Evaluating multi-locus phylogenies for species boundaries determination in the genus Diaporthe

Introduction

Species in the ascomycete genus Diaporthe have been identified all over the world. Typically, Diaporthe species are saprobes, endophytes, or plant pathogens (Webber & Gibbs, 1984; Boddy & Griffith, 1989; Udayanga et al., 2011). Some plant pathogenic Diaporthe species are associated with cankers, diebacks, rots, spots and wilts on a wide range of plants, some of which are of economic importance as is the case of citrus, cucurbits, soybeans, eggplant, berries and grapevines (Backman, Weaver & Morgan-Jones, 1985; Merrin, Nair & Tarran, 1995; Farr, Castlebury & Rossman, 2002; Farr et al., 2002; Shishido et al., 2006). Less frequently, Diaporthe species can also cause lupinosis and other health problems in humans and other mammals (Van Warmelo & Marasas, 1972; Sutton et al., 1999; Battilani et al., 2011; Garcia-Reyne et al., 2011).

Distinction between Diaporthe species has historically been based on an approach that combined morphological information, cultural characteristics, and host affiliation (Udayanga et al., 2011). This approach made it difficult to reliably discriminate between the various members of the genus, because many of these fungi are asexual with low host specificity (Rehner & Uecker, 1994; Murali, Suryanarayanan & Geeta, 2006). As a consequence, an unnecessary increase in the number of proposed Diaporthe species occurred. This number currently stands at 977 and 1,099 for Diaporthe and 980 and 1,047 for Phomopsis in Index Fungorum (http://www.indexfungorum.org/) and Mycobank (http://www.mycobank.org/), respectively (both accessed 14 November 2016). The extinction of the dual nomenclature system for fungi raised the question about which generic name to use, Diaporthe or that of its asexual morph Phomopsis. Given that both names are well known among plant pathologists, and have been equally used, Rossman et al. (2015) proposed that the genus name Diaporthe should be retained over Phomopsis because it was introduced first and therefore has priority.

The problem of incorrect species attribution has practical consequences for the study of this genus, because accurate species identification is essential for understanding the epidemiology, for controlling plant diseases, and to guide the implementation of international phytosanitary measures (Santos & Phillips , 2009; Udayanga et al., 2011). Therefore, there was an urgent need to reformulate species identification in the genus Diaporthe (Santos & Phillips , 2009).

Advances in the areas of gene sequencing and molecular evolution over the last 50 years have led to the notion that ribosomal genes can be used to distinguish between species and study their molecular evolution (Woese & Fox, 1977). The choice of these genes comes from the fact that their function is conserved over all living organisms, which has been assumed to imply that their evolutionary rate should be roughly constant over the tree of life.

The molecular evolution studies mentioned have been used to develop general fungal classifications (Shenoy, Jeewon & Hyde, 2007) and have also been used for species reclassification in the genus Diaporthe (Santos & Phillips , 2009; Santos et al., 2011; Thompson et al., 2011; Baumgartner et al., 2013; Gomes et al., 2013; Huang et al., 2013; Tan et al., 2013; Gao et al., 2014; Udayanga et al., 2014a; Udayanga et al., 2014b). In fact, recently the ITS region of the ribosomal genes has been accepted as the official fungal barcode (Schoch et al., 2012), and its sequence is frequently used for molecular phylogeny analysis of Diaporthe species.

However, assuming that ribosomal gene sequences evolve at a uniform rate, independent of species is sometimes incorrect (Anderson & Stasovski, 1992; O’Donnell, 1992; Carbone & Kohn, 1993). In addition, due to the strong constraints imposed by ribosome function on the mutations in the sequence of ribosomal genes, close microbial species may have identical rDNA sequences, while having clearly different genomes. For example, a comparison between Cladosporium, Penicillium and Fusarium species at the NCBI Genome and GenBank databases (Schoch et al., 2012) will confirm this statement. Such considerations suggested that phylogenetic trees based on sets of genes are potentially more powerful in solving species boundaries than phylogenetic trees based on any single genes, as the former trees contain information about the simultaneous evolution of various biological processes (Olmstead & Sweere, 1994; Rokas et al., 2003).

The possibility of using full genomes to create phylogenetic trees becomes more feasible as the number of fully sequenced genomes increases. For example, the full genomic complement of genes/proteins involved in metabolism have been used to reconstruct phylogenies that provide information regarding the evolution of metabolism in various species (Heymans & Singh, 2003; Ma & Zeng, 2004; Forst et al., 2006; Oh et al., 2006). This type of genome wide phylogeny reconstruction is impossible for organisms that have not had their genomes fully sequenced and annotated. This is the case for the genus Diaporthe, for which the first genome sequencing project started in 2013 (GOLD project Gp0038530) and until now only Diaporthe species have their genome sequenced (Phomopsis longicolla, Diaporthe aspalathi, Diaporthe ampelina and Diaporthe helianthi) (Li et al., 2015; Baroncelli et al., 2016; Li et al., 2016; Savitha, Bhargavi & Praveen, 2016).

Although full genome sequences are still forthcoming for Diaporthe species, current species identification and phylogeny reconstruction in the genus are already largely dependent on molecular sequences (Santos, Correia & Phillips, 2010). The sequences more frequently used for these studies are: large subunit (LSU) of the ribosomal DNA, intergenic spacers (IGS) of the ribosomal DNA, internal transcribed spacer (ITS) of the ribosomal DNA, translation elongation factor 1-α (TEF1) gene, ß-tubulin (TUB) gene, histone (HIS) gene, calmodulin (CAL) gene, actin (ACT) gene, DNA-lyase (APN2) gene, 60s ribosomal protein L37 (FG1093) gene and mating type genes (MAT-1-1-1 and MAT-1-2-1) (Farr, Castlebury & Rossman, 2002; Farr, Castlebury & Rossman, 2002; Castlebury et al., 2003; Pecchia, Mercatelli & Vannacci, 2004; Schilder et al., 2005; Van Rensburg et al., 2006; Kanematsu, Adachi & Ito, 2007; Santos, Correia & Phillips, 2010; Santos et al., 2011; Thompson et al., 2011; Grasso et al., 2012; Sun et al., 2012; Udayanga et al., 2012; Baumgartner et al., 2013; Bienapfl & Balci, 2013; Gomes et al., 2013; Huang et al., 2013; Sun et al., 2013; Tan et al., 2013; Vidić et al., 2013; Gao et al., 2014; Udayanga et al., 2014a; Udayanga et al., 2014b; Wang et al., 2014).

However, multi-locus phylogenies for the genus Diaporthe have only been developed in the last few years (Schilder et al., 2005; Van Rensburg et al., 2006; Udayanga et al., 2012; Baumgartner et al., 2013; Gomes et al., 2013; Huang et al., 2013; Tan et al., 2013; Gao et al., 2014; Udayanga et al., 2014a; Udayanga et al., 2014b; Wang et al., 2014). In fact, creating phylogenies that include several loci is still possible only for a limited set of species from the genus Diaporthe, because not all genes have been sequenced for all tentative species. This is due to many reasons, among which the lack of resources that prevents unlimited sequencing of samples. Nevertheless, a multi-locus approach should always be used for accurate resolution of species in the genus Diaporthe.

In recent studies the maximum number of loci used was to create multi loci phylogenies seven (TEF1, TUB, HIS, CAL, ACT, APN2 and FG1093), simultaneously sequenced across approximately 80 isolates from 9 Diaporthe species (Udayanga et al., 2014a). These loci were used to establish the specific limits of D. eres. This work provides a good example of how to establish the boundaries for one species within the genus Diaporthe. However, if this is to be extended to the other species of the genus, it is important to determine which loci are the most informative to be sequenced and used in a much wider range of Diaporthe species.

With this in mind we asked which combination of frequently sequenced loci better discriminate species boundaries in Diaporthe. To answer this question, we considered the ITS, TEF1, TUB, HIS and CAL loci, which had been sequenced for 96 different Diaporthe species. This paper ranks these loci according to their contribution for improving/decreasing the resolution of Diaporthe species determination, as they are added/removed from multi-locus phylogenies analysis.

Materials & Methods

Results

We analyse 142 isolates from 96 Diaporthe species for which the ITS, CAL, TUB, HIS, and TEF1 loci had been sequenced (Table S1). The alignments for each locus were then concatenated in all possible 31 combinations of one, two three, four and five genes. Alignment characteristics for this study are reported in Table 4. Each combination was used to build a ML and ME phylogenetic trees. Each tree was bootstrapped 1,000 times and every tree used is a condensed tree with a 90% cut-off. Alignments and trees were deposited in TreeBase (Study Accession: S20343).

Best and worst resolving phylogenetic trees

The “quality” (resolution) of the individual phylogenetic trees was determined as described in methods.

Figures 1 and 2 present the condensed MP and ML trees build from the concatenated multiple alignments of the five loci, respectively. Phylograms showing all complete trees are given as (Figs. S1 and S2, respectively). These trees are the best resolving trees built for each method, as indicated by the scores shown in Table 2 for MP trees and in Table 3 for ML trees.

The increase in tree length (Table 2) and log-likelihood scores (Table 3) of the trees with the increase in number of loci indicates that resolution of the trees is directly correlated with the number of loci used to build them. This is also true for the tree scores and log-likelihood scores normalized by alignment length. Thus, the worst trees are built using the multiple alignments for only one locus. Within the one-locus trees, the best MP (Fig. 3 and Fig. S3) and ML (Fig. 4 and Fig. S4) condensed trees are shown in Figs. 3 and 4. TEF1 trees have the highest values for length and-log likelihood.

Phylogenetic informativeness and identification of species boundaries

Figure 5 shows that the TEF1 sequence is the most informative for species separation, both globally and per alignment site. In addition, we also see that the ITS sequence is the least informative to resolve Diaporthe species (Fig. 5). The five loci can be ranked from most to least informative for Diaporthe species separation as follows: TEF1 > HIS > CAL > TUB > ITS.

The dataset we used for this analysis is as close as we currently can get to a standard set of well separated Diaporthe species, taking into account that the five loci we analyse needed to be sequenced for all individuals in the set. Taking this into account, an inspection of the trees is required to understand, on top of all the statistical analyses, if species are well separated or not.

We see that, in general, the addition of a new locus to the alignment decreases the number of isolates from the same species that do not cluster together (separation errors). Therefore, the tree of 5 loci has less separation errors than 4-loci trees, which in turn have less separation errors than the 3-loci trees, and so on. As expected from our previous analysis, the TEF1 tree provides the best single locus ML tree, TEF1-TUB tree provides the best 2-loci ML tree, TEF1-TUB-CAL the best 3-locus ML tree. The results from the MP trees are qualitatively similar although, in general, these trees have more separation errors that the ML ones.

Discussion

Identifying species boundaries in organisms is a difficult task, as theoretical and practical definitions of species are not always consistent with each other (Doolittle & Zhaxybayeva, 2009; Giraud et al., 2008). While Woese & Fox (1977) suggested using ribosomal sequences to define species borders, such sequences are not always the best choice. For example, searching GenBank will reveal that some Cladosporium, Penicillium and Fusarium species cannot be differentiated using ITS (Schoch et al., 2012).

More recent work suggests that trees based on multi-loci sequence analysis (MLSA) provide more accurate estimations of phylogeny than single gene trees, if appropriate loci are used (Gadagkar, Rosenberg & Kumar, 2005; Mirarab, Bayzid & Warnow, 2016). Briefly, MLSA concatenates sequence alignments from multiple genes and uses the concatenated sequences to determine phylogenetic relationships. This method appears to more optimally resolve the phylogenetic position of species in the same or in closely related genera (Hanage, Fraser & Spratt, 2006). An increase in the number of loci used to build MLSA phylogenetic trees positively correlates to sensitivity and accuracy in species separation (Rokas et al., 2003; Udayanga et al., 2011). In contrast, increasing the number of species in the alignment leads to a decrease in the ability to separate them accurately, unless a higher number of appropriate loci are used to maintain the quality of that separation (Bininda-Emonds et al., 2001; Kim, 1998; Poe & Swofford, 1999; Rokas et al., 2003; Udayanga et al., 2011). The choice of appropriate loci to be used in such trees can be optimized in genera with a large number of sequenced genomes, because in such cases it is possible to make full genome studies to identify the best set of loci to separate species. Nevertheless, the amount of information that must be analysed for doing so could become prohibitive (Thangaduras & Sangeetha, 2013).

The choice of appropriate loci that optimizes species separation is harder when fully sequenced genomes are not available, as is the case for the genus Diaporthe. Nevertheless, MLSA phylogenetic studies of Diaporthe species have been done using loci that have been chosen in a more or less ad hoc manner, by taking into account how conserved they were in different fungal genus (Baumgartner et al., 2013; Gao et al., 2014; Gomes et al., 2013; Huang et al., 2013; Schilder et al., 2005; Tan et al., 2013; Udayanga et al., 2012; Udayanga et al., 2014a; Udayanga et al., 2014b; Van Rensburg et al., 2006; Wang et al., 2014). In general, these studies show that MLSA phylogenetic trees provide higher resolution for Diaporthe species than single locus phylogenetic trees (Huang et al., 2013; Udayanga et al., 2012; Van Rensburg et al., 2006).

The current study addresses the problem of which loci are best for accurate species separation in the genus Diaporthe in a systematic manner. Walker et al. (2012) performed a similar study. While we use five non-coding loci to study species separation in Diaporthe, those authors employed two single copy protein-coding genes (FG1093 and MS204) to study species separation in Sordariomycetes. While Walker et al. (2012) analysed various aspects of codon conservation and substitution rates, these analyses are meaningless for our sequence dataset. The use of non-coding sequences is favoured in Diaporthe species separation because coding sequences are typically too conserved to allow for appropriate separation within the genus.

The major contributions of this paper are two-fold. First, our work confirms that the quality of species separation in phylogenetic trees increases with the number of loci used to build phylogenetic trees. Second and more importantly it identifies the best combination of loci that one should use for building those phylogenetic trees, if only one, two, three, or four loci can be sequenced. To achieve this, we took the most commonly sequenced loci for 142 Diaporthe isolates and studied which loci optimize species differentiation in the genus. We chose only loci that are commonly sequenced for members of the genus. Then, we selected a sequence dataset that was experimentally validated by others (Castlebury et al., 2003; Van Niekerk et al., 2005; Santos et al., 2011; Gomes et al., 2013) before being deposited in GenBank. Whenever possible we favoured sequences from ex-type isolates and produced via low throughput, high fidelity, sequencing methods. In addition, our sequence selection maximized intraspecific sequence variation, which in turn maximizes the possibility that intra-specific hyperdiversity could be higher than interspecific diversity. Thus, species separation through phylogenetic trees in our sample is made more difficult by our sequence selection, making our analysis more robust. In this paper we only show and analyse condensed MP and ML trees, using a cut-off of 90%, which means that our trees are very robust to gene order, as a significant amount of bootstrapping was used to calculate them. In fact, to test that, we performed a side experiment where we changed the order of the locus sequences in the alignments and recalculated the trees (Fig. S5).

We found that species differentiation is optimized by creating phylogenetic trees built from the multiple sequence alignment of five loci: TEF1-TUB-HIS-CAL-ITS. However, little information is lost when ITS locus is removed and only the other four loci are used to simultaneously build the phylogeny. In addition, we also provide researchers with a ranking of best loci to sequence if only one, two, three or four of the loci can be sequenced.

It may be surprising that the ribosomal ITS locus is the least informative of the five loci when it comes to separating Diaporthe species. However, Santos, Correia & Phillips (2010) found that the ITS region in Diaporthe is evolving at much faster rates than TEF1 or even MAT genes. Hence, what seems to be happening is that ITS sequences present a wider variation than is advisable for creating precise species boundaries. Therefore a slowly evolving gene region should be utilized in order to establish precise species limits (Udayanga et al., 2012).

DNA barcoding (Kress et al., 2014) refers to the use of standard short gene sequences to identify species. The use of DNA barcoding implies that an effort should be made to standardize the use of the loci for phylogenetic studies. ITS is the official DNA barcode region in fungi (Schoch et al., 2012). This work supports previous studies whose results suggest that using ITS as a standard for species separation in fungi should be discontinued (Gomes et al., 2013; Thangaduras & Sangeetha, 2013). Our results strongly recommend that TEF1 should be used instead, at least in the genus Diaporthe. This is consistent with and further develops previously published results, which proposed either TEF1, HIS, or APN2 as alternative locus for barcoding in the genus (Santos, Correia & Phillips, 2010; Udayanga et al., 2014b). However, Gomes et al. (2013), using Bayesian analysis, consider HIS and TUB as best resolving genes. Nevertheless, considering that Gomes et al. (2013) use shorter sequences than those used here, one is tempted to cautiously analyse and reinterpret their conclusions.

Despite the TEF tree appears to be a better species separator than the 5-loci tree, the true is that, the alignment used to build the 5-loci tree is roughly five times larger than that for the TEF tree. This means that, with a larger number of positions, there is bound to be more variability in the bootstrapping of the 5-loci tree than in the bootstrapping of the TEF tree. Hence, the observation that the TEF give better resolution than 5 loci results from a statistical artefact. This fact occurs when focusing on the D. eres complex clade. For example, in the case of the D. eres complex, all the species are grouped in the same clade in both cases (D. alleghaniensis, D. alnea, D. celastrina, D. bicincta, D. eres, D. neilliae and D. vaccinii). However, in the 5-loci trees the resolution of this species complex is better. This is especially important as phylogenetic analyses of the D. eres complex often revealed ambiguous clades with short branch and moderate statistic supports due to their high variability. Udayanga et al. (2014a) studied this problematic by using different genes, whose sequences are not available for the other Diaporthe species we consider. Therefore, we could not incorporate their data in our study. We also note that one possible explanation for the observation that some species of the D. eres complex do not “group” in the same clade could be due to the fact that they are not really D. eres. However, to test that, we would need to actually obtain samples of the complex, re-sequence and analyse them in order to clarify the species boundaries in this group.

The problem of species boundary identification is very relevant in the genus Diaporthe, where a general taxonomic revision based on molecular analysis is probably overdue. Such a revision could then be used to improve the annotation of sequences in public databases, such as GenBank. For example, many of the sequences we use in our analysis are still assigned to species that have already been reclassified. This also emphasizes that a standard procedure with minimal information required for submitting new Diaporthe species needs to be put in place in order to avoid unnecessary creation of new species (Udayanga et al., 2014b). Furthermore, as also suggested by Gomes et al. (2013) we feel that this revision should be made using molecular data. Any new Diaporthe species report should be accompanied by molecular data that supports the identification of the individual as a new species. In addition, we feel that a proper taxonomic revision of the genus should also consider morphological descriptions and epitypification of species as previously suggested (Gomes et al., 2013; Udayanga et al., 2014b).

Conclusions

Our results indicate that:

• In order of effectiveness the best sets of loci for resolving Diaporthe species are TEF1-TUB-CAL-HIS-ITS, TEF1-TUB-CAL-HIS, TEF1-TUB-CA L, TEF1-TUB and TEF1.

• The TEF1 locus is a better candidate for single locus DNA barcoding in the genus Diaporthe than the ITS locus.

• Multi-loci DNA barcoding will provide a more accurate species separation in the genus than single locus barcoding. Furthermore, a four loci barcoding including TEF1-TUB-HIS-CAL will be almost as effective as a five loci barcoding including ITS-TEF1-TUB-HIS-CAL.

Supplemental Information