Environmental Phenols and Growth in Infancy: The Infant Feeding and Early Development Study

Abstract

Context

Higher mean and rapid increases in body mass index (BMI) during infancy are associated with subsequent obesity and may be influenced by exposure to endocrine-disrupting chemicals such as phenols.

Objective

In a prospective US-based cohort conducted 2010-2014, we investigated associations between environmental phenol exposures and BMI in 199 infants.

Methods

We measured 7 urinary phenols at ages 6-8 and 12 weeks and assessed BMI z-score at up to 12 study visits between birth and 36 weeks. We examined individual and joint associations of averaged early infancy phenols with level of BMI z-score using mean differences (β [95% CI]) and with BMI z-score trajectories using relative risk ratios (RR [95% CI]).

Results

Benzophenone-3, methyl and propyl paraben, and all phenols jointly were positively associated with higher mean BMI z-score (0.07 [−0.05, 0.18], 0.10 [−0.08, 0.27], 0.08 [−0.09, 0.25], 0.17 [−0.08, 0.43], respectively). Relative to a stable trajectory, benzophenone-3, 2,4-dichlorophenol, 2,5-dichlorophenol, and all phenols jointly were positively associated with risk of a rapid increase trajectory (1.46 [0.89, 2.39], 1.33 [0.88, 2.01], 1.66 [1.03, 2.68], 1.41 [0.71, 2.84], respectively).

Conclusion

Early phenol exposure was associated with a higher mean and rapid increase in BMI z-score across infancy, signaling potential long-term cardiometabolic consequences of exposure.

There are 2 related yet distinct aspects of infant growth that may be relevant for later cardiometabolic health: the level (mean amount, eg, high or low) as well as the trajectory (change over time, eg, increasing or decreasing) of body mass index (BMI) during infancy. Infant BMI correlates with fat mass and cardiometabolic health in childhood and adulthood (1), and higher infant BMI confers excess risk of developing obesity later in life (2). Furthermore, an infant growth trajectory characterized by rapid or excessive growth is associated with continued disordered growth and obesity development (1-5). These associations are observed regardless of the metric used for defining infant growth trajectories (ie, weight, length, weight for length, BMI, or fat mass), and have even been observed among infants born at term with an appropriate for gestational age birthweight (6). Perturbations in the early life environment may impact both the level and trajectory of BMI in infancy and, thereby, subsequent obesity and its cardiometabolic sequelae. Most research has focused on the infant nutritional environment (7), with environmental chemical exposures receiving less attention despite their ubiquity and potential to disrupt human development (8, 9).

Phenols are highly prevalent endocrine-disrupting chemicals commonly used in consumer and industrial products (10). Exposure to environmental phenols, most notably bisphenol A (BPA), has been associated with numerous adverse health effects, and phenols have been shown to influence epigenetic, physiologic, and biologic pathways (10-15). Observational studies in pregnant persons suggest inverse associations of urinary gestational dichlorophenol concentrations with fetal growth outcomes (10, 16), and observational studies in children suggest positive associations of urinary dichlorophenol and bisphenol concentrations with obesity and growth outcomes (17). Only 3 studies have examined phenol exposures during infancy (18-20) despite the importance of this timeframe for development and health.

Of those 3 studies, 2 measured concentrations of bisphenols in breastmilk and found that BPA was inversely associated with infant growth rate (18, 20). The third study measured BPA in neonatal dried blood spots and found that BPA was positively associated with rapid infant weight gain and early childhood obesity (19). No prior studies have examined other phenols such as dichlorophenols or parabens in relation to infant growth outcomes nor have studies assessed infant exposure through measures of phenol biomarkers directly in urine, which is the preferred matrix for phenol assessment when available (21, 22). Further, because phenol biomarkers demonstrate high temporal variability (23), assessing phenol exposure at multiple timepoints can reduce bias (24, 25).

To address this literature gap, we examined the association between urinary phenol concentrations and BMI z-score in infancy using data from the Infant Feeding and Early Development Study (IFED).

Materials and Methods

Study Sample and Design

The IFED study was a prospective cohort of mothers and their infants recruited at delivery between August 2010 and November 2013 from 8 regional hospitals around Philadelphia, PA. Details of the motivation for and design of IFED have been previously published (26). Briefly, mothers were eligible for participation if they spoke English, were at least 18 years of age, had no endocrine disorders, and needed no steroids, immunosuppressants, or hormones to maintain the pregnancy. Infants were eligible if they were healthy singletons born at 37 to 42 weeks’ gestation with a weight 2500 to 4500 g. Families chose and adhered to a planned infant feeding method (soy-milk formula [90-100% of calories from soy-milk formula], cows-milk formula [90-100% of calories from cow’s milk formula and ≤1% of calories from soy-milk formula], or breastmilk [90-100% of calories from breastmilk and ≤1% of calories from soy-milk formula]) as previously described (26). Thresholds for feeding regimen adherence were relaxed across the study to accommodate supplementation, but at least 80% of calories had to be from the planned infant feeding method to be retained in the study. Among those who completed the study, adherence to planned feeding method was high. Potentially eligible mother–infant pairs were identified from medical charts and approached around or before delivery to determine study eligibility and obtain written informed consent. Institutional review boards at the Children's Hospital of Philadelphia, Virtua Hospitals, Abington Memorial Hospital, the National Institute of Environmental Health Sciences, and Copernicus Group approved the study protocol. The analysis of deidentified samples at the Centers for Disease Control and Prevention (CDC) laboratory was determined not to constitute engagement in human subjects research.

Birth visits were completed within 72 hours of delivery. Infants were followed across infancy until the last study visit at 28 weeks (±30 days) for boys and 36 weeks (±30 days) for girls. The parent study followed girls longer than boys due to prior work showing a longer minipuberty period in infant girls. Other study visits were scheduled at 2 and 4 weeks (±4 days); 6, 8, and 12 weeks (±10 days); 16, 20, 24, and 28 weeks (±14 days); and, for girls, 32 weeks (±14 days). In June 2012, the study visit at 6 weeks was discontinued. During each study visit, mothers completed questionnaires, and study staff obtained biospecimens and anthropometric measures from infants.

Because soy-milk formula was hypothesized to have endocrine-disrupting effects, exposure to environmental phenols was only assessed in the breastmilk and cow-milk formula–fed infants.

Environmental Phenol Assessment

Seven environmental phenols were quantified in urine samples collected at the 6-8 week (6 or 8 week visit) and 12 week visits. Urine samples were collected in Tushies diapers, pediatric urine bags (Hollister U-Bag, Libertyville, IL; BPA free), directly into a polypropylene cup during study visits, or some combination of the above (27, 28). Phenol concentrations did not significantly differ by collection type. To collect in pediatric urine bags, standard hospital procedures were used to cleanse the infant, and pediatric urine bags were applied and left on for the duration of the study visit. Urine was then transferred into a clean, sterile, polypropylene urine cup and stored in a study site refrigerator at 2 to 8°C before being processed. Processing occurred within 6 hours of collection, and staff created as many aliquots as possible using 5-mL polypropylene cryovials (Nalgene). Specimens were then frozen, and bulk shipped overnight on dry ice in an insulated shipping box monthly for storage at the National Institute of Environmental Health Sciences at −80 °C. Urine creatinine concentrations were measured using an Olympus AU400e analyzer (Beckman Coulter Inc. Irving, TX).

For phenol measurements, urine specimens were shipped overnight on dry ice to the CDC laboratory. Solid phase extraction coupled to high-performance liquid chromatography tandem mass spectrometry (27-29) was used to determine concentrations of BPA, benzophenone-3, 2,4-dichlorophenol, 2,5-dichlorophenol, methyl paraben, propyl paraben, and triclosan. The limits of detection (LODs) were 0.1 ng/mL (BPA, 2,4- and 2,5-dichlorophenol, propyl paraben), 0.2 ng/mL (benzophenone-3), and 1.0 ng/mL (methyl paraben, triclosan). Details of quality assurance and assessment procedures are provided elsewhere (Supplementary 1 (30)).

Growth Assessment

Infant weight and recumbent length were assessed by trained study staff at every study visit. Details of procedures including quality assurance and assessment are described elsewhere (Supplementary 2 (30)). Sex- and age-specific z-scores for weight and BMI (weight [kg]/length [m]²) were calculated from the World Health Organizations Growth Standards (31). Infant BMI z-score was our primary study outcome due to its association with infant fat mass and cardiometabolic health (1-5), and its ability to estimate adiposity in infants (32, 33). All infants had at least 4 BMI measures across infancy, with a median of 10 measures (25th and 75th percentile: 9, 12) (Fig. 1).

Figure 1.

Infant BMI z-score by age at each study visit among the infant feeding and early development study sample (N = 199), 2010-2014. In girls and boys, study visits were scheduled at 2 and 4 weeks (±4 days); 6, 8, and 12 weeks (±10 days); 16, 20, 24, and 28 weeks (±14 days). In girls, 2 additional study visits were scheduled at 32 weeks (±14 days) and 36 weeks (±30 days). In June 2012, the study visit at 6 weeks was discontinued. Dashed y-axis reference lines were placed at a BMI z-score of −1.96, 0, and 1.96.

Covariates

Covariates were selected a priori and included infant sex assigned at birth (female, male), infant diet (breastmilk, cow’s milk), infant race (Black, White, All other persons), maternal education (high school or less, some college or more), maternal age (20 or less, 21-25, 26-30, and 31 or above years), smoking during pregnancy (yes, no), alcohol during pregnancy (yes, no), and gestational age in weeks at birth (37, 38, 39, 40, 41, 42). These covariates were collected via maternal report and medical record. We selected covariates that were confounders based on observed associations with exposure and outcome status in IFED infants or the literature (5). Infant sex assigned at birth was included as a precision variable.

