Abstract

   Glucocorticoid resistance (GR) is associated with exposure to chronic
   stress and an increased risk of metabolic and inflammatory disorders in
   both animal and human populations. Studies on ethnic disparities
   highlight the African-American (AA) population as having a high
   propensity to both GR and chronic stress exposure. Glucocorticoids and
   inflammation play a very important role in pregnancy outcome and fetal
   development. To date, however, the metabolites and metabolic pathways
   associated with GR during pregnancy have not been identified, obscuring
   the mechanisms by which adverse health consequences arise, and thus
   impeding targeted therapeutic intervention. The objective of this study
   was to perform untargeted high-resolution metabolomics (HRM) profiling
   on 273 pregnant AA women, to identify metabolites and metabolic
   pathways associated with GR during the first trimester of pregnancy and
   to evaluate their cross-sectional association with birth outcomes and
   psychosocial variables related to chronic stress exposure. For this
   study, GR was determined by the concentration of dexamethasone required
   for 50% inhibition (Dex IC50) of the cytokine tumor-necrosis factor
   alpha (TNF-alpha) release in vitro in response to a standard dose of
   lipopolysaccharide. The results for Metabolome-Wide Association Studies
   (MWAS) and pathway enrichment analysis for serum metabolic associations
   with Dex IC50, showed energy (nicotinamide and TCA cycle), amino acid,
   and glycosphingolipid metabolism as top altered pathways. Bioinformatic
   analysis showed that GR, as indicated by elevated Dex IC50 in the
   pregnant women, was associated with increased inflammatory metabolites,
   oxidative stress related metabolites, increased demand for functional
   amino acids to support growth and development, and disruption in
   energy-related metabolites. If confirmed in future studies, targeting
   these physiologically significant metabolites and metabolic pathways
   may lead to future assessment and intervention strategies to prevent
   inflammatory and metabolic complications observed in pregnant
   populations.

   Keywords: Metabolomics, Glucocorticoid resistance, Health disparity,
   Pregnancy, Stress, Metabolites, Inflammation

Highlights

     * •
       GR is associated with chronic stress and is a risk factor for
       adverse health outcomes, especially among African Americans.
     * •
       Metabolites and metabolic pathways associated with GR relate to
       energy production, amino acid metabolism, and inflammation.
     * •
       Findings provide a foundation for future studies investigating risk
       factors in this health disparity population.

1. Introduction

   Glucocorticoids (GC) are steroid hormones under the influence of the
   hypothalamic-pituitary-adrenal (HPA) axis. GC mediate their actions by
   binding to glucocorticoid receptors, becoming a ligand dependent
   transcription factor that regulates key physiological events through
   target gene modulation in inflammation, glucose and fat metabolism, and
   cellular differentiation [[34]1]. An important clinical pathological
   outcome observed is glucocorticoid resistance (GR). GR is identified in
   vitro by an increase in the dose of dexamethasone required to inhibit
   50% (i.e., the IC[50]) of pro-inflammatory cytokines such as tumor
   necrosis factor alpha (TNF-α), in the presence of the inflammatory
   agent bacterial lipopolysaccharide (LPS) [[35][2], [36][3], [37][4]].
   Mechanisms for the manifestation of GR are still under investigation,
   however physiological stimuli such as chronic stress can activate the
   HPA axis thereby regulating plasma GC levels and GC action
   [[38]5,[39]6]. Genetic and epigenetic studies have shown that early
   exposure to altered glucocorticoid response (as early as beginning in
   utero) programs changes in neuroendocrine and immune mechanisms
   [[40]7,[41]8]. These altered glucocorticoid responses may further lead
   to increased vulnerability to adverse health outcomes later in life
   such as asthma [[42]7]; insulin resistance [[43]9]; inflammatory
   disease [[44]10], metabolic syndrome [[45]11], and potentially, adverse
   pregnancy outcomes [[46]12,[47]13]. Hence understanding what
   metabolites and metabolic pathways are associated with GR becomes
   necessary, in order ultimately to delineate mechanisms that lead to GR
   and identify therapeutic targets to prevent GR and its consequences.

   Recent advances in metabolomics offer an opportunity to take this next
   step. Metabolomics includes multiple technologies that allow the
   identification of specific metabolites and complex metabolic pathways
   associated with health outcomes of interest [[48]14,[49]15]. A recent
   state-of the-science White Paper, emphasizes that an individual’s
   metabolic state, that is, the metabolites and metabolic pathways
   present, reflects the existing complex interactions between genome,
   proteome, diet, environmental exposures, and the gut microbiome
   [[50]16]. These metabolites and pathways provide a real-time phenotype
   that can be screened to identify metabolic trends associated with
   adverse outcomes, potentially leading to early and targeted assessment
   and intervention strategies that influence those metabolites and
   pathways [[51]15]. Metabolomics has been applied previously to identify
   biomarkers that associate with adverse pregnancy outcomes, including
   preterm birth, gestational diabetes mellitus, preeclampsia and small
   for gestational age (SGA) [[52][17], [53][18], [54][19], [55][20]], but
   has not been utilized to identify pathways related to GR or other
   measures of chronic stress exposure in pregnant women.

   Importantly, ethnic disparities have been observed in glucocorticoid
   induced insulin resistance, wherein African Americans were more insulin
   resistant than Caucasian Americans under both placebo and dexamethasone
   treatments [[56]9]. Compared to Caucasians, the National Academy of
   Medicine [[57]21] and others (Kramer et al., 2011a; Mutambudzi et al.,
   2017; Parker and Douglas, 2010), have also identified stress as a key
   contributor to the health disparity in pregnancy outcomes experienced
   by AA women. Based on these ethnic disparities, and the future risk for
   GR-induced symptoms of metabolic syndrome in AA women, the present
   study was focused on utilizing metabolomics to identify metabolites and
   metabolic pathways associated with graded Dex IC50 response within the
   first trimester in AA women and to evaluate cross-sectional
   associations in light of personal factors such as age, body mass index
   (BMI), parity, and insurance status, and psychosocial measures of
   stress including adverse childhood experiences and exposure to
   discrimination. The potential to expose metabolites and metabolic
   pathways associated with GR and other measures of stress exposure
   during the first trimester of pregnancy, and to investigate their
   associations with pregnancy outcomes, will ultimately increase our
   ability to develop, deliver, and evaluate targeted clinical
   interventions that will improve maternal and fetal/infant health
   outcomes.

2. Materials and methods

   Study population. Participants were drawn from women enrolled in the
   Emory University African American Vaginal, Oral, and Gut Microbiome in
   Pregnancy Cohort [[58]22], for which pregnant US-born women who
   self-identify as African American (AA) are recruited from two hospitals
   in Atlanta, GA: a private hospital affiliated with Emory University
   that provides services for a socioeconomically and educationally
   diverse population and a county-supported hospital that provides
   services to the low-income and underserved. Both are staffed by Emory
   Obstetrics and Gynecology faculty. AA women presenting to the
   hospital-affiliated clinics for prenatal care between 8-14 weeks
   gestation, as determined by standard criteria based upon last menstrual
   period and/or first trimester ultrasound, were offered enrollment if
   they met the following inclusion criteria: 1) US-born black by
   self-report; 2) between 8-14 weeks gestation with a singleton pregnancy
   (verified by clinical record); 3) able to comprehend written and spoken
   English; 4) between 18-40 years of age; and 5) experiencing no chronic
   medical condition or taking prescribed chronic medications aside from
   prenatal vitamins (verified by clinical record). Women selected for
   inclusion in this study (N ​= ​273) are a subsample of those enrolled
   into the Emory University African American Vaginal, Oral, and Gut
   Microbiome in Pregnancy Cohort between June, 2014, and January, 2017,
   for whom sufficient venous blood sample was available for determination
   of GR and metabolomics assays. The present study was approved by the
   Emory University Institutional Review Board; all participants provided
   written informed consent prior to enrollment and participation.

   Sample collection and clinical data. At the 8-14 week enrollment
   encounter, participants completed questionnaires and provided
   biological samples as described in detail previously [[59]22]. Briefly,
   women were informed of the study during their first prenatal visit by
   the prenatal clinic receptionist and those interested in learning more,
   then met with the study coordinator. After hearing details of the study
   and initial inclusion criteria, women who continued to express interest
   in participating provided informed consent after which they were
   accompanied to the clinical laboratory where a venous blood sample, in
   addition to that required for their initial prenatal visit, was
   collected for later analysis of in vitro WBC Dex IC50, an indicator of
   GR [[60]3,[61]23,[62]24] and; (5) high-resolution serum metabolomics.
   After completing all biological sample collections, women returned to
   the waiting room where they next completed demographic and health
   questionnaires providing information on: (1) Maternal race, based on
   self-report; (2) Maternal age, parity, prenatal health insurance status
   (Medicaid, private) at the time of enrollment; (3) Psychosocial
   measures relevant to the analyses of this study including participants’
   exposure to chronic stress and racial discrimination; (4) Fetal age at
   the time of the 1st trimester prenatal exam as determined by fetal
   ultrasound and/or last menstrual period; (5) Infant gestational age at
   delivery or the occurrence of a spontaneous abortion (e.g.,
   miscarriage), defined as the loss of a fetus before the 20th week of
   pregnancy, and ascertained after delivery from the prenatal record by
   medical chart abstraction using a standardized chart abstraction tool;
   and (6) Body Mass Index (BMI) calculated from patients’ measured height
   at the first prenatal visit and patient-report of pre-pregnancy weight
   at the first visit.

   In Vitro Dexamethasone Resistance Assay. Blood collected at the first
   prenatal visit (weeks 8–14 of gestation) was processed and analyzed to
   evaluate GR as described previously [[63]25]. Briefly, whole blood was
   diluted 10:1 with sterile phosphate buffered saline, and 450 ​μL of
   this mixture was incubated in duplicate with LPS (from Escherichia Coli
   Serotype 055:B5, Millipore-Sigma; final concentration 30 ​ng/ml) for
   24 ​h in the absence or presence of serial dilutions of dexamethasone
   (Millipore-Sigma, 10^-9 to 10^-6 ​M) in 24-well cell culture plates at
   37 ​°C in 5% CO[2]. Culture well contents were centrifuged at 1000 X g
   for 10 ​min, and supernatants aliquoted and stored at −80 ​°C until
   assayed by ELISA (BD Biosciences) for TNF-α, a frequently used
   indicator cytokine, including in 50% of the GR-related studies
   identified in a recent meta-analysis [[64]4]. The DEX concentration
   required to suppress 50% of the LPS-induced cytokine response was
   determined from supernatant concentrations of TNF-α using dose-response
   curves fitted by Prism (GraphPad Software, La Jolla, CA).

   High-Resolution Metabolomics. High-resolution metabolomics (HRM) was
   completed using established methods [[65]26,[66]27]. A total of 50 ​μL
   of serum sample was treated with 100 ​μL of ice-cold LC-MS grade
   acetonitrile (Sigma Aldrich) containing 2.5 ​μL of internal standard
   solution with eight stable isotopic chemicals selected to cover a range
   of chemical properties. Following addition of acetonitrile, serum was
   equilibrated for 30 ​min on ice, and the precipitated proteins were
   removed by centrifugation (16.1 × g at 4 ​°C for 10 ​min). Resulting
   supernatant was maintained at 4 ​°C until analysis. Serum samples were
   analyzed in batches of 20; each batch included duplicate analysis of
   pooled human plasma (QStd) for quality control purposes.

   Samples were analyzed in triplicate with 10 ​μL injections using
   hydrophilic interaction liquid chromatography (HILIC) column in
   positive electrospray ionization (ESI) mode (Thermo Scientific Dionex
   Ultimate 3000RSLCnano, Thermo Scientific Orbitrap Fusion Tribrid Mass
   Spectrometer). The high-resolution mass spectrometer was operated in
   full scan mode at 120,000 resolution and mass-to-charge ratio (m/z)
   range 85–1275. Raw data files were extracted and aligned using apLCMS
   [[67]28] with modifications by xMSanalyzer [[68]29]. Uniquely detected
   ions consisted of accurate mass m/z, retention time and ion abundance,
   referred to as m/z features. Prior to data analysis, m/z features were
   batch corrected using ComBat [[69]30].

   Metabolite Annotation. Metabolic features were annotated using with
   xMSannotator using HMDB database, wherein the confidence scores for
   annotation are derived from a multistage clustering algorithm [[70]31].
   Further identification of the selected metabolites by criteria of
   Schymanski [[71]32] in combination with xMSannotator [[72]31], included
   metabolite identification through matching to both m/z and retention
   time of a previously characterized authenticated chemical standard
   (level 1); annotated compounds matching m/z from databases wherein M+H
   is present and isotopes co-elute (level 2); annotated compounds
   matching m/z from databases wherein M+H is present and pathway
   metabolites present (level 3); m/z match from database within 5ppm
   tolerance with unique identifier (level 4). Database used for
   annotation include HMDB (Human Metabolome Database,
   [73]http://www.hmdb.ca/) [[74]33] and KEGG (Kyoto Encyclopedia of Genes
   and Genomes, [75]http://www.genome.jp/kegg/) [[76]34]. Additional
   manual search was done using METLIN
   ([77]http://metlin.scripps.edu/index.php) at 5 ​ppm tolerance [[78]35].

   Bioinformatics analysis. A total of 16,481 ​m/z features were detected,
   which were further filtered to retain only features with non-zero
   values in >80% in all individuals and >50% in all technical replicates.
   The remaining 8830 ​m/z features were log2 transformed, quantile
   normalized and used for further regression based statistical analysis.
   A linear regression model adjusted for demographic covariates such as
   age, BMI and parity, was used to select metabolic features associated
   with increasing GR (Dex IC50) in a metabolome-wide-association study
   (MWAS). Threshold at P ​≤ ​0.05 without adjustment for multiple
   comparisons and for q ​≤ ​0.05 after correction for false discovery
   rate (FDR) was applied. FDR correction protects against type 1 error
   but not type 2 statistical error. Significant features (raw P ​< ​0.05)
   were further studied by pathway enrichment analyses using mummichog
   1.0.9 [[79]36]. This approach protects against type 2 statistical error
   and protects against type 1 statistical error by permutation testing in
   pathway enrichment analysis by including all features at P ​< ​0.05
   [[80]37]. Hierarchical cluster analysis (HCA) and partial least square
   discriminant analysis (PLS-DA) were used for comparison of the 1094
   significant features obtained from lmreg model and plotted for quintile
   1 (n ​= ​55, Dex IC50 ​< ​5348 pM) and quintile 5 (n ​= ​55, Dex
   IC50 ​> ​12898 pM) (raw P ​< ​0.05) using Metaboanalyst [[81]38].
   Boxplots were plotted in R v.3.2.3.

   Statistical Analyses. Descriptive analyses were conducted for the study
   population with median (range) and percentages (n) computed for
   demographic variables including age, BMI, parity and insurance status,
   each previously shown to impact pregnancy outcomes [[82][39], [83][40],
   [84][41], [85][42], [86][43]]. Spearman correlation was used to test
   for associations between Dex IC50 and demographic data, while crude
   odds ratio (OR) and adjusted OR were calculated using logistic
   regression modeling for pregnancy outcomes and psychosocial variables.
   Dex IC50 was used as a continuous variable in the linear regression
   analyses.

3. Results

3.1. Demographics and clinical parameters

   A total of 273 AA women at 8-14 weeks of gestation met criteria for
   inclusion in this study. Demographic and clinical characteristics for
   all 273 AA pregnant women, including serum Dex IC50 levels, maternal
   age in the first trimester, body mass index, parity, insurance status
   and education are summarized in [87]Table 1. The concentration of
   dexamethasone required to suppress 50% of the LPS-induced TNF-α
   response (Dex IC50), a measure for GR, is depicted as a bar plot for
   each of the participating subjects ([88]Fig. 1) and the range of values
   are presented as quintiles in [89]Table 1. The range of Dex IC50 values
   for the study population was 125-fold and the median (range) was 9524
   pM (246 - 30,888 pM). The median Dex IC50 for Quintile Q1, was 3593 pM
   and highest quintile, Q5 was 16,504 pM. Demographic survey results show
   that the median age for the study sample was 24 years; 17.9% of the
   subjects were greater than 30 years of age, and the majority were in
   the age range of 20–29 years (69.6%). The median BMI was 26.1 ​kg/m^2
   and ranged from 16.5 to 56.5 ​kg/m^2; with 54.6% of the study
   population overweight. The study sample comprised of 48.7% subjects
   without previous birth history. A total of 77.3% of the study
   population had medical insurance from the government and 22.7% had
   private medical insurance. The majority of subjects either graduated
   high school (35.9%) or attended some college/technical school (86%).
   None of the demographic variables reported were correlated with serum
   Dex IC50 values.

