory studies of trauma patients [12] or experimental studies
conditions and an evolving immunologic milieu. The goal of of healthy subjects exposed to endotoxin [8,11]. Unlike in
reducing genome-wide signals to individual biomarkers, or these cases, the time of onset of the inflammatory stimulus
even groups of biomarkers used in combination, might in sepsis cannot be accurately known, resulting in consid-
therefore be difficult to achieve. Some investigators have erable uncertainty regarding the timing of sampling with
instead focused on developing gene signatures that com- respect to the ebb and flow of the immunological response.
bine signals from dozens or hundreds of genes to be used as Many studies include protocols to collect samples for gene
a filter in identifying patients with sepsis from those with expression analysis within 24 hours of admission to the
non-infectious SIRS, and in predicting outcomes of sepsis ICU; however, patients are admitted to the ICU at various
syndromes. In one such study, a 138-gene signature dis- stages of sepsis and can transition from one stage to the
tinguished sepsis from SIRS with sensitivity and specifi- next even within the first 24 hours of their stay. The extent
city in the validation cohort of 81% and 79%, respectively to which gene expression is being compared across similar
[16]. genomic, molecular, and pathophysiologic epochs in these
Translating genome-wide assays for clinical use studies is thus uncertain.
involves developing tools that can be deployed in the The importance of timing in sepsis gene expression
unpredictable and dynamic clinical environment of the analysis is illustrated by one recent study of five pediatric
emergency department or ICU. New methods of data patients with severe sepsis and septic shock due to menin-
representation are needed to convey key signals contained gococcal meningitis [65]. In this work, expression levels of
within high-dimensional data to front-line clinicians both key genes differed between patients at various time points.
quickly and unambiguously. Along these lines, investiga- Further, the overall contour of expression trajectories of
tors have tested the use of novel data visualization meth- key genes across the entire 48-hour period also differed.
ods such as ‘gene expression mosaics’ that convey high- These differential trajectories were seen for some genes
dimensional gene expression data in 2D colored patterns that have been investigated as biomarkers in sepsis, sug-
[62]. Expression mosaics have been developed in the study gesting that certain biomarkers might be more useful in
of pediatric septic shock subclasses and have been shown to some patients than in others, and might be more useful at
be useful in both computer-based and clinician-based inter- certain stages of illness than at others.
pretation of expression patterns [49,63]. Among clinicians A number of statistical methods have been developed to
without specific training in the interpretation of these pat- analyze time course gene expression data. Approaches
terns, gene expression mosaics were sorted according to include Markov models [66–68], ANOVA [69], and the
sepsis subclass (A, B, or C) with overall k value for agree- use of cubic splines to model changing expression levels
ment of 0.81 [63]. over time [70]. Time course gene expression data from
In an effort to develop more scalable solutions for gene trauma and burn patients have been used to develop
expression analysis in acute care, investigators have also statistical methods for the analysis of leukocyte gene
used multiplexed color-coded probes, so-called ‘molecular expression over time, such as the riboleukogram, which
barcodes’, to directly measure mRNA transcript abun- uses principal components analysis to graphically repre-
dance in samples of interest (NanoString nCounter sys- sent a patient’s genomic trajectory over time [36,71].
