Horizontal or lateral gene transfer (HGT or LGT) is the transmission of portions of genomic DNA between organisms through a process decoupled from vertical inheritance. In the presence of HGT events, different fragments of the genome are the result of different evolutionary histories. This can therefore complicate investigations of the evolutionary relatedness of lineages and species. Also, as HGT can bring into genomes radically different genotypes from distant lineages, or even new genes bearing new functions, it is a major source of phenotypic innovation and a mechanism of niche adaptation. For example, of particular relevance to human health is the lateral transfer of antibiotic resistance and pathogenicity determinants, leading to the emergence of pathogenic lineages.
Inferring horizontal gene transfer through computational identification of HGT events relies upon the investigation of sequence composition or evolutionary history of genes. Sequence composition-based ("parametric") methods search for deviations from the genomic average whereas evolutionary history-based ("phylogenetic") approaches identify genes whose evolutionary history significantly differs from that of the host species. The evaluation and benchmarking of HGT inference methods typically rely upon simulated genomes, for which the true history is known. On real data, different methods tend to infer different HGT events, and as a result it can be difficult to ascertain all but simple and clear-cut HGT events.
Overview Horizontal gene transfer was first observed in 1928, in Frederick Griffith's experiment: showing that virulence was able to pass from virulent to non-virulent strains of Streptococcus pneumoniae, Griffith demonstrated that genetic information can be horizontally transferred between bacteria via a mechanism known as transformation. Similar observations in the 1940s and 1950s showed evidence that conjugation and transduction are additional mechanisms of horizontal gene transfer.
To infer HGT events, which may not necessarily result in phenotypic changes, most contemporary methods are based on analyses of genomic sequence data. These methods can be broadly separated into two groups: parametric and phylogenetic methods. Parametric methods search for sections of a genome that significantly differ from the genomic average, such as GC content or codon usage. Phylogenetic methods examine evolutionary histories of genes involved and identify conflicting phylogenies. Phylogenetic methods can be further divided into those that reconstruct and compare phylogenetic trees explicitly, and those that use surrogate measures in place of the phylogenetic trees.
The main feature of parametric methods is that they only rely on the genome under study to infer HGT events that may have occurred on its lineage. It has been a considerable advantage at the early times of the sequencing era, when few closely related genomes were available for comparative methods. However, because they rely on the uniformity of the host's signature to infer HGT events, not accounting for the host's intra-genomic variability will result in overpredictions—flagging native segments as possible HGT events. Similarly, the transferred segments need to exhibit the donor's signature and to be significantly different from the recipient's. Furthermore, genomic segments of foreign origin are subject to the same mutational processes as the rest of the host genome, and so the difference between the two tends to vanish over time, a process referred to as amelioration. This limits the ability of parametric methods to detect ancient HGTs.
Phylogenetic methods benefit from the recent availability of many sequenced genomes. Indeed, as for all comparative methods, phylogenetic methods can integrate information from multiple genomes, and in particular integrate them using a model of evolution. This lends them the ability to better characterize the HGT events they infer—notably by designating the donor species and time of the transfer. However, models have limits and need to be used cautiously. For instance, the conflicting phylogenies can be the result of events not accounted for by the model, such as unrecognized paralogy due to duplication followed by gene losses. Also, many approaches rely on a reference species tree that is supposed to be known, when in many instances it can be difficult to obtain a reliable tree. Finally, the computational costs of reconstructing many gene/species trees can be prohibitively expensive. Phylogenetic methods tend to be applied to genes or protein sequences as basic evolutionary units, which limits their ability to detect HGT in regions outside or across gene boundaries.
Because of their complementary approaches—and often non-overlapping sets of HGT candidates—combining predictions from parametric and phylogenetic methods can yield a more comprehensive set of HGT candidate genes. Indeed, combining different parametric methods has been reported to significantly improve the quality of predictions. Moreover, in the absence of a comprehensive set of true horizontally transferred genes, discrepancies between different methods might be resolved through combining parametric and phylogenetic methods. However, combining inferences from multiple methods also entails a risk of an increased false-positive rate.
Parametric methods Parametric methods to infer HGT use characteristics of the genome sequence specific to particular species or clades, also called genomic signatures. If a fragment of the genome strongly deviates from the genomic signature, this is a sign of a potential horizontal transfer. For example, because bacterial GC content falls within a wide range, GC content of a genome segment is a simple genomic signature. Commonly used genomic signatures include nucleotide composition, oligonucleotide frequencies, or structural features of the genome.
To detect HGT using parametric methods, the host's genomic signature needs to be clearly recognizable. However, the host's genome is not always uniform with respect to the genome signature: for example, GC content of the third codon position is lower close to the replication terminus and GC content tends to be higher in highly expressed genes. Not accounting for such intra-genomic variability in the host can result in over-predictions, flagging native segments as HGT candidates. Larger sliding windows can account for this variability at the cost of a reduced ability to detect smaller HGT regions.
Just as importantly, horizontally transferred segments need to exhibit the donor's genomic signature. This might not be the case for ancient transfers where transferred sequences are subjected to the same mutational processes as the rest of the host genome, potentially causing their distinct signatures to "ameliorate" and become undetectable through parametric methods. For example, Bdellovibrio bacteriovorus, a predatory δ-Proteobacterium, has homogeneous GC content, and it might be concluded that its genome is resistant to HGT. However, subsequent analysis using phylogenetic methods identified a number of ancient HGT events in the genome of B. bacteriovorus. Similarly, if the inserted segment was previously ameliorated to the host's genome, as is the case for prophage insertions, parametric methods might miss predicting these HGT events. Also, the donor's composition must significantly differ from the recipient's to be identified as abnormal, a condition that might be missed in the case of short- to medium-distance HGT, which are the most prevalent. Furthermore, it has been reported that recently acquired genes tend to be AT-richer than the recipient's average, which indicates that differences in GC-content signature may result from unknown post-acquisition mutational processes rather than from the donor's genome.
Nucleotide composition
Bacterial GC content falls within a wide range, with Ca. Zinderia insecticola having a GC content of 13.5% and Anaeromyxobacter dehalogenans having a GC content of 75%. Even within a closely related group of α-Proteobacteria, values range from approximately 30% to 65%. These differences can be exploited when detecting HGT events as a significantly different GC content for a genome segment can be an indication of foreign origin.
Oligonucleotide spectrum The oligonucleotide spectrum (or k-mer frequencies) measures the frequency of all possible nucleotide sequences of a particular length in the genome. It tends to vary less within genomes than between genomes and therefore can also be used as a genomic signature. A deviation from this signature suggests that a genomic segment might have arrived through horizontal transfer.
The oligonucleotide spectrum owes much of its discriminatory power to the number of possible oligonucleotides: if n is the size of the vocabulary and w is oligonucleotide size, the number of possible distinct oligonucleotides is nw; for example, there are 45=1024 possible pentanucleotides. Some methods can capture the signal recorded in motifs of variable size, thus capturing both rare and discriminative motifs along with frequent, but more common ones.
Codon usage bias, a measure related to codon frequencies, was one of the first detection methods used in methodical assessments of HGT. This approach requires a host genome which contains a bias towards certain synonymous codons (different codons which code for the same amino acid) which is clearly distinct from the bias found within the donor genome. The simplest oligonucleotide used as a genomic signature is the dinucleotide, for example the third nucleotide in a codon and the first nucleotide in the following codon represent the dinucleotide least restricted by amino acid preference and codon usage.
It is important to optimise the size of the sliding window in which to count the oligonucleotide frequency: a larger sliding window will better buffer variability in the host genome at the cost of being worse at detecting smaller HGT regions. A good compromise has been reported using tetranucleotide frequencies in a sliding window of 5 kb with a step of 0.5kb.
A convenient method of modelling oligonucleotide genomic signatures is to use Markov chains. The transition probability matrix can be derived for endogenous vs. acquired genes, from which Bayesian posterior probabilities for particular stretches of DNA can be obtained.
Structural features Just as the nucleotide composition of a DNA molecule can be represented by a sequence of letters, its structural features can be encoded in a numerical sequence. The structural features include interaction energies between neighbouring base pairs, the angle of twist that makes two bases of a pair non-coplanar, or DNA deformability induced by the proteins shaping the chromatin.
The autocorrelation analysis of some of these numerical sequences show characteristic periodicities in complete genomes. In fact, after detecting archaea-like regions in the thermophilic bacteria Thermotoga maritima, periodicity spectra of these regions were compared to the periodicity spectra of the homologous regions in the archaea Pyrococcus horikoshii. The revealed similarities in the periodicity were strong supporting evidence for a case of massive HGT between the bacteria and the archaea kingdoms.
Genomic context