The disparity of priapulid, archaeopriapulid and palaeoscolecid worms in the light of new data


Correspondence: Matthew A. Wills, Department of Biology & Biochemistry, The University of Bath, 4 South, The Avenue, Claverton Down, Bath BA2 7AY, UK.

Tel.: 01225 383504; fax: 01225 386779; e-mail:


Priapulids and their extinct relatives, the archaeopriapulids and palaeoscolecids, are vermiform, carnivorous ecdysozoans with an armoured, extensible proboscis. These worms were an important component of marine communities during the Palaeozoic, but were especially abundant and diverse in the Cambrian. Today, they comprise just seven genera in four families. Priapulids were among the first groups used to test hypotheses concerning the morphological disparity of Cambrian fossils relative to the extant fauna. A previous study sampled at the generic level, concluding that Cambrian genera embodied marginally less morphological diversity than their extant counterparts. Here, we sample predominantly at the species level and include numerous fossils and some extant forms described in the last fifteen years. Empirical morphospaces for priapulids, archaeopriapulids and palaeoscolecids are relatively insensitive to changes in the taxon or character sample: their overall form has altered little, despite the markedly improved sampling. Cambrian and post-Cambrian genera occupy adjacent rather than broadly overlapping regions of these spaces, and Cambrian species still show lower morphological disparity than their post-Cambrian counterparts. Crucially, the significance of this difference has increased with improved taxon sampling over research time. In contrast with empirical morphospaces, the phylogeny of priapulids, archaeopriapulids and palaeoscolecids derived from morphological characters is extremely sensitive to details of taxon sampling and the manner in which characters are weighted. However, the extant Priapulidae and Halicryptidae invariably resolve as sister families, with this entire clade subsequently being sister group to the Maccabeidae. In our most inclusive trees, the extant Tubiluchidae are separated from these other living taxa by a number of small, intervening fossil clades.


As clades evolve, there are usually changes in the number and identity of their constituent lineages (diversity) and in the variety of morphologies realized by those lineages (disparity) (Neige, 2003). Diversity change is commonly investigated by determining species richness or its proxies (e.g. numbers of supraspecific taxa) and by calculating rates of origination, extinction and turnover. Morphological disparity is less routinely considered and is quantified using a variety of indices of morphological difference or distinctness (Wills, 1998a). Plots of disparity through time can be especially helpful for illuminating the macroevolutionary dynamics (rates and modes of evolution) of clades (Wills et al., 1994; Foote, 1997a, 1999; Wills, 1998a; Wills & Fortey, 2000; Erwin, 2007; Brusatte et al., 2008; Ruta, 2009), particularly when combined with the corresponding diversity profiles. Disparity and diversity are not necessarily correlated (Wills, 2001). At one extreme, it is perfectly possible for a large number of morphologically conservative and similar species to have low disparity, whereas at the other extreme, a small number of species can exemplify a large proportion of the major ‘design’ options explored throughout the entire history of a clade (Foote, 1997b). Studies of diversity and disparity have been important tools for exploring major diversifications and radiations, most notably in debates concerning the magnitude of the Cambrian explosion (Briggs et al., 1992; Gould, 1993; Wills et al., 1994; Fortey et al., 1996; Monge-Najera & Hou, 2000; Valentine & Jablonski, 2003; Briggs & Fortey, 2005). Two ecdysozoan groups (animals that shed their cuticular exoskeleton) are most often cited in this latter context: arthropods (Briggs et al., 1992; Wills et al., 1994; Waggoner, 1996) and priapulid worms (Wills, 1998b). Both are reported to have comparable morphological disparity in the Cambrian and the Recent. This is important for our understanding of the manner in which metazoans radiated, because it implies that Cambrian animals had already explored a variety of ‘design’ options similar to that realized by their present-day counterparts. This means that the apparent morphological eccentricity of many Cambrian fossils is more than simply a function of their relative unfamiliarity (Wills & Fortey, 2000). It also means that Cambrian fossils cannot be dismissed as mere stem group plesions: early organisms of little morphological distinctiveness and merely en route to their more derived and differentiated crown-group counterparts (Briggs & Fortey, 2005). This challenges the traditional ‘cone of increasing diversity’ (or, more precisely, it shifts this evolutionary model back in time) in favour of an approximately cylindrical model of bodyplan diversity from the Cambrian to the Recent (Wills & Fortey, 2000). It also implies that the magnitude of Cambrian morphological diversity requires some form of explanation: whether in terms of Pre-Cambrian evolution at a small size (Fortey et al., 1996), the gradual Pre-Cambrian differentiation of internal bodyplans decoupled from the appearance of external (and fossilizable) characters in the Cambrian (Parker, 2004), or some other mechanism. Whereas early molecular clock estimates for the period of Pre-Cambrian clandestine cladogenesis were half a billion years or more (Wray et al., 1996), with the majority in the range of 100–400 My (Levinton, 2008), recent relaxed clock calibrations may narrow the gap even more (Erwin et al., 2012). Finally, estimates of Cambrian disparity for arthropods and priapulids challenged the more radical ‘inverted cone’ model of Gould (1989, 1993), who argued that levels of bodyplan variety were much greater in the Cambrian than at any time since.

