Prebiotics increase iron absorption and reduce the adverse effects of iron on the gut microbiome and inflammation: a randomized controlled trial using iron stable isotopes in Kenyan infants

Abstract

Background

Iron fortificants tend to be poorly absorbed and may adversely affect the gut, especially in African children.

Objective

We assessed the effects of prebiotic galacto-oligosaccharides/fructo-oligosaccharides (GOS/FOS) on iron absorption and gut health when added to iron-fortified infant cereal.

Methods

We randomly assigned Kenyan infants (n = 191) to receive daily for 3 wk a cereal containing iron and 7.5 g GOS/FOS (7.5 g+iron group), 3 g (3-g+iron group) GOS/FOS, or no prebiotics (iron group). A subset of infants in the 2 prebiotic+iron groups (n = 66) consumed 4 stable iron isotope–labeled test meals without and with prebiotics, both before and after the intervention. Primary outcome was fractional iron absorption (FIA) from the cereal with or without prebiotics regardless of dose, before and after 3 wk of consumption. Secondary outcomes included fecal gut microbiota, iron and inflammation status, and effects of prebiotic dose.

Results

Median (25th–75th percentiles) FIAs from meals before intervention were as follows: 16.3% (8.0%–27.6%) without prebiotics compared with 20.5% (10.4%–33.4%) with prebiotics (Cohen d = 0.53; P < 0.001). FIA from the meal consumed without prebiotics after intervention was 22.9% (8.5%–32.4%), 41% higher than from the meal without prebiotics before intervention (Cohen d = 0.36; P = 0.002). FIA from the meal consumed with prebiotics after intervention was 26.0% (12.2%–36.1%), 60% higher than from the meal without prebiotics before intervention (Cohen d = 0.45; P = 0.007). After 3 wk, compared with the iron group, the following results were observed: 1) Lactobacillus sp. abundances were higher in both prebiotic+iron groups (P < 0.05); 2) Enterobacteriaceae sp. abundances (P = 0.022) and the sum of pathogens (P < 0.001) were lower in the 7.5-g+iron group; 3) the abundance of bacterial toxin-encoding genes was lower in the 3-g+iron group (false discovery rate < 0.05); 4) fecal pH (P < 0.001) and calprotectin (P = 0.033) were lower in the 7.5-g+iron group.

Conclusions

Adding prebiotics to iron-fortified infant cereal increases iron absorption and reduces the adverse effects of iron on the gut microbiome and inflammation in Kenyan infants.

This trial was registered at clinicaltrials.gov as NCT03894358.

Introduction

Iron deficiency anemia is highly prevalent in African infants, and fractional iron absorption (FIA) from iron-fortified foods and supplements for infants is generally low [1,2]. Thus, most of the iron fortificant passes unabsorbed into the colon where it favors growth of enteropathogens that require iron for replication and virulence. By contrast, important commensal barrier strains, such as bifidobacteria and lactobacilli require little or no iron for growth [[3], [4], [5], [6]]. In African infants and children in settings with poor hygiene, iron fortification may adversely affect the gut microbiota, decrease beneficial bifidobacteria and lactobacilli, and increase enteropathogens and gut inflammation [[7], [8], [9], [10]] although not all studies agree [11,12]. These adverse changes in the gut microbiota likely contribute to the increased risk of diarrhea when iron is provided to infants and children in low-resource settings [3,13].

Promising strategies to reduce the adverse effects of iron on the gut are to use the lowest possible iron dose with proven efficacy and the coprovision of prebiotic fibers [9], which can selectively enhance the growth of beneficial gut commensal bacteria [14,15] and protect from overgrowth of enteropathogens [14,15]. In a 4-mo trial in Kenyan infants, the addition of prebiotic galacto-oligosaccharides (GOS) to an iron-fortified micronutrient powder (MNP) mitigated most adverse effects of iron on the gut [9]. An additional benefit of providing prebiotics with iron is that they may increase iron absorption [[16], [17], [18]]. In anemic Kenyan infants who consumed MNP containing 5 mg iron and 7.5 g GOS or the same MNP without GOS daily for 3 wk, GOS consumption increased iron absorption by 62% [16]. However, the design of that study could not distinguish whether the enhancing effect of GOS on iron absorption was an acute effect (from addition to the test meal) or a conditioning effect (previous consumption of GOS daily for 3 wk), or both [16].

Previous studies in Kenyan infants used only GOS in combination with iron [16,19]. The addition of a prebiotic mixture of short-chain GOS and long-chain fructo-oligosaccharides (scGOS/lcFOS) with a 9:1 ratio to infant formula can shift the gut microbiota toward a Bifidobacterium sp. rich community and lower gut pH while suppressing the growth of enteropathogens [20,21]. GOS and FOS have generally recognized as safe status in the United States [22] and have been approved for infant formula use in Europe [23]. Therefore, in this study, we assessed the effect of the addition of 2 different doses of scGOS/lcFOS on iron absorption from a low-dose (3.6 mg iron) infant cereal formulated for complementary feeding in Africa. The primary outcome was FIA from the cereals with or without prebiotics, measured before and after the cereals were consumed daily for 3 wk.

Methods

Study design and participants

This study was an intervention trial in infants conducted in Kwale County in south Kenya. We compared the effects of daily consumption of a wheat-based instant cereal containing 3.6 mg iron and 2 different doses of scGOS/lcFOS at a ratio of 9:1 or no prebiotics. In a random subset of infants in the intervention study, we measured iron absorption using stable iron isotopes from the cereal with and without the prebiotics, before and after the 3-wk intervention. Our primary outcome was FIA from the fortified cereal with or without prebiotics, regardless of dose, before and after 3 wk of consumption, measured by erythrocyte isotope incorporation. The study design is shown in Figure 1. Secondary outcomes were the effect of the cereal, with or without prebiotics, before and after 3 wk of consumption, on the following: 1) fecal bifidobacteria abundance, fecal microbiome composition, and fecal pH; 2) hemoglobin (Hb), plasma ferritin (PF), and soluble transferrin receptor (sTfR) concentrations; and 3) plasma C-reactive protein (CRP), α-1-acid glycoprotein, intestinal fatty acid binding protein, and fecal calprotectin. An additional secondary outcome was the effect of prebiotic dose (3.5 g compared with 7 g) on FIA and the above secondary outcomes.

FIGURE 1.

Study overview.

