I. The Methods of Molecular Phylogenetics           

Molecular phylogenetics refers to any method of inferring evolutionary relationships from similarities or differences in molecular structure.

  • The goal of phylogenetics, whether based on molecules or morphology, is to reconstruct the evolutionary history of groups of organisms.
  • Molecular phylogenetics is no different in principle from inferring phylogeny from the similarities in morphology. Many of the same methods are applied to both molecules and morphology.
  • Since molecular changes underlie all inherited morphological changes, molecular phylogenetics can be viewed as simply a more direct approach to morphological phylogeny.

Molecular characters suffer from problems that also afflict morphological characters. For example, neither molecules nor morphology may be able of resolving the phylogeny of evolution that was both ancient and rapid, as in the Cambrian Explosion.

  • Just as a telescope is incapable of producing a clear image of a cell, techniques for looking at remote phylogenetic changes are not able to resolve the small details of what occurred during a short time.
  • Molecular phylogenetics has so far proved incapable of resolving branching patterns among some clades such as spiralians.

Another problem shared by molecular and morphological characters is homoplasy (nonhomologous characters appearing to be similar in different taxa).

  • The same base or amino acid can occur homoplasiously at a position on molecular sequences from two taxa, tending to make the two taxa appear to be more closely related than they really are. Such homoplasy is especially likely for DNA, because it has only four different DNA bases. Adenine (A), for example, could occur at the same position in two sequences either because there had been no change at the position or because there had been two or more changes (for example, A to C to A).
  • Two homologous DNA sequences that are saturated with mutations will be identical at one-fourth of their positions merely by homoplasy.
  • Some molecular characters are virtually immune to homoplasy. These include mitochondrial gene rearrangements and short and long interspersed elements (SINEs and LINEs).

Other problems shared by molecular and morphological phylogenetics arise from polymorphism (homologous characters appearing differently in the same species). Because of polymorphism, the time of divergence may appear to be earlier than it was.

  • The occurrence of two or more forms within a species (polymorphism) indicates that evolution has occurred before speciation. If different forms of a molecule were present in two populations that later diverged into species, the time of divergence inferred from the molecule will appear to be earlier than it actually was.
  • As the molecules continue to evolve separately, however, the original differences between them will become negligible compared with the changes following speciation. Consequently, this problem can be disregarded at higher taxonomic levels.

Polymorphism can also result in the incorrect phylogenetic sequence.

  • Consider hypothetical species A, B, and C', the prime indicating that a character in species C' differs from that in A and B. Whether morphological or molecular, the character difference would tend to suggest that species A and B are closer to each other than either one is to C' (Fig. 1a). In fact, however, B and C' might be sister groups that diverged from a polymorphic ancestor, with B and C' each inheriting a different form of the character (Fig. 1b).

Figure 1. Polymorphism can lead to incorrect molecular or morphological trees. (a) Taxa A and B have inherited one form of a molecule, while C' has inherited a different form of the homologous molecule, leading to the inference that A and B are sister groups to the exclusion of C'. (b) In fact, B and C' might be sister groups that inherited different forms of the molecule from a polymorphic ancestor.

  • This kind of problem is thought to be responsible for conflicting molecular phylogenies for humans, chimpanzees, and gorillas (Graur and Li 2000, p. 222). Again, however, this is not likely to be a problem at higher taxonomic levels, since evolution of the character subsequent to speciation will obscure the relatively small differences that existed before speciation.

Similar problems result from different copies of duplicated genes.

  • If a gene has been duplicated in an ancestor, the descendants will have two types of homologs of the gene or gene product: orthologous (derived from the same ancestral copy) and paralogous (derived from different ancestral copies).
  • Paralogous copies may cause problems similar to those of polymorphism, because, like polymorphisms, they are different versions of the same gene.

Another problem with molecular phylogenetics is long-branch attraction: the tendency of fast-evolving molecules to appear more closely related than they actually are.

