1. Introduction to Phylogenetic Systematics
Mark Fishbein
Dept. Biological Sciences
Mississippi State University
13 October 2003

2. Which of these critters are most closely related?

3. alligator
gila monster
purple gallinule ?
gopher tortoise
kingsnake

4. Phylogeny Branching history of evolutionary lineages
New branches arise via speciation
Speciation occurs when gene flow is severed between populations
Phylogenetic relationships depicted as a tree

5. © W. S. Judd, et al., Plant Systematics

6. © W. S. Judd, et al., Plant Systematics

7. } Phylogenetic data Morphology Secondary chemistry Cytology
Allele frequencies
Protein sequences
Restriction sites
DNA sequences }
“Molecular” data

8. Molecular (genetic) data
Proteins
Serology (immunoassay)
Isozymes (electrophoretic variants)
Amino acid sequences
DNA
Structural (translocations, inversions, duplications)
Restriction sites
DNA sequences
Substitutions
Insertions/Deletions

9. What are genes?
From Raven et al. (1999), Biology of Plants

10. Genomes All of the genes within a cell are the genome
Genes located in the nucleus are the nuclear genome
Other genomes (organellar)
Mitochondrion: mitochondrial genome
Chloroplast: plastid genome

11. nucleus chloroplast mitochondrion
From Raven et al., 1999, Biology of Plants

12. Comparison of Genomes Nuclear Mitochondrial Plastid Size Large Small
Number
Multiple
Single
Shape of Chromosomes
Linear
Circular
Ploidy
Diploid
Haploid
Inheritance
Biparental
Uniparental

13. Structural rearrangements
Inversion
Crossing over, duplication, and loss
From Freeman and Herron (1998), Evolutionary Analysis

14. Chemistry of Genes DNA Parallel strands linked together
Linear array of units called nucleotides
Phosphate
Sugar: deoxyribose
One of four bases
Adenine (“A”)
Cytosine (“C”)
Guanine (“G”)
Thymine (“T”)

15. From Raven et al. (1999), Biology of Plants

16. DNA structure Paired strands are linked by bases
A must bond with T
G must bond with C
Each link is composed of a purine and a pyrimidine
A & G are purines
C & T are pyrimidines

17. DNA function
DNA is code for making proteins (and a few other molecules)
Proteins are the structures and enzymes that catalyze biochemical reactions that are essential for the function of an organism
DNA code is read and converted to protein in two steps
Transcription: DNA is copied to messenger RNA
Translation: messenger RNA is template for protein

18. DNA code
A gene is a code composed of a string of nucleotide bases (A’s, C’s, G’s, T’s)
A protein is composed of a string of amino acids (there are 20)
How does the DNA code get translated into protein?

19. DNA code
Each amino acid is coded for by at least one triplet of nucleotide bases in DNA
Each triplet is called a codon
There are 64 possible codons (4 bases, 3 positions = 43)

20. From Raven et al. (1999), Biology of Plants

21. DNA functional classes
Coding
Proteins (exons)
Ribosomes (RNA)
Transfer RNA
“Non-coding”
Introns
Spacers

22. From Raven et al. (1999), Biology of Plants

23. Homology in Molecular Systematics
Assess orthology
Align sequences
Homology is often implicit (is this a good thing?)

24. DNA Sequences and Homology
Homology: similarity due to common descent
How do we assess homology of DNA sequences?
Levels of homology
Locus
Allele
Nucleotide position

25. From W. P. Maddison (1997), Systematic Biology 46:527

26. Orthology vs. Paralogy
DNA sequences that are at homologous loci are orthologous
DNA sequences that are similar due to duplication but are at different loci are paralogous
Orthology may be best detected with a phylogenetic analysis of all sequences

27. From Martin & Burg (2002), Systematic Biology 51:578

28. Multiple Sequence Alignment
Goal: create data matrix in which columns are homologous positions
Problem: sequences vary in length
Why?
Insertions
Deletions

29. Simple Sequence Alignment
Taxon 1 GTACGTTG
Taxon 2 GTACGTTG
Taxon 3 GTACGTTG
Taxon 4 GTACATTG
Taxon 5 GTACATTG
Taxon 6 GTACATTG

30. Simple Sequence Alignment
Taxon 1 GTACGTTG
Taxon 2 GTACGTTG
Taxon 3 GTACGTTG
Taxon 4 GTACATTG
Taxon 5 GTACATTG
Taxon 6 GTACATTG

31. DNA Sequence Data MatrixG T A C T2 T3 T4 T5 T6

32. Slightly Less Simple Sequence Alignment
Taxon 1 AGAGTGAC
Taxon 2 AGAGTGAC
Taxon 3 AGAGTGAC
Taxon 4 AGAGGAC
Taxon 5 AGAGGAC
Taxon 6 AGAGGAC

33. Slightly Less Simple Sequence Alignment
Taxon 1 AGAGTGAC
Taxon 2 AGAGTGAC
Taxon 3 AGAGTGAC
Taxon 4 AGAG-GAC
Taxon 5 AGAG-GAC
Taxon 6 AGAG-GAC

34. Alignment Gaps
Gaps are inserted to maximize homology across nucleotide positions
Gaps are hypothesized indels
Inserting a gap assumes that an indel event is a better explanation of the differences among sequences than nucleotide substitution

35. Taxon 1 AGAGTGAC Taxon 2 AGAGTGAC Taxon 3 AGAGTGAC Taxon 4 AGAGGAC
3 substitutions
0 indels
0 substitutions
1 indels

36. Ambiguous Alignment with a Single-Base Indel
Taxon 1 GGTCAG
Taxon 2 GGCCAA
Taxon 3 AGCTAA
Taxon 4 AGCAA
Taxon 5 AGCAA
Taxon 6 AGCAA

