Preterm birth (PTB), defined as birth prior to 37 weeks gestational age (GA), occurs in approximately 11% of live births and accounts for over half of infant mortality cases worldwide [1]. If a premature infant survives the immediate postnatal period, they face increased risk of developing major short- and long-term health problems including cerebral palsy, chronic lung disease, visual and hearing impairments, and adult metabolic diseases [2–6]. This elevated risk is attributable to organ immaturity, as well as an increased risk of medical complications linked to oxidative stress and inflammation during the neonatal period [7–9].

The immune system is not spared from the effects of PTB. The composition and function of hematopoietic cell populations change dramatically throughout gestation as the embryonic and fetal immune system mature. Premature interruption of the immunologically protected intrauterine environment results in an extremely fragile infant whose immune system is unprepared for the microbe-ridden external environment. A variety of systemic and cell-specific alterations in immune function have been identified in preterm infants that greatly increase their vulnerability to infection [10–12].

The importance of DNA methylation (DNAm) in the process of hematopoietic cell lineage commitment is well established [13, 14], and multiple studies have found differential methylation between cord blood of preterm and term infants [15–17]. However, these studies have used whole blood, which is a mixed-cell sample in which overall DNAm levels are influenced by cell composition [18, 19]. As a result, these studies cannot distinguish prematurity-associated DNAm patterns due to differences in cell composition from DNAm patterns reflecting developmental changes in immune function.

Using the Illumina Infinium Human Methylation 450 Bead Chip (450K array), we provide genome-wide DNAm profiles of T cells, monocytes, granulocytes, and nucleated red blood cells (nRBCs) collected from cord blood of infants born at term or highly preterm (<31 weeks GA). These DNAm profiles were compared between cell types and across GA to evaluate an epigenetic basis for altered neonatal immune function with prematurity. This study provides important insights into the role of DNAm in early hematopoietic system maturation in humans.

Previous DNAm studies using cord blood have identified significant differences between preterm and term infants [15–17]; however, interpretation of these studies is limited by the confounding factor of cord blood cell composition. Granulocytes and T cells are the two most abundant cell types in whole blood and thus are the most likely to influence overall DNAm, but cell type proportions show considerable inter-individual variability and also change with gestational age [12]. Some DNAm changes previously associated with prematurity may simply reflect these changes in cell composition with GA. This study is the first evaluation of the epigenetic impact of PTB in hematopoietic cell populations isolated from the same individuals.

An important question is the functional role of these DNAm changes in hematopoietic cell populations during ontogeny. There was a notable difference in the number of cell-specific prematurity-associated DM sites in each cell type, ranging from 273 in T cells to 9258 in nRBCs (Table 2). The number of prematurity-associated DM sites in a given cell type may relate to the magnitude of phenotypic differences between preterm and term populations. For example, we and others have reported major functional differences in dendritic cells and macrophages between preterm and term infants [37–40]. In contrast, fewer functional differences have been observed between preterm and term T cells [41, 42]. For granulocytes, much less is known regarding gestational differences. The high number of prematurity-associated DM sites we observed in granulocytes (987) suggests a more dynamic maturation across late gestation than for either monocytes or T cells. Alternatively, it is possible that these DNAm changes may reflect differences in the composition of granulocyte subsets, including a mixture of eosinophils, basophils, and mast cells, between age groups. However, this is less likely given that our granulocytes were overwhelmingly represented by neutrophils in both preterm and term samples (>95%; data not shown). Given the extent of DNAm differences between preterm and term granulocytes, functional studies may provide new insight into the limitations of the preterm immune system.

Our findings showed moderate overlap with previous studies of prematurity-associated DNAm in whole cord blood, with approximately 30% of the DM sites from each comparison study also discovered in at least one of our cell types (Fig. 3a). This is a greater amount of overlap than the 161 of 1347 CpG sites Fernando et al. [16] found in common with Cruickshank et al. [15] and another study not evaluated in this paper [34]. This increased overlap with other studies may reflect reduced noise our data due to eliminating variation due to cell composition differences. Alternatively, since we compared four sets of prematurity-associated DM sites (one per cell type), all of which were of fairly large size, we may have increased our chance of overlap just by having a greater number of hits.

When the cell type-DM sites discovered in this study are overlapped with the three comparison studies, the highest numbers of overlapping sites were observed with T cell- and nRBC-specific DM sites, the two cell types with the strongest cell-specific DNAm patterns (Fig. 3b). We additionally compared our cell type-DM sites to a subgroup of 196 CpG sites identified by Fernando et al. to be associated with PTB but not GA—and thus thought to reflect systematic differences due to prematurity rather than cell composition—and found almost no overlap (Fig. 3b). This supports their assertion that those 196 CpG sites are more likely related to the molecular mechanisms of PTB than cell composition, compared to the 1151 CpG sites they identified as associated with both PTB and GA. Thus, subsequent studies of PTB may be able to work around concerns of variability in cell composition by considering prematurity as a separate variable from GA.

Two CpG sites highlighted by Fernando et al. as being of potential interest for preterm delivery were not replicated in our study. One site, in MYL4 (myosin light chain 4), was the only DM site identified by Fernando et al. [16], Cruickshank et al. [15], and Schroeder et al. [34]; the other, ESR1 (estrogen receptor) was observed in both Fernando et al. [16] and Schroeder et al. [34]. Fernando et al. suggested that these sites may be related to the labor process, since MYL4 activity is involved in the myometrial contraction pathway [43], and upregulation of ESR1 leads to the increase in estrogen activity required for contractions [44]. The lack of replication in our data may be a consequence of all of our subjects being born by caesarean section, whereas all of the comparison studies included at least some subjects born by vaginal delivery. Notable genes in which we replicated the differential methylation found by other studies include ADORA2A, which has been associated with the inflammatory pathway in the myometrium [45], and GABBR1, which encodes a gamma-aminobutyric acid receptor and has been associated with chemotaxis in cord blood-derived stem cells [46]. These two genes were identified as differentially methylated in both Fernando et al. [16] and Cruickshank et al. [15].

