Iron from nanostructured ferric phosphate: absorption and biodistribution in mice and bioavailability in iron deficient anemic women

Food fortification with iron nanoparticles (NPs) could help prevent iron deficiency anemia, but the absorption pathway and biodistribution of iron-NPs and their bioavailability in humans is unclear. Dietary non-heme iron is physiologically absorbed via the divalent metal transporter-1 (DMT1) pathway. Using radio- iron isotope labelling in mice with a partial knockdown of intestine-specific DMT1, we assessed oral absorption and tissue biodistribution of nanostructured ferric phosphate (FePO4-NP; specific surface area [SSA] 98 m2g-1) compared to to ferrous sulfate (FeSO4), the reference compound. We show that absorption of iron from FePO4-NP appears to be largely DMT1 dependent and that its biodistribution after absorption is similar to that from FeSO4, without abnormal deposition of iron in the reticuloendothelial system. Furthermore, we demonstrate high bioavailability from iron NPs in iron deficient anemic women in a randomized, cross-over study using stable-isotope labelling: absorption and subsequent erythrocyte iron utilization from two 57Fe-labeled FePO4-NP with SSAs of 98 m2g−1 and 188 m2g−1 was 2.8-fold and 5.4-fold higher than from bulk FePO4 with an SSA of 25 m2g−1 (P < 0.001) when added to a rice and vegetable meal consumed by iron deficient anemic women. The FePO4-NP 188 m2g-1 achieved 72% relative bioavailability compared to FeSO4. These data suggest FePO4-NPs may be useful for nutritional applications.

Iron deficiency anemia (IDA) is a major global health burden and women of childbearing age are particularly vulnerable 1 . Iron fortification of foods is a recommended strategy to prevent IDA, but iron fortification remains a challenge, because well absorbed, water-soluble iron compounds, e.g. ferrous sulfate (FeSO 4 ), often cause unacceptable sensory changes when added to foods 2 . Poorly acid-soluble iron compounds, such as ferric phosphate (FePO 4 ) in bulk form, are stable in foods but their absorption in humans is too low to have nutritional value 3 . However, the dissolution and absorption of poorly-soluble iron compounds is inversely related to particle size and nanostructured iron compounds may be useful as food fortificants or supplements 4 . We have previously shown that scalable nanostructured FePO 4 (FePO 4 -NP) produced by flame spray pyrolysis (FSP) dissolve rapidly in gastric conditions 5 . In an anemic rat model, iron bioavailability (determined as the ability to rapidly replete hemoglobin (Hb) when given orally) from two FePO 4 -NP compounds (FePO 4 /Fe 2 O 3 , specific surface area [SSA] 197 m 2 /g and FePO 4 /Zn 3 (PO 4 ) 2 , SSA 191 m 2 /g) was 75 and 95%, respectively, compared to the ionic reference compound FeSO 4 5 . However, rodents absorb iron efficiently because they endogenously synthesize ascorbic acid, have lower duodenal pH, and are less affected by dietary absorption inhibitors than humans 6,7 . Thus, whether the Results Iron nanoparticle production and characterization. Two different sized ferric orthophosphate dihydrate nanoparticles (FePO 4 -NPs) with SSAs of 188 m 2 g −1 and 98 m 2 g −1 were produced by flame spray pyrolysis (FSP) at ETH Zurich as previously described 15 with adaptations (see "Methods"). FSP is a scalable production process 16 for tailor-made particles with high SSA and well-defined chemical composition 17 . As indicated by transmission electron microscopy (TEM) images and X-ray diffraction (XRD), the two FePO 4 -NPs are almost spherical particles with a calculated size (dBET) of 11 nm and 21 nm, respectively ( Fig. 1a-c) and XRD-amorphous (Fig. 1d) . Amorphous ferric orthophosphate with a SSA of 25 m 2 g −1 served as reference compound for BET and XRD measurements. FePO 4 -NPs for the human study were produced from stable isotope ( 57 Fe) enriched precursors, bulk FePO 4 and FeSO 4 from ( 58 Fe) enriched precursors. For the human study, bulk size FePO 4 with a SSA of 27 m 2 g −1 was produced as previously described 18 with adaptations (see "Methods"). As indicated by TEM, these particles have a size above 100 nm and an irregular shape. The calculated primary particle size of these FePO 4 particles with an SSA of 27 m 2 g −1 would be 77 nm; however, as shown in the TEM images, the assumption of sphericity is not valid and we therefore refer to these particles as bulk FePO 4 . For the methods of SSA, hydrodynamic diameter, XRD and TEM, see "Methods". In ultrapure water, bulk FePO 4 and the 98 m 2 g −1 and 188 m 2 g −1 FePO 4 -NPs formed large agglomerates with a mean diameter and polydispersity index (PDI) of 859 ± 152 nm (PDI 0.63) at pH 6.7, 1191 ± 603 nm (PDI 0.34) at pH 6.1 and 1149 ± 132 nm (PDI 0.41) at pH 6.0 and a Zeta potential of -5.3 ± 1.6 mV, −2.1 ± 0.1 mV and -2.5 ± 0.1 mV. The same FePO 4 -NPs were shown to form stable agglomerates for up to 72 h in aqueous dispersions with added protein content (10% FCS) with a mean diameter of 231-615 nm 19 . Particle characterization followed MIRIBEL guidelines 20 . Characterization of intestine-specific DMT1 knockdown model. We chose DMT1 int/int and DMT1 fl/fl littermates (controls) as the model to study absorption and biodistribution of nanosized FePO 4 (SSA 98 m 2 g −1 ). DMT1 int/int and DMT1 fl/fl mice were bred on a C57BL/6 background by crossing floxed DMT1 (courtesy of Nancy Andrews, Duke University, USA 21 ) with villin-Cre transgenic mice. Figure 2a shows a schematic illustration of the floxed allele and the primer positions used for genotyping of the DMT1 int/int and DMT1 fl/fl mice. Homozygously floxed DMT1 fl/fl mice that do not carry the Cre recombinase were chosen as controls in all mice experiments to account for a potential effect of floxing on iron absorption and biodistribution. For details of breeding, genotyping and housing see "Methods".
