Plasma concentrations of perfluoroalkyl substances and risk of inflammatory bowel diseases in women: a nested case control analysis in the Nurses’ Health Study cohorts

Abstract

BACKGROUND

Perfluoroalkyl substances (PFASs) are synthetic compounds used in a wide variety of industrial and consumer applications. An association between PFAS exposure and risk of ulcerative colitis (UC) has been reported in a highly exposed population. However, data are limited on risk of inflammatory bowel diseases (IBD) among individuals with background population levels of PFAS exposure.

OBJECTIVES

We set out to examine the association between plasma PFAS concentrations and risk of IBD among women in two population-based, prospective cohort studies in which pre-diagnostic blood specimens were available.

METHODS

We conducted a nested case-control study in the Nurses' Health Study and Nurses' Health Study II cohorts. We identified 73 participants with incident Crohn’s disease (CD) and 80 participants with incident UC who had provided blood samples before diagnosis. Cases were matched 1:2 to IBD-free controls. Plasma concentrations of five major PFASs were measured by liquid chromatography and tandem mass spectrometry. We used conditional logistic models to estimated odds ratios for risk of IBD according to log₁₀-transformed PFAS concentrations, adjusting for potential confounders.

RESULTS

In multivariable models, we observed inverse associations between plasma concentrations of three PFASs and risk of CD (all P≤0.012 for a standard deviation increase in log₁₀PFAS). The inverse association with CD was strongest for perfluorodecanoate, where, compared to the lowest tertile, the odds ratio (OR) for the highest tertile was 0.39 (95% confidence interval, 0.17-0.92). No associations were observed between PFAS concentrations and UC risk.

DISCUSSION

Our results do not support the hypothesis that elevated PFAS exposure is associated with higher risk of UC. Contrary to expectation, our data suggest that circulating concentrations of some PFASs may be inversely associated with CD development.

INTRODUCTION

The incidence of inflammatory bowel diseases (IBD), comprising Crohn’s disease (CD) and ulcerative colitis (UC), has increased in parallel with urbanization in historically low-incidence regions such as Asia, South America and the Middle East.¹ Multiple environmental factors have been hypothesized to account for these trends in IBD incidence, including the “westernization” of diet, rising prevalence of obesity, physical inactivity, and improved sanitation and hygiene.^2,3 Moreover, owing to the association between IBD and urban living in some studies, environmental pollutants have also been considered as exposures that might explain temporal and geographic incidence trends.^4,5 A small number of studies have examined associations between ambient air pollution and IBD with mixed results,^4-6 however, persistent organic pollutants have received little attention as IBD risk factors.

Perfluoroalkyl substances (PFASs) are synthetic compounds used in a wide variety of industrial, manufacturing, and consumer applications, including production of fluoropolymers (e.g. polytetrafluoroethylene), surfactants, and oil and water repellent coatings for food packaging and textiles.⁷ PFAS are resistant to degradation and are highly persistent in the environment.⁸ Human exposure to PFAS occurs through contaminated water, food, and indoor and outdoor air.^9-11 Elimination half-lives for the most common PFASs in humans are estimated to be 2.5 – 8 years and almost the entire U.S. population has detectable serum PFOA and other PFAS in the ng/mL range.^8,12,13 The two most-produced PFASs are perfluorooctane sulfonate (PFOS) and perfluorooctanoic acid (PFOA, also known as C8). Production of PFOS was curtailed in the U.S. in 2002 and a Global PFOA Stewardship Program was initiated in 2006.¹⁴ Although serum concentrations of legacy PFASs have decreased in the US population in recent years,^15,16 PFAS exposure remains a public health concern on account of the indefinite persistence and long elimination half-lives of these substances.

An association between PFOA exposure and risk of UC has been reported in a highly exposed US population.^17,18 In contrast, no increase in IBD risk was observed in a Swedish cohort exposed to PFAS-contaminated drinking water.¹⁹ Only one study has examined IBD risk among individuals exposed to PFAS at levels typical of those experienced by the general population.²⁰ In that study, patients with UC were found to have higher mean concentrations of PFOA in post-diagnostic blood samples compared to CD patients or controls. However, concentrations of other PFASs were higher among UC cases than controls. We therefore set out to examine the association between plasma concentrations of PFASs and risk of CD and UC among women in two population-based prospective cohort studies in which blood specimens were obtained many years prior to IBD diagnosis.

