Asthma is a complex chronic disease that is influenced by multiple factors including genetics and epigenetics. Here, we have identified 221 unique CpGs that significantly associate with PC scores generated from an advanced, comprehensive panel of 35 clinical markers of asthma. Of these, 190 have previously been found to associate with asthma and asthma-related phenotypes, while the other 31 CpGs are novel findings. Two of these CpGs reside within close proximity (1 MB) to 17 SNPs previously identified in a large GWAS of asthma. We then performed a longitudinal assessment in methylation data from whole blood samples collected approximately 10 years and identified 49 unique significantly associated CpGs and obtained a correlation of the estimates across the two timepoints of 0.82 from the 271 (222 T1 + 49 T2) significant associations.
Historically, asthma as a disease has been challenging to identify the root cause of and to treat effectively [32]. The disease heterogeneity and mixed involvement of environmental stimuli and the genome have largely contributed to this difficulty. Based on GWA studies, certain loci within the genome are associated with most forms of asthma, while specific subtypes of asthma are associated with distinct genetic loci [33, 34]. Further, genetic pleiotropy is most certainly at play within asthma, and other complex diseases, leading to increased complexity when attempting to determine genetic causality. While the genetic involvement of this heterogeneity has started to become elucidated, much of our understanding about variability in phenotype remains to be discovered.
We utilized measured clinical markers of asthma in young adults and their middle-aged parents from families with a history of asthma where at least one individual within the family had been diagnosed with asthma. For each individual, an extensive number of clinical markers related to asthma were captured. This comprehensive dataset, while valuable in its depth, is too large -given the sample size- to glean the valuable nuances through individual assessments of each specific marker. Additionally, some markers that were collected lacked a normal distribution, which is challenging in a traditional EWAS approach. For these reasons, we applied PCA to reduce the dimensionality of the 35 different variables, into 10 PCs with each successive PC capturing a lesser and different portion of the overall variance within the data set, making comparisons of it with a rich DNA methylation dataset much more feasible. Importantly, the PCs maintained a much more normal distribution, making them ideal for EWAS. This method of phenotype data reduction and handling has been previously implemented in a GWAS of rice variants, where GWASs were successfully performed using the PC scores from multiple phenotype variables [35]. To our knowledge, this is the first instance of implementing it for epigenome-wide association studies.
Through our EWASs of each of the 10 PCs, we identified 222 CpGs to be significantly associated with one of the PCs derived from the 35 clinical asthma markers. Of these 222 associations, only a single CpG was found to be replicated across multiple principal components, leaving 221 unique CpGs to be significantly associated. The single overlapping CpG, cg07329820, was identified with PCs 9 and 10. These two PCs align with asthma markers that are very similar (largely allergen response measures), which could explain why they identified the same CpG.
A breakdown of the identified CpGs from each PC shows some intriguing details. PC7 had the greatest number of CpG associations with 204. While this PC accounts for only 4.3% of the total clinical marker variance, it almost exclusively represents the variance associated with blood eosinophil counts and blood immunoglobulin levels. This large number of associations could be indicative of the eosinophilic inflammatory response typically seen in allergic asthma. Previous literature has shown that methylation experiments in whole blood may capture more of the Th2-mediated immune response due to the nature of the cell population, making it the ideal sample source for identifying this type of association [36,37,38]. It is critical to note that because the proportions of cells themselves in a sample can drastically influence the measurement of methylation, we made sure to incorporate cell proportion estimates as a covariate in our analyses. Without this inclusion, it could be argued that the findings identified here are only due to differences in cellular composition. With this correction in mind, it may be the case that the differential methylation identified here is something global, across multiple immune cell types, either in response to or to promote the eosinophilic inflammatory pathway. A query of the EWAS Atlas of these 204 CpGs showed that, almost exclusively, the CpGs we identified have a previous association with asthma and measurements of asthma (Supplemental Fig. 4). In total 190 of the 204 CpGs had some form of previous association (93.1%), leaving 14 novel associations. Since the majority of overlap of our findings with previous EWASs of asthma, asthma phenotypes, and symptoms are captured within this EWAS of PC7, we postulate that many of the previous findings generated in studies utilizing whole blood samples could be identifying associations with the underlying influences of increased eosinophil counts/activation and increased immunoglobulin levels.
An additional investigation of these 204 CpGs via the eFORGE online database identified multiple significantly associated tissue-specific regulatory elements (Supplemental Fig. 5). The greatest number of associations were from subpopulations of white blood cells in enhancers and weak transcriptional activators. This is to be expected considering our results were generated from samples of whole blood. Many of the subpopulations identified here are lymphocytes, potentially indicating that the methylation profiles of these adaptive immune cells may be altered due to the changes in eosinophil counts and immunoglobulin levels. The tissue group with the next largest number of significant associations was the digestive tract, with six significant associations to various subpopulations of cells. These associations could hint at the conserved role that all mucosal and epithelial barriers play in pathogen protection and the body’s response to the external environment. It has also been previously shown that respiratory viral infections can cause dramatic changes on both the presentation of asthma symptoms as well as DNA methylation [39]. In a previous study, Zhu et al. identified 33 differentially methylated CpGs (DMCs) associated with bronchiolitis severity [39]. An assessment for any overlap between these DMCs and our significantly associated CpGs did not identify any overlap, though it could be possible that pathogen exposure could still be playing a role in the DNA methylation signature associated with asthma.
A KEGG pathway enrichment analysis (conducted via the EWAS Atlas) of the 221 unique CpGs highlighted the FoxO signaling pathway, showing that 7 of the CpGs (all found within the PC7 EWAS) have been previously identified to associate with this pathway. This pathway is of particular interest due its significant role in regulating apoptosis, glucose metabolism, oxidative stress, and longevity, which has led to findings associating its dysregulation with other illnesses such as Alzheimer’s disease, type 2 diabetes, and cancer [40, 41]. It has been previously shown that FoxO1 modulates IL-9 generating Th9 cells [42]. Further, overexpression of IL-9 in the lungs can result in eosinophilic inflammatory infiltration and mucus secretion [42, 43]. Additionally, FoxO3 has been found to be expressed in airway epithelium playing a critical role in controlling the innate immune response to airway infections [43]. It could be, based on these intertangled findings of DNA methylation with eosinophilic presence and FoxO signaling, that epigenetic modulation of FoxO signaling plays a role in its contribution to asthma and airway inflammation.
Principal component 1 identified a single significantly associated CpG, cg18182148. While this principal component does capture the largest amount of variance (35.3%), the variance it captures is largely from one specific subset of measurements, most notably those measuring general lung functionality such as vital capacity, peak flow, forced expiratory volume, and methacholine challenge response. This is continued with PC2, PC3, and PC4 which were significantly associated with 4, 2, and 1 CpGs, respectively. These PCs, like PC1, capture a significant amount of variance from these same lung function measurements. This may indicate that these measurements of lung function, though important in characterizing symptoms of asthma, may have limited identifiable associations with DNA methylation. The specific CpG significantly associated with PC1, cg18182148, is the only CpG of these identified here related to immune regulation. It has not been previously shown to associate with asthma phenotypes, though it has been previously identified in studies of various cancers of the prostate, liver, and colon (via the EWAS Atlas) [44,45,46,47]. These associations are likely due to its placement within the transcription start site (TSS) of GFI-1, a strong oncogenic and hematopoietic regulator. GFI-1 plays a critical role in lymphoid differentiation, which could be the reason for its association here due to the large influence the immune system can play on lung functionality and, more generally, asthma as a whole [48]. While analyzing a whole blood sample, as done in this study, is a reasonable approach for investigating immune-mediated conditions like asthma, examining the relationship between lung function measurements and DNA methylation from lung epithelial tissue may reveal additional insights not captured here. This could offer a deeper understanding of the local cellular environment influencing specific lung function differences.
Principal components 5 and 6 capture variance associated with bronchial hyperreactivity, coughing, reversibility (of lung function following the administering of ventolin), and Tiffenau Index. PC5 was significantly associated with 2 CpGs, while PC6 did not have any significant associations. The two CpGs identified here, cg07454584 and cg18561513, have not been previously shown to have any associations with asthma or other immune-mediated illnesses (via the EWAS Atlas). Like the measurements captured in PCs 1–4, these asthma markers, while important for asthma diagnosis and the understanding of symptoms, may not have large quantities of associations with DNA methylation measured via whole blood.
Principal components 8, 9, and 10 contain much of the variance from the numerous allergen tests included in this study (containing both skin prick test results and serum specific IgE measurements). These PCs were significantly associated with one, two, and 5 CpGs, respectively. Of the 8 CpGs identified here, all but one has not been previously shown to associate with chronic illnesses or immune-mediated diseases. Cg14161241, which was significantly associated with PC10, is the lone exception with previous associations with obesity and type 2 diabetes [49]. Like many of the other markers represented via PCs 1–6, the effect on DNA methylation measurable via a whole blood sample is likely limited with these markers.
Due to the well-documented genetic contributions to asthma, we next compared our findings within the methylome to those previously discovered in the genome via large GWASs [10, 50]. Based on the findings of Demenais et al., we searched for significantly associated SNPs within 1 MB of the 221 unique CpGs that we identified. Of the significantly associated SNPs (p < 5 × 10–8, N = 892 SNPs), we found 17 (1.9%) to be within 1 MB of 2 of the 221 CpGs that we identified (cg02046836, cg23515090). Interestingly, neither of these two CpGs have been previously identified as an mQTL [31]. The proximity of these SNPs to the identified CpGs could be caused from multiple factors outside of known standard mQTL associations. It could be that these CpGs do have some genetic influence acting on their methylation status that has not yet been characterized. Additionally, it is possible that the SNPs and CpGs that are in close proximity to one another is simply a random occurrence, and their presence and downstream effects are independent of one another. To summarize, we found limited overlap of our EWAS signal with top loci from a previously performed GWAS of asthma and that these significant CpGs nearby these GWAS SNPs were not directly affected by mQTLs, which suggests that our EWAS largely captures independent epigenetic signal at loci that also harbor small amounts of genetic variants that influence genetic susceptibility to asthma.
Thus far, we have shown significant associations with clinical markers of asthma from a cross-sectional timepoint that corroborates other previous findings from similar studies. However, an increase in our understanding of asthma may lie in investigating the disease to some longitudinal capacity. Two previous studies in the Isle of Wight Birth Cohort (IOWBC) were conducted in 2022, which investigated the association of DNA methylation with asthma acquisition across adolescence and adulthood [51, 52]. In each of these studies, several CpGs, with some residing in immune regulatory genes, were identified to be associated with asthma acquisition [51, 52]. The findings from these two studies highlight promising insight into the potential longitudinal persistence of DNA methylation signatures associated with asthma.
After our initial findings, with the availability of an additional sample collected after 10 years for many of the individuals, we investigated the longitudinal persistence of the DNA methylation signature that was identified to associate with asthma markers at the first timepoint. This analysis was based on the same PC data from the original 35 clinical markers measured at baseline. In total, these 10 EWASs identified a total of 49 significantly associated CpGs, without replication across multiple PCs. When looking within a 100 kb window of proximity, we found that 3 CpGs identified at timepoint 2 (found via PCs 2, 3, and 7) were within 100 kb of 4 CpGs identified at T1 (all found via PC7). A query of the EWAS Atlas, however, using these 49 CpGs identified at T2 showed that 3 and 1 CpGs have been previously associated with FENO and allergic sensitization, respectively (Fig. 3). This highlights that there may be some longitudinal persistence of an asthmatic epigenetic signature. We estimated, when comparing the estimates of all 271 significant associations (222 + 49) across the two timepoints, a strong correlation of 0.82 adding to the notion that some long-term persistence of the asthma-associated methylation signature is occurring. It would have been of value if our study of asthma clinical markers had been reassessed along with the longitudinal blood sampling, as changes in asthma prevalence and presentation could lead to changes in DNA methylation, which could be better captured in future studies. Additional studies investigating the long-term lung functionality following asthma diagnosis, even in cases of remission, could benefit from investigating DNA methylation longitudinally in a more nuanced, complete manner.
Our study does have some limitations that should be considered. First, all individuals included in this study are from the Netherlands and of European descent. Asthma prevalence can differ greatly depending on the local environment of the individual and their ethnicity. Specifically, the environment of the Netherlands and their clinical diagnostic criteria could influence an individual’s asthmatic status. Including a diverse population could improve the generality of the findings to other groups of individuals. It is well documented that research involving non-European populations is lacking greatly, and we strongly advocate for research within these populations to expand the generality and applicability of research findings to these populations of individuals. The results generated from the second time point contained a lower number of individuals, who took part in the NTR biobank study and did not have clinical marker measurements, which could have shed some insight on the persistence of the asthma symptoms and severity at that time. Future studies investigating asthma longitudinally could benefit greatly from having longitudinal DNA methylation and clinical marker data. Overall, increasing the number of individuals at both timepoints using multiple individuals from varying backgrounds would be the most optimal strategy and offer the most comprehensive assessment of asthma.
Comments (0)