Despite their similarities, findings for arthropods and priapulids differ in one important respect. Whereas Cambrian and Recent arthropods are typified as occupying broadly overlapping regions of anatomical ‘design space’ (or morphospace), priapulids and their allies occur in adjacent regions and appear to have extemporized on different ‘design’ themes (Wills, 2001). In the case of priapulids, these separate regions of morphospace may correspond to two distinct clades.

In the time since the disparity of Cambrian and Recent priapulids and their allies was first investigated (Wills, 1998b), a number of important new fossils have been described (Dong et al., 2005; Han et al., 2006, 2007c; Huang et al., 2006a; Zhang et al., 2006; Maas et al., 2007; Conway Morris & Peel, 2010; Topper et al., 2010), and the phylogeny of the group has been reappraised on at least two occasions (Dong et al., 2005; Harvey et al., 2010). Does this considerable expansion of our knowledge influence the conclusions of the original paper (Wills, 1998b)? We therefore ask to what extent the findings of the initial study are sensitive to factors such as improvements in taxon sampling, the discovery of new fossils through research time and differences in the characters coded. Several additional analyses of the original data matrix are compared with those of the significantly expanded matrix of Dong et al. (2005), in addition to a new data set compiled here for all species known in sufficient detail.

It is well known that subtle changes in taxon and character sampling can yield radically different phylogenetic trees under all models of character evolution (Cobbett et al., 2007; Carlson & Fitzgerald, 2008; Heath et al., 2008; Li et al., 2008; O'Leary & Gatesy, 2008; Panero & Funk, 2008). The level of sensitivity is, among other things, a function of the amount of homoplasy in the data. Amounts of homoplasy are, in turn, influenced by the intensity of taxon sampling (although this relationship is not a straightforward one). Does this sensitivity also apply to discrete character-based morphospace studies, given that some indices of morphological disparity are strongly influenced by sample size differences? If so, to what extent are conclusions about the manner in which clades explore design possibilities through time influenced by the precise sampling regimes of particular investigators?