Infant race was included in models as a proxy for unmeasured factors which may influence exposures (eg, marketing of consumer products) (34) and outcomes (eg, racism-related stress of caretakers, health care quality) (35). Mothers reported infant race and selected all that applied from the following categories: American Indian or Alaskan Native, Asian, Black or African American, Native Hawaiian or Other Pacific Islander, White, or Unknown or Not reported. We aggregated race as Black, White, or All other persons, with the last including groups with small sample sizes (Asian and American Indian or Alaskan Native), Multiracial individuals, and 1 infant with race not reported.

Descriptive Analyses and Data Curation

We performed statistical analyses using SAS 9.4 (Cary, NC) and R version 4.0.4 (Vienna, Austria). We examined distributions and descriptive statistics for covariates and outcomes using means ± SD and n (%).

One sample had creatinine below the lower limit of quantification (LLOQ; 1.0 mg/dL). With the exception of triclosan, detected in 64% across all samples, phenols were detected in >90% of samples. Missing covariates (n = 1 missing smoking during pregnancy), and creatinine and phenol concentrations below the LOD or LLOQ were imputed (36, 37). We generated 10 imputed datasets using 20 chained iterations per dataset including all exposures and covariates from our main models as well as sample collection characteristics (infant age in weeks, timing [Am or Pm], season [winter, spring, summer, fall], and batch). Creatinine and phenol concentrations were imputed from a left censored log-normal regression model with the maximum imputed value set to the LLOQ or LOD (38).

After imputation, the O’Brien method of standardization (39) was used to correct phenols for urinary dilution. Models were fit to predict creatinine from infant sex, infant diet, infant race, maternal education, maternal age, smoking during pregnancy, alcohol during pregnancy, gestational age at birth, and sample collection characteristics (infant age in weeks, timing, season, and batch).

Urinary phenols have short biological half-lives and exposures tend to be episodic in nature, contributing to within-subject concentration variability (24). To obtain a more stable estimate of a participant's exposure, we averaged up to 2 repeated measures of each creatinine-standardized phenol at 6 to 8 and 12 weeks using the geometric mean (n = 46 contributed 1 sample and 153 contributed 2 samples). We regarded this average as an estimate of early infancy exposure to environmental phenols.

We used the median and quartiles (25th and 75th percentile) to examine distributions and describe creatinine-standardized phenol concentrations. We used Spearman's rank correlations to examine associations between averaged creatinine-standardized phenol concentrations and visualized them using a heatmap. Average creatinine-standardized phenol concentrations were then log-transformed for analyses.

With the exception of tests for interaction terms, we focus on the direction and magnitude of observed associations rather than P values based on recommendations from the American Statistical Association (40) and the guidelines for Strengthening Reporting of Observational Studies in Epidemiology (41).

Determining Associations With Level of Infant Growth

We used linear mixed effects models (42) to examine individual associations between phenols and level of BMI z-score. These models included a cubic b-spline for age with 4 knots at 3, 8, 16, and 24 weeks as these split the age distribution into nearly equal numbers of observations. We used random intercepts to accommodate repeated measures. From these models, we estimated the difference in mean BMI z-score across infancy (β [95% CI]) for an interquartile range (IQR) increase in phenol concentration. We tested whether an interaction term between exposure and splined age was significant (P ≤ .01); P values of interaction terms from these models were all above .45, suggesting associations did not change over infancy, so we proceeded with models without interaction terms. We ran separate models for each phenol.

To examine joint associations for multiple chemicals, we ran quantile g-computation with parametric generalized linear regression models and participant-specific random effects (36). These models estimated the mean BMI z-score difference (β and 95% CI) for a simultaneous IQR increase in all phenol concentrations (ie, the phenol mixture) and included the same spline terms for infant age as our single-chemical models. Analyses were run both with and without covariate adjustment; adjusted models were the primary analysis.

Determining Associations With Heterogeneous Trajectories of Infant Growth

To identify BMI z-score trajectories, we first centered each infant's BMI z-scores by subtracting subject-specific means based on recommendations from Heggeseth et al (43). We then fit a Gaussian growth mixture model to these centered measures to identify groups of infants whose infant BMI z-score trajectories have different shapes. Gaussian growth mixture models are an extension of linear mixed models that assign individuals in the population to a finite number of groups, each characterized by a specific trajectory. Models included a cubic b-spline for age. For a range of knot numbers between 3 and 6, we chose knot locations to split the age distribution into nearly equal numbers of observations; 4 knots minimized the sum of squared errors. We used random intercepts to accommodate repeated measures. We determined the optimal number of trajectory groups based on Akaike information criterion, Bayesian information criteria, entropy, and visual inspection of the plotted trajectories (44). To ensure adequate sample size, groups were required to include at least 8% of infants.

We examined individual associations between phenol concentrations and BMI z-score trajectory with multinomial logistic regression. These models estimated relative risk ratios (RR [95% CI) for an IQR increase in phenol concentrations in a given trajectory compared to a common reference trajectory. We ran separate models for each phenol.

To examine joint associations for multiple chemicals, we ran quantile g-computation with generalized log-binomial regression (36). These models estimated the marginal RR (95% CI) of belonging to a given trajectory for a simultaneous IQR increase in all phenol concentrations (ie, the phenol mixture). We ran separate models comparing each trajectory of interest to a common reference trajectory.

