A step by step guide to phylogeny reconstruction

A step by step guide to phylogeny reconstruction
The Plant Journal (2006) 45, 561–572
doi: 10.1111/j.1365-313X.2005.02611.x
A step by step guide to phylogeny reconstruction
C. Jill Harrison and Jane A. Langdale*
Department of Plant Sciences, University of Oxford, South Parks Road, Oxford, OX1 3RB, UK
Received 13 July 2005; revised 20 September 2005; accepted 29 September 2005.
For correspondence (fax þ44 1865 275147; e-mail [email protected]).
The aim of this paper is to enable those who have never reconstructed a phylogeny to do so from scratch. The
paper does not attempt to be a comprehensive theoretical guide, but describes one rigorous way of obtaining
phylogenetic trees. Those who follow the methods outlined should be able to understand the basic ideas
behind the steps taken, the meaning of the phylogenetic trees obtained and the scope of questions that can be
answered with phylogenetic methods. The protocols have been successfully tested by volunteers with no
phylogenetic experience.
Keywords: beginners, PAUP*, protocol, phylogeny.
Aims of phylogeny reconstruction
A phylogeny is the evolutionary history of a group of
entities. Given that this can only truly be known in
exceptional circumstances, the main aim of phylogeny
reconstruction is to describe evolutionary relationships in
terms of relative recency of common ancestry. These
relationships are represented as a branching diagram, or
tree, with branches joined by nodes and leading to terminals at the tips of the tree (Figure 1). The three main
types of relationship distinguished are monophyly, paraphyly and polyphyly (Hennig, 1966). Monophyletic and
paraphyletic groups have a single evolutionary origin.
Monophyletic groups include all the descendants from a
single ancestor, as well as that ancestor. If one lineage
emerging from a monophyletic group is removed, a paraphyletic group remains. In contrast, polyphyletic groups
result from convergent evolution, and the characters that
support the group are absent in the most recent common
ancestor (Kitching et al., 1998). In gene families these
principles approximate to orthology and paralogy (Fitch,
1970). Orthology refers to groups of genes that reveal
species phylogeny. Thus, within a monophyletic gene
group each species is represented by a single orthologue.
In contrast, paralogues reveal the history of a gene family.
Thus, within a gene group each species may be represented by a number of paralogues.
ª 2006 The Authors
Journal compilation ª 2006 Blackwell Publishing Ltd
Overview of phylogenetic analysis
Choosing a study group
Before embarking on phylogeny reconstruction it is
important to think about the specific biological question to
be answered. It is advisable to sample as densely as possible, to reduce artefactual associations between terminals
(Hillis, 1996). For example, if the aim of the reconstruction is
to examine when duplications occurred within a gene family
from a single species it is appropriate to sample the gene
family from that species comprehensively. However, if the
aim is to understand how a gene family evolved it is
important to sample as widely as possible, not only within
but also between species. A good first step is to survey the
literature pertaining to the group of interest. This will inform
the choice of species and genes to be included in the analysis and will indicate clades in which relationships are likely
to be resolved and statistically supported in the resulting
phylogeny. It will also highlight clades that should benefit
from greater sampling or more attention in alignments.
Sampling the group of interest
There are four main sources of data for reconstructing the
phylogeny of a gene family: published sequences of characterized genes (which may not always clearly appear in
562 C. Jill Harrison and Jane A. Langdale
Figure 1. Some phylogenetic terminology illustrated using a phylogeny of ARP genes redrawn
from Harrison et al. (2005) and rooted on a
Selaginella ARP gene, SkARP.
In the phylogeny, Osrs2 is the sister terminal to
Zmrs2, and together these form a monophyletic
group. PsPHAN and AtAS1 are paraphyletic with
respect to the monphyletic sister group containing AmPHAN, NtPHAN and LePHAN. ‘Clade’ and
‘monophyletic group’ can be used interchangeably, as can ‘grade’ and ‘paraphyletic group’.
(a) Terminals, branches, nodes, a clade and a
grade are indicated.
(b) To date, Antirrhinum is the only species
reported to have two ARP genes, AmPHAN1 and
AmPHAN2, and these are paralogues. The box
indicates a monophyletic (orthologous) gene
database searches, but should be familiar to the researcher
from the literature); gene databases such as NCBI; EST
project databases; and unpublished data from colleagues.
The advantage of including unpublished sequences from
EST databases is that it increases species sampling and also
increases the possibility of deducing the point at which gene
duplications occurred within the family of interest. In addition to searching a number of different databases, sampling
is optimized by using the TBLASTX option at the sequence
retrieval stage in BLAST searches (Altschul et al., 1997). As
opposed to other options, this program translates the nucleotide sequence in all six frames and compares the output
against all the translated sequences in the database. It
therefore maximises the potential for retrieving sequences
similar to the gene of interest.
Sequence retrieval
The number of sequences retrieved from BLAST searches
varies depending on the size of the gene family, and what is
chosen for inclusion in further analyses will vary accordingly. It is feasible to download all of the sequences for a
small gene family, but with a large gene family some
selection is required. It is important to be pragmatic at this
stage, because analysis of 100 sequences can take days to
compute. When sequences are retrieved from BLAST searches they are allocated an e-score, which is an indication of
the degree of similarity between the initial sequence used
for searches and the sequence retrieved. The closer the
e-value is to 0, the higher the degree of similarity between
the two sequences. For large gene families there may be a
clear cut-off point between the e-scores of the group of
interest and further gene family members. If there is no clear
cut-off, the literature can be used to identify genes in the list
of retrieved sequences that, on the basis of their function,
are likely to be outside the group of interest. The e-scores of
those sequences can then be used as a guide for the cut-off
point (also see Hall, 2001).
Amino acid or nucleotide data?
Both amino acid and nucleotide data can be analysed to
generate a phylogeny and there has been much debate
about which is best (e.g. Simmons et al., 2002a,b). The main
argument for using amino acid data to infer phylogeny is
that there are more possible character states for amino acids
as opposed to nucleotides (20 versus 4). For the same reason, alignment of amino acid sequence data is also generally
easier. However, the increased number of characters in
nucleotide sequences can lead to better resolution of the
tree, for example when the gene family of interest has just
one highly conserved domain that is barely divergent at the
amino acid level. The argument against using nucleotides is
that with only four possible character states at one position
there is a possibility that terminals may share a character
state by chance. However, such ‘saturation’ at a particular
position in the alignment is minimized by thorough sampling (Källersjö et al., 1999). Fortunately, many alignment
programs allow transition between sequence types, so both
possibilities can be tried with ease.
