When Plaquing Is Not Possible: Computational Methods for Detecting Induced Phages

Abstract

High-throughput sequencing of microbial communities has uncovered a large, diverse population of phages. Frequently, phages found are integrated into their bacterial host genome. Distinguishing between phages in their integrated (lysogenic) and unintegrated (lytic) stage can provide insight into how phages shape bacterial communities. Here we present the Prophage Induction Estimator (PIE) to identify induced phages in genomic and metagenomic sequences. PIE takes raw sequencing reads and phage sequence predictions, performs read quality control, read assembly, and calculation of phage and non-phage sequence abundance and completeness. The distribution of abundances for non-phage sequences is used to predict induced phages with statistical confidence. In silico tests were conducted to benchmark this tool finding that PIE can detect induction events as well as phages with a relatively small burst size (10×). We then examined isolate genome sequencing data as well as a mock community and urinary metagenome data sets and found instances of induced phages in all three data sets. The flexibility of this software enables users to easily include phage predictions from their preferred tool of choice or phage sequences of interest. Thus, genomic and metagenomic sequencing now not only provides a means for discovering and identifying phage sequences but also the detection of induced prophages.

Keywords: genomics; induction; metagenomics; prophage; temperate phages.

Figure 1

Schematic of PIE software. Required input files include raw sequencing reads (paired-end or single-end) and prophage sequence predictions (FASTA format file). Raw reads are trimmed and assembled (❶) and categorized via BLAST as prophage or bacterial in origin (❷). After categorization, sequence coverage is computed for both the prophage sequences (❸) and bacterial sequences (❹). For the prophage sequences, an additional metric is computed: evenness. Prophage sequences that do not have an evenness ≥90% are not considered further (red stop light). Prophage sequences meeting the evenness threshold (green stop light) are compared with the threshold for the distribution of bacterial contig coverage values (shown as 99% in the figure). Prophage sequences with a coverage exceeding the threshold are called as induced phages (“√” in green box); prophage sequences that do not meet this threshold are not called as induced phages (“X” in red circle).

Figure 2

Induced prophages identified for a single bacterium (A) and mock communities (B–D). In the first test (A), a single prophage sequence with 100× genome coverage was added to bacterial reads producing a PtoH ratio of 10,000:1, 1000:1, 100:1, 10:1, and 1:1. (B) The same prophage sequence used in our prior test, again represented by reads at a 100× genome coverage, were added to reads from seven bacterial strains; each strain independently was sampled to produce the 5 different PtoH ratios tested. (C) A total of 10 prophage sequences and the seven bacterial genomes were examined at five different PtoH ratios. (D) 34 prophage sequences and the seven bacterial genomes were examined at five different PtoH ratios. For all charts, if the prophage(s) were identified and met/exceeded the 99% threshold, they are shown in orange. If they did not meet/exceed the threshold, they are represented by green.

Figure 3

Density distribution of the coverage of the bacterial contigs from the (A) mixed community and (B) urine metagenome. The bacterial contig coverage values are shown by the density plot, where the width of the plot corresponds to the number of contigs with that coverage. The 99% threshold is shown by the red line. Predicted prophage sequences exceeding this threshold are shown as red circles, while predicted prophage sequences that do not meet this threshold are shown in blue circles. Density plots, such as those shown here, are automatically produced by PIE.