CORRESPONDENCE
Molecular Biomarker to Assist in Diagnosing
Abdominal Sepsis upon ICU Admission
To the Editor:
Abdominal infection is the second most common cause of sepsis.
Although procalcitonin (PCT) has been widely studied as a
biomarker for detecting sepsis (1), studies on the value of this
protein in diagnosing abdominal sepsis have yielded conflicting
results (2). We analyzed blood leukocyte transcriptomes of
critically ill patients to identify a molecular biomarker for
the rapid assistance in diagnosing abdominal sepsis at ICU
admission.
Abdominal sepsis was defined as an intra-abdominal infection
with a post hoc infection likelihood of definite or probable
according to previously described criteria (3), combined with at
least one parameter from the 2001 International Sepsis Definitions
Conference (4); patients who underwent abdominal surgery for a
noninfectious condition were used as controls. RNA harvest,
processing, and expression analyses were performed as described,
using U219 96-array plates (Affymetrix) (5, 6). Validation of RNA
biomarkers was performed by quantitative reverse transcription
polymerase chain reaction analysis, as described (5), using the
following primer pairs: NLRP1, encoding NLR family pyrin domain
containing 1 (forward 59 GGCTGTCGTTACACTGTGTCTG;
reverse 59 TTCACTTGCAGCCTGATCC); IDNK, encoding IDNK,
gluconokinase (forward 59 AAAAGGCATACCGCTCAATG; reverse
59 GGCTGAACAGGCTAGAACCA); and PLAC8, encoding
placenta-specific 8 (forward 59 CAGAACTCCAACTGGCAGAC;
reverse 59 ATCAGCTGCAACTTGACACC). Results were normalized
to HPRT1, encoding hypoxanthine phosphoribosyltransferase 1
(forward 59 GGATTTGAAATTCCAGACAAGTTT; reverse
59 GCGATGTCAATAGGACTCCAG). PCT was measured on
Kryptor (Thermo Fisher). Statistics and bioinformatics were
performed as described (5).
Genomewide gene expression profiles were generated from
blood collected within 24 hours after ICU admission from
60 surgical patients with abdominal sepsis and 42 patients after
abdominal surgery for a noninfectious condition (hereafter referred
to as post-op GI, for postoperative gastrointestinal, patients).
Leukocyte transcriptomes from both patient groups were compared
with those of 42 healthy subjects. Patients were similar in
demographics and comorbidities (Table 1), and white blood cell,
monocyte, and neutrophil counts were not different (data not
shown). Lymphocyte counts were slightly higher in post-op
GI patients as compared with patients with abdominal sepsis
(Wilcoxon test P = 0.038), with median values equating to 1.04 3
109 cells/L and 0.71 3 109 cells/L, respectively. Differences in
Supported by the Center for Translational Molecular Medicine, project MARS
(Molecular Diagnosis and Risk Stratification of Sepsis) (grant 04I-201).
Author Contributions: Conception and design: B.P.S. and T.v.d.P.; data
analysis and interpretation: B.P.S., A.J.H., A.M.K., and T.v.d.P.; clinical data
collection: M.A.W., L.A.v.V., J.H., M.J.B., M.J.S., and O.L.C.; microarray
scans: M.F., M.R.T., and P.N.; and drafting the manuscript for important
intellectual content: B.P.S. and T.v.d.P.
Originally Published in Press as DOI: 10.1164/rccm.201707-1339LE on
October 3, 2017
1070 American Journal of Respiratory and Critical Care Medicine Volume 197 Number 8 | April 15 2018
clinical severity indices were evident between the two patient
cohorts, with higher Acute Physiology and Chronic Health
Evaluation IV scores and more shock in patients with abdominal