To validate the model and to determine the optimal age window for the feeding and radio-isotope experiments, we assessed Hb concentration trajectories of DMT1 int/int and DMT fl/fl mice fed an iron-sufficient (35 ppm iron as ferrous citrate) or iron deficient (3 ppm native iron) diet from postnatal day (PND) 24 (weaning) until 42. PND 42 was set as endpoint, as it was the age when first mice reached a Hb < 4 g/dL, which we defined as the humane endpoint (see "Methods" for definition) 22 . We measured Hb in tail blood spots at PND 24,27,30,33,36,39,and 42. Using repeated-measures ANOVA with time as within-group factor, and dietary iron content (35 ppm vs. 3 ppm) and genotype (DMT1 int/int vs. DMT fl/fl ) as between-group factors, we found a significant effect of iron content (p = 0.023) for lower Hb in mice receiving the 3 ppm iron diet, and of genotype (p = 0.006) for lower Hb in DMT1 int/int mice (Fig. 2b). There was no significant iron content x genotype interaction, indicating that both the 3 ppm iron diet and the intestine-specific DMT1 knockdown independently lowered Hb across PND 24-42. However, we obtained no significant time x iron content (p = 0.087) or time x genotype (p = 0.090) interactions, indicating that the Hb-lowering effect of the 3 ppm diet and the intestine-specific DMT1 knockdown was not time-dependent during this age period. These results confirm that the DMT1 int/int mice bred for our studies have significantly lower Hb concentrations from PND 24-42, both when fed an iron-sufficient or iron deficient diet, but there was no progressive decline in Hb concentrations, suggesting upregulation of iron absorption from a partly intact DMT1 pathway.
We further determined expression of DMT1 mRNA harboring an iron-responsive element (IRE) in its 3'-terminal exon and the upstream 5' exon1A in crude homogenates prepared from whole duodenum (not isolated enterocytes), colon and liver of DMT int/int and DMT fl/fl mice fed an iron deficient diet from PND 24-42 (Fig. 2c). The DMT1-IRE and DMT1-exon1A isoforms were previously implicated in iron regulation, specifically in the duodenum 23 . DMT1 mRNA levels are expected to be increased in the duodenum of iron deficient mice with a functional DMT1 gene 23 . We showed that expression of both exon1A and the IRE of DMT1 was significantly lower in the duodenum of DMT1 int/int compared to DMT1 fl/fl mice fed an iron deficient diet. However, the reduction in DMT1 mRNA expression was partial, suggesting incomplete knockdown of DMT1 in the DMT1 int/int mice. This is supported by the observation that expression of the DMT1-IRE isoform in the duodenum was positively associated with Hb concentrations measured at PND42 in DMT1 fl/fl mice receiving 3 ppm and 35 ppm iron (standardized β = 0.623, p = 0.026; regression models controlled by sex), with the same trend observed in DMT1 int/int mice (standardized β = 0.522, p = 0.079) (data not shown). There were no differences in expression of DMT1 mRNA between DMT1 int/int and DMT1 fl/fl mice in colonic and liver tissue, confirming the intestinespecificity of the partial knockdown.
Eighteen-day FePO 4 -NP feeding study in DMT1 int/int and DMT1 fl/fl mice. To determine whether FePO 4 -NP is absorbed via DMT1 and utilized for erythropoiesis, we compared Hb trajectories between DMT1 int/ int and DMT1 fl/fl (control) mice fed with FePO 4 -NP (SSA 98 m 2 g −1 ) or FeSO 4 as the reference compound from PND 24-42 (18 days). We hypothesized that (1) intestine-specific DMT1 knockdown would lower Hb equally in mice fed FePO 4 -NP or FeSO 4 and, (2) that Hb would be comparable in DMT1 fl/fl mice fed iron in the form of nanosized FePO 4 -NP or FeSO 4 . We randomly allocated a total of 28 male and female DMT1 int/int and DMT1 fl/ fl mice to receive either the purified AIN-93G diet containing 35 ppm iron as FePO 4 -NP or FeSO 4 from PND 24-42. The design of the study is shown in Fig. 3a. For more details see "Methods". We measured Hb in tail blood spots at PND 24,27,30,33,36 and 42. Using repeated-measures ANOVA with age as within-group factor, and iron compound (FePO 4 -NP vs. FeSO 4 ) and genotype (DMT1 int/int vs. DMT fl/fl ) as between-group factors, www.nature.com/scientificreports/ we found a significant effect of genotype (p = 0.042) for lower Hb concentrations in DMT1 int/int mice, and a nonsignificant effect of iron compound (p = 0.067) for lower Hb in mice fed FePO 4 -NP (Fig. 3b). Although the interaction between genotype and iron compound was not significant (p = 0.140), the Hb-lowering effect of the DMT1 knockdown appeared to be compound-specific, in that Hb trajectories were similar between compounds in DMT1 fl/fl mice, but Hb concentrations were lower in DMT1 int/int mice fed FePO 4 -NP than fed FeSO 4 . The data in Fig. 3b suggest the FeSO 4 was nearly as effective at increasing Hb in DMT1 int/int mice as it was in DMT1 fl/fl mice, whereas for FePO 4 -NP, Hb concentrations