37. Ambiguous Alignment with a Single-Base Indel
Taxon 1 GGTCAG GGTCAG
Taxon 2 GGCCAA GGCCAA
Taxon 3 AGCTAA AGCTAA
Taxon 4 AG-CAA AGC-AA
Taxon 5 AG-CAA AGC-AA
Taxon 6 AG-CAA AGC-AA
4 substitutions
1 indels
4 substitutions
1 indels

38. Gap Number and Length
All else being equal, is it better to assume fewer longer gaps, or more shorter gaps?
In other words, what is more likely:
For a new indel to occur?
For an existing indel to lengthen?
There is no general answer!
Alternate alignments are explored algorithmically

39. Alignment Algorithms
Typically built up from pairwise alignments, using assumed gap costs
Problem: most algorithms require an initial tree to define alignment order--bias
Solution: simultaneous tree estimation and alignment optimization
Problems: costly, unjustifiable parameters

40. Clustal Alignment Algorithm
Creates alignment based on penalties for gap opening (number of gaps) and gap extension (gap length)
Multiple alignment built according to guide tree determined by pairwise alignments
Order of adding sequences determined by a guide tree

41. Clustal Alignment Algorithm
Distance matrix calculated from pairwise comparisons
Dendrogram calculated from from distance matrix
Additional sequences are added according to dendrogram, until all sequences are added
Alignment calculated for most similar pair of sequences, based on alignment parameters

42. Tree-Based Alignment
Simultaneous tree and alignment estimation using parsimony
TreeAlign
MALIGN
Implement similar gap opening/extension costs
These applications are very slow!

43. Alignment in the Future?
Incorporate a more sophisticated understanding of molecular evolution in parameterization
For example, what are realistic values of gap costs? Are they universal?
Can phylogeny estimation proceed without optimizing alignments?
Likelihood based methods can sum over all alignments
Will require major contribution of biologists

44. Methods of tree estimation
Character based
Maximum parsimony (MP)
Fewest character changes
Maximum likelihood (ML)
Highest probability of observing data, given a model
Bayesian
Similar to ML, but incorporates prior knowledge
Distance based
Minimum distance
Shortest summed branch lengths

45. Major classes of data Character-based Distance-based Bird A G T
Alligator
Lizard C
Snake
Tortoise
Character-based
Distance-based
Alligator
Lizard
Snake
Tortoise
Bird
0.20
0.60
1.00
0.00
0.80

46. Minimum Distance Alligator Lizard Snake Tortoise Bird 0.20 0.60 1.00
0.00
0.80

47. Maximum Parsimony 3, 5 are slightly more 2: C complicated... 1: A 4: G
Bird A G T
Alligator
Lizard C
Snake
Tortoise
3, 5 are slightly more
complicated...
2: C
1: A
4: G

48. Parsimony Criterion j = character L = tree length
N = number of characters
w = character weight
diff (x1, x2) = number of steps along branch
L = tree length
= topology
k = branch
B = number of branches

49. Parsimonious Character Reconstruction
To evaluate the parsimony of a tree, each character is optimized (then the sum is computed)
Several parsimony algorithms have been developed that optimize character reconstructions
Algorithms differ in assumptions about permissible transformations between character states

50. Likelihood Criterion L = tree likelihood = topology
j = character (site)
l = site likelihood

51. Site Likelihoods
lk, the probability of the nucleotides of each sequence at a given site, is the product of probabilities along the branches of the tree
The probability along a branch is the product of
Probability of a substitution
Branch length
Summed over ancestral states

52. Substitution Model Many models have been proposed
Elements are the rate of substitution of one base for another, per site
Rates are instantaneous (probability of change in a short period of time)
Rates may be allowed to vary among sites

53. Maximum Likelihood 1 2 3 4 5
Bird A G T
Alligator
Lizard C
Snake
Tortoise
Tree is selected that maximizes likelihood of observed sequences, given a model of substitution

54. The Molecular Systematics Revolution
Dramatic increase in the size of data sets
Characters, taxa
Increased confidence in homology assessment?
Computational advances
Technology
Algorithms and software

55. Large Data Sets
Pre-1990: up to ~25 taxa (rarely 100), 1 gene or up to 100 morphological characters
1998: 2538 rbcL sequences of green plants; entire mtDNA sequences (15,000 bp) in animals
2004: ??

56. The Large Data Set Headache
Problem: the large number of sequences, not the size of sequences
Application of phylogenetic optimality criteria requires evaluation of all possible trees
Algorithms guaranteed to find optimal solutions have limited applicability

57. The Problem of Finding Optimal Trees
There are too many trees to evaluate!
The number of possible topologies increases very rapidly with the number of taxa/samples
There are [(2m - 5)!] / [2m-3(m-3)!] unrooted trees , where m = number of taxa

58. Taxa Trees
,135
stars in the universe
atoms in the universe
From Hillis et al. (1996), Applications of Molecular Systematics

59. Heuristic Methods Starting trees followed by rearrangement
Starting trees sample “tree space”
Rearrangements search for local optima
How to get starting trees?
How to rearrange trees?
These methods are prone to getting trapped on local optima

60. Cutting-edge Methods Ratchet Annealing algorithms Genetic algorithms
Temporarily “warps” the character space
Annealing algorithms
Accept suboptimal trees and gradual movement towards optima
Genetic algorithms
Analogous to evolution by natural selection

61. Using Phylogenies
Why are birds so different than their closest relatives?
What genes are involved in the origin of novel traits (like wings)?
Is the rate of molecular evolution in birds especially high?