Microbiotyping the Sinonasal Microbiome

Abstract

This study offers a novel description of the sinonasal microbiome, through an unsupervised machine learning approach combining dimensionality reduction and clustering. We apply our method to the International Sinonasal Microbiome Study (ISMS) dataset of 410 sinus swab samples. We propose three main sinonasal "microbiotypes" or "states": the first is Corynebacterium-dominated, the second is Staphylococcus-dominated, and the third dominated by the other core genera of the sinonasal microbiome (Streptococcus, Haemophilus, Moraxella, and Pseudomonas). The prevalence of the three microbiotypes studied did not differ between healthy and diseased sinuses, but differences in their distribution were evident based on geography. We also describe a potential reciprocal relationship between Corynebacterium species and Staphylococcus aureus, suggesting that a certain microbial equilibrium between various players is reached in the sinuses. We validate our approach by applying it to a separate 16S rRNA gene sequence dataset of 97 sinus swabs from a different patient cohort. Sinonasal microbiotyping may prove useful in reducing the complexity of describing sinonasal microbiota. It may drive future studies aimed at modeling microbial interactions in the sinuses and in doing so may facilitate the development of a tailored patient-specific approach to the treatment of sinus disease in the future.

Keywords: 16S rRNA gene; chronic rhinosinusitis; microbiome; microbiotype; next-generation sequencing; paranasal sinuses; sinus.

Figure 1

Beta diversity ordination plots. (A) Weighted UniFrac PCoA - by Diagnosis. (B) Weighted UniFrac PCoA - by Centre. (C) Jensen-Shannon PCoA - by Diagnosis; (D) Jensen-Shannon PCoA - by Centre.

Figure 2

Microbiotyping the sinonasal microbiome. (A) Illustration of the assigned microbiotypes on the Jensen-Shannon PCoA biplot. Arrows were used to depict the projection of the genera onto the PCoA matrix. Each arrow is indicated by the color of the genus according to the Legend. (B) Taxonomic composition of the three microbiotypes at the genus level. (C) Histograms demonstrating the relative abundance of Corynebacterium and Staphylococcus. (D) Subgroups of microbiotype 3 (hierarchical density-based clustering). (E) Distribution of staphylococcal species (mean relative abundance).

Figure 3

Prevalence and distribution of the microbiotypes.

Figure 4

Validation of microbiotyping approach on Dataset Two. (A) Independent K-Means clustering of Dataset Two samples using our described K-means microbiotyping approach “Unsupervised.” (B) Prediction of microbiotypes on Dataset Two samples using the K-means model fitted on the Main Dataset “Semi-supervised.” (C) Taxa composition of Dataset Two samples as per the “Unsupervised approach.” (D) Taxa Composition of the combined Main Dataset and Dataset Two samples as per the “Semi-supervised approach.”