© 2007 Society of Systematic Biologists
Fossils Impact as Hard as Living Taxa in Parsimony Analyses of Morphology
Edited by Jack Sullivan
1 Department of Biology and Biochemistry, The University of Bath The Avenue, Claverton Down, Bath, BA2 7AY, UK E-mail: M.A.Wills{at}bath.ac.uk
2 Department of Zoology, The Natural History Museum Cromwell Road, South Kensington, London, SW7 5BD, UK
 |
Abstract |
|---|
 Top Abstract Material and MethodsResults and DiscussionConclusionsReferences |
|---|
Systematists disagree whether data from fossils should be included in parsimony analyses. In a handful of well-documented cases, the addition of fossil data radically overturns a hypothesis of relationships based on extant taxa alone. Fossils can break up long branches and preserve character combinations closer in time to deep splitting events. However, fossils usually require more interpretation than extant taxa, introducing greater potential for spurious codings. Moreover, because fossils often have more "missing" codings, they are frequently accused of increasing numbers of MPTs, frustrating resolution and reducing support. Despite the controversy, remarkably little is known about the effects of fossils more generally. Here we provide the first systematic study, investigating empirically the behavior of fossil and extant taxa in 45 published morphological data sets. First-order jackknifing is used to determine the effects that each terminal has on inferred relationships, on the number of MPTs, and on CI' and RI as measures of homoplasy. Bootstrap leaf stabilities provide a proxy for the contribution of individual taxa to the branch support in the rest of the tree. There is no significant difference in the impact of fossil versus extant taxa on relationships, numbers of MPTs, and CI' or RI. However, adding individual fossil taxa is more likely to reduce the total branch support of the tree than adding extant taxa. This must be weighed against the superior taxon sampling afforded by including judiciously coded fossils, providing data from otherwise unsampled regions of the tree. We therefore recommend that investigators should include fossils, in the absence of compelling and case specific reasons for their exclusion.
Keywords: Fossils; leaf stability; relationships; resolution; MPTs
Received January 23, 2007; Revised March 19, 2007; Accepted June 4, 2007
The vast majority, perhaps 99%, of all the species that have ever existed are extinct (Nee and May, 1997), and the living biota is a relatively well-studied but diminutive slice of this total diversity. The morphological gaps between extant species and between higher groups can at once be convenient, when our concern is utilitarian classification, and frustrating, when our attention turns to phylogeny (Wiens, 2006). In general, the greater the morphological distance between an organism and other closely related living taxa, the more problems attend its phylogenetic placement (e.g., sea spiders [Dunlop and Arango, 2005] and turtles [Iwabe et al., 2005; Harris et al., 2007]). Because fossils can sample lineages close in time to relatively deep splitting events, they may bridge problematic morphological gaps and display combinations of character states that are not found in the extant biota (Gauthier et al., 1988; Donoghue et al., 1989; Santini and Tyler, 2004; Stockley et al., 2005; Wilson, 1992), thereby subdividing long branches (Anderson, 2001; Poe, 2003; Wiens, 2005) and improving phylogenetic accuracy. They also offer a snapshot of morphologies that are temporally closer to the splitting of lineages (Novacek, 1992), may actually introduce less homoplasy than their extant counterparts (Wagner, 1999), and can alter radically inferences about character evolution and trends (Finarelli and Flynn, 2006). Moreover, because our knowledge of the extant biota is better than our knowledge of the biota at any other point in history, new fossil finds are more likely to sample a hitherto unknown major branch of the tree of life than discoveries of additional living taxa. Very few living species first described in recent decades were thought to belong to hitherto unknown major groups (possibly equivalent to orders, classes or phyla) (e.g., Kristensen, 2002), whereas this phenomenon has been much more common in paleontology (Briggs and Fortey, 2005).
Some systematists appear ambivalent or even hostile towards the inclusion and importance of fossils in phylogenetic analyses (Patterson, 1981; Ax, 1987). Whereas living taxa can often be dissected and studied exhaustively (at least in principle), fossils are usually fragmentary and subject to variable preservation, often leading to substantial nonrandomly distributed "missing" or "uncertain" data (Wiens, 1998). This can result in an increase in the number of optimal trees (Gauthier et al., 1988) and a concomitant decrease in the resolution of some consensus trees that can obscure which relationships are unambiguously supported by the parsimonious interpretation of the data from extant taxa alone (Wilkinson, 1995; Anderson, 2001). A related issue is that the potentially greater relative instability of incompletely known fossil taxa can obscure the levels of support for inferred relationships estimated using, for example, bootstrapping and decay analysis (Wilkinson and Benton, 1996; Horovitz 1999; Wilkinson, 2003). Nonetheless, recent simulation work demonstrates that analyses including taxa with abundant missing data can sometimes have better phylogenetic resolution than analyses excluding them (although including additional, more complete taxa can yield better resolution still) (Wiens, 2005, 2006). Critical, in this regard, is the nature and amount of data that are present, rather than the proportion that is missing. Another objection is that fossils often require greater and more uncertain interpretation. Fossils from the deepest radiations are among the most problematic, but frustratingly, often cited as the most pivotal (Conway Morris, 1993; Briggs and Fortey, 2005). Many espouse their contribution as vital given the limited grip of sequence data on the oldest and most rapid radiations (Smith and Littlewood, 1994; Wills and Fortey, 2000; Rokas et al., 2003; Smith and Turner, 2005). Although rare genomic changes are sufficiently conservative to preserve phylogenetic signals over hundreds of millions of years, they are typically too infrequent to capture numerous lineage splits in a short time. Only fossils can preserve past character states unchanged, irrespective of their rates of evolution. Others regard fossil data (and morphological data more generally [Scotland et al., 2003]) as being of lower "quality" (Nixon, 1996; Grande and Bemis, 1999; O'Leary, 1999), offering a weak phylogenetic signal at the cost of too much interpretative noise (Patterson, 1981; Ax, 1987).
Here, we present an empirical study of 45 published matrices comprising morphological data for both fossil and extant taxa. Using a variety of metrics, we measure the impact of fossil and of extant taxa upon inferred phylogenetic relationships by comparing the results of parsimony analyses of the full data with comprehensive analyses in which a single fossil or extant taxon is deleted from the data. We also use bootstrap leaf stabilites (Thorley and Wilkinson, 1999; Wilkinson, 2006) to compare the support for the placements of fossil and extant taxa. Our aim is to evaluate alternative perspectives on the importance of fossils using empirical data, and for each data set we seek to test a series of null hypotheses that fossils are no more or less significant or problematic for phylogenetic inference than are extant taxa. We demonstrate that, in several important respects, the behavior of fossil taxa in the majority of our 45 matrices is indistinguishable from that of their extant counterparts. As an empirical study, we do not address the issue of phylogenetic accuracy, which is unknowable in our case studies. This has been investigated elsewhere in a series of simulation studies (Wiens, 1998, 2003a, 2003b, 2005, 2006). Rather, we focus on the impact of taxa.
|
Material and Methods |
|---|
|
TopAbstractMaterial and MethodsResults and DiscussionConclusionsReferences |
|---|
Data Sets and Parsimony Analyses
Forty-five morphological data matrices were sourced from the literature (Table 1). Papers were identified using the online ISI Science database, Biosis, and Google Scholar, searching the years 1980 to 2006 with the keywords "fossil," "extant," and "phylogeny." We also searched several journals manually: Palaeontology, Journal of Vertebrate Paleontology, Biological Journal of the Linnean Society, Zoological Journal of the Linnean Society, and American Museum Novitates. We selected only matrices including both fossil and extant taxa, containing eight or more taxa in total, with at least three taxa in the smallest group (the minimum number capable of providing a statistically significant difference in their behaviors: Mann-Whitney test yields P = 0.037 where three fossil and five living taxa (or vice versa) are ranked with no overlap). We did not include "total evidence" data sets that combined morphological and molecular characters. Fossils in such matrices typically have many times more "missing data" than their extant counterparts. Data sets covered a range of tree sizes and ratios of fossil to extant taxa and contained fossils of all ages (Table 1). Where multiple data sets from the same study group were available (e.g., Cetacea), we avoided those that were very obviously derivative of others. Matrices were retyped and reanalyzed to ensure that the same number of trees of the same length as originally reported could be reproduced. All character weights and orderings applied by the original authors were also implemented in this study. All parsimony analyses and calculations of tree-to-tree distances were performed with PAUP* 4.0b10 (Swofford, 2002). One hundred random addition sequences followed by TBR branch swapping were found to be effective in recovering the original MPTs, and these options were used in all subsequent analyses.