tem) [64]. On the basis of microarray gene expression Results from these studies suggest that genomic profiles
210
Review Trends in Molecular Medicine April 2014, Vol. 20, No. 4
Table 1. Statistical methods and other analytic tools used in gene expression profiling of sepsis
Analysis Description Refs
Single time point data
Univariate Student’s t-test A t-test is performed for each gene represented in the experiment to identify those that are [15,16,45,46]
differentially expressed between two groups (e.g., ‘sepsis’ and ‘control’). Correction of significance
level is required to reflect the testing of multiple hypotheses
Significance analysis of Identifies genes that are differentially expressed between two or more groups. SAM uses gene- [8]
microarrays (SAM) specific t-tests to assign a score based on the differences in expression levels between groups, relative
to the standard deviation [73]. A tuning parameter is used to select a tolerable false discovery rate
K-means clustering Samples are partitioned into a user-specified number of clusters according to their proximity to one [9,12,24]
another in n-dimensional gene space (where is n is the number of genes whose expression levels are
used in the analysis)
Extraction of Differential Uses an optimal discovery procedure to identify genes that are differentially expressed between user- [12,36]
Gene Expression (EDGE) specified groups [74]
Linear mixed models Each gene is modelled individually with expression levels as the dependent variables and any number [18]
of phenotypic features (such as group assignment, age, and day of sample) used as independent
variables. Differential expression between conditions of interest can be inferred, controlling for
potential confounders
Hierarchical clustering Similar expression patterns are grouped, forming a dendrogram that can be used to select clusters. [23,24,45,46]
Clustering can be done according to similarity between genes, between samples, or both. Results are
often depicted as a heatmap
Riboleukogram This approach uses a mathematical technique similar to principal components analysis in order to [36,71]
reduce the dimensionality of the data and compare patients based on average expression vectors over
time
Classification and Optimal predictors and cutoff values are determined by means of an algorithm that evaluates all [59]
regression trees (CART) possible combinations. CART has been used to identify a diagnostic panel of serum biomarkers based
on findings from whole-genome expression profiling
Multiple time point data
Timecourse ANOVA Accommodates multifactorial data to determine whether variations in gene expression over time are [69]
(TANOVA) related to the condition of interest, or an independent factor (e.g., age)
Average time curve This method involves determining whether the population average time course is best represented by [70]
a flat line, suggesting no difference in expression over time, or by a curve (cubic splines), indicating a
significant change over time
Visualization
Gene expression mosaics The Gene Expression Dynamics Inspector (GEDI) platform is used to generate a color representation [49,63]
of gene expression patterns based on self-organizing maps [58]. These expression mosaics lend
themselves to human pattern recognition, as well as computer-based recognition
in sepsis oscillate around a baseline immune attractor sampling and tissue source used further complicates direct
state. Early results support an increased between-patient comparisons between studies. Importantly, many analyses
variance in gene expression at the height of the acute have been based on small sample sizes and need to be
inflammatory phase, with differences between individuals confirmed in larger cohorts. As microarray technology and
diminishing as patients return to a baseline state of health bioinformatics methods evolve, concordance is likely to
[71]. As these statistical and computational methods increase.
evolve, comparison of gene expression trajectories in sepsis Additional challenges remain (Box 1), including the
may provide even greater insight into the molecular phy- need for more comprehensive clinical data by which to
siology of sepsis than comparison of gene expression at a annotate gene expression patterns, and for more reliable
single time point. diagnostic categories with which to label patients with
sepsis syndromes. These should not only include basic
Concluding remarks and future perspectives ‘case/control’ and ‘survivor/non-survivor’ categories but
The advent of high-throughput genomic technologies such
as gene expression microarrays have made possible the Box 1. Outstanding questions
study of the complex, dynamic changes in the host tran-
scriptome as it responds to severe infection. Initial studies Which source tissue (i.e., whole blood, neutrophils, or peripheral
blood mononuclear cells) is best suited for generating the gene
have added substantially to our understanding of sepsis expression data needed to address a particular biological
pathophysiology, identified different molecular pheno- hypothesis?
types of sepsis, and suggested novel targets for new sepsis How should time series gene expression data be collected,
therapies. Nonetheless, conclusions from these initial stu- analyzed, and merged with clinical outcome data?
What are the optimal transcriptome-based definitions and classi-
dies have been far from unanimous. Discrepancies might
fications of sepsis syndromes?
be attributable to various technical factors, including lack Can gene expression events early in the course of sepsis predict
of agreement between microarray platforms in earlier later transcriptional events, response to therapies, and clinical
studies, and an abundance of different statistical methods outcome?
and bioinformatics pathways used to conduct analyses What is the role of next-generation sequencing technologies in
the transcriptome profiling of sepsis syndromes?