METHODS

Study populations

The Nurses’ Health Study (NHS) is a nationwide prospective cohort that was initiated in 1976 with the enrollment of 121,701 female registered nurses, aged 30–55 years at baseline, who responded to a postal health questionnaire. Follow-up questionnaires have been administered every two years to update information on health and lifestyle factors. The Nurses’ Health Study II (NHS II) is a parallel cohort established in 1989 with the enrollment of 116,686 female nurses, aged 25–42 years. Participants in the NHS II have been similarly followed using biennial questionnaires. Follow-up exceeds 85% for both cohorts.²¹

Ascertainment of IBD

We have previously reported in detail the ascertainment of IBD cases in the NHS and NHS II cohorts.^22,23 In brief, from baseline, participants in both cohorts had the opportunity to report diagnoses of UC or CD through an open-ended response option on biennial questionnaires. From 1982 onward in the NHS, a diagnosis of UC has been specifically queried, and from 1992 onward for CD. In the NHS II, diagnoses of UC and CD have both been queried specifically since 1993. When a participant in either cohort reported a diagnosis of IBD, a supplementary questionnaire was mailed, and permission requested for access to related medical records. Approximately 80% of participants contacted gave consent for record review. Two gastroenterologists, blinded to exposure information, independently reviewed the records (PL, HK, ANA, KEB, JMR). A diagnosis of CD or UC was assigned based on established endoscopic, surgical, radiologic, and histologic findings.^24,25 Instances of disagreement on case definition were resolved by consensus.

Blood collection and identification of cases and controls

Blood collection in the NHS cohorts has been described previously.^26,27 Briefly, between 1989 and 1990, 32,826 NHS participants (aged 43–69 years) volunteered to provide a blood sample and complete a short questionnaire. Blood specimens were obtained similarly from 29,611 NHS II participants (aged 32–54 years) between 1996 and 1999. Participant blood specimens were shipped overnight on ice and centrifuged promptly on arrival. Aliquots of plasma were transferred to continuously-monitored liquid nitrogen freezers. PFASs are extremely stable and their concentrations are unaffected by long-term storage.²⁸

The current case-control analysis utilized 73 cases of CD (58 from NHS and 15 from NHS II) and 80 cases of UC (66 from NHS and 14 from NHS II) with available plasma, obtained from blood samples collected before diagnosis or indexing. Each case was matched to two control individuals who were free of IBD at the time of diagnosis of their respective case. Matching was based on cohort, age, month of blood collection, fasting status, menopausal status, and menopausal hormone therapy use at the time of blood collection.²⁹ Menopausal status and hormone use were considered potential confounders based on prior studies in these cohorts.³⁰ For a small number of cases (CD n=3, UC n=1), plasma was only available from one of the two matched controls.

Assessment of plasma PFAS concentrations

Based on a previous analysis of plasma PFAS concentrations in the NHS II,³¹ we chose to measure the concentrations of the five most abundant PFASs (all >99% detection rate). PFOA, perfluorohexanesulfonate (PFHxS), perfluorononanoic acid (PFNA), perfluorodecanoic acid (PFDA), and PFOS [branched (br-PFOS) and linear (n-PFOS) isoforms] were measured by online solid-phase extraction and analysis using liquid chromatography coupled to a triple quadropole tandem mass spectrometer, as described previously.³² The analyses were performed at the Environmental Medicine Laboratory at the University of Southern Denmark. Concentrations of the two PFOS isoforms were very strongly correlated (Spearman r=0.84; p<0.001) and we therefore chose to use total PFOS, the sum of PFOS isoform concentrations, for all analyses.

Plasma specimens of matched case–control pairs were analyzed in the same analytical run. Blinded duplicate quality control (QC) samples were included in each batch to monitor the quality of these assays. Overall, within batch coefficients of variation (CV) ranged from 4.0% - 8.7% for PFOA, 4.5% - 12.3% for PFHxS, 5.9% - 9.5% for PFNA, 6.2% - 11.3% for PFDA, and 2.3% - 10.6% for total PFOS. External QC samples showed recoveries for all PFASs of 92.6 to 108.7%. The limit of detection (LOD) for all analytes was 0.03 ng/mL.

