## Introduction

Selenium is an essential micronutrient for humans1, with 25 genes expressing selenoproteins in the human genome2. These include iodothyronine deiodinases, thioredoxin reductases and glutathione peroxidases (GPx) which have critical roles in thyroid function, redox homeostasis and antioxidant defence and can be compromised by Se deficiency1,2,3. Selenium deficiency also affects immune responses and is linked to lower CD4+ T cell counts, disease progression and mortality among individuals infected with HIV-1 (ref. 4). Several biomarkers are used to define human Se status, including its concentration in blood fractions and urine1. Whole-plasma GPx3 activity saturates at ~100 μg Se L−1, corresponding to habitual Se intakes of ~1 μg Se kg−1 body mass d−1 and this relationship is used by many expert bodies to set dietary recommendations of 25–75 μg Se person−1 d−1 at an individual level1. Few studies have been conducted on Se nutrition in Southern Africa, although Se deficiency is probably widespread based on intake data and extrapolation. For example, intakes of 17 μg Se d−1 were reported for adults in rural Burundi5 and 15–21 μg Se d−1 for children in rural areas of Zomba District, Malawi6. The latter study corresponds to low plasma Se status (typically <60 μg L−1) among adults in this area7,8,9. In a recent spatial survey of soil and maize grain in Malawi, >90% of the population were estimated to consume <7.5 μg Se person−1 d−1 from maize grain10. Maize provides >50% of dietary energy supply in Malawi based on retail-level food balance sheets11 and household12 surveys, but contributes more in some groups. On calcareous soils classified as Eutric Vertisols, grain Se concentrations were >10-fold higher than on other soil types10 due to increased soil-to-crop Se transfer (Figure 1). This is likely to be due mainly to greater stability of soluble Se(VI) species at high pH, i.e. the most plant-available form of Se13 and decreased strength of Se(IV) adsorption on soil colloids and differences in soil clay mineralogy10. However, soils of the Eutric Vertisols type comprise only 0.5% of the land area of Malawi and Se deficiency risk is likely to be high at a population level.

The aims of this study were (1) to test whether spatial variation in soil-to-crop transfer of Se due to soil pH corresponds with Se intake and Se status in individuals and (2) to determine the risk of dietary Se inadequacy from net food availability and food composition tables. Where Se deficiency risks are high, it may be feasible to adopt agricultural-based programmes to enhance the mineral composition of foodstuffs, for example, by enriching fertilisers with Se14,15,16.

## Results

Soil pH markedly affected dietary Se intake and biomarkers of Se status in Malawi, as predicted from spatial surveys of maize10. Women from villages with acid soils in Zombwe Extension Planning Area (EPA) had median dietary Se intake of 6.5 μg Se d−1 (standard deviation (SD) 9.4, range 1.1–62.3 μg Se d−1, n = 56) from all dietary sources including water (Figure 2a; Supplementary Tables 1, 2). Selenium intake was eight-fold higher in villages with proximal calcareous soils in Mikalango EPA (median 55.3 μg Se d−1, SD 44.9, range 5.8–192 μg Se d−1, n = 58). Plasma Se concentration in Zombwe EPA (median 53.7 μg L−1, SD = 9.7, range 32.3–78.4, n = 60) was less than half of those in Mikalango EPA (median 117 μg L−1, SD = 22.5, range 82.6–204, n = 60; Figure 2b). Urine Se concentration in Zombwe EPA (median 7.3 μg L−1, SD = 2.0, range 4.1–13.3, n = 59) was one third that of Mikalango EPA (median 25.3 μg L−1, SD = 18.9, range 12.4–106, n = 56; Figure 2c). Variation between EPAs was much greater than that between villages within an EPA for dietary Se intake and plasma and urine Se concentrations. EPA + village terms explained 43 + 2, 79 + 4 and 46 + 9% of the total variation in dietary Se intake, plasma and urine Se concentrations, respectively. Differences between EPAs were highly significant for dietary Se intake and plasma and urine Se concentrations (P0.001). Variation between villages within specific EPAs also occurred for plasma (P = 0.004) and urine (P = 0.021) Se concentration, but not for dietary Se intake (P = 0.943). GPx3 activity also differed between EPAs (P = 0.002). However, only 8% of the total variation in GPx3 activity was explained by EPA, whereas 46% of the variation occurred between villages within an EPA (Figure 2d). Thus, plasma GPx3 activity in Zombwe EPA (median 162 nmol min−1 mL−1, SD 24.1, range 116–207, n = 60) was lower than in Mikalango EPA (median 177 nmol min−1 mL−1, SD 25.6, range 113–230, n = 53). Residual variation was due to individual-level variation within villages. Given that approximately half of the variation in dietary Se intake, urine Se concentration and plasma GPx3 activity occurred among individual volunteers, it seems that plasma Se concentration is the most robust biomarker of Se status in these settings. Urine Se concentration will be more sensitive to short-term variation in Se intake than blood plasma Se concentration, whereas GPx3 may also vary due to oxidative stress conditions1. However, urine Se concentration measurements (Figure 3) could provide an effective non-invasive method for identifying Se deficiency risk in populations, once cut-off points to assess the severity of Se deficiency have been established, as is used routinely for iodine17,18.

We estimated per capita supply of dietary Se available for human consumption in 46 African countries using an approach described previously for magnesium19, by integrating Food and Agriculture Organization (FAO) Food Balance Sheets (FBSs) for 2007 and a food Se composition table (Supplementary Table 3). Risk of inadequate Se intake was estimated using an EAR cut-point method19,20. Mean Se supply in 2007 for Africa was 71 μg person−1 d−1, ranging from 27 μg d−1 in Djibouti to 264 μg d−1 in Ghana (Figure 4; Supplementary Table 4). Selenium supply tended to be lower in Southern and Mid Africa. The risk of inadequate Se intake in Africa, is therefore 22% overall representing 230 M people (Figure 4). It must be stressed that the use of FBSs and food Se composition tables to estimate Se supply (i.e. and thereby infer intake and deficiency risks) must be conducted with great caution. For example, Se concentrations of fresh coconut in this analysis is from a single West Africa food composition table21 which reports a very high value of 810 μg 100 g−1 FW. This value is much higher than that reported for coconut by the US Department of Agriculture22 of 10.1 μg 100 g−1 FW. So in Ghana, for example, which has a FBS value for 2007 of 7.3 kg coconut person−1 y−1, the estimated supply of Se from coconut could be either 162 or 2 μg Se person−1 d−1 depending on which source is used. Here, we elected to use the higher value, as it was more geographically relevant, but we note that our decision may underestimate the risk of Se deficiency. In a recent study of Se intakes of children (ages 12–15) in 3 residential care orphanages in Ghana from duplicate diets23, the mean dietary intakes of Se were 58, 82, 92 μg Se person−1 d−1. These are less than the supply figure of 264 μg Se person−1 d−1 derived from FBSs for Ghana, but interestingly, are greater than reported Se intakes in Malawi in our study. So whilst Se supply is likely be lower than suggested in Figure 4 for countries where coconut consumption is high, an approach based on FBSs and food composition tables can still identify populations at higher risks of deficiency and also guide efforts to improve food composition tables.