Fertilizer Type Affects Stable Isotope Ratios of Nitrogen in Human Blood Plasma—Results from Two-Year Controlled Agricultural Field Trials and a Randomized Crossover Dietary Intervention Study

The stable nitrogen isotope ratio δ15N is used as a marker of dietary protein sources in blood. Crop fertilization strategies affect δ15N in plant foods. In a double-blinded randomized cross-over dietary intervention trial with 33 participants, we quantified the effect of fertilizer type (conventional: synthetic fertilizer and organic: animal or green manure) on δ15N in blood plasma. At study baseline, plasma δ15N was +9.34 ± 0.29‰ (mean ± standard deviation). After 12 days intervention with a diet based on crops fertilized with animal manure, plasma δ15N was shifted by +0.27 ± 0.04‰ (mean ± standard error) compared to synthetic fertilization and by +0.22 ± 0.04‰ compared to fertilization with green manure (both p < 0.0001). Accordingly, differences in the δ15N values between fertilizers are propagated to the blood plasma of human consumers. The results indicate a need to consider agricultural practices when using δ15N as a dietary biomarker.

evaluated by plotting and by calculating and testing Pearson's correlation. The source of any deviation was investigated and addressed in a sensitivity analysis; results from the modified model were compared with those of the original model, and any major deviations reported. If no major deviations were observed, only the results from the main model representing the entire study population were reported.
Crops: Three main effects, production system (CON, OA, OB), year (2007,2008), and field replicate (1, 2), without interactions, were used to model crop stable isotope ratios. Crop type was specified as a random effect, which implies that the crops included were regarded as representing a larger population of crops that could potentially be included in diets. For δ 15 N, faba beans were excluded from this model due to a different mode of nitrogen uptake compared with all other crops included (atmospheric fixation compared with uptake from soil). For δ 15 N and δ 13 C, heterogeneity of variance between crops was included in the statistical model.
Intervention diets: Four fixed effects, system, year, field replicate, and menu, without interactions, were used to model the stable isotope composition of the intervention diets. For 2007, pooled samples of menu 1 and 2 were analyzed, so menu was included as a numerical covariate with a value of 1 for menu 1 (in 2008), 2 for menu 2 (in 2008), and 1.5 for pooled menus (in 2007). Sample ID was specified as random factor, to account for the technical-analytical duplicate.
Blood plasma: Five fixed effects, i.e., system, year, field replicate, day (0, 12), and carryover, and the interaction (system * day), were used to model the stable isotope composition in blood plasma. Carryover was a composite factor with seven levels, describing the position of the intervention in the sequence (1 st , 2 nd , or 3 rd ) and, for the on the shift in the plasma stable isotope composition from day 0 to day 12. A test of this interaction corresponds to a test of the main hypothesis of this study.
Due to an expected autoregressive element in the data and varying length of washout periods, a first-order continuous autoregressive (CAR(1)) covariance structure with constant variances was expected a priori to best describe the data. A covariate representing the chronological time since study baseline ("chron") was included in the model for fitting the CAR(1) covariance structure. This structure was compared with three alternative structures: compound symmetry (CS), first-order autoregressive (AR(1)), and unstructured (UN).
Heterogeneous variances with respect to day were considered; plausibly, the variance could be smaller on day 12 compared to day 0, because study participants consumed more similar diets during the 12-day intervention than habitually or during washout periods. All combinations of covariance and variance structures were considered.
As random factor, the subject ("subject") was included.
For δ 15 N, the selected covariance structure was CAR(1) with homogeneous variances. Plotting fitted values against standardized residuals clearly identified one outlying subject. All model assumptions tested were fulfilled upon removal of this subject from the model. This removal had only minor effects on the estimates of coefficients and on statistical significance.
For δ 13 C, the selected covariance structure was CAR(1) with heterogeneous variances with respect to day. A deviation from normal distribution of error terms was indicated, apparently due to day 0 values for certain subjects. A slight negative correlation between random effects and error terms was also indicated (r=-0.15, p=0.041). Removing five subjects with clearly outlying data points in the quantilequantile-plot led to normally distributed error terms, but a slight negative correlation between random effects and error terms remained (r=-0.21, p=0.0053). Removing these five subjects from the dataset had only minor effects on the estimates of coefficients and on statistical significance. Any vs first intervention range -0.07 to -0.12 range -0.08 to -0.15 Stable isotope ratios as a function of production system x day interaction, production system, year, field replicate, day, and a carryover factor. The carryover factor has seven levels. A pairwise comparison of all these is not of interest, so only the study baseline is compared with all other levels. CON= conventional, OA and OB = organic fertilized with pig manure and legumes, resp. SE -standard error