Assessment of covariates

For body weight, menopausal status, and menopausal hormone use, data were extracted from short questionnaires administered at the time of blood collection. For all other covariates, including smoking status, leisure-time physical activity, oral contraceptive (OC) use, parity, and state of residence, information was obtained from the standard biennial follow up questionnaires returned closest to the time of blood collection. We use cumulative average Alternative Healthy Eating Index 2010 (AHEI) as a measure of diet quality since AHEI scores are strongly associated with the risk of chronic diseases.³³ AHEI score was derived using data from semi-quantitative, validated, food frequency questionnaires administered every 2 – 4 years.³⁴

Statistical analyses

Plasma PFAS concentrations that were lower than the LOD of the assay were replaced by the LOD divided by 2 (i.e., 0.015 ng/mL). This applied to only a single control participant for PFDA. To control for imprecision introduced by batch effects, PFAS concentrations were batch-corrected using the method described previously by Rosner et al.³⁵ Batch-correction used a linear model to regress PFAS concentrations on batch indicator variables. PFAS concentrations were therefore normalized by log₁₀ transformation, given their right-skewed distributions.³¹ PFAS concentrations were then recalibrated by subtracting the difference between the coefficient of each batch and the average of the coefficients across all batches.

A chi-square test or Wilcoxon rank-sum tests was used to compare categorical or continuous features between cases and controls. We created tertiles for plasma PFAS for CD and UC case-control populations separately, based on the distribution of PFAS concentrations in controls. We used the lowest tertile as referent and median value for each tertile as an ordinal variable to test for linear trend. We also generated a continuous variable for log-transformed PFAS values where one unit change was equal to one standard deviation (SD). Odds ratios and 95% confidence intervals (95% CIs) were computed using conditional logistic regression models, stratified on match. The basic model controlled for the matching factors of cohort, age, month of blood draw, menopausal status, and menopausal hormone use. A multivariable model adjusted for potential confounding by BMI (continuous), smoking status (never/former/current), OC use (ever/never), parity (continuous), physical activity (continuous), diet quality (continuous AHEI score), and region of residence [classified as coastal (AL, AK, CA, CT, DE, FL, GA, HI, ME, MD, MA, NH, NJ, NC, OR, RI, SC, VA, and WA), Great Lakes region (IL, IN, MI, MN, NY, OH, PA, and WI), or inland (all other states), as previously described³¹]. Statistical interaction was evaluated using the Wald test on multiplicative interaction terms included in the multivariable model. Given that we evaluated multiple exposures for two outcomes, we decided a priori to limit interaction analyses to age and the interval between blood collection and diagnosis. Interaction terms were generated using binary variables for the potential effect modifier, with cut points at the approximate median value for the CD or UC case-control population, and continuous log-transformed PFAS concentrations. We conducted sensitivity analyses excluding participants where the date of blood collection was less than 2 years before the date of IBD diagnosis or indexing. PFAS exposure has been reported to vary with socioeconomic status,³⁶ we therefore additionally ran multivariable models for each continuous log-transformed PFAS including adjustment for median census tract household income, in tertiles. Given that seafood consumption is a source of PFAS exposure, and fish or marine fatty acids intake may be associated with IBD risk,³⁷ we also conducted sensitivity analyses adjusting for cumulative average daily number of portions of fish and shellfish consumed, using prospectively-collected food frequency questionnaire data.³⁸

Approvals

The study protocol was approved by the institutional review board of the Brigham and Women’s Hospital and the Human Subjects Committee Review Board of Harvard T.H. Chan School of Public Health.

RESULTS

Participant characteristics

Characteristics of IBD cases and matched controls at the time of blood collection are shown in Table 1. As expected, there were no significant differences between cases and controls for the matching factors of age, menopausal status, menopausal hormone use, and time interval between blood collection and diagnosis/indexing (all P_difference ≥0.85). Overall, the median time interval between blood collection and diagnosis or indexing was 10.5 years (interquartile range 6.3-14.8 years). The interval was longer for UC cases and controls (11.5 years) compared to CD cases and controls (9.6 years). No statistically significant differences were observed for BMI, physical activity, smoking status, OC use, parity, and time interval between blood collection and diagnosis or indexing. Compared to CD cases, a smaller proportion of CD controls were from coastal states (P_difference=0.028). The median plasma concentrations for PFOA and PFDA were somewhat higher among CD controls compared cases (4.16 ng/ml vs. 3.64 ng/ml, P_difference=0.04 for PFOA and 0.19 ng/ml vs. 0.14 ng/ml P_difference=0.01 for PFDA). PFAS concentrations did not differ significantly between UC cases and controls (all P≥0.08).

Table 1.

