Gestational phthalate levels and biomarkers of aging in infants and children from New York City

Abstract

Background:

Phthalates, a group of endocrine disrupting chemicals used in plastics, are ubiquitous in the environment and may influence cellular processes related to child health and development. Mitochondrial DNA copy number (mtDNAcn), telomere length (TL), and DNA methylation-based epigenetic gestational age acceleration (EGAA) are accessible biomarkers associated with biological aging and development. Little is known about how gestational phthalate exposure, individually and in mixtures, may impact biomarkers of biological aging through childhood.

Methods:

We used data and samples from the Columbia Children’s Center for Environmental Health (CCCEH) Mothers and Newborns Cohort, which recruited women in the third trimester from Northern Manhattan and the South Bronx, New York. Children returned for follow up at ages 5 and 7. We measured phthalate metabolites in third trimester maternal urine, relative mtDNAcn in cord blood (n=262) or childhood whole blood (n=340), relative leukocyte TL (rLTL) in cord blood (n=151) or childhood whole blood (n=155), and cord blood DNA methylation (n=188). Phthalates were assessed as metabolites or summed into parent compounds. Associations between phthalate exposure and biomarkers of aging were modeled with generalized linear models. Phthalate mixture effects were assessed using quantile g-computation.

Results:

Associations between phthalate metabolites and mtDNAcn and rLTL may reflect altered cellular health but were not robust to multiple comparison correction. At birth, a 10% increase in monobutyl phthalate was positively associated with mtDNAcn (β=1.96, 95% CI: 0.40, 3.52) while a 10% increase in mono-iso-butyl phthalate was negatively associated with mtDNAcn (β=−1.51, 95% CI: −2.90, −0.12). A 10% increase in mono-(3-carboxypropyl) phthalate was associated with a 1.06 percent increase in rLTL at birth (β=1.06, 95% CI: 0.17, 1.95). No significant associations were seen with phthalate exposure and EGAA, nor the phthalate mixture.

Conclusion:

We observed low evidence for the impact of phthalate metabolites on biomarkers of aging.

1. Introduction

Humans are exposed to a variety of chemical additives in consumer goods (1). Phthalates are a class of synthetic chemicals used primarily as plasticizers in an assortment of products, including food packaging, housing materials, and as scent stabilizers in personal care products and home goods (2). As such, phthalate exposure occurs through dermal absorption, ingestion of foods that have come into contact with plastics, and inhalation of household dust (2). The half-lives of phthalates in the human body are relatively short, but phthalate exposure is ubiquitous and difficult to avoid in the built environment resulting in pseudochronic exposure (3,4). Phthalates pose a concern to human health due to their similarity in chemical structure to endogenous hormones, which can mimic or block normal hormone signaling (3). Notably, phthalates can cross the placenta, leading to direct gestational exposure and subsequent impacts on fetal development and short- and long-term health (5,6). Previous studies have shown that gestational exposure to phthalates may be associated with lower birth weight, neurodevelopmental disorders, and allergic diseases (7-9).

Aging, or the biological processes that underly the loss of homeostatic mechanisms and lead to aging-related diseases, has become a recent topic of interest in the context of environmental exposures (10,11). While phenotypes of aging are generally observed in geriatric populations, studies of early life stress and adversity provide evidence that aging trajectories are determined early in life (12-16). Accelerated aging processes are hypothesized to mediate associations between early life adversity and poor physical and mental health outcomes in adolescence and adulthood, including greater risk for chronic diseases (15,17,18). Hallmarks of aging—which include DNA instability, telomere attrition, epigenetic changes, proteostasis loss, deregulated nutrient-sensing, mitochondrial dysfunction, cellular senescence, stem cell exhaustion, altered intercellular communication, disabled macroautophagy, chronic inflammation, and dysbiosis—may be identifiable through specific biomarkers as early as infancy (19). To date, several biomarkers have been identified as indicative of these underlying processes, each representing different constructs of aging.

One such biomarker is mitochondrial DNA copy number (mtDNAcn). Mitochondria are responsible for energy production in the cell and have their own circular genome. Insults to both mtDNA and nuclear DNA can adversely affect mtDNA integrity and lower mtDNAcn (20-22). Several studies have shown that mtDNAcn, which is reflective of mitochondrial dysfunction, declines with age and aging-related disease states in adults (23-30). Alternatively, increases in mtDNAcn may also be indicative of mitochondrial dysfunction, potentially to compensate for lower energy production (21). In infants and children, mtDNAcn is sensitive to environmental exposures experienced in utero, including manganese and tobacco smoke (31).

Telomere attrition, another hallmark of aging, can be determined by longitudinally measuring telomere length (TL). Telomeres are regions of repetitive DNA sequences located at chromosome ends. These regions shorten with each cell division until cellular senescence is triggered to prevent the accumulation of mutations (32). Telomerase—the enzyme that maintains TL— expression peaks during gestation and most inter-individual TL variability originates early in life (33,34). After birth and throughout childhood, telomere attrition increases and then slows (35). In animal models, early life TL is shown to be predictive of lifespan (36,37). In humans, TL measured in adulthood has been associated with several cardiovascular and neurodegenerative diseases related to aging (38,39). Emerging evidence supports that TL at birth may be influenced by exposure to environmental and social stressors in utero (40-44).

Finally, epigenetic clocks were developed to predict biological aging from widely-available DNA methylation microarray data (45,46). Though originally developed in adults, these have been extended to gestational age using DNA methylation measured in umbilical cord blood (47,48). By comparing the estimated biological/epigenetic gestational age (EGA) of infants to their chronological gestational age, we can estimate epigenetic gestational age acceleration (EGAA). Several studies have examined EGAA in association with maternal chemical exposures, maternal mental health, and pregnancy complications, although findings have been mixed (49-55). Further, the utility of EGAA to predict aging outcomes in adulthood remains underexplored, given the length of time required for such a study. However, epigenetic age acceleration (EAA) in adulthood, has been associated with neurodegenerative diseases, cancer, frailty, and other aging-related diseases in multiple geriatric cohorts, depending on the type of clock and the goal of the study (56-60).

