Many biologists agree that a phylogenetic tree of relationships should be the central underpinning of research in many areas of biology. Comparisons of plant species or gene sequences in a phylogenetic context can provide the most meaningful insights into biology. This important realization is now apparent to researchers in diverse fields, including ecology, molecular biology, and physiology (see recent papers in Plant Physiology, e.g. Hall et al., 2002a; Doyle et al., 2003). Examples of the importance of a phylogenetic framework to diverse areas of plant research abound (for review, see Soltis and Soltis, 2000; Daly et al., 2001). One obvious example is the value of placing model organisms in the appropriate phylogenetic context to obtain a better understanding of both patterns and processes of evolution. The fact that tomato (Lycopersicon esculentum) and other species of this small genus actually are embedded within a well-marked subclade Solanum (and, hence, are more appropriately referred to as species of Solanum; tomato has been renamed as of Solanum lycopersicon; e.g. Spooner et al., 1993; Olmstead et al., 1999) is a powerful statement that is important to geneticists, molecular biologists, and plant breeders in that it points to a few close relatives of S. lycopersicon (out of a genus of several hundred species) as focal points for comparative genetic/genomic research and for crop improvement. Snapdragon (Antirrhinum majus) was historically part of a broadly defined Scrophulariaceae, a family that is now known to be grossly polyphyletic (i.e. not a single clade). Phylogenetic studies indicate that Scrophulariaceae should be broken up into several families (Olmstead et al., 2001), and snapdragon and its closest relatives are part of a clade recognized as the family Plantaginaceae.

A phylogenetic framework has revealed the patterns of evolution of many morphological and chemical characters, including complex pathways such as nitrogen-fixing symbioses, mustard oil production, and chemical defense mechanisms (for review, see Soltis and Soltis, 2000; Daly et al., 2001). However, the importance of phylogeny reconstruction applies not only to the organisms that house genes but also to the evolutionary history of the genes themselves. For example, are the genes under investigation the members of a single well-defined clade, all members of which appear to descend from a recent common ancestor as a direct result of speciation (orthologous genes), or do the sequences represent one or more ancient duplications (paralogous genes; see also Doyle and Gaut, 2000)? Gene families are, of course, the norm in studies of nuclear genes, but investigators are often bewildered by the diversity of genes encountered in a survey of a family of genes from a diverse array of plants. Phylogenetic methodology offers several solutions by permitting inferences of putative orthology among a set of sequences.

Examples of the phylogenetic analysis of gene families abound (e.g. genes encoding: heat shock proteins, Waters and Vierling, 1999; phytochrome, Kolukisaoglu et al., 1995; Mathews and Sharrock, 1997; and actin, McDowell et al., 1996). A noteworthy recent example involves MADS box genes, which encode transcription factors that control diverse developmental processes in plants. Some of the best known examples of MADS box genes include the A, B, and C class floral genes that control the identity of floral organs (for review, see Ma and dePamphilis, 2000). Phylogenetic analyses indicate that a minimum of seven different MADS box gene lineages were already present in the common ancestor of extant seed plants approximately 300 million years ago (mya; Becker et al., 2000). Thus, a diverse tool kit of MADS box genes was available before the origin of the angiosperms.

A phylogenetic perspective also provides the basis for comparative genomics (e.g. Soltis and Soltis, 2000; Walbot, 2000; Daly et al., 2001; Kellogg, 2001; Hall et al., 2002a; Mitchell-Olds and Clauss, 2002; Pryer et al., 2002; Doyle and Luckow, 2003). However, obtaining the appropriate phylogenetic perspective may be difficult: What phylogenetic hypotheses are already available for the group of interest? Are phylogenetic studies underway on a particular group, and is it possible to obtain unpublished trees? Is the phylogenetic underpinning for a lineage of interest sound enough for use in comparative genetic/genomic analyses? Not all phylogenetic trees are of equal quality, and the most fruitful phylogenomic comparisons will be those based on the strongest phylogenetic inferences.

