Biofortified and fortified maize consumption reduces prevalence of low milk retinol, but does not increase vitamin A stores of breastfeeding Zambian infants with adequate reserves: a randomized controlled trial

ABSTRACT

Background

Replacement of conventional staples with biofortified or industrially fortified staples in household diets may increase maternal breast milk retinol content and vitamin A intakes from complementary foods, improving infant total body stores (TBS) of vitamin A.

Objectives

To determine whether biofortified or industrially fortified maize consumption by Zambian women and their breastfeeding infants could improve milk retinol concentration and infant TBS.

Methods

We randomly assigned 255 lactating women and their 9-mo-old infants to a 90-d intervention providing 0 µg retinol equivalents (RE)/d as conventional maize or ∼315 µg RE/d to mothers and ∼55 µg RE/d to infants as provitamin A carotenoid-biofortified maize or retinyl palmitate–fortified maize. Outcomes were TBS, measured by retinol isotope dilution in infants (primary), and breast milk retinol, measured by HPLC in women (secondary).

Results

The intervention groups were comparable at baseline. Loss to follow-up was 10% (n = 230 mother-infant pairs). Women consumed 92% of the intended 287 g/d and infants consumed 82% of the intended 50 g/d maize. The baseline geometric mean (GM) milk retinol concentration was 1.57 μmol/L (95% CI: 1.45, 1.69 μmol/L), and 24% of women had milk retinol <1.05 μmol/L. While mean milk retinol did not change in the biofortified arm (β: 0.11; 95% CI: −0.02, 0.24), the intervention reduced low milk retinol (RR: 0.42; 95% CI: 0.21, 0.85). Fortified maize increased mean milk retinol (β: 0.17; 95% CI: 0.04, 0.30) and reduced the prevalence of low milk retinol (RR: 0.46; 95% CI: 0.25, 0.82). The baseline GM TBS was 178 μmol (95% CI: 166, 191 μmol). This increased by 24 µmol (± 136) over the 90-d intervention period, irrespective of treatment group.

Conclusions

Both biofortified and fortified maize consumption improved milk retinol concentration. This did not translate into greater infant TBS, most likely due to adequate TBS at baseline. This trial was registered at clinicaltrials.gov as NCT02804490.

vitamin A, biofortification, fortification, breast milk, total body stores

Introduction

See corresponding article on page 1077.

See corresponding article on page 1322.

An estimated 48% of infants and young children in sub-Saharan Africa are affected by vitamin A deficiency (1). High-dose vitamin A supplementation strategies have been widely implemented, likely preventing blinding xerophthalmia on the continent and contributing to declines in under-5 mortality over the past 2 decades (2). However, less progress has been achieved towards ensuring adequate dietary vitamin A intakes (3). Both industrial-scale food fortification and breeding programs to improve the provitamin A carotenoid content of staple crops are important public health strategies, intended to replace nutrient-poor staples, thereby filling the dietary gap. Current evidence to support the scale-up of industrial fortification as a vitamin A deficiency control intervention is somewhat limited (4). Greater attention has been paid to biofortification, where significant improvements have been demonstrated in serum provitamin A carotenoid concentrations (5, 6) and total body vitamin A stores in preschool-aged children (7), as well as evidence of improvements in breast milk retinol with a short-term feeding intervention (8).

Infancy is a period of particular public health concern. While vitamin A status often goes unmeasured in this age group, it is well accepted that infants are born with low vitamin A stores, regardless of their mother's status (9). They are heavily reliant on breast milk retinol and, to a lesser extent, vitamin A provided by complementary foods, to build the stores necessary to support their immune defenses during 1 of the highest-risk periods for infectious morbidity. However, in populations with poor status, mothers often have low concentrations of vitamin A in their milk (10). This is often coupled with inadequate infant intakes from other dietary sources (10). WHO guidelines call for vitamin A supplementation of infants 6–11 mo of age (11); however, this 105 µmol dose, generally provided at 6 mo of age, is thought to have only a transient impact on status (12). We designed the present trial to test whether the replacement of conventional maize with biofortified or industrially fortified maize as the staple in diets of lactating Zambian mothers and their infants may improve total body stores (TBS) of vitamin A in the latter half of infancy.

Methods

Subjects

This research was carried out in the Mkushi District in the Central Province of Zambia. Mkushi was selected due to its high prevalence of undernutrition and poverty, and its heavy reliance on maize as a staple crop. A survey carried out in 2009 indicated that ∼42% of children aged 2–5 y in Mkushi had low serum retinol concentrations (<0.70 μmol/L), after adjusting for elevated acute phase proteins, and that 15.6% (3.0 SE) of children had modified relative dose response values >0.06, indicating inadequate liver vitamin A stores (13). Data from our previous trials in the Mkushi District also indicated a 63% prevalence of marginal vitamin A status among children 4–8 y of age (5) and a 56% prevalence of low breast milk retinol concentrations among lactating women aged 18–48 y (8). This trial was carried out in the District's commercial center.

We registered all infants attending under-5 clinic visits at the Mkushi District Hospital over the period from July 2015 to September 2016. We approached a subset of families, based on residence in the highest population-density neighborhoods, for consent and enrollment into a surveillance cohort. We visited these families monthly to collect data on maternal and infant diet and morbidity. Data were collected using electronic forms developed with OpenDataKit version 1.1.7 (www.opendatakit.org). Mothers were also asked to bring their infants to our study clinic for immunizations, vitamin A supplementation (105 µmol at 6 mo and 210 µmol at 1 y), and growth monitoring. While vitamin A capsules were provided as a service, in accordance with WHO guidelines and Zambian policy, integration of this activity into the research protocol enabled study clinic staff to tightly control and document the timing of supplementation relative to the trial activities. In the first 100 women recruited into the surveillance cohort, we collected milk samples to assess whether statuses were similar in the somewhat larger and more diverse study area targeted in the present trial. This was done by asking the subject to feed her infant for 1 min and then collecting a 10-mL midstream sample from the left breast after 1 min. Samples were stored on ice prior to analysis.

Only apparently healthy, singleton infants who received their 6-mo vitamin A capsule (105 µmol) from our study clinic staff were considered potentially eligible for the trial. We visited these families in their homes roughly 1 wk prior to the infant's 9-mo birthday to obtain informed consent for additional screening procedures and, if eligible, trial enrollment. We verified maternal age, intention to remain in the study area, current breastfeeding status, and intention to continue breastfeeding throughout the trial period. A finger-prick blood sample was taken from women and infants to test for hemoglobin (HemoCue Hb 201+; Hemocue) and malaria parasitemia (rapid diagnostic test; SD Bioline Malaria Ag P.f, 05FK50; Standard Diagnostics). A spot urine test was used to test for human chorionic gonadotropin. Nonsevere anemia and/or malaria were treated in accordance with Zambian guidelines (iron supplementation for anemia; artemether/lumefantrine for uncomplicated malaria). We provided referrals for the following: pregnancy, severe anemia (Hb < 8.0 g/dL for women; Hb < 7.0 g/dL for infants), signs of severe malaria, or other severe illness.

