Comparison of Total Body Vitamin A Stores Using Individual versus Population ¹³C-Natural Abundance of Serum Retinol in Preschool Children and Women Residing in 6 Diverse African Countries

Abstract

Background

Stable isotope techniques using ¹³C to assess vitamin A (VA) dietary sources, absorption, and total body VA stores (TBSs) require determination of baseline ¹³C abundance. ¹³C-natural abundance is approximately 1.1% total carbon, but varies with foods consumed, supplements taken, and food fortification with synthetic retinyl palmitate.

Objectives

We determined ¹³C variation from purified serum retinol and the resulting impact on TBSs using pooled data from preschool children in Burkina Faso, Cameroon, Ethiopia, South Africa, Tanzania, and Zambia and Zambian women.

Methods

Seven studies included children (n = 639; 56 ± 25 mo; 48% female) and one in women (n = 138; 29 ± 8.5 y). Serum retinol ¹³C-natural abundance was determined using GC-C-IRMS. TBSs were available in 7 studies that employed retinol isotope dilution (RID). Serum CRP and α₁-acid-glycoprotein (AGP) were available from 6 studies in children. Multivariate mixed models assessed the impact of covariates on retinol ¹³C. Spearman correlations and Bland–Altman analysis compared serum and milk retinol ¹³C and evaluated the impact of using study- or global-retinol ¹³C estimates on calculated TBSs.

Results

¹³C-natural abundance (%, median [Q1, Q3]) differed among countries (low: Zambia, 1.0744 [1.0736, 1.0753]; high: South Africa, 1.0773 [1.0769, 1.0779]) and was associated with TBSs, CRP, and AGP in children and with TBSs in women. ¹³C-enrichment from serum and milk retinol were correlated (r = 0.52; P = 0.0001). RID in children and women using study and global estimates had low mean bias (range, −3.7% to 2.2%), but larger 95% limits of agreement (range, −23% to 37%).

Conclusions

¹³C-natural abundance is different among human cohorts in Africa. Collecting this information in subgroups is recommended for surveys using RID. When TBSs are needed on individuals in clinical applications, baseline ¹³C measures are important and should be measured in all enrolled subjects.

Introduction

Objective biomarkers of dietary intake and accurate nutritional status assessment are 2 goals of the US NIH strategic plan for realizing the vision of precision nutrition [1,2]. Critically, these biomarkers need to be accurate and accessible at the individual level rather than accurate only at the population level. Stable isotope-based assessment methods can be used to estimate relative dietary contributions, particularly across trophic levels [3,4]. Stable isotope tracers can also be used to assess nutrient absorption [5], metabolism, and total body content using the principle of isotope dilution [6].

Ratios of stable isotopes of nitrogen (¹⁵N) and carbon (¹³C) can be used to assess differences in intake of meat, fish, and plant-based foods [[7], [8], [9], [10], [11]]. ¹³C makes up approximately 1.1% of all natural carbon, but in humans, it is highly dependent on the foods and supplements consumed. ¹³C is fractionated (separated) from ¹²C during specific biological processes, resulting in reduced or increased ¹³C enrichment that is specific to certain organisms or metabolic processes [12]. For example, there is a large difference between C3 and C4 plants, whereby C4 plants (e.g., maize, rice, sugarcane, and sorghum) fractionate less ¹³C, resulting in a higher ¹³C-natural enrichment within the plant than C3 plants (most vegetables consumed by humans). This difference can be exploited to predict added sugar or sugar-sweetened beverage intakes and evaluate interventions to change intakes of these foods [13,14]. Furthermore, nutrient-specific intakes can be traced, such as the nonprovitamin A carotenoid lutein [15] and biofortified provitamin A maize as a contribution to total vitamin A (VA) intakes using retinol ¹³C enrichments [16, 17].

Provitamin A carotenoids synthesized by C4 plants will also have a higher ¹³C enrichment than C3 plants, which is measurable in animals and humans consuming them as an increase in circulating retinol ¹³C-natural abundance. Based on the amount of shift in baseline ¹³C enrichment, the relative contribution of C4 provitamin A sources (e.g., biofortified maize) to the total VA intake can be determined [16,17]. This could be used to assess coverage of provitamin A interventions or to determine the contribution of biofortified maize to VA stores relative to other sources in effectiveness studies. Applications to indicate metabolite dietary origins require enough sensitivity, accuracy, and precision for ¹³C/¹²C determination. Isotope ratio MS has the advantage of high sensitivity and is commonly used when working with enrichments close to natural abundance.