|
Taxon Deletion
Impact on inferred relationships
A central question is whether fossils are more or less likely than extant taxa to overturn relationships inferred in their absence. To test whether fossils and extant taxa differ in their impact upon relationships inferred for other taxa, we used a first-order taxon jackknife (Fig. 1), comparing the results of phylogenetic analyses with and without each taxon. Trees that differed only in the inclusion/exclusion of a single taxon were compared by pruning that taxon from the larger tree (i.e., the tree inferred from the full data). Where there were multiple trees after pruning, these were condensed to remove duplicates. To measure change in inferred relationships due to a given taxon we used (1) RF, the Robinson and Foulds (1981) tree-to-tree distance (= symmetric difference on full splits) and (2) D, the number of leaves pruned from the maximum agreement subtree (Finden and Gordon, 1985). The two measures capture different aspects of differences in relationships in compared trees. Figure 2 illustrates a tree (the original) compared with three others that can be obtained by swapping two similarly sized branches on the original. In the first, the swap is between adjacent terminal taxa separated by just two internal nodes (close and shallow). In the second, the swap is again between single terminal taxa, but this time separated by six internal nodes (distant and shallow). In the third derivative, the swap is between relatively deep adjacent branches, separated by just two nodes (close and deep). The symmetrical difference distance considers derivatives 1 and 3 to be equally different from the original (RF = 2), despite the difference in depth. Derivative 2, by contrast, is considered much more different from the original (RF = 10), reflecting the number of nodes over which the swap is made. The maximum agreement subtree distance provides a very different signal, being sensitive to the size and depth of the branches being swapped, as well as distance, and being greatest for the deep swap (D = 3). This measure places greater emphasis on differences between major branches than the symmetrical-difference distance. The fourth alternative (distant and deep) is not readily illustrated on such small trees.
|
|
Most analyses yielded more than one MPT, necessitating measurement of the distance between two sets of trees. A crude measure of the distance between the two sets could be inferred by deriving a consensus tree for each set and calculating the distance between the pair of consensus trees. However, sets of MPTs that differ greatly from one another can yield identical consensus trees, and sets of MPTs that differ very little from one another can yield quite dissimilar consensus trees, depending upon the methods used. An alternative would be to measure the mean distance between every tree in one set and every tree in the other. Unfortunately, this has the disadvantage that two identical sets of trees will yield a mean distance closely approximate to the mean distance between all possible pairs of trees within each group (the within-group mean may be slightly greater than the between-group mean because the former does not include the distances between each tree and itself—all by definition zero—in the calculations). In order to derive a measure that is zero for two identical sets of MPTs, we have measured the mean distance between each tree in each group and its nearest neighbor (on each of the distance measures) in the other group (Fig. 3).
|
Impact on numbers of trees and on resolution of the strict consensus
A second question is whether fossil or extant taxa are more likely to increase the number of MPTs. We used the first-order jackknife to test this directly by comparing numbers of MPTs inferred with and without each taxon. We also compared the latter to the number of distinct trees when the taxon is pruned from the MPTs for the full data, which allowed us to determine whether data from fossils or extant taxa are more likely to increase the number of most parsimonious interpretations of the interrelationships of taxa other than themselves. A closely related issue is the impact of taxa on the resolution of the strict consensus tree. The number of nodes in the strict consensus can be reduced to one even with just two fundamental trees. We first calculated the number of nodes in the strict consensus of MPTs found with the a priori first-order jackknifing of each taxon. We also compared each of these with the number of nodes in the corresponding strict consensus trees obtained after pruning each taxon from the set of MPTs for all taxa.
Impact on levels of homoplasy
A third question concerns the extent to which fossils and extant taxa are likely to introduce more or less homoplasy. Here we used differences in the modified consistency index (CI') (i.e., excluding uninformative characters) (Kluge and Farris, 1969) and the retention index (RI) (Farris, 1989) when taxa are included or not as measures of the homoplasy introduced by a taxon.
To summarize the effects of jackknifing fossil and living taxa upon CI' values within a single data set, we first calculated the CI' when jackknifing each taxon via the searched route (one value for each taxon). These were compared with the mean CI' (there could be different values for each tree) when pruning each taxon (again, yielding one CI' value for each taxon). The mean CI' values from the pruned route were always less than or equal to those from the searched route (i.e., pruned trees were longer than or as long as the searched trees), and the difference between them provided an index of the impact of each taxon. Finally, we calculated the mean difference in index (impact) for fossil and living taxa (two values for each data set). In order to compare these values over all 45 data sets, we used binomial and Wilcoxon tests. A similar procedure was used for the RI.
Leaf Stability
Another concern regarding fossils is that their inclusion can result in low levels of support for otherwise well-supported relationships (Wilkinson, 2003). Most support measures (e.g., bootstrap proportions, decay indices, etc.), traditionally calculated for clades, can also be determined for other relationships such as triplets (three taxon statements/subtrees, such as "A and B are more closely related to each other than to C"). In rooted trees, the leaf stability (LS) (Thorley and Wilkinson, 1999; Thorley and Page, 2000; Wilkinson, 2006) of a taxon is the average of the support for the triplets including that taxon (Fig. 4). The rationale is that stable taxa will contribute to well-supported triplets. Taxa with lower leaf stabilities are more likely to impact negatively upon apparent support. Thus we can use leaf stabilities to measure directly the extent to which relationships of a given fossil or extant taxon to the other taxa are well supported, and use this as a proxy for the likely impact on apparent support. Here we used bootstrapping to assess support and the bootstrap maximum (the highest bootstrap support value for any resolution of a triplet) as the measure of triplet support. Bootstrapping can be prohibitively time consuming for larger, "noisy" matrices, and we implemented two compromise search strategies here. The first used rather "dirty" heuristics: closest addition, retaining just one tree at each step, with MulTrees deactivated, but with 10,000 bootstrap replicates. The second used 100 random additions, followed by TBR branch swapping retaining just one tree at each step and with MulTrees deactivated. Just 200 bootstrap replicates were made. RadCon (Thorley and Page, 2000) was used to calculate leaf stabilities from the bootstrap treefiles. In practice, processing limitations of our version of RadCon prohibited calculating leaf stability values for 9 of our 45 data sets, and no values are available for those data sets as a result (indicated with an asterisk in online Appendix 1, http://www.systematicbiology.org).
|
Detecting Differences between the Impact of Fossil and Living Taxa
The absolute impact of adding/removing a taxon depends upon the rest of the data (Kearney, 2002). Hence, the number of taxa, the number of characters, the distribution of character states and the topologies of the MPTs can all influence the probable magnitude of taxon deletion effects. Although it is not so straightforward to compare the effects of deleting taxa from different matrices, most of the confounding factors are controlled when comparing the effects of removing taxa from the same matrix. We tested for differences in the behavior of fossil and extant taxa within each data set using Mann-Whitney U tests. Differences in the impact of fossil and extant taxa over all data sets (a single hypothesis for each measure) were tested by first calculating mean values for fossil and extant taxa in each data set (45 pairs of values, except for leaf stability). The null hypotheses of no difference in the impact of the two groups were then tested using both binomial (assuming equal numbers of data sets with higher means for fossils or for extant taxa) and Wilcoxon tests on matched pairs.
Are Living and Fossil Taxa Sampled Differently?
Phylogenetic analyses rarely include all available species of a particular clade, and workers must make subjective decisions regarding the selection and coding of taxa. Even when phylogenies are "complete" (including all known species descended from some ancestor), records of living species might be expected to sample true extant diversity better than the fossil record samples true extinct diversity. The environmental conditions necessary for fossilization are extremely uncommon, so that the number of fossils from any group is probably vanishingly small relative to the total number of individuals that existed within it. Species containing many individuals with mineralized hard parts, and living in environments susceptible to rapid sedimentation and burial are those most likely to have left fossils. Such characteristics are very often clade specific (e.g., bivalve mollusks, trilobites) such that some groups are likely to be sampled better than others, and many clades will leave little or no record (e.g., most meiofaunal phyla [Fortey et al., 1996]). These differences in sampling suggest that extant taxa may often have more similar codings or be phenetically closer on average to their nearest neighbors than fossils. However, because "complete" sampling is impractical when investigating large clades and deep relationships, many workers select a rarefied set of terminals in a manner designed to capture the maximum range of morphological variation. This may entail including a subsample of living taxa alongside all of the fossil data. Taxa may be selected as putatively plesiomorphic or "typical" exemplars from their parent group (e.g., one species from each genus or family), or putative monophyla (genera, species, or higher taxa in some cases) can be coded as a whole, sometimes using polymorphic codings. The coding of inferred groundplans is also possible. Insofar as the original taxonomy provides a very approximate guide to morphological diversity, all of these sampling schemes are defensible at some level. Whatever the approach, however, the relative scarcity of fossil taxa means that they cannot be sampled as densely as it is possible to sample extant forms. However, this does not imply that living taxa are necessarily sampled more densely than fossils in particular cases.
