Maternal serum concentrations of per- and polyfluoroalkyl substances and childhood acute lymphoblastic leukemia

Abstract

Background

Per- and polyfluoroalkyl substances (PFAS) are widespread and environmentally persistent chemicals with immunotoxic properties. Children are prenatally exposed through maternal transfer of PFAS to cord blood, but no studies have investigated the relationship with childhood leukemia.

Methods

We measured maternal serum levels of 19 PFAS in first-trimester samples collected in 1986-2010 and evaluated associations with acute lymphoblastic leukemia in full-term offspring (aged younger than 15 years) for 400 cases and 400 controls in the Finnish Maternity Cohort, matched on sample year, mother’s age, gestational age, birth order, and child’s sex. We analyzed continuous and categorical exposures, estimating odds ratios (ORs) and 95% confidence intervals (CIs) via conditional logistic regression adjusted for maternal smoking and correlated PFAS (ρ ≥ ±0.3). We also stratified by calendar period, mean diagnosis age, and the child’s sex.

Results

N-methyl-perfluorooctane sulfonamidoacetic acid was associated with acute lymphoblastic leukemia in continuous models (per each doubling in levels: OR_perlog2 = 1.22, 95% CI = 1.07 to 1.39), with a positive exposure-response across categories (OR_{>90th percentile} = 2.52, 95% CI = 1.33 to 4.78; P_trend = .01). Although we found no relationship with perfluorooctane sulfonic acid overall, an association was observed in samples collected in 1986-1995, when levels were highest (median = 17.9 µg/L; OR_perlog2 = 4.01, 95% CI = 1.62 to 9.93). A positive association with perfluorononanoic acid was suggested among first births (P_interaction = .06). The N-methyl-perfluorooctane sulfonamidoacetic acid association was mainly limited to children diagnosed before age 5 years (P_interaction = .02). We found no consistent patterns of association with other PFAS or differences by sex.

Conclusions

These novel data offer evidence of a relationship between some PFAS and risk of the most common childhood cancer worldwide, including associations with the highest levels of perfluorooctanesulfonic acid and with a precursor, N-methyl-perfluorooctane sulfonamidoacetic acid.

Per- and polyfluoroalkyl substances (PFAS) are man-made chemicals widely used since the 1940s in industrial and commercial applications, including electronics, textiles, nonstick cookware, and fire retardants (1). PFAS are water soluble, and many are resistant to degradation in the environment. Humans are exposed primarily from contaminated food, drinking water, house dust, and occupational exposures. The half-lives of PFAS in serum vary, with estimated ranges of 3-8 years (2,3). Use of one of the most prevalent long-chained PFAS, perfluorooctanesulfonic acid, was restricted in the European Union in 2009; perfluorooctanoic acid was more recently banned in 2020 (4). Although published data for serum per- and polyfluoroalkyl substance levels in the Finnish population are limited (5), levels of these 2 legacy compounds have declined in Swedish mothers (6). Corresponding increases in other long-chained PFAS (eg, perfluorononanoic acid, perfluorodecanoic acid, and perfluoroundecanoic acid) have been observed in Norwegian women of reproductive age (7), although temporal trends have varied across European populations (8,9).

Children experience greater per- and polyfluoroalkyl substance serum burdens than their mothers and other adults (10,11). Prenatal and postnatal exposure to PFAS occur through cord blood (12) and breast milk (13), respectively, thus, parity and duration of breastfeeding are key determinants of maternal levels (14-16). The Norwegian Mother, Father and Child Cohort Study identified other factors associated with higher maternal levels including diet (eg, high seafood consumption), older age, higher education and income, and urbanicity, although each explained only a small percent of the variance (eg, 4% for diet) (16). Serum PFAS concentrations in Nordic women are somewhat lower than in American women (17), but exposure is similarly ubiquitous, with 100% detection rates for the most abundant PFAS in the Finnish population in 2007 (18).

The International Agency for Research on Cancer recently changed the classification of perfluorooctanoic acid, determining it to be a human carcinogen on the basis of strong mechanistic data, sufficient evidence in experimental animals, and limited human evidence (19). Among the newer human evidence is that perfluorooctanoic acid may increase the risk of non-Hodgkin lymphoma (20,21) and kidney (22) and breast (23,24) cancers in adults. Perfluorooctanesulfonic acid is newly classified as a possible human carcinogen (19) and evidence of adult cancer associations for other PFAS has also been accumulating, but the potential cancer risks in children are not known.

Immune dysregulation in utero and in early childhood is a hypothesized mechanism for the development of leukemia (25), the most commonly occurring childhood cancer (26). The US National Toxicology Program concluded that perfluorooctanesulfonic acid and perfluorooctanoic acid are immunotoxic on the basis of antibody suppression in animal and human (including children) studies (27). In the Mother, Father and Child Cohort Study, higher prenatal serum concentrations of 4 common PFAS were associated with lower levels of antirubella antibodies after vaccination (28). Established food and water intake levels for perfluorooctanesulfonic acid in Europe are based in part on immunotoxicity in children (29).

Despite observed cancer risk from PFAS in adult populations and biological plausibility, there are no epidemiologic studies assessing the relationship between in utero PFAS exposure and childhood cancers. We evaluated associations between PFAS and childhood acute lymphoblastic leukemia (ALL), the predominant leukemia subtype in children, within a large, nationally representative cohort of women and their children in Finland.

