In BMJ open
OBJECTIVE : The human gut microbiota plays important roles in human health but is also known to be highly diverse between populations from different regions. Yet most studies inadequately account for this regional diversity in their analyses. This study examines the extent to which geographical variation can act as a confounding variable for studies that associate the microbiota with human phenotypic variation.
DESIGN : Population-based study.
SETTING : China.
PARTICIPANTS : 2164 participants from 15 province-level divisions in China.
PRIMARY AND SECONDARY OUTCOME MEASURES : We analysed the impact of geographic location on associations between the human gut microbiota and 72 host factors representing a wide variety of environmental-level, household-level and individual-level factors.
RESULTS : While the gut microbiota varied across a wide range of host factors including urbanisation, occupation and dietary variables, the geographic region (province/megacity) of the participants explained the largest proportion of the variance (17.9%). The estimated effect sizes for other host factors varied substantially by region with little evidence of a reproducible signal across different areas as measured by permutational multivariate analysis of variance and random forest models.
CONCLUSIONS : Our results suggest that geographic variation is an essential factor that should be explicitly considered when generalising microbiota-based models to host phenotype across different populations.
Sun Shan, Wang Huijun, Tsilimigras Matthew Cb, Howard Annie Green, Sha Wei, Zhang Jiguo, Su Chang, Wang Zhihong, Du Shufa, Sioda Michael, Fouladi Farnaz, Fodor Anthony, Gordon-Larsen Penny, Zhang Bing
Gut microbiota, geographic differences, machine learning, microbiota-host associations