Phylogenetic methodology relies on the assumption that
the characters used to generate trees are homologous. For
gene family phylogenies careful alignment of sequence
data fulfils this requirement. Sequence alignment can be
achieved automatically or manually. Automatic alignments
may fail to correctly identify regions of conservation within
a gene, whereas manual alignments allow this but are
more labour intensive (see example in Figure 2 and further
discussion in Baldauf, 2003). Using published and auto-
ª 2006 The Authors
Journal compilation ª 2006 Blackwell Publishing Ltd, The Plant Journal, (2006), 45, 561–572
Building phylogenies 563
Figure 2. The effect of misalignment on tree topology.
(a) Part of a KNOX protein sequence alignment generated by CLUSTALW.
(b) Part of an equivalent sequence alignment generated manually.
(c) One of eight most parsimonious trees (L ¼ 12949) generated by analysis of the CLUSTALW alignment in (a).
(d) One of 124 most parsimonious trees (L ¼ 1892) generated by analysis of manually aligned data in (b).
Trees are rooted on a mouse homeodomain protein sequence. The origin of class I (I), and class II (II) KNOX clades are indicated on the trees. Members of the outgroup are indicated in red and genes from species that are taxonomically basal are indicated in blue. Although there were fewer trees generated from the CLUSTALW
alignment and the consensus is better resolved, there are major problems with the tree. The main problem is that the in-group is not monophyletic, as indicated by
the interspersed placement of the four out-group sequences. Furthermore, the placement of representatives from the most basal species is widespread, whereas
they should be together.
matic alignments as a guide to manual alignment offers a
good compromise, and is more rigorous than automatic
alignment alone. There are many programs for manual
alignment, including Se-Al (Rambaut, 1996) and BioEdit
(Tom Hall, Ibis Therapeutics, Carlsbad, CA, USA). These
have the benefit of allowing easy transition between nucleotide and translated sequence formats and also output
files in useful formats. Both are available free on the web.
ª 2006 The Authors
Journal compilation ª 2006 Blackwell Publishing Ltd, The Plant Journal, (2006), 45, 561–572
564 C. Jill Harrison and Jane A. Langdale
in question should collapse. Hence majority rule trees are
not informative about phylogeny.
Method of analysis
Once data are aligned, there are many different types of
phylogenetic analysis that can be implemented (recently
reviewed by Holder and Lewis, 2003). The type of analysis
used will be determined by compromise between the length
of computational time and the degree of rigour required. The
main techniques are distance, parsimony and likelihood
(including Bayesian analysis). All three can be performed
using PAUP* software (Swofford, 2003) that is available from
Sinauer Associates Inc. Publishers, Sunderland, MA, USA.
There are many alternative programs that perform the same
functions and are equally valid to use (see web resource list
at the end of this paper).
Distance. Distance methods [e.g. neighbour joining (NJ),
distance and minimum evolution] calculate pairwise distances between sequences, and group sequences that are
most similar. This approach has potential for computational
simplicity and therefore speed. However, distance methods
do not allow an analysis of which characters contribute to
particular groupings. As with other methods, the outcome
may depend on the order in which entities are added to the
starting tree, but because only one tree is outputted it is not
possible to examine conflicting tree topologies. Although
distance methods are often useful for making an initial tree,
they should be used for final trees with caution. Instead,
parsimony and likelihood are preferred because they have
the potential to rigorously explore the relationship between
the tree and the entities included. Parsimony and likelihood
use different criteria to choose the best trees. In these analyses the branches of a starting tree are rearranged to form
the tree that minimises the number of character state
changes (parsimony) or the tree that best fits the data (likelihood).
Parsimony. Parsimony assumes that shared characters in
different entities result from common descent. Groups are
built on the basis of such shared characters, and the simplest
explanation for the evolution of characters is taken to be the
correct, or most parsimonious one. With multiple characters, different groupings may be equally plausible, or
equally parsimonious, and therefore multiple trees are
generated. In such cases, a strict consensus tree should be
derived that includes only topologies that are not contradicted in any of the initial trees. If the strict consensus tree is
unresolved there is no congruence between initial trees, and
thus it is likely that the data used to build the tree are phylogenetically uninformative. A majority rule consensus tree
shows nodes that are consistent in half to all of the most
parsimonious trees and the percentage of trees in which a
given topology exists is shown on the branches. However,
since by definition all most parsimonious trees are considered equally good, if any one contradicts the others the node
Likelihood methods. In contrast to parsimony, maximum
likelihood analyses compute the probability that a data set fits
a tree derived from that data set, given a specified model of
sequence evolution. A good first step is to compare the data
against a set of models of sequence evolution and choose the
one that best describes the observed pattern of sequence
variation. Two programs in which this can be performed are
Modeltest (Posada and Crandall, 1998) and MrModeltest
(Nylander, 2004). Alternatively a user-specified model may
be chosen. This model of sequence evolution is then used in
the likelihood analysis. The analysis starts with a specified
tree derived from the input dataset (for example a NJ tree)
and swaps the branches on the starting tree until the tree with
the highest likelihood score (i.e. the best probability of fitting
the data) is gained. This score is a function both of the tree
topology and the branch lengths (number of character state
changes). Likelihood analysis therefore allows an explicit
examination of the assumptions made about sequence evolution. Likelihood methods are the most computationally
demanding techniques for phylogenetic analysis. Currently
only nucleotide data sets can be used to perform maximum
likelihood (ML) analyses in PAUP*. Bayesian inference is another likelihood method that is gaining popularity, but this
cannot yet be implemented in PAUP*. Instead, the program
MrBayes should be used (Huelsenbeck and Ronquist, 2001;
Huelsenbeck et al., 2001). In Bayesian analysis, a further set of
assumptions (termed priors) are inputted into the original
model and the branch swapping algorithms differ. Likelihood
methods produce a number of trees, one of which is usually
found to be the most likely tree.