Methods

Study population

The Finnish Maternity Cohort (FMC) is a population-based study including serum samples from approximately 2 million pregnancies (30), covering more than 90% of all pregnant Finnish women in 1983-2016. Blood collection began as a nationwide effort organized by the Finnish Institute for Health and Welfare. After registration of each new pregnancy (usually at 10-14 weeks’ gestation), a blood sample was drawn at municipal maternity care units for routine screening tests; remaining sera were stored as 1 aliquot at −25°C in a biorepository at Biobank Borealis, Oulu, Finland. Informed consent operated via an opt-out principle.

Finland has personal identification numbers that enable follow-up and linkages with nationwide registries, including the Finnish Population Registry for emigration, vital status, biological parents and siblings, and the Medical Birth Registry for the child’s gestational age, delivery complications, birth weight, and sex, and maternal information on parity, body mass index (BMI), and smoking status during pregnancy. Cancer cases were identified by linking the child’s personal identification number to the nationwide Finnish Cancer Registry.

Case and control selection

We limited the cohort from which cases and controls were sampled for a nested case-control analysis to the offspring of mothers with no history of cancer and whose first or second pregnancies resulted in a full-term (37-42 week), singleton live birth, and with adequate serum (400 µL). We excluded children born with Down syndrome (International Classification of Diseases version 10 Q90), an established risk factor for childhood leukemia. Because maternal PFAS concentrations decrease with parity via their transfer to the fetus (16,31), we selected all cases of ALL among first live births born to women with adequate sera (n = 243) and sampled their matched controls (n = 243), then randomly sampled the rest of the cases and controls from second births (n = 157 cases; 157 controls). Cases were diagnosed between ages 0 and 14 years from children born between 1987 and 2010 whose mothers were ages 18-39 years at sample collection (1986-2010), because the Medical Birth Registry includes essential information about the pregnancy from 1987 onward. We matched cases and controls by calendar year of sample collection (approximately 5-year increments), mother’s age (approximately 3-year increments), birth order, gestational age at birth (within 2 weeks), and child’s sex. Approval to use the FMC, cancer registry, and population data for this study was granted from the Finnish Institute for Health and Welfare.

Laboratory assays

PFAS levels were measured by the Department for Health Security at the Finnish Institute of Health and Welfare Environmental Health Unit laboratory using 200 µL of serum; matched cases and controls were analyzed in the same batch. We quantitated 19 PFAS: perfluorooctanoic acid, perfluorooctanesulfonic acid, perfluorononanoic acid, perfluorodecanoic acid, perfluoroundecanoic acid, perfluorohexane sulfonic acid, N-methyl-perfluorooctane sulfonamidoacetic acid, N-ethyl-perfluorooctane sulfonamidoacetic acid, perfluoroheptanoic acid, perfluoroheptane sulfonic acid, perfluorotetradecanoic acid, 6:2 polyfluoroalkyl phosphoric acid diesters, perfluorohexanoic acid, perfluorodecane sulfonic acid, N-methyl-perfluorooctane sulfonamide, perfluorododecanoic acid, perfluorotridecanoic acid, perfluorooctane sulfonamide, and N-ethyl-perfluorooctane sulfonamide. We separately quantified 15 other persistent organic pollutants in serum and evaluated them as potential confounders. The laboratory assays are further detailed in the Supplementary Material (Supplementary Methods; Supplementary Table 1, available online).

Statistical analysis

Assay values below the limit of quantitation were imputed between zero and the limit of quantitation using the lognormal distribution of quantified values among controls (32). For PFAS with at least 50% detections in controls (perfluorooctanoic acid = 100%; perfluorooctanesulfonic acid  = 99.8%; perfluorononanoic acid = 87%; perfluorodecanoic acid = 64%; perfluorohexane sulfonic acid = 81.5%; N-methyl-perfluorooctane sulfonamidoacetic acid  = 66.8%; N-ethyl-perfluorooctane sulfonamidoacetic acid  = 58%; Supplementary Table 2, available online), we evaluated exposures continuously (log₂-transformed) and categorically in quartiles (Q) and with an additional split of Q4 at the 90th percentile (>75th-90th, >90th) based on levels in controls. The referent group for categorical analyses of perfluorooctanoic acid, perfluorooctanesulfonic acid, perfluorononanoic acid, and perfluorohexane sulfonic acid (all detected >70%) included Q1. For perfluorodecanoic acid, N-methyl-perfluorooctane sulfonamidoacetic acid, and N-ethyl-perfluorooctane sulfonamidoacetic acid (detection proportions between 50% and 70%), we evaluated associations with levels above the limit of quantitation in categories compared with the group with levels below the limit of quantitation. For the remaining PFAS with detection proportions below 50% and a minimum of 5 cases with quantifiable levels, we estimated associations in the group with quantifiable levels compared with those below the limit of quantitation. We estimated odds ratios (ORs) and 95% confidence intervals (CIs) using conditional logistic regression. Models implicitly controlled for matching factors and were adjusted for maternal smoking status (did not smoke, quit in first trimester, smoked after the first trimester, unknown or missing). BMI was not included in models because of a high proportion (>50%) of missing values. We estimated individual PFAS associations mutually adjusted for correlated PFAS (Spearman ρ > ±0.30). We stratified models by sex, calendar year (1986-1995, 1996-2000, 2001-2005, 2006-2010), birth order (first or second live birth), and mean age at case diagnosis (younger than 5 years and 5 years and older). We also conducted analyses of perfluorooctanoic acid and perfluorooctanesulfonic acid tertiles cross-classified with the perfluorooctanesulfonic acid precursor N-methyl-perfluorooctane sulfonamidoacetic acid using a common referent group reflecting the lowest exposed category of each chemical (tertile 1 for perfluorooctanoic acid and perfluorooctanesulfonic acid and below the limit of quantitation for N-methyl-perfluorooctane sulfonamidoacetic acid). We evaluated interactions and linear trends (parameterizing category medians as continuous variables) with Wald statistics. Statistical analyses were conducted in SAS (v 9.4), and tests were 2-sided with a threshold for statistical significance set at an α  of  0.05.

