implement and require significant expertise to distinguish
inconsistency resulting from HGTs from other sources.
12.3.4 Methods Based on Compositional Biases
12.3.4.1 Principle
These methods rely on the fact that genomes are more than a
collection of genes. They are dynamic entities whose
sequence evolves constantly through the mutational process.
Depending on the constraints acting on genomes, some types
of mutations are fixed more or less frequently. The main
consequence of this phenomenon is that each genome
harbors relatively homogeneous characteristics. In contrast,
pressures acting on genomes vary from one organism to
another, and thus genomes from different organisms usually
present different features, which may constitute signatures.
Genomic signatures can be of different types. The most
obvious is the oligonucleotide composition of genomic
sequences (Burge et al. 1992; Gautier 2000). It has been
noticed very early that the frequencies of oligonucleotides
are homogeneous within a genome but variable from one
genome to another. Mononucleotides composition (i.e., the
frequency of bases A, C, G, and T) is the easiest to measure
and is usually expressed in terms of G + C content. In
present-day genomes, the G + C content varies from 16 to
75 % (cf. Chap. 6). Di-, tri-, and tetra-nucleotides (i.e., words
of size two, three, or four) are also frequently used to
characterize genomes. Intuitively, it is obvious that the
frequencies of the different types of oligonucleotides
are correlated. For instance, the G + C content will influence
the frequency of dinucleotides. Accordingly, under the
assumption that nucleotides are randomly distributed in
genomes, a depletion of A- and T-containing dinucleotides
(i.e., AA, TT, AT, and TA) and, conversely, an excess of
G- and C-containing dinucleotides (i.e., CC, CG, GG, and
GG) is expected in G + C rich genomes. However, genome
analyses reveal differences between expected and observed
oligonucleotide frequencies. For instance, while the genome
of B. subtilis contains 43.5 % of G + C, the dinucleotides
AA, GC, and TT are more frequent than expected by chance,
and the dinucleotides TA, GT, and AC are underrepresented
(Rocha et al. 1998). The over- and underrepresentation of
oligonucleotides vary among genomes. This suggests that
nucleotides are not randomly distributed within genomes.
This also supports that the frequencies of oligonucleotides of
length equal or greater than 2 can be used for the fine and
reliable detection of atypical genomic regions.
A second type of signature is based on the genetic code
usage. Indeed, because of the degeneration of the genetic
code, most amino acids can be coded by different codons
(Fig. 12.14). This does not mean that when an amino acid is
coded by different codons, each codon is equally used.
Instead, it is frequent that one or two codons are used
preferentially to others. For instance, among the two codons
encoding glutamine, CAG is two times more represented
than CAA in E. coli APEC O1 (Gammaproteobacteria)
genes (Fig. 12.14). Several hypotheses have been proposed
to explain codon usage biases, such as the G + C content
of genomes or the copy number of genes encoding each
tRNA. Regardless of its origin, it has been noticed very
early that the codon usage varies from one organism to
another. The comparison of E. coli APEC O1 and Staphylococcus aureus Mu50 (Firmicutes) revealed that these bacteria do not use each codon at the same frequency. For
instance, while CAA (Glu) is seven times more frequently
used than CAG in S. aureus, E. coli APEC O1 uses preferentially CAG (Fig. 12.14). The most obvious example
regards the codons coding for leucine: CTG is used preferentially by E. coli APEC O1 while TTA is majority in
S. aureus Mu50 (Fig. 12.14). The difference of codon
usage in these two bacteria may reflect the fact that the
genomes of Firmicutes are A and T rich compared to that
of E. coli APEC O1 (the genomic G + C of these two
bacteria is 32 % and 50.5 %, respectively). Therefore, it is
not surprising that the A + T rich codons are more frequently represented in the genome of S. aureus Mu50.
Finally, the amino acid composition of proteins is also
distinctive among prokaryotes. The frequency of each amino
acid varies from a given proteome to another. For instance,
UGG (the only codon for tryptophan) is almost two times
more abundant in E. coli APEC O1 than in S. aureus Mu50
(Fig. 12.14). This indicates that the proteins of the
gammaproteobacterium are enriched with tryptophan compared to that of the firmicute. Conversely, proteins of
S. aureus Mu50 are richer in lysine (codons AAA and
AAG) than those of E. coli APEC O1. The mechanisms at
the origin of amino acid composition variations are not fully
understood. It has been hypothesized that these variations
reflect, at least partially, environmental selective constraints.
For example, it has been shown that the proteomes of hyperthermophilic organisms are enriched in isoleucine, valine,
tyrosine, tryptophan, arginine, leucine, and glutamate
(Zeldovich et al. 2007). Conversely, they tend to be depleted
in amino acids such as asparagine. Similarly, it has been
shown that cytoplasmic proteins from halophiles are
enriched in acidic amino acids and depleted in basic amino
acids (Kennedy et al. 2001).
Nucleotide composition, codon usage, and amino acid
composition biases are well documented (Gautier 2000).
They can also be used for detecting HGT events based on
the fact that if the donor and the recipient organisms belong
to different species, their composition biases may be different. At the time of transfer, the transferred DNA harbors
genomic signatures which are specific of the donor genome
whereas the receptor cell will present different genomic
12 Horizontal Gene Transfer in Microbial Ecosystems
461
inconsistency resulting from HGTs from other sources.
12.3.4 Methods Based on Compositional Biases
12.3.4.1 Principle
These methods rely on the fact that genomes are more than a
collection of genes. They are dynamic entities whose
sequence evolves constantly through the mutational process.
Depending on the constraints acting on genomes, some types
of mutations are fixed more or less frequently. The main
consequence of this phenomenon is that each genome
harbors relatively homogeneous characteristics. In contrast,
pressures acting on genomes vary from one organism to
another, and thus genomes from different organisms usually
present different features, which may constitute signatures.
Genomic signatures can be of different types. The most
obvious is the oligonucleotide composition of genomic
sequences (Burge et al. 1992; Gautier 2000). It has been
noticed very early that the frequencies of oligonucleotides
are homogeneous within a genome but variable from one
genome to another. Mononucleotides composition (i.e., the
frequency of bases A, C, G, and T) is the easiest to measure
and is usually expressed in terms of G + C content. In
present-day genomes, the G + C content varies from 16 to
75 % (cf. Chap. 6). Di-, tri-, and tetra-nucleotides (i.e., words
of size two, three, or four) are also frequently used to
characterize genomes. Intuitively, it is obvious that the
frequencies of the different types of oligonucleotides
are correlated. For instance, the G + C content will influence
the frequency of dinucleotides. Accordingly, under the
assumption that nucleotides are randomly distributed in
genomes, a depletion of A- and T-containing dinucleotides
(i.e., AA, TT, AT, and TA) and, conversely, an excess of
G- and C-containing dinucleotides (i.e., CC, CG, GG, and
GG) is expected in G + C rich genomes. However, genome
analyses reveal differences between expected and observed
oligonucleotide frequencies. For instance, while the genome
of B. subtilis contains 43.5 % of G + C, the dinucleotides
AA, GC, and TT are more frequent than expected by chance,
and the dinucleotides TA, GT, and AC are underrepresented
(Rocha et al. 1998). The over- and underrepresentation of
oligonucleotides vary among genomes. This suggests that
nucleotides are not randomly distributed within genomes.
This also supports that the frequencies of oligonucleotides of
length equal or greater than 2 can be used for the fine and
reliable detection of atypical genomic regions.
A second type of signature is based on the genetic code
usage. Indeed, because of the degeneration of the genetic
code, most amino acids can be coded by different codons
(Fig. 12.14). This does not mean that when an amino acid is
coded by different codons, each codon is equally used.
Instead, it is frequent that one or two codons are used
preferentially to others. For instance, among the two codons
encoding glutamine, CAG is two times more represented
than CAA in E. coli APEC O1 (Gammaproteobacteria)
genes (Fig. 12.14). Several hypotheses have been proposed
to explain codon usage biases, such as the G + C content
of genomes or the copy number of genes encoding each
tRNA. Regardless of its origin, it has been noticed very
early that the codon usage varies from one organism to
another. The comparison of E. coli APEC O1 and Staphylococcus aureus Mu50 (Firmicutes) revealed that these bacteria do not use each codon at the same frequency. For
instance, while CAA (Glu) is seven times more frequently
used than CAG in S. aureus, E. coli APEC O1 uses preferentially CAG (Fig. 12.14). The most obvious example
regards the codons coding for leucine: CTG is used preferentially by E. coli APEC O1 while TTA is majority in
S. aureus Mu50 (Fig. 12.14). The difference of codon
usage in these two bacteria may reflect the fact that the
genomes of Firmicutes are A and T rich compared to that
of E. coli APEC O1 (the genomic G + C of these two
bacteria is 32 % and 50.5 %, respectively). Therefore, it is
not surprising that the A + T rich codons are more frequently represented in the genome of S. aureus Mu50.
Finally, the amino acid composition of proteins is also
distinctive among prokaryotes. The frequency of each amino
acid varies from a given proteome to another. For instance,
UGG (the only codon for tryptophan) is almost two times
more abundant in E. coli APEC O1 than in S. aureus Mu50
(Fig. 12.14). This indicates that the proteins of the
gammaproteobacterium are enriched with tryptophan compared to that of the firmicute. Conversely, proteins of
S. aureus Mu50 are richer in lysine (codons AAA and
AAG) than those of E. coli APEC O1. The mechanisms at
the origin of amino acid composition variations are not fully
understood. It has been hypothesized that these variations
reflect, at least partially, environmental selective constraints.
For example, it has been shown that the proteomes of hyperthermophilic organisms are enriched in isoleucine, valine,
tyrosine, tryptophan, arginine, leucine, and glutamate
(Zeldovich et al. 2007). Conversely, they tend to be depleted
in amino acids such as asparagine. Similarly, it has been
shown that cytoplasmic proteins from halophiles are
enriched in acidic amino acids and depleted in basic amino
acids (Kennedy et al. 2001).
Nucleotide composition, codon usage, and amino acid
composition biases are well documented (Gautier 2000).
They can also be used for detecting HGT events based on
the fact that if the donor and the recipient organisms belong
to different species, their composition biases may be different. At the time of transfer, the transferred DNA harbors
genomic signatures which are specific of the donor genome
whereas the receptor cell will present different genomic
12 Horizontal Gene Transfer in Microbial Ecosystems
461