Streamlining CRISPR spacer-based bacterial host predictions to decipher the viral dark matter

Abstract

Thousands of new phages have recently been discovered thanks to viral metagenomics. These phages are extremely diverse and their genome sequences often do not resemble any known phages. To appreciate their ecological impact, it is important to determine their bacterial hosts. CRISPR spacers can be used to predict hosts of unknown phages, as spacers represent biological records of past phage–bacteria interactions. However, no guidelines have been established to standardize host prediction based on CRISPR spacers. Additionally, there are no tools that use spacers to perform host predictions on large viral datasets. Here, we developed a set of tools that includes all the necessary steps for predicting the hosts of uncharacterized phages. We created a database of >11 million spacers and a program to execute host predictions on large viral datasets. Our host prediction approach uses biological criteria inspired by how CRISPR–Cas naturally work as adaptive immune systems, which make the results easy to interpret. We evaluated the performance using 9484 phages with known hosts and obtained a recall of 49% and a precision of 69%. We also found that this host prediction method yielded higher performance for phages that infect gut-associated bacteria, suggesting it is well suited for gut-virome characterization.

A catalog of tens of thousands of viruses from human metagenomes reveals hidden associations with chronic diseases

“…Matching bacterial CRISPR spacer sequences to phage genomes is one way to determine whether a bacterial lineage has previously been exposed to a particular phage. Advances in cataloging of CRISPR spacers from bacterial genomes and optimization of phage/host matching pipelines allowed the association of most of the phages discovered in this project to bacterial hosts ( http://crispr.genome.ulaval.ca/ ) ( 63 ). Specifically, 31,259 of the 45,033 virus sequences had at least one CRISPR spacer match from a known bacterium or multiple bacteria, with 369,465 total spacers matched to unique loci in CHVD sequences ( Dataset S2 ).…”

“…Bacteria-Encoded CRISPR Spacer Analysis. CrisprOpenDB (https://github.com/ edzuf/CrisprOpenDB) was used (commit 04e4ffcc55d65cf8e13afe55e081-b14773a6bb70) to assign phages to hosts based on CRISPR spacer match using BLASTN (63). Three mismatches were allowed for hits.…”

Prophages in the infant gut are largely induced, and may be functionally relevant to their hosts

“…Bacteroides and Bifidobacterium are well established as key members of the infant gut microbiota at one year of age in our samples [14] as well as in others [3,4,78], and so it would make sense for a large proportion of the prophages to infect these bacteria. The high proportion of Salmonella predictions may be an overestimation due to the tool used, which due to the high number of Salmonella spacers available may struggle to distinguish between Escherichia and Salmonella as predicted hosts and lead to an overprediction in the amount of Salmonella-infecting phages [63]. It is therefore important to remember that these are only predictions and whilst they largely make sense in the given environment there is no experimental evidence for hosts of these novel phages.…”

“…Bacterial hosts of the phages were predicted using CrisprOpenDB (v1.0) [63] with 1 mismatch allowed, and the host with the most prophages predicted to infect them were identified in R and the host with more than 1% of the prophages predicted to infect them were visualised.…”

Hundreds of viral families in the healthy infant gut

“…Bacterial hosts for the vOTUs were predicted using 318k CRISPR spacers from our metagenome assembled genomes 30 , CRISPRopenDB 32 and WIsH 33 , and the three predictions were integrated by taking the last common ancestor (LCA). 63% of the vOTUs yielded host predictions at the bacterial genus-level, while 77% were covered at the bacterial order-level (Figure 2A) and 79% at the phylum level.…”