(Table 1). A lack of standardization in terms of timing of
211
Review Trends in Molecular Medicine April 2014, Vol. 20, No. 4
also more nuanced labels related to illness trajectory and randomised, double-blind, placebo-controlled trial. Lancet 379, 2072–
2078
response to therapeutic interventions. New methods con-
27 Prasad, A.S. et al. (2007) Zinc supplementation decreases incidence of
tinue to be developed, including RNA sequencing, micro- infections in the elderly: effect of zinc on generation of cytokines and
RNA sequencing [72], and evaluation of microbial nucleic oxidative stress. Am. J. Clin. Nutr. 85, 837–844
acid signals. 28 Heyland, D. et al. (2013) A randomized rrial of glutamine and
antioxidants in critically ill patients. N. Engl. J. Med. 368, 1489–
1497
References 29 Carcillo, J.A. et al. (2012) The randomized comparative pediatric
1 Alizadeh, A.A. et al. (2000) Distinct types of diffuse large B-cell critical illness stress-induced immune suppression (CRISIS)
lymphoma identified by gene expression profiling. Nature 403, 503–511 prevention trial. Pediatr. Crit. Care Med. 13, 165–173
2 van’t Veer, L.J. et al. (2002) Gene expression profiling predicts clinical 30 Vanlaere, I. and Libert, C. (2009) Matrix metalloproteinases as drug
outcome of breast cancer. Nature 415, 530–536 targets in infections caused by gram-negative bacteria and in septic
3 Paez, J.G. (2004) EGFR mutations in lungcvancer: Correlation with shock. Clin. Microbiol. Rev. 22, 224–239
clinical response to gefitinib therapy. Science 304, 1497–1500 31 Nagase, H. (1999) Matrix metalloproteinases. J. Biol. Chem. 274,
4 Levy, M.M. et al. (2003) 2001 SCCM/ESICM/ACCP/ATS/SIS 21491–21494
international sepsis definitions conference. Crit. Care Med. 31, 32 Solan, P.D. et al. (2012) A novel role for matrix metalloproteinase-8 in
1250–1256 sepsis. Crit. Care Med. 40, 379–387
5 Dombrovskiy, V.Y. et al. (2007) Rapid increase in hospitalization and 33 Yazdan-Ashoori, P. et al. (2011) Elevated plasma matrix
mortality rates for severe sepsis in the United States: a trend analysis metalloproteinases and their tissue inhibitors in patients with
from 1993 to 2003. Crit. Care Med. 35, 1244–1250 severe sepsis. J. Crit. Care 26, 556–565
6 Hartman, M.E. et al. (2013) Trends in the epidemiology of pediatric 34 Cuenca, A.G. et al. (2013) Development of a genomic metric that can be
severe sepsis. Pediatr. Crit. Care Med. 14, 686–693 rapidly used to predict clinical outcome in severely injured trauma
7 Sørensen, T.I. et al. (1988) Genetic and environmental influences on patients. Crit. Care Med. 41, 1175–1185
premature death in adult adoptees. N. Engl. J. Med. 318, 727–732 35 Spielman, R.S. et al. (2007) Common genetic variants account for
8 Calvano, S.E. et al. (2005) A network-based analysis of systemic differences in gene expression among ethnic groups. Nat. Genet. 39,
inflammation in humans. Nature 437, 1032–1037 226–231
9 Fitting, C. et al. (2004) Compartmentalization of tolerance to 36 McDunn, J.E. et al. (2008) Plasticity of the systemic inflammatory
endotoxin. J. Infect. Dis. 189, 1295–1303 response to acute infection during critical illness: Development of the
10 Ge, Y. et al. (1997) Relationship of tissue and cellular interleukin-1 and riboleukogram. PLoS ONE 3, e1564
lipopolysaccharide after endotoxemia and bacteremia. J. Infect. Dis. 37 Barnato, A.E. et al. (2008) Racial variation in the incidence, care, and
176, 1313–1321 outcomes of severe sepsis. Am. J. Respir. Crit. Care Med. 177, 279–
11 Talwar, S. (2006) Gene expression profiles of peripheral blood 284
leukocytes after endotoxin challenge in humans. Physiol. Genomics 38 Wynn, J.L. et al. (2011) The influence of developmental age on the early
25, 203–215 transcriptomic response of children with septic shock. Mol. Med. 17,
12 Xiao, W. et al. (2011) A genomic storm in critically injured humans. J. 1146–1156
Exp. Med. 208, 2581–2590 39 Bouchon, A. et al. (2001) TREM-1 amplifies inflammation and is a
13 Phua, J. et al. (2013) Characteristics and outcomes of culture-negative crucial mediator of septic shock. Nature 410, 1103–1107
versus culture-positive severe sepsis. Crit. Care 17, R202 40 Gibot, S. (2005) Clinical review: role of triggering receptor expressed on
14 Palmer, C. et al. (2006) Cell-type specific gene expression profiles of myeloid cells-1 during sepsis. Crit. Care 9, 485–489
leukocytes in human peripheral blood. BMC Genomics 7, 1–15 41 Turnbull, I.R. et al. (2009) Effects of aging on the immunopathologic
15 Tang, B.M.P. et al. (2007) The use of gene-expression profiling to response to sepsis. Crit. Care Med. 37, 1018–1023
identify candidate genes in human sepsis. Am. J. Respir. Crit. Care 42 Fermin, D.R. et al. (2008) Sex and age dimorphism of myocardial gene
Med. 176, 676–684 expression in nonischemic human heart failure. Circ. Cardiovasc.