We conducted the study from July 2019 to January 2020. Inclusion criteria were as follows: 1) age 6–11 mo; 2) no acute or chronic illness; 3) z-scores for weight-for-age and weight-for-length ≥ −3; 4) Hb ≥ 70 g/L; 5) no intake of iron-containing mineral and vitamin supplements in the previous 2 mo; and 6) no antibiotic use in the previous 1 mo. Caregivers gave informed consent with a written signature or a fingerprint. The study was approved by ethics committees of the ETH Zurich, Switzerland (EK 2018-N-84), and Jomo Kenyatta University of Agriculture and Technology, Kenya (JKU/2/4/896B) and registered at clinicaltrials.gov as NCT03894358.

Study procedures

Randomization

After screening for eligibility, we individually randomly assigned infants to receive a 3-wk intervention of the cereal containing 3.6 mg iron as ferrous fumarate (FeFum) and ascorbic acid at a 4:1 ascorbic acid:iron molar ratio and one of the following: 1) 7.5 g of the prebiotics mixture (7.5-g+iron group); 2) 3 g of the same prebiotics (3-g+iron group); or 3) no prebiotics (iron group). Moreover, within these groups, infants were individually randomly assigned to the iron absorption study or not (Figure 1). Randomization was done using a computer-generated (Excel; Microsoft Office 2016) list and color codes. Caregivers were blinded to the intervention assigned to their infants.

Intervention and absorption studies

Infants consumed the cereal daily for 3 wk at home. We provided caregivers with measuring and feeding cups and instructions to prepare and feed the infant 2 portions (24 g + 90 mL water per portion) per day at home. Before and after the 3-wk intervention, we measured height and weight; collected whole blood by venepuncture for the determination of Hb, iron and vitamin A status, systemic inflammation, and gut integrity; and collected fecal samples for gut inflammation, microbiota composition, and functional analyses. We performed weekly visits during which we completed a morbidity questionnaire, assessed compliance, collected the leftover cereal, and provided the subsequent week’s supply.

Seventeen days before the start of the intervention, the subset of infants in the prebiotic+iron groups who participated in the iron absorption study consumed 2 test meals of the wheat-based cereal, with and without one of the prebiotic doses, labeled with stable iron isotopes (⁵⁸FeFum and ⁵⁷FeFum) on 2 mornings separated by 72 h (Figure 1). Details of the test meals and feeding procedures are given in Supplemental Material. Two weeks after the second test meal, we collected venous blood for the determination of Hb and the erythrocyte incorporation of the stable iron isotope labels, measured height and weight, and collected fecal samples. For collection of fecal samples, mothers were provided with collection materials and instructed to collect the samples on the morning of the study visit. These blood and fecal samples served as baseline samples for the 3-wk intervention. During the 3-wk intervention, the subset of infants in the absorption study received the same prebiotic dose as in the absorption study. After 3 wk, we again collected venous blood and fecal samples and redetermined the erythrocyte iron–isotopic ratio as a new baseline for the second set of test meals. Then, the infants received the same 2 labeled test meals as at baseline, with and without prebiotics. Two weeks after the last test meal, we collected venous blood for the determination of Hb and the erythrocyte incorporation of the stable iron isotopes. Any infants who remained anemic at study end were treated according to local guidelines.

Infant cereal

Instant wheat-flour based cereal products were produced by Danone at Villefranche-sur-Saone, France. The description of the products is provided in Supplemental Material, and their detailed nutritional composition is shown in Supplemental Table 1.

Stable-isotope labels

The ⁵⁷FeFum and ⁵⁸FeFum were prepared by Dr. Paul Lohmann GmbH from ⁵⁷Fe-enriched and ⁵⁸Fe-enriched elemental iron (95.55% and 99.89% isotopic enrichment, respectively; Chemgas). We analyzed the labeled iron compounds for iron–isotopic composition and the tracer iron concentration using isotope-dilution mass spectrometry, as described further.

Biologic sample collection and laboratory analyses

We collected venous blood samples into lithium-and-heparin–coated vacutainers and measured Hb using a HemoCue Hb 301 analyzer (Angelholm), with quality controls provided by the manufacturer. Anemia was defined as Hb of <110 g/L [24] . We prepared whole blood and plasma aliquots and froze them at −20 °C. We collected human milk samples for human milk oligosaccharides (HMO) analysis by manual milk expression by the mother into a clean plastic container; after homogenization, 1–2 mL portions were stored at −20 °C. The caregivers were carefully instructed to collect the infants’ fecal samples on the evening preceding or on the morning of the study visit. Details of the stool processing, DNA extraction, and qPCR analyses are described in Supplemental Material and in Supplemental Table 2. We analyzed gut microbiota by qPCR for total bacteria, Bifidobacterium spp., Lactobacillus/Pediococcus/Leuconostoc spp., and Enterobacterium spp. and for selected enteropathogenic bacteria, such as Campylobacter spp., Salmonella spp., Clostridium difficile, Clostridium perfringens, enteropathogenic Escherichia coli, and enterotoxigenic E. coli. We chose these pathogens because they are common in Kenyan infants in this area [8,9]. We performed 16S rRNA gene sequencing at baseline and end point on all subjects and shotgun metagenomics on all infants participating in the absorption study plus a subset of age-matching and sex-matching children from the iron-only group (n = 105). Full methods and primers for gut microbiota analysis are described in Supplemental Material.

In whole blood, we determined FIA by measuring erythrocyte incorporation of the stable iron isotopes. Details of the isotopic analyses are given in Supplemental Material. We measured iron isotope ratios using an inductively coupled plasma mass spectrometer (Neptune; Thermo Finnigan) [25] and calculated the amount of ⁵⁷Fe and ⁵⁸Fe isotopic labels in blood 14 d after the administration of the second and fourth labeled test meals, based on the shift in iron–isotopic ratios and the estimated amount of circulating iron [26] and considering that iron–isotopic labels were not monoisotopic [27,28]. We analyzed the plasma for PF, soluble transferrin receptor, CRP, α-1-acid glycoprotein, retinol-binding protein [29], plasma hepcidin, and intestinal fatty acid binding protein. PF was adjusted for inflammation using the BRINDA method [30]. In feces, we measured pH and calprotectin. In breast milk samples, we measured HMOs and classified the samples into 4 groups defined based on secretor (Se) and Lewis (Le) polymorphisms. Methods and reference values for all these assays are shown in the Supplemental Material.

