Detection of an intestinal cell DNA methylation signature in blood samples from neonates with necrotizing enterocolitis

ABSTRACT

Background

Necrotizing enterocolitis (NEC) is an often fatal intestinal injury that primarily affects preterm infants for which screening tools are lacking. We performed a pilot analysis of DNA methylation in peripheral blood samples from preterm infants with and without NEC to identify potential NEC biomarkers.

Methods

Peripheral blood samples were collected from infants at NEC diagnosis (n = 15) or from preterm controls (n = 13). Targeted genome-wide analysis was performed to identify DNA methylation differences between cases and controls.

Results

Broad differences between NEC cases and controls were identified in distinct genomic elements. Differences between surgical NEC cases and controls were frequently associated with inflammation. Deconvolution analysis to identify cell type-specific DNA signatures revealed increases in ileal, vascular endothelial, and cardiomyocyte cell type proportions and decreases in colonic and neuronal cell type proportions in blood from NEC cases relative to controls.

Conclusions

We identified marked differences in DNA methylation of peripheral blood samples from preterm infants with and without NEC. Increased ileal cell-specific methylation signatures in the blood of infants with NEC relative to controls, with a marked increase seen in surgical cases, provides rationale for further analysis of intestinal DNA methylation signatures as biomarkers of NEC.

1. Introduction

Necrotizing enterocolitis (NEC) is an intestinal inflammatory disease that is a leading cause of death in premature infants [1,2]. NEC develops suddenly in ~7% of preterm infants who are born weighing <1500 grams [3,4]. It occurs in an unpredictable manner, and, in severe cases, can rapidly induce irreversible intestinal injury, resulting in the need for emergency surgery or death within hours of diagnosis. Unfortunately, the mortality rate remains high at up to 30–50%, with death or profound morbidity, including neurodevelopmental impairment, observed in ~70% of infants with surgical NEC [5]. The etiology of these poor outcomes is multifactorial, but a failure to improve diagnostic tools or provide patients with targeted therapies is likely a predominant contributor. In fact, the survival rate for infants afflicted by NEC has not significantly improved over the past 50 years, and associated costs nationally have been estimated to be $2–3 billion per year [6–8].

The typical clinical presentation of NEC in the neonatal period can include abdominal distension, grossly bloody stools, feeding intolerance, laboratory abnormalities, and the presence of pneumatosis intestinalis (air within the bowel wall) on abdominal imaging. With the exception of intramural gas, these signs are nonspecific and overlap with other clinical entities that are common amongst preterm infants. This can make the diagnosis of NEC extremely challenging. It is not uncommon for a premature infant to undergo treatment for presumed NEC, given the lack of availability of definitive diagnostic tools. In addition, once an infant is diagnosed with NEC, due to the lack of knowledge regarding underlying pathophysiology and mechanisms that contribute to the development of NEC, no specific or targeted therapies are currently available. At this time, treatment options include bowel rest, broad-spectrum antibiotics, and, in severe cases, resection of necrotic bowel. The ability to accurately identify which infants will develop NEC, predict their disease course, and provide a definitive diagnosis would be revolutionary in the field of neonatology.

This bleak picture highlights the urgent need to develop diagnostic and predictive biomarkers that accurately identify premature infants who are at risk of NEC, with the goal of early detection and intervention to improve clinical outcomes. This cross-sectional pilot study represents a first step toward the development of DNA methylation-based biomarkers for NEC.

Methylated DNA has several features that make it an attractive target for the development of biomarkers utilizing a precision medicine approach to clinical treatments. For example, there are ample data showing DNA methylation signatures provide us with detailed insight into the molecular phenotype of their cell type/tissue of origin. Alterations in the epigenomic molecular phenotype at the level of DNA methylation are intimately associated with the pathobiology of disease and exposure to environmental stressors [9,10]. Furthermore, DNA methylation is highly tissue and cell-type specific [11]. We can use these features to better understand the cell/tissue of origin when sampling from a mixed sample, such as blood, where multiple cell types contribute to the overall methylation pattern. Lastly, DNA is a highly stable analyte and, therefore, is a robust target for clinical chemistry protocols that might involve transportation and/or delay between receipt and analysis.

Previous research has demonstrated that altered DNA methylation is found in the ileum, colon, and stool of neonates with NEC [12–14]. In this pilot study, our objective was to build upon these prior observations by executing an in-depth analysis of DNA methylation in blood samples from infants with NEC and their control counterparts. Peripheral blood is routinely acquired in neonates in the intensive care unit for clinical purposes, and it can serve as an excellent source for biomarker identification when coupled with technology that requires minimal blood volumes and utilizes discarded blood samples. Identifying a methylation signature in the blood of infants with NEC relative to controls is a first step in developing urgently needed high-quality predictive and diagnostic biomarkers to dramatically improve medical care provided to preterm neonates.

