Early-life exposure to perfluorinated alkyl substances modulates lipid metabolism in progression to celiac disease

OBJECTIVES: Celiac disease (CD) is a systemic immune-mediated disorder with increased frequency in the developed countries over the last decades implicating the potential causal role of various environmental triggers in addition to gluten. Herein, we apply determination of perfluorinated alkyl substances (PFAS) and combine the results with the determination of bile acids (BAs) and molecular lipids, with the aim to elucidate the impact of prenatal exposure on risk of progression to CD in a prospective series of children prior the first exposure to gluten (at birth and at three months of age). METHODS: We analyzed PFAS, BAs and lipidomic profiles in 76 plasma samples at birth and at 3 months of age in the Type 1 Diabetes Prediction and Prevention (DIPP) study (n=17 progressors to CD, n=16 healthy controls, HCs). RESULTS: Plasma PFAS levels showed a significant inverse association with the age of CD diagnosis in infants who later progressed to the disease. Associations between BAs and triacylglycerols (TGs) showed different patterns already at birth in CD progressors, indicative of different absorption of lipids in these infants. DISCUSSION: PFAS exposure may modulate lipid and BA metabolism, and the impact is different in the infants who develop CD later in life, in comparison to HCs. The results indicate more efficient uptake of PFAS in these infants. Higher PFAS exposure during prenatal and early life may accelerate the progression to the disease in the genetically disposed children.


Introduction
Celiac disease (CD) is a systemic immune-mediated disorder, which is triggered by gluten and other prolamins in genetically susceptible individuals1. The frequency of CD has increased in the developed countries over the last decades, implicating the potential causal role of various environmental triggers in addition to gluten. Similarly to CD, the incidence of other autoimmune diseases such as type 1 diabetes (T1D) has also increased during the last decades2,3. CD and T1D share common predisposing alleles in the class II HLA-region4,5, and approximately 10% of patients with T1D also develop clinical CD6,7, while subjects with CD are at-risk for developing T1D before 20 years of age8. The possible triggers that have been indicated to affect the onset of CD include the composition of the intestinal microbiota, infant feeding, and the use of antibiotics1,9,10.
The hygiene hypothesis, stating that a decrease of the infectious burden is associated with the rise of allergic and autoimmune diseases, has also been proposed in CD because several studies have shown that CD is more common in developed countries11-13. Another possible explanation for varying incidence in different populations implicates the role of different infant feeding patterns (including the amount and timing of gluten introduction) in families with low socioeconomic status12. Socioeconomical status (SOS) may also have a broader role in increasing the risk for CD, such as different exposures to environmental pollutants. Higher SOS has been in several studies linked with higher burdens of several persistent organic pollutants (POP), particularly for perfluorinated alkyl substances (PFAS) 14,15. Exposure to phospholipids, e.g., lysophosphatodylcholines (LPCs), phosphatodylcholines (PCs) and sphingomyelins (SMs). Similarly, we have recently identified systematic differences in plasma lipid profiles between children who later progressed to clinical CD during the follow-up, when compared to children who remained healthy5. Importantly, these differences were observed prior to the first introduction of the gluten in the diet and before the first signs of CD-associated autoimmunity. Similar result were recently reported in an Italian cohort study21.
In line with the observations of intestinal dysbiosis in CD1, changes in lipid metabolism5,21 as well as the documented link of CD with liver disorders22, it has been shown that circulating BAs are elevated in CD, including in children23,24. BAs not only facilitate the digestion and absorption of dietary lipids in the small intestine, they are also important metabolic regulators involved in the maintenance of lipid and glucose homeostasis25. BAs are produced in the liver, and their homeostasis is maintained through tightly controlled enterohepatic circulation. Moreover, there is a close interplay between BA and gut microbiota. Gut microbiota is involved in the biotransformation of secondary BAs, while BAs can modulate the microbial composition in the gut due to their antimicrobial activity as well as through regulation of the composition of the intestinal microbiota through farnesoid X receptor (FXR) and Takeda G-protein receptor 5 (TGR-5)26,27. PFAS, on the other hand, have been shown to undergo similar enterohepatic circulation as the BAs, and PFAS can also suppress the BA biosynthesis in the liver28.

