Quantifying PFAS-Omics Burden Scores for Nontargeted Analysis Using Multidimensional Item Response Theory: An Exploratory Analysis of Novel and Legacy PFAS in Cord Blood

Abstract

Fetal development is a vulnerable period for exposure to per- and polyfluoroalkyl substances (PFAS). However, certified analytical standards do not exist for many PFAS, limiting our ability to quantify overall exposure burden to PFAS as a chemical class. PFAS-focused nontargeted analysis (NTA) enables detection of PFAS for which chemical standards may not exist. The overall objectives of this study were to provide a more comprehensive picture of PFAS exposure in cord blood, develop cumulative exposure burden scores for the PFAS detected, and evaluate differences in the infant’s PFAS burden score with respect to mother’s parity. We measured PFAS using targeted and NTA methods in cord blood samples collected between 2003 and 2006 in the HOME Study (Cincinnati, Ohio). Using NTA, we putatively identified 42 PFAS in cord blood, 4 of which were also detected in targeted analysis. We summarized an infant’s overall prenatal exposure burden to PFAS using item response theory methods. We constructed two scores, one based on PFAS concentrations from targeted analysis (“PFAS exposure burden scores”), and one based on relative abundance from NTA (“PFAS-omics scores”). As expected, infants with multiparous mothers had significantly lower PFAS exposure burden scores than those with nulliparous mothers, but these disparities were not present when comparing their PFAS-omics scores. Our results show that infants are exposed to a wide range of PFAS, including perfluorinated chemicals, polyfluorinated chemicals, and fluorotelomers, before birth. Further, PFAS-focused NTA can help estimate total exposure to PFAS. Lastly, reported disparities in PFAS exposure burden across parity may depend on the panel of assessed PFAS and their half-lives.

Graphical Abstract

INTRODUCTION

Human exposure to per- and polyfluoroalkyl substances (PFAS) begins prior to birth; PFAS have been detected in cord serum, placenta, and amniotic fluid.^1–3 PFAS, defined as chemicals with at least one aliphatic perfluorocarbon moiety, are a growing class of thousands of chemicals that bioaccumulate in the environment and human tissues.^4–7 PFAS have been manufactured since the late 1940s⁸ and humans now encounter PFAS everyday—they are used in products ranging from oil/water repellant textiles (e.g., carpets) to household goods to food packaging.^4,9,10 Widespread use of PFAS in industrial applications and household products lead to human exposure through contaminated food, water, air, consumer products and house dust.^11,12 Most PFAS are referred to as “forever chemicals” because they have half-lives in the human body and the environment ranging from months to years¹³ due to the strength of the carbon–fluorine bond. Consequently, >98% of the US population had detectable levels of some PFAS in their blood.^14–16 PFAS accumulated in the mother’s body can be transferred to the fetus during pregnancy, resulting in fetal exposure during critical windows of development. Measuring PFAS in cord blood therefore provides an important snapshot of the prenatal exposure burden.

Gestational PFAS exposure is linked with a range of adverse health outcomes in children, including low birth weight, preterm birth, altered DNA methylation, gestational weight gain, reduced vaccine response, and metabolic alterations.^17–22 The American College of Obstetricians and Gynecologists states that reducing exposure to toxic environmental agents, such as PFAS, is a “critical area of intervention”.²³ Therefore, it is important to identify populations with higher prenatal PFAS exposure to prevent exposure and associated adverse health sequelae such as metabolic syndrome, reduced kidney function, disruption of thyroid function, and adverse pregnancy outcomes.³ A major goal of precision environmental health²⁴ is primary prevention through interventions targeted at groups with elevated exposure burden. Researchers have proposed using exposure burden scores to summarize overall exposure burden for biomonitoring and identification of at-risk groups.^24–26

However, quantifying the totality of an individual’s exposure burden to individual PFAS is challenging because of the vast number of PFAS used in commerce. Because of substitutions, in which some PFAS are phased out of production and replaced with other similar PFAS, due to shifting industry practices or regulatory pressure, researchers are interested in studying human exposure to PFAS as an entire chemical class. While serum concentration of two legacy PFAS, perfluorooctanoic acid (PFOA) and perfluorooctanesulfonic acid (PFOS), decreased in the US population after their voluntary phase-out in the 2000s,^27–29 exposure to many replacement PFAS, including GenX and ADONA, is increasing in the environment and in humans.^15,30 A major challenge to assessing exposure to these replacement PFAS is that they are not routinely quantified in traditional targeted analyzes. For example, only a small percentage of total PFAS concentration can often be explained by targeted approaches,³¹ with some studies showing 10% or less coverage in total concentration using targeted methods for environmental samples^32–34 and 50% or less in human serum or plasma samples.³⁵ To allow precision interventions to reduce exposure, it is important to quantify a person’s exposure burden to as many PFAS as possible, which can be addressed using PFAS-focused nontargeted analysis (NTA), and to evaluate and assess their toxicological relevance.

Because there is no gold standard for measuring a person’s cumulative exposure to PFAS, it may be helpful to have different cumulative PFAS metrics that can help researchers understand different aspects of a person’s total body burden of PFAS exposure. The National Academies of Science, Engineering and Medicine recently proposed a cumulative PFAS metric,³⁶ using a simple additive sum of targeted PFAS concentrations, as a clinical biomonitoring tool, showing the importance of a cumulative exposure metric. However, the NASEM sum only includes seven targeted PFAS. Since PFAS are a large chemical class comprised of thousands of compounds, and due to shifting time trends of PFAS use over time, there may be limitations to how well the sum represents a person’s total cumulative PFAS. We previously used item response theory (IRT)^37–40 to construct PFAS exposure burden scores^37,39 for human biomonitoring, estimating a person’s underlying cumulative exposure to both measured and similar unmeasured PFAS. We showed that PFAS exposure burden scores can enable cross-study comparisons when studies may measure different sets of PFAS analytes, by placing burden scores onto a common scale across studies. IRT allows use of all available PFAS data, even if different studies measure different sets of PFAS, which is important because laboratories are now measuring a larger panel of targeted PFAS. However, both the NASEM sum and the PFAS burden score only include PFAS data for those PFAS for which analytical standards exist. Meanwhile, a PFAS-omics burden score, computed using NTA PFAS data, can provide a cumulative exposure to all putatively identified PFAS in a biospecimen sample, beyond those for which analytical standards exist, which can help us understand a person’s cumulative body burden of exposure to PFAS at a snapshot in time. A PFAS-omics burden score can be useful to simplify complexity, and a single score can be more interpretable for clinicians and the general public to understand how a person’s exposure compares relative to that of the general population. It can also be used to conduct risk stratification to identify more highly exposed people to target for precision interventions, such as potentially reducing exposure to PFAS sources. PFAS-omics scores can help researchers more comprehensively understand a person’s cumulative exposure to PFAS, capturing exposure to newer and unknown PFAS structures.

Cumulative cord blood PFAS levels can serve as a direct biomarker of in utero exposure burden and provide insight into early life health risks. Identifying sources of variability in fetal PFAS burden can help researchers understand fetal exposure, and to identify newborns with higher exposure who might be at increased risk of adverse health outcomes. Prior works suggests that transplacental transfer is an elimination pathway for PFAS, and several studies have described differences in newborn PFAS exposure in relation to maternal parity, and potentially systematically lower legacy PFAS levels in multiparous women. For example, studies from the Healthy Pregnancy, Healthy Baby (HPHB) cohort⁴¹ and the Study of Women’s Health Across the Nation (SWAN)⁴² have documented significantly higher levels of several legacy PFAS in nulliparous women. These findings highlight the need to consider parity when estimating cumulative exposure burden. While prior work has focused on legacy PFAS, we extend prior findings by investigating whether similar patterns exist for previously unknown and replacement PFAS identified through NTA. This enables us to determine if differences in PFAS exposure by parity persist across the evolving PFAS landscape.

In this study, we used previously collected targeted PFAS data in cord blood in the HOME Study and collected new NTA data on the samples, and then applied IRT methods to summarize overall exposure burden to PFAS in cord blood. The HOME Study is a prospective pregnancy and birth cohort that enrolled 468 pregnant women living in the Cincinnati, Ohio metropolitan area from March 2003 to January 2006,^43,44 a time period that coincides with the phase-out of legacy PFAS such as PFOA and PFOS, and the emergence of replacement PFAS. Thus, the cohort is able to capture both legacy and emerging PFAS compounds in a population with elevated exposure, possibly due to industrial pollution. In the Ohio River Valley, elevated levels of PFOA in drinking water have been detected due to the DuPont Washington Works plant located 250 miles upstream of Cincinnati, Ohio.^45,46 In the HOME Study, maternal serum PFOA levels were twice as high as a nationally representative sample of US pregnant women during the same time (5.5 versus 2.2 ng/L, respectively).⁴⁷ This population may also have elevated levels of lesser-studied PFAS, a goal of this paper is to use using PFAS-specific NTA to capture all putative PFAS.

This cohort is well suited for NTA because: (1) its archived cord serum samples allow assessment of in utero exposure; (2) it has extensive targeted PFAS measurements, covariate data, and biospecimens, allowing for future integration of PFAS-specific NTA data with health outcomes; and (3) it provides the opportunity to develop PFAS-omics burden scores for the first time in this cohort. The HOME Study’s detailed follow-up to young adulthood will enable future analyzes that link PFAS-omics burden scores to child health outcomes, during a time period of legacy PFAS phase-out and increased demand for replacement PFAS, helping researchers better understand the PFAS exposure landscape. Using HOME Study data, we constructed two IRT burden scores, one based on PFAS concentrations from targeted analysis (“PFAS exposure burden scores”), and one based on relative abundance from NTA (“PFAS-omics scores”). We evaluated differences in the two exposure burden scores across parity, comparing PFAS assayed through a targeted method versus PFAS-focused NTA. This study is one of the first to characterize both targeted and nontargeted PFAS features in cord blood, including legacy and replacement PFAS compounds, enabling future analyzes of their impacts on health outcomes.

MATERIALS AND METHODS

Study Participants

Our study included cord serum samples from mother-infant dyads in the Health Outcomes and Measures of the Environment (HOME) Study,⁴⁴ a prospective pregnancy and birth cohort that enrolled pregnant women from the greater Cincinnati, Ohio region from 2003 to 2006. This study protocol was reviewed and approved by the Institutional Review Board of the Cincinnati Children’s Hospital (IRB # 01-8-5; 2008–0022; 2015–6165). All participants gave informed consent. Of the 420 mother-infant dyads in the HOME Study, we included participants who had targeted PFAS analysis data collected in the cord serum (216 mother-infant dyads) and nontargeted data (120 mother-infant dyads) that met our inter- and intrabatch retention time criteria for PFAS (retention time drift less than 6 s), resulting in a final sample size of 74 (Figure S1).

