The participants (infants/toddlers [n = 20]) in this study were recruited from the ongoing study “A phase 2, single-centre, double blind, randomized, placebo-controlled study testing the primary prevention of persistent asthma in high risk children by protection against acute respiratory infections during early childhood using OM-85” (OMPAC). The inclusion/exclusion criteria for the OMPAC study are shown in Table S1.

For the analysis of the long-term variation in K_fb, participants were asked to donate samples twice for the present study. The first sample collection occurred in April 2014 while the second occurred in September 2014. At each sampling time, one plasma sample of ~0.5 mL and faecal samples from 1 to 2 bowel movements (~2 g dry weight [dw]) were collected from each participant, and a questionnaire regarding living and eating habits was completed by their parents. Participants were not required to fast before the blood samples were conducted. Thirteen participants completed two sample periods. Four participants provided a sample only in the first sample period, while three participated only in the second.

If a faeces sample was smaller than 1 g dw, it was combined with the other faeces sample from the same sampling day prior to extraction. Otherwise the two faeces samples from the same sampling day (from participants No. 5, 20 and 22 from the first sampling, and participants No. 2, 3, 8, 10, 15 and 16 from the second sampling) were analysed separately to provide information regarding the short-term variation in K_fb.

This study received ethics approval by the University of Queensland Ethics Committee (approval number H/308NRCET/00) and Children’s Health Services Queensland Human Research Ethics Committee (HREC reference number: HREC/12/QRCH/119). The Centers for Disease Control and Prevention (CDC) were determined not to be engaged in human subject research since no personally identifiable information was made available to CDC researchers.

Blood was collected into a polypropylene tube with 100 units of preservative free heparin. It was then processed as follows to obtain plasma: (i) centrifuging at 700 ×g for 10 min at room temperature; (ii) collecting the plasma layer down to within 0.5–1.0 cm of the top of the red cell layer (avoiding disturbance of white cells) using a sterile transfer pipette; (iii) centrifuging the plasma again at 700 ×g for 10 mins to remove platelets; (iv) transferring the plasma sample into a brown Eppendorf tube and storing at −80 °C.

Faecal samples for each participant were collected from one to two defecation events directly before and after the blood sampling. At each sampling point, only faecal material that had not been in contact with the inner liner of the diaper was transferred to a sheet of aluminium foil. The folded aluminium foil sheet was then sealed inside a plastic resealable bag and stored in a freezer at the participant’s home until transportation to the laboratory (on ice, in a cool bag). A detailed description of faecal sample collection and transportation has been provided previously (Chen et al., 2015). Faecal samples were stored at −20 °C until analysis.

Analysis included four chlorinated pesticides (p,p′ DDE, HCB, β-HCH, γ-HCH), five PCB congeners (PCB28, 118, 138, 153 and 180) and five PBDE congeners (PBDE 47, 99, 100, 153 and 154) in plasma and faecal samples. All brown Eppendorf tubes containing plasma samples were sent to the Centers for Disease Control and Prevention (CDC), USA, on dry ice for analysis. POPs were measured in individual plasma samples using a methodology published previously (Jones et al., 2012; Sjodin et al., 2004). POPs were measured in faeces using a previously described method (Chen et al., 2015). Detailed description of the methods can be found in the Table S2.

Quality control procedures for plasma and faeces samples included method blanks and spiked samples. Additional details of quality control procedures and determination of limits of detection (LOD) and quantification (LOQ) are provided in Table S3. The average recoveries of the labelled standards ranged from 61% to 93% for all chemicals (Table S4). Only results that were above the LOQ were used in the regression analysis. A more conservative criterion for quantification was used for faecal samples because the variability in matrix properties contributes to a higher uncertainty in the measurements. For summarizing data, results for plasma below the LOQ were imputed as LOQ/(2)^1/2; results for faeces below the LOD were imputed as LOD/(2)^1/2 and for faeces between the LOD and the LOQ were imputed as the average of LOD and LOQ.

Concentrations of target POPs in faeces were determined on both a dry-weight (dw) and lipid-weight basis. In those cases where two samples from the same day were analysed (Table S5), the dry-weight based concentrations of POPs in a given individual were more consistent than lipid-weight based concentrations (paired t-test done on percent differences of both lipid-based and dry-weight based data, P = 0.005, n = 38). Therefore, the dry-weight based concentrations in faecal samples were used for the data analysis. In the following text, C_b and C_f represent the lipid-weight normalized concentration in plasma and dry-weight normalized concentration in faeces, respectively.

