Of the 101 children (51 boys, 50 girls) examined, 21 were formula-fed, and 80 were breastfed for at least 4 months, including 27 children from a dioxin hotspot outside Berlin (former copper smelter without filter system). None of the children were subsequently excluded from the study. The median age of the children was 351 days (50.1 weeks, range 48.7–52.7 weeks). Basic characteristics and body measurements for the formula-fed and breastfed children are summarized in Table 1. The distribution of body measurements in the study corresponded to expectations for healthy German children, with all anthropometric data between the 3rd and 97th percentiles, except for one breastfed girl with a body weight of 7.30 kg at the age of 11.4 months slightly below the 3rd percentile.
The 21 ‘formula-fed’ infants were never breastfed (n = 11) or breastfed for maximally 2 weeks (n = 10). The 80 breastfed children had a mean duration of exclusive breastfeeding of 5.5 months. Thirty of them were exclusively breastfed for 6 months, and eight for longer than 6 months (max 10.8 months). Twenty-nine children were still partially breastfed at the time of the examination. As an estimate for the total transfer of milk and contaminants from mother to child, the equivalent duration of exclusive breastfeeding was calculated from the information at which time each meal of mother’s milk was replaced by complementary foods (e.g. 2 months with three breastfeeding meals and three other meals were calculated as 1 month of exclusive breastfeeding). This value corresponds to a duration of breastfeeding with abrupt weaning and was on average 7.4 months (= median, range 3.6–11.1 months). Thirty-six mothers had already breastfed at least one other child, on average with a total equivalent duration of exclusive breastfeeding of 7.3 months for all siblings (range 1.8–18.5 months). Compared to the children from Berlin, the 27 children from a dioxin hotspot outside Berlin were breastfed significantly shorter in terms of the equivalent duration of exclusive breastfeeding (mean 6.6 vs. 7.8 months in breastfed children from Berlin) and the time of weaning. Anthropometric data of the children were not significantly different in the two regions.
Signs of infection and sensitive CRP
All children were healthy (i.e. good general health condition, no fever, no illness specific symptoms) apart from minimal signs of airway infections (e.g. sneezing or coughing not more than twice within one hour, slight nasal obstruction without runny nose) observed in 42 children. As expected, the number of previous infections (up to 10) was mainly influenced by the number of siblings in toddler age (median: three airway infections without and six with one or more siblings). Only a relatively small number of 12 children already attended a kindergarten or a childminder. There were no significant group differences between breastfed and formula-fed children with regard to the infections they had suffered until then.
The CRP values determined by routine methods for the identification of children with inflammatory processes were all below the usually stated limit of 6 mg/L. Using a more sensitive method, sCRP revealed values below the detection limit of 0.2 mg/L in 33 children and a median of all children of 0.26 mg/L. Twenty-two children were above the triple value of the median of 0.75 mg/L (maximum 8 mg/L). Surprisingly, even at this low level, a partly highly significant association of sCRP values with parameters of the acute phase reaction was found: lower prealbumin, retinol, iron, and free thyroxine, higher thyroxine-binding protein (Abraham et al. 2003). The reason for higher sCRP values is presumably the activation of granulocytes caused by bacterial superinfections often developing in small children with viral airway infections. This was not apparent from the clinical investigation: 33% of children with and 14% of those without mild signs of infection had sCRP values above 0.75 mg/L.
Reproducibility of results for biological parameters
The second blood sampling carried out in 36 children on average 30 days after the first one allowed repeating measurements of laboratory parameters to check their biological reproducibility over time. This information was rated relevant as it makes sense to consider an impact of a persistent contaminant on a biological parameter only if the parameter has a low intra-individual variability within weeks, as compared to the inter-individual variability. As a measure of reproducibility, the average absolute difference between the two measurements was divided by the standard deviation (SD) of the parameter calculated from the measurement in all children. These values are compiled for all parameters in the project report (Abraham 2000). For the parameters identified in the following as possibly influenced by the levels of PFASs, low intra-individual variability was observed for specific antibody levels (between 19 and 26% of SD, due to huge inter-individual variability) as well as for the lymphocyte subpopulation CD27- as percentage of CD8 cells (22% of SD). A higher intra-individual variability was found for CD45RO+CD45RA− as percentage of CD8 cells (60% of SD). The value for the IL-10 production of ex-vivo lymphocytes after PHA stimulation was one of the highest (89% of SD).
