Use of Model-Based Compartmental Analysis and a Super-Child Design to Study Whole-Body Retinol Kinetics and Vitamin A Total Body Stores in Children from 3 Lower-Income Countries

ABSTRACT Background Model-based compartmental analysis has been used to describe and quantify whole-body vitamin A metabolism and estimate total body stores (TBS) in animals and humans. Objectives We applied compartmental modeling and a super-child design to estimate retinol kinetic parameters and TBS for young children in Bangladesh, Guatemala, and the Philippines. Methods Children ingested [13C10]retinyl acetate and 1 or 2 blood samples were collected from each child from 6 h to 28 d after dosing. Temporal data for fraction of dose in plasma [13C10]retinol were modeled using WinSAAM software and a 6-component model with vitamin A intake included as weighted data. Results Model-predicted TBS was 198, 533, and 1062 μmol for the Bangladeshi (age, 9–17 mo), Filipino (12–18 mo), and Guatemalan children (35–65 mo). Retinol kinetics were similar for Filipino and Guatemalan groups and generally faster for Bangladeshi children, although fractional transfer of plasma retinol to a larger exchangeable storage pool was the same for the 3 groups. Recycling to plasma from that pool was ∼2.5 times faster in the Bangladeshi children compared with the other groups and the recycling number was 2–3 times greater. Differences in kinetics between groups are likely related to differences in vitamin A stores and intakes (geometric means: 352, 727, and 764 μg retinol activity equivalents/d for the Bangladeshi, Filipino, and Guatemalan children, respectively). Conclusions By collecting 1 or 2 blood samples from each child to generate a composite plasma tracer data set with a minimum of 5 children/time, group TBS and retinol kinetics can be estimated in children by compartmental analysis; inclusion of vitamin A intake data increases confidence in model predictions. The super-child modeling approach is an effective technique for comparing vitamin A status among children from different populations. These trials were registered at www.clinicaltrials.gov as NCT03000543 (Bangladesh), NCT03345147 (Guatemala), and NCT03030339 (Philippines).


Supplemental Methods
Subjects. Bangladeshi infants were recruited from the urban slum areas of Kamrangir Char and Hazaribagh in Dhaka; study procedures were approved by the Institutional Review Board of the International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b). In Guatemala, subjects were recruited from several urban centers (Santa Catarina Pinula and La Comunidad, Mixco) and from 2 semi-urban communities (San Miguel Dueñas and Santa María de Jesus); procedures were approved by the Institutional Ethics Committee of INCAP. Filipino children were recruited from low-income neighborhoods in Mandaluyong City within Metro-Manila; the study protocol was approved by the Institutional Review Board at the University of California, Davis and by the Research Ethics Board of the University of the Philippines, Manila. All studies were also approved by the Research Ethics Board of Newcastle University.
Assessment of vitamin A intake. Vitamin A intake was estimated for the Bangladeshi children based on measured breastmilk vitamin A concentrations and published estimates of breastmilk intake (Supplemental References 1,2) and by doing 24 h dietary recalls on 2 separate days and a 30 d food frequency questionnaire. Estimates for dietary intake were determined using a food composition table from the Institute of Nutrition and Food Science at the University of Dhaka and included both preformed vitamin A and provitamin A carotenoids [as RAE based on the IOM conversion factors (Supplemental Reference 3)]. For the Guatemalan children, dietary vitamin A intake was assessed by using a 30 d food frequency questionnaire as well as 24 h recalls done on 3 separate days during the study; the latter included weighing mother-reported portions of consumed foods. Estimated intake, based on the INCAP Food Composition Table, included both preformed vitamin A and provitamin A carotenoids [as RAE using the IOM conversion factors (Supplemental Reference 3)]. For the Filipino children, dietary vitamin A intake was assessed by conducting 24 h recalls on 4 -5 separate days during the study, a 12 h in-home observation to weigh and record the child's food intake (with 12 h recall to complete the 24 h period), and a 30 d questionnaire to assess use of vitamin A-containing supplements. For breastfeeding children, breast milk intake was measured using the 2 H2O dose-to-mother method (Supplemental References 4,5). Results were combined to estimate usual vitamin A intake using the US National Cancer Institute method (Supplemental Reference 6); the estimate included preformed vitamin A and provitamin A carotenoids [as RAE based on the IOM conversion factors (Supplemental Reference 3)].
Calculation of vitamin A intake for modeling. Because assessed vitamin A intake included both preformed and provitamin A, we adjusted the value for provitamin A carotenoids (as RAE) to the equivalent amount of preformed vitamin A. That is, because the conversion of provitamin A carotenoids to RAE [using the IOM conversion factors (Supplemental Reference 3)] includes absorption and conversion efficiency, and we assumed an absorption efficiency of 80% for preformed vitamin A for modeling, the value for vitamin A intake used as weighted input in the model was calculated as: preformed vitamin A + (provitamin A / 0.8). More specifically, for the Bangladeshi and Guatemalan children, the individual assessed preformed and provitamin A intakes were used in the above calculation; the geometric mean of the individual adjusted estimates was used in the model. For the Filipino children, the geometric mean value for assessed preformed and provitamin A intake was used in the above calculation. Preformed vitamin A contributed, on average, >95% of total vitamin A intake for the Bangladeshi subjects, >90% for the Filipinos, and ~76% for the Guatemalan children.
Retinol analyses. Retinol was extracted from plasma using a modification of the method of Aebischer et al. (16). Briefly, 400 µL of plasma was diluted with 400 µL diH20 before addition of 800 µL ethanol containing 50 pmol [ 12 C]retinyl acetate as internal standard. The denatured sample was extracted twice with 2 mL of hexane and hexane in the combined extracts was evaporated under N2. The residue was resuspended in 100 µL of ethanol and 10 µL was used for LC-MS/MS analysis. Liquid chromatographic conditions were adapted from Kane and Napoli (17) to a Shimadzu Prominence HPLC equipped with a 100 mm x 2.1 mm 3 µm Supelcosil ABZ PLUS column (Supelco) and a 4 x 2 mm SecurityGuard (Phenomenex) C18 cartridge maintained at 25°C. The binary mobile phase system consisted of 0.1% formic acid aqueous (A) and 0.1% formic acid in acetonitrile (B); separation of analytes was done at a flow rate of 400 µL/min and a linear gradient of 60% B to 95% B from 0.0 -6.0 min; 100% B from 6.1 -13.0 min; 60% B from 13.0 -14.0 min; and then 60% B from 14.0 -17.0 min to re-equilibrate the column. The HPLC was coupled to an API 4000 MS/MS system (AB SCIEX) with atmospheric pressure chemical ionization in positive ion mode. Optimized atmospheric pressure chemical ionization source parameters were: CUR 20 psig; GS1 40 psig; GS2 40 psig; CAD 6 psig; TEM 300°C; NC 5 µA. Selected reaction monitoring and quantitation of [ 12 C]retinol, [ 12 C]retinyl acetate, and [ 13 C10]retinol was done as previously described (18).

