Extracellular vesicles (EVs), nanoparticles secreted by both gram-negative and gram-positive bacteria, carry various biomolecules and cross biological barriers. Gut microbiota-derived EVs are currently being investigated as a communication mechanism between the microbiota and the host. Few clinical studies, however, have investigated gut microbiota-derived EVs. Here, we show that machine learning models were able to accurately distinguish gut microbiota and respective microbiota-derived EV samples according to their taxonomic composition both within each data set (area under the curve [AUC] 0.764-1.00) and in a cross-study setting (AUC 0.701-0.997). These results show that gut microbiota-derived EVs form a distinct taxonomic entity from gut microbiota. Thus, conventional gut microbiota composition may not correctly reflect communication between the gut microbiota and the host unless microbiota-derived EVs are reported separately.IMPORTANCEGut microbiota-derived extracellular vesicles (EVs) have been suggested to be a communication mechanism between the gut microbiota and the human body. However, the data on EV secretion from the gut microbiota remain limited. To investigate and compare the composition of gut microbiota-derived EVs to gut microbiota composition, we used a machine learning approach to classify 16S rRNA gene sequencing data in seven clinical data sets incorporating both gut microbiota and gut microbiota-derived EV samples. The results of the study show that microbiota-derived EVs form a separate taxonomic entity from the gut microbiota. Gut microbiota-derived EVs should be included in clinical studies that investigate gut microbiota to gain more comprehensive insight into gut microbiota-host communication.

Keywords: 16S RNA; bacteria; extracellular vesicles; gut; gut microbiome; gut microbiota.