2. Materials and methods

2.1. Study population and selection criteria

Sample collection for this study was previously approved by the University of Pittsburgh Institutional Review Board (IRB) (Protocol: PRO09110437). Infants with NEC (n = 15) or premature control infants (n = 13) without NEC (Table 1) were recruited under Protocol PRO09110437 at either the Children’s Hospital of Pittsburgh (CHP) of University of Pittsburgh Medical Center (UPMC) or Magee-Women’s Hospital Neonatal Intensive Care Units (NICUs). Consent for participation in the study was obtained from their parent or legal guardian per the University of Pittsburgh IRB regulations. Controls were chosen to match the gestational age at birth and post-menstrual age at sample collection for infants with NEC as closely as possible, given the small cohort size. Exclusion criteria included major congenital anomalies such as gastroschisis, omphalocele, congenital diaphragmatic hernia, or cyanotic congenital heart disease.

Table 1.

Cohort description.

Characteristics	NEC (n = 15)	Control (n = 13)	P value
Male sex	11 (73%)	5 (38%)	0.063†
Gestational age at birth (weeks)	27 0/7 (24 5/7–33 0/7)	30 1/7 (28 1/7–32 3/7)	0.299‡
Birth Weight (g)	891 (660–2004)	1340 (1113–1713)	0.185‡
Age at enrollment (days)	21 (7–29)	9 (5.5–16.5)	0.042‡
Postmenstrual age at NEC diagnosis (weeks)	33 0/7 (29 4/7–34 2/7)	–
Postmenstrual age at enrollment (weeks)	–	31 5/7 (30 1/7–33 4/7)
NEC Severity
Medical NEC	11 (73%)	–
Surgical NEC	4 (27%)	–
Comorbitidies
BPD	9 (60%)	1 (7.7%)	0.004†
IVH	8 (53%)	0 (0%)	0.002†
ROP	8 (53%)	4 (31%)	0.229†
Age at NICU discharge (days)	89 (30–124)	37 (28–52)	0.012‡

Date are representated as median (interquartile range) or n (%).

NEC: Necrotizing Enterocolitis, NICU: Neonatal Intensive Care Unit

BPD: Bronchopulmonary Dysplasia, IVH: Intraventricular hemorrhage, ROP: Retinopathy of Prematurity

^†Chi-squared test

^‡Mann-Whitney U-test

2.2. Sample collection and storage

Blood was collected from infants at the time of enrollment, which, for infants with a recent NEC diagnosis, was at the time of their next scheduled blood draw, or after consent was obtained for controls. This was prior to surgery in 3 of the 4 patients with surgical NEC. Blood samples (250 ul volume) were collected into purple top BD blood collection tubes (Becton, Dickson and Company, Franklin Lakes, NJ) and stored at 4°C until daily collection by the research team, and subsequently stored at −80°C until thawed for analysis.

2.3. Sequencing of DNA

DNA methylation sequencing was conducted as previously described [13,14]. In summary, DNA extracted from blood samples was fragmented to approximately 175 base pairs (bp) via mechanical shearing technology (Covaris, Woburn, MA). Libraries were then prepared utilizing the KAPA HyperPrep Kit (Roche, Pleasanton, CA). Post-ligation, the libraries were subjected to bisulfite conversion using the EZ DNA Methylation-Direct Kit (Zymo Research, Irvine, CA) and subsequently amplified through 13–15 cycles of PCR. For targeted enrichment, the bisulfite-converted DNA libraries underwent capture using the SeqCap Epi CpGiant Enrichment Kit (Roche, Pleasanton, CA) and were further amplified for 10 cycles. Sequencing was performed on an Illumina HiSeq 2500 platform, generating 100 bp paired-end reads.

Quality control measures included trimming of low-quality sequences and removal of adapter sequences via Trim-Galore (https://www.bioinformatics.babraham.ac.uk/projects/trim_galore/). The processed reads were aligned to the human reference genome (GRCh38/hg20) using Bismark Bisulfite Read Mapper [15] in paired-end, bowtie2 mode, with unmapped reads subsequently aligned in single-end, bowtie2 mode. Duplicate reads were removed, and methylation calling was performed on both paired-end and single-end files, which were then merged to ascertain the methylation status of each CpG site. Methylation signatures were identified using the beta-binomial test implemented in the R packages methylSig [16] and DSS [17].

2.4. Identification of differentially methylated CpG sites

Principal component analysis (PCA) was performed on the CpGs located within promoters (defined as upstream 1500 bp to downstream 500 bp of a transcription start site), CpG islands (CGIs), introns, exons, and intergenic regions of autosomes (i.e., non-sex chromosomes). We also undertook PCA using methylation levels of CpGs located only within previously identified cell type-specific methylation biomarkers [18]. Specifically, we calculated the standard deviation of CpG methylation rates across all samples for each CpG. Only CpGs with a standard deviation above the median were selected. Pairwise plots of the first three principal components were generated to examine the similarity among the samples. To identify differentially methylated CpGs between two groups of samples, we used a negative binomial test, as implemented in the R package methylKit [19], with the false discovery rate (FDR) controlled at ≤ 0.05.

