The chemical tags in question are methyl groups, small molecular attachments that bind to specific positions in the genome called CpG sites, where a cytosine nucleotide sits next to a guanine. DNA methylation is a classic epigenetic mark: it does not change the underlying genetic code, but it can dial gene activity up or down. Because the epigenome responds to environmental stimuli and physiological stress, the researchers reasoned that the massive injury caused by an aneurysm rupture might reshape methylation patterns in ways that influence the biological processes driving DCI, including vascular dysregulation, microthrombosis, cortical spreading depolarizations, and neuroinflammation. Previous candidate-gene studies had pointed to a handful of genes such as INSR, ITPR3, and HAMP, but genome-wide searches had been rare.

The team adopted a two-stage design that has become the gold standard in human genetics. In the discovery stage, they profiled methylation across roughly 450,000 CpG sites using the Illumina Infinium HumanMethylation450 BeadChip in blood samples drawn from 68 aSAH patients within 48 hours of their hemorrhage, before any sign of DCI. This timing matters: by collecting samples before the complication appeared, the investigators ensured that any methylation differences they found could not simply be a consequence of DCI itself. The patients, recruited through the University of Pittsburgh Medical Center Neurovascular Intensive Care Unit, had a mean age of 53.7 years, were predominantly female, and were followed for up to one year to track both DCI occurrence and long-term recovery using the Glasgow Outcome Scale and Modified Rankin Score.

The statistical machinery behind the discovery scan was deliberately rigorous. The researchers applied functional normalization to correct batch effects and technical noise, filtered out probes overlapping common genetic variants, sex chromosomes, and cross-reactive regions, and retained 418,247 CpG sites for analysis. They used M-values, which have more favorable statistical properties than raw beta values, and fitted linear models with empirical Bayes moderation while adjusting for age and sex. Crucially, they also incorporated surrogate variables derived from surrogate variable analysis to absorb the effects of cell-type heterogeneity, a notorious confounder in blood-based methylation studies, since different white blood cell populations carry distinct methylation signatures. The significance threshold for the epigenome-wide scan was set at a stringent 2.4 × 10⁻⁷.

No CpG site crossed that threshold. But the strongest signal in the entire scan landed at a site called cg18031596, annotated to a gene with an obvious biological connection to the disease process: ANGPT1, located on chromosome 8, which encodes angiopoietin-1, a protein central to blood vessel repair and stability. The p-value there was 2.3 × 10⁻⁶, and a post hoc power calculation showed the study had 94.6 percent power to detect a true difference of the observed magnitude. Angiopoietin-1 acts through the Tie2 receptor on endothelial cells, maintaining the integrity of the vascular lining, blocking leakage, and exerting anti-inflammatory effects. In mouse models, overexpression of the protein reduces ischemic lesion volume after arterial occlusion, and earlier clinical work had found that aSAH patients who later developed DCI showed decreased serum levels of angiopoietin-1 early after admission. All of this made ANGPT1 a compelling suspect.

To test the signal properly, the team turned to a replication cohort of 175 additional aSAH patients recruited under identical criteria at the same center. Rather than re-running the full array, they used targeted bisulfite pyrosequencing to measure methylation at five CpG sites across the ANGPT1 locus, including cg18031596 and four neighboring sites. Fifty-eight of the original discovery samples were re-assayed with the same sequencing method, bringing the combined dataset to 233 individuals, of whom 80 developed DCI. The sequencing measurements correlated well with the original array data, with Spearman coefficients ranging from 0.64 to 0.86, and the five targeted CpGs were highly correlated with one another, as expected for sites packed into a single regulatory region.

Then came the twist that gives this study its scientific drama. In the replication cohort, four of the five CpG sites were indeed significantly associated with DCI, passing the Bonferroni-adjusted threshold of 0.045. But the direction of the effects was flipped. In the discovery scan, higher methylation at cg18031596 had appeared to raise DCI risk; in the replication samples, higher methylation was associated with lower risk, with DCI cases showing a mean methylation of 4.84 percent versus 5.56 percent in controls. In the pooled analysis of all 233 patients, three of the five sites remained significant, and the strongest signal, at p = 0.004, came from a CpG that had not even been measured in the original array. By the strict logic of replication, the study failed: a consistent, reproducible effect of ANGPT1 methylation on DCI risk could not be demonstrated.

Rather than burying the inconsistency, the researchers dissected it with unusual transparency. They examined whether smoking, a known driver of methylation changes and a risk factor for aSAH, could explain the flip, but the pattern was messy, with some sites flipping only in smokers and others only in non-smokers. They tested whether the use of surrogate variables in the discovery stage, which could not be replicated in the smaller sequencing dataset, accounted for the discrepancy; this explained the sign reversal at four of the five sites but not at cg18031596 itself. They also noted that the two measurement technologies, array and pyrosequencing, though correlated, were not identical, leaving room for platform-specific artifacts. Finally, they ran a striking simulation: 10,000 random splits of the combined sample into discovery-sized and replication-sized groups. Among the 1,229 simulated discovery sets that produced a significant result, eight showed effect estimates as extreme as the real one, and in every one of those eight, the replication coefficient came out in the opposite direction. In other words, once a small discovery sample stumbles onto a chance extreme result, a direction-flipped replication is almost guaranteed, a phenomenon with a correlation of −0.87 between discovery and replication coefficients in the simulations.

The study also explored whether ANGPT1 methylation might predict recovery rather than DCI itself. In the discovery data, lower methylation at cg18031596 was associated with better Glasgow Outcome Scale scores at three months, with a borderline association at twelve months, and in the sequencing data higher methylation was linked to a lower risk of death at three months. None of these associations replicated cleanly, however, echoing a previous report that high serum angiopoietin-1 predicts better functional outcome. The authors caution that the relationship between methylation and gene expression is tissue-specific and dynamic, and it remains unknown whether methylation at cg18031596 actually regulates ANGPT1 expression in the relevant biological context, though public datasets suggest a correlation in several tissues.

The honest conclusion is that ANGPT1 methylation remains an open question rather than a validated biomarker. The study’s strengths include its longitudinal design, which rules out reverse causality, its careful control of technical artifacts and cell-type heterogeneity, and its willingness to publish a negative and confusing result in full detail, something the field needs more of. Its limitations are equally instructive: small samples inherent to a low-incidence, high-mortality disease, the inability to adjust for methylation quantitative trait loci without genotype data, and the open debate over whether blood is the right tissue for studying brain conditions, even though immune cells in blood do reflect the inflammatory responses implicated in DCI. For now, the message for patients and clinicians is that no epigenetic test can yet predict delayed cerebral ischemia after aneurysm rupture. But the study lays out a clear roadmap: larger, well-powered cohorts with complete genetic and clinical data are needed before ANGPT1, or any methylation marker, can move from statistical curiosity to bedside tool.

Subject of Research: DNA methylation at the ANGPT1 gene as a potential biomarker for delayed cerebral ischemia after aneurysmal subarachnoid hemorrhage

Article Title: ANGPT1 methylation and delayed cerebral ischemia in aneurysmal subarachnoid hemorrhage patients

Article References: Liu, D., Arockiaraj, A. I., Shaffer, J. R., Poloyac, S. M., Sherwood, P. R., Alexander, S. A., Crago, E. A., Weeks, D. E., & Conley, Y. P. (2021). ANGPT1 methylation and delayed cerebral ischemia in aneurysmal subarachnoid hemorrhage patients. Epigenetics Communications, 1(1), Article 3. https://doi.org/10.1186/s43682-021-00001-7

Image Credits: AI Generated

DOI: 10.1186/s43682-021-00001-7

Keywords: aneurysmal subarachnoid hemorrhage, delayed cerebral ischemia, DNA methylation, ANGPT1, epigenome-wide association study, angiopoietin-1, biomarkers, neurocritical care, pyrosequencing, replication failure, surrogate variable analysis, stroke