sepsis, which should be considered a limitation of our study. We
found 1,196 overexpressed genes and 686 underexpressed genes
in abdominal sepsis relative to post-op GI (Figure 1A). From these,
we derived a 134-gene signature with a misclassification error rate
of 10%, using support vector machine and bootstrap resampling
(Figure 1B). Subsequently, we applied a validated combinatorial
strategy (5, 7) to the 134-gene signature, that is, calculating all
possible two-gene expression ratios ranked by Wilcoxon’s rank sum
test adjusted probabilities (adjusted for the 35,644 comparisons),
as well as receiver operator characteristic (ROC) area under
the curve (AUC). In so doing, we identified a sepsis NLRP1,
IDNK, and PLAC8 gene expression score [sNIP score, equation:
(NLRP1 2 IDNK)/PLAC8] that stratified patients with abdominal
sepsis (median = 0.17; quartile 1 [Q1]–Q3 = 0.09 to 0.28) and post-
op GI control patients (median = 20.28; Q1–Q3 = 20.4 to 20.19;
adjusted P value = 3 3 10211; Figure 1C) with ROC AUC of 0.97
(95% confidence interval [CI], 0.94–0.99). NLRP1 and PLAC8 have
roles in the innate immune response, whereas IDNK has been linked
to cellular metabolism. To evaluate the potential clinical utility of
the sNIP candidate biomarker, a numerical threshold was defined
at 20.12 (Figure 1C), which favored high sensitivity (93%), but at
the expense of specificity (86%). We reasoned that by favoring a
high sensitivity, we address the potentially serious consequences of
false-negative predictions. At this threshold, the sNIP score yielded
90.2% accuracy (95% CI, 84.3–95.1%). The threshold-dependent
positive and negative likelihood ratios were 6.53 and 0.078,
respectively.
We tested the robustness of the sNIP score in an independent
cohort of 46 patients with abdominal sepsis and 27 post-op GI
control patients by quantitative reverse transcription polymerase
chain reaction (Table 1). Applying the sNIP score showed
significant discrimination of abdominal sepsis and post-op GI
control patients (Wilcoxon’s test P value = 1.3 3 10210), with ROC
AUC of 0.91 (95% CI, 0.84–0.97) (Figure 1D). The threshold-
dependent accuracy was 0.88 (95% CI, 0.82–0.940), sensitivity
of 0.95 (95% CI, 0.9–1), specificity of 0.79 (95% CI, 0.67–0.9).
Positive and negative predictive values were 0.86 (95% CI,
0.8–0.94) and 0.92 (95% CI, 0.83–1), respectively. We also explored
the sNIP score in publicly available data from external cohorts. In
a cohort encompassing patients with microbiologically proven
sepsis (n = 10) and postsurgical noninfectious patients (n = 11)
(GSE28750) (8), application of the sNIP score resulted in favorable
discrimination of patients with sepsis with a ROC AUC of 0.98
(95% CI, 0.92–1) (Figure 1D). In another cohort consisting
of patients with postsurgical sepsis (n = 42) and postsurgical
noninfectious patients (n = 23; E-MTAB-1548) (9), evaluation
of the sNIP score yielded ROC AUC of 0.86 (95% CI, 0.75–0.95)
(Figure 1D). Finally, we compared the sNIP score with plasma
PCT and other candidate gene expression biomarkers previously
reported to assist in detecting all-cause sepsis; namely, the sepsis
MetaScore (10) and the Septicyte score (7), as well as a candidate
community-acquired pneumonia biomarker, FAIM3 (Fas apoptotic
inhibitory molecule 3):PLAC8 ratio (5) (Figure 1E). PCT was not
suitable as a biomarker in this setting. ROC analysis using the
microarray data from our discovery cohort showed that the AUCs
CORRESPONDENCE