Levels of contaminants
Levels of the contaminants measured are given in Table 2 for the 21 formula-fed and the 80 breastfed children. PFOA and PFOS were quantifiable in all the 101 children, whereas in case of PFHxS and PFNA, one and 28 samples were below the LOQ, respectively. The other five PFASs (PFBS, PFHxA, PFDA, PFDoDA and ADONA) were found to be exclusively or predominantly below the LOQ. Of the lipophilic contaminants already measured at the end of the 1990s, 22 compounds were quantifiable—with a few exceptions—in all breastfed children: twelve 2378-substituted PCDD/Fs (calculated as I-TEq value), four coplanar/mono-ortho PCBs (No. 126, 169, 118, 156, considered together with PCDD/Fs as WHO-TEq (1998)), three ndl-PCBs (No. 138, 153, 180, considered as sum) as well as ß-HCH, HCB and pp-DDE. Furthermore, the levels of mercury, cadmium and lead are compiled in Table 2.
The correlation of PFASs with the equivalent duration of exclusive breastfeeding was highest for PFOA (r = 0.68), as displayed in Fig. 1, followed by PFNA (r = 0.63), PFOS (r = 0.59), and PFHxS (r = 0.32). The high correlation coefficients were comparable to those of the lipophilic contaminants ndl-PCBs (r = 0.72), ß-HCH (r = 0.70), pp-DDE (r = 0.64), and I-TEq (r = 0.62). The coefficients for the correlations of the different contaminants among each other are given in the supplemental Table S1. For PFOA, highest correlation coefficients were observed with ndl-PCBs (r = 0.72), PFNA (r = 0.72), I-TEq (r = 0.67) and PFOS (0.67).
Depending on the degree of accumulation, an inverse association between the maternal levels about 1 year after delivery and the equivalent duration of exclusive breastfeeding was found for PFOA (r = − 0.48), as also displayed in Fig. 1, followed by PFNA (r = − 0.32), PFHxS (r = − 0.28), and PFOS (r = − 0.18). Therefore, the levels of the mothers who did not breastfed the child (n = 21) are expected to be the best measure for the background levels of PFASs at the end of the 1990s in Germany, and were 4.9 ± 1.5 µg/L for PFOA, 17.2 ± 6.1 µg/L for PFOS, 1.8 ± 0.9 µg/L for PFHxS, and 0.4 ± 0.2 µg/L for PFNA (mean ± SD). Detailed data on the different POP kinetics will be published in a separate paper.
Evaluation of the vaccine antibodies anti-Hib (IgG), anti-tetanus (IgG and IgG1) and anti-diphtheria (IgG) in the 1-year-old children was restricted to those vaccinated two or three times against Hib (n = 98) and tetanus/diphtheria (n = 100). Tetanus IgG antibodies were measured as IU/mL (mean 0.97 IU/mL, median 0.50 IU/mL, range 0.03–18.4 IU/mL). Tetanus IgG and IgG1 antibodies strongly correlated (r = 0.992); therefore, the following evaluations were made for tetanus IgG1 antibodies only. Basic data of Hib, tetanus IgG1 and diphtheria antibodies are compiled in Table 3. The time since the last vaccination was 23 ± 7 weeks (mean ± SD, range 2–36 weeks) for all these vaccinations.
Levels of vaccine antibodies were strongly right skewed and were therefore log-transformed (base 10) for further statistical analyses. Linear regression analysis revealed a strong influence of the time since last vaccination (Fig. 2, roughly indicating half-lives between 5 and 7 weeks on average for the three antibody levels), and a much smaller but significant influence of the number of vaccinations in case of tetanus only. For the subsequent evaluations, the antibody levels were adjusted for the time since last vaccination and additionally for the number of vaccinations in case of tetanus only.
Correlation between adjusted antibody levels and PFOA levels revealed significant associations for Hib (r = − 0.32, p = 0.001), tetanus IgG1 (r = − 0.25, p = 0.01) and diphtheria (r = − 0.23, p = 0.02). In contrast, no significant associations were observed in case of PFOS for Hib (r = − 0.05, p = 0.66), tetanus IgG1 (r = − 0.07, p = 0.52) and diphtheria antibodies (r = − 0.02, p = 0.84). A further adjustment (Spearman partial correlation) for the equivalent duration of exclusive breastfeeding revealed no relevant influence of this parameter. There were no significant correlations of levels of PFHxS and PFNA with levels of the vaccine antibodies (data not shown).
To quantify the PFOA concentration-dependent lowering of adjusted antibody levels, individual data are displayed for the three antibodies in Fig. 3a–c, together with the moving average revealing a ‘decrease’ roughly above concentrations of 15 µg/L. As these results support an interpretation of an effect occurring above a certain PFOA level, different approaches were used to estimate a NOAEC and the effect size above this level. Fitting a function consisting of two linear segments, a horizontal segment and—after a sharp bend—a decreasing segment (‘knee’ function, see “Methods”), revealed ‘knee’ levels (corresponding to NOEACs) for the different vaccinations between 12.2 and 16.9 µg/L, and PFOA levels between 24.7 and 29.7 µg/L corresponding to an average ‘decrease’ of antibody levels by one standard deviation of the values of children with PFOA levels below the ‘knee’ levels (see Fig. 3 and Table 3).
As can be seen especially from the individual data for antibodies against diphtheria in Fig. 3c, a higher proportion of the adjusted levels (i.e. regressed values at the time of vaccination) was below a certain level at higher PFOA levels: of the 11 children below a limit of 1 IU/mL (0 on the log-scale), just one child had a PFOA level below the ‘knee’ at 16.2 µg/L, whereas 10 children had higher PFOA levels.
To allow viewing on the data from a different perspective, quintiles and deciles for the three levels of adjusted antibodies in relation to the PFOA and PFOS levels are displayed as boxplots in supplemental Figure S1. To illustrate normal distribution within the PFOA quintiles, cumulative distribution curves are presented in supplemental Figure S2. NOAECs (between 18.9 and 22.4 µg/L) were also derived as PFOA mean of the quintiles/deciles below the one with significant difference from the first quintile/decile (t-test, Table 3). Testing was only done in case of a significant ANOVA, i.e. for PFOA and all three vaccine antibodies (for details see supplemental Table S2); in case of PFOS, ANOVA analyses did not reveal significant differences for the quintile/decile distributions. Comparing the means of adjusted antibody levels (i.e. the regressed values at the time of vaccination after back transformation to the linear scale) for the PFOA quintiles Q1 and Q5, a reduction on the linear scale by 86, 54 and 53% was observed in case of Hib, tetanus and diphtheria, respectively.
Regarding the other contaminants measured (Table 2) possibly acting as important confounders, correlation of antibody levels (adjusted for the time since last vaccination and the number of vaccinations in case of tetanus) with the contaminant levels revealed significant associations for I-TEq (r = − 0.22, p = 0.03) and ndl-PCBs (r = − 0.23, p = 0.02) in case of Hib only, likely resulting from the high correlation of these contaminant groups with PFOA (see above and Table S1); a multivariate analysis using the function ‘stepAIC’ in R with inclusion of I-TEq and ndl-PCBs in addition to PFOA and PFOS revealed a (highly significant) influence of PFOA only. No significant associations were observed between other contaminants and other antibody levels.
