13

HOMOLOGY AND DNA

SEQUENCE DATA

WARD WHEELER

Department of Invertebrate Zoology, American Museum of Natural History, New York, NY 10024

INTRODUCTION

The phylogenetic analysis of DNA sequences, like that of all other comparative
data, is based on schemes of putative homology which are then tested via
congruence to determine synapomorphy schemes and cladistic relationships.
Unlike some other data types, however, the matrix of putative homologies or
"characters" is not directly observable. When sequences are unequal in length,
the correspondences among sequence positions are not preestablished and some
sort of procedure is required to determine which positions are "homologous."
This is the traditional province of multiple sequence aligrmaent (= alignment
here). Alignment generates a collection of column vectors through the insertion
of gaps, which form the character set. Whether accomplished manually, or via
some computational algorithm, these characters are then submitted to
phylogenetic analysis in the same manner as other forms of data. This scheme of
correspondences or putative homologies has two salient features. First,
alignment precedes the phylogenetic analysis (i.e., cladogram search) and is
The Character Concept in Evolutionary Biology
Copyright 9 2001 by Academic Press. All right of reproduction in any form reserved. 303
304 WARD WHEELER

never revised in light of systematic hypotheses. Second, alignment-based

homology schemes rest on a notion of base-to-base homology where individual
nucleotide bases transform among five states (A, C, G, T/U, and gap) within a
single character. Two methods have recently been proposed ("Optimization-
Alignment," Wheeler, 1996 and "Fixed-State Optimization," Wheeler, 1999)
which avoid multiple alignment altogether and question these two tenets of
sequence analysis. Although these approaches are parsimony methods, and rely
on testing homology through synapomorphy, they differ in the entities they
propose for testing and this has implications for the interpretation of DNA
sequence homology.
In discussing these concepts, a shorthand will be used. To describe those
correspondences among states frequently referred to as putative homologies, the
lowercase "homology" will be used. To describe those correspondences that
have been tested through congruence on a cladogram (i.e., synapomorphy), the
uppercase "Homology" will be used. The discussion here is mainly concerned
with methods of deriving homology statements, but all of these would then be
tested with other data to determine which homologies are Homologies.

STATIC VERSUS DYNAMIC HOMOLOGY

The standard precursor to the phylogenetic analysis of DNA sequences is

alignment. This procedure takes the unequal length strings of nucleotide bases
and inserts place-holding gaps ("-") to make the corresponding (homologous)
bases line up into intelligible columns. These columns (characters) comprise the
data used to reconstruct cladograms. However this alignment is created, once
phylogenetic analysis has begun, it will not be revised. That is, the homologies
explicitly defined in the alignment will not be reexamined during the cladogram-
search process.
Consider four sequences: I ~ , II GGG, III GAAG, and IV GAA. An
alignment can be generated to be supplied to standard phylogenetic analysis. In
this case, insertion-deletion events are given a cost of two and base substitutions
one. The most parsimonious (minimum cost) cladogram relating these four taxa
would be that which holds I and III to be sister taxa with an overall length of six
(1 indel and 4 base changes--Fig. 1.) Given this alignment, the two other
phylogenetic scenarios are less favored (7 and 8 steps). There is another
alignment, however, which generates the same minimum length for topology (((I
III) II) IV) yet yields the same length (6 steps) for one of the other two possible
topologies (Fig. 2). Using this alignment, two of the topologies are equally
parsimonious.
13. HOMOLOGY AND DNA SEQUENCE DATA 305

Topology

Alignment ((I ID HI) IV) (((I Ill) II) I V ) (((I IV) II) lid
I GGGG 7 6 8
II --GGG
HI GAAG
IV --GAA
Insertion-Deletion events cost 2
Base changes cost I

F I G U R E 1 Possible alignment for four simple sequences and the cladogram cost (length) for the
possible topologies for these taxa.

