Epigenetic biomarker for preeclampsia-associated preterm birth and potential preventative medicine

Abstract

Preterm birth (PTB) has dramatically increased within the population (i.e. >10%) and preeclampsia is a significant sub-category of PTB. Currently, there are no practical clinical parameters or biomarkers which predict preeclampsia induced PTB. The current study investigates the potential use of epigenetic (DNA methylation) alterations as a maternal preeclampsia biomarker. Non-preeclampsia term births were compared to preeclampsia PTBs to identify DNA methylation differences (i.e. potential epigenetic biomarker). Maternal buccal cell cheek swabs were used as a marker cell for systemic epigenetic alterations in the individuals, which are primarily due to environmentally induced early life or previous generations impacts, and minimally impacted or associated with the disease etiology or gestation variables. A total of 389 differential DNA methylation regions (DMRs) were identified and associated with the presence of preeclampsia. The DMRs were genome-wide and were predominantly low CpG density (<2 CpG/100 bp). In comparison with a previous PTB buccal cell epigenetic biomarker there was a 15% (60 DMR) overlap, indicating that the majority of the DMRs are unique for preeclampsia. Few previously identified preeclampsia genes have been identified, however, the DMRs had gene associations in the P13 K-Akt signaling pathway and metabolic gene family, such as phospholipid signaling pathway. Preliminary validation of the DMR use as a potential maternal biomarker used a cross-validation analysis on the samples and provided 78% accuracy. Although prospective expanded clinical trials in first trimester pregnancies and clinical comparisons are required, the current study provides the potential proof of concept a preeclampsia epigenetic biomarker may exist. The availability of a preeclampsia PTB maternal susceptibility biomarker may facilitate clinical management and allow preventative medicine approaches to identify and treat the preeclampsia condition prior to its occurrence.

Introduction

Preterm birth (PTB) is the world’s leading cause of death among children <5 years of age. In the USA, preterm delivery (< 37 weeks) now occurs in over 10% of all pregnancies [1]. The impact of PTB on the offspring health and later life development and pathology of the individual is dramatic [2]. In addition to lower birth weights, respiratory distress, and hypoglycemia [3], there are many long-term adverse medical and social consequences of PTB [4]. A significant decline in mathematical and IQ scores with every week <38 weeks suggests a failure to achieve full neurodevelopment by the time of birth creates a lifelong impact on all infants born prematurely, especially those born <34 weeks gestation [5]. Hypertensive disorders of pregnancy (including preeclampsia), complicates 10% of pregnancies and accounts for 8–10% of all PTBs [6]. Preeclampsia is also increasing in frequency with obesity having become the leading attributable risk (30% of obese pregnancies) [7]. Diabetes is also a risk factor for preeclampsia occurring in 15–20% of pregnancies in women with type 1 diabetes and 10–14% of pregnancies in women with type 2 diabetes [8]. Preeclampsia is a clinically diagnosed progressive disorder comprised of onset of hypertension, with or without proteinuria, and less frequent liver and kidney dysfunction. Other hypertensive disorders of pregnancy (chronic hypertension and gestational hypertension) can evolve into preeclampsia if proteinuria or other severe features develop: (i.e. neurological symptoms, pulmonary edema, thrombocytopenia, liver dysfunction and placental insufficiency). The onset of preeclampsia most often occurs after 20-week gestation and except for family history or history of prior hypertensive disorder of pregnancy, obstetrical clinicians cannot easily predict that a pregnancy is at risk for preeclampsia. There are few other clinical parameters or biomarkers for the disease. These include the prediction model of sFlt-1/PIGF ratios and use of ambulatory blood pressure monitoring in preeclampsia [9, 10]. Clinical risk factors for preeclampsia currently accepted include: previous preeclampsia, chronic renal disease, chronic hypertension, diabetes mellitus, systemic lupus erythematosus or antiphospholipid syndrome, first pregnancy, maternal age >35 years, body mass index >35 kg/m², inter-pregnancy interval >10 years, and family history of preeclampsia [11]. Few molecular biomarkers have been studied or identified [12]. Placental growth factor predicts time to delivery in women with signs or symptoms of early preterm preeclampsia using a prospective multicenter study [12]. On a physiological level, there appears to be inadequate vascularization in the developing placenta and uterine environment, along with the proposed involvement of abnormal immune parameters [13]. However, these specific proteins or lipid metabolites have not been found to be useful as a preeclampsia diagnostic. The links between high blood pressure and preeclampsia to PTB have also not been clarified on a molecular or physiological level. Many individuals with chronic high blood pressure do not have PTBs [14]. Therefore, the molecular etiology and clinical factors associated with PTB remain to be elucidated. A recent review of 125 studies found that “no single molecular biomarker presents sufficient clinical sensitivity and specificity for screening late-onset and term preeclampsia” [15].

