1. Metagenomic analysis of bacteriophage populations in the human gut
APC Bioinformatics Workshop 23/03/2018
Metagenomic analysis of bacteriophage populations in the human gut
Andrey Shkoporov,
Gut Phageomics Spoke 1

2. Why study gut phageome? Bacteria Phages Prevotella
How gut phages impact on gut bacteriome and host physiology?
Human gut virome is composed of 1015 viruses, >98% are bacteriophages.
6-20% of faecal DNA has bacteriophage origin.
Majority of bacterial strains contain prophages.
Little is known if and how bacteriophages regulate composition and functional repertoire of gut bacteriome.
Evidence of sterile-filtered faecal water being able to recapitulate the effect of FMT. Role of phage?
Phage transplants as alternatives to FMT? Resistant to damaging factors, easy to store, no need for anaerobic conditions.
Phages
Bacteria
Prevotella
Storage at RT
Repeated freeze-thaw cycles
Can phages be used as biomarkers of bacteriome composition?
Faecal bacteriome but not phageome composition is strongly influenced by colonic residence time.
Faecal phage composition is resistant to storage, repeated freeze-thaw cycles.
Faecal phages can serve as biomarkers of microbiota in the upper gut.

3. Gut phage metagenomics:
typical workflow
Isolation
of VLPs
Purification of
viral nucleic acids
Library
preparation
Next generation
sequencing (NGS)
Processing of reads,
assembly and filtration
of NGS data*
Alignment of reads
to closed reference DB
DB of viral
contigs
Analysis of
specific genes
Host prediction
Viral taxonomy
Alignment of reads to
the DB of
assembled contigs
Count
matrix
Statistical tests to
identify
drivers of diversity
Analysis of
α- and
β-diversity

4. Isolation of VLPs
Original faecal sample homogenated in SM buffer
Different morphologies and physical properties of viral particles impact on isolation efficiency
Centrifugation at 5 krpm (x2)
Filtration through 0.45μm pore filter (x2)
Precipitation with PEG/NaCl, extraction with chloroform
(may result in removal of enveloped viruses)
Bacterial DNA removal is never complete
Ribosomes may be co-purified with virions
Treatment with DNase and RNase

5. Purification of nucleic acids and library prep
VLP fraction
Viral genomes can be dsDNA, ssDNA, dsRNA, ssRNA, circular, linear or fragmented.
Nucleic acid yields from faecal VLP are typically low
Traditionally used Illumina kits (Truseq Nano, Nextera XT) require WGA (typically MDA) to be done.
MDA introduces significant bias
Alternative kits optimized for low input of dsDNA/ssDNA should be considered
Extraction of nucleic acids (dsDNA, ssDNA, RNA, circular or linear)
Reverse transcription, WGA
Fragmentation and library prep

6. Processing of NGS reads (Illumina)
Illumina Miseq: 2x300 bp (Nextera XT library)
Illumina Hiseq 2500: High Output 2x150 bp (TruSeq Nano or Swift Biosciences Accel-NGS kit)
Typically >1M reads per sample
Adaptor and quality trimming + filtering increases mapping rates and assembly efficiency
Raw Illumina Miseq reads
Reads after trimming and filtration
FastQC: quality control checks on raw and processed .fastq files
Kraken: fast classification of reads against human database
Cutadapt: removal of adaptors
Trimmomatic: removal of adaptors, quality trimming, cropping and filtering.

7. Assembly of NGS reads Closed reference database vs. assembled contigs
as template read mapping.
Bacteriophages are most predominant biological entities on earth and outnumber bacterial 10:1.
Despite that, bacteriophages are under-represented in sequence DBs.
Mapping of reads to viral reference DBs allows to identify <2% of real diversity in a sample.
Assembly-based approach uses same reads to build contigs, which are then used as template for read alignment.
Bacteria: 134,397 genomes of 13,537 species
Viruses: 7,490 genomes of 4,853 species
Mapping reads to the database of contigs results in recruitment of much higher fraction of reads.
“Viral dark matter” - uncharacterised component of the phageome.
Differentiation between viral and non-viral DNA can be challenging

8. Challenges associated with assembly of phage metagenomic reads
Assembly of NGS reads
atgcat
agtcta
...
Samples:
Sample n
spAdes
megahit
IDBA
Pooled samples
Pooled contigs
Nonredundant
contigs DB
BLAST contigs all-vs-all
Remove shorter contigs from pairs with >90% homology and >90% overlap
Challenges associated with assembly of phage metagenomic reads
Variable complexity of samples ( virotypes)
Highly uneven relative abundance of viruses
Assemblers tend to fragment genomes in case coverage is either to low or too high
Strain-level variations hard to capture
Best performing assemblers: spAdes, metaspAdes, Megahit. Velvet, mira, Vicuna, SOAPdenovo and Abyss performed badly.
Other considerations
Individual sample and pooled assemblies could both be used
Different assembler programs
Different fixed k-mer lengths
Random sub-setting of reads to improve assembly of highly over- represented sequences

9. Improving virome assembly through hybrid sequencing
Technologies such as single molecule real-time (SMRT) sequencing, generate long, low-coverage reads with relatively high error rates.
Hybrid sequencing is an approach which uses these long reads in conjunction with accurate short reads to generate high quality assemblies and addresses the individual failings of each sequencing technology.
As a proof of concept, we obtained improved assembly of putative viral genomes from a human gut virome using a hybrid sequencing approach.
Thomas Sutton
Bioinformatics
PhD student
PacBio Sequel

10. Removing non-viral contigs and generating count matrix
Nonredundant
contigs DB
Viral RefSeq
Virsorter
pVOGs DB
>2 hits per 10kb
>3 hits in total
Circular
Selected
viral contigs
Contigs >3 kb with no hits
to NCBI nt DB > 100 nt
“Viral dark matter”
As contamination with bacterial DNA is inevitable procedure to remove contigs of non- viral origin is employed.
Selecting for contigs hitting viral RefSeq DB, positive by Virsorter classifier, having >2 hits to pVOGs DB per 10 kb, as well as all circular contigs.
Viral dark matter is also included (contigs negative by other tests but not hitting anything in NCBI nt).
Mapping reads
back to
selected contigs
Bowtie2 in end-to-end mode
Count matrix
Bar chart of viral contigs
Ordination plots, α- and β-diversity analyses

11. Viral taxonomy and gene annotation
Feargal Ryan
Bioinformatics
Post-Doc
Demovir
Performs homology searches of protein encoded by contigs using Usearch against viral portion of UniProtKB/TrEMBL database.
Uses voting approach to determine taxonomy of a contig.
Extremely accurate (99% in 10-fold cross- validation test with UniProtKB/TrEMBL database).
Enrique Gonzales
Tortuero
Bioinformatics
Post-Doc
Viga
Predicts open reading frames, tRNA, rRNA genes and repeats by wrapping Prodigal, ARAGORN, INFERNAL, PILERCR, TRF and IRF tools.
Assign functions to protein-coding genes using BLAST/HMM searches against nr, UniProtKB and PFAM databases.
Convenient output in .gbk format.

12. Acknowledgements Gut Phageomics Lab PIs: Prof. Colin Hill
Prof. Paul Ross
Bioinformaticians
Dr. Feargal Ryan
Dr. Adam Clooney
Dr. Enrique Gonzales Tortuero
Thomas Sutton, PhD student
“Wet lab” scientists
Dr. Lorraine Draper
Dr. Stephen Stockdale
As well as all other members of Gut Phageomics Lab and the APC community...