Subjects

In this nested cross-sectional analysis, pregnant women between the ages of 18-40 years provided amniotic samples. Amniotic fluid samples were obtained from subjects in spontaneous term (≥37 ^0/7 weeks) labor (defined as the presence of strong, regular uterine contractions at a minimum frequency of 2 contractions/10 min, followed by changes in cervical effacement and dilatation that led to delivery at term (≥37 ^0/7 ). Amniotic fluid samples were also collected from women, not in labor, undergoing elective cesarean deliveries. Gestational age was determined by the last menstrual period, and was corroborated by ultrasound biometry. Subjects with multiple gestations, preeclampsia, placenta previa, fetal anomalies, gestational diabetes mellitus, and/or other medical/surgical complications of pregnancy were excluded. Subjects who were treated for preterm labor or for suspected intraamniotic infection and delivered at term were excluded from the study. Details of this cohort and samples can be found in our other publications.

Collection of samples

For vaginal deliveries, amniotic fluid samples were collected during labor, immediately before artificial rupture of the membranes, by transvaginal amniocentesis of intact membranes, using a 22-gauge needle through the dilated cervical os. In cases undergoing cesarean delivery, samples were collected by transabdominal amniocentesis. Amniotic fluid samples were centrifuged immediately for 10 minutes at 2000 g to remove cellular and particulate matter and supernatant aliquots were processed rapidly and stored in the dark at –80°C in filled tubes, to minimize autooxidation until analysis. We studied 29 term labor and 33 term not in labor, which provided >80% power to detect the difference in analyte concentrations between the groups, at a .05 significance level for each analyte in their log-transformed form.

Demographic data were collected from patient interviews and clinical data were abstracted from the medical records of the patients. Data collected included age, socioeconomic status (education, annual income, insurance status, and marital status), behavioral factors (smoking, alcohol and drug use during pregnancy), prepregnancy body mass index (BMI), and a complete medical and obstetrical history.

Documentation of ultrastructural morphology of term chorioamniotic membranes by transmission electron microscopy

Transmission electron microscopy (TEM) was performed to determine the presence of SA morphologic changes in chorioamniotic membranes. For this, we used images from our ongoing studies and the detailed protocol for TEM studies are described in our prior publications. Chorioamniotic membranes from term births (n = 6 each from labor and not in labor) were fixed, stained, and embedded in PolyBed 812 (Polysciences, Washington, PA). Both amnion and chorion layers were evaluated separately and also as a unit. Briefly, initial fixation was for 24 hours at 4°C in a fixative with 2.5% paraformaldehyde, 0.2% glutaraldehyde, and 0.03% picric acid in 0.05 mol/L cacodylate buffer. After fixation, samples were rinsed 3 times with cacodylate buffer and postfixed with 1% osmium tetroxide in 0.1 mol/L cacodylate buffer. Osmicated tissue was rinsed twice with deionized water and stained en bloc with 2% aqueous uranyl acetate for 1 hour at 60°C. The samples were then dehydrated by a series of ethanol-water solutions (50%, 75%, 95%, and 100% ethanol for 3 exchanges). Dehydrated tissue was infiltrated with 2 exchanges of propylene oxide, then with propylene oxide–diluted PolyBed resin at a 1:1 ratio, a 1:2 ratio, then twice with pure PolyBed 812. Finally, the samples were embedded in PolyBed 812, and cured overnight at 60°C. Since precise tissue orientation could not be maintained during curing of the resin, the first resin blocks were cut to give a wide flat face of the desired sectioning plane, replaced into new embedding molds, and cured again. Samples were cut as 90-nm sections, placed on formvar-coated slotted grids, and poststained for 3 minutes with a solution of Reynold lead citrate. Images were taken with a JEM 1400 electron microscope (JEOL, Tokyo, Japan).