Support for accelerated aging beginning early in life comes primarily from studies of exposure to early life adversity, but it is posited that environmental pollutants can act through similar biological pathways (13,61,62). Many biomarkers of aging, including mtDNAcn, TL, and EGAA, are sensitive to gestational exposure to environmental and social stressors (31,63). Additionally, biological aging processes are intertwined with hormone signaling, suggesting that phthalates and other endocrine disrupting compounds can influence aging (64,65). Evidence suggests that phthalate metabolites may be associated with aging biomarkers, though existing literature highlights that observed associations are specific to the timing of exposure and type of phthalate. One prior study found that prenatal exposure to specific phthalates was positively associated with relative leukocyte TL (rLTL) in newborns (43). A different study found that prenatal exposure to specific phthalates was negatively associated with male EAA in childhood (66). Literature on gestational phthalate exposure and mtDNAcn is limited, though one study examined exposure to bisphenol S, another endocrine disrupting chemical, and found it was positively associated with mtDNAcn in children (67). Alterations to these biomarkers have been associated with risk of future disease such as cancer, cardiovascular disease, and neurodegeneration (20,28,39,57,68). To date, no study has examined the associations of gestational phthalate exposure with this compendium of biomarkers of aging in early life. As this is a time of rapid development, there is a need to better understand variation in these early life biomarkers and their influence on lifelong aging trajectories (69). Here we explore potential associations of gestational phthalate exposure with EGAA at birth and relative mtDNAcn and rLTL in newborns and children. Given that the health impacts of phthalates and other endocrine disrupting chemicals are often sex-specific, we also investigate the sex-specific effects of phthalates on these biomarkers.

2. Methods

2.1. Study population

We leveraged existing participant information from the Columbia Center for Children’s Environmental Health (CCCEH) Mothers and Newborns Cohort, a longitudinal birth cohort focused on minority populations that are under-represented in research (70). Cohort characteristics have been previously described in detail (70). Briefly, 727 pregnant individuals that self-identified as African American or Dominican were recruited between 1998 and 2006 from Northern Manhattan and the South Bronx, New York, USA. Individuals were between 18-35 years of age at the time of enrollment, in their third trimester, and lived in Northern Manhattan/Bronx for at least one year prior to enrollment. Participants were excluded if they smoked during their current pregnancy, had their first prenatal visit after the 20th gestational week, were diagnosed with gestational diabetes, hypertension, or HIV, or reported illicit substance abuse. The Institutional Review Board of Columbia University approved study protocols. Before each visit, mothers were informed of all study procedures and provided written informed consent. At age seven, children began providing informed assent.

Of the 727 mother-child dyads, 10 participants had a cord blood cotinine level greater than 25 ng/mL, suggesting active smoking, and were excluded from the present analysis (71). Figure 1 presents a flowchart outlining the number of samples in each sub-analysis. Four hundred and thirty-five mothers provided a urinary sample during the third trimester of pregnancy from which phthalate metabolites could be measured. Given the overlap between all outcomes and phthalates was low (n=104 participants), we elected to use different study samples for each outcome to maximize the sample size in each analysis. Of the participants with phthalate data, DNA methylation was measured in 188 cord blood samples. mtDNAcn was measured in 262 cord blood samples and in 284 and 56 peripheral blood samples at ages five and seven, respectively. rLTL was measured in 151 cord blood samples and in 155 peripheral blood samples (128 at age five and 123 at age seven) and 96 participants had measurements at both childhood timepoints. Missingness of covariate data was low (≤3%), so participants with missing demographics data were retained and data was imputed with multiple imputations of chained equations (see Covariate Selection below).

Figure 1. Flowchart of participant inclusion.

Abbreviation: mitochondrial DNA copy number (mtDNAcn)

2.2. Phthalate metabolite measures

Spot urine samples were collected during the third trimester of pregnancy and sent to the Centers for Disease Control and Prevention (CDC) for analysis. Phthalate metabolites were measured across two batches, from 2008-2009 and 2010-2013. On-line solid-phase extraction (SPE) method, coupled with isotope dilution high-performance liquid chromatography/tandem mass spectrometry (HPLC/MS/MS) was used to measure phthalate metabolites as previously described (72). Each analytical run included calibration standards, reagent blanks and quality control samples (72,73). Intraclass correlation coefficients (ICCs) were previously calculated across batches and ranged from 0.27 to 0.77 after adjusting for specific gravity (74). The following phthalate metabolites were measured: monobenzyl phthalate (MBzP), the main metabolite of butylbenzyl phthalate (BBzP); mono-iso-butyl phthalate (MiBP), the main metabolite of di-isobutyl phthalate (DiBP); monoethyl phthalate (MEP), the metabolite of diethyl phthalate (DEP); monomethyl phthalate (MMP), the main metabolite of dimethyl phthalate (DMP); and several other high molecular weight phthalates; mono-(3-carboxypropyl) phthalate (MCPP) a metabolite of di-n-octyl phthalate; monobutyl phthalate (MBP) a primary metabolite of dibutyl phthalate (DBP); and the sum of four metabolites of DEHP [ΣDEHP; mono-2-ethylhexyl phthalate (MEHP), mono-2-ethyl-5-hydroxyhexyl phthalate (MEHHP), mono-2-ethyl-5-oxohexyl phthalate (MEOHP), and mono-2-ethyl-5-carboxpentyl phthalate (MECPP)]. MMP was only detected in 30% of the study sample and was excluded from further analysis. The sum of the metabolite concentrations of DEHP was used in place of individual metabolites due to the high correlation between the individual components, and was calculated by converting metabolite concentrations (in ng/mL) into their respective molar concentrations and summing these values (75). Metabolites of diisononyl phthalate (DiNP)—mono-carboxyoctyl phthalate (MCOP) and mono-carboxynonyl phthalate (MCNP)—were only measured in a subset of participants (n=163), and therefore ΣDiNP was excluded from our analyses. When phthalate metabolites concentrations were below the limit of detection ($\text{LOD}$), we assigned a value of $\text{LOD}∕\sqrt{2}$. To correct for urinary dilution, we adjusted for specific gravity ($\text{SG}$) that was measured using a handheld refractometer (Atago PAL 10-S; Atago U.S.A. Inc., Bellevue, WA) using the following formula:

$$E_{SG}=E_{O}x\frac{S{G}_{median}}{S{G}_o}$$

where ${\mathrm{E}}_{\mathrm{o}}$ is the observed unadjusted phthalate metabolite concentration of the sample, ${\text{SG}}_{\text{median}}$ is the study population median specific gravity, ${\text{SG}}_{\mathrm{o}}$ is the observed specific gravity of the sample, and ${\mathrm{E}}_{\text{SG}}$ is specific gravity-adjusted phthalate metabolite concentration of the sample (76).

2.3. Biomarker Measures

2.3.1. Blood Collection and DNA Isolation

Blood was collected from child participants by trained study staff at the age five and seven visits and from umbilical cord blood at birth. Following buffy coat separation, DNA was extracted from samples using a standard phenol-chloroform protocol and stored at −80°C until analysis.

2.3.2. Mitochondrial DNA Copy Number

mtDNAcn was measured in samples collected at birth and at age five or seven. Relative mtDNAcn was measured using a duplex qRT-PCR assay as described previously (63). Briefly, all samples and standards were analyzed in triplicate. Samples were distributed across plates balanced by several confounding factors, and all plates contained three no template controls and a standard curve composed of a pooled sample of all study samples at equal mass. Calibration to the standard curve revealed no plate effects. Both mtDNA and nDNA were quantified, and relative mtDNAcn was then calculated as the ratio of mtDNA quantity to nDNA quantity. mtDNA measures were log2-transformed for statistical analyses.

2.3.3. Relative Leukocyte Telomere Length

Telomere length was measured in samples collected at birth and at age five and/or seven. Telomere length was quantified as telomere repeat sequence copy number (T) relative to single gene (S: albumin) copy number. The ratio was measured using an adapted approach to the multiplex monochrome quantitative polymerase chain reaction (MMqPCR) protocol, described previously (77). For all cells sampled, rLTL is measured proportionally to the average TL across chromosomes. This protocol produces estimates that are highly correlated with Southern blot (77). A detailed description of this approach, including primer sequences, plating schemes, and serial dilutions is available elsewhere (35). To mitigate potential plate effects for cord blood samples, a normalizing factor was calculated by averaging the pooled sample across plates and then dividing the plate-specific pool by this average. Samples across all plates were then divided by the normalizing factor (78). rLTL measures were log2-transformed for statistical analyses.

2.3.4. DNA Methylation and Epigenetic Gestational Age Acceleration

DNA methylation was measured at birth in 176 cord blood samples using the Infinium HumanMethylation450 BeadChip (450K) array (485,577 CpG sites) and in 272 cord blood samples using the Illumina MethylationEPIC BeadChip (EPIC) array (866,895 CpG sites) (Illumina, Inc., San Diego, CA, USA). Samples were assessed for genomic DNA contamination using both quality control and sex checks implemented with the EWAStools R package (79). Twenty-six samples failed control checks and 13 samples showed sex mismatches and were removed from the study. We further tested sample contamination using the SNP outliers function in EWAStools and a cutoff SNP outlier score ≤ −4. We also examined 10 CpG sites shown to be indicative of maternal blood contamination in cord blood samples, as described by Morin and colleagues, and used a cutoff of four out of 10 CpG contamination threshold (80). Samples that had a SNP outlier test score of ≤ −3 but passed the CpG contamination test were included in further analysis (n=15). Samples that had a CpG contamination score of five out of 10 CpGs but passed the SNP outlier test were also included (n=8). Samples that failed both tests or failed one test at a more extreme threshold (SNP outlier score > −3 or CpG contamination score ≥ 6), were removed from further analysis (450k n=7, EPIC n=21, total n=28).

We considered two measures of EGA for our analyses: the Knight and Bohlin clocks, which use 148 and 96 CpG sites to predict GA, respectively (47,48). Of the participants with methylation measured via 450k, missingness ranged from 1-4 CpGs for the Bohlin clock estimate and 1 CpG for the Knight clock. Of the participants with EPIC data, missingness ranged from 1-15 CpGs for the Bohlin clock and all participants were missing between 1-5 CpGs for the Knight clock. Missing data at any of these CpG sites was imputed using the impute.knn function (81). Previous studies support that up to 30% imputed values do not statistically impact Horvath’s mAge clock estimation, and this approach has proven more robust than alternate imputation methods (82,83). To confirm that CpG missingness in samples would not confound our analysis, we tested associations of CpG missingness with chronological GA, sample storage time, and gestational urinary phthalate concentrations and observed no significant associations. Chronological GA in the cohort was primarily extracted for participants from medical records based on best obstetric practices (84). We did not identify statistically significant associations with CpG missingness (data not shown). Clock performance was evaluated using the average age acceleration defined as the arithmetic mean of the set of signed differences between DNA methylation GA and chronological GA, the median absolute difference between DNA methylation GA and chronological GA, and the spearman correlation coefficient between DNA methylation GA and chronological GA (45,85). GA predicted by the Bohlin clock performed better across all criteria for both array types and was used to calculate EGA. Clock performance differed between the 450K array (mean signed difference=0.14, median absolute difference=0.12, ρ=0.44) and the EPIC array (mean signed difference=−1.03, median absolute difference=−0.75, ρ=0.53). We therefore adjusted models for array as a batch covariate. EGAA was calculated as the residuals from a linear model of EGA regressed on chronological GA (86). EGAA residuals were used for analyses. Recent work has shown that epigenetic clock performance may differ by array type. To provide additional context for our results, we examined the Spearman correlations of beta values across the 450K and EPIC platforms for CpGs included in both clocks using the Cross-Array Comparison and Testing Interface (87). This tool reports correlations of beta values from different array platforms using data from 30 samples run on each array type. The mean Spearman correlation coefficients for CpGs in the Bohlin clock and Knight clock were 0.50 and 0.58, respectively.

2.4. Covariate selection

Cohort participants completed a maternal interview during their gestational visit and follow-up interviews at each subsequent visit during which extensive demographic information was collected. Confounders hypothesized to impact phthalate exposure (2,7-9) and biomarkers of interest (35,63,88,89) were selected for inclusion in models. Across all biomarkers, models were adjusted for maternal age (in years), maternal ethnicity (Dominican/African American), pre-pregnancy BMI (in kg/m²), parity (nulliparous/multiparous), receipt of public assistance (yes/no), infant sex (male/female), and other phthalate metabolite levels. Analyses of mtDNAcn and rLTL additionally adjusted for DNA storage time, as this was previously established as an important consideration in these analyses (63). EGAA analyses additionally adjusted for batch and cell type composition, which was estimated from cord blood DNA methylation using the Houseman method (90). We imputed missing covariate data based on the dyads included in the current analysis (N=398) for maternal BMI (N=11, 2.8%), receipt of public assistance (N=3, <1%), gestational age (N=7, 1.8%), birthweight (N=11, 2.8%), and previous birth (N=2, <1%) with logistic regression (binary variables) or predictive mean matching (continuous variables) using the Multivariate Imputations by Chained Equations (MICE) R package and 10 imputations (91). Models were run on each imputed dataset and the results were pooled using Rubin’s rules (92).

2.5. Statistical analyses

Spearman correlation coefficients were used to measure correlations between all phthalate metabolites (individual metabolites and the ΣDEHP metabolites) as well as the correlations between the biomarkers of interest.

We first considered potential non-linear associations of the phthalate metabolites and each outcome using generalized additive models. However, we did not observe evidence of non-linear relationships (data not shown), and therefore used generalized linear models (GLMs) with an identity link function to model the exposures and outcomes. To account for skewed distributions leading to influential outliers, we log2-transformed the phthalate metabolite concentrations prior to analyses. Effect estimates and confidence intervals for models with log-transformed exposures and outcomes (mtDNAcn and rLTL) were transformed to represent the relative percent change in relative biomarker for a relative 10 percent increase in phthalate concentration. This was done by calculating 1.1 to the power of the effect estimate or confidence interval bound, subtracting 1, and multiplying by 100.(93) Statistical significance was determined as a p-value less than 0.05, though following best practice we considered the direction, magnitude, and precision of the effect estimates rather than relying solely on p-values when interpreting results (94,95). P-values were corrected for multiple comparisons using a false discovery rate (FDR) with q < 0.05 considered significant. As phthalate exposure often occurs as a mixture, we tested whether including all five phthalate metabolites and ΣDEHP in the same model would result in multicollinearity with variance inflation factors (VIF). For each outcome, the VIFs when including all metabolites and ΣDEHP in the model were below four, and as such we chose to model all exposures in the same models, rather than separately (96). mtDNAcn in childhood was measured at either five or seven years of age, with no overlap. Therefore, we combined these measures into a single childhood mtDNAcn outcome in a GLM, per the original study design (63). The two-way mixed effects ICC did not suggest a strong relationship between mtDNAcn at birth and childhood (ICC=0.31) (97) so mtDNAcn measured in cord blood and at childhood were analyzed separately. rLTL was also measured at ages five and seven, with some overlap across samples. We therefore modelled childhood rLTL using a linear mixed effects model with a random effect for participant. The ICC for rLTL at birth and at ages five (ICC=0.01) and seven (ICC=0.11) was low, so cord blood rLTL was analyzed in a separate model.

To investigate if the association between gestational phthalate exposure and aging biomarkers was modified by child sex, we repeated main models with an interaction term specific for phthalate metabolite*sex. Across each biomarker and timepoint, the significance of the interaction term for phthalate metabolite*sex was used to judge the presence of effect modification by sex. We also assessed this relationship via sex-stratified models to extract strata specific estimates.

Quantile g-computation was used to estimate the association between the gestational phthalate mixture on the biomarkers of interest using the qgcomp R package (98). We estimated the differences in outcome per quartile of joint phthalate biomarker concentration as this provided estimates similar in direction to GLMs and did not overly divide the relatively small sample size. We estimated standard errors for these models using 200 non-parametric bootstrapped samples. Separate models were constructed for each biomarker at each timepoint with ΣDEHP, MiBP, MEP, MCPP, MBzP, and MBP as the mixture components, and adjusted for the same covariates as the GLMs. Significance of the overall mixture effect was assessed using a p-value of 0.05. A positive or negative scaled effect estimate is presented as each metabolite’s contribution to the observed overall mixture effect.

Additional Analyses

Previous research suggests that mtDNAcn and rLTL are impacted by cell type composition. As this data was only available in cord blood, we subset our sample to participants with DNA methylation measures and mtDNAcn (n = 184) or rLTL (n = 112) at birth to evaluate the effect of adjustment for cell-type composition on these biomarkers. Models identical to our main analyses were used with the addition of cell type composition estimated from cord blood DNA methylation using the Houseman method (90). We also assessed mtDNAcn and rLTL measured at ages five (mtDNAcn_age5 n=284, rLTL_age5 n=128) and seven (mtDNAcn_age7 n=56, rLTL_age7 n=123) separately. Additionally, to ensure comparability in sub-samples of the cohort, we subset our sample to only the participants with all biomarkers of aging and re-ran models.

To explore which individual chemicals might warrant further investigation for regulatory relevance we also assessed potential differences in the relationship between phthalates and biomarkers of aging when the five metabolites and ΣDEHP were modeled separately rather than in the same linear model. Following the structure of the main models, mtDNAcn and rLTL at birth as well as mtDNAcn at childhood were modeled with a GLM.

In addition to the quality control measures taken above testing associations of CpG missingness with several covariates, we stratified models by methylation array to evaluate if estimates were comparable between EGAA measured with 450k and EPIC assays.

3. Results

3.1. Descriptive statistics of the study population

Study participants were either Dominican (67%) or African American (33%) with an average age of 25.4 ± 4.8 years and an average BMI of 25.8 ± 6.0 kg/m² (Table 1). At birth, over 40% of mothers in our study population reported receipt of public assistance (42.5%). Slightly more than half of mothers (53.5%) had given birth previously. Average birth weight of the newborns was 3,380 ± 478 grams with an average gestational age of 39.3 ± 1.3 weeks. Gestational age ranged from 35-42 weeks; 14 infants were born before 37 weeks and 5 infants were born after 41 weeks gestation. There was a higher proportion of female infants (52.8%) compared to males (47.2%). The average age of children at the year 5 visit was 5.05 ± 0.20 years and the average age at the year 7 visit was 7.07 ± 0.21 years. For mtDNAcn analyses where the two timepoints were combined, the average age of participants was 6.03 ± 1.03 years. We did not observe any major differences in our population subset comparing those with mtDNAcn or rLTL measured at birth to those with the same biomarker in childhood (Supplementary Table 1).

Table 1.

Characteristics of mother-child pairs included in one or more analyses versus the full cohort

	Dyads included in current analyses (n=398)	Mothers and Newborns Cohort (n=717)a	P valueb
Maternal characteristics	N (%)/Mean (SD)	N (%)/Mean (SD)
Self-reported ethnicity
Dominican	266 (66.8%)	470 (65.6%)	0.47
African American	132 (33.2%)	247 (34.4%)
Public assistance (yes)	169 (42.5%)	295 (41.1%)	0.53
Previous birth (yes)	213 (53.5%)	393 (54.8%)	0.47
Age (years)	25.4 (4.76)	25.2 (4.94)	0.21
Pre-pregnancy BMI (kg/m2)	25.8 (6.03)	25.6 (6.03)	0.55
Child characteristics	N (%)/Mean (SD)b	N (%)/Mean (SD)c
Sex assigned at birth
Female	210 (52.8%)	372 (51.9%)	0.65
Male	188 (47.2%)	345 (48.1%)
Gestational age (weeks)	39.3 (1.27)	39.3 (1.39)	0.76
Birth weight (grams)	3380 (478)	3370 (478)	0.68
Age at year 5 visit (years)	5.02 (0.17)	5.05 (0.20)	<0.001
Age at year 7 visit (years)	7.06 (0.20)	7.07 (0.21)	0.13
Biomarkers	Median (25th, 75th percentile), n	Median (25th, 75th percentile), n
Birth
Relative mtDNA copy number	0.68 (0.02, 4.02), n=262	0.64 (0.02, 4.02), n=442	0.002
Relative telomere length	1.78 (0.45, 6.77), n=151	1.71 (0.45, 6.77), n=222	<0.001
Epigenetic gestational age acceleration	−0.15 (−2.20, 2.19), n=188	0.01 (−0.52, 0.57), n=342	0.03
Childhood
Relative mtDNA copy number (age 5 or 7)	1.21 (0.12, 2.80), n=340	1.23 (0.11, 2.81), n=509	0.173
Relative telomere length (age 5)	1.01 (0.20, 3.25), n=128	0.90 (0.17, 3.25), n=199	<0.001
Relative telomere length (age 7)	1.14 (0.49, 3.43), n=123	1.06 (0.24, 3.43), n=187	<0.001

Abbreviations: Body mass index (BMI); mitochondrial DNA (mtDNA)

^aMothers who had a cord blood cotinine level > 25 ng/mL are not included (n=10)

^bP-value calculated with t-test for continuous outcomes and chi-squared test for categorical outcomes

The distribution of urinary phthalate metabolite concentrations from mothers included in one or more analyses, adjusted for specific gravity, were similar in the different analysis subpopulations (Table 2). Phthalate metabolite concentrations were moderately correlated (ρ=0.36-0.74), with the strongest correlation detected between MiBP and MBP (Supplementary Figure 1a). Aging biomarkers were not correlated with one another, with the exception of rLTL measured at ages 5 and 7 (ρ=0.54) (Supplementary Figure 1b). The median values of each biomarker were generally similar across participants included in the analyses compared to the full study population (Table 1). Median rLTL values at all time points were slightly higher in the analysis subsets compared to the full study population. rLTL was higher at birth than in childhood, while mtDNAcn was lower at birth than in childhood. Median EGAA differed between the 450K array (−0.45) and EPIC array (0.52) compared to the combined EGAA (−0.15).

Table 2.

Distribution of urinary phthalate metabolite concentrations from birthing parent included in one or more analyses, adjusted for specific gravity.

	Geometric mean (GSD)
Phthalate metabolite (ug/L) or parent compound (nmol/L)	Birth			Childhood (5-7 years)
	mtDNAcn (n=262)	rLTL (n=151)	DNAm (n=188)	mtDNAcn (n=340)	rLTL (n=155)	All biomarkers (n=104)
ΣDEHPa	0.28 (3.29)	0.30 (3.28)	0.27 (3.29)	0.29 (3.23)	0.29 (3.39)	0.30 (3.35)
MEHHP	20.96 (3.67)	22.64 (3.69)	20.10 (3.69)	21.95 (3.57)	21.97 (3.80)	21.89 (3.82)
MECPP	37.49 (3.09)	40.63 (3.06	35.41 (3.07)	38.85 (3.07)	38.94 (3.15)	39.92 (3.04)
MEOHP	17.39 (3.59)	18.91 (3.59	16.49 (3.58)	18.28 (3.52)	18.25 (3.72)	18.34 (3.70)
MEHP	5.18 (3.91)	5.41 (3.91)	4.83 (3.78)	5.28 (3.75)	5.30 (4.02)	5.10 (3.91)
MBP	37.43 (2.95)	37.27 (3.1)	36.76 (2.92)	37.58 (2.87)	35.92 (3.11)	37.96 (3.02)
MBzP	14.12 (4.21)	12.78 (4.15	14.25 (4.21)	13.48 (3.79)	12.68 (4.08)	12.96 (4.27)
MCPP	1.85 (3.09)	1.93 (3.15)	1.80 (3.30)	2.08 (2.88)	1.90 (3.07)	1.91 (3.21)
MEP	163.98 (3.57)	160.30 (3.47)	172.63 (3.73)	161.28 (3.45)	156.48 (3.50)	176.29 (3.60)
MiBP	8.47 (2.92)	9.34 (2.87)	8.35 (2.85)	9.24 (2.84)	8.94 (2.87)	9.41 (2.71)

Abbreviations: mitochondrial DNA copy number (mtDNAcn); relative leukocyte telomere length (rLTL); DNA methylation (DNAm); monobenzyl phthalate (MBzP), mono-iso-butyl phthalate (MiBP), the main metabolites of butylbenzyl phthalate (BBzP) and di-isobutyl phthalate (DiBP), respectively; monoethyl phthalate (MEP), the metabolite of diethyl phthalate (DEP) and several other high molecular weight phthalates; mono-(3-carboxypropyl) phthalate (MCPP) a metabolite of di-n-octyl phthalate; monobutyl phthalate (MBP) a primary metabolite of dibutyl phthalate (DBP); and the sum of four metabolites of DEHP [ΣDEHP; mono-2-ethylhexyl phthalate (MEHP), mono-2-ethyl-5-hydroxyhexyl phthalate (MEHHP), mono-2-ethyl-5-oxohexyl phthalate (MEOHP), and mono-2-ethyl-5-carboxpentyl phthalate (MECPP)].

^aThe parent phthalate concentration of DEHP was calculated by converting concentrations (in ng/mL) into their respective molar concentrations and summing these values.

3.2. Associations of gestational phthalates with biomarkers of biological aging

Associations between phthalate metabolites and biomarkers of aging were mixed (Figure 2). After accounting for multiple comparisons, no significant associations were seen with phthalate exposure and biomarkers of aging (Supplementary Table 2). Across the analyses detailed below, we report nominal p-values. Effect estimates for mtDNAcn and rLTL can be interpreted as the relative percent change in relative biomarker for a relative 10% increase in phthalate levels. For EGAA, effect estimates can be interpreted as a one unit change per doubling in phthalate concentration. The interpretation of effect direction varies based on the biomarker of interest. Lower rLTL or higher EGAA signify accelerated aging. Both lower and higher mtDNAcn may indicate mitochondrial dysfunction.

Figure 2.

Linear model estimates for a) relative percent change in relative mitochondrial DNA copy number (mtDNAcn) and leukocyte telomere length (rLTL) per relative 10% higher concentration of phthalate b) one unit change in epigenetic gestational age acceleration (EGAA) per doubling of phthalate. All models adjusted for maternal age, maternal ethnicity, pre-pregnancy BMI, parity, receipt of public assistance, infant sex, and other phthalate concentrations. Analyses of mtDNAcn and rLTL additionally adjusted for DNA storage time. Epigenetic age analyses additionally adjusted for batch and cell type composition, which was estimated from cord blood DNA methylation using the Houseman method. Points represent coefficients and lines represent the 95% CI for the estimate.

3.2.1. Associations of gestational phthalates with mtDNAcn

At birth, a 10% increase in MiBP was associated with a 1.43% decrease in mtDNAcn (β=−1.43, 95% CI=−2.73, −0.12, Figure 2, Supplementary Table 2) and a 10% increase in MBP was associated with an 1.90% increase in mtDNAcn (β=1.90, 95% CI=0.39, 3.43, Figure 2, Supplementary Table 2). The direction of effect was consistent for these associations when mtDNAcn was measured in childhood, though they were not significant. In the single phthalate models, a 10% increase in MBP remained significantly associated with mtDNAcn at birth (β=0.91, 95% CI=0.12, 1.71, Supplemental Figure 2). Additionally, a 10% increase in MBzP was associated with a 0.72% increase in mtDNAcn (β=0.72, 95% CI=0.12,1.32, Supplemental Figure 2) at birth. In the childhood sex-stratified analysis, associations might be more pronounced in males, but due to large confidence intervals we cannot draw the conclusion that these estimates are different from one another (Supplemental Figure 3). Mixture models demonstrated no association with mtDNAcn at birth (Ψ=0.07, 95% CI = −0.01, 0.24) nor at childhood (Ψ=0.06, 95% CI= −0.05, 0.17), though individual metabolite weights in the mixture models were consistent with results from the linear models (Table 3). MiBP received the largest negative weight at birth (−0.76), which is in line with the negative association observed in the main linear analysis, in which all phthalate metabolites (individual metabolites and the ΣDEHP) were also included in the model. In the additional analysis adjusting for cord blood cell types, associations with cord blood mtDNAcn were attenuated, suggesting that the observed associations may be driven by differences in cell type composition (Supplementary Figure 4). Examining mtDNAcn at age five only slightly increased the effect estimate for MiBP and the association reached borderline significance (β=−1.11, 95% CI=−2.14, −0.01) compared to our main model (β=−0.82, 95% CI=−1.74, 0.11). The sample size of children with mtDNAcn measured at age seven only was too small (n=56) to interpret meaningful effect estimates (data not shown). When restricted to the subset of participants with all biomarkers, effect estimates were similar in direction and magnitude to the main models (Supplementary Figure 5).

Table 3.

Quantile q-computation model coefficients for the associations between the phthalate mixture and biomarkers of aging. Each metabolite’s contribution to the observed overall mixture effect is reported as the individual metabolite or parent compound’s weight and standard deviation (SD) to the corresponding positive and negative scaled effect estimates across 10 imputations.

	Mixture effect (95% CI)	Weight of phthalate metabolite or parent compound contributing to mixture effect (SD)
		DEHPa	MBP	MBzP	MCPP	MEP	MiBP
Birth
Relative mtDNA copy number	0.07 (−0.1, 0.24)	0.13 (0.01)	0.54 (0.01)	0.31 (0.01)	0.01 (0.02)	−0.24 (0)	−0.76 (0.02)
Relative telomere length	0.03 (−0.09, 0.16)	−0.57 (0.03)	−0.2 (0.02)	−0.22 (0.01)	0.87 (0.01)	0.13 (0.01)	−0.01 (0.01)
Epigenetic gestational age acceleration	0.03 (−0.08, 0.14)	0.53 (0.01)	0.3 (0.01)	0.17 (0.01)	−0.88 (0.02)	−0.1 (0.01)	−0.02 (0.03)
Childhood
Relative mtDNA copy number	0.06 (−0.05, 0.17)	−0.28 (0.01)	0.41 (0.03)	−0.38 (0.03)	0.48 (0.03)	0.11 (0)	−0.34 (0.03)
Relative telomere length	−0.01 (−0.1, 0.07)	−0.59 (0)	−0.32 (0.01)	0.07 (0.01)	−0.09 (0.01)	0.03 (0)	0.9 (0.01)

Abbreviations: mitochondrial DNA (mtDNA); monobenzyl phthalate (MBzP), mono-iso-butyl phthalate (MiBP), the main metabolites of butylbenzyl phthalate (BBzP) and di-isobutyl phthalate (DiBP), respectively; monoethyl phthalate (MEP), the metabolite of diethyl phthalate (DEP) and several other high molecular weight phthalates; mono-(3-carboxypropyl) phthalate (MCPP) a metabolite of di-n-octyl phthalate; monobutyl phthalate (MBP) a primary metabolite of dibutyl phthalate (DBP); and the sum of four metabolites of DEHP [ΣDEHP; mono-2-ethylhexyl phthalate (MEHP), mono-2-ethyl-5-hydroxyhexyl phthalate (MEHHP), mono-2-ethyl-5-oxohexyl phthalate (MEOHP), and mono-2-ethyl-5-carboxpentyl phthalate (MECPP)].

^aThe parent phthalate concentration of DEHP was calculated by converting concentrations (in ng/mL) into their respective molar concentrations and summing these values.

3.2.2. Associations of gestational phthalates with rLTL

At birth, a 10% increase in MCPP was associated with an 1.03% increase in rLTL (β=1.03, 95% CI=0.16, 1.90, (Figure 2, Supplementary Table 2)). In the single phthalate models, the directionality of the association between MCPP at birth with rLTL persisted but was not statistically significant (β=0.53, 95% CI=−0.08, 1.14, Supplementary Figure 2). We did not observe any statistically significant associations of gestational phthalate exposure with rLTL at birth or in childhood when stratified by sex (Supplementary Figure 3). The mixture model demonstrated no significant association with rLTL at birth (Ψ=0.03, 95% CI = −0.09, 0.16) or childhood (Ψ=−0.01, 95% CI= −0.10, 0.07). Again, individual metabolite weights were consistent with results from the linear models. MCPP received the largest positive weight in the qgcomp mixture model for rLTL at birth (0.87, Table 3). In the additional analysis including cell type, estimates of association remained consistent in magnitude and direction (Supplementary Figure 4). Assessment of rLTL in the subset of participants with all biomarkers did not meaningfully change the estimates from our main model (Supplementary Figure 5). Separate models at ages five and seven also did not differ from the main model estimates as none of these associations showed statistical significance (data not shown).

3.2.3. Associations of gestational phthalates with EGAA

No phthalate metabolite was significantly associated with EGAA in our main model, and estimates were imprecise (Figure 2). Findings from the single phthalate models also showed null results (Supplementary Figure 2). In the sex-stratified analysis, MBzP was associated with increased EGAA among females (β=0.10, 95% CI=−2.5x10⁻³, 0.20) but the confidence interval was quite large suggesting imprecision (Supplementary Figure 3). The mixture model did not demonstrate a significant association with EGAA at birth (Ψ=0.03, 95% CI= −0.08, 0.14) (Table 3). In the subset of participants with all biomarkers, EGAA estimates remained non-significant (Supplementary Figure 5). Models stratified by methylation platform show no significant differences between the associations observed with combined, 450k, and EPIC estimated EGAA (Supplementary Figure 6).

4. Discussion

Gestational exposure to phthalates has been previously associated with several adverse health outcomes through childhood (7-9). As such, our study sought investigate associations between maternal urinary phthalate metabolite concentrations in pregnancy and biomarkers of aging in early life, specifically mtDNAcn, rLTL, and EGAA. Our preliminary findings suggest that certain gestational phthalates may impact molecular biomarkers of aging at birth and early childhood. However, we did not observe consistent associations between any one phthalate with multiple biomarkers, suggesting that phthalate exposure may not have a major impact on biological aging in early life, or that these biomarkers represent different underlying molecular processes. We observed significant associations between gestational MBP and MiBP exposure with mtDNAcn at birth, but not childhood. Additionally, we found MCPP was associated with increased rLTL at birth. After multiple testing, no significant associations were observed.

Phthalates and other endocrine disrupting compounds may exert their effects on mtDNAcn by increasing oxidative stress and altering energy homeostasis (88). Due to limited repair mechanisms, mtDNAcn can vary in response to increased oxidative stress, with both high and low levels associated with adverse health outcomes (21,99). However, interpreting mtDNAcn requires caution, as its value as a biomarker indicative of mitochondrial function is contested and context dependent (100). Altered mtDNAcn has been associated with changes in ATP production, oxidative stress, lipid metabolism, and aging related disorders including neurodegeneration in human population studies (27). We found that gestational exposure to MBP was significantly associated with increased mtDNAcn and MiBP with decreased mtDNAcn at birth. Though not significant, the direction of these relationships persisted throughout childhood, and the contribution of these metabolites to the overall mixture effect was consistent in direction. MBP and MiBP are both primary metabolites of structurally similar low-molecular-weight phthalates (DBP and DiBP, respectively) used in personal care products, adhesives, and plastics. These compounds may exert opposite effects on mtDNAcn due to differences in parent compound structure and receptor binding affinities. DBP has a straight-chain butyl group while DiBP has a branched chain isobutyl group, resulting in different pharmacokinetics (101,102). Additionally, the metabolic pathways of the compounds differ; DBP is more estrogenic than DiBP (103). Animal models suggest DBP disrupts mitochondrial function and induces oxidative stress, which could lead to compensatory increases in mtDNAcn (104). MiBP may elicit a weaker or distinct mitochondrial response though is less well studied. Studies examining exposure to phthalates and mtDNAcn specifically are limited. One previous study demonstrated that exposure to phthalates, specifically MCNP, was positively associated with mtDNAcn in sperm (105). These findings highlight the need for further research on the potential impact of this endocrine disruptor on mitochondrial DNA dynamics.

Furthermore, mtDNAcn is known to differ based on cell type composition, specifically platelet count. Although complete blood count data were unavailable, we estimated cord blood cell type composition based on DNA methylation data. We could not assess childhood cell type composition, as DNA methylation was not measured at this time, but associations between phthalates and mtDNAcn at birth were attenuated when cell type composition was included in analyses. This suggests that cell type composition might be important in the study of phthalates on mtDNAcn. Future studies on mtDNAcn should include information on platelet count, which we were unable to estimate from DNA methylation data as platelets do not have nuclei. There is also little known on trajectories of mtDNAcn throughout childhood. Prior work in adults found that mtDNAcn remains stable or decreases through midlife with stronger declines after age 50 (106), though most studies have been conducted cross-sectionally. Less is known on mtDNA dynamics in childhood, especially with respect to the impact of prenatal factors on the measure. To our knowledge, one study investigated mtDNAcn across childhood and found that relative copy number was lowest at birth, increased into childhood, and decreased slightly in adolescence (63). While our study aimed to explore the influence of gestational phthalates on mtDNAcn, there is also a need to better characterize changes in mtDNAcn during key phases of growth and development to clearly interpret the measure and identify children at greater risk of disease.

We also found that gestational exposure to MCPP was associated with increased rLTL at birth, although no association was observed in childhood. Previous studies found no association between gestational MCPP exposure and rLTL at birth (43). In a Chinese population, exposure to MEP, MECPP, MEOHP, MEHHP, and MBP early in gestation was associated with shortened TL, and some associations were specific to girls (107). A study in the US found several phthalates were associated with an increase in TL in boys (108). The conflicting findings may result from differences in study populations and timing of exposure (109). Phthalates have short half-lives in the human body, and the use of phthalate containing products may vary throughout pregnancy based on a mother’s needs or preferences (110). The impact of gestational phthalate exposure on TL remains unclear, and warrants further research, specifically on windows of susceptibility.

Additionally, the biological mechanisms that impact the rate of rLTL attrition are not well understood. Evidence supports that TL at birth may be influenced by exposure to environmental and social stressors in utero (40-44). Further, in animal and human studies, early life TL was shown to be predictive of lifespan (36,37,111). Our study found that gestational MCPP exposure might be associated with longer rLTL at birth. One possible mechanism resulting in longer rLTL is that phthalate exposure may induce c-myc expression. Within the human telomerase complex, c-myc is a regulator gene that can activate telomerase reverse transcriptase and telomerase activity (112). Nevertheless, telomere dysfunction is associated with human pathologies (39). More research on phthalate exposure with TL would aid in understanding if current phthalate regulations are protective enough for sensitive populations such as children as well as help to identify populations at greater risk of disease.

In our study population, gestational phthalate exposure was not associated with EGAA. This is in contrast to a previous study that suggested an association. In a population of predominantly low-income Mexican families, MBzP was marginally associated with lower EGAA at birth, and a phthalate mixture was negatively associated with decreased EGAA at birth for male offspring (113). While we did not observe sex-specific effects in our study, likely because of the small sample size and null effect estimates, epigenetic changes independent of those captured by the Bohlin clock might exist. Additionally, we observed that the median EGAA differed significantly between samples measured on the 450K array in comparison to the EPIC array. Differences in epigenetic age estimation due to technical variability across array platforms has been previously observed (87). Studies of epigenetic aging using clocks developed for adult populations have mitigated these differences by performing principal component analysis on the CpGs in each clock, then conducting biological age prediction using the principal components, rather than CpG-level data (87,114). However, such principal components-based clocks have not been developed for clocks estimating gestational age. The discordance between array platforms in our analysis highlights a promising area for future larger studies investigating EGAA from methylation data across multiple array types. In our analysis, we adjusted for array type in our batch variable and additionally stratified models by array type, though this resulted in small sample sizes. Increased EGAA has been linked to other prenatal environmental exposures, like tobacco smoke and other endocrine disrupting chemicals, like BPA (89,115). The biological implication of accelerated EGAA in children is not well understood. Research in adult populations strongly suggest that EAA is related to biological aging and disease states (56-60). In contrast, the relationship between EGAA and aging in pediatric populations requires further investigation (113). The influence of gestational environmental exposures on EGAA in early life is not well characterized and serves as an important topic for future studies.

The differences in our findings, across phthalate metabolites and biomarkers, highlight that each phthalate metabolite may have different mechanisms, and act differently for each biomarker of interest. The phthalate metabolites associated with mtDNAcn and rLTL in this study, MBP, MiBP, and MCPP, exhibit endocrine disrupting properties. MBP can both inhibit and stimulate steroidogenesis by up or downregulating the expression of steroidogenic acute regulatory protein (StAR). StAR facilitates cholesterol transport into mitochondria during steroid hormone production (116,117). Also, some evidence supports that the parent compound of MiBP can activate the estrogen receptor alpha (118). MiBP has also been shown to interfere with sex hormone-binding globulin (SHBG) by binding to the protein and altering the bioavailability of sex hormones like androgens and estrogens (119). MCPP, as a large contributing part of a mixture, may also impact SHBG levels (120). Biological aging processes are intertwined with hormone signaling, suggesting that these phthalates may influence biological aging as measured with mtDNAcn and rLTL.

The variability in findings across previous studies and our work may result from differences in phthalate measurements and exposure composition. One previous study compared geometric mean phthalate concentrations across 10 pregnancy cohorts, as well as National Health and Nutrition Examination Survey (NHANES), a nationally representative survey (121). The geometric means of phthalate concentrations for the Mothers and Newborns Cohort were highest amongst all cohorts for MEP, MEHHP, and MBzP. MiBP was the only phthalate for which the Mothers and Newborns participants fell towards the middle of the compared values. This might be explained in part by the sociodemographic characteristics of the population, as use of phthalate-containing products have been proven to differ based on socioeconomic status (122-124) and if a person is of a particular marginalized race or ethnicity (125). Of the 10 pregnancy cohorts and NHANES compared in the previous study, the phthalate metabolite profile of women in the Mothers and Newborns cohort appeared most similar to that of women in PROGRESS and CANDLE, pregnancy cohorts in Mexico and Tennessee, respectively. For example, concentrations of MEP were highest in Mothers and Newborns (176.3 μg/L), compared to PROGRESS (144 μg/L) and CANDLE (126 μg/L). Similarly elevated concentrations in Mothers and Newborns were observed for MEHHP (21.9 μg/L vs. 19.7 μg/L in PROGRESS and 6.67 μg/L in CANDLE), MECPP (39.9 μg/L vs. 44.3 μg/L in PROGRESS and 13.1 μg/L in CANDLE), and MBzP (13.0 μg/L vs. 1.41 μg/L in PROGRESS and 1.51 μg/L in CANDLE). For MEHP, levels in Mothers and Newborns (5.10 μg/L) were higher than CANDLE (2.14 μg/L) but lower than PROGRESS (20.9 μg/L). Though measured with a different urinary dilution method, geometric means of phthalate metabolites in our study sample were higher across all chemicals than those measured in women of child-bearing age in NHANES (2011-2012) as well as NYU CHES, a cohort of pregnant women and their babies in New York City (2016-2018) (121). This may be due to the timing of phthalate measurement. In 2008, US Congress passed the Consumer Product Safety Improvement Act (126) that banned the sale of childcare products containing lead and certain phthalates. Due to this, the profile of common phthalates has changed in recent years and should be considered in comparability and generalizability of findings.

While we sought to perform a thorough analysis of gestational phthalate exposure and biomarkers of aging by including multiple biomarkers, we acknowledge these markers do not measure the same underlying processes. Altered mtDNAcn may indicate a biological response to excess reactive oxygen species production and reflect mtDNA damage and dysfunction, with prior studies linking it to changes in energy metabolism (88). TL is associated with the onset of cellular senescence, and ultimately longevity and age-related disease risk (35). EGAA is calculated using DNA methylation at birth, an epigenetic alteration that affects gene expression, and may capture trajectories of in utero growth and development (66). It is unclear if our conflicting results suggest that phthalates have a minimal impact on aging related biomarkers, or rather that these biomarkers are simply measuring different molecular processes and should be interpreted separately.

Our work is not without its limitations. Analytical power was limited due to a relatively small sample size, resulting in imprecise effect estimates with wide confidence intervals. To increase statistical power, we combined methylation data across two array platforms and may have introduced error in EGAA estimates. However, extensive quality control work was completed on the data, and while samples were missing CpGs across platforms and clocks, missingness was not associated with covariates of interest and should not bias the results. To maximize the sample size, we also included childhood samples for mtDNAcn and rLTL collected across two timepoints. As participants only had overlap at these timepoints for rLTL measures, data was modeled differently across aging biomarkers. Given that the analytical samples vary across timepoints and aging biomarkers, associations between specific phthalate metabolites and aging biomarkers at one age but not another require caution in interpretability.

Previous research has also shown that DNA storage time impacts measurement of mtDNAcn and rLTL, as DNA degrades at different rates (35,127). While we were able to estimate the storage time for all samples and adjust for this in our analyses, samples were collected chronologically so other temporal factors may have also impacted our findings. Also, phthalates were measured at one timepoint, across multiple batches. Though batch effects should be minimal, this data was not available. However, a prior analysis investigated the ICCs for phthalate metabolite levels from 48 women that provided urine samples biweekly over 6 to 8 weeks in late pregnancy. ICCs were assessed across 135 samples with two to four repeats per woman. ICCs were 0.77 for MBzP, 0.65 for MnBP, and 0.60 for MiBP and ranged from 0.27 to 0.42 for DEHP metabolites after adjusting for specific gravity (74). This suggests moderate reliability over time. Nonetheless, it is unclear whether this single measure adequately represents the individual's average exposure throughout pregnancy. Exposure to phthalates is typically ubiquitous given routine personal care product use resulting in pseudochronic exposures. However, phthalates have short half-lives in the human body and previous studies indicate that phthalate concentrations vary throughout pregnancy and throughout the day, suggesting that a single measurement reflects a narrow window of exposure (110,121,128). A limitation of this study is reliance on a single measurement of phthalate metabolites during pregnancy, which may have introduced exposure measurement error resulting in non-differential misclassification of the exposure. Additionally, phthalates postnatally and in early childhood might influence measures of biological aging throughout early life that we were unable to measure. It is also possible that the LOD drifted towards more sensitive measurements over time that could impact our associations. Lastly, while the cohort collected extensive information on demographics that could be leveraged in analyses, there may be unmeasured confounders in the study, specifically those related to unique sources of exposure to phthalates that we found were significantly associated with mtDNAcn and rLTL.

Nevertheless, our analysis has numerous strengths. This work was focused on Dominican and African American participants, traditionally underrepresented populations in research and at greater risk of chemical exposures (125,129-132). Also, while there was missing data in the collected demographic information, we were able to use multiple imputation to avoid loss of data and maximize the study’s power. Lastly, the measures of mtDNAcn and rLTL at birth and in childhood serve as a strength of this study, allowing us to examine associations at different developmental stages. These findings can inform further research on the impact of endocrine disrupting chemicals and other developmental toxicants on biological aging.

In this exploratory analysis, we aimed to investigate information on multiple biomarkers of aging to provide a more comprehensive understanding of the impact of gestational phthalate exposure on early life molecular biomarkers. Mostly null results were observed. We did not identify large consistencies across the impact of gestational phthalates on aging related biomarkers, but phthalate metabolites may exert their effect via different mechanisms and impact each outcome differently. Our findings do not provide strong evidence of an association between gestational exposure to phthalates and biomarkers of aging; however, they suggest the need for future research to better understand these relationships across larger, diverse populations during critical windows of development.