Priapulids, archaeopriapulids and palaeoscolecids

Priapulids are members of the Scalidophora, a group that also contains the Kinorhyncha and Loricifera. Together with Nematoida (Nematoda plus Nematomorpha), they constitute the Cycloneuralia, the probable sister group of arthropods (Telford et al., 2008). However, relationships within this portion of the metazoan tree are not entirely settled (Peel, 2010), with some authors placing Tardigrada as the sister group of Nematoida (Dunn et al., 2008; Pick et al., 2010). The bodies of extant priapulids consist of an anterior proboscis and a posterior cylindrical trunk (with or without caudal appendages) (Baltzer, 1931; Adrianov & Malakhov, 2001; Adrianov & Malakhov, 1995). The proboscis can be extended out of (or invaginated back into) the trunk, and a protrusible, cone-like pharynx is equipped with concentric rings of spines. Cambrian archaeopriapulids have a similar bodyplan, but all lack a caudal appendage. The precise relationships between extant priapulids and their fossil archaeopriapulid relatives remain unclear. Whereas extant species (plus the Carboniferous Priapulites (Schram, 1973; Huang et al., 2004b)) almost certainly constitute a clade (Wills, 1998b; Conway Morris & Peel, 2010; Harvey et al., 2010), the Cambrian archaeopriapulids almost certainly do not. The picture is further complicated by the palaeoscolecids (Whittard, 1953; Mueller & Hinz-Schallreuter, 1993; Conway Morris, 1997; Han et al., 2007b,2007d; Conway Morris & Peel, 2010), another group of Cambrian to latest Silurian cycloneuralian ecdysozoan worms with a bodyplan ostensibly similar to that of the priapulids (Conway Morris & Peel, 2010). Palaeoscolecids probably constitute a paraphyletic group, as there are few (if any) characters common to all species (Harvey et al., 2010). Their armoured proboscis is shared with all priapulids and archaeopriapulids, whereas their annulated trunks are also present in many other genera. Palaeoscolecids are unique in having circular patterns of tessellating phosphatic plates down their trunks (Ivantsov & Wrona, 2004). Not all putative palaeoscolecid characters are shared by all palaeoscolecid species, however. This may often be the result of poor preservation: indeed assemblages of phosphatic plates are frequently found in the absence of other, more volatile soft tissues (Conway Morris & Peel, 2010). Some authors propose a close relationship between palaeoscolecids and nematomorph worms (Budd, 2001; Maas et al., 2007; Sorensen et al., 2008), although this may result from limited sampling of the priapulid stem (Harvey et al., 2010). An alternative hypothesis – namely that palaeoscolecids resolve close to the base of the Ecdysozoa – is not supported by numerical cladistic analyses of large-scale character matrices. Rather, it derives from particular hypotheses about the nature of the ancestral ecdysozoan (Budd & Jensen, 2000, 2003; Budd, 2001, 2003; Zrzavy, 2003; Webster et al., 2006) and the putatively symplesiomorphic nature of the characters that priapulids and palaeoscolecids share (Conway Morris & Peel, 2010). In the analyses of Wills (1998b), Dong et al. (2005) and Harvey et al. (2010), palaeoscolecids, priapulids and archaeopriapulids have usually been resolved as a clade (Cycloneuralia) to the exclusion of other ecdysozoan groups. We do not force this assumption here. Irrespective of the debate over monophyly, a unified empirical morphospace for priapulids, archaeopriapulids and palaeoscolecids is defensible inasmuch as they constitute a grade of organization, share many similarities, are efficiently described by a common set of characters and may also represent an ecological unit.

Materials and methods

The data set

Our list of characters (Appendix S1) and data matrix (Appendix S2) were synthesized primarily from Conway Morris (1977), Wills (1998b), Lemburg (1999), Dong et al. (2004, 2005) and Harvey et al. (2010), and augmented with additional data from the primary literature. We coded all those extant and fossil species known in enough detail to permit suitable comparisons across our data matrix. As with many analyses that include fossil taxa, the use of missing entries and ambiguous codings was unavoidable. These were often concentrated in characters pertaining to details of the introvert. Where this was severe, taxa were necessarily omitted. For example, we did not include codings for the palaeoscolecids Sahascolex (Ivantsov & Wrona, 2004), Protoscolex (Conway Morris et al., 1982), Gamascolex and Plasmuscolex (Kraft & Mergl, 1986). Nor did we attempt to code the five diagnosed species of Markuelia (Dong et al., 2010; Cheng et al., 2011) separately, but rather followed Dong et al. (2010) in using a composite coding for this genus. Harvey et al. (2010) used a composite coding for palaeoscolecids sensu stricto, defined as those taxa unambiguously preserving trunk ornamentation comprising polymorphic, tessellating plates (i.e. Palaeoscolex piscatorum, P. sinensis, Sahascolex, Protoscolex, Gamascolex, Plasmuscolex, Hadimopanella, Milaculum and Kaimenella). This is entirely legitimate, but we note that the incomplete and fragmentary nature of many of these specimens means that the compositely coded ‘Palaeoscolecida s. s.’ differs very little from the generic Palaeoscolex coded by Dong et al. (2005). Moreover, the suite of characters utilized by Harvey et al. (2010), Dong et al. (2010) and herein do not code the sort of fine structural details that would differentiate between all or many genera. Hence, coding Palaeoscolecida s. s., Palaeoscolex or even Palaeoscolex piscatorum all resulted in similar suites of character scores in practice.