Targeted PFAS Analysis

Figure S2 shows the goals of this study. The targeted data used in this study has been previously described and summarized (including sample preparation, PFAS quantitation methods, and quality control (QC) and blank samples).^48–51 The cord blood samples were analyzed at the Center for Disease Control and Prevention (CDC) laboratory using isotope dilution and in-line solid-phase extraction liquid chromatography with tandem mass spectrometry.⁵⁰ Targeted PFAS analysis followed all QC procedures recommended by the CDC lab.^52–54 The following 8 PFAS were quantified: 2-(N-Ethyl-perfluorooctane sulfonamido) acetic acid (Et-PFOSA-AcOH), N-methylperfluoroctane sulfonamidoacetic acid (Me-PFOSA-AcOH), perfluorodecanoic acid (PFDeA), perfluorohexanesulfonic acid (PFHxS), perfluorononanoic acid (PFNA), perfluorooctanoic acid (PFOA), perfluorooctanesulfonic acid (PFOS), and perfluorooctane-sulfonamide (PFOSA). The lower limits of detection (LLOD) for targeted PFAS and their detection rates are shown in Table S1. Concentrations below the LLOD were imputed by the LLOD divided by square root of 2. As previously discussed, targeted PFAS analysis followed all QC procedures recommended by the CDC lab.⁵⁰ Each batch included QC and blank samples, and coefficients of variation (CVs) for QC materials were approximately 6%.⁵⁵ The detection levels were 0.09 ng/mL for meFOSAA, 0.2 ng/mL for PFOS, and 0.1 ng/mL for all other PFAS.⁵⁵

NTA Analysis Using FluoroMatch

The sample extraction and data collection processes for these samples have been previously described, and additional details are provided in Table S2.⁵⁵ Briefly, frozen serum samples thawed on ice and then extracted by protein precipitation with acetonitrile containing a mixture of stable isotope standards. The supernatant was transferred by pipet to an LC-HRMS analysis vial. A pooled sample (5 μL of each sample extract aliquoted into one analysis vial) was created in each analytical batch for identification purposes and used to collect MS² data. The extracts were analyzed in triplicate with 10 μL injection volumes on a Thermo QExactive Q-Exactive HF-X Orbitrap MS equipped with a Vanquish ultrahigh-performance liquid chromatograph. We collected data in negative mode using a C-18 column. The mobile phases were UHPLC-MS grade water and acetonitrile containing 1 mM ammonium acetate. Mass spectral data were collected in full scan for all samples (120,000 resolution and mass-to-charge ratio (m/z) range 85–1275) and in full-scan data-dependent (dd) MS² for the pooled samples. Further details of the analytical methods, including the chemicals and reagents used, the chromatography scheme, and analytical sequence and QC description, are provided in the Table S2.

Data Processing.

Mass spectral data files were saved in the .RAW file format and converted to.cdf format using Xcalibur File Converter. The converted files were peak-picked, noise-filtered, and aligned using apLCMS.⁵⁶ The apLCMS output was a table of uniquely detected ions, or features, consisting of m/z, retention time, and ion abundance. The feature table was limited to features with coefficient of variation (CV) < 30% for triplicate injections using xMSanalyzer.⁵⁷ Prior to batch correction, missing data were imputed using the minimum area within a feature divided by the square root of 2 and then the area was log2-transformed to account for non-normality of the residuals. We then performed batch correction on feature data using the WaveICA 2.0 package⁵⁸ in R. Because PFOS and PFOA were consistently detected in our samples and QA/QC, we retained batches in this analysis where the retention times of these two chemicals was within 0.05 min of the first batch’s retention time (4.26 and 7.23 min for PFOA and PFOS, respectively). The batch-corrected feature table was annotated with FluoroMatch Modular version 2.6^59,60 on the University of Michigan’s Great Lakes computing cluster. Annotation included the pooled samples with MS² data collected in each analytical batch. FluoroMatch provides a fully automated approach compiling all mass spectrometric evidence (e.g., retention time, isotopic pattern, fragmentation, and accurate mass) to discern PFAS homologous series and structures from nontargeted data sets. For matching formulas and identifications, we used ±5 ppm mass error. A letter-based annotation confidence scheme (A-E) was assigned by FluoroMatch, where A is equivalent to a Level 2 or 3 on the Schymanski scale,⁶¹ B is equivalent to a Level 3–5, and C, D, and E are equivalent to Level 5. We carried out a sequence of steps to identify PFAS with high-confidence annotations using the “series” or groups of PFAS clustered by FluoroMatch. First, series that had no suspected PFAS identified by FluoroMatch (i.e., the entire series had no ID, formula, or structural information assigned) were removed as these features did not have any database matches either by accurate mass or MS/MS. Then, we required the series to have at least one FluoroMatch Score of B- or greater. Finally, the homologous series of PFAS were examined—if a feature included in the series did not follow retention time order, it was removed from the data set. The final data set contained 42 confirmed or putatively identified PFAS, which we grouped by the functional group of the homologous series (Table 2). Twelve of these putative identifications were not in a homologous series. Confidence levels for each annotation were assigned using the scale by Charbonnet et al.⁶² The NTA Study Reporting Tool (SRT) was used in the preparation of this manuscript and more details are provided in the Supporting Information.⁶³

Table 2.

Identified and Tentatively Identified PFAS Detected in Cord Serum Using FluoroMatch: The HOME Study^a

family	group	chemical name	formula	FluoroMatch (confidence level)	m/z	retention time (min)
perfluorinated	sulfonic acid	PFPeS	C5HF11O3S	D− (3a)	349.09612	3.13
	sulfonic acid	PFHxS	C6HF13O3S	A* (1)	398.93406	4.95
	sulfonic acid	PFHpS	C7HF15O3S	B+ (2c)	448.93261	5.78
	sulfonic acid	PFOS	C8HF17O3S	C+* (1)	498.92977	7.23
	sulfonic acid	PFNS	C9HF19O3S	C+ (3d)	548.9289	8.2
	carboxylic acid	PFPeA	C5HF9O2	D− (3d)	262.97773	3.43
	carboxylic acid	PFHxA	C6HF7O2	B− (2c)	312.97714	4.06
	carboxylic acid	PFHpA	C7HF9O2	E (3d)	363.12748	4.15
	carboxylic acid	PFOA	C8HF15O2	B−* (1)	412.96633	4.26
	carboxylic acid	PFDeA	C10HF19O2	D−* (1)	512.95666	4.83
	polyfluoroalkyl phosphate ester (PAP)	trifluoromethyl dihydrogen phosphate	CH2F3O4P	B− (3c)	164.95509	0.93
	polyfluoroalkyl phosphate esters (PAP)	perfluoroethyl dihydrogen phosphate	C3H2F7O4P	D− (4)	264.95321	1
	polyfluoroalkyl phosphate esters (PAP)	perfluoropropyl dihydrogen phosphate	C4H2F9O4P	D− (4)	314.94655	1.04
	polyfluoroalkyl phosphate esters (PAP)	perfluorononyl dihydrogen phosphate	C9H2F19O4P	D− (4)	564.93089	2.22
	polyfluoroalkyl phosphate esters (PAP)	perfluoroundecyl dihydrogen phosphate	C10H2F21O4P	D− (4)	614.92973	4.55
	imidazole	2-(perfluoromethyl)-1H-imidazole	C4H3F3N2	D− (4)	135.06659	7.23
	imidazole	2-(perfluoroethyl)-1H-imidazole	C5H3F5N2	D− (4)	185.01248	9.04
	imidazole	2-(perfluoropropyl)-1H-imidazole	C6H3F7N2	B− (3c)	235.01114	11.77
	phenol ether	4-(trifluoromethoxy)phenol	C7H5F3O2	D− (4)	177.02275	1.71
	phenol ether	4-(perfluoroethoxy)phenol	C8H5F5O2	B− (4)	227.01442	5.78
	no group	3-(difluoro(trifluoromethoxy)methoxy)-1,1,2,2,3,3-hexafluoropropane-1-sulfonic acid	C5HF11O5S	B− (4)	380.93176	1.64
	no group	2-(perfluoroethoxy)ethan-1-ol	C4H5F5O2	B− (4)	179.01428	2.45
	no group	2,2,2-trifluoroacetamide	C2H2F3NO	B− (4)	112.00331	9.19
polyfluorinated	no group	2-((N-ethyl-1,1,1-trifluoromethyl)sulfonamido)ethyl (2-((N-ethyl-1,1,2,2,2-pentafluoroethyl)sulfonamido)ethyl) hydrogen phosphate	C11H19F8N2O8PS2	A (2a)	553.01108	0.89
	no group	2,3,3,3-tetrafluoro-1-(2-hydroxyphenyl)propan-1-one	C9H6F4O2	B (3a)	221.02446	2.37
	no group	5-chloro-2,2,3,3,4-pentafluoropent-4-enoic acid	C5H2ClF5O2	B (3c)	222.95696	7
	no group	2-((1,1,2,3,3,4,4,5,5,6,6,6-dodecafluorohexyl)oxy)acetic acid	C8H4F12O3	B− (3a)	374.9923	5.17
	no group	1-(1,1,2,2-tetrafluoroethyl)cyclohexane-1-carboxylic acid	C9H12F4O2	B− (4)	227.07043	9.71
	no group	Dimethyl (trifluoromethyl)phosphonate	C3H6F3O3P	B+ (4)	176.99617	3.64
fluorotelomer	alcohol substituted fluorotelomer sulfonic acids	3,3,4,4,5,5,5-heptafluoro-1-hydroxypentane-1-sulfonic acid	C5H4F7O4S	B− (4)	292.96421	2.66
	alcohol substituted fluorotelomer sulfonic acids	3,3,4,4,5,5,6,6,6-nonafluoro-1-hydroxyhexane-1-sulfonic acid	C6H5F9O4S	E (4)	342.88414	3.48
	alcohol substituted fluorotelomer sulfonic acids	3,3,4,4,5,5,6,6,7,7,7-undecafluoro-1-hydroxyheptane-1-sulfonic acid	C7H5F11O4S	B− (3c)	392.96575	4.85
	fluorotelomer sulfonic acids	1:2 Fluorotelomer sulfonic acid	C3H5F3O3S	B− (3a)	176.98374	1.02
	fluorotelomer sulfonic acids	2:2 Fluorotelomer sulfonic acid	C4H5F5O3S	B− (4)	226.97781	2.51
	fluorotelomer sulfonic acids	3:2 Fluorotelomer sulfonic acid	C5H5F7O3S	D− (4)	276.95601	3.25
	fluorotelomer sulfonic acids	4:2 Fluorotelomer sulfonic acid	C6H5F9O3S	D− (4)	326.93265	4.48
	fluorotelomer carboxylic acids (x:1)	1:1 FTCA	C3H3F3O2	C− (3c)	127.00083	8.36
	fluorotelomer carboxylic acids (x:1)	2:1 FTCA	C4H3F5O2	C− (3c)	176.99722	11.88
	fluorotelomer carboxylic acids (x:1)	3,4,4,4-Tetrafluoro-3-(trifluoromethyl)butanoic acid	C5H3F7O2	B (2)	226.99499	12.03
	no group	4,4,5,5,5-pentafluoropentane-1-thiol	C5H7F5S	B− (4)	193.00934	6.45
	no group	1:2 FTCA	C4H5F3O2	B− (3c)	141.0169	7.73
	no group	3,4,4,4-tetrafluorobut-2-enoic acid	C4H2F4O2	A− (2a)	156.99089	8.3

^aConfidence levels for each annotation were assigned using the scale by Charbonnet et al.⁶²

^*indicates these PFAS were confirmed with a reference standard.