The K_fb was calculated according to the following formula: Kfb=Cf(ng/gdw)Cb(ng/glipid)

Due to the varying number of samples that were collected from each participant, the data from one sampling time point from an individual (who participated at both sampling time points) was randomly chosen to form a restricted cross-sectional dataset (n = 20) for the linear-regression analyses of C_b and C_f. Short-term (< 24 h) variability in K_fb was assessed using data from a sub-group of participants (n = 7) who provided two faecal samples within a 24-hour period. For each of these individuals, the same C_b was used to derive both K_fb values. The intra-individual variance in K_fb (the percent difference between the two values) was then calculated as well for this sub-group. To assess the relative contributions of intra-individual variation and inter-individual variation to short-term variability in K_fb, the intra-class correlation coefficient (ICC) was determined. ICC is calculated as the ratio of the between-individual variance to the total variance (sum of between- and within-individual variance) in a set of data with repeated measures for individuals. ICC values provide a measure of reproducibility for repeated measures and vary between 0 and 1, with 0 indicating no reproducibility and 1 indicating perfect reproducibility. The relative contributions of intra- and inter-individual variation to long-term variability in K_fb was also assessed by determining the ICC in a subgroup of participants (n = 13) who provided faecal and blood samples at two sampling periods (~5 months apart).

Linear-regression analyses of the relationship between C_b and C_f, and ANOVA of K_fb among chemicals were performed using GraphPad Prism 6 (GraphPad Software, Inc., USA). The intra-class correlation coefficient (ICC) was determined using SPSS Statistics 23 (Chicago, USA). Other statistical evaluation of the analytical and demographical data was performed using Microsoft Excel 2013 (Microsoft, Redmond, WA, USA).

The participants were evenly distributed in terms of sex and delivery type. The majority of participants (60%) were weaned before the first sampling period. The average age of the participants was 13 months (ranging from 5.6 months to 24 months) for the cross-sectional dataset. One participant took antibiotics within a week before or after sampling, and one participant took antibiotics on the sampling day. No bowel complaints were reported for any participant on the sampling days. Additional details regarding the participating infants and toddlers are presented in Tables S8, S9, and S10.

The cross-sectional dataset (n = 20) that was used for the linear-regression analyses is summarized in Table S11. The C_b of four chlorinated pesticides (p,p′-DDE, HCB, β-HCH and γ-HCH), five PCB congeners (PCB28, 118, 138, 153 and 180) and five PBDE congeners (BDE 47, 99, 100, 153 and 154) from 20 infants/toddlers are summarized in Table 1 (more details are summarized in Table S12). In this cross-sectional dataset, p,p′ DDE was the only compound quantified in all the plasma samples. HCB, PCB118, 138, 153 and 180 were quantified in 55–75% of the samples. Other analysed chemicals were quantified in < 50% of the samples. p,p′ DDE had the highest C_b, followed by HCB and then the PCB congeners. Arithmetic mean values of C_b were similar to mean lipid-normalized concentrations in the pooled serum sample collected in 2006–2007 from Australian infants, except for the BDEs 47, 99 and 100, which had lower concentrations in our dataset (Toms et al., 2009) (Fig. S1).

C_b was not normally distributed for any of the chemicals (D’Agostino & Person normality test). The C_b of p,p′-DDE ranged by more than two orders of magnitude (9.6–3400 ng/g lipid) among participants. The geometric means of C_b of PCB 138, 153, 180 and HCB were 2.7, 3.3, 2.0, and 8.8 ng/g lipid, respectively. Concentrations of DDE, HCB, PCB118, 138 and 153 in plasma are plotted against participant age in Fig. S2.

The C_f of four chlorinated pesticides (p,p′-DDE, HCB, β-HCH and γ-HCH), five PCB congeners (PCB28, 118, 138, 153 and 180) and five PBDE congeners (BDE 47, 99, 100, 153 and 154) in faecal samples from 20 infants/toddlers are also summarized in Table 1. p,p′-DDE, HCB and BDE47 were quantified above LOQ in 90% of faecal samples, followed by PCB118 (80%), PCB138 and PCB153 (both 75%). Other analysed chemicals were quantified in < 50% of the samples. p,p′-DDE had the highest concentration in faeces (mean 1.7 ng/g dw), followed by BDE47 (0.37 ng/g dw) and then PCB 153, HCB and other PCB congeners (0.084–0.20 ng/g dw).