Rooting trees
If direct evidence of ancestor–descendant relationships is
absent, the direction of change must be inferred by rooting
the trees. Unless they are rooted, phylogenetic methods give
rise to branching diagrams from which it is impossible to
examine the direction in which traits change. In some instances it may not be necessary to root the generated trees.
For example, if the hypothesis is to test whether a group of
genes are orthologous, and those genes are dispersed
amongst other genes on the tree, the hypothesis is essentially refuted. However, in most cases knowledge of the
direction of change is fundamental to our understanding of
evolutionary processes.
There are two good ways to root molecular trees. Outgroup rooting (Maddison et al., 1984) compares the character
states in the group of interest (the in-group) with those in a
group that is closely related to, but definitely not in, the ingroup (the out-group). These differences are used to infer the
direction of character change in the resultant tree.
ª 2006 The Authors
Journal compilation ª 2006 Blackwell Publishing Ltd, The Plant Journal, (2006), 45, 561–572
Building phylogenies 565
Out-groups can be selected on the basis of prior knowledge of
the group of interest, or may become apparent during
alignment (see Figure 3). In gene families, appropriate outgroups share at least one conserved domain with the group
of interest. It is often not trivial to find an appropriate outgroup because in candidate sequences, domains that can be
adequately aligned for phylogenetic analysis may not provide sufficient variance and other domains may be too variant
to align. Performing two separate rounds of analysis can
solve this problem (Fitter et al., 2002; Harrison et al., 2005).
The first round of the analysis should sample as widely as
possible within the gene superfamily and should be carried
out using only sequence from the most conserved domains.
This will allow identification of gene clades that are most
closely related to the clade of interest and are potential outgroups for a second round of analysis. The second step
should include sequences from the group of interest, plus
place-holder representatives from the most closely related
clades identified in the first step as potential out-groups. The
root should be placed between the most distant of these
(selected as the out-group) and the rest (Figures 3, 4). At this
stage, it may also be possible to include extra sequence
outside the most conserved domain. If the group of interest
shares common descent, the sequences chosen as outgroups will all fall outside the in-group (Figure 4).
The second way of rooting is to use duplicated genes,
where sequences from one gene clade are used to root
another (Simmons et al., 2000). Duplicate gene rooting has
the advantage that it can reveal unexpected relationships
Clade of
Clade of interest
clades for
Other out-group
identified in the
first round of
Most distant
Out-group rooted
Figure 4. Out-group rooting for a gene family tree.
amongst the species or genes in the main clades where there
has previously been ambiguous rooting (Brown and Doolittle, 1995; Gogarten et al., 1989; Iwabe et al., 1989; Mathews
and Donoghue, 1999).
Rooting adds an extra node at the base of the tree, and by
convention rooted trees are drawn with a stalk at their base.
In trees with out-group rooting, the first or specified number
of branches on the tree is part of the out-group. Be aware
that as unrooted trees also branch at their base, they can be
easily mistaken for rooted trees.
Statistical support for trees
As phylogenetic trees represent historical patterns of relationship that are generally incompletely sampled, they are
Figure 3. Candidate out-group sequences.
Potential out-group sequences are indicated by a drop in sequence identity at the bottom of the alignment. In this instance the last five sequences are likely to
comprise the out-group, and a good placement for the root would be between the Gossypium (CO490788) sequence and the rest.
ª 2006 The Authors
Journal compilation ª 2006 Blackwell Publishing Ltd, The Plant Journal, (2006), 45, 561–572
566 C. Jill Harrison and Jane A. Langdale
hypotheses of relationship not factual depictions of evolutionary relationships. It is therefore hard to assess how
accurately they reflect true biological events. However, there
are ways in which the robustness of the data used for making the trees can be tested (note that this is not the same as
accuracy). The quantifiers most commonly used are bootstrap and jack-knife support values. The bootstrap value
shows the percentage of times that a clade appears when
individual characters in the data set are randomly removed
and replaced with data from another character from the
same data set, and the analysis performed again for a specified number of replications (Felsenstein, 1985). The jackknife value shows the percentage of times that a clade appears when a specified percentage of characters are randomly removed from the data set and the analysis
performed again (Farris et al., 1996). Reasonable support is
roughly 70% and high support 95% (Felsenstein, 1985; Hillis
and Bull, 1993).
In ML analyses, tree support can be evaluated by bootstrapping (which is more time-consuming than with parsimony analyses), and conflict between alternative tree
topologies can be examined with less likely trees. In
contrast, Bayesian analysis outputs both a tree and the
support for that tree together. This is conceptually equivalent to ML with bootstrapping. The support value shown on
the nodes of the tree is termed the posterior probability. The
posterior is the actual probability of a node being correct.
This sounds great, but given that the topology may be prone
to sampling and analytical artefacts, and that the probability
is dependent on the model used for analysis, it may not
reflect evolutionary history!
How to build phylogenetic trees
Database searches and sequence retrieval
(i) Search NCBI by opening the BLAST page on the NCBI
website (http://www.ncbi.nlm.nih.gov/BLAST/). Paste
the sequence that you wish to place in the context of
a phylogeny into the box that opens and select the
TBLASTX option from the ‘choose a translation’ menu.
Use the ‘nr’ databases. Scroll to the bottom of the page
and make sure that the ‘sequence retrieval’ box is ticked
in the format section. Select ‘BLAST’ and then ‘format’ to
start the search. When the results are returned, scroll
through the list to check e-scores and then go through
selecting individual sequences to retrieve by checking
the boxes next to the gene identifier numbers. Do not
select genomic sequences or those with poor e-scores
(values closest to 0 reflect greatest similarity to the
search sequence). When all the required boxes have
been checked, click on ‘get selected sequences’. On the
next page change ‘display’ to FASTA, and then ‘send’ all
to file. Save the file to the desktop, open in Word and
then save as a text only (.txt) file. Examine the file and
then using a combination of published data and the
individual records on NCBI (open sequence file and
select ‘cds’ to get coding sequence), identify the start
codon in each sequence. Delete sequence 5¢ to the ATG
from all sequences in the text file, to ensure that all
sequences start at the beginning of the first open
reading frame.
(ii) Search EST databases at http://www.plantgdb.org/
PlantGDB-cgi/blast/PlantGDBblast. Unless you specifically want to, do not search the databases from species
that are already genetically well characterized for the
gene of interest, as sequences retrieved are likely to be
identical to those retrieved from NCBI. Instead access
individual databases and check ‘EST’ and ‘cDNA’ boxes
for each species to be examined. Select the TBLASTX
option, paste in the sequence of interest and run the
search. Display the retrieved sequences and copy and
paste into the text file.
(iii) Search algal EST databases on http://www.kazusa.or.jp/
en/plant/database.html and http://www.chlamy.org/
chlamydb.html. Select TBLASTX or BLASTN, choose the
database of interest and run the search. Add the
sequences retrieved to the text file.
(iv) Add any unpublished sequences that you have to the
text file.
Formatting retrieved sequences
(i) The text file must be formatted so that it is compatible
with alignment programs. Open the file in Word and
first remove all of the spaces and returns using the
‘replace’ function under the edit menu.
(ii) Scroll through the sequences and insert a return
between the end of each identifier line and the
sequence, and between the end of each sequence and
the next identifier line. Delete any unnecessary information in the identifier line, any polyadenylation
sequences and any genomic sequences (Figure 5). Do
not remove the species name or the GenBank accession
(iii) Save as a text file.
(iv) Next, identify the correct reading frame for each
sequence. To do this, open the translation program at
http://us.expasy.org/tools/dna.html and copy and paste
each sequence from the text file in turn into the box.
Select ‘compact’ as the output format and select ‘translate sequence’. Examine the output to identify the
correct frame (informed by an idea of what conserved
motifs are present in the gene family). If necessary,
copy and paste the sequence into the reverse complement program at http://www.bioinformatics.vg/
ª 2006 The Authors
Journal compilation ª 2006 Blackwell Publishing Ltd, The Plant Journal, (2006), 45, 561–572
Building phylogenies 567
Figure 5. Sequence format for alignment.
(a) Downloaded sequence format.
(b) Modified format.
copy and paste the output reverse sequence into the text
file in place of the original sequence. Delete nucleotides
as appropriate in the text file to adjust the reading frame
of each sequence to 5¢–3¢ frame 1. If conserved motifs
cannot be identified in any of the six frames, delete the
sequence from the file.
(v) To identify and remove duplicate sequences, open
CLUSTALW (Ramu et al., 2003) at http://www.ebi.ac.uk/
clustalw/ (It is equally valid to use CLUSTALX, but we find
that versions that are currently freely accessible on the
web are unreliable). Copy and paste the sequences
from the text file into the alignment window and run the
program. When the data are returned, open the alignment file, examine the sequences and identify duplicates. Go back to the initial text file and delete all
duplicated sequences, saving the longest in each case
(this is tedious to do manually but we have been unable
to find a program that does pairwise alignments of the
whole set and deletes the shortest sequence every time
it finds an identical match.
Aligning data
Remember that this step is the assessment of character
homology and is therefore fundamentally important to the
output. Rigorous alignment of a large data set can take
months; it is also subjective and therefore it is always worth
getting a second opinion. The steps outlined are optimized
for the alignment of coding sequences, but can equally be
applied to other nucleotide sequences by omitting instructions that refer to amino acid alignment. Alignment protocols differ slightly depending on whether you are using a
Macintosh or a PC. Both approaches are outlined below.
(i) Download Se-Al from http://evolve.zoo.ox.ac.uk/
software.html?id¼seal. To load your sequences into
the Se-Al program, open the program, choose ‘open’
under the file menu and select your text file. Files do not
open if they are not formatted exactly as shown in
Figure 5. Save this Se-Al alignment file as version 1 – it
will be the only one where you can revert to DNA
(ii) To automatically align sequences in amino acid format, first change the sequence format to translated
sequence by changing the alignment type to ‘amino
acid’ under the alignment menu. Select ‘export’ under
the file menu, select ‘FASTA’ under file format and
check ‘export alignment as displayed’. Save the FASTA
file. Open CLUSTALW again, copy and paste the data
from the FASTA file into the alignment window, select
‘pir’ under output format and run the program. When
the data are returned, open the alignment file. Select
‘save page as’ from the browser file menu and save to
the desktop. Reopen the file from within Word and
save as a .txt file. Finally, open the text file from within
Se-Al. Save the Se-Al file as version 2. You now have
the alignment in a form that can be further edited
(iii) Spend time scanning the alignment for regions of
conservation that CLUSTALW may have missed (it is
easier to scan the sequence if you select ‘use block
colours’ under the alignment menu). If necessary,
manually adjust the alignment (see Baldauf, 2003 for
further guidelines). To move entire sequences, double
click on the sequence and drag it in either direction.
To move blocks of sequence, select the block and drag it.
(iv) When the alignment is finished, copy and paste the
conserved regions into a new file and save it as version
3. Keep the alignment file for reference and publication.
If the tree needs to be built using DNA data, go back to
the Se-Al version 1 file and manually edit it to look
identical to the version 3 file. Change alignment type to
‘DNA’ under the alignment menu and then save the file
as version 4.
ª 2006 The Authors
Journal compilation ª 2006 Blackwell Publishing Ltd, The Plant Journal, (2006), 45, 561–572
568 C. Jill Harrison and Jane A. Langdale
(v) To transfer Se-Al alignment version 3 or 4 files to PAUP*,
select ‘export’ under the file menu. Select ‘NEXUS’
under file format and check ‘export alignment as
displayed’. Save the NEXUS file.
(i) Download BioEdit alignment software from (http://
To load sequences into the program, choose ‘open’
under the file menu and select your text file. Save this
BioEdit file as alignment version 1 – it will be the only
one where you can revert to DNA sequence.
(ii) Select names of sequences with the mouse and change
data to amino acid format by selecting ‘translate or
reverse translate (permanent)’ under sequence menu.
Save the file as a FASTA file.
(iii) To automatically align sequences in amino acid format,
open CLUSTALW, upload the FASTA file, select ‘pir’ as
the output option and run the program. When the data
are returned, open the alignment file, select ‘save as’
from the browser file menu and save as a text file to the
desktop. Finally, open the text file from within BioEdit.