PTB primarily manifests during the third trimester of pregnancy, often with no major clinical parameter associated with the PTB [16]. Although the presence of obesity and high blood pressure are risk factors that can be linked to preeclampsia PTB, there are no significant factors that can be used as a diagnostic or predictive tool in the first or second trimester periods [17]. The potential presence of an early-stage diagnostic could be used to improve the clinical management of the pregnancy and delay or prevent the onset of the PTB [18]. The types of preventative clinical management can include novel therapeutics and healthcare management strategies [18]. Preventative medicine approaches for PTB include anticipatory treatments, such as close maternal/fetal monitoring, bedrest, and administration of neuroprotective magnesium and prenatal steroids, have been shown to markedly improve neonatal outcomes, especially for extreme premature infants. These preventive and mitigating interventions remain unused in a persistently high number of preterm infants in whom preeclampsia was not detected until it was too late to do more than an emergency delivery. There has been significant effort to identify specific genetic markers [19], metabolites [20], and blood diagnostics [21] with minimal success to use molecular biomarkers for PTB and allow a preventative medicine approach.

Epigenetic alterations occur at very high frequency compared to genetic mutations that generally are represented in at best 1% of the diseased population [22]. Epigenetic biomarkers for various diseases have now been developed and can provide much higher (e.g. 90%) frequency predictions of disease susceptibility [23]. Epigenetic analysis allows for the use of systemic cells, such as buccal cheek cells, to be used to assess early life or previous generation impacts that promote disease susceptibility [23]. The systemic cell is defined as a cell type that is not associated with the disease etiology, can be impacted through epigenetic inheritance [23], and is less responsive to environmental factors in regard to its epigenetics. For example, fibroblasts or skin cells are generally not responsive to medications, toxicants, or gestational factors [23]. The buccal cell has been shown to be a useful systemic biomarker cell to identify various disease susceptibilities [23], and is not responsive to physiological variables such as gestation. The alterations in buccal cell epigenetics will be derived primarily from early life developmental exposures or through epigenetic transgenerational inheritance of previous generation exposures [23, 24]. Systemic biomarker cells, like buccal cells, can be potentially used to reduce confounder clinical impacts since they are not involved in the disease etiology. Recently, a potential epigenetic maternal biomarker for PTB has been developed [25]. The presence of such an epigenetic biomarker can now be potentially used to facilitate preventative medicine approaches in the early stages of pregnancy to identify those susceptible for PTB later in pregnancy. The current study was designed to identify epigenetic biomarkers for a subset of PTBs that are due to preeclampsia. A previous study [25] had a small number of preeclampsia patients, so the current study was performed with a larger preeclampsia cohort to determine if a preeclampsia maternal biomarker may exist that is distinct from the previous PTB biomarker. Although the current study provides the proof of concept that such an epigenetic biomarker may exist, expanded prospective clinical trials and first trimester testing is now needed to extend the current study and demonstrate the potential utility of the epigenetic biomarker. The current study with post-birth samples is required to assess the potential presence of a preeclampsia biomarker. A future study with the use of first trimester buccal analysis is now needed to validate the preeclampsia biomarker.

The maternal marker cell used for the current study involved the cheek swab buccal cell, which is not involved in the etiology of preeclampsia. Therefore, observations indicate a systemic impact of environmentally induced early life or ancestral origin of the pathology. Therefore, the disease susceptibility appears to be derived from environmentally induced systemic impacts on the mother, or through epigenetic inheritance through the female germline. This preliminary proof of concept study and observations suggests an expanded prospective clinical study is now needed to validate and improve the potential preeclampsia susceptibility biomarker.

Results

The clinical sites were at the Indiana University Health and Franciscan Health Hospitals, Indianapolis Indiana. Cheek swab buccal samples were collected, as a systemic biomarker somatic cell, from 78 mothers following delivery. A population of term births with no preeclampsia (n = 13) and PTBs (n = 23) control (no preeclampsia) maternal buccal cell samples were collected and stored at −20 C° until shipment. Case samples consisted of PTB (n = 13) with preeclampsia were also collected. A number of preterm and term births without preeclampsia, but with the presence of gestational hypertension or abnormal blood pressure were also analyzed. The maternal buccal cell sample list and clinical information is presented in Table 1. The patient age, other clinical morbidities, and pregnancy outcomes are presented. Samples were collected and stored at −20 C° at the clinical sites. Once sample collection was completed, all samples were shipped in a single shipment to Washington State University on dry ice for storage at −80 C° until analysis. Ideally, the buccal cells provide a systemic biomarker cell that is affected by early life or previous generation impacts, and so is not influenced dramatically by variable clinical or gestation conditions associated with the disease etiology [23].

Table 1.

