Bioactive glycans in a microbiome-directed food for children with malnutrition

Childhood undernutrition remains a major global health challenge, with 149 million children under 5 stunted and 45 million wasted in 2020. Gut microbiome development typically matures by the end of the second postnatal year; in malnourished children, this maturation is perturbed and resembles younger-age microbiota, with associated metabolic immaturity. Prior gnotobiotic animal studies indicate that microbiota from malnourished children can transmit impaired growth and metabolic abnormalities. The research question is whether a microbiome-directed complementary food (MDCF-2) exerts superior effects on ponderal growth by targeting specific gut microbial taxa and functions, and which microbial pathways and glycans mediate this response. The purpose of this study was to identify MAGs associated with WLZ, determine treatment-induced expression changes in metabolic pathways, quantify glycans in MDCF-2 versus a conventional RUSF, and link microbial glycan utilization to clinical growth outcomes in Bangladeshi children with MAM.

Previous studies have shown: (1) gut microbiomes from malnourished children can transmit growth impairments to gnotobiotic animals; (2) microbiome-directed foods can improve outcomes in animals and undernourished children; (3) community development trajectories are well-characterized in early life, with specific taxa like Bifidobacterium decreasing during complementary feeding; and (4) polysaccharide-utilization loci (PULs) in Bacteroidota, including Prevotella copri, govern glycan detection and metabolism. These findings underpin the rationale for using MDCF-2 to repair microbiota immaturity and guide exploration of specific glycan-microbe interactions relevant to growth.

Design: A 3-month randomized, controlled feeding study in Mirpur, Dhaka, Bangladesh enrolled 12–18-month-old children with moderate acute malnutrition (MAM; WLZ −2 to −3). Participants received twice-daily supplementation with either MDCF-2 or RUSF (~220–250 kcal/day). Fifty-nine children in each arm completed the intervention plus a 1-month follow-up. Fecal samples were collected every 10 days during the first month and every 4 weeks thereafter; blood and anthropometry were collected per schedule. Intake of supplements, meal frequency, acceptable diet, and breastmilk consumption did not differ between arms. Metagenomics: DNA from 942 fecal samples (7–8 per participant) underwent short-read shotgun sequencing; a subset (n=15, upper quartile WLZ responders to MDCF-2, t=0 and 3 months) included long-read sequencing. Participant-wise assemblies were generated and contigs binned into MAGs. Quality filters yielded 1,000 high-quality MAGs (≥90% complete, ≤5% contamination) covering ~66% of reads. Taxonomy was assigned using GTDB-Tk. Abundances were computed for 707 fecal samples spanning baseline through 1-month post-intervention. Statistical modeling: From 837 MAGs passing abundance/prevalence filters, linear mixed-effects models identified MAGs associated with WLZ across time (β(MAG), FDR q<0.05), yielding 222 WLZ-associated MAGs (75 positive, 147 negative). Additional models assessed treatment group, study week, and their interaction on MAG abundance trajectories; GSEA tested enrichment of WLZ-associated MAGs among those changing faster per treatment. Metabolic reconstruction: Using a subsystems approach (mcSEED), 199,334 proteins across MAGs were annotated into 1,308 non-redundant functions and presence/absence predictions for 106 metabolic pathways were inferred, generating MAG metabolic phenotypes. GSEA assessed pathway enrichment among WLZ-associated MAGs and treatment-responsive changes. Glycomics of foods: UHPLC-QqQ-MS quantified 14 monosaccharides and 49 glycosidic linkages; polysaccharide content was defined by nonenzymatic cleavage to oligosaccharides followed by structural characterization. Ingredients analyzed: MDCF-2 (chickpea flour, soybean flour, peanut paste, mashed green banana pulp) and RUSF (rice, lentil, milk powder). Comparative analyses identified differential monosaccharides, linkages, and polysaccharide classes (e.g., galactans, mannans, starch, cellulose). Metatranscriptomics: Microbial RNA-seq on 350 fecal samples from baseline, 1 month, and 3 months. Reads were mapped to MAGs; filtering removed low-contribution MAGs (benchmarked via simulated data). PCA assessed variance structure (DNA-based abundance vs RNA-based transcriptome). Differential expression used negative binomial generalized linear models ranking transcripts by treatment, time, WLZ-response quartiles, and their interactions; GSEA identified enriched pathways and leading-edge transcripts. PUL analysis and isolates: PULs were identified in 11 P. copri MAGs; conservation relative to MAG Bg0019 was assessed as functionally conserved, structurally distinct, or absent. Correlation between PUL conservation and WLZ association was tested. Six P. copri isolates from Bangladeshi children representing varied PUL repertoires were selected; five were used in defined-medium growth assays with purified glycans representative of MDCF-2 (e.g., sugar beet arabinan, wheat arabinoxylan, barley β-glucan, potato galactan, carob galactomannan, soybean rhamnogalacturonan, tamarind xyloglucan, beechwood xylan; chondroitin sulfate control). OD600 tracked growth; MS quantified monosaccharide consumption. Fecal glycosidic linkages: UHPLC-QqQ-MS measured 49 linkages in feces at baseline and 3 months from MDCF-2 upper vs lower WLZ-response quartiles (n=30). Linear mixed-effects models compared changes over time by WLZ-response quartile (q<0.05). Food frequency questionnaires assessed background diet correlations with fecal linkages. Ancillary: PCA enrichment of taxa via GSEA; evaluation of CAZyme transcripts from key PULs; linkage of clinical WLZ changes with microbial and glycomic metrics.

• MDCF-2 improved WLZ rate of change compared with RUSF despite 15% lower caloric density (reported in prior clinical outcomes; current study links mechanisms).
• 1,000 high-quality MAGs reconstructed; 837 analyzed for abundance/prevalence; 222 MAGs significantly associated with WLZ over time (75 positive, 147 negative; q<0.05). Positively associated genera included Agathobacter, Blautia, Faecalibacterium, Prevotella; negatively associated included Bacteroides, Bifidobacterium, Streptococcus.
• Treatment dynamics: MAGs increasing faster with MDCF-2 were significantly enriched for those positively associated with WLZ (q=3.41×10−3, GSEA). MAGs with higher mean abundance or increasing faster with RUSF were enriched for negatively WLZ-associated MAGs (q=1.57×10−4 and q=3.41×10−3).
• Pathway-level enrichment: Carbohydrate utilization pathways predominated among WLZ-associated and MDCF-2-responsive functions (P=0.006, Fisher’s test). In metatranscriptomes, MDCF-2 was associated with enrichment of α-arabinooligosaccharide (αAOS), arabinose, fucose utilization, plus de novo synthesis of arginine, glutamine, lysine, and folate (q<0.1, GSEA). No pathways were significantly enriched with RUSF.
• Two P. copri MAGs (Bg0018, Bg0019), the only P. copri positively associated with WLZ, contributed 11/14 leading-edge αAOS transcripts and the majority of leading-edge transcripts across enriched carbohydrate pathways; over half (67/99; 68%) of leading-edge transcripts in upper-quartile WLZ responders derived from these two MAGs. These MAGs belong to clade A.
• PUL conservation: Ten PULs shared between Bg0018 and Bg0019 (7 conserved, 3 structurally distinct) varied among the other nine P. copri MAGs. Degree of PUL conservation correlated with strength of WLZ association across P. copri MAGs (Pearson r = −0.79 between Euclidean distance from Bg0019 PUL profile and β(WLZ); P=0.0035). Five of the seven conserved PULs target mannan/galactan, enriched in MDCF-2.
• Food glycomics: Compared to RUSF, MDCF-2 had significantly more L-arabinose, D-xylose, L-fucose, D-mannose, D-galacturonic acid (P<0.05), and higher abundance of galactans and mannans (P<0.05), whereas RUSF had more starch and cellulose (P<0.05). Arabinan structures differed by ingredient sources.
• Isolate phenotypes: P. copri isolates whose PUL profiles most closely matched Bg0018/Bg0019 (e.g., BgF5_2, Bg2C6, Bg2H3) showed strongest growth on MDCF-2-representative glycans enriched/unique to MDCF-2 (arabinans, galactans, mannans). Growth phenotypes aligned with predicted PUL substrate specificities and CAZyme repertoires.
• Fecal glycosidic linkages: Fourteen linkages increased significantly more over time in upper- vs lower-quartile WLZ responders (q<0.05); none increased more in lower-quartile responders. These linkages are present in MDCF-2 and are consistent with predicted products of CAZymes in Bg0018/Bg0019 PULs (e.g., 4,6-mannose from soybean galactomannan cleavage by PUL7/PUL8 endo-1,4-β-mannosidases). CAZyme transcripts from key PULs (e.g., GH51, GH26, GH26|GH5_4, GH130, CE7) were detectable and modestly higher in upper-quartile responders.
• Background diet: Consumption of legumes and nuts correlated positively with fecal levels of several linkages (e.g., t-Araf, 5-Araf, 2,3-Araf, t-GalA, 2,4,6-glucose), and distinguished upper- from lower-quartile responders, suggesting background diet can augment MDCF-2 effects.
• Overall, results identify two WLZ-associated P. copri strains as principal mediators of MDCF-2 glycan metabolism, linking microbial carbohydrate processing to improved ponderal growth.

Findings support a model in which MDCF-2, enriched in specific plant glycans (galactans, mannans, arabinans), promotes fitness and glycan-metabolic activity of key growth-associated taxa, notably P. copri clade A strains possessing conserved PULs aligned to these glycans. Transcriptomic enrichment of αAOS, arabinose, and fucose utilization pathways with MDCF-2, and the dominance of P. copri leading-edge transcripts, indicate targeted microbial responses. The correlation between P. copri PUL conservation and WLZ association suggests that specific strain-level genomic features underlie beneficial outcomes. Increased fecal linkages in upper-quartile responders and isolate growth phenotypes on MDCF-2-representative glycans further connect dietary glycans, microbial metabolism, and clinical response. Background diet (legumes/nuts) appears to amplify these effects. Collectively, the results address the research question by identifying microbial mechanisms and taxa that respond to MDCF-2 and relate to improved WLZ, highlighting the importance of strain-level specificity and glycan composition in microbiome-directed foods.

This study delineates the microbial targets and glycan-mediated mechanisms underlying the superior weight gain observed with MDCF-2 compared to RUSF in Bangladeshi children with MAM. Two P. copri clade A MAGs with conserved PULs for mannan/galactan/arabinans are principal contributors to MDCF-2-induced carbohydrate utilization, and their PUL conservation correlates with positive WLZ association. MDCF-2 is enriched in bioactive glycans that engage these taxa, and fecal glycosidic linkage profiles reflect enhanced glycan processing in higher WLZ responders. These insights can guide the optimization of microbiome-directed foods by identifying bioactive glycan structures and effector taxa, and inform development of synbiotic strategies (e.g., pairing P. copri-like strains with specific glycans). Future directions include controlled clinical trials testing probiotic/synbiotic formulations with WLZ-associated strains, dose optimization of MDCF-2, detailed dietary intake quantification, and reverse-translation gnotobiotic models to dissect causal pathways and host-microbe metabolic interactions.

Causality between P. copri glycan metabolism and improved ponderal growth is not formally established. Background diet variability may confound fecal glycan signals and microbial responses. Contributions of other community members and non-carbohydrate nutrients to weight gain remain unclear. Some transcriptomic differences were modest and did not meet strict significance thresholds. Strain-level inferences from MAGs have inherent limitations, and not all MAGs had cultured representatives. The study duration was limited to 3 months with 1-month follow-up; longer-term growth and developmental outcomes were not assessed.