Content uploaded by Rob Edwards
Author content
All content in this area was uploaded by Rob Edwards on Aug 22, 2019
Content may be subject to copyright.
Chapter 17
Phage Genome Annotation Using the RAST Pipeline
Katelyn McNair, Ramy Karam Aziz, Gordon D. Pusch, Ross Overbeek,
Bas E. Dutilh, and Robert Edwards
Abstract
Phages are complex biomolecular machineries that have to survive in a bacterial world. Phage genomes
show many adaptations to their lifestyle such as shorter genes, reduced capacity for redundant DNA
sequences, and the inclusion of tRNAs in their genomes. In addition, phages are not free-living, they
require a host for replication and survival. These unique adaptations provide challenges for the bioinfor-
matics analysis of phage genomes. In particular, ORF calling, genome annotation, noncoding RNA
(ncRNA) identification, and the identification of transposons and insertions are all complicated in phage
genome analysis. We provide a road map through the phage genome annotation pipeline, and discuss the
challenges and solutions for phage genome annotation as we have implemented in the rapid annotation
using subsystems (RAST) pipeline.
Key words Phage, Genome annotation, RAST, Functional annotation, Gene predictions
1 The Steps of Phage Genome Annotation
The essential steps in annotating any genome, whether phage,
bacterial, or eukaryotic, consist of identifying the features in the
genome and assigning terms describing roles or functions to those
features. Typical features that can be found in a phage genome
include protein-encoding genes, noncoding RNA genes, insertion
elements and transposons, direct and indirect repeats, origins of
replication, and attachment or integration sites. Annotations are
routinely only added to protein and RNA-encoding genes, labels
are often provided for insertion elements or transposons. Specific
for phages, they are fundamentally dependent on a cellular host to
replicate, and the functions on its genome can only be completely
understood in the context of the genome of the host. Thus, identi-
fication of prediction of the bacterial or archaeal host is an impor-
tant part of phage annotation. Together, these features provide the
core annotation of phages and this annotation provides the first
steps to understanding the function of the phage as it interacts with
Martha R.J. Clokie et al. (eds.), Bacteriophages: Methods and Protocols, Volume 3, Methods in Molecular Biology, vol. 1681,
https://doi.org/10.1007/978-1-4939-7343-9_17,©Springer Science+Business Media LLC 2018
231
its host (Fig. 1). We discuss the approaches to identify and annotate
each of these features below, and discuss how these annotations are
performed in the Rapid Annotation Using Subsystems Technology
approach (RAST) [1,2].
Protein-encoding genes are the focus of most automated anno-
tation systems, and more algorithms have been designed to handle
these features than other features. Generally a protein-encoding
gene can be identified as a long stretch of sequence in one reading
frame that can be translated into protein sequence without includ-
ing one of the three stop codons; these long stretches are called
Open Reading Frames (ORFs). In gene calling, the stop codons are
obvious because there is a choice of three codons to choose from
and they are all stop codons (unless the phage encodes a suppressor
tRNA which we do not discuss here). Most algorithms attempt to
identify the longest nonoverlapping ORFs in a genome, based on
the theory that the longer the open reading frame the less likely it is
to occur by chance. There are many alternative gene-finding algo-
rithms that have been developed over the last two decades,
Fig. 1 Pipeline of phage genome annotation starting with DNA sequences and
ending with an annotated genome
232 Katelyn McNair et al.
including CRITICA [3], GeneMark [4,5], GISMO [6], Glimmer
[7,8], MetaGeneAnnotator [9], and Prodigal [10]. Most of the
gene-finding algorithms find the same large genes because these are
obvious and have high confidence. The algorithms may differ in the
particular start sites that they identify; there may be multiple methi-
onine (ATG) or valine (GTG) codons that could all be used as the
start codon, and predicting exactly which start codon is the correct
one for a given gene is difficult without a priori knowledge of the
translation boundaries of the gene. In addition, the gene callers also
differ in their ability to identify small protein-encoding genes.
Short genes are statistically difficult to separate from the back-
ground noise of stretches of nucleotides that do not encode a
stop codon, and often gene calling algorithms use an artificial cut
off of (for example) 75 amino acids. It remains to be determined
how many small proteins are encoded in phage genomes, and this is
unlikely to be approached from a pure bioinformatics standpoint, as
it will require biological validation of bioinformatics predictions or
large-scale proteomic studies.
Most bacterial genomes are not thought to contain overlapping
open reading frames, and these shadow ORFs are removed during
the annotation step [10]. In viruses, including phages, however,
there are several well-known examples of two different genes from
the same stretch of DNA, such as the Rz/Rz1 system [11]. One
study even suggests that new genes may be born via this process,
providing evidence from the comparative genomics of Rhabdovir-
idae genomes [12]. These overlapping regions are generally not
predicted using most bioinformatics approaches, as adding over-
lapping ORFs to gene prediction algorithms would include an
enormous number of false positives to compensate for only a few
false negatives. Therefore, most phage protein prediction schemes
ignore overlapping proteins.
Following ORF identification, most bioinformatic gene predic-
tion tools assign a confidence score to the ORFs using a model of
what a gene is expected to look like, based on its nucleotide usage
statistics. These statistics are specific for a species, and depend on
properties like the codon usage and GC content of the genome. In
bacterial genomes, the RAST pipeline starts by identifying highly
conserved genes that are present in nearly every genome. The
statistics from those genes are then used to build a genome-specific
model for open reading frame identification that is applied to the
rest of the genome. In phage genes, there are typically very few, if
any, highly conserved genes, and never enough to build a reliable
gene model. Therefore, most gene calling is performed by a generic
model that is not trained on the specific genome being annotated
but on the genomes of all phages. By default, the RAST pipeline
uses Glimmer to identify the open reading frames, but options are
available to use MetaGeneAnnotator [9], GeneMark [4], or
Prodigal [10].
Phage Genome Annotation Using the RAST Pipeline 233
The functional annotation of protein-encoding phage genes is
usually based on homology searches against existing phages. His-
torically, phage genes were named with a single letter starting at
gpA, and either proceeding along the genome or assigning names
based on the order in which the genes or their products were
found. This resulted in several unrelated proteins from different
phages all having the same names. For example both terminases and
DNA replication initiation proteins have been annotated as gpA in
different phage genomes available from GenBank. This confusion,
amplified by the explosion of genome sequences in recent years, led
to efforts to categorize phage proteins into either phage ortholo-
gous groups (POGs) [13] or subsystems [14] that have unified the
annotation of many phage proteins. These common, descriptive,
names provide a framework for comparing annotations among
different phage genomes. The RAST system uses a combination
of homology, chromosomal clustering, and subsystems to assign
functions to proteins. First, proteins are annotated on the basis of
homology to known proteins. If this initial search yields matches to
proteins that are a component of a subsystem, RAST then tries to
find other members of the subsystem that should be present in the
same genome based on information from the previously annotated
genomes. The advantage of this approach is that the RAST system
can strengthen otherwise weak assertions of homology, based on
predictions from subsystem annotations. Of note, the RAST tools
allow the analysis of proteins in their chromosomal context, which
sometimes helps determine the roles of proteins with unknown
functions based on the functions of their chromosomal neighbors
(e.g., protein subunits encoded by different genes, members of
operons, or transporters of metabolites whose metabolizing
enzymes are encoded on the same cluster). Phage genomes, like
bacterial genomes, also order some of their genes, and this infor-
mation can be leveraged to identify clusters of genes. For example,
the small and large terminase (TerS and TerL) are frequently adja-
cent on the genome, and the identification of one leads to the
identification of the other.
A major difficulty in the functional annotation of protein-
encoding genes on phage genomes by homology searches is the
fact that most proteins have no close homologs in the reference
databases. Especially for novel phages, this results in the majority of
encoded ORFs having no annotated function, or a hypothetical
function at best. A possible solution includes homology-indepen-
dent annotation, based on amino acid usage profiles of the proteins.
One such approach, iVIREONS (https://vdm.sdsu.edu/ivireons/
) uses machine learning to “learn” the characteristics of manually
annotated phage proteins and then tests unknown proteins to see if
they have similar characteristics [15].
Noncoding RNA (ncRNA) genes. Although Ribosomal RNAs
have not yet been found in phage genomes, most pipelines,
234 Katelyn McNair et al.
including RAST, look for them anyway as the pipelines have been
developed for bacterial genome annotation and the computational
cost of looking for rRNA genes is a minimal addition to the pipe-
line. Ribosomal RNA genes are highly conserved and are identified
by extrinsic gene calling—using a database of known RNA genes to
compare against. In contrast to rRNA genes that are recognized by
homology, tRNA genes are recognized by intrinsic gene calling—
using only features of the sequence. They are typically identified by
computational tools built specifically to recognize the secondary
structure of the tRNA molecule [16]. As with tRNAs, the function
of other non-protein coding RNA genes also depends on the
structure of the folded RNA molecule rather than the nucleotide
sequence. Therefore, other noncoding RNA genes are also recog-
nized by their conserved secondary structure rather than homology
to existing sequences [17]. The RAST pipeline uses a manually
curated database of ribosomal RNA genes to find them in a
genome, and uses tRNAScan-SE [16] to identify tRNA genes.
Many phages encode tRNA genes, and it has been proposed that
these may supplement host-encoded tRNAs in translating phage
proteins for anticodons that are insufficiently covered by the bacte-
rial tRNAs [18]. These tRNA genes are also often used as phage
integration sites in the host’s genome (attP). Integration of the
phage disrupts the host gene, and thus carrying complete, or near
complete, tRNA genes allows the phage to reconstitute a tRNA
into which it can integrate [19]. There has been little exploration of
the role of ncRNA in phage lifestyle. Recent work with CRISPR/
Cas systems have identified the presence of these systems in phage
genomes [20] and metagenomes [21], and it is thought that they
are being used to attack other phages that may be infecting the
same host.
Insertion elements and transposons are currently identified by
annotations of protein-encoding genes. Transposases (Tn) are
readily identified as protein-encoding genes, and the similarity
between members of the transposase family, and with other recom-
binases, is high enough that they usually receive accurate annota-
tion. However, the repeats flanking the insertion sequence or
transposon are not typically automatically annotated. There are
boutique databases of these problematic mobile elements
[22,23], but often the classification of insertion (IS) elements is
dependent on one or a few residues. Typically automatic annotation
systems identify the Tn or IS elements but cannot identify the fine
details responsible for the accurate categorization of these ele-
ments. More work is required to accurately denote the ends of
these mobile elements in automatic phage annotation systems.
Direct and indirect repeats are usually used to identify the ends of
insertion elements and transposons [22], and to predict the ends of
prophages that have been found in bacterial genomes [13]. Stan-
dard informatics approaches can easily identify repeats longer than
Phage Genome Annotation Using the RAST Pipeline 235
approximately 14 nucleotides in a phage genome. Below that
length, repeats are found too frequently to ascertain whether they
are indeed the correct flanking repeats, or randomly occurring
repeated sequence elements. A few websites can be used to identify
repeats in DNA sequences (e.g., [24,25]).
Phage attachment sites are impossible to detect de novo if only
the phage is known, but if the phage and the host genome
sequences are known, they are trivial to find. The phage carries
the attachment site P (attP) that has sequence homology to the
bacterial attachment site B (attB). Integration is initiated by recom-
bination between attP and attB, resulting in attL and attR sites
that flank the nascent prophage.
Accurately Annotating Phage Metadata. Annotating genomic
metadata is a general challenge to genomics and metagenomics.
With bacteriophages, this issue is even more problematic, given the
lack of systematic nomenclature for viruses (as opposed to the
binomial system used for cellular organisms, see Chapter 15 of this
book). Some attempts were made to suggest systematic nomencla-
ture for viruses similar to those used for plasmids [26], but they are
not widely applied or enforced. In addition to accurate taxonomic
descriptions of viruses, including metadata associated with the virus
(e.g., its morphology, actual host, host range, and lifestyle) is
equally important. These make comparative genomics studies pos-
sible, enable predictive tools such as those that identify the host of
unknown phages [27], or predict the lifestyle of new phages [14]
and improve metagenomic/microbiomic annotations. Other
important types of metadata can be computed from the genomic
information, e.g., a genome’s length, %G+C, and codon usage
[28]. These too have quite powerful applications in comparative
genomics, prophage finding, and metagenomics. For example,
information content of phage genomes has improved prophage
finding [29] and is proposed to improve metagenomic analysis
[30]. As with gene annotation, metadata annotation needs to use
a controlled vocabulary (which has to be consistent but not neces-
sarily rigid or hierarchical). Spelling inconsistencies (e.g., firmicutes
vs. Firmicutes vs. gram-positive bacteria) or terminology inconsis-
tencies (e.g., temperate vs. lysogenic lifestyles) are all obstacles
against computational analysis and data propagation.
To summarize, phage annotation involves the identification
and functional description of several types of features, including
protein-encoding genes, RNA genes, insertion elements and trans-
posons, repeats, and attachment sites. Moreover, phage–host asso-
ciations are an important part of understanding phage biology that
can be predicted using a range of computational tools [27]. The
RAST pipeline provides an automated approach to phage genome
annotation. The pipeline currently uses bacterial ORF-finding algo-
rithms to identify the proteins in the genome, and a combination of
236 Katelyn McNair et al.
homology-based and subsystems-based approaches to decorate
those proteins with their functional annotation. RNA genes are
detected by a combination of extrinsic and intrinsic gene calling
methods. There remain several hurdles to accurate phage genome
annotation, especially the assignment of functions to unknown
proteins, the identification of small proteins in the genome, and
the correct and unambiguous identification of insertion elements
and transposons. The combinations of bioinformatics advances and
a better understanding of phage biology will help to improve phage
genome annotation, making this field a fertile area for further
exploration.
Acknowledgments
This work was supported by grants from the National Science
Foundation MCB-1330800 and DUE-1323809 to RAE. BED
was supported by the Netherlands Organization for Scientific
Research (NWO) Vidi grant 864.14.004.
References
1. Aziz RK, Bartels D, Best AA, DeJongh M,
Disz T, Edwards RA, Formsma K, Gerdes S,
Glass EM, Kubal M, Meyer F, Olsen GJ,
Olson R, Osterman AL, Overbeek RA, McNeil
LK, Paarmann D, Paczian T, Parrello B, Pusch
GD, Reich C, Stevens R, Vassieva O,
Vonstein V, Wilke A, Zagnitko O (2008) The
RAST Server: rapid annotations using subsys-
tems technology. BMC Genomics 9:75
2. Brettin T, Davis JJ, Disz T, Edwards RA,
Gerdes S, Olsen GJ, Olson R, Overbeek R,
Parrello B, Pusch GD, Shukla M, Thomason
Iii JA, Stevens R, Vonstein V, Wattam AR, Xia F
(2015) RASTtk: A modular and extensible
implementation of the RAST algorithm for
building custom annotation pipelines and
annotating batches of genomes. Sci Rep
5:8365
3. Badger JH, Olsen GJ (1999) CRITICA: cod-
ing region identification tool invoking compar-
ative analysis. Mol Biol Evol 16:512–524
4. Borodovsky M, Mclninch JD, Koonin EV,
Rudd KE, Me
´digue C, Danchin A (1995)
Detection of new genes in a bacterial genome
using Markov models for three gene classes.
Nucleic Acids Res 23:3554–3562
5. Lukashin AV, Borodovsky M (1998) Gene-
Mark.hmm: new solutions for gene finding.
Nucleic Acids Res 26:1107–1115
6. Krause L, McHardy AC, P€
uhler A, Stoye J,
Meyer F (2007) GISMO - Gene identification
using a support vector machine for ORF classi-
fication. Nucleic Acids Res 35:540–549
7. Delcher AL, Harmon D, Kasif S, White O,
Salzberg SL (1999) Improved microbial gene
identification with GLIMMER. Nucleic Acids
Res 27:4636–4641
8. Kelley DR, Liu B, Delcher AL, Pop M, Salz-
berg SL (2012) Gene prediction with Glimmer
for metagenomic sequences augmented by
classification and clustering. Nucleic Acids Res
40:e9–e9
9. Noguchi H, Taniguchi T, Itoh T (2008) Meta-
GeneAnnotator: Detecting species-specific pat-
terns of ribosomal binding site for precise gene
prediction in anonymous prokaryotic and
phage genomes. DNA Res 15:387–396
10. Hyatt D, Chen G-L, LoCascio PF, Land ML,
Larimer FW, Hauser LJ (2010) Prodigal: pro-
karyotic gene recognition and translation initi-
ation site identification. BMC Bioinformatics
11:119
11. Summer EJ, Berry J, Tran TAT, Niu L, Struck
DK, Young R (2007) Rz/Rz1 lysis gene
equivalents in phages of Gram-negative hosts.
J Mol Biol 373:1098–1112
12. Walker PJ, Firth C, Widen SG, Blasdell KR,
Guzman H, Wood TG, Paradkar PN, Holmes
EC, Tesh RB, Vasilakis N (2015) Evolution of
genome size and complexity in the Rhabdovir-
idae. PLoS Pathog 11:e1004664
Phage Genome Annotation Using the RAST Pipeline 237
13. Kristensen DM, Waller AS, Yamada T, Bork P,
Mushegian AR, Koonin EV (2013) Ortholo-
gous gene clusters and taxon signature genes
for viruses of prokaryotes. J Bacteriol
195:941–950
14. McNair K, Bailey BA, Edwards RA (2012)
PHACTS, a computational approach to classi-
fying the lifestyle of phages. Bioinformatics
28:614–618
15. Seguritan V, Alves N, Arnoult M, Raymond A,
Lorimer D, Burgin AB, Salamon P, Segall AM
(2012) Artificial neural networks trained to
detect viral and phage structural proteins.
PLoS Comput Biol 8:e1002657
16. Lowe TM, Eddy SR (1997) tRNAscan-SE: a
program for improved detection of transfer
RNA genes in genomic sequence. Nucleic
Acids Res 25:955–964
17. Nawrocki EP (2014) Annotating functional
RNAs in genomes using Infernal. Methods
Mol Biol 1097:163–197
18. Bailly-Bechet M, Vergassola M, Rocha E
(2007) Causes for the intriguing presence of
tRNAs in phages. Genome Res 17:1486–1495
19. Williams KP (2002) Integration sites for
genetic elements in prokaryotic tRNA and
tmRNA genes: sublocation preference of inte-
grase subfamilies. Nucleic Acids Res
30:866–875
20. Seed KD, Lazinski DW, Calderwood SB,
Camilli A (2013) A bacteriophage encodes its
own CRISPR/Cas adaptive response to evade
host innate immunity. Nature 494:489–491
21. Cassman N, Prieto-Davo
´A, Walsh K, Silva
GGZ, Angly F, Akhter S, Barott K, Busch J,
McDole T, Haggerty JM, Willner D,
Alarco
´n G, Ulloa O, DeLong EF, Dutilh BE,
Rohwer F, Dinsdale EA (2012) Oxygen mini-
mum zones harbour novel viral communities
with low diversity. Environ Microbiol
14:3043–3065
22. Aziz RK, Breitbart M, Edwards RA (2010)
Transposases are the most abundant, most
ubiquitous genes in nature. Nucleic Acids Res
38:4207–4217
23. Riadi G, Medina-Moenne C, Holmes DS
(2012) TnpPred: a web service for the robust
prediction of prokaryotic transposases. Comp
Funct Genomics 2012:678761
24. Benson G (1999) Tandem repeats finder: a
program to analyze DNA sequences. Nucleic
Acids Res 27:573–580
25. Volfovsky N, Haas BJ, Salzberg SL (2001) A
clustering method for repeat analysis in DNA
sequences. Genome Biol 2:RESEARCH0027
26. Kropinski AM, Prangishvili D, Lavigne R
(2009) Position paper: the creation of a ratio-
nal scheme for the nomenclature of viruses of
Bacteria and Archaea. Environ Microbiol
11:2775–2777
27. Edwards RA, McNair K, Faust K, Raes J,
Dutilh BE (2016) Computational approaches
to predict bacteriophage–host relationships.
FEMS Microbiol Rev 40:58–72
28. Aziz RK, Dwivedi B, Akhter S, Breitbart M,
Edwards RA (2015) Multidimensional metrics
for estimating phage abundance, distribution,
gene density, and sequence coverage in meta-
genomes. Front Microbiol 6:381
29. Akhter S, Aziz RK, Edwards RA (2012)
PhiSpy: a novel algorithm for finding pro-
phages in bacterial genomes that combines
similarity- and composition-based strategies.
Nucleic Acids Res 40:e126–e126
30. Akhter S, Bailey BA, Salamon P, Aziz RK,
Edwards RA (2013) Applying Shannon’s infor-
mation theory to bacterial and phage genomes
and metagenomes. Sci Rep 3:1033
238 Katelyn McNair et al.