We cannot address all of the crucial issues relating to the importance of phylogeny in a comprehensive fashion and, therefore, will focus on a few main topics. We provide: (a) phylogenetic summaries and references for major clades of land plants, with an emphasis on angiosperm model systems; (b) a “primer” of phylogenetic methods, including evaluation of parsimony, distance, maximum likelihood (ML), and Bayesian methods, the importance of measures of internal support in phylogenetic inference, and methods of analysis of large data sets; and (c) use of molecular data to estimate divergence times of genes or organisms. A major goal is to foster increased interaction and communication between phylogeneticists and physiologists/molecular geneticists by providing contacts and references for those requiring a phylogenetic backbone for analyses.

One question that systematists are frequently asked is: Where would I find the most recent phylogenetic tree for group (fill in the blank)? We provide a brief summary of relevant trees below, with a focus on land plants. In addition, selected trees for angiosperms can be found at http:/www.mobot.org/MOBOT/research/APweb//, http://www.flmnh.ufl.edu/deeptime/ and http://plantsystematics.org/). Researchers can also consult Tree of Life (http://tolweb.org/tree/phylogeny.html) and TreeBASE (http://www.treebase.org/treebase). Phylogenetic questions can also be posed directly to experts working on various groups of plants; a partial list of phylogenetic consultants is provided in Table I (for a larger list, see also http://www.flmnh.ufl.edu/deeptime/).

Table I. Partial list of phylogenetic experts for various clades of land plants.

Figure 1. Summary of phylogenetic relationships among major lineages of embryophytes (land plants). Charales are the sister group of the embryophytes. Within the embryophytes, liverworts, hornworts, and mosses are the basal most lineages; however, their precise (more ...)

Plants colonized the land approximately 450 mya. Within the land plants, the three lineages long known as the “bryophytes” (liverworts, hornworts, and mosses) do not form a single clade in most analyses but instead form a grade that subtends the tracheophytes (Fig. 1). Furthermore, the precise branching order of the three “bryophyte” lineages remains ambiguous, with different topologies suggested by various data sets. A branching order of liverworts, hornworts, and mosses has emerged as one favored arrangement (e.g. Karol et al., 2001); other data suggest that hornworts, followed by a clade of mosses + liverworts, are the basal branches of the embryophytes (Renzaglia et al., 2000).

However, the sister group relationship of Gnetales and angiosperms has not been supported by most molecular analyses. Analyses of combined data sets of multiple genes representing all three plant genomes (plastid, mitochondrion, and nucleus) have found strong support for a clade of extant gymnosperms (Fig. 1; e.g. Bowe et al., 2000; Chaw et al., 2000; Pryer et al., 2001; Soltis et al., 2002). However, some extinct gymnosperms (e.g. Caytoniales and Bennettitales) may be more closely related to angiosperms than to any lineage of living gymnosperm (Donoghue and Doyle, 2000). Cycads and Ginkgo biloba are sisters to the remaining living gymnosperms. The relationship between cycads and Ginkgo biloba is unclear; in some analyses, cycads and Ginkgo biloba are successive sisters to a clade of conifers and Gnetales, whereas in others, Ginkgo biloba and cycads form a clade that is sister to other extant gymnosperms. Some molecular analyses support a surprising placement of Gnetales within conifers as sister to Pinaceae (Bowe et al., 2000; the “gne-pine” hypothesis of Chaw et al., 2000).

The placement of Gnetales within conifers is an excellent example of a molecular phylogenetic result that must be viewed with caution, for several reasons. First, the placement of Gnetales within conifers is supported largely by mitochondrial genes; genes from other genomes do not place Gnetales within conifers. Furthermore, there is conflict between first and second versus third codon positions of cpDNA genes, with different positions supporting different placements of Gnetales. In addition, because most analyses of seed plants have involved small numbers of taxa, the gne-pine hypothesis may be an artifact of inadequate taxon sampling in some analyses. Our current interpretation of relationships among extant seed plants, showing Gnetales as sister to all conifers, is depicted in Figure 1. Analysis of extant gymnosperms exemplifies the complexities inherent in phylogenetic analysis of ancient lineages that have undergone significant extinction.

Figure 2. Summary tree of angiosperm relationships based on Soltis et al. (2000, 2003), with basal angiosperm relationships modified following Zanis et al. (2002). Numbers are jackknife values. Values for basal angiosperms are from Zanis et al. (2002); the (more ...)

Perhaps the best known classification of angiosperms is that of Cronquist (1981), who recognized six subclasses of dicots, Magnoliidae, Hamamelidae, Rosidae, Dilleniidae, Caryophyllidae, and Asteridae, and five subclasses of monocots, although these were followed less frequently. Molecular phylogenies indicate that these subclasses, like the classes Magnoliopsida and Liliopsida, should also be abandoned. The Magnoliidae are paraphyletic, and both the Hamamelidae and Dilleniidae are grossly polyphyletic, with constituent members appearing throughout much of the angiosperm tree. Thus, “Magnoliidae,” “Hamamelidae,” and “Dilleniidae” do not refer to monophyletic groups, and these names are no longer valid. Cronquist's concepts of Rosidae, Asteridae, and Caryophyllidae must be expanded and revised to correspond to monophyletic groups; these clades are the rosids, asterids, and Caryophyllales sensu (APG II, 2003). Although Caryophyllales are recognized at the ordinal level (see APG II, 2003), both rosids and asterids are supraordinal groups that are not assigned a Linnean rank in the APG II classification.

It is important to note that deep-level angiosperm phylogeny is not yet resolved. Relationships among the major clades of eudicots (e.g. rosids, asterids, Caryophyllales, Saxifragales, Santalales, and a few smaller clades) are unresolved (Fig. 2), presenting a limitation for many areas of comparative biology, including comparative genomics.

The phylogenetic trees available for many families of angiosperms facilitate interpretation of the evolution of diverse characters (molecular, physiological, and genetic). These trees also aid in the appropriate choice of representative taxa for comparative studies (see also Daly et al., 2001; Hall et al., 2002a); it is often useful to choose representative taxa from across the breadth of a clade and not simply one or two taxa from only a small part of the diversity of that clade.

Because trees depicting organismal phylogenies have accumulated so rapidly, it is often difficult for the nonexpert to know how to obtain a tree for a group of interest. Unfortunately, there is no single source that serves as a compendium of all intrafamilial phylogenetic trees. Judd et al. (2002) provide trees and relevant references for many families of tracheophytes. However, because it is an entry level textbook, many families are not covered. Therefore, we provide a short list of experts (Table I) who can assist with phylogenetic questions for major groups of embryophytes. A larger list is available on the Deep Time website.

Recent phylogenetic studies of Arabidopsis and relatives (Koch et al., 1999, 2001, 2003; Koch, 2003; O'Kane and Al-Shehbaz, 2003) have provided an initial tree for Brassicaceae sensu stricto and identified an Arabidopsis clade that contains the closest relatives of Arabidopsis. However, a more comprehensive analysis of the family is required and is well underway (M. Beilstein, E. Kellogg, and I. Al-Shehbaz, personal communication).

Table II. Comparison of methods of phylogeny reconstruction

Although there is a desire among many investigators for rapid phylogeny reconstruction and “instant tree,” it may be prudent to explore several methods (e.g. Swofford et al., 1996; Doyle and Gaut, 2000; Nei and Kumar, 2000). There remains a tendency to place more trust in phylogenetic results supported by multiple approaches (Doyle and Gaut, 2000). Regardless of method of phylogenetic inference, however, some measure of internal support (e.g. bootstrap, jackknife, and posterior probabilities; see below) is essential.

Many non-systematists employ NJ to the exclusion of other methods (Nei and Kumar, 2002). The distance measures used in NJ and other distance methods are typically based on models of nucleotide substitution. The NJ algorithm is fast and readily available in software packages such as MEGA (http://www.megasoftware.net/) and PAUP*. However, it also has important weaknesses. For example, NJ provides only a single tree, precluding comparison with other topologies. In reality, many optimal trees may be found in MP and ML analyses, depending on the data set, and these methods allow all optimal or near-optimal trees to be compared. Furthermore, different trees can be obtained with NJ depending on the entry order of the taxa (Farris et al., 1996; see Table II). One solution is to run multiple NJ analyses with different random entry orders of the taxa, accompanied by bootstrap or jackknife analysis (see below). Finally, because sequence differences are summarized as distance values, it is impossible to identify the specific character changes that support a branch. Although proponents of NJ, Nei and Kumar (2000) nonetheless argue for a pluralistic approach. Other methods of phylogenetic inference should be explored in addition to NJ.

MP is preferred by many phylogeneticists because of its theoretical basis and the diagnosable units it produces. The advantages of parsimony over NJ are several (Table II), an important one being that parsimony seeks to recover all shortest trees. Depending on the data set, a parsimony search may yield one (or a few) to hundreds or thousands of equally short trees. These shortest trees can be summarized in a strict consensus tree, which depicts only the nodes present in all equally short trees. In addition, MP analysis provides diagnoses (i.e. specific sets of characters) for each clade and branch lengths in terms of the number of steps (or changes) on each branch of a tree.

Statistical methods of phylogeny reconstruction, incorporating models of nucleotide (or amino acid) substitution, are preferred by many molecular phylogeneticists (see Lewis, 1998). Both ML and BI rely on such models to reconstruct both topology and branch lengths and, thus, are computationally intensive. ML analysis finds the likelihood of the data, given a tree and a model of molecular evolution. Like ML, BI has had a long tenure in statistics. However, it has only recently been introduced into phylogenetics (see Huelsenbeck et al., 2001, 2002). Although BI uses the same models of evolution as some other methods of phylogenetic analyses (e.g. ML and NJ), it represents a powerful tool and perhaps the wave of the future in phylogenetic inference. BI is based on a quantity referred to as the posterior probability of a tree, a value that can be interpreted as the probability that a tree is correct, given the data. BI uses a likelihood function to compute the posterior probability. Although BI allows the researcher to specify a prior belief in relationships (Table II; Huelsenbeck et al., 2001, 2002), this option has not been explored extensively to date, and Bayesian analyses typically assign equal prior probability values to all possible trees. Whereas ML is not feasible for large data sets (more than perhaps 50 taxa), BI (as implemented in Mr-Bayes; see Huelsenbeck et al., 2001) incorporates a faster search strategy (using Markov chains) and can be used on data sets of several hundred taxa to find tree, branch lengths, and support (but see Suzuki et al., 2002).

Certainly a frustrating aspect of phylogenetic analysis to those outside of the field is the number of inference methods available. NJ is widely used, in part, because of its speed and ready availability in computer packages such as MEGA. It also is part of alignment packages such as MegAlign (http://www.dnastar.com/cgi-bin/php.cgi?r10.php). However, parsimony can be readily implemented using PAUP* (Swofford, 1998; NJ and ML are also part of the PAUP package). PAUP* is often not employed by molecular biologists, however, because the user friendly version with pull-down menus is made for Macintosh, not Windows, operating systems.

Interpretations of bootstrap and jackknife values vary (for review, see Soltis and Soltis, 2003), although few view these values in a strict statistical sense. Bootstrap values are conservative, but biased, measures of phylogenetic accuracy (Hillis and Bull, 1993), with values of 70% or greater corresponding to “true” clades in experimental phylogenies (Hillis and Bull, 1993). Thus, some consider values of 70% or more as indicators of strong support, whereas others reserve “strong support” for values of 90 or 95% and above. Although different phylogenetic methods may yield different optimal topologies, the differences generally involve poorly supported clades. Those clades that are strongly supported generally appear in topologies regardless of the method of phylogenetic inference. Additional measures of support include the decay index or Bremer support (Bremer, 1994) for parsimony analyses and the posterior probabilities generated in BI.

Measures of internal support indicate those relationships in which we should, and should not, have confidence. A recently identified clade of MADS-box genes appears as the sister group to the well-known B class floral genes that specify the identity of petals and stamens in Arabidopsis and snapdragon. Becker et al. (2002) termed this new clade B_sister and determined that these genes are present in diverse seed plants. Although the monophyly of the B_sister clade received 92% bootstrap support, the placement of the B_sister clade as sister to the clade of B class genes received only 77% bootstrap support. With this level of support, it is reasonable to question whether the B_sister clade is really the sister group of the clade of B class genes. Increased sampling of B_sister genes from additional taxa and more rigorous analyses are needed to establish with certainty the placement of the B_sister clade within the MADS box genes of plants.

Many efforts to date evolutionary divergences using a molecular clock have yielded age estimates that are grossly inconsistent with the fossil record, regardless of method of tree construction. For example, molecular-based estimates of divergence times in plants reveal a vast range of dates. Using molecular data, the age of the angiosperms has been estimated as 350 to 420 mya, greater than 319, 200, to 140 to 190 mya (for review, see Sanderson and Doyle, 2001). However, the oldest unequivocal angiosperm fossils are 125 to 135 mya (for review, see Soltis et al., 2002).

Many sources of error and bias can affect molecular-based estimates of divergence times (see Sanderson and Doyle, 2001; Soltis et al., 2002). Obviously, an incorrect topology will yield erroneous estimates, with the magnitude of the problem depending on the extent of the topological error (Sanderson and Doyle, 2001). Inaccurate calibration will bias the resulting estimates. Also problematic are heterogeneous rates of evolution among lineages (see Sanderson and Doyle, 2001; Soltis et al., 2002). Inadequate taxon sampling can compound the problem. Estimates of divergence times can also vary among genes or other data partitions (e.g. among codon positions). Another potential source of error is the method used to estimate divergence dates. Sanderson and Doyle (2001) used molecular data to examine angiosperm divergences and found that the age of crown group angiosperms ranges from 68 to 281 mya, depending on data, tree, and assumptions, with most estimates falling between 140 and 190 mya.

Given that rate heterogeneity among lineages is common in most molecular-based trees, can we reliably use molecular data to estimate divergence times? Simple clock-based approaches to estimating divergence times are not likely to yield meaningful estimates. However, several approaches have been proposed when the assumption of rate constancy is violated: linearized trees (Takezaki et al., 1995), nonparametric rate smoothing (Sanderson, 1997, 1998), penalized likelihood (Sanderson, 2002), Bayesian approaches (e.g. Huelsenbeck et al., 2002; Thorne and Kishino, 2002), and “PATH” (Britton et al., 2002; for review of methods and instructions for implementing nonparametric rate smoothing, see http://www.flmnh.ufl.edu/deeptime/). Although methods to accommodate deviations from a steady molecular clock are still under development, it is nonetheless possible to estimate dates of divergence, given: (a) a reliable calibration point or points, (b) adequate sampling of taxa and characters, and (c) a method that is robust to rate heterogeneity. Confidence intervals for the estimated dates and consistency with the fossil record provide means for assessing the reliability of age estimates. Despite attempts to accommodate deviations from constant evolutionary rates, however, confidence intervals are typically large, and divergence times should be interpreted carefully.

An exciting recent development is the merging of phylogenetics and genomics. Phylogenetic hypotheses have become the framework for the choice of organisms in genomic analyses, and more and more molecular biologists are using phylogenetic trees to guide their sampling of taxa for comparative research. This trend will continue. Systematics is moving rapidly; therefore, molecular biologists are encouraged to contact systematics “experts” for help in obtaining the best supported trees for a given clade of interest. We stress the importance of a rigorous phylogenetic analysis of data. It is ironic, for example, that researchers may spend years gathering gene sequence data, but then want an immediate phylogenetic “answer” within seconds or minutes. A thorough phylogenetic analysis, evaluating alternative alignments, exon versus intron boundaries, using different phylogenetic methods, and obtaining estimates of internal support, may take several weeks or more, and this should not be considered an unreasonable investment of time. Our review of issues relating to phylogeny reconstruction also illustrates the need for more “quick courses” in phylogeny reconstruction for molecular biologists interested in constructing gene trees.

We thank Jeff Doyle, Bernie Hauser, Alice Harmon, and two anonymous reviewers for helpful comments on earlier drafts of this paper.