Bayesian multivariate image analysis for studying oral microbiome biogeography

点击翻译按钮获取中文摘要

Periodontitis and caries are highly prevalent oral biofilm diseases. Reducing the societal burden of these polymicrobial diseases will require a better understanding of the human-microbe superorganism and interactions among microbial species. A critical barrier in microbiology has been a near total lack of knowledge and tools to examine the spatial organization of microbial communities at the ten- to 100-micron scale. To meet this need, we recently developed an imaging technique, Combinatory Labeling and Spectral Imaging Fluorescence in Situ Hybridization (CLASI-FISH). CLASI-FISH images display the abundance of up to 28 taxa in each region of a sample while also displaying how cells of each taxon (taxonomic unit) are located relative to each other and relative to host cells. However, the quantitative methods that have been used to analyze spectral imaging data thus far are limited to describing spatial patterns of one or two taxa at a time. Moreover, they lack the ability to address challenges raised specifically by biofilm architecture, such as how to incorporate shapes (information needed to infer cell-to-cell contact), how to model spatial distributions of up to 28 taxa simultaneously, and how to combine data from multiple images. We propose three aims that address these limitations and, in doing so, advance the field of spatial statistics for the analysis of complex image data in general: 1) extend spatial statistics techniques to account for bacterial taxa’s shape and abundance in modeling joint spatial patterns; 2) develop a multivariate Bayesian log-Gaussian Cox process model that extends to multiple images and non-spatial covariates, such as host characteristics; and 3) develop a Bayesian paradigm to model and quantify corncob-like arrangements of two taxa, accounting for shapes. The core innovation proposed is to develop and apply statistical methods that go beyond analyzing measures of abundance and composition to quantify spatial relationships among microbes in biofilm images. This flexible modeling framework will allow testing of hypotheses regarding microbe-microbe interactions and associations with host characteristics. This is a fundamental shift for how such images will be analyzed, potentially providing new insights into the role of microbes in the oral cavity. To test the methods’ performance, we will perform simulation studies and compare oral biofilm image data from subjects with and without periodontitis. We will make software available for the routine application of these methods by microbiologists. We anticipate wide use of these novel methods and software, which will find broad application to other human biofilm diseases and to biogeography in general. Elucidating the spatial distribution of oral microbes is required to determine the role of biofilm in human oral health and disease. The methods we develop will lead to the identification of key bacterial interactions that may serve as novel targets for the prevention or treatment of periodontitis and other oral diseases.