Table 1. Characteristics and Outcomes of Abdominal Sepsis and Postoperative Gastrointestinal (Post-op GI) Noninfectious Control
Patients (Discovery and Validation Cohorts)

Discovery Cohort Validation Cohort

Post-op GI Abdominal Sepsis P Post-op GI Abdominal Sepsis P
Parameter Patients (n = 42) Patients (n = 60) Value Patients (n = 26) Patients (n = 46) Value

Demographics
Age, yr 64.5 (54–73) 63.5 (55.5–70.3) 0.76* 59 (50–67) 60 (49–67) 0.91*
Sex, male, % 64 53 0.31† 77 59 0.13†
Chronic comorbidities, n (%)
Diabetes mellitus 9 (21) 8 (13) 0.29† 7 (27) 8 (17) 0.38†
COPD 2 (5) 3 (5) 1† 2 (8) 4 (9) 1†
Hypertension 16 (38) 15 (25) 0.19† 5 (19) 14 (30) 0.41†
Congestive heart failure 3 (7) 1 (2) 0.3† 0 0 1†
Admission diagnosis, surgery
Primary sepsis diagnosis,
n (%)
Primary peritonitis — 1 (2) n.a. — — n.a.
Secondary peritonitis — 53 (88) n.a. — 38 (83) n.a.
Tertiary peritonitis — — n.a. — 1 (2) n.a.
Biliary tract infection — 4 (7) n.a. — 2 (4) n.a.
Pancreatic infection — — n.a. — 2 (4) n.a.
Intra-abdominal abscess — 2 (3) n.a. — 3 (6.5) n,a.
Admission diagnosis controls,
n (%)
GI bleeding 3 (4) — n.a. 3 (11) — n.a.
GI cancer 28 (67) — n.a. 16 (59) — n.a.
GI other‡ 11 (26) n.a. 7 (27) n.a.
Severity on ICU admission
APACHE IV score 53 (38–66) 76.5 (65–91) 1.7 3 1025* 54 (44–66) 73 (54–91) 0.002*
SOFA score 5 (2–7) 7 (6–10) 0.0004* 4 (3–5) 7 (5–9) 0.0002*
Shock, n (%) 0 27 (45) 2.8 3 1028† 0 16 (35) 0.0003†
Outcome, n (%)
ICU mortality 1 (2) 6 (10) 0.24† 0 7 (15) 0.044†
30-d mortality 4 (9.5) 10 (16) 0.39† 1 (4) 8 (17) 0.14†

Definition of abbreviations: APACHE = Acute Physiology and Chronic Health Evaluation; COPD = chronic obstructive pulmonary disease; GI =
gastrointestinal; n.a. = not applicable; SOFA = Sequential Organ Failure Assessment.
Values represent the median and lower–upper quartiles in parentheses, unless otherwise indicated. Significance was demarcated at P , 0.05.
*Wilcoxon rank sum test probability.
†
Fisher’s exact test probability.
‡
Other: GI surgery (nonnfectious) (6), cholecystectomy (3), fistula (noninfectious) (2), GI perforation (noninfectious) (2), and GI vascular ischemia (5).

of the sepsis MetaScore, Septicyte, and FAIM3:PLAC8 scores were Our study has a limitation; that is, a lack of more appropriate
significantly lower than the sNIP score in diagnosing patients with patient matching.
abdominal sepsis on ICU admission. The fact that the AUCs of the We derived a three-gene score (sNIP) that may assist in the rapid
sepsis MetaScore, Septicyte, and FAIM3:PLAC8 scores were lower diagnosis of abdominal sepsis on admission to the ICU. Future
than reported earlier (5, 7, 10) is most likely because of the design prospective validation in subjects with suspected abdominal sepsis
of the current investigation, wherein surgical patients with to evaluate the clinical value of the sNIP score are warranted. n
abdominal sepsis were compared with patients with a resembling
clinical scenario (i.e., ICU patients who had been subjected to Author disclosures are available with the text of this letter at
abdominal surgery, but for a noninfectious condition). We www.atsjournals.org.
specifically selected this strategy considering that sepsis biomarkers
are likely to be used within a certain clinical context; for example, Brendon P. Scicluna, Ph.D.
Maryse A. Wiewel, M.D.
in patients suspected to have peritonitis or pneumonia. Earlier Lonneke A. van Vught, M.D.
studies reporting on sepsis biomarkers compared patients with all- Arie J. Hoogendijk, Ph.D.
Augustijn M. Klarenbeek, B.Sc.
cause sepsis with a variety of noninfectious ICU patients, in which
University of Amsterdam
control patients differed strongly in their clinical diagnosis and Amsterdam, the Netherlands
presentation from patients with sepsis (1, 7, 10). The notion that
the value of sepsis biomarkers may vary in different clinical Marek Franitza, Ph.D.
Mohammad R. Toliat, Ph.D.
contexts is supported by the inadequacy of the FAIM3:PLAC8 ratio, Peter Nürnberg, Ph.D.
previously derived for diagnosis of patients with suspected University of Cologne
community-acquired pneumonia (5), to diagnose abdominal sepsis. Cologne, Germany