Similar to the results for plasma, C_f was not normally distributed for any of the chemicals (D’Agostino & Person normality test). The geometric mean C_f of p,p′ DDE was 0.61 ng/g dw, while for the PCBs ranged from 0.019 to 0.068 ng/g dw. The geometric mean C_f of HCB was 0.069 ng/g dw, while for the PBDEs it ranged from 0.0019 to 0.20 ng/g dw. The concentrations of DDE, HCB, PCB118, 138 and 153 in faecal samples are plotted for every participant along with their ages in Fig. S3.

As both C_b and C_f are more nearly log-normal than normal, the association between log₁₀C_b and log₁₀ C_f was assessed for p,p′-DDE, HCB, PCB118, 138 and 153 using linear regression (see Tables S13 and S14). The regressions were significant with values of r² ranging from 0.55 to 0.77, except for HCB (r²of 0.022). These relationships are illustrated in Fig. 1.

The K_fb values for all plasma-faeces sample pairs are shown in Table S15. Summary statistics for the faeces:blood concentration ratios (K_fb) determined from the cross-sectional dataset are shown in Tables S16 and 2. The K_fb values were normally distributed for p,p′-DDE, HCB and PCB138 (D’Agostino & Pearson normality test). The mean K_fb values for p,p′-DDE, HCB and PCB138 were 0.0081, 0.024 and 0.018, respectively. The coefficient of variation (CV) of K_fb was 47%, 66% and 57% for p,p′-DDE, HCB, and PCB138, respectively. The K_fb values for PCB 118 and PCB 153 ranged from 0.010 to 0.11 and 0.0072 to 0.073, respectively.

The influence of short-term variation in K_fb was assessed using a sub-group of participants (n = 7) who provided two faeces samples collected successively within a day. These samples were analysed separately. The percentage difference in C_f between successively collected samples was as high as 48% (Table S5). These C_f data were used to calculate two K_fb values for each individual, whereby C_b was assumed to be the same for the two successively collected faecal samples (see Table 2). Two-way random absolute agreement ICC coefficients were calculated for the paired K_fb values for the seven individuals in this subgroup. For HCB and the PCBs, the ICCs ranged from 0.70 to 0.95, indicating that most of the observed variability in K_fb was due to inter-individual differences, while 5–30% was due to short-term variability between the measurements of two successive faecal samples. A lower value of 0.50 was obtained for p,p′-DDE, indicating that half of the observed variability in K_fb was due to short term variability between two successive faecal-sample measurements.

A second sub-group of the participants (n = 13) provided samples collected ~5 months apart, which allowed investigation of the long-term intra-individual variation in K_fb. The percentage difference between the two K_fb values intra-individually varied from 18 to 112% for p,p′-DDE, 39 to 100% for HCB, 19 to 59% for PCB118, 13 to 82% for PCB138, and 0 to 75% for PCB153 (see Table 2). The two-way random absolute agreement ICC values were above 0.84 and 0.85 for PCB 118 and 153, respectively, indicating that ~15% of the observed whole variability in K_fb was due to intra-individual variability. The ICC values for DDE, HCB and PCB 138 were 0.32, 0.50 and 0.47, respectively, indicating that 50% or more of the total variability was due to intra-individual variability for these chemicals.

For the five compounds with high quantification frequencies in both matrices, the log-transformed C_b and C_f values were significantly and linearly correlated for p,p′-DDE, PCB118, 138, and 153. The 95% confidence intervals of the slope of the log-log regression included one for all four chemicals, consistent with a linear relationship between C_b and C_f. For HCB the correlation was not significant. This may be due to the fact that HCB had the narrowest ranges in C_b and C_f and all of the data were close to the LOD/LOQ (Fig. 1). The K_fb value for p,p′-DDE was significantly different from the K_fb values for the three PCB congeners (one-way ANOVA, Friedman test, P < 0.001). This may reflect the different sorptive capacities of the faeces for the different compounds. This indicates that chemical-specific K_fb values are needed for biomo-nitoring purposes.

The utility of faeces as a biomonitoring tool is dependent on having an accurate and precise estimate of K_fb. The overall variability in K_fb was high; the variation of K_fb in the cross-sectional dataset ranged from a CV of 0.46 to 0.70. ICC assessment showed that both short-term intra-individual variability and long-term intra-individual variability contribute significantly to the overall variability in K_fb. The low variation of replicate C_f measurements in the QA sample (a pooled faecal sample with concentrations similar to the real samples; Table S4 and Fig. S5) indicates that the short-term variability is not largely attributable to analytical uncertainty. Rather, short-term variability may be attributable to variability in the digestive process on the scale of hours/meals (because these variables could affect the sorption properties of the lumen contents). This short-term variability could be reduced by pooling samples over longer periods of time. In previous studies with adults, faeces samples were pooled over 3 days, and the inter-individual variability in K_fb was smaller than in this study. For instance, the CV of K_fb for PCB153 was 0.44 for adults (Moser and McLachlan, 2001), while it was 0.70 for infants/toddlers in this study.