These sampling differences may have implications for the probable impact of a given taxon on the global phylogeny. Take the hypothetical case where two species share identical codings for all characters in a matrix. Deleting either one of these alone cannot affect MPT length or relationships, because the other taxon behaves identically in all respects (Wilkinson, 1995). Moreover, if both taxa belong to the same group (fossil or extant), there are two taxa in one group whose removal appears to have no impact. Removing both taxa simultaneously might have a huge impact on both relationships and length, but this is never revealed by a first-order jackknife. More generally, other things being equal, the more similar the codings for two taxa, the less likely the removal of either will influence topology or length. Thus, if differences in the sampling of fossils and extant taxa make it more likely that members of one group (fossils or extant taxa) have more similar "nearest neighbors," this could result in bias. For example, studies that were initiated with fossil or extant taxa may contain only a token selection of representatives of the other group. Alternatively, studies of a large crown group may code just a selection of exemplars (e.g., one from each subclass) but all of the fossil species where these are few in number. To investigate possible differences between fossil and extant taxa, we measured the minimum character state distance (corrected for missing entries) between each taxon and the taxon most similar to it (its nearest neighbor), irrespective of group (fossil or extant). Although it is not possible to control for all preservational and sampling biases, to the extent that our study asks questions concerning the impact of taxa in published matrices, such differences may be integral to the nature of fossil and extant taxa as commonly conceived and implemented.
|
Results and Discussion |
|---|
|
TopAbstractMaterial and MethodsResults and DiscussionConclusionsReferences |
|---|
Fossil and Extant Taxa Have a Similar Impact on Inferred Relationships
The majority of the data matrices (39 from 45) showed no significant difference in the impact of fossil and extant taxa on either RF or D measures (Table 1). Two data sets showed a significantly greater impact for extant taxa according to the Mann-Whitney tests: "Jurassic lizards" (Evans and Chure, 1998) (P = 0.009 using RF, P = 0.014 using D) and "Cetacea 2" (Kimura and Ozawa, 2002) (P = 0.036 using RF and D). Four data sets showed a significantly greater impact for fossil taxa: "snakes 2" (Lee and Scanlon, 2002) (P = 0.022 using RF and D), "Microcavi" (Ubilla et al., 1999) (P = 0.043 using D), "Cretaceous teiids" (Nydam and Cifelli, 2002) (P = 0.011 using D), and "hedgehogs" (P = 0.001 using RF, P = 0.000 using D). If significant and nonsignificant differences are considered, there were 22 data sets where the extant taxa had a mean RF greater than the fossils, and 20 data sets where the reverse was true (binomial test, indistinguishable from random with P = 0.439). There were three cases of draws. Using D, there are 21 cases where extant taxa had the higher mean impact, 21 cases where fossil taxa had higher mean impact, and three cases of draws. Over all 45 data sets, Wilcoxon tests indicated that the mean impact of fossil and living taxa was not significantly different, whether using the RF (P = 0.438) or D (P = 0.349).
Within particular data sets, individual taxa often have a particularly strong effect on relationships, despite a nonsignificant difference between the impact of fossil and extant taxa overall. For example, within the edentates matrix (Gaudin, 1995), omission of the fossil Megalonyx resulted in a mean RF value of 12.0 (Fig. 5). This is much greater than the next highest ranked, the living Eremotherium, with a mean value of 3.6, although a Mann-Whitney U test for this matrix detected no significant difference between fossil and extant impacts overall.
|
." The symmetrical-difference distance between the topologies in a and b is given simply by the sum of unique branches (RF = 12 in this case). (c) One of six maximum agreement subtrees that can be derived by combining the pruned and searched trees. Each contains 19 terminals from a possible 28. The maximum agreement subtree distance is given simply by the number of terminals omitted (D = 28 – 19 = 9).Fossil and Extant Taxa Are Equally Likely to Increase Numbers of MPTs and Reduce Resolution of the Strict Consensus
The number of MPTs was marginally but not significantly more likely to be smaller when jackknifing fossil taxa compared with their extant counterparts. Mann-Whitney tests identified significant differences for just six from 45 matrices. These were "Carsosaurs" (P = 0.035), "Cetacea 2" (P = 0.030), "crown carnivores" (P = 0.006), "P haerodus" (P = 0.021), "Jurassic lizards" (P = 0.015), and "snakes 1" (P = 0.011). In just two of these cases ("Cetacea 2" and "Jurassic lizards"), jackknifing extant taxa yielded fewer MPTs, on average, than jackknifing fossils. In the other four cases, the reverse was true. If significant and nonsignificant differences are considered, jackknifing extant taxa yielded fewer MPTs in 20 data sets, whereas jackknifing fossil taxa yielded fewer MPTs in 24 data sets (a nonsignificant difference, binomial test, P = 0.326). Finally, a Wilcoxon matched-pairs test on the mean number of MPTs produced when jackknifing fossil or extant taxa showed no significant difference (P = 0.852). Strict consensus resolution was also similar when jackknifing fossils compared with extant taxa. Mann-Whitney tests identified significant differences for just five from 45 matrices. For four of these, jackknifing fossils resulted in greater resolution. These were: "apternodus shrews" (P = 0.042), "crown carnivores" (P = 0.017), "Phaerodus" (P = 0.001), and "snakes 1" (P = 0.021). For one data set, Jurassic lizards, strict consensus resolution was better when jackknifing extant taxa (P = 0.030). Over all 45 data sets, the mean consensus resolution when jackknifing fossils was similar to that when jackknifing extant taxa (Wilcoxon matched-pairs test, P = 0.635).
Concerns that fossils might inflate numbers of MPTs and reduce consensus resolution therefore arise from particular case studies. A greater proportion of missing entries (Huelsenbeck, 1991) or a reduction in the absolute number of scored cells (Wiens, 2003a, 2003b) in fossil taxa has usually been blamed. Wilkinson (1995) found substantially more missing entries in purely fossil than purely neontological data sets. Our sample of matrices confirmed the significantly greater abundance of missing data in fossil than in extant taxa for "mixed" data sets (means of 27.1% versus 4.0% missing entries) (Wilcoxon matched-pairs test across all data sets, P = 0.000). However, the extent to which missing entries promote instability depends on their distribution more than their absolute number (Wilkinson, 1995; Kearney and Clark, 2003; Wiens, 2003a, 2003b, 2006). We investigated the relationship between amounts of missing data and MPT number by calculating Spearman's rank correlation between the per taxon percentage of missing data and the number of MPTs upon that taxon's deletion for each data set. None of our cladograms yielded P < 0.081, and correlation was weak in all cases. Simulations by Wiens (2003b) demonstrated that the important variable across data sets is number of nonmissing data cells, as opposed to the fraction of missing ones. However, within a given data set, the absolute number of characters is constant, and so the complement of the fraction of missing entries is proportional to the number of scored entries.
Our study did not primarily address the effects of missing data, which have been investigated much more directly and thoroughly elsewhere (Wilkinson, 1995, 2003; Wiens, 1998, 2003a, 2003b, 2005, 2006). Huelsenbeck (1991), Wills and Fortey (2000), and others advocate the inclusion of fossils chiefly because they sample extinct lineages from closer in time to key cladogenetic events. As such, we considered the amount of missing data to be a conflating variable. However, we did not set out to sample data sets in which the codings for fossils were relatively complete but rather to sample as many as we could find. The exclusion of "total evidence" data sets (combining morphology with molecules) from our sample limited the number and proportion of missing codings for fossils. Because the number of molecular characters typically greatly exceeds the number of morphological characters, fossils are often largely represented by missing entries in combined matrices. We note that the most fragmentary fossils are unlikely to have been included, or may never have been considered for inclusion, in published matrices.