Statistical analysis

Using G∗Power Statistical Program (v.3.1.3), we calculated the sample sizes necessary to detect a 1-tailed 42% difference in FIA between the 2 arms and a 1-tailed 30% difference in FIA within the infants (without compared with prebiotics and acute compared with chronic effect). Based on a SD of 0.228 from log-transformed FIA from previous studies at the ETH Zurich and assuming a type I error rate of 5% and power of 80%, our sample size calculation indicated that 29 infants were needed in each arm for the stable iron isotope study. Anticipating a dropout rate of 18%, we aimed to enroll 35 infants per group. The power calculation for the intervention study was based on the data reported by Paganini et al. [9] , in which Kenyan infants were administered either iron-only or iron with prebiotics daily for 4 mo. They estimated that a sample size of 35 infants in each group would be sufficient to detect a 2-tailed difference of 0.85 log number of copies/g feces in bifidobacteria, considering an SD of 1.25 in log abundance, with a type I error rate of 5% and power of 80%. However, considering multiple testing of outcomes, the number of participants per group was increased to 50 in our study. Anticipating a dropout rate of 30%, we aimed to enroll 65 infants per group in the intervention study.

We performed the statistical analyses using the R statistical programming environment v.4.0.2. We checked the data for normality using the Shapiro–Wilk W test and Q–Q plots. Values in the text and in tables are presented as means ± SD for normally distributed data and as medians (25th–75th percentiles) for nonnormally distributed data. When data were not normally distributed, appropriate transformation of values was performed before statistical analysis. We tested for homogeneity of variances before performing relevant statistical tests. We calculated z-scores for weight-for-age, weight-for-length, and length-for-age using WHO Anthro software v.3.2.2. We used Pearson χ² tests to compare categorical variables between groups at baseline, and where the sample size was not sufficient, we used Fisher exact tests. We used independent-samples t tests to compare continuous variables between the 2 groups (arms 1 and 2) at baseline. For comparisons of FIA without and with prebiotics, regardless of dose (our prespecified primary outcomes), we used paired-samples t tests for normally distributed data and related samples Wilcoxon signed rank tests for not normally distributed data. We calculated Cohen d to estimate effect sizes on the primary outcomes. Because of between-group differences in iron status and the influence of iron status on absorption, we adjusted FIA to a corrected PF of 12 μg/L for between-group comparisons. We used Bonferroni adjustment to correct our results for multiple comparisons (level of significance: P < 0.017). We used linear mixed-effect model analyses to assess whether the dose of the prebiotic affects the following: 1) acute prebiotic effect [dependent variable: FIA before intervention without and with prebiotic; fixed effects: prebiotic (without/with), dose (7.5 g/3 g)]; 2) chronic prebiotic effect [dependent variable: FIA without prebiotic before and after intervention; fixed effects: time (before intervention/after intervention), dose (7.5 g/3 g)]; and 3) combined acute and chronic prebiotic effect [dependent variable: FIA before intervention without prebiotic and after intervention with prebiotic; fixed effects: acute+chronic prebiotic (yes/no), dose (7.5 g/3 g)]. We added serum ferritin and CRP concentrations measured at the time of the absorption studies as covariates to these models. In the intervention study, for continuous variables, between-group differences before the intervention were tested using 1-factor analysis of variance. Between-group differences after the intervention were tested using analysis of covariance with values before the intervention as covariates. We assessed the intervention effect on gut microbiota analyzed by qPCR by fitting linear mixed-effects models, with detailed methods described in the Supplemental Material. We defined the fixed effects on the variance as time, group, time-by-group, age; the random structure was defined as the subject. If the interaction term was significant, a post hoc analysis was performed to investigate the effect of 1 factor within levels of the other by applying Bonferroni correction. P values of <0.1 were considered not statistically significant trends. P values of <0.05 were considered statistically significant.

For 16S rRNA gene amplicon and shotgun sequencing, statistical analyses and graphs were performed with R software (version 3.6.0). α-Diversity was analyzed as change from baseline using linear mixed model (nlme) or using Mann–Whitney or Wilcoxon signed rank test. Beta-diversity was analyzed using PERMANOVA (vegan) [31]. Differential analyses were performed with DESeq2 (version 1.24.0) on 16S based and metagenomic data sets. The prediction of virulence factors, specifically bacterial toxins, was performed using PathoFact, a pipeline dedicated to pathogenesis analysis from metagenomics data sets [32]. Fold changes were evaluated with the Wald test [false discovery rate (FDR)-adjusted P < 0.1]. All methods are detailed in the Supplemental Material.

Results

Participants

We recruited study participants from July to November 2019. The intervention period ran in parallel from July 2019 to December 2020. We screened 258 infants and randomly assigned 191 eligible infants to the 3 study groups (Figure 1). In the 7.5-g+iron group and 3-g+iron group, 32 of 63 infants and 34 of 64 infants, respectively, were then randomly assigned to the iron absorption study. For the 7.5-g+iron group, the 3-g+iron group, and the iron group, 50, 58, and 59 infants, respectively, finished the intervention (Figure 1); reasons for attrition during the study are given in the Supplemental Material. Compliance with cereal consumption was 93%, 90%, and 87%, respectively.

Iron absorption study

Table 1 presents the baseline characteristics of the infants in the iron absorption study. Overall, 74.6% of the infants were anemic, 79.4% were iron deficient, 38.1% were vitamin A deficient, and 17.5% experienced inflammation. There were no significant baseline group differences, with the exception that PF (both unadjusted and adjusted) was lower in the 7.5-g+iron group than the 3-g+iron group (P = 0.018 and P = 0.020, respectively). Because they had CRP concentration of >5 mg/L at the time of test meal administration, we excluded FIA values from 6 infants before intervention and from 5 infants after intervention because of a potential bias from inflammation on iron absorption. We also excluded 1 FIA value from 1 infant in the 7.5-g+iron group who vomited the test meal. Therefore, to determine the acute and chronic effect of prebiotics on FIA, in the 7.5-g+iron group and in the 3-g+iron group, 25 and 28 infants completed the absorption study, respectively, and were included in the analyses.

TABLE 1.

Baseline characteristics of the Kenyan infants participating in the iron absorption study, by prebiotic dose group^∗.

	3.0 g Prebiotics	7.5 g Prebiotics	P
n	34	29
Female/male	17 (50.0)/17 (50.0)	14 (48.3)/15 (51.7)	0.891
Age, mo1	8.0 ± 1.3	7.8 ± 1.3	0.594
Body length, cm	68.8 ± 3.0	68.3 ± 3.0	0.450
Body weight, kg	7.9 ± 1.1	7.9 ± 1.3	0.826
Weight-for-length z-score	−0.35 ± 1.11	−0.11 ± 1.44	0.453
Weight-for-age z-score	−0.51 ± 0.97	−0.41 ± 1.33	0.730
Length-for-age z-score	−0.35 ± 0.76	−0.49 ± 1.07	0.532
Hemoglobin, g/L	106 (100–115)	103 (98–108)	0.457
Anemia1	23 (67.6)	24 (82.6)	0.170
Plasma ferritin, μg/L	14.2 (8.5–23.4)2	7.2 (3.1–16.6)	0.018
<12 μg/L	13 (39.4)2	19 (65.5)	0.040
Plasma ferritin adjusted, μg/L3	11.0 (8.1–19.0)2	6.4 (2.9–12.5)	0.020
<12 μg/L	19 (57.6)2	21 (72.4)	0.223
Soluble transferrin receptor, mg/L	11.0 (9.5–16.1)2	11.0 (8.9–16.2)	0.570
>8.3 mg/L	26 (78.8)2	23 (79.3)	0.960
Iron deficiency4	27 (81.8)2	23 (79.3)	0.803
Iron deficiency anemia5	19 (57.6)2	22 (75.9)	0.153
C-reactive protein, mg/L	0.24 (0.02–2.41)2	0.15 (0.03–1.14)	0.965
0.05–4.99	17 (51.5)2	16 (55.2)	0.773
≥5	3 (9.1)2	3 (10.3)	0.868
α-1-acid glycoprotein, g/L	0.67 (0.45–0.91)2	0.63 (0.49–0.77)	0.520
≥1	7 (21.2)2	3 (10.3)	0.312
Inflammation6	7 (21.2)2	4 (13.8)	0.445
Retinol-binding protein, μmol/L	0.74 ± 0.152	0.77 ± 0.18	0.552
Vitamin A deficiency7	13 (39.4)2	11 (37.9)	0.906

Independent-samples t tests were used to compare continuous variables. Pearson χ² or Fisher exact tests were used to compare categorical variables.

^∗Values are given as mean ± SD, median (25th–75th percentiles), or n (%).

¹<110 g/L [16].

²n = 33.

³Adjusted for inflammation using BRINDA correction [15].

⁴Adjusted plasma ferritin <12.0 μg/L and/or soluble transferrin receptor >8.3 mg/L [14].

⁵Anemia and iron deficiency [14].

⁶C-reactive protein ≥ 5 mg/L and/or α-1-acid glycoprotein ≥ 1 g/L [14].

⁷Retinol-binding protein < 0.70 μmol/L [14].

In a pooled analysis assessing FIA from both prebiotic doses (Table 2), the addition of prebiotics to the test meal at baseline resulted in a 26% higher median FIA than the test meal without prebiotics [16.3% (8.0%–27.6%) compared with 20.5% (10.4%–33.4%); Cohen d = 0.53; P < 0.001]. Median FIA from the test meal consumed without prebiotics after intervention was 41% higher than the median FIA from the test meal without prebiotics before intervention [16.3% (8.0%–27.6%) compared with 22.9% (8.5%–32.4%); Cohen d = 0.36; P = 0.002). Median FIA from the test meal consumed with prebiotics after intervention was 60% higher than the median FIA from the test meal without prebiotics before intervention [16.3% (8.0%–27.6%) compared with 26.0% (12.2%–36.1); Cohen d = 0.45; P = 0.007] (Supplemental Figure 1). The analyses assessing pooled FIA and FIA by prebiotic dose are summarized in Table 2. In the linear mixed-effect model analysis, there was no significant effect of the prebiotic dose on FIA: 1) when prebiotics were added to the test meal at baseline before intervention (P = 0.990); 2) from the test meals without prebiotics after intervention compared with those without prebiotics before intervention (P = 0.625); or 3) on FIA from the test meals consumed with prebiotics after intervention compared with those without prebiotics before intervention (P = 0.826). In a pooled analysis assessing the effect on FIA from both prebiotic doses, there was no effect of HMO type (P = 0.368) or secretor status (P = 0.112) on the enhancing effect of prebiotics on FIA before intervention and no effect of HMO type (P = 0.688) or secretor status (P = 0.238) on FIA after the 3-wk intervention.

TABLE 2.

Fractional iron absorption (%) from a wheat-based instant cereal containing 3.6 mg iron as ferrous fumarate without and with 2 doses of prebiotics (3.0 or 7.5 g of galacto-oligosaccharides and fructo-oligosaccharides, ratio 9:1), before and after 3 wk of daily consumption of the cereal containing the prebiotics^∗.

	Test meal	Pooled prebiotic doses	3.0 g Prebiotics	7.5 g Prebiotics
Iron absorption before 3-wk prebiotic intervention, %	Without prebiotic	16.3 (8.0–27.6) (n = 54)	13.4 (6.1–26.8) (n = 28)	18.6 (11.7–28.7) (n = 25)
	With prebiotic	20.5 (10.4–33.4)1 (n = 53)	17.6 (6.9–31.4) (n = 28)	25.8 (14.2–35.6)2 (n = 25)
Iron absorption after 3-wk prebiotic intervention, %	Without prebiotic	22.9 (8.5–32.4)3 (n = 53)	20.5 (9.5–33.1)4 (n = 28)	23.9 (14.0–38.1)4 (n = 25)
	With prebiotic	26.0 (12.2–36.1)2 (n = 53)	25.7 (7.6–31.4) (n = 28)	29.4 (9.3–33.4) (n = 25)

The pooled doses were compared by paired sample t tests, with Bonferroni adjustment. The data by dose was compared by linear mixed-effect models.

^∗Values are given as median (25th–75th percentiles).

¹P < 0.001; compared with the test meal without prebiotic before intervention.

²P < 0.01; compared with the test meal without prebiotic before intervention.

³P < 0.005; compared with the test meal without prebiotic before intervention.

⁴P < 0.05; compared with the test meal without prebiotic before intervention.

Intervention study

Table 3 summarizes before and after the 3-wk intervention sex, age, anthropometrics, iron and vitamin A status, systemic inflammation, hepcidin, and gut inflammation and integrity of the study population. There were no significant baseline group differences. After 3 wk, compared with the iron-only group, the following results were observed: 1) median CRP was lower in the 3-g+iron group [0.74 mg/L (0.26–3.83 mg/L) compared with 0.27 mg/L (0.04–1.35 mg/L); P = 0.008]; 2) median fecal pH was lower in the 7.5-g+iron group [5.5 (5.1–5.9) compared with 5.0 (4.6–5.4); P = 0.001] (Figure 2A); and 3) median fecal calprotectin, a measure of gut inflammation, was lower in the 7.5-g+iron group [361.0 μg/g (168.2–725.4 μg/g) compared with 198.4 μg/g (102.5–463.0 μg/g); P = 0.033]: it increased ∼10% in the iron group but decreased ∼20% in the 7.5-g+iron group (Figure 2B).

TABLE 3.

Gender, age, anthropometrics, iron and vitamin A status, systemic inflammation, hepcidin, and gut inflammation and integrity in Kenyan infants before and after 3 wk of consuming a wheat-based instant cereal containing 3.6 mg iron as ferrous fumarate without and with 2 doses of prebiotics (3.0 or 7.5 g of galacto-oligosaccharides and fructo-oligosaccharides, ratio 9:1).

	Fe	Fe + 3 g Prebiotics	Fe + 7.5 g Prebiotics	P
n	61	60	53
Female/male, n	33/28	37/23	28/25	0.582
Age, mo1	8.1 ± 1.3	8.3 ± 1.4	8.5 ± 1.5	0.272
Body length, cm
Baseline	68.5 ± 3.3	68.7 ± 3.2	69.5 ± 2.8	0.171
End point	69.9 ± 2.92	69.9 ± 3.23	70.7 ± 3.14	0.784
Body weight, kg5
Baseline	7.8 (7.1–8.7)6	7.9 (7.2–8.6)	7.9 (7.1–8.7)	0.999
End point	8.2 (7.2–8.9)2	8.0 (7.4–8.7)3	8.1 (7.4–8.8)4	0.992
Weight-for-length z-score
Baseline	−0.36 (−0.84–0.87)6	−0.15 (−0.97–0.66)	−0.49 (−1.28–0.24)	0.168
End point	−0.12 (−1.09–0.73)2	−0.07 (−0.98–0.54)3	−0.37 (−1.88–0.89)4	0.890
Weight-for-age z-score
Baseline	−0.61 (−1.18–0.35)6	−0.34 (−1.29–0.14)	−0.60 (−1.41–0.07)	0.800
End point	−0.48 (−1.11–0.57)6	−0.23 (−1.22–0.17)	−0.56 (−1.35–0.21)	0.970
Length-for-age z-score
Baseline	−0.52 ± 1.26	−0.50 ± 0.97	−0.31 ± 0.99	0.552
End point	−0.38 ± 1.102	−0.44 ± 1.023	−0.29 ± 1.104	0.837
Hemoglobin, g/L
Baseline	108 (101–116)	110 (103–117)	105 (100–110)	0.125
End point	108 (103–116)2	109 (103–115)3	108 (99–114)4	0.997
Plasma ferritin, μg/L
Baseline	12.2 (6.7–31.4)	16.6 (8.6–32.3)	10.9 (5.8–17.9)	0.201
End point	13.7 (9.4–27.2)2	14.6 (8.3–27.7)3	10.8 (6.3–18.6)4	0.737
Plasma ferritin adjusted, μg/L8
Baseline	10.7 (5.4–20.5)	12.9 (7.6–21.2)	8.5 (5.4–14.1)	0.188
End point	11.1 (7.2–22.5)2	12.4 (6.7–20.9)3	9.5 (5.3–14.9)4	0.491
Soluble transferrin receptor, mg/L
Baseline	10.3 (8.6–14.0)	10.9 (9.1–16.0)	12.0 (8.9–15.4)	0.303
End point	11.3 (8.8–14.8)2	12.5 (9.8–16.3)3	13.0 (9.0–17.3)4	0.714
C-reactive protein, mg/L
Baseline	1.16 (0.25–3.41)	0.38 (0.04–3.58)	1.16 (0.05–4.21)	0.301
End point	0.74 (0.26–3.83)2,b	0.27 (0.04–1.35)3,a	0.63 (0.04–2.25)7,a,b	0.008
α-1-acid glycoprotein, g/L
Baseline	0.70 (0.57–1.12)	0.70 (0.50–1.17)	0.68 (0.50–1.00)	0.298
End point	0.80 (0.60–0.98)2	0.68 (0.54–0.93)3	0.69 (0.51–1.26)7	0.591
Hepcidin, ng/mL9
Baseline	—	1.5 (0.6–4.5)	2.0 (0.9–6.1)	0.616
End point	—	1.6 (0.5–4.4)	1.4 (0.6–2.9)	0.615
Retinol-binding protein, μmol/L
Baseline	0.7 (0.6–0.9)	0.8 (0.7–0.9)	0.7 (0.6–0.9)	0.051
End point	0.8 (0.7–0.9)2	0.9 (0.7–1.0)3	0.8 (0.7–0.9)7	0.096
Fecal calprotectin, μg/g
Baseline	327 (162.0–517.5)6	361.7 (169.9–671.3)2	248.2 (86.9–424.5)	0.082
End point	361.0 (168.2–725.4)3,b	367.7 (190.0–601.4)3,a,b	198.4 (102.5–463.0)4,a	0.033
Fecal pH
Baseline	5.2 (4.7–5.7)6	5.2 (4.8–5.7)2	5.0 (4.6–5.5)	0.064
End point	5.5 (5.1–5.9)10,c	5.8 (5.3–6.7)3,b	5.0 (4.6–5.4)7,a	<0.001
Intestinal fatty acid binding protein, ng/mL
Baseline	0.96 (0.71–1.41)3	1.04 (0.74–1.33)11	0.96 (0.72–1.30)7	0.461
End point	1.03 (0.72–1.43)10	1.03 (0.67–1.47)3	0.99 (0.64–1.28)7	0.756

For continuous variables, between-group differences before the intervention were tested using 1-factor analysis of variance. Between-group differences after the intervention were tested using analysis of covariance with values before the intervention as covariates. For categorical variables, between-group differences were tested using Pearson χ² tests. Across rows, different letter superscripts indicate significant differences: P < 0.05.

¹Mean ± SD all such values.

²n = 59.

³n = 58.

⁴n = 50.

⁵Median (25th–75th percentiles) all such values.

⁶n = 60.

⁷n = 49.

⁸Adjusted for inflammation [29].

⁹Only measured in 47 infants participating in the stable iron isotope study (arm 1: n = 23, arm 2: n = 24).

¹⁰n = 57.

¹¹n = 55.

FIGURE 2.

Fecal pH and calprotectin, Enterobacteriaceae and sum of the genes of the 7 targeted pathogens in Kenyan infants (n = 191) receiving daily a cereal containing 3.6 mg iron as ferrous fumarate and either 7.5 g of galacto-oligosaccharides/fructo-oligosaccharides (GOS/FOS) (Fe+7.5 g GF group) (n = 52), or 3 g GOS/FOS (Fe+3 g GF group) (n = 61), or no prebiotics (Fe alone group) (n = 60). (A) Fecal pH by group at baseline and after the 3-week intervention (end point). (B) Fecal calprotectin, a measure of gut inflammation, by group at baseline and end point. (C) Log gene copies/g feces of Enterobacteriaceae spp. by group at baseline and end point. (D) Log gene copies/g feces of the sum of all pathogens by group at baseline and end point. We assessed the effects of the intervention on fecal pH, calprotectin, Enterobacteriaceae and sum of pathogen genes by plotting linear mixed models. We defined the fixed effects on the variance as time, group, time-by-group, and age. The random structure was defined as the subject. The boxplots indicate the median and interquartile range (IQR), and the whiskers indicate the minimum and maximum values. The asterisks indicate significant difference based on a P value (∗∗∗P < 0.001, ∗∗ P < 0.01, ∗ P < 0.05).

Effect of intervention on gut microbiota composition and function

The linear mixed-effects models assessing the intervention effect on gut microbiota analyzed by qPCR are summarized in Table 4. During the intervention, there was no significant group-by-time interaction on total Bifidobacterium spp.; this lack of effect was confirmed with an additional series of qPCR analyses using a different protocol and primers (described in Supplemental Material). Enterobacteriaceae decreased in the 7.5-g+iron group and increased in the iron group (group-by-time interaction, P = 0.079) (Figure 2C). The summed variable of all pathogens was also decreased in the 7.5-g+iron group and increased in the iron group (group-by-time interaction, P = 0.09) (Figure 2D).

TABLE 4.

Fecal microbiota analyzed by qPCR before and after 3 wk of consuming a wheat-based instant cereal containing 3.6 mg iron as ferrous fumarate without or with 2 doses of prebiotics (3.0 or 7.5 g of galacto-oligosaccharides and fructo-oligosaccharides, ratio 9:1).

	Fe alone	Fe + 3 g Prebiotics	Fe + 7.5 g Prebiotics	Time	Group	Time by group
n	60	61	52
Eubacteria, log gene copies/g feces
Baseline	10.180 (9.714, 10.491)	10.166 (9.775, 10.3999)	10.015 (9.607, 10.202)	0.554	0.027	0.797
End point	10.108 (9.806, 10.360)	10.121 (9.739, 10.541)	9.905 (9.588, 10.203)
Bifidobacterium spp., log gene copies/g feces
Baseline	9.611 (9.068, 10.028)	9.657 (9.285, 10.071)	9.480 (9.143, 10.028)	0.139	0.75	0.857
End point	9.535 (9.076, 10.029)	9.599 (8.964, 10.114)	9.396 (9.119, 9.834)
Lactobacillus/Pediococcus/Leuconostoc spp. log gene copies/g feces
Baseline	8.489 (7.664, 8.873)	8.289 (7.822, 9.106)	8.792 (7.981, 9.425)	0.551	0.118	0.692
End point	8.388 (7.903, 8.953)	8.535 (7.946, 9.033)	8.637 (8.105, 8.927)
Enterobacteriaceae spp, log gene copies/g feces
Baseline	8.308 (7.891, 8.935)	8.271 (7.684, 8.699)	8.298 (7.778, 8.904)	0.266	0.022	0.079
End point	8.468 (7.992, 8.934)	8.263 (7.730, 8.794)	7.873 (7.352, 8.338)
Campylobacter spp., log gene copies/g feces
Baseline	7.989 (6.274, 8.690)	7.471 (6.166, 9.050)	7.047 (5.962, 8.432)	0.259	0.159	0.665
End point	8.432 (6.967, 9.392)	9.011 (6.826, 9.493)	7.172 (6.170, 8.937)
Salmonella spp., log gene copies/g feces
Baseline	2.420 (1.845, 2.874)	2.519 (2.095, 3.287)	2.325 (1.770, 2.990)	0.341	0.074	0.844
End point	2.102 (1.760, 2.828)	2.513 (1.857, 2.929)	2.258 (1.840, 3.356)
Enteropathogenic Escherichia coli (EPEC), log gene copies/g feces
Baseline	7.103 (6.203, 8.523)	6.812 (6.050, 8.150)	6.728 (5.301, 8.131)	0.531	0.551	0.801
End point	7.040 (5.924, 8.120)	7.013 (5.963, 8.188)	7.070 (5.523, 7.862)
Enterotoxigenic Escherichia coli (ETEC), log gene copies/g feces
Baseline	4.142 (2.929, 6.078)	3.833 (3.228, 6.061)	3.940 (3.098, 5.985)	0.437	0.982	0.854
End point	4.135 (3.134, 7.317)	4.762 (3.505, 6.407)	4.027 (3.149, 6.385)
Clostridium difficile, log gene copies/g feces
Baseline	6.327 (3.270, 7.998)	6.657 (4.412, 7.790)	5.913 (3.570, 7.864)	0.286	0.001	0.239
End point	6.584 (3.836, 7.660)	7.592 (6.282, 8.310)	5.857 (3.301, 6.795)
Clostridium perfringens, log gene copies/g feces
Baseline	5.817 (4.379, 6.694)	5.918 (4.429, 6.884)	5.665 (4.590, 7.037)	0.706	0.812	0.634
End point	6.025 (4.595, 7.019)	5.465 (4.521, 6.627)	5.572 (4.452, 6.377)
Summed variable of all pathogens, log gene copies/g feces
Baseline	8.874 (8.319, 9.304)	8.654 (8.113, 9.137)	8.485 (8.008, 9.152)	0.216	<0.001	0.09
End point	8.999 (8.510, 9.557)	8.762 (8.265, 9.224)	8.194 (7.425, 8.749)

Data are presented as median (IQR). Linear mixed-effects models: fixed effects on the variance as time, group, time-by-group age; the random structure was defined as the subject. Significance was set as P < 0.05.

To provide context for the intervention effects, Figure 3 shows family-level microbiota composition at baseline in all infants in a heatmap representing the 25 most abundant bacterial families detected in each stool sample by 16S rRNA gene sequencing. At baseline, members of Bifidobacteriaceae, Bacteroidaceae and Prevotellaceae were most abundant in the study population. During the intervention, there was a significant increase of genus-based α-diversity (using Shannon index) in the 3-g+iron group compared with baseline in the same group (P = 0.004) (Figure 4A). As seen with qPCR, no significant changes were observed in the abundance of Bifidobacterium in both prebiotic groups compared with the iron group at end point (data not shown). There was a significant increase in Bacteroides sp. abundance in the 3-g+iron group compared with the iron group at the end of intervention (FDR = 0.013) (Figure 4B). There was a significant increase in the abundance of Prevotella spp. in 7.5-g+iron group compared with that in the iron group (FDR = 0.00005) (Figure 4C), and there were significant increases of Lactobacillus sp. abundance in both prebiotic groups compared with those in the iron group (3g+iron FDR = 0.00048, 7.5g+iron FDR = 0.042) (Figure 4D).

FIGURE 3.

Family-level microbiota composition at baseline in Kenyan infants. Heatmap representing the relative of the 25 most abundant bacterial families detected in each baseline stool sample combined with UPGMA dendrograms based on Bray–Curtis distances.

FIGURE 4.

Effect of intervention on gut microbiota (16S rRNA gene sequencing). (A) Comparison of alpha diversity (genus level) among the 3 study arms using Shannon diversity indexes at baseline samples and end point samples. Violin plots include the median, 95% CI, IQR, and density plot where the width of the plots indicate frequency. A significant difference in Shannon diversity was observed in 3 g prebiotic group at the end of the intervention compared with that at the baseline in the same group (linear mixed model, P = 0.004). (B–D) Differential analysis using Deseq2 on most abundant genera to study the effect of intervention between groups: (B) Bacteroides, (C) Prevotella, (D) Lactobacillus based on 16S rRNA gene sequencing of subjects from the 3 study arms (ITT population; n = 173) between baseline and end point samples. ∗∗∗P < 0.001, ∗∗P < 0.01, ∗P < 0.05). The statistics are based on the differential expression (DESeq2) between groups after intervention using subject as a covariate.

By shotgun metagenomics, after 3 wk of intervention, there was higher species-based α-diversity in the iron group but not in prebiotic groups (Richness, P = 0.0019; Shannon, P = 0.0073) (Figure 5A,B). In addition, a lower intrasubject Bray–Curtis dissimilarity to baseline was observed in both prebiotic groups compared with that in the iron group (7.5 g+iron group, P = 0.0028; 3 g+iron group, P = 0.048) (Figure 5C). After the 3-wk intervention, using DEseq2 with subject as a covariate, there was a significantly lower abundance of some bacterial toxins-encoding genes in the 3-g+iron group compared with that in the iron group but not in the 7.5-g+iron group compared with that in the iron group (FDR < 0.1). Among the most abundant genes, the reduction of a zinc-dependent phospholipase C encoding gene was observed in the 3-g+iron group compared with that in the iron group after the 3-wk intervention (FDR = 0.005) (Figure 5D).

FIGURE 5.

Effect of intervention on gut microbiota composition (species level) and function by shotgun metagenomics. (A) Species-based alpha-diversity Shannon index (linear mixed model); (B) species-based alpha-diversity Richness (linear mixed model ∗∗P < 0.01); (C) within-subject Bray–Curtis dissimilarity to baseline (Mann–Whitney test, ∗P < 0.05, ∗∗P < 0.01); (D) DESeq2 normalized abundance of Zn-DEP-PLCP toxin gene before and after intervention (log count). ∗∗FDR < 0.01 (interaction between groups).

Discussion

The main findings of this study in anemic Kenyan infants fed iron-fortified infant cereal are as follows: 1) addition of scGOS/lcFOS increased FIA both from single meals before (+26%) and after 3 wk of feeding (+60%), and 2) feeding of the prebiotic+iron-fortified cereal for 3 wk led to a 41% increase in FIA in test meals given without scGOS/lcFOS, indicating a conditioning effect; and 3) the prebiotic effect on FIA was independent of the prebiotic dose. After 3 wk of the intervention, the following were observed: 4) there were no significant group differences in total Bifidobacterium spp.; 5) Lactobacillus sp. abundances were higher in both prebiotic+iron groups compared with those in the iron group; 5) Enterobacteriaceae and the summed variable of all pathogens were lower in the 7.5-g prebiotic+iron group than those in the iron group; 6) the abundance of specific bacterial toxins-encoding genes were lower in the 3-g prebiotic+iron group than those in the iron group, 7) intrasubject gut microbiome dissimilarity (distance) to baseline was higher in the iron group than that in the prebiotic+iron groups; and 8) fecal pH and gut inflammation were lower in the 7.5-g prebiotic+iron group than those in the iron group.

In this study at baseline, addition of scGOS/lcFOS to the infant cereal, regardless of dose, increased iron absorption by 26%. Our findings differ somewhat from a previous study in Kenyan infants fed maize meals fortified with 5 mg iron as FeFum + NaFeEDTA, or FeSO4, without or with 7.5 g GOS, which showed a nonsignificant increase in iron absorption with GOS [19]. However, our findings are consistent with previous single-meal iron absorption studies in iron-depleted Swiss women, where addition of 7 and 15 g GOS to a test meal containing 14 mg iron as FeFum increased iron absorption by 26% and 28%, respectively [18,33] and 15 g GOS or FOS given with a 100-mg iron supplement as FeFum increased iron absorption by 40%–50% [34]. Prebiotics may enhance iron absorption when given in the same test meal by increasing iron solubility through chelation or iron reduction and/or might increase gastric residence time allowing for greater iron dissolution [17,35].

Notably, in our study, iron absorption from the iron-fortified infant cereal consumed without prebiotics after 3 wk of prebiotic feeding was 41% higher than absorption from the infant cereal without prebiotics before the intervention. Our findings agree with a previous study in Kenyan infants who consumed MNP containing 5 mg iron and 7.5 g GOS or the same MNP without GOS daily for 3 wk: GOS consumption increased iron absorption by 62% [16]. However, that study could not distinguish a conditioning effect because the test meal fed after 3 wk also contained added GOS [16]. In this study, our design, which included test meals with and without scGOS/lcFOS after 3 wk of feeding, demonstrated that conditioning of the infant gut by feeding prebiotics increases iron absorption, independent of the acute enhancing effect when these components are coadministered in the same meal. Several mechanisms might explain this conditioning effect [17]. First, prebiotics might increase short-chain fatty acid and lactic acid production by gut commensal bacteria, thereby decreasing distal gut luminal pH and increasing iron dissolution and absorption [36,37], although colonic iron absorption in humans is believed to be minimal. Second, animal studies have shown that prebiotics can stimulate proliferation of enterocytes creating a greater surface for iron absorption and/or increase gene expression of enterocyte proteins involved in iron absorption [[38], [39], [40]]. Third, in this study, Lactobacillus sp. abundance increased significantly in both prebiotic groups compared with that in the iron group, and in mice, microbiota metabolites, particularly from Lactobacillus species, may modulate iron absorption [41]. Finally, prebiotics may facilitate iron absorption by reduction of gut and/or systemic inflammation and may increase Hb concentration [38,[42], [43], [44], [45], [46]].

Previous studies in Ivorian schoolchildren [7] and Kenyan infants [[8], [9], [10]] have reported that iron fortificants can decrease gut Bifidobacteriaceae and Lactobacillaceae and increase Enterobacteriaceae, enteropathogens, and gut inflammation. Similar effects have also been seen in healthy, iron-sufficient Swedish infants, where supplemental iron drops lowered abundance of Lactobacillus spp. and increased abundances of Clostridium and Bacteroides spp. [47]. However, not all studies agree [11,12]. Our findings of a protective effect of prebiotics on iron-induced dysbiosis are similar to a previous 4-mo controlled intervention trial in Kenyan infants provided iron in MNPs [9]. In this study, in the prebiotic+iron groups compared with the iron group, fecal pH and gut inflammation were lower, Lactobacillus ap. abundances were higher, whereas abundances of Enterobacteriaceae and the summed variable of all pathogens were lower. Moreover, the abundance of specific bacterial toxins-encoding genes was lower in the 3-g prebiotic+iron group than that in the iron group. A previous controlled study in Dutch infants reported that 6 wk of consuming infant formula containing GOS/FOS resulted in greater abundance of bifidobacteria, lower fecal pH and an increased ratio of fecal acetate to propionate [21]. Overall, our findings confirm and extend previous work in this area and show that a mixture of scGOS/lcFOS added to an iron-fortified infant cereal can mitigate many of the adverse effects of iron fortification on the infant gut microbiome.

Strengths of this study include the following: 1) a large sample size of mostly anemic African infants, a key target population; 2) use of erythrocyte incorporation of stable iron isotopes to quantify iron absorption from multiple test meals; 3) a study design that distinguished the acute and conditioning effects of prebiotics on iron absorption; and 4) rigorous analyses of fecal gut microbiota by qPCR, 16S rRNA gene, and shotgun sequencing. Limitations of the study include the following: 1) a relatively short intervention period that did not allow us to assess potential effects of prebiotics on iron status or Hb; 2) we did not directly supervise consumption of the infant cereals or collection of the stool samples into the OMNIgene tubes in the home but rather depended on the mother for these, therefore we cannot ascertain the effect of these factors on variability of the fecal microbial profiles; 3) for the primary outcome of FIA, we had a single study group (the pooled intervention groups consuming fortified cereal, with 2 different doses of prebiotics), and because we did not measure FIA in the iron-only group before and after the intervention, we cannot rule out that other mechanisms could potentially explain the beneficial effect of feeding the prebiotic-containing infant cereal for 3 wk on iron absorption; 4) during the intervention, there was significant increase of genus-based α-diversity in the 3-g+iron group compared with that in baseline, but not in the 7.5-g+iron group, which is counterintuitive; and iv) we had a fairly high dropout rate of 18%–20% in the absorption study.

Although providing additional iron to infants and children in low-income countries can reduce anemia, it may also increase diarrheal disease: a systematic review found a 15% increased risk of diarrhea (relative risk, 1.15; 95% CI: 1.06, 1.26) when iron was provided at ≥80% of the WHO recommended dietary allowance [13]. Thus, iron strategies should provide the lowest effective iron dose that minimizes adverse effects on the infant gut, and iron absorption be maximized. To our knowledge, this study demonstrated for the first time that not only a single dose of prebiotics but also previous conditioning of the infant gut with prebiotics facilitates increased iron absorption. This effect may be important in that it suggests that overall dietary iron absorption in infancy might be increased by provision of prebiotics. In addition, we confirm earlier findings describing prebiotic mitigation of adverse iron effects from MNPs on the African infant gut and extend these findings to iron-fortified infant cereals, which are provided to many infants worldwide.

Acknowledgments

We thank C Zeder, A Krzystek, T Christ, A Minder, S Kobel (ETH Zurich, Switzerland), and J Erhardt (Willstaett, Germany) for supporting the laboratory and data analysis. We thank M Bergeonneau (Danone Nutricia Africa & Overseas, Limonest, France) for product development; R Berends, A Botma, and L Kaptein for clinical study support; and S Tims, H de Weerd, T van Eijndthoven, E Balder, and A Kakourou (Danone Nutricia Research, Utrecht, Netherlands) for computational analyses, database management, and statistics. We thank R Fristedt and AS Sandberg (Chalmers University, Sweden) for performing phytic acid measurements in the infant cereal products.

Author contributions

The authors’ responsibilities were as follows – MBZ, RBS, NM, MAU, NUS: designed the study, NM, MAU, NUS, SK, EW, MBZ: conducted the study; NM, MAU, NUS, GR, IK, GP, EC, MD: performed the data analyses, all authors: participated in the data interpretation; NM, MAU, RBS, MBZ: wrote the first draft of the manuscript; and all authors: edited the manuscript and read and approved the final version of the manuscript.

Conflict of interest

MD, IK, GR, and RBS are Danone Nutricia Research employees. Other authors have no conflict of interest.

Funding

This study was funded by Danone Research and the ETH Zurich, Switzerland.

Data availability statement

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

Appendix A. Supplementary data

The following is the Supplementary data to this article: