Phylogenetic Systematic analysis:
    an example using molecular data

  Why use molecules?

      Molecules provide an independent estimate of phylogeny
         Avodis a circular argument:
            Morphology is used to create a classification,
       then the classification is interpreted to explain evolution
       Ex.: Chinese water deer (Hydropotes) are the only antlerless deer
            => placed in a separate subfamily
                    & assumed to be ancestral type
                 But, molecular analysis shows antlers were lost secondarily

      Molecules provide large numbers of characters for analysis
       Homo has ca. 200 bones and 3,000,000,000 nucleotide pairs
            Typical morphological study involves <100 characters
            Typical molecular study involves >1,000

      Patterns of molecular evolution are understood
        Transitions (Ts) are more frequent than Transversions (Tv)
            [recall Bio2250: Ts = CT or AG, Tv everything else
         'silent' >> 'replacement' substitutions
         3rd position >> 2nd & 1st substitutions (usually)

      Relative importance of characters is easier to judge
         Is the # of toes more important than # of teeth?
         Are scales versus feathers more important than # of temporal openings?
       But: Any one nucleotide position is more or less like any other
 

1. Defining the systematic problem: "Is the Giant Panda a bear or a raccoon?"

   Evolutionary relationships of the Giant Panda (Ailuropoda)
      Ailuropoda has been considered to be either a bear (Ursidae) or a raccoon (Procyonidae)
      General morphology suggests ursid ancestors:
         Details of skull, diet, biogeography seem to suggest procyonid ancestors
       Ex.: alar canal is present in Ursidae (including Ailuropoda),
                                       absent in Procyonidae, except lesser panda (Ailurus)

2. Collecting the data:

   Measure homologous characters in a set of taxa:
      with DNA, each nucleotide position is a separate character

   mitochondrial DNA (mtDNA) is used in many systematic studies
      "Small circular molecule ...", 16Kbp, maternally-inherited (cytoplasmic)
      13 protein loci (fast), 2 rDNA & 22 tDNA genes (slow), control region (very fast)
      'Universal primers' permit PCR & DNA sequencing from many taxa
      cytochrome b gene is widely used:
            Large data base for comparison
            1140 bp in most vertebrates; we examine 401 bp in lab

3. Analyzing the data:

   Phenetic (how similar are taxa?)
      versus cladistic (how closely related are taxa?) criteria

   These criteria agree, iff rates of evolution are constant
      If evolutionary rates differ, closely related organisms may appear different
       Ex.: Crocodiles are more closely related to birds, but more similar to lizards
                 Crocodiles resemble lizards more than birds
                 because birds rapidly evolved specializations for flight

   A. Phenetic analysis

      Simplest measure is % sequence similarity (S)
                                          p-distance = (1 - S) x 100

      Patterns of similarity can be inferred from cluster analysis
         Most widely used is UPGMA [Unweighted Pair Group Method with Averaging],
            a Sequential Agglomerative Hierarchical Nesting (SAHN) algorithm
            [algorithm = a set of instructions for doing a repetitive task]
          In (n) x (n) matrix, join the most similar pair
             re-calculate (n-1) x (n-1) matrix, re-join,
                and so on, until last pair is joined
      Results are show as a phenogram:
         a diagram of phenetic relationships
       UPGMA method assumes that rates of evolution are equal
            so branch tips "come out even" (contemporaneous)

      Some alternatives:
       Neighbour-Joining (NJ) analysis does not assume rate equality
                 large evolutionary rate differences lead to incorrect trees
             NJ allows branch lengths proportional to change: tips come out uneven
               [algorithm joins nodes, rather than tips]
            This method is more realistic, computationally harder
                    [see www.megasoftware.net for free software]

       Differential weighting of nucleotide substitutions
            accord greater 'significance' to 'important' changes
         Ex.: Kimura 2-parameter distance (K2P) model treats Ts & Tv separately
                transition bias  = [Ts] / [Tv]
                        There are twice as many kinds of transversions as transitions:
                            expected K = 0.5
                But: Tv are rare for close comparisons,
                                       more common for distant relationships
                K is variable according to the evolutionary problem under consideration:
                     K > 6 for close comparisons

B. Cladistic Analysis

     Principles of homology & analogy can be applied to nucleotide changes
         We rely only on shared derived (synapomorphic) nucleotide sites,
         & avoid shared ancestral (symplesiomorphic) nucleotide sites,
                       and changes unique to single taxa (autapomorphies),
                       and convergent nucleotides between unrelated taxa.

     Choice of preferred hypothesis is made on the Principle of Parsimony
          In general: parsimony means that the simpler hypothesis is to be preferred
                          complex hypotheses are less probable
       Evolutionary parsimony:
                a hypothesis that requires fewer character changes is preferred
            Ex.: to explain the origin of a complex structure
                            it is more parsimonious to hypothesize that it has evolved only once
                In molecular systematics, these changes are nucleotide substitutions [DNA mutations]

      The "Four-Taxon Problem" and the "Three-Taxon Statement":
         Among four taxa A, B, C, & D, there are three hypotheses of relationship:
            either A is most closely related to B, or to C, or to D
      We want to be able to evaluate hypotheses of the form:
       "X and Y are more closely related to each other than either is to Z"
            The alternative hypotheses can be shown as networks with branches and an internode

If (for example), A is most closely related to B
       A & B will share characters inherited from their common ancestor
       These changes will occur on internode between the pairs

Seven classes of nucleotide sites can be identified
    (for details, see Notes on Parsimony Analysis)
    [Note: "Type" terminology is unique to this course]


 Types 1 - 4 are uninformative:
           They give no information about relationships, because
   all hypotheses require the same number of changes,
                   so none is more parsimonious than the others.

   Type 1 is invariant: no changes are required.

   Type 2 indicates only that one taxon is unique wrt the others:
                all hypotheses require a single nucleotide change.

   Type 3 indicates that all taxa are distinct & unique:
                all hypotheses require three nucleotide changes.

   Type 4 indicates that two taxa are similar,
                but not whether this similarity is ancestral or derived:
                    shared a could be either
                    hypothesis requires two changes
                Alternative hypotheses also require two nucleotide changes.
                     [a '+' indicates a change along a particular network branch]

Types 5, 6 & 7 are informative:
          They give information about relationships,
                because one hypothesis requires fewer changes than the others
                  & is therefore more parsimonious than the others

   Type 5 indicates that A & B are most closely related:
      The first hypothesis can explain the distribution of nucleotides with a single change,
         the latter two require two changes each. [See also lab #5]
     The first hypothesis is a more parsimonious explanation of the data than the others.

By the same logic:
   Type 6 indicates that A & C are most closely related.
   Type 7 indicates that A & D are most closely related.
    [Homework: for the three networks above,
        sketch the changes required by sites of types 6 & 7]

A cladistic analysis counts the number of
  informative characters favouring each hypothesis
      The hypothesis with the "highest score" requires the fewest changes
         and is therefore the 'most parsimonious' explanation.
         This is also called the 'minimum length' solution.

    Cladistic analyses may also be weighted:
            Ex.: Count Tv:Ts as 3:1 Tv are 3x as meaningful
              or, count Tv only (Transversion parsimony) for "deep" analyses
              or, count 1st & 2nd position substitutions >> 3rd

C. Placing the root & Inferring the direction of evolutionary change

   Suppose the first hypothesis (A & B are most closely related) is most parsimonious
      Ex.: In Lab #5, we found that the majority of sites were of type #5. We said:
       "Ailuropoda & Ursus are more closely related to each other
            than either is to Procyon (or Martes)."
               The hypothesis can be drawn as an unrooted network
       But: this evidence can also be used to argue
         "Procyon & Martes are more closely related to each other
                than either is to Ursus (or Ailuropoda)."
        To resolve this, we need to know where their common ancestor fits in.

 There are four branches and one internode in this network
  An evolutionary tree is a network with a root:
      The root indicates the relationship with the common ancestor
          A 'root' can be placed on any of the branches or the internode.
          So, there are five possible rooted trees for this unrooted network.
              All are equally parsimonious:
              not all place A & B as each other's closest relatives.
              Some of these make shared charactes symplesiomorphic

   There are several ways to placement the root

      (1) Outgroup rooting:
         Include a taxon that is known to be less closely related
            to any of the ingroup taxa than they are to each other.
            Such a taxon is called an outgroup or sister taxon.
       Ex.: Lynx (Feloidea) is an outgroup to the Canoidea
              (Note that this tree is equivalent to the NJ phenogram)

      (2) Midpoint rooting:
         Place the root halfway between the two most different taxa.
            This assumes that molecular evolution is clock-like.
            (Here, this places the root on the internode).

      (3) Character Polarity:
         If the character state of the ancestor is known (or can be inferred).
         Root the tree accordingly

         Use of polarity is usually not possible with molecular data
            Any nucleotide can mutate to any other, in either direction
                any a c g t  looks exactly like any other a c g t
                [Some models allow for differential probabilities of mutation]
            Homologous nucleotide in ancestor has most likely mutated

         Use of polarity with morphological data is standard
              Ex.: In an analysis of the evolution of the number of heart chambers in
                codfish (2), lizard (3), crocodile (4), & bird (4)
                we know that the evolutionary order is 3 4
                     (this is called a transformation series)
                The root will be placed on the codfish branch,
                        because we know the codfish most resembles the ancestor.
                Crocs & Birds have a recent common ancestor with a four-chambered heart.

 D. Statistical tests determine confidence in branching order
          Bootstrap Analysis: a re-sampling technique
                statistical tests usually involve obtaining replicates / repeating experiment
                this is inconvenient with DNA data: $$$
            Suppose existing data set (401bp) is a random sample of parametric data set (complete genome)
                  re-sample existing n sites 1000 times, repeat phylogenetic analysis:
                        how often do same clades / clusters appear?
                    "50% bootstrap support" indicates particular group
                              occurs more frequently than all others combined
                      95% criterion is desirable, not often obtained with small data sets


What does this analysis explain about the biology & evolution of Pandas?

   1. Ailuropoda and Ursus are each others' closest relatives:
           The Giant Panda is a highly derived  bear, not a raccoon.
       Ailuropoda should be classified in Ursidae.

     2. Similarities of Ailuropoda and Ailurus are convergent (analogous):
            these represent parallel feeding specializations.
         Ex.: "Hypertrophied masticatory apparatus" permits feeding on bamboo:
            (expanded zygomatic arch, high mandibular ramus, and molariform teeth)
         Jaw articulation above toothrow gives mechanical advantage:
            (similar modifications occur in Hyaena for crushing bones).

     3. Some similarities between Ailuropoda and other ursids are ancestral homologies:
         Bears (including pandas) have short gestation and tiny neonates.
            In most bears, gestation & birth occur during winter hibernation:
                Hypothesis: early parturition (birth) gives access to milk, when no other food is available
       But: Pandas do not hibernate, young are carried during foraging:
           Why have altricial (underdeveloped) young when food is readily available?

"Small young could be explained if the suite of physiological and behavioural adaptations associated with the production of small neonates were established before splitting of the panda and ursid lines." (Ramsay & Dunbrack, 1987)

            That is, tiny neonates are a conserved ancestral condition rather than a a contemporary adaptive response.

     4. Panda evolution may have proceeded quickly:
       Ailuropoda & Ursus are about as similar genetically as dog & fox.
         Development & growth of cranial versus axial skeleton in pandas resembles Hyaenas and boxer dogs:
            heavy crania,  less-developed post-cranial (axial) skeletons.
            Selection may operate on similar, hypothetical 'growth fields'

"The basic adaptive transition from Ursus to Ailuropoda required the changing of very few genetic messages [during an] origin by way of a very small population occupying a local bamboo forest." (Stanley 1979)

    Pandas may be a textbook case of quantum speciation:
          the origin of a new adaptive type in one or a few speciation events.


Text material © 2010 by Steven M. Carr