2.5. Cell type proportion estimation by deconvolution analysis of DNA methylation data

Available cell type-specific whole-genome DNA methylation data (GSE186458) was used to estimate cell type proportions in each sample [20]. We used the cell type reference atlas from Loyfer et al. (2023), consisting of 205 tissue samples for 39 cell types [18]. Non-negative least squares were used to regress observed methylation fractions at cell type maker regions onto the cell type signature matrix. Resulting estimates were normalized to sum to one. Markers with fewer than 30 counts in any sample were excluded from the analysis.

Cell types whose proportions differed between the blood of stratified/non-stratified patients with NEC and controls were identified by performing Kruskal–Wallis tests and Wilcoxon rank sum tests, respectively, on each cell type and controlling the FDR using the Benjamini–Hochberg procedure [21].

3. Results

3.1. Cohort characteristics

The cohort consisted of premature infants enrolled at NEC diagnosis (n = 15) or premature infants without NEC (controls, n = 13, Table 1) hospitalized at the Children’s Hospital of Pittsburgh or Magee-Womens Hospital NICUs. Blood was collected after enrollment at the next scheduled blood draw after a diagnosis of NEC, or after consent for controls. There were four infants with surgical NEC and 11 infants with medical NEC (Bell’s Stage 2 or higher). There were no significant differences in sex, gestational age at birth, or birth weight between the groups (Table 1). Although the median gestational age and birth weight were lower for the studied infants with NEC, this difference was not statistically significant, given the considerable overlap in the interquartile range between the groups. Infants who developed NEC were enrolled at a significantly older age after delivery (NEC: 21 days vs. Control: 9 days, p < 0.05 via Mann-Whitney), had a significantly higher rate of bronchopulmonary dysplasia (BPD) (NEC: 60% vs. Control: 7.7%, p < 0.01 via Chi-Square test) and intraventricular hemorrhage (IVH) (NEC: 53% vs. Control: 0%, p < 0.01 via Chi-Square test), as well as a significantly prolonged hospitalization relative to the control infants (Median days of hospitalization for NEC: 89 vs. Median days for control: 37, p < 0.05, Mann-Whitney, Table 1). NEC has been associated with worse outcomes for preterm neonates, including an increased rate of IVH, BPD, and prolonged hospitalization [22]. Thus, these differences are to be expected when comparing a cohort of infants with and without NEC.

3.2. Analysis of differential methylation in the blood of infants with NEC and controls

Bisulfite conversion followed by solution-phase hybridization was used to sequence approximately 5.5 million CpG sites within ~80 megabytes (Mb) of the human genome overlapping with promoters, exons, introns, CpG islands (CGIs), CGI shores, and enhancers. An average of ~27 gigabytes (Gb) of data per sample was generated, corresponding to a mean sample read depth of ~57×.

Broad differences in DNA methylation patterns for autosomes in the blood of infants with NEC compared to controls were analyzed. Principal component analysis (PCA) revealed a visible distinction between surgical and medical NEC cases and controls when focusing on CpG sites located in promoters, CGIs, introns, exons, intergenic regions, and ~10,000 short sequences identified as cell type-specific methylation biomarkers previously identified by Loyfer et al. (2023) [18] (Supplementary Figure S1A-F).

Next, genome-wide CpG sites whose methylation levels differed significantly between cases and controls were identified. We detected only 23 CpG sites that were differentially methylated when controlling for FDR at a q value of ≤ 0.05 (Supplementary Table S1). Thus, analysis was repeated with stratification of patients based on NEC severity. Comparison between samples obtained from controls and infants with surgical NEC identified 10,114 differentially methylated CpG sites with a methylation difference of 10% or more after controlling for multiple testing comparisons (q ≤ 0.05) (Supplementary materials 1). Of these, 1,839 overlapped with promoters, 5,493 with introns, 1,648 with exons, 1,027 CGIs, 2,340 with CGI shores, and 54 with enhancers. CpG site-specific differences between control and medical NEC blood samples were much rarer, with only 12 CpG sites found to differ with a methylation difference of 10% or more after controlling for multiple testing comparisons (q ≤ 0.05) (Supplementary Table S2). The comparison between medical and surgical NEC samples identified 1474 differentially methylated CpG sites with a methylation difference of 10% or more after controlling for multiple testing comparisons (q ≤ 0.05) (Supplementary Table S3).