Quantifying PFAS Burden Using Unidimensional IRT Model

Following our previous work,^37,39 we used deciles to categorize the 8 individual targeted PFAS analytes and calculated a latent PFAS burden to quantify the exposure to PFAS mixtures by applying item response theory (IRT).^64–76 IRT is a flexible modeling approach, similar to confirmatory factor analysis, which can describe the relationships of measured items (e.g., each PFAS analytes) as indicators for an unobserved latent trait (e.g., latent PFAS burden).

We fitted graded response IRT model (GRM), which can account for ordinal polytomous data, to calculate a latent PFAS burden from the 8 targeted PFAS analytes using targeted PFAS data from 216 participants. Specifically, we constructed the GRM model as follows: There are m = 1, …, 8 PFAS analytes, and we calculated decile cutoffs for each PFAS analyte. Suppose for a given subject i, x_i,m represents their polytomous level on the mth PFAS analyte, such that x_i,m can take on K_m possible values, k_m = 1, ⋯, K_m. θ_i denotes the latent PFAS burden for individual i. The GRM relates an individual’s observed level of a PFAS analyte x_i,m, to an individual’s latent PFAS burden θ_i via the following formulas.

The probability of scoring at or above the given response option, k, given the level of θ_i is

$$Pr\left(x_{i,m}\ge k\mid {\theta }_i\right)=\frac{\text{exp}\left[{\alpha }_m{\theta }_i+{\beta }_{m,k}\right]}{1+\text{exp}\left[{\alpha }_m{\theta }_i+{\beta }_{m,k}\right]}$$

where α_m is a discrimination parameter such that larger values indicate a stronger relationship with the latent variable and β_m,k is an extremity parameter that indicates the boundaries between adjacent categories.

The probability of scoring at the current response option, k, is

$$Pr\left(x_{i,m}=k\mid {\theta }_i\right)=\frac{\text{exp}\left[{\alpha }_m{\theta }_i+{\beta }_{m,k}\right]}{1+\text{exp}\left[{\alpha }_m{\theta }_i+{\beta }_{m,k}\right]}-\frac{\text{exp}\left[{\alpha }_m{\theta }_i+{\beta }_{m,k+1}\right]}{1+\text{exp}\left[{\alpha }_m{\theta }_i+{\beta }_{m,k+1}\right]},\text{where}{\beta }_1>{\beta }_2>\dots >{\beta }_{K_m}$$

We estimated an expected a posteriori estimate (EAP) $\hat{\theta }$, which is quantified as the most plausible value of the latent PFAS burden for subject i, given subject i’s levels of PFAS analytes (x_i), a standard normal prior distribution g(θ), and the graded response IRT model

$$\widehat{{\theta }_i}=\int \theta p\left(\theta \mid x_i;\alpha ,\beta \right)g(\theta )d\theta$$

Quantifying PFAS-omics Scores Using Multidimensional IRT Model

We used quartiles to categorize the 42 PFAS analytes included in the NTA and calculated a latent PFAS-omics burden score to quantify the exposure to the PFAS putatively identified in NTA using an exploratory multidimensional IRT model for the 120 participants. Specifically, a multidimensional IRT model allows multiple latent variables to underlie values of the PFAS analytes. For a model that specifies D latent variables, a multidimensional GRM gives the probability of scoring at or above the given response option as

$$\text{Pr}\left(x_{i,m}\ge k\mid {\theta }_i\right)=\frac{\text{exp}\left[\sum _d^D{\alpha }_{dm}{\theta }_{di}+{\beta }_{mk}\right]}{1+\text{exp}\left[\sum _d^D{\alpha }_{dm}{\theta }_{di}+{\beta }_{mk}\right]}$$

and the probability of a value in category k found, as before, by Pr(x_i,m ≥ k|θ_i) – Pr(x_i,m ≥ k + 1 |θ_i). If it is known in advance which analytes correspond to which latent variables, it is possible to estimate the parameters of a confirmatory IRT model by constraining α_dm = 0 wherever analyte m is not an indicator of dimension d. If instead, these relationships are not known in advance, we may wish to estimate every α_dm value in an exploratory IRT model (also known as item factor analysis).^77,78 However, this model is not mathematically identified if D > 1.⁷⁹ Therefore, we identify each multidimensional model by fixing one α_dm = 0 and then transforming the item parameter estimates to improve interpretability using oblimin rotation, which allows latent variables to be correlated. Notably, this rotation does not affect the model-data fit or the amount of information conveyed by the model. Instead, rotation may improve our ability to assign subsets of items as the primary indicators of each latent variable. For each item, we also calculated its h2 value, a quantity that is invariant to rotation and which represents the proportion of variance in the observed data that is associated with the latent factors. The h2 values indicate the proportion of shared variance of the indicator with the latent factor, and higher h2 values suggest that a great proportion of the variance in the analyte is explained by the latent variable. In this paper, we consider h2 values less than 0.1 as indication that that factor does not closely represent variability in the analyte since less than 10% of its variance is associated with the latent factors.⁸⁰ Multidimensional EAP scores may be calculated using the same formula stated above, where θ_i is a vector of dimension D and g(θ) is a matrix that gives the correlations among dimensions after rotation.

We first explored the number of dimensions by comparing IRT models from 1 to 6 dimensions. We selected the model with lowest BIC as the best model using the BIC criterion.⁸¹ We estimated the EAP score for each dimension of the best fitting model after using oblimin rotation, and then calculated a standardized average burden score as the PFAS-omics score. We also calculated a weighted PFAS-omics score, which used the sum of squared loadings (SS loadings) of each factor as weights in the secondary analyzes. The SS loadings of each factor represented the variance of each factor. We first multiplied the factor scores by the SS loading of each factor, calculated the sum of the weighted scores, then divided the weighted sum by the total SS loadings of the three factors.

Our IRT models were estimated using “mirt” package⁸² (1.40 version) in R (4.1.1 version), running under Windows system.

PFAS Exposure Differences across Parity

Based on parity, we categorized the mothers into three groups (nulliparous = 0, primiparous = 1, multiparous = 2 and above). We used the Kruskal–Wallis rank sum test (for comparing multiple groups) and Wilcoxon Rank Sum Test (for comparing two groups) to examine whether there were differences in exposure to individual PFAS or PFAS mixtures across parity groups. We calculated the robust Cohen’s d with 500 bootstrap samples using WRS2 package⁸³ in R to measure the effect size of differences between parity groups for PFAS burden, PFAS-omics scores, and PFAS analytes that were found to have significant differences in the Kruskal–Wallis rank sum test. Robust Cohen’s d measures the effect size when comparing two samples and can be interpreted as none-small (0–0.10), small-medium (0.1–0.25), and medium-large (0.25–0.40).

We also calculated the association of PFAS exposures and parity using regression models which adjusted for mother’s age at delivery, race, and household income.

Secondary Analysis of Accounting for Measurement Errors in the IRT Model

Since the burden scores estimated from IRT models have standard errors, we conducted a secondary analysis using a resampling approach to more appropriately account for the uncertainty in latent burden scores. First, we imputed the PFAS burden and PFAS-omics scores using 1000 sets of imputed plausible values⁸⁴ from the IRT models. With each set of plausible values, we investigated the differences of the new PFAS burden and the new PFAS-omics scores across parity groups using Wilcoxon Rank Sum Test, assessed the effect sizes of differences by calculating the robust Cohen’s d, and calculated the adjusted associations using the new estimated burden scores.

Exploratory secondary analysis of prior breastfeeding duration and PFAS scores:

We conducted an additional secondary analysis that examined the association between breastfeeding duration and cord blood PFAS levels, among newborns born to parous women, using Spearman correlation.

RESULTS AND DISCUSSION

Sample Characteristics

Table 1 presents the socio-demographic characteristics of the participants. The study sample that had both targeted PFAS and nontargeted PFAS measurements was predominantly non-Hispanic White (65.2%), and 28.4% were non-Hispanic Black. 45.9% of the infants were identified as female at birth. Median age of the mothers was 31.4 (Q1 was 26.0 and Q3 was 33.5), and median family income of the sample was $65,000 (Q1 was $35,000 and Q3 was $75,000). 39.4% of the mothers were nulliparous, 42.3% were primiparous, and 18.3% were multiparous.

Table 1.