Table 1.

   Demographic and clinical characteristics of the study population.
   (Pregnant AA women, n ​= ​273, ∗P ​< ​0.05).
   Characteristics Results Quintile 1 Quintile 5
   Number of subjects 273 55 55
   DexIC[50] (pM)[90]^a 9524
   (246–30,888) 3593
   (246–5348) 16504 ∗
   (12898–30888)
   Age, y[91]^a 24 (18–35) 24 (19–35) 23 (18–35)
    <20 y[92]^b 34 (12.5%) 5 8
    20–29 y[93]^b 190 (69.6%) 31 37
    ≥30 y[94]^b 49 (17.9%) 14 8
   BMI (kg/m^2)[95]^a 26.1 (16.5–56.5) 26.6 (18.4–56.5) 24.6 (16.5–45.6)
   Parity[96]^b
    0 133 (48.7%) 22 28
    1 79 (28.9%) 20 14
    2 39 (14.3%) 9 6
    3 18 (6.6%) 3 6
    4 4 (1.5%) 1 1
   Insurance Status[97]^b
    Government Insurance 211 (77.3%) 44 42
    Private Insurance 62 (22.7%) 11 13
   [98]Open in a new tab
   ^a

   median (range).
   ^b

   n (%).

Fig. 1.

   [99]Fig. 1
   [100]Open in a new tab

   Glucocorticoid resistance in pregnant African-American women. Dex IC50
   in pM concentration required to suppress 50% of the LPS-induced
   cytokine response is plotted for each of the 273 pregnant
   African-American women. Each bar represents an individual, with the
   lowest Dex IC50 value recorded at 246 pM and highest at 30,888 pM.
   Details on the Dex IC50 values for each individual along with age, BMI
   and parity are provided in [101]Supplemental Table 1.

3.2. Metabolome wide association study of glucocorticoid resistance

   In order to investigate the global serum metabolic alterations
   associated with GR, we performed a metabolome wide association study of
   untargeted serum metabolic features identified by HRM. Using a linear
   regression model with Dex IC50 values as a continuous variable, a total
   of 75 out of 8830 were found as discriminatory features at a false
   discovery rate (FDR) of 0.25 and 1092 ​m/z features at a raw P
   value ​< ​0.05. This selection was adjusted for demographic covariates
   such as age, BMI and parity. The Dex IC50 values, age, BMI and parity
   for each individual subject is provided in [102]Supplemental Table 1.
   The type 1 Manhattan plot (-log[10] ​P versus m/z) shows that these
   discriminatory features comprise a broad range of m/z ([103]Fig. 2A),
   wherein red represents features that are positively and blue represents
   features negatively associated with Dex IC50. The top dotted horizontal
   line marks FDR 0.25 threshold and the bottom dotted line represents raw
   P value ​< ​0.05 threshold ([104]Fig. 2). Two thirds of all the 1092
   discriminatory features were negatively associated with Dex IC50. Also
   73% of features with an m/z less than 400 were positively associated
   with Dex IC50, while 84% of features greater than m/z 400 were
   negatively associated. The type 2 Manhattan plot (-log[10] ​P versus
   chromatographic retention time) showed that 62% of the discriminatory
   features elute within 60 ​s ([105]Fig. 2B). 78% of these early eluting
   features were negatively associated with Dex IC50, while the later
   eluting features comprised mostly of positively associated metabolic
   features, consistent with an amino acid elution pattern. The mass to
   charge, retention time and P-value for the 1092 discriminatory
   metabolites is provided in [106]Supplemental Table 2.

Fig. 2.

   [107]Fig. 2
   [108]Open in a new tab

   Metabolome wide association study of chronic stress. (A) Type 1
   Manhattan plot shows the negative log P (-log P) for each metabolite
   (m/z features) as a function of the mass to charge ratio (m/z). (B)
   Type 2 Manhattan plot shows the -log P for each metabolite as a
   function of retention time in seconds. Positively associated m/z
   features are represented as red dots and negative associations are
   represented as blue dots. The top dashed horizontal line represents an
   FDR cutoff of 25% using Benjamini-Hochberg correction and the lower
   dashed horizontal line represents a cutoff at raw P ​< ​0.05. Details
   on the discriminatory metabolites are provided in [109]Supplemental
   Table 2. (For interpretation of the references to colour in this figure