were increased in the DMT1 fl/fl mice but were unchanged in the DMT1 int/int mice. This pattern is supported by the liver iron concentrations (Fig. 3c) at PND 42 in DMT1 int/ int and DMT1 fl/fl mice fed FePO 4 -NP or FeSO 4 . We found a significant effect of genotype (p = 0.009) for lower liver iron concentrations in DMT1 int/int mice. Specifically, FePO 4 -NP feeding resulted in significantly lower liver iron concentrations in DMT1 int/int compared to DMT1 fl/fl mice, but FeSO 4 did not. The lacking effect of the intestinal DMT1 knockdown on Hb trajectories and liver iron concentrations in mice fed a diet containing iron as FeO 4 -NP may be explained by the high solubility of FeSO 4 at low gastric pH and rodents' high efficiency in absorbing iron 6,7 . Considering that our model was not a complete intestinal DMT1 knockout model (described above), it can be speculated that rapid dissolution in the proximal gut and absorption of FeSO 4 in mice with a partial intestinal DMT1 knockdown was enough to maintain erythropoiesis and liver iron stores during the experimental period (PND [24][25][26][27][28][29][30][31][32][33][34][35][36][37][38][39][40][41][42]. This may also explain why the DMT int/int mice fed (and bred on) ferrous citrate in the characterization experiment were more anemic from PND 24-42 than the DMT int/int mice fed (and bred on) FeSO 4 in the feeding experiment. Taken together, these data suggest that compared to FeSO 4 , FePO 4 -NP were less well absorbed in DMT1 int/int compared to DMT1 fl/fl mice, suggesting that absorption of iron from FePO 4 -NP is more sensitive to partial knockdown of DMT1. is DMT1-dependent and biodistribution comparable to FeSO 4 in iron deficient anemic conditions, we administered a single oral dose of irradiated FePO 4 -NP (SSA 98 m 2 g −1 ) or FeSO 4 labelled with radioactive iron ( 59 Fe) to iron deficient anemic DMT1 int/int and DMT1 fl/fl control mice and measured tissue distribution of 59 Fe 24 h after administration. Based on the findings from the feeding study, we hypothesized that less iron from a single oral dose of FePO 4 -NP would be absorbed in DMT1 int/int than DMT1 fl/fl mice. We further hypothesized that tissue distribution of iron taken up and absorbed from the single oral dose of FePO 4 -NP would be equal to FeSO 4 in both iron deficient anemic DMT1 int/int and DMT1 fl/fl control mice. A total of 27 male and female DMT1 int/int and DMT1 fl/fl mice were randomly allocated by genotype to receive radiolabeled FePO 4 -NP (SSA 98 m 2 g -1 ) or FeSO 4 and were placed on an iron deficient diet (3 ppm native iron) from PND 21 until the end of the experiment. At PND 24, mice were transported to the Nuclear Energy Corporation South Africa (NECSA) and acclimatized until administration of radiolabeled FePO 4 -NP or FeSO 4 by oral gavage at PND 30. The design of the study is shown in Fig. 4a. The day before compound administration, Hb was measured in a tail blood spot, and 2 h before administration mice were fasted. A single dose of ~ 50 µg iron (mean ± SEM administered: 47.4 ± 2.9 µg) in the form of the allocated compound labelled with 59 Fe (FePO 4 -NP: 0.08 ± 0.01 MBq; FeSO 4 : 0.52 ± 0.2 MBq) was orally gavaged (in 100 µL saline containing 0.1% bovine serum albumin) and total dose administered determined by measuring syringe activity before and after gavage using a dose calibrator (Capintec CRC-15R, Capintec Inc., Ramsey, NJ, USA). Mice were then placed into a clean metabolic cage and received ad libitum access to iron deficient diet (3 ppm iron) one hour after oral gavage. After 24 h, mice were euthanized and individual were determined by two-sided independent t-tests. The results are shown as means ± SEM and differences were considered significant at p < 0.05. www.nature.com/scientificreports/ tissues were dissected and analyzed for 59 Fe content using a gamma-counter. All counts were adjusted for decay. For more details see "Methods". DMT1 int/int mice had significantly lower Hb than DMT1 fl/fl mice (5.7 ± 0.5 vs. 8.9 ± 0.7 g/dL, p = 0.01) the day before compound administration. As shown in Fig. 4b, 24 h after administration, significantly less 59 Fe from FePO 4 -NP was present in whole blood and heart (expressed as % initial 59 Fe dose/g tissue) in DMT1 int/int compared to DMT fl/fl mice. These data are consistent with lower absorption of FePO 4 -NP in DMT1 int/int compared to DMT fl/fl mice shown in the feeding study above (Fig. 3b,c). Notably, in DMT1 int/int and DMT1 fl/fl mice (Fig. 4c), we found no significant difference in tissue 59 Fe distribution between FePO 4 -NP and FeSO 4 . The only exception was in DMT1 fl/fl mice, where significantly more 59 Fe from FePO 4 -NP was present in femoral bone marrow than from FeSO 4 (Fig. 4c,ii). Overall, the results from this study indicate that in iron deficient anemic mice, iron from a single oral dose of FePO 4 -NP displays a similar biodistribution as FeSO 4 .