Mother/infant pairs were considered eligible for the trial if the mother was aged 18–45 y, had a hemoglobin concentration ≥8.0 g/dL, was free from chronic health conditions (i.e., any issue requiring regular medical visits), was breastfeeding and planning to continue through ≥12 mo postpartum, was not currently pregnant, and was not planning to relocate during the timeframe of the trial. Infants had to have a hemoglobin concentration ≥7.0 g/dL and have received a vitamin A capsule (105 µmol) at 6 mo of age. The exclusion criteria were a chronic health condition in the mother or infant; severe anemia in the mother (Hb < 8.0 g/dL) or infant (Hb < 7.0 g/dL); pregnancy; not currently breastfeeding or planning to cease breastfeeding prior to the infant's first birthday; infant not receiving the 105 µmol dose of vitamin A at ∼6 mo; or intent to move from the study area.

We individually randomized mother/infant pairs at a 1:1:1 ratio to receive: 1) conventional low-carotenoid white maize (WM); 2) biofortified orange maize (OM); or 3) retinyl palmitate–fortified white maize (FM). The randomization scheme was prepared by an independent statistician using SAS version 12.3 (SAS Institute). A random number ranging from 0 to 1 was drawn from a uniform distribution and assigned to each of 255 sequential, unique identifiers. These were then sorted by their random numbers and divided by tertiles, with the first third assigned to group 1, the second third to group 2, and the final third to group 3. This blocking was done to ensure adherence to the 1:1:1 ratio and achieve groups equal in size. The randomization scheme was maintained by the project's data manager in an access-restricted encrypted database, concealed from all staff involved in enrollment, data collection, or intervention delivery. Consent forms were received by the data manager at the end of each workday, with unique identifiers for consenting mother/infant pairs entered, in order of receipt, into the project database. This registration of consent for each mother/infant pair triggered their assignment to the next available treatment allocation in the database. Allocations were then transmitted to the intervention delivery team. Staff responsible for collecting data/biospecimens remained blinded to the intervention assignment.

The protocol for this trial was reviewed and approved by the Institutional Review Boards of Eres Converge in Zambia, the Johns Hopkins Bloomberg School of Public Health, and the University of California, Davis. Authority to conduct research was obtained from the Ministry of Health. The trial was registered at clinicaltrials.gov as NCT02804490.

Sample size

Sample size estimates were based on the expected difference in the infant's TBS over a 90-d period, from ∼9 to 12 mo of age. We first estimated infant vitamin A intakes based on: 1) reported values for breast milk intake (14); 2) the vitamin A concentration of breast milk previously measured at this site (8); 3) vitamin A intake from the assigned maize intervention, as detailed below; and 4) an assumed intake of ∼100 µg retinol equivalents (RE)/d from nonstudy foods. This yielded vitamin A intake estimates of 300 µg RE/d in the WM group and 411 µg RE/d in the OM and FM groups. We then estimated the effects of these estimated dietary vitamin A intakes on TBS in the WM, OM, and FM groups, assuming that 70% of dietary vitamin A was retained (7) and that 2.2%/d of TBS was catabolized (15, 16). Based on these estimates, we expected to observe a 10.5 µmol difference in TBS between the OM or FM groups and the WM group. A sample size of 195 mother-infant pairs, or 65 per group, was calculated to detect a 10.5 µmol difference in TBS (effect size = 0.55) between the OM or FM groups and the WM group, assuming a pooled SD of 18.9 µmol, a 2-sided α of 0.05, and power of 80%. The sample size was increased to 85 per group to account for losses to follow-up. With this sample size, and assuming a SD of 0.70 µmol/L for breast milk retinol (8), we calculated that we would be able to detect a 0.20 µmol/L or greater change in mean breast milk retinol. In order to identify 255 participants, we aimed to enroll 500 mother/infant pairs into the surveillance cohort.

Intervention

Adult women in rural Zambia consume, on average, 287 g dry weight/d of maize (13). Estimates for the 6–12-mo age group are ∼50 g/d (M Angel, HarvestPlus, personal communication, 2015). Biofortified maize bred for nutrition efficacy trials ranges from 15 to 20 μg provitamin A carotenoid per g dry weight, primarily in the form of β-carotene. Assuming 17.5 μg/g, 12:1 bioconversion, and 75% retention poststorage and cooking, average daily consumption would provide ∼315 μg RE/d for women and ∼55 μg RE/d for infants, or 1.09 μg RE/g per maize meal. The fortified maize meal was designed to provide an equivalent amount of preformed vitamin A, factoring in 5% overage to account for losses during milling.

Biofortified maize was produced by HarvestPlus for the purposes of this trial. We purchased white maize from the same harvest season for the WM and FM groups. Maize was milled into 25-kg bags of breakfast meal (64% extraction) at a commercial mill in Lusaka equipped with a microdoser. Fortificant (dry retinyl palmitate; 250 S/N-B) was donated by DSM Nutritional Products, Inc., and premix was produced from a refined breakfast meal filler at the Zambian National Institute of Scientific and Industrial Research in Lusaka. Maize meal was transported to Mkushi on the day of milling and stored in a refrigerated container (−22°C) to prevent carotenoid and fortificant degradation. Maize meal and prepared food samples were shipped on ice to Craft Technologies in Wilson, NC, for analysis by HPLC, as per published methods. Bags were delivered to the kitchen on Monday mornings, and leftover maize meal was discarded on Saturday afternoons.

Trained cooks prepared meals in a central kitchen. Women and infants received breakfast and lunch of maize meal nshima (boiled until stiff) or porridge prepared with WM, OM, or FM for 6 d/wk for 90 d. Meals were prepared in accordance with a standardized, rotating menu of infant porridges and nshima with relish for mothers. WM, OM, and FM were stored and prepared in separate rooms to prevent mixing. Servings were weighed and sealed into a thermos container for infant porridge or into reusable food containers for nshima and relish. Containers were placed in color-coded, insulated bags and delivered to households by project staff. Upon delivery, women were given time to consume the meal and to feed their infant. They were asked not to share food, but to consume as much as possible. Distributors directly observed, at random, at least 1 breakfast and 1 lunch meal each week. Otherwise, meals were left at the household and the distributor returned to retrieve the containers and any leftovers. Leftovers were weighed in their containers and recorded.

Data collection

Trial enrollment began in March 2016 and ended when the sample size was attained in January 2017. Follow-up activities were completed in June 2017. The key activities that follow are summarized in Table 1. At the baseline consent visit (9 mo – 1 wk), we treated women for intestinal helminthiasis (400 mg albendazole) and collected stool samples from infants. Infants testing positive were treated with 100 mg mebendazole. At 9 mo, we collected a 7-d morbidity history for women and infants. Infants with diarrhea or fever were rescheduled for the following week. We collected dietary data by a 24-h dietary recall for women (17) or a 24-h semiquantitative FFQ for infants (18). We measured height or length (Shorrboard portable stadiometer), weight (Seca 874 digital scale), and mid-upper arm circumference (insertion tape) using standard protocols (19). We administered [¹³C₁₀]-retinyl acetate (400 μg in 162 μL sunflower oil; Buchem B.V.) by positive displacement pipet into the infant's mouth and asked women to breastfeed to facilitate isotope absorption. At 4 d postdosing (9 mo + 4 d), we repeated diet and morbidity assessments. A nurse collected a 5-mL infant blood sample by antecubital venipuncture (4 mL EDTA-treated; 1 mL nontreated) and stored the sample on ice packs. Mother/infant pairs returned to the clinic at 10 d postdosing (9 mo + 10 d). Women were asked to breastfeed and, after the breastfeeding episode, were asked to refrain from breastfeeding for 1 h. During this period, we repeated maternal and infant diet and morbidity assessments. We used a digital pupillometer to measure the mother's pupillary response to light stimuli under dark-adapted conditions (20). A 10-mL fasted maternal blood sample (9 mL EDTA-treated; 1 mL nontreated) and a 5-ml infant blood sample (4 mL EDTA-treated; 1 mL nontreated) were collected, and we administered the 9-mo measles vaccine to infants. After 1 h, we asked women to breastfeed from the right breast and, after 1 min, collected all milk from the left breast using a manual breast pump. The collection time and weight of the sample were recorded, and a 10-mL sample was retained for analysis.

TABLE 1

Summary of key dosing or biospecimen collection contacts for women and infants enrolled in a provitamin A carotenoid-biofortified maize or retinyl palmitate–fortified maize intervention trial

Time points		Mother	Infant
6 mo	—	—	Vitamin A supplement (105 µmol)
Baseline: 9 mo	−1 wk	Urine (spot); Capillary blood	Capillary blood; Stool
	+0 d	—	Isotope dosing
	+4 d	—	Venous blood (5 mL)
	+10 d	Venous blood (10 mL); Breast milk (10 mL)	Venous blood (5 mL); Measles vaccination
Endline: 12 mo	−1 wk	Urine (spot); Venous blood (10 mL); Breast milk (10 mL)	Capillary blood; Stool
	+0 d	—	Venous blood (5 mL); Isotope dosing
	+4 d	—	Venous blood (5 mL)
	+10 d	—	Venous blood (5 mL); Vitamin A supplement (210 µmol)

After the 90-d intervention (12 mo − 1 wk), mothers provided a spot urine sample to test for human chorionic gonadotropin and an infant stool sample to test for helminths. Infants were treated, as necessary, with 100 mg mebendazole. We collected maternal and infant diet and morbidity histories, a full milk sample, anthropometric measurements, and pupillometry measures, as described above. A 10-mL fasted venous blood sample was taken from women and a finger-prick sample was taken from infants. The latter was used solely for malaria and hemoglobin testing. One week later (12 mo), we repeated infant diet and morbidity assessments, rescheduling any infants reporting a history of fever or diarrhea to the following week. We collected a 5-mL venous blood sample (4 mL EDTA-treated; 1 mL nontreated) from infants and then administered the second dose of [¹³C₁₀]-retinyl acetate, as described above. At 4 d postdosing (12 mo + 4 d), we collected data on infant morbidity and collected a 5-mL venous blood sample (4 mL EDTA-treated; 1 mL nontreated) from infants. These same activities were repeated at 10 d postdosing (12 mo + 10 d), after which we administered a vitamin A supplement (210 µmol) as per Zambian guidelines.

For all venous blood draws, samples were tested for malaria parasitemia and thick and thin smears were prepared for microscopy. Malaria-positive cases were treated with artemether/lumefantrine. All blood and breast milk samples were maintained in dim-light conditions and on ice packs for transport to the project laboratory.

Laboratory analyses

Upon receipt, laboratory technicians centrifuged infant and maternal (where applicable) whole blood samples at 1500 × g at room temperature for 15 min to separate plasma or serum for analysis. Plasma aliquots were prepared for measurement of [¹³C₁₀]-retinol and [¹²C]-retinol (infants only) and plasma retinol (mothers and infants). Serum aliquots were prepared for a multiplex analysis of protein biomarkers. For all breast milk samples, we measured milk fat by filling 3 capillary tubes (75 μL) with well-mixed fresh milk and centrifuging samples using a Creamatocrit Centrifuge (Separation Technology), as previously described (21). The mean of 3 measures was used to characterize the milk fat content (g/L). Breast milk was then apportioned into aliquots for analysis. For milk samples collected from the subsample of women in the surveillance cohort, we measured retinol concentration using the iCheck FLUORO portable fluorometer (Bioanalyt GmbH). Biospecimens were stored at −20°C until shipped on dry ice to Newcastle University (infant plasma); the VitMin Lab in Willstaett, Germany (maternal and infant serum); or the University of California, Davis (maternal plasma and breast milk) for analysis.

For maternal samples, we used published HPLC methods to analyze plasma retinol and β-carotene (22) and breast milk retinol and β-carotene (23). Paired baseline and endline samples were analyzed together. In each batch, 3 aliquots of pooled plasma were analyzed to assess precision of the measurements. Retinol and β-carotene concentrations of certified Control Serum (Standard Reference Material 1950; National Institute of Standards and Technology) were analyzed to assess accuracy and were 100% and 98%, respectively, of the certified values. Within-day and between-day CVs for retinol and β-carotene measurements were <5%.

We measured [¹³C₁₀]-retinol and [¹²C]-retinol in grouped baseline and endline infant plasma samples (9 mo + 10 d, 12 mo, and 12 mo + 10 d preferentially; or 9 mo + 4 d, 12 mo, and 12 mo + 4 d) using LC-MS/MS, as previously described (24). The retinol concentration of certified Control Serum (Standard Reference Material 1950; National Institute of Standards and Technology) was analyzed to assess accuracy and precision of the measurements. The measured retinol concentration was 101% of the certified value of the control serum. The within-day and between-day CVs for the retinol measurements were 2.3% and 4.4%, respectively.

Serum concentrations of C-reactive protein (CRP), α₁-acid glycoprotein (AGP), retinol-binding protein, ferritin, and soluble transferrin receptor were measured in duplicate in both maternal and infant samples using a combined sandwich ELISA method (25).

Statistical analysis

We used Stata Version 13.1 (StataCorp LP) for the data analysis. Hypotheses were tested using 2-sided tests. A P value of 0.05 was considered statistically significant. As our outcomes were complementary measures of maternal and infant vitamin A status, no mathematical correction was made for multiple comparisons. We reviewed frequencies of categorical variables, which we present as n (%), and distributions of continuous variables. For the latter, histograms and normal probability plots were used to assess normality. Normally distributed variables are presented as means ± SDs. Skewed variables were natural log-transformed and, where applicable, are presented as geometric means (GMs) and 95% CIs. We evaluated associations between transformed biomarker concentrations by calculating Pearson product-moment correlation coefficients. We tested for comparability between intervention groups at baseline and intervention compliance using an ANOVA for normally distributed variables, the Kruskal-Wallis test for non–normally distributed variables, and the chi-square test for categorical variables. We flagged baseline variables that differed between groups and were associated with outcomes at P < 0.2 as potential confounders to be tested in multivariate models. Intervention compliance was characterized by: 1) the number of meals consumed as a proportion of total meals served (2 meals/d for 90 d); and 2) the total amount of maize consumed. For the latter, we subtracted the median weight of the container (the weight of the infant thermoses varied by number of washings) to estimate leftovers, calculated the dry weight of leftovers (1 g dry = 3.6 g wet for nshima; 1 g dry = 7.08 g wet for porridge), and subtracted the dry weight of leftovers from the serving size (287 g dry weight/d for women; 50 g dry weight for infants). Compliance was compared between observed and unobserved meals by a t-test.

For plasma retinol, we used cutoffs of 0.70 μmol/L or 1.05 μmol/L to indicate deficient (at 12 mo of age) and marginal vitamin A (mothers) statuses, respectively (26). We used methods published by Larson et al. (27) to derive inflammation-adjusted retinol concentrations. As breast milk retinol is found in milk fat and fat content is highly variable, we estimated retinol per gram of milk fat by dividing concentrations, in nmol/L, by the milk fat content of the sample (g/L). We used cutoffs of 1.05 μmol/L or 28 nmol/g fat to indicate low milk retinol (28). The following cutoffs were used for inflammatory and nutrient biomarkers: >5 mg/L for CRP; >1 g/L for AGP; <12 μg/L or <15 μg/L for ferritin in infants and mothers, respectively; and >8.3 mg/L for soluble transferrin receptor. We calculated z-scores for infant anthropometric indices using the WHO Growth Standards (29). BMI in kg/m² was calculated as weight (kg)/height (m²). We used the following equation to calculate body surface area (BSA) (30):

$$\begin{eqnarray*} {\rm{BSA}}\,\,\left({{{\rm{m}}^2}} \right) = {\rm{sqrt}}\left[ {{\rm{weight}}\,\,\left( {{\rm{kg}}} \right) \times {\rm{height}}\,\,\left( {{\rm{cm}}} \right)/3600} \right] \end{eqnarray*}$$

(1)

TBS of vitamin A were estimated for all infants who had paired plasma samples from either the +10 d (n = 142) or +4 d contacts (n = 19) after dosing with [¹³C₁₀]-retinyl acetate. We used a simplified retinol isotope dilution equation based on compartmental modeling of whole-body retinol kinetics in children, as described previously (31, 32):

$$\begin{eqnarray*} {\rm{TBS}}\,\,\left( {\mu {\rm{mol}}} \right) = {\rm{\mathit{ Fa}}} \times {\rm{\mathit{ S}}}\,\, \times \left( {1/{\rm{S}}{{\rm{A}}_{\rm{p}}}} \right) \end{eqnarray*}$$

(2)

Here, Fa is the fraction of dose of [¹³C₁₀]-labeled vitamin A that is absorbed and found in the storage compartment at time t; S is the ratio of the retinol-specific activity in plasma to that in the storage compartment at time t, and SA_p is retinol-specific activity (tracer/total tracee) in plasma, calculated as [(¹³C₁₀)-retinol/(¹³C₁₀)-retinol + (¹²C)-retinol]/dose (µmol) at time t.

Values of 1.58 and 0.578 were used as the values for Fa × Sat 4 d and 10 d after dosing, respectively; these values were derived from a compartmental model of whole-body retinol kinetics for Bangladeshi infants and children 9–17 mo of age (32).

Because [¹³C₁₀]-retinyl acetate was administered at both the 9-mo and 12-mo time points, we needed to account for the contribution of the baseline [¹³C₁₀]-retinol dose remaining in plasma at the time of the endline measurement (i.e., at 12 mo). For the +10-d paired samples, we developed a prediction equation for plasma retinol specific activity at 12 mo +10-d that would be attributable to the baseline [¹³C₁₀]-retinyl acetate dose. First, we plotted the natural log of values of plasma retinol specific activity at 9 mo + 10 d and 12 mo − 1 wk in the negative control group versus time, which yielded a slope of −0.012. Assuming this 1.2%/d decline, we then predicted the plasma retinol–specific activity at 12 mo + 10 d that was attributable to the first [¹³C]-retinyl acetate dose. We subtracted predicted values from measured values. This approach could not be used for the +4-d paired samples, as the final exponential term of a plasma retinol kinetic model in young children does not begin until ≥8–10 d after dosing (15). Instead, we subtracted measured values at 12 mo + 4 d from the measured values at 12 mo + 10 d (i.e., 4 d after receipt of the second [¹³C₁₀]-retinyl acetate dose). After this correction, retinol-specific activity was <0.0010 for 8 measurements, resulting in negative or extreme (>750 μmol) estimates of TBS. These were regarded as implausible and excluded from the analysis. The liver vitamin A concentration was estimated under the assumption that liver represents 80% of TBS and using the equation derived by Yoshizumi et al. (30) for infants and young children:

$$\begin{eqnarray*} {\rm{liver}}\,\,{\rm{weight}}\left( {\rm{g}} \right) = 772\left({{\rm{g}}/{{\rm{m}}^2}} \right) \times {\rm{BSA}}\,\,\left({{{\rm{m}}^2}} \right) - 38 \end{eqnarray*}$$

(3)

Our a priori hypothesis specified that infants in the biofortified and fortified maize groups would have a greater change in TBS over the course of the trial (primary outcome), as a result of higher vitamin A intakes from trial foods and higher vitamin A intakes from breast milk. The latter was dependent on our second a priori hypothesis, that women in the biofortified and fortified maize groups would have higher breast milk retinol concentrations (secondary outcome), controlling for baseline values, than those in the conventional maize group following the 90-d intervention. As our prior research suggested a greater benefit among women with lower milk retinol (8), we also assessed the impact on the prevalence of a low milk retinol concentration (secondary outcome). We present other infant vitamin A status measures (plasma and liver retinol) and maternal vitamin A status measures or indicators (plasma retinol and milk retinol per gram milk fat; low plasma retinol and low milk retinol per gram milk fat) to enable comparisons with other studies in this field. Of these, only maternal plasma retinol concentration was a prespecified secondary outcome. All others would be classified as exploratory. Data on other biomarkers of inflammation or nutritional status in women and infants are similarly provided for the purposes of contextualizing our results and comparing across studies, and are considered exploratory.

All analyses were done on an intention-to-treat basis. Analyses included only cases for which we had measures (breast milk retinol for maternal models and TBS for infant models) at both the baseline and endline visits. We used linear regression modeling for an analysis of continuous outcomes. Models for baseline biomarker concentration included dummy independent variables representing treatment allocation. All models for endline biomarker concentration included treatment allocation and baseline biomarker concentration as independent variables. To account for variability in the breast milk fat content, the milk fat concentration was also included as an independent variable in the model for milk retinol in μmol/L. We employed a robust regression method, including all observations in our analysis, but enabling the model to remove high-leverage outliers (Cook's D > 1) and applying an M-estimator to down-weight observations with large absolute residuals (33, 34). We used a modified Poisson regression with a robust error variance for an analysis of binary outcomes (35), defined by the cutoffs specified above. Baseline prevalence models included only treatment allocation. Endline prevalence models included treatment allocation and baseline biomarker concentration as independent variables. Finally, in all multivariate models, we replaced dummy variables for allocation with a continuous allocation variable to test for a dose response across treatment groups.

Results

We registered a total of 3093 infants and recruited a surveillance cohort of 500 mother/infant pairs (Figure 1). The GM breast milk retinol concentration measured in the first 20% of women enrolled into this cohort was 19.1 nmol/g fat (95% CI: 17.1, 21.3 nmol/g); 77% had low milk retinol concentrations. Of mother/infant pairs in the surveillance cohort, 172 (34%) were deemed ineligible for the trial due to infants turning 9 mo of age prior to the start of trial activities, mothers not planning to continue breastfeeding, or infants missing their 6-mo vitamin A supplement. A further 19 refused participation (4%), 47 moved outside of the study area prior to the infant's 9-mo birthday (9%), and 7 infants died (1%). We recruited and randomly assigned 255 mother/infant pairs for the trial. Of these, 230 (90%) completed the full 90-d intervention, with 18 (7%) and 7 (3%) lost to follow-up due to refusals or moving out of the study area, respectively. Loss to follow-up was not different by treatment group (P = 0.23). We collected and analyzed paired maternal biospecimens from 216 women (92%, 82%, and 80% from the conventional, biofortified, and fortified groups, respectively; P = 0.08). Paired, analyzable samples were available from 161 infants (68%, 60%, and 61% for the conventional, biofortified, and fortified groups, respectively; P = 0.49). There were minor differences between infants for whom we had biochemical data, compared with those missing these measures (Supplemental Table 1), in terms of household bicycle ownership (30.9% vs. 45.3%, respectively; P = 0.02), maternal education [7 y (25th percentile, 9; 75th percentile, 11) vs. 7 y (25th percentile, 9; 75th percentile, 12), respectively; P = 0.05], and infant length (67.0 cm ± 2.9 vs. 67.7 cm ± 2.6, respectively; P = 0.05).

FIGURE 1

Enrollment, participation, and losses to follow-up of women and infants enrolled in a provitamin A carotenoid-biofortified maize or retinyl palmitate–fortified maize intervention trial.

Baseline characteristics of households, mothers, and infants did not differ significantly by allocation in the overall sample (Table 2). More than half of families (61%) in the study owned their own home, although with more limited access to electricity (42%), clean water (55%), and improved sanitation (25%). Roughly 80% of women were married, 48% had at least a primary education, and the majority reported unpaid family labor (64%) or a small business (27%) as their occupation. Nutritional status was characterized by a mean ± SD BMI of 22 ± 4 kg/m² and a 19% anemia prevalence among mothers. Among infants, the prevalence of anemia was 47% at baseline; 33% of infants were stunted and 8% were underweight. While there were no between-group differences at the time of enrollment, we did find the following minor differences when restricting to cases with paired maternal biospecimens: women in the conventional maize arm were less educated [7 y (25th percentile, 4; 75th percentile, 9) vs. 8 y (25th percentile, 6; 75th percentile, 9) in both the biofortified and fortified arms, respectively; P = 0.04] and more likely to be employed outside of the home (46%, vs. 24% in the biofortified and 37% in the fortified arms; P = 0.02). There were also differences in anemia prevalence by group: 17%, 10%, and 27% in the conventional, biofortified, and fortified maize arms, respectively (P = 0.04). Of these variables, only maternal education met our definition as a potential confounder: milk fat (P = 0.10), milk retinol (0.15), and infant TBS (P = 0.17) and liver retinol (P = 0.09) tended to be positively associated with years in school.

TABLE 2

Baseline characteristics¹

	Conventional maize	Biofortified maize	Fortified maize
	n = 85	n = 85	n = 85	P2
Household characteristics
Family asset ownership
Home	46 (54.1)	54 (63.5)	55 (64.7)	0.30
Bicycle	34 (40.0)	36 (42.4)	32 (37.6)	0.82
Mobile telephone	69 (81.2)	71 (83.5)	69 (81.2)	0.90
Radio	41 (48.2)	50 (58.8)	45 (52.9)	0.38
Television	43 (50.6)	42 (49.4)	45 (52.9)	0.90
Access to electricity	37 (43.5)	35 (41.2)	36 (42.4)	0.95
Access to safe drinking water	45 (52.9)	48 (56.5)	48 (56.5)	0.87
Access to improved sanitation	29 (34.1)	16 (18.8)	19 (22.4)	0.06
Maternal characteristics
Married	66 (77.6)	70 (82.4)	70 (82.4)	0.56
Education, y	7 [4, 9]	8 [6, 9]	8 [6, 9]	0.06
Occupation
Unpaid family labor	48 (56.5)	62 (72.9)	54 (63.5)	0.07
Self-employed/small business	24 (28.2)	20 (23.5)	25 (29.4)
Farming/farm labor	3 (3.5)	0 (0.0)	3 (3.5)
Other	10 (11.8)	3 (3.5)	3 (3.5)
Age, y	26 [22, 33]	26 [22, 30]	27 [23, 32]	0.89
Nutritional status
Height, cm	155.3 ± 5.5	155.9 ± 5.5	155.3 ± 8.2	0.81
Weight, kg	54.2 ± 10.3	54.1 ± 10.1	55.7 ± 10.0	0.62
BMI, kg/m2	22.3 ± 3.9	22.4 ± 4.1	22.8 ± 3.7	0.83
Hemoglobin, g/dL	13.1 ± 1.5	13.3 ± 1.4	13.0 ± 1.5	0.48
Anemic3	14 (16.5)	11 (12.9)	22 (26.5)	0.06
Infant characteristics
Male sex	46 (54.1)	38 (44.7)	42 (49.4)	0.47
Nutritional status
Length, cm	67.5 ± 3.0	67.8 ± 2.6	67.2 ± 2.5	0.41
Length-for-age Z-score < −2	26 (32.1)	27 (33.8)	27 (34.2)	0.96
Weight, kg	8.1 ± 1.1	8.1 ± 0.9	8.0 ± 1.1	0.72
Weight-for-age Z-score < −2	6 (7.4)	5 (6.3)	7 (8.9)	0.82
Hemoglobin, g/dL	10.0 ± 1.2	10.0 ± 1.3	9.8 ± 1.1	0.33
Anemic3	35 (41.7)	38 (45.2)	45 (54.2)	0.25

1

Data are from households and mother/infant pairs enrolled in a 90-d feeding intervention providing 0 μg RE/d as conventional maize or ∼315 μg RE/d to mothers and ∼55 µg RE/d to infants as either provitamin A carotenoid-biofortified maize or retinyl palmitate–fortified maize, by intervention group. Values are n (%), median [25th percentile, 75th percentile], or mean ± SD. RE, retinol equivalents.

2

P values are based on a Χ² test, ANOVA, or Kruskal-Wallis test. P < 0.05 was considered statistically significant.

3

Defined as hemoglobin <12 g/dL for women and <10 g/dL for infants.

The carotenoid and vitamin A profiles of biofortified and fortified maize, both as maize flour and prepared meals, are shown in Supplemental Table 2. The provitamin A content of the biofortified maize meal was lower than anticipated (12.3 μg/g vs. 17.5 μg/g, respectively), yielding an estimated 1.02 μg RE/g, compared to 1.11 μg RE/g in the fortified maize. Losses during cooking were also greater in the biofortified maize: the vitamin A contents of biofortified porridge and nshima were both 0.47 μg RE/g, while those of fortified porridge and nshima were 0.69 μg RE/g and 0.86 μg RE/g, respectively. This would equate to ∼135 μg RE/d for women and ∼25 μg RE/d for infants in the biofortified maize arms, or ∼200 μg RE/d for women and ∼40 μg RE/d for infants in the fortified maize arm. Overall, women consumed a median of 264 g/d (25^th percentile, 236; 75^th percentile, 277), or 92% of the intended 287 g/d, and infants consumed 41 g/d (25th percentile, 33; 75th percentile, 46), or 82% of the intended 50 g/d. Compliance did not differ significantly by group (Table 3). Observation of meals did not appear to influence the amount consumed by women or infants, although the large difference in variances between observed and unobserved measures precluded the use of nonparametric analyses to test these differences.

TABLE 3

Intervention adherence¹

	Conventional maize	Biofortified maize	Fortified maize
	n = 78	n = 70	n = 68	P2
Attendance, %	93.1 [10.2, 96.9]	90.9 [13.1, 94.2]	91.2 [12.7, 95.3]	0.08
Mean daily intake,3 g dry weight/d
Mothers	252.4 [35.1, 266.3]	246.9 [42.5, 255.6]	254.1 [39.9, 265.3]	0.23
Infants	37.6 [10.7, 40.4]	36.2 [12.2, 39.8]	39.8 [9.9, 43.1]	0.13
Mean daily intake, %
Mothers	88.0 [12.2, 92.8]	86.0 [14.8, 89.1]	88.5 [13.9, 92.4]	0.23
Infants	75.2 [21.4, 80.7]	72.4 [24.5, 79.7]	79.6 [19.8, 86.1]	0.13

1

Data are from mother/infant pairs enrolled in a 90-day feeding intervention providing 0 μg RE/d as conventional maize or ∼315 μg RE/d to mothers and ∼55 µg RE/d to infants as either provitamin A carotenoid-biofortified maize or retinyl palmitate–fortified maize, by intervention group. Includes only subjects for whom paired baseline and endline breast milk samples were available. Values are median [25th percentile, 75th percentile]. RE, retinol equivalents.

2

P values are based on the Kruskal-Wallis test. P < 0.05 was considered statistically significant.

3

Dry weight was calculated based on inputs: 3.6 g prepared nshima = 1 g dry weight; 7.1 g prepared porridge = 1 g dry weight.

Baseline measures of vitamin A status in women did not differ by group (Table 4). The overall GM plasma retinol concentration was 1.21 μmol/L (95% CI: 1.16, 1.27 μmol/L) and 29% of women could be classified as having a marginal status (<1.05 μmol/L) based on this biomarker. As there was evidence of degradation in carotenoid concentrations (Supplemental Figure 1), we do not report those results here. The GM breast milk retinol concentration was 1.57 μmol/L (95% CI: 1.45, 1.69 μmol/L) or, expressed as per gram milk fat, 47.3 nmol/g fat (95% CI: 44.1, 50.6 nmol/g). As shown in Table 5, approximately one-quarter (24%) of women had low milk retinol concentrations (<1.05 μmol/L). Milk retinol was strongly associated with both the plasma retinol concentration and breast milk fat concentration (Supplemental Table 3), for which the GM was 33.1 g/L (95% CI: 31.3, 35.0 g/L). Neither the maternal plasma nor breast milk retinol concentration was significantly associated with the inflammatory protein concentration (Supplemental Table 3). After the 90-d intervention, we found no differences in mean plasma retinol in either the biofortified or fortified maize groups (Table 4). Breast milk retinol concentrations tended to increase in both the biofortified and fortified groups, although the mean concentration was significantly greater at endline only in the fortified maize group (P = 0.01). There was a dose-response relation, with milk retinol increasing in a stepwise manner in response to the biofortified and fortified maize (P < 0.05). Both the biofortified and fortified maize interventions reduced the prevalence of a low milk retinol concentration, by 58% (P = 0.02) and 54% (P = 0.01), respectively. Neither intervention impacted concentrations of other biomarkers of nutritional status or inflammation measured in maternal plasma (Supplemental Tables 4 and 5).

TABLE 4

Comparisons of maternal plasma and breast milk retinol concentrations at baseline and endline visits¹

		Baseline2	Endline	Change
	n	GM (95% CI)	GM (95% CI)	Mean ± SD	β (95% CI)3	P	P-trend4
Plasma retinol, μmol/L
Conventional	78	1.20 (1.13, 1.29)	1.17 (1.09, 1.26)	−0.03 ± 0.28	Reference	—	0.85
Biofortified	69	1.20 (1.10, 1.31)	1.21 (1.12, 1.30)	0.00 ± 0.34	0.02 (−0.05, 0.09)	0.64
Fortified	68	1.25 (1.15, 1.34)	1.19 (1.10, 1.30)	−0.04 ± 0.31	−0.01 (−0.08, 0.06)	0.84
Milk retinol, μmol/L
Conventional	78	1.67 (1.46, 1.91)	1.65 (1.45, 1.87)	−0.05 ± 1.36	Reference	—	0.01
Biofortified	70	1.45 (1.27, 1.66)	1.67 (1.49, 1.87)	0.18 ± 1.31	0.11 (−0.02, 0.24)	0.09
Fortified	68	1.57 (1.37, 1.81)	1.85 (1.63, 2.11)	0.29 ± 1.50	0.17 (0.04, 0.30)	0.01

1

Data are from lactating women enrolled in a 90-d mother/infant feeding intervention providing 0 μg RE/d as conventional maize or ∼315 μg RE/d to mothers and ∼55 µg RE/d to infants as either provitamin A carotenoid-biofortified maize or retinyl palmitate–fortified maize. Analysis includes only subjects with paired baseline and endline milk retinol values. Biomarker concentrations were natural log-transformed to approximate a normal distribution; changes in biomarker concentrations from baseline to endline were normally distributed. P < 0.05 was considered statistically significant. GM, geometric mean; RE, retinol equivalents.

2

No significant differences in analyte concentrations at baseline visit, based on simple linear regression.

3

Analyte concentrations measured at endline visit were compared using robust linear regression, including baseline concentrations and maternal education as covariates. The model for milk retinol also included endline milk fat concentration.

4

Dummy variables for allocation were replaced with a continuous allocation variable in adjusted models to test for a dose-response relation; P-trend is based on the Wald F test for this linear term.

TABLE 5

Prevalence and prevalence ratios for low maternal plasma or breast milk retinol concentrations at baseline and endline visits¹

		Baseline2		Endline
	n	n	%	n	%	RR (95% CI)	P3	P-trend4
Plasma retinol <1.05 μmol/L
Conventional	78	22	28.2	22	28.2	Reference	—	0.67
Biofortified	69	23	33.3	23	33.3	1.06 (0.66, 1.69)	0.82
Fortified	68	17	25.0	21	30.9	1.11 (0.69, 1.77)	0.67
Milk retinol <1.05 μmol/L
Conventional	78	15	19.2	17	21.8	Reference	—	0.01
Biofortified	70	19	27.1	10	14.3	0.42 (0.21, 0.85)	0.02
Fortified	68	17	25.0	9	13.2	0.46 (0.25, 0.82)	0.01

1

Data are from lactating women enrolled in a 90-d mother/infant feeding intervention providing 0 μg RE/d as conventional maize or ∼315 μg RE/d to mothers and ∼55 µg RE/d to infants as either provitamin A carotenoid-biofortified maize or retinyl palmitate–fortified maize. The analysis includes only subjects with paired baseline and endline milk retinol values. P < 0.05 was considered statistically significant. RE, retinol equivalents.

2

No significant differences in prevalence estimates at the baseline visit, based on modified Poisson regression.

3

Prevalence estimates for low plasma or breast milk retinol at the endline visit were compared using modified Poisson regression, including baseline concentrations and maternal education as covariates. The model for milk retinol also included endline milk fat concentration.

4

Dummy variables for allocation were replaced with a continuous allocation variable in adjusted models to test for a dose-response relation; the P-trend is based on the Wald F test for this linear term.

At baseline, the overall GMs of infant vitamin A biomarkers were 0.82 μmol/L (95% CI: 0.78, 0.86 μmol/L) for plasma retinol, 178 μmol (95% CI: 166, 191 μmol) for TBS, and 0.54 μmol/L (95% CI: 0.50, 0.58 μmol/L) for liver retinol concentration. Infant plasma retinol was not associated with either the TBS or liver retinol concentration (Supplemental Table 6). However, there were strong associations between all maternal vitamin A status biomarkers and both the infant TBS (P < 0.01) and liver retinol concentration (P < 0.001). We found no significant effect of either biofortified or fortified maize meal consumption on infant plasma retinol (Table 6). While infant plasma retinol was strongly associated with inflammatory protein concentrations (P < 0.001), results of the inflammation-adjusted model did not differ from the unadjusted model (data not shown). While, overall, TBS increased by a mean ± SD of 24 ± 136 μmol over the 90-d intervention period, there were no differences in the mean TBS or liver retinol concentration in either the biofortified or fortified maize groups at endline (Table 6). We found no effect of the feeding intervention on any secondary infant analytes reflective of inflammation or nutritional status (Supplemental Tables 7 and 8).

TABLE 6

Comparisons of infant vitamin A status measures at baseline and endline visits¹

		Baseline2	Endline	Change
	n	GM (95% CI)	GM (95% CI)	Mean ± SD	β (95% CI)	P3	P-trend4
Plasma retinol, μmol/L
Conventional	58	0.80 (0.73, 0.87)	0.80 (0.73, 0.87)	0.00 ± 0.25	Reference	—	0.49
Biofortified	51	0.86 (0.80, 0.93)	0.83 (0.77, 0.90)	−0.03 ± 0.21	−0.01 (−0.11, 0.10)	0.92
Fortified	52	0.81 (0.74, 0.89)	0.84 (0.78, 0.91)	0.02 ± 0.28	0.04 (−0.07, 0.14)	0.47
Total body stores, μmol
Conventional	55	169 (147, 194)	189 (164, 219)	26 ± 156	Reference	—	0.57
Biofortified	49	177 (160, 195)	196 (177, 218)	21 ± 97	0.04 (−0.14, 0.22)	0.65
Fortified	49	190 (170, 213)	205 (178, 236)	27 ± 147	0.05 (−0.13, 0.24)	0.57
Liver retinol concentration, μmol/g liver
Conventional	52	0.51 (0.44, 0.59)	0.53 (0.45, 0.62)	0.03 ± 0.47	Reference	—	0.45
Biofortified	49	0.54 (0.49, 0.59)	0.55 (0.49, 0.61)	0.02 ± 0.28	0.04 (−0.15, 0.22)	0.69
Fortified	47	0.59 (0.52, 0.67)	0.60 (0.52, 0.69)	0.02 ± 0.45	0.07 (−0.12, 0.26)	0.45

1

Data are from breastfeeding infants enrolled in a 90-d mother/infant intervention providing 0 μg RE/d as conventional maize or ∼315 μg RE/d to mothers and ∼55 µg RE/d to infants as either provitamin A carotenoid-biofortified maize or retinyl palmitate–fortified maize. Analysis includes only subjects with paired baseline and endline plasma retinol values. Biomarker concentrations were natural log-transformed to approximate a normal distribution; changes in biomarker concentrations from baseline to endline were normally distributed. P < 0.05 was considered statistically significant. GM, geometric mean; RE, retinol equivalents.

2

No significant differences in analyte concentrations at the baseline visit, based on simple linear regression.

3

Analyte concentrations measured at the endline visit were compared using robust linear regression, including baseline concentrations and maternal education as covariates.

4

Dummy variables for allocation were replaced with a continuous allocation variable in adjusted models to test for a dose-response relation; the P-trend is based on the Wald F test for this linear term.

Discussion

We demonstrated that biofortified or fortified maize consumption can improve the vitamin A content of breast milk, as indicated by greater mean milk retinol in the fortified maize arm and a reduction in the prevalence of low milk retinol seen with both interventions. This did not translate into improved infant TBS. An increase in milk retinol of ∼10–20 µmol/L, with wide interindividual variation in responses, is consistent with our prior short-term biofortified maize intervention (8), as well as with research on orange-fleshed sweet potatoes (36). The milk retinol concentration was notably, and unexpectedly, higher in our present study [1.59 µmol/L vs. 0.95 µmol/L in our prior short-term biofortified maize intervention (8) and 0.63 µmol/L in research on orange-fleshed sweet potatoes (36)]. This may be attributable to strong seasonal fluctuations in the intake of vitamin A–rich foods (37); trial data collection extended across all seasons, compared to early, lean-season assessments in our prior study (8) and our pretrial screening. Further research is warranted on vitamin A intakes and statuses in this setting to inform safe and effective targets to address dietary gaps under varying scenarios.

Our study further demonstrates the efficacy of an industrially fortified maize product, adding to a recent review of the literature (4). The increase in milk retinol is similar to that seen with low-dose daily vitamin A supplementation (36) or a high-dose capsule (38–42). Given the homeostatic control of plasma retinol and its marginal baseline status, that the biofortified maize did not alter plasma retinol is unsurprising. While there is prior evidence of a protective benefit of preformed vitamin A–fortified wheat products on serum retinol (43, 44), the baseline status was poorer in prior trials and the impact was generally limited to subjects with lower status (43).

Daily consumption of biofortified maize did not increase estimated infant TBS or liver vitamin A concentrations, nor did it reduce the prevalence of low plasma retinol in infants. We attribute this to 2 factors. First, we provided a high-dose vitamin A supplement (105 µmol) to infants at 6 mo of age, which was 3 mo prior to the intervention. High-dose capsules are thought to meet infant vitamin A needs for ∼3 mo in populations at risk of deficiency (12). Studies of preschool-aged children in Peru and Zambia also suggested that younger children utilize vitamin A more rapidly than adults, at a rate of ∼2.0%/d (15, 45). Thus, we expected that the capsule given at 6 mo of age would have had little effect on infant TBS 3 mo later. Second, vitamin A intakes from breast milk and complementary foods may have been sufficient to meet the 190 µg RE/d estimated average requirement for infants 7–12 mo of age (46). Based on the observed mean milk vitamin A concentration at 9 mo of age and an assumed breast milk intake of 730 mL/d (47), infants in our study were likely consuming ∼345 µg RE/d from breast milk between 6 and 9 mo of age. Although we do not have quantitative dietary intake data, our recall data did indicate that more than half of infants were regularly receiving foods such as dark green leafy vegetables and eggs.

Taken together, if infant vitamin A intakes from 6 to 9 mo of age are adequate, vitamin A provided by the high-dose supplement would be retained to a greater extent and utilized to a lesser extent, as newly absorbed dietary vitamin A is utilized preferentially (48). This is supported by our baseline TBS data. Assuming ∼345 µg RE/d from breast milk and a vitamin A utilization rate of ∼0.5%/d [based on the estimated fractional catabolic rate for infants 9–17 mo of age (32)], we would expect infant TBS to be ∼173 µmol at 9 mo of age. Our overall GM TBS at baseline was 177 µmol, similar to the GMs of Bangladeshi infants specifically screened for vitamin A adequacy (SM Ahmad, MNA Afsa, MJ Alam, A Oxley, MJ Haskell, JL Ford, MH Green, G Lietz, unpublished results, 2019). Because the intervention provided small amounts of vitamin A daily, the expected increase in infant TBS would be relatively small in relation to the estimated TBS at 9 mo of age, requiring a much larger sample size to detect any difference in the change in TBS between groups. Also, the variability in the response of TBS to the interventions was higher than expected. It should be noted that this wide interindividual variation in responses was similar across biomarkers. The lack of effect on infant plasma retinol concentrations is consistent with results on estimated infant TBS and liver vitamin A concentrations.

The primary strength of our study was the use of multiple biomarkers. We would argue that this approach is crucial in vitamin A intervention studies. There is considerable debate regarding the use and interpretation of vitamin A status measures in an inflamed population (49, 50). Isotope dilution is thought to provide the most robust measure in this context. However, it would likely depend on when inflammation occurred relative to administration of labeled vitamin A, and to when plasma retinol–specific activity is measured. Absorption of labeled vitamin A was significantly reduced in 1 preschool-aged Zambian child with a recent fever (45). In rats, inflammation reduced both mobilization of retinol from the liver and plasma retinol–specific activity (51). Thus, acute inflammation could potentially lower plasma retinol–specific activity and lead to an overestimation of TBS. The magnitude of effect is likely related to the nature of the stressor, as well as to its timing (52). Chronic, low-grade inflammation may have less of an impact on estimates of total body vitamin A stores. If low-grade inflammation occurs after the dose of labeled vitamin A has mixed with vitamin A stores, the plasma retinol concentration may decrease, but the plasma retinol–specific activity may not be affected. Retinol isotope dilution is unlikely to be informative for research in lactating women at present, as researchers are unable to account for the loss of labeled isotopes in breast milk. A prior epidemiological study suggests that inflammation also has less influence on the breast milk retinol concentration compared with the plasma retinol concentration (53). Our data support this. Our study further benefited from having a positive control arm. This supported the validity of our trial design, but also provided evidence regarding an industry-driven strategy to sustainably improve the food supply.

Our trial did have a number of limitations. Despite prior evidence of deficiency in the study area (8) and pretrial screening, the baseline status of women was better than anticipated and infants had adequate vitamin A reserves. This may have limited our ability to detect an intervention impact and to draw conclusions about efficacy for populations with a higher vitamin A deficiency prevalence and less variability in baseline vitamin A status. For infants, a considerably larger sample size would have been required to detect the effect of a relatively low dose of vitamin A on status. In addition, our assumptions regarding provitamin A carotenoid content (17.5 µg/g) and retention (75%) proved to be overestimates. Considerable attrition was a further challenge, although we found no evidence of a differential loss to follow-up. Maternal carotenoid concentrations were only ∼30% of what we previously reported (5, 6). Unlike retinol, for which stability at −25°C has been demonstrated for ≤3 y (54), we concluded that carotenoid measures were unreliable due to long-term storage (55). Finally, our approach for analyzing multiple, complementary biomarkers has implications for type I errors. Significant findings for our secondary outcomes should therefore be interpreted with caution.

Overall, our study suggests that biofortification and industrial fortification strategies may help to improve the statuses of lactating mothers, reducing the risk of low breast milk retinol concentrations. Breast milk is the most critical source of dietary vitamin A for breastfeeding infants. While increased milk retinol did not influence infant TBS in our study, this was likely due to the high-dose vitamin A capsule provided at 6 mo of age. As countries reconsider the continued need for and sustainability of capsule delivery, we would suggest that food fortification strategies—whether via carotenoid breeding programs or industry-driven efforts—be carefully considered to ensure the continued protection of breastfeeding infants. Monitoring of the vitamin A intake and/or status over time, as well as adherence of programs to targeted breeding or fortification levels, is also recommended to track the safety and effectiveness of these interventions (2, 56).

ACKNOWLEDGEMENTS

We thank Luka Mwamba and Christopher Chibuye from Mkushi District for their support. We thank Arthur Musonda and Penyani Zimba at our Mkushi field office for managing the feeding intervention and data collection, as well as Lee Wu at the Johns Hopkins University for statistical support. We further thank V.C. Kalaivanan from Dariyaye Millers, Grace Munthali from the Zambian National Institute for Scientific and Industrial Research, and Klaus Kraemer from Sight and Life for their assistance on fortification, as well as Neal Craft and Jeurgen Erhardt for analyses of food samples and nutrient biomarkers, respectively. We also thank Erick Boy, Fabiana Moura, and Keith West for their input on the study design.

The authors’ responsibilities were as follows—ACP, MJH: designed the research, wrote the manuscript, and had primary responsibility for the final content; and all authors: conducted the research and read and approved the final manuscript. The authors report no conflicts of interest.

Data Availability

Data described in the manuscript, code book, and analytic code will be made available upon request pending application and approval.

Notes

Financial support for this study was provided by HarvestPlus (www.HarvestPlus.org). In-kind support was provided by DSM Nutritional Products, Inc. (fortificant) and Bioanalyt GmbH (iCheck Fluoro and consumables). Additional funding was provided by the Sight and Life Global Nutrition Research Institute at Johns Hopkins University, with support from the Christian Blind Mission.

The views expressed do not necessarily reflect those of HarvestPlus.

Supplemental Tables 1–8 and Supplemental Figure 1 are available from the “Supplementary data” link in the online posting of the article and from the same link in the online table of contents at https://academic.oup.com/ajcn/.

Abbreviations used: AGP, α₁‐acid glycoprotein; BSA, body surface area; CRP, C‐reactive protein; FM, fortified maize; GM, geometric mean; OM, orange maize; RE, retinol equivalents; TBS, total body stores; WM, white maize.