The functional significance of differentially methylated CpG sites was examined using pathway analysis. Genes containing CpG sites found to be differentially methylated between surgical NEC cases and controls in defined genomic elements (introns, exons, promoters, CGIs, CGI shores) were evaluated for their propensity to be enriched within known biological pathways. These results are summarized in Table 2. For CpG sites in promoters, we found 4 of the 5 top-identified pathways directly related to the inflammatory response. These 5 were, Molecular Mechanisms of Cancer (p = 3.3 × 10⁻⁸), Neutrophil Degranulation (p = 1.6 × 10⁻⁷), Th2 Pathway (p = 4.32 × 10⁻⁷), Th1/Th2 Activation Pathway (p = 1.18 × 10⁻⁶), and PD-1/PD-L1 cancer immunotherapy pathway (p = 4.4 × 10⁻⁶). The top predicted upstream regulators of these genes support a central role for the inflammatory response and include interleukin-4 (IL4, p = 2.65 × 10⁻¹⁵), interleukin-15 (IL15, p = 7.04 × 10⁻¹²), T cell receptor complex (TCR, p = 1.53 × 10⁻¹¹), immunoglobulin (p = 4.61 × 10⁻¹¹), transforming growth factor beta (TGFβ, p = 5.69 × 10⁻¹¹), lipopolysaccharide (LPS, p = 1.81 × 10⁻¹⁰), tumor necrosis factor alpha (TNFα, p = 4.72 × 10⁻¹⁰), and the transcription factor signal transducer and activator of transcription 3 (STAT3, p = 5.82 × 10⁻¹⁰). We also found a central role in the modification of methylation in genes involved in inflammation upon examination of patterns in exons. The top 5 identified pathways were Leukocyte Extravasation Signaling (p = 5.82 × 10⁻⁶), Th2 Pathway (p = 4.41 × 10⁻⁵), Neutrophil Degranulation (p = 4.72 × 10⁻⁵), Th1 and Th2 Activation Pathway (p = 5.25 × 10⁻⁵), and endothelial nitric oxide synthase (eNOS) signaling (p = 5.53 × 10⁻⁵). The top upstream regulators were found to be the transcriptional regulator LMO2 (p = 1.24 × 10⁻¹⁴), immunoglobulin complex (p = 1.18 × 10⁻¹³), lipopolysaccharide (LPS, p = 6.77 × 10⁻¹³), dexamethasone (p = 6.77 × 10⁻¹³), and transforming growth factor beta (TGFβ, p = 1.85 × 10⁻¹²). A similar analysis of medical NEC cases and controls was not performed due to insufficient numbers of significantly differentially methylated CpG sites to support pathway analysis.

Table 2.

Ingenuity pathways analysis of differentially methylated CpG sites in the blood of infants with surgical nec relative to controls.

Genomic Elements	Pathway	P value
Introns	RHO GTPase Cycle	9.32 × 10−13
	Role of NFAT in Cardiac Hypertrophy	6.93 × 10−12
	Molecular Mechanisms of Cancer	1.05 × 10−11
	Signaling by VEGF	5.20 × 10−10
	Axonal Guidance Signaling	8.64 × 10−10
Top Upstream Regulator	TGFβ1	4.48 × 10−19
Exons	Leukocyte Extravasation Signaling	5.82 × 10−6
	Th2 Pathway	4.41 × 10−5
	Neutrophil Degranulation	4.72 × 10−5
	Th1 and Th2 Activation Pathway	5.25 × 10−5
	eNOS signaling	5.53 × 10−5
Top Upstream Regulator	LMO2	2.65 × 10−14
Promoters	Molecular Mechanisms of Cancer	3.31 × 10−8
	Neutrophil Degranulation	1.60 × 10−7
	Th2 Pathway	4.32 × 10−7
	Th1 and Th2 Activation Pathway	1.18 × 10−6
	PD-1, PD-L1 Cancer Immunotherapy Pathway	4.40 × 10−6
Top Upstream Regulator	IL4	2.65 × 10−15
CpG Islands	Potassium Channels	2.65 × 10−4
	Calcium Signaling	9.35 × 10−4
	AMPK Signaling	2.00 × 10−3
	Oxytocin Signaling Pathway	2.21 × 10−3
	Endocannabinoid Neuronal Synapse Pathway	2.93 × 10−3
Top Upstream Regulator	GLI1	1.38 × 10−5
CpG Island Shores	Human Embryonic Stem Cell Pluripotency	8.40 × 10−7
	Chronic Myeloid Leukemia Signaling	2.71 × 10−6
	RHO GTPase Cycle	3.58 × 10−6
	ID1 Signaling Pathway	1.25 × 10−5
Top Upstream Regulator	IL4	1.01 × 10−11

