Methods for the inference of protein phylogeny

By Fred R. Opperdoes

Introduction

Once a multiple sequence alignment has been prepared, such alignment may serve the purpose of further evolutionary analyses. The final goal of such an analysis is to prepare an evolutionary tree describing the relationship of the various taxa with respect to each other. In order to understand the terminology used in the area of phylogeny, study the hypothetical tree shown below:

Example of a phylogenetic tree (for a normal size figure click here)

There exist various methods for the preparation of evolutionary trees: These are "Distance Methods" based on a matrix containing pair wise distance values between all sequences in the alignment, and "Character-Based Methods" that carry out calculations on each of the individual residues of the sequences. In general, distance methods are fast, while character-based methods are much slower, because they are CPU (central processing unit) intensive.

Back to the Table of Contents.

Methods available for tree construction

• Distance methods (in the public domain)
  • UPGMA,
  • Transformed Distance Methods,
  • Least Squares Methods. Fitch and Kitsch of the Phylip package (Jo Felsentein, Univ. Washington) are least squares method (Fitch and Margoliash)
  • Neighbor-joining methods
    • Neighbor of the Phylip package (Jo Felsentein, Univ. Washington),
    • ClustalW (D. Higgins, EMBL (check),
    • Distnj in Protml package (Adachi and Hasegawa, Univ. Tokyo)
  • Darwin (Gaston Gonner, ETH, Zurich, via mail server or WWW)
• Character-based methods
  • Protein Maximum likelihood (in the public domain)
    • Protml (Adachi and Hasegawa, Univ. Tokyo) (very cpu intensive)
    • Puzzle (Korbinian Strimmer and Arndt von Haeseler, Univ. Munich) (A heuristic method much faster than Protml)
  • Protein maximal parsimony (in the public domain)
    • Protpars (Jo Felsentein, Univ. Washington)
    • Paup (David Swofford, latest version will be commercial)

Back to the Table of Contents

Distance Matrix Methods

• UPGMA (Unweighted Pair Group with Arithmetic Mean) uses real (uncorrected) distance values and a sequential clustering algorithm. (This method of tree construction is very sensitive to differences in branch length or unequal rates of evolution. Therefore, it should only be used with closely related OTUs, or when there is constancy of evolutionary rate. The method is often used in combination with isoenzyme or restriction site data or with morphological criteria)
• Transformed Distance Methods. Corrections of the observed distance values, according to an evolutionary model, may be introduced to obtain trees with true evolutionary distances (PAM values, Kimura's formula), or corrections are carried out with reference to an outgroup (Farris, 1971; Klotz et al, 1979). This method is recommended when evolutionary distant organisms are included in the data set.
• Neighbors Relation Methods.
  • FITCH (Fitch, 1981).
  • Neighbor-Joining method, (Saitou and Nei, 1987).

  Should all be used with corrected (see above) distance matrices.
  

NB:These methods, which all assume an evolutionary model, result in only one best possible tree.

Character-based methods

• Maximum Parsimony
  • Is character-based and uses sequence information itself rather than distance information.
  • The information content used by this method is not necessarily larger than for the distance matrix methods, since there is only a limited number of informative sites. For more information on what is an informative site, click here.
  • Calculates for all possible trees the tree that represents the minimum number of substitutions at each informative site.
  • Does not assume an evolutionary model.

  
  NB: Many equally likely trees can be found

• Maximum likelihood .
  This is a method for the inference of phylogeny. It evaluates a hypothesis about evolutionary history in terms of the probability that the proposed model and the hypothesized history would give rise to the observed data set. The supposition is that a history with a higher probability of reaching the observed state is preferred to a history with a lower probability. The method searches for the tree with the highest probability or likelihood.
  The following programs are available from the web:
  
  • DNAML (DNA data only. By Joe Felsenstein in the Phylip package)
  • FastDNAML (DNA data only. A faster algorithm applied by Gary Olsen to Joe Felsenstein's DNAML program )
  • ProtML (DNA and protein. By Adachi and Hasegawa, 1992)
  • Puzzle (DNA and protein. By Strimmer and von Haeseler, 1995). This program applies a heuristic method and is much faster than PROTML, but does not guarantee to find the best tree.

NB: For more information on the maximum likelihood methods: click here

.Back to the Table of Contents

How to root an unrooted tree?

• Most methods for the inference of phylogeny yield trees that are unrooted. Thus from a tree by itself it is impossible to tell which of the OTUs branched off before all the others.
• To root a tree one should add an outgroup to the data set. An outgroup is an OTU for which external information (eg. paleontological information) is available that indicates that the outgroup branched off before all other taxa.

Some good advice to root a tree

• Do not choose an outgroup that is very distantly related to your taxa. This may result in serious topolocical errors because sites may have become saturated with multiple mutations by which information may have been erased.
• Do not choose either an outgroup that is too closely related to the taxa in question. In this case it may not be a true outgroup.
• The use of more than one outgroup generally improves the estimate of tree topology.
• In the absence of a good outgroup the root may be positioned by assuming approximately equal evolutionary rates over all the branches. In this way the root is put at the midpoint of the longest pathway between two OTUs. This way of rooting is called mid-point rooting.

• DNA or Protein used in the analysis
• Distance or Parsimony methods applied
• The number of OTUs included in the alignment
• The order of the OTUs in the alignment
• The selection of an appropriate outgroup

NB: None of the methods may guarantee the one tree with the correct topology.

.Back to the Table of Contents

How to use the tree-construction programs?

So as to have an idea about the reliability of the topology of the resulting tree, one should do one or all of the following:

• Apply more than one of different methods (distance, maximum parsimony, maximum likelihood) to the data set.
• Vary the parameters used by the different programs, such as the seed value and jumble factor for the order of OTU addition.
• When in doubt, apply various evolutionary models for matrix construction.
• Add or remove one or more OTUs and see how this influences tree topology.
• Try to include an outgroup that may serve as a root for your tree.
• Apply "Bootstrap" or "Jacknife" analyses to your data set and prepare a consensus tree of 100 - 1000 replicas (depending on size of the data set and on computer power). Keep in mind that in the case of bootstrap analysis only nodes that occur in more than 95% of the cases are reliable.

NB: Only when widely different methods provide you with similar or identical tree topologies and such topologies are supported by good bootstrap values (> 95%) the trees can be considered reliable.

.Back to the Table of Contents

Limitations of the various methods

• Distance approaches (UPGMA, corrected distances and neighbor-joining) do not use the original (sequence) data, but derived distance information. Some information is said to be lost.
• Character-state approaches (maximum parsimony, maximum likelihood) are said to be more powerful than distance methods because they use the raw data. However, this is usually a small fraction of the data. Maximum parsimony uses only the relevant sites. So when the number of informative sites is not large, this method is often less efficient than distance methods (Saitou and Nei, 1986). Maximum parsimony is notorious for its sensitivity to codon bias and unequal rates of evolution.
• None of the methods is reliable when OTUs with highly unequal evolutionary separation are included in the data set.

.Back to the Table of Contents

Complication of paralogous genes

The presence of more than one homologue of a certain gene, or of different members of a gene family, in one and the same organism may complicate considerably phylogenetic analyses. When such a situation is encountered one speaks of the presence of paralogous genes.

Two genes are said to be orthologous if they diverged after a speciation event, whereas they are said to be paralogous if they diverged after a duplication event.

Let's take the example of the mammalian lactate dehydrogenase isoenzymes M and L. In the case of mouse and rat the isoenzymes are the result of a gene duplication that took place well before the separation of these two species.

where () indicates speciation and X indicates gene duplication.

Here one says that the LDH_M gene family is paralogous to the LDH_L gene family. In the case one is not aware of the presence of paralogous genes and isoenzymes in the organisms, because for instance only one sequence for each organism (e.g. LDH_L for mouse and LDH_M for rat) is available and isoenzyme data are missing, then the resulting phylogeny would suggest a much earlier separation of mouse and rat. This will inevitably lead to erroneous phylogenetic trees.

Here you'll be able to find more explanation on orthologous and paralogous genes.

Definition: Paralogy is the construction of a phylogenetic tree from a mixture of genes generated by duplications.

To find out what is the difference between Cladistics and Phenetics and to read about the differences between the various phylogeny inference methods, click here.

Back to the Table of Contents

Still to add:

some literature examples

Origin of chloroplasts in J. Molec. Evolution, 1995

Back to the Table of Contents

Last updated: 25 September 1997.