Results

Distributions of child’s sex, sample year, birth order, and maternal age were the same for ALL cases and controls (Table 1), as expected from matching. Mean age was similar between cases and controls. Approximately 57% of the children were male, and by design, the majority (61%) were from a first live birth. The median age group of the mothers was 27-29 years, and most did not smoke. PFAS correlations across the full study period were weak to moderate, including between perfluorooctanesulfonic acid and perfluorooctanoic acid (ρ = 0.54) and between perfluorooctanesulfonic acid and N-methyl-perfluorooctane sulfonamidoacetic acid (ρ = 0.42) and N-ethyl-perfluorooctane sulfonamidoacetic acid (ρ = 0.64; Supplementary Table 3, available online).

Table 1.

Characteristics of cases and controls in the nested case-control study of maternal PFAS and childhood acute lymphoblastic leukemia PFAS in the Finnish Maternity Cohort, 1986-2010

Characteristic	Acute lymphoblastic leukemia cases, No. (%)	Controls, No. (%)
	(n = 400)a	(n = 400)a
Child’s age at diagnosis, mean (SD), y	5.4 (3.5)	5.6 (3.9)
Child’s sex
Male	227 (56.8)	227 (56.8)
Female	173 (43.3)	173 (43.3)
Sample year
1986-1990	22 (5.5)	22 (5.5)
1991-1995	88 (22.0)	88 (22.0)
1996-2000	100 (25.0)	100 (25.0)
2001-2005	109 (27.3	109 (27.3
2006-2010	81 (20.3)	81 (20.3)
Birth order
First live birth	243 (60.8)	243 (60.8)
Second live birth	157 (39.3)	157 (39.3)
Maternal age at sample collection, y
18-20	26 (6.5)	26 (6.5)
21-23	58 (14.5)	58 (14.5)
24-26	78 (19.5)	78 (19.5)
27-29	100 (25.0)	100 (25.0)
30-32	77 (19.3)	77 (19.3)
33-35	38 (9.5)	38 (9.5)
36-39	23 (5.8)	23 (5.8)
Maternal smoking during pregnancyb
Did not smoke	324 (81.2)	322 (80.5)
Quit during first trimester	13 (3.3)	14 (3.5)
Smoked after first trimester	53 (13.3)	54 (13.5)
Unknown or missing	10 (2.5)	10 (2.5)
Maternal body mass index, mean (SD), kg/m2	23.9 (4.5)	23.4 (4.5)
Missing	266 (66.5)	279 (69.8)

a

Groups may not sum to 100% because of rounding. PFAS = per- and polyfluoroalkyl substances.

b

P = .99 from χ² test.

In models adjusted for smoking status, continuous perfluorononanoic acid and perfluorodecanoic acid levels showed suggestive associations with childhood ALL (OR_perlog2 = 1.06, 95% CI = 0.94 to 1.19, and OR = 1.09, 95% CI = 0.97 to 1.22, respectively; Table 2). Odds ratios for both were nonmonotonically elevated across most exposure categories, and these patterns were largely unchanged with adjustment for correlated PFAS. Risk of ALL was increased in smoking-adjusted models with a doubling of N-methyl-perfluorooctane sulfonamidoacetic acid levels (OR_perlog2 = 1.18, 95% CI = 1.04 to 1.34) and was more than twofold higher in the group with higher than 90th percentile of exposure vs below the limit of quantitation (OR = 2.39, 95% CI = 1.29 to 4.43), with a significant trend (P_trend = .01). These associations were not attenuated following adjustment for correlated PFAS, including perfluorooctanesulfonic acid, perfluorohexane sulfonic acid, and N-ethyl-perfluorooctane sulfonamidoacetic acid (>90th percentile OR = 2.52, 95% CI = 1.33 to 4.78; P_trend = .01). N-ethyl-perfluorooctane sulfonamidoacetic acid was inversely associated with ALL in fully adjusted models, without trend across categories (P_trend = .17). We observed no clear patterns in associations with perfluorooctanoic acid, perfluorooctanesulfonic acid, or perfluorohexane sulfonic acid. In cross-classified analyses, the N-methyl-perfluorooctane sulfonamidoacetic acid association was strongest among those jointly exposed to the highest tertile of N-methyl-perfluorooctane sulfonamidoacetic acid and the lowest tertile of perfluorooctanoic acid (OR = 2.69, 95% CI = 1.21 to 6.00; P_interaction = .06); a similar pattern was evident for perfluorooctanesulfonic acid (P_interaction = 0.75; Supplementary Table 4, available online). We observed a suggestive increase in ALL risk with detectable vs not detectable perfluorohexane sulfonic acid (OR = 1.68, 95% CI = 0.88 to 3.19) but no associations with detectable perfluoroheptanoic acid, perfluorohexane sulfonic acid, perfluoroundecanoic acid, or perfluorotetradecanoic acid (Supplementary Table 5, available online).

Table 2.

Odds ratios and 95% confidence intervals for associations between PFAS in maternal serum and risk of childhood acute lymphoblastic leukemia in the Finnish Maternity Cohort

PFAS	µg/L	Cases, No.	Controls, No.	OR (95% CI)b	Ptrend	OR (95% CI)c	Ptrend
Perfluorooctanoic acid	<2.32	100	100	1.00 (referent)	.55	1.00 (referent)	.28
	≥2.32-3.17	107	100	1.06 (0.71 to 1.59)		0.98 (0.65 to 1.49)
	>3.17-4.21	106	100	1.04 (0.68 to 1.60)		0.94 (0.58 to 1.52)
	>4.21-5.32	48	60	0.76 (0.45 to 1.29)		0.68 (0.38 to 1.23)
	>5.32	39	40	0.93 (0.54 to 1.62)		0.77 (0.40 to 1.47)
	Continuousa	400	400	1.12 (0.90 to 1.39)		1.11 (0.85 to 1.45)
Perfluorooctanesulfonic acid	<7.16	113	100	1.00 (referent)	.91	1.00 (referent)	.82
	≥7.16-11.83	94	100	0.76 (0.49 to 1.18)		0.63 (0.39 to 1.02)
	>11.83-17.36	87	100	0.67 (0.40 to 1.12)		0.58 (0.33 to 1.02)
	>17.36-24.58	61	60	0.79 (0.44 to 1.41)		0.63 (0.31 to 1.26)
	>24.58	45	40	0.89 (0.46 to 1.73)		0.80 (0.36 to 1.81)
	Continuousa	400	400	1.01 (0.82 to 1.25)		0.96 (0.73 to 1.26)
Perfluorononanoic acid	<0.24	85	100	1.00 (referent)	.62	1.00 (referent)	.71
	≥0.24-0.42	104	100	1.29 (0.84 to 2.04)		1.23 (0.78 to 1.93)
	0.42-0.68	108	100	1.36 (0.87 to 2.10)		1.25 (0.77 to 2.01)
	>0.68-1.02	67	60	1.45 (0.86 to 2.45)		1.25 (0.69 to 2.27)
	>1.02	36	40	1.14 (0.61 to 2.11)		0.89 (0.42 to 1.88)
	Continuousa	400	400	1.06 (0.94 to 1.19)		1.02 (0.83 to 1.24)
Perfluorodecanoic acid	Below the limit of quantitation	129	144	1.00 (referent)	.25	1.00 (referent)	.49
	≥0.15-0.20	54	64	0.97 (0.62 to 1.53)		0.99 (0.62 to 1.58)
	>0.20-0.27	79	64	1.51 (0.96 to 2.37)		1.48 (0.91 to 2.42)
	>0.27-0.38	67	64	1.31 (0.81 to 2.10)		1.27 (0.75 to 2.14)
	>0.38-0.54	46	39	1.48 (0.86 to 2.56)		1.43 (0.76 to 2.70)
	>0.54	25	25	1.25 (0.64 to 2.45)		1.23 (0.57 to 2.66)
	Continuousa	400	400	1.09 (0.97 to 1.22)		1.12 (0.90 to 1.39)
Perfluorohexane sulfonic acid	<0.28	108	100	1.00 (referent)	.86	1.00 (referent)	.74
	≥0.28-0.47	101	100	0.91 (0.61 to 1.37)		0.89 (0.58 to 1.35)
	>0.47-0.69	88	100	0.80 (0.53 to 1.21)		0.77 (0.50 to 1.18)
	>0.69-0.97	59	60	0.89 (0.57 to 1.40)		0.86 (0.53 to 1.38)
	>0.97	44	40	1.00 (0.48 to 1.71)		0.95 (0.54 to 1.66)
	Continuousa	400	400	1.04 (0.93 to 1.15)		1.01 (0.87 to 1.18)
N-methyl-perfluorooctane sulfonamidoacetic acid	Below the limit of quantitation	117	133	1.00 (referent)	.01	1.00 (referent)	.01
	≥0.15-0.26	62	67	1.20 (0.73 to 2.00)		1.25 (0.75 to 2.06)
	>0.26-0.40	73	67	1.47 (0.91 to 2.37)		1.50 (0.91 to 2.45)
	>0.40-0.64	58	67	1.23 (0.73 to 2.09)		1.28 (0.75 to 2.19)
	>0.64-0.94	44	40	1.51 (0.85 to 2.69)		1.65 (0.90 to 3.02)
	>0.94	46	26	2.39 (1.29 to 4.43)		2.52 (1.33 to 4.78)
	Continuousa	400	400	1.18 (1.04 to 1.34)		1.22 (1.07 to 1.39)
N-ethyl-perfluorooctane sulfonamidoacetic acid	Below the limit of quantitation	184	168	1.00 (referent)	0.26		.17
	≥0.15-0.33	63	58	0.66 (0.37 to 1.18)		0.58 (0.31 to 1.07)
	>0.33-0.61	52	58	0.49 (0.25 to 0.94)		0.43 (0.21 to 0.87)
	>0.61-1.05	52	58	0.47 (0.24 to 0.92)		0.39 (0.19 to 0.80)
	>1.05-2.18	26	35	0.37 (0.17 to 0.81)		0.30 (0.13 to 0.70)
	>2.18	23	23	0.51 (0.22 to 1.16)		0.39 (0.16 to 0.97)
	Continuousa	400	400	0.87 (0.78 to 0.98)		0.87 (0.78 to 0.97)

a

Continuous models reflect association per 1-unit increase in serum PFAS on the log-base 2 scale, corresponding to an approximate doubling in analyte levels. CI = confidence interval; OR = odds ratio; PFAS = per- and polyfluoroalkyl substances.

b

Adjusted for smoking status during pregnancy (did not smoke, quit in first trimester, smoked after first trimester, missing or unknown).

c

Adjusted for smoking status and mutually adjusted for correlated PFAS based on Spearman ρ ≥ ±0.30: perfluorooctanoic acid models are additionally adjusted for perfluorooctanesulfonic acid, perfluorohexane sulfonic acid, N-methyl-perfluorooctane sulfonamidoacetic acid, and N-ethyl-perfluorooctane sulfonamidoacetic acid. Perfluorooctanesulfonic acid models are additionally adjusted for perfluorooctanoic acid, perfluorohexane sulfonic acid, N-methyl-perfluorooctane sulfonamidoacetic acid, and N-ethyl-perfluorooctane sulfonamidoacetic acid. Perfluorononanoic acid models are additionally adjusted for perfluorodecanoic acid and N-ethyl-perfluorooctane sulfonamidoacetic acid. Perfluorodecanoic acid models are additionally adjusted for perfluorononanoic acid and N-ethyl-perfluorooctane sulfonamidoacetic acid. Perfluorohexane sulfonic acid models are additionally adjusted for perfluorooctanoic acid and perfluorooctanesulfonic acid. N-methyl-perfluorooctane sulfonamidoacetic acid models are additionally adjusted for perfluorooctanesulfonic acid, perfluorooctanoic acid, and N-ethyl-perfluorooctane sulfonamidoacetic acid. N-ethyl-perfluorooctane sulfonamidoacetic acid models are additionally adjusted for perfluorooctanoic acid, perfluorooctanesulfonic acid, perfluorononanoic acid, perfluorodecanoic acid, and N-methyl-perfluorooctane sulfonamidoacetic acid.

Median levels of perfluorooctanesulfonic acid and perfluorooctanoic acid were highest in the earliest sample years and declined over time, although for perfluorooctanoic acid more modestly; rates of detection of N-methyl-perfluorooctane sulfonamidoacetic acid and N-ethyl-perfluorooctane sulfonamidoacetic acid also declined with calendar period. Perfluorononanoic acid levels increased over time (Table 3). Consistent with these secular trends in levels and detection proportions, the correlations between PFAS also varied over time (Supplementary Table 3, available online). Serum perfluorooctanesulfonic acid levels were more than 70% higher in samples collected in 1986-1995 (median = 17.90 µg/L) vs 2006-2010 (median = 4.79 µg/L). In analyses by sample year, a strong positive association with perfluorooctanesulfonic acid was observed during the 1986-1995 period (OR_perlog2 = 4.01, 95% CI = 1.62 to 9.93; Table 3) that was not apparent overall. A suggestive association with perfluorononanoic acid was observed in later years (2006-2010) when levels were at their highest, although the estimate was imprecise (OR = 1.64, 95% CI = 0.86 to 3.10). The association with N-methyl-perfluorooctane sulfonamidoacetic acid was consistent across time periods.

Table 3.

Odds ratios and 95% confidence intervals for associations between PFAS in maternal serum and risk of childhood acute lymphoblastic leukemia in the Finnish Maternity Cohort, by sample year

PFAS	Sample year	No. detected (%)	Median (IQR)a	Cases, No.	Controls, No.	OR (95% CI)b,c
Perfluorooctanoic acid	1986-1995	110 (100)	3.51 (2.45-4.67)	110	110	0.40 (0.20 to 0.81)
	1996-2000	100 (100)	3.27 (2.51-4.29)	100	100	1.21 (0.65 to 2.23)
	2001-2005	109 (100)	3.16 (2.29-4.05)	109	109	1.56 (0.90 to 2.69)
	2006-2010	81 (100)	2.72 (1.96-3.44)	81	81	1.43 (0.76 to 2.68)
Perfluorooctanesulfonic acid	1986-1995	110 (100)	17.90 (13.36-24.76)	110	110	4.01 (1.62 to 9.93)
	1996-2000	100 (100)	13.66 (10.71-17.77)	100	100	1.34 (0.67 to 2.68)
	2001-2005	109 (100)	9.30 (7.24–13.04)	109	109	0.72 (0.42 to 1.26)
	2006-2010	80 (99.8)	4.80 (3.34-6.45)	81	81	0.96 (0.56 to 1.66)
Perfluorononanoic acid	1986-1995	55 (50.0)	0.32 (0.17-0.47)	110	110	1.14 (0.77 to 1.69)
	1996-2000	48 (52.0)	0.33 (0.18-0.51)	100	100	0.91 (0.61 to 1.35)
	2001-2005	79 (72.5)	0.54 (0.28-0.80)	109	109	0.82 (0.55 to 1.22)
	2006-2010	74 (91.4)	0.71 (0.49-0.99)	81	81	1.64 (0.86 to 3.10)
Perfluorodecanoic acid	1986-1995	91 (82.7)	0.15 (0.10-0.20)	110	110	1.33 (0.87 to 2.02)
	1996-2000	88 (88.0)	0.15 (0.10-0.21)	100	100	1.32 (0.82 to 2.13)
	2001-2005	90 (82.6)	0.26 (0.14-0.35)	109	109	1.37 (0.87 to 2.18)
	2006-2010	57 (70.4)	0.34 (0.25-0.48)	81	81	0.50 (0.26 to 0.96)
Perfluorohexane sulfonic acid	1986-1995	93 (84.6)	0.47 (0.30-0.69)	110	110	1.10 (0.81 to 1.50)
	1996-2000	81 (81.0)	0.50 (0.34-0.73)	100	100	0.84 (0.58 to 1.22)
	2001-2005	96 (88.1)	0.50 (0.29-0.79)	109	109	0.92 (0.67 to 1.27)
	2006-2010	78 (96.3)	0.34 (0.10-0.53)	81	81	1.02 (0.74 to 1.42)
N-methyl-perfluorooctane sulfonamidoacetic acid	1986-1995	79 (71.8)	0.24 (0.13-0.45)	110	110	1.26 (0.98 to 1.62)
	1996-2000	95 (95.0)	0.47 (0.31-0.67)	100	100	1.30 (0.98 to 1.71)
	2001-2005	81 (74.3)	0.28 (0.15-0.51)	109	109	1.18 (0.91 to 1.53)
	2006-2010	12 (14.8)	0.11 (0.06-0.14)	81	81	1.25 (0.87 to 1.80)
N-ethyl-perfluorooctane sulfonamidoacetic acid	1986-1995	109 (99.1)	0.72 (0.50-1.52)	110	110	0.83 (0.64 to 1.07)
	1996-2000	90 (90.0)	0.44 (0.29-0.89)	100	100	0.89 (0.71 to 1.12)
	2001-2005c	32 (29.4)	0.09 (0.08-0.18)	109	109	0.33 (0.15 to 0.74)
	2006-2010d	1 (1.23)	0.07 (0.03-0.11)	81	81	—

a

Distributions are based on values in controls. CI = confidence interval; IQR = interquartile range; OR = odds ratio; PFAS = per- and polyfluoroalkyl substances.

b

Associations are per 1-unit increase in serum PFAS on the log-base 2 scale, corresponding to an approximate doubling in analyte levels.

c

Adjusted for smoking during pregnancy (did not smoke, quit in first trimester, smoked after first trimester, missing or unknown) and mutually adjusted for correlated PFAS during each time interval based on Spearman ρ ≥ ±0.30 (correlations available in Supplementary Table 2, available online). Mutual adjustments for N-ethyl-perfluorooctane sulfonamidoacetic acid in 2001-2005 and 2006-2010 used the binary variable (detect or nondetect) because the detection proportion was less than 50% in each of these periods.

c

Detect or nondetect analysis because of detection less than 50% in this period.

d

Not estimated because of detection proportion approximately 1% in this period.

In sex-specific models, the pattern of association with N-methyl-perfluorooctane sulfonamidoacetic acid was similar for boys and girls (P_interaction = .91), although the odds ratio above the 90th percentile was slightly stronger among males (OR = 3.00, 95% CI = 1.21 to 7.42; P_trend < .01) than females (OR = 2.39, 95% CI = 0.93 to 6.18; P_trend = .32; Supplementary Table 6, available online). Associations with other PFAS also showed no clear differences by sex. Median levels of most PFAS were slightly higher among women with no previous live births, except for N-ethyl-perfluorooctane sulfonamidoacetic acid (0.22 µg/L in first vs 0.28 µg/L for second births) and for N-methyl-perfluorooctane sulfonamidoacetic acid (0.28 µg/L for both; Supplementary Table 7, available online). A positive association with ALL was suggested for perfluorononanoic acid among firstborn children (P_interaction = .06) and for perfluorooctanoic acid among second-born children (P_interaction = .41), but there was no evidence of statistical interaction. Associations with N-methyl-perfluorooctane sulfonamidoacetic acid were apparent among both first- and second-born children, although they were stronger and statistically significant only in the latter group (OR = 1.38, 95% CI = 1.10 to 1.73; P_interaction = .08). We found similar distributions of most of the PFAS among children in strata of mean diagnosis age. A suggestive positive association with perfluorooctanoic acid was apparent for those diagnosed aged younger than 5 years (P_interaction = .04), and the association with N-methyl-perfluorooctane sulfonamidoacetic acid was also stronger and statistically significant only among children diagnosed before age 5 years (OR_<5 years = 1.35, 95% CI = 1.12 to 1.63 vs OR_≥5 years = 1.09, 95% CI = 0.90 to 1.33; P_interaction = .02). In contrast, a suggestive positive association with perfluorodecanoic acid was apparent only in older children, although without evidence of statistical interaction (P_interaction = .61; Supplementary Table 8, available online).

Discussion

We observed a strong and statistically significant association between N-methyl-perfluorooctane sulfonamidoacetic acid levels in women’s serum during pregnancy and risk of childhood ALL in their offspring. Although we found no overall association with perfluorooctanesulfonic acid or perfluorooctanoic acid—the most abundant legacy compounds—ALL risk increased markedly with perfluorooctanesulfonic acid concentrations in 1986-1995, a period corresponding to peak perfluorooctanesulfonic acid serum levels in the population. Suggestive associations were also apparent with perfluorononanoic acid levels among first births and in later years when levels were at their highest.

ALL comprises approximately 27% of childhood cancers diagnosed in Finland (33), similar to other European countries, the United States, and China (34). It predominantly occurs in early childhood, with more than half of cases diagnosed before age 5 years (26). Incidence trends for childhood leukemia in Europe show increases that are not fully explained by diagnostic patterns or improvements in cancer registration (35). Established risk factors for childhood ALL include immune dysregulation, genetics, and environmental exposures (25,26). However, genetic causes explain only a small proportion of childhood cancers, and only maternal pesticide exposure during pregnancy and low-dose ionizing radiation have been consistently linked with ALL risk (25,26,34). The likelihood of higher PFAS exposures in children and their demonstrated immunosuppression effects in pediatric populations (28,36) were strong prior hypotheses motivating this investigation.

Perfluorooctanesulfonic acid was added to the Stockholm Convention rule on persistent organic pollutants in 2009, which recommended against its use. Although serum in our study was predominantly collected prior to this implementation, the principal worldwide manufacturer of perfluorooctanesulfonic acid announced a voluntary phaseout in use of a major precursor compound, perfluorooctanesulfonyl fluoride, in 2000. Serum perfluorooctanesulfonic acid levels in our controls showed corresponding declines over time, with near fourfold higher exposure levels during the earliest sample period vs the latest. We found that perfluorooctanesulfonic acid levels were associated with childhood ALL in this early period, when levels were at their highest. Perfluorooctanesulfonic acid levels in our controls in 1996-2000 (geometric mean = 13.4 µg/L) were approximately 50% lower than in women in the US National Health and Nutrition Examination Survey (NHANES) during the closest comparable period, 1999-2000 (geometric mean = 28.0 µg/L) (37). Levels of N-methyl-perfluorooctane sulfonamidoacetic acid, a perfluorooctanesulfonic acid precursor, likewise followed a similar pattern (geometric mean = 0.45 µg/L in 1996-2000 in FMC vs 0.9 µg/L in women of the 1999-2000 NHANES) (37). In contrast, we observed less of a decline in perfluorooctanoic acid levels, perhaps because of the more recent restriction in its use. Levels of perfluorooctanoic acid in FMC mothers were more like levels in the earliest available measurements in the US general population (geomean = 3.5 µg/L in NHANES 1999-2000 vs 2.8 µg/L in FMC in 1996-2000). Thus, our data also demonstrate that the mixture of PFAS exposures varied over time, an important consideration in the interpretations of our findings. We also saw the strongest suggestive increased risk of ALL in association with perfluorononanoic acid in a later time period (2006-2010) corresponding to when its levels were at their highest in the maternal sera, similarly supporting plausibility. It is not clear why we observed inverse associations with N-ethyl-perfluorooctane sulfonamidoacetic acid in the main analyses. However, by 2001-2005, N-ethyl-perfluorooctane sulfonamidoacetic acid was detected in fewer than 50% of samples, and associations were not statistically significant in the earlier periods when detection rates and serum levels were higher.

Another major finding of our study was the strong association of ALL with N-methyl-perfluorooctane sulfonamidoacetic acid, an N-alkyl sulfonamino oxidation by-product of N-methyl perfluorooctanesulfonamidoethanol, which was used primarily in surface treatment applications (eg, protectants for carpets, textiles) (38,39). N-ethyl-perfluorooctane sulfonamidoacetic acid is an oxidation product of N-ethyl perfluorooctanesulfonamidoethanol, which was primarily used in paper and packaging protectant applications. Both N-methyl-perfluorooctane sulfonamidoacetic acid and N-ethyl-perfluorooctane sulfonamidoacetic acid can be considered markers of consumer-related exposure to perfluorooctanesulfonyl fluoride-based products and can metabolize to perfluorooctanesulfonamidoacetate and perfluorooctane sulfonamide, which in turn can metabolize to perfluorooctanesulfonic acid as a terminal end product (39). N-methyl-perfluorooctane sulfonamidoacetic acid has been detected in multiple environmental media, including in wastewater, soil, sediments, and precipitation, as well as in wildlife (38). Like perfluorooctanesulfonic acid, levels of N-methyl-perfluorooctane sulfonamidoacetic acid and N-ethyl-perfluorooctane sulfonamidoacetic acid have been detected in the serum of populations around the world, with levels of both declining because of the phaseout of perfluorooctanesulfonyl fluoride use in consumer products. Our data demonstrated small and nonmonotonic declines in levels and larger declines in detection rates of both PFAS, more markedly for N-ethyl-perfluorooctane sulfonamidoacetic acid. This difference in serum levels may be explained by the prior use of N-methyl perfluorooctanesulfonamidoethanol (which degrades into N-methyl-perfluorooctane sulfonamidoacetic acid) on products such as textiles and furnishings that remain in the home and may result in repeated exposure vs the use of N-ethyl perfluorooctanesulfonamidoethanol (which degrades to N-ethyl-perfluorooctane sulfonamidoacetic acid) on the paper and other products with which people come into more transient contact. To our knowledge, this is the first report of any positive cancer association with N-methyl-perfluorooctane sulfonamidoacetic acid. Our cross-classified analyses showed the N-methyl-perfluorooctane sulfonamidoacetic acid association with ALL was evident even when levels of perfluorooctanesulfonic acid were low, suggesting that the N-methyl-perfluorooctane sulfonamidoacetic acid association is independent of perfluorooctanesulfonic acid. Whether the immunotoxic and immune-dysregulation properties of perfluorooctanesulfonic acid are directly translatable to this precursor compound is unknown.

We also observed an association with perfluorononanoic acid by birth order, with the strongest suggested association among firstborn children and no clear association in second-born offspring. This finding is plausible given that median perfluorononanoic acid levels were higher in the mothers of firstborn children and maternal PFAS concentrations decrease with parity (16,31), which could lead to firstborn children having higher in utero exposures compared with subsequent offspring. However, although we saw a similar but weaker pattern for perfluorooctanesulfonic acid, the relationship was the opposite for perfluorooctanoic acid, which yielded a suggestive association with ALL only among second births. These findings are intriguing, but as we had fewer second-born children in our analysis and time period–specific associations evident in the main analysis, their interpretation may be limited. Our analyses also yielded different patterns of association by age group, with the most notable the positive relationships with perfluorooctanoic acid and N-methyl-perfluorooctane sulfonamidoacetic acid among children diagnosed before 5 years. Given the varying prevalence of ALL molecular subtypes by age (40), these findings may provide clues to differential effects of PFAS on ALL subtypes.

A major strength of our study is the quantification of maternal PFAS levels during early pregnancy, a potentially critical exposure period for the developing fetus. Other strengths include the population-based FMC study design and large number of cases, which enabled us to evaluate associations by important risk factors and potential modifiers of PFAS exposure. Although we adjusted for smoking status during pregnancy, smoking frequency and history prior to pregnancy were not available. Likewise, although continued exposure to N-methyl-perfluorooctane sulfonamidoacetic acid and perfluorooctanesulfonic acid through sources in the home is plausible, we lacked information about the child’s exposures after birth, such as from dietary and drinking water sources. Similarly, we had no information on breastfeeding, but breastfeeding rates in Finland are high (92% at 1 month and 66% at 6 months) (41). Therefore, we would expect limited heterogeneity in postnatal exposure from breast milk and for PFAS levels in pregnancy to correlate with postpartum levels. We also expect limited influence on PFAS levels from hemodynamic changes during pregnancy because all samples were collected in the first trimester. Although cancer registration in Finland is of high quality (42), a previous FMC study of childhood leukemia (1983-2006) found that 74% of cases were linked to an index mother with available serum (43). However, bias due to differential cohort participation by both PFAS concentrations and ALL risk seems unlikely. Consistent with other populations, 80% of leukemia cases were ALL (43), suggesting that the case ascertainment is representative of the general Finnish population, although information on specific molecular subtypes was not available. Finally, although we matched on most ALL risk factors, maternal BMI data were insufficient for analysis. We also lacked information on parental occupation and the geographic location of cases and controls, therefore we acknowledge the potential for confounding by these factors in our analyses.

We found that in utero exposure to some PFAS is positively associated with childhood ALL. Specifically, N-methyl-perfluorooctane sulfonamidoacetic acid showed a strong and consistent dose-response relationship, and the highest concentrations of mother’s serum perfluorooctanesulfonic acid were associated with the development of ALL in their children. Given the ubiquity of PFAS exposure and limited established childhood leukemia risk factors, these findings have important public health implications.

Data availability

Study data are not publicly available owing to data privacy issues. Analytic code and a data dictionary can be requested after publication from Northern Finland Biobank Borealis. More information is available at https://oys.fi/biopankki/.

Author contributions

Rena R. Jones, PhD, MS (Conceptualization; Data curation; Formal analysis; Funding acquisition; Investigation; Methodology; Project administration; Resources; Writing—original draft; Writing—review & editing), Jessica M. Madrigal, PhD (Data curation; Formal analysis; Writing—review & editing), Rebecca Troisi, ScD (Conceptualization; Data curation; Writing—review & editing), Heljä-Marja Surcel, PhD (Conceptualization; Data curation; Methodology; Project administration; Writing—review & editing), Hanna Öhman, PhD (Conceptualization; Data curation; Methodology; Writing—review & editing), Juha Kivelä, PhD (Data curation; Project administration; Writing—review & editing), Hannu Kiviranta, PhD (Formal analysis; Writing—review & editing), Panu Rantakokko, PhD (Formal analysis; Writing—review & editing), Jani Koponen, PhD (Formal analysis; Investigation; Writing—review & editing), Danielle Medgyesi, MS (Conceptualization; Writing—review & editing), Katherine McGlynn, PhD (Conceptualization; Writing—review & editing), Joshua Sampson, PhD (Conceptualization; Methodology; Writing—review & editing), Paul Albert, PhD (Methodology; Writing—review & editing), and Mary H. Ward, PhD (Conceptualization; Data curation; Formal analysis; Funding acquisition; Investigation; Methodology; Project administration; Writing—review & editing).

Funding

This work was supported by the Intramural Research Program of the National Cancer Institute (ZIA CP010125–28).

Conflicts of interest

All authors declare no competing interests.

Acknowledgements

The funder had no role in the design or conduct of the study, including the collection, management, analysis, and interpretation of the data; it reviewed and approved the manuscript for submission.