Fossil and Extant Taxa Have a Similar Impact on Levels of Homoplasy
Seven of the 45 data sets showed a significant difference (Mann-Whitney tests) in the extent to which jacknifing fossil or extant taxa impacted upon the CI' and/or RI (online Appendix 1, http://www.systematicbiology.org). In four of these seven cases, removing fossils caused more improvement in the CI' and/or RI than removing extant taxa. These were "Cretaceous teiids" (P = 0.004 for CI'), "Phaerodus" (P = 0.003 and 0.001 for CI' and RI, respectively), "snakes 2" (P = 0.004 and 0.017), and "swans" (P = 0.012 for both CI' and RI). In the remaining three data sets, the reverse was true: "Cetacea 2" (P = 0.009 and 0.014), "elasmobranchs" (P = 0.032 and 0.027), and "Jurassic lizards" (P = 0.003 and 0.005). Binomial and Wilcoxon tests over all 45 matrices indicated no significant differences (binomial, P = 0.082 and 0.215 for CI' and RI, respectively; Wilcoxon, P = 0.401 and 0.857). Taken at face value, this indicates that fossil and extant taxa are equally likely to increase overall homoplasy.
Fossils May Be Less Stable Than Extant Taxa
Leaf stability results were very closely correlated between the two bootstrapping approaches (rs = 0.962, P = 0.000). Mann-Whitney tests detected no significant differences in the leaf stabilities of fossil and extant taxa in 27 of the 36 data sets investigated (online Appendix 1, www.systematicbiology.org). However, in all of the nine remaining cases ("armadillos," P = 0.037; "Cryptodira," P = 0.013; "dolichosaurs," P = 0.009; "elasmobranchs," P = 0.010; "Gonorynchiformes," P = 0.006; "hedgehogs," P = 0.001; "Ornithurae," P = 0.001; "snakes 2," P = 0.003; "teleosts," P = 0.006), extant taxa were significantly more stable than fossils (online Appendix 1, http://www.systematicbiology.org). If significant and nonsignificant differences are considered, there were 30 cases where extant taxa had a mean stability greater than fossils, and 6 cases where the reverse was true (a highly significant difference: binomial test, P = 0.000). Moreover, mean LS values for fossil and extant taxa over the 36 data sets indicated a highly significant difference (Wilcoxon matched-pairs test, P = 0.000). The inclusion of unstable taxa can reduce the support for otherwise well-supported relationships. Because fossils are more likely to be unstable, they are also more likely to reduce support in the rest of the tree. However, the degree of instability does not appear to be a simple function of the amount of missing data. Only one data set ("paddlefish") had a significant correlation between the leaf stability of taxa (whether fossil or extant) and their proportion of missing codings (rs = –0.70, P = 0.043).
Fossils Tend to Resolve Closer to the Roots Than Extant Taxa
Taxa might be expected to impact most strongly upon relationships close to their phylogenetic position. Taxa separated from the root by many nodes would be less likely to affect relationships among deep branches than taxa placed closer to the root. Fossils can resolve anywhere in a tree, but to the extent that the order of cladistic branching and stratigraphic order tend to correlate (Norell and Novacek, 1992; Benton et al., 1999; Wills, 1999; Wagner and Sidor, 2000), older fossils are more likely to branch closer to the root than younger fossils or extant taxa. The precise positioning of fossils depends on the taxonomic scope of the study (e.g., even an early branching, plesiomorphic mammal will resolve many nodes up from the root in a tree of all tetrapods). The maximum agreement subtree distance (D) is sensitive to "deeper" changes in inferred relationships (Fig. 2). Here the "height" of taxa in a majority-rule consensus of the original trees was approximated using both the patristic distance and the number of nodes between each ingroup taxon and the outgroup taxon closest to the ingroup (online Appendix 1, http://www.systematicbiology.org). There were significant differences in the height of fossil and extant taxa on one (or more usually, both) of these measures in 13 from 45 cases. Significant data sets using patristic (where one P-value is quoted) or patristic and nodal distances respectively (where two P-values values) were: "African endemic mammals," P = 0.005 and 0.001; "anurans," P = 0.006; "Birgeria," P = 0.008 and 0.037; "Cetacea 1," P = 0.034 and 0.024; "Cetacea 2," P = 0.012; "Cretaceous teiids," P = 0.001 and 0.011; "Crocodylia," P = 0.0006 and 0.011; "dolichosaurs," P = 0.030 and 0.018; "edentates," P = 0.010 and 0.009; "Gnoristinae," P = 0.018 and 0.011; "mammal origins," P = 0.011 and 0.009; "Ornithurae," P = 0.001 and 0.001; "snakes 2," P = 0.003. In all cases, the fossils were closer to the root than the extant taxa (although we note that taxa with greater amounts of missing data are likely to have shorter patristic distances in a parsimony framework). Across all 45 data sets, neither the mean patristic nor nodal distances for fossil and extant taxa were significantly different (Wilcoxon matched-pairs tests, P = 0.596 and P = 0.319, respectively). Despite the deeper placement of fossils in 13 data sets (coupled with the prediction that removing deeper branching taxa is more likely to impact on inferred relationships), in only 3 of these 13 matrices was there a significant difference in the impact on relationships of fossil and extant taxa ("Cetacea 2" with more influential living taxa; "Cretaceous teiids" and "snakes 2" with more influential fossils). Moreover, even if (under some circumstances) fossils have enhanced impact in part because they are closer to the root, this placement is part of what makes them intrinsically valuable (Huelsenbeck, 1991; Wills and Fortey, 2000).
Fossils Are Closer to Their Nearest Phenetic Neighbors Than Extant Taxa
The taxa that differ most in their codings from the other taxa in a matrix might be expected to have the greatest impact on relationships, CI', tree number, and strict consensus resolution upon their inclusion/exclusion. We might also expect fossil and extant taxa to have nearest neighbors with differing proximities. This seems especially likely where a small number of token fossils have been coded in a matrix of predominantly extant taxa, or vice versa. A group with much denser taxon sampling (which correlates with the densest sampling of morphology) should also contain the taxa with the nearest phenetic neighbors. Mean pairwise character distances between taxa were calculated in PAUP, preserving any orderings and weightings used by the original authors, and adjusting for missing data by averaging over only those character comparisons scored as not missing for both taxa (online Appendix 1, http://www.systematicbiology.org). Mann-Whitney U tests detected no significant differences within any of the 45 matrices (P 
0.092). However, over all 45 data sets, fossils had closer nearest neighbors than extant taxa on average in 32 cases, with six ties (binomial test, P = 0.000). By contrast, a Wilcoxon matched-pairs test of mean nearest neighbor distances for fossil and living taxa over all data sets demonstrated no significant difference (P = 0.000). This is the opposite result from that predicted on the basis of sampling considerations. The extinct biota is subject to all the biases inherent in the fossil record, whereas all living species can be sampled, at least in principle. However, it is often claimed that fossils are morphologically intermediate between living forms (Briggs and Fortey, 2005), and this result is consistent with a diverging model of clade evolution in which many extant taxa are morphologically isolated from one another after the winnowing extinction of fossil branches.
Any Taxon Can Have Unpredictable and Pronounced Effects on Relationships
The unpredictable effects of omitting fossils are exemplified in the phylogeny of scalidophoran worms and their allies provided by Dong et al. (2005). These authors focused on the phylogenetic position of a key fossil find, Markuelia. In their analysis, this worm resolved close to the root in both of two MPTs, but more closely related to all other cephalorhynchs than nematodes, nematomorphs, Ancalagon and possibly Fieldia (Fig. 6a). Omitting the well-studied fossil priapulid Ottoia (Conway Morris, 1977; Wills, 1998b) had the greatest mean impact on topology overall (RF = 17.8 and D = 8.4). As noted by Dong et al. (2005), omitting other taxa, particularly the early-branching plesions Fieldia and Ancalagon, also had a marked implact on relationships. Ottoia by contrast is highly derived, and separated from Markuelia by 10 or 11 nodes (only two taxa, Louisella and Scolecofurca, were more distant). Despite a large proportion of missing codings in Ottoia (38% missing or inapplicable), its deletion resulted in an increase from 2 to 85 MPTs. Markuelia (among other taxa) becomes more unstable and the strict consensus is highly unresolved (Fig. 6b), but in the majority of the 85 MPTs, Markuelia nests much further from the root. This example also illustrates that the mean impact is a conservative index. Maximum impact (the difference between the two most dissimilar trees from the pruned and searched sets following deletion of Ottoia) was much higher (RF = 32, D = 11) (Fig. 6c). Hence the omission of one taxon has the potential to radically change inferred relationships, influence the position of distantly related taxa, and reduce resolution of consensus trees. Comparison of character change lists for the two most different MPTs from the searched and pruned routes revealed no straightforward reason for the impact of Ottoia. Considering all of our data sets, the most influential taxa are no more or less often fossil than extant (chi-square test for RF yields P = 0.727, and for D yields P = 0.211).
|
|
Conclusions |
|---|
|
TopAbstractMaterial and MethodsResults and DiscussionConclusionsReferences |
|---|
This study strongly supports the inclusion of fossils in phylogenetic studies. In most respects, the behavior of the fossils in our data sets was similar to that of their extant counterparts. There was no significant difference in the extent to which they were likely to affect inferences of relationships (measured by their impact on two indices of tree distance), numbers of MPTs, resolution of strict consensus trees, and levels of homoplasy (measured using the CI' and RI). The only detrimental effect was that fossils were significantly more likely to reduce overall support for the tree than extant taxa, although the difference was not great.
