SRI Supplement to Reproductive Sciences - Volume 25 Number 1 - March 2018 - 244A

244A

Reproductive Sciences Vol. 25, Supplement 1, March 2018

known. A possible pathway could be fetal gender dependent maternal
immune activation. In this (ongoing) study we analyzed the influence of
fetal gender on the maternal immune system.
METHODS: First-trimester decidual tissue was obtained from surplus
tissue at chorionic villus sampling (CVS), which was performed between
10 and 12 weeks of gestation for maternal age indication or serum
screening related risk for aneuploidy. Decidual tissue was mechanically
separated from the villi and stored until analysis. Using rtPCR, mRNA
expression of T cell transcription factors and several cytokines was
analyzed in 18 primigravid pregnancies, 9 male and 9 female fetuses,
with uncomplicated pregnancy outcomes. mRNA data were normalized
to HPRT mRNA expression. Data was compared using Mann-Whitney
test ; p<0.05.
RESULTS: Early pregnancy decidual mRNA expression of FOXP3 and
IFN gamma was respectively 6-fold (p<0.01) and 6-fold (p<0.01) higher
in pregnancies of female fetuses. Moreover, mRNA expression of IL6
2.5-fold (p= 0.07) higher in female fetuses.
CONCLUSION: Our results show that fetal gender modulates the
maternal immune response in early pregnancy. The lower mRNA
expression of FOXP3 in decidual tissue of male fetuses could indicate a
possible Treg pathway responsible for the altered pregnancy outcome in
male fetuses. Samples of decidual tissue from term pregnant and other
Treg cell markers are currently analyzed.

F-190
Maternal Immune and Pro-Coagulant Pathway Activation in a Rat
Model of Recurrent Miscarriage and Preeclampsia. Eugenia Mata
Greenwood, Tino Sanchez†, Denise Bellinger*, Carlos Casiano*. Loma
Linda University, Loma Linda, CA, United States.
INTRODUCTION: We have found that Brown Norway rats present
spontaneous recurrent fetal losses, gestational hypertension and
proteinuria making it a novel model of recurrent miscarriage and
preeclampsia. The aim of this study was to study the plasma proteomic
profile of this rat throughout gestation to identify mechanisms of disease.
METHODS: Heparinized plasma from Brown Norway rats (disease
phenotype) and Lewis rats (control phenotype) were studied for
differential protein profiles at gestational day 13 using 2D DIGE protein
expression profiling. In-gel analysis of protein differences between plasma
samples identified more than 75 proteins of interest. 54 Proteins were
extracted and identified by Mass Spectrometry, and key findings were
validated by ELISA.
RESULTS: Brown Norway pregnancies were characterized by increased
plasma levels of complement C4, prothrombin and C-reactive protein
compared to non-pregnant Brown Norway rats. Furthermore, our control
rat strain, Lewis, showed decreased TNF-alpha, IL-2, and total IgG plasma
levels compared to Brown Norway rats at pregnancy days 9, 13 and 17.
Lewis pregnancies also showed lower plasma levels of complement C4,
prothrombin, and C-reactive protein, but higher levels of transthyretin
(TTR) compared to Brown Norway pregnancies.
CONCLUSION: Brown Norway rat pregnancies show maternal
activation of immune (Th1-mediated) and pro-coagulant pathways in
association with recurrent miscarriage and preeclampsia-like symptoms.

F-191
Metabolomic Identification of Novel Diagnostic Biomarkers in Ectopic
Pregnancy. Onur Turkoglu†,1 Ayse Citil,3 Ceren Katar,3 Ismail Mert,2
Praveen Kumar,1 Ali Yilmaz,1 Dilek S Uygur,3 Salim Erkaya,3 Stewart
Graham,1 Ray O Bahado-Singh*.1 1Beaumont Health, Royal Oak, MI,
United States; 2Mayo Clinic, Rochester, MN, United States; 3Zekai Tahir
Burak Women's Health Research and Education Hospital, Ankara, Turkey.
INTRODUCTION: Ectopic pregnancy (EP) is a potentially lifethreatening condition and early diagnosis still remains a challenge,
causing a delay in management leading to tubal rupture. We performed
metabolomic analysis on plasma to develop diagnostic biomarkers for
ectopic pregnancy (EP).
METHODS: Proton Nuclear Magnetic Resonance ( 1 H NMR)
metabolomics profiling was performed on 39 EP cases and 89 early
intrauterine pregnancy controls. To avoid over-fitting, datasets were