Correspondence 1071
 CORRESPONDENCE

A B C
Adj. P-value = 3x10–11
25 686 1,196 0.4
Adjusted P-value (–log10)

20
0.2
15

sNIP score
10 0.0

134 genes
5
–0.2

0
–2 0 2 4
–0.4
Fold expression (log2)
over-expressed in abdominal sepsis (fold change > 1.5)
under-expressed in abdominal sepsis (fold change < –1.5)
scaled expression –0.6
healthy subjects
–4 –2 0 2 4
post-op Gl patients post-op GI patients
abdominal sepsis patients abdominal sepsis patients

D E
1.0 1.0 sNIP score FAIM3:PLAC8 score **
qRT-PCR validation AUC 0.97 AUC 0.85
AUC 0.91 95% CI: 0.94–0.99
0.8 0.8 95% CI: 0.77–0.92
95% CI: 0.84–0.97
Sutherland et.al. sepsis MetaScore * PCT **
Sensitivity

Sensitivity

0.6 GSE28570 (8) 0.6 AUC 0.89 AUC 0.83

AUC 0.98 95% CI: 0.83–0.95 95% CI: 0.74–0.91
0.4 95% CI: 0.92–1 0.4 SeptiCyte test *
Almansa et.al. AUC 0.89
0.2 E-MTAB-1548 (9) 0.2 95% CI: 0.82–0.95
AUC 0.86
0.0 95% CI: 0.75–0.95 0.0
1.0 0.8 0.6 0.4 0.2 0.0 1.0 0.8 0.6 0.4 0.2 0.0
Specificity Specificity
Figure 1. Identification of the sepsis NLRP1, IDNK, and PLAC8 (sNIP) gene expression score. (A) Volcano plot depicting the differences in
blood gene expression in patients with abdominal sepsis as compared with postoperative gastrointestinal (post-op GI) patients. (B) Heat map
representation of a 134-gene expression signature that best discriminates abdominal sepsis and post-op GI patients. (C) Dot plot showing sNIP
scores in abdominal sepsis and post-op GI patients. Horizontal lines in dot plots denote medians. Horizontal line across graph depicts the test
threshold. Probabilities were calculated by Wilcoxon’s rank sum test and adjusted for multiple comparisons in sNIP score derivation (35,644
comparisons). (D) Evaluation of the sNIP score by receiver operator characteristic area under the curve (AUC) in an independent cohort by qRT-PCR
and additional publicly available cohorts. GSE28570 (8), postsurgical noninfectious controls compared with culture-proven sepsis (cases). E-MTAB-
1548 (9), postsurgical sepsis compared with postsurgical noninfectious patients. (E) Receiver operator characteristic AUC of the FAIM3:PLAC8
score, sepsis MetaScore (10), Septicyte score (7), and procalcitonin (PCT) compared with the sNIP score. *P , 0.01; **P , 0.001. CI = confidence
interval; FAIM3 = Fas apoptotic inhibitory molecule 3; IDNK = IDNK, gluconokinase; NLRP1 = NLR family pyrin domain containing 1; PLAC8 =
placenta-specific 8.

Janneke Horn, M.D. Tom van der Poll, M.D.

University of Amsterdam University of Amsterdam
Amsterdam, the Netherlands Amsterdam, the Netherlands

Marc J. Bonten, M.D.

University Medical Center Utrecht References
Utrecht, the Netherlands
1. Wacker C, Prkno A, Brunkhorst FM, Schlattmann P. Procalcitonin as a
Marcus J. Schultz, M.D. diagnostic marker for sepsis: a systematic review and meta-analysis.
University of Amsterdam Lancet Infect Dis 2013;13:426–435.
Amsterdam, the Netherlands 2. Watkins RR, Lemonovich TL. Serum procalcitonin in the diagnosis and
management of intra-abdominal infections. Expert Rev Anti Infect Ther
Olaf L. Cremer, M.D. 2012;10:197–205.
University Medical Center Utrecht 3. Klein Klouwenberg PM, Ong DSY, Bos LDJ, de Beer FM, van Hooijdonk
Utrecht, the Netherlands RTM, Huson MA, et al. Interobserver agreement of Centers for Disease

1072 American Journal of Respiratory and Critical Care Medicine Volume 197 Number 8 | April 15 2018
CORRESPONDENCE
Control and Prevention criteria for classifying infections in critically ill
patients. Crit Care Med 2013;41:2373–2378.
4. Levy MM, Fink MP, Marshall JC, Abraham E, Angus D, Cook D, et al;
SCCM/ESICM/ACCP/ATS/SIS. 2001 SCCM/ESICM/ACCP/ATS/SIS
International Sepsis Definitions Conference. Crit Care Med 2003;31:
1250–1256.
5. Scicluna BP, Klein Klouwenberg PM, van Vught LA, Wiewel MA, Ong DS,
Zwinderman AH, et al. A molecular biomarker to diagnose community-
acquired pneumonia on intensive care unit admission. Am J Respir
Crit Care Med 2015;192:826–835.
6. van Vught LA, Klein Klouwenberg PM, Spitoni C, Scicluna BP, Wiewel MA,
Horn J, et al.; MARS Consortium. Incidence, risk factors, and
attributable mortality of secondary infections in the intensive care unit
after admission for sepsis. JAMA 2016;315:1469–1479.
7. McHugh L, Seldon TA, Brandon RA, Kirk JT, Rapisarda A, Sutherland AJ,
et al. A molecular host response assay to discriminate between sepsis
and infection-negative systemic inflammation in critically ill patients:
discovery and validation in independent cohorts. PLoS Med 2015;12:
e1001916.
8. Sutherland A, Thomas M, Brandon RA, Brandon RB, Lipman J, Tang B,
et al. Development and validation of a novel molecular biomarker
diagnostic test for the early detection of sepsis. Crit Care 2011;15:
R149.
9. Almansa R, Heredia-Rodrı́guez M, Gomez-Sanchez E, Andaluz-Ojeda D,
Iglesias V, Rico L, et al. Transcriptomic correlates of organ failure
extent in sepsis. J Infect 2015;70:445–456.
10. Sweeney TE, Shidham A, Wong HR, Khatri P. A comprehensive
time-course-based multicohort analysis of sepsis and sterile
inflammation reveals a robust diagnostic gene set. Sci Transl Med
2015;7:287ra71.
Copyright © 2018 by the American Thoracic Society
Plasma Exosomes Disrupt the Blood–Brain Barrier
in Children with Obstructive Sleep Apnea and
Neurocognitive Deficits
To the Editor:
Obstructive sleep apnea (OSA) is a prevalent condition in children and is
associated with a significant constellation of morbidities, including
neurocognitive, cardiovascular, and metabolic dysfunction (1–3).
Activation and propagation of multiple inflammatory pathways, altered
lipid metabolism, and oxidative stress mechanisms have all been
implicated in end-organ morbidity (4, 5). In two recent papers, Lim
and Pack and our group have proposed that disruption of the
blood–brain barrier (BBB) may underlie the cognitive impairments
associated with OSA (6, 7). As corroborative evidence, studies in mice
exposed to intermittent hypoxia have shown increases in brain
parenchymal water, along with alterations in aquaporin expression,
indicating increased BBB permeability (8, 9). BBB permeability changes
The authors are supported by NIH grant HL130984 (L.K.-G.).
Author Contributions: A.K. performed experiments, analyzed data, and
drafted components of the manuscript; D.G. participated in the conceptual
framework of the project, provided critical input in all phases of the
experiments, analyzed data, and edited versions of the manuscript; L.K.-G.
provided the conceptual framework of the project, analyzed data, drafted
components, and finalized the manuscript and is responsible for the financial