3.3. Deconvolution analysis identifies differences between NEC cases and controls

Epigenomic analysis of whole blood may be confounded by differencing frequencies of heterogenous cell populations of unknown proportions. Thus, deconvolution analysis focused specifically on CpG sites located on chromosomes 1 and 2 using a widely available cell type-specific DNA methylation reference data repository [18]. Although Loyfer et al. characterized whole genome DNA methylation for 39 different flow-sorted cell types, we excluded their colon and ileal epithelial cell reference data, and instead, we used only 37 of their cell type reference data sets. We utilized our previously published genome-wide DNA methylation analysis of colon and ileal cells from infants with and without NEC, given the similarity of the patient population to the population in our current study [13,14]. We focused our analysis on chromosomes 1 and 2 since these chromosomes contain a sufficient number of CpG sites to perform adequate deconvolution analysis. As the two largest chromosomes, the data are more consistent compared to analysis of smaller chromosomes. Non-negative least squares were used to regress observed methylation fractions at previously characterized cell type marker regions [18] onto the cell type signature matrix. The resulting estimates were normalized to sum to one. Markers with fewer than 30 counts in any sample were excluded from the analysis. Cell types whose proportions differed between the blood of stratified/non-stratified NEC patients and controls were identified by performing Kruskal–Wallis tests and Wilcoxon rank sum tests, respectively, on each cell type and controlling the FDR using the Benjamini–Hochberg procedure [21].

As expected, leukocytes, including granulocytes, monocytes/macrophages, T lymphocytes, and natural killer (NK) cells, were detected in the blood of both controls and patients with NEC. However, the difference in cell type proportions was not significant between NEC cases and controls (Figure 1). The overall frequency of NK cells in the blood of infants with NEC was decreased, although the difference was not statistically significant when considering CpG sites located on chromosome 1 (Wilcoxon test, p = 0.088, Adjusted p = 0.24) and chromosome 2 (Wilcoxon test, p = 0.098, Adjusted p = 0.27) (Figure 1(g,h)).

Figure 1.

Similar frequencies of leukocyte subsets were detected in the peripheral blood of infants with and without NEC. Cell type proportions in the blood calculated using comparative deconvolution analysis of methylation patterns are shown for (a,b) granulocytes, (c,d) monocytes/macrophages, (e,f) T lymphocytes, and (g,h) natural killer cells. Boxes represent the 25th −75th percentile, with a horizontal line indicating the median. Points represent individual patients. Significance determined via Wilcoxon test.

We previously detected differential methylation in the ileum and colon as well as the stool of infants with NEC relative to controls [12–14]. In this study, genome equivalents originating from ileal and colonic cells in blood samples were detected for both NEC cases and controls (Figure 2(a-d)). Interestingly, the proportions of ileal cells were increased in the blood of infants with NEC (Figure 2(a,b)), with significant differences before FDR control detected for CpG sites on both chromosome 1 (Wilcoxon test, p = 0.033; Adjusted p = 0.20) and chromosome 2 (Wilcoxon test, p = 0.029; Adjusted p = 0.17) (Figure 2(a,b)). In contrast, markers of colonic cells trended in the opposite direction, with NEC cases showing lower cell type proportions than controls, although these changes were not statistically significant (Figure 2(c,d)).

Figure 2.

Intestinal cells are detectable in the blood of preterm infants with an increased frequency of cells from the ileum for infants with nec relative to controls. Cell type proportions in the blood were calculated using comparative deconvolution analysis of methylation patterns, shown for the (a,b) ileum and (c,d) colon. Boxes represent the 25th −75th percentile, with a horizontal line indicating the median. Points represent individual patients. Significance determined via Wilcoxon test.

To determine the impact of disease severity on the results of the deconvolution analysis, surgical NEC cases were separated from medical NEC cases and compared with controls in a three-way comparison (Figures 3, 4, and Supplementary Table S4). As in Figure 1, there were no statistically significant differences in cell type proportions between NEC cases and controls for granulocytes, monocytes/macrophages, or T cells (Figure 3(a-f)). NK cell frequencies were lower, although the difference was not statistically significant in NEC cases compared to controls, with a further reduction in NK cell proportions between medical and surgical NEC (Figure 3(g,h).).

Figure 3.

Stratification of infants with NEC based on disease severity revealed similar frequencies of leukocyte subsets in the peripheral blood for infants with surgical NEC, medical NEC, and controls. Cell type proportions in the blood calculated using comparative deconvolution analysis of methylation patterns are shown for (a,b) granulocytes, (c,d) monocytes/macrophages, (e,f) T lymphocytes, and (g,h) natural killer cells. Boxes represent the 25th −75th percentile, with a horizontal line indicating the median. Points represent individual patients. Significance was determined via Kruskal-Wallis.

Figure 4.

Infants with surgical NEC exhibit significantly increased proportions of cells from the ileum and decreased cells from the colon in the peripheral blood. Cell type proportions in the blood calculated using comparative deconvolution analysis of methylation patterns are shown for the (a,b) ileum and (c,d) colon for infants with medical NEC, surgical NEC, and controls. Boxes represent the 25th −75th percentile, with a horizontal line indicating the median. Points represent individual patients. Significance was determined via Kruskal-Wallis.

When patients were stratified based on NEC severity, analysis of ileal and colonic cell proportions in the blood revealed a significant difference for infants with medical NEC, surgical NEC, and controls (Figure 4). The difference in the frequency of ileal cells was highly significant for CpG sites on both chromosomes 1 (Kruskal–Wallis test, p = 0.0041; Adjusted p = 0.025) and 2 (Kruskal–Wallis test, p = 0.0039; Adjusted p = 0.024), with the highest levels approaching 20% in blood samples from surgical NEC cases (Figure 4(a,b)) and Supplementary Table S4). The differences in proportions of colonic cells in the blood were not as significant for CpG sites on both chromosomes 1 (Kruskal–Wallis test, p = 0.023; Adjusted p = 0.070) and 2 (Kruskal–Wallis test, p = 0.035; Adjusted p = 0.11), with the lowest levels approaching 0% in blood samples from surgical NEC cases and loss of significance when controlling for multiple comparisons (Figure 4(c,d)).

The deconvolution analysis was repeated using all 39 reference cell types reported by Loyfer et al. (2023) [18] without including our previously published ileal and colon DNA methylation data. The results of these analyses were broadly consistent with those described above without significant differences in leukocyte populations, including granulocytes, monocytes/macrophages, and T lymphocytes between NEC samples and controls (Supplementary Figure S2A-F and Supplementary Table S5). As before, we saw a trend toward lower NK proportions in cases of surgical NEC compared to medical NEC, with the highest levels in controls (Kruskal-Wallis, p = 0.146 for chromosome 2) (Supplementary Figure S2G, H and Supplementary Table S5). Notably, although the Loyfer et al. (2023) [18] cell type reference data did include ileal and colonic epithelial cells from adults, our deconvolution analysis using those data exclusively did not detect these cell types in these whole blood samples. This is likely reflective of the fact that the Loyfer et al. dataset included samples from adults, while the ileal and colonic epithelial cell reference data from our previous studies was generated from preterm neonates, as described above.

We observed the presence of DNA methylation signatures from several other notable cell types, particularly vascular endothelium, which was considerably elevated in surgical NEC (chromosome 1: Kruskal–Wallis test, p = 0.0097; Adjusted p = 0.088, chromosome 2: Kruskal–Wallis test, p = 0.032; Adjusted p = 0.14) (Figure 5(a,b) and Supplementary Table S5). Similar results were observed for methylation biomarkers in cardiomyocytes, which were present in all three groups and significantly increased in surgical NEC samples (chromosome 1: Kruskal–Wallis test, p = 0.044; Adjusted p = 0.16, chromosome 2: Kruskal–Wallis test, p = 0.015; Adjusted p = 0.14) (Figure 5(c,d)). Also of interest was the presence of neuronal cell markers, which were reduced with borderline significance in surgical NEC samples (chromosome 1: Kruskal–Wallis test, p = 0.054; Adjusted p = 0.16, chromosome 2: Kruskal–Wallis test, p = 0.08; Adjusted p = 0.24) (Figure 5(e,f)).

Figure 5.

Increased frequencies of vascular endothelial cells and cardiomyocytes but decreased neurons were detected in the blood of infants with surgical NEC relative to controls. Comparative deconvolution analysis of DNA methylation in blood samples from medical NEC cases, surgical NEC cases, and controls was performed using reference data from 39 flow-sorted reference cell types [18]. Cell type proportions using CpG site methylation data on chromosomes 1 and 2 are shown for (a,b) vascular endothelium, (c,d) cardiomyocytes, and (e,f) neurons. Boxes represent the 25th −75th percentile, with a horizontal line indicating the median. Points represent individual patients. Significance was determined via Kruskal-Wallis.

4. Discussion

This manuscript presents a preliminary analysis of methylation patterns in the peripheral blood of a cohort of preterm infants with and without NEC using whole-genome bisulfite sequencing. We provide an overview of global methylation patterns and use deconvolution analysis to identify specific frequencies of cell types in the peripheral blood. Most importantly, we demonstrate that ileal cell methylation signatures are detectable in the blood of infants with NEC at the time of diagnosis at levels significantly higher than controls. In contrast, colon-associated methylation patterns are significantly reduced. These data provide support for further examination of prospective methylation patterns in the blood as predictive and diagnostic biomarkers for NEC.

We determined that NEC severity influenced global methylation patterns with 10,114 differentially methylated CpG sites identified when comparing samples from patients with surgical NEC to controls. This is in contrast to 23 CpG sites, which were different when comparing all patients with NEC to controls. These findings likely indicate the degree of systemic and localized inflammation of the intestine for patients with surgical NEC compared with medical NEC. This difference in disease severity is also reflected by the presence of 1474 differentially methylated CpG sites between patients with medical and surgical NEC. The robust inflammatory response in surgical NEC is observed in the biological pathway analysis, which demonstrated that pathways related to immune activation and inflammation are involved in disease. The most significant difference in this inflammatory response was seen between infants with surgical NEC and controls. These findings may also indicate that blood methylation analysis could potentially be used to reflect disease severity and, in the future, predict disease course. There are currently no predictive NEC biomarkers being utilized clinically in neonatal medicine, but there is a predictive scoring tool that integrates clinical and laboratory data from the electronic medical record [23]. Many candidate predictive biomarkers utilizing blood, urine, and stool have been previously identified, although all have significant limitations that prevent clinical use, and this remains an area of active investigation [24–27].