To account for uncertainty in the assignment of participants to BMI z-score trajectories, we created a weighted pseudo-cohort for these analyses. Each participant had records equal to the number of trajectories and a corresponding weight proportional to their probability of being assigned to that trajectory. We ran the multinomial logistic and quantile g-computation models with weighting on this pseudo-cohort (45). Analyses with outcomes were run both with and without covariate adjustment; adjusted models were the primary analysis.

Secondary and Sensitivity Analyses

We had 2 secondary analyses conducted to assess the impact of features of the IFED study design (differential follow-up by infant sex and grouping by infant feeding method) on associations as well as to provide comparisons with prior literature examining infant breastmilk exposures (18, 20) and sexually dimorphic associations (8, 46) These models estimated sex- and diet-specific associations through methods described elsewhere (Supplementary 3 (30)). We additionally carried out several sensitivity analyses to assess robustness of our results as described elsewhere (Supplementary 4 (30)).

Results

Descriptive Analyses

Among the 410 infants enrolled in the IFED Study at delivery, urine samples were available from 199 infants fed breastmilk or cow’s milk and were analyzed for phenol concentrations. This sample was predominantly (61%) Black and 48% of mothers reported having at most a high school education (Table 1). Infants contributed 186 samples at 6 to 8 weeks and 166 samples at 12 weeks, corresponding to 33 infants with samples at 6 to 8 weeks only, 13 with samples at 12 weeks only, and 153 with samples at both timepoints (Table 2). Phenol concentrations were low to moderately correlated (Fig. S2 (30)).

Table 1.

Descriptive characteristics in the infant feeding and early development study sample (N = 199), 2010-2014

Characteristic	n (%)
Infant sex assigned at birth
Female	98 (49)
Male	101 (51)
Infant diet
Breastmilk	90 (45)
Cow’s milk	109 (55)
Infant race
Black	122 (61)
White	49 (25)
All other personsa	28 (14)
Maternal education
High school or less	95 (48)
Some college or more	104 (52)
Maternal age
20 or less	34 (17)
21-25	58 (29)
26-30	58 (29)
31+	49 (25)
Smoking during pregnancy
Yes	54 (27)
No	144 (73)
Alcohol during pregnancy
Yes	62 (31)
No	137 (69)
Gestational age at delivery (weeks)
37	16 (8)
38	41 (21)
39	58 (29)
40	52 (26)
41	32 (16)

Missing data include 1 infant missing smoking during pregnancy.

^a All other persons category includes groups with small sample sizes and those reported as belonging to multiple groups including American Indian or Alaskan Native (1); Asian (2); American Indian or Alaskan Native, Black or African American (4); Asian, Black or African American (3); Asian, White (3); Black or African American, Native Hawaiian or Other Pacific Islander (2); Black or African American, White (7); American Indian or Alaskan Native, Black or African American, White (4); Asian, Black or African American, White (1); Unknown or Not reported (1).

Table 2.

Distribution of creatinine-standardized phenol concentrations (ng/mL) in the infant feeding and early development study sample (N = 199), 2010-2014

Phenol	LOD (ng/mL)	% >LOD	Median concentration (25th percentile, 75th percentile)
			6-8 Weeks (N = 186)	12 Weeks (N = 166)	Early infancya (N = 199)
BPA	0.1	96	0.7 (0.4, 1.4)	0.9 (0.5, 1.7)	0.9 (0.5, 1.5)
Benzophenone-3	0.2	98	3.9 (1.7, 12.8)	5.5 (2.3, 13.9)	4.8 (2.0, 12.1)
2,4-Dichlorophenol	0.1	95	0.4 (0.2, 0.7)	0.5 (0.2, 1.1)	0.5 (0.3, 0.8)
2,5-Dichlorophenol	0.1	91	1.3 (0.5, 3.9)	2.2 (0.7, 6.5)	1.6 (0.6, 4.6)
Methyl paraben	1.0	99	397.8 (101.2, 905.3)	301.1 (65.4, 1146.4)	321.9 (102.3, 952.2)
Propyl paraben	0.1	98	9.4 (2.8, 74.2)	9.0 (2.6, 60.1)	11.0 (3.7, 46.1)
Triclosan	1.0	64	4.4 (2.0, 11.6)	5.4 (2.3, 14.2)	5.0 (2.3, 11.2)

Phenol assessment included 352 urine samples from 199 infants at 6-12 weeks of age; 33 infants had samples at 6 to 8 weeks only, 13 at 12 weeks only, and 153 had samples at both timepoints. The percent of concentrations above LOD was calculated for all samples. Creatinine and phenol concentrations below LOD were imputed from a left censored log-normal regression model with the maximum imputed value set to the LOD. Phenols were then creatinine-standardized using the O’Brien method with creatinine predicted from infant sex, infant diet, infant race, maternal education, maternal age, smoking during pregnancy, alcohol during pregnancy, gestational age at birth, age in weeks of collection, timing of collection, season of collection, and batch.

Abbreviations: BPA, bisphenol A; LOD, limit of detection.

^a Early infancy exposure concentrations are the median (25th percentile, 75th percentile) for the study sample after calculating the subject-specific geometric mean of early infancy creatinine-standardized phenol concentrations from repeated visits at 6-8 and 12 weeks. For 46 infants, this geometric mean includes samples at 1 timepoint. For 153 infants, this geometric mean includes samples at both timepoints.

Associations With Level of Infant Growth

In unadjusted and adjusted models, most phenol concentrations were positively associated with mean BMI z-score, though effect estimates were small and imprecise (Table S1 (30)). After covariate adjustment, an IQR increase in benzophenone-3, methyl paraben, and propyl paraben concentrations were associated with mean BMI z-score differences of 0.07 (95% CI −0.05, 0.18), 0.10 (95% CI −0.08, 0.27), and 0.08 (95% CI −0.09, 0.25), respectively (Fig. 2). An IQR increase in the phenol mixture was associated with a mean BMI z-score difference of 0.17 (95% CI −0.08, 0.43).

Figure 2.

Associations between level of BMI z-score and phenol concentrations in the infant feeding and early development study sample (N = 199), 2010-2014. Abbreviations: BPA, bisphenol A. Dashed x-axis reference lines represent the null. We estimated single chemical individual associations with linear mixed effects regression and multichemical (Mixture) joint associations with parametric generalized linear regression and participant-specific random effects in quantile g-computation. Model covariates included infant sex assigned at birth, infant diet, infant race, maternal education, maternal age, smoking during pregnancy, alcohol use during pregnancy, and gestational age at birth. Phenol concentrations were creatinine-standardized, averaged using the geometric mean from up to 2 repeated measures, and log-transformed for analyses.

Results of secondary and sensitivity analyses for associations with the level of infant growth are presented elsewhere (Tables S2-S4 (30)). Secondary analyses suggested phenols were usually positively associated with mean BMI z-score in female but not male infants (Table S2) though CIs were wide. Associations were generally consistent between diet strata (Table S3). Compared with our primary analyses, associations were attenuated for sensitivity analyses investigating weight-for-age z-score as an outcome as well as noncreatinine-standardized phenol concentrations as an exposure (Table S4). The magnitude of mean BMI z-score differences was larger in sensitivity analyses limiting to study visits at or after 12 weeks. Analyses for exposures at 6 to 8 weeks and 12 weeks suggested that neither time period was associated with greater risk than the other.

Associations With Heterogeneous Trajectories of Infant Growth

Four trajectories of BMI z-score were identified (Fig. 3). The largest trajectory group (n = 82) was labeled “stable” and exhibited little change in BMI z-score across infancy with centered BMI z-score staying around 0. We identified 2 trajectory groups with BMI z-score increasing across infancy, one which we labeled “rapid increase” containing 47 infants with BMI z-score increases in early infancy and the other we labeled “delayed increase” containing 27 infants with BMI z-score increases in late infancy. Notably, infants in the rapid increase group had the lowest mean BMI z-score at birth (−0.57 [SD 0.82]) and the highest mean BMI z-score at the end of follow-up (1.37 [SD 0.85]). Finally, the fourth trajectory group contained 43 infants and was labeled “decrease” as infants exhibited BMI z-score decreases across infancy. Based on model fit statistics, these 4 trajectories were consistently identified as the optimal descriptors for infant growth across almost all analyses (Tables S5-S7, Figs. S2-S5 (30)). The stable trajectory was the reference category in analyses investigating associations with exposures because it had the largest sample size. Given its association with later cardiometabolic outcomes (1-5), our primary comparison of interest was rapid increase relative to stable trajectory.

Figure 3.

BMI Z-score trajectory groups in the infant feeding and early development study sample (N = 199), 2010-2014. Before running growth mixture models to identify trajectory groups, BMI z-scores were centered by subtracting participant-specific mean BMI z-scores. (A) Plot of observed and predicted centered BMI z-score trajectory groups. Observed values indicated with shapes and colors according to group with highest probability of assignment. Predicted group trajectory and 95% CIs indicated with lines and colors according to group with highest probability of assignment. (B) Spaghetti plots of observed BMI z-scores for each assigned trajectory group. Trajectories are labeled and colored according to trajectory group with the highest probability of assignment. Dashed y-axis reference lines were placed at a BMI z-score of −1.96, 0, and 1.96.

Almost all phenols were positively associated with the rapid increase trajectory in unadjusted and adjusted models (Table S8 (30)). After covariate adjustment, the risk of rapid increase relative to stable trajectory for an IQR increase in phenol concentrations was 1.46 (95% CI 0.89, 2.39) for benzophenone-3, 1.33 (95% CI 0.88, 2.01) for 2,4-dichlorophenol, and 1.66 (95% CI 1.03, 2.68) for 2,5-dichlorophenol (Fig. 4). An IQR increase in the phenol mixture was associated with a 1.41-fold (95% CI 0.71, 2.84) increased risk of Rapid Increase relative to stable trajectory. There was evidence of inverse associations between some phenols and a delayed increase relative to stable trajectory. For instance, BPA, and 2,4- and 2,5-dichlorophenol were associated with lower risk of delayed increase relative to stable trajectory (0.64 [95% CI 0.40, 1.01], 0.81 [95% CI 0.62, 1.05], and 0.81 [95% CI 0.57, 1.13], respectively).

Figure 4.

Associations between centered BMI z-score trajectory groups and phenol concentrations and in the infant feeding and early development study sample (N = 199), 2010-2014. Abbreviations: BPA, bisphenol A. The reference group for analyses was the “stable” trajectory. Dashed x-axis reference lines represent the null. We estimated single chemical individual associations with multinomial logistic regression and multichemical (mixture) joint associations with log-binomial regression in quantile g-computation. Models adjusted for infant sex assigned at birth, infant diet, infant race, maternal education, maternal age, smoking during pregnancy, alcohol use during pregnancy, and gestational age at birth. Phenol concentrations were creatinine-standardized, averaged using the geometric mean from up to 2 repeated measures, and log-transformed for analyses.

Results of secondary and sensitivity analyses for associations with trajectories of infant growth are shown elsewhere (Tables S9-S14 (30)). The magnitude of positive associations between phenols and rapid increase relative to stable trajectory appeared stronger in female than male infants (Table S9). In diet-stratified secondary analyses, there was a suggestion of higher risk of delayed increase relative to stable trajectory among infants fed a cow’s-milk diet; other associations were generally consistent with our primary analyses (Table S10). Associations were generally consistent with our primary analysis in sensitivity analyses for BMI z-score (Table S11) and centered weight for age z-score (Table S12). Analyses for exposure at 6 to 8 weeks and 12 weeks did not suggest one exposure timeframe conferred greater risk as associations were similar to one another and to our main analysis (Table S13). Sensitivity analyses for noncreatinine adjusted phenol concentrations exhibited attenuated estimates compared to primary analyses (Table S14).

Discussion

Due to their widespread use, exposure to environmental phenols is ubiquitous in the US population including infants. Potential sources of infant exposure may range from indirect household and caregiver exposures (eg, breastmilk transfer of caregiver exposures) to direct exposures from the use of products targeted towards infants (eg, baby oils, lotions, wipes) (47-50). Endocrine-disrupting chemicals such as phenols can disturb hormonal functions and perturb normal health and development (12, 13). Infants likely experience heightened susceptibility to endocrine-disrupting chemicals, even at low concentrations (15). Relative to adults, infant metabolic pathways are less mature, infant livers are less efficient at inactivating phenols via conjugation, and smaller infant sizes lead to higher body burdens and doses of chemical exposures (51). Though the exact mechanisms by which endocrine disrupting chemicals influence health outcomes is not known, experimental studies have shown that some phenols can impact numerous pathways that play key roles in human growth and cardiometabolic development including adipogenesis, thyroid hormones, the hypothalamic–pituitary axis, pancreatic β-cell function, and inflammatory pathways (10, 12-15, 52). The impact of endocrine-disrupting chemicals on growth outcomes may be sexually dimorphic given sex-based differences in hormones and growth (8), but further studies are needed to better elucidate what, if any, sex-specific associations between phenols and growth exist. Due to concerns about their endocrine-disrupting properties and impacts on health, regulations banning the use of some phenols in products have been passed in the European Union (eg, 2014 ban on parabens from use in products applied to the nappy area of children (53, 54)) and the United States (eg, 2016 ban on use of triclosan in soaps (55)). Several regulations have also been passed limiting the amount of phenols used in certain products or suggesting tolerable daily intake levels (56, 57). Assessing the latter, however, is difficult. Though many countries implement population-level biomonitoring for environmental phenols, infants are frequently excluded from routine biomonitoring. Our study suggests this exclusion may be an oversight given the relatively high urinary concentrations of some environmental phenols in a healthy, diverse cohort of US infants. Further, we show that exposure to environmental phenols or their precursors in early infancy was associated with higher mean and rapid growth in infant BMI z-score. These findings are concerning given the known consequences these infant growth characteristics pose for later cardiometabolic health outcomes such as obesity (5).

Our study sought to investigate 2 related yet distinct characteristics of infant growth: the level (mean) as well as the trajectory (change over time) of BMI z-score. To examine associations with the level of infant BMI z-score, we employed linear mixed effects models as is commonly done in environmental studies with longitudinal growth outcomes. Point estimates from these models suggested positive associations (though with wide CIs) of mean BMI z-score in infancy with the phenol mixture and with benzophenone-3 and parabens individually. Imprecision when investigating mean level of BMI z-score may be enhanced by failing to account for heterogeneity in growth trajectories that contributes additional variability among individuals. For instance, an infant may be born small but rapidly increase in size, spending most of infancy with a high BMI z-score (rapid increase trajectory), or may be born average and experience a delayed increase in size (delayed increase trajectory), spending late infancy with a high BMI z-score; in both cases the infant would have a high mean BMI z-score. Phenols may have heterogenous associations with these trajectories of growth which are not captured by linear mixed effects models. We therefore also sought to investigate associations with infant BMI z-score trajectory using growth mixture models followed by estimation of relative risk ratios. Our growth mixture models identified 4 trajectories of infant BMI z-score which bear some similarities to trajectories identified in prior studies (5, 58, 59) though differences in growth measurements used, study follow-up time, and sample characteristics preclude direct comparisons. However, prior studies have identified infant growth trajectories described as stable (5, 58, 59), which may be viewed as normative as these infants have a mean birth size z-score of 0 and experience little deviations from this z-score across infancy. We also observed a rapid increase trajectory, which has also been previously described in the literature (2-5, 58-60) and which in our cohort was characterized by low (<0) mean birth z-scores followed by rapid increases until a high (>1) mean z-score was reached at the end of follow-up around 9 months. This pattern of infant growth has been implicated in the development of obesity and cardiometabolic diseases (2-4, 60). Our other patterns (delayed increase and decrease) are less consistently observed, and the clinical relevance of these infant growth patterns is unknown. Findings from our growth mixture models followed by estimation of relative risk ratios analysis implicated a mixture of phenols and, individually, benzophenone-3 and dichlorophenols as positively associated with rapid infant growth. This study is the first to investigate urinary phenol concentrations during infancy and growth outcomes. Given our findings and those from experimental research (10, 12-15), further observational studies of associations between exposure to environmental phenols during infancy are needed.

BPA is ubiquitously used as a plasticizer in consumer products and food packaging and is the most commonly studied environmental phenol (12). One review of observational studies concluded that BPA exposure is not strongly associated with fetal growth (10), whereas another concluded that a positive association exists between BPA exposure and pediatric growth outcomes (17). However, most prior observational studies were cross-sectional. We identified 3 cohort studies of BPA exposure and growth outcomes in children (46, 61, 62). Of those, 2 evaluated early childhood (1-2.5 years) exposure periods. In 2014, Braun et al reported that the average of 2 early childhood urinary BPA concentrations was not associated with mean BMI z-score but was associated with an increased BMI z-score slope in 297 American children from 2 to 5 years of age (46). In 2016, Vafeiadi et al reported that urinary BPA concentrations at 2.5 years were not associated with BMI z-score at 4 years in 235 children in a cohort in Greece (61). Two additional studies have investigated exposure to bisphenols in breastmilk and infant growth. In 2020, Jin et al reported in that BPA in breastmilk was inversely correlated with rate of weight gain in 190 exclusively breastfed Chinese infants (18). Similarly, Çiftçi et al reported an inverse correlation between BPA concentrations in breastmilk and infant weight in 80 exclusively breastfed Turkish infants in 2021 (20). Assessing BPA exposure in dried blood spots, Yeung et al reported in 2019 that BPA exposure was positively associated with rapid infant weight gain and early childhood obesity in 1954 singleton and 966 twin American 3 year olds (19). None of these studies found strong evidence that associations differed by infant sex. In the current study, early infancy urinary BPA was inversely associated with a delayed increase relative to stable trajectory of centered BMI z-score; no other associations were noted overall or in sex- and diet-stratified analyses. Given these equivocal findings and the lack of prospective cohort studies with repeated urinary exposure measures, further research is needed to better understand whether and how BPA exposure affects growth in humans.

Other environmental phenols of interest include benzophenone-3, dichlorophenols, parabens, and triclosan. Benzophenone-3 is commonly used in sunscreens and food packaging materials as it absorbs and scatters ultraviolet radiation. 2,4-Dichlorophenol and 2,5-dichlorophenol are metabolites of chemicals used in a variety of products including dyes, agricultural, and pharmaceutical products as well as pesticides, deodorizers, herbicides, and antiseptics; they are also byproducts of water chlorination. Parabens are primarily used as antimicrobial preservatives in personal care products, pharmaceuticals, and food products. Triclosan is an antimicrobial and antifungal agent commonly found in personal care and hygiene products. Few studies have investigated associations between these environmental phenols and growth; and, to our knowledge, no prior study has focused on exposures in infancy. Reviews of observational studies in humans have concluded that concentrations of benzophenone-3, parabens, and triclosan in the gestational parent are not strongly associated with fetal growth, but inverse associations may exist between concentrations of dichlorophenols in the gestational parent and birth size (10). A recent review and meta-analysis reported positive associations between dichlorophenols and pediatric growth; none of the other environmental phenols were associated with postnatal growth in this synthesis (17). Only 1 prospective cohort was included (63). In this US-based study published in 2017, urinary phenol concentrations measured in a single spot urine sample in 6- to 8-year-old girls were examined in relation to BMI and other measures of adiposity at 7-15 years. The only associations reported were positive associations between 2,5-dichlorophenol and study outcomes. In the current study, we report positive associations between benzophenone-3, dichlorophenols, and parabens and growth outcomes (level and trajectory of BMI z-score) in infants. Our findings do not suggest notable sex- or diet-specific associations. We suggest future studies investigating environmental phenols and growth should focus on 2,5-dichlorophenol given consistency in the epidemiologic literature regarding its potential obesogenic effect. Further, 2,5-dichlorophenol and its precursors have been shown to impact liver and thyroid outcomes (64, 65); these organ systems play key roles in in regulating infant growth and metabolic outcomes (66, 67).

Study Limitations

Our study was limited by a small sample size (n = 199) and a lack of generalizability (ie, our sample comprised relatively healthy, term, singleton infants weighing 2500-4500 g). Nevertheless, rapid infant growth is associated with excess cardiometabolic risks in later life, even among term infants with appropriate for gestational age birthweights (6). Thus, despite our highly selected sample, our findings still may have implications for infant health. This was a secondary data analysis and the underlying study had longer follow-up for female infants and a required feeding regimen for all infants; these study design characteristics may have influenced our findings. We undertook several secondary and sensitivity analyses which collectively suggested these study characteristics did not influence study conclusions. Due to small sample sizes in secondary analyses, results of these analyses should be interpreted with caution. We assume that the subject-specific geometric means of phenol concentrations measured at 6 to 8 and 12 weeks represented an “average early infancy” exposure. Our primary analysis therefore investigated BMI z-score co-occurring with this time period (ie, including BMI z-score measures before 12 weeks). In a sensitivity analysis restricted to BMI z-score assessed at 12 weeks or later, we observe larger positive associations between phenol concentrations and the mean BMI z-score in infancy, but growth mixture models were unable to uncover multiple distinct growth trajectories. These analyses helped ensure temporality of the exposure–outcome association and reduced the potential for reverse causality. Finally, there is potential for residual confounding, as we were unable to adjust for all potential confounders (eg, parental BMI), as well as confounding by in utero exposure, as early infancy phenol concentrations may be expected to correlate with gestational phenol concentrations and gestational chemical exposures may be associated with infant growth. However, prior studies suggest that maternal–offspring phenol concentrations are only weakly correlated and concentrations of urinary phenols in IFED infants were dissimilar to those observed in US women and pregnant women during a similar timeframe (68-71). Previous studies have observed higher infant phenol concentrations than in adults (50), potentially reflecting infant exposure to sources specific to the postnatal period (eg, baby oils, lotions, or wipes) (47-50). Given that even low doses of endocrine disrupting chemicals may have important effects (72) and that infants may be more susceptible to exposures (15, 51), it is encouraging that governments have begun regulating use of these chemicals in infant products (53-57).

Study Strengths

There were several strengths to this study including repeated exposure and outcome assessments, as well as standardized laboratory and study visit procedures that promoted high quality measures. Further, our study sample was racially and socioeconomically diverse relative to other US cohorts, which allowed us to investigate this research question in a population potentially experiencing exposure and health inequities (73). We standardized phenol concentrations using the O’Brien method, which takes into account factors such as infant age and diet as well as sample characteristics when adjusting for urinary dilution (39). Our 2 methodologic approaches to investigating growth provide complementary information about how phenols may influence growth. Investigations of level of BMI z-score enabled us to focus on interpretable mean differences across infancy. Growth mixture models have several strengths including their ability to identify data-based clusters of participants who share similarly shaped growth trajectories (46, 63, 74). They also have limitations including uncertainty in assigning individuals to trajectory groups—an issue that we dealt with by incorporating weights in subsequent analyses (75, 76)—and in the current study, small sample sizes within some trajectory groups. We observed patterns of infant growth that were consistent across most secondary and sensitivity analyses and that replicated well-documented patterns of infant growth (2-4, 6, 60). Finally, we implemented several sensitivity analyses which suggested findings were mostly consistent across various methods for examining exposures and outcomes.

Conclusions

In a US-based prospective cohort study, we observed positive associations between early infancy phenol exposure assessed via urinary biomarkers and higher mean (level) and rapid growth (trajectory) in infant BMI z-score. Ours is the first study to investigate associations between infant urinary phenol biomarker concentrations and growth outcomes. This study provides novel evidence suggesting that phenol exposure during this period of heightened susceptibility may dysregulate infant growth. Our findings are concerning in light of the known significance of dysregulated infant growth for later cardiometabolic health (1-5). However, given the small sample size, limited generalizability of our study sample, and differential follow-up time in male and female infants, we encourage replication in other cohorts. Further studies should also investigate dose–response relationships, whether some infants may experience excess susceptibility to exposure (77), and the potential long-term health effects of exposure in early infancy to endocrine-disrupting chemicals such as phenols.

Abbreviations

BMI

body mass index

BPA

bisphenol A

IQR

interquartile range

LLOQ

lower limit of quantification

LOD

limit of detection

Funding

This research was supported in part by the Intramural Research Program of the National Institute of Environmental Health Sciences (Z01-ES044006) and award number 1K99ES035107-01. Data collection at the Children's Hospital of Philadelphia (CHOP) was supported through Subcontract PHR-SUPS2-S-09-00196 under Contract HHSN291200555546C between the National Institute of Environmental Health Sciences and Social & Scientific Systems Inc.

Disclosures

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data Availability

Some or all datasets generated during and/or analyzed during the current study are not publicly available but are available from the corresponding author on reasonable request.

Disclaimer

The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the Centers for Disease Control and Prevention (CDC). Use of trade name is for identification only and does not imply endorsement by the CDC, the Public Health Service, or the U.S. Department of Health and Human Services.