The study focused on DNA methylation, a chemical modification in which methyl groups are attached to cytosine bases at specific locations in the genome, known as CpG sites. Methylation patterns act as a dynamic layer of gene regulation, influencing which genes are switched on or off without altering the underlying genetic code. Because these patterns can shift in response to disease processes, researchers have long suspected that methylation signatures in peripheral blood might reveal the biological pathways that go awry in preeclampsia, or even serve as biomarkers for predicting risk, identifying disease subtypes, or monitoring response to therapy. What set this investigation apart was its longitudinal design. Most previous epigenetic studies of preeclampsia sampled blood only after symptoms had appeared, capturing the aftermath of disease rather than its origins. By collecting blood samples in each trimester from women before any signs of illness, the team hoped to catch the earliest molecular ripples of the disorder.

The discovery phase drew on participants from the Prenatal Exposures and Preeclampsia Prevention Project, a NICHD-funded cohort recruited at UPMC Magee-Women’s Hospital between 2008 and 2014. The researchers assembled 56 women, half of whom went on to develop preeclampsia and half of whom remained normotensive throughout pregnancy, delivered at term, and birthed normally grown infants without proteinuria. Critically, each case was matched one-to-one with a control on self-identified race, pre-pregnancy body mass index, smoking history, and gestational age at sample collection, minimizing the influence of factors known to confound methylation measurements. Genome-wide methylation data were generated from the longitudinal blood samples using the Illumina Infinium MethylationEPIC Beadchip, an array that interrogates more than 850,000 CpG sites across the genome. After rigorous quality control removed problematic probes and samples, more than 700,000 sites remained available for analysis.

To squeeze signal out of noisy high-dimensional data, the team employed surrogate variable analysis, a statistical technique that identifies and removes unwanted sources of variation such as batch effects and the shifting proportions of different white blood cell types across pregnancy. This cell-type heterogeneity is a notorious pitfall in blood-based epigenetic studies, because methylation patterns differ sharply between immune cell populations, and pregnancy itself remodels the maternal immune system. The epigenome-wide association study was conducted separately within each trimester, regressing methylation values on preeclampsia status while adjusting for the surrogate variables. The results were sobering but not empty: no individual CpG site reached the stringent genome-wide significance threshold of 9 multiplied by 10 to the power of negative 8, yet 16 sites, distributed as five in the first trimester, one in the second, and ten in the third, met the suggestive threshold of 1 multiplied by 10 to the power of negative 5.

Recognizing that testing each CpG in isolation sacrifices statistical power, the researchers also searched for differentially methylated regions, contiguous stretches of the genome where neighboring CpG sites show consistent methylation differences between cases and controls. Using the dmrff method, which aggregates association statistics across nearby sites, they identified three significant regions in each trimester, nine in total, all surviving strict Bonferroni correction for multiple testing. In the first trimester, the top region mapped to RAB18 on chromosome 10; in the second, the strongest signal fell within NCOA4, a gene involved in iron metabolism, on the same chromosome; and in the third trimester, the leading region encompassed TRAF3IP2 and its antisense long noncoding RNA TRAF3IP2-AS1 on chromosome 6. That final region contained the two strongest individual hits from the third-trimester analysis, making it the natural candidate to carry forward into replication testing.

The biology of TRAF3IP2 makes it a compelling suspect in preeclampsia. The gene encodes an adaptor protein that links two potent inflammatory signaling pathways, interleukin-17 and transforming growth factor beta, both of which have been implicated in the maladaptive immunomodulation thought to underlie the disease. Recent work has also shown that TRAF3IP2-AS1 regulates interleukin-17A signaling and TRAF3IP2 expression through recruitment of the splicing factor SRSF10. Adding weight to the connection, an earlier study found that TRAF3IP2 was differentially methylated in umbilical cord blood from pregnancies affected by early-onset preeclampsia. Other suggestive hits from the discovery scan also pointed toward genes with plausible roles in pregnancy biology, including NOTCH4, a regulator of placental vascular development; COQ10A, involved in coenzyme Q metabolism; AQP11, an aquaporin water channel; SEPN1, a selenoprotein; and TFF3, a trefoil factor. Four of these six genes had previously been linked to preeclampsia in earlier research.

Replication, however, proved elusive. Because only third-trimester samples were available from an independent cohort, the team focused their replication efforts on the two top third-trimester CpG sites, cg16155413 and cg21882990, within the TRAF3IP2 region. They first validated their discovery findings by re-measuring methylation at these sites in the original discovery samples using pyrosequencing, a different laboratory platform. The two technologies correlated well, with Pearson correlations of 0.86 and 0.82 for cg16155413 and its adjacent site, and the validation sample showed significantly higher methylation in preeclampsia cases at both sites, mirroring the discovery results. But when the researchers turned to the independent replication sample of 114 women from earlier iterations of the same cohort, the association vanished. Methylation levels were actually slightly lower, though not significantly so, in cases compared with controls, and the direction of effect reversed relative to the discovery sample.

The failure to replicate does not necessarily condemn the original finding as a false positive. The discovery and replication cohorts differed in ways that could plausibly affect methylation signals. The discovery sample was composed predominantly of women who self-identified as Black, roughly 72 to 75 percent, while the replication sample was overwhelmingly composed of women who self-identified as White, around 90 percent. The replication participants also had lower average body mass index, greater maternal age, and different distributions of gestational age at birth. Because DNA methylation is well documented to vary by self-reported race and ancestry, these demographic differences could easily mask or reverse a genuine signal. The researchers also noted that the replication analyses could not apply the same surrogate variable adjustment for cell-type heterogeneity, since genome-wide methylation data were not available for the independent samples, leaving a potentially important confounder uncorrected.

The study’s strengths are considerable. It is the largest investigation to date of blood-based DNA methylation in preeclampsia, and its longitudinal design, careful one-to-one matching, stringent quality control, and sophisticated statistical treatment of cell-type heterogeneity and multiple testing set a high methodological bar. Yet the same rigor carries a cost: strict correction for multiple testing may have obscured real but modest signals in a modestly sized sample. The authors also emphasize that preeclampsia is not one disease but a constellation of subtypes, including early-onset and late-onset forms with distinct underlying mechanisms. Pooling all cases into a single group, as this study necessarily did given its size, may dilute subtype-specific methylation signatures and complicate the search for clinically useful biomarkers. Future cohorts will need to be substantially larger and stratified by subtype to untangle this heterogeneity.

For now, the nine differentially methylated regions that survived Bonferroni correction stand as the study’s most durable contribution, a roadmap of genomic neighborhoods, spanning chromosomes 2, 5, 6, 10, 14, 15, 17, and 19, that deserve systematic follow-up in larger and more diverse populations. The TRAF3IP2 region in particular remains a biologically plausible candidate that warrants re-examination with genome-wide data in independent cohorts. The broader lesson may be that a single universal methylation biomarker for preeclampsia is unlikely to exist, and that progress will depend on embracing the syndrome’s complexity rather than averaging over it. As the researchers conclude, additional research is warranted not only for the suggestive hits in the first and second trimesters, which could not be tested in replication, but for the full set of significant regions identified across pregnancy. Each of these epigenetic waypoints may ultimately help illuminate the pathways that transform a healthy pregnancy into a hypertensive crisis, and perhaps, one day, allow clinicians to intervene before symptoms ever begin.

Subject of Research: Longitudinal epigenome-wide association study of blood DNA methylation in preeclamptic versus normotensive pregnancy

Article Title: A longitudinal epigenome-wide association study of preeclamptic and normotensive pregnancy

Article References: A longitudinal epigenome-wide association study of preeclamptic and normotensive pregnancy. (n.d.). https://doi.org/10.1186/s43682-022-00014-w

Image Credits: AI Generated

DOI: 10.1186/s43682-022-00014-w

Keywords: preeclampsia, DNA methylation, epigenetics, EWAS, pregnancy, TRAF3IP2, biomarkers, differentially methylated regions, longitudinal study, hypertensive disorders of pregnancy, surrogate variable analysis, maternal health