Clinical sample information

			Preterm or term
Subj ID	Age	Other clinical morbidities	Preterm	Term	Diabetes type	Abnormal Blood pressure	Hypertensive Disorder of Pregnancy	Hypertension type	Preeclampsia	Preeclampsia (Severe)
Preterm birth no preeclampsia controls
458	23	Anemia, asthma, depression	Y		n/a	Yes	No	n/a	No	No
459	35	Airway disease, hypothyroidism	Y		n/a	Yes	No	n/a	No	No
460	20	Kidney disease, seizures, autism	Y		n/a	Yes	No	n/a	No	No
464	29	None	Y		n/a	Yes	No	n/a	No	No
465	20	ADHD, allergic rhinitis, asthma	Y		n/a	Yes	No	n/a	No	No
469	31	None	Y		n/a	Yes	No	n/a	No	No
470	23	None	Y		n/a	Yes	No	n/a	No	No
471	25	None	Y		n/a	Yes	No	n/a	No	No
477	27	Convulsions, anxiety, reflux	Y		n/a	Yes	No	n/a	No	No
480	26	Cholangitis, colitis, liver disease	Y		n/a	No	No	No	No	No
445	34	None	Y		n/a	Yes	No	n/a	No	No
444	31	None	Y		n/a	No	No	n/a	No	No
400	31	MTHFR	Y		n/a	Yes	No	n/a	No	No
403	21	Beta-thalassemia	Y		n/a	Yes	No	n/a	No	No
402	34	None	Y		n/a	Yes	No	n/a	No	No
407	25	Cervical cancer, depression	Y		n/a	Yes	No	n/a	No	No
409	29	None	Y		n/a	No	No	n/a	No	No
417	26	Cauda equina syndrome, neurogenic bladder	Y		n/a	Yes	No	n/a	No	No
408	29	Hypothyroidism, bipolar disease, schizophrenia	Y		n/a	Unknown	No	n/a	No	No
411	34	PCOS, anxiety	Y		n/a	Unknown	No	n/a	No	No
Term birth no preeclampsia controls
457	22	Anemia, hyperthyroidism		Y	n/a	Yes	No	n/a	No	No
454	26	None		Y	n/a	No	No	n/a	No	No
453	38	None		Y	n/a	Yes	No	n/a	No	No
456	32	None		Y	n/a	Yes	No	n/a	No	No
432	20	HSV history		Y	n/a	Yes	No	n/a	No	No
436	27	hypothyroidism		Y	n/a	Yes	No	n/a	No	No
421	38	Depression, asthma		Y	n/a	Yes	No	n/a	No	No
437	26	Multiple sclerosis		Y	n/a	Yes	No	n/a	No	No
440	33	Hypothyroidism		Y	n/a	Yes	No	n/a	No	No
425	27	Asthma		Y	n/a	Yes	No	n/a	No	No
424	31	Migraines		Y	n/a	Yes	No	n/a	No	No
441	31	PCOS history		Y	n/a	Yes	No	n/a	No	No
433	18	Anemia, chronic sinusitis		Y	n/a	Yes	No	n/a	No	No
426	37	None		Y	n/a	Yes	No	n/a	No	No
PTBs diabetic no preeclampsia controls
463	24	None	Y		Gestational DM	Yes	no	n/a	No	No
474	27	HSV history	Y		Gestational DM	Yes	No	n/a	No	No
476	27	PCOS, obesity, asthma, GERD	Y		DM Type 2	Unknown	No	n/a	No	No
PTB preeclampsia cases
462	32	Thyroid disease, depression, anxiety	Y		DM Type 1	No	Yes	Chronic	Yes	Yes
467	34	Gluten allergy	Y		n/a	No	Yes	Gestational	Yes	Yes
475	36	Anxiety, asthma	Y		n/a	No	Yes	Chronic	Yes	Yes
479	24	None	Y		n/a	No	Yes	Gestational	Yes	Yes
482	26	None	Y		Gestational DM	No	Yes	Chronic	Yes	Yes
483	32	None	Y		n/a	No	Yes	Gestational	Yes	No
485	24	PCOS, hypokalemia, hypercalcemia	Y		Gestational DM	No	Yes	Gestational	Yes	Yes
447	28	PCOS, obesity, skeletal dysplasia	Y		n/a	No	Yes	Gestational	Yes	No
410	30	Anxiety, depression	Y		DM Type 1	No	Yes	Gestational	Yes	Yes
412	25	None	Y		n/a	No	Yes	Gestational	Yes	Yes
416	20	None	Y		n/a	No	Yes	Gestational	Yes	Yes
461	28	Depression, obesity	Y		n/a	No	Yes	Chronic	Yes	No
484	20	None	Y		DM Type 2	Yes	Yes	Chronic	Yes	No

Sample group for samples with identification number and correlated clinical information

The samples were thawed and DNA isolated and fragmented by sonication to approximately 200 bp, as described in the “Methods” section. The samples for individual patients were then used for a methylation DNA immunoprecipitation (MeDIP) procedure and then bar coded and used in Illumina next-generation sequencing, as described in the “Methods”. The DNA methylation information for samples was then used to identify differential DNA methylation regions (DMRs) between the clinical groups. The PTB- non-preeclampsia versus preterm preeclampsia were compared for DMR analysis and a small number of DMRs were identified that did not show statistical significance, having a false discovery rate (FDR) >0.1. However, sample size may have contributed to this comparison, so future-expanded clinical trials are needed. A comparison of term non-preeclampsia versus preterm preeclampsia groups identified at an edgeR P-value of P < 1e-04 a total of 389 DMRs with a statistical significance and FDR of <0.05, Fig. 1a. This preterm preeclampsia comparison with non-preeclampsia provides a useful comparison to identify potential preeclampsia biomarker sites, so was the focus of the current study. The majority of DMR involved a single 1 kb window with 15 DMRs having 2 or 3 kb size, Fig. 1a. The preeclampsia DMR chromosomal locations shown in Fig. 1b demonstrated a genome-wide distribution on all chromosomes with only one cluster of DMRs on chromosome 20, shown with black box, Fig. 1b. The genomic feature for CpG density for the preeclampsia DMRs was predominantly 1 or 2 CpG/100 bp, Fig. 2a. The DMR length was predominantly 1 or 2 kb with some DMR >3 kb in length, Fig. 2b.

Figure 1.

DMR identification and numbers. (a) The number of DMRs found using different P-value cutoff thresholds. The All Window column shows all DMRs. The Multiple Window column shows the number of DMRs containing at least two significant windows (1000 bp each). The number of DMRs with the number of significant windows (1000 bp per window) at an edgeR P-value threshold of P < 1e-04 for DMR is presented with FDR ≤0.05. Term non-preeclampsia versus preterm preeclampsia DMRs (b) DMR Chromosomal locations. The DMR locations on the individual chromosomes are represented with an arrowhead and a cluster of DMRs with a black box. All DMRs containing at least one significant window at a P-value threshold of P < 1e-04 for DMR are shown. Term non-preeclampsia versus preterm preeclampsia DMR

Figure 2.

DMR genomic features (CpG density and length). (a) Term non-preeclampsia versus preterm preeclampsia DMR CpG density. (b) Term non-preeclampsia versus preterm preeclampsia DMR length. (c) DMR principal component analysis (PCA). The first two principal components are used. The underlying data are the RPKM read depth for DMR genomic windows.

A principal component analysis (PCA) was used to compare the normalized reads per kilobase per million (RPKM) read depth for the DMR genomic windows between the term non-preeclampsia and preterm preeclampsia samples, Fig. 2c. Good separation of the two groups was observed for the principal components 1 and 2. This is expected due to the DMR statistical analysis and assists in the subsequent analysis of outlier samples. The preeclampsia DMR name, genomic location, statistics, size, CpG density, maximum fold change, and gene associations are presented in Supplementary Table S1.

Further analysis of the preeclampsia DMRs involved a comparison with a previously identified buccal cell preterm epigenetic biomarker [25]. This previous PTB DMR biomarker contained 165 DMRs at a similar edgeR P < 1e-04 statistic [25]. It should be noted that several of the non-preeclampsia samples used in the current study were also used in the previous PTB study. Since the preterm non-preeclampsia versus preterm preeclampsia did identify 35 DMR at edgeR P < 1e-04 that had an FDR >0.1, Supplementary Table S2, a comparison was made with the term non-preeclampsia versus preterm preeclampsia with the 389 preeclampsia DMRs, Fig. 3. A Venn diagram of the P < 1e-04 DMRs demonstrated a negligible overlap, Fig. 3a. An extended overlap analysis was performed comparing each P < 1e-04 DMR group with all other DMR groups at P < .05 statistical comparison. The previous PTB 165 DMR had minimal overlap with the other DMR groups, Fig. 3b. Interestingly, the preterm non-preeclampsia versus preterm preeclampsia had good overlap with the term non-preeclampsia versus preterm preeclampsia, with 40% overlap (14 DMR), Fig. 3b. This may be influenced by the common preterm preeclampsia samples in the two analyses. The highest level of overlaps with the previously reported PTB DMR was demonstrated between the term non-preeclampsia versus the preterm preeclampsia 389 DMRs and the published PTB DMRs at P < .05 with 60 DMRs in common (15.4%), Fig. 3b. These DMRs in common are listed in the Supplementary Table S1. The comparison of the 389 term non-preeclampsia versus the preterm preeclampsia DMRs had 161 DMRs or 41% in common, Fig. 3b. This was expected due to some common preeclampsia samples between the analyses. Therefore, the non-significant preterm non-preeclampsia versus preeclampsia DMR analysis at a lower P < .05 significance had a good overlap with the term non-preeclampsia versus preterm preeclampsia DMR, with 161 of the 389 DMRs, Fig. 3b. This is further discussed in the “Discussion” section.

Figure 3.

DMR overlap. (a) Venn diagram of all DMR overlap (P < 1e-04). (b) Extended overlaps (P < 1e-04 versus P < 0.05). The number of DMRs overlapped and percentage presented.