Summary Statistics of HOME Study Participants with Targeted PFAS and Non-targeted (NTA) PFAS Measures

(A) Demographics of the sample
	sample with targeted PFAS measures (n = 216)	sample with NTA PFAS measures (n = 120)	sample with NTA PFAS and targeted PFAS measures (n = 74)
child sex (%)
male	50.5	52.6	54.1
female	49.5	47.4	45.9
maternal race group (%)
non-hispanic white	65.1	70.2	65.2
non-hispanic black	28.8	23.7	28.4
hispanic	6.0	6.1	5.4
parity (%)
nulliparous (0)	43.8	44.9	39.4
primiparous (1)	31.9	36.4	42.3
multiparous (2 and above)	24.3	18.7	18.3
mother’s age at delivery (median, [25th, 75th])	31.04 [26.22, 33.39]	31.15 [26.24, 33.44]	31.35 [25.99, 33.45]
household Income (dollars) (median, [25th, 75th])	55,000 [27,500, 85,000]	55,000 [35,000, 77,500]	65,000 [35,000, 75,000]

Median (interquartile range) of targeted PFAS in cord serum
cord serum PFAS (median, [25th, 75th])	sample with targeted PFAS measures (n = 216)	sample with NTA PFAS and targeted PFAS measures (n = 74)
Et-PFOSA-AcOH (ng/mL)	0.07 [0.07, 0.07]	0.07 [0.07, 0.07]
Me-PFOSA-AcOH (ng/mL)	0.26 [0.17, 0.52]	0.26 [0.17, 0.38]
PFDeA (ng/mL)	0.07 [0.07, 0.07]	0.07 [0.07, 0.07]
PFHxS (ng/mL)	0.70 [0.40, 1.10]	0.68 [0.40, 1.10]
PFNA (ng/mL)	0.40 [0.30, 0.60]	0.50 [0.30, 0.60]
PFOA (ng/mL)	3.30 [2.49, 4.60]	3.28 [2.60, 4.40]
PFOS (ng/mL)	4.40 [3.11, 6.20]	4.40 [3.15, 6.35]
PFOSA (ng/mL)	0.07 [0.07, 0.07]	0.07 [0.07, 0.07]

Non-targeted Analysis

Of the 42 confirmed or putatively identified PFAS detected in the cord samples, 19 were polyfluorinated or fluorotelomers and 23 were perfluorinated (Table 2, Figure S3). Of these PFAS, confirmed structures (Level 1 on the Schymanski Scale) included PFOA, PFHxS, PFDeA and PFOS. Other high confidence annotations (A FluoroMatch Score) included a hydrogen substituted unsaturated perfluorocarboxylic acid (H-PFUCAs) (C₄H₂F₄O₂) and a substituted phosphate ester of N-ethyl perfluorooctane sulfonamido ethanol (SAmPAP) (C₁₁H₁₉F₈N₂O₈PS₂). We detected several PFAS homologous series in cord blood with at least three homologues detected, including perfluorocarboxylic acids (PFCAs), perfluorosulfonic acids (PFSAs), polyfluoroalkyl phosphate ester (PAPs), perfluorinated imidazoles, Alcohol substituted fluorotelomer sulfonic acids, Fluorotelomer sulfonic acids (FTSs), Fluorotelomer carboxylic acids (FTCAs, X:1) (Table 2).

Correlations between Targeted PFAS and Nontargeted PFAS

Four PFAS (PFDeA, PFHxS, PFOA and PFOS) were annotated in both targeted analysis and NTA (Figure 1). Concentrations in targeted analysis were correlated with log abundances in NTA for PFOA (spearman ρ = 0.52, p < 0.001), PFOS (spearman ρ = 0.74, p < 0.001), and PFHxS (spearman ρ = 0.64, p < 0.001). No correlation was found between PFDeA concentration and PFDeA abundance (p = 0.35). Four PFAS detected in the targeted analysis, including, Et-PFOSA-AcOH, Me-PFOSA-AcOH, PFNA, and PFOSA, were not annotated in the NTA feature table (Table S1). Me-PFOSA-AcOH and Et-PFOSA-AcOH were detected above the LLOD in 93% and 23% of the targeted data, respectively; however, in NTA using FluoroMatch, the m/z for Me-PFOSA-AcOH was not detected within a reasonable retention time compared to the other PFAS detected and the m/z for Et-PFOSA-AcOH was annotated with a score of “D-” and did not have a reasonable retention time. Thus, both were filtered from the feature table prior to statistical analysis. This is likely due to the relatively low concentrations of these PFAS in these samples (Table 1B). Further, no MS/MS was acquired at high enough sensitivity to provide structural information. Intelligent data-acquisition methods of data-independent analysis approaches in combination with FluoroMatch could improve coverage in the future.^85,86 The detection frequency of PFOSA in the targeted data set was only 6% above LLOD, thus it is likely that this PFAS was not detected in NTA because the detection rate and abundance of this chemical was too low.

Figure 1.

Correlation between concentration and relative abundance in PFOA, PFOS, PFDeA, and PFHxS in HOME Study participant cord serum samples. (A) PFDeA concentration vs abundance (B) PFHxS concentration vs abundance (C) PFOA concentration vs abundance (D) PFOS concentration vs abundance.

Estimating PFAS Burden from 8 PFAS in the Targeted Analysis

We calculated decile cutoffs (Table S3) for the 8 PFAS analytes in the targeted analysis and fit a unidimensional IRT model. The decile cutoffs were the same for certain deciles of Et-PFOSA-AcOH, Me-PFOSA-AcOH, PFDeA, PFNA, and PFOSA; therefore, these PFAS had fewer than ten categories, as there was less variability in the distribution of these analyte concentrations in the population. Table S4 presents item parameters of the unidimensional IRT model calibrated from 8 PFAS analytes in the targeted analysis. Using this IRT model, we estimated a PFAS burden score to quantify exposure to the 8 PFAS mixtures which were measured in the targeted analysis for each participant.

Estimating PFAS-Omics Score from 42 PFAS in the NTA

We calculated quartile cutoffs (Table S5) for the 42 PFAS analytes in the NTA and compared the model fits of multidimensional IRT models. Table S6 presents the fit statistics and M2 statistics of the IRT models from 1 factor to 6 factors which were calculated from the log-transformed abundances of the 42 PFAS in the nontargeted analysis. We selected the 3-factor model as the best fitting model which had the lowest BIC. Coefficients, factor loadings, and item fits of the 3-factor IRT model are presented in Tables S7, S8, and S9. The 6 most informative PFAS analytes in the 3-factor IRT model were putatively identified as 2-(perfluoroethoxy)ethan-1-ol, Perfluoroundecyl dihydrogen phosphate, Perfluorononyl dihydrogen phosphate, 2,3,3,3-tetrafluoro-1-(2-hydroxyphenyl)propan-1-one, Perfluoroethyl dihydrogen phosphate, 1:2 Fluorotelomer sulfonic acid, with h2 values of 0.951, 0.899, 0.898, 0.706, 0.695, and 0.629. Seventeen PFAS analytes were less informative in the model (with h2 values less than 0.1).

Table S10 shows the covariances of the 3 latent factors. We calculated an averaged burden from the 3 latent factors as the PFAS-omics score. Figure 2 shows the correlations between PFAS-omics score and the 8 PFAS assessed as part of the targeted analysis. PFAS burden calculated from 8 PFAS analytes in the targeted analysis was not correlated with PFAS-omics score calculated from 42 PFAS analytes in the NTA (p = 0.32). No correlation was found between PFAS-omics score and any PFAS analyte in the target analysis.

Figure 2.

Scatter plots of PFAS-omics score vs targeted PFAS concentrations in HOME Study participant cord serum samples (A) Scatter plots of PFAS-omics score vs PFAS burden (B)Scatter plots of PFAS-omics score vs PFOA concentration (C) Scatter plots of PFAS-omics score vs PFOS concentration.

Figure S4 shows the exposure patterns of the 42 PFAS analytes in the NTA analysis. Abundance of the 42 PFAS analytes for each participant are shown in quartiles. Each participant had unique exposure patterns of the 42 putatively identified PFAS in this study sample. There was no obvious relationship between PFAS burden and PFAS abundance in the NTA, except that participants with higher PFAS burden tended to have higher PFOS in the NTA. (Figure S4A) On the other hand, as shown in Figure S4B, participants with higher PFAS-omics score tended to have higher abundance of many PFAS analytes in the NTA.

PFAS Exposure Differences across Parity Groups

Table 3 presents the median and interquartile range of the two burden scores and each PFAS analyte in the targeted and nontargeted analyzes across the three parity groups. PFAS burden (p = 0.002) and concentrations of PFHxS (p = 0.015), PFOA (p < 0.001), and PFOS (p = 0.003) in the targeted analysis were significantly different across parity groups. PFAS-omics score was not different across the parity groups (p = 0.848). Of the 42 NTA PFAS analytes, PFOA (p = 0.02) and perfluoroundecyl dihydrogen phosphate (p = 0.02) differed across parity groups. Table S11 presents the findings in the sample having both targeted and nontargeted PFAS measures.

Table 3.