Systematists routinely make subjective decisions regarding the sampling of taxa in their analyses. There are many valid reasons why some taxa should be included and others omitted. The most often cited are schemes that sample "typical" or plesiomorphic representatives of all the putative monophyla of interest, those that sample "typical" or plesiomorphic representatives of groups at some arbitrary taxonomic level, and those that attempt to sample uniformly morphological diversity. All are justifiable. However, we find no reason why the data from fossils should be singled out a priori as inferior: indeed, the inclusion of fossil species may actually help to satisfy some of the above criteria. We also note that our results came from a sample of matrices that already included fossil taxa. These, by definition, derived from a subset of practitioners who already considered fossils in their taxon sampling schemes. It does not follow that fossil and extant taxa will always have a comparable impact when added to all data sets. Most matrices produced by neontologists are intended to offer reasonably even sampling (however defined) of the extant biota, but include no fossils. These are progressively less likely to be affected by the addition of more extant terminals as the taxon sampling density increases. The addition of small numbers of fossil taxa, by contrast, is much more likely to sample an otherwise overlooked branch of the tree of life.
Each taxon should be considered individually, and omitted from analyses only if it can be demonstrated to introduce particular problems. Taxa should not be omitted a priori because of characteristics that they supposedly possess by dint of being members of particular groups. Of the several valid reasons for omitting particular taxa, all apply to extant as well as fossil forms. The most commonly cited is where the number or distribution of missing data codings in a taxon is such that its inclusion prohibitively extends search times, increases numbers of optimal trees, or otherwise obfuscates well-supported relationships. Simulations have demonstrated that the proportion of missing codings is less critical in this regard than the number of characters actually coded (Wiens, 2005, 2006). Moreover, even large proportions of missing data need not result in spurious inferences of relationships (Wiens, 2003a, 2003b; Wiens et al., 2006)
In some cases the excluson of taxa can be justified when it is clear from their lack of unique combinations of character states that their inclusion cannot alter the parsimonious interpretation of relationships among the other taxa (Wilkinson, 1995). More generally, methods commonly used for measuring support, such as bootstrapping and decay analysis, can be extended using reduced consensus methods to overcome any obfuscatory effect of particular taxa, whether they are fossil or extant (Wilkinson, 1996; Wilkinson et al., 2000). Our results suggest that such methods should be explored when resolution or support appears poor as an alternative to excluding particular taxa.
A related but potentially more damaging issue is where the states of many of the coded characters are open to multiple interpretations. This can either happen when the study material is indistinct or fragmentary, or where the homology of structures is particularly unclear (e.g., Dunlop and Arango, 2005; Glenner et al., 1995). Frustratingly, the more unusual and interesting the material, then the greater the temptation to overinterpret in order to place the taxon phylogenetically (Briggs and Fortey, 2005). The dangers of introducing combinations of characters that do/did not, in fact, exist are potentially much greater than those of incorporating missing data (Wagner, 2000). Evaluating putative homologies can be problematic for both fossil and living taxa (Richter, 2005). Admittedly, implementing the usual "tests" for homology can become more difficult as the morphological difference between species increases, and some of the oldest fossils do indeed attract controversy in this regard (Brasier and Antcliffe, 2004; Budd and Jensen, 2000; Chen et al., 2004; Ramskold, 1992). The very oldest metazoan fossils are separated from their living descendants (or the last common ancestor of both) by, perhaps, 600 million years (Newman et al., 2006) (or 600 million years "worth" of evolution). However, because the majority of phyla are recognizable by the Middle Cambrian at the latest (Conway Morris, 2006; Marshall, 2006), the living representatives of different phyla will have diverged and evolved independently for at least 513 million years (or around 1 billion years "worth"), and possibly much longer (Blair and Hedges, 2005). Clearly, time does not correlate simply with morphological difference or the difficulty of establishing homologies, but by the same token, neither does the age of a particular fossil. Neontologists have all the molecular and developmental techniques of "evo-devo" at their disposal, and these offer impressive tests of homology at a variety of levels (Rutishauser and Moline, 2005; Jager et al., 2006). However, such approaches are not a panacea (Cracraft, 2005), and their application is typically uncommon relative to the number of characters. Neither are problems of homoplasy limited to morphological data (Brocchieri, 2001).
Another possible justification for omitting particular taxa is when the developmental stage of material is uncertain. Because there may easily be fewer adults than earlier life history stages in a stable population, species known from small numbers of specimens are quite likely to be juveniles. If juveniles are coded as adults (or vice versa), ontogenetic character change can be conflated with evolutionary change, and this may yield spurious relationships. Problems of this kind are less frequent for most extant taxa, because the juvenile forms of common extant species can help to determine whether rarer living material represents juveniles or not (e.g., the hemichordate Planctosphaera pelagica[Hart et al., 1994], the gastropod Balea costigera [Preece and Gittenberger, 2003], and the ephemeropteran Behningia baei [McCafferty and Jacobus, 2006]). By contrast, specialists may disagree over the ontogenetic status of even reasonably abundant ancient fossil material because there are no closely related species with known developmental histories with which to compare them (e.g., several Cambrian crustaceans [Schram, 1986; Walossek and Muller, 1992]). We suggest that such taxa should be included experimentally, and if their impact on inferred relationships is significant, results should be published both with and without them (Wills, 1998a).
Many authors have highlighted the importance of adequate taxon sampling in attempts to answer the deepest (and most interesting?) phylogenetic questions (e.g., Graybeal, 1998; Poe, 1998; Hillis et al., 2003; Pollock et al., 2002; Rydin and Kallersjo, 2002; Zwickl and Hillis, 2002; Lee, 2005). The inclusion of fossils is the only way to sample morphological change in many parts of the tree of life, and their contribution may be vital for breaking up long branches (Wills and Fortey 2000; Wiens, 2005). Fossils are also unique in providing snapshots of morphology from closer in time to the cladogenetic events that systematists attempt to reconstruct (Smith, 1998). As our examples confirm, the omission/inclusion of any taxon can have a marked impact on phylogenetic hypotheses. We therefore recommend that fossils should be treated in the same way as any other taxa: omitting them only where there is specific justification and not merely because they are extinct.
|
Acknowledgements |
|---|
We thank Rob Beck, Olaf Bininda-Emonds, and John Wiens for constructive reviews that allowed us to significantly improve the quality of this work. We also thank David Gower, Ronald Jenner, and Andras Kosztolanyi for their helpful comments. We are grateful to all those who contributed data sets. A. Cobbett was supported by a BBSRC studentship and M. A. Wills's work was supported by BBSRC grant BB/C006682/1.
|
References |
|---|
|
TopAbstractMaterial and MethodsResults and DiscussionConclusionsReferences |
|---|
-
Anderson J. S. The phylogenetic trunk: Maximal inclusion of taxa with missing data in an analysis of the Lepospondyli (Vertebrata, Tetrapoda). Syst. Biol. (2001) 50:170–193.
Asher R. J., McKenna M. C., Emry R. J., Tabrum A. R., Kron D. G. Morphology and relationships of Apternodus and other extinct zalambdodont placental mammals. Bull. Am. Mus. Nat. Hist. (2002) 273:1–117.[CrossRef]
Asher R. J., Novacek M. J., Geisler J. H. Relationships of endemic African mammals and their fossil relatives based on morphological and molecular evidence. J. Mamm. Evol. (2003) 10:131–194.[CrossRef]
Ax P. The phylogenetic system (1987) Chichester: John Wiley and Sons.