Immune parameters directly related to the vaccine response
Besides antibody production, also the production of IFNɣ and the proliferation of ex-vivo lymphocytes after stimulation with the toxoids were available for the evaluation of the immune response to tetanus and diphtheria vaccination. These three independent biological parameters revealed significant correlations between each other (at least r = 0.28) for tetanus and diphtheria, respectively, with highest correlation coefficients for production of IFNɣ and proliferation (r = 0.65 for tetanus and r = 0.64 for diphtheria), both determined in the same cell culture. Levels of PFOA significantly correlated with the production of IFNɣ of ex-vivo lymphocytes after stimulation with tetanus (r = − 0.33, p = 0.01) and diphtheria (r = − 0.24, p = 0.08) toxoid (n = 55 only, see also Table 3). For the PFOS levels, associations were missing (r = − 0.10 for tetanus) and or weaker (r = − 0.21 for diphtheria). Comparing the means of IFNɣ levels for the PFOA tertiles Q1 and Q3 (n = 18 each), a reduction by 64 and 59% was observed in case of tetanus and diphtheria, respectively. No relevant associations were observed between PFOA/PFOS and lymphocyte proliferation after specific and unspecific stimulation (Table 4).
Other immune parameters
Correlation coefficients for the various biological parameters measured are compiled in Table 4. No significant influence was observed for the main groups of immunoglobulins (IgG, IgA, IgM, IgE) and IgG subclasses. The same holds true for white blood cell count, lymphocyte main populations (T cells, CD4/CD8 cells, B cells, NK cells), ex-vivo lymphocyte proliferation after unspecific stimulation, and granulocyte function.
Regarding the cytokine production of ex-vivo lymphocytes after unspecific stimulation with PHA, a significantly inverse association with PFOA was found for IL-10 (but not for IL-5 and IFNγ, Table 4). However, this parameter was found to have a low biological reproducibility (see above). Of the lymphocyte subpopulations investigated, significantly positive correlations with levels of PFOA were found for CD45RO+CD45RA− cells (r = 0.23) and for CD27− cells as percentage of CD8+ T cells (r = 0.27), however, for the latter, the correlation coefficient was below 0.2 after adjustment for levels of pp-DDE. Both surface markers were found to correlate between each other (r = 0.34).
Infections and PFOA/PFOS levels
The correlations of anamnestic items regarding infections in the first year of life with PFOA and PFOS are compiled in supplemental Table S3, adjusted for the number of siblings and the equivalent duration of breastfeeding. The evaluation revealed no influence of PFOA and PFOS on infections during the first year of life. Furthermore, no significant associations were observed between levels of PFOA/PFOS and the occurrence of atopic skin diseases (data not shown) as well as the levels of CRP (supplemental Table S3).
Other biological parameters: clinical chemistry, red blood count, thyroid status
Three nutrition-depending parameters of clinical chemistry (total protein, albumin, iron) were found to be negatively associated with PFOA, but not with PFOS (Table 4). This is likely due to the high correlation of PFOA with the duration of breastfeeding which is associated with lower protein and iron intake. The lower iron intake by breastfeeding can lead to an iron deficiency, measurable as hypochromic, microcytic anemia with higher red blood cell distribution width (RDW). Indeed, these parameters were also negatively (MCH, MCV) or positively (RDW) associated with PFOA (and in part with PFOS). An adjustment of these correlations for the equivalent duration of exclusive breastfeeding revealed disappearance of the significant associations in all the cases described. The same holds true for the inverse association between bilirubin and PFOA (data not shown).
Regarding other parameters under discussion to be effected by PFOA and/or PFOS, no significant associations were found for cholesterol, liver enzymes or thyroid parameters, apart from a significantly positive association of PFOS and the thyroid stimulating hormone (TSH, Table 4). Pathologically high cholesterol levels of more than 200 mg/dL were found in six children (5.9%, range 208–309 mg/dL) and are likely to be genetically determined; no associations of these values with PFOA/PFOS levels were observed. An exclusion of these children did not change the missing correlation of cholesterol and PFOA/PFOS. Values of cholesterol (mean ± SD) for PFOA quintiles Q1 and Q5 were 151 ± 19 and 153 ± 39 mg/dL, respectively.