Further analysis examined the gene associations within the 389 DMR of the term non-preeclampsia versus preterm preeclampsia data set, Fig. 4. The DMRs in Supplementary Table S1 with gene associations are compiled in Fig. 4a for gene categories. The metabolism and signaling categories were the most common observed. The gene pathway correlations for the DMR associated genes are presented in Fig. 4b and also demonstrate metabolism, with 11 genes being present, and signaling, with 5 genes being present. Few previous genes have been associated with preeclampsia. Any associations are limited to maternal buccal cells and need to consider that these are a systemic marker cell type, which are not directly relevant to preeclampsia etiology. All the DMR associated genes are presented in Supplementary Table S1. However, the presence of a systemic cell having epigenetic alterations suggests a potential generational or early life environmental impact may be associated with preeclampsia etiology. Therefore, the previous generation preeclampsia susceptibility is likely inherited through epigenetic transgenerational inheritance to the current generation to increase preeclampsia susceptibility. The systemic buccal cell biomarker can be used to identify this susceptibility for preeclampsia.

Figure 4.

DMR gene associations. (a) DMR gene functional categories. Preterm no preeclampsia versus preterm preeclampsia (PT.NPE versus PT.PE) and term birth no preeclampsia versus preterm preeclampsia (TN.PE versus PT.PE). (b) Term non-preeclampsia versus preterm preeclampsia KEGG gene pathway associations. The number of DMR-associated genes for each pathway are listed in brackets.

A validation analysis examined the potential use of the 389 DMRs as preeclampsia epigenetic biomarkers using a cross validation procedure [26]. The procedure repeated the DMR analysis 10 times, leaving a portion of the samples out of each analysis for use as a test set. A previously published and established random forest classifier was then trained on the samples used in the analysis to predict the group of the remaining samples [27]. The accuracy across all 10 cross-validation test sets was 78%. See the “Methods” section for details. Observations suggest an adequate accuracy with the preeclampsia biomarkers, but expanded studies with larger numbers of clinical samples are required to improve the accuracy of the analysis.

Discussion

The development of epigenetic biomarkers for pathology and disease susceptibility has recently been considered to allow preventative medicine approaches to delay onset or prevent the onset of the disease [23, 25]. Epigenetic alterations occur in high frequency with the majority of the individuals with the pathology having the epigenetic biomarkers [28]. Previous studies have developed potential epigenetic biomarkers for human male infertility and therapeutic responsiveness [29], the detection of female rheumatoid arthritis [30] and the identification of the paternal transmission of autism susceptibility to offspring [31]. While genetic mutations through genome-wide association studies (GWAS) for disease biomarkers were originally sought, the low frequency (i.e. <1%) of associated genetic mutations limits the use of such genetic biomarkers [22]. In contrast, the epigenetic biomarkers have shown highfrequency (i.e. >90%) associations with individuals having the disease, which suggests they may be far more useful in promoting preventative medicine applications [23]. The current study used a systemic maternal biomarker cell that is efficient to assess early-life or generational impacts on the epigenetics. The systemic cell is less responsive to the disease etiology variables of the current study, such as gestational development timing, hypertension, or preterm versus term parameters [23]. Therefore, variables such as sex of offspring or gestational effects would not be anticipated to impact the buccal cell systemic biomarker cell. The current study does not address disease etiology, but rather disease susceptibility with the use of a systemic biomarker cell [23]. Environmental impacts early in life or ancestrally are anticipated to be more impactful and easily identified with the use of systemic cells not associated with the disease etiology.

PTB is increasing within the population with most populations now >10% of all pregnancies being preterm. Previous health conditions such as obesity or family history of PTB have been observed, but few clinical parameters can be used to predict PTB in the early stages of pregnancy. Although clinical management and preventative medicine approaches can be used to delay the onset or prevent PTB [32], currently prediction of PTB susceptibility is difficult and limited. These preventative and palliative medicine approaches include novel therapeutics and clinical management [33]. The key to the use of these preventative approaches is the early detection of PTB susceptibility, prior to its onset.

Preeclampsia often results in PTB and is the most important member of hypertensive disorders of pregnancy category complications associated with PTB. The clinical conditions of preeclampsia involve abnormal blood pressure, hypertension, and uterine circulation abnormalities [34]. Preeclampsia is a clinically diagnosed progressive disorder comprised of onset hypertension, with or without protein urea, and less frequent liver and kidney dysfunction. Other hypertensive disorders of pregnancy (chronic hypertension and gestational hypertension) can develop into preeclampsia, if other severe features develop (i.e. neurological symptoms, pulmonary edema, thrombocytopenia, liver dysfunction and placental insufficiency). Table 1 provides information on other clinical morbidities within the patient population and was similar for the control and preeclampsia populations. Preeclampsia has familial associations similar to PTB in general. The impacts of preeclampsia are often more severe than non-preeclampsia PTB conditions, due to vascular issues in the placenta and uterus compounding the PTB phenotype. Although some immune and vascular genes have been associated with the preeclampsia phenotype, few biomarkers for preeclampsia have been developed [35, 36]. If such a biomarker existed, then preventative medicine approaches could be used to reduce the impact of the clinical condition.

The recent development of a potential epigenetic PTB biomarker supports the concept that early-stage diagnosis may be possible to allow preventative medicine approaches to delay or prevent the onset of the PTB [25]. This observation suggests the potential to identify preeclampsia with an epigenetic biomarker may be feasible. The previous PTB study and the current study used buccal cells as a systemic purified marker cells for the identification of a general epigenetic shift in all cells that will lead to various disease susceptibilities [25]. Since each cell type has a unique epigenetic profile that determines its cell specificity, systemic effects can be detected that are due to early life or ancestral environmentally induced epigenetic inheritance and familial ancestral exposures [23, 37, 38]. As mentioned, systemic cells (e.g. buccal) may not be influenced by disease parameters such as hypertension, fetal sex, or gestational age but provide a more stable susceptibility biomarker cell [23].

The current study identifies a potential preeclampsia biomarker that involved 389 DMRs with high significance (P < 1e-04) for the preeclampsia, Fig. 1. The genomic features of low-density CpG and genome-wide distribution of the DMRs were observed, Figs 1 and 2. The DMR associated genes also involved PTB-associated gene pathways, Fig. 4. The signaling, cytoskeleton, transport, and transcription-related genes were most frequent. The metabolic, neuroactive signaling, and PI3K-Akt signaling pathways were predominant, Fig. 4. To obtain this epigenetic biomarker, the term birth with no preeclampsia mothers were compared to PTB preeclampsia mothers. This provides a preliminary (i.e. pilot) study comparison for a systemic maternal biomarker approach following the disease occurrence. Interestingly, a comparison of the PTB non-preeclampsia versus preeclampsia mothers did not identify a significant epigenetic biomarker. However, a future study with larger sample number and statistical assessment is needed to extend this study. When a comparison of the term non-preeclampsia versus preterm preeclampsia, 389 DMR epigenetic biomarkers was made with the previous published PTB analysis, 60 (15%) of the newly identified DMRs were found to be present in the prior analysis at a P < .05 threshold, Fig. 3. Observations suggest that an expanded clinical trial with larger numbers of preterm non-preeclampsia patients versus preeclampsia preterm may lead to an efficient epigenetic maternal biomarker. Clearly, expanded clinical trials in different populations will be needed to further optimize and validate the preeclampsia epigenetic biomarker, but the current study indicates that such a preeclampsia epigenetic biomarker exists with this preliminary study.

A preeclampsia or general PTB epigenetic maternal biomarker will ideally be used in the early stages of pregnancy to identify the potential susceptibility to have a preeclampsia condition develop and PTB susceptibility. Ideally, a first trimester analysis that allows preventative medicine approaches to be considered will need to be the future focus and future expanded clinical study. However, the current study provides the proof of concept such a study is needed. Such a preeclampsia biomarker can then be used to facilitate the clinical management of the pregnancy and delay the onset or prevent the PTB and the development of the preeclampsia condition. Expanded clinical trials are needed for the first trimester period, and this will allow more efficient clinical management of pregnancy. The types of clinical management that could be initiated include therapies, therapeutics, and clinical management. The result of this type of prevention medication or therapy following the use of the PTB and preeclampsia maternal biomarkers would potentially result in a dramatic reduction in PTB frequency. Since PTB has negative impacts on the child as they mature and age, the result of reducing the 10% of the population PTBs and preeclampsia will significantly improve the populations later life health. Although further clinical trials and expanded first trimester studies are needed, the impact of the use of epigenetic biomarkers to improve health and detect disease susceptibility for preventative medicine application may be significant.

Methods

Clinical sample collection and analysis

This study is an extension of our previous study of epigenetic biomarkers of PTB study [25] to further identify preeclampsia epigenetic-associated DMR as a potential maternal biomarker. The Indiana University Health (IUH) Hospitals (Riley Hospital for Children, IUH Methodist, IUH North) and Franciscan Health, Indianapolis, Indiana, USA provided maternal buccal cell samples for the current study. Informed consent and HIPAA authorization was obtained from all participants prior to the clinical sample collection. The study protocol was approved by the Indiana University Institutional Review Board (IRB) #1901985132 and the Franciscan IRB, #1489434. All methods were performed in accordance with the relevant guidelines and regulations. Informed consent and HIPAA authorization were obtained from all participants prior to sample collection. All study samples were collected ∼9 days following birth and stored at −20°C at the clinical sites. Maternal samples were categorized by gestation day at delivery, hypertensive disorders of pregnancy, and maternal diabetes. The case group consisted of preterm deliveries (<37 weeks gestation) and maternal preeclampsia. The control groups consisted of term deliveries (≥37 weeks gestation) with no maternal preeclampsia, and preterm deliveries (<37 weeks gestation) with no preterm preeclampsia. The demographic data for these subjects are presented in Supplementary Table S1. Following shipment, buccal cells were stored at −80°C until use.