Our data demonstrate that deconvolution analysis could potentially be used to identify frequencies of leukocyte subsets in the peripheral blood of preterm infants. Although these populations did not significantly differ between the groups except for a trend toward reduced NK cells, they did provide detailed insight into peripheral blood methylation patterns in preterm infants, which can be used as a robust dataset for further exploration. Notably, the observation that NK cell proportions were reduced in NEC samples in our data is consistent with previous findings [28]. Our data are also in line with a study examining the abundance of T cell subsets, including regulatory T cells and T helper 17 cells, using methylation analysis of dried blood spots in preterm infants, which failed to detect a difference in these number of cells in preterm infants with NEC and controls [29]. Although there was no difference between the groups in that study, it did provide support for the possibility of analyzing methylation patterns from the minute volume of blood present in a dried blood spot, which is an essential factor for the implementation of a clinical biomarker in premature infants.

The most significant finding in our study was the detection of ileal and colonic cell-specific methylation patterns in the blood of preterm infants with and without NEC. Strikingly, we found an increase in the frequency of cells from the ileum of infants with NEC relative to controls, which was accentuated when infants were stratified based on NEC severity. The increased detection of ileal cell signatures in the blood of patients with surgical NEC is likely reflective of the severity of inflammation and reduction in gut barrier integrity in these patients. These patients often have severe and irreversible injury to their intestinal mucosa accompanied by a robust inflammatory cell infiltrate. This cellular signature potentially results from a combination of engulfment of damaged epithelial cells by circulating leukocytes and epithelial cell extravasation in the setting of increased intestinal and vascular barrier permeability. A possible explanation for the decrease in colon-associated signatures in the blood of patients with surgical NEC is that the presence of these cells in the blood is reflective of a homeostatic state, which is disrupted in NEC. It is possible that intestinal cells are present in the blood during development as a part of normal cellular turnover that occurs during the period of rapid intestinal development in utero. An alternative explanation is that the increased frequency of cells from the colon is a mathematical phenomenon reflective of the increase in cells from the ileum or other blood cell types when the total cellular frequency is set to 1 using deconvolution analysis, as in these studies.

We also detected significantly increased vascular endothelial and cardiomyocyte signatures in the blood of patients with surgical NEC and a trend toward decreased neuronal signatures. The presence of vascular endothelial cells could possibly result from injury to the intestinal vasculature in the setting of systemic inflammation, which then leads to engulfment by immune cells or transit across the leaky vascular barrier. Patients with surgical NEC are often severely ill and require hemodynamic support with vasopressor(s) infusions. With this degree of illness, it is feasible for cardiac injury, as evidenced by the presence of a cardiomyocyte signature, to occur. Lastly, the decrease in neuronal cells in surgical NEC may reflect the loss of a homeostatic state characterized by glial cell turnover during normal development, as well as the intestinal glial cell loss previously demonstrated to be associated with the development of NEC [30]. Future mechanistic studies are needed to test these theories.

Clinical tests that analyze circulating DNA in the peripheral blood have been implemented and are well-characterized in adult medicine to determine disease risk [31]. For example, cell-free nucleic acid-based testing (cfDNA), which detects fetal DNA fragments in the blood of a pregnant patient, is utilized to non-invasively identify chromosomal abnormalities in the fetus [32–34]. Interestingly, an extension of that technology, cell-free RNA (cfRNA) testing, can identify changes in organ-specific fetal gene expression in maternal blood throughout gestation [35]. Blood methylation patterns, both globally and associated with specific genes, are used to non-invasively determine colorectal cancer risk in adults [36–38]. Of note, as discussed in the above studies, there is a distinct difference between cell-free nucleic acids in plasma and nuclear DNA extracted from circulating cells in blood, as in our data. The former is from apoptotic and necrotic cells, and it is well known that plasma contains cell-free DNA, which circulates systemically. Although our samples were prepared from whole blood, the fraction of cell-free DNA contributed by plasma is so small relative to high molecular weight nuclear DNA that it is unlikely to have contributed significantly to our results. Therefore, in contrast to the above, this study examines cell-associated DNA in the context of necrotizing enterocolitis in premature infants. When we detect intestinal cells, vascular endothelium, and neurons, this indicates these cells are most likely circulating from their organ of residence. The mechanisms behind these findings will need to be examined, but these studies demonstrate that methylation analysis of the peripheral blood can provide valuable insight into organ-specific pathology in preterm neonates.

