Inflammation biomarkers in vaginal fluid and preterm delivery

Abstract

STUDY QUESTION

Which inflammation biomarkers detected in the vaginal fluid are most informative for identifying preterm delivery (PTD) risk?

SUMMARY ANSWER

Elevated interleukin (IL)-6 at mid-trimester was associated with increased odds of spontaneous PTD at <35 weeks and with PTD plus histologic chorioamnionitis (HCA), and had the greatest sensitivity for detecting these two PTD subtypes.

WHAT IS KNOWN ALREADY

Maternal and/or fetal inflammation play a role in some preterm deliveries, therefore inflammation biomarkers might help to identify women at greater risk.

STUDY DESIGN, SIZE, DURATION

We examined 1115 women from the Pregnancy Outcomes and Community Health Study, a cohort study conducted from September 1998 through June 2004, for whom data were available on mid-pregnancy inflammatory biomarkers.

PARTICIPANTS/MATERIALS, SETTING, METHODS

At enrollment at 16–27 weeks gestation, vaginal fluid samples were collected from a swab and 15 eluted biomarkers were measured using the Meso Scale Discovery multiplex electrochemiluminescence platform. Associations of biomarkers with PTD were examined, according to clinical circumstance, week at delivery and presence/absence of HCA. Weighted logistic regression was used to determine odds ratios (OR) and 95% confidence intervals (CI) adjusted for race. Sensitivity and specificity were compared between individual and multiple biomarkers, identified by a bootstrapping method.

MAIN RESULTS AND THE ROLE OF CHANCE

Elevated IL-6 (>75th percentile) displayed the strongest association with spontaneous PTD <35 weeks (OR 2.3; CI 1.3–4.0) and PTD with HCA (OR 2.8; CI 1.4–6.0). The sensitivity of IL-6 to detect spontaneous PTD <35 weeks or PTD with HCA was 0.43 and 0.51, respectively, while specificity was 0.74 and 0.75, respectively. IL-6 plus IL1β, IL-6r, tumor necrosis factor-alpha or granulocyte-macrophage colony-stimulating factor increased specificity (range 0.84–0.88), but decreased sensitivity (range 0.28–0.34) to detect both PTD subtypes. Results were similar when a combination of IL-6 and bacterial vaginosis (BV) was explored. Thus, the use of multiple biomarkers did not detect PTD subtypes with a greater sensitivity than IL-6 alone, and IL-6 is a specific but non-sensitive marker for the detection of spontaneous PTD.

LIMITATIONS, REASONS FOR CAUTION

Our ability to find small effect size associations between PTD and inflammation biomarkers (OR <2.0) might have been limited by the modest number of less common PTD subtypes in our population (e.g. spontaneous delivery <35 weeks, PTD accompanied by HCA) and by relatively higher variability for some cytokines, for example tumor necrosis factor-α, IL-12p70, IL-10 and granulocyte-macrophage colony-stimulating factor, that are less stable and commonly undetectable or detectable at low levels in human vaginal secretions.

WIDER IMPLICATIONS OF THE FINDINGS

Larger studies are needed to further explore a role of inflammation biomarkers in combination with other risk factors, including specific BV-associated organisms, for the prediction of PTD subtypes.

STUDY FUNDING/COMPETING INTEREST(S)

This work was supported by the National Institute of Child Health and Human Development, National Institute of Nursing, March of Dimes Foundation, Thrasher Research Foundation and Centers for Disease Control and Prevention. The authors have no conflicts of interest.

Introduction

Preterm delivery (PTD), delivery before 37 weeks of gestation, is the leading cause of neonatal morbidity and mortality and has been linked to long-term mental and physical disabilities (Goldenberg et al., 2008; Moster et al., 2008). Approximately 12.0% of infants in the USA are born preterm (Hamilton et al., 2011). In addition, the overall PTD rate in the USA remains higher than in previous decades. Due to the serious complications that result from PTD, public health professionals have focused on identifying biomarkers to predict women at risk for delivering preterm.

The pathways leading to PTD are complex and have not been completely elucidated. The maternal immune system plays important roles throughout pregnancy with respect to maternal immune tolerance towards the fetus and protection against infection (Witkin et al., 2011). Additionally, immunological processes are involved in maintenance of gestational tissue growth, remodeling and differentiation (Witkin et al., 2011). Inflammation has been implicated as a component in one or more pathways to PTD (Goldenberg et al., 2008). Human studies have reported associations between spontaneous PTD, histologic chorioamnionitis (HCA) and serum, vaginal and amniotic fluid levels of cytokines (Gargano et al., 2008; Puchner et al., 2011; Simhan et al., 2011), soluble proteins that form complex networks mediating inflammation (Hames and Glover, 2000). In animal models, inflammation at the chorio-decidua leads to ripening of the cervix, rupture of membranes and preterm labor (Grigsby et al., 2010; Kemp et al., 2010). Further, the expression of cytokines, including interleukin (IL)-1β, IL-6, IL-8 and tumor necrosis factor (TNF)-α, have been linked to premature parturition through the up-regulation of prostaglandins, matrix metalloproteinase (MMP) and other uterine and cervical factors (Kemp et al., 2010). Investigators of PTD have long considered inflammation biomarkers as potentially useful for: (i) examining the nature of the host inflammatory response in relation to PTD; and (ii) identifying women at increased risk for PTD (Menon et al., 2011). However, results from studies examining inflammation biomarkers and PTD have been inconsistent, and none have identified a biomarker that accurately and reliably predicts PTD (Dudley et al., 1994; Goepfert et al., 2001; Andrews et al., 2006; Gargano et al., 2008; Puchner et al., 2011; Simhan et al., 2011). In part, this may be related to the complexity of the inflammatory process, its biomarkers and to the heterogeneity of PTD pathways.