Iron bioavailability of two 57 Fe-labeled FePO 4 -NP in iron deficient anemic women. To quantify
iron absorption and erythrocyte incorporation (bioavailability) from two FePO 4 -NPs (SSA of 98 m 2 g −1 and 188 m 2 g −1 ) in iron deficient, mostly anemic women, we fortified two standardized rice test meals with FePO 4 -NPs labeled with stable iron isotopes and compared their performance to labeled bulk FePO 4 with a SSA of 25 m 2 g −1 (as the negative reference compound) and dried FeSO 4 (as the positive reference compound) in a randomized, cross-over study. We hypothesized that: (1) an increase in SSA of the FePO 4 -NP compound would significantly increase iron bioavailability, and (2) iron bioavailability from FePO 4 -NP SSA 188m 2 g −1 would not differ significantly compared to iron bioavailability from FeSO 4 . Iron bioavailability was estimated by using stable-isotope techniques measuring the incorporation of 57 Fe and 58 Fe into erythrocytes 14 days after administration 24 . Fe (% of initial 59 Fe/g tissue) from a single dose (~ 50 µg) of FePO 4 -NP (SSA 98 m 2 g −1 ) after oral gavage in iron deficient anemic DMT1 fl/fl (n = 7 male; n = 5 female) and DMT1 int/int (n = 7 male, n = 8 female) mice. Differences in tissue 59 Fe distribution by genotype (DMT1 int/int vs. DMT1 fl/fl ) were determined by two-sided independent t-tests. The results are shown as means ± SEM and differences were considered significant at p < 0.05. (c) Biodistribution of 59 Fe from FePO 4 -NP and FeSO 4 (reference compound) 24 h after oral gavage in iron deficient anemic (i) DMT1 int/int mice and (ii) DMT1 fl/fl mice. Differences in tissue 59 Fe distribution by iron compound (FePO 4 -NT vs. FeSO 4 ) were determined by two-sided independent t-tests. The results are shown as means ± SEM and differences were considered significant at p < 0.05. The study was carried out in 18 Thai women who provided informed written informed consent. Main inclusion criteria were: (1) female aged 18 to 49 years and (2) Hb ≥ 80 g/L and plasma ferritin < 25 µg/L (see "Methods" for additional inclusion criteria and details on methods). We chose Thailand as our study site because, worldwide, the greatest number of women with anemia live in the WHO South and Southeast Asia Region, and in Thailand, 32% of women of reproductive age are anemic 25 . The study had a cross-over design in which all women consumed four labeled test meals over a 31-day period (Fig. 5a). The order of the four test meals was randomly assigned to each subject. In a first phase, the women presented fasting and consumed the first test meal between 7.00 and 9.00 am on day 1 and the second test meal between 7.00 and 9.00 am on day 2; in the second phase, 14 days later, they consumed the remaining 2 test meals between 7.00 and 9.00 am on days 16 and 17. Consumption of meals was directly supervised and all women completed all meals. No intake of food and fluids was allowed for 2 h afterward. The standardized test meal consisted of rice with vegetables (see "Methods") fortified with 2 mg of isotopically labeled compound and a glass of deionized water (200 ml). The two FePO 4 -NPs were labeled with 57 Fe, while the bulk FeSO 4 and FePO 4 were labeled with 58 Fe (see "Methods"). We calculated the amounts of 57 Fe and 58 Fe labels in blood 14 d after the 2nd and 4th test meals based on the shift in iron-isotopic ratios and the estimated amount of iron circulating in the body 24 . For details of laboratory analyses, calculation of iron absorption and statistical analyses see "Methods".

Scientific Reports
Baseline characteristics of the 18 women are shown in Table 1; 14 were iron deficient, 12 had iron deficiency anemia, 15 had normal hemoglobin A (HbA) and three had HbE trait. As shown in and FeSO 4 , respectively, with significant differences between all groups (for all, p < 0.005). Thus, particle size reduction of FePO 4 to SSA 98 m 2 g −1 and 188 m 2 g −1 increased iron bioavailability 2.8 and 5.4-fold compared to bulk FePO 4 , and achieved a relative bioavailability of 34% and 72% compared to FeSO 4 . In separate linear regressions including C-reactive protein and plasma ferritin (iron status) as independent variables and iron absorption from bulk FePO 4 and FeSO 4 combined (standardized β = −0.393, SE = 0.008, p = 0.075) and on iron absorption from the two FePO 4 -NPs combined (standardized β = −0.575, SE = 0.014, p = 0.0027) as outcome variables, plasma ferritin was a stronger negative predictor of iron absorption from the two FePO 4 -NPs (p = 0.007) than from the larger compounds (p = 0.027). Because iron absorption through the usual DMT1 mediated pathway is regulated by iron status 10 , but absorption through simple translocation of nanoparticle iron would likely not be, this is consistent with iron absorption from FePO 4 -NPs being at least partially dependent on the DMT1 mediated pathway.

Conclusions
Previous in vitro studies suggested that gastrointestinal uptake of iron NPs could occur through several mechanisms 13,26,27 . Our in vivo data suggest that absorption of FePO 4 -NPs is largely DMT1 dependent (Fig. 3). Supporting our findings, uptake in cell models of the same FePO 4 -NP SSA 188 m 2 g −1 used in this study was reduced by DMT1 inhibitors or siRNA targeting of DMT1 14 . However, our findings do not rule out a DMT1 independent route for absorption of NP iron, as our DMT1 int/int mice had reduced but still detectable expression and FeSO 4 as negative and positive reference compounds, respectively. Meal sequence was randomized across all women. Pairwise comparisons were performed using two-sided paired t-tests with Bonferroni adjustment for multiple testing. Boxes indicate median and interquartile ranges, whiskers describe the range of the data (min to max). Differences were considered significant at p < 0.05. CRP C-reactive protein, Hb hemoglobin. . We did not determine the expression of other DMT1 isoforms, which could have been responsible for the observed iron absorption, nor measure DMT1 protein levels in duodenal enterocytes to confirm the absence of DMT1 in DMT1 int/int knockdown mice. Importantly, our mouse data (Fig. 4) suggest that the biodistribution of FePO 4 -NPs after uptake in anemic DMT1 int/int and DMT1 fl/fl mice is comparable to FeSO 4 , and there was no abnormal tissue deposition of iron absorbed from FePO 4 -NP in the reticuloendothelial system (spleen, Peyer's Patches) or in the kidney or liver. In our human study, plasma ferritin was a strong negative predictor of iron absorption from the two FePO 4 -NPs (SSA of 188 m 2 g −1 and 98 m 2 g −1 ), suggesting physiologic regulation by body iron stores. Taken together, these data argue against, but do not rule out, unregulated translocation and biodistribution of NP iron from gastrointestinal exposure.
Other food-relevant nanoparticles, e.g. SiO 2 , TiO 2 ZnO and Al 2 O 3 , induce adverse effects in human intestinal cells or experimental animals [28][29][30][31][32][33][34][35][36][37] . In contrast, gastrointestinal exposure to a variety of iron NPs has not been associated with measurable toxicity 5,19,26,38-44 . Short-term feeding of iron deficient rats with diets containing iron NPs 5,38 or iron NPs stabilized on β-lactoglobulin fibrils 43 did not cause histopathology or oxidative stress. We previously performed extensive toxicity testing of the FePO 4 NPs used in this study 19 . In rats fed diets containing these iron NPs for 90 days (at doses at which FeSO 4 has been shown to induce adverse effects), feeding did not cause toxicity, including oxidative stress, organ damage, abnormal tissue iron accumulation or histological changes 19 . Furthermore, they were taken up in vitro by gastrointestinal cells without reducing cell viability or inducing oxidative stress 19 . In these studies, FePO 4 nanoparticles appeared to be as safe for ingestion as FeSO 4 .
Conflicting information regarding the toxicity of food relevant nanoparticles and regulatory issues have so far hampered applications of nanotechnology in the food sector. However, as described above, in vitro and in vivo studies have consistently shown that iron NPs do not show measurable toxicity on gastrointestinal exposure. Here, we demonstrate that absorption of FePO 4 -NPs is largely via the physiological DMT1 pathway and biodistribution patterns resemble FeSO 4 . In addition, we show that nanostructuring of FePO 4 sharply increases bioavailability in anemic women and achieves a relative bioavailability of 72% compared to FeSO 4 , confirming previous absorption data in rats 5,38 . Together, these findings suggest the promise of iron NPs as novel food fortificants. However, further research is needed to better clarify the uptake pathways of iron and other mineral NPs from gastrointestinal exposure before they can be recommended for nutritional applications.

Methods
Nanoparticle production and characterization. For the production of nanostructured ferric phosphate (FePO 4 -NP) by flame spray pyrolysis (FSP), iron nitrate nonahydrate (purity ≥ 97.0%, Sigma-Aldrich, Buchs, Switzerland) and tributyl phosphate (97%, Sigma-Aldrich) were dissolved in a 1:1 mixture by volume of ethanol (denat. 2% 2-butanone, Alcosuisse) and 2-ethylhexanoic acid (purity ≥ 99%, Sigma-Aldrich) at a total metal concentration of 0.4 mol L −1 or 0.5 mol L −1 for the two compounds. This precursor solution was fed at 2 or 7 mL min −1 into the FSP spray nozzles by a syringe pump (Lambda, VIT-FIT) and atomized by co-flowing 5 or 7 L min -1 of oxygen (purity ≥ 99.5%, Pangas) at 1.5 bar pressure drop. The spray was ignited by a methane/ oxygen (2.5 L min −1 ) ring-shaped flame 45 . Using a vacuum pump (Busch, Mink MM1202 AV), product particles were collected on water-cooled Teflon membrane-filters (1TMTF700WHT, BHA Technologies AG) placed at least 70 cm above the burner. Pharmaceutical grade (German Pharmacopoeia DAB, Erg. B.6, no. 505033001, Lohmann) amorphous ferric orthophosphate with a SSA of 25 m 2 g −1 served as reference compound for BET and XRD measurements. FePO 4 -NPs for the human study were produced from stable isotope ( 57 Fe) enriched precursors, bulk FePO 4 and FeSO 4 from ( 58 Fe) enriched precursors (Chemgas). SSA was determined by N 2 adsorption (Micromeritics Tristar 3000, Micromeritics Instruments Corp) at 77 K in the relative pressure range p/p0 = 0.05-0.25 and calculated using Brunauer-Emmett-Teller (BET) theory. Assuming dense spherical particles, the particle diameter (d BET ) was calculated from the measured SSA according to d BET = 6/(ρ·SSA), where ρ is  46 . For transmission electron microscopy (TEM) analysis, the powders were deposited on a parlodion foil supported on a copper grid and analyzed on a CM12 microscope (FEI, LaB6 cathode, operated at an acceleration voltage of 100 kV). The crystallinity of the powders was investigated by X-ray diffraction (XRD) on a AXS D8 Advance diffractometer (Bruker) operating with a Cu-Kα radiation. Hydrodynamic diameter was determined by dynamic light scattering using a Zetasizer Nano ZS (Malvern).
Animal studies. Intestine-specific DMT1 knockdown (DMT1 int/int ) model. Intestine-specific DMT1 knockdown (DMT1 int/int ) mice were bred by crossing floxed DMT1 (DMT fl/fl ) mice on a homogenous C57BL/6 strain (courtesy of Nancy Andrews, Duke University, USA 21 ) with villin-Cre transgenic mice on the C57BL/S6J strain (ETH Zurich). The floxed DMT1 mice were back-crossed to a C57BL/6J background for six generations to establish the DMT1 fl/fl mouse breeding colony at ETH Zurich. The 6th generation was then shipped to the Vivarium at NWU to establish the intestine-specific DMT1 knockdown and villin-Cre mouse breeding colony for the experiments described in this article. To obtain DMT1 int/int mice, the animals needed to be homozygously floxed and have the villin-Cre transgene (Cre-positive). The homozygously floxed littermates (DMT fl/fl ) that did not have the villin-Cre transgene (Cre-negative) served as control in all mice experiments to account for a potential effect of floxing on iron absorption and biodistribution. To breed these mice, we mated heterozygously floxed males that have the villin-Cre (Cre-positive) transgene with homozygoulsy floxed females without the villin-Cre transgene (Cre-negative). Theoretically, this results in an approximate yield of 25% DMT1 KO and 25% controls per litter. In order to obtain the required number of mice for the different experiments, we continuously bred, genotyped and enrolled mice into the experiments. Thus, mice in the different experimental groups are from different litters.
Genotyping. The genotyping method to screen each mouse bred from the floxed DMT1 and villin-Cre mice colony was as follows. Briefly, a tissue sample (distal tail sample [≤ 2 mm]) was collected at PND 10-21 47 . gDNA isolation was done using the GenElute™ Mammalian Genomic DNA (gDNA) Miniprep Kit (Sigma Aldrich) following the manufacturers protocol. The quality of DNA was assessed and quantified using the NanoDrop™ spectrophotometer (ND-1000, Wilmington, DE, USA). Gene specific polymerase chain reaction (PCR) was performed and amplicons were visualized using ethidium bromide stained gel electrophoresis. For PCR amplification of the DMT1 gene the forward primer 5'-atgggcgagttagaggcttt-3' and the reverse primer 5'-cctgcatgtcagaaccaatg-3' were used. For PCR amplification of the villin-Cre gene the forward primer 5'-gtgtgggacagagaacaaacc-3' and reverse primer 5'-acatcttcaggttctgcggg-3' were used together with an endogenous control (MyD88) primer pair 5'-agacaggctgagtgcaaacttgtgctg-3' and 5'-ccggcaactagaacagacagactatcg-3' . Control gDNA with known genotype (ETH) as well as none-template control samples were included in each PCR run. Characterization of intestine-specific DMT1 knockdown model. A total of 10 DMT1 int/int (n = 6 male; n = 4 female) and 21 DMT1 fl/fl (n = 10 male, n = 11 female) mice, born to dams that were kept on the purified AIN-93G diet containing 35 ppm iron (as ferrous citrate) ad libitum, were randomly allocated in pairs to receive an iron deficient (3 ppm native iron) or iron-sufficient (35 ppm ferrous citrate) diet ad libitum from PND 24 to PND 42. PND 42 was set as endpoint, as this was the time point when first mice reached an Hb < 4 g dL −1 , which we defined as humane endpoint 22 (a humane endpoint is the earliest scientifically justified point at which pain or distress in an experimental animal can be prevented, terminated, or relieved, while meeting the scientific aims and objectives of the study 50 ). Hemoglobin (Hb) concentrations were measured in tail blood spots (20 µg dL −1 ) at PND 24, 27, 30, 33, 36, 39 and 42 using a calibrated Hb 201 + HemoCue® system (HemoCue Angelholm, Sweden). At PND 42, mice were euthanized by decapitation, and liver, duodenum and colon tissue immediately removed, snap frozen in liquid nitrogen, and stored at −80 °C until analysis.
Expression of DMT1 mRNA harboring an iron-responsive element (IRE) in its 3'-terminal exon and the upstream 5' exon1A in duodenum, colon and liver was analyzed with qPCR. PCR ready Syber green primers were synthesized by IDT (WhiteHead Scientific, South Africa). Primer pair sequences were adapted from Hubert and Hentze (2002) 23 . Total RNA was isolated by using Trizol reagent following the standarized protocol. cDNA synthesis was performed using 10 mM oligo-dT 18mer and random hexamer primer mix with Superscript II reverse transcriptase (Qiagen), following the method prescribed by the manufacturer. A total of 50 ng cDNA was used Eighteen-day FePO 4 -NP feeding study in DMT1 int/int and DMT1 fl/fl mice. A total of 15 DMT1 int/int (n = 6 male; n = 9 female) and 13 DMT1 fl/fl (n = 6 male, n = 7 female) mice, born to dams that were kept on the standardized AIN-93 G diet containing 35 ppm iron (as ferrous sulphate [FeSO 4 ]) ad libitum, were randomly allocated in pairs to receive a diet fortified with 35 ppm FePO 4 -NP (SSA 98 m 2 g −1 ) or FeSO 4 (reference compound) added to an iron-free AIN93-G diet from PND 24 to PND 42 ad libitum. Hb concentrations were measured in tail blood spots at PND 24, 27, 30, 33, 36 and 42 using the Hb 201 + HemoCue® system (HemoCue Angelholm, Sweden). At PND 42, mice were euthanized by decapitation and liver tissue immediately removed, snap frozen in liquid nitrogen, and stored at −80 °C until analysis. Liver tissue samples were homogenized and digested with nitric acid according to Erikson et al. (1997) 51 , and total iron concentrations were measured by using the hydrogen reaction mode on an Agilent 7900 quadrupole ICP-MS at the Central Analytical Facilities, Stellenbosch University, South Africa. Samples were introduced via a 0.4 ml/min micromist nebulizer into a peltier-cooled spray chamber at a temperature of 2 °C. The instrument was optimized for analysis in high matrix introduction (HMI) mode, and all samples and standards were diluted with argon gas to minimize matrix load to the analyzer. The instrument was calibrated using a National Institute of Standards and Technology (NIST) traceable standard (Inorganic Ventures, USA). NIST-traceable quality control standards at high and low concentration levels (De Bruyn Spectroscopic Solutions, Bryanston, South Africa) were analyzed to verify the accuracy of the calibration before sample analysis commenced and this was repeated for every 12 samples to monitor drift. A germanium (Ge) internal standard was introduced online to monitor instrument drift and correct for matrix differences between samples and standards. During the course of the analysis, internal standard recovery was between 90 and 110% for all samples, and recovery for drift monitor standards between 95 and 105%. Oxide formation was less than 0.3%. Three replicate measurements were completed for each sample.

Absorption and biodistribution of radiolabeled FePO 4 -NP and FeSO 4 from an acute oral dose in iron deficient anemic DMT1 int/int and DMT1 fl/fl mice.
A total of 13 DMT1 int/int (n = 7 male; n = 5 female) and 15 DMT1 fl/fl (n = 7 male, n = 8 female) mice, born to dams that were kept on the standardized AIN-93 G diet containing 35 ppm iron (as FeSO 4 ) ad libitum, were placed on an iron deficient diet (3 ppm native iron) from PND 21 and throughout the entire experiment. Mice were randomly allocated to receive radiolabeled FePO 4 -NP (SSA 98 m 2 g −1 ) or FeSO 4 . At PND 24, mice were transported to the South African Nuclear Energy Corporation South Africa (NECSA) and left to acclimatize to the new environment until administration of radiolabeled FePO 4 -NP or FeSO 4 by oral gavage at PND 30 (29)(30)(31). The day before compound administration, Hb was measured in a tail blood spot.
Mice were fasted for 2 h before administration of the oral gavage during which they were acclimatized to metabolic cages (3701M081; Tecniplast). Then, a single dose of ~ 50 µg iron in the form of the allocated compound labelled with 59 Fe (mean activity: 0.30 MBq) was orally gavaged (in 100 µL saline containing 0.1% bovine serum albumin) using disposable flexible gavage needles. Total dose administered was determined by measuring syringe activity before and after gavage using a dose calibrator (Capintec CRC-15R, Capintec Inc., Ramsey, NJ, USA). Mice were then placed into a clean metabolic cage and received ad libitum access to iron deficient diet (3 ppm iron) one hour after oral gavage. After 24 h, the mice were euthanized by decapitation and the following individual organs were dissected and analyzed for 59 Fe content and weighed using an automated Hidex®600 SL gamma-counter (Hidex Oy, Finland). All counts were adjusted for decay: whole blood, heart, lung, liver, spleen, stomach, duodenum, peyer's patches, ileum, jejenum, colon, kidneys, femur, as well as feces and urine. Chyme was separated from the intestinal segments by washing with 1% phosphate-buffered saline.
A total of 15 mg of the nanostructured FePO 4 (SSA 98 m 2 g -1 ) was irradiated in the SAFARI-1 20 MW research reactor in a hydraulic position at a neutron flux of 1 × 10 14 n/cm 2 s for 16 days with a 3 day cooling period. Irradiation provided 512 MBq 59 Fe (t ½ = 44.5 days) per 1 mg of FePO 4 -NPs 52 . The color of the FePO 4 -NP (SSA 98 m 2 g −1 ) remained yellowish during the irradiation given confidence that the particles retained their nano structure. This was further confirmed by measuring the surface area using a Tristar 3000 BET surface area and porosity analyzer (Micromeretrics, Norcross, USA). Using a special small volume adapter the SSA of the irradiated particles was determined to be 79.1 m 2 g −1 which mimicked the SSA determined prior to irradiation. Special care was taken to remove the static nature of the NP after irradiation using a Antistatic Ionizer (RADAWG, Radom, Poland), before opening and during handling of the radioactive labelled NPs. FePO 4 -NP were decayed a further 30 days prior to animal administration to allow for the co-activated 32 P (t ½ = 14 days) to decay. 32 P is a pure β emitter and hence did not interfere with the determination using gamma spectrometry at the > 1000 keV range 52 . 59 Fe-labelled FeSO 4 was purchased from American Radiolabeled Chemicals, Inc. (St. Louis, USA) with a specific activity > 185 GBq/g. In order to mimic the chemical administrated dose used for the NP's (50 µg and 0.3 MBq), non radioactive FeSO 4 was added to a subset of the purchased stock solution. 59 Fe radioactivity (for administration) was detected by a dose calibrator (Capintec CRC-15R, Capintec Inc., Ramsey, NJ, USA), while samples were analyzed using an automated Hidex ® 600 SL gamma-counter (Hidex Oy, Finland) making use of the 1099 keV 56.5% gamma emission. Cross calibration between the two instruments were obtained by using a 59 Fe standard curve after a series of dilutions. The irradiated particles were dispersed in in saline containing 0.1% bovine serum albumin by mixing for 30 s on a vortex followed by sonication for 10 min in a Sonorex Digitec waterbath (Bandelin Electronic) at 35 kHz and 80 W and were gavaged within 15 min after sonication. We acknowledge that we did not determine absorption and biodistribution of FePO 4 -NP with an SSA of 188 m 2 g −1 in our mouse www.nature.com/scientificreports/ experiments because we were not able to radio-label this compound without colour and structural changes (visual inspection) after irradiation in the reactor as opposed to the 98 m 2 g −1 where this was not observed.
Statistical analysis. Data were analyzed using IBM SPSS Statistics software (version 24). Data were examined for normality of distribution (using q-q plots, histograms, and Shapiro-Wilk test) and the presence of outliers (using box plots). Homogeneity of variance was examined by the Levene's test. Variables that significantly deviated from normality and/or variance of homogeneity were transformed prior to interferential statistical analysis. Differences in Hb trajectories over time  by genotype (DMT1 int/int vs. DMT1 fl/fl ) and by dietary iron content (35 ppm vs. 3 ppm) or iron compound (FePO 4 -NT vs. FeSO 4 ) were determined using repeated measures ANOVA. Differences in tissue iron concentrations, gene expression and percentage initial 59 Fe dose/g tissue by genotype and by iron compound were determined by two-sided independent t-tests. The results were expressed as means ± SEM and differences were considered significant at p < 0.05.
Sample size calculations indicated that 16 women should be included based on 80% power to detect a 40% difference in iron bioavailability within subjects, an SD of 8.2% for log-transformed absorption data from previous absorption studies with the same meal and iron source/compound in a similar population of Thai women, and a type I error rate of 5%. We anticipated a dropout rate of 10% and therefore recruited 18 women.
Ethics. The study has been carried out according to the guidelines laid down in the Declaration of Helsinki, and all procedures involving human study participants were approved by the Ethics Committee of ETH Zurich (Zurich, Switzerland) and the Mahidol University Central Institutional Review Board (Salaya, Nakhon Pathom, Thailand). The study was registered at clinicaltrials.gov as NCT03660462 on 06/11/2018. Detailed oral and written information explaining the study purposes and potential risks and benefits were provided to the interested volunteers. Written informed consent was obtained from all participants. The oral administration of stable isotopes does not present any health risk. All data were coded and treated confidentially.