16 Tang, B.M.P. et al. (2009) Gene-expression profiling of peripheral blood Genet. 1, 117–125
mononuclear cells in sepsis. Crit. Care Med. 37, 882–888 43 Jenner, R.G. and Young, R.A. (2005) Insights into host responses
17 Vartanian, K. et al. (2009) Gene expression profiling of whole blood: against pathogens from transcriptional profiling. Nat. Rev.
comparison of target preparation methods for accurate and Microbiol. 3, 281–294
reproducible microarray analysis. BMC Genomics 10, 2 44 Yu, S-L. et al. (2004) Differential gene expression in gram-negative and
18 Parnell, G.P. et al. (2013) Identifying key regulatory genes in the whole gram-positive sepsis. Am. J. Respir. Crit. Care Med. 169, 1135–
blood of septic patients to monitor underlying immune dysfunctions. 1143
Shock 40, 166–174 45 Tang, B.M.P. et al. (2008) Gene-expression profiling of gram-positive
19 Prabhakar, U. et al. (2005) Correlation of protein and gene expression and gram-negative sepsis in critically ill patients. Crit. Care Med. 36,
profiles of inflammatory proteins after endotoxin challenge in human 1125–1128
subjects. DNA Cell Biol. 24, 410–431 46 Ramilo, O. et al. (2007) Gene expression patterns in blood leukocytes
20 Tang, B.M. et al. (2010) Genome-wide transcription profiling of human discriminate patients with acute infections. Blood 109, 2066–2077
sepsis: a systematic review. Crit. Care 14, R237 47 Marshall, J.C. (2007) Sepsis: rethinking the approach to clinical
21 Hotchkiss, R.S. et al. (2013) Immunosuppression in sepsis: a novel research. J. Leukoc. Biol. 83, 471–482
understanding of the disorder and a new therapeutic approach. Lancet 48 Vincent, J-L. et al. (2013) Sepsis definitions: time for change. Lancet
Infect. Dis. 13, 260–268 381, 774–775
22 Shankar, A.H.A. and Prasad, A.S.A. (1998) Zinc and immune function: 49 Wong, H.R. et al. (2011) Validation of a gene expression-based
the biological basis of altered resistance to infection. Am. J. Clin. Nutr. subclassification strategy for pediatric septic shock. Crit. Care Med.
68, 447S–463S 39, 2511–2517
23 Wong, H.R. et al. (2007) Genome-level expression profiles in pediatric 50 Maslove, D.M. et al. (2012) Identification of sepsis subtypes in critically
septic shock indicate a role for altered zinc homeostasis in poor ill adults using gene expression profiling. Crit. Care 16, R183
outcome. Physiol. Genomics 30, 146–155 51 Wong, H.R. et al. (2012) Interleukin-27 is a novel candidate diagnostic
24 Wong, H.R. et al. (2009) Identification of pediatric septic shock biomarker for bacterial infection in critically ill children. Crit. Care 16,
subclasses based on genome-wide expression profiling. BMC Med. 7, 34 R213
25 Srinivasan, M.G. et al. (2012) Zinc adjunct therapy reduces case fatality 52 Wong, H.R. et al. (2013) Interleukin 27 as a sepsis diagnostic biomarker
in severe childhood pneumonia: a randomized double blind placebo- in critically ill adults. Shock 40, 382–386
controlled trial. BMC Med. 10, 14 53 Calfee, C.S. et al. (2010) Plasma interleukin-8 is not an effective risk
26 Bhatnagar, S. et al. (2012) Zinc as adjunct treatment in infants aged stratification tool for adults with vasopressor-dependent septic shock.
