100
D.E. Soltis and P.S. Soltis
placeholders used for larger, well-supported subclades. Saxifragales, Malvales,
Sapindales, Brassicales (glucosinolate clade), Fabaceae, Cornales, and
Hydrangeaceae are a few of the well-supported clades of flowering plants that have
been examined with more taxa than employed by Soltis et al. (submitted). The
existing topologies for these well-studied and well-supported clades could be
"grafted" onto the three-gene tree, replacing the original placeholders. Donoghue
et al. (1998) employed this basic approach in establishing a tree for Asteridae for
use in analyses of floral evolution. Working from the tree for Asteridae s.l. provided by Olmstead et al. (1993) based on rbeL sequences, subclades better sampled
by other investigators, such as Apiales and Cornales, were grafted onto the "parent" tree, in place of the initial exemplars.
7 Other Approaches: Compartmentalization
An alternative to analysis of an entire large data set involves the partitioning of
taxa to permit analyses of subsets of the data, a general method of analysis referred
to as "compartmentalization" (Mishler 1994). Known monophyletic groups (i.e.,
compartments) are represented by an inferred hypothetical ancestor in more inclusive analyses. The character states of this ancestor are based on those of all the taxa
that compose the monophyletic group, rather than a single exemplar, and the ancestor will likely differ from all real taxa in the group. The position of each monophyletic group (as represented by the hypothetical ancestor) is free to move relative
to other taxa and groups in the analysis. Mishler (1994) recommended a three-step
procedure: (1) a global analysis of all taxa to identify well-supported groups (i.e.,
compartments); (2) local analyses within compartments; and (3) further global analysis, with compartments represented by hypothetical ancestors or with compartment
topologies constrained as found in local analyses. Of course, the most difficult aspect of a compartmentalized analysis such as this is the recognition of groups sufficiently well supported to be considered compartments. To provide reliable estimates of clades in a large data set, the global analysis in step (1) will likely require
extensive computation. A significant benefit of a compartmentalized analysis is
improved homology assessment within compartments, whether the data are molecular or morphological. Consequently, inferred relationships within compartments
are more reliable, with the usc of additional characters, and homoplasy among
compartments is reduced.
A compartmentalized analysis (Mishler et al. 1998) was applied to a data set of
100 18S rDNA sequences for land plants and green-algal outgroups (Soltis et aI., in
press). Unconstrained parsimony analyses of this data set recovered trees that portray clearly spurious relationships, and bootstrap support for most clades was low
(Soltis et aI., in press). Compartmentalization, recognizing eight clades as compartments (algae, hornworts, liverworts, mosses, lycophytes, ferns, conifers, and
angiosperms), improved resolution among these groups, produced trees that are
more consistent with other hypotheses of land-plant relationships (e.g., Kenrick
D.E. Soltis and P.S. Soltis
placeholders used for larger, well-supported subclades. Saxifragales, Malvales,
Sapindales, Brassicales (glucosinolate clade), Fabaceae, Cornales, and
Hydrangeaceae are a few of the well-supported clades of flowering plants that have
been examined with more taxa than employed by Soltis et al. (submitted). The
existing topologies for these well-studied and well-supported clades could be
"grafted" onto the three-gene tree, replacing the original placeholders. Donoghue
et al. (1998) employed this basic approach in establishing a tree for Asteridae for
use in analyses of floral evolution. Working from the tree for Asteridae s.l. provided by Olmstead et al. (1993) based on rbeL sequences, subclades better sampled
by other investigators, such as Apiales and Cornales, were grafted onto the "parent" tree, in place of the initial exemplars.
7 Other Approaches: Compartmentalization
An alternative to analysis of an entire large data set involves the partitioning of
taxa to permit analyses of subsets of the data, a general method of analysis referred
to as "compartmentalization" (Mishler 1994). Known monophyletic groups (i.e.,
compartments) are represented by an inferred hypothetical ancestor in more inclusive analyses. The character states of this ancestor are based on those of all the taxa
that compose the monophyletic group, rather than a single exemplar, and the ancestor will likely differ from all real taxa in the group. The position of each monophyletic group (as represented by the hypothetical ancestor) is free to move relative
to other taxa and groups in the analysis. Mishler (1994) recommended a three-step
procedure: (1) a global analysis of all taxa to identify well-supported groups (i.e.,
compartments); (2) local analyses within compartments; and (3) further global analysis, with compartments represented by hypothetical ancestors or with compartment
topologies constrained as found in local analyses. Of course, the most difficult aspect of a compartmentalized analysis such as this is the recognition of groups sufficiently well supported to be considered compartments. To provide reliable estimates of clades in a large data set, the global analysis in step (1) will likely require
extensive computation. A significant benefit of a compartmentalized analysis is
improved homology assessment within compartments, whether the data are molecular or morphological. Consequently, inferred relationships within compartments
are more reliable, with the usc of additional characters, and homoplasy among
compartments is reduced.
A compartmentalized analysis (Mishler et al. 1998) was applied to a data set of
100 18S rDNA sequences for land plants and green-algal outgroups (Soltis et aI., in
press). Unconstrained parsimony analyses of this data set recovered trees that portray clearly spurious relationships, and bootstrap support for most clades was low
(Soltis et aI., in press). Compartmentalization, recognizing eight clades as compartments (algae, hornworts, liverworts, mosses, lycophytes, ferns, conifers, and
angiosperms), improved resolution among these groups, produced trees that are
more consistent with other hypotheses of land-plant relationships (e.g., Kenrick