A simple, fast, and accurate method of phylogenomic inference.
Wu M, Eisen JA. · 2008
Read A simple, fast, and accurate method of phylogenomic inference. by Wu M, Eisen JA. (2008) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc2760878.txt.
Introduction
Wu and Eisen presented AMPHORA in 2008, an automated pipeline for phylogenomic inference that builds organism trees from many protein markers at once. Microbial classification long rested on a single ribosomal RNA gene, the gold standard, yet single genes can misplace distant organisms due to composition bias. Whole genomes now supply every protein gene, but cloning and aligning many markers by hand was slow.
AMPHORA automates marker choice, alignment, and tree building for large data sets, using protein sequences that are less prone to nucleotide bias. It lets researchers corroborate ribosomal trees with genome-wide evidence at scale. The pipeline aims to make robust, multi-gene phylogeny a routine step in microbial systematics and ecology rather than a bespoke effort.
Catalog record
| CREATOR | Wu M, Eisen JA. |
|---|---|
| YEAR | 2008 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 46 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A simple, fast, and accurate method of phylogenomic inference By Martin Wu, Jonathan A Eisen Published in: Genome Biology DOI: 10.1186/gb-2008-9-10-r151 ABSTRACT An automated pipeline for phylogenomic analysis (AMPHORA) is presented that overcomes existing limits to large-scale protein phylogenetic inference. BACKGROUND Since the 1970s the use of small subunit (SSU) rRNA (SSU rRNA) sequences has revolutionized microbial classification, systematics, and ecology. The SSU rRNA gene has become the most sequenced gene, with hundreds of thousands of its sequences now deposited in public databases. It has become the current 'gold standard' in microbial diversity studies, and for good reasons. For one, it is present in all microbial organisms. For another, the gene sequence is highly conserved at both ends. This enables one to obtain nearly full-length SSU rRNA gene sequences by polymerase chain reaction amplification using 'universal' primers and without having to isolate and culture the organism in question. Until very recently, the vast majority of microbes were identified and classified only by recovering and sequencing their SSU rRNA genes. This single sequence of approximately 1.5 kilobases is often the only information we have about the organism from which it came - the only way we know that it exists in the natural environment. Although the SSU rRNA gene has been extremely valuable for phylogenetic studies, it has its limitations. For example, it has been well documented that evolutionarily distant SSU rRNA genes that are similar in nucleotide composition have been consistently - but nevertheless incorrectly - placed close together in phylogenetic trees [1,2,1]. Furthermore, inferring the phylogeny of organisms from any single gene carries some risks and must be corroborated by the use of other phylogenetic markers. Many researchers turned to protein encoding genes such as EF-Tu, rpoB, recA, and HSP70 [3]. Because protein sequences are conserved at the amino acid level instead of at the nucleotide level, phylogenetic analyses of protein sequences are in general less prone to the nucleotide compositional bias seen in SSU rRNA [2,4-6]. In addition, the less constrained variation at the third codon position allows these genes to be used in studies of more closely related organisms. However, because of difficulties in cloning protein encoding genes from diverse species, SSU rRNA remained the gold standard. The situation changed with the advent of genomic sequencing. Each complete genome sequence brings with it the sequences for all protein encoding genes in that organism. Now, not only can one build gene trees based on a favorite protein encoding gene, but also one has the option to concatenate multiple gene sequences to construct trees on the 'genome level'. Possessing more phylogenetic signals, such 'genome trees' or 'super-matrix trees' are less susceptible to the stochastic errors than those built from a single gene [7]. Recent studies attempting to reconstruct the tree of life have demonstrated the power of this approach [8,9] (for review [10]). Likewise, genome trees have also been used successfully to reassess the phylogenetic positions of individual species [11,12]. It is worth pointing out, however, that the genome trees are still susceptible to systematic errors caused by compositional biases, unrealistic evolutionary models, and inadequate taxonomic sampling [7,13,14]. Despite its demonstrated usefulness, phylogenetic inference based on protein markers has been limited in application, mainly because of the formidable technical difficulties inherent in this approach. Typically, molecular phylogenetic inference involves three steps: retrieval of homologous sequences, creation of multiple sequence alignments, and phylogenetic tree construction. Because only characters of common ancestry can be used to infer the evolutionary history, the most critical step is sequence alignment, in which sequences are overlaid horizontally on each other in such a way that, ideally, each column in the alignment would only contain homologous characters (amino acids or nucleotides). To ensure this positional homology, the alignments must be curated - a process that evaluates the probable homology of each column or position in the alignments. Positions for which the assignment of homology is uncertain are then excluded from further analysis by masking [15]. Judicious masking increases the signal-to-noise ratio and often improves the discriminatory power of the phylogenetic methods [16]. Unfortunately, curation requires skilled manual intervention, thus making it impractical to process suitably the massive amount of genome sequence data now available. Frequently, masking is simply ignored. Automated masking to remove alignment positions that contain gaps or that have a low degree of conservation has not been satisfactory. For example, using these criteria and given a set of ad hoc parameters such as the minimum block length, GBLOCKS automatically selects conserved blocks from multiple sequence alignments for phylogenetic analysis. However, trees constructed using GBLOCKS-treated alignments have been found to have dramatically weaker support, possibly because of excessive removal of informative sites by the program [17]. In addition, although many programs are available to automate the creation of multiple sequence alignments, their use for the de novo alignment of a large protein family is still fairly time consuming. To overcome these problems, we have developed an automated pipeline for building concatenated genome trees using multiple protein markers, thus making this powerful method applicable on a larger scale. Our pipeline can rapidly and accurately generate highly reproducible multiple sequence alignments for a set of selected phylogenetic markers. More importantly, unlike previous automated methods [9] it can mask the alignments with quality equivalent to that of curation by humans. The same pipeline can also be applied to metagenomic data analyses. In metagenomics or environmental genomics, natural populations of microbes are collected from the environment; their DNAs are cloned and directly sequenced. One fundamental goal of metagenomics is to determine who is present in the community and what they are doing. Phylogenetic analysis of markers present in these collected samples can be very informative in revealing who is there. If the marker happens to be part of a larger assembled sequence fragment, then the entire fragment can be anchored by that marker to a specific taxonomic clade. In this way, environmental shotgun sequences can be sorted into taxon-specific 'bins' in silico, thereby allowing us to determine who is doing what. The most striking finding to date from this approach was the discovery of a proteorhodopsin gene in bacteria, a homolog of the bacteriorhodopsin gene previously found only in some archaea. In this case, the gene could be anchored within the bacteria because it was found to be associated with a bacterial SSU rRNA gene [18]. However, because the SSU rRNA gene constitutes only a tiny fraction of any genome, the probability that any given sequence fragment can be anchored to a specific taxonomic clade by using this one gene is small. Thus, phylotyping of metagenomic data can greatly benefit from the use of alternative phylogenetic markers such as the multiple protein markers described below. In this paper, we introduce AMPHORA (a pipeline for AutoMated PHylogenOmic infeRence) and demonstrate two significant applications: building a genome tree from 578 complete bacterial genomes that are available at the time of the study and identifying bacterial phylotypes from metagenomic data collected from the Sargasso Sea. RESULTS AND DISCUSSION The AMPHORA pipeline Introduction With the rapid increase in available genomic sequence data, there is an ever-urgent need for automated phylogenetic analyses using protein sequences. However, automation is frequently accompanied by reduced quality. We