Save the BioEdit file as alignment version 2. You now
have the alignment in a form that can be further edited
(iv) Spend time scanning the alignment for regions of
conservation that CLUSTALW may have missed (it is
easier to scan the sequence if you select ‘inverse colours’
under the view menu). If necessary, manually adjust the
alignment. Move sequences along the alignment individually by clicking on the sequence with the mouse,
unclicking and then dragging the sequence across. As it
is impossible to move entire sequences bidirectionally,
ascertain which sequence has regions of conservation
most 3¢ and align other sequences with respect to that.
(v) When the alignment is finished, save the file again as
alignment version 2. Delete unaligned domains by
changing the mode window to ‘edit’, highlighting the
block to be deleted and using the ‘delete’ key on the
keyboard. Save the modified file as BioEdit alignment
version 3. Keep the alignment file for reference and
publication. If the tree needs to be built using DNA data,
go back to the BioEdit version 1 file, select names of
sequences with the mouse and change data to amino
acid format by selecting ‘toggle translation’ under the
sequence menu. Manually edit the alignment to look
identical to the version 3 file. Change back to DNA
sequence by selecting names of sequences with the
mouse and selecting ‘toggle translation’ again under
the sequence menu. Save the file as BioEdit alignment
version 4.
(vi) To transfer BioEdit version 3 or 4 files to PAUP*, select
‘file’, ‘export’, ‘sequence alignment’ and then ‘nex’.
Running a parsimony analysis in PAUP*
The number of trees resulting from a parsimony search
increases hugely with increasing numbers of entities included in the analysis (Felsenstein, 1978). With large data sets it
is computationally unfeasible to find all possible trees. Instead, heuristic search strategies are used to examine a
subset of the trees. The starting parameters of a heuristic
search can affect the outcome of the search, and search
strategies can be modified to affect the stringency of the
analysis. Changeable parameters include the sequence of
addition of terminals to the starting tree, branch swapping
algorithms, and the number of replicates of the analysis run.
The significance of these and the effects of their alteration
are discussed in detail in Felsenstein (2004). The search
strategy of Catalán et al. (1997) is reasonably stringent and is
recommended here, as the order in which entities are added
to the starting tree is randomized, emphasis is placed on
branch swapping at deep rather than shallow nodes, and the
whole search is replicated 1000 times. Again, protocols differ
for Macintosh and PC users.
(i) Open the NEXUS file in PAUP* and click on ‘execute’
under the file menu. Problems in this step arise from an
incorrectly formatted NEXUS file; if new entities (taxa) or
characters have been added, the parameters of the
NEXUS file (Ntax, Nchar) need to be changed too.
(ii) Under the analysis menu select ‘parsimony’ and ‘heuristic’ search.
(iii) Set the parameters for the heuristic search. In the box
that opens select ‘best trees’ under keep and then select
‘maxtrees’. In the menu that appears, select ‘automatically increase by 100’ and select OK. Next change the
main menu from ‘general search options’ to ‘stepwise
addition options’. In the next screen, select ‘random’
under addition sequence, set replicates to ‘1000’ and
select hold ‘1’ tree at each step. Next change the main
menu to ‘branch swapping options’ and select ‘TBR’.
Select save no more than ‘2’ trees ‡ score ‘5’ for each
replicate and ‘swap on best trees’ only. Select search to
start the analysis.
(iv) When the analysis has finished, note the number of
trees, the length of the shortest tree, and save the trees
using the ‘save trees to file’ option under the ‘trees’
menu. The trees are saved as data, not as a pictorial
representation of the tree. If a pictorial representation is
required, follow the steps outlined in 7. To reopen the
.tre file at a later date, first reopen the original .nex file
by selecting ‘open’ under the file menu and then
selecting ‘execute’. Under the trees menu, select ‘get
trees from file’, select the .tre file of interest and then
select ‘get trees’.
ª 2006 The Authors
Journal compilation ª 2006 Blackwell Publishing Ltd, The Plant Journal, (2006), 45, 561–572
Building phylogenies 569
(v) If there is more than one tree, select ‘compute the
consensus tree’ under the trees menu. Select both the
strict and majority rule (50%) consensus options.
(vi) Root the trees by selecting ‘root trees’ under the trees
menu and then ‘rooting options’. On the screen that
appears select ‘root tree at internal node with basal
polytomy’. Then select ‘define outgroup’ and on the
next screen, use the mouse to highlight one sequence
from the out-group. Select ‘to outgroup’, then ‘OK’ in
this and the next screen and ‘root’ in the final screen. Do
not use the entire out-group as the root, as this forces
the putative in-group to form a monophyletic group.
(vii) Go to the ‘print trees’ or ‘print consensus trees’ option
under the trees menu and either print the tree, or select
‘preview’ and save as a PICT file for manipulation in
other programs.
ets). Similarly, the out-group identifier should be inserted minus the square brackets.
Run the analysis by selecting ‘execute’ under the file
menu. The data will be returned in the two elected files.
To view the trees, download the TreeView program
from http://taxonomy.zoology.gla.ac.uk/rod/treeview.
html. Open TreeView, and open the tree file from within
Use the icons at the top of the tree window to choose
the plot type (including plots with branch lengths).
Root the trees by selecting ‘define outgroup’ and ‘root
with outgroup’ from the ‘tree’ menu.
To save the tree for manipulation in other graphics
programs select ‘print trees’ under the trees menu and
then select ‘picture’ to save the file.
Statistical support for trees
PC. To perform optimality analyses in PAUP* on the PC,
commands are required to set the search strategy. Commands are listed in the PAUP* user manual (commands reference document), available at http://paup.csit.fsu.edu/
downl.html#Anchor-58521. Commands can be typed into
the command line at the bottom of the display window, or
can alternatively be pasted directly into the NEXUS file. The
commands listed below instigate analyses equivalent to
those outlined for Macintosh users. To change search
strategy parameters refer to the PAUP* manual.
(i) Open the NEXUS data file from within PAUP* in the ‘edit’
(ii) To set analysis parameters for parsimony analysis,
scroll to the bottom of the data matrix and after
‘End;’ hit return and type the following commands:
begin paup;
set criterion¼parsimony maxtrees¼100
Hsearch start¼stepwise addseq¼random
nreps¼1000 savereps¼yes nchuck¼2
chuckscore¼5 dstatus¼none;
savetrees file¼[the elected name1] brlens¼yes;
set root¼outgroup;
outgroup [the name of the out-group taxon that will be
used as the root]/only;
gettrees file¼[the elected name1];
contree all/majrule¼yes treefile¼[the elected
Note that square brackets make a command invisible to
PAUP*. The commands listed will result in the production of most parsimonious trees (saved to the elected
file) plus a strict consensus and a majority rule tree
compiled from the most parsimonious trees (saved to
another elected file). The elected names should be
inserted as indicated above (minus the square brack-
Bootstrap and jack-knife analyses resample the original data
set used for phylogenetic inference, and rerun analyses.
Thus, the trees obtained may not be as good an estimate of
phylogeny as the original trees obtained. Therefore, support
values obtained should be manually superimposed on
appropriate nodes of parsimony or ML trees.
(i) To obtain bootstrap and jack-knife values open the
NEXUS file that was used to make the original tree in
PAUP* and ‘execute’ it.
(ii) Select the bootstrap or jack-knife option under the
analysis menu. Enter ‘500 replicates’ for each analysis,
with ‘50%’ deletion for jack-knife (see Farris et al. (1996)
and Felsenstein (2004) for a discussion of this).
(iii) Run the parsimony analysis by repeating steps (ii)–(vii)
above, selecting ‘print bootstrap consensus’ at the last
stage. The topology of the trees may differ from the
original tree. Manually transfer the bootstrap or jackknife percentages across from their respective trees to
those clades which are also present in the original tree.
(i) To perform a bootstrap analysis with equivalent search
parameters to those specified for Macintosh users,
open the NEXUS file from within PAUP* in the ‘edit’
mode, scroll to the end of the data matrix and type the
following commands after ‘End;’:
begin paup;
set maxtrees¼100 increase¼auto;
bootstrap nreps¼500 conlevel¼50 treefile¼[the
elected name3] keepall¼yes cutoffpct¼50/
start¼stepwise addseq¼random nreps¼1000
savereps¼yes nchuck¼2 chuckscore¼5
ª 2006 The Authors
Journal compilation ª 2006 Blackwell Publishing Ltd, The Plant Journal, (2006), 45, 561–572
570 C. Jill Harrison and Jane A. Langdale
savetrees file¼[the elected name3];
set root¼outgroup;
outgroup [one of the outgroup taxon names]/only;
Note that the elected name should be different from
those used in the original analysis. Select ‘execute’
under the file menu to run the analysis.
(ii) To perform a jack-knife analysis with equivalent search
parameters to those specified for Macintosh users,
open the original NEXUS file as above and type the
following commands at the end of the file:
begin paup;
set maxtrees¼100 increase¼auto;
jackknife pctdelete¼50 nreps¼500 conlevel¼50
treefile¼[the elected name4] keepall¼yes
cutoffpct¼50 grpfreq¼yes/start¼stepwise
addseq¼random nreps¼1000 savereps¼yes
nchuck¼2 chuckscore¼5 dstatus¼none;
set root¼outgroup;
outgroup [one of the outgroup taxon names]/only;
savetrees file¼[the elected name4];
Again, the elected name should be different from those
used previously. Select ‘execute’ under the file menu to
run the analysis.
(iii) The display window will show statistical support for
nodes of the trees. The topology of the trees may differ
from the original tree and therefore bootstrap or jackknife percentages have to be manually transferred to
those clades which are also present in the original tree.
Bootstrap and jack-knife values will not transfer to
TreeView and thus trees must be printed from the PAUP*
display window using ScreenGrab software (http://
or equivalent. This can involve more than one screen
grab per tree!
Likelihood analysis
For ML trees, the first step is to find a model of sequence
evolution that fits the DNA changes in the aligned sequences
that are being used. This can be done using the Modeltest
program (Posada and Crandall, 1998) or MrModeltest
(Nylander, 2004) in conjunction with PAUP*. We advocate
using Modeltest, as it tests more models of evolution than
MrModeltest. Note that only DNA data matrices can be
analysed with ML in PAUP*.
(i) Download Modeltest from http://darwin.uvigo.es/.
(ii) Open the NEXUS file in PAUP* in the ‘edit’ mode.
(iii) After the ‘End;’ command at the end of the data matrix,
type in a return and then:
default lscores longfmt¼yes;
Select ‘execute’ under the file menu.
(iv) Open the modeltest folder, the PAUPblock folder within
it and the ‘modelblock PAUPb10’ program. Select ‘execute’ under the file menu to run the program. The
results will automatically be saved as ‘modelscores’ in
the ‘PAUPblock’ folder.
(v) Open the modeltest folder, the BIN folder within it and
the modeltest program. In the window that appears,
select ‘file’ and then browse to select the modelscores
file (the argument box remains empty.) Run the
program (click OK). This instructs the program to run
through all the different models of sequence evolution
featured by the program and to test sequence changes
within the data against each of them.
(vi) When the data are returned, scroll through the file to the
Akaike information criterion (AIC) section. Copy the
lines of text in this section that start with ‘begin paup’
and end with ‘End;’. Paste this text into the end of the
NEXUS file.
(vii) To run the ML analysis, select ‘execute’ under the file
menu. Select ‘likelihood’ and ‘heuristic search’ under
the analysis menu. In the window that opens, change
‘general search options’ to ‘starting tree options’ and
select ‘neighbour joining’. Next, change the main
menu to ‘branch swapping options’ and select ‘TBR’.
Select save no more than ‘2 trees ‡score 5’ for each
replicate and ‘swap on best trees’ only. Select ‘search’
to start the analysis.
(viii) Root the tree and display it as before [steps (vi) & (vii)
under parsimony].
(i) Download Modeltest from http://darwin.uvigo.es/.
(ii) To perform a likelihood analysis with equivalent
search parameters to those specified for Macintosh
users, open the NEXUS file in PAUP* in the ‘edit’
mode. Remove or bracket out all of the commands
that were used for parsimony analyses and then after
the ‘End;’ command at the end of the data matrix
default lscores longfmt¼yes;
Save the file as ‘likelihood start’ and then save a
second version as ‘likelihood analysis’.
(iii) Open the ‘likelihood start’ NEXUS alignment in the
‘edit’ mode of PAUP*. After the ‘End;’ command at the
end of the data matrix, paste in the text from the
model fit file available at http://www.plants.ox.ac.uk/
(iv) Execute the file to see which model best fits the data.
The results will be saved automatically as a ‘model
scores’ file in the folder from which the program was
ª 2006 The Authors
Journal compilation ª 2006 Blackwell Publishing Ltd, The Plant Journal, (2006), 45, 561–572
Building phylogenies 571
(v) Close the likelihood start file.
(vi) Download the program mtgui from http://www.genedrift.org/mtgui.php. Open the mtgui program, click
‘select file’ and open ‘model’. In the window that
appears, select ‘modeltest’. Scroll through the window
that opens to the ‘AIC’ parameters and copy the ‘Lset’
commands. Close the modeltest file.
(vii) Open the likelihood analysis NEXUS file in the edit
mode of PAUP* and type the following commands at the
end of the file.
Begin paup;
Set criterion¼likelihood;
[paste the Lset commands from the previous step here]
Hsearch start¼nj nchuck¼2 chuckscore¼5
savetrees format¼nexus brlens¼yes append¼yes
file¼[the elected name5];
lscores 1/scorefile¼[the elected name5].sf
set root¼outgroup;
outgroup [the name of the required taxon]/only;
showtrees all;
(viii) ‘Execute’ the NEXUS file. The tree produced is a rooted
ML tree.
Description of trees
There are a number of numerical outputs from phylogenetic analyses that describe how the data used to infer
phylogeny fit the resultant tree. In particular, the number of
trees and tree length, are generally reported. The number of
trees is greater with more conflict between possible outcomes. The tree length shows the number of character
changes in a given tree. These values can be found in the
PAUP* display window once an analysis is completed and
should also be recorded in the saved tree file. Tree length is
informative when considering the evolution of particular
characters within a group, and therefore is particularly
useful where critical evaluation of character evolution is
the case of insufficient phylogenetic signal from the data,
it is possible that the alignment stringency was too high,
especially if only the most conserved domains have been
accepted. The alignment should be checked to see where it
might be possible to include more data. If protein sequence was used to infer phylogeny, the analysis could be
rerun using the nucleotide data. In instances of poor
resolution, it is also common for the statistical support to
be very low. Look at the support values for the clades. If
the support values are good to high, you can be reasonably confident that derived tree reflects the data used to
generate it.
The next step is to compare the strict consensus
parsimony and ML trees to see if they are congruent.
Trees that are congruent have the same topology, so that
regardless of the order in which terminals appear on the
page the branching order and sister group relationships
are equivalent. If they are not congruent, it is likely that
one or both of the techniques used is suffering from an
artefact, such as ‘long branch attraction’ (Sanderson et al.,
2000). This arises when groups are formed on the basis of
similarity rather than homology, and may be difficult to
detect, but can be minimized by maximising sampling. The
outcome of a likelihood analyses may also be subject to
artefacts, as it is dependent on the model of molecular
evolution specified.
If the parsimony and ML trees are congruent, look at
the relationships between the terminals of interest and
see how they answer the biological question that was
originally posed in terms of orthology and paralogy, and
in terms of monophyly, paraphyly and polyphyly (Figure 1). Another useful concept is that of ‘sister groups’,
which refers to clades that are most closely related to
each other. For example, in Figure 1 the monocot
sequences form a sister group to the eu-dicot sequences.
Note that sister group relationships are not dependent on
the graphical representation: the sister group relationship
is the same whether the monocot or the eu-dicot clade
comes at the top of the tree. Remember to bear in mind
that the initial taxon sampling and alignment steps can
have a significant effect on the quality of the tree
Interpreting a phylogenetic tree
Alignment and tree presentation
Once a phylogenetic tree is generated, what inferences can
be drawn from it? First look at the strict consensus parsimony tree. A totally unresolved tree (the branches all
come out from the same node) may indicate a rapid
radiation (e.g. Richardson et al., 2001), insufficient phylogenetic information in the original alignment or incongruence between tree topologies. It should be possible to
distinguish between the latter two possibilities by looking
at both the alignment and the individual tree topologies. In
Once alignments and trees are satisfactorily understood
they can be prepared for presentation. Alignment and tree
files should be placed in the public domain, either as an
image file in publications or as an electronic file on a suitable web-based database such as TreeBASE (http://
www.treebase.org/treebase/). Trees should be displayed
and annotated such that they clearly convey the understanding that has been gained from phylogeny reconstruction.
ª 2006 The Authors
Journal compilation ª 2006 Blackwell Publishing Ltd, The Plant Journal, (2006), 45, 561–572
572 C. Jill Harrison and Jane A. Langdale
Further reading
A good review of phylogenetic principles and interpretation
of gene family trees is provided by Thornton and DeSalle
(2000). An accessible and an in-depth theoretical guide to
phylogeny reconstruction are given by Page and Holmes
(1998) and Felsenstein (2004) respectively. Finally, Hall
(2001) provides a basic practical guide to navigating some
phylogenetic programs, with minimal theoretical coverage.
Useful web resources
Joe Felstenstein’s website (http://evolution.genetics.
washington.edu/phylip/software.html) has a reasonably
comprehensive list of links to phylogeny programs available
on the web. Some further resources that the testers of this
protocol found useful are found on: http://www.soe.ucsc.
We thank Daniel Barker, Sheila McCormick, Elizabeth Moylan, Robert
Scotland, Andrew Smith, and two anonymous reviewers for constructive comments on the manuscript. Work in the Langdale group
is supported by the BBSRC and the Gatsby Charitable Foundation.
Altschul, S.F., Madden, T.L., Schäffer, A.A., Zhang, J., Zhang, Z.,
Miller, W. and Lipman, D.J. (1997) Gapped BLAST and PSIBLAST: a new generation of protein database search programs.
Nucl. Acids Res. 25, 3389–3402.
Baldauf, S.L. (2003) Phylogeny for the faint of heart: a tutorial.
Trends Genet. 19, 345–351.
Brown, J.R. and Doolittle, W.F. (1995) Root of the universal tree of
life based on ancient amino-acyl-tRNA synthetase gene duplications. Proc. Natl Acad. Sci. USA 92, 2441–2445.