In the application of ¹³C-retinol to assess VA status using retinol isotope dilution (RID), a baseline blood sample is required to assess the natural abundance of ¹³C in serum retinol. Some studies have eliminated the need for this baseline blood sample by obtaining a group level estimate from separate participants [18,19]. To attempt to optimize study designs and reduce the number of required blood samples, we performed a pooled analysis of 8 studies with VA ¹³C-natural abundance data from 6 African countries to provide estimates of within- and between-country variation.

Our primary objectives were to: 1) describe reference ranges and covariates of retinol ¹³C-natural abundance so they could be used to assess changes in dietary contributions from foods and supplements over time, and 2) compare RID-derived VA status using individual or population retinol ¹³C-natural abundance to inform RID study designs and population evaluations seeking to minimize blood sampling requirements. We hypothesized that ¹³C-natural abundance would differ within and among countries because of the diversity in plant and animal foods consumed, supplementation usage, and country-level fortification mandates.

Methods

Data sources

We conducted an analysis of available baseline data from 8 studies (total n = 781). Four outliers were excluded based on a retinol ¹³C-natural abundance >3 SDs from the mean: 3 from women in Zambia and 1 from a child in Ethiopia. Five studies among preschool children (n = 467) were part of a regional Technical Cooperation project supported predominantly by the International Atomic Energy Agency. Studies were conducted in Burkina Faso, Cameroon, Ethiopia, South Africa [20], and Tanzania [21]. Two studies in Zambia were part of biofortification feeding trials in preschool children (n = 172) [22,23], and one additional study assessed VA status in Zambian lactating and nonlactating women (n = 138) [24]. Consistent exclusion criteria for the children were active fever >38 °C, severe anemia cutoff of <70 g/L (except in South Africa, which used a moderate anemia cutoff of <90 g/L), or severe wasting (weight-for-height z-score <−3). Pregnant women were excluded in Zambia.

Ethics

Each country was required to seek independent ethical approval and encouraged to register their trials. Individual study ethics committee approvals were reported in their respective publications [[21], [22], [23], [24]] and one combined analysis reporting on the remaining studies [25] (Supplementary Table 1). Written informed consent was obtained from the parents or guardians of all children enrolled or the women themselves.

Laboratory and biomarker analysis

HemoCue (Hb-301) was used to measure hemoglobin concentration using venous blood. CRP, α₁-acid-glycoprotein (AGP), and RBP levels were analyzed using ELISA by either the VitMin Lab [26] or at the University of Wisconsin-Madison using commercially available kits (CRP: Cayman Chemical Company; AGP: Abcam; and RBP: Arbor Assays). Serum retinol was determined using HPLC and purified commercial standards (Sigma Aldrich). Serum retinol and milk VA ¹³C enrichment analysis was performed in the laboratory of SAT using GC-C-IRMS as published previously [23,25]. Total body VA stores (TBSs) and total liver VA reserves (TLRs) were determined using RID and the tracer-to-tracee ratio [23,27]. Assumptions used for calculating TBSs and TLRs are reported in their respective articles and include dose absorption, ratio between serum and liver enrichment, dose catabolism, liver to body weight ratio, and fraction of TBSs stored in liver.

Statistical methods

Data are reported as medians (Q1 and Q3) or as means ± SDs. Data were analyzed using SAS version 9.4 (SAS Institute). Nonnormal covariates (CRP and AGP) were natural log-transformed for analysis. Spearman correlations were determined among outcomes. Differences in ¹³C-natural enrichment among studies were determined with study as a fixed effect. Lactating and nonlactating women from the same study were treated as 2 separate groups. Multivariate models included age, sex, and other covariates as fixed effects and study as a random effect. Post hoc letter groupings among studies were determined using the macro PDMIXED with a Tukey adjustment [28]. TBSs and TLRs were calculated using the RID equation [29] with study-specific assumptions as reported by the authors. To evaluate the impact of using population estimates on ¹³C TBS calculations, each participant’s individual, within-study, and within-life stage (children and women) median retinol ¹³C-natural abundance was used to calculate separate TBS estimates, referred to as TBS_individual, TBS_study, and TBS_global, respectively. Resulting TLRs for 50% and 80% of TBSs in liver were calculated with the same RID assumptions (i.e., TLR_50_sudy, TLR_50_global, TLR_80_study, and TLR_80_global). Bland–Altman analysis was used for method comparisons: 1) serum compared with milk ¹³C-enrichment, and 2) TBS_individual and TLR_individual compared with their respective study and global estimates. Statistical significance was considered at P value of ≤0.05.

Results

Study inclusion and participant data

The 8 included studies contained 777 participants with data on serum retinol ¹³C-natural abundance. Of the 7 studies in children, 47.6% of participants were female and median ages ranged from 45.2 to 70.0 months (Table 1). Median VA TBSs ranged from 105 μmol in Ethiopia to 700 μmol in Zambia. From the study of women of reproductive age (range, 15–48 y), 54.4% were lactating at the time of sampling.

TABLE 1.

Baseline characteristics of children and women (n = 775) enrolled in studies in 6 countries¹

	Burkina Faso	Cameroon	Ethiopia	South Africa	Tanzania	Zambia 2010	Zambia 2012	Zambia 2019
								Nonlactating	Lactating
n	104	45	130	93	95	39	133	63	75
Sex, female (%)	46.2	53.3	44.6	54.8	47.4	46.2	45.1	100	100
Age (mos)	49.1 (42.8, 52.9)	60.0 (48.0, 60.0)	48.0 (41.0, 54.0)	45.2 (39.2, 52.4)	47.8 (41.2, 53.3)	54.5 (46.0, 60.0)	70.0 (66.0, 77.0)	360 (276, 456)	336 (270, 396)
Weight (kg)	14.7 (13.6, 16.1)	17.0 (15.0, 19.0)	14.0 (12.7, 15.4)	13.4 (12.2, 14.8)	14.4 (13.4, 15.7)	14.3 (13.3, 16.2)	16.9 (15.8, 18.7)	57.7 (51.1, 67.0)	56.7 (52.0, 64.3)
Height (cm)	98.1 (93.8, 102)	107 (101, 111)	96.5 (92, 103)	94.9 (92.0, 98.2)	96.3 (91.0, 100)	98.8 (91.0, 103)	107 (104, 112)	157 (153, 161)	157 (153, 164)
Retinol (μmol/L)2	0.833 (0.668, 0.959)	0.787 (0.679, 0.932)	0.778 (0.635, 0.941)	1.07 (0.890, 1.29)	1.04 (0.845, 1.22)	—	0.959 (0.762, 1.17)	1.64 (1.33, 2.09)	1.55 (1.31, 1.94)
RBP (μmol/L)2	0.818 (0.675, 1.09)	0.875 (0.665, 0.990)	0.860 (0.685, 1.03)	0.935 (0.745, 1.24)	0.805 (0.667, 1.01)	—	0.93 (0.697, 1.35)	—	—
CRP (mg/L)2	0.120 (0.010, 1.012)	3.07 (0.535, 16.2)	0.230 (0.010, 0.930)	0.140 (0.010, 0.890)	0.235 (0.010, 1.250)	1.37 (0.773, 2.43)	1.36 (0.433, 7.02)	—	—
AGP (g/L)2	0.705 (0.538, 1.12)	1.12 (0.675, 1.48)	0.705 (0.550, 1.19)	0.630 (0.445, 0.915)	0.645 (0.460, 0.915)	1.12 (0.861, 1.64)	2.39 (1.956, 2.88)	—	—
TBS (μmol)	431 (290, 613)	644 (519, 922)	105 (60, 154)	561 (418, 678)	217 (147, 307)	—	700 (536, 852)	451 (319, 609)	285 (209, 459)
TLR (50%)3	0.489 (0.339, 0.742)	0.632 (0.538, 0.853)	0.121 (0.072, 0.176)	0.684 (0.548, 0.863)	0.251 (0.182, 0.350)	—	0.652 (0.528, 0.845)	0.165 (0.111, 0.213)	0.102 (0.076, 0.160)
TLR (80%)4	0.783 (0.542, 1.187)	1.01 (0.861, 1.37)	0.193 (0.115, 0.282)	1.10 (0.877, 1.381)	0.401 (0.291, 0.560)	—	1.04 (0.845, 1.35)	0.264 (0.178, 0.341)	0.163 (0.121, 0.255)

AGP, α-1-acid-glycoprotein; TLR, total liver vitamin A reserve; TBS, total body vitamin A store; —, not measured or data not available.