The intra-individual variability observed in individuals who provided samples 5 months apart could be due to the short-term variability discussed above. It is also possible that it was partly due to changes occurring over a longer time period. Long-term intra-individual variability was observed in our previous study, where K_fb of PCB153, BDE 47 and p,p′-DDE in a single infant decreased by more than an order of magnitude between the age of 4 months and 12 months. It was hypothesized that the change in diet during weaning changed the properties of faeces and consequently contributed to this decrease (Chen et al., 2016). In the current study, the mean K_fb values of p,p′-DDE declined slightly with age (Fig. S6), but no significant trends with age were found for the other studied chemicals. Meanwhile, the mean K_fb was lower for the weaned group (except for HCB), but the difference was not significant (Fig. S7). There is a good agreement between K_fb for participants that were aged > 8 months in this study and the earlier study (Fig. S8). Variability in K_fb could possibly be reduced by focusing on toddlers with a narrower age range and more clearly defined weaning status.

The ICC analysis indicates that some of the variability in K_fb is due to inter-individual differences. A variety of factors could contribute to inter-individual variability in K_fb, including long-term differences in dietary composition that could influence faecal properties (e.g., high fiber vs. low fiber) or differences in gut microflora. This study was not designed to investigate the sources of inter-individual variability in K_fb. However, our results provided some insight into whether microflora composition has a strong influence on K_fb.. Two participants (No. 15 and 23) had taken antibiotics shortly before the sampling time point. Although this likely strongly affected the gut microflora, the K_fb values were similar to both the values from the second sampling of the same participants and to the values of the other participants (with one exception: participant No. 15 had a markedly lower K_fb for p,p′-DDE). These two cases suggest that microflora composition may not strongly influence K_fb.

Previously, Sahlstrom et al. (2015) investigated the feasibility of faeces for biomonitoring of PBDEs in toddlers. Tri-decaBDEs were determined in faeces from 22 toddlers and compared to matched serum samples. They developed linear regression models of ln transformed C_b versus ln transformed C_f. In contrast to this study, the slopes of their ln-ln regressions deviated markedly from 1 for many of the PBDE congeners, suggesting that C_b was not linearly related to C_f. As the detection frequency of PBDE congeners in plasma was low in the current study, we could not confirm these findings, as we were unable to assess the association of C_b and C_f for PBDEs. On the other hand, the variation in K_fb was similar in the two studies. For instance, the coefficient of variation of K_fb values was 0.39 for BDE47 and 0.66 for BDE207 in Sahlstrom et al. (2015).

The number of participants in this study was relatively small, and in some contexts might be considered a “pilot scale” study. However, the value of this limited sample size must be understood in the context of the ethical and logistical constraints associated with repeated blood collection from infants, and also in the context of the previous research on faeces:blood partitioning of POPs: to our knowledge, no previous study has had > 22 participants, adult or infant, and none has provided repeated evaluations in individuals over time. This study provides insight into the relative importance of intra- and inter-individual variability in K_fb, but it does not identify the factors responsible for these forms of variability. More work is needed to understand this variability, perhaps structured around hypotheses about the relevance of specific factors. As our understanding of this variability grows, there will be a need to test the methodology on more diverse populations and a broader selection of contaminants.

The large proportion of samples with concentrations below the LOQ in plasma was a pronounced limitation in this study. This was partly due to the small plasma volume, which was a consequence of the difficulty in obtaining large volumes of blood from infants and toddlers. This difficulty is indeed a major motivation for this research aimed at developing an alternative biomonitoring method to the analysis of blood, but it also poses a dilemma, constraining the development of alternative biomonitoring methods. Future work could focus on more highly exposed individuals where detection of contaminants in blood may be increased.

Despite these limitations, this work provides insight into the potential of faeces for biomonitoring POPs in infants and toddlers. The presence of strong correlations between C_f and C_b demonstrate the promise of this technique for p,p′-DDE and several PCB congeners. The considerable variability in K_fb indicates that there are limits to the technique as we applied it. However, our analysis of the sources of variability suggests that there are ways to improve the precision and make faeces a more powerful biomonitoring tool.