Baseline characteristics of cases and controls at time of blood collection

	CD			UC
	Cases (n=73)	Controls (n=143)	p- valuea	Cases (n=80)	Controls (n=159)	p- valuea
Cohort
NHS	58	113		66	131
NHSII	15	30		14	28
Age, years (SD)	53.6 (8.2)	53.5 (8.3)	0.98	52.9 (7.3)	53.1 (7.3)	0.85
BMI, Kg/m2 (SD)	26.3 (5.3)	26.0 (5.2)	0.69	25.8 (4.0)	25.3 (4.6)	0.42
Physical activity, MET-h/wk (SD)	14.3 (17.0)	15.4 (14.7)	0.61	13.5 (14.7)	16.3 (22.0)	0.24
AHEI dietary score (SD)	52.7 (8.8)	50.9 (7.5)	0.13	50.5 (8.3)	50.6 (9.6)	0.90
Fish consumption (SD), portions/day	0.26 (0.18)	0.26 (0.18)	0.97	0.27 (0.19)	0.25 (0.19)	0.99
Smoking status (%)
Never	31 (42)	67 (47)		32 (40)	76 (48)
Former	26 (36)	58 (41)	0.20	39 (49)	68 (43)	0.52
Current	16 (22)	18 (13)		9 (11)	15 (9)
Pre-menopausal (%)	23 (32)	45 (31)	0.99	27 (34)	54 (34)	0.97
Oral contraceptive use (%)b	42 (58)	88 (62)	0.57	53 (66)	90 (57)	0.15
Post-menopausal hormone use (%) c	20 (27)	40 (28)	0.93	22 (28)	44 (28)	0.98
Parity (SD)	2.7 (1.6)	3.1 (1.8)	0.14	2.9 (1.8)	2.9 (1.7)	0.92
Region of residenced (%)
Coastal	26 (36)	32 (22)	0.028	22 (28)	45 (28)	0.81
Great lakes	40 (55)	80 (56)		50 (63)	102 (64)
Inland/other	7 (9.6)	31 (22)		8 (10)	12 (7.6)
Census tract median household income (IQR), 1000 USD	59 (47-74)	59 (46-72)	0.76	55 (45-73)	59 (47-76)	0.59
Interval from blood collection to diagnosis or indexing (IQR), years	9.7 (4.8-13.4)	9.6 (4.7-13.7)	0.96	11.5 (7.0-15.8)	11.5 (7.3-15.8)	0.96
Median plasma PFAS concentration (IQR), ng/ml
PFOA	3.64 (2.33-5.35)	4.16 (3.02-5.20)	0.04	3.97 (2.85-4.83)	4.11 (3.13-5.26)	0.23
PFHxS	1.74 (0.91-2.39)	1.59 (1.11-2.90)	0.60	1.61 (1.14-2.70)	1.97 (1.25-3.34)	0.08
PFNA	0.62 (0.45-0.88)	0.65 (0.47-0.94)	0.37	0.68 (0.48-1.05)	0.68 (0.51-0.97)	0.96
PFDA	0.14 (0.09-0.22)	0.19 (0.12-0.26)	0.015	0.20 (0.12-0.27)	0.19 (0.13-0.27)	0.86
Total PFOS	22.59 (14.63-29.66)	23.93 (16.69-36.12)	0.19	23.68 (17.13-30.54)	24.51 (18.42-32.50)	0.36

Numbers of participants and percentages are given for categorical variables. Percentages may not sum to 100 due to rounding. For continuous variables, the mean value is presented with standard deviation (SD) in parentheses. For time interval, household income, and PFAS concentrations, the median and interquartile range (IQR) are given.

^aThe p-value for difference was computed by Pearson chi-square test for categorical variables, a t-test for continuous variables, and Wicoxon rank sum test for PFAS concentrations.

^bEver use of oral contraceptive use.

^cCurrent use of hormonal therapy among post-menopausal women.

^dCoastal states: AL, AK, CA, CT, DE, FL, GA, HI, ME, MD, MA, NH, NJ, NC, OR, RI, SC, VA, and WA; Great Lakes region states: IL, IN, MI, MN, NY, OH, PA, and WI; and inland/other states: all other states.

AHEI, alternative healthy eating index (including alcohol); BMI, body mass index; IQR, interquartile range; MET, metabolic equivalent task; PFAS, perfluoroalkyl substances; PFDA, perfluorodecanoic acid; PFHxS, perfluorohexane sulfonic acid; PFNA, perfluorononanoic acid; PFOA, perfluorooctanoic acid; PFOS, perfluorooctane sulfonate (total of linear and branch forms); SD, standard deviation, USD, United States dollars.