Our matrix consisted of 45 putative priapulid, archaeopriapulid and palaeoscolecid species plus 10 outgroup taxa coded for 88 characters. Whereas the matrix of Wills (1998b) coded autapomorphic characters explicitly, the matrices of Dong et al. (2005) and Harvey et al. (2010) did not. In order to make the results from all of our analyses comparable in this regard, autapomorphic characters and unique states were removed prior to all of our (disparity) analyses. Secondly, the increased taxon sampling and larger outgroups used here and in the study by Dong et al. (2005) unsurprisingly influenced the inferred phylogeny. In all of our trees, and in most of those of Dong et al. (2005), the Kinorhyncha and Loricifera resolved in a more derived position than at least Ancalagon, Fieldia and Markuelia (Dong, 2007). However, as noted by Cobbett et al. (2007) and Dong et al. (2010), this arrangement was extremely labile in single taxon deletion experiments. We note that Harvey et al. (2010) placed Ancalagon (Walcott, 1911b; Conway Morris, 1977; Adrianov & Malakhov, 1995) and Fieldia (Conway Morris, 1977; Forney et al., 1977) higher in their trees, whereas the inclusion of data on Markuelia (Dong et al., 2010) returned it to a position below Kinorhyncha and Loricifera.

Phylogenetic analyses

For the data sets of Wills (1998b) and Dong et al. (2005) we implemented relatively straightforward parsimony searches in PAUP* version 4.0b10 (Swofford, 2002) in order to replicate the original results of these authors. These consisted of 1,000 random additions of terminals, followed by tree bisection–reconnection (TBR) branch swapping. We then condensed the resulting set of most parsimonious trees (MPTs) by collapsing all branches with a minimum length of zero (the ‘amb-’ option: Goloboff et al., 2000) and eliminating duplicate trees. Wills (1998b) experimented with the ordering of a small number of characters, whereas Dong et al. (2005) presented trees inferred without Ancalagon, Fieldia or both. Differences in each case were not great. Here, we included minimal assumptions and compared the trees produced with all characters treated as unordered and (in the first instance) with the most inclusive set of taxa. In analysing the data from Wills (1998b), Kinorhyncha and Loricifera were designated as the outgroup. In analysing the matrix from Dong et al. (2005) and the data presented herein, trees were rooted using the Gastrotricha. For the analyses of both the Wills (1998b) and Dong et al.'s (2005) data sets, there appeared to be only a single island of MPTs. Bremer support values were calculated using TreeRot v.3 (Sorenson & Franzosa, 2007) in conjunction with PAUP*, and using the same search parameters as above. In addition, for these two data sets, we investigated the effects of first-order jack-knifing each terminal on the pattern of inferred relationships among the remaining terminals (Cobbett et al., 2007). This necessitated a series of new parsimony searches (as above) omitting each terminal in turn. For each jack-knife, strict and 50% majority rule consensus trees were calculated from the resulting MPTs (the ‘searched’ route, illustrating the effects of excluding that taxon). In order to quantify the effects of each terminal upon inferred relationships, these consensus trees were compared with consensus trees from the original set of MPTs (i.e. those obtained including all taxa in the parsimony searches). However, because each jack-knifed consensus tree contained one fewer terminal than the trees from the full data set, the corresponding terminal was pruned from all the latter MPTs before removing duplicate trees and computing the consensus (‘pruned’ route). The difference in inferred relationships between each pair of consensus trees (searched and pruned) was calculated in terms of both the symmetric difference distance (Robinson & Foulds, 1981) and the maximum agreement subtree distance (Finden & Gordon, 1985). The entire process was automated in PAUP* with a script written using the program DelBat (available from MAW), which also summarized the PAUP* output into a spreadsheet.

