Antenatal iron supplementation, FGF23, and bone metabolism in Kenyan women and their offspring: Secondary analysis of a randomized controlled trial

Background: Fibroblast growth factor-23 (FGF23) regulates body phosphate homeostasis primarily by increasing phosphaturia. It also acts as a vitamin D-regulating hormone. Maternal iron deficiency is associated with perturbed expression and/or regulation of FGF23 and hence might be implicated in the pathogenesis of hypophosphatemia-driven rickets in their offspring. Objectives: We aimed to determine the effect of antenatal oral iron supplementation on FGF23 concentration and maternal and infant markers of bone-mineral regulation. Methods: We performed a secondary analysis of a trial in which 470 rural Kenyan women with singleton pregnancies and hemoglobin concentrations ≥ 90 g/L were randomly allocated to daily, supervised supplementation with 60 mg elemental iron as ferrous fumarate or placebo from 13-23 weeks of gestation until 1 mo postpartum. As previously reported, iron supplementation improved iron status in mothers and neonates. For the present study, we reanalyzed all available plasma samples collected in mothers and neonates at birth, with primary outcomes being concentrations of FGF23, measured by 2 assays: 1 that detects intact hormone and C-terminal cleavage products (total-FGF23) and another that detects the intact hormone only (intact-FGF23). Results: Analysis was performed on 433 women (n = 216, iron group; n = 217, placebo group) and 414 neonates (n = 207, iron group; n = 207, placebo group). Antenatal iron supplementation reduced geometric mean total-FGF23 concentrations in mothers and neonates by 62.6% (95% CI: 53.0%, 70.3%) and 15.2% (95% CI: -0.3%, 28.4%, P = 0.06), respectively. In addition, it increased geometric mean neonatal intact-FGF23 concentrations by 21.6% (95% CI: 1.2%, 46.1%), increased geometric mean maternal hepcidin concentrations by 136.4% (95% CI: 86.1%, 200.3%), and decreased mean maternal 25-hydroxyvitamin D concentrations by 6.1 nmol/L (95% CI: -11.0, -1.2 nmol/L). Conclusions: Analysis of this randomized trial confirms that iron supplementation can reverse elevated FGF23 production caused by iron deficiency in iron-deficient mothers and their neonates. Further investigations are warranted to assess to what extent iron supplementation can prevent FGF23-mediated hypophosphatemic rickets or osteomalacia. The present study used samples and data that were collected as part of a randomized placebo-controlled trial in rural Kenya ({“type”:”clinical-trial”,”attrs”:{“text”:”NCT01308112″,”term_id”:”NCT01308112″}}NCT01308112) originally designed to measure the effect of antenatal iron supplementation on maternal Plasmodium infection risk, maternal iron status, and neonatal outcomes. The recruitment for the original trial was conducted between 2011 and 2013. Study details and main results have been published elsewhere (12). In brief, pregnant women aged 15–45 y with singleton pregnancies, gestational age of 13–23 wk (determined by obstetric ultrasonography), and hemoglobin concentration ≥ 90 g/L were individually randomly assigned at a 1:1 allocation ratio to 60 mg of elemental iron as ferrous fumarate or placebo until 1 mo postpartum. The supplements contained no other micronutrients and their neonates did not receive any additional supplements. From screening until the end of the intervention, local mill operators added fortificant iron (target dose: 20 mg/kg flour) to grain routinely brought for milling by homestead members of all the participating women. Based on weighed intake studies, we estimate that fortification supplied on average 5.7 mg of elemental iron as ferric sodium ethylenediaminetetraacetate daily to the pregnant women in both arms of the intervention. Participants and field staff were blinded to the randomization and intervention until data analysis. All women provided written informed consent. Approval for the original study was obtained from ethics committees at the London School of Hygiene and Tropical Medicine, United Kingdom, and the Kenyatta National Hospital/University of Nairobi (ethics approval number KNH-ERC/A/453, 6/01/2010); additional approval was obtained from the latter committee (KNH-ERC/A/272, P280/05/2017) to perform additional analyses on archived plasma samples to generate the data presented in this report. Maternal weight and height were measured at baseline, and neonatal weight, length, and head circumference were measured at birth. At baseline, we collected venous blood samples in EDTA-coated tubes from women. We also collected maternal venous blood, neonatal cord blood, and placental biopsies within 1 h postpartum. For home deliveries, samples were collected within 2 h postpartum. Plasma was stored within 2 h of blood collection in liquid nitrogen (−196°C) and subsequently frozen (−80°C) until analysis. We performed ELISA tests at the Medical Research Council Elsie Widdowson Laboratory, Cambridge, United Kingdom to measure the following markers in plasma collected at delivery: total-FGF23 using the C-terminal assay (60-6100) which detects both the C-terminal fragment and the intact hormone, and intact-FGF23 using the intact-FGF23 assay (60-6600) which detects only the intact form of the hormone (Immutopics); 1,25(OH)2D (the active metabolite of vitamin D; Immunodiagnostic Systems); β-C-terminal telopeptide (β-Crosslaps—a marker of bone resorption that is elevated in increased bone resorption; Immunodiagnostic Systems); and hepcidin (a regulator of iron metabolism that is low in iron deficiency and raised by inflammation; Hepcidin-25, Bioactive). Hemoglobin concentration was measured in the field by photometer (HemoCue 301, Radiometer) and plasma α1-acid glycoprotein (a marker of inflammation) and soluble transferrin receptor (sTfR—a marker of iron status) were measured at Meander Medical Centre, Amersfoort, Netherlands (UniCel DxC 880i analyzer, Beckman Coulter) as part of the original trial (12). Plasma concentrations of ferritin, C-reactive protein (CRP—a marker of inflammation), 25-hydroxyvitamin D [25(OH)D—marker of vitamin D status], total alkaline phosphatase, intact parathyroid hormone (PTH—a primary regulatory hormone of calcium that is elevated in calcium deficiency), cystatin C (a marker that is elevated in kidney dysfunction), and phosphate were measured at Meander Medical Centre, Amersfoort, Netherlands (Architect c16000 and Architect i2000SR, Abbott). ELISA assay accuracy was monitored across the working range of assays using kit controls supplied by the manufacturers. The UK laboratory was accredited by the Vitamin D External Quality Assessment Scheme (http://www.deqas.org/) and the UK National External Quality Assessment Service (https://ukneqas.org.uk/) and the laboratory in the Netherlands by the External Quality Assurance System (https://www.eurl-ar.eu/eqas.aspx) and German Society for Clinical Chemistry and Laboratory Medicine (DGKL) (https://www.dgkl.de/en/). In addition, an aliquot of a pooled plasma sample was assayed in each batch to monitor possible drift in measurements over time. Intra- and interassay CVs were <7% and <6% for total-FGF23, intact-FGF23, β-Crosslaps, 1,25(OH)2D, 25(OH)D, PTH, ferritin, CRP, phosphate, total alkaline phosphatase, and cystatin C and <5% and <17% for hepcidin, respectively. Estimated glomerular filtration rate (eGFR—a measure of kidney glomerular function) was calculated using equations approved by the National Kidney Foundation (https://www.kidney.org/professionals/KDOQI/gfr)—the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) cystatin C equations for those aged >18 y: eGFR, mL·min−1·1.73 m−2 = 123.9 × [cystatin C (mg/L)/0.8]−1.328 × 0.996age and for those aged ≤18 y: eGFR, mL·min−1·1.73 m−2 = 70.69 × [cystatin C (mg/L)]−0.931. At baseline, dipstick tests (Access Bio) were used to detect histidine-rich protein-2 and lactate dehydrogenase specific either to P. falciparum or to nonfalciparum human Plasmodium species (12). qPCR was used to detect P. falciparum–specific DNA in erythrocytes, and mothers were also tested for HIV infection. HIV-infected mothers continued or were offered antiretroviral treatment as part of their standard clinical care. The primary analysis concerned group differences in plasma concentrations of total-FGF23 and intact-FGF23 in maternal blood samples at delivery and neonatal cord blood samples. As secondary outcomes, we studied other biomarkers of bone metabolism and kidney function, namely, plasma concentrations of phosphate, PTH, total alkaline phosphatase, β-Crosslaps, 25(OH)D, 1,25(OH)2D, cystatin C, and eGFR. Sample size requirements were calculated based on the effect of iron supplementation on Plasmodium infection risk (12) as part of the original trial design; because they are not relevant to the current study, they are not reported here. The current study performed analysis on all available plasma samples from mothers at birth (n = 433) and from infant cord blood (n = 414) at delivery. Anthropometric z scores for the neonates at birth were derived with Kenyan children as a reference (19). The following definitions were used: prematurity–being born at or before 37 completed weeks of gestation (<259 days of gestation as calculated by early pregnancy ultrasound); anemia: hemoglobin concentration < 110 g/L for pregnant women (19); iron deficiency (depleted iron stores)–plasma ferritin concentration <15 µg/L for women and <12 µg/L for neonates (20); inflammation–plasma CRP > 10 mg/L (21); vitamin D insufficiency–25(OH)D concentration <50 nmol/L; and vitamin D deficiency–25(OH)D concentration <30 nmol/L (22). There are no established thresholds for intact-FGF23 or total-FGF23 concentrations or for eGFR in pregnancy and neonates, or for anemia in neonates, and so these variables were not dichotomized. Plasmodium infection was defined as past or present maternal infection assessed at parturition, regardless of species, as indicated by ≥1 positive test results for the presence of Plasmodium lactate dehydrogenase or P. falciparum–specific histidine-rich protein-2 in plasma or by placental histopathology or P. falciparum DNA in maternal erythrocytes from venous or placental blood by a PCR test. Statistical analysis was performed using Stata version 16.0 (StataCorp). We visually inspected histograms to assess the shape of the distribution and to identify possible outliers. Skewed data were normalized by log transformation as appropriate. Groups were described using mean ± SD for normally distributed data and geometric mean ± geometric SD (GSD) for log-transformed data. GSD was calculated as the exponentiated SD of the log-transformed variable. It is a dimensionless factor that indicates variation that is equivalent to subtraction or addition of 1 SD on a log-transformed scale. For plasma CRP concentration at baseline, we computed descriptive statistics with a Tobit model to account for data being left-censored at the limit of quantification (1 mg/L). Plasma concentrations of ferritin and soluble transferrin receptor (sTfR) are affected by infection and inflammation independently of iron status. We used multiple regression models to adjust the iron markers for such effects (Supplemental Methods). We used adjusted plasma iron markers to calculate the body iron index, i.e., the ln of the ratio of the adjusted ferritin concentration to the adjusted sTfR concentration. This indicator has been shown to be linearly associated with quantitative estimates of the size of the body iron store in iron-replete adults, and with the size of the functional deficit that would need to be corrected before iron could again be accumulated in the store in iron-deficient individuals (23). Crude intervention effects on continuous outcomes were estimated by simple linear regression. To assess the potential role of confounding due to imbalances in baseline variables, we used multiple fractional polynomial regression to estimate intervention effects adjusted for maternal characteristics assessed at randomization, i.e., hemoglobin concentration, body iron index, age, BMI, gestational age at delivery (calculated from early pregnancy ultrasonography), parity, HIV infection, and Plasmodium infection. Both in simple linear regression models and in fractional polynomial models, we accounted for heteroscedasticity of the error terms as appropriate (P values for Breusch–Pagan/Cook–Weisberg tests < 0.05). Intervention effects are reported as absolute differences in means for normally distributed outcomes, or as relative differences in geometric means for log-transformed outcomes. For unadjusted prevalence differences, we used Newcombe's method to estimate 95% CIs and “N−1” chi-square tests to compute P values (24). We used log-binomial regression models (adjrr command in Stata package st0306.pkg) to estimate prevalence differences adjusted for baseline characteristics, and to estimate unadjusted prevalence differences when contingency tables contained cells with expected values <10. In a preplanned analysis, we used multiple fractional polynomial regression analysis to explore to what extent iron status at baseline modified the magnitude of the effect of iron supplementation on FGF23 and selected markers of bone metabolism at delivery, anticipating that iron absorption and thus the response to administered iron would be larger in iron-deficient women than in their iron-replete peers. We examined such effect modification with a single independent variable (body iron index) and 10 outcomes (see the Outcomes section), in both mothers and neonates, which resulted in 20 analyses. We used the mfpi procedure in Stata with the “flex(3)” specification to define the flexibility of the main effects and interaction models with adjustment for potentially influential maternal characteristics assessed at randomization, i.e., hemoglobin concentration, BMI, gestational age, parity, HIV infection, and Plasmodium infection. We used a nominal significance level of 0.05 for selection of variables and power functions; selection of linear, first-degree, or second-degree polynomials was based on the lowest value for Akaike's information criterion. To check for possible overfitting in the interaction models, we examined to what extent possible trends in intervention effects observed across quintiles of body iron index were consistent with effect modification as measured by fractional polynomial regression.