Regional Variations In Human Gut Microbiome And Its Impact On Disease Models

BackgroundDysbiosis,departure of the human microbiome from the healthy state,is a powerful biomarker for disease incidence and progression.The gut microbiota has potential for diagnosing and predicting disease,but models trained in one geographic region may not work worldwide.Generalization of healthy baseline and disease models among different locations in a smaller spatial scales is rarely evaluated.ObjectivesTo study the relationship between human gut microbiome and metadata of hosts,we conduct the Guangdong Gut Microbiome Project.Also,we can find out the relationship between location and human gut microbiome through the rigorous design of GGMP.Methods1.We randomly choose,using PPS sampling method,7009 samples among 14 cities in Guangdong.We collected stool samples,which process by Illumina Hiseq 2500 sequencing,as well as host metadata by questionnaire and blood and physical test.2.We use PERMANOVA,PCoA and random forest to study the relationship between host metadata and gut microbiome.3.We study the interpolation and the extrapolation of disease models by using machine learning model to integrate microbiome features.Results1.In this study,the gut microbiome consist mostly of Bacteroidetes,Firmicutes and Proteobacteria.Gut microbiome variation and host phenotypes are widely correlated and regional variation explained the most.2.Gut microbiome dysbiosis has been observed in metabolic disease including metabolic syndrome,diabetes,fatty liver and obesity,which characterized with increase of Proteobacteria and decline of Bacteroidetes.Almost all the disease-related OTUs are regionally varied.The regional variations are greater than health-disease variations in these disease-related OTUs.3.Gut microbiome can diagnose metabolic disease with Machine Learning model.The average AUC of metabolic disease of one district are 70.5%(67.1%to 75.9%),73.8%(62.8%to 84.6%),74.2%(67.6%to 84.3%),72.6%(66.3%to 77.0%)in metabolic syndrome,diabetes,fatty liver and obesity respectively.But due to the regional variation,the applying AUCs of these four disease decrease to 56.2%(51.3%to 58.1%),52.1%(50.3%to 54.5%),55.8%(51.3%to 59.1%),56.9%(52.1%to 59.5%).4.A difficulty gradient was implied by this analysis,in that interpolating within a smaller scale yields much better results than extrapolating,and interpolating within a larger scale yields results of intermediate quality.Conclusions1.The human gut microbiome is widely associated with host phenotypes.In this study,we find geographical distribution play an important role in the gut microbiome association.2.The impact of regional variance on gut microbiome may surpass the impact of disease between different locations.And the cross-applying of different disease models should assess the regional variation.3.Our data reinforce the necessity of building localized reference baselines for gut microbiota with consistent sampling protocols,and applicability of disease models to new populations must be tested explicitly rather than assumed.