Study procedures. Women were recruited by screening at the Institute of Nutrition at Mahidol University and at the Khlong Yong Health Promoting Hospital in Nakhon Pathom, Thailand. All details of the study were explained to them, and if they were interested in participating in the study, they were asked to sign the written informed consent form. Then, weight and height were measured (to calculate BMI), and a venous blood sample collected to determine Hb, serum ferritin, and C-reactive protein concentrations. Women who fulfilled all the inclusion criteria were instructed to not eat red meat, fish or poultry four days prior to the scheduled test meals.
In a first phase, the women consumed two randomly assigned test meals (A, B, C or D) between 7.00 and 9.00 am after an overnight fast on study day 1 and 2. In the second phase, the women consumed the remaining randomly assigned test meals (A, B, C or D) between 7.00 and 9.00 am after an overnight fast on day 16 or 17. To distinguish between the absorption of the two forms of 57 Fe-labeled FePO 4 -NP (SSA 98 m 2 g −1 and SSA 188 m 2 g −1 ) and between 58 Fe-labeled FeSO 4 and FePO 4 , the subjects consumed one Fe compound labeled with 57 Fe and one with 58 Fe in each phase. The test meals (see details below) were fortified with 2 mg of the respective isotopically labelled iron compound and administered with a glass of deionized water (200 ml). Fourteen days after the last teast meal administration in the first phase (day 16) and again fourteen days after the last test meal administration in the second phase (day 31), a venous blood sample was taken to determine iron absorption.
Composition of the test meal. The test meal was composed of steamed white rice (50 g dry weight), which was served with a vegetable soup prepared from local vegetables (50 g white cabbage, 50 g Chinese cabbage, 30 g Thai mushrooms and 20 g steamed carrots) in 120 mL of water. All ingredients were purchased in bulk and used for the entire study. The food portions were kept frozen until use, and each portion was microwaved on the day of feeding.
Stable-isotope labels. 58 FeSO 4 were prepared from 58 Fe-enriched elemental iron (> 99.8% isotopic enrichment; Chemgas) by dissolution in 0.1 mol/L sulfuric acid. 58 FePO 4 was prepared from 58 Fe-enriched elemental iron (> 99.8% isotopic enrichment; Chemgas). The two FePO 4 -NP (SSA 98 m 2 g −1 and 188 m 2 g −1 ) were prepared from 57 Fe-enriched elemental iron (> 99.8% isotopic enrichment; Chemgas) by flame spray pyrolysis as described previously 5 . We analyzed the labeled iron compounds for iron isotopic composition and the tracer iron concentration via isotope-dilution mass spectrometry as described below. www.nature.com/scientificreports/ Laboratory analyses. Venous blood samples were drawn into EDTA-treated tubes. We measured Hb immediately after blood draw with the use of hematology analyzer (Sysmex, Kobe, Japan) and quality controls provided by the manufacturer before each assessment. Hemoglobin typing for β-globin abnormality was done by using HPLC (Variant Hemoglobin Testing System; BioRad, Hercules. CA) with calibrators and controls provided by the manufacturer. DNA analysis for α-globin abnormalities was done by using a GeneAmp PCR System (Applied Biosystem, Foster City, CA) and a Gel Doc 2000 Gel Documentation System (BioRad, Hercules, CA). A 500 µL aliquot of whole blood was frozen for isotopic analysis (see below). The remaining blood was centrifuged and the plasma aliquoted and stored at − 20 °C. Whole blood and plasma were shipped frozen to the ETH Zurich, Switzerland, for analysis of iron (plasma ferritin and soluble transferrin receptor) and inflammation (C-reactive protein, alpha-1-acid glycoprotein) parameters using a multiplex immunoassay 53 . Anemia was defined as Hb < 120 g/L 54 . Iron deficiency was defined as plasma ferritin < 12 mg/L and/or soluble transferrin receptor > 8.3 µg/mL 53 , and iron deficiency anemia was defined as Hb < 120 g/L 54 and plasma ferritin < 12 mg/L and/or soluble transferrin receptor concentration > 8.3 µg/ml 53 . Normal C-reactive protein and alpha-1-acid glycoprotein concentrations for this assay in healthy adults are < 5 mg/L and < 1 g/L, respectively 53 .
Calculation of iron absorption. Whole blood samples were mineralized in duplicate with the use of a nitric acid and microwave digestion followed by separation of the iron from the blood matrix via anion-exchange chromatography and a subsequent precipitation step with ammonium hydroxide. We measured iron isotope ratios by using an inductively coupled plasma mass spectrometer (Neptune, Thermo Finnigan, Germany) equipped with a multicollector system for simultaneous iron beam detection 55 . We calculated the amounts of 57 Fe and 58 Fe isotopic labels in blood 14 days after administration of the test meals based on the shift in iron-isotopic ratios and the estimated amount of iron circulating in the body. We remeasured the baseline isotopic composition in blood at the start of the second phase of test meals. Circulating iron was calculated based on the blood volume that was estimated from body length and weight at endpoint measurement according to Linderkamp et al. 56 and measured Hb (mean Hb from baseline and endpoint). The calculations were based on the methods described by Turnlund et al. 57 and Cercamondi et al. 58 , taking into account that iron isotopic labels are not monoisotopic.
Data and statistical analysis. Data were analyzed using SPSS (IBM SPSS statistics, version 22.0). Normally distributed data were presented as mean ± SD, and not normally distributed data as median (IQR). Repeatedmeasures ANOVA was used to assess the effect of iron compound on square root-transformed fractional iron absorption. Fractional iron absorption was the dependent variable and the iron compound was added to the model as the independent variable; pairwise comparisons were performed using two-sided paired t-tests with Bonferroni adjustment for multiple testing. Separate linear regressions were done to compare predictors of iron absorption from the nano-sized iron compounds and the bulk sized compounds. C-reactive protein and serum ferritin (iron status) were added as independent variables with the dependent variable being (1) iron absorption from bulk FePO 4 and FeSO 4 and (2) iron absorption from the two FePO 4 -NPs. Significance was set at P < 0.05.

Data availability
All relevant data are included in the manuscript. These are also available by the authors upon request.