Background Pneumonia remains the leading cause of infectious mortality in children under five, with the highest burden in sub-Saharan Africa. Dysbiosis in nasopharyngeal (NP) microbiota may influence pneumonia susceptibility and progression, but little is known about its composition or clinical relevance in low- and middle-income countries (LMICs). We characterized the NP microbiota of children hospitalized with severe pneumonia in East Africa and investigated associations with clinical outcomes.

Methods We performed 16S rRNA partial gene sequencing of NP swabs collected at hospital admission from 876 children enrolled in the COAST trial across five sites in Kenya and Uganda. Clinical, demographic, and virological data were prospectively collected. Microbial profiles were analysed using hierarchical clustering, non-metric multidimensional scaling (NMDS), and multivariable regression to assess associations with respiratory viral infections, sepsis, cyanosis, bacteraemia, coma, HIV status, malnutrition, sickle cell disease (SCD), malaria, and mortality.

Findings The nasopharyngeal microbiome was structured in six distinct clusters, each dominated by different genera, including Staphylococcus, Streptococcus, Haemophilus, Dolosigranulum, Corynebacterium, and Moraxella. NMDS revealed significant alignment between different microbiome clusters and key clinical outcomes: clusters dominated by Corynebacterium and Dolosigranulum were directionally associated with mortality (p < 0.001). Notably, Corynebacterium abundance was elevated in children who died within 48 hours of admission, then declined over longer survival intervals, approaching levels observed in survivors.

Interpretation These data provide one of the largest high-resolution surveys of the paediatric upper airway microbiome in Africa and identify microbial patterns associated with viral infection, HIV status, early death and bacteraemia. The unexpected association between Corynebacterium and mortality may reflect distinct species composition in LMIC settings, warranting further investigation using species-resolved metagenomics. These findings lay the groundwork for microbiome-based risk stratification in childhood pneumonia.

The authors have declared no competing interest.

This study is part of the EDCTP2 programme (grant number RIA-2016S-1636-COAST-Nutrition) supported by the European Union, and UK Joint Global Health Trials scheme: Medical Research Council, Department for International Development Wellcome Trust Grant Number MR/L004364/1, UK.

I confirm all relevant ethical guidelines have been followed, and any necessary IRB and/or ethics committee approvals have been obtained.

Yes

The details of the IRB/oversight body that provided approval or exemption for the research described are given below:

Ethical approval was obtained from the Scientific and Ethics Review Unit of the Kenya Medical Research Institute (KEMRI) and the Uganda National Council for Science and Technology (UNCST).

I confirm that all necessary patient/participant consent has been obtained and the appropriate institutional forms have been archived, and that any patient/participant/sample identifiers included were not known to anyone (e.g., hospital staff, patients or participants themselves) outside the research group so cannot be used to identify individuals.

Yes

I understand that all clinical trials and any other prospective interventional studies must be registered with an ICMJE-approved registry, such as ClinicalTrials.gov. I confirm that any such study reported in the manuscript has been registered and the trial registration ID is provided (note: if posting a prospective study registered retrospectively, please provide a statement in the trial ID field explaining why the study was not registered in advance).

Yes

I have followed all appropriate research reporting guidelines, such as any relevant EQUATOR Network research reporting checklist(s) and other pertinent material, if applicable.

Yes

The 16S rRNA sequence data generated from MiSeq sequencing in this study is available at NCBI SRA under accession number PRJNA1285267.

All analyses were performed using R (v4.5.0). Raw paired-end reads were processed using the DADA2 pipeline (v1.22.0). Quality filtering and adapter trimming were performed using filterAndTriminterface of the fastqPairedFilter function, followed by error model learning with learnErrors and denoising via dada function. Forward and reverse reads were merged using mergePairs, and chimeras were removed using removeBimeraDenovo functions. Taxonomic assignment of amplicon sequence variants (ASVs) was performed using the assignTaxonomy function and the SILVA reference database (v138). The resulting ASV table was aggregated to genus level, and samples were normalized to relative abundance. Microbiome composition was assessed using hierarchical clustering of Bray–Curtis dissimilarities computed from genus-level relative abundance profiles. Clustering was performed using Ward’s method and visualised using dendrograms and stacked bar plots of the top 50 genera. Samples were assigned to six compositional clusters, identified using the NbClust R package based on the community microbiome structure. To identify dominant taxa within each cluster, average genus abundance per cluster was computed, and genus names were rendered as word clouds, scaled by mean relative abundance. To explore overall microbial structure in relation to clinical and demographic metadata, two-dimensional non-metric multidimensional scaling (NMDS) was applied to Bray–Curtis dissimilarities. The envfit function from the vegan package was used to fit clinical and virological covariates as vectors into NMDS space with 999 permutations to assess significance. Significant genus vectors were overlaid on the NMDS ordination to examine alignment of different taxa and microbial clusters with ordination gradients. Associations between individual microbial taxa and clinical or virological variable were also tested using the Maaslin2 R package (v1.21.0). Relative abundances at the genus level were used as outcomes, and different clinical features (sepsis, cyanosis, death, coma, severe malaria, HIV status, SCD, asthma history, diarrhoea, malnutrition, influenza A, RSV) were passed as fixed effects in multivariable models. Significant associations were defined using an FDR-adjusted q-value < 0.25. Results were visualised using coefficient plots (with standard errors) and annotated boxplots showing genus-level relative abundance stratified by metadata variable.