Scientific Abstracts

randomly divided into a discovery group (26 EP vs 60 controls) and
an independent test group (13 EP and 30 controls). Logistic regression
models including metabolites only or in combination with clinical risk
factors were performed using discovery group. The discovery regression
models were subsequently validated in the test group. Area under the
Receiver Operating Characteristics curve (AUC), 95% confidence interval
(CI), sensitivity, and specificity values were calculated. Metabolite set
enrichment analysis (MSEA) was used to elucidate altered biochemical
pathways in EP.
RESULTS: There were no significant difference between mean
gestational age of EP vs controls in validation (p=0.34) and independent
test (p=0.11) groups. In 13 of 43 (30.3%) metabolites serum concentrations
were significantly altered in EP (p<0.05). Metabolites significantly
separated EP and controls groups (p<0.05) (graphic1). In the independent
validation group a two-metabolite model (lactic acid + acetic acid)
achieved high diagnostic accuracy AUC (95% CI) =0.990 with a
sensitivity and specificity of 92% and 93%, respectively for EP detection.
MSEA revealed significant (p<0.05) perturbation of multiple metabolite
pathways including gluconeogenesis, pyruvate, butyrate, tyrosine, and
ketone body metabolisms (graphic2). Clinical history did not improve
predictive accuracy over metabolomic markers alone.
CONCLUSION: EP is an important cause of maternal death and
morbidity. In the first report of its kind, novel metabolomic biomarkers
accurately predicted EP in an independent validation group. If validated
in larger patient groups, this could result in improved early diagnosis,
accuracy and reduced health care cost from monitoring, hospitalization
and unnecessary surgery in suspected EP cases.
*Figure(s) will be available online.

F-192
Automating Segmentation of Placenta Tissue Images for Computer
Aided Diagnosis and Creating Objective Measures for Microscopic
Components. Anika Mukherjee†, 2,3 Sina Salsabili†, 1 Adrian DC
Chan*,1 Eran Ukwatta*,1 Michal Leckie†,2,3 David Grynspan*,2 Shannon
Bainbridge*.3 1Carleton University, Ottawa, ON, Canada; 2Children's
Hospital of Eastern Ontario, Ottawa, ON, Canada; 3University of Ottawa,
Ottawa, ON, Canada.
INTRODUCTION: Manual image analysis of tissue images is critical
for pathological diagnosis but is a tedious, costly, and time-consuming
process with high inter- and intra- individual variability. Past attempts
have been made at automating tissue and cellular image analysis, but the
task is complicated by 2 factors: 1) complexity of tissue (cells touching,
structures in process of maturation, varying sizes/staining, artefacts) and
2) lack of gold standard definitions of visual characteristics in tissue. We
hypothesize that given a set of visual features (size, color, convexity),
a computer can learn visual characteristics to recognize placenta tissue
and its microscopic components. This would improve clinical practice
by reducing hospital costs and helping to create objective image feature
standards that can inform diagnoses. Our objective is to develop a program
to automatically segment placental villi accurately.
METHODS: Experiments were performed on high resolution scans
of placental biopsy slides (purchased from the RCWIH biobank) from
healthy controls and pregnancies complicated by preeclampsia We
manually segmented 1-3 rectangles of 2740 by 3964 pixels from one
representative slide 9 clinical categories using the annotate tool in Aperio
Imagescope. We then developed a program in MatLab which used visual
features that define individual villi in placenta tissue. Segmentation criteria
were: density of syncytial knots between two villi >25%, orientation of
knots following perimeter of villi, <25% blood vessels or mesenchymal
connective tissue, and >60% concavity between villi.
RESULTS: Manually segmented images were used as ground truth and
compared with predictions on identical images generated by our model.
Accuracy = 77.3%, precision = 80.1%, and recall = 85.6 (on 2 areas from
each of 3 test slides).
*Figure(s) will be available online.
Figure 1. A. Ground truth B. Automatic segmentation
CONCLUSION: Our data shows that it is possible to automatically
segment placental villi with a high level of accuracy. False positives and

Table of Contents for the Digital Edition of SRI Supplement to Reproductive Sciences - Volume 25 Number 1 - March 2018