In a recently published review of 217 studies focused on PTD and biomarkers, including inflammation biomarkers, Menon et al. (2011) noted that study methodologies varied widely. There were, however, some common limitations across many of the studies, such as combining all PTDs into a single group, limited rationale for biomarker selection, inadequate description of study design and participant characteristics, and problems with analytic methods and interpretation. The authors further concluded that a single biomarker will not be sufficient for predicting all PTD and inflammation biomarkers, specifically cytokines, should be further explored as predictors of PTD. Our study is responsive to the critique outlined by Menon et al. (2011). We examined multiple inflammation biomarkers in mid-pregnancy vaginal fluid to determine their associations with PTD grouped by clinical circumstance (spontaneous versus medically indicated), by week at delivery (<35, 35–36) and by presence/absence of HCA. In addition, we compared the sensitivity and specificity of a combination of these vaginal inflammation biomarkers with that of a single biomarker.

Methods

Study design

The Pregnancy Outcomes and Community Health (POUCH) study was designed to investigate pathways to PTD as well as other adverse pregnancy outcomes. The study received institutional review board approval at Michigan State University, Michigan Department of Community Health and nine community hospitals. Women in community clinics who met eligibility criteria and expressed interest constituted the sampling frame. A random, stratified sampling scheme was used to sample women for the POUCH Study cohort and sub-cohort (described below). Certain groups of interest were over-sampled, a technique frequently employed in national surveys, such as NHANES (1994). Particularly, when the total sample size is limited by available resources, oversampling of specific strata improves the precision of risk estimates for the oversampled strata. In analyses of the entire sample, sampling weights are used and individuals in oversampled strata are assigned lower sampling weights so that, among the whole, each stratum represents its original proportion in the sampling frame.

POUCH study cohort and sub-cohort

Eligibility criteria for the POUCH Study cohort included prenatal care at one of 52 clinics in 5 Michigan Communities during 8 September 1998 through 15 June 2004, 16–27th week of pregnancy, maternal serum alpha-fetoprotein (MSAFP) screening, singleton pregnancy with no known chromosomal abnormality or birth defect, maternal age >15 years, no pre-pregnancy diabetes mellitus and proficiency in English. Women who met the eligibility criteria and expressed interest in the study constituted the ‘sampling frame’. Those who had normal MSAFP levels were stratified by race/ethnicity and randomly sampled into the cohort. In addition, all interested women with unexplained high MSAFP levels of greater than two multiples of the median were invited to participate because this prenatal screening biomarker has been consistently linked to risk of PTD and was of particular interest in the POUCH Study (Holzman et al., 2001). Of the 3038 women recruited into the cohort, 3019 were followed (99%) through delivery (Fig. 1). Overall 7% of the cohort had high MSAFP levels compared with the typical 3–5% in the screened population. In cohort analyses, after sampling weights are applied, women with elevated MSAFP account for only 3–5% of the cohort, typical of any unselected pregnancy cohort. Thus, the POUCH Study cohort can be thought of as a survey sample intended to represent the original sampling frame. A comparison of maternal characteristics between POUCH Study participants and all mothers delivering within the five study communities during the same period (data from birth certificates) showed there was little difference, with the exception that older African-American mothers were slightly under-represented in the POUCH Study. The weighted percentage of PTD in the POUCH Study cohort was 10.7%, similar to that found in the community populations.

Figure 1.

Flow chart of inflammatory marker samples.

In order to conserve resources, some costly data elements (e.g. placental examinations, medical record abstraction, assays in stored biologic samples) were obtained from only a sub-sample of POUCH Study participants, referred to as the sub-cohort. The sub-cohort of 1371 women was assembled by using a random stratified sampling design, sampling from among the cohort and again oversampling from certain strata. There were eight sampling strata defined by race/ethnicity (African American, non-African American), MSAFP levels (high, normal) and pregnancy outcome (preterm, term). The sub-cohort included 100% of cohort women who delivered preterm (n = 335) and 100% of women who delivered at term with high MSAFP (n = 165). In the remaining strata of women with normal MSAFP and term deliveries, the study sampled 72% of African Americans (n = 422) and 23% of non-African Americans (n = 449). In all sub-cohort analyses, sampling weights were used to account for the oversampling of women with high MSAFP into the cohort and the oversampling of particular maternal characteristics (i.e. high MSAFP, PTD, African-Americans with term deliveries) into the sub-cohort. Each sub-cohort woman is assigned a sampling weight that reflects her representation of similar women in the original sampling frame (described further in analytic section).

Vaginal inflammation biomarkers

At enrollment (16–27 weeks’ gestation) cohort women met with a study nurse, signed consent forms, completed in-person interviews and self-administered questionnaires, and had biological samples collected. Vaginal fluid samples were collected via a fetal fibronectin specimen collection kit (Adeza International, Sunnyvale, CA, USA) with a sterile Dacron swab. After inserting a vaginal speculum, the study nurse swabbed the area just below the cervix, and then placed the swab in 1 ml extraction buffer (Adeza International, Sunnyvale, CA, USA) which was immediately refrigerated (4°C) for at least 24 h. After the refrigeration period, buffer was expressed from the swab, the specimen was filtered using a 10.25 mm × 10.2 cm serum filter (Fisherbrand Serum Filter System, Fisher Scientific, Pittsburgh, PA, USA), divided into 0.5 ml aliquots and stored at −80°C.

For this analysis we included 1115 women from the sub-cohort (Fig. 1) with data on vaginal inflammation biomarkers. Pro-inflammatory markers IL1β, TNFα, IL-6, RANTES and IL-8 and the receptors [IL-1 receptor antagonist (IL-1ra), IL-6 receptor (IL-6r), TNF-receptor 1 (TNF-r1), TNF-receptor 2 (TNF-r2)], which have been implicated in preterm parturition (Romero et al., 2007) were included in this study. We also include IL-10, as the down-regulation of this anti-inflammatory cytokine has been suggested to play an important role in the onset of labor (Romero et al., 2007). Additionally, we included IL-12p70, a cytokine that can promote type 1 T-helper (Th1) response (Enright et al., 2011), MMP-9 which contributes to cervical ripening, membrane rupture, matrix remodeling and placental separation (Stygar et al., 2002; Goldman et al., 2003), granulocyte–macrophage colony-stimulating factor (GM-CSF), which is a regulator of neutrophil production and function during infection and has been implicated in fetal inflammatory response syndrome (Chaiworapongsa et al., 2011), intercellular adhesion molecule 1 (ICAM1) which is an adhesion molecule involved in inducing inflammation of the chorion, amnion and decidua (Steinborn et al., 1999), and C-reactive protein (CRP), a marker of systemic inflammation that has been implicated as a potential predictor of PTD (Vogel et al., 2005).