Association between PFAS concentrations and risk of IBD

For all five PFASs, the multivariable ORs for CD were numerically lower for the second and third tertiles of PFAS concentrations compared to the lowest tertile (Table 2). However, there was a statistically significant trend only for PFDA (tertile 2 vs.1 OR=0.59; 95% CI 0.29-1.20; tertile 3 vs. 1 OR=0.39; 95% CI, 0.17-0.92; P_trend=0.027). In multivariable models where log-transformed PFAS concentrations were modelled as continuous variables, statistically significantly lower odds of CD were observed for PFOA (OR for 1 SD increase = 0.63; 95% CI, 0.46-0.87; P=0.005), PFDA (OR for 1 SD increase = 0.52; 95% CI, 0.35-0.78; P=0.002), and PFOS (OR for 1 SD increase = 0.64; 95% CI, 0.45-0.90; P=0.012). A statistically non-significant trend toward lower odds of CD was observed for PFNA (multivariable OR for 1 SD increase = 0.73; 95% CI, 0.50-1.05; P=0.089).

Table 2.

Association between plasma PFAS levels and risk of Crohn’s disease

Exposure	Median (range), (ng/ml)	Cases (n)	Controls (n)	Basic model OR (95% CI)	Multivariable model OR (95% CI)
PFOAa
Tertile 1	2.45 (0.89-3.32)	33	47	referent	referent
Tertile 2	3.99 (3.36-4.77)	19	48	0.60 (0.31-1.17)	0.54 (0.25-1.16)
Tertile 3	6.48 (4.84-73.12)	21	48	0.66 (0.34-1.29)	0.59 (0.28-1.25)
p-trendb				0.16	0.12
Log10PFOAc				0.73 (0.55-0.96)	0.63 (0.46-0.87)
p-value				0.027	0.0047
PFHxSa
Tertile 1	0.84 (0.17-1.25)	29	47	referent	referent
Tertile 2	1.73 (1.29-2.30)	24	48	0.86 (0.45-1.64)	0.77 (0.36-1.63)
Tertile 3	4.13 (2.33-85.28)	20	48	0.70 (0.35-1.37)	0.63 (0.29-1.36)
p-trendb				0.30	0.24
Log10PFHxSc				0.91 (0.69-1.19)	0.82 (0.59-1.15)
p-value				0.48	0.25
PFNAa
Tertile 1	0.44 (0.10-0.54)	31	47	referent	referent
Tertile 2	0.65 (0.55-0.82)	20	48	0.65 (0.32-1.29)	0.53 (0.23-1.19)
Tertile 3	1.12 (0.83-4.35)	22	48	0.68 (0.33-1.42)	0.54 (0.24-1.23)
p-trendb				0.28	0.15
Log10 PFNAc				0.90 (0.65-1.23)	0.73 (0.50-1.05)
p-value				0.50	0.089
PFDAa
Tertile 1	0.10 (0.04-0.13)	33	47	referent	referent
Tertile 2	0.17 (0.13-0.23)	23	48	0.69 (0.36-1.31)	0.59 (0.29-1.20)
Tertile 3	0.32 (0.24-1.21)	17	48	0.49 (0.23-1.04)	0.39 (0.17-0.92)
p-trendb				0.057	0.027
Log10 PFDAc				0.67 (0.49-0.93)	0.52 (0.35-0.78)
p-value				0.016	0.0016
Total PFOSa
Tertile 1	13.31 (2.03-18.63)	26	47	referent	referent
Tertile 2	23.74 (18.72-31.40)	30	48	1.06 (0.52-2.16)	0.93 (0.42-2.06)
Tertile 3	40.91 (31.57-117.97)	17	48	0.58 (0.25-1.33)	0.48 (0.18-1.26)
p-trendb				0.19	0.14
Log10PFOSc				0.74 (0.55-1.00)	0.64 (0.45-0.90)
p-value				0.052	0.012

The basic model conditioned on matching factors, including age, month of sample collection, fasting status, menopausal status and postmenopausal hormone use. The multivariable model, additionally adjusted for baseline BMI (continuous), every use of oral contraceptives, parity, states of residence (coastal, Great Lakes, or inland), smoking status (never, former, or current), physical activity (continuous), and AHEI score (continuous). AHEI, alternative healthy eating index; BMI, body mass index; CI, confidence interval; OR, odds ratio; PFAS, perfluoroalkyl substances; PFDA, perfluorodecanoic acid; PFHxS, perfluorohexane sulfonic acid; PFNA, perfluorononanoic acid; PFOA, perfluorooctanoic acid; PFOS, perfluorooctane sulfonate (total of linear and branch forms)

^aOdds ratios according to batch-corrected PFAS tertiles estimated using conditional logistic models.

^bp-values for linear trend were calculated by modeling the median PFAS values for each tertile as a continuous variable.

^cORs for a 1-SD increase in Log₁₀-transformed PFAS, which were 0.24 for PFOA, 0.37 for PFHxS, 0.26 for PFNA, 0.27 for PFDA, and 0.27 for total PFOS.

