Transcript Lecture5

Molecular
Phylogenetics
Phylogenetic trees are about visualizing
evolutionary relationships
“Nothing in Biology
Makes Sense Except in
the Light of Evolution”
Theodosius Dobzhansky (1900-1975)
Phylogeny
Hypothesis of evolutionary relationships
 Phylogenetic tree = graphical summary
of evolutionary history
 We have been using trees throughout
the semester

Now we will examine how to construct
them
 Phylogeny is only an estimate

Phylogenetics

Under Darwin’s hypothesis of common
descent  Species in the same genus
stem from a recent ancestor

Hierarchical classification reflects not a
mystical ordering of the universe, but
rather a real historical process
Phylogenies

Species tree (how are my species related?)




contains only one representative from each
species
when did speciation take place?
all nodes indicate speciation events
Gene tree (how are my genes related?)


normally contains a number of genes from a single
species
nodes relate either to speciation or gene
duplication events
Phylogenetic Trees

Diagram consisting of branches and nodes
A
B
C
D
E
terminal node
interior node
split (bipartition)
also written AB|CDE
or portrayed **---
branch (edge)
root of tree
Unrooted vs. rooted trees
Rooting a Phylogeny

Several methods used to identify polarity





Most commonly used is the outgroup
method
The character state of the target taxa is
compared with that of a relative that
diverged earlier
Outgroup represents the ancestral state
Identify outgroup from other phylogenetic
studies or fossil data
Good to use several outgroups at once
Rooting Using an Outgroup
1.
The outgroup should be a sequence (or set of sequences or
taxon) known to be less closely related to the rest of the
sequences (taxa) than they are to each other
2. It should ideally be as closely related as possible to the rest
of the sequences (taxa) while still satisfying condition 1
The root must be somewhere between the outgroup and the
rest (either on the node or in a branch)
The POINT of rooting (using an outgroup) is to include the
ancestor of the group of interest in the phylogeny!
Terms
Clade: A set of species (or
sequences) which includes all of the
species (or sequences) derived from a
single common ancestor
 Monophyly
 Polyphyly
 Paraphyly

Cladograms VS. Phylograms

Cladogram



Only shows you the relationships between taxa
Branch lengths provide no data!
Phylogram


Shows you relationships AND the amount of change
(evolution) inferred along each branch
Therefore, branch lengths are very important!
Cladogram
Phylogram [sometimes Phenogram]
(branch lengths mean something)
Cladograms VS. Phylograms
Species A
Species A
Species B
Species B
Species C
Species C
Species D
Species E
Species D
Species E
Species F
Species F
5 changes
Phylogenetics Terms

Monophyletic Group


Paraphyletic Group


All members are believed to stem
from a single common ancestor,
and the group includes this
common ancestor
Group that is monophyletic except
that some descendents of the
common ancestor have been
removed
Polyphyletic Group

consisting of unrelated lineages,
each more closely related to other
lineages not placed in the taxon
Cladistic Methods
Techniques that identify monophyletic groups
based on synapomorphies
 Synapomorphies define evolutionary
branching points
 Autapomorphies and ancestral characters do
not
 Must be able to identify homology of traits
and direction of change through time
(Polarization)

Homology

The features of organisms almost
always evolve from pre-existing features
of their ancestors

Unlikely that features arise de novo from
nothing…
Homology

Homologous features are derived from a
common ancestor


Organs of 2 organisms are homologous if they have
been inherited (& perhaps modified) from a single
organ of a common ancestor
A character may be homologous among species
but a character state may not


5 toed state is homologous in humans and lizards but
the 3 toed state is not homologous in Guinea pigs and
Sloths
The wings of birds and those of bats are not
homologous, although their forelimbs in general are
homologous structures (convergent evolution)
Maximum Parsimony (Cladistic)
Occam’s Razor
Entia non sunt multiplicanda praeter
necessitatem.
William of Occam (1300-1349)
The best tree is the one which requires the least
number of substitutions
Parsimony and Phylogeny

Most closely related taxa should have the
most traits in common
Assume that traits are independent, heritable, and
variable in target taxa
 Traits may be DNA sequence, presence or absence
of skeletal elements or floral parts, mode of
embryonic development, etc.
 Traits scored in different taxa must be
homologous