The point of this example is that the alignment process yields static
homology schemes which is not optimized for any particular topology. Once the
alignment is determined, all testing of the alignment itself stops. Although
homologies are tested on each cladogram, there may be no single homology
matrix which optimizes Homology (yields the most parsimonious result) for each
cladogram. In order to give each topology its shortest length, homologies need
to be generated which are optimal for that particular topology. It is this need
which motivates the method of"Optimization-alignment" (Wheeler, 1996).

Topology
Alignment ((III) HI) IV) (((I m) II) IV) (((I IV) II) III)
I GGGG 7 6 8
II --GGG
III GAAG
IV --GAA

I GGGG 6 6 8
II GGG--
HI GAAG
IV GAA--

Insertion-Deletion events cost 2

Base changes cost 1

FIGURE 2 Comparison of the implications of two different alignments on the cladogram costs for
the sequences in Fig. 1.
306 WARD WHEELER

In the method of optimization-alignment, the diagnosis of each cladogram

attempts to find the lowest cost hypothetical ancestral sequences possible. This
is accomplished by examining all possible homologies between the nucleotide
bases of the two descendent nodes. Dynamic programming is used (in a step akin
to pairwise sequence alignment) to optimize each HTU sequence for the
minimum weighted number of insertion-deletion events and base substitutions
(Fig. 3). At each node down the cladogram, all possible hypothetical ancestral
sequences are implicitly constructed and their costs determined. The minimum
cost sequence is retained and used to optimize the next node down the
cladogram. This algorithm is greedy (only using descendent sequence
information) and can overestimate the real cladogram length (as described by
Wheeler, 1996, 2000). The case of the four example sequences shows the local
optimization of homology (Fig. 4). Since no a priori homology statements are
made, each cladogram can create those homologies most advantageous to its
topology. Hence, the two best cladograms at length 6 arise without fuss.

Hypothetical
GAAG AncestralSequence Cost

GAA(G) 2

4
GAAG "

G--AA ,.._ G ,~A ,JtA - 5

GAAG "

GA--A ~ G A ,~A ~j ~ 3
GAAG r

GAb,-- ~ GAA(G) 2
GAAG "

FIGURE 3 An example of HTU sequence optimization at an internal node via the optimization-
alignment (Wheeler, 1996) procexiurc.
 13. HOMOLOGYAND DNA SEQUENCE DATA 307

Topology ((I n) HI) IV) (((I HI) II)IV) (((I IV) II)HI)

IGGGGI I GGG I IGGGGI [GAAGI IGGGGI I GAA 'l

Diagnosis GGG(G) IGAAGI GRRG I GGG'"I GRR(G) [~GGG I

GILRG I GAA 1 (G)GGG I GAA l GGG

GAACG) G~t~ GRRCG)

Length 2 indels + 2 substitutions = 6 1 indel + 4 substitutions = 6 2 indels + 4 substitutions =

I ( ) denotes insertion--default events * denotes base substitutions r - - i denotes observed sequences" ]

FIGURE 4 Optimization results for the three topologies and sequences of Fig. 1. Note the possible
ambiguities in HTU sequence determination.

Searching for Homology can directly test this dynamic optimization scheme.
Since the method attempts to more efficiently derive homologies, and
Homologies, it can be tested by parsimony. Cladograms should be more
parsimonious and homoplasy less prominent with dynamic homology. Although
this will have to be tested by multiple data sets, the cases presented by Wheeler
and Hayashi (1998) on chelicerates and the examples in Wheeler (2000) and here
show this pattern of more parsimonious solutions for dynamic homology than for
static alignments.

BASE-TO-BASE VERSUS FRAGMENT HOMOLOGY

Both the static and dynamic homologies mentioned earlier rely on a notion
of homology which derives from nucleotide base correspondences. This need
not be the case, however. Homology could be viewed as a phenomenon existing
at the level of the sequences themselves as opposed to base-to-base statements.
This view sees entire strings of DNA nucleotides as characters. Such entities as
the small subunit rDNA locus could be a single character. The locus would then
be homologous among all the taxa, and the actual observed sequences themselves
would constitute the character states. In the context discussed earlier, this would
constitute a "static" homology scheme since the character vectors would be
preordained. The complexity of the sequences would allow for homology
statements more like those of other forms of character analysis, where position
and complexity aid in character delimitation.
308 WARD WHEELER

Seq. 1 Seq. 2 Seq. 3

Seq. 1 0 C21 C31

Seq. 2 Cl2 0 C32 Cij = Cji

Seq. 3 El3 C23 0

FIGURE 5 Matrixof minimum transformation cost between sequence pairs.

When employing blocks of contiguous sequence as characters, with

observed sequences as states, dynamic programming methods must be used to
optimize cladograms and determine their length. The procedure is identical to
the optimization of Sankoff-style characters ("step-matrix" characters), just
modified for large numbers of states (nstates <- ntaxa). This approach relies on
the postulate that only observed sequences may be optimized to hypothetical
ancestors. This restricts the possible world of reconstructed sequences, but also
requires that these sequences exist. Each sequence becomes a state in an
extremely complex character. The first step in the optimization procedure is the
determination of the transformation cost matrix among all the states. This is
defined as the minimum transformation cost (including all forms of base
substitutions and insertion-deletion costs) between each pair of states (Fig. 5).
Once these transformation costs are known, standard dynamic programming
implicitly examines the assignment of each of these states to each internal node
and determines the optimal set of states and cladogram length (Fig. 6).

I I

I1,1, ~0 . . .

Length = Minimum
FIGURE 6 An example of down-pass cladogram optimization via the fixed-state approach.
13. HOMOLOGY AND DNA SEQUENCE DATA 309

Given that the cladogram length is based on nucleotide sequence, it might

seem strange to say that the bases themselves are not homologous. This effect is
derived from the pairwise nature of the character transformation matrix.
Consider three sequences I AAATTT, II TTT, and III AAA. When
transformation costs are determined, the first "T" in sequence II (position 1)
corresponds with the first "T" of sequence I (position 4). This same "T" in
sequence I also corresponds with the first "A" of sequence III (position 1). If our
logic were transitive, this would imply that position 4 of sequence I would
correspond to position 1 of sequence III. It does not. Position 1 of sequence III
("A") corresponds to position 4 of sequence I (also "A"). No circle of
correspondence can be drawn among these nucleotides describing state
transformations. They are not homologous (Fig. 7).

I AAATTT L AAATI'T
'I I TTT
II TTT IL AAA
'I II TTT
III AAA III/ AAATTT
'I AAA

FIGURE 7 Schemeof base correspondencesimpliedby the fixed-state approach.

310 WARD WHEELER

Two other aspects of this approach affect ideas of homology. One of the
salient features of base-to-base methods, whether built upon static or dynamic
homologies, is difficulty in tracing complex homologies through the cladogram
(or alignment)---in other words, messy data. When there is extensive sequence
length variation coupled with base changes, tremendous uncertainty in homology
can occur in both multiple-aligmnent and optimization alignment. The
requirement that such variation be accommodated over the entire cladogram can
make local tmcertainties propagate throughout the analysis. Since the sequence
level homology approach transforms the complex states with their variations in
length and nucleotide base composition into simple numbered states with
pairwise costs, this problem does not occur. Such seemingly confusing variation
patterns will certainly lead to longer cladograms, but the homologies (at the
fragment level) will remain clear.
A second feature of fragment level homology is the requirement that the
character homologies be defined a priori. Whether entire loci, structurally or
functionally defined regions are employed as homologies, they are determined by
the investigator. This is akin to the delimitation of variation in complex
morphological features. Are complex structures such as complete development
in the endopterygote insects single or multiple characters? As with all such
seemingly arbitrary decisions, what matters most is the effect of changing these
character delimitations on phylogenetic results.
The notion of synapomorphy as a shared derived feature might also seem to
be altered by the homology concept implicit in sequence fragment comparisons.
Since each taxon may well express a unique character state, it might appear that
synapomorphy (as a shared state) would be impossible. This criticism would
only apply if the characters were completely unordered. State transformation
costs are not equal among states, hence are more akin to synapomorphy in the
context of ordered characters. Two taxa might present states 1 and 2 of an
ordered series 0 --~ 1 ---, 2. These taxa are united by the transformation implied
by the ordering with 1 and 2 sharing special derived similarity not found in 0
(Platnick, 1979). The concept of synapomorphy (or Homology) is unaffected by
the fixed-state approach.

COMPARISONS

For these distinctions (static versus dynamic; base-to-base versus fragment)

to be anything more than nomenclature, some means of comparing these methods
and judging superiority must be offered. Congruence could be that measure.
When analyzing single data sets, via whatever method, the best solution is that
which minimized discord among data (i.e., characters). This may be measured by
simplicity (parsimony) or with respect to complex statistical models (likelihood).
The three methods of viewing homology here define characters in somewhat
different ways, hence simple counting of change for single data sets (i.e.,
13. HOMOLOGY AND DNA SEQUENCE DATA 311

cladogram length) cannot be used. The things that are counted are just not the
same. This notion of character congruence, however, can be extended to the
broader concept of congruence among data sets. Character congruence has been
used to discriminate among analysis parameters (Wheeler, 1995; Whiting et al.,
1997; Wheeler and Hayashi, 1998) and could reasonably be used to compare the
behavior of methods (although numerous other means could also be employed).
Two types of congruence measures can be used: character based and
topological. The relative merits and demerits of these approaches have been
explored in the literature (Mickevich and Fan'is, 1981; Wheeler, 1995) and
character congruence will be used here due to its link with parsimony and
combined data analysis. Phylogenetic methods are judged to be superior if they
accommodate variation in multiple data sets efficiently as measured by the
Mickevich-Farris incongruence length metric (Mickevich and Farris, 1981).

EXAMPLE--ARTHROPODS

In order to compare these three homology-determination methods, the

arthropod data of Wheeler et aL (1993) are used. These data consist of 100
morphological characters, -~650 18S rDNA nucleotides, and 228 Ubiquitin
nucleotides. To these data-~350 28S rDNA nucleotides were added. The 18S
and Ubiquitin data were determined for 25 extant taxa and the morphological
data scored for these taxa and "Trilobita," an extinct clade. The 28S rDNA data
were determined for 15 of the extant taxa (Table I).

TABLE I Taxon List

Mollusca
Cephalopoda Loligo pealei
Polyplacophora Lepidochiton cavernae
Annelida
Polycheata Glycera sp.
Oligocheata Lumbricus terrestris
Hirudinea Haemopis marmorata
Onychophora
Peripatoidae Peripams trinitatis
Peripatopsidae Pe ripatoides novozealandia
Trilobita groundplan of Ramsk61d and Edgecombe, 1991.
(morphological analysis only)
Chelicerata
Pycnogonida Anoplodactylus portus
Xiphosura Limulus polyphemus
Seorpiones Centruroides hentzii
Uropygi Mastogoprocms giganteus
Araneae Nephila clavipes
Araneae Peucetia viridans
Crustacea
312 WARD WHEELER

Cirrepedia Balanus sp.

Malacostraca CalBnectes sp.
Myriapoda
Chilopoda Scutigera coleoptrata
Diplopoda Spirobolus sp.
Hexapoda
Zygentoma Thermobius sp.
Ephemerida Heptagenia sp.
Odonata Libellula pulchella
Odonata Dorocordulia lepida
Dictyoptera Mantis reli giosa
Auchenorrhyncha Tib ice n sp.
Lepidoptera Papi li o sp.
Diptera Drosophila me lanogas te r

Three analyses were performed. In each case, the insertion-deletion cost was
set at two and all base substitutions set at one. When morphological characters
were used, character transformations were set at two. In the first analysis, the
data were aligned (via MALIGN; Wheeler and Gladstein, 1994) and
phylogenetic analysis was performed using PHAST (Goloboff, 1996). The
second analysis employed optimization-alignment as implemented in POY
(Gladstein and Wheeler, 1996). The third used the fixed-state optimization
technique also as implemented in POY. Gaps/indels were included and given the
same weight (2) in all length calculations. All searches employed TBR branch
swapping and 10 random addition sequences. The results of the individual data
partitions, combined results, and congruence calculations are stanmarized in
Table II and Figs. 8-10.

TABLE II Comparison of Methodologies

Data Alignment Method Fixed-state

Optimization-alignment

18S rDNA 503 501 584

Ubiquitin 3871 392 484
28S rDNA 919 848 943
Morphology 2522 252 252
Combined 2123 2007 2271
Incongruence3 0.0292 0.00698 0.00352

This length of 387 steps is shorter than that of the optimization-alignmentpurely due to the treatment
of ambiguities. When all ambiguities are treated as missing data, both alignment (MALIGN-PHAST)
and optimization-alignment (POY) yield the same length of 387 steps.
2This length is 2 times the length of 126 steps.
3Calculated as (Combined - 18S rDNA - Ubiquitm - 28S rDNA - Morphology)/Combined.
13. HOMOLOGYAND DNA SEQUENCE DATA 313

FLepidochiton
~--Loligo
I F Glycera
I FL-H er.opi
II i--Lumbdcus
tJ ,---[- Peripatoides
II L-Peripatus
II _.I -mril~
H I F -'An~176
I H Fumulus
[ L.~ r Centruroides
Y ~ - - Mastogoproctus
Peucetia
t _ Nephila
Callinectes
Balanus
Scutigera
Spirobolus
Thermobius
Heptagenia
Dorocordulia
Libellula
Mantis
Tibicen
Papilio
Drosophila

FIGURE 8 Morphologicallybasedcladogramof arthropodrelationships(Wheeleret al., 1993).

The dynamic homology approach of optimization-alignment resulted in

more parsimonious cladograms in all the cases where sequences were unequal in
length. This is due, no doubt, to the simultaneous optimization of synapomorphy
and homology uniquely for each topology. The cladograms derived from the
fixed-state approach were the longest. The restriction on the possible range of
internal node (HTU) sequences is responsible for this. Since internal node
sequences are chosen from the range of observed terminal sequences, longer
cladograms frequently arise (Wheeler, 1999). Overall character incongruence
was lowest (0.00352 vs. 0.00698 and 0.0292) for the fixed-state analysis.
314 WARD WHEELER

Lepidochiton A Lepidochiton B
Glycera -- Lepidochiton C
Glycera -.: Loligo
Loligo Loligo -- Peripatoides
Haernopis Haemopis
Lumbricus ~- Peripatus
Lumbricus -- Balanus
I ~ Peripatoides r-I-- Peripatoides
I --" Callinectes
I I =- Peripatus I I " - Peripatus -- Scutigera
L.~ ~ Callinectes L~ ,.-r- Callinectes -- Spirobolus
I I L._ Balanus I I - Balanus -- Glycera
L~ !'- Anoplodactylus i ~ F Anoplodactylus Anoplodactylus
! I r - Limulus I I f-~mulus Limulus
t ~ ~l._Mastogoproctus l_q ~ Mastogoproctus Centruroides
I I I.a- Centruroides I I E! m Centruroides Mastogoproctus
L.J L F - Peucetia I I LF Peucetia Peucetia
I L_. Nephila t_~ t_. Nephila Nephila
I ...J-" Scutigera ,--l'- Scutigera
I __~ Haamopis
U L_ Spirobolus II -Spirobolus Lumbricus
] I"- Thermobius II F "ThermObius
t.~ f _ Heptagenia -~9 Ir--Heptagenia ~ L ~Tibicen
Papilio
U F Mantis L~ ~--Mantis Drosophila
I r " E l - Dorocordulia I ~ Tibicen Mantis
L.~ L_ Libellula Papilio Dorocordulia
I I'- Tibicen I L_ Drosophila Libellula
Papilio L . F Dorocordulia Thermobius
Drosophila L - - Libellula Heptagenia

~ S L ~tpAnOopdlauycslt D
-- Anoplodactylus F
Mastogoproctus Scutigera
Papilio Scutigera
~ ~ S Tibicen
CPBaAnOpoldacytuls E
Balanus ,.4-- Nephila
Libellula ..j t_ Peucetia
Centruroides Papilio
Limulus L r - Spirobolus
Mantis L.. Balanus
Nephila Callinectes -- Mastogoproctus
Peucetia Drosophila " Centruroides
utigera stogoproctus t _ Limulus
irobolus entruroides .._[-- IJbellula
Callinectes Limulus
ellula I L_FPapitio
irobolus .~ "--Drosophila
Mantis /anus I I-" Tibicen
L_I - Drosophila Nephila Callinectes
r Tibicen Peucetia L _ Mantis

-- Lepidochiton G I-- Lepidochiton H -- Lepidochiton

-- Loligo ~.- Loligo -- Glycera
-- Haemopis J i-- Haemopis -- Lumbricus
--Centruroides I I I-'Peripatus -- Drosophila
-- Drosophila L~ I - E l - " Peripatoides -- Haemopis
-- Limulus I I ~ Mastogoproctus -- Peripatoides
-- Callinectes I I "--- Peucetia -- Peripatus
-- Libellula I I [ " Papilio -- Loligo
-- Balanus L~ ,~ _]-.. Thermobius -- Mastogoproctus
--Oorocordulia I I I I-L I-Scutigera -- Peucetia
--Anoplodactylus II ILl '--Spirobolus -- Nephila
-- Peripatoides ] I ~ Mantis -- Tibicen
-- Mastogoproctus U L r - Heptagenia -- Mantis
-- Peucetia 9 ~ Nephila -- Papilio
-- Peripatus I ~ Tibicen -- Anoplodactylus
--r-Glycera i r - An~ -- Limulus
~-Lu~,cus II F "Dr~176 -- Centruroides

~ - Papilio
[ " Thermobius
I--I-" Scutigera
"7 L'-Spir~176
~ r l _ J - - Centruroides
I I -l_r- Glycera
L- i -- Lurnbricus
I !"-Limulus
--

--
Callinectes
Balanus
-- Dorocordulia
-- Libellula
l _ r -Mantis i ~ ,-.!'- Balanus _~_ Scutigera
Heptagenia L~ L_ Dorocordulia Spirobolus
--L.J-- Nephila L J - Callinectes . _ ~ Thermobius
L _ Tibicen L - - Libellula Heptagenia

FIGURE 9 Cladograms of individual data partitions when subjected to different analytical

techniques. A. I SS rDNA and multiple sequence alignment. B. 18S rDNA and optimization-
alignment. C. 18S rDNA and fixed-state optimization. D. 28S rDNA and multiple sequence
alignment. E. 28S rDNA and optimization-alignment. F. 28S rDNA and fixed-state optimization. G.
Ubiquitint and multiple sequence alignment. H. Ubiquitin and optimization-alignment. I. Ubiquitin
and fixed-state optimization.
13. HOMOLOGYAND DNA SEQUENCE DATA 315

. ~ ~ ~'~ ~.~

~~.~ ~~~~~.w
"~ ~ ~ ~ ~ ~""[ ~
~.~-~.= _~-~., ~ . ~ I I I I
~ ~ ~. U ! I
I
"-~ 1 ]1 -
,--9 ,,~1 ,. ,
..,I
9 ,...I 1

t~

.,

~L .... I
9,.,I . . J . I

~ ~a.~ -~.o~ I I
~~~-=~_~- ~~e LI-~ ~ iJ

.~ L r J L J I....
~.~L I .. I'
I I ,~ I

FIGURE 10 Cladograms of combined data (18S rDNA, Ubiquitin, 28S rDNA, morphology) for
arthropod taxa when subjected to different analytical techniques. A. Multiple sequence aligmnent. B.
Optimization-alignment. C. Fixexi-statr optimization.
316 WARD WHEELER

DISCUSSION

Clearly, the way we view sequence homology has tremendous implications

for the elucidation of phylogenetic pattern. The three modes discussed here
(static, dynamic, and fragment level) imply different patterns of relationship in
the small test case used here. Furthermore, since the reconstructions of
hypothetical ancestral sequences vary with the method, the types of evolutionary
events reconstructed on these patterns differ as well.
An additional feature of the base-to-base methods, the nonindependence of
the nucleotide characters, remains largely unexplored. In both alignment and
optimization-alignment, the homology scheme for each nucleotide is determined
in concert with all the other bases that surround it. The relative position and
number of mdels and nucleotide substitutions in adjacent sequence positions
ftmdamentally affect positional homology. Clearly, such character statements are
not independent. However, when cladograms are constructed, the cost of
changes and indels are summed linearly over the data--an assumption of rigid
character independence. Since the individual bases play no role in homology
with the fixed-state approach, this character dependence problem vanishes. The
indels and base substitutions only determine the cost of transformation between
states, there is no requirement that these changes be independent. This
inconsistency of base-to-base homology is avoided.
With character incongruence levels at half or lower levels than the other
techniques and character nonindependence removed, the fixed-state approach to
sequence homology is clearly worth consideration.

ACKNOWLEDGMENTS

I would like to acknowledge the contributions of Daniel Janies, Gonzalo Giribet, Norman
Platnick, Lorenzo Prendini, Randall Schuh, Susanne Schulmeister, and Mark Williams to this work
through discussion and abuse. I would also like to thank Portia Rollins for expert art work.

LITERATURE CITED

Gladstein, D. S., and Wheeler W. C. (1997). "POY: The Optimization of Alignment Characters."
Program and Documentation. New York, NY. Available at "ftp.amnh.org"/pub/molecular.
Goloboff, P. (1996). PHAST. Program and Documentation. Version 1.5.
Miekevich, M. F., and Farris, S. J. (1981). The implications of congruence in Menidia. Syst. Zool.
30:351-370.
Platnick, N. I. (1979). Philosophy and the transformation ofcladistics. Syst. ZooL 28:537-546.
Ramsk61d, L., and Edgecombe, G. D. (1991). Trilobite monophyly revisited. Hist. Biol. 4:267-283.
Wheeler, W. C. (2000). Heuristic reconstruction of hypothetical-ancestral DNA sequeces: sequence
alignment versus direct optimization. In "Homology and Systematics: Coding Characters for
Pylogenetic Analysis" (R. W. Scotland, ed.), pp. 106-113. Taylor and Francis, London.
13. HOMOLOGY AND DNA SEQUENCE DATA 317

Wheeler, W. C. (1999). Fixed character states and the optimization of molecular sequence data.
Cladistics 15:379-385.
Wheeler, W. C. (1996) Optimization alignment: the end of multiple sequence alignment in
phylogentics? Cladistics 12:1-10.
Wheeler, W. C. (1995). Sequence alignment, parameter sensitivity, and the phylogentic analysis of
molecular data. Syst. Biol. 44:321-332.
Wheeler, W. C., and Gladstem, D. S. (1994). MALIGN: A multiple sequence alignment program. J.
Hered. 85:417.
Wheeler, W. C., and Gladstein, D. M. (1992-1996). Malign: A Multiple Sequence Alignment
Program. Program and Documentation. New York, NY. available ftp,amnh.org
/pub/molecular/malign
Wheeler, W. C., and Hayashi, C. Y. (1998). The phylogeny of the chelicerate orders. Cladistics
24:173-192.
Wheeler, W. C., Cartwright, P., and Hayashi, C. (1993). Arthropod phylogenetics: a total evidence
approach. Cladistics 9:1-39.
Whiting, M. F., Carpenter, J. C., Wheeler, Q. D., and Wheeler, W. C. (1997). The Strepsiptera
problem: phylogeny of the holometabolous insect orders inferred from 18S and 28S ribosomal
DNA sequences and morphology. Syst. Biol. 46:1-68.