For UC, there was no trend toward lower odds associated with increasing tertiles of PFAS concentrations (Table 3; all multivariable P_trend≥0.20). No statistically significant associations were seen for UC and continuous log-transformed PFAS concentrations (all multivariable P_trend≥0.20).

Table 3.

Association between plasma PFAS levels and risk of ulcerative colitis

Exposure	Median (range), (ng/ml)	Cases (n)	Controls (n)	Basic model OR (95% CI)	Multivariable model OR (95% CI)
PFOA a
Tertile 1	2.76 (0.34-3.38)	31	53	referent	referent
Tertile 2	4.13 (3.39-4.78)	28	53	0.91 (0.48-1.74)	1.01 (0.51-2.02)
Tertile 3	6.15 (4.79-23.39)	21	53	0.68 (0.35-1.35)	0.78 (0.38-1.59)
p-trendb				0.28	0.49
Log10PFOA c				0.87 (0.64-1.18)	0.91 (0.66-1.25)
p-value				0.38	0.55
PFHxS a
Tertile 1	1.09 (0.20-1.46)	32	53	referent	referent
Tertile 2	1.90 (1.47-2.65)	28	53	0.86 (0.43-1.70)	0.87 (0.42-1.80)
Tertile 3	4.84 (2.69-47.48)	20	53	0.62 (0.31-1.25)	0.62 (0.29-1.31)
p-trend b				0.18	0.20
Log10PFHxS c				0.82 (0.61-1.11)	0.82 (0.60-1.12)
p-value				0.20	0.21
PFNA a
Tertile 1	0.44 (0.11-0.56)	28	53	referent	referent
Tertile 2	0.69 (0.57-0.83)	24	54	0.84 (0.43-1.67)	0.84 (0.40-1.75)
Tertile 3	1.26 (0.84-19.55)	28	52	1.03 (0.51-2.09)	1.14 (0.53-2.48)
p-trend b				0.91	0.69
Log10PFNA c				1.02 (0.77-1.36)	1.10 (0.80-1.50)
p-value				0.88	0.56
PFDA a
Tertile 1	0.11 (0.01-0.15)	30	53	referent	referent
Tertile 2	0.19 (0.15-0.23)	19	53	0.64 (0.33-1.26)	0.69 (0.34-1.41)
Tertile 3	0.31 (0.23-2.86)	31	53	1.09 (0.56-2.13)	1.21 (0.59-2.48)
p-trend b				0.86	0.66
Log10PFDA c				1.06 (0.80-1.40)	1.13 (0.83-1.54)
p-value				0.71	0.43
Total PFOS a
Tertile 1	15.41 (2.32-21.09)	34	53	referent	referent
Tertile 2	24.73 (21.16-29.57)	22	53	0.58 (0.29-1.19)	0.57 (0.27-1.19)
Tertile 3	36.03 (29.59-106.90)	24	53	0.65 (0.33-1.29)	0.67 (0.33-1.37)
p-trend b				0.22	0.26
Log10 PFOS c				0.85 (0.60-1.20)	0.88 (0.62-1.25)
p-value				0.35	0.47

The basic model conditioned on matching factors, including age, month of sample collection, fasting status, menopausal status and postmenopausal hormone use. The multivariable model, additionally adjusted for baseline BMI (continuous), every use of oral contraceptives, parity, states of residence (coastal, Great Lakes, or inland), smoking status (never, former, or current), physical activity (continuous), and AHEI score (continuous). AHEI, alternative healthy eating index; BMI, body mass index; CI, confidence interval; OR, odds ratio; PFAS, perfluoroalkyl substances; PFDA, perfluorodecanoic acid; PFHxS, perfluorohexane sulfonic acid; PFNA, perfluorononanoic acid; PFOA, perfluorooctanoic acid; PFOS, perfluorooctane sulfonate (total of linear and branch forms)

^aOdds ratios according to batch-corrected PFAS tertiles estimated using conditional logistic models.

^bp-values for linear trend were calculated by modeling the median PFAS values for each tertile as a continuous variable.

^cORs for a 1-SD increase in Log₁₀-transformed PFAS, which were 0.24 for PFOA, 0.37 for PFHxS, 0.26 for PFNA, 0.27 for PFDA, and 0.27 for total PFOS.