GO pathway analyses of the prematurity-associated DM sites highlighted potentially important differences in gene regulation that are unique to each cell type (Additional file 3). For instance in granulocytes, prematurity-associated DM sites were significantly enriched for genes associated with the Ras-Raf-MEK-ERK cascade. Defects in this pathway have been associated with impaired neutrophil extracellular trap formation and with respiratory burst in neutrophils [47, 48], both of which are also deficient in preterm infants [49, 50]. The prevalence of prematurity-associated DM sites in genes associated with these functions could reflect either reduced functional ability in preterm neutrophils or a low proportion of neutrophils within the preterm granulocyte population. Some of our findings point towards novel pathways potentially involved in the maturation of hematopoietic cells, such as the enrichment for prematurity-associated DM sites in genes associated with placental development in T cells and with dermal development in monocytes. In nRBCs, prematurity-associated DNAm changes were widespread and associated with GO terms related to the cytoskeleton, membrane composition and cell-cell junctions, and motility. This may reflect the large-scale structural changes that occur in erythroblasts as they mature and prepare to extrude their nucleus.

Based on gestational age differences, DNAm conformed to the epigenetic profile of the dominant hematopoietic organ when evaluated in source-DM sites [28]: the liver in mid-gestation, and the bone marrow in late gestation (Fig. 5). Despite these candidate sites being identified exclusively in nRBCs derived ex vivo from hematopoietic stem cells, the DNAm trends in this study were consistent across both nRBCs and WBCs. Thus, our findings indicate that hematopoietic sources have epigenetic signatures that are shared across multiple cell lineages derived from that organ. Additionally, our analysis of these hematopoietic source-DM sites revealed that nRBCs actually gain DNAm with increasing GA in functionally relevant regions of the genome, specifically the fetal liver-hypomethylated sites. This is a rare occurrence, as the overwhelming trend is for nRBCs to become demethylated both during erythropoiesis [29, 30] and as the fetus approaches term. This novel observation is important to our understanding of hematopoiesis during ontogeny since it indicates that although nRBC demethylation is largely global and passive [29, 30], it also has some selectivity, with certain CpG sites protected from the widespread DNAm loss.

The main limitation in our study is the small sample size. Other studies evaluating prematurity-associated DNAm had sample sizes ranging from 22 [16] to 50 [17]. With only ten subjects, our study had reduced power to detect changes in DNAm. Considering the large epigenetic differences between cell lineages, we expect that our study was sufficiently powered to compare cell types. However, differential methylation associated with prematurity is expected to be of a smaller scale than cell type differences, so this may have led to an underestimation of prematurity-associated DM sites. There was also an increased chance that genetic factors impacted our findings, since some CpG sites are methylation quantitative trait loci (mQTLs), or sites where DNAm is more strongly associated with individuals than cell type [20, 51]. We mitigated this concern by performing an additional probe filtering step to remove suspected mQTLs, as described in the Supplementary Methods (Additional file 1).

It is possible that heterogeneity in our subjects’ clinical characteristics reduced our ability to detect prematurity-associated differential methylation. All births were caesarean sections with no indications of infection; however, one preterm case was attributed to preeclampsia, and four of the five preterm births were multiples (Table 1). This raises the concern that multiplicity in the preterm subjects may have confounded our results. There is limited information on how the immune system differs with multiple births: it has been shown that intrauterine infection occurs more often in preterm births with dizygotic twins compared to monozygotic twins or singletons, but no differences in postnatal outcome have been associated with zygosity [52]. Additionally, CD4⁺ T cell activity has been observed to be significantly lower in preterm dizygotic twins than in preterm singletons [53]. For DNAm studies, the effect of twin births has only been assessed within twin pairs [54], not between twins and singletons. For our study, considering the extreme difference in GA of our preterm and term cases (<31 versus >38 weeks), we expect that prematurity will have a much greater effect on DNAm than differences due to multiplicity. This is in keeping with Fernando et al.’s [16] observation of more distinct clustering of DNAm in extreme PTB cases compared to intermediate PTB cases. Despite these limitations, we identified prematurity-DM sites that showed reasonable overlap with prior studies [15–17] and cell-specific DNAm patterns that were consistent with our previous findings in cord blood cell populations [20, 32].

The preterm immune system differs from that of the term neonate in both cell composition and function, resulting in heightened vulnerability to infection in preterm infants. We identified epigenetic markers of immune system differences with prematurity by comparing DNAm of major cord blood hematopoietic cell populations across gestation. Changes in DNAm between preterm and term hematopoietic cells in our study likely reflect a shift from the liver to the bone marrow as the predominant hematopoietic source with advancing gestational age. Granulocytes were identified as a candidate cell population of particular interest in preterm infants’ susceptibility to infection, due to their relatively high number of prematurity-associated DM sites and the enrichment of these sites for GO terms related to the Ras-Raf-MEK-ERK cascade. Our findings provide important insights into the epigenetic regulation of hematopoietic cell-specific functions during fetal development. These data may have clinical implications, as they highlight gene regulatory mechanisms on both cell-specific and systemic levels that are involved in neonatal immune system maturity. Larger samples will be required to determine the potential impact of cause of PTB (such multiple gestations or preeclampsia) on these epigenetic profiles.