Senescence detection by SA β-gal staining

Chorioamniotic membranes were dissected from placenta after normal term (labor and not in labor) deliveries (3 samples collected from Nashville, TN, and 6 from University of Texas Medical Branch, Galveston, TX). Tissue samples were collected and processed as described in prior publications. Briefly, midzone portion of the chorioamniotic membranes were dissected from the placenta, rinsed in phosphate-buffered saline, and decidua and blood clots were scraped off using cotton gauze. Multiple scrapings without peeling the membranes were performed to obtain chorioamniotic membranes used in this study. Decidua was removed to determine the senescence of chorioamniotic membranes, since decidual SA with preterm labor has already been shown in animal models. Tissues were embedded in OCT (Sakura Finetek, Torrance, CA) and frozen until tissues were sectioned for staining. Cryostat sectioning was performed on tissues (6 μm thickness) and mounted onto glass slides prior to staining. Senescent cells were identified by SA β-gal staining in a distinct pH (6.0) using a histochemical staining kit in frozen tissue sections (senescence detection kit; BioVision Inc, Milpitas, CA). Frozen tissue sections were fixed with 0.5 mL of fixative solution for 15 minutes, at room temperature. Staining was performed on frozen section as per manufacturer’s protocol and counter-stained lightly with eosin to better visualize cellular morphology. Number of senescent cells per high-power field was examined and a semiquantitative assessment of the number of cells staining for SA β-gal (blue staining) was performed. Two independent assessments were made by investigators who were blinded to the outcome. Ten high-power fields were examined for both amnion and chorion layers separately and number of positive (blue staining) and negative (only eosin stained) cells were counted. Statistical difference in the percentage of cells positive for SA β-gal was calculated between term labor and term not in labor samples.

Selection of biomarkers

Fourteen biomarkers that belong to cytokines and growth factors (IL-15, monocyte chemoattractant protein [MCP]-2, granulocyte macrophage colony-stimulating factor [GM-CSF], IL-6, IL-13) chemokines (eotaxin-1, IL-8), angiogenic factors (angiogenin, vascular endothelial growth factor), matrix-degrading enzymes (MMP-1) and its inhibitor (tissue inhibitor of metalloproteinase-1), and apoptotic factors (soluble Fas ligand [sFasL], osteoprotegerin [OPG], intercellular adhesion molecule [ICAM]-1) were chosen for this study, based on their potential role at term pregnancies ( Table 2 ). We selected markers that perform major biological function in the cell cycle (proliferation, arrest, or apoptosis) and inflammation (chemotaxis, tissue/extracellular matrix destruction, and remodeling).

FAST Quant-microspot assays for biomarker measurement

All samples included in this study were assayed simultaneously to avoid variability introduced during assay procedures. The assays were performed on amniotic fluid samples using the FAST Slide protein microarray platform developed and manufactured by Maine Manufacturing (now GVS Life Sciences, Sanford, ME). Briefly, the FAST Slide surface uses a nitrocellulose-based 3-dimensional polymer coating to provide reproducibility and sensitivity for microarray assays. The polymer binds proteins in a noncovalent, irreversible manner, and can be probed using the same method as in traditional blots. The 3-dimensional nature of the matrix allows for the retention of arrayed protein in near-quantitative fashion resulting in linear increases in signal intensities in response to increased concentration of antigen/analyte, over a very wide range of concentrations. Sixteen-pad nitrocellulose-coated glass slides were arrayed in triplicate with reference markers and capture antibody for analytes in that array, using a piezo-electric spotter. Lack of cross-reactivity among these markers was established. The detailed protocols for the assay can be found at http://www.gvs.com/flex/cm/pages/ServeBLOB.php/L/ENn/IDPPagina/611 . Briefly, slides were blocked by incubating in 70-μL protein array blocking buffer for 15 minutes at room temperature; the blocking buffer was removed, and 70-μL samples or standards added to the appropriate pad and incubated overnight. The slides were then washed 5 times in 1X protein array wash buffer, and 70 μL of biotinylated detection antibody cocktail was added and incubated for 1 hour. Following a further 5 washes, 70 μL streptavidin-Cy5 solution was added and the slides were incubated for 1 hour in the dark, washed 5 times, and allowed to dry. The slides were imaged in an Axon GenePix 4200A fluorescent imaging system (Axon Instruments, Foster City, CA). Images were analyzed using Imaging Research ArrayVision software (Imaging Research, St. Catharines, Ontario, Canada). Briefly, spot intensities were determined by subtracting background signal. Spot replicates from each sample condition were averaged and then compared to the appropriate standard curves. Accuracy of Fast Quant protein arrays is comparable to the correspondent enzyme-linked immunosorbent assay determinations with a similar linear range. The advantages of protein array as compared to conventional enzyme-linked immunosorbent assay are the assessment of multiple cytokines using smaller volume and different specimens at the same time. Samples were diluted until the sample’s readings fell within the antigen’s standard curve, then the sample’s cytokine concentrations were determined by correcting these values for the dilution.