For the new data set presented herein, straightforward parsimony searches as outlined above were extremely slow, became stuck in local optima and located many islands of trees. Our initial PAUP* searches incorporated the parsimony ratchet, implemented using a batch file produced by PRAP (Müller, 2004). We used 1000 reweightings at twice the base weight for 20% of the characters in our data set (i.e. 18 from 88). Each reweighted data set was then subjected to 50 random additions with TBR branch swapping, retaining only representatives of the sets of shortest trees. This procedure was effective at finding the MPT length for the unweighted data, but not necessarily at finding all MPTs. We subsequently reran the flat-weighted data, but retained no more than 1000 trees longer than the minimum established above. Once this minimum length was achieved, all trees were retained. We ran 1000 random additions, followed by TBR swaps. We also explored the effects of successively reweighting according to the maximum value of the rescaled consistency index found across this initial set of trees, and of implied weighting with = 3 (in both cases utilizing conventional heuristic searches: 1000 random additions, followed by TBR swaps and condensing the resulting trees using ‘amb-’) (Goloboff et al., 2000). Finally, we used the program Ghosts (Wills, 1999; Wills et al., 2008) to calculate stratigraphic congruence for each of the MPTs obtained from the flat-weighted analysis. We then computed a consensus including only those trees with the maximal gap excess ratio (GER) (Wills, 1999; Wills et al., 2008). This uses stratigraphic congruence as an ancillary criterion to filter the MPTs, rather than to influence relationships a priori. Because most of the species considered here first appear either in the Cambrian or in the Recent, we utilized a vastly simplified stratigraphy (Yunnan, Sirius Passet, St David's, Pennsylvanian, Early Cretaceous, Late Cretaceous, Recent). This avoids excessively heavy penalties for sister group pairs from the Cambrian and Recent. Our data file is included with the electronic version of the paper (Appendix S3).

Disparity analyses

The methods of disparity analysis differed somewhat from those used by Wills (1998b). Firstly, a matrix of generalized Euclidean distances (GED) was derived between all taxa using the program Matrix (Wills, 1998a). All characters were treated as unordered. Secondly, this triangular matrix was ordinated in Ginkgo (Bouxin, 2005) using principal coordinates analysis and Cailliez's method for negative eigenvalue correction (Cailliez, 1983). Minimum spanning trees (MSTs) were calculated using the algorithm Spantree in the ‘R’ language and environment for statistical computing and graphics (available at Coordinate scores for all taxa were extracted from the ordinations, on sufficient axes to capture 90% of the total variance (as indicated by the cumulative eigenvalues). The number of axes required differed for the three data sets. The coordinate scores were then used to calculate the disparity of Cambrian and post-Cambrian taxa. Two indices were used: sum of ranges and sum of variances (Wills et al., 1994). Range-based measures are expected to be more sensitive to differences in sample sizes and the positions of outliers than those derived from variances. In order to control for the effects of sample size and to test for differences in the disparity of the Cambrian and post-Cambrian samples, measures were rarefied at all sample sizes up to and including the number of taxa. The program Rare (Wills, 1998a) was used to bootstrap 1,000 samples and to calculate the upper and lower limits (bound) of the 90% confidence interval. After Foote (1992), the two groups were considered to have significantly different disparity where the mean value for the smaller group lay outside the 90% confidence interval for the larger one.

Spatial clustering

In order to visualize regional differences in the density of taxa on the first two axes of our character morphospaces, we used two-dimensional point pattern analysis in Spatstat v. 1.21-2 (Baddeley & Turner, 2005), an open source library in ‘R’. We first estimated the spatial bounds of the point distribution with the Ripley–Rasson algorithm (Ripley & Rasson, 1977). We then produced a kernel density estimate of the probability distributions of taxa within this window, which are illustrated using bivariate contour density plots.