Study design
The current study is a part of the Type 1 Diabetes Prediction and Prevention study in Finland (DIPP), which is an ongoing prospective birth cohort study initiated in 1994. In DIPP, parents of newborn children at the university hospitals of Turku, Tampere and Oulu in Finland are asked for permission to screening for T1D-conferring HLA risk alleles in the umbilical cord blood. Families of children identified as having an increased HLA-conferred risk for T1D are invited to join the study. This study included children born at Oulu and Tampere University Hospitals between 1994 and 2003. Our analysis included children born at Tampere University Hospital between August 1999 and September 2005. During that period, 23,839 children were screened at birth for increased risk of T1D, and 2,642 eligible children were enrolled in the follow-up and had at least two visits to the study clinic. These children carry the high-risk HLA DQB1*02/*03:02 genotype or the moderate-risk HLA-DQB1*03:02/x genotype (x ≠ DQB1*02, 03:01, or 06:02). More than 1,200 of these children took part in the DIPP-CD study. Because of the enrolment criteria 60% of the children were male. At each visit, the families were interviewed for dietary changes, infections, growth, important family related issues and the children gave a non-fasting venous blood sample. The children were followed for 4 CD-related antibodies: anti-transglutaminase 2 (anti-TG2), anti-endomysium (EMA), antigliadin (AgA-IgG and AgA-IgA) and anti-reticulin (ARA) antibodies, and for 4 T1D-associated autoantibodies: islet cell autoantibodies (ICA), autoantibodies against insulin (IAA), tyrosine phosphatase-like protein (IA-2A) and glutamate decarboxylase (GADA). IgA deficiency was excluded. All children participating in this study were of Caucasian origin. None of the mothers had CD. The maternal health care system in Finland gives recommendations as to the starting ages for different solid foods for infants and these are generally followed. For this reason, exclusive breast feeding and addition of new food ingredients to the infants' diets were very uniform in this study. The children of the CD follow-up cohort were annually screened for anti-TG2 antibodies using a commercial kit (Celikey Pharmacia Diagnostics, Freiburg, Germany). If a child's sample was found positive, all previous and forthcoming samples were analyzed for the All rights reserved. No reuse allowed without permission. the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint (which was not peer-reviewed) is . https://doi.org/10.1101/2020.04.02.20051359 doi: medRxiv preprint entire set of CD-related antibodies. A duodenal biopsy was recommended for all anti-TG2positive children. If the biopsy was consistent with the ESPGHAN criteria of 1990, a glutenfree diet (GFD) was recommended. None of the children participating in this study had any T1D-associated autoantibodies in any samples during the follow-up. The first exposure to gluten in this study was at the median age of five months ( Table 1).
We have randomly selected 17 children with biopsy proven CD (progressors) and a matched control for each progressor, with the similar risk HLA alleles, born within ±1 month of each other, having had each sample taken within ±1 month of each other and living in the same region of the country throughout the whole follow-up period. Samples at birth (cord blood) and at 3 months of age were analyzed. Altogether, 76 plasma samples from children developing CD and from their HCs were analyzed. The clinical and genetic data of the participants are found in Table 1.
The ethics committees of the participating university hospitals approved the study. Written informed consent was obtained from the parents for HLA-screening, autoantibody analysis and intestinal biopsies.

Analysis of PFAS and BAs
The BAs and PFAS were analyzed using the established method as described previously29. In brief, the sample preparation was done with 25 mg Ostro Protein Precipitation and Phospholipid Removal 96-well plate (Waters Corporation, Milford, USA), using 30 L of serum and a set of BA and PFAS internal standards. Matrix-matched calibration standards were made using newborn bovine serum. Analyses were performed on an Acquity UPLC system coupled to a triple quadrupole mass spectrometer (Waters Corporation, Milford, USA) with an atmospheric electrospray interface operating in negative ion mode. Aliquots of 10 L of samples were injected into the Acquity UPLC BEH C18 2.1 mm × 100 mm, 1.7 m column (Waters Corporation). A trap column (PFC Isolator column, Waters Corporation) was installed All rights reserved. No reuse allowed without permission. the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint (which was not peer-reviewed) is

Lipidomic analyses
The lipidomic analyses were performed as described previously5. The plasma samples (10 L) were extracted using a modified version of the previously published Folch procedure30. The samples were analyzed using an ultra-high-performance liquid chromatography quadrupole time-of-flight mass spectrometry method (UHPLC-Q-TOF-MS from Agilent Technologies (Santa Clara, CA, USA). The analysis was done on an ACQUITY UPLC® BEH C18 column (2.1 mm × 100 mm, particle size 1.7 m) by Waters (Milford, USA). Internal standard mixture was used for normalization and lipid-class specific calibration was used for quantitation as previously described. Quality control was performed throughout the dataset by including blanks, pure standard samples, extracted standard samples and control plasma samples.
Relative standard deviations (%RSDs) for lipid standards representing each lipid class in the control plasma samples (n=8) and in the pooled serum samples (n = 20) were on average 11.7% (raw variation). The lipid concentrations in the pooled control samples was on average 8.4% and 11.4% in the standard samples. This shows that the method is reliable and repeatable throughout the sample set. MS data processing was performed using open source software MZmine 2.1831.

Statistical analyses
All rights reserved. No reuse allowed without permission. the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
Previously published lipidomics dataset5 was obtained from MetaboLights with the study identifier MTBLS729. The data was log-transformed and missing values were imputed by half of the row's minimum. The datasets were auto-scaled prior to multivariate analyses.
To integrate across different data types, we applied sparse generalized correlation discriminant analysis via the DIABLO framework, part of mixOmics package (v6.10.8)33. The method constructs components across different data blocks, by maximizing their covariance with each other and a given response (Y) variable. Heterogenous data such as PFAS, BAs and lipid levels were partitioned into three different blocks and regressed to a binary response variable, i.e., CD progressors (CD) or HCs. Regularized sparse partial least squares discriminant (sPLS-DA)33,34 models were fitted. The optimal number of components that achieve the best performance based on the overall error rate or Balanced Error Rate (BER) were determined.
Block sPLSDA models were developed at two time-points, i.e., at birth (cord blood) and at 3 months of age. Moreover, these models were cross-validated 35 by 5-fold cross-validation (CV) with (N =100 repeats). The final model performances were assessed by area under the curve (AUC), overall misclassification error rate, and BER generated using The key predictors/contributors that are jointly associated with the response variable of interest across all the input data matrices were identified by their Variable Importance (VI) score, i.e (absolute weighted loadings). The direction (up or down) of contribution of a particular predictor was determined by estimating the mean relative log fold change in the CD progressors vs. HCs at a particular time-point.
In order to understand which variables influence the outcomes, multivariate correlations were performed by a method described in33. In addition, bivariate correlation was performed using Pearson correlation. Debiased Sparse Partial Correlation algorithm (DSPC) was used for All rights reserved. No reuse allowed without permission.
the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint (which was not peer-reviewed) is . https://doi.org/10.1101/2020.04.02.20051359 doi: medRxiv preprint estimating partial correlation networks, visualized by the MetaboAnalyst 436 version with cutoff values off correlations between +/-0.22 to 0.75. Univariate analysis (unpaired two sample t-test and paired t-test) using the 't.test' function, was deployed to identify mean differences in the concentration of individual PFAS and BAs between CD progressors vs. HCs at birth and at 3 months of age. Libraries/packages such as 'Heatmap.2', 'mixOmics', 'boxplot', 'beanplot', 'gplot' and 'ggplot2' were used for data visualization.

Levels of PFAS and BAs in the infants
Seven PFAS compounds were detected in the samples, namely PFHpA, PFHxS, PFOA, PFNA, PFOS, PFDA and PFUnDA, both at birth and at three months of age (Supplementary Table S2). PFOS and PFOA had highest concentrations and they were detected in the majority of samples, whilst PFDA and PFNA were detected in less than 10% of the samples and were excluded from further analyses. The levels of PFAS were lower in the cord blood (CB) than at the age of three months ( Figure 1A-C). No significant differences were observed between cases and controls either at birth or at 3 months of age ( Figure 1A-C). Interestingly, the levels of the total PFAS were elevated from birth to 3 months of age was significant only for the CD group (fold change (FC) 3 months vs. at birth = 1.72, p = 0.003 Supplementary Table S3) although also controls had an increasing trend (FC = 1.02, p = 0.35). Among the individual PFAS, PFOA showed nearly 2-fold increase, being significant in both groups ( Figure 1A-C) Table S3).

(Supplementary
In addition to the two primary BAs, we measured five primary/conjugated and secondary BAs (Supplementary Table S2). The levels of secondary BAs were mostly below the detection limits, particularly in the cord blood samples, where the total BA pool consisted mainly of GCA and GCDCA. Similarly, as for PFAS, no significant differences were observed between cases All rights reserved. No reuse allowed without permission. the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint (which was not peer-reviewed) is . https://doi.org/10.1101/2020.04.02.20051359 doi: medRxiv preprint and controls either at birth or at 3 months of age (Figure 1D-F). The BA profiles at birth and at 3 months of age were markedly different in both groups (Figure 1D-F). The total BA pool showed significant, almost 7-fold upregulation (p = 0.0001). Particularly, GHCA and CDCA, which were close the limits of detection at birth, have increased substantially with age.

Associations between PFAS, BA and lipid levels
Partial correlation network analysis between PFAS, BAs and lipids (classes) measured in the cord blood showed different pattern of association in HCs (Figure 2A the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint (which was not peer-reviewed) is . https://doi.org/10.1101/2020.04.02.20051359 doi: medRxiv preprint level ( Figure 3C). Only weak correlation between the levels of PFAS and LPCs was observed while overall the PFAS levels were not associated with either lipids or BAs. In CD progressors, the patterns were different from the controls also at three months of age ( Figure 3D)
Next, we selected those lipids that showed significant differences between CD and HCs in infancy in our previously reported study5. Together these lipids with the BAs and PFAS were subjected to multiblock (MB) analysis. MB analysis identified PFAS, BA and molecular lipid species as predictors (discriminative features), that help to classify CD progression vs. HCs.
The association between top predictors (high variable importance scores) are shown in (Figure 4A, D). All rights reserved. No reuse allowed without permission. the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint (which was not peer-reviewed) is . https://doi.org/10.1101/2020.04.02.20051359 doi: medRxiv preprint In cord blood, the levels of PFOA and PFOS were identified as the top linear predictors of CD progression ( Figure 4B). On the other hand, the levels of primary BAs such as GCA and GHCA are the key predictors of CD progression ( Figure 3C) (Figures 4A and 5A). were identified as key contributors, that were downregulated in the CD progressors as compared to the HCs (Figures 4A and 5A).
At three months of age, concentration of PFHxS was marked as a top predictor associated with CD progression, along with levels of PFOA and PFOS. Moreover, the level of PFHxS was elevated in the CD progressors as compared to HCs (Figures 1A, 4E and 5B). At this age, elevated levels of secondary BAs such as DCA and CDCA were marked as the key predictors for CD progression ( Figure 4F). As stated before, the level of PFOA was negatively correlated with that of DCA ( Figure 2B), presumably, there is an interaction between the PFAS and BAs in the CD progressors. Moreover, PFOA is inversely linked to age of diagnosis of CD ( Figure   2B and Figure 6). Besides, the MB model identified several PCs, SM(d36:2), PC(O-38:4,5) as key predictors that discriminate the lipid profiles of CD progressors from the HCs (Figure 5B). which showed increased LPCs at those 4 month old infants that progressed later to CD21. Furthermore, our results showed that the associations between BAs and complex lipids showed different patterns already at birth in those infants that later developed CD as compared with healthy controls. Particularly, significant correlations were observed between the TGs and specific BAs already at birth in the CD progressors. At three months of age, both groups showed significant correlation between the lipids and BAs, however, the patterns were different. Particularly, at this later time point, CD group showed significant positive correlations between the different classes of TGs while the healthy controls showed the opposite correlations. Thus, the data agree well with our earlier study, in which we observed distinct lipid changes in children that later developed CD, even prior to the first exposure to dietary gluten, particularly in the TG class. The data suggested that the specific TGs, found elevated in CD progressors, may be due to a host response to compromised intake of essential lipids in the small intestine, requiring de novo lipogenesis. The current study suggests that the changes in TG metabolism are also related to alteration in BA metabolism, and that early-life exposure to PFAS may contribute to these changes. Currently, there is very limited amount of data on BA and lipid metabolism in infants. The BA pool in the fetus and in newborns is unique.
Over 90% of fetal BAs are conjugated forms of the primary BAs, cholic acid (CA) and chenodeoxycholic acid (CDCA), because the intestinal bacteria necessary to transform primary BAs into secondary BAs are thought to be absent in utero, or based on recent findings, very different than later in life38,39. The enterohepatic circulation of BAs in utero is also minimal.
Instead, there is a transplacental gradient for BAs in the fetal-to-mother direction, except for secondary and tertiary BAs, which are more abundant in maternal serum40. PFAS exposure has been shown to impact BA metabolism through several mechanisms. BAs and PFAS have similar enterohepatic circulation25,28, and it has been indicated that 7-alphahydroxylase (CYP7A1), which catalyzes the first and rate-limiting step in the classical pathway of formation of BAs from cholesterol, may be down-regulated by PFAS25,41. This may lead to increased re-uptake of BAs, which would generate negative feedback loops via the farnesyl-All rights reserved. No reuse allowed without permission. the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint (which was not peer-reviewed) is . https://doi.org/10.1101/2020.04.02.20051359 doi: medRxiv preprint X-receptor and subsequently reduce their de novo synthesis25,42. It has also been shown that PFOA inhibits the function of the hepatocyte nuclear factor 4α43, which plays a central role in the regulation of BA metabolism in the liver, and is linked both with the synthesis and conjugation of primary BAs. However, in neonates, the alternative (acidic) pathway is the major pathway for bile acid synthesis and this pathway is governed by mitochondrial sterol 27hydroxylase (CYP27A1), which can initiate a process independent of CYP7A1. Only after weaning, CYP7A1 is expressed and the classic pathway becomes the major pathway for bile acid synthesis in adult liver25. There is currently no data on the impact of PFAS exposure on the regulation of the CYP27A1. On the other hand, the impact of PFAS on the TG metabolism may be modulated through the bile acids, as the bile acid receptor FXR has also a regulatory role in triglyceride metabolism44-46. In humans, the recent data suggests that FXR is not activated directly by PFAS44. In the liver, FXR activation, by, e.g., bile acids, would result in downregulation of CYP7A1, which in addition to inhibition of the classical BA synthetic pathway also reduces the expression of several genes mediating free fatty acid synthesis, thereby attenuating de novo lipogenesis25. Thus, activation of FXR modulates free FA oxidation and TG clearance to the circulation.
The acylalkylPCs type of lipids that showed significant association with the CD progression were positively associated with the levels of total BAs and PFHxS, indicating potentially disturbed hepatic synthesis of these lipids, particularly as the intra-lipid regulation between these lipids and particularly TGs was disturbed in CD progressors. These ether lipids, in addition to their structural roles in cell membranes, are thought to function as endogenous antioxidants, and emerging studies suggest that they are involved in cell differentiation and signaling pathways47. Interestingly, these lipids have shown to be endogenous antigens to activate invariant natural killer T cells (iNKT)48, which are subset of innate immune cells.
Recently, therapeutic potential of iNKT cell antigens against autoimmunity have been suggested49. Thus, the reduced levels of the alkyl ether lipids could suggest compromised response to oxidative stress. Interestingly, these lipids have previously been found to be All rights reserved. No reuse allowed without permission. the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint (which was not peer-reviewed) is . https://doi.org/10.1101/2020.04.02.20051359 doi: medRxiv preprint associated with disease development of pediatric CD5,21 and type 1 diabetes50, thus indicating that they may play an important role in the development of an autoimmune disorders. Due to their role in cell membranes, the decrease in ether-linked PCs particularly, PC(O-38:4,5) might also attribute to compromised intestinal permeability in the CD progressors.
Taken together, our results show that PFAS exposure may modulate lipid and BA metabolism, and the impact is different in the infants who develop CD later in life, in comparison with healthy controls. Although we did not observe any significant differences in the levels of PFAS exposure in those children that later developed CD, which may be due to the small sample size, we did observe a significant increase in their levels in progressors to CD from the time of birth to three months of age, suggesting more efficient uptake of PFAS. Furthermore, the age of diagnosis was strongly associated with the PFAS exposure. Our study thus suggests that further investigations of the impacts of exposures to environmental chemicals are merited.   the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

Conflicts of interest
The copyright holder for this preprint (which was not peer-reviewed) is . https://doi.org/10.1101/2020.04.02.20051359 doi: medRxiv preprint mean difference between case-control at a particular time-point was tested by unpaired twosample t test. Mean difference between cord blood and 3 months samples obtain from the same infants were tested by paired t test. CD, HCs and BA denotes CD progressors, healthy controls and bile acids respectively. All rights reserved. No reuse allowed without permission. the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint (which was not peer-reviewed) is . https://doi.org/10.1101/2020.04.02.20051359 doi: medRxiv preprint the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint (which was not peer-reviewed) is . https://doi.org/10.1101/2020.04.02.20051359 doi: medRxiv preprint the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint (which was not peer-reviewed) is . https://doi.org/10.1101/2020.04.02.20051359 doi: medRxiv preprint the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint (which was not peer-reviewed) is . https://doi.org/10.1101/2020.04.02.20051359 doi: medRxiv preprint Figure 5. Key lipid contributors/predictors, that jointly (together with PFAS and BAs), aided in the separation of CD progressors vs. HCs, at birth and at 3 months of age. The direction of regulation of the predictors are shown with light brown and blue colors that corresponds to the CD progressors or HCs respectively; wherever the mean intensity of the predictor is maximum. All rights reserved. No reuse allowed without permission. the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint (which was not peer-reviewed) is . https://doi.org/10.1101/2020.04.02.20051359 doi: medRxiv preprint Figure 6. Correlation of PFOA concentration at birth and the age of diagnosis.
All rights reserved. No reuse allowed without permission. the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint (which was not peer-reviewed) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint (which was not peer-reviewed) is  the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
The copyright holder for this preprint (which was not peer-reviewed) is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.