Automated Phylogenetic Taxonomy: An Example in the Homobasidiomycetes (Mushroom-Forming Fungi)

David S. Hibbett, R. Henrik Nilsson, Marc Snyder, Mario Fonseca, Janine Costanzo, Moran Shonfeld

Systematic Biology · 2005 · 75 citations · 29 references

DOIFull text

Open access

Concepts

Abstract

Botanical Institute, Göteborg University, Box 461, 405 30, Göteborg, Sweden; E-mail: henrik.nilsson@botany.gu.se Systematists confront a truly formidable task, which is to construct a comprehensive, phylogenetically accurate classification for all of life. Progress toward this goal has accelerated in recent years through analyses of DNA and protein sequences. The dramatic growth of molecular phylogenetics prompts us to ask whether established taxonomic practices are equipped to deal with the rapidly accumulating data. Here, we consider the current status of phylogenetics and classification in the homobasidiomycetes (mushroom-forming fungi), which includes roughly 17,000 described species (Kirk et al., 2001). We find that the available data are not being integrated, and that there is a significant gap between current taxonomy and understanding of phylogenetic relationships. To close this gap, we suggest that there is a need for new approaches that use the tools of bioinformatics to automate the process of phylogenetic analysis and classification, and we describe a prototype software package that we have developed for this purpose. The tools that we have created are applied to homobasidiomycetes, but can be adapted for use in any group of organisms. Taxonomy of homobasidiomycetes has been revolutionized through the use of molecular techniques. The most widely sampled locus for studies at high to moderate taxonomic levels (i.e., genera and higher) in homobasidiomycetes is a region of about 1000 base pairs at the 5′ end of the nuclear-encoded large subunit ribosomal DNA (nuc-lsu rDNA), which can be aligned across distantly related taxa, but can often distinguish closely related species (e.g., Moncalvo et al., 2002). As of November 2004, GenBank (http://www.ncbi.nlm.nih.gov/Genbank/index.html) contained about 4204 nuc-lsu rDNA sequences of homobasidiomycetes, representing 2167 distinct names (not all of the sequences are identified to the species level). However, the most comprehensive analyses so far include only 877 sequences (representing 877 species; Moncalvo et al., 2002) or 656 sequences (representing approximately 646 species; Binder et al., 2005), which indicates that the currently available data are not being integrated in comprehensive, simultaneous analyses. This is unfortunate, because comprehensive analyses are needed to determine the composition of clades and develop detailed phylogenetic hypotheses. Comprehensive analyses are also needed to detect mislabeled sequences, which are common in groups like fungi, where identification is difficult (Binder et al., 2005; Bridge et al., 2003; Vilgalys, 2003). Finally, analyses with multiple accessions of morphologically defined species are needed to reveal cryptic lineages. To enable comprehensive phylogenetic analyses in homobasidiomycetes, it is important that the database of nuc-lsu rDNA sequences be expanded and integrated. However, nuc-lsu rDNA alone is not sufficient to reconstruct homobasidiomycete phylogeny in all of its details, as has been shown in many studies (e.g., Binder and Hibbett, 2002). Fortunately, analyses of multilocus data sets (supermatrices) are becoming more common (e.g., Matheny et al., 2002; Wang et al., 2004), and these are providing improved resolution of major clades of homobasidiomycetes and other groups of fungi (Binder and Hibbett, 2002; Lutzoni et al., 2004). Numerous single-gene and multilocus studies have contributed to our current understanding of the major groups of homobasidiomycetes. In 2001, Hibbett and Thorn reviewed the results of 26 studies and proposed a “preliminary phylogenetic outline,” in which the homobasidiomycetes were divided into eight mutually exclusive major clades that were given informal names (e.g., “euagarics clade”). Subsequently, several other putatively independent clades have been discovered, including the “athelioid clade,” “corticioid clade,” “trechisporoid clade,” and “Gloeophyllum clade” (Hibbett and Binder, 2002; Langer, 2002; Larsson et al., 2004; Binder et al., 2005). There may be other major clades waiting to be discovered. For example, Jaapia argillacea and several other species of resupinate basidiomycetes cannot be placed into any of the twelve clades named above based on currently available sequences (Larsson et al., 2004; Binder et al., 2005). Analyses of environmental sequences also suggest that there are major groups of homobasidiomycetes that have not yet been described (Vandenkoornhuyse et al. 2002; Schadt et al., 2003). This is not surprising, given that only about 70,000 species of fungi have been described (Kirk et al., 2001), whereas it is estimated that there are as many as 1.5 million extant species (Hawksworth, 2001). The fungal taxonomic literature is highly dispersed, which makes it difficult to assess the extent to which recent advances in phylogenetic reconstruction have been incorporated into currently accepted classifications. However, there is one major reference that provides a measure of the current status of fungal taxonomy. The Dictionary of the Fungi series presents a unified classification for all groups of fungi down to the level of genera, and is now in its ninth edition (Kirk et al., 2001). Between the eighth and ninth editions, the classification of homobasidiomycetes in the Dictionary was revised extensively, in response to progress in phylogenetic reconstruction (Hawksworth et al., 1995; Kirk et al., 2001). The eighth edition placed the homobasidiomycetes in 26 orders in the subclass Holobasidiomycetidae, whereas the ninth edition placed most of the homobasidiomycetes in eight orders in the subclass Agaricomycetidae. The ninth edition of the Dictionary of the Fungi represented a major, episodic advance toward a phylogenetic classification of homobasidiomycetes. Nevertheless, based on current views of homobasidiomycete phylogeny (Hibbett and Thorn, 2001; Weiss and Oberwinkler, 2001; Larsson et al., 2004; Binder et al., 2005), the Agaricomycetidae and at least four of the orders within the Agaricomycetidae are probably not monophyletic (Table 1). The most problematic order is the Polyporales, which includes representatives of nine different clades. Comparison of higher-level classification of homobasidiomycetes in the Dictionary of the Fungi versus current hypotheses of phylogenetic relationships, with exemplar genera. Genera followed by question marks are of uncertain placement. Comparison of higher-level classification of homobasidiomycetes in the Dictionary of the Fungi versus current hypotheses of phylogenetic relationships, with exemplar genera. Genera followed by question marks are of uncertain placement. It is obviously unfair to expect the classification in the Dictionaryof the Fungito reflect the most recent phylogenetic analyses, especially those that appeared after the latest edition of the Dictionary was published. Indeed, the discrepancy between the Dictionary classification and our current understanding of phylogeny argues for the abandonment of print as a medium for communicating the most up-to-date taxonomy (in fact, on-line classifications already exist for both Basidiomycota, http://www.mycokey.com/AAU/Systematics/SystematicsBasi.html, and Ascomycota, http://www.umu.se/myconet/Myconetxxxx.html). There is evidence that the limitations of print are not the only impediments to translation of phylogenetic trees into taxonomy, however. The International Code of Botanical Nomenclature (Greuter et al., 2000) may also be part of the problem (Hibbett and Donoghue, 1998). The constraints on taxonomic progress imposed by the Code (Greuter et al., 2000) are illustrated by the controversy over the classification of the genus Coprinus, a familiar assemblage of “inky cap” mushrooms. In 1994, and repeatedly thereafter, Coprinus was shown to be polyphyletic (Hopple and Vilgalys, 1994; Redhead et al., 2001, and references therein). Redhead et al. (2001) suggested that Coprinus s. lat. should be divided into Coprinus s. str., Coprinellus, Coprinopsis, and Parasola, but this proposal has become mired in a nomenclatural debate concerning conservation and typification of generic names (Jørgensen et al., 2001; also see discussion at http://www.cbs.knaw.nl/nomenclature/index.htm). Consequently, Coprinus s. lat. remains an “accepted” taxon (Kirk et al., 2001) fully a decade after it was shown that it is not a clade. The example of Coprinus demonstrates that the challenges of translating phylogenetic hypotheses into taxonomy can be especially difficult at taxonomic levels at or below the rank of family, because of nomenclatural complexities. Problems such as the one posed by Coprinus will only increase in number as more genera are sampled intensively. For example, in the study of Binder et al. (2005), 34 genera were resolved as nonmonophyletic, and in the study of Moncalvo et al. (2002), 44 genera were resolved as nonmonophyletic. In summary, homobasidiomycete taxonomy is blessed with an abundance of data and analyses, but lacks efficient mechanisms to integrate the emerging data into comprehensive phylogenetic trees, or translate trees into classifications. Over 90% of the homobasidiomycete nuc-lsu rDNA sequences in GenBank have been published since 1998, and there is no reason to expect the rate of data acquisition to decline in the near term. As the sequence database grows, there is a danger that the gap between taxonomy and understanding of phylogenetic relationships will widen. To prevent this from happening, systematists may wish to consider new tools that automate the process of phylogenetic classification. We have developed a Perl script package called mor that performs automated phylogenetic taxonomy in homobasidiomycetes. In brief, mor is a Web-accessible, interactive program that compiles and screens nuc-lsu rDNA sequences from GenBank, constructs phylogenetic trees, and translates trees into classifications (Fig. 1). Central to mor is the use of node-based phylogenetic taxon definitions, which allow the composition of taxa (clades) to be continuously updated as the underlying phylogeny changes. The mor package is available on an open-source basis and can be modified to work with any gene and any group of organisms. The following paragraphs provide an overview of the operations performed by mor. More detailed information is provided in the program documentation available at the mor website (http://mor.clarku.edu). Simplified flow chart representing operations performed by mor, including sequence acquisition and screening (left), alignment and phylogenetic analyses (center), and extraction of clade composition information (right). The mor software runs in a Linux environment (presently on a 1.6-GHz AMD Athlon PC). Upon being evoked on a weekly basis by the operating system, mor uses BioPerl routines (http://bioperl.org) to connect to GenBank and retrieve new sequences using a Boolean search string tailored to match only homobasidiomycete nuc-lsu rDNA sequences between 800 and 1500 base pairs. Downloaded sequences are screened according to several criteria, including percentage of ambiguous sites (sequences with more than 3% of the positions reported as “n” are rejected on the assumption that they are of low quality), and occurrence of putatively conspecific sequences with at least 99.5% identity in the dataset (values of these screening parameters are arbitrary and can be adjusted by modifying the Perl script). The latter criterion is intended to prevent redundant sequences from accumulating in the data set, while still allowing sequences representing cryptic species to be added. Sequences that pass these tests are subjected to a final screening procedure using a profile hidden Markov model constructed with HMMER 2.3.2 (Eddy, 1998; http://hmmer.wustl.edu/) based on a reference alignment of 1275 homobasidiomycete nuc-lsu rDNA sequences. For each candidate sequence, HMMER calculates a bit score, which reflects whether the candidate sequence is a better match to the profile model or a null model of nonhomologous sequences, and an E-value, which reflects the number of false positives expected at that bit score. Threshold values for bit scores and E-values were determined by matching 200 other homobasidiomycete sequences and 200 nonhomobasidiomycete sequences to the reference alignment using the hmmpfam component of HMMER. Following the screening procedure, the accepted and rejected sequences are stored in separate Web-accessible MySQL databases (http://www.mysql.com). Reasons for rejection are noted in a rejected sequences viewer. Accepted sequences are aligned using Clustal W (Thompson et al., 1994) in profile mode (optionally, Clustal W can be set to perform a global alignment). Aligned sequences are analyzed by PAUP* 4.0b10 (Swofford, 2003) using an unconstrained 37% jacknife neighbor-joining (NJ) analysis, which provides a crude measure of clade support, followed by a constrained equally weighted heuristic maximum parsimony (MP) analysis. The jacknife NJ analysis uses 100 replicates and maximum likelihood distances calculated with the GTR model. The MP analysis uses a 202-sequence backbone monophyly constraint (Fig. 2) that was constructed by hand using MacClade 4.0 (Maddison and Maddison, 2001), and which reflects major groupings based on prior studies, including multilocus analyses (Hibbett and Thorn, 2001; Binder and Hibbett, 2002; Larsson et al., 2004; Lutzoni et al., 2004; Binder et al., 2005). The backbone constraint tree does not include representatives of the athelioid clade, trechisporoid clade, or Auriculariales s. lat., because their higher-level placements are poorly resolved. Two hundred two-sequence backbone monophyly constraint tree used by mor. Major clades are bracketed and labeled. Several of the clades delimited in mor (Fig. 3) are not represented in this constraint tree, including the trechisporoid clade, athelioid clade, Auriculariales, and Sebacinales. These groups were excluded from the backbone constraint because their higher-level placements are uncertain. For the MP analysis, mor uses a “twin constraint” search strategy, which eliminates the need to construct a starting tree from scratch through stepwise taxon addition and to perform the early rounds of branch-swapping in each iteration of the program. At the outset, this search protocol requires the presence of a data matrix, one treefile containing a “general constraint” tree (i.e., the 202-sequence backbone constraint tree described above), and another treefile containing a “temporary constraint,” which is one of the most parsimonious trees generated in the previous iteration of mor. The search proceeds as follows: Update the starting dataset (i.e., download, screen, and add new sequences to the existing alignment). Execute the data set and load the temporary constraint tree as a backbone monophyly constraint; add the new sequences to this tree by stepwise taxon addition, but do not perform branch swapping. Load the general constraint (Fig. 2) as a backbone monophyly constraint (i.e., replace the current constraint); perform a heuristic search using the tree in memory as a starting tree for branch swapping (presently mor is set to perform TBR for 24 hours). Write one of the resulting trees to the temporary constraint treefile (i.e., overwrite the old temporary constraint treefile); this becomes the starting tree for the next run of the program. Calculate a 50% majority-rule consensus tree. Go to step 1. The majority-rule consensus MP tree and the jackknife NJ tree are converted to tree graphics in ASCII characters and are made available on the mor site through the Apache Web server (http://httpd.apache.org). Terminals in the trees are labeled with complete genus and species names and GenBank accession numbers, and are provided with links that allow viewers to toggle between the parsimony and NJ trees, view the GenBank accession directly, make and view comments, or search for images on the internet. Each week, the current alignment and MP and NJ trees are archived on the mor website (examples of these files have been deposited at the Systematic Biology website http://www.systbiol.org). The routines described above produce comprehensive, continuously updated phylogenetic trees, but they do not generate taxonomy. For this purpose, mor includes a “clade parser,” which delimits the contents of individual clades, and thus translates trees into classifications. The clade parser takes as its input a tree in Newick format (http://evolution.genetics.washington.edu/phylip/newicktree.html), representing the 50% majority-rule consensus tree from the MP analysis, and sets of up to 20 sequences identified by GenBank accession numbers that act as “specifiers” in node-based taxon definitions (de Queiroz and Gauthier, 1992). clade is thus defined as the most recent common of its and all of its (i.e., the least monophyletic group that all of the To determine the contents of each clade, the clade parser the and in the Newick from the the clade parser the tree in a to the each that is the clade parser the tree in a toward the starting from the in the Newick a that was on the is that is as the most recent common of the clade. is to all of the in the clade, which are in a “clade is which presents a tree (in the as the MP and NJ a of all sequences links to literature and and a to a format (Maddison et al., with the aligned sequences and tree in Newick format for the clade. The tree on the clade is using the component of the package The information on the clade is updated each as the underlying phylogeny changes. of recent in the of all clades is and the of such are archived and the mor In addition, an is which all of the sequences that are not placed in any defined clade. may add new clades to mor by a of with a for the clade and a set of As of this mor has sequences, which is of the homobasidiomycete nuc-lsu rDNA sequences in The homobasidiomycete nuc-lsu rDNA sequences were not because they do not the criterion base The screening of of the sequences, most often because they are more than 99.5% to conspecific sequences already in the dataset or include many ambiguous sites Two hundred sequences have been aligned and which species in genera of the described species of and parsimony trees of have been clades have been based on major clades by Hibbett and Thorn Larsson et al. and Binder et al. (Fig. The clades that have been delimited in mor have been given informal with their use in the of Hibbett and Thorn Larsson et al. and Binder et al. it should be to clades in mor in a that will to taxon definitions the of the MP tree by mor on The clades that have been delimited in mor are labeled and the that are with in are the numbers of sequences in each The trechisporoid clade is within clade, and the athelioid clade is within the clade. The positions of the group and the group are not defined are probably as in the fully expanded view of the tree is available at of the by to be which is not the of the dataset and the limitations of the analysis. For example, several of the genera and which are and and which are are placed as close of the Auriculariales s. lat., which is highly (Fig. Nevertheless, in the composition of the clades with their as estimated in the phylogenetic studies The clade of homobasidiomycetes delimited in mor is the clade which is represented by sequences, whereas the is the athelioid clade, which is represented by sequences. The mor demonstrates that the of phylogenetic taxonomy can be This also demonstrates an of taxonomy to taxonomy, which is that taxa in a are to the groups delimited by mor be using However, that another of classification, which is that the names of taxa are in the of in tree This is for automated taxonomic because trees may from to week, resulting in that the of be that detect between the of and the relationships clades, but automated to such are probably not a shown by the Coprinus taxonomy is to be fully it will be to a of classification. that updated taxonomy because the of taxa be their phylogenetic definitions be This problem be by trees, as is in mor. Systematists have to to taxonomic in the of a phylogenetic tree, identified by its of This not be so different from the in which the of taxa often cannot be a taxonomic work is example, et al. is not to Kirk et al. are in the alignment and phylogenetic reconstruction used in mor. For example, the parsimony analysis in mor is to being by it to one heuristic and should be by that have been for large such as the parsimony with such mor is to provide analyses, because of the imposed by such a large data set, as as its on a locus (nuc-lsu Nevertheless, mor can into phylogeny resulting from more analyses, including multilocus studies, which can be used to of the general of mor is that it does not the large database of sequences from the of is more than nuc-lsu rDNA and it has become a region for studies at low taxonomic levels and for identification in molecular and 2001). As of November 2004, there were at least homobasidiomycete sequences in of these data to mor will resolution of the clades of homobasidiomycetes. However, is to be aligned across all groups of homobasidiomycetes, so it will be to data sets and trees, and with the nuc-lsu rDNA tree using et al., 1998; 2004). major of such analyses will be to develop automated for of and nuc-lsu rDNA sequences. with the of the classification generated by mor be to that are represented by sequences in This also be because taxonomic have an tree which be with trees using of sequences and named species will be in where the has not been in it should be to develop automated that produce continuously updated classifications that the of and based on molecular data and taxonomy. We are to the which this work through the the of including for that and We also Binder, and an for their comments, and the of and of for to

References

29