Baez A. M., Harrison T. A new pipine frog from an Eocene crater lake in north-central Tanzania. Palaeontology (2005) 48:723–737.[CrossRef][Web of Science]
Baez A. M., Pugener L. A. New Paleogene pipid frog from northwestern Patagonia. J. Vertebr. Paleontol. (1998) 18:511–524.
Baskin J. A. Bassariscus and Probassariscus (Mammalia, Carnivora, Procyonidae) from the early Barstovian (middle Miocene). J. Vertebr. Paleontol. (2004) 24:709–720.[CrossRef]
Benton M. J., Hitchin R., Wills M. A. Assessing congruence between cladistic and stratigraphic data. Syst. Biol. (1999) 48:581–596.
Blagderov V., Grimaldi D. Fossil Sciaroidea (Diptera) in Cretaceous ambers, exclusive of Cecidomyiidae, Sciaridae, and Keroplatidae. Am. Mus. Novit. (2004) 3433:1–76.[CrossRef]
Blair J. E., Hedges S. B. Molecular phylogeny and divergence times of deuterostome animals. Mol. Biol. Evol. (2005) 22:2275–2284.
Boisserie J. R., Lihoreau F., Brunet M. The position of Hippopotamidae within Cetartiodactyla. Proc. Natl. Acad. Sci. USA (2005) 102:1537–1541.
Brasier M., Antcliffe J. Decoding the Ediacaran enigma. Science (2004) 305:1115–1117.
Briggs D. E. G., Fortey R. A. Wonderful strife: Systematics, stem groups, and the phylogenetic signal of the Cambrian radiation. Paleobiology (2005) 31:94–112.
Brocchieri L. Phylogenetic inferences from molecular sequences: Review and critique. Theor. Popul. Biol. (2001) 59:27–40.[CrossRef][Web of Science][Medline]
Budd G. E., Jensen S. A critical reappraisal of the fossil record of the bilaterian phyla. Biol. Rev. (2000) 75:253–295.
Caldwell M. W. Squamate phylogeny and the relationships of snakes and mosasauroids. Zool. J. Linn. Soc. (1999) 125:115–147.[CrossRef]
Caldwell M. W., Carroll R. L., Kaiser H. The pectoral girdle and forelimb of Carsosaurus marchesetti (Aigialosauridae), with a preliminary phylogenetic analysis of mosasaurids and varanoids. J. Vertebr. Paleontol. (1995) 15:516–531.
Chen J. Y., Waloszek D., Maas A. A new "great-appendage" arthropod from the Lower Cambrian of China and homology of chelicerate chelicerae and raptorial antero-ventral appendages. Lethaia (2004) 37:3–20.[Web of Science]
Conway Morris S. Fossil priapulid worms. Spec. Pap. Palaeontol. (1977) 20:1–155.
Conway Morris , S. The fossil record and the early evolution of the Metazoa. Nature (1993) 361:219–225.[CrossRef]
Conway Morris , S. Darwin's dilemma: The realities of the Cambrian "explosion." Phil. Trans. R. Soc. Lond. B Biol. Sci. (2006) 361:1069–1083.
Cozzuol M. A. A "northern" seal from the Miocene of Argentina: implications for phocid phylogeny and biogeography. J. Vertebr. Paleontol. (2001) 21:415–421.[CrossRef]
Cracraft J. Phylogeny and evo-devo: Characters, homology, and the historical analysis of the evolution of development. Zoology (2005) 108:345–356.[CrossRef][Web of Science][Medline]
Dong X. P., Donoghue P. C. J., Cunningham J. A., Liu J., Cheng H. The anatomy, affinity and phylogenetic significance of Markuelia. Evol. Dev. (2005) 7:468–482.[CrossRef][Web of Science][Medline]
Donoghue M. J., Doyle J. A., Gauthier J., Kluge A. G., Rowe T. The importance of fossils in phylogeny reconstruction. Annu. Rev. Ecol. Syst. (1989) 20:431–460.[CrossRef][Web of Science]
Dooloey A. C., Fraser N. C., Luo Z. The earliest known member of the rorqual-gray whale clade (Mammalia: Cetacea). J. Vertebr. Paleontol. (2004) 24:453–463.[CrossRef]
Dunlop J. A., Arango C. P. Pycnogonid affinities: A review. J. Zool. Systemat. Evol. Res. (2005) 43:8–21.[CrossRef]
Evans S. E., Chure D. C. An unusual braincase (?Stokesosaurus clevelandi) from the Cleveland Lloyd Dinosaur Quarry, Utah (Morrison Formation; Late Jurassic). J. Vertebr. Paleontol. (1998) 18:115–147.
Farris J. S. The retention index and the rescaled consistency index. Cladistics (1989) 5:417–419.[CrossRef][Web of Science]
Finarelli J. A., Flynn J. J. Ancestral state reconstruction of body size in the Caniformia (Carnivora, Mammalia): The effects of incorporating data from the fossil record. Syst. Biol. (2006) 55:301–313.
Finden C. R., Gordon A. D. Obtaining common pruned trees. J. Classif. (1985) 2:255–276.[CrossRef]
Fortey R. A., Briggs D. E. G., Wills M. A. The Cambrian evolutionary "explosion": Decoupling cladogenesis from morphological disparity. Biol. J. Linn. Soc. (1996) 57:13–33.[CrossRef][Web of Science]
Gaffney E. S. The postcranial morphology of Meiolania platyceps and a review of the Meiolaniidae. Bull. Am. Mus. Nat. Hist. (1996) 229:3–166.
Gao K.-Q., Cheng Z. W. A new lizard from the Lower Cretaceous of Shandong, China. J. Vertebr. Paleontol. (1999) 19:456–465.
Gao K.-Q., Wang Y. Mesozoic anurans from Liaoning Province, China, and phylogenetic relationships of archaeobatrachian anuran clades. J. Vertebr. Paleontol. (2001) 21:460–476.[CrossRef]
Gaudin T. J. A re-examination of elasmobranch monophyly and chondrichthyan phylogeny. Neues Jahrbuch. Geol. Palaontol. (1991) 182:133–160.
Gaudin T. J. The ear region of edentates and the phylogeny of the Tardigrada (Mammalia: Xenarthra). J. Vertebr. Paleontol. (1995) 15:672–705.
Gaudin T. J., Wible J. R. The phylogeny of living and extinct armadillos (Mammalia, Xenarthra, Cingulata): A craniodental analysis. In: Amniote paleobiology: Perspectives on the evolution of mammals, birds and reptiles—Carrano M. T., Gaudin T. J., Blob R. W., Wible J. R., eds. (2005) Chicago, Illinois: University of Chicago Press. 153–198.
Gauthier J., Kluge A. G., Rowe T. Amniote phylogeny and the importance of fossils. Cladistics (1988) 4:105–209.[Web of Science]
Gentry A. W. The subfamilies and tribes of the family Bovidae. Mamm. Rev. (1992) 22:1–32.[CrossRef]
Glenner H., Grygier M. J., Hoeg J. T., Jensen P. G., Schram F. R. Cladistic analysis of the Cirripedia-Thoracica. Zool. J. Linn. Soc. (1995) 114:365–404.[CrossRef]
Goodwin H. T., Hayes F. E. Morphologically derived ground squirrels from the Borchers local fauna, Meade County, Kansas, with a redescription of ?Spermophilus cragini. J. Vertebr. Paleontol. (1994) 14:278–291.
Gould G. C. Hedgehog phylogeny (Mammalia, Erinaceidae)—the reciprocal illumination of the quick and the dead. Am. Mus. Novit. (1995) 3131:1–45.
Grande L., Bemis W. E. Interrelationships of Acipenseriformes with comments on "Chondrostei." In: Interrelationships of fishes—Stiassny M. L. J., Parenti L. R., Johnson G. D., eds. (1996) London: Academic Press. 85–115.
Grande L., Bemis W. E. A comprehensive phylogenetic study of amiid fishes (Amiidae) based on comparative skeletal anatomy. An empirical search for interconnected patterns of natural history. Society of Vertebrate Paleontology Memoir. (1998) 4: i–x, 1–690; Suppl to J. Vertebr. Paleontol.
Grande L., Jin F., Yabumoto Y., Bemis W. E. 
Protopsephurus liui, a well-preserved primitive paddlefish (Acipenseriformes: Polyodontidae) from the Lower Cretaceous of China. J. Vertebr. Paleontol. (2002) 22:209–237.[CrossRef]
Grande T., Poyato-Ariza F. J. Phylogenetic relationships of fossil and recent gonorynchiform fishes (Teleostei: Ostariophysi). Zool. J. Linn. Soc. (1999) 125:197–238.[CrossRef]
Graybeal A. Is it better to add taxa or characters to a difficult phylogenetic problem? Syst. Biol. (1998) 47:9–17.
Guo-Quing L., Wilson M. V. H. Phylogeny of Osteoglossomorpha. In: Interrelationships of fishes—Stiassny M. L. J., Parenti L. R., Johnson G. D., eds. (1996) London: Academic Press. 163–174.
Harris S. R., Pisani D., Gower D. J., Wilkinson M. Investigating stagnation in morphological phylogenetics using consensus data. Syst. Biol. (2007) 56:125–129.
Hart M. W., Miller R. L., Madin L. P. Form and feeding mechanism of a living Planctosphaera pelagica. Marine Biol. (2003) 120:521–533.[CrossRef]
Harvey P. H., May R. M., Nee S. Phylogenies without fossils. Evolution (1994) 48:523–529.[CrossRef][Web of Science]
Hillis D. M., Pollock D. D., McGuire J. A., Zwickl D. J. Is sparse taxon sampling a problem for phylogenetic inference? Syst. Biol. (2003) 52:124–126.
Horovitz I. A phylogenetic study of living and fossil platyrrhines. Am. Mus. Novit. (1999) 3269:1–40.
Huelsenbeck J. P. When are fossils better than extant taxa in phylogenetic analysis. Syst. Zool. (1991) 40:458–469.[Abstract]
Iwabe N., Hara Y., Kumazawa Y., Shibamoto K., Saito Y., Miyata T., Katoh K. Sister group relationship of turtles to the bird-crocodilian clade revealed by nuclear DNA-coded proteins. Mol. Biol. Evol. (2005) 22:810–813.
Jager M., Murienne J., Clabaut C., Deutsch J., Le Guyader H., Manuel M. Homology of arthropod anterior appendages revealed by Hox gene expression in a sea spider. Nature (2006) 441:506–508.[CrossRef][Web of Science][Medline]
Jiang-Chong Z. New fossil osteoglossomorph from Ningxia, China. J. Vertebr. Paleontol. (2004) 24:515–524.[CrossRef]
Kearney M. Fragmentary taxa, missing data, and ambiguity: Mistaken assumptions and conclusions. Syst. Biol. (2002) 51:369–381.
Kearney M., Clark J. M. Problems due to missing data in phylogenetic analyses including fossils: A critical review. J. Vertebr. Paleontol. (2003) 23:263–274.[CrossRef]
Kimura T., Ozawa T. A new cetothere (Cetacea: Mysticeti) from the early Miocene of Japan. J. Vertebr. Paleontol. (2002) 22:684–702.[CrossRef]
Kluge A. G., Farris J. S. Quantitative phyletics and evolution of anurans. Syst. Zool. (1969) 18:1–32.
Kristensen R. M. An introduction to Loricifera, Cycliophora, and Micrognathozoa. Integr. Comp. Biol. (2002) 42:641–651.
Lee M. S. Y. Squamate phylogeny, taxon sampling, and data congruence. Organ. Divers. Evol. (2005) 5:25–45.[CrossRef]
Lee M. S. Y., Scanlon J. D. Snake phylogeny based on osteology, soft anatomy and ecology. Biol. Rev. (2002) 77:333–401.[Medline]
Li G., Grande L., Wilson M. V. H. The species of Phaerodus (Teleostei: Osteoglossidae) from the Eocene of North America and their phylogenetic relationships. J. Vertebr. Paleontol. (1997) 17:487–505.
Livezey B. C. A phylogenetic analysis of geese and swans (Anseriformes: Anserinae), including selected fossil species. Syst. Biol. (1996) 45:415–450.
Lucas S. G., Luo Z. Adelobasileus from the Upper Triassic of West Texas: The oldest mammal. J. Vertebr. Paleontol. (1993) 13:309–334.
Marshall C. R. Explaining the Cambrian "explosion" of animals. Annu. Rev. Earth. Planet. Sci. (2006) 34:355–384.[CrossRef][Web of Science]
Mayr G. Phylogeny of early Tertiary swifts and hummingbirds (Aves: Apodiformes). Auk (2003) 120:145–151.[CrossRef][Web of Science]
McCafferty W. P., Jacobus L. M. A new oriental species of Behningia Lestage (Ephemeroptera: Behningiidae). J. Insect Sci. (2006) 6. Art. No. 47. http://www.insectscience.org/6.47/.
Nee S., May R. M. Extinction and the loss of evolutionary history. Science (1997) 278:692–694.
Newman S. A., Forgacs G., Muller G. B. Before programs: The physical origination of multicellular forms. Int. J. Dev. Biol. (2006) 50:289–299.[CrossRef][Web of Science][Medline]
Nixon K. C. Paleobotany in cladistics and cladistics in paleobotany: Enlightenment and uncertainty. Rev. Palaeobot. Palynol. (1996) 90:361–373.[CrossRef][Web of Science]
Norell M. A., Novacek M. J. The fossil record and evolution—Comparing cladistic and paleontological evidence for vertebrate history. Science (1992) 255:1690–1693.
Norell M. A., Wheeler W. Missing entry replacement data analysis: A replacement approach to dealing with missing data in paleontological and total evidence data sets. J. Vertebr. Paleontol. (2003) 23:275–283.[CrossRef]
Novacek M. J. Fossils, topologies, missing data and the higher-level phylogeny of eutherian mammals. Syst. Biol. (1992) 41:58–73.
Nydam R. L., Cifelli R. L. A new teiid lizard from the Ceder Mountain Formation (Albian-Cenomanian boundary) of Utah. J. Vertebr. Paleontol. (2002) 22:276–285.[CrossRef]
O'Leary M. A. Parsimony analysis of total evidence from extinct and extant taxa and the Cetacean-Artiodactyl question (Mammalia, Ungulata). Cladistics (1999) 15:315–330.[CrossRef][Web of Science]
Patterson C. Significance of fossils in determining evolutionary relationships. Annu. Rev. Ecol. Systemat. (1981) 12:195–223.[CrossRef]
Pierce S. E., Caldwell M. W. Redescription and phylogenetic position of the Adriatic (Upper Cretaceous; Cenomanian) dolichosaur Pontosaurus lesinensis (Kornhuber, 1973). J. Vertebr. Paleontol. (2004) 24:373–386.[CrossRef]
Poe S. Sensitivity of phylogeny estimation to taxonomic sampling. Syst. Biol. (1998) 47:18–31.
Poe S. Evaluation of the strategy of long-branch subdivision to improve the accuracy of phylogenetic methods. Syst. Biol. (2003) 52:423–428.
Pollock D. D., Zwickl D. J., McGuire J. A., Hillis D. M. Increased taxon sampling is advantageous for phylogenetic inference. Syst. Biol. (2002) 51:664–671.
Preece R. C., Gittenberger E. Systematics, distribution and ecology of Balea (= Tristania) (Pulmonata: Clausiliidae) in the islands of the Tristan-Gough group. J. Molluscan Stud. (2003) 69:329–348.
Ramskold L. Homologies in Cambrian Onychophora. Lethaia (1992) 25:443–460.[Web of Science]
Richter S. Homologies in phylogenetic analyses—Concept and tests. Theor. Biosci. (2005) 124:105–120.
Rieppel O., Kluge A. G., Zaher H. Testing the phylogenetic relationships of the Pleistocene snake Wonambi naracoortensis Smith. J. Vertebr. Paleontol. (2002) 22:812–829.[CrossRef]
Robinson D. F., Foulds L. R. Comparison of phylogenetic trees. Math. Biosci. (1981) 53:131–147.[CrossRef][Web of Science]
Rokas A., King N., Finnerty J., Carroll S. B. Conflicting phylogenetic signals at the base of the metazoan tree. Evol. Dev. (2003) 5:346–359.[CrossRef][Web of Science][Medline]
Rosenberg M. S., Kumar S. Incomplete taxon sampling is not a problem for phylogenetic inference. Proc. Natl. Acad. Sci. USA (2001) 98:10751–10756.
