Changes of trace element status during aging: results of the EPIC-Potsdam cohort study

All the samples (baseline and follow-up) were blinded and mixed in a single large set and were frozen at − 80 °C until 1 day prior to measurement. Same instruments, pipettes, operators, vessels, and chemicals were used for all the samples. Samples were kept at 4 °C until the next day for measurement. For TE-profiling, the method published in the Journal of Trace Elements in Medicine and Biology was employed [29]. In brief, 50 µL of sample was diluted with 440 µl of a diluent solution containing 5 vol.% 1-butanol (99%, Alfa Aesar, Karlsruhe, Germany), 0.05 m % Na-EDTA (Titriplex^® III, pro analysis, Merck, Darmstadt, Germany), 0.05 vol.% Triton™ X-100 (10% in H₂O, Merck-Sigma Aldrich, Steinheim, Germany), and 0.25 vol% ammonium hydroxide (puriss. p.a. plus, 25% in H₂O, Fluka, Buchs, Germany). As internal standard and for isotope dilution analysis 10 µL of a solution containing 50 µg/L ⁷⁷Se (prepared from isotopically enriched ⁷⁷Se standard: 97.20 ± 0.20% ⁷⁷Se; 0.10% ⁷⁴Se; 0.40 ± 0.10% ⁷⁶Se; 2.40 ± 0.10% ⁷⁸Se; 0.10% ⁸⁰Se; 0.10% ⁸²Se as certified by Trace Sciences International, Ontario, Canada, purchased from Eurisotop SAS, Saarbrücken, Germany) and 5 µg/L Rh (diluted from 1000 mg/L single element stock solution, Carl Roth, Karlsruhe, Germany) was added to give a total volume of 500 µL. This solution was directly subjected to analysis via inductively coupled plasma tandem mass spectrometry (ICP-MS/MS) (Agilent ICP-QQQ-MS 8800, Agilent Technologies, Waldbronn, Germany). ICP parameters and monitored isotopes can be found in Supplemental Table 1. The instrument was optimized daily for maximum sensitivity. For external calibration (all elements except Se), standards were prepared matrix-matched in the diluent solution using 1000 mg/L single element stock solutions, purchased from Carl Roth (Karlsruhe, Germany). Se was determined using isotope dilution analysis. For quality control, reference material RECIPE^® ClinChek^® serum control lyophilized (Ref. 8880-8882, Lot 347) (both levels) was measured in triplicate daily. Mean recoveries were Mn: 106.0% ± 8.9%, Fe: 105.4% ± 6.4%, Cu: 101.1% ± 6.4%, Zn: 91.6% ± 8.4%, I: 92.0% ± 9.0%, Se: 76.5% ± 20.3%. Furthermore, sufficient blank samples (distilled H₂O) were carried along to determine limits of detection (LOD, 3ϭ-criterion) and quantification (LOQ, 10ϭ-criterion) on a daily basis. Measurement was done in seven single-day batches with 70 samples each, except for the last one, which only had 54 left.

Across the 220 selected participants, one individual with a concentration of I at the second time point ten times higher than the P99 was excluded from the analysis. The final sample included 131 men and 88 women.

For Mn, 8 concentration values were below the limit of detection (LOD) and 19 below the limit of quantification (LOQ) at baseline, and 5 concentration values were below the LOD and 36 below the LOQ at the second time point. Left-censored data were handled by substituting by LOD/√2 for censored values less than LOD and by LOQ/√2 for censored values less than LOQ. In addition, for all TE, except Se, since it was determined by isotope dilution analysis, data were also missing due to insufficient recovery of the internal standard, indicating strong matrix effects in these samples, thus rendering the obtained values unreliable (10 missing at baseline and 8 at 20 years of follow-up). Missing values were handled using single imputation based on the fully conditional specification method [30].

Baseline, follow-up concentrations and differences for each TE (ΔTE, defined as the differences between the follow-up concentration and the baseline concentration, at the individual level) were analyzed as continuous variables. As most TE variables at the two time points showed non-normality, concentrations were presented as median (IQR), and differences in TE profiles over time were evaluated using Wilcoxon signed-rank test. In addition, medians of the relative differences (IQR), expressed as %, were computed to estimate the effect size of the differences. Serum Cu to Zn ratio as well as serum Cu to Se ratio was calculated as they may constitute novel biomarkers for aging [21] and resistance to thyroid hormone β [31], respectively.

We also calculated the proportion of subjects (in %) with concentrations within normal reference ranges, at the two time points [32‐37]. Of note, reference ranges for Mn vary among laboratories [38]. We further calculated the percentage of participants at both time points with optimal Se status (defined as Se concentrations > 100 µg/L, based on both saturation of glutathione peroxidase activity and selenoprotein P [39]), as well as the percentage of participants with Cu to Zn ratio ≤ 2 (a ratio > 2 expressing inflammatory reaction or inadequate Zn status [40]). The percentages at baseline and follow-up were compared using the matched pair McNemar’s test. P values were adjusted for multiple testing using the Benjamini–Hochberg procedure.

Spearman’s correlations were undertaken for investigating the relationships between changes in individual TE. Associations within differences in TE were further examined using Spearman-adjusted partial correlation coefficients, adjusted for all other TE differences.

Principal component analysis (PCA) was performed on the age-related differences in each individual TE (i.e. 6 variables constituting the absolute differences in concentration of each TE between the two time points), using the PROC FACTOR^® procedure in SAS. This method allows to generate independent linear combinations of the initial variables (here the age-related differences in TE concentrations), thereby maximizing the explained variance. The factors were orthogonally rotated using the ‘VARIMAX option’ in SAS. The number of factors retained was based on eigenvalue > 1, a scree-test, and the interpretability of factors.

For comparability, TE differences as well as continuous independent variables were standardized. Assumptions of linear regressions were checked using diagnostic plots (fit plot, residual plot, and a diagnostics panel from the PROC REG^® procedure in SAS). They were not verified for Mn, I and Se, Se to Cu ratio and, therefore, confidence intervals for the parameters for these TE differences need to be taken cautiously. Multivariable linear regression models were then used to analyze sociodemographic, lifestyle and anthropometric characteristics associated with longitudinal TE variability.

The following baseline factors were included into the models: concentration of the respective TE, sex, educational level (no vocational training/vocational training, technical college, university), age, waist circumference, overall leisure-time physical activity (defined as the sum of sports, biking, and gardening in h/week), and dietary quality (as assessed by the Mediterranean diet score). In addition, for variables available at both baseline and follow-up, change over time was also considered, i.e. change in mineral and vitamin use (non-use/use at baseline/use at follow-up/constant use), change in hypertension status (non-hypertensive/hypertensive/newly hypertensive), change in alcohol consumption (no or low intake/constant high intake/stop drinking/higher intake at follow-up/lower intake at follow-up), change in amount of time devoted to sports, and change in waist to hip ratio during aging. For all statistical analyses, a P value < 0.05 was considered statistically significant. All statistical analyses were performed using the statistical software package SAS (version 9.4, Enterprise Guide 6.1, SAS Institute Inc., Cary, NC, USA), with a significance level of 0.05 for 2-sided tests.

Most studies investigating age-specific differences in TE status have been conducted cross-sectionally [13‐15, 22, 23, 41‐45] and comparison with the present work is, therefore, somewhat limited. However, overall, age-related changes in TE observed in our study were consistent with the literature. Two national surveys have examined predictors of blood Mn concentrations [23, 44]: one was conducted in a representative sample of Korean adults (KNHANES) and the other study used data from the US National Health and Nutrition Examination Survey. In the US survey [23], higher blood Mn concentration was associated with younger age, whereas the highest Mn concentrations were observed for individuals in the 30–39 age range in the Korean study [44]. In a Brazilian study [45] conducted among 947 adults, aged 40 year or older, blood concentrations of Mn were significantly lower with higher age. Possible comparison with our data is limited, as the above mentioned studies differ in the study design, in the population studied (e.g. younger or/and Asian individuals) and more importantly, in the assessment of Mn status. In these studies [23, 44, 45], Mn measurements were based on blood concentrations, and Mn whole-blood concentrations have been shown to be higher than serum Mn concentration as a considerable amount of Mn is bound to hemoglobin in erythrocytes [38]. In line with our findings, in the ZincAge study carried out among 1090 healthy elderly individuals aged of 74 years on average, negative correlations were observed between age and plasma Zn [22]. In that study based on data from five European countries, age was the most important predictor of Zn differences. According to the authors, this difference was due to physiological variations occurring during aging rather than dietary intake (assessed by dietary Zn intake and a Mediterranean diet score). Elevated plasma or serum Cu to Zn ratio has been suggested to represent a marker of health status and a predictor of all-cause mortality in elderly population [40]. Consistent with the literature [22, 40, 42], we observed that Cu to Zn ratio significantly increased with aging. In the present work, elevated Fe concentrations were observed with advancing age. These results contrast somehow with work indicating high prevalence of Fe deficiency anemia in older age [24], but appear in accordance with findings from the US Framingham Heart Study cohort which showed that white Americans aged 67–96 year old were at lower risk of Fe deficiency but had rather elevated Fe stores, based on multiple Fe measures [41]. However, it should be noted that serum Fe is not a good indicator of Fe stores and not a reliable tool for measuring Fe deficiency. Concerning Se status, our results are consistent with some previous studies, observing negative relationships between Se concentrations and age [13, 14, 43], while another study observed this association only in women [15]. In our study, median Se concentrations decreased from 85.19 (17.15) µg/L to 79.28 (17.69) µg/L after ~ 20y. This is consistent with the decline observed in the longitudinal EVA study of French elderly after 9 years of follow-up [46]. Se to Cu ratio has been suggested as a novel sensitive biomarker for resistance to thyroid hormone [31], as it allows to identify subjects with resistance to thyroid hormones due to mutations of the receptors. In our study, Se to Cu ratio significantly decreased over time, from 0.09 to 0.08. We observed herein a slight increase in serum concentrations of I during the period considered (1994–1996; 2004–2006). To our knowledge, no other study has evaluated age-specific differences in I using serum concentrations, limiting the comparability of the present results. Studies are rather based on other indicators such as urinary iodine concentration, or blood concentrations of thyroid stimulating hormone and thyroglobulin [47]. This trend is, therefore, rather difficult to interpret but might be seen in light of the fact that obligatory salt iodization has been stopped just after reunification in East Germany [48], and in turn the urinary I excretion has diminished at the beginning of the 1990s. This, however, needs to be considered with caution as improvement of I supply has been observed as of 1994 and variations existed across East Germany [48]. Of note, the recommendation for I intake does not vary for middle-age adults and older adults, notably because data are insufficient to support the need for deriving specific dietary reference values for older adults [49].

Despite knowing that TE interact with each other and work in parallel, very few studies have looked at the interdependence between TE based on serum concentrations, limiting comparability of the present data with other studies. We investigated herein the correlations between differences in TE concentrations over time. In our study, low-to-mild correlations were observed between age-related differences in TE. The first factor identified by PCA reflected the correlated decline in both Mn and Zn over time while the second factor reflected the observed (on average) increase in both Cu and I over time. It should be noted that the two extracted-factors, based on only six variables, only explained 42% of the variance. Interestingly, increase in Cu concentrations over time were significantly positively correlated with increase in I concentrations. Although this needs further investigation, a possible hypothesis could be variation in food habits between the two time points. Thus, for example, seafood and fish are good sources of both I and Cu, and in turn the latter correlation may reflect changes in food intake over time. However, it cannot be determined whether the factors obtained reflect nutritional or/and metabolism change during aging. Further research is required to identify possible food determinants associated with specific TE patterns.

In the present work, we evaluated a large number of factors in relation to longitudinal TE variability, including factors related to change. Overall, the studied factors (mostly anthropometrics and lifestyle characteristics) were not strong determinants of TE longitudinal variability, except, predictably, due to the regression to the mean effect, the respective TE baseline concentration (higher TE concentration at baseline were related to higher decrease over time). Compared to men, we observed that women had lower increase in Cu and higher increase in Fe over time. Sex-specific differences in serum Cu concentrations have been repeatedly reported in earlier studies, with higher concentrations detected among women [42, 50]. Gender differences have also been observed for Fe. Interestingly, we also observed significant associations between change in vitamin and mineral use and longitudinal changes in Fe, Cu, I, and Se, suggesting that dietary supplement use may significantly affect overall TE status. To our knowledge, only one French study has investigated various factors in relation to longitudinal TE status, but limited to plasma Se [46]. In that study, in line with our results, sociodemographic and lifestyle characteristics had no effect on the slope of Se decline. However, hypertension had no effect on Se decline while surprisingly in our study we observed a negative relationship between Se decline and being hypertensive. The study observed an association of Se decline with occurrence of cardiovascular disease during follow-up. This finding cannot be confirmed by us as we excluded participants with cardiovascular disease from our study population. Herein, heavy alcohol consumers compared to low consumers had higher Fe increase during aging. This seems in line with studies showing that alcohol intake enhances Fe status. Higher dietary quality (as expressed by higher Mediterranean score) at baseline was related to higher Mn decrease. Further research investigating determinants of TE status should include nutritional factors, including food consumption and dietary supplement use as well as metabolic markers to disentangle their possible role and respective contribution to overall TE profiles.

Inadequate supply of TE has been linked to several pathologic conditions [1]. In our study, participants were characterized by an overall good TE status, as assessed by the serum concentrations, except regarding Zn and Se, in particular at an advanced age. A low percentage of individuals in our study had concentrations outside reference ranges regarding Mn, in line with the fact that Mn deficiency is rare [1]. However, participants with Mn concentrations above the reference range were more frequent than those with low Mn concentrations (data not shown). In the present work, as highlighted above, a large number of individuals had their serum Fe concentrations within the reference ranges. Among the few individuals who had abnormal Fe concentrations in our study, rather high than low serum Fe concentrations were detected (data not shown), although not reflecting toxicity. This is noteworthy, as elevated Fe stores in middle-aged and older individuals have been linked to several chronic diseases [51]. However, as already mentioned, serum Fe concentration is less reliable to assess Fe status [51] than, e.g. plasma/serum ferritin [52]. Participants of our study had overall a good Cu status, as reflected by normal serum Cu concentrations. Serum Cu, which was used to assess Cu status in our study, is one of the most frequently used biomarkers for Cu status, despite its lack of specificity and limited sensitivity [52]. Based on nutritional intakes, the prevalence of Zn and Cu deficiency in healthy, middle-aged and older Europeans has been estimated to be < 11% for Zn and < 20% for Cu [21], contrasting somehow with our findings. In our study, a large percentage of the participants had Zn concentrations outside the reference ranges at both time points (almost half at baseline and 60% and after 20 years). It should be borne in mind that plasma Zn concentrations do not accurately reflect changes in Zn intake and status. However, despite its known weaknesses, circulating Zn remains widely used to determine Zn status. Given that serum Zn has limited reliability, developing more reliable biomarkers for Zn status is, therefore, still of public health relevance [1, 52], since mild Zn deficiency is very common [1]. The percentage of individuals with normal Se concentrations significantly decreased over time in our study. The optimal concentration for Se status, based on cut-off derived from saturation of selenoprotein P and glutathione peroxidase activity [39], was reached for only 13% of the participants at baseline and 6% at follow-up. These data may suggest insufficient intake of Se in this population of elderly German adults, in particular at advanced age. These figures are consistent, although slightly lower, with those observed in the SU.VI.M.AX study carried out cross-sectionally among middle-aged French individuals [15]. Increment of Cu to Zn ratio above 2 usually indicates inflammation or inadequate nutritional Zn status [40]. In our study, unsurprisingly, the percentage of individuals with Cu to Zn above this cutoff significantly increased with aging, reaching almost 20% at follow-up, indicating increased inflammatory response.