DNA preparation

Frozen human buccal samples were thawed for analysis. Genomic DNA from buccal samples was prepared as follows: the buccal brush was suspended in 750 μl of cell lysis solution and 3.5 µl of Proteinase K (20 mg/ml). This suspension was incubated at 55°C for 3 h, then vortexed and centrifuged briefly. The lysis solution was then transferred to a new 1.5 µl microcentrifuge tube. The microcentrifuge tube with the buccal brush was centrifuged again to obtain any remaining solution which was combined with the transferred lysis solution. The buccal brush was discarded and 300 µl of protein precipitation solution (Promega, A795A, Madison, WI, USA) was added to the lysis solution. The sample was incubated on ice for 15 min, then centrifuged at 4°C for 30 min. The supernatant was transferred to a fresh 2 ml microcentrifuge tube and 1000 µl ice cold isopropanol was added along with 2 µl Glycoblue (Thermo Fisher, Waltham, MA, USA). This suspension was mixed thoroughly and incubated at −20°C overnight. The suspension was then centrifuged at 4°C for 20 min, the supernatant was discarded, and the pellet (visualized with Glycoblue) was washed with 75% ethanol, then air-dried and resuspended in 100 μl H₂O. DNA concentration was measured using a Nanodrop (Thermo Fisher, Waltham, MA, USA).

Methylated DNA immunoprecipitation

MeDIP with genomic DNA was performed. Individual DNA samples (2–4 µg of total DNA) were diluted to 130 μl with 1× Tris-EDTA (TE, 10 mM Tris, 1 mM EDTA) and sonicated with the Covaris M220 using the 300 bp setting. Fragment size was verified on a 2% E-gel agarose gel. The sonicated DNA was transferred from the Covaris tube to a 1.7 ml microfuge tube, and the volume was measured. The sonicated DNA was then diluted with TE buffer (10 mM Tris HCl, pH7.5; 1 mM EDTA) to 400 μl, heat-denatured for 10 min at 95°C, then immediately cooled on ice for 10 min. Then 100 μl of 5X IP (immunoprecipitation) buffer and 5 μg of antibody (monoclonal mouse anti-5-methyl cytidine; Diagenode #C15200006) were added to the denatured sonicated DNA. The DNA-antibody mixture was incubated overnight on a rotator at 4°C. The following day magnetic beads (Dynabeads M-280 Sheep anti-Mouse IgG; 11201D) were prewashed as follows: the beads were resuspended in the vial and the appropriate volume (50 μl per sample) was transferred to a microfuge tube. The same volume of Washing Buffer (at least 1 ml 1XPBS with 0.1% BSA and 2 mM EDTA) was added and the bead sample was resuspended. The tube was then placed into a magnetic rack for 1–2 min and the supernatant was discarded. The tube was removed from the magnetic rack and the beads were washed once. The washed beads were resuspended in the same volume of 1xIP buffer (50 mM sodium phosphate ph7.0, 700 mM NaCl, 0.25% TritonX-100) as the initial volume of beads. Fifty microliters of beads were added to the 500 μl of DNA-antibody mixture from the overnight incubation, then incubated for 2 h on a rotator at 4°C. After the incubation, the bead–antibody–DNA complex was washed three times with 1X IP buffer as follows: The tube was placed into a magnetic rack for 1–2 min and the supernatant was discarded, then the magnetic bead antibody pellet was washed with 1xIP buffer three times. The washed bead antibody DNA pellet was then resuspended in 250 μl digestion buffer with 3.5 μl Proteinase K (20 mg/ml). The sample was incubated for 2–3 h on a rotator at 55°C, then 250 μl of buffered Phenol–Chloroform–Isoamylalcohol solution was added to the sample, and the tube was vortexed for 30 s and then centrifuged at 14 000 rpm for 5 min at room temperature. The aqueous supernatant was carefully removed and transferred to a fresh microfuge tube. Then 250 μl chloroform was added to the supernatant from the previous step, vortexed for 30 s and centrifuged at 14 000 rpm for 5 min at room temperature. The aqueous supernatant was removed and transferred to a fresh microfuge tube. To the supernatant, 2 μl of glycoblue (20 mg/ml), 20 μl of 5 M , and 500 μl ethanol were added and mixed well, then precipitated at −20°C for 1 h to overnight. The precipitate was centrifuged at 14,000 rpm for 20 min at 4°C and the supernatant was removed, while not disturbing the pellet. The pellet was washed with 500 μl cold 70% ethanol in a −20°C freezer for 15 min then centrifuged again at 14 000 rpm for 5 min at 4°C and the supernatant was discarded. The tube was spun again briefly to collect residual ethanol to the bottom of the tube and as much liquid as possible was removed with the gel loading tip. The pellet was air-dried at RT until it looked dry (about 5 min) then resuspended in 20 μl H₂O or TE. DNA concentration was measured in a Qubit device (Life Technologies) with ssDNA kit (Molecular Probes Q10212).