Rutishauser R., Moline P. Evo-devo and the search for homology ("sameness") in biological systems. Theor. Biosci. (2005) 124:213–241.
Rydin C., Kallersjo M. Taxon sampling and seed plant phylogeny. Cladistics (2002) 18:485–513.[CrossRef][Web of Science]
Salisbury S. W., Molnar R. E., Frey E., Willis P. M. A. The origin of modern crocodyliforms: New evidence from the Cretaceous of Australia. Proc. R. Soc. B (2006) 273:2439–2448.[CrossRef][Medline]
Santini F., Tyler J. C. The importance of even highly incomplete fossil taxa in reconstructing the phylogenetic relationships of the Tetraodontiformes (Acanthomorpha: Pisces). Integr. Comp. Biol. (2004) 44:349–357.
Schram F. R. Crustacea (1986) Oxford: Oxford University Press.
Scotland R. W., Olmstead R. G., Bennett J. R. Phylogeny reconstruction: The role of morphology. Syst. Biol. (2003) 52:539–548.
Smith A. B. What does palaeontology contribute to systematics in a molecular world? Mol. Phylogenet. Evol. (1998) 9:437–447.[CrossRef][Web of Science][Medline]
Smith A. B., Littlewood D. T. J. Paleontological data and molecular phylogenetic analysis. Paleobiology (1994) 20:259–273.[Abstract]
Smith N. D., Turner A. H. Morphology's role in phylogeny reconstruction:Perspectives from paleontology. Syst. Biol. (2005) 54:166–173.
Steppan S. J., Pardinas U. F. J. Two new fossil muroids (Sigmodontinae: Phyllotini) from the early Pleistocene of Argentina: Phylogeny and paleoecology. J. Vertebr. Paleontol. (1998) 18:640–649.
Stockley B., Smith A. B., Littlewood T., Lessios H. A., Mackenzie-Dodds J. A. Phylogenetic relationships of spatangoid sea urchins (Echinoidea): Taxon sampling density and congruence between morphological and molecular estimates. Zool. Scripta (2005) 34:447–468.[CrossRef][Web of Science]
Swofford D. L. PAUP* 4.0b10: Phylogenetic analysis using parsimony (*and other methods) (2002) Sunderland, Massachusetts: Sinauer Associates.
Thewissen J. G. M., Williams E. M., Roe L. J., Hussain S. T. Skeletons of terrestrial cetaceans and the relationships of whales to artiodactyls. Nature (2001) 413:277–281.[CrossRef]
Thorley J. L., Page R. D. M. RadCon: Phylogenetic tree comparison and consensus. Bioinformatics (2000) 16:486–487.
Thorley J. L., Wilkinson M. Testing the phylogenetic stability of early tetrapods. J. Theor. Bio. (1999) 200:343–344.[CrossRef][Web of Science][Medline]
Tshudy D., Sorhannus U. Jagtia kunradensis, a new genus and species of clawed lobster (Decapoda: Nephropidae) from the Upper Cretaceous (Upper Maastrichtian) Maasticht Formation, the Netherlands. J. Paleontol. (2000) 74:224–229.
Ubilla M., Piñeiro G., Quintana C. A. A new extinct species of the Genus Microcavia (Rodentia, Caviidae) from the Upper Pleistocene of the northern basin of Uruguay, with paleobiogeographic and paleoenvironmental comments. Stud. Neotrop. Fauna Environ. (1999) 34:141–149.[CrossRef]
Wagner P. J. The utility of fossil data in phylogenetic analyses: A likelihood example using Ordovician-Silurian species of the Lophospiridae (Gastropoda: Murchisoniina). Am. Malacol. Bull. (1999) 15:1–31.
Wagner P. J. Exhaustion of morphologic character states among fossil taxa. Evolution (2000) 54:365–386.[CrossRef][Web of Science][Medline]
Wagner P. J., Sidor C. A. Age rank/clade rank metrics—Sampling, taxonomy, and the meaning of "stratigraphic consistency." Syst. Biol. (2000) 49:463–479.
Walossek D., Muller K. J. The Alum Shale Window—Contribution of Orsten arthropods to the phylogeny of Crustacea. Acta Zool. (1992) 73:305–312.
Wible J. R. Origin of Mammalia: The craniodental evidence reexamined. J. Vertebr. Paleontol. (1991) 11:1–28.
Wiens J. J. Does adding characters with missing data increase or decrease phylogenetic accuracy? Syst. Biol. (1998) 47:625–640.
Wiens J. J. Incomplete taxa, incomplete characters, and phylogenetic accuracy: Is there a missing data problem? J. Vertebr. Paleontol. (2003a) 23:297–310.[CrossRef]
Wiens J. J. Missing data, incomplete taxa, and phylogenetic accuracy. Syst. Biol. (2003b) 52:528–538.
Wiens J. J. Can incomplete taxa rescue phylogenetic analyses from long-branch attraction? Syst. Biol. (2005) 54:731–742.
Wiens J. J. Missing data and the design of phylogenetic analyses. J. Biomed. Informatics (2006) 39:34–42.[CrossRef][Web of Science][Medline]
Wiens J. J., Fetzner J. W., Parkinson C. L., Reeder T. W. Hylid frog phylogeny and sampling strategies for speciose clades. Syst. Biol. (2006) 54:719–748.[CrossRef]
Wilkinson M. Coping with abundant missing entries in phylogenetic inference using parsimony. Syst. Biol. (1995) 44:501–514.
Wilkinson M. Majority-rule reduced consensus methods and their use in bootstrapping. Mol. Biol. Evol. (1996) 13:437–444.[Abstract]
Wilkinson M. Missing entries and multiple trees: Instability, relationships, and support in parsimony analysis. J. Vertebr. Paleontol. (2003) 23:311–323.[CrossRef]
Wilkinson M. Identifying stable reference taxa for phylogenetic nomenclature. Zool. Scripta (2006) 35:109–112.[CrossRef][Web of Science]
Wilkinson M., Benton M. J. Sphenodontid phylogeny and the problems of multiple trees. Phil. Trans. R. Soc. Lond. B Biol. Sci. (1996) 351:1–16.
Wilkinson M., Thorley J. L., Upchurch P. A chain is no stronger than its weakest link: Double decay analysis of phylogenetic hypotheses. Syst. Biol. (2000) 49:754–776.
Wills M. A. A phylogeny of recent and fossil Crustacea derived from morphological characters. In: Arthropod relationships—Fortey R. A., Thomas R. H., eds. (1998a) London: Chapman and Hall. 189–209.
Wills M. A. Cambrian and recent disparity: The picture from priapulids. Paleobiology (1998b) 24:177–199.[Abstract]
Wills M. A. Congruence between phylogeny and stratigraphy: Randomization tests and the gap excess ratio. Syst. Biol. (1999) 48:559–580.
Wills M. A., Fortey R. A. The shape of life: How much is written in stone? Bioessays (2000) 22:1142–1152.[CrossRef][Web of Science][Medline]
Wilson M. V. H. Importance for phylogeny of single and multiple stem-group fossil species, with examples from fresh-water fishes. Syst. Biol. (1992) 41:462–470.
Zhang J. Morphology and phylogenetic relationships of Kuntulunia (Teleostei: Osteoglossomorpha). J. Vertebr. Paleontol. (1998) 18:280–300.
Zhou Z., Zhang F. A beaked basal ornithurine bird (Aves, Ornithurae) from the Lower Cretaceous of China. Zool. Scripta (2006) 35:363–373.[CrossRef][Web of Science]
Zwickl D. J., Hillis D. M. Increased taxon sampling greatly reduces phylogenetic error. Syst. Biol. (2002) 51:588–598.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  J. J. Wiens Paleontology, Genomics, and Combined-Data Phylogenetics: Can Molecular Data Improve Phylogeny Estimation for Fossil Taxa? Syst Biol, May 19, 2009; (2009) syp012v1. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. A. Wills, P. M. Barrett, and J. F. Heathcote The Modified Gap Excess Ratio (GER*) and the Stratigraphic Congruence of Dinosaur Phylogenies Syst Biol, December 1, 2008; 57(6): 891 - 904. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. A Jenner and D.T. J Littlewood Problematica old and new Phil Trans R Soc B, April 27, 2008; 363(1496): 1503 - 1512. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||