In order to quantify the morphospace clustering of Cambrian and post-Cambrian taxa in all dimensions, we implemented the nearest-neighbour analysis (NNA) described in the study by Foote (Foote, 1990). This approach computes the shortest distance, di, between a given specimen, i, and all others, and compares it with the expected shortest distance, ri, under a null model of random, uniform distribution with the same constraints as the empirical distribution (i.e. the same number of neighbours and the same spatial bounds). From these, we computed pi, that is, the proportional difference between observed and expected distances (pi = (d− ri)/ri). The mean pi value math formula is the mean clustering intensity. If math formula , then points are clustered, whereas if math formula , then points are negatively contagious. If math formula is indistinguishable from 0, then the spatial distribution cannot be distinguished from a random one. Clustering intensity has been computed for Cambrian species, post-Cambrian species and all species considered together.

In addition to the above, we used a principal point–based method (Flury, 1990, 1993; Klingenberg & Froese, 1991) to describe the clustering of Cambrian and post-Cambrian species. The k principal points of a clustered multivariate distribution are the set of points minimizing the expected Euclidean distance between each observation and its nearest principal point. The sample mean squared deviation (SMSD: used as a measure of the performance of the k-mean clustering for a given set of k principal points) is the average squared Euclidean distance of each observation to its nearest principal point. The SMSD value of the empirical distribution is computed for k ranging from 1 to N, where N is the number of observations. Thus, SMSD can range from the variance of the sample (multiplied by (− 1)/N) down to zero. Scaling the SMSD values between zero and one allows distributions with different overall levels of disparity to be compared. The pattern of SMSD decay (as k ranges from 1 to N) is compared with the pattern obtained under a null model of multivariate uniform distribution constrained to have the same spatial bounds as the empirical distribution investigated (20 replications were made to ensure the stability of estimates).


Phylogenetic analyses

Seven MPTs with an ensemble consistency index (CI) of 0.598 and ensemble retention index (RI) of 0.672 were obtained from the data of Wills (1998b) (Fig. 1). Ancalagon and Fieldia resolved either as a clade or a paraphyletic series immediately below (or as sister group to) all other fossil and extant priapulids. Two other large clades emerged from all the MPTs: one comprising all remaining Cambrian taxa (Louisella (Walcott, 1911a; Conway Morris, 1977), Scolecofurca (Conway Morris, 1977), Ottoia (Walcott, 1911b; Conway Morris, 1977; Conway Morris & Robison, 1986), Cricocosmia (Han et al., 2007a,2007b), Palaeoscolex (Whittard, 1953; Conway Morris & Robison, 1986; Conway Morris, 1997), Maotianshania (Sun & Hou, 1987; Zhang et al., 2006; Maas et al., 2007) and Selkirkia (Melville, 1985; Conway Morris & Robison, 1986; Zhang et al., 2006)) and the other comprising all extant taxa plus the Carboniferous Priapulites (Schram, 1973), but with the possible exception of the Tubiluchidae (Tubiluchus and Meiopriapulus). The Tubiluchidae were variously placed as sister taxon to either the ‘extant’ clade or the Cambrian clade. Taxon jack-knifing experiments revealed several terminals with considerable impact, notably Scolecofurca, Maccabeus, Tubiluchus and Palaeoscolex. As such, this tree is highly labile with small perturbations of the taxon sample.

Figure 1.

Majority rule consensus of seven most parsimonious trees (MPTs) (CI' = 0.598, RI = 0.672, 119 steps) resulting from the parsimony analysis of the data in the study by Wills (1998b), with all characters treated as unordered. Daggers (†) denote fossil taxa. Internal branches marked with heavy circles are also present in the strict consensus, and numbers within these indicate Bremer support. Numbers in light circles indicate the percentage of fundamental trees supporting a given branch (100% for those with Bremer support values). Histograms to the right of the cladogram illustrate the impact of removing individual taxa upon the inferred relationships of those remaining (Cobbett et al., 2007). Sets of MPTs from each run are summarized as both the strict and the majority rule consensus, whereas the distances between trees are calculated as both the symmetrical difference distance (RF) (Robinson & Foulds, 1981) and the maximum agreement subtree distance (d1) (Finden & Gordon, 1985). The inclusion/exclusion of Scolecofurca, Louisella and Maccabeus has the greatest overall impact upon inferred relationships.

Dong et al. (2005) also resolved Ancalagon and Fieldia low in their topology, but additionally placed the then newly described genus, Markuelia, in this position. These were followed by a pectinate succession of the ‘outgroup’ taxa Kinorhyncha and Loricifera. As in the study by Wills (1998b), there were two major clades above this. The first comprised all extant taxa (invariably inclusive of the Tubiluchidae) plus Priapulites, whereas the second comprised all Cambrian taxa (many new to this analysis). In our reanalysis (Fig. 2), we recovered two MPTs with a CI of 0.561 and RI of 0.735. First-order taxon jack-knifing revealed a number of strongly influential taxa, including Sicyophorus, Yunnanpriapulus, Xiaoheiqingella (Han et al., 2004; Huang et al., 2004b), Ottoia, Cricocosmia, Kerygmachela and Maccabeus. The taxa with the highest impact were therefore predominantly fossils.

Figure 2.

Strict consensus of two most parsimonious trees (MPTs) (CI' = 0.561, RI = 0.735, 195 steps) resulting from the parsimony analysis of the data in the study by Dong et al. (2005), with all characters treated as unordered. Daggers (†) denote fossil taxa. Numbers within circles indicate Bremer support. Histograms to the right of the cladogram illustrate the impact of removing individual taxa upon the inferred relationships of those remaining (Cobbett et al., 2007). Sets of MPTs from each run are summarized as both the strict and the majority rule consensus, whereas the distances between trees are calculated as both the symmetrical difference distance (RF) (Robinson & Foulds, 1981) and the maximum agreement subtree distance (d1) (Finden & Gordon, 1985). The taxa with the greatest impact therefore differ from those highlighted in the analysis of Wills (1998b)).

The data set compiled in this paper yielded 416 distinct MPTs (after removing duplicates and collapsing branches with a minimum length of zero: Goloboff's ‘amb-’), with a modified ensemble consistency index (CI') of 0.504 and retention index (RI) of 0.797. A majority rule consensus (plus compatible groupings) is illustrated (Fig. 3a), although a few of the clades recovered were only present in a substantial minority of the fundamentals. Successive approximations reweighting converged on stable weights and the same 8 distinct MPTs after reweighting twice. The strict consensus of these trees is illustrated (Fig. 3b). Implied weighting with = 3 yielded 6 distinct MPTs, all different from those obtained by successive approximations (strict consensus as Fig. 3c). Finally, when the 416 trees from the flat-weighted analyses were filtered by their stratigraphic congruence, 32 (8%) were retained with the maximum GER (0.623). These are summarized as a majority rule consensus (Fig. 3d), which differs only slightly from the consensus for all 416 MPTs (Fig. 3a).

Figure 3.

Trees resulting from the analysis of our more inclusive data set, with all characters treated as unordered. Daggers (†) denote fossil taxa. (a) Flat-weighted (‘unweighted’) analysis yielded 416 distinct most parsimonious trees (MPTs) (CI'  =  0.504, RI = 0.797) after removing duplicates and collapsing branches with a minimum length of zero (Goloboff's ‘amb-’). A majority rule consensus (plus compatible groupings) is illustrated. Numbers on internal branches indicate the percentage of the fundamentals containing each clade. Unnumbered branches were present in all fundamentals (100% support). (b) Strict consensus of eight trees resulting from two consecutive rounds (converged) of successive approximations reweighting (using the maximum value of the rescaled consistency index across all trees). (c) Strict consensus of six trees resulting from implied weights analysis (= 3). (d) Majority rule consensus (plus compatible groupings) of the 32 flat-weighted trees from ‘a’ with the maximum stratigraphic congruence (gap excess ratio of 0.623) (Wills, 1999).