between 7 and 120 days with probable serious bacterial infection: a Crit. Care Med. 38, 1436–1441
212
Review Trends in Molecular Medicine April 2014, Vol. 20, No. 4
54 Pachot, A. et al. (2006) Systemic transcriptional analysis in survivor 65 Kwan, A. et al. (2013) Transcriptional instability during evolving
and non-survivor septic shock patients: a preliminary study. Immunol. sepsis may limit biomarker nased risk stratification. PLoS ONE 8,
Lett. 106, 63–71 e60501
55 Pachot, A. et al. (2008) Decreased expression of the fractalkine receptor 66 Zeng, Y. and Garcia-Frias, J. (2006) A novel HMM-based clustering
CX3CR1 on circulating monocytes as new feature of sepsis-induced algorithm for the analysis of gene expression time-course data.
immunosuppression. J. Immunol. 180, 6421–6429 Comput. Stat. Data Anal. 50, 2472–2494
56 Nowak, J.E. et al. (2010) Admission chemokine (C-C motif) ligand 4 67 Geng, H. et al. (2011) Applications of hidden Markov models in
levels predict survival in pediatric septic shock. Pediatr. Crit. Care microarray gene expression data. In Hidden Markov Models, Theory
Med. 11, 213–216 and Applications (Dymarski, P., ed.), pp. 249–268, InTech
57 Standage, S.W. and Wong, H.R. (2011) Biomarkers for pediatric sepsis 68 Ming, Y. and Kendziorski, C. (2006) Hidden Markov models for
and septic shock. Expert Rev. Anti Infect. Ther. 9, 71–79 microarray time course data in multiple biological conditions. J.
58 Kaplan, J.M. and Wong, H.R. (2011) Biomarker discovery and Am. Stat. Assoc. 101, 1323–1332
development in pediatric critical care medicine. Pediatr. Crit. Care 69 Zhou, B. et al. (2010) Analysis of factorial time-course microarrays with
Med. 12, 165–173 application to a clinical study of burn injury. Proc. Natl. Acad. Sci.
59 Wong, H.R. et al. (2012) The pediatric sepsis biomarker risk model. U.S.A. 107, 9923–9928
Crit. Care 16, R174 70 Storey, J.D. et al. (2005) Significance analysis of time course
60 Wong, H.R. et al. (2014) Testing the prognostic accuracy of the updated microarray experiments. Proc. Natl. Acad. Sci. U.S.A. 102, 12837–
pediatric sepsis biomarker risk model. PLoS ONE (in press) 12842
61 Wong, H.R. et al. (2014) A multi-biomarker-based outcome risk 71 Polpitiya, A.D. et al. (2009) Using systems biology to simplify complex
stratification model for adult septic shock. Crit. Care Med. 9, e86242 disease: immune cartography. Crit. Care Med. 37, S16–S21
http://dx.doi.org/10.1097/CCM.0000000000000106 72 Ma, Y. et al. (2013) Genome-wide sequencing of cellular microRNAs
62 Eichler, G.S. et al. (2003) Gene Expression Dynamics Inspector (GEDI): identifies a combinatorial expression signature diagnostic of sepsis.
for integrative analysis of expression profiles. Bioinformatics 19, 2321– PLoS ONE 8, e75918
2322 73 Tusher, V.G. et al. (2001) Significance analysis of microarrays applied
63 Wong, H.R. et al. (2010) Toward a clinically feasible gene expression- to the ionizing radiation response. Proc. Natl. Acad. Sci. U.S.A. 98,
based subclassification strategy for septic shock: Proof of concept. Crit. 5116–5121
Care Med. 38, 1955–1961 74 Storey, J.D. et al. (2007) The optimal discovery procedure for large-
64 Geiss, G.K.G. et al. (2008) Direct multiplexed measurement of gene scale significance testing, with applications to comparative microarray
expression with color-coded probe pairs. Nat. Biotechnol. 26, 317–325 experiments. Biostatistics 8, 414–432
213