Single Cell Transcriptomics Reveals the Hidden Microbiomes of Human Tissues

Abstract

The human microbiome has been studied extensively across sites in the body that are readily accessible to sampling. Internal organs and tissues, however, have remained largely unexplored and, in the absence of infectious disease, are widely assumed to be free of microorganisms. Using single-cell transcriptomic data from the Tabula Sapiens spanning 15 human organ donors, 20 tissues, 400,000+ annotated cells, 100+ cell types, and ~70 billion sequences, we created an atlas of the healthy human tissue microbiome with cell type resolution. In order to construct this atlas, we developed a computational pipeline which identifies high-confidence microbial sequences from three domains of life within the sea of human transcripts in individual cells, and show that this is a non-trivial search for one in a million sequences. Together with data from a separate validation cohort of 8 additional donors, we identify more than a thousand diverse bacterial, viral and fungal species in human tissues. In characterizing the human tissue microbiome, we demonstrate that individuals harbor unique species associated not just with each tissue but often with particular cell types, and that many of these species remain uncharacterized. Moreover, combining our data with the Human Microbiome Project, the tumor microbiome dataset by Nejman et al., and the PATRIC database, we map the likely microbial flow routes from external-facing microbiomes to internal tissues and tumors, and show the existence of many unexpected routes taken by both commensals and pathogens. We find that ~30% of bacterial species found in tumors are detectable across healthy tissues, suggesting that tumor microbiomes are in part sourced from healthy tissues, even those from distant sites. Increasing the resolution of sampling from tissues to cell types, we quantify the microbial load and diversity across different cell types to reveal a network of host cell type and microbe associations. For example, we identified traces of both latent and active Epstein Barr Virus infections in various cell types such as splenic plasma cells. Broad exploration of the healthy tissue microbiome may provide insights which ultimately are of clinical importance.

Unsupervised reference-free inference reveals unrecognized regulated transcriptomic complexity in human single cells

“…65.99% of reads assigned to Rfam-annotated compactors, Prokaryotes (2.1M reads; 20.64%), Archaea (871K reads; 8.28%), and Microsporidia (432K reads; 4.11%). Some detected rRNA could represent contamination or microbiome composition, as has also been reported by a recent microbial analysis of human single cells (Mahmoudabadi, Tabula Sapiens Consortium, and Quake, n.d.) (Supplement). The most abundant non-ribosomal noncoding RNA was U6-snRNA (28K reads; 0.26%), a small nuclear RNA and one of the core components of the spliceosome (Supplement).…”

“…Table 1). Some detected rRNA could represent contamination or microbiome composition, as has also been reported by a recent microbial analysis of human single cells (Mahmoudabadi, Tabula Sapiens Consortium, and Quake 2023). The most abundant non-ribosomal noncoding RNA was U6-snRNA (RNU6) (22K reads; 0.22% of all Rfam-annotated reads), a small nuclear RNA involved in the spliceosome.…”

cellsnake: a user-friendly tool for single-cell RNA sequencing analysis

“…Recent studies have shown that the heterogeneity in microbiota and the present cell types along with their functions are co-dependent 25 . Cell-associated microbial reads can be identified in scRNA-seq data 26 .…”

Machine learning models of human tissue microbiomes for tissue-of-origin prediction

“…As detailed in the previous study, we devoted five 10X lanes to deeply sequence ~200 negative control samples derived from reagents and instruments used during the experimental protocols 23 . While these samples were collected post tissue processing experiments, they were acquired from the same batch or from the same exact sample used in the experiments that were then stored in freezers.…”

“…Study design. The Tabula Sapiens Microbiome (TSM) dataset derived from searching through human tissue derived droplets (empty or containing cells) as previously described by Mahmoudabadi et al 23 In this study various feature tables (shown in blue) are constructed where the droplets are rows and taxa are columns. The table holds the relative abundance of each taxa in each droplet.…”