MeDIP-seq analysis

The analysis for DNA methylation used in this study involves MeDIP-Seq which examines potentially 95% of the epigenome, compared to Bisulfite-Seq at 40–50% epigenome, or tiling arrays at <2% epigenome [39]. All these procedures have been validated, but the main difference is the percent of the genome investigated. The MeDIP DNA samples (50 ng of each) were used to create libraries for next-generation sequencing (NGS) using the NEBNext Ultra RNA Library Prep Kit for Illumina (San Diego, CA, USA) starting at step 1.4 of the manufacturer’s protocol to generate double-stranded DNA. After this step, the manufacturer’s protocol was followed. Each sample received a separate index primer. NGS was performed at WSU Spokane Genomics Core using the Illumina HiSeq 2500 with a PE50 application, with a read size of ∼50 bp and a minimum of 11 million reads per sample, and 11 sample libraries each were run in one lane.

Molecular bioinformatics and statistics

Basic read quality was verified using information produced by the FastQC program [40]. Reads were filtered and trimmed to remove low-quality base pairs using Trimmomatic [41]. The reads for each sample were mapped to the GRCh38 human genome using Bowtie2 [42] with default parameter options. The mapped read files were then converted to sorted BAM files using SAMtools [43]. To identify DMR, the reference genome was broken into 1000 bp windows. The MEDIPS R package [44] was used to calculate differential coverage between control and exposure sample groups. The edgeR P-value [45] was used to determine the relative difference between the two groups for each genomic window. Windows with an edgeR P-value < 10⁻⁴ were considered DMRs The DMR edges were extended until no genomic window with an edgeR P-value < 0.1 remained within 1000 bp of the DMR CpG density and other information was then calculated for the DMR based on the reference genome. DMR were annotated using the NCBI provided annotations. The genes that overlapped with DMR were then input into the KEGG pathway search [46, 47] to identify associated pathways. The DMR-associated genes were then sorted into functional groups by reducing Panther [48] protein classifications into more general categories. All MeDIP-Seq genomic data obtained in the current study have been deposited in the NCBI public GEO database (GEO #: GSE259360).

To estimate whether the DMRs identified could be potentially used as a biomarker, a 10-fold cross-validation analysis was performed. This involved splitting the samples in each sample group into 10 parts. DMRs were identified using samples from nine of the parts. The tenth part was used as a test set to determine whether the DMRs identified could be used to classify the remaining samples into treatment groups. The procedure was repeated a total of 10 times, using each of the 10 parts as the test sets. A standard random forest classifier was used and trained on the test samples using read depths at DMR sites and used to classify the test samples (https://cran.r-project.org/web/packages/randomForest/index.html). An accuracy of 78% was observed that represented in the term birth with no preeclampsia, 12 accurately assessed as no preeclampsia and 4 inaccurately assessed as preeclampsia, and for the preeclampsia 2 inaccurately assessed as no preeclampsia and 9 accurately assessed as preeclampsia.

Acknowledgements

We acknowledge Ms Donna Watkins, Ms Leah Engelstad, Ms Dianne Herron, and Mr Jeffrey Joyce at Indiana University for clinical recruitment and sample collection assistance and, Dr Millissia Ben Maamar, and Ms Sarah De Santos for technical assistance. We acknowledge Ms Heather Johnson for assistance in preparation of the manuscript. We thank the Genomics Core laboratory at WSU Spokane for sequencing data. This study was supported by The Libra Foundation (Grant #GF007237) (https://www.thelibrafoundation.org) (last accessed 11 January 2024) and The John Templeton Foundation (Grant # 50183 and 61174) (https://templeton.org) (last accessed 11 January 2024) grants to M.K.S. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Author contributions

Paul Winchester (Patient’s recruitment, Clinical sample collection oversight, Data analysis, Editing manuscript), Cathy Proctor (Clinical sample collection and recruitment, Editing manuscript), Eric E. Nilsson (Sample processing, Data analysis, Editing manuscript), Daniel Beck (Bioinformatics, Data analysis, Editing manuscript), and Michael K. Skinner (Conceptualization, Data analysis, Funding acquisition, Writing and editing manuscript).

Supplementary data

Supplementary data is available at EnvEpig online.

Conflict of interest:

The authors report no conflict of interest.

Funding

This study was supported by The Libra Foundation (Grant #GF007237) (https://www.thelibrafoundation.org) (last accessed 11 January 2024) and The John Templeton Foundation (Grant # 50183 and 61174) (https://templeton.org) (last accessed 11/ January 2024) grants to M.K.S. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Data availability

All molecular data have been deposited into the public database at the NCBI repository, (GEO # GSE259360), and R code computational tools are available at GitHub (https://github.com/skinnerlab/MeDIP-seq) (last accessed 11 January 2024) and www.skinner.wsu.edu (last accessed 11 January 2024).

Ethics approval and consent to participate

Ethics approval and consent to participate were carried out on all participants. Approvals to conduct the study were obtained from Indiana University Institutional Review Board (IRB) #1901985132 and the Franciscan Institutional Review Board (IRB), #1489434. All methods were performed in accordance with the relevant guidelines and regulations.

Consent for publication

Not applicable.