Interrelationships and determinants of aging biomarkers in cord blood

Our study initially enrolled 199 eligible mother-newborn pairs with singleton newborns in the ENVIRONAGE birth cohort [42]. These mother-newborn pairs were recruited as a subset of the ongoing prospective cohort between July 2014 and June 2015 at the East-Limburg Hospital in Genk, Belgium. This study was conducted according to the principles outlined in the Helsinki Declaration [43] after approval by the Ethical Committee of Hasselt University and the East-Limburg Hospital in Genk. Written informed consent was obtained from all participating mothers at recruitment. Epigenome-wide methylation status of the CpG sites was retrieved from cord blood samples in the framework of the EXPOsOMICS project (FP7) [44]. Data on relative TL were available for 198 neonates. For six of these neonates’ data on mtDNA content was missing. Finally, two samples were removed from the analysis because they were classified as extreme outliers concerning the calculated DNAm age [> 3 × interquartile range (IQR) below the first quartile or above the third quartile], using the R package NCmisc 1.1.6. Therefore, the final sample size in this study was 190 (Fig. 2).

Maternal body mass index (BMI) was determined during the first antenatal visit (weeks 7–9 of pregnancy) by dividing weight in kilograms by height in meters squared. The conception date was estimated based on the first ultrasonographic examination. After delivery, mothers filled out a study questionnaire, which collected detailed information about sociodemographic and lifestyle factors for both mothers and fathers. Paternal age was missing in nine cases and imputed with the mean paternal age at birth. Parity was classified into three categories for mothers having their first, second, and third or more newborns. Maternal educational level was coded as “low" for mothers who did not obtain any diploma, “middle” when they obtained a high school diploma, and “high” when they obtained a college or university degree. Maternal smoking status was categorized as “never smoker,” when the mother never smoked before or during pregnancy, “former smoker” when the mother had quit smoking before pregnancy, and “smoker” when the mother continued to smoke during pregnancy. Newborn ethnicity was categorized based on the grandparents' origin and was classified as European when two or more grandparents were European and non-European when at least three grandparents were of non-European origin. The season of delivery was divided into the cold season (October 1—March 31) and the warm season (April 1—September 30).

Cord blood samples were collected directly after delivery in BD Vacutainer^® Lithium Heparin, Plus Plastic K2EDTA Tubes (BD, Franklin Lakes, NJ, USA) and centrifuged at 3200 rpm for 15 min. After that, buffy coat and plasma were separated and frozen instantly at − 80 °C.

Cord blood DNA was extracted and processed at the Epigenomics and Mechanisms Branch (formerly Epigenetics Group), International Agency for Research on Cancer (IARC). In detail, after thawing and extraction with the QIAamp DNA mini Kit (Qiagen Ltd, Manchester, UK), DNA was first bisulfite-converted using the Zymo EZ DNA methylation^™ kit (Zymo, Irvine, CA, USA), consequently hybridized to Illumina Infinium Human Methylation 450K BeadChip arrays [45] and scanned using the Illumina HiScanSQ system. After background subtraction with Illumina GenomeStudio, the raw intensity data were preprocessed, including the calculation of the methylation level at each CpG as the beta-value, the normalization using the funnorm normalization of the minfi package [46], and quality control employing in-house software within the R statistical computing environment. Samples underwent further quality control employing Illumina’s detection p-value > 0.05 and bead count lower than 3, excluding failed samples. Additionally, background subtraction and dye bias correction were performed on Infinium II probes. Finally, data were also trimmed for outliers containing values more than three interquartile ranges below the first quartile or above the third quartile so that 485,512 probes remained for analysis.

The quantity and purity of the extracted DNA were assessed by spectrometric analysis using the Nanodrop 1000 spectrophotometer (Isogen, Life Science, Belgium), and integrity was evaluated using agarose gel electrophoresis. All measurements were performed in triplicate on a 7900HT Fast Real-Time PCR System (Applied Biosystems, Foster City, CA, USA) using a 384-well format. DNA quantity was determined through the Quant-iT™ PicoGreen® dsDNA Assay Kit (LifeTechnologies, Europe), to ensure a uniform DNA input of 5 ng/PCR reaction. Average relative telomere length and mtDNA content were measured by a modified quantitative real-time PCR (qPCR) protocol as described previously [5, 47, 48] and in detail provided in Additional file 2: Text S1. PCR cycles are described in detail in Additional file 3: Tables S2–S4. Telomere assay‑precision expressed by the intra-class correlation coefficient (ICC) [49] was 0.936 (95% CI: 0.808 to 0.969) for the inter-assay ICC and 0.952 (95% CI: 0.947 to 0.956) for the intra-assay ICC was. Cycle thresholds of the telomere and mtDNA amplifications were normalized relative to the cycle thresholds of the single-copy gene amplifications Additional file 3: Tables S2–S4 using the qBase software (Biogazelle, Zwijnaarde, Belgium). Relative average telomere lengths and mitochondrial content were expressed as the individual relative ratio to the average ratios of the entire sample set.

DNAm age was calculated according to the epigenetic clock developed by Horvath using the Bioconductor package “methylclock” [50] with the cell count reference option according to Bakulski [51]. The degree of methylation was expressed as the percentage of methylated cytosines over the sum of methylated and unmethylated cytosines. Global DNA methylation was calculated by calculating the study population's arithmetic mean of the beta-values (epigenome-wide average DNA methylation). Additionally, mean methylation of the three functional gene regions, promoter (n = 140,003 probes), gene body (n = 158,210 probes), and intergenic regions (n = 187,299 probes) was calculated. TL and mtDNA were log₁₀ transformed to improve normal distribution.

In the statistical analysis, Pearson correlation was applied to address the relationship between the four aging biomarkers. In the next step, path analysis, adjusted for a priori selected covariates, was performed to assess the associations between the aging biomarkers and establish significant early-life aging determinants. This form of structural equation modeling (SEM) is characterized by multiple linear regression equations with simultaneous estimation of regression coefficients for all hypothesized relations between the variables. The path analysis model included sex, gestational age newborn ethnicity, birthweight, maternal smoking, maternal education, maternal early-pregnancy BMI, and parity as covariates. For TL and mtDNA content, additionally, white blood cell count, season of delivery, and parental age, and for global methylation cell-type distribution according to Bakulski [51] were included as covariates. The presence of random effects was taken into account by using (i) the residuals of DNAm age and global methylation regressed on array chip and array position and (ii) the residuals of TL and mtDNA content regressed on sample storage time. The following assumptions regarding the direction of the associations were made: (i) TL affected mtDNA content and was in turn affected by DNAm age and global methylation, (ii) DNAm age was affected by mtDNA content and global methylation, (iii) global methylation was affected by mtDNA content. The path analysis was accomplished with the lavaan package, version 0.6–5 [52]. Statistical significance was defined as p < 0.05. All data analyses were performed in RStudio using R 3.5.2.

Demographic characteristics and perinatal factors of the mother-newborn pairs are reported in Table 1. The newborns in this study were mostly of European origin (89%) with a mean (SD) gestational age of 39.15 (± 1.54) weeks and a mean (SD) birthweight of 3393 (± 484) g. The TL and mtDNA content range was between 0.51–1.58 and 0.25–4.35, respectively, and the mean (± SD) DNAm age was 0.46 (± 0.26) years. The mothers were on average 29 (± 4.5) years old and had a pre-pregnancy BMI of 24.39 (± 4.34) kg/m².

Figure 3 shows the inter-correlation of the different biological markers of aging. Cord blood DNAm age showed a significant, inverse correlation with all other aging biomarkers, except with TL where the correlation was not significant. The strength of correlation for DNAm age was in descending order with cord blood mean (i) gene body methylation, r = − 0.65, (ii) global methylation r = − 0.64, (iii) intergenic region methylation r = − 0.63 and (iv) promoter methylation r = − 0.57 (all p < 0.001) and with mtDNA content (r = − 0.16, p = 0.027). Cord blood TL and mtDNA content were significantly correlated (r = 0.26, p < 0.001). Global methylation was significantly correlated with promoter, gene body and intergenic region methylation (r = 0.92, r = 0.97 and r = 1, all p < 0.001).

When stratified for sex, these correlations remained significant (Additional file 4: Figure S1 and Additional file 5: Figure S2). For newborn girls, the correlation between TL with mean global, body, intergenic methylation, and DNAm age changed direction, as did the correlation between mtDNA content and mean promoter methylation in boys (all remaining non-significant).

The path analysis model representing the causal assumptions of the relationships between the aging biomarkers and a priori selected covariates showed a good overall fit (χ2 = 23.00, degrees of freedom = 27, p = 0.69; Root Mean Square Error of Approximation = 0.00, p = 0.963, Standardized Root Mean Square Residual = 0.034, Comparative Fit Index = 1.00). A graphical indication of the significant associations with standardized estimates is shown in Fig. 4.

Assuming all other variables were held constant, the strongest effect was shown for global methylation on DNAm age with a decrease of 0.64 standard deviations (SD) in DNAm age for each SD (0.01%) increase in global methylation (p < 0.001). The unstandardized coefficient translated to a decrease of 0.21 years of DNAm age for an IQR increase (0.011 units) of global methylation. For each SD increase in log₁₀ transformed TL, the mean level of log₁₀ transformed mtDNA content increased by 0.31 SD (p < 0.001) [for the unstandardized coefficients mtDNA content increased by 6.96% (95% CI: 4.97%, 8.95%) for a 10% increase in TL].

Concerning the covariates, season of delivery was associated with a decrease of 0.21 SD in log₁₀ transformed TL and 0.24 SD in log₁₀ transformed mtDNA content (p = 0.02 and p = 0.003, respectively) for the warmer half of the year compared to the cold half of the year. This corresponds for the unstandardized coefficients to a decrease of − 7.10% (95% CI: − 3.44%, − 10.63%) in TL and − 16.44% (95% CI: − 8.64%, − 23.58%) in mtDNA content respectively. Furthermore, an increase of one SD in gestational age (1.54 years) was associated with an increase of 0.20 SD in DNAm age (p = 0.017), indicating a 1.25 weeks higher DNAm age for each additional week of gestation. Newborns girls had on average 5.68% (95% CI: 2.95%–8.48%) longer telomeres, corresponding to an increase of 0.19 SD in mean log₁₀ transformed cord blood TL (p < 0.001), or unstandardized to an increase of 6.4% cord blood TL (95% CI: 4.38–8.42). Additionally, female sex was associated with a 0.3 unit higher global methylation (0.27 SD, p < 0.001), and a 11.49% (95% CI: − 9.43%, − 13.55%) lower mtDNA content (0.16 SD, p = 0.016). Moving from a lower maternal educational level to a higher one was associated with a decrease of 0.21 SD (p = 0.014) in DNAm age or a decrease of about 2.77 weeks in DNAm age for the unstandardized coefficient. An increase of one SD (5.34 years) in paternal age was linked to an increase of 0.15 SD (p = 0.027) in log₁₀ transformed mtDNA content, or for the unstandardized coefficient to an increase of 1.20% mtDNA content for an increase in one year of paternal age.

With this study on newborns, we demonstrate an absence of correlation between TL and epigenetic age consistent with findings in adults [31] (Fig. 1 and Additional file 1: Table S1). A cross-sectional study including 800 middle-aged persons found no significant correlation between blood TL and DNAm age (r = − 0.05, p = 0.17) [32]. Furthermore, in 773 participants (mean chronological age = 69.68) of the LipidCardio Study no significant association between TL and DNAm age (β = 3.00, p = 0.18) was demonstrated [27]. Besides, the dynamics of both factors have also been shown to change throughout life. Telomere attrition is higher in young children when environmental influences exert the most impact on inter-individual variation in telomere length, which is subsequently preserved throughout life [21]. Likewise, the DNA methylation-based biological clock "ticks" differently over the life course. The Horvath clock shows a non-linear rate of the clock ticking faster than chronological aging during childhood and adolescence and a linear association with chronological years during adulthood [53]. In older people, an increase in DNAm age occurs even at a slower rate than chronological age [54]. Findings concerning the correlation between blood TL and DNAm age in adult and elderly populations are, therefore, not automatically transferrable to early childhood. However, this is the period that is presumably the most sensitive to environmental influences and sets the base for later life [21]. Our findings are important as they indicate that cord blood TL and DNAm age not only relate to different pathways/mechanisms of biological aging in adults but also in neonates for the period of fetal programming. Although telomere length and the epigenetic clock in newborns indicate different aging measures, this does not mean that DNA methylation is not linked with TL. Distinct epigenetic signatures were identified, and epigenetic regulation of newborn TL was reported previously [55] involving CpGs distinct from the CpG-sets used to predict biological age by epigenetic age clocks. Moreover, variants in Telomerase Reverse Transcriptase (TERT) gene on chromosome 5, associated with Horvath DNAm age derived intrinsic epigenetic age acceleration, were also found to be associated with longer telomeres, indicating that hTERT expression is required for DNAm aging in human primary fibroblast [56].

The finding of a correlation between mtDNA content and TL in cord blood is in line with previous findings of a positive association in 613 cord blood samples of the ENVIRONAGE cohort, showing a 5.22% (95% CI: 3.26 to 7.22; p < 0.0001) higher mtDNA content for a 10% increase in TL [57]. Furthermore, also in healthy adults (r = 0.120, p < 0.001) [35], older women (r = 0.39, p < 0.0001) [28] and 166 non-smoking elderly (r = 0.23, p = 0.0047) [29] TL and mtDNA content were positively associated, conform with the mitochondrial-telomere axis of aging [23, 24, 57].

We also detected a correlation between global DNA methylation and three functional gene regions (promoter, gene body, intergenic) with DNAm age. In adults, no evidence for such an association was found in a previous pooled analysis of 479 individuals from the Australian Mammographic Density Twins and Sisters and 3354 individuals from the Melbourne Collaborative Cohort Study (r = 0.01, p > 0.19) [36]. This could be explained by the different dynamics of global methylation and DNAm age in newborns and adults. Global levels of DNA methylation increase over the first few years of life [58], remain relatively stable during adulthood, and then decrease beginning in late adulthood [59].

Another finding of our study is the inverse correlation between DNAm age and mtDNA content in cord blood. This finding is corroborated by a recent study in 812 older males of the Veteran Affairs Normative Aging Study, where the mtDNA copy number was negatively associated with cross-sectional, though not with prospective measures of DNAm age (p = 0.03 and p = 0.33 respectively) [30] (Fig. 1). In contrast, in a case–control study, DNAm age-derived AA and mtDNA content were not significantly correlated in the entire group of patients with bipolar disorder (BD), their siblings, and healthy age-matched controls (r = 0.038, p = 0.780), though positively correlated within the older (33-51 years) group of BD patients (r = 0.697, p < 0.001) [37] (Fig. 1). The different directions of correlation in our study of healthy newborns and the case–control study of BD patients may be partially explained by underlying differences in the biochemical mechanisms and etymology of BD.

The results of the path analysis confirmed previous findings in cord blood that at delivery, female sex is positively associated with telomere length [5, 21, 22, 60, 61] and global methylation [62]. In another study applying pyrosequencing of LINE-1 as a proxy of global methylation [63], an inverse association with the female sex was reported. Also, our finding of a negative association between female sex and cord blood mtDNA content is in line with previous findings. A study of the FLEHS III birth cohort in Flanders, Belgium, also found a trend for newborn girls having, on average, 5.96% less mtDNA content than boys (p = 0.08) [64].

Furthermore, delivery during the warm season was associated with shorter telomeres and less mtDNA content, which is in line with earlier findings of a longitudinal study including 2,827 residents of Costa Rica with a baseline age of 60 and older of longer telomeres in blood collected in October-December [65]. Moreover, a recent study in the ENVIRONAGE birth cohort found prenatal temperature exposure above a certain threshold associated with shorter cord blood TL [66]. We also observed a negative association between the warm season and mtDNA content, which is in contrast to findings in the same cohort on placental tissue mtDNA content, reporting a negative association between mtDNA content and the cold season (β = -0.243 ± 0.040, p < 0.0001) [48]. A possible explanation for the discordance in the mtDNA content of the two tissues may originate from their different biological function, which has been postulated previously [67].

Our finding of an inverse association between maternal educational level with DNAm age is in accordance with previous findings on the relation between cord-blood DNAm age and maternal socioeconomic status (SES), another measure of maternal educational attainment. [68]. Furthermore, the association between paternal age at birth and mtDNA content corroborates an earlier report in adults [69].

With regard to the causal assumptions in the path analysis, different possibilities concerning the direction of effect between the aging biomarkers are reasonable. On the one hand, telomere-dependent growth arrest is associated with increased mitochondrial dysfunction [70] through suppression of PGC-1α and PGC-1β promoters, impairing mitochondrial biogenesis and function [71]; on the other hand, mitochondrial dysfunction leads to telomere attrition and genomic instability via the increase in oxidative stress [70, 72]. Furthermore, global hypomethylation is linked with biological aging via the loss of constitutive heterochromatin integrity, a hallmark of aging in eukaryotes leading to global 5mC with increased genetic instability as a result [17, 73]. The generation of important co-substrates required for histone phosphorylation, acetylation and deacetylation processes, such as adenosine triphosphate (ATP), acetyl CoA, flavin adenine dinucleotide, and nicotinamide adenine dinucleotide depends again on mitochondrial activity [74]. In a study of the ENVIRONAGE cohort investigating the relationship between epigenome-wide methylation with cord blood insulin and mtDNA content, several pathways and differentially methylated regions (DMRs) also pointed in the direction of histone modification as one of the underlying mechanisms connecting these factors [75]. Dysfunctions in mitochondrial activity represented by alterations in mtDNA content may, therefore, have direct effects on global DNA methylation profiles [76, 77]. These molecular changes are not limited to the postnatal period, as environmental influences during pregnancy have been shown to already determine the genome methylation status and telomere length at birth [78, 79].

The significance of cord blood aging biomarkers lies in their predictive value for age-related NCD long before actual health effects become visible. We recently found that telomere length in childhood and early adulthood is highly determined by cord blood telomere length [21]. Therefore, observed health and disease conditions related to shorter telomeres, may find to some extent their onset at birth. Understanding the interrelationships and determinants of aging biomarkers is essential from a prevention point of view, as these insights enable the early recognition of individuals with increased risks and the development of personalized treatment plans.

Telomere length at birth determines the natural life span [80], and it has been estimated, that at an age of 40 years, the difference in life expectancy between people with an 1 SD shorter or longer TL than the population mean amounts to 2.5 years [81]. Studies directly investigating the predictive power of aging biomarkers at birth with life-span and age-related disease in adults and older people are lacking. However, their relationship with childhood health measures may act as a surrogate for their association with health outcomes later in life, as previous studies demonstrated a link between health outcomes in childhood and adulthood [82‐84].

Cord blood aging biomarkers have been linked with deviating health measures in infancy and childhood, showing associations with neurocognitive [85‐87], cardiovascular [88, 89] and metabolic outcomes [60, 75, 90‐93], pubertal onset [94, 95] and the immune system [96, 97] (Fig. 5). In this context we recently demonstrated that telomere length at birth was significantly related to childhood diastolic blood pressure at the age of four [98]. Despite the lack of studies in many fields of age-related health measures, the fact that metabolic outcomes are related to all four cord blood aging biomarkers (Fig. 5) stresses the importance of alterations in the childhood metabolism linking cord blood aging biomarkers with adverse health outcomes at old age. Deviations from childhood health measures can be predictive of adverse health outcomes in adulthood and old age for example lower birthweight or weight at the age of 1 year have been associated with later life cardiovascular disease [99‐101], diabetes type II [102], and frailty [103]. Furthermore, childhood BP was predictive of cardiovascular health in adulthood [104], pubertal timing was associated with multiple morbidities and lifespan in men [105], and infant eczema predicted adult asthma [106].

Concerning the determinants of aging biomarkers, identified in this study, their effect could translate to an altered disease susceptibility later in life. Children born in the warmer half of the year have on average shorter telomeres and lower mitochondrial DNA content, predisposing them to adverse infancy and childhood health measures such as lower birthweight, lower neurocognitive performance and higher blood pressure. On a population level, this could in turn result in a higher susceptibility to cardiovascular disease, diabetes type II, and frailty, therefore lowering life expectancy at old age.

Our study has several strengths and limitations. This is, to our knowledge, the first study to investigate the correlation between TL, DNAm age, global methylation, and mtDNA content in neonates, which is essential to disentangle their relationships in this crucial developmental stage. We also used data from a birth cohort that reflects the physiological ranges of the measured variables. On the other hand, the sample size of 190 newborns may have been too small to sustain significant results. Furthermore, the analysis of epigenetic age was confined to the algorithm published by Horvath [9]; other algorithms like the Hannum predictor [107] or various “gestational clocks” [108, 109] were not investigated as our focus was on the aging processes. We did, however, include gestational age in the path analysis. Regarding the Horvath clock, age acceleration is often studied besides age methylation Our study could not access a potential age acceleration as newborns have the same chronological age. Furthermore, path analysis made it necessary to hypothesize about the direction of effects between the aging biomarkers. Based on previous literature findings, evidence for both possible directions could often be found, making it necessary to choose one direction despite plausible reasons for another assumption.