The limit of detection was calculated as the average value from 0 antigen samples plus 3 SD, extrapolated from the standard curves calculated by ArrayVision. Limits of detection are conservative estimates based on development of FAST Quant arrays using buffer-diluted antigens. These values are guidelines; actual detection limits may be lower ( Table 3 ).

Statistical analyses of SASP markers

For our primary analysis, we considered SASP markers as our exposure variables and term labor vs term not in labor (reference group) as our outcome. To identify baseline characteristics associated with term labor, variables including maternal age, race, education, marital status, BMI, smoking, parity, gestational age, infant sex, and delivery type were examined using multivariate logistic regression to derive odds ratios, 95% confidence intervals, and P values. As SASP markers were not normally distributed, we used nonparametric test to examine correlation among markers as well as differences in SASP marker distributions between groups. Correlation among senescence markers was determined by Spearman correlation test and Wilcoxon-Mann-Whitney was used to examine differences in the distribution of SASP markers between women at term labor and term not in labor.

To analyze associations between SASP markers and term labor vs term not in labor, regression modeling was utilized. First, given that dichotomizing a continuous exposure leads to loss of statistical power and often relies on the use of an arbitrary cut-point for our primary analysis, we choose to examine SASP markers as continuous variables. We used generalized additive models with smoothing splines to explore relationships between SASP markers and term labor (term not in labor as reference). A generalized additive model does not assume a parametric relationship between an exposure and outcome and therefore is a flexible nonparametric regression technique. This type of exploratory modeling approach can be used to determine appropriate transformations and for building parametric models. There were significant nonparametric trends between all SASP markers and term labor and we subsequently transformed SASP markers to the log base 2 (creating a normal distribution) and used logistic regression to examine associations between SASP marker concentrations and term labor. Thus, a 1-U increase in a SASP marker (in log scale) can be interpreted as a doubling of intensity. To derive inferences about odds of term labor when SASP markers are elevated, we dichotomized SASP markers by the median and calculated odds ratios and 95% confidence intervals using logistic regression. Potential covariates were included in the full model and once removed, any variables that changed the coefficient by >10% was considered a confounder. Parity, BMI, cigarette smoking, socioeconomic factors (defined by education, income, marital and insurance status) appeared to have an effect on some markers (all contributors of oxidative and psychosocial stress during pregnancy that can impact senescence), and were included in the final regression model. We used Bonferroni correction for multiple comparisons and a P value < .0018 was considered statistically significant. All analyses were conducted using SAS V9.2 (SAS Institute Inc, Cary, NC).

Analysis of interaction between biomarkers using multifactor dimensionality reduction analysis

We used multifactor dimensionality reduction (MDR) to explore interactions among SASP markers that influence term labor. We are the first to report biomarker interactions in preterm birth using this approach. MDR is considered a powerful approach to detect nonlinear interactions by combining categorical attribute selection, attribute construction, and classification with cross-validation and permutation testing ( Epistasis.org ). MDR has been used to examine gene-gene and gene-environment interactions. However, we have successfully used MDR to explore interactions among biological markers. To run the MDR software, we first had to exclude patients with missing data (n = 3) and then categorize SASP in quartiles. We then used default MDR settings, with 1 exception where we used a Fisher exact test set at 0.01, rather than tie cells, which require an even number of cases and controls. MDR is a method that categorized combinations of biomarker models by risk to produce models that most consistently predict disease determined by cross-validation consistency (CVC). CVC or testing accuracy ranges from 0-1, where the highest CVC indicated the most accurate model. We then incorporated permutations (MDR permutation application) to determine a P value for statistical significance.