Laboratory analyses of vaginal inflammation biomarkers

All vaginal protein biomarkers were measured in a central laboratory accredited by the College of American Pathologists (Laboratory of Genital Tract Biology, Brigham and Women's Hospital) using standardized operational procedures. Samples were analyzed in duplicate using the Meso Scale Discovery (MSD) multiplex electrochemiluminescence (ECL) platform and Sector Imager 2400 (MSD, Gaithersburg, MD). The MSD ECL detection system has been validated for cytokine quantification in the vaginal fluid matrix by comparisons with traditional enzyme-linked immunosorbent assay (ELISA) and other platforms in a multicenter study (Fichorova et al., 2008). Two custom multiplex assays (i) IL6, IL10, IL12p70, TNFα, GMCSF, ICAM-1 and RANTES, with detection cut-offs of 0.6, 0.6, 9.8, 0.6, 0.6, 9.8 and 2.4 pg/ml, respectively, and (ii) IL-1β, IL-8, IL-6R, CRP, MMP9, TNFR1 and TNFR2 with cut-off of 2, 1.2, 1.8, 22, 1746, 19 and 1.4 pg/ml, respectively, were optimized to allow the detection of each biomarker within the linear concentration range of the vaginal samples. IL-1ra was measured by Quantikine ELISA (R&D Systems, Minneapolis, IL, USA) and a Victor² reader (Perkin Elmer, Boston, MA, USA). Supplementary data, Table SI shows the intra-class and the inter-class coefficient of variation (CV) obtained from the analysis of a split quality control sample in each assay plate. Inter-class CVs range from 0.10 (RANTES) to 0.37 (TNF-α), while intra-class CVs ranged from 0.027 (IL-8) to 0.157 (TNF-α).

Outcome assessment

PTD was defined as birth before 37 completed weeks' gestation. Gestational age was determined by the last menstrual period (LMP) or by ultrasound data when the LMP-derived gestational age differed from the ultrasound estimate by at least 2 weeks. Based on the information abstracted from labor and delivery medical records, PTD was divided into two groups: (i) spontaneous PTD included women with preterm labor defined as regular contractions that led to cervical change (≥2 cm dilatation), or spontaneous premature rupture of membranes and (ii) medically indicated PTD included women who had labor induced or who were delivered by Caesarian section before either preterm labor or premature rupture of membranes.

PTD with HCA was also analyzed as an outcome among 901 sub-cohort women with vaginal inflammation biomarker and placental pathology data (Fig. 1). Details of the placenta histopathology protocol have been published previously (Holzman et al., 2007). Briefly, nine tissue sections, representing the placental disc, cord and extraplacental membranes, were examined microscopically by a placental pathologist who was blinded to gestational age at delivery and all clinical data. The microscopic evaluation used a detailed descriptive rather than diagnostic approach, which allowed a number of potential definitions to be tested. For the present analysis, HCA was defined as polymorphonuclear leukocyte inflammatory pattern in the chorionic plate and/or extraplacental membrane chorion and amnion, plus karyorrhexis and or necrotizing inflammation. In our previous work these pathology-based criteria for HCA constituted a threshold for positive associations with risk of PTD (Holzman et al., 2007).

At enrollment, study participants met with a study nurse and provided information about demographics, current pregnancy, reproductive history, health behaviors such as smoking, and psychosocial factors. Potential confounding factors and known mediators for PTD were considered. Self-reported data on maternal age (modeled as a dichotomous variable), race (Black, non-Hispanic White/others), parity (no previous live births, one or more births) and Medicaid use (yes, no) were examined. Smoking was included as a four-level variable (smoked but quit before enrollment, smoking less than half a pack of cigarettes at enrollment, smoking half a pack or more at enrollment and no smoking in pregnancy). Pre-pregnancy BMI was also a four-level variable using Centers for Disease Control cut points: underweight (BMI <18.5 kg/m²), normal weight (BMI 18.5–24.9 kg/m²), overweight (BMI 25.0–29.9 kg/m²) and obese (BMI ≥ 30 kg/m²). Vaginal swabs were examined to determine the presence of bacterial vaginosis (BV) as scored according to Nugent's criteria (Nugent et al., 1991). A score of 0–3 is considered normal vaginal flora, 4–6 is an intermediate score and a score of 7 or greater is defined as BV positive.

Analytical strategy

Percentages and weighted percentages of important covariates (described above) were examined in the sample of women with inflammation biomarker data (n = 1115) and in the slightly smaller sample of women with both inflammation biomarker and placental pathology data (n = 901). These percentages were compared with those in the entire sub-cohort to assess possible selection bias.

Correlations among vaginal inflammation biomarkers, modeled as continuous variables, were examined using Pearson's correlation coefficients. Biomarkers were also dichotomized as >75th percentile versus ≤75th percentile; the upper quartile cut-point was based on the biomarker distribution among women with normal MSAFP levels who delivered at term (Supplementary data, Table SII). In the first set of models, weighted logistic regression was used to assess relations between each dichotomized biomarker and a five-level pregnancy outcome variable (term as referent, medically indicated <35 weeks, medically indicated 35–36 weeks, spontaneous <35 weeks and spontaneous 35–36 weeks). In a second set of analyses, the outcome was changed to a combination of PTD (no/yes) and HCA (no/yes) and modeled as a four-level variable (term and no HCA as referent, and three comparison levels of term and HCA, PTD and no HCA and PTD and HCA). Associations among maternal characteristics and inflammation biomarkers were evaluated as potential confounders or key covariates (i.e. BV). All regression results are presented unadjusted and adjusted for race/ethnicity.

After modeling each inflammation biomarker separately, bootstrapping was used to determine which inflammation biomarkers were consistently associated with PTD. Cases and controls were each randomly sampled for bootstrapping with 200 replications. The case group was alternately defined as spontaneous PTD <35 weeks’ gestation or PTD with HCA, because these specific outcomes had positive results with individual inflammation biomarkers in the previous set of analyses. The control group consisted of all term deliveries. The number of positive associations out of 200 bootstrap samples was counted and biomarkers with the highest number of positive associations were used to create a list of top multiple biomarkers associated with PTD groups. Weighted logistic regression models were repeated for the single most consistent biomarker (based on bootstrapping) and for a combination of biomarkers selected by bootstrapping. These models incorporated the five-level and four-level outcome variables described in the previous paragraph.

Analyses were then conducted to determine the sensitivities and specificities of a single biomarker and a set of biomarkers in relation to the two main groups of PTD modeled in bootstrapping, that is those that were spontaneous <35 weeks’ gestation and those that were <37 weeks’ gestation and accompanied by HCA. In these sensitivity/specificity analyses the outcome was dichotomized, for example spontaneous <35 weeks’ gestation versus all other outcomes including term and other preterm. This is in keeping with prospective screening practices to detect a PTD subtype when the outcome is not yet known. Weights were incorporated in all regression models to reflect oversampling of high MSAFP into the cohort and the sub-cohort sampling scheme. All logistic regression analyses were conducted by using ProcSurveylogistic procedures with SAS, version 9.2 (SAS Institute, Inc., Cary, NC, USA).

Results

In the two samples (sub-cohort women with vaginal inflammation biomarker data and the subset with vaginal biomarker plus placenta data) just over 57% were between the ages of 20–29 years, 47–49% had >12 years of education, 51–54% were non-Hispanic White, 71–72% were in their 20–24th week of pregnancy at enrollment, 59–60% had a parity of ≥1, 71–72% never smoked, 44–45% had a normal pre-pregnancy BMI and 78–79% were negative for BV (Table I). These percentages were similar to those observed in the entire cohort.

Table I.

Maternal characteristics of POUCH study participants: data are presented for the total sub-cohort, the sub-cohort with inflammation measures and the sub-cohort with both inflammation measures and placental measures.

Sub-cohort
	Total (n = 1371)			Subset with inflammation markers (n = 1115)			Subset with inflammation markers with placenta data (n = 901)
	n	%	Weighted %	n	%	Weighted %	n	%	Weighted %
Maternal age (years)
<20	243	17.7	14.7	191	17.1	13.9	142	15.8	13.3
20–29	776	56.6	57.2	644	57.8	58.5	517	57.4	57.7
≥30	352	25.7	28.1	280	25.1	27.6	242	26.9	29.0
Maternal education (years)
<12	317	23.1	18.9	266	23.9	19.5	195	21.6	18.3
12	399	29.1	27.8	327	29.3	28.1	263	29.2	27.8
>12	655	47.8	53.3	522	46.8	52.4	443	49.2	53.8
Race/ethnicity
White	692	50.5	65.8	573	51.4	66.4	488	54.2	66.1
African American	579	42.2	24.6	464	41.6	24.6	345	38.3	24.6
Others	100	7.3	9.6	78	7.0	9.0	68	7.5	9.3
Gestational age at enrollment
<20 weeks	224	16.3	15.7	188	16.9	15.6	137	15.2	13.8
20–24 weeks	965	70.4	71.1	793	71.1	73.0	651	72.3	74.0
25–27 weeks	182	13.3	13.2	134	12.0	11.4	113	12.5	12.1
Medicaid insurancea
No	586	42.8	50.9	464	41.7	49.9	398	44.2	50.8
Yes	783	57.2	49.1	650	58.3	50.1	502	55.8	49.2
Paritya
None	577	42.1	41.5	456	40.9	39.5	357	39.7	38.7
At least one	793	57.9	58.5	658	59.1	60.5	543	60.3	61.3
Smoking
Never	979	71.4	72.5	791	70.9	72.1	651	72.3	73.1
Stopped before enrollment	132	9.6	9.9	109	9.8	9.9	83	9.2	9.3
<½ pack per day at enrollment	182	13.3	11.5	149	13.4	11.7	114	12.7	11.3
≥½ pack per day at enrollment	78	5.7	6.1	66	5.9	6.3	53	5.9	6.3
Pre-pregnancy BMI
Low	65	4.7	3.9	52	4.7	3.8	39	4.3	3.8
Normal	609	44.4	46.6	490	43.9	46.3	402	44.6	45.7
High	303	22.1	23.0	245	22.0	22.7	200	22.2	23.7
Obese	394	28.7	26.4	328	29.4	27.2	260	28.9	26.8
BV (Nugent score)b
Negative	878	78.0	80.6	867	78.0	80.4	711	79.2	80.9
Intermediate	103	9.1	9.2	102	9.2	9.3	81	9.0	9.3
Positive	145	12.9	10.2	142	12.8	10.3	106	11.8	9.8

^aIn the inflammation marker sample, 1 missing for Medicaid insurance and parity; 4 missing for BV.

^bWeighted percentages reflect sampling design for the cohort and sub-cohort (see Materials and methods section of this paper).

The correlation between each pair of inflammation biomarkers varied with a range of 0.06 (IL6 and ILra) to 0.85 (IL10 and IL12; Table II). A total of 34 cytokine pairs were correlated at R >0.50. Among these pairs, 17 were considered highly correlated at R >0.70.

Table II.

Pearson correlations between inflammation markers sampled at mid-pregnancy (weighted).