support of the project and the manuscript content; and all authors have
reviewed and approved the final version of the manuscript.
Originally Published in Press as DOI: 10.1164/rccm.201708-1636LE on
October 20, 2017
Correspondence
have also been inferred in adult patients with OSA (10). In this setting,
it has been reported that endothelial cells secrete exosomes, and several
reports show that endothelial cells can also be targeted by exosomes
derived from different cell types. The tight junction complex is critically
involved in the exchange of ions, solutes, and cells that travel across
BBB paracellular spaces and tight junctions. Zonula occludens-1
(ZO-1) is one of several protein families that are essential for tight
junction formation (11), and it is now established that stressful
conditions can disrupt brain endothelial tight junctions and affect
cognition via exosome-related biological activities affecting the BBB (12).
To examine the potential contribution of circulating exosomes to
BBB disruption in the context of pediatric OSA, we explored, using an
in vitro BBB system (13), the effect of plasma-derived exosomes from
children with polysomnographically determined OSA with evidence
of neurocognitive deficits (NC1; n = 12); age-, sex-, ethnicity-, body
mass index z-score–, apnea–hypopnea index–matched children with
OSA and no evidence of cognitive deficits (NC2; n = 12); and
control children without OSA or cognitive deficits (CO; n = 6). The
characteristics of the subjects are shown in Table 1. All subjects
underwent overnight polysomnography, which was scored as per
current American Academy of Sleep Medicine guidelines, and in
the morning after the sleep study, fasting blood was drawn into
ethylenediaminetetraacetic acid tubes, and plasma was immediately
separated by centrifugation and stored until assay, immediately
followed by cognitive test batteries (1). The presence of cognitive
deficits (NC1) was defined as the presence of two or more cluster
subtests that were more than 1 SD below the mean, as previously
described (14). Plasma exosomes were isolated and purified as
previously described (2) and fulfilled all the required criteria as
specified by the current consensus of the International Society for
Extracellular Vesicles (15).
Using an immortalized human brain microvascular endothelial
cell model (hCMEC/D3; Cat# SCC066, EMD Millipore), the
effects of equivalent numbers of exosomes from each subject on
transcellular electrical impedance of a hCMEC/D3 monolayer were
evaluated by electric cell-substrate impedance-sensing arrays. As
previously described (2), exosomes were added in duplicate wells
and changes in impedance across the hCMEC/D3 monolayer were
continuously monitored in the electric cell-substrate impedance
sensing instrument (Applied Biophysics Inc.) for up to 48 hours.
Appropriate internalization of the exosomes by human brain
microvascular endothelial cells was verified in a preliminary set
of experiments using time-lapse confocal microscopy. Of note,
the resistance across the hCMEC/D3 monolayer at confluence
was measured at more than 800 V $ cm2 (13). In addition,
immunofluorescence staining of confluent hCMEC/D3 endothelial
cell monolayers that were grown on 12-well cover slips for 24 hours
in Dulbecco’s modified Eagle medium containing 10% fetal
bovine serum were also performed. Isolated exosomes from subjects
were added individually to cover slips for 24 hours. Cells were fixed
with 4% (wt/vol) paraformaldehyde in phosphate-buffered saline
(PBS) for 20 minutes at room temperature and then washed again
with PBS. The cell membranes were permeabilized by incubation
with 0.25% (vol/vol) Triton-X-100 in PBS for 10 minutes at room
temperature. After washing with PBS, the samples were blocked with
3% (wt/vol) bovine serum albumin in PBS for 45 minutes at room
temperature, followed by overnight incubation at 48 C with ZO-1
antibody (1:400; Life Technologies). Alexa 488 was used as
secondary antibody (1:400; Life Technologies), and nuclear staining
1073