Considering the natural history of IBD and the potential delay in obtaining a formal diagnosis, we ran a sensitivity analysis excluding cases and matched controls where the date of IBD diagnosis or indexing was within 2 years of the date of blood collection. For CD, this left 66 cases and 130 matched controls. Although precision was lower, results were generally consistent with those of our main analysis (Supplementary Table 1). The multivariable association with increasing PFDA tertiles was attenuated somewhat (tertile 2 vs.1 OR=0.64; 95% CI 0.30-1.36; terile 3 vs. 1 OR=0.46; 95% CI, 0.18-1.17; P_trend=0.084). The association for continuous PFOS was also slightly weaker (OR for 1 SD increase = 0.70; 95% CI, 0.49-1.01; P=0.056). However, estimates for continuously modelled PFOA (OR for 1 SD increase = 0.69; 95% CI, 0.50-0.95; P=0.022) and PFDA (OR for 1 SD increase = 0.54; 95% CI, 0.35-0.84; P=0.006) were similar to those in our main analysis.

Among 76 UC cases and 151 controls diagnosed or indexed more than 2 years from blood collection (Supplementary Table 2), the estimates were also similar to those of our main analysis, with no statistically significant association between PFAS concentrations modelled as tertiles (P_trend≥0.12) or as continuous variables (P_trend≥0.11).

Additional adjustment of our multivariable models for census tract-level median household income and seafood consumption did not meaningfully impact the estimates obtained in our main analyses (data not shown).

Exploratory analyses

We examined whether age and time interval from blood collection to diagnosis or indexing modified the association between PFAS concentrations and IBD risk. In age stratified multivariable models, there was a suggestion of heterogeneity in the estimates for CD between age strata (<54 years or ≥54 years) for PFDA (Supplementary Table 3). For a 1 SD increase in PFDA concentration, the OR for the lower age group was 0.22 (95% CI, 0.08-0.60) compared 0.73 (0.43-1.25) for the higher age group (P_interaction=0.047). No statistically significant differences were observed between age strata for other PFASs (all P_interaction≥0.14), nor for any PFAS and time interval from blood collection (all P_interaction≥0.44). For UC no statistical evidence of effect modification was apparent for age or time interval from blood collection (Supplementary Table 4; all P_interaction≥0.11).

DISCUSSION

In a case-control study nested in two population-based cohorts of female nurses, we observed statistically significant inverse associations between plasma PFOA, PFDA and PFOS concentrations and risk of CD. Blood was drawn a median of approximately 10 years before diagnosis or indexing. No associations were observed between PFAS concentrations and risk of UC. Although we considered our stratified analyses exploratory, the inverse association between PFDA and CD risk appeared strongest among younger women.

Data from animal models and epidemiologic studies have implicated PFAS exposure in a variety of adverse outcomes including thyroid dysfunction, obesity, cancers, and reproductive or developmental harm.³⁹ The European Food Safety Authority recently conducted a detailed evaluation of the risks of PFAS exposure to human health and concluded that immunotoxicity was the critical risk at background levels of exposure.⁴⁰ An association between PFAS exposure and IBD was first suggested by a study conducted among residents of the Mid-Ohio Valley, West Virginia, who were exposed to high levels of PFOA through contaminated drinking water.⁴¹ A community cohort of 28,541 adults was combined with a cohort of 3,713 workers who had been employed at a local PFOA production plant, the source of the environmental emissions.¹⁸ Yearly serum PFOA was estimated based on residential address, timing and amount of PFOA discharged, and prevailing environmental conditions at the time. In a retrospective time-event analysis, cumulative PFOA exposure was associated with risk of UC, but not CD. Compared to the lowest quartile of predicted PFOA concentration, the highest quartile had a risk ratio (RR) for UC of 2.96 (95%CI, 1.65-4.96; P_trend <0.001).¹⁸ A similar association with UC was observed in a subsequent analysis restricted to the worker cohort.¹⁷ In contrast, no convincing association between PFAS exposure and IBD was detected in a Swedish study conducted in the Ronneby cohort,¹⁹ where elevated PFAS exposure had occurred in the past through contaminated drinking water.⁴² In a sub-study, serum PFAS concentrations and fecal biomarkers were measured in 189 individuals from Ronneby and a control municipality. No association was detected between serum PFAS concentrations and fecal zonulin or calprotectin, both markers of mucosal inflammation.¹⁹ To our knowledge, only one other study has examined IBD risk among individuals with PFAS exposure levels similar to the general population. In a case control study based on an ambulatory clinic population, serum PFAS concentrations were measured in blood samples from 114 patients with UC, 60 patients with CD, and 75 unaffected family members or friends. Blood was drawn within one year following diagnosis for cases and most patients were in the pediatric age range. An association between serum PFOA concentration and UC was observed (OR for 1 unit log increase was 2.00; 95% CI, 1.08-3.67), but no monotonic increase in risk was detected over quintiles of serum PFOA. For PFHxS, PFNA, and PFOS, average serum concentrations were actually lower in UC cases compared to controls.²⁰

There are several reasons why our results may differ from the studies described above. Plasma PFAS concentrations in our study participants were considerably lower than in the highly exposed populations previously studied. For example, median PFOA concentrations in the Mid-Ohio Valley population cohort were approximately 6 times higher than in our study and 80 times higher in the worker cohort.¹⁸ It is therefore possible that a positive association with UC is only apparent with very high cumulative PFOA exposure. Compared to the hospital-based pediatric case control study, our participants were considerably older. Nevertheless, our interaction analyses suggested that the inverse association between PFDA and CD risk was stronger among younger women and no statistically significant difference in risk according to age group was observed for UC.

There are biologically plausible mechanisms through which PFAS exposure may influence IBD risk. Evidence from animal models and epidemiologic studies suggests that PFOA and PFOS exposure impairs adaptive immunity and suppresses T-cell dependent antibody responses.^43-46 PFOS and PFOA-associated immunotoxicity may be mediated, in part, by activation of peroxisome proliferator activated receptor alpha (PPARα).⁴⁴ PPARα agonists are generally considered to exert anti-inflammatory effects,^47,48 and experimental PPARα activation has been reported to reduce colitis severity in a mouse model of UC.⁴⁹ Indeed, PPARs are considered potential therapeutic targets for IBD.⁵⁰ Acute and chronic PFOS exposure has been demonstrated to shift the balance of T_H1/T_H2 responses in mice,^51,52 and has been hypothesized to influence macrophage phenotype.¹⁸ Furthermore, PFOS concentrations have been found to be inversely associated with multiple sclerosis, an immune-mediated disease which shares epidemiologic and immunologic similarities with CD.⁵³ Data on the influence of PFAS exposure on gut mucosal immune function are limited. In the mouse Citrobacter rodentium model of intestinal immune responses, PFOS exposure augmented IL-22 production by innate lymphoid cells, limiting growth of the C. rodentium. However, in later phase infection, PFOS exposure contributed to mucin depletion, dysbiosis, and failure to clear the pathogen.⁵⁴ Thus, some PFASs may have pleiotropic effects on immune function that are relevant to the pathogenesis of IBD.

Our study has several strengths. Cases and controls were selected from two population-based cohorts, minimizing potential selection biases. All IBD cases were confirmed by record review, thereby reducing outcome misclassification. Controls were carefully matched to cases on age and reproductive factors, and we were able to additionally control for multiple potential confounders, such as parity, diet quality, smoking, and physical activity, using prospectively collected data. Blood samples were drawn many years before IBD diagnosis or indexing, mitigating reverse causality. We were also able to demonstrate the robustness of our results by excluding cases diagnosed within first two years of blood draw. We recognize that our study also has limitations. Identification of IBD cases relied on self-reporting. It is possible that CD ascertainment was not as complete in the NHS in the 2-4 years following blood collection, before to CD diagnoses were specifically queried. We do not expect the likelihood of reporting a diagnosis of IBD to be associated with exposure status. Furthermore, while legacy PFAS concentrations in blood have fallen over time,¹⁶ cases and controls were matched on age and time of blood draw. We therefore do not believe that case ascertainment method would have generated bias.

In common with prior studies of PFAS exposure and IBD risk,^17,18,20 we measured circulating PFAS concentrations. The relationship between blood PFAS levels and concentrations in the mucosal immune compartment is unclear. Nevertheless, based on dosing studies in animals, we anticipate that plasma PFAS concentrations are likely to correlate strongly with relevant tissue concentrations.⁵⁵

As with all observational analyses, we cannot exclude the possibility that unmeasured or residual confounding biased our findings. Our analysis was limited to NHS and II participants with available plasma, and power was limited for subgroup analyses, which we considered exploratory. Findings from our study population of predominantly white, female nurses, may not be generalizable to men and racially or ethnically more diverse populations.

In conclusion, our findings do not support the previously reported association between high PFOA exposure and risk of UC. Indeed, our data suggest that, at background levels of exposure, inverse associations may exist between concentrations of several major PFASs and CD risk. Further population-based studies would be valuable in confirming these associations.

DATA SHARING

Further information including the procedures to obtain and access data from the Nurses’ Health Studies is described at https://www.nurseshealthstudy.org/researchers (contact nhsaccess@channing.harvard.edu).