Variable	Label	crp	gmcsf	icam1	il1b	il1ra	il6	il6r	il8	il10	il12p70	mmp9	rantes	tnfa	tnfr1	tnfr2
crp	CRP, pg/ml (mean)	1
gmcsf	GM-CSF, pg/ml (mean)	0.20	1
icam1	ICAM1, pg/ml (mean)	0.45	0.41	1
il1b	IL-1b, pg/ml (mean)	0.46	0.32	0.56	1
il1ra	IL-1RA, pg/ml (mean)	0.39	0.24	0.25	0.33	1
il6	IL-6, pg/ml (mean)	0.16	0.27	0.61	0.38	0.06	1
il6r	IL-6R, pg/ml (mean)	0.46	0.37	0.61	0.77a	0.29	0.41	1
il8	IL-8, pg/ml (mean)	0.45	0.27	0.56	0.74a	0.27	0.42	0.73a	1
il10	IL-10, pg/ml (mean)	0.20	0.49	0.56	0.29	0.11	0.44	0.33	0.31	1
il12p70	IL-12p70, pg/ml (mean)	0.26	0.49	0.71a	0.35	0.17	0.48	0.39	0.35	0.85a	1
mmp9	MMP-9, pg/ml (mean)	0.39	0.27	0.51	0.68	0.25	0.41	0.81a	0.76a	0.32	0.36	1
rantes	RANTES, pg/ml (mean)	0.23	0.33	0.45	0.34	0.17	0.22	0.40	0.44	0.38	0.42	0.40	1
tnfa	TNF-a, pg/ml (mean)	0.33	0.35	0.75a	0.55	0.24	0.62	0.51	0.51	0.60	0.66	0.45	0.33	1
tnfr1	TNF-R1, pg/ml (mean)	0.52	0.35	0.51	0.79a	0.33	0.31	0.84a	0.84a	0.29	0.33	0.82a	0.46	0.44	1
tnfr2	TNF-R2, pg/ml (mean)	0.33	0.26	0.55	a0.73a	0.16	0.50	0.77a	0.74a	0.31	0.37	0.74a	0.38	0.49	0.79a	1

Bold indicates correlation >0.5.

^aIndicates highly correlated (>0.7).

Elevated IL-6 levels (>75th percentile) were significantly associated with spontaneous delivery at <35 weeks, adjusted odds ratio (aOR) = 2.3, 95% confidence interval (CI) 1.3–4.0 and with PTD accompanied by HCA, aOR = 2.8, 95% CI 1.4–6.0 (Table III). In addition, some biomarkers produced aORs of 1.5 or greater but the CIs included 1.

Table III.

Inflammation markers >75th percentile in relation to preterm birth subtypes defined by clinical circumstance (spontaneous versus medically indicated) by week at delivery (<35, 35–36) and by presence/absence of HCA.

	n	aORa (95% CI)
		CRP	GM-CSF	ICAM1	IL-1β	IL-1 ra	IL-6	IL-6R	IL-8
PTD
Term delivery (ref.)	848	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0
Medically indicated <35 weeks	32	0.3 (0.1, 1.1)	0.9 (0.4, 2.2)	0.9 (0.4, 2.2)	0.7 (0.3, 1.7)	0.1 (0.0, 0.8)	0.8 (0.3, 1.9)	0.7 (0.3, 1.7)	0.5 (0.2, 1.4)
Medically indicated 35–36 weeks	54	1.1 (0.6, 2.0)	0.7 (0.4, 1.4)	0.7 (0.3, 1.4)	0.9 (0.5, 1.8)	0.7 (0.3, 1.3)	1.0 (0.5, 2.0)	0.8 (0.4, 1.6)	0.7 (0.3, 1.4)
Spontaneous <35 weeks	55	0.9 (0.5, 1.8)	0.9 (0.5, 1.8)	1.1 (0.6, 2.1)	1.5 (0.8, 2.6)	0.8 (0.4, 1.7)	2.3 (1.3, 4.0)	1.2 (0.6, 2.2)	1.0 (0.5, 1.9)
Spontaneous 35–36 weeks	126	1.4 (0.9, 2.1)	1.2 (0.8, 1.9)	1.4 (0.9, 2.2)	1.1 (0.7, 1.7)	1.4 (0.9, 2.1)	1.0 (0.6, 1.5)	1.1 (0.7, 1.8)	1.0 (0.6, 1.6)
PTD and HCA
Term and no HCA (ref.)	595	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0
Term and HCA	75	1.4 (0.8, 2.7)	0.7 (0.3, 1.4)	0.7 (0.3, 1.5)	0.7 (0.4, 1.5)	1.0 (0.5, 1.8)	0.6 (0.3, 1.3)	1.0 (0.5, 1.9)	0.6 (0.3, 1.3)
PTD and no HCA	199	1.0 (0.7, 1.5)	0.8 (0.5, 1.2)	0.8 (0.5, 1.2)	0.9 (0.6, 1.3)	0.8 (0.5, 1.2)	0.9 (0.6, 1.3)	0.8 (0.5, 1.1)	0.8 (0.5, 1.1)
PTD and HCA	32	0.8 (0.3, 2.0)	1.7 (0.8, 3.7)	1.4 (0.6, 3.2)	1.7 (0.8, 3.8)	0.8 (0.3, 2.0)	2.8 (1.4, 6.0)	1.9 (0.9, 4.0)	0.8 (0.3, 1.9)
		IL-10	IL-12p70	MMP9	RANTES	TNFα	TNF-R1	TNF-R2
PTD
Term delivery (ref.)	848	1.0	1.0	1.0	1.0	1.0	1.0	1.0
Medically indicated <35 weeks	32	1.1 (0.5, 2.5)	1.0 (0.4, 2.3)	0.8 (0.3, 1.9)	1.2 (0.5, 2.6)	1.3 (0.6, 2.8)	0.8 (0.4, 1.9)	0.9 (0.4, 2.0)
Medically indicated 35–36 weeks	54	1.2 (0.6, 2.2)	1.1 (0.6, 2.1)	0.7 (0.4, 1.5)	0.5 (0.2, 1.2)	1.0 (0.5, 1.8)	0.6 (0.3, 1.3)	0.5 (0.2, 1.1)
Spontaneous <35 weeks	55	0.8 (0.4, 1.6)	1.2 (0.6, 2.2)	1.0 (0.5, 1.8)	1.4 (0.7, 2.5)	1.1 (0.6, 2.1)	1.2 (0.6, 2.2)	1.3 (0.7, 2.4)
Spontaneous 35–36 weeks	126	0.9 (0.6, 1.5)	1.3 (0.8, 2.0)	1.0 (0.6, 1.6)	0.9 (0.5, 1.4)	1.4 (0.9, 2.1)	1.1 (0.7, 1.6)	0.9 (0.6, 1.4)
PTD and HCA
Term and no HCA (ref.)	595	1.0	1.0	1.0	1.0	1.0	1.0	1.0
Term and HCA	75	0.7 (0.4, 1.5)	1.1 (0.5, 2.1)	1.0 (0.5, 2.0)	1.5 (0.8, 2.8)	0.8 (0.4, 1.6)	0.8 (0.4, 1.5)	0.7 (0.4, 1.5)
PTD and no HCA	199	0.8 (0.6, 1.2)	1.0 (0.7, 1.5)	0.7 (0.5, 1.1)	0.9 (0.6, 1.4)	1.0 (0.7, 1.4)	0.8 (0.5, 1.2)	0.8 (0.5, 1.1)
PTD and HCA	32	0.8 (0.3, 2.0)	1.3 (0.6, 3.0)	1.0 (0.4, 2.4)	1.1 (0.5, 2.6)	1.8 (0.8, 3.9)	1.1 (0.5, 2.5)	1.5 (0.7, 3.3)

^aOdds ratio adjusted for race.

In further analyses, we examined associations among biomarkers and BV. IL-6 was associated with increased odds of BV (OR 1.4, 95% CI 0.9–2.2), although this did not reach statistical significance. Women with both increased IL-6 and BV had increased odds of PTD with HCA (aOR 4.0, 95% CI 1.5–10.9), but the association with PTD <35 weeks (aOR 2.0, 95% CI 0.8–4.8) was not statistically significant.

We also ran exploratory analyses to determine whether a disruption in the balance of inflammatory markers, i.e. increased IL1β expression (>75th percentile) and decreased IL1ra expression (<50th percentile); or increased IL6 expression (>75th percentile) and decreased IL6r expression (<50th percentile), were associated with PTD subtypes. Race-adjusted ORs were consistently above 1 (1.8 and 2.0 for spontaneous PTD <35 weeks, 2.3 and 1.4 for PTD with HCA), but all CIs included 1. These results may be explained by the limited sample size for examining biomarker interactions.

Our bootstrapping result showed that in repeated sampling, the top quartiles of two cytokines, IL-6 (positive 159 times out of 200) and IL-1β (positive 63 times out of 200), were most consistently associated with spontaneous delivery <35 weeks. Women with elevated IL-6 levels in mid-pregnancy vaginal fluid had an aOR of 2.3 (95% CI 1.3–4.0) for this subtype of PTD. Women with elevated IL-6 and IL-1β had a very similar aOR of 2.2 (Table IV). Elevated IL-6 had a weighted sensitivity of 0.43 and specificity of 0.74 for spontaneous delivery <35 weeks. The combination of elevated IL-6 and IL1-β increased the specificity (0.85) but considerably lowered the sensitivity (0.28). When we excluded women with HCA from the analysis (data not shown), the aOR for spontaneous delivery <35 weeks was attenuated and no longer statistically significant among women in the top quartile of IL-6 (aOR 1.6, 95% CI 0.8–3.1) and among women in the top quartiles of IL-6 and IL-1β (aOR 1.4, 95% CI 0.7–3.0).

Table IV.

Association between inflammation marker level >75th percentile and preterm delivery subgrouped by clinical circumstance (spontaneous versus medically indicated) and by week at delivery (<35, 35–36).

	≤75th percentile	>75th percentile			Sensitivity	Specificity	Weightedb	Weightedb
	n (ref.)	n	OR (95% C. I.)	aORa (95% C. I.)			Sensitivity	Specificity
IL-6
Term delivery (ref.)	639	209	1.0	1.0
Medically indicated <35 weeks	25	7	0.8 (0.3, 1.9)	0.8 (0.3, 1.9)
Medically indicated 35–36 weeks	40	14	1.0 (0.5, 1.9)	1.0 (0.5, 2.0)
Spontaneous <35 weeks	31	24	2.2 (1.2, 3.9)	2.3 (1.3, 4.0)	0.44	0.75	0.43	0.74
Spontaneous 35–36 weeks	94	32	1.0 (0.6, 1.5)	1.0 (0.6, 1.5)
IL-6 and IL1β
Term delivery (ref.)	720	128	1.0	1.0
Medically indicated <35 weeks	27	5	1.0 (0.4, 2.7)	1.0 (0.4, 2.8)
Medically indicated 35–36 weeks	48	6	0.7 (0.3, 1.7)	0.7 (0.3, 1.7)
Spontaneous <35 weeks	39	16	2.3 (1.2, 4.2)	2.2 (1.2, 4.0)	0.29	0.85	0.28	0.85
Spontaneous 35–36 weeks	106	20	1.1 (0.6, 1.9)	1.1 (0.6, 1.8)

^aOdds ratio adjusted for race.

^bWeighted percentages reflect sampling design for the cohort and sub-cohort (see Materials and methods section of this paper).

A second set of bootstrapping analyses suggested that PTD with HCA was also associated with elevated IL-6 levels in mid-pregnancy vaginal fluid (positive 156 times out of 200). In addition elevations in four other inflammation biomarkers were implicated as potentially relevant; TNF-α  (positive 90 times out of 200), IL-6r (positive 68 times out of 200), GM-CSF (positive 63 times out of 200) and IL-1β (positive 53 times out of 200). As before, when IL-6 was assessed alone and then in combination with each of the other inflammation biomarkers, the aORs for PTD with HCA were similar (aOR range 2.2–3.1) with overlapping CIs (Table V). The upper quartile of IL-6 had a weighted sensitivity of 0.51 and specificity of 0.75. The addition of the other inflammation biomarkers lowered the sensitivity to 0.29–0.38 and increased the specificity to 0.84–0.88. Results were similar for spontaneous PTD <35 weeks and PTD with HCA when we examined the combination of elevated IL-6 and BV (sensitivity 0.12 and 0.21, specificity 0.94 and 0.95). We had a total of eight women for whom the interval between vaginal fluid sampling and delivery was ≤4 weeks. In a sensitivity analysis, removal of these women had no major impact on results in any of our analyses (data not shown).

Table V.

Association between inflammation marker level >75th percentile subgrouped by presence/absence of HCA.

	≤75th percentile	>75th percentile			Sensitivity	Specificity	Weightedb	Weightedb
	n (ref.)	n	OR (95% CI)	aORa (95% CI)			Sensitivity	Specificity
IL-6
Term and no HCA (ref.)	444	151	1.0	1.0
Term and HCA	61	14	0.6 (0.3–1.3)	0.6 (0.3–1.3)
PTD and no HCA	151	48	0.9 (0.6–1.3)	0.9 (0.6–1.3)
PTD and HCA	17	15	2.8 (1.3–5.9)	2.8 (1.4–6.0)	0.47	0.75	0.51	0.75
IL-6 and TNFα
Term and no HCA (ref.)	500	95	1.0	1.0
Term and HCA	64	11	0.7 (0.3, 1.6)	0.7 (0.3, 1.5)
PTD and no HCA	167	32	0.9 (0.6, 1.5)	0.9 (0.6, 1.5)
PTD and HCA	22	10	2.4 (1.1, 5.4)	2.2 (1.0, 5.1)	0.31	0.84	0.32	0.84
IL-6 and GMCSF
Term and no HCA (ref.)	518	77	1.0	1.0
Term and HCA	68	7	0.6 (0.2, 1.7)	0.6 (0.2, 1.6)
PTD and no HCA	177	22	0.8 (0.5, 1.3)	0.8 (0.5, 1.3)
PTD and HCA	23	9	2.7 (1.1, 6.3)	2.6 (1.1, 6.0)	0.28	0.88	0.29	0.88
IL-6 and IL-6r
Term and no HCA (ref.)	497	98	1.0	1.0
Term and HCA	65	10	0.7 (0.3, 1.7)	0.7 (0.3, 1.7)
PTD and no HCA	172	27	0.8 (0.5, 1.2)	0.8 (0.5, 1.2)
PTD and HCA	21	11	3.0 (1.4, 6.7)	3.1 (1.4, 6.7)	0.34	0.84	0.38	0.84
IL-6 and IL1β
Term and no HCA (ref.)	504	91	1.0	1.0
Term and HCA	66	9	0.7 (0.3, 1.7)	0.7 (0.3, 1.6)
PTD and no HCA	171	28	0.9 (0.6, 1.5)	0.9 (0.6, 1.5)
PTD and HCA	21	11	3.3 (1.5, 7.3)	3.1 (1.4, 6.8)	0.34	0.85	0.36	0.85

^aOdds ratio adjusted for race.

^bWeighted percentages reflect sampling design for the cohort and sub-cohort (see Materials and methods section of this paper).

Discussion

Women with vaginal fluid IL-6 levels in the upper quartile at mid-trimester had increased odds of spontaneous delivery at <35 weeks’ gestation and PTD accompanied by HCA. Elevated levels of inflammatory markers TNF-α, GM-CSF, IL-1β and IL-6r tended to associate with increased odds of PTD subtypes, but these associations were not statistically significant. Results from this study showed that IL-6 had the greatest sensitivity for detecting two PTD subtypes. While the use of two inflammation biomarkers slightly increased the specificity for the PTD subtypes, it did so at the cost of decreasing sensitivity.

During pregnancy cytokines are suggested to be involved in implantation, cervical ripening, membrane rupture and enhancement of uterine contractions (Orsi and Tribe, 2008; Challis et al., 2009). Therefore, alterations in inflammatory responses may be responsible for deleterious outcomes during pregnancy. Our results showing a positive relation between IL-6 levels and spontaneous delivery at <35 weeks’ gestation are consistent with two prospective studies and a systematic review reporting that IL-6 measured in the cervico-vaginal fluid at mid-trimester is associated with spontaneous PTD (Lockwood et al., 1994; Paternoster et al., 2002; Wei et al., 2010). Using a higher cut-point, the Preterm Prediction Study (125 matched case–control pairs) found that IL-6 (>90th or 95th percentile) was no longer significantly associated with spontaneous delivery at <35 weeks after adjustments for confounding factors (aOR 1.8, 95% CI 0.8–4.3), positing the study may have been underpowered (Goepfert et al., 2001).

In our study, IL-6 displayed the strongest association with PTD accompanied by HCA. Further, the association between IL-6 and spontaneous PTD <35 weeks was attenuated after we excluded women with HCA. The lack of associations between mid-pregnancy IL-6 levels and later spontaneous PTD (35–36 weeks’ gestation) is perhaps because the measurable changes in markers of inflammatory response in the vaginal fluid have not yet occurred, a transition from acute to chronic inflammation has occurred under the influence of IL-6, or because different underlying causes predominate in the late versus early spontaneous PTD.

In vitro and in vivo studies have shown that IL-6 expression is enhanced in decidual cells following chorioamnionitis (Dudley et al., 1994; Maeda et al., 1997; Lockwood et al., 2010). Bacterial infections can lead to inflammation and have been found to be associated with HCA in preterm deliveries (Andrews et al., 2006). Therefore, the IL-6 association with PTD accompanied by HCA may suggest an increased inflammatory response following microbial infection of the vaginal tract. Rizzo et al. (1996) and Holst et al. (2005) showed that cervical IL-6 is a marker for intra-amniotic infection and inflammation among women in preterm labor. Our finding that an increased vaginal IL-6 level was not strongly linked to BV was consistent with the results from the Preterm Prediction Study (Goepfert et al., 2001). We did not examine the presence/absence of specific micro-organisms which may be more closely tied to vaginal cytokine profiles.

Cytokines interact with one another for the overall balance of inflammatory responses. For example, IL-6 function is mediated through interaction with IL-6r, and IL-1ra regulates inflammation by acting as an endogenous inhibitor of IL-1β. Thus, detecting disruption in the balance of certain inflammation biomarkers may be most informative for understanding and/or predicting PTD subtypes. Following this line of thinking, we examined combinations of high IL-1β with low IL-1ra, and high IL-6 with lowIL-6r but the numbers of women with these combinations were low and therefore we lacked statistical power to detect even moderate associations. IL1-β has been demonstrated to interact with IL-6, enhancing IL-6 mRNA and protein expression when incubated with cultured human decidual cells in the hormonal milieu of pregnancy (Lockwood et al., 2010). IL-6 can be regulated by a variety of other factors associated with inflammation and PTB. Higher IL-6 levels have been associated with cervical ectopy, and lower levels have been associated with prevalence of some BV-associated bacteria, such as Gardnerella vaginalis (Kyongo et al., 2012). The multitude of these factors may explain why IL-6 levels did not show a high correlation with IL-1β in our sample. On the other hand, TNFα, which similarly to IL-1β, up-regulates proinflammatory cascades leading to IL-6 production, showed a stronger correlation with IL-6 in our cohort and has been shown to up-regulate IL-6 levels in human cervical uterine cells (Fichorova and Anderson, 1999).

Recently, Menon et al. conducted a systematic review of 217 studies and 116 different biomarkers measured in the blood, amniotic fluid, cervical/vaginal secretions, saliva or urine in relation to PTD and was unable to identify any single biomarker that could reliably predict PTD (Menon et al., 2011). These results were similar to a meta-analysis of 72 studies and 30 biomarkers that was unable to find any clinically useful marker for predicting PTD (Conde-Agudelo et al., 2011). We favored examining inflammation biomarkers in the vaginal fluid because sampling is relatively easy and minimally invasive. In addition, if inflammation is localized and not systemic, the vaginal fluid may be more informative than maternal blood. Detection of elevated risk at mid-trimester might allow time for intervention before the onset of PTD. However, in our sample, IL-6 levels lacked adequate sensitivity despite the moderately high specificity. These results are similar to those of other studies indicating that IL-6 is a specific but non-sensitive marker for the detection of spontaneous PTD (Lockwood et al., 1994; Goepfert et al., 2001; Paternoster et al., 2002). The high specificity may mean that IL-6, along with other biomarkers, could be used to establish risk of early spontaneous PTD. For example, women who have low fetal fibronectin and IL-6 below the established 75th percentile threshold may be considered at low risk for early spontaneous PTD. Future studies are needed to explore the utility of IL-6 in combination with other biomarkers.

PTD pathways are complex and influenced by environmental factors, genetics and population characteristics. Therefore, it is suggested that a single biomarker cannot accurately predict all PTD subtypes in all populations (Conde-Agudelo et al., 2011; Menon et al., 2011). We examined the use of multiple inflammation biomarkers for detecting PTD subtypes. Our results showed that the addition of elevated (>75th percentile) TNF-α, GM-CSF, IL-1β or IL-6r to IL-6 was able to slightly increase the specificity to detect PTD with HCA. A similar pattern was noted when elevated (>75th percentile) IL-1β was added to elevated IL-6 for detecting spontaneous PTD at <35 weeks. However, our pairs of inflammation biomarkers were no more sensitive that IL-6 alone to detect PTD subtypes. Vaginal inflammation biomarkers could perhaps be combined with other biomarkers or risk factors, such as microbial factors, for better prediction of PTD. Although the combination of elevated IL-6 and BV did not increase sensitivity in our study, this avenue of research should be further explored with specific BV-related organisms.

Our study had several strengths. We were able to prospectively examine inflammation biomarker levels in women with PTD subtypes defined by clinical circumstance, week of delivery and placental pathology. Placental examination was blinded to the outcome measures as well as other clinical variables. The POUCH study recruited women from multiple communities including rural, urban and suburban settings. However, our sample size was modest for less common PTD subtypes (i.e. spontaneous delivery <35 weeks, PTD accompanied by HCA), and limited our statistical power to detect significant associations at OR <2.0. Other analytic methods, such as ROC curves, can be used to identify a diagnostic threshold and examine the predictive power of inflammation biomarkers. We chose not to apply the ROC method because our small sample sizes would lead to inaccurately estimated validation parameters in ROC analyses, especially when poorly distributed biomarkers are used (Hanczar et al., 2010). In addition, our ability to find small effective size associations between PTD and inflammation biomarkers (OR >2.0) might have been limited by the relatively higher variability for some cytokines (e.g. TNFα, IL-12p70, IL-10 and GM-CSF) that are less stable and have commonly been found either undetectable or unreliably detectable at low levels (median <1 pg/ml) in human vaginal secretions (Scott et al., 2011; Kyongo et al., 2012), which diminishes their value as biomarkers.

Our results suggest that cytokines may play a role in PTD accompanied by HCA. However, elevated vaginal IL-6 alone or IL-6 in combination with other elevated vaginal cytokines measured at mid-trimester are insufficient predictors of PTD subtypes. Large prospective studies or aggregation of stored samples from multiple studies will be needed to further explore inflammation biomarkers in combination with other risk factors for the prediction of PTD subtypes.

Supplementary data

Supplementary data are available at http://humrep.oxfordjournals.org/.

Authors’ roles

B.T. was involved in the concept of the study, analysis and interpretation of the data, and wrote the manuscript. C.H. was involved in the design and concept of this study and the original POUCH study, data acquisition, analysis and interpretation of data and provided significant revisions. R.F. was involved in the concept of the study, acquisition of the inflammation biomarker data, interpretation of data and provided significant revisions. Y.T. was involved in the data analysis, drafting portions of the methods section and significant revisions. N.J. was involved in the concept of the study and significant revisions. W.F. contributed to the analysis of the data and provided significant revisions. P.K. was involved in the design and concept of the original POUCH study, acquired the placental histology data and provided significant revisions. All authors approved the final version of the paper.

Funding

This work was supported by the National Institute of Child Health and Human Development grant number R01 HD034543, National Institute of Nursing Research (Renewal NIH POUCH) grant number R01 HD34543, March of Dimes Foundation (Perinatal Epidemiological Research Initiative Program) Grants 20-FY98-0697 through 20-FY04-37, Thrasher Research Foundation grant 02816-7 and Centers for Disease Control and Prevention grant U01 DP000143-01.

Conflict of interest

None declared.