Plasma volume estimation.
Calculation of fraction of dose in plasma required an estimate of plasma volume for each child. For this work, we estimated plasma volume using the regression equation of Linderkamp et al. (19). Specifically, blood volume (L) was estimated based on body weight (kg) and height (cm); then, hemoglobin concentration was used to estimate hematocrit as hemoglobin × 3 (Supplemental Reference 7), and plasma volume was estimated as blood volume × (1hematocrit).
Rationale for updated hypotheses in the compartmental model. The model shown in Figure 1 was adapted from a previously-published model for vitamin A kinetics in children (8). Specifically, in addition to irreversible loss from the larger storage compartment 6, the current model includes an additional site of loss from extrahepatic tissues that do not recycle retinol to plasma (component 8). The hypothesis that 2 tissue sites of irreversible loss are required was supported by modeling tracer data for plasma and loss in rats [unpublished observations, MH Green, based on data presented in (Supplemental Reference 8)] in which, to adequately fit the loss data, the model required output from both an exchangeable and a nonexchangeable extravascular compartment (analogous to compartment 6 and component 8, respectively). In that analysis, a delay of 75 min was required to fit the loss data. Thus, we fixed the delay time in component 8 [DT (8)] at 75 min for all 3 groups in the current analysis. Note that the value assigned to DT(8) does not influence model results and was chosen to provide a more physiologically-sound model. Since further testing using theoretical data (unpublished observations, JL Ford and MH Green) indicated that, without loss data, one cannot accurately distinguish the proportion of output from these 2 sites, for the current work, we partitioned the rate of loss from tissues so that 50% of the output (i.e., disposal rate) was from tissue component 8 and 50% was from storage compartment 6.

Supplemental Discussion
Future recommendations for the super-child approach. Based on the current analysis, the following revised design is recommended for future super-child studies (Supplemental Figure  1). We recommend that there be a sample size of at least 60 children per group and that 2 blood samples be collected from each child at times (n=13) from 5 h to 42 d. Based on sensitivity analysis performed using WinSAAM (Supplemental Reference 9), the recommended sampling times are 5, 8, and 12 h, and 1, 2, 4, 7, 10, 14, 21, 28, 35, and 42 d. All children should be sampled at 7 d (see below) and at one additional time, randomly assigned at recruitment, such that 5 children will be sampled at each of the 12, non-7 d times.
Accuracy related to administration of the labeled oral dose is crucial. That is, the time and volume or, preferably, the weight of the dose administered needs to be precisely recorded at the time of dosing (d 0). The amount of label administered to each child must be adequate to ensure accurate detection 42 d later, based on detection capability of the equipment that will be used for analysis.
For the current modeling analysis, absorption efficiency for preformed vitamin A was assumed to be 80% (24,27). To improve the accuracy of predictions by the model, investigators should consider actually determining absorption of vitamin A. For example, the dual isotope absorption method (Supplemental Reference 10) can be used to measure vitamin A absorption by administering an oral dose of labeled vitamin A followed ~3 h later by an intravenous dose of a differently-labeled vitamin A in a physiological lipid emulsion; absorption can be calculated as the ratio of fraction of dose of orally ingested vitamin A / intravenously injected vitamin A at a time after the ratio in plasma has stabilized (e.g., 14 d in rats).
In addition to predicting a group mean value for TBS from modeling the composite dataset, data from the plasma samples collected on d 7 can be used in an RID equation to estimate TBS in individual children. Since in more recent work (15), we found that 7 d was a better time than 4 d for doing RID in children, we recommend the former time (7 d) for future studies. Specifically, the super-child model can be used to calculate time-variant values for the equation coefficients (Fa and S) to be applied in the equation along with each child's retinol specific activity in plasma (SAp) at the corresponding time (d 7). Note that values for the coefficients can be calculated at all times and values from 4 to 42 d can be used in the equation to predict TBS for individuals sampled at those times. It is important to mention that, while we hypothesize that 7 d will be more accurate than 4 d for estimating the absolute value for TBS in individual children by RID when a group model-predicted value for FaS is used, TBS predictions for the group should be approximately the same at both times. In addition, use of 4 d data can still be valuable for ranking subjects within a group from low to high TBS and estimating the distribution and range of TBS for the group.  Table 2 for statistical uncertainties for these parameters. The model is shown in Figure 1. TBS, total body stores.  Figure 1.    (7))*(F(5)/M(5))/ ((F(6)+F (7))/(M(6)+M (7)