Parsimony and Phylogeny
Shared derived characters (ONLY) are
used to deduce the branching patterns
of the tree
 Synapomorphy


Synapomorphies are used to attach two
branches at a NODE on the tree
Molecular Synapomorphies
Molecular Synapomorphies
Parsimony and Phylogeny

Traits may revert to ancestral form
because of mutation or selection
This may destroy phylogenetic signal and
lead to reconstruction of misleading
relationships
 Reversal


Convergence and Reversal and
collectively known as Homoplasy
Molecular Homoplasy via Reversal
Parsimony and Phylogeny

Homoplasy
Creates noise in the data
 Some characters give conflicting
information about relationships
 Systematists try to minimize homoplasy in
a data set
 Choose characters that evolve slowly
relative to age of taxa

Parsimony and Phylogeny
Parsimony minimizes total amount of
evolutionary change in a tree
 Synapomorphies are usually more
common than convergence and reversal
 Most parsimonious trees minimize
homoplasy to give best estimate of
phylogeny

Fitch (equal-weighted) parsimony
Data for site 1 shown on tree topology for all 16 possible combinations of
states at the 2 interior nodes. Character length is 2 for this site.
Tree length (or tree score)
Total steps = 2 + 1 + 2 + 2 +
Character length
from site 1
C
A
B
A
B
B
D
C
D
D
C
225
(best)
+ 1 = 237
Character length
from site 2
A
237
...
241
(worst)
This value is used to
compare this tree
topology to other
tree topologies
(smaller is better)
Phylogenetic Characters

Which characters should be used to
reconstruct the correct phylogeny?

Morphological characters




ie, Skeleton
For fossils only morphological characters can be used
Morphological characters difficult to use because
taxonomic expert needed
Molecular characters



Allozymes, RFLPs, DNA sequences
MUST CHOOSE MOLECULAR MARKER THAT IS
APPROPRIATE
Best molecular marker is one which has plenty of
variation (=phylogenetic signal) yet not too much
homoplasy (not too variable!).
Phylogenetic Characters

Which characters should be used to
reconstruct the phylogeny?
Molecular data has the advantage that they
can be rapidly collected and scored
 However, homoplasy difficult to indentify



Only four bases: G, A, T, C
Multiple types of data (including
multiple gene sequences) often the best
What sequences should I use for
organism phylogenies?





Slowly evolving / Fast evolving
rRNA
mitochondrion
Nuclear
chloroplast
Other Phylogenetic methods
Parsimony is not the only method for estimating
phylogenetic relationships!…
Some pitfalls of Parsimony…
It can take quite a long time to
compute a Parsimony estimate of a
phylogeny…
 Also, parsimony may be very error
prone when:

rates of evolution are variable
 very divergent species (or OTUs) are
compares because it does not deal well
with accounting for homoplasy…

Other Phylogenetic methods

Other reconstruction methods

Distance (Phenetic) methods
e.g.: Neighbor joining and UPGMA
 Based on clustering technique
 Based on overall similarity



Not a cladistic method
Uses differences (distances) among character states to
group taxa
Using Distance
Methods to
Reconstruct
Phylogenetic
Relationships
Species with the LEAST
genetic distance (or
other distance)
between them are
assumed to be CLOSE
relatives
However, there are
MANY cases where this
may NOT be true!
Distance-Based Methods
(UPGMA, Neighbor Joining, etc..)
Distance methods are typically very
very fast and easy to use to estimate a
phylogenetic tree
 However, they are not cladistic because
they do not look for synapomorphies,
but rather overall similarity…


This means this method is also susceptible
to lots of error when a dataset has lots of
homoplasy…
Distance methods
Normally fast and simple
 e.g. UPGMA, Neighbour Joining,
Minimum Evolution, Fitch-Margoliash

Correction for multiple hits





Only differences can be observed directly –
not distances
All distance methods rely (crucially) on this
A great many models used for nucleotide
sequences (e.g. JC, K2P, HKY, Rev, Maximum
Likelihood)
AA sequences are infinitely more complicated!
Accuracy falls off drastically for highly
divergent sequences
Distance methods
Attempts to account for multiple hits using models
in distance methods
(observed vs. estimated amount of evol. distance)
Other Phylogenetic methods

Maximum likelihood assumes a particular model of
sequence evolution and calculates how likely each branch
arose based on the character data



Uses all data, even autapomorphies and invariant sites
Uses models of evolution designed to capture a pattern of change
across characters (e.g., DNA)
Allows us to account for complex patterns of nucleotide evolution
across regions of genes that may evolve very differently (thus, not
all types of changes are weighted evenly in determining the
phylogeny…)
Lets look at an example… although we will save more heated
discussions of patterns for Bayesian MCMCMC methods….
1.8
2
Gene 1
1.8
1.6
Gene 2
C-G
1.4
1.2
A-C
1
0.8
0.6
0.4
0.2
285
240
195
Length Along Genome
150
105
60
15
504
441
378
315
252
189
126
F
63
00
C
A-T
0
Relative Rate of Substitution (G-T = 1)
Within vs. Between Gene Variation
Transversions
Maximum Likelihood Methods
Likelihood methods are among the most accurate
methods to reconstruct phylogenies!
 However, they are VERY VERY computationally
intensive  a tree with 30 species may take several
days, with 100 species may take several months!
 New likelihood methods employing Bayesian
statistics along with Marcov Chain Monte Carlo
algorithms are helping to solve this problem and are
the cutting edge of phylogeny reconstruction these
days…

Likelihood Methods




Requires a model of evolution
Each substitution has an associated likelihood
given a branch of a certain length
A function is derived to represent the likelihood
of the data given the tree, branch-lengths and
additional parameter
So, the tree we get from ML is “the phylogeny
that is most likely to have produced the observed
data (under the model of evolution selected)”
The Likelihood Criterion
Given two trees, the one maximizing the
probability of the observed data is best
Site likelihood - probability of the data for one
site conditional on the assumed model of
evolution
 Tree score - sum of site log-likelihoods (term
score also general term for the derivative of the
lnL)
 Unlike parsimony tree lengths, log-likelihoods
are comparable across models as well as trees

Models can be made more parameter rich to
increase their realism

The most common additional parameters are:




A correction to allow different substitution rates for
each type of nucleotide change
A correction for the proportion of sites which are
unable to change
A correction for variable site rates at those sites
which can change
The values of the additional parameters will be
estimated in the process (e.g. PAUP)
A gamma distribution can be used to
model site rate heterogeneity
Long Branches Attract
In a set of sequences evolving at different
rates the sequences evolving rapidly are
drawn together
Distance methods are VERY VERY prone to
making this error
Parsimony is also prone to this error
Likelihood methods employ an ‘informed’ view
of character change (a model) which helps
identify situations which probably represent
homoplasy, thus decreasing LBA
Phylogenetic Methods…
It is useful to use a variety of tree
reconstruction methods
 If methods are congruent you have more
confidence in your reconstructions!

Reconstructing Phylogenies

Phylogenies can be useful tools to
answer important evolutionary
questions

One must always question the methods
used to reconstruct the phylogeny to be
confident in the results
Comparison of methods
Inconsistency
 Neighbour Joining (NJ) is very fast but depends on
accurate estimates of distance. This is more difficult
with very divergent data
 Parsimony suffers from Long Branch Attraction. This
may be a particular problem for very divergent data
 NJ (and less so, MP) can suffer from Long Branch
Attraction
 Parsimony is also computationally intensive
 Codon usage bias can be a problem for MP and NJ
 Maximum Likelihood is the most reliable but depends
on the choice of model and is very slow
 Methods may be combined

Finding the best tree…??

How do we find the best tree?

With a small number of taxa you can evaluate all
possible trees




Exhaustive search
With more taxa the amount of possible trees
increase exponentially
For 8 taxa in artiodactyl tree there are 10,395
possibilities
Must use a shortcut method to evaluate trees
Tree Space Search Strategies
Exhaustive
 Branch and Bound
 Heuristic Branch Swapping

Exhaustive Search
Sequences
Number of unrooted, binary trees
4
3
5
15
6
105
7
945
8
10,395
9
135,135
10
2,027,025
11
34,459,425
12
654,729,075
13
13,749,310,575
14
316,234,143,225
15
7,905,853,580,625
16
213,458,046,676,875
17
6,190,283,353,629,375
18
191,898,783,962,510,625
19
6,332,659,870,762,850,625
20
221,643,095,476,699,771,875
This run would just
about be finished had
we started it at the
time prokaryotes
diverged from eukaryotes (about 2.5
billion years ago!)
Branch-and-bound
At start, we know this tree has length 1982 steps
Lineage out
of contention
(> 1982 steps)
Lineage out
of contention
(> 1982 steps)
(best)
Ok, still under 1982 steps
Theoretically predicted to always find the best tree
-if using MP with a B&B search for 50 taxa, plan on between
2 days to 2 weeks
Tree Islands
This landscape has 5 peaks, only 1 of which represents
the global optimum.
best tree (globally optimal tree)
locally
optimal tree
Imagine this depicts “tree space”
Heuristic search algorithms are
“hill climbers” – they only climb up
-However, this type of search is
typically the only choice we have for
even small datasets…
Solution  Do 100-1000 replicates
starting in different parts of tree
space to find global optimum
Heuristic search started here will not
find global optimum
How confident am I that my
tree is correct?
Bootstrap values
Bootstrapping is a statistical
technique that can use random
resampling of data to determine
sampling error for tree topologies
How Reliable is a Phylogeny?

How do we evaluate confidence in a
tree?
Bootstrap values are percentages of the
number of times the same branch arose
after repeated sampling
 Bootstrap support over 70% indicates that
the correct relationship was probably found
 Investigators usually report any bootstrap
value over 50%

Bootstrapping phylogenies
Characters are resampled with replacement to
create many bootstrap replicate data sets
 Each bootstrap replicate data set is analysed (e.g.
with parsimony, distance, ML etc.)
 Agreement among the resulting trees is
summarized with a majority-rule consensus tree
 Frequencies of occurrence of groups, bootstrap
proportions (BPs), are a measure of support for
those groups

Bootstrapping – How reliable is our phylogeny or
part of our phylogeny??
Bootstrap
replicate
#1…
…repeat this
random resampling
of date lots of times
and see how many
of these ‘pseudodatasets estimate a
particular
relationship..
Bootstrap - interpretation

Bootstrapping is a very valuable and widely used technique (it is
demanded by some journals)

BPs give an idea of how likely a given branch would be to be
unaffected if additional data, with the same distribution, became
available

BPs are not the same as confidence intervals. There is no simple
relationship between bootstrap values and confidence intervals.
There is no agreement about what constitutes a ‘good’ bootstrap
value (> 70%, > 80%, > 85% ????)

Some theoretical work indicates that BPs can be a conservative
estimate of confidence intervals

If the estimated tree is inconsistent all the bootstraps in the world
won’t help you…..
Bootstrap - interpretation


Lets consider why bootstraps for a particular
relationship may be low….
This can mean 2 things:


There is conflicting signal in the data whereby some evidence
supports one relationship while other evidence supports
another
Could be, simply, very little evidence overall (not conflicting
evidence, but only one or two characters overall support this
relationship)


This is often the case if speciation is rapid and lineages split from
one another (or a common ancestor) rapidly
In general, rapid evolutionary radiations are nearly impossible to
accurately estimate without enormous amounts of characters
(because of low phylogenetic signal or evidence for exclusive
common ancestry)
Bootstrapping
59
71
Ochromonas
Symbiodinium
Prorocentrum
Loxodes
Tracheloraphis
Spirostomumum
Euplotes
Tetrahymena
Gruberia
16
59
26
71
16
21
Ochromonas
Symbiodinium
Prorocentrum
Loxodes
Spirostomumum
Tetrahymena
Euplotes
Tracheloraphis
Gruberia
Majority-rule consensus (with minority components)
Wim de Grave et al. Fiocruz bioinformatics training course
MP – Tree length based tests of
hypotheses

Templeton test
Compare the length (# steps) of the
optimal tree vs. the length of tree that
would result from the topological
hypothesis you want to test
 Do this statistically using a 1-tailed
Wilcoxan signed-rank test

Likelihood-based tests of
topologies

Kishino-Hasegawa test




Trees specified apriori
KH can be used to test whether two competing
hypotheses have significantly different likelihood
NB should not be used to test trees that have
been chosen on the basis of the data!
Shimodaira-Hasegawa test

Can be used to test confidence of ML tree
compared to related trees (e.g. second most likely
tree from the data)