Predicting plasmid host range with dinucleotide analysis

1
Predicting plasmid host range with dinucleotide
analysis Haruo Suzuki, Masahiro Sota, Celeste J.
Brown, and Eva M. Top Department of Biological
Sciences, University of Idaho, Moscow, Idaho 83844
R-071
Contact Information Department of Biological
Sciences, PO Box 443051, University of Idaho,
Moscow, ID 83844-3051, Phone [phone]
Fax [phone], E-mail hsuzuki_at_uidaho.edu

• INTRODUCTION
• Bacterial plasmids are ubiquitous mobile elements
  that serve as a pool of many host-beneficial
  traits such as antibiotic resistance 1. To
  understand the role of plasmids in horizontal
  gene transfer, we need to gain insight into the
  life history of the plasmids, i.e., the range
  of hosts in which they have evolved. Since
  extensive data support the proposal that foreign
  DNA acquires the host's nucleotide composition
  during long-term residence (amelioration) 2,
  comparison of nucleotide composition of plasmids
  and chromosomes could shed light on a plasmids'
  life history.
• The average absolute difference, ?, has been
  commonly used to measure differences in
  dinucleotide relative abundance or 'genomic
  signature' between bacterial chromosomes and
  plasmids 3. Here, we introduced Mahalanobis
  distance, D2, which accounts for the
  variance-covariance structure of the chromosomal
  signatures.

Was virulence plasmid pO157 acquired from
Yersinia pestis? Bacterial strains most similar
in dinucleotide composition to plasmid pO157 from
enterohemorrhagic Escherichia coli O157H7 EDL933.
Mahalanobis distance identifies plausible hosts
of Bacillus anthracis virulence plasmid
pXO1 Five highest ranking bacterial strains
based on Mahalanobis distance D2 (A) and average
absolute difference ? (B) for plasmid pXO1.
GOAL To compare performance of Mahalanobis
distance and commonly used ?-distance in
identifying plasmids hosts based on dinucleotide
composition similarity

• Yersinia pestis, agent of the bubonic plague,
  were identified as potential long-term hosts of
  plasmid pO157.

Inferring potential hosts of a plasmid from
unknown host Bacterial strains most similar in
dinucleotide composition to broad-host-range
plasmid pB10 captured directly from environmental
samples 4.

• The D2 value ranked plausible hosts (B. cereus
  sensu lato) first, while the ? value ranked
  unlikely hosts first.
• Different species comprising the B. cereus sensu
  lato group are largely defined by differences in
  plasmids, while the chromosomes are highly
  similar.
• The known natural host of pXO1, B. anthracis,
  ranked 5 based on the D2 value, and only 10 based
  on the ? value.

• METHODS
• Analyses were implemented using G-language Genome
  Analysis Environment, available at
  http//www.g-language.org. Complete genome
  sequences of 195 chromosomes and 412 plasmids of
  Bacteria were retrieved from the NCBI ftp site
  (ftp//ftp.ncbi.nih.gov/genomes/Bacteria).
• Dinucleotide relative abundance (xij) is defined
  as
• where fi and fj denote the frequency of the
  mononucleotide i and j respectively (i,j ?
  A,C,G,T) and fij the frequency of the
  dinucleotide ij.
• We compared two distance measures to quantify the
  dinucleotide relative abundance difference
  between DNA sequences.
• Average absolute difference, ?, used previously
• where xij and yij denote the relative abundance
  value of the dinucleotide ij for the plasmid and
  the chromosomal segment, respectively.
• Mahalanobis distance, D2, introduced in this
  study

Mahalanobis distance performs better to identify
plasmids hosts Distributions of ranks of
dinucleotide relative abundance similarities
between plasmids and their known natural hosts
based on Mahalanobis distance D2 (A) and average
absolute difference ? (B)

• Members of the ß-Proteobacteria were identified
  as long-term evolutionary hosts of plasmid pB10,
  which is in agreement with experimental evidence
  of the plasmids replication range.

• CONCLUSIONS
• Mahalanobis distance performs better than
  commonly used ?-distance to identify known
  natural hosts of plasmids based on dinucleotide
  composition similarity.
• Dinucleotide analysis with Mahalanobis distance
  can also be used to infer potential hosts of
  plasmids, which can then be experimentally tested.

_
Acknowledgements We thank members of IBEST
(Initiative for Bioinformatics and Evolutionary
Studies) at the University of Idaho for useful
discussions. This work was supported by National
Science Foundation (EF-0627988) National
Institutes of Health (P20RR016454,
P20RR16448). References 1 Frost LS et al.
(2005) Nat Rev Microbiol, 3, 722-732. 2
Lawrence JG and Ochman H (1997) J Mol Evol, 44,
383-397. 3 Campbell A et al. (1999) Proc Natl
Acad Sci USA, 96, 9184-9189. 4 Schlüter A et
al. (2003) Microbiology 1493139-53.

• The plasmid-host pairs tended to rank higher when
  using the D2 value compared to the ? value
  (Wilcoxon signed rank test p lt 10-15).