  • Because of homoplasy, long branches (molecular sequences that have evolved rapidly or for a long time) appear to be more closely related to each other than do sequences that have evolved slowly or for less time. When long branches are mixed with short ones, the long branches tend to join one another during tree reconstruction. This problem is called long-branch attraction.
  • Long-branch attraction can be avoided by eliminating from the study group taxa in which molecular sequences have evolved more rapidly than in other taxa, and by eliminating parts of sequences that have evolved more rapidly than other parts of the same molecule.

Molecular phylogenetics has gained wide acceptance in spite of these and other problems because it provides a large amount of evidence that is independent of morphology, as well as other advantages.

  • In any two taxa there are many more homologous molecules than there are homologous morphological characters, especially if the taxa are as different as, say, sponges and insects.
  • Every difference in a molecule is potentially an independent character, so one gene or protein may provide dozens or hundreds of characters. The gene for the RNA in the smaller subunit of the ribosome, for example, contains more than 1,700 bases.
  • In contrast to morphological characters, which can be influenced by environment, molecules are for the most part strictly inherited.
  • Many molecular characters, such as the presence of a particular base or amino acid at a given position, are strictly binary. In contrast, many morphological characters vary continuously: one must set an arbitrary criterion for whether, for example, a bird's beak is short or long.
  • Some molecules may evolve at a regular rate, so it is sometimes possible to estimate the time of divergence of two groups from their degree of molecular difference.

Several kinds of experiments support the validity of molecular phylogenetics.

  • The molecular phylogeny of 10 strains of laboratory mice inferred from chromosomal differences agreed exactly with the known phylogeny (Fitch and Atchley 1987). In contrast, phylogenies based on morphology (lower jaw structure) or life-history traits (litter size, body mass at different ages, etc.) gave conflicting phylogenies, none of which was correct.
  • Molecular phylogenetics correctly reconstructed the branching pattern and branch lengths for a virus serially propagated in the presence of a mutagen (Hillis et al. 1992; Hillis et al. 1994).
  • Phylogenies of birds and mammals based on different molecules were more nearly in agreement with each other than were phylogenies based on different morphological characters (Bledsoe and Raikow 1990).

Molecular characters can be of two types: discrete (qualitative) differences in molecular sequence and continuous (quantitative) distance between molecules.

  • The following are examples of discrete characters: differences in base or amino-acid sequences, gene rearrangements and duplications, and the position of transposable elements on chromosomes.
  • The following kinds of data provide distance measures: degree of immunological compatibility, electrophoresis of proteins, the number of discrete differences, and DNA-DNA hybridization.

The first step in molecular phylogenetics is to select a suitable molecule that is homologous in all the taxa to be included in the phylogeny.

  • The molecule must occur in all taxa to be studied (the study group).
  • The molecule must be large enough to provide a sufficient number of differences for comparison.
  • For a phylogeny of higher taxonomic categories (kingdom, phylum, class), the molecule should have evolved slowly, since these taxa have had more time to evolve. One example of such a highly conserved molecule is rDNA - the DNA that encodes one of the ribosomal RNAs.
  • For lower taxonomic categories a fast-evolving molecule is needed to ensure that it is sufficiently different among taxa. Mitochondrial DNA (mtDNA) is an example of a fast-evolving molecule.

Many molecular characters are much less susceptible to homoplasy and long-branch attraction than are nucleic-acid sequences. These characters include amino-acid sequences from proteins, the positions of short and long interspersed elements, and Hox genes.

Elongation factors, actin, and tubulins are among the widely used proteins.

  • Since there are so many different amino acids, the problem of homoplasy and long-branch attraction are less troublesome in proteins than in nucleic acids.
  • Elongation factors are proteins involved in protein synthesis in all organisms. One of the most widely used proteins in molecular phylogenetics is elongation factor-1∝ (EF-1∀).

The position of short and long interspersed elements (SINEs and LINEs) are another increasingly common source of discrete characters.

  • SINEs and LINEs are highly repetitive DNA sequences that occupy much of the genomes of animals (more than a third in humans). Their only known function is to make copies of themselves to be inserted into the genome, so they are characterized as "junk DNA" as well as "selfish DNA."
  • Since SINEs and LINEs are transposed to random positions in the genome, the occurrence of a particular SINE or LINE at the same location in two different organisms is likely to be a synapomorphy rather than a homoplasy.
  • SINEs and LINEs do not occur broadly across taxa, however, so they have been used mainly to resolve relationships among lower taxonomic categories.

Hox genes have also been used to infer phylogenetic relationships.

  • Hox genes occur in clusters and encode transcription factors that regulate development. They are best characterized in segmented animals such as insects, where they function as homeotic genes determining the identity of each segment depending on its location along the antero-posterior axis. Mutations in Hox genes may be involved in evolutionary changes in body plans.
  • Hox genes occur in Cnidaria and all Bilateria where they have been sought. In most animals there is only one cluster of Hox genes, but in most vertebrates there are four duplicated clusters. Orthologous and paralogous Hox genes can be identified from one taxon to another by comparing nucleotide sequences.
  • Animals with similar clusters of Hox genes can be inferred to be closely related.

The most commonly used molecular data for higher taxonomic levels are base sequences from genes that encode ribosomal RNA, especially 18S rDNA.

Nucleic-acid sequences must be aligned before they can be compared.

  • Alignment is necessary to ensure that homologous base positions are being compared. Alignment is done either by inspection or by means of computer algorithms.
  • Numerous mutations as well as insertions or deletions of bases make alignment difficult
    (Fig. 2).

    Figure 2. Homologous sequences of DNA bases from two taxa (1 and 2). An insertion, dele-tions, and a large number of mutations in the middle portion of the sequence make alignment ambiguous for that region.

  • Secondary structure is sometimes used to aid alignment. For example, antiparallel complementary segments of RNA may be used for alignment since they are likely to be more stable than loops.
  • Ambiguous segments are often discarded from analysis.

Assumptions may be needed about the probabilities of different molecular changes.

  • For DNA sequences one may need to allow for the fact that transitions (pyrimidine changing to pyrimidine or purine changing to purine) are more likely than transversions (pyrimidine changing to purine or vice versa). Transversions are typically assumed to be several times less likely than transitions and are therefore weighted more heavily. First or second positions in codons may also be weighted more heavily than third positions, where synonymous substitutions are less-rigorously selected against.

Molecular relationships are represented as trees constructed of branches with nodes at both ends of every branch.

  • A terminal node represents an operational taxonomic unit (OTU), which is a presumptive taxon (Fig. 3). (From now on in this discussion, OTUs will be assumed to be extant taxa.)
  • An internal node represents the hypothetical ancestor of two or more taxa.
  • Branches may be unscaled to show only phylogenetic relationships, or they may be scaled by making the length of each branch proportional to some distance measure, such as the number of base differences.

    Figure 3. A simple tree with some of the nodes and branches indicated. Terminal nodes rep-resent OTUs (presumptive taxa) 1 through 4. Internal nodes represent hypothetical ancestors. The branches are scaled in this example.

Inferring (reconstructing) a phylogeny consists of generating or selecting one tree out of perhaps millions of possible ones.

  • Only three different trees represent all possible relationships of four taxa (Fig. 4).

    Figure 4. The three different trees possible with four taxa. Branches are unscaled.

  • As the number of taxa increases, the possible trees become a forest. The number of possible strictly bifurcating trees with n taxa is {(2n-5)!}/{2n-3(n-3)!}. Even with only 10 taxa this equals more than 2 million possible trees!
  • Phylogenetic reconstruction consists of either generating a single tree according to some al-gorithm or selecting one tree according to an optimality criterion.
  • The most widely used algorithm for generating a single tree is neighbor-joining. The most widely used methods for selecting an optimal tree is maximum parsimony and maximum likelihood.

The neighbor-joining method (NJ) is an algorithm that generates one tree with the shortest total branch length.

  • NJ begins by assuming that all taxa are joined at a single node. It then sequentially joins one pair of taxa at a time to find the combination that gives the shortest total branch length (Fig. 5).

Figure 5. Neighbor-joining applied to the four taxa from Figure 3 illustrated by a graphical procedure called star deconstruction. (a) All the unscaled branches are joined at a single internal node. (b and c) The first (and in this simple case, the only) internal branch is added, with each possible pair of taxa joined as neighbors at one end (dashed, red lines) and the remaining taxa joined at the node at the other end. Only two of the three possibilities are shown here.

  • For each of the trees with a different pair joined as neighbors, the two neighbors are combined to form a composite taxon. The length of the branch to that composite taxon is set so that the average distance from the two neighbors to every other taxon is the same as in the original scaled tree (Fig. 6). The neighboring pair of taxa that give the shortest total branch length are assumed to be neighbors in the final tree. In Figure 6, the tree in (a) is shortest, so taxa 3 and 4 would be joined as neighbors. NJ would therefore construct the tree shown in Figure 5b.

    Figure 6. The trees shown in Figure 5b and c after combining the first pair of neighbors into one branch (dashed, red) and rescaling. The tree in (a) has a shorter total branch length than the tree in (b) (as well as the other alternative, not shown).

  • After the first pair of neighbors and the first internal branch are found, the procedure is repeated with the first pair of taxa represented by one branch. The second internal branch is then found, and so on. (With only four taxa, of course, there would be only one internal branch.) Finally, the scaled tree is reconstructed using the internal branches that were found.
  • ADVANTAGE: NJ takes relatively little computational effort.
  • DISADVANTAGES: NJ generates only one tree, which may not be vastly superior to an alternative. If sequences are short, statistical errors increase. Long distances are likely to be underestimated because of multiple substitutions at the same positions. NJ also looks only at the number, not the nature, of changes.

The maximum parsimony method (MP) selects the cladogram with the minimum number of changes in character state.

  • When applied to molecular-sequence data, MP begins by identifying informative sites. An informative site is one in which there are at least two different character states, at least two of which occur in more than one taxon (Fig. 7).
1) T T C G A C C G T
2) C T T A A C T G T
3) C T A T G C T G G
4) C T G T G C C G G
         
x
  
y
  
z

Figure 7. Aligned homologous DNA sequences from four taxa. Informative sites are indicated by letters x, y, and z. Only positions with two or more different bases, at least two of which occur in more than one taxon, are informative.

  • The MP method searches all possible trees to find the one that requires the smallest number of changes for each informative position. The tree requiring the fewest changes is the most parsimonious and therefore preferred, since it requires the fewest hypotheses about evolutionary change (in accordance with Occam's Razor).
  • The total number of changes is the length of the tree. For example, based on Figure 7 above, the tree shown in Figure 8a would be shorter than the tree in Figure 8b, requiring four rather than five base substitutions.

    Figure 8. Two of the three possible trees for taxa 1 through 4 with DNA sequences shown in Figure 7. (a) A total of only four base substitutions at the informative sites x, y, and z are re-quired with this tree. (b) Five base substitutions are required with this tree, which is therefore longer and less parsimonious. The remaining tree, grouping 1 with 3 and 2 with 4, also requires five substitutions.
  • In the example in Figure 8, all substitutions are assumed to be equally likely. This is called unweighted parsimony. Often more weight is given to transversions, which are less likely than transitions, or only transversions may be counted. If only transversions were counted in Figure 7, only position z would be informative. In that case, the tree in Figure 8a would still be preferred, requiring only one transversion compared with two in b.
  • With 12 or more taxa an exhaustive search of the more than 13 billion trees is impractical, so the number of trees to be examined for length must first be reduced by some other method. One approach to reducing the number is to first perform what is called a heuristic search of the most likely trees. In a heuristic search, NJ or some other method is first used to find a provisional tree. Branches are then rearranged and examined by MP to try to find a shorter tree. If a shorter one is found, all others are ignored, and the process is repeated.
  • ADVANTAGE: Unlike NJ, MP uses information about the type of change at each informative site and not merely the number of changes.
  • DISADVANTAGES: By using only informative sites, MP still uses only a small portion of sequence information. MP also has the disadvantage that it often recovers a number of equally parsimonious trees. Both of these problems are minimized by using long sequences with many informative positions. MP produces only cladograms, which are, of course, unscaled phylogenetic trees. The most serious limitation of MP is that with more than 12 taxa an exhaustive search of all possible trees is impractical, so there is no certainty that the most parsimonious tree will be found.

The maximum likelihood method (ML) begins with an explicit model of evolution and possible trees, then it attempts to find the tree that is most likely with the given data.

  • With ML, one must first estimate the probability of each kind of change in character state (for example, the probability of no change in a base, a transition, or a transversion). The likelihood Ln for the bases at each position n and for each tree is then calculated from these probabilities. The logarithm of these values of L are then added to get the log likelihood (ln L) of each tree. The tree with the highest (least-negative) value of ln L is taken to be the most likely.
  • Suppose we estimate or assume that the probability of a nucleotide base remaining unchanged is 0.7, the probability of a transition is 0.2, and the probability of a transversion is 0.1. We can now apply these probabilities to calculate the likelihoods of the trees in Figure 8 given the sequences in Figure 7. Figure 9 shows the possible changes in the tree shown in Figure 8a that could have led to the bases at the first position. Table 1 shows how the likelihood is calculated.

    Figure 9. An illustration of ML for the first position in the sequence in Figure 7 and the tree in Figure 8a. For taxon 1 the base at the first position is T, and for the other three taxa the base is C. X and Y represent the bases at the first position for the two ancestral taxa. One explana-tion for the bases at the position in these four taxa is that both X and Y inherited C from their common ancestor, and there was a transition from C to T in the evolution of taxon 1. Another possibility is that was a transition in the divergence of X and Y, so that X became T and Y be-came C, and this was followed by a transition from T to C in the evolution of taxon 2. It is also possible, but less likely, that X was A or G, and there were two transversions in the evolution of taxa 1 and 2. Similarly, Y was most likely C, but it could have been any of the other three bases. Therefore there are 16 (4 x 4) different ways that the bases at this first position could have occurred with this tree. Each way has a different probability. The likelihood of these bases occurring at this position with this tree is the sum of all these 16 probabilities. Let us assume the probability of no change is 0.7, the probability of a transition is 0.2, and the probability of a transversion is 0.1. If X and Y were both C, then there were four branches with no change and one with a transition at that site, so the probability of each of the four bases being what they are is 0.74 x 0.2 = 0.04802. If Y had been C and X had been T, A, or G, the probabilities would have been 0.01372, 0.00049, and 0.00049, respectively. These four probabilities with Y = C are shown in the top row of Table 1. Making Y one of the other bases gives the other three rows in the table. Adding all 16 of the probabilities gives the likelihood (L1,) and the log likelihood (ln L1 ) for the bases at the first position given the data. This procedure would be repeated for every position in the sequence. Adding all these log likelihoods gives the log likelihood (ln L ) for the tree and data. This procedure would be carried out for all trees to find the one with the maximum log likelihood.

    Table 1. An application of ML to the first position in the sequence in Fig. 7 and the tree in Fig. 8a. X represents the base at the first position for the ancestor of sister taxa 1 and 2, and Y represents the base for the ancestor of sister taxa 3 and 4. Each row shows the probability of the bases occurring at the first position in the four taxa if the bases at that position in X and Y are as shown, assuming that the probability of no change is 0.7, the probability of transition is 0.2, and the probability of transversion is 0.1. In the first row and first column, for example, if both X and Y had C as the base at the first position in the sequence, then there would have been no change at that site for three of the taxa or for X and Y, and there would have been one transition for taxon 1, giving a probability of 0.04802. If Y had C, and X had A (first row, third column), there would have been no change for two branches, and a transversion for each of the branches X-Y, X-1, and X-2, for a probability of 0.72x0.13 = 0.00049. The sum of all 16 probabilities gives the likelihood L1 = 0.06858 and ln L1 = -2.680 that this tree correctly represents the phylogeny given the bases at this position. This procedure would be repeated for every position in the sequence. Adding all the ln L values for each position gives the total log likelihood ln L for the tree. For the tree in Fig. 8a and the sequences in Fig. 7, the log likelihood is -24.716. The log likelihood of other trees would be calculated similarly. The log likelihood for the tree in Fig. 8b is -27.732, and the log likelihood for the third possible tree (not shown) is -28.490. The tree with the highest ln L is considered the most likely. Thus, the tree in Fig. 8a is the most likely of the three, as was also shown with MP.

    X = C
    X = T
    X = A
    X = G
    Y = C
    0.04802
    0.01372
    0.00049
    0.00049
    Y = T
    0.00112
    0.00392
    0.00004
    0.00004
    Y = A
    0.00014
    0.00014
    0.00007
    0.00002
    Y = G
    0.00014
    0.00014
    0.00002
    0.00007
    L1 = 0.06858; ln L1 = -2.680

  • ADVANTAGE: Unlike NJ and MP, ML uses all the character data and not simply the number of character changes or a few informative positions.
  • DISADVANTAGES: The main criticism of ML is that the likelihood of each kind of base substitution, and therefore the total likelihood for each tree, depends on explicit assumptions about their probabilities. Another criticism of ML is that, unlike NJ and MP, it cannot be used with morphological characters, since one cannot estimate the probability of changes in character state. ML is also limited by the amount of computer time and memory available to examine every possible tree and calculate the likelihoods for each one. It is often necessary to first perform a heuristic search to narrow the number of trees (as in MP), and thus the tree with the maximum likelihood may be missed. Perhaps the best use of ML is in finding the most likely among several competing hypothetical trees, rather than trying to search all possible ones.

With more than a few taxa, any method requires a computer.

  • The computer time and memory required increase rapidly with the number of taxa. The analysis places a large burden on computer resources and limits the number of taxa that can be considered simultaneously. Some analyses, especially with ML, may require months of computer time or may terminate prematurely with a fatal "out of memory" error.
  • More than 150 different computer programs are available. For a list and links to many of them, see http://phylogeny.arizona.edu/tree/programs/programs.html.

To show the temporal sequence of divergence, trees have to be rooted. The root represents the most recent common ancestor of the study group.

  • Figure 3 is an example of an unrooted tree. It represents relationships and distances among the four taxa of the study group, but it does not show the sequence of evolutionary divergences, since it lacks a temporal reference.
  • Molecular phylogenetic trees are usually rooted by using molecular information from one or more outgroups that are believed from paleontological or other evidence to be outside the study group (Fig. 10a). Ideally, the outgroup used for rooting is the sister group of the study group.
  • Alternatively, the root can be placed at the midpoint of the longest pathway separating two taxa in the study group (Fig. 10b). This assumes that the two most distant taxa diverged earliest from their most recent common ancestor, and each branch thereafter evolved at about the same rate.

Figure 10. Unscaled phylogenetic trees resulting from the rooting of the tree in Figure 3 by two different methods. (a) If an outgroup were thought to be close to 4, the root would have been placed on the branch terminating in 4, resulting in this rooted phylogenetic tree. (b) Without an outgroup, the root would have been placed at the midpoint on the longest pathway between two taxa (between 2 and 3 in Figure 3), resulting in a different phylogenetic tree.

  • The number of possible rooted trees is the same as the number of unrooted trees with the number of taxa increased by one, since rooting is equivalent to adding a new taxon to the study group.

For convenience in printing large trees, branches are often represented as horizontal lines joined by vertical lines representing internal nodes. Branches may be unscaled, or they may be scaled according to some distance measure.

  • In an unscaled phylogenetic tree, the terminal nodes are aligned, and the positions of internal nodes represent the order of divergence (Fig. 11a). In a scaled phylogenetic tree, the branches are proportional to the degree of molecular difference or some other distance measure (Fig. 11b).

Figure 11. Phylogenetic trees in Figure 3 with horizontal branches. (a) The unscaled tree rooted as in Figure 10a. (b) The scaled tree rooted as in Figure 10b. The distance between two taxa is found by measuring along the horizontal branches connecting them, ignoring the lengths of vertical branches, which represent nodes.

Phylogenies reconstructed by different methods are generally similar to each other.

  • Figure 12 shows a comparison of phylogenetic analyses of Platyhelminthes using NJ, MP, and ML.

Figure 12. Phylogenetic trees for Platyhelminthes using the same 18S rDNA se-quences analyzed by NJ, MP, and ML, modified from Figure 2 of Katayama, Nishioka, and Yamamoto (1996). Note that the topologies (branching patterns) for the trees produced by the three methods are all similar. For simplicity, branches for individual species were collapsed to one branch for each order. Yeast (S. cerevisiae) was used to root the tree, and four diploblasts were used as outgroups. The scales for NJ and ML show the number of base substitutions per sequence position. Small numbers for NJ and MP are bootstrap values indicating the reliability of each branch. (See next section.) Bootstrapping was not done for ML because of the large amount of computer time required.

Confidence in an internal branch can be tested by bootstrapping.

  • Bootstrapping is done by randomly sampling the data and replacing them so that some data are ignored and others represented more than once. A new tree is then reconstructed from the pseudoreplicated data. This is typically done hundreds of times, and the percentage of time an internal branch occurs in the trees is the bootstrap value of the branch. A bootstrap value of more than 90% or 95% is regarded as strong support for the branch. Bootstrap values are shown in Figure 12 on the previous page. Note that branches with high bootstrap values, such as the branch for Acoela, occur by all three methods of tree reconstruction.
  • In some situations a method called parametric bootstrapping is more appropriate. In para-metric bootstrapping, numerical simulation based on a model of evolution is used to produce the pseudoreplicate samples.
  • Bootstrapping tests the precision, not the accuracy, of the branch. That is, it indicates the ability of the data to recover the branch, but not whether the branch is correct.
  • A similar but less-used procedure is jackknifing, in which data are not replaced after sampling and each datum is therefore used only once.

A branch with low bootstrap support may be collapsed.

  • Collapsing a branch consists of joining the two nodes of the branch (Fig. 13).

Figure 13. Collapsing branches that are poorly supported. (a) The original tree with one branch having a bootstrap value of only 45%. (b) After collapsing the poorly supported branch there is an unresolved trichotomy for branches (1 + 2), 3, and 4.

A consensus tree can be created by collapsing branches that are not supported in all trees created by different methods of analysis. A consensus tree can also be produced by comparing molecular and morphological trees.

Molecular and morphological data can be combined to create a "total-evidence tree."

  • One difficulty with combining molecular and morphological characters for total-evidence analysis is that the former are typically so abundant that they may overwhelm the morphological characters.

Because of long-branch attraction, differences in sequence alignment, limitations in the size of study groups, and different methods of tree reconstruction, conflicting molecular phylogenies have been proposed. As techniques have improved and more molecules from more species have been sequenced, many of the past conflicts have been resolved.

  • Still, one should not accept any phylogenetic tree, whether based on molecules or morphol-ogy, at face value. Phylogenetic trees are hypotheses to be tested.
  • The advantage of molecular phylogenetics is not that it is infallible, but that it provides a completely independent means of testing morphological hypotheses.
  • There is now a broad agreement among molecular phylogeneticists about the main outlines of animal phylogeny.