Median (IQR) of PFAS Analytes Across Parity Groups^ab

name	parity = 0	parity = 1	parity = 2 and above	p
Burden score
PFAS burden	0.39 [−0.46, 0.89]	−0.08 [−0.66, 0.45]	−0.13 [−0.98, 0.17]	0.002
PFAS-omics score (42 NTA PFAS)	0.05 [−0.63, 0.65]	0.15 [−0.96, 0.94]	0.08 [−0.43, 0.77]	0.848
PFAS-omics score (25 NTA PFAS)	0.00 [−0.71, 0.77]	0.26 [−0.92, 0.92]	0.27 [−0.33, 0.63]	0.806
weighted PFAS-omics score (42 NTA PFAS)	−0.02 [−0.50, 0.42]	0.11 [−0.61, 0.61]	0.04 [−0.35, 0.47]	0.835
weighted PFAS-omics score (25 NTA PFAS)	−0.09 [−0.61, 0.43]	0.05 [−0.61, 0.54]	0.09 [−0.33, 0.34]	0.735
Targeted PFAS analytes (ng/mL)
Et-PFOSA-AcOH	0.07 [0.07, 0.07]	0.07 [0.07, 0.10]	0.07 [0.07, 0.07]	0.496
Me-PFOSA-AcOH	0.26 [0.17, 0.52]	0.26 [0.17, 0.41]	0.35 [0.17, 0.61]	0.461
PFDeA	0.07 [0.07, 0.07]	0.07 [0.07, 0.07]	0.07 [0.07, 0.07]	0.702
PFHxS	0.80 [0.50, 1.20]	0.60 [0.40, 0.90]	0.60 [0.40, 1.05]	0.015
PFNA	0.45 [0.38, 0.60]	0.40 [0.30, 0.52]	0.40 [0.30, 0.58]	0.286
PFOA	3.83 [2.90, 5.76]	3.00 [2.20, 3.92]	3.10 [2.22, 4.03]	<0.001
PFOS	5.35 [3.40, 6.93]	4.10 [2.95, 6.20]	3.80 [2.42, 5.15]	0.003
PFOSA	0.07 [0.07, 0.07]	0.07 [0.07, 0.07]	0.07 [0.07, 0.07]	0.317
NTA PFAS analytes (relative abundance)
PFPeS	43.92 [43.53, 44.35]	44.07 [43.58, 44.97]	43.98 [43.25, 44.72]	0.512
PFHxS	46.12 [45.47, 46.48]	45.81 [45.41, 46.25]	45.71 [45.15, 46.26]	0.238
b-PFHpS	41.77 [40.83, 42.43]	41.56 [40.47, 42.47]	41.61 [40.88, 42.10]	0.752
PFOS	48.42 [47.84, 49.00]	48.36 [47.74, 49.03]	47.99 [47.83, 48.33]	0.159
PFNS	44.13 [42.35, 45.79]	43.91 [40.89, 48.01]	44.80 [42.01, 46.06]	0.905
PFPeA	33.18 [31.15, 34.79]	34.38 [31.78, 39.44]	33.69 [30.56, 35.21]	0.377
PFHxA	48.65 [48.24, 48.78]	48.63 [48.44, 48.91]	48.71 [48.60, 49.20]	0.377
PFHxA	48.65 [48.24, 48.78]	48.63 [48.44, 48.91]	48.71 [48.60, 49.20]	0.377
PFHpA	43.31 [42.59, 43.86]	43.12 [42.60, 43.75]	43.43 [42.66, 43.92]	0.624
PFOA	47.01 [46.44, 47.36]	46.52 [46.02, 46.97]	46.55 [46.11, 46.93]	0.021
PFDeA	46.82 [46.40, 47.36]	47.01 [46.60, 47.51]	46.93 [46.56, 47.45]	0.485
trifluoromethyl dihydrogen phosphate	38.18 [36.51, 43.55]	41.94 [36.87, 44.84]	38.93 [32.88, 40.78]	0.211
perfluoroethyl dihydrogen phosphate	46.77 [46.43, 47.38]	46.89 [46.60, 47.55]	46.99 [46.47, 47.46]	0.51
perfluoropropyl dihydrogen phosphate	39.26 [38.67, 39.98]	39.20 [38.40, 39.84]	39.45 [39.09, 40.22]	0.41
perfluorononyl dihydrogen phosphate	46.76 [46.33, 47.21]	47.03 [46.57, 47.62]	47.07 [46.74, 47.41]	0.103
Perfluoroundecyl dihydrogen phosphate	48.66 [48.32, 48.93]	49.11 [48.65, 49.36]	48.89 [48.44, 49.23]	0.02
2-(perfluoromethyl)-1H-imidazole	46.98 [46.63, 47.38]	46.95 [45.92, 47.90]	47.27 [46.98, 47.91]	0.25
2-(perfluoroethyl)-1H-imidazole	50.18 [50.07, 50.49]	50.18 [50.04, 50.27]	50.25 [50.14, 50.35]	0.359
2-(perfluoropropyl)-1H-imidazole	43.15 [42.80, 43.67]	42.95 [42.75, 43.50]	43.41 [42.69, 43.87]	0.464
4-(trifluoromethoxy)phenol	42.84 [41.39, 44.42]	42.14 [39.79, 43.48]	42.78 [38.96, 43.65]	0.204
4-(perfluoroethoxy)phenol	38.51 [37.75, 39.25]	38.44 [37.42, 39.55]	38.61 [37.83, 39.64]	0.96
NTA PFAS analytes (relative abundance)
3,3,4,4,5,5,6,6,7,7,7-undecafluoro-1-hydroxyheptane-1-sulfonic acid	43.26 [42.31, 43.66]	43.02 [42.71, 43.64]	43.52 [42.98, 43.86]	0.232
3,3,4,4,5,5,6,6,6-nonafluoro-1-hydroxyhexane-1-sulfonic acid	39.57 [38.64, 41.62]	40.73 [38.70, 42.41]	39.48 [38.70, 41.27]	0.379
3,3,4,4,5,5,5-heptafluoro-1-hydroxypentane-1-sulfonic acid	44.47 [44.27, 44.65]	44.37 [43.96, 44.69]	44.45 [44.18, 44.64]	0.548
1:2 Fluorotelomer sulfonic acid	54.62 [54.41, 54.87]	54.63 [54.34, 54.72]	54.61 [54.43, 54.81]	0.759
2:2 Fluorotelomer sulfonic acid	46.18 [45.71, 46.95]	45.98 [45.66, 46.55]	46.28 [45.66, 46.58]	0.406
3:2 Fluorotelomer sulfonic acid	46.27 [46.03, 46.61]	46.40 [46.21, 46.84]	46.37 [46.19, 46.70]	0.322
4:2 Fluorotelomer sulfonic acid	42.26 [41.72, 42.68]	42.33 [41.43, 42.63]	42.24 [41.34, 42.75]	0.876
1:1 FTCA	59.93 [59.73, 60.08]	59.80 [59.55, 59.97]	59.83 [59.80, 60.01]	0.083
2:1 FTCA	33.61 [32.55, 35.46]	33.70 [32.85, 34.91]	34.13 [32.72, 35.84]	0.882
3,4,4,4-Tetrafluoro-3-(trifluoromethyl)butanoic acid	37.09 [35.25, 41.46]	37.78 [34.56, 41.47]	37.69 [33.99, 43.00]	0.999
2-((N-ethyl-1,1,1-trifluoromethyl)sulfonamido)ethyl (2-((N-ethyl-1,1,2,2,2-pentafluoroethyl) sulfonamido)ethyl) hydrogen phosphate	46.28 [45.86, 46.67]	46.24 [45.74, 46.66]	46.44 [45.90, 46.85]	0.873
3-(difluoro(trifluoromethoxy)methoxy)-1,1,2,2,3,3-hexafluoropropane-1-sulfonic acid	40.39 [38.54, 42.69]	40.08 [38.43, 41.81]	41.12 [39.34, 43.00]	0.404
2,3,3,3-tetrafluoro-1-(2-hydroxyphenyl)propan-1-one	48.87 [48.58, 49.16]	48.63 [48.35, 49.01]	48.77 [48.42, 49.11]	0.144
2-(perfluoroethoxy)ethan-1-ol	56.99 [56.44, 57.13]	56.66 [56.24, 57.12]	56.99 [56.49, 57.23]	0.351
Dimethyl (trifluoromethyl)phosphonate	49.29 [49.13, 49.54]	49.31 [48.95, 49.45]	49.35 [49.01, 49.51]	0.807
2-((1,1,2,3,3,4,4,5,5,6,6,6-dodecafluorohexyl)oxy)acetic acid	54.47 [54.11, 54.74]	54.54 [54.22, 54.71]	54.68 [54.48, 54.81]	0.181
4,4,5,5,5-pentafluoropentane-1-thiol	48.98 [48.86, 49.09]	48.97 [48.90, 49.04]	49.00 [48.90, 49.12]	0.773
5-chloro-2,2,3,3,4-pentafluoropent-4-enoic acid	40.89 [40.12, 41.44]	40.94 [40.45, 42.00]	41.35 [40.67, 41.79]	0.3
2,2,2-trifluoroacetamide	51.27 [51.16, 51.48]	51.10 [50.98, 51.30]	51.26 [50.98, 51.48]	0.039
1-(1,1,2,2-tetrafluoroethyl)cyclohexane-1-carboxylic acid	48.39 [48.04, 48.79]	48.73 [48.11, 48.87]	48.76 [48.22, 48.86]	0.27
1:2 FTCA	66.80 [66.63, 66.94]	66.72 [66.59, 66.96]	66.78 [66.62, 66.90]	0.942
3,4,4,4-tetrafluorobut-2-enoic acid	57.12 [56.67, 57.44]	57.05 [56.78, 57.33]	57.25 [56.80, 57.58]	0.482

^aP-values of Kruskal–Wallis rank sum test are shown in the table.

^b*Sample size for targeted PFAS: n = 92 for nulliparous, n = 67 for primiparous, n = 51 for multiparous. *Sample size for NTA PFAS: n = 48 for nulliparous, n = 39 for primiparous, n = 20 for multiparous.

Table S12 presents the robust Cohen’s “d” statistic for the significant comparisons. Targeted PFHxS (robust Cohen’s d = −0.44, 95% CI: −0.88, −0.03 for multiparous vs nulliparous; robust Cohen’s d = −0.51, 95% CI: −0.89, −0.14 for primiparous vs nulliparous), PFOA (robust Cohen’s d = −0.52, 95% CI: −0.88, −0.26 for multiparous vs nulliparous; robust Cohen’s d = −0.53, 95% CI: −0.81, −0.26 for parity primiparous vs nulliparous), PFOS (robust Cohen’s d = −0.61, 95% CI: −0.94, −0.24 for multiparous vs nulliparous; robust Cohen’s d = −0.33, 95% CI: −0.63, −0.03 for primiparous vs nulliparous) were significantly lower both in the multiparous and primiparous. NTA PFOA (robust Cohen’s d = −0.55, 95% CI: −1.24, −0.01 for multiparous vs nulliparous; robust Cohen’s d = −0.58, 95% CI: −1.04, −0.12 for parity primiparous vs nulliparous) was significantly lower both in the multiparous and primiparous groups, while perfluoroundecyl dihydrogen phosphate was significantly higher in the primiparous group (robust Cohen’s d = 0.64, 95% CI: 0.14, 1.14).

PFAS burden scores were lower in infants whose mothers were multiparous (Wilcoxon Rank Sum test p = 0.001; robust Cohen’s d = −0.54, 95% CI: −0.90, −0.18) and in children whose mothers were primiparous (Wilcoxon Rank Sum test p = 0.012; robust Cohen’s d = −0.40, 95% CI: −0.77, −0.13) compared with infants whose mothers were nulliparous. The difference was null when comparing their PFAS-omics scores (robust Cohen’s d = 0.18, 95% CI: −0.34, 0.73 for multiparous vs nulliparous; robust Cohen’s d = 0.08, 95% CI: −0.57, 0.42 for parity primiparous vs nulliparous). (Table S12 and Figures 3 and 4) Findings were consistent when using weighted PFAS-omics scores. The findings were consistent in the sample that had both targeted PFAS and nontargeted PFAS measurements. (Table S13)

Figure 3.

Box plots of PFAS burden and PFAS-omics scores across parity groups. P-values of pairwise comparisons between the parity groups are shown in the figures.* Sample size for PFAS burden: n = 92 for parity = 0, n = 67 for parity = 1, n = 51 for parity = 2+. *Sample size for PFAS-omics score: n = 48 for parity = 0, n = 39 for parity = 1, n = 20 for parity = 2+.

Figure 4.

Effect sizes of differences of comparing PFAS burden and PFAS-omics scores across parity groups *Sample size for PFAS burden: n = 92 for parity = 0, n = 67 for parity = 1, n = 51 for parity = 2+. *Sample size for PFAS-omics score: n = 48 for parity = 0, n = 39 for parity = 1, n = 20 for parity = 2+.

Table 4 shows the adjusted association of PFAS exposures and parity. After adjusting for mothers’ age, household income, and race, compared with the nulliparous, targeted PFOA (β = −1.37, p = 0.001), nontargeted PFOA (β = −0.45, p = 0.016), 1:1 FTCA (β = −0.17, p = 0.026) and 2,3,3,3-tetrafluoro-1-(2-hydroxyphenyl)propan-1-one (β = −0.25, p = 0.049) were significantly lower in the primiparous; and targeted PFOA (β = −1.47, p = 0.002) and PFOS (β = −1.63, p = 0.002), nontargeted PFOA (β = −0.54, p = 0.032) were significantly lower in the multiparous; while nontargeted 1-(1,1,2,2-tetrafluoroethyl)cyclohexane-1-carboxylic acid (β = 0.48, p = 0.029) was significantly higher in the multiparous.

Table 4.

