Prenatal Exposure to Perfluoroalkyl Substances Associated With Increased Susceptibility to Liver Injury in Children

Study Subjects

This study was part of the Human Early-Life Exposome project (HELIX) project,⁽³¹⁾ a collaboration across six established and ongoing longitudinal, population-based birth-cohort studies in Europe: the BiB (Born in Bradford) study in the United Kingdom, the EDEN (Étude des Déterminants pré et postnatals du développement et de la santé de l’Enfant) study in France, the INMA (INfancia y Medio Ambiente) cohort in Spain, the KANC (Kaunas cohort) in Lithuania, the MoBa (Norwegian Mother, Father and Child Cohort Study),⁽³²⁾ and the RHEA Mother Child Cohort study in Crete, Greece. Across these cohorts, a subcohort of 1,301 mothers and their singleton children (~200 children in each cohort) was followed in 2014-2015 for a clinical examination, a computer-assisted interview with the mother, and the collection of additional biological samples. Data collection was standardized across cohorts and performed by trained staff. The full HELIX protocol and database are described elsewhere.⁽³¹⁾

Our study population consisted of 1,105 (85%) mothers and their children (median age, 8.2 years; interquartile range, 6.6-9.1) from the HELIX subcohort, included based on availability of information on PFAS exposure during pregnancy and on liver enzyme levels and serum metabolomic profiling in childhood. All participating families provided written informed consent. Approval for the HELIX project was obtained from the local ethical committees at each site. Additionally, the current study was approved by the University of Southern California Institutional Review Board.

Plasma PFAS Concentrations in Pregnancy

Maternal blood samples were collected in mid-pregnancy for the INMA (mean gestational age [SD], 13.7 [2.0] weeks), MoBa (18.7 [0.9] weeks), and RHEA cohorts (14.1 [3.7] weeks) and in late pregnancy for the BiB (26.6 [1.4] weeks), EDEN (26.1 [1.2] weeks), and KANC cohorts (39.4 [1.3] weeks). Measurement of PFAS was performed at the Department of Environmental Exposure and Epidemiology at the Norwegian Institute of Public Health (NIPH) in plasma or serum samples for BiB, in serum samples for EDEN and RHEA, in whole-blood samples for KANC, and in plasma samples for MoBa. For INMA, PFAS were measured in plasma samples at the Institute for Occupational Medicine, RWTH Aachen University in Germany. Concentrations of PFOS, PFOA, PFNA, PFHxS, and perfluoroundecanoate (PFUnDA) were determined using column-switching liquid chromatography (LC) coupled to a triple-quadrupole mass spectrometer in serum or plasma and online solid-phase extraction and ultra-high-performance LC coupled with tandem mass spectrometry (MS) in whole blood.⁽³³⁾ A ratio of 1:1 was assumed for serum and plasma, whereas 1:2 ratios were assumed for whole-blood versus serum/plasma.⁽³³⁾ The limit of detection (LOD) was 0.02 μg/L for samples assessed at NIPH, whereas for samples assessed at RWTH Aachen University, LODs were 0.05 µg/L for PFNA and 0.1 µg/L for the other PFAS. We used all available PFAS measures and applied quantile regression imputation of left-censored data (QRILC) to obtain singly imputed values for samples with concentration below LOD (ranging from 0.3% to 4.7% across PFAS). QRILC imputes missing elements using random draws from a truncated distribution estimated by a quantile regression and has been shown to have high imputation accuracy.⁽³⁴⁾ Details about quality assurance and quality control have been reported⁽³³⁾ and are briefly described in Supporting Text S1.

Blood Sample Collection in HELIX Children

Children provided blood samples during the subcohort follow-up visit at the end of the clinical examination after a median (5th, 95th percentile) fasting time of 3.3 (2.2, 5.9) hours. Blood samples were collected and processed according to identical predefined standardized protocols across all six cohorts.

Liver Enzyme Levels in Childhood

We assessed levels of ALT, aspartate aminotransferase (AST), and gamma-glutamyltransferase (GGT) in child serum at the Biochemistry Laboratory of the Clínica Universidad de Navarra using homogenous enzymatic colorimetric methods on a Colorimetry Cobas 8000 analyzer, according to the manufacturer’s instructions (Roche Diagnostics GmbH, Mannheim, Germany). All coefficients of variation (CVs) were <3%.

Metabolite Profiling in Childhood

Child metabolite levels were quantified in serum at Imperial College of London (London, UK), using the targeted metabolomics AbsoluteIDQ p180 Kit (Biocrates Life Sciences AG, Innsbruck, Austria). The kit allows the targeted analysis of 188 metabolites of different classes, including amino acids, biogenic amines, acylcarnitines, glycerophospholipids, sphingolipids, and sum of hexoses, thus covering a wide range of analytes and metabolic pathways in one targeted assay. Of the total 188 metabolites, 42 were analyzed quantitatively by LC-electrospray ionization (ESI)-MS/MS with the use of external calibration standards at seven different concentrations and isotope-labeled internal standards for most analytes. The other 146 metabolites were analyzed by flow injection analysis-ESI-MS/MS using a 1-point internal standard calibration with 14 representative internal standards. We excluded 11 serum metabolites having a CV >30% and >30% of the data below LOD, thus leaving 177 metabolites to be used for further analysis. Median CV across these metabolites was 11.9%. Details about the assessment of serum metabolites in HELIX children have been published.⁽³⁵⁾

Statistical Analyses

We defined liver injury risk as having any liver enzyme concentration above the 90th percentile for the study population (ALT, ≥22.7 IU/L; AST, ≥41.4 IU/L; or GGT, ≥17.1 IU/L). Maternal concentrations of PFAS were right-skewed and log₂-transformed to improve model fit. Spearman’s correlation coefficients were computed to assess pair-wise correlations between individual PFAS compounds.

We examined the association of prenatal PFAS mixture exposure with liver injury risk using Bayesian kernel machine regression (BKMR), a nonparametric flexible modeling approach that can accommodate for both correlation and nonlinearity when evaluating the PFAS exposure response. Details about the BKMR models used in our study can be found in Supporting Text S2. Information on covariates was obtained through interviews, self-administered questionnaires, ad-hoc measurements, or medical records. We identified potential confounders and predictors of the outcomes of interest based on previous knowledge and a directed acyclic graph approach (Supporting Fig. S1). We included the following covariates in the models: cohort of inclusion, maternal age (in years), maternal education level (low, middle, or high), maternal prepregnancy body mass index (BMI; in kg/m²), child ethnicity (white, other), child age (in years), and child sex.

We performed several sensitivity analyses in the PFAS mixture-response association. First, we made further adjustment for child plasma levels of PFAS. Second, we additionally adjusted for child weight status (normal weight vs. overweight or obese) based on the age- and sex-specific BMI cutoffs proposed by the World Health Organization.⁽³⁶⁾ Third, we adjusted for gestational weight gain, available food indicators of maternal diet quality (consumption of fish, fruits, and vegetables; in times per week), child sedentary behavior (minutes per week), and food indicators of child diet quality (fish intake, total fruit and vegetable intakes, and total sugar-sweetened beverage consumption; in times per week). Fourth, we repeated analysis while excluding one cohort at a time to assess whether a specific cohort had a marked influence on the overall mixture effect. Fifth, we repeated analysis after stratifying by sex, given that metabolic effects in children of prenatal PFAS exposure have been previously suggested to differ by sex.⁽³⁷⁾ Sixth, we conducted stratified analyses by mid- and late-pregnancy period of PFAS assessment to assess the extent to which maternal physiological differences between these periods can affect the results.

Serum concentrations of metabolites in childhood were right-skewed and log₁₀-transformed to improve normality. We then applied xMWAS, an automated framework for data-driven integration and differential network analysis,⁽³⁸⁾ to identify metabolites that were differentially associated with PFAS between children at high and low risk for liver injury. Details about this approach are provided in Supporting Text S3. In each group of children, we used xMWAS to first select the metabolites with the largest association to PFAS and then generate a global association network. We then compared the eigenvector centrality of each metabolite between the high- and low-risk networks to identify those with differences in their network topology. This can gauge the relative importance of a metabolite in discriminating biological processes between the two risk groups.^{(39, 40)} If a metabolite was not present in one of the networks, its eigenvector centrality was considered 0. Metabolites with an absolute value of difference in their eigenvector centrality >0.2 were considered to have differential contribution to, and thus disparate importance in, the high-risk group as compared to low risk.⁽³⁹⁾ Upon identification of the differentiating metabolites in the high- versus low-risk group, we annotated them to their Kyoto Encyclopedia of Genes and Genomes (KEGG) and Chemical Entities of Biological Interest identifiers and performed pathway overpresentation analysis, using ConsensusPathDB,⁽⁴¹⁾ to characterize dysregulated metabolic pathways that are both associated with PFAS exposure and risk for liver injury. KEGG pathways with more than two annotated metabolites and a P value < 0.05 based on the hypergeometric test were kept for further evaluation.

Last, to provide insight into the joint contribution of prenatal PFAS and the metabolites to liver injury risk, we performed an integrated latent variable analysis to estimate latent unknown clusters⁽⁴²⁾ of children associated with increased susceptibility to liver injury by incorporating information on their prenatal PFAS mixture exposure and the metabolites identified from the network analysis. Details about this method are provided in Supporting Text S4.

All analyses were done using R software (version 3.5.3; R Foundation for Statistical Computing, Vienna, Austria). We used the software packages bkmr for BKMR analysis, xMWAS for network analysis, and LUCIDus for the integrated latent variable analysis.

Study Population Characteristics

No differences in liver enzyme levels were observed between boys and girls. A total of 253 children (22.8%) were classified as being at risk for liver injury. The percentage of children at risk for liver injury was highest in the Greek cohort (31.6%) and lowest in the Lithuanian cohort (4.4%; Table 1). Children at risk for liver injury were more likely to be nonwhites, overweight or obese, and born from mothers with low educational status and higher BMI prepregnancy.

Maternal blood PFAS concentrations of the study population are displayed in Supporting Table S1. Median (25th, 75th percentile) concentrations of PFOA, PFOS, PFHxS, PFNA, and PFUnDA in pregnancy were 2.38 (1.45, 3.45), 6.74 (4.43, 10.35), 0.59 (0.34, 0.93), 0.72 (0.47, 1.11), and 0.20 (0.13, 0.30) ng/mL, respectively. Pair-wise correlations between the PFAS compounds revealed moderate-to-high correlations, with PFOS-PFHxS and PFOA-PFHxS correlations being the highest (Spearman’s rho = 0.70 and 0.65, respectively; Supporting Table S1).

Prenatal PFAS Exposure and Liver Injury Risk in Children

Higher prenatal exposure to PFAS mixture was associated with increased risk for liver injury during childhood (Fig. 1). In Fig. 1A, we display the estimated change in the risk as maternal concentrations of all five PFASs were simultaneously fixed at the same percentile (over a range between 0.3 and 0.75, with 0.05% increments) and compared to when those PFAS concentrations were each at their 25th percentile. PFNA and PFOA had the greatest contribution to the mixture effect, as suggested by their posterior inclusion probabilities in the mixture-response function for liver injury risk (Fig. 1A). Children who were highly exposed to prenatal PFAS mixture (75th percentile) had 2-fold higher risk for liver injury (odds ratio [OR], 2.0; 95% confidence interval [CI], 1.53-2.61) compared to children with low levels of exposure (25th percentile). Similar positive associations with the prenatal PFAS mixture exposure were found when we analyzed the three liver enzymes as separate outcomes (Fig. 1B-D). When exploring the effect of each of the separate PFAS while holding the remaining PFAS within the mixture constant, increased risks were also observed for PFNA and PFOA (Supporting Fig. S2A). In further analyses examining synergism between PFAS within the mixture, we observed that the effects of PFHxS, PFOA, PFOS, and PFUnDA were more pronounced at higher levels of PFNA (Supporting Fig. S2B).

Correlations between maternal and child blood PFAS were low to moderate, with the highest correlations being observed for maternal PFOS with child PFHxS and PFOS (Spearman’s rho = 0.53 and 0.50, respectively). Effect estimates for the prenatal PFAS mixture on liver injury risk did not materially change when we adjusted for child PFAS levels (Supporting Fig. S3A). Effect estimates also did not materially change when we took into account weight status in childhood (Supporting Fig. S3B). Additional adjustment for gestational weight gain, maternal diet-quality indicators, and child lifestyle risk factors for NAFLD, including sedentary behavior and diet-quality indicators, did not change considerably the results (Supporting Fig. S3C). Significant associations of prenatal PFAS mixture exposure with increased liver injury risk were also observed when we repeated the analysis excluding one cohort at a time, as well as when we excluded the two cohorts contributing most to the cases of increased liver enzymes (Supporting Fig. S3D). In stratified analysis by sex, we found that the effect of prenatal PFAS mixture was greater in girls than in boys, although effect estimates were in the same direction and had overlapping CIs (Supporting Fig. S3E). When we stratified by trimester of pregnancy for PFAS assessment, we observed that effect estimates were slightly stronger when maternal PFAS concentrations were assessed in mid-pregnancy (first to second trimester) compared to late pregnancy, but estimates had the same direction and overlapping CIs (Supporting Fig. S3F).

Prenatal PFAS Exposure and the Child Serum Metabolome

We observed that the network of children at high risk for liver injury involved 66 metabolites that were associated with PFAS (Fig. 2A; Supporting Table S2). In comparison, the network of low risk involved 34 metabolites (Fig. 2B; Supporting Table S3). A total of 29 metabolites had the greatest change in eigenvector centrality across the two networks (|delta eigenvector centrality| ≥0.2). These metabolites included five amino acids, one biogenic amine, 19 glycerophospholipids, three sphingomyelins, and one hexose and exhibited both positive and negative associations with maternal serum PFAS in children at high risk for liver injury (Fig. 2C).

Pathway analysis of these metabolites showed that the most dysregulated pathways were related to protein and amino acid metabolism (Fig. 2D). Specifically, six pathways, including protein digestion and absorption, aminoacyl-tRNA (transfer RNA) biosynthesis, valine, leucine, and isoleucine biosynthesis, valine, leucine, and isoleucine degradation, and phenylamine, tyrosine, and tryptophan biosynthesis, were over-represented in the network of high liver injury risk. PFAS-associated metabolic perturbations were also observed for some lipid metabolism pathways, including glycerophospholipid metabolism, ether lipid metabolism, and sphingolipid signaling pathway.

Identification of Children at Risk for Liver Injury Based on Prenatal PFAS Exposure and Serum Metabolome

The integrated latent variable analysis estimated two subgroups of children (Fig. 3A). The high-risk subgroup had 56% increased risk for liver injury (OR, 1.56; 95% CI, 1.21-1.92) compared to the low-risk subgroup. This group was characterized by increased prenatal PFAS levels and increased child serum levels of branched-chain amino acids (BCAAs; valine, leucine, and isoleucine), the aromatic amino acids (AAAs) tryptophan and phenylalanine, biogenic amine acetylornithine, and glycerophospholipids phosphatidylcholine (PC) aa C36:1 and Lyso-PC a C18:1 (Fig. 3B; Supporting Table S4).