We recognize this study has limitations, including the small cohort and the blood collection at the time of diagnosis of NEC. Given the small size of this pilot study, we were unable to control for the variety of maternal and infant factors that may potentially influence DNA methylation patterns, and thus, there may be biological differences between groups that impacted the results of the study. Future studies will need to be conducted in a larger cohort of infants to optimize the matching of potential confounding variables and determine if methylation changes can predict NEC risk prior to disease onset and accurately prognosticate potential disease severity. A multi-center study would be optimal to assess the generalizability of these findings to a larger patient population. Finally, although DNA methylation analysis provides broad insight into biological dysregulation, its impact on RNA expression is highly nuanced. Although, in many instances, increased methylation is associated with reduced RNA expression, and vice versa, there are an abundance of examples where this is not the case [39–42]. Therefore, the pathway analysis we undertook combined, both in increased and decreased methylation changes, so as not to mislead the reader into assuming that these analyses were based upon a prior understanding of inverse correlations between DNA methylation and RNA expression.

5. Conclusion

In conclusion, we present the first analysis of peripheral blood methylation patterns in a cohort of preterm infants with and without NEC. We used deconvolution analysis to determine that the frequency of cells from the ileum significantly increased, whereas cells from the colon were significantly decreased in blood samples from patients with surgical NEC relative to their control counterparts. This novel observation will provide the basis for further studies examining tissue-specific methylation signatures in the blood of preterm infants with and without NEC and supports the possibility for methylation patterns to be used as a biomarker for this devastating disease.

Funding Statement

The work was supported by the Center for Gastrointestinal Biology and Disease, School of Medicine, University of North Carolina at Chapel Hill [N/A]; Chan Zuckerberg Initiative [2022–316749]; Eunice Kennedy Shriver National Institute of Child Health and Human Development [R01HD105301]; National Institute of Diabetes and Digestive and Kidney Diseases [DP1DK140012]; School of Medicine, University of North Carolina at Chapel Hill [N/A]; Thrasher Research Fund [N/A].

Article highlights

• In this pilot study, peripheral blood methylation patterns were examined in a cohort of preterm infants that included controls and infants with medical NEC or surgical NEC.

• Deconvolution analysis demonstrated that infants with NEC have a methylation signature in the peripheral blood that is consistent with an increased frequency of cells from the ileum and a decreased frequency of cells from the colon relative to controls.

• Differences in methylation signatures were most pronounced when comparing controls to infants with surgical NEC.

• This is the first study demonstrating that organ-specific cellular signatures in the peripheral blood can be used to successfully identify infants with and without NEC, and raises the possibility that DNA methylation signatures have future utility as a biomarker for NEC.

Competing interest disclosure

Declaration of Interests: David Peters and Tianjiao Chu are founders and equity holders of Signature Diagnostics (SDx). SDx played no role in this study. The other authors have no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants, or patents received or pending, or royalties.

Writing assistance disclosure

No writing assistance was utilized in the production of this manuscript.

Financial disclosure

This manuscript was funded by the following sources. Lauren Frazer is supported by a K08DK140604 from the National Institutes of Health (NIH), a Thrasher Research Fund Early Career Award, UNC School of Medicine Physician Scientist Training Program Faculty Award, a UNC Center for Gastrointestinal Biology and Disease Pilot Award, and a UNC Children’s Development Early Career Investigator Grant through the generous support of donors to UNC. Misty Good is supported by NIH grants R01DK118568, R01HD105301, R44HD110306, DP1DK140012, the Chan Zuckerberg Initiative Grant number 2022–316749, and the University of North Carolina at Chapel Hill Department of Pediatrics. David Peters and Misty Good are supported by R01DK124614 from the NIH. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Ethical conduct of research statement

After consent was obtained from parents or legal guardians of all infants at enrollment, samples were collected from infants admitted to the Children’s Hospital of Pittsburgh or Magee-Womens Hospital Neonatal Intensive Care Units of the University of Pittsburgh Medical Center per the guidelines of the University of Pittsburgh Institutional Review Board (IRB) (Protocol Number PRO09110437).

Data availability statement

The data discussed in this publication have been deposited in NCBI’s Gene Expression Omnibus (Edgar et al., 2002) and are accessible through GEO Series accession number GSE213704 https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE213704

Author contributions

Study design by Misty Good and David Peters. Manuscript preparation by Lauren Frazer, Misty Good, Tianjiao Chu, Caitlin Green, and David Peters. Critical review of the manuscript by all authors. Data generation and sample processing by Patricia Shaw, Camille Boufford, Lucas Naief, Michaela Ednie, Caitlin Green, and Laken Ritzert. Data analysis by Misty Good, Tianjiao Chu, and DP. Research coordination and sample acquisition by Misty Good.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/17501911.2025.2459552