¹Data are % or median (Q1, Q3).

²Analytes measured in serum.

³Total liver reserves assuming 50% of total body vitamin A in liver.

⁴Total liver reserves assuming 80% of total body vitamin A in liver.

Serum retinol ¹³C-natural abundance among countries

The serum and breastmilk VA ¹³C-natural isotopic abundance among studies is reported in Figure 1. The overall significance of differences in natural abundance by study was P < 0.0001. Although most studies did not differ from each other, Burkina Faso and South Africa had significantly higher enrichment than both studies of children in Zambia. The Zambian 2012 study had significantly lower enrichments than all other studies.

FIGURE 1.

Serum vitamin A ¹³C natural isotopic abundance in humans by study and lactation status from 6 African countries. ¹³C isotopic abundance is the fraction of ¹³C out of total carbon (%). Data are mean ± SD. Serum retinol and milk vitamin A ¹³C isotopic abundance means without a common letter differ, P < 0.05.

Correlates of VA ¹³C enrichment

Spearman correlations of serum and milk ¹³C enrichment with age and biomarkers for VA status (i.e., serum retinol, RBP, and TBSs) and inflammation (i.e., CRP and AGP) are shown in Table 2. Data from serum enrichment were analyzed separately between children and women. For children, age, TBSs, CRP, and AGP had significant inverse associations with retinol ¹³C enrichment. In a mixed model with age and sex as fixed effects and study as a random effect, neither age nor sex were significant.

TABLE 2.

Spearman correlations among serum and milk vitamin A ¹³C-natural abundance with age and serum nutritional and inflammatory biomarkers in children and women from 6 African countries

	Age	Retinol1	RBP1	CRP1	AGP1	TBSs
Serum retinol 13C-natural enrichment
Children	−0.35∗∗∗	0.049	0.031	−0.16∗∗∗	−0.34∗∗∗	−0.096∗
n	632	574	534	615	618	570
Women	−0.025	0.086	—	—	—	0.51∗∗∗
n	131	135	—	—	—	136
Milk vitamin A 13C-natural enrichment
Women	0.0096	0.21	—	—	—	0.28†
n	47	46	—	—	—	47

Significant correlations are indicated by: ∗ P < 0.05, ∗∗∗ P < 0.001. † P = 0.06

AGP, α-1-acid-glycoprotein; TBS, total body vitamin A store.

¹Analytes measured in serum.

In women, correlates with serum retinol ¹³C enrichment had similar significance patterns between lactating and nonlactating women (data not shown); correlations reflected pooled data from all women. ¹³C enrichment was significantly and positively associated with TBSs. Milk VA ¹³C enrichment did not have any significant associations; the correlation estimate with TBSs was also positive with P = 0.06.

Comparison of VA ¹³C enrichment in paired serum and milk

In the subgroup of lactating women in Zambia, serum and milk VA ¹³C enrichment were compared. Overall, there was a statistically significant, moderate association between serum and milk enrichment (Figure 2A; Spearman correlation coefficient 0.52, P = 0.0001). Bland–Altman analysis revealed an isotopic abundance (%) mean bias of −0.00060 with 95% limits of agreement of −0.00325 and 0.00206 (Figure 2B). There was no notable pattern of bias related to ¹³C isotopic abundance.

FIGURE 2.

Comparison of paired serum and milk vitamin A ¹³C natural abundance in lactating women in Zambia (n = 48). (A) Scatter plot with best fit line (solid) and line of identity (dashed). (B) Bland–Altman plot with mean bias (solid line) and upper and lower 95% limits of agreement (dashed lines).

Impact of individual vs. population natural abundance on RID

Mean bias, bias fraction of reference mean (%), and 95% limits of agreement for comparing TBS and TLR using individual compared with study or global population retinol ¹³C-natural abundance are presented in Table 3. In children, study-level ¹³C-natural abundance had a mean bias percentage range of −1.7% to −0.7%, whereas global-level outcomes had a mean bias percentage range of 1.4%–2.2% for RID outcomes. In women, study- and global-level ¹³C-natural abundance had a consistent direction of mean bias, which ranged from −3.7% to −1.6%. Limits of agreement were wider using global estimates compared with study estimates in children, whereas they were more comparable in women. Limits of agreement ranged from −23% to 37% in children and −21% to 14% in women.

TABLE 3.

Comparison of vitamin A total body stores and estimated liver concentrations using individual and population retinol ¹³C-natural abundance in children and women from 6 African countries

	TBS-study1	TBS-global1	TLR-study	TLR-global	TLR-study	TLR-global
	(μmol)		50% of TBS in liver (μmol/g)		80% of TBS in liver (μmol/g)
Children
Mean bias	−17	14	−0.018	0.011	−0.029	0.017
95% LoA	(−318, 283)	(−370, 398)	(−0.343, 0.306)	(−0.390, 0.411)	(−0.549, 0.490)	(−0.623, 0.657)
Mean bias % of reference mean2	−1.7	2.2	−0.7	1.4	−1.1	2.2
95% LoA	(−23, 20)	(−32, 37)	(−15, 13)	(−20, 23)	(−23, 21)	(−32, 37)
n	570		568		568
Women
Mean bias	−31	−25	−0.010	−0.008	−0.016	−0.012
95% LoA	(−202, 139)	(−180, 130)	(−0.063, 0.043)	(−0.056, 0.041)	(−0.100, 0.069)	(−0.090, 0.066)
Mean bias % of reference mean2	−3.7	−2.5	−2.3	−1.6	−3.7	−2.5
95% LoA	(−21, 14)	(−20, 14)	(−13, 9)	(−12, 9)	(−21, 14)	(−20, 14)
n	136		136		136

LoA, limits of agreement; TBS, total body vitamin A store; TLR, total liver vitamin A reserve.

¹Study and global refer to TBS or TLR outcomes calculated using study and global population-level retinol ¹³C-natural abundance values, respectively.

²Mean bias, mean bias % of individual reference, and 95% limits of agreement were determined relative to vitamin A outcomes calculated using individual retinol ¹³C-natural abundance values within children or women subgroups.

For children, covariates associated with error in TBS_study and TBS_global compared with TBS_individual (all μmol) were: TBSs (β = 70.8 and 85.9, respectively, both P < 0.001, Figure 3A, 3B) and a TBSs by retinol ¹³C-natural abundance percentage interaction (β = −65.9, −79.8, respectively, both P <0.001). The effect of retinol ¹³C-natural abundance was significant for TBS_global only (β = 3481, P = 0.08; β = 6949, P = 0.021; respectively), as was ln CRP, but with a small impact on error (mg/L, β = −1.2, P = 0.30; β = −0.88, P = 0.012; respectively). Study effects with P value of <0.05 included Cameroon, South Africa, and Zambia 2012 (Z = 35.1, 43.6, and −57.5, respectively). Serum retinol, RBP, and AGP levels were not significantly associated with either TBS_study or TBS_global.

FIGURE 3.

Comparison of vitamin A total body stores (TBSs) using individual and population retinol ¹³C natural abundance in children and women from 6 African countries. Bland–Altman analysis compared the difference in vitamin A TBSs calculated using individual (TBS_individual) and study population level (TBS_study) retinol ¹³C natural abundance against TBS_individual as the reference. Relative TBS_study error (%) was calculated as (TBS_study – TBS_individual)/TBS_individual. (A) Children with available TBS_individual data (n = 570) and (B) relative TBS_study error (%). (C) Lactating and nonlactating women with available TBS_individual (n = 136) and (D) relative TBS_study error (%).

For women, error in TBS_study and TBS_global was associated with TBSs (β = 83.2, 83.9, Figure 3C, 3D), retinol ¹³C-natural abundance (β = 19,089, 17,250), and their interaction (β = −77.4, −78.0), all respectively and P < 0.001; serum retinol was not significantly associated with either TBS_study or TBS_global.

Discussion

This study demonstrated within- and between- country differences in VA ¹³C-natural enrichment, which had significant associations with age, inflammation, and TBSs. Comparing TBSs calculated using individual or population estimates of retinol ¹³C-natural abundance demonstrated low mean bias but wide 95% limits of agreement. We conclude that variation in retinol ¹³C-natural abundance should be expected in other populations or population subgroups. Using study-specific or population retinol ¹³C-natural abundance estimates would be expected to provide reasonably accurate estimates of continuous RID outcomes, but individual-level natural abundance measurements would be necessary for accurate determination of VA stores and resulting status (deficient, adequate, or excessive) at the individual level.

This analysis included a relatively large sample size by combining VA ¹³C-natural isotopic enrichment among 8 studies in 6 African countries. Analysis of ¹³C-natural abundance for all studies was conducted in the same laboratory. The use of GC-C-IRMS provides high sensitivity and precision for determining ¹³C abundance at natural enrichment levels [30], which is required for determining subtle differences in enrichments among populations, as shown in this study. Furthermore, we observed that age, TBSs, and inflammation were all inversely associated with retinol ¹³C abundance in children and positively associated with TBSs in women. The comparison between paired serum and breastmilk VA ¹³C-natural abundance demonstrated that these 2 sources provide comparable results, which is in agreement with a prior report [24].

The inverse association of retinol ¹³C enrichment with age is consistent with age-related changes in hair ¹³C enrichment following weaning in children [31] and in RBC ¹³C and ¹⁵N in adolescents compared with older adults [32]. Both instances are hypothesized to be because of life stage-associated dietary patterns; differences in subsistence-based compared with market-based diets [32], and the transition from breastfeeding through weaning [31,32]. These factors could explain the similar effect of age that we observed in this study and are reflected in the lack of significance of an age effect on retinol ¹³C enrichment when between-study variability was controlled.

Breastfeeding acts as an additional trophic level, resulting in enriched ¹³C in the child relative to the mother. This difference slowly equilibrates to others consuming similar diets by 4 y of age [31]. We noticed a similar decrease in enrichment in children spanning this age range, which would be consistent with weaning following breastfeeding. Children in these studies would have had access to both high-dose VA supplements before the age of 5 y and fortification of foods, such as sugar, which would decrease ¹³C in the retinol body pool. However, in our study, we observed that lactating women had very similar serum and milk VA ¹³C enrichments, indicating that this stage is not responsible for significant ¹³C fractionation for VA. It is possible that different dietary forms of VA (e.g., provitamin A carotenoids, preformed VA from animal source foods, or supplements) could undergo differential ¹³C fractionation during digestion, absorption, and metabolism because of the nutrient form and food matrix.

In women, TBSs were positively associated with retinol ¹³C enrichment. This could result from those with higher TBSs consuming VA-rich foods or supplements that have higher ¹³C enrichment. Additionally, it could be that through continued metabolism and recycling of VA in the body, the VA pool is slowly enriched over time. VA is fat-soluble and primarily stored in the liver; however, it recycles extensively between liver, blood, and extrahepatic target tissues. It has a relatively long system half-life of approximately140 d in humans with adequate to high VA stores, and a tracer dose was detectible for 5 y in a human female [33], providing ample time for fractionation to take place and accumulate during these metabolic processes in those with higher stores.

A limitation of this analysis is that we were not able to assess biomarkers of interest (e.g., CRP and AGP) in all studies. It remains to be answered if the same relationship between ¹³C enrichment and inflammation would hold in women. We cannot determine any causal directionality in this relationship; one explanation for the association observed could be that altered VA metabolism during inflammation could result in human metabolic isotopic fractionation. In rats, induced inflammation reduced hepatic VA mobilization, which resulted in a decoupling of VA tracer enrichment between liver VA stores and more recently ingested VA circulating in blood. A similar effect in humans could bias circulating retinol ¹³C enrichment toward the enrichment of recently consumed VA, which would then equilibrate toward enrichment of VA stores when inflammation resolves. This is consistent with the inverse association of CRP with retinol ¹³C-natural abundance and the hypothesis of gradual enrichment of the VA storage pool because of metabolic fractionation over time. If unaccounted for, this could produce errors in both retinol ¹³C-natural enrichment and RID outcomes. We note that the rat study induced significant inflammation, and these studies excluded participants with fever. We only noticed a minor significant effect of CRP on TBS_global estimates in children but not on any study population-level estimates. It could also be that differences in VA enrichments are due to dietary pattern differences, which in turn alter immune function and inflammatory response. One analysis in children in Bangladesh revealed that hair ¹³C enrichment was lower in stunted children, hypothesized to be because of diets lower in animal-sourced foods [31].

Additionally, analysis of food samples for VA ¹³C enrichment would allow further explanation of differences among and between countries. Determining the ¹³C enrichment patterns of plant and animal source foods, fortificants, and supplements could help explain the natural variation observed between studies included in this analysis. Furthermore, this could allow estimation of relative contributions of each of these components to VA status, provided they had enough variation in natural enrichments. This has been demonstrated between a C3 plant (carrot), a C4 plant (maize), and a fortificant in a gerbil model [16] and evaluated in humans following a biofortified maize feeding trial [17]. Nonetheless, the difference between the South African and Zambian child cohorts is likely a direct result of the diets consumed. The children in the area studied in South Africa consumed sheep liver, which would have a higher enrichment due to the C4 grasses that the sheep consume [20]. This is in contrast to the Zambian children who consumed many orange fruits and vegetables, which are all C3 plants [23,34]. The synthetic retinyl palmitate from mandated fortified sugar and World Health Organization recommended high-dose supplements typically have a lower enrichment of ¹³C [35].

Future research directions include assessing inflammatory biomarkers in lactating and nonlactating women to see if the same relationship between ¹³C enrichment and inflammation observed in children is also observed in women. This could provide insights as to what physiologic states or metabolic processes impact VA ¹³C stable isotope ratios so these covariates can be accounted for in analysis of dietary intakes or tracer studies. Further modeling studies could assess the impact of RID calculation error from study- or population-level estimates on the diagnostic accuracy to categorize VA status as deficient, adequate, or excessive. This study shows that there are within- and between-population differences in VA ¹³C enrichment that should be considered for VA applications utilizing ¹³C stable isotope methodology. In the application of the RID test, clinical studies evaluating individual VA status should always measure the baseline ¹³C enrichments. Even individual differences are significant and dependent on foods consumed and supplement usage [17]. Restriction of foods and supplements high in VA during the RID mixing period is often used to eliminate influence on the serum to liver tracer ratio when individual VA status is important [36,37]. However, in population surveys, analysis of a subgroup or a nondosed group is recommended to reduce evaluation costs, and dietary restriction is usually not feasible.

Biomarkers of dietary intake are important to address compliance to interventions. Although the concept of precision or personalized nutrition is relatively new and has some strengths in regard to health benefits [38], more research is needed to ensure that methods to assess dietary intake and nutritional status are accurate and accessible at the individual level. Having biomarkers available to support survey tools such as dietary recalls and food frequency questionnaires will be essential for randomized controlled trials designed to evaluate individualized dietary recommendations.

Funding

Manuscript preparation was supported by NIH-R01 DC019357 and Particles for Humanity. Sample collection was supported by the regional Technical Cooperation project RAF6047 from the International Atomic Energy Agency (IAEA) and the IAEA Coordinated Research Project E4.30.30, contract number 18842 (to MEvS and MAD); the Tanzania Food and Nutrition Center (EMU); and the Ethiopian Public Health Institute (to THB and MW). The study of Zambian women was funded in part by the University of Wisconsin-Madison Global Health Institute visiting scholar fellowship (to CK and SAT), an endowment entitled “Friday Chair for Vegetable Processing Research” (to SAT), and HarvestPlus. NIH grant T32-DK007665 (BMG) and HarvestPlus also supported the field expenses for the studies of Zambian children.

Author disclosures

BMG, OOS, ANZ, GMN, THB, MW, MEvS, MAD, EMU, CK, JC, NK, CRD, MG, and SAT, no conflicts of interest.

Appendix A. Supplementary data

The following is the Supplementary data to this article: