A microbiome-directed therapeutic food for children recovering from severe acute malnutrition

Abstract

Globally, severe acute malnutrition (SAM), defined as a weight-for-length z-score more than three SDs below a reference mean (WLZ < -3), affects 14 million children under 5 years of age. Complete anthropometric recovery after standard, short-term interventions is rare, with children often left with moderate acute malnutrition (MAM; WLZ -2 to -3). We conducted a randomized controlled trial (RCT) involving 12- to 18-month-old Bangladeshi children from urban and rural sites, who, after initial hospital-based treatment for SAM, received a 3-month intervention with a microbiome-directed complementary food (MDCF-2) or a calorically more dense, standard ready-to-use supplementary food (RUSF). The rate of WLZ improvement was significantly greater in MDCF-2-treated children (P = 8.73 × 10^-3), similar to our previous RCT of Bangladeshi children with MAM without antecedent SAM (P = 0.032). A correlated meta-analysis of plasma levels of 4520 proteins in both RCTs revealed 215 positively associated with WLZ (largely representing musculoskeletal and central nervous system development) and 44 negatively associated (primarily related to immune activation). Moreover, the positively associated proteins were significantly enriched by MDCF-2 (q = 1.1 × 10^-6). Characterizing the abundances of 754 bacterial metagenome-assembled genomes in serially collected fecal samples disclosed the effects of acute rehabilitation for SAM on the microbiome and how, during treatment for MAM, specific strains of Prevotella copri function at the intersection between MDCF-2 glycan metabolism and anthropometric recovery. These results provide a rationale for further testing the generalizability of MDCF efficacy and for identifying biomarkers to define treatment responses.

Fig. 1 –. Study design and anthropometric analyses across sites and treatment groups.

(A) Study design. (B) Hospital-based acute nutritional rehabilitation protocol for SAM. (C) Accounting of participants enrolled, dropped out, flood-affected, and completing the study in Dhaka or Kurigram. Only children residing in Kurigram were affected by flooding during the trial. (D) Results of linear mixed effects modeling of the study-wide effect of MDCF-2 versus RUSF treatment on the primary outcome measure - rate of change of WLZ over time. Flood-affected participants were excluded from this analysis. The P-value corresponds to the effect of the term representing the ‘treatment group x study week’ interaction. (E) Results of modeling the primary outcome measure, plus additional anthropometric measures, across both sites (left), in Kurigram (middle), and in Dhaka (right). Each data point indicates the β-coefficient for the effect of the ‘treatment group x study week’ term on each of the indicated anthropometric measures. The upper and lower 95% confidence intervals for each coefficient estimate are indicated as error bars. The RUSF treatment group was used as the reference value, hence positive coefficients indicate increases in anthropometric measurements associated with MDCF-2 treatment.

Fig. 2 –. Correlated meta-analysis of plasma proteins.

(A) Schematic of the Correlated Meta-Analysis (CMA). Black arrows and boxes indicate a protein with a WLZ-association P-value < 0.05, whereas grey arrows indicate a P-value > 0.05. (B) Plasma protein WLZ-associations obtained from CMA, with an over-representation analysis performed on proteins positively or negatively associated with WLZ. The colored circles indicate the Gene Ontology Biological Processes, with data points (proteins) colored by their membership in each GO Biological Process term. The number in each colored circle represents the number of proteins assigned to the GO BP. (C) The 30 most significantly WLZ-associated proteins from the CMA. The columns on the right summarize the assignments of proteins to GO Biological Processes.

Fig. 3 –. WLZ association of MAGs during acute rehabilitation for SAM and subsequent treatment for MAM with MDCF-2 or RUSF.

Principal Component Analysis (PCA) was performed on variance-stabilized MAG abundances in fecal samples serially collected from participants through the course of the study. (A) Projections of fecal microbiome configurations along PC1 (representing 18.9% of total variance), separated by timepoint and study site during the SAM phase, and additionally by treatment group during the MAM phase and 1-month follow-up. Plots depict the median, first, and third quartiles; whiskers extend to the largest value within 1.5 × the interquartile range. (B) Species-level taxonomic assignments of MAG ‘drivers’ of variance, defined as the top 1% of contributors to projection along each of the first three principal components. The percent variance explained by each principal component is indicated at top. (C) Results of enrichment analysis of MAGs, ranked by the strength and direction of their association with WLZ during acute rehabilitation for SAM. The horizontal axis indicates the normalized enrichment score, whereas the size of each point represents the number of MAGs assigned to each enriched species. (D) WLZ-associated MAGs identified in the MAM phase of the study.

Fig. 4 –. Phylogenetic and functional characteristics of MAGs associated with WLZ in the SAM and MAM phases of the trial.

(A) Phylogenetic analysis of the relationships of WLZ-associated P. copri and other P. copri MAGs from the current study, P. copri MAGs identified in our prior study of similarly aged (12–18-month-old) Bangladeshi children with primary MAM, and selected P. copri isolates cultured from the study population. The black boxes in the grid denote MAGs that displayed significant WLZ-associations during the indicated phases of this or the prior MDCF-2 study (threshold cutoff, q < 0.05 in the primary-MAM study and SAM-phase of the current study, and q < 0.1 in the MAM phase the current study). Tree tip colors on the left side of the panel refer to the clade assignment of each MAG whereas the shape of the tips (square, triangle) indicate the origin of each MAG. (B) Relationship between the presence of mcSEED metabolic pathways and the WLZ associations of MAGs possessing each pathway. Points are colored based on the prevalence of each metabolic pathway across the set of 613 MAGs. Pathways whose presence or absence were significantly associated (q < 0.001) with the MAG WLZ-association coefficient (b3 in Eq. 2) are listed on the right in order of decreasing association strength. The dashed line represents the significance threshold (q < 0.05). (C) The presence or absence of mcSEED metabolic pathways in MAGs positively associated with WLZ during the MAM phase (outer four rings) compared to negatively WLZ-associated MAGs (next eight inner rings). The second innermost ring denotes the prevalence of each metabolic pathway in positively versus negatively WLZ-associated MAGs. The innermost ring indicates the strength of the association of each predicted phenotype with the WLZ association of each MAG. Only pathways with a significant β₁ coefficient are included (q < 0.1). (D) PUL conservation in WLZ-associated P. copri MAGs identified in the current and prior MDCF-2 trial compared to other P. copri MAGs and isolates cultured from the study population. The color-coded matrix in the center indicates the extent of conservation each PUL in each MAG relative to Bg0019 (the MAG with the strongest WLZ association in the fecal microbiomes of participants in the primary MAM trial). The known or predicted polysaccharide substrates of these PULs are described (see table S16B for details of PUL annotation and numbering convention).