Catalán, P., Kellogg, E.A. and Olmstead, R.G. (1997) Phylogeny of
Poaceae subfamily Pooideae based on cp ndhF gene sequences.
Mol. Phyl. Evol. 8, 150–166.
Farris, J.S., Albert, V.A., Källersjö, M., Lipscomb, D. and Kluge, A.G.
(1996) Parsimony jacknifing outperforms neighbour-joining.
Cladistics, 12, 99–124.
Felsenstein, J. (1978) The number of evolutionary trees. Syst. Zool.
27, 27–33.
Felsenstein, J. (1985) Confidence limits on phylogenies: an approach using the bootstrap. Evolution, 39, 783–791.
Felsenstein, J. (2004) Inferring Phylogenies. Sunderland Massachusetts: Sinauer Associates.
Fitch, W.M. (1970) Distinguishing homologous from analogous
proteins. Syst. Zool. 19, 99–113.
Fitter, D.W., Martin, D.J., Copley, M., Scotland, R.W. and Langdale,
J.A. (2002) GLK gene pairs regulate chloroplast development in
diverse plant species. Plant J. 31, 713–727.
Gogarten, J.P., Kibak, H., Dittrich, P. et al. (1989) Evolution of the
vacuolar Hþ-ATPase: implications for the origin of eukaryotes.
Proc. Natl Acad. Sci. USA 86, 6661–6665.
Hall, B.G. (2001) Phylogenetic Trees Made Easy: A How-to Manual
for Molecular Biologists. Sunderland Massachusetts: Sinauer
Harrison, C.J., Corley, S.B., Moylan, E.C., Alexander, D.L., Scotland,
R.W. and Langdale, J.A. (2005) Independent recruitment of a
conserved developmental mechanism during leaf evolution.
Nature, 434, 509–514.
Hennig, W. (1966) Phylogenetic Systematics. Urbana, Illinois: University of Illinois Press.
Hillis, D.M. (1996) Inferring complex phylogenies. Nature, 383, 130–
Hillis, D.M. and Bull, J.J. (1993) An empirical test of bootstrapping as
a method for assessing confidence in phylogenetic analysis. Syst.
Biol. 42, 182–192.
Holder, M. and Lewis, P.O. (2003) Phylogeny estimation: traditional
and Bayesian approaches. Nat. Rev. Genet. 4, 275–284.
Huelsenbeck, J.P. and Ronquist, F. (2001) MRBAYES: Bayesian
inference of phylogenetic trees. Bioinformatics, 17, 754–755.
Huelsenbeck, J.P., Ronquist, F., Nielsen, R. and Bollback, J.P. (2001)
Bayesian inference of phylogeny and its impact on evolutionary
biology. Science, 294, 2310–2314.
Iwabe, N., Kuma, K.-I., Hasegawa, M., Osawa, S. and Miyata, T.
(1989) Evolutionary relationship of archaebacteria, eubacteria,
and eukaryotes inferred from phylogenetic trees of duplicated
genes. Proc. Natl Acad. Sci. USA 86, 9355–9359.
Källersjö, M., Albert, V.M. and Farris, J.S. (1999) Homoplasy
increases phylogenetic structure. Cladistics, 15, 91–93.
Kitching, I.J., Forey, P.L., Humphries, C.J. and Williams, D.M. (1998)
Cladistics (2nd edn): The Theory and Practice of Parsimony
Analysis. Oxford, UK: Oxford University Press.
Maddison, W.P., Donoghue, M.J. and Maddison, D.R. (1984) Outgroup analysis and parsimony. Syst. Zool. 33, 83–103.
Mathews, S. and Donoghue, M.J. (1999) The root of angiosperm
phylogeny inferred from duplicate phytochrome genes. Science,
286, 947–950.
Nylander, J.A.A. (2004) MrModeltest v2. Program Distributed by the
Author. Uppsala University: Evolutionary Biology Centre.
Page, R.D.M. and Holmes, E.C. (1998) Molecular Evolution: A Phylogenetic Approach. Oxford: Blackwell Science Ltd.
Posada, D. and Crandall, K.A. (1998) Modeltest: testing the model of
DNA substitution. Bioinformatics, 14, 817–818.
Rambaut, A. (1996) Se-Al: Sequence Alignment Editor (Se-Al
v2.0a11). (Available at: http://evolve.zoo.ox.ac.uk/).
Ramu, C., Sugawara, H., Koike, T., Lopez, R., Gibson, T.J., Higgins,
D.G. and Thompson, J.D. (2003) Multiple sequence alignment with
the Clustal series of programs. Nucl. Acids Res. 13, 3497–3500.
Richardson, J.E., Pennington, R.T., Pennington, T.D. and Hollingsworth, P.M. (2001) Rapid diversification of a species-rich genus of
neotropical rain forest trees. Science, 293, 2242–2245.
Sanderson, M.J., Wojciechowski, M.F., Hu, J.-M., Khan, T.S. and
Brady, S.G. (2000) Error, bias, and long-branch attraction in data
for two chloroplast photosystem genes in plants. Mol. Biol. Evol.
17, 782–797.
Simmons, M.P., Bailey, D.C. and Nixon, K.C. (2000) Phylogeny
reconstuction using duplicate genes. Mol. Biol. Evol. 17, 469–473.
Simmons, M.P., Ochoterena, H. and Freudenstein, J. (2002a) Amino
acid vs. nucleotide characters: challenging preconceived notions.
Mol. Phyl. Evol. 24, 78–90.
Simmons, M.P., Ochoterena, H. and Freudenstein, J. (2002b) Conflict between amino acid and nucleotide characters. Cladistics, 18,
Swofford, D.L. (2003) PAUP*: Phylogenetic Analysis Using Parsimony (* and Other Methods), Version 4.0b 10. Sunderland,
Massachusetts: Sinauer Associates.
Thornton, J.W. and DeSalle, R. (2000) Gene family evolution and
homology: genomics meets phylogenetics. Annu. Rev. Genomics
Hum. Genet. 1, 41–73.
ª 2006 The Authors
Journal compilation ª 2006 Blackwell Publishing Ltd, The Plant Journal, (2006), 45, 561–572
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF