Compartmentalized profiling of amniotic fluid cytokines in women with preterm labor

Objective Amniotic fluid cytokines have been implicated in the mechanisms of preterm labor and birth. Cytokines can be packaged within or on the surface of extracellular vesicles. The main aim of this study was to test whether the protein abundance internal to and on the surface of extracellular vesicles changes in the presence of sterile intra-amniotic inflammation and proven intra-amniotic infection in women with preterm labor as compared to the women with preterm labor without either intra-amniotic inflammation or proven intra-amniotic infection. Study design Women who had an episode of preterm labor and underwent an amniocentesis for the diagnosis of intra-amniotic infection or intra-amniotic inflammation were classified into three groups: 1) preterm labor without either intra-amniotic inflammation or proven intra-amniotic infection, 2) preterm labor with sterile intra-amniotic inflammation, and 3) preterm labor with intra-amniotic infection. The concentrations of 38 proteins were determined on the extracellular vesicle surface, within the vesicles, and in the soluble fraction of amniotic fluid. Results 1) Intra-amniotic inflammation, regardless of detected microbes, was associated with an increased abundance of amniotic fluid cytokines on the extracellular vesicle surface, within vesicles, and in the soluble fraction. These changes were most prominent in women with proven intra-amniotic infection. 2) Cytokine changes on the surface of extracellular vesicles were correlated with those determined in the soluble fraction; yet the magnitude of the increase was significantly different between these compartments. 3) The performance of prediction models of early preterm delivery based on measurements on the extracellular vesicle surface was equivalent to those based on the soluble fraction. Conclusions Differential packaging of amniotic fluid cytokines in extracellular vesicles during preterm labor with sterile intra-amniotic inflammation or proven intra-amniotic infection is reported herein for the first time. The current study provides insights into the biology of the intra-amniotic fluid ad may aid in the development of biomarkers for obstetrical disease.

Proteins present in amniotic fluid, in particular cytokines, are key regulators of parturition, and labor-associated changes in their concentrations, with or without infection at both term and preterm, have been well characterized . Until recently, regulatory activity of these proteins was considered to be mediated via soluble autocrine [63][64][65][66], paracrine [63,65,67], and endocrine [68-70] signaling pathways, by direct engagement with cell-surface receptors. However, it is now established that such mediators are also associated with extracellular vesicles (both ectosomes and exosomes) and are present both on the surface and within the lumen of vesicles [71][72][73][74]. Extracellular vesicle-associated proteins, therefore, represent an additional, and as yet uncharacterized, pathway that may contribute to the initiation of labor and delivery at both term and preterm.
Labor and delivery are associated with significant changes in the protein complement of amniotic fluid extracellular vesicles [86]. There is a paucity of data, however, about the role of amniotic fluid extracellular vesicle-associated proteins in preterm labor and delivery and whether their presentation on extracellular vesicles changes in association with intra-uterine inflammation (with or without proven infection).
Herein, we tested the hypothesis that preterm labor with sterile intra-amniotic inflammation and preterm labor with intra-amniotic infection are associated with an increased expression of cytokines in extracellular vesicles present in amniotic fluid, and that such changes will differ between the different extracellular vesicle compartments (internal, surface) and the soluble fraction. delivery between 32 and 36 6/7 weeks of gestation. Term delivery was defined as birth � 37 weeks of gestation. Intra-amniotic inflammation was diagnosed when the concentration of interleukin 6 (IL-6) in the amniotic fluid was � 2.6 ng/ml [92,93]. Microbial invasion of the amniotic cavity was detected from a positive amniotic fluid culture and/or polymerase chain reaction/electrospray ionization-mass spectrometry (PCR-ESI/MS) result [18,19,[94][95][96][97][98][99][100][101]. Based on the results of the amniotic fluid culture, PCR-ESI/MS, and amniotic fluid IL-6 concentration, patients were classified into the following groups: 1. Preterm labor without either intra-amniotic inflammation or detectable microbes in the amniotic cavity (control group, n = 88); 2. Preterm labor with intra-amniotic inflammation but without detectable microbes in the amniotic cavity (sterile intra-amniotic inflammation [SIAI] group, n = 19); and 3. Preterm labor with detectable microbes in the amniotic cavity and intra-amniotic inflammation (proven intra-amniotic infection [IAI] group, n = 33).

Amniotic fluid samples
Amniotic fluid was retrieved either by transabdominal amniocentesis under antiseptic conditions using a 22-gauge needle monitored by ultrasound, or by amniocentesis during cesarean delivery under antiseptic conditions. Amniotic fluid samples were transported in capped, sterile syringes to the clinical laboratory and were cultured for aerobic and anaerobic bacteria as well as for genital mycoplasmas, as previously detailed [35]. At the time of collection, an amniotic fluid white blood cell count [102,103], a glucose concentration [104] and a Gram stain [105] were performed. Concentrations of IL- 6 [92] were determined either at the time of collection or from frozen plasma samples. Amniotic fluid samples used in this study were collected between 17.1 and 36.4 weeks of gestation and were stored at −70˚C until analysis.

Preparation of extracellular vesicle fractions
Amniotic fluid samples were thawed at room temperature and treated with Exoquick-TC™ (System Biosciences, SBI, Palo Alto, CA, USA) to sediment the extracellular vesicles, according to the manufacturer's instructions. The resulting supernatants free of extracellular vesicles, and extracellular vesicle pellets re-suspended in the original starting volume, were collected for subsequent cytokine measurement.

Cytokine measurement
The concentration of 38 cytokines (see S1 Table) was determined with an in-house multiplexed bead-based assay, using methods similar to those previously described [61]. All antibody pairs and protein standards were purchased from R&D Systems (Minneapolis MN, USA), except those for IFN-β (Abcam, Cambridge, UK). Magnetic beads (Luminex Corporation, Austin, TX, USA) with distinct spectral signatures (regions) were coupled to analyte-specific capture antibodies, according to the manufacturer's recommendations, and stored at 4˚C. Samples and protein standards for the supernatant fluid and intact vesicles were prepared in assay buffer (1% bovine serum albumin in PBS), and lysed samples and standards were prepared in assay buffer with Triton™ X-100 at a final concentration of 0.5%. Samples and standards were combined with analyte-specific capture antibody coupled bead mixtures in 96-well flat bottom plates (Nunc, ThermoFisher Scientific, Waltham, MA, USA) and incubated at 4˚C overnight. Plates were washed on a magnetic plate washer (405 TS, Biotek Winooski, VT, USA), followed by incubation with polyclonal biotinylated anti-analyte antibodies for 1 hour at room temperature. Plates were washed and incubated for 30 minutes with 16 μg/ml of streptavidin-phycoerythrin (ThermoFisher Scientific, Waltham, MA, USA) in PBS. Plates were washed and beads were re-suspended in PBS and read on a Luminex 200 analyzer (Luminex Corporation, Austin, TX, USA) with acquisition of 100 beads for each region and analyzed using Bioplex Manager software (Bio-Rad Laboratories, Hercules, CA, USA). Analyte concentrations (pg/ml) were determined using 5P regression algorithms and expressed as the mean pg/ml ± S.E. Concentrations were corrected for dilution by ExoQuick-TC™ or Triton™ X-100. Extracellular vesicle luminal content was calculated as [analyte content of lysed vesicle] − [analyte content of intact vesicles].

Statistical analyses
Analysis of the demographic data. Continuous variables were compared among multiple groups using the Welch's one-way analysis of variance (ANOVA) [106] or Kruskal-Wallis test, as appropriate. The Fisher's exact test was used to compare proportions in the analysis of contingency tables.
Differences in protein concentration among groups by amniotic fluid compartment. Protein concentrations were offset by adding 1 unit and then log 2 transformed to improve normality before analysis [107]. The transformed concentrations were then compared between pairs of the preterm labor groups, using the Wilcoxon rank sum test. Fold change in protein concentrations between groups was determined based on the median log transformed values in each group.
To test for differences in the magnitude of change among the preterm labor groups between different compartments (extracellular vesicle surface, extracellular vesicle internal, and amniotic fluid soluble fraction), a quantile regression model for repeated measures [108] was used for analysis-the transformed protein concentration was treated as the dependent variable, while the preterm labor group, the amniotic fluid compartment and gestational age at amniocentesis were treated as fixed effects. Differences in the effect of infection or inflammation on the cytokine abundance between compartments were tested by allowing for interaction terms in the regression models. The R package, rqpd, was used to estimate the model coefficients and their significance.
The Benjamini-Hochberg procedure [109] was employed to correct for multiple comparisons, and a q-value (corrected p-value) of less than 0.05 was considered a statistically significant result.
All analyses were performed using software packages within the R statistical environment [110].
Prediction of preterm delivery. An additional analysis was conducted to determine whether amniotic fluid compartmentalized protein abundance measured before 32 weeks of gestation is predictive of the time of delivery (<32 vs �32 weeks of gestation). Random forest prediction models [111] were built for each of the three amniotic fluid compartments separately, as well as using data from all three amniotic fluid compartments at the same time. The randomForest in R was used to fit the models while allowing for 1,000 decision trees in each random forest model. Leave-one-out cross validation was used to estimate prediction performance indicators, including the area under the receiver operating characteristic (ROC) curves (AUC), sensitivity, and specificity. history of previous preterm delivery, and fetal gender were found among the 3 preterm labor groups. Gestational age at delivery, birthweight, 1-and 5-minute Apgar scores, frequency of clinical chorioamnionitis, amniotic fluid white blood cell count, and amniotic fluid glucose concentration varied significantly among the groups, consistent with clinical presentation. Among the patients with preterm labor with intra-amniotic infection, microorganisms were detected in the amniotic fluid through cultivation in 16 (48.5%) cases and through PCR/ ESI-MS in 29 (88%) cases (S2 Table). The most common microorganisms detected included Sneathia spp. (n = 6), Fusobacterium nucleatum (n = 6), Ureaplasma parvum (n = 4), and Ureaplasma urealyticum (n = 4).

Differences in protein concentration by amniotic fluid compartment among the groups
Protein concentrations by preterm labor group and amniotic fluid compartment are summarized as box plots in S1 Fig and data is available in S3 Table. Intra-amniotic inflammation, regardless of the detection of microorganisms, was associated with a pronounced expression of cytokines on the extracellular vesicle surface and in the soluble fraction of amniotic fluid (Fig 1). Pair-wise comparisons of protein concentration by amniotic fluid compartment are presented in Table 2 and illustrated in volcano plots (Fig 2). For some comparisons, even though the median protein concentrations were identical (log 2 fold change = 0) in the comparison groups, the Wilcoxon rank sum test returned a significant p-value (e.g. IL-1α: preterm labor with intra-amniotic infection vs preterm labor without either intra-amniotic inflammation or proven intra-amniotic infection in the internal compartment of vesicles Table 2). In most cases, this was due to non-detection of protein (S4 Table) in majority of the samples resulting in median concentrations being zero for the two groups being compared. However, as was the case for IL-1α (S1 Fig), among the samples in which the protein was detected, the concentrations were higher in one group compared to the other. Therefore, these comparisons were retained. Extracellular vesicle surface. On the surface of extracellular vesicles, median concentrations of 37 proteins were significantly increased (q-value <0.05) in preterm labor with intraamniotic infection compared to preterm labor without either intra-amniotic inflammation or proven intra-amniotic infection. The highest fold change (FC) was observed for MIP-3α (log 2 FC = 10.5) followed by CXCl6 (log 2 FC = 8.6) and Calgranulin C (log 2 FC = 8.6). Median concentrations of 28 cytokines were significantly higher in preterm labor with sterile intra-amniotic inflammation compared to preterm labor without either intra-amniotic inflammation or proven intra-amniotic infection, with the highest fold changes being observed for Calgranulin C (log2 FC = 7.7), MIP-3α (log2 FC = 6.8), and IL-6 (log2 FC = 5.6). The comparison of intraamniotic infection to sterile intra-amniotic inflammation groups identified 28 mediators with increased concentration [e.g., IL-33 (log2 FC = 5.8), IFN-λ (log2 FC = 5.6), and MIP-1α (log2 FC = 4.5), among others].
Extracellular vesicle internal compartment. In the internal compartment of the extracellular vesicles, the median concentrations of 24 cytokines were significantly higher in preterm labor with intra-amniotic infection compared to preterm labor without either intra-amniotic inflammation or proven intra-amniotic infection. MIP-3α (log 2 FC = 8.7), Calgranulin A (log 2 FC = 3.5), and IL-1β (log 2 FC = 2.3) showed the highest increase in concentration in the presence of intra-amniotic infection.

Compartment-dependent differences in protein abundance among three preterm labor groups
Quantile regression models were used to assess if the differences in protein abundance among preterm labor groups were significantly different among the three amniotic fluid compartments. For each pair-wise comparison among the study groups, the differences in log 2 -fold changes between the pairs of amniotic fluid compartments are provided in Table 3. Significant differences highlighted in Fig 3 (scatter plots of paired log 2 fold changes on the differences observed among the amniotic fluid compartments), are summarized below: The extracellular vesicle surface compartment versus the extracellular internal compartment. The changes in protein abundance with intra-amniotic inflammation, regardless of the detection of microorganisms, were more pronounced on the extracellular vesicle surface

IAI vs Control SIAI vs Control IAI vs SIAI IAI vs Control SIAI vs Control IAI vs SIAI IAI vs Control SIAI vs Control IAI vs SIAI
Log 2 fold changes in protein abundance among the three preterm labor groups are presented for each compartment. Statistical significance was assessed by Wilcoxon rank sum test. IAI, preterm labor with intra-amniotic infection; SIAI, preterm labor with sterile intra-amniotic inflammation; Control, preterm labor without either intra-amniotic inflammation or proven intra-amniotic infection Significance code for p-values:  Table 3). For example, when comparing preterm labor with intra-amniotic infection vs preterm labor without either intra-amniotic inflammation or proven intra-amniotic infection, fold changes of 16 proteins were higher when assessed on the extracellular vesicle surface than internal to extracellular vesicles. The extracellular vesicle surface compartment versus the amniotic fluid soluble fraction. A comparison of the data collected on the extracellular vesicle surface and in the soluble fraction of amniotic fluid revealed differences in the fold changes in protein abundance with intra-amniotic infection and with sterile intra-amniotic inflammation as follows: some cytokines showed significantly stronger associations on the extracellular vesicle surface and others in the soluble fraction of amniotic fluid (top panel in Fig 3 and Table 3). For example, when comparing preterm labor with intra-amniotic infection vs preterm labor without either intraamniotic inflammation or proven intra-amniotic infection, fold changes of 15 proteins were significantly higher when assessed on the surface of extracellular vesicles compared to the Volcano plot of-log 10 transformed p-values against log 2 transformed fold changes illustrating the changes in protein abundance with intra-amniotic infection and with sterile intra-amniotic inflammation. Differentially abundant proteins (q-value < 0.05) are highlighted in red. Changes in protein concentrations were most evident when comparing preterm labor with intra-amniotic infection to preterm labor without either intra-amniotic inflammation or proven intra-amniotic infection on the extracellular vesicle surface or in the soluble fraction of amniotic fluid. Control: preterm labor without either intra-amniotic inflammation or proven intra-amniotic infection.

Soluble vs Surface
Gro-α/CXCL-1 Differences in the magnitude of changes in cytokine concentration with intra-amniotic infection and with sterile intra-amniotic inflammation were assessed between pairs of amniotic fluid compartments. Data represent differences in log 2 fold changes (Δ log 2 FC) reported in Table 2. Statistical significance was assessed using quantile regression models. PTL: preterm labor. Significance code for p-values:  Table 3). For e.g., the fold changes in MIP-3α with intra-amniotic infection and with sterile intra-amniotic inflammation were higher within vesicles compared to the soluble fraction of amniotic fluid (preterm labor with intra-amniotic infection vs preterm labor without either intra-amniotic inflammation or proven intra-amniotic infection: Δ log 2 FC = 3.5; preterm labor with sterile intra-amniotic inflammation vs preterm labor without either intraamniotic inflammation or proven intra-amniotic infection: Δ log 2 FC = 1.4). Similarly, for IL-4, the change in protein abundance with intra-amniotic infection vs sterile intra-amniotic inflammation was higher within extracellular vesicles compared to that in the soluble fraction of amniotic fluid (Δ log 2 FC = 1.4).

Prediction of preterm delivery based on compartmentalized proteomic profiles of amniotic fluid
Of the 138 cases for which gestational age at delivery was available, 22 delivered at term (� 37 weeks), 61 delivered early preterm (< 32 weeks), and the 55 delivered late preterm (at or after 32 but before 37 weeks).

Fig 3. Scatter plot of fold changes in protein abundance with intra-amniotic infection and with sterile intraamniotic inflammation between pairs of amniotic fluid compartments.
Proteins for which the changes were significantly different between compartments are highlighted in red. Data represents log 2 fold changes between groups. Control: preterm labor without either intra-amniotic inflammation or proven intra-amniotic infection. For the subset of cases where amniocentesis was performed before 32 weeks of gestation, the predictive value of the compartmentalized cytokine concentrations for early preterm delivery was evaluated by building univariate and multivariate prediction models. The performance of prediction models is summarized in Table 4 that gives the AUC statistic of univariate models by amniotic fluid compartment, in Table 5 that provides the performance metrics (sensitivity, and specificity) of multivariate models, and

Discussion
Pathologic insults to the feto-placental unit are often clinically expressed as complications of pregnancy, such as preterm labor with intact membranes and preterm prelabor rupture of the membranes [113][114][115]. A contemporary paradigm of the initiation of labor and delivery, both at term and preterm, involves the activation of intrauterine inflammatory signaling pathways and a shift to a more pro-inflammatory intra-amniotic environment [35, 60, 61, 96-101, 103, 116-128].
To gain insight into the contribution of proteins packaged in extracellular vesicles, we established the profile and compartmentalization of extracellular vesicle-associated proteins in amniotic fluid obtained from women with preterm labor. We evaluated the distribution of 38 cytokines between the vesicle surface and intra-vesicle compartments. Their concentration in extracellular vesicle-free amniotic fluid was also determined for each sample. These measurements allowed us to test whether there was an advantage to measuring cytokine concentrations in the extracellular vesicle compartments, as opposed to unfractionated amniotic fluid, in predicting preterm delivery. We have specifically addressed the following research questions:

Does the cytokine abundance internal to and on the extracellular vesicle surface change with sterile intra-amniotic inflammation and with intraamniotic infection in women with preterm labor?
We found that intra-amniotic inflammation, regardless of detected microbial invasion of the amniotic fluid cavity, was associated with an increased abundance of cytokines internal to and on the surface of extracellular vesicles. Earlier we reported on the general increase of cytokines [61] in unfractionated amniotic fluid obtained from preterm labor cases with intra-amniotic infection and sterile intra-amniotic inflammation. These results are reconfirmed herein, but here, for the first time we report on changes in cytokines quantified internally and on the surface of extracellular vesicles in this subset of women with preterm labor. Development of the pro-inflammatory vesicle surface was most evident in intra-amniotic infection, for which the surface expression of 37 cytokines was increased. Of these, 36 cytokines were also increased in the soluble fraction of amniotic fluid, and 24 had increased concentrations within the extracellular vesicles.
The top ranked proteins based on the magnitude of change in their concentrations with sterile intra-amniotic inflammation were Calgranulin C, MIP-3α, IL-16, MIP-1b, Groα/ CXCL1, CXCL6, MCP-1, IL-8, CXCL13, and IL-6 on the surface of extracellular vesicles and MIP-3α, and Calgranulin C in the lumen of extracellular vesicles. The overexpression of these inflammatory mediators in the absence of microbes is initiated upon recognition of damageassociated molecular patterns (DAMPs or alarmins) by pattern recognition receptors (e.g. Toll-like receptors, NLRs, etc.) [61, 165,[168][169][170][171]. In this study, we observed that concentrations of alarmins, IL-1α [172], and S100 family proteins (Calgranulin A, and Calgranulin C) [173] were significantly increased in the soluble fraction of the amniotic fluid in preterm labor cases with sterile intra-amniotic inflammation. These findings are in line with previous studies showing that the intra-amniotic administration of alarmins induces sterile intra-amniotic inflammation and preterm birth in mice [174,175]. The mechanisms whereby DAMPs induce sterile intra-amniotic inflammation and preterm birth involve the activation of the NLRP3 inflammasome [99, 124,[175][176][177][178][179] The association of intra-amniotic inflammation with or without detected microbes with preterm parturition syndrome and the role of cytokines in these processes are well documented [60, 61, 93, 96, 97, [180][181][182]. However, this study shows that there is up-regulation in inflammatory mediators secreted not only as free molecules in amniotic fluid but also as part of the cargo carried by extracellular vesicles. The observed expression and the change in abundance of not only free (soluble) cytokines but also extracellular vesicle associated cytokines with intra-amniotic inflammation/infection may represent urgency and redundancy in the feto-placental unit's response to environmental stimuli.
While cytokine changes with inflammation, regardless of detected microbial invasion of the amniotic fluid cavity, on the surface of extracellular vesicles correlated with those determined in the soluble fraction of amniotic fluid, the magnitude of the increase was significantly different between these compartments, suggesting that they include different information about the amniotic fluid milieu. Overall, although protein changes in extracellular vesicle surface and the soluble fraction in intra-amniotic inflammation, regardless of detected micro-organisms, were higher than those within the extracellular vesicles, the magnitude of changes within the extracellular vesicles was higher compared to the soluble fraction of amniotic fluid for MIP-3α.
Cytokines contained within the internal compartment are not accessible to routine immunoassay quantification; therefore, this contribution has not been incorporated in contemporary models of inflammation-associated parturition. While the relative amount of internal mediators is low, their biological effect may be significant as vesicles can deliver these mediators to the very vicinity of the target cells. The roles of these extracellular vesicle -associated mediators in parturition remain to be established. Previously, we and other investigators have established that encapsulation within extracellular vesicles protects against degradation and affords alternative pathways for engagement with target cells and transport pathways [183].

Is compartmentalized amniotic fluid protein concentration profiling useful in the development of biomarkers for early preterm delivery?
To determine the potential of amniotic fluid compartmentalized protein profiling for developing new biomarkers of preterm labor and other obstetrical complications, we have developed univariate and multi-variate prediction models for early preterm delivery (gestational age �32 weeks). For all but one cytokine (TRAIL), the AUC estimates obtained from either the extracellular vesicle surface or the amniotic fluid soluble fraction concentrations were significantly above 0.5. The AUC statistic for univariate models based on data collected on the extracellular vesicle surface for IL-6, IL-8, MCP-1, MIP-1α, and MIP-1β was greater than 0.8. This is consistent with our previous report of the association between unfractionated amniotic fluid concentrations of these proteins and early preterm delivery in patients diagnosed with a short cervix [62]. Although prediction performances (AUC) for individual cytokines were different between compartments, the overall performance of the multivariate models was statistically similar among the different amniotic fluid compartments. This finding can be explained in part by the redundancy of the cytokine network, but it does not preclude that compartmentalized profiling could improve prediction performance for other phenotypes.

Strengths and limitations
The main strength of this study is its novelty component, which is the compartmentalized profiling of proteins in the amniotic fluid for three phenotypes of preterm labor. Another strength is the moderate sample size that enabled not only detection of differences among groups within a given compartment, but also detection of significant differences in the magnitude of changes between groups across the different compartments. The use of cross-validation to avoid overfitting in multi-variate models was also a strength of the analysis. Limitations are related to the variability in the gestational ages at sampling of amniotic fluid, and the lack of amniotic fluid samples collected during second or early third trimester from mothers without preterm labor.

Conclusions
The data obtained in this study are consistent with our hypothesis that amniotic fluid proteins are differentially expressed and grouped within the extracellular vesicle and soluble amniotic fluid fraction, and compartment-specific profiles characterize clinical subgroups of preterm labor. Intra-amniotic inflammation with and without detectable microorganisms was associated with the differential packaging of extracellular vesicles cytokines. Proteins packaged within the extracellular vesicles contribute to total amniotic fluid concentrations. Although the current study points to possible improvement in biomarker prediction based on compartmentalized profiling, further de-convolution of the heterogeneity of preterm labor and delivery is warranted.