share
I Introduction
There is a general agreement that horizontal gene transfer (HGT) is important in genome evolution. To which degree is still a matter of debate. The discussion oscillates between two extreme positions: first, the idea that the rate of transfer and its impact are of such a magnitude as to be the “essence of phylogeny” -at least for Prokaryotic organisms [Doolittle, 1999] and, on the other hand, the researchers who opine that the role of HGT in evolution has been overestimated [Kurland et al., 2003]
and that, while a matter of interest, it is not relevant when compared to other causes of genomic evolution like paralog duplication and secondary gene looses. Most probably, the real weight of HGT in evolution is posed somewhere between these extremes. Independently of the outcome of the discussion, there is a general agreement that it is relevant to detect events of HGT.
There are several proposed methods to detect HGT. They can be classified into four categories: deviant composition, anomalous phylogenetic distribution, abnormal sequence similarity and incongruent phylogenetic trees
[Eisen, 2000, Ragan, 2001, Philippe and Douady, 2003].Deviant composition methods are based on the different phenotypic characteristics among divergent genomes. They are mainly focused on bias in GC contents or codon usage [Mrazek and Karlin, 1999] and bias in the nucleotide composition in the third and first codon position [Lawrence and Ochman, 1997]. Deviant genes might exist for reasons other than HGT, and only recently transferred genes would be detected by this method [Eisen, 2000, Hooper and Berg, 2002]. Also, this group of methods normally does not detect transferred genes from phenotypically similar genomes.
Anomalous phylogenetic distribution is based on the identification of homologous genes shared by genomes in disjunct phylogenetic lineages and its absence in close relatives (in one or both lineages). However, polyphyletic gene looses and rapid sequence divergence can mislead the identification of HGT [Eisen, 2000].
Abnormal sequence similarity is based on the assumption of overall similarity as a measure of phylogenetic relatedness. Usually, BLAST searches (or other similar algorithms) are used to detect sequences in one genome more similar to sequences in divergent genomes than those sequences found in phylogenetic closer genomes (the phylogenetic relationships between genomes are set prior the analysis according to some other criterion like rRNA phylogeny). While these methods work fast, they are not fully reliable because the similarity between a gene in two different species can be explained by a number of phenomenons besides HGT. For instance, evolutionary rate variation can lead to misleading results in the identification of HGT genes, both as false positives and false negatives [Eisen, 2000].
Phylogenetic analysis is often considered to be perhaps the best way to investigate the occurrence of HGT because it remains the only one to reliably infer historical events from gene sequences [Eisen, 2000]; Accordingly, incongruent phylogenetic trees between different families of genes will be caused by HGT; however conflicting phylogenies can be a result of either artifacts of phylogenetic reconstruction, HGT or unrecognized paralogy [Zhaxybayeva et al., 2004].
In this study, we propose a method that does not depend on DNA functional traits and due to this reason is no longer correct to say that it detects HGT because it might also detect transfer of non-coding DNA (ncDNA). From now on, we will refer to horizontal genetic material transfer (HGMT). We use eight variables that can be measured over a set of windows covering whole genomes and combine them with an Artificial Neural Network (NN). The field of using DNA measurables other than those derived from protein coding requirements have been largely ignored. Up to our knowledge, there are no methods based exclusively on structural DNA traits to detect neither HGT or HGMT events.
Ii Methods
Our approach was to take a pair of DNA sequences –genomes in case of prokaryotes, chromosomes for eukaryotes-, the first one is the donor genome and the second one the acceptor. The data were taken from Genbank release 24.
To calculate the variables that characterize locally the DNA sequences, a window is placed over the chromosomes (Figure 1). The window can slide over the sequences or can be put randomly (see below).
We used a ”classic NN approach” –a backpropragation multilayer perceptron (MLP), very similar to the model reported by Uberbacher
[Uberbacher et al., 1991] (Figure 2)
The novelty and main contribution of this paper is the set of measurables we use to evaluate a DNA sequence. Following the nomenclature of Uberbacher we will call them sensors.
For the training stage of the MLP A window of fixed length (300bp unless otherwise stated) was placed repeatedly over both genomes at random independent positions and every time eight sensors were evaluated over the subsequence in the window and were used feed and train a MLP with binary output; corresponding to acceptor DNA sequences and to donor ones. For the prediction stage the window was allowed to slide along the acceptor sequence and a plot of its position against the outcome was obtained.
Ii-a Definition of the sensors
We worked with a total of eight sensors divided in three groups. The first one includes traditional measures of DNA variance, the second reflects the DNA local correlations structure and the third one is a measure of the DNA spatial structure according to the dinucleotide distribution:
-
This group comprises CG and CpG contents. There is a number of publicactions reporting the bias of these measures among different organisms. CpG is well known to tell appart prokaryal and eukaryal lineages [Shimizu et al., 1997]. Even if the underlying reasons are still unclear [Wang et al., 2004], it gives a good first clue.
-
In 1995, we proposed an index of DNA heterogeneity that disclosed different styles of genomic structural organization [Miramontes et al., 1995]. Given a DNA sequence, it is translated into the three possible binary sequences using the groupings purine-pyrimidine (), weak-strong () and amino-keto ().
the following index is calculated over each binary derivative
(1) Where stands for the number of bases followed by the base, where and are zero or one. The phenomenology behind this index is discussed in [Miramontes et al., 1995].
-
In 1992 the group of R.E. Dickerson [Quintana, 1992] reported the variability in the DNA structural angles depending on the dinucleotide steps. Their results can be summarized in the Table LABEL:Tab:01 (page 345 of the cited reference)
A C G T A L I L I C V L V L G H H L I T V H V L Matrix of angular variability according to one dimer step along the DNA sequence
H stands for high twist profile steps for two stacked consecutive base pairs along the double helix. They are characterized for having high twist, positive cup and negative roll angels. The parameters L, I, and V are the Low, Intermediate and Variable twist.
Iii Results
One of the open problems in designing a MLP for pattern detection is to set the number of neurons in the hidden layer (the number of neurons in the output layer is determined by the number of classes to classify). In order to find out the best suited to our interests, we ran several configurations and tested the resulting output with an artificial problem: To detect a fragment of
E. coli (donor) inserted in silico in a mouse (Mus musculus) chromosome (acceptor). To this end, a set of 20000 fragments of length 300 of both genomes was picked up randomly to train the MLP. The network configuration 8-5-1 yielded the results shown in Figure 3. With a 300bp sliding window and an overlap of 30bp the response of the MLP is steadily while the sliding window travel across the acceptor chromosome and then jumps to when it enters the donor insertion and goes back to for the rest of the acceptor sequence.
Once the MLP was well tuned, we carried out three case studies:
-
To detect a prokaryal insertion in a prokaryal genome. In this case we selected, for no particular reasons, the genome of Archeoglobus fulgidus as the acceptor genome and Pseudomonas aeruginosa as the donor species in a second in silico experiment. Figure 4 shows that the results are good enough to encourage one step further.
![]()
Fig. 4: Archeoglobus fulgidus (acceptor) in the horizontal axis with a sequence of Pseudomonas aeruginosa (donor) inserted in the middle. The ordinate can only be or depending whether the output of the MLP classify the sequence as an acceptor or as a donor. -
Next attempt is the detection of a real horizontal gene transfer already reported in the literature [Kroll, 1998]. It is a gene (Cu-Zn-superoxide dismutase) of Haemophylus ducreyi inserted in the Neisseria mengiditis genome. Figure 5 shows a clear detection.
![]()
Fig. 5: Neisseria mengiditis (acceptor) in the horizontal axis with a sequence of Haemophylus ducreyi (donor) inserted in the middle. The ordinate can only be or depending whether the output of the MLP classify the sequence as an acceptor or as a donor. The horizontal thick line emphasizes the region of insertion. -
Last example is a case of organelle to nuclear genome transfer. Figure 6 shows chromosome 2 of Arabidopsis thaliana as the acceptor sequence for its own mitochondrion. As the horizontal scale unit is 30bp, the inserted fragment goes from 107,653 to 116,985 which correspond to the nucleotides from approximately 3,230,000 to 3,510,000 there is then is a clear inserted fragment of the order of 270kb which coincides with the unexpected case of an organelle to nuclear transfer event [Xiaoying et al., 1999]. This case is important for our claims because approximately more than sixty percent of the Arabidopsis thaliana’s mitochondrial genome is not translated into aminoacids [Unseld et al., 1997]. Notwithstanding, our method clearly detects the transfer.
![]()
Fig. 6: Arabidopsis thaliana (acceptor) in the horizontal axis with a sequence of its own mitochondrial genome (donor) inserted in the middle. The horizontal scale unit is 30bp
Iv Discussion
In this paper we show the feasibility of using variables not related to the DNA function as measurables that can be used to detect horizontal exchange of genetic material between different species. The results are very good and encourage the further development of this line of research. It is a matter of future work to build on a complete set of variables with the minimum size but having the maximum power of resolution.
This paper also contributes to clarify the biology behind the phenomenon of horizontal gene transfer. For instance, the degree of amelioration of horizontally transferred genes is somehow related to the accuracy to which different methods detects xenologous genes. i,e., the sequencing of the chromosome 2 from Arabidopsis thaliana, has revealed a large and unexpected organellar-to-nuclear genetic material transfer event of the mitochondrial genome (20). The sequence in the nucleus is 99% identical to the mitochondrial genome, suggesting that the transfer event is very recent. Therefore, amelioration has been negligible, and the MLP clearly detects the transferred DNA. On the other hand, as shown in Figure 5, the method presented here, effectively identifies DNA that has been transferred from Haemophilus sp. (a Gamma-proteobacteria) to Neisseria meningitidis, (a member of the Beta-proteobacteria subdivision) [Kroll, 1998]. However, the high degree of spreading of the points in Figure 5 suggest that accumulated mutations since the horizontal gene transfer event, have likely ameliorated sodC in N. meningitidis. There are 136 nucleotide differences between H. ducreyi sodC and the homologous gene from N. meningitidis. If we assume equal rates of substitutions among the two sequences, then an approximate of 68 new mutations have accumulated since the horizontal transfer event in each sequences (out of 561 nucleotides in N. meningitidis sodC). This is a substantial amount of change if we compare to the number of differences among 16S rRNA in the two species (83 differences accumulated in each gene, out of 1544 nucleotides in N. meningitidis 16S rRNA) and if we take into account that the transfer event must have happened after the divergence of the two species. The extent to which the NN can go in detection when amelioration occurs will be reported elsewhere.
Acknowledgement
This research was partially supported by PAPIIT-UNAM grant IN115908. The investigation is part of Citlali’s B.Sc. dissertation in Computer Science at UNAM.
References
- [Carbone et al., 2003] Carbone, A., Zinovyev, A. and Képès F. (2003) Codon adaptation index as measure of dominating codon bias. Bioinformatics, 19, 2005-2015.
- [Doolittle, 1999] Doolittle, WF. (1999) Phylogenetic classification and the universal tree. Science, 284, 2124-2129.
- [Eisen, 2000] Eisen, JA. (2000) Horizontal gene transfer among microbial genome: new insights form complete genome analysis. Curr Opin Genet Dev, 10, 606-611.
- [Hooper and Berg, 2002] Sean D. Hooper and Otto G. Berg (2002) Detection of genes with atypical nucleotide sequence in microbial genomes. J Mol Evol 54, 365-375.
- [Kroll, 1998] Kroll JS, Wilks KE, Farrant JL, Langford PR. (1998) Natural genetic exchange between Haemophilus and Neisseria: intergeneric transfer of chromosomal genes between major human pathogens. Proc Natl Acad Sci USA, 95, 12381-5.
- [Kurland et al., 2003] Kurland, C.G., Canback, B. And O. Berg. (2003). Horizontal gene transfer: A critical view. Proc Natl Acad Sci USA, 100, 9658-9662.
- [Lawrence and Ochman, 1997] Lawrence, JG. and Ochman, H (1997) Amelioration of bacterial genomes: rates of change and exchange. J Mol Evol, 44, 383-397.
- [Mahadevan and Ghosh, 1994] Mahadevan I, Ghosh I. (1994) Analysis of E.coli promoter structures using neural networks. Nucleic Acids Res, 22, 2158-65.
- [Miramontes et al., 1995] Miramontes P, Medrano L, Cerpa C, Cedergren R, Ferbeyre G and Cocho G. (1995) Structural and thermodynamic properties of DNA uncover different evolutionary histories. J Mol Evol, 40, 698-704.
- [Mrazek and Karlin, 1999] Mrazek J, Karlin S. Detecting alien genes in bacterial genomes (1999). Ann N Y Acad Sci, 18, 314-29.
-
[Narayanan et al., 2002]
Narayanan A, Keedwell EC, Olsson B. (2002) Artificial intelligence techniques for bioinformatics.
Appl Bioinformatics, 1,1 91-222. - [Parbhane et al., 2000] Parbhane RV, Tambe SS, Kulkarni BD.(2000) ANN modeling of DNA sequences: new strategies using DNA shape code. Comput Chem, 24, 699-711.
- [Philippe and Douady, 2003] Philippe, H., Douady, CJ. (2003) Horizontal gene transfer and phylogenetics. Curr Opin Microbiol, 6, 498-505.
- [Quintana, 1992] Quintana, JR., Grzeskowiak, K., Yanagi, K., Dickerson RE. (1992) Structure of a B-DNA decamer with a central T-A step: C-G-A-T-T-A-A-T-C-G. J Mol Biol, 225, 379-95.
- [Ragan, 2001] Ragan, MA. (2001) Detection of lateral gene transfer among microbial genomes. Curr Opin Genet Dev, 11, 620-626.
- [Shimizu et al., 1997] Shimizu TS, Takahashi K, Tomita M. (1997) CpG distribution patterns in methylated and non-methylated species. Gene, 205, 103-7.
- [Uberbacher et al., 1991] Uberbacher EC, Mural RJ. (1991) Locating protein-coding regions in human DNA sequences by a multiple sensor-neural network approach. Proc Natl Acad Sci USA, 88, 11261-5.
- [Unseld et al., 1997] Unseld, M., Marienfeld, RM., Brandt P, Brennicke, A. (1997) The mitochondrial genome of Arabidopsis thaliana contains 57 genes in 366,924 nucleotides. Nat Genet, 15, 57-61.
- [Wang et al., 2004] Wang Y, Rocha EP, Leung FC, Danchin A. (2004) Cytosine methylation is not the major factor inducing CpG dinucleotide deficiency in bacterial genomes. J Mol Evol, 58, 692-700.
- [Wu, 1997] Wu CH. (1997) Artificial neural networks for molecular sequence analysis. Comput Chem, 21, 237-56.
- [Xiaoying et al., 1999] Xiaoying, L. et al. (1999) Sequence and analysis of chromosome 2 of the plant Arabidopsis thaliana. Nature, 402, 761-768.
- [Zhaxybayeva et al., 2004] Zhaxybayeva, O., Lapierre, P., Gogarten, P. (2004) Genome mosaicism and organismal lineages. Trends Genet, 20, 254-260.
Comments
There are no comments yet.