Adjusted Associations of Serum PFAS Concentrations and Non-targeted PFAS Relative Abundance with Parity

PFAS	parity (ref = 0)		race group (ref = NH white)		income (increment per 10k)	mom age (years)
	parity = 1	parity = 2 and above	NH Black	hispanic
Burden Scores
PFAS burden	−0.316 [−0.596, −0.037], p = 0.027	−0.54 [−0.858, −0.221], p = 0.001	−0.172 [−0.497, 0.153], p = 0.297	0.242 [−0.261, 0.745], p = 0.345	0.012 [−0.026, 0.049], p = 0.544	0.028 [0.001, 0.056], p = 0.044
PFAS-omics score (42 NTA PFAS)	−0.129 [−0.584, 0.326], p = 0.575	0.106 [−0.509, 0.721], p = 0.733	0.569 [−0.006, 1.145], p = 0.052	1.043 [0.163, 1.922], p = 0.021	0.024 [−0.04, 0.087], p = 0.462	0.035 [−0.018, 0.089], p = 0.195
PFAS-omics score (25 NTA PFAS)	−0.108 [−0.56, 0.344], p = 0.636	0.069 [−0.542, 0.68], p = 0.824	0.607 [0.036, 1.179], p = 0.038	1.108 [0.234, 1.982], p = 0.013	0.02 [−0.043, 0.083], p = 0.533	0.045 [−0.008, 0.099], p = 0.097
Weighted PFAS-omics score (42 NTA PFAS)	−0.033 [−0.332, 0.266], p = 0.828	0.071 [−0.333, 0.476], p = 0.728	0.382 [0.004, 0.761], p = 0.048	0.728 [0.149, 1.306], p = 0.014	0.015 [−0.026, 0.057], p = 0.464	0.028 [−0.008, 0.063], p = 0.126
Weighted PFAS-omics score (25 NTA PFAS)	−0.055 [−0.357, 0.247], p = 0.717	0.086 [−0.322, 0.495], p = 0.677	0.361 [−0.021, 0.743], p = 0.064	0.684 [0.100, 1.269], p = 0.022	0.017 [−0.025, 0.059], p = 0.428	0.022 [−0.014, 0.058], p = 0.224
Targeted PFAS analytes
Et-PFOSA-AcOH	−0.001 [−0.019, 0.018], p = 0.958	−0.013 [−0.035, 0.008], p = 0.219	−0.013 [−0.035, 0.009], p = 0.236	0.022 [−0.012, 0.056], p = 0.203	0 [−0.002, 0.003], p = 0.804	0 [−0.002, 0.002], p = 0.985
Me-PFOSA-AcOH	0.003 [−0.089, 0.096], p = 0.942	0.093 [−0.012, 0.197], p = 0.084	−0.018 [−0.125, 0.089], p = 0.736	0.134 [−0.031, 0.3], p = 0.112	0.006 [−0.006, 0.018], p = 0.344	−0.006 [−0.015, 0.003], p = 0.189
PFDeA	0 [−0.013, 0.013], p = 0.979	0.007 [−0.008, 0.021], p = 0.386	0.013 [−0.002, 0.028], p = 0.087	0.008 [−−0.016, 0.031], p = 0.517	0.002 [0, 0.004], p = 0.037	0.001 [0, 0.002], p = 0.204
PFHxS	−0.076 [−0.287, 0.134], p = 0.475	−0.184 [−0.423, 0.056], p = 0.133	−0.422 [−0.666, −0.177], p = 0.001	0.156 [−0.222, 0.535], p = 0.417	−0.011 [−0.039, 0.018], p = 0.463	0.013 [−0.008, 0.034], p = 0.222
PFNA	−0.036 [−0.108, 0.036], p = 0.323	−0.04 [−0.121, 0.042], p = 0.339	0.028 [−0.055, 0.111], p = 0.504	0.082 [−0.047, 0.211], p = 0.212	0 [−0.01, 0.009], p = 0.954	0.008 [0.001, 0.015], p = 0.026
PFOA	−1.372 [−2.18, −0.564],p = 0.001	−1.472 [−2.393, −0.552],p = 0.002	−1.016 [−1.955, −0.077],p = 0.034	−0.852 [−2.306, 0.602], p = 0.249	−0.064 [−0.173, 0.044], p = 0.246	0.082 [0.002, 0.161], p = 0.044
PFOS	−0.746 [−1.665, 0.173], p = 0.111	−1.625 [−2.672, −0.579],p = 0.002	0.085 [−0.983, 1.153], p = 0.875	−0.023 [−1.677, 1.63], p = 0.978	0.069 [−0.055, 0.193], p = 0.272	0.094 [0.003, 0.184], p = 0.043
PFOSA	−0.004 [−0.009, 0.001], p = 0.135	−0.003 [−0.009, 0.003], p = 0.282	−0.003 [−0.009, 0.003], p = 0.33	−0.001 [−0.01, 0.009], p = 0.895	0 [−0.001, 0.001], p = 0.597	0 [0, 0.001], p = 0.097
NTA PFAS analytes
PFPeS	0.756 [−0.793, 2.306], p = 0.335	0.306 [−1.79, 2.402], p = 0.773	1.873 [−0.087, 3.834], p = 0.061	1.765 [−1.233, 4.763], p = 0.246	0.172 [−0.044, 0.387], p = 0.117	−0.031 [−0.214, 0.153], p = 0.742
PFHxS	−0.113 [−0.515, 0.289], p = 0.578	−0.369 [−0.913, 0.175], p = 0.181	−0.637 [−1.146, −0.128],p = 0.015	−0.257 [−1.035, 0.521], p = 0.513	−0.029 [−0.085, 0.027], p = 0.305	0.01 [−0.037, 0.058], p = 0.669
b-PFHpS	−0.951 [−2.026, 0.125], p = 0.082	−0.258 [−1.713, 1.196], p = 0.725	−0.309 [−1.67, 1.052], p = 0.653	0.245 [−1.835, 2.325], p = 0.815	−0.01 [−0.159, 0.14], p = 0.897	0.109 [−0.018, 0.237], p = 0.092
PFOS	−0.004 [−0.396, 0.387], p = 0.982	−0.442 [−0.972, 0.089], p = 0.102	−0.429 [−0.925, 0.067], p = 0.089	0.208 [−0.55, 0.967], p = 0.587	−0.03 [−0.085, 0.024], p = 0.276	0.005 [−0.042, 0.051], p = 0.843
PFNS	−0.183 [−3.149, 2.784], p = 0.903	1.534 [−2.479, 5.548], p = 0.45	0.226 [−3.528, 3.979], p = 0.905	−1.185 [−6.923, 4.554], p = 0.683	0.304 [−0.109, 0.717], p = 0.147	−0.149 [−0.5, 0.203], p = 0.403
PFPeA	1.682 [−1.679, 5.043], p = 0.323	0.291 [−4.256, 4.838], p = 0.899	−3.275 [−7.527, 0.978], p = 0.13	−0.949 [−7.451, 5.553], p = 0.773	−0.081 [−0.549, 0.387], p = 0.732	−0.108 [−0.506, 0.29], p = 0.592
PFHxA	0.013 [−0.281, 0.307], p = 0.929	0.273 [−0.124, 0.671], p = 0.176	−0.103 [−0.475, 0.269], p = 0.585	0.084 [−0.485, 0.652], p = 0.771	−0.015 [−0.056, 0.026], p = 0.472	−0.03 [−0.065, 0.005], p = 0.089
PFHpA	−0.461 [−1.241, 0.319], p = 0.243	−0.091 [−1.146, 0.964], p = 0.865	−0.098 [−1.085, 0.889], p = 0.845	−0.292 [−1.801, 1.217], p = 0.702	−0.06 [−0.169, 0.048], p = 0.272	0.016 [−0.076, 0.109], p = 0.724
PFOA	−0.451 [−0.816, −0.087],p = 0.016	−0.541 [−1.034, −0.048],p = 0.032	−0.416 [−0.877, 0.045], p = 0.077	0.004 [−0.702, 0.709], p = 0.992	−0.078 [−0.128, −0.027],p = 0.003	0.008 [−0.036, 0.051], p = 0.727
NTA PFAS analytes
PFDeA	−0.063 [−0.605, 0.478], p = 0.817	0.284 [−0.449, 1.017], p = 0.444	0.232 [−0.454, 0.918], p = 0.504	0.196 [−0.853, 1.244], p = 0.712	0.052 [−0.023, 0.128], p = 0.172	−0.024 [−0.088, 0.04], p = 0.465
Trifluoromethyl dihydrogen phosphate	0.44 [−3.245, 4.126], p = 0.813	−2.728 [−7.714, 2.257], p = 0.28	4.255 [−0.408, 8.918], p = 0.073	3.695 [−3.434, 10.824], p = 0.306	0.163 [−0.35, 0.675], p = 0.531	0.195 [−0.242, 0.631], p = 0.378
Perfluoroethyl dihydrogen phosphate	0.119 [−0.264, 0.502], p = 0.54	0.209 [−0.31, 0.728], p = 0.426	0.336 [−0.149, 0.821], p = 0.173	0.925 [0.184, 1.667], p = 0.015	0.029 [−0.024, 0.082], p = 0.282	0.008 [−0.037, 0.054], p = 0.716
Perfluoropropyl dihydrogen phosphate	−0.628 [−2.204, 0.947], p = 0.431	0.613 [−1.519, 2.744], p = 0.57	− 1.483 [−3.476, 0.511], p = 0.143	1.279 [−1.769, 4.327], p = 0.407	−0.069 [−0.288, 0.15], p = 0.535	−0.023 [−0.21, 0.163], p = 0.804
Perfluorononyl dihydrogen phosphate	0.165 [−0.193, 0.523], p = 0.364	0.333 [−0.151, 0.817], p = 0.176	0.265 [−0.188, 0.718], p = 0.248	0.771 [0.078, 1.463], p = 0.03	0.025 [−0.025, 0.074], p = 0.328	0.008 [−0.034, 0.05], p = 0.707
Perfluoroundecyl dihydrogen phosphate	0.247 [−0.095, 0.589], p = 0.154	0.286 [−0.176, 0.748], p = 0.222	0.172 [−0.261, 0.604], p = 0.433	0.43 [−0.231, 1.091], p = 0.199	0.007 [−0.04, 0.055], p = 0.761	−0.001 [−0.041, 0.04], p = 0.975
2-(perfluoromethyl)-1H-imidazole	−0.418 [−1.324, 0.489], p = 0.363	0.433 [−0.793, 1.659], p = 0.485	−1.205 [−2.352, −0.059],p = 0.04	0.275 [−1.478, 2.028], p = 0.756	−0.094 [−0.22, 0.032], p = 0.142	−0.007 [−0.114, 0.101], p = 0.9
2-(perfluoroethyl)-1H-imidazole	−0.137 [−0.29, 0.016], p = 0.078	−0.184 [−0.391, 0.022], p = 0.08	−0.053 [−0.246, 0.14], p = 0.588	−0.193 [−0.489, 0.102], p = 0.197	−0.021 [−0.042, 0], p = 0.055	0.019 [0.001, 0.037], p = 0.039
2-(perfluoropropyl)-1H-imidazole	−0.182 [−0.437, 0.074], p = 0.162	−0.038 [−0.383, 0.308], p = 0.829	0.081 [−0.242, 0.405], p = 0.619	−0.184 [−0.679, 0.31], p = 0.461	−0.03 [−0.066, 0.006], p = 0.098	0.009 [−0.022, 0.039], p = 0.578
4-(trifluoromethoxy)phenol	−2.349 [−5.572, 0.874], p = 0.151	−2.325 [−6.685, 2.036], p = 0.293	0.907 [−3.171, 4.985], p = 0.66	2.735 [−3.5, 8.971], p = 0.386	−0.084 [−0.533, 0.364], p = 0.71	0.069 [−0.313, 0.451], p = 0.721
4-(perfluoroethoxy)phenol	−0.078 [−1.808, 1.653], p = 0.929	0.172 [−2.169, 2.513], p = 0.884	0.5 [−1.689, 2.69], p = 0.651	2.865 [−0.482, 6.213], p = 0.093	0.009 [−0.232, 0.25], p = 0.94	−0.17 [−0.375, 0.035], p = 0.103
3,3,4,4,5,5,6,6,7,7,7-undecafluoro-1-hydroxyheptane-1-sulfonic acid	0.56 [−0.215, 1.336], p = 0.155	0.907 [−0.142, 1.956], p = 0.089	−0.424 [−1.405, 0.557], p = 0.393	−0.154 [−1.654, 1.346], p = 0.839	0.002 [−0.106, 0.109], p = 0.978	−0.043 [−0.135, 0.049], p = 0.356
3,3,4,4,5,5,6,6,6-nonafluoro-1-hydroxyhexane-1-sulfonic acid	0.981 [−1.29, 3.251], p = 0.393	2.082 [−0.99, 5.153], p = 0.182	−2.028 [−4.901, 0.845], p = 0.164	0.384 [−4.008, 4.776], p = 0.863	−0.032 [−0.348, 0.283], p = 0.839	−0.322 [−0.591, −0.053],p = 0.019
3,3,4,4,5,5,5-heptafluoro-1-hydroxypentane-1-sulfonic acid	−0.263 [−0.733, 0.207], p = 0.27	−0.198 [−0.834, 0.438], p = 0.539	0.265 [−0.33, 0.86], p = 0.379	0.093 [−0.817, 1.003], p = 0.84	0.038 [−0.028, 0.103], p = 0.253	−0.016 [−0.072, 0.039], p = 0.564
1:2 Fluorotelomer sulfonic acid	−0.095 [−0.289, 0.099], p = 0.334	−0.149 [−0.411, 0.114], p = 0.263	−0.043 [−0.289, 0.202], p = 0.727	−0.043 [−0.419, 0.332], p = 0.82	−0.013 [−0.04, 0.014], p = 0.326	0.015 [−0.008, 0.038], p = 0.197
2:2 fluorotelomer sulfonic acid	−0.168 [−0.504, 0.168], p = 0.324	−0.026 [−0.481, 0.429], p = 0.91	−0.035 [−0.461, 0.39], p = 0.87	−0.137 [−0.788, 0.513], p = 0.676	0.033 [−0.014, 0.08], p = 0.163	0.01 [−0.03, 0.05], p = 0.608
3:2 fluorotelomer sulfonic acid	0.116 [−0.217, 0.45], p = 0.49	0.117 [−0.334, 0.569], p = 0.607	0.318 [−0.104, 0.74], p = 0.139	0.322 [−0.324, 0.967], p = 0.325	0.038 [−0.009, 0.084], p = 0.11	0.003 [−0.036, 0.043], p = 0.874
4:2 fluorotelomer sulfonic acid	−0.145 [−0.616, 0.327], p = 0.544	−0.175 [−0.813, 0.463], p = 0.587	0.142 [−0.454, 0.739], p = 0.637	−0.832 [−1.744, 0.08], p = 0.073	0.041 [−0.025, 0.106], p = 0.221	0.002 [−0.054, 0.058], p = 0.947
1:1 FTCA	−0.169 [−0.317, −0.021],p = 0.026	−0.1 [−0.301, 0.1], p = 0.324	−0.105 [−0.292, 0.083], p = 0.271	−0.031 [−0.317, 0.256], p = 0.833	−0.006 [−0.027, 0.014], p = 0.55	0.012 [−0.005, 0.03], p = 0.176
2:1 FTCA	0.493 [−1.556, 2.541], p = 0.634	0.062 [−2.709, 2.833], p = 0.965	−0.747 [−3.339, 1.844], p = 0.568	0.796 [−3.166, 4.758], p = 0.691	−0.143 [−0.428, 0.142], p = 0.32	0.078 [−0.164, 0.321], p = 0.524
3,4,4,4-tetrafluoro-3-(trifluoromethyl)butanoic acid	−0.139 [−2.771, 2.492], p = 0.917	−0.887 [−4.447, 2.674], p = 0.622	2.899 [−0.431, 6.229], p = 0.087	5.328 [0.237, 10.419],p = 0.04	0.223 [−0.143, 0.589], p = 0.23	0.162 [−0.15, 0.474], p = 0.305
2-((N-ethyl-1,1,1-trifluoromethyl)sulfonamido)ethyl (2-((N-ethyl-1,1,2,2,2-pentafluoroethyl)sulfonamido)ethyl) hydrogen phosphate	0.055 [−0.288, 0.398], p = 0.751	−0.115 [−0.579, 0.348], p = 0.622	0.012 [−0.422, 0.445], p = 0.958	−0.795 [−1.458, −0.133],p = 0.019	−0.03 [−0.077, 0.018], p = 0.22	0.036 [−0.004, 0.077], p = 0.078
3-(difluoro(trifluoromethoxy)methoxy)-1,1,2,2,3,3-hexafluoropropane-1-sulfonic acid	0.464 [−1.899, 2.826], p = 0.698	2.705 [−0.491, 5.901], p = 0.096	−4.13 [−7.119, −1.14], p = 0.007	0.341 [−4.229, 4.912], p = 0.883	−0.028 [−0.357, 0.301], p = 0.866	−0.154 [−0.434, 0.126], p = 0.277
NTA PFAS analytes
2,3,3,3-tetrafluoro-1-(2-hydroxyphenyl)propan-1-one	−0.245 [−0.489, −0.001],p = 0.049	−0.1 [−0.43, 0.23], p = 0.548	0.351 [0.042, 0.659], p = 0.026	0.324 [−0.148, 0.796], p = 0.176	0.026 [−0.008, 0.06], p = 0.126	0.006 [−0.023, 0.035], p = 0.69
2-(perfluoroethoxy)ethan-1-ol	−0.195 [−0.439, 0.048], p = 0.115	−0.004 [−0.333, 0.325], p = 0.98	0.32 [0.012, 0.628], p = 0.042	0.37 [−0.101, 0.841], p = 0.122	−0.005 [−0.038, 0.029], p = 0.789	0.016 [−0.013, 0.045], p = 0.277
Dimethyl (trifluoromethyl)phosphonate	−0.269 [−0.592, 0.054], p = 0.102	−0.185 [−0.622, 0.253], p = 0.404	0.183 [−0.226, 0.592], p = 0.376	0.425 [−0.2, 1.05], p = 0.18	0.021 [−0.024, 0.066], p = 0.365	0.004 [−0.034, 0.042], p = 0.832
2-((1,1,2,3,3,4,4,5,5,6,6,6-dodecafluorohexyl)oxy)acetic acid	0.183 [−0.256, 0.622], p = 0.411	0.326 [−0.269, 0.92], p = 0.28	0.02 [−0.536, 0.575], p = 0.944	0.006 [−0.844, 0.855], p = 0.989	−0.021 [−0.082, 0.04], p = 0.49	0.006 [−0.046, 0.058], p = 0.822
4,4,5,5,5-pentafluoropentane-1-thiol	0.008 [−0.053, 0.07], p = 0.787	0.055 [−0.027, 0.138], p = 0.187	−0.08 [−0.157, −0.002],p = 0.043	0.023 [−0.095, 0.142], p = 0.698	0 [−0.009, 0.008], p = 0.967	−0.007 [−0.014, 0], p = 0.056
5-chloro-2,2,3,3,4-pentafluoropent-4-enoic acid	0.761 [−0.03, 1.552], p = 0.059	0.646 [−0.424, 1.716], p = 0.234	−0.538 [−1.539, 0.463], p = 0.289	−1.503 [−3.033, 0.028], p = 0.054	0.094 [−0.016, 0.204], p = 0.092	−0.079 [−0.173, 0.014], p = 0.096
2,2,2-trifluoroacetamide	−0.133 [−0.274, 0.008], p = 0.063	−0.021 [−0.211, 0.17], p = 0.831	−0.065 [−0.243, 0.113], p = 0.471	0.058 [−0.214, 0.33], p = 0.675	0.014 [−0.005, 0.034], p = 0.151	−0.003 [−0.02, 0.014], p = 0.708
1-(1,1,2,2-tetrafluoroethyl) cyclohexane-1-carboxylic acid	0.214 [−0.099, 0.528], p = 0.178	0.475 [0.05, 0.9], p = 0.029	−0.091 [—0.488, 0.306], p = 0.649	0.039 [−0.568, 0.646], p = 0.899	0.04 [−0.004, 0.083], p = 0.075	−0.034 [−0.072, 0.003], p = 0.07
1:2 FTCA	0.011 [−0.169, 0.192], p = 0.901	−0.002 [−0.246, 0.242], p = 0.987	0.026 [−0.202, 0.255], p = 0.819	−0.011 [−0.36, 0.338], p = 0.951	0.019 [−0.006, 0.044], p = 0.142	0.005 [−0.017, 0.026], p = 0.653
3,4,4,4-tetrafluorobut-2-enoic acid	−0.27 [−0.602, 0.062], p = 0.11	−0.104 [−0.553, 0.346], p = 0.648	−0.015 [−0.435, 0.405], p = 0.944	0.197 [−0.445, 0.839], p = 0.544	−0.011 [−0.057, 0.035], p = 0.635	0.02 [−0.02, 0.059], p = 0.325

Sensitivity Analysis for PFAS-Omics Score Calculations

As a sensitivity analysis, we also fitted IRT models from the 25 nontargeted PFAS analytes that had h2 values greater than 0.1. (Table S14 and Table S15) PFAS exposure differences across parity using the PFAS-omics score calculated from 25 NTA PFAS analytes were consistent with those for PFAS-omics score calculated from 42 NTA PFAS analytes. PFAS-omics score of 25 NTA PFAS analytes was not different across the parity groups (Wilcoxon Rank Sum test p = 0.806; robust Cohen’s d = 0.22, 95% CI: −0.25, 0.73 for multiparous vs nulliparous; robust Cohen’s d = −0.12, 95% CI: −0.63, 0.37 for parity primiparous vs nulliparous).

Secondary Analysis of Accounting for Measurement Errors in the IRT Model

Difference in cord blood PFAS burden across parity groups persisted in the secondary analysis accounting for measurement errors of the IRT model. PFAS burden scores were different across the parity groups (Kruskal–Wallis rank sum test p < 0.05 in 977 of the 1000 plausible values) and were lower in the primiparous (robust Cohen’s d ranged from −0.67 to −0.14 in 1000 resamples) and multiparous (Cohen’s d ranged from −0.86 to −0.27 in 1000 resamples) compared with the nulliparous. (Figure S5) No difference was detected for PFAS-omics score in any of the 1000 resamples.

Exploratory Secondary Analysis of Prior Breastfeeding Duration and PFAS Scores

Among parous women, n = 116 had targeted PFAS data and breastfeeding duration (in weeks) reported; n = 61 had NTA PFAS data and breastfeeding duration reported. Only the targeted PFAS burden score was associated with breastfeeding duration among parous women, with a Spearman correlation of −0.326 (p < 0.001). (Table S16)

Our findings demonstrate the application of exploratory IRT methods for NTA PFAS data to construct PFAS-omics scores. Cord serum was found to have more PFAS beyond the compounds routinely measured in targeted chemical analyzes. NTA allowed us to detect additional putatively identified PFAS that are rarely targeted due to lack of standards or access to sufficiently sensitive instrumentation. NTA, which postulates unknown compounds without suspect lists, includes suspect screening analysis (SSA), which compares molecular features against chemical databases.⁸⁷

In one prior cord blood SSA study where PFAS-focused data analysis workflows were not used, detected PFAS were limited to only legacy compounds, including PFHxS, PFOS, PFDeA, PFUnA, and PFNA.⁸⁸ SSA that focused on PFAS identified or putatively identified a broader range of PFAS in cord sera, including unsaturated perfluoroalkyl alcohols, H-substituted perfluorocarboxylic acids, H-substituted perfluorinated sulfonic acids, perfluorononenyl benzenesulfonate, ether-perfluorosulfonic acids, Cl-substituted perfluoroether sulfonic acids, Cl-substituted perfluorinated carboxylic acids, ketone-perfluorosulfonic acids, fluorotelomer sulfonates, fluorotelomeric carboxylates, and unsaturated perfluorocarboxylic acids.⁸⁹ Several of these replacement PFAS were detected in this study, including fluorotelomer sulfonates and fluorotelomeric carboxylates. Because suspect screening approaches are limited to a predetermined list of candidates,⁸⁷ a PFAS-focused NTA approach expands the range of PFAS that can potentially be detected.

Importantly, many of the putatively identified PFAS in this study have not been well characterized in human biomonitoring studies and do not have commercially available analytical standards. As a result, their structural elucidation is challenging, and toxicological profiles and health implications remain largely unknown. Further investigation of these compounds is needed to confirm their identities, establish their prevalence, and understand their relevance as potential biomarkers of exposure, as well as to assess their potential health effects and inform future risk assessments. Our findings underscore using nontargeted approaches in exposomics to identify previously unknown or replacement PFAS, and the need for ongoing research to characterize these newly discovered chemicals, including developing analytical standards and conducting targeted studies to assess their impact on human health.

Several studies have reported higher PFAS in nulliparous women compared with multiparous women.^41,90,91 Studies have found disparities in PFAS burden by parity but have primarily focused on targeted PFAS data that tends to assay PFAS with analytical standards readily available or those included in the analytical methods published by government agencies (such as EPA Method 1633 and CDC 6304.09). While a consistent link has been reported between parity and lower serum concentrations of PFOA,^92–105 PFOS,^{92–99,101,102,104–106} PFNA^{92–97,99,101,102,104,105} and PFHxS,^{92–101,104,105} we did not identify significant disparities between overall PFAS-omics scores and parity. In our study sample, infants born to mothers who were primiparous or multiparous had lower concentrations of PFHxS, PFOA, PFOS, PFAS burden in the targeted analysis, and lower relative abundance of PFOA in the NTA, compared to infants born to mothers who were nulliparous. Studies have suggested that breastfeeding may be an important factor of eliminating PFAS.^{101,107–109} However, our exploratory secondary analysis did not find associations between prior breastfeeding duration and PFAS-omics scores; only the targeted PFAS burden score was associated with breastfeeding duration among parous women.

Targeted PFAS analytes were not associated with the PFAS-omics scores. While many PFAS had higher h2 levels, indicating a stronger “contribution” to the PFAS-omics score, there were several NTA PFAS with low contribution to the PFAS-omics scores because they were uncorrelated with the relative abundance of many other PFAS. Further research is needed to investigate how to optimally include PFAS that may be uncorrelated with the other PFAS into scoring. Further investigation is also needed to understand whether PFAS subscores are needed, and whether subscores should be created based on chemical structure similarities, or exposure source similarities.¹¹⁰ Here, we created PFAS-omics scores as an average of three correlated factor subscores. As an exploratory secondary analysis, we further explored the individual factor loadings to better characterize the factors and what they potentially represent. Multidimensional IRT allows PFAS features to load onto more than one factor. To characterize the individual factors, we examined the features with a larger proportion of variance explained by the latent factor (e.g., h2 > 0.2). Hence the factors appear to reflect different chemical groupings, as follows: Factor 1: Mixed-class PFAS containing sulfonic, carboxylic, and phosphoric acid functionalities; Factor 2: Fluorotelomer- and FTCA-dominated PFAS with primarily carboxylic and heterocyclic structures; Factor 3: Functionally diverse PFAS, including alcohols, sulfonamides, thiols, and ethers. While these groupings do not directly map onto distinct exposure sources or biological effects, they provide some insight into the chemical diversity represented by each factor. Future work is needed to potentially link these latent dimensions more directly to exposure sources or health outcomes.

This study has some limitations. First, the cord serum targeted and NTA PFAS analyzes were performed in different laboratories, which have different mass spectrometers and used different analytical conditions. We were unable to adjust for all potential differences in laboratory procedures. For example, the CDC lab used in-line solid-phase extraction (SPE), which was not available at the NTA laboratory and could have reduced matrix interference. The analysis is comparable across different laboratories for these specific compounds as targeted and NTA concentrations/abundances of PFOA and PFOS were still correlated. However, since this evaluation was limited to two compounds and did not address the greater variability inherent in nontargeted analyzes, further work is needed before generalizing laboratory comparability to nontargeted data. Additionally, while our batch correction procedure addressed run-to-run differences within the lab where NTA was performed, batch correction does not account for sample-to-sample variability such as matrix effects. The IRT-based approach in this study is specific to human serum, and future applications should consider strategies to better quantify and address these sources of variability and potential interlaboratory differences. The overall burden score, estimated using both individual PFAS components as well as exposure patterns to the PFAS mixture, can be estimated with less error compared with using the individual components of the PFAS mixture.³⁷

Second, our PFAS-omics score, which included PFAS of varying chain lengths and varying half-lives, did not adjust for persistence of PFAS. Thus, the PFAS-omics scores can be interpreted as a snapshot in time of the overall PFAS levels. Our comparison of disparities in the population should be limited to the study sample, and further research is needed to understand national level disparities. These findings may not be generalizable because the HOME Study sample was recruited from participants in the greater Cincinnati area (2003–2006) known to have PFAS (particularly PFOA) contamination of public drinking water supplies due to industrial pollution from the DuPont Washington Works plant located 250 miles upstream of Cincinnati, Ohio.^45,46

In conclusion, we found 42 PFAS in cord blood using a NTA approach, many of these PFAS are not commonly screened in targeted panels. Because analytical standards are not available for purchase for most of the 10,000+ compounds on the EPA Master List, we constructed burden scores using IRT to summarize and quantify total exposure to PFAS before birth. The two scores were the “PFAS exposure burden score,” based on PFAS concentrations from targeted analysis, and the “PFAS-omics scores”, based on relative abundance from NTA. Infants with primiparous and multiparous mothers had significantly lower PFAS exposure burden scores than infants with nulliparous mothers, but these differences were null when comparing their PFAS-omics scores. Future studies should consider total PFAS exposure burden to better understand the potential health impacts of PFAS, and further investigate the putatively identified PFAS from this study, particularly those without available analytical standards, to determine their relevance as biomarkers and potential effects on human health.

Supplementary Material

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.est.5c06490.

Figure S1. flowchart of the study sample Figure S2. flowchart of the study goals Table S1: Lower limits of detection (LLODs) of the 8 targeted PFAS analytes Table S2: instrument parameters used for LC-HRMS with negative ESI MS/MS Spectra Study Reporting Figure S3. Box-and-whisker plot of the PFAS detected in NTA. relative abundances of the NTA PFAS were log2-transformed and batch corrected Table S3: decile cutoffs of the 8 PFAS analytes in the targeted analysis Table S4: item parameters of the unidimensional IRT model for calculating PFAS burden Table S5: quartile abundance cutoffs for 42 PFAS analytes in the NTA Table S6: explore the number of dimensions of IRT model to quantify PFAS-omics score from 42 PFAS analytes in NTA Table S7: oblimin-rotated coefficients of the IRT model calculating PFAS-omics score Table S8: oblimin-rotated factor loadings of the IRT model calculating PFAS-omics score Table S9: item fits of the IRT model calculating PFAS-omics score Table S10: covariances of the 3 latent factors Figure S4: Heatmap of the exposure pattern of the 42 NTA PFAS in the sample Table S11: median (IQR) of PFAS analytes across parity groups for participants both having targeted and nontargeted PFAS measurements Table S12: robust Cohen’s d (95% CI) for comparing burden scores across parity groups and for comparisons that are significantly different across parity groups Table S13: robust Cohen’s d (95% CI) of comparisons in the sample both having targeted and nontargeted PFAS analytes Table S14: oblimin-Rotated coefficients of the IRT model calculating PFAS-omics score from 25 NTA PFAS that had h2 values > 0.1 Table S15: oblimin-Rotated factor loadings of the IRT model calculating PFAS-omics score from 25 NTA PFAS that had h2 values > 0.1 Table S16: Spearman correlation of breastfeeding duration and PFAS burden scores Figure S5: comparing exposure to PFAS mixtures using 1000 plausible values Table S16: Spearman correlation of breastfeeding duration and PFAS burden scores (PDF)