The strict consensus of all 416 trees from the flat-weighted analysis (not shown, but implicit in Fig. 3a) shared several features with the cladograms of Wills (1998b), Dong et al. (2005, 2010) and Harvey et al. (2010). Above Gastrotricha was a polytomy comprising the lobopodians (namely Peripatus and two small clades of Tardigrada + Microdictyon and Aysheaia + Kerygmachela) and a clade of all other taxa. In the strict consensus, the next branch up supported a small clade comprising Nematoida (Nematomorpha and Nematoda) plus Palaeoscolex piscatorum. At the crown of the tree were the extant Tubiluchidae as sister group to the fossils Selkirkia and Paraselkirkia, and a clade of Priapulidae (Priapulus, Priapulopsis, Acanthopriapulus and the extinct Priapulites) as sister group to the Halicryptidae, followed by Maccabeidae. Between the basal plesions and the crown, however, there were significant differences in inferred relationships; notably in this new flat-weighted analysis, the Kinorhyncha and Loricifera resolved in a clade containing several archaeopriapulids (Palaeopriapulites, Paratubiluchus, Sicyophorus, Ancalagon and Fieldia) rather than resolving close to the root of the tree.

Unsurprisingly, there was a great deal more resolution in the majority rule tree (Fig. 3a), but many clades also appeared in the strict consensus of the trees from the successive reweighting (Fig. 3b), as well as in that from the implied weights analysis (Fig. 3c). ‘Above’ the Nematoida plus Palaeoscolex piscatorum were two plesions, Xystoscolex and Markuelia. The position of the latter close to the root of the tree was consistent with its initial description as a basal form by Dong et al. (2005), Dong (2007) and in some of the trees of Dong et al. (2010). Agreement between the variously weighted trees (Fig. 3a-c) was greatest for post-Cambrian taxa. In all three (and in Fig. 3d), relationships within the clade of ((Priapulidae, Halicryptidae), Maccabeidae) were highly similar. Acanthopriapulus horridus (Van Der Land, 1970; Amor, 1975) resolved within the genus Priapulus (Menzies, 1959; Murina & Starobogatov, 1961; Shapeero, 1962; Rauschert, 1986; Lemburg, 1999; Webster et al., 2007; Wennberg et al., 2009), which was sister group to a clade or paraphylum of Priapulopsis (Von Salvini-Plawen, 1973; Storch et al., 1995). The Carboniferous plesion Priapulites konecniorum (Schram, 1973; Huang et al., 2004b) either constituted the sister group to ((Priapulus Acanthopriapulus), Priapulopsis), or resolved within a paraphyletic Priapulopsis (the successively reweighted tree; Fig 3b). Clades of the genera Halicryptus (Lemburg, 1995a,b; Shirley & Storch, 1999; Janssen et al., 2009) and Maccabeus (Por & Bromley, 1974; Por, 1980) followed as successive sister groups.

All of our analyses retrieved the Tubiluchidae as a clade. This comprised all species of Tubiluchus (Calloway, 1975; Kirsteuer, 1976; Van Der Land, 1982, 1985; Adrianov et al., 1989; Adrianov & Malakhov, 1991; Todaro & Shirley, 2003; Rothe et al., 2006) and Meiopriapulus (Morse, 1981; Storch et al., 1989; Paulay & Holthuis, 1994) (i.e. Tubiluchidae). All trees also resolved Meiopriapulus as sister taxon to Tubiluchus. In the flat- and successively weighted consensus trees, a small fossil clade of Selkirkia and Paraselkirkia (Zhang et al., 2006) constituted the sister group to Tubiluchidae, whereas these same fossils emerged as sister group to the clade ((Priapulidae, Halicryptidae), Maccabeidae) in the implied weights consensus. In this latter tree, it was Paratubiluchus that resolved as sister group to Tubiluchidae, a position consistent with the affinities postulated by Han et al. (2004). Paratubiluchus was also the Cambrian taxon patristically and phenetically closest to living species, consistent with its apparent mixture of extant and Cambrian characters (Han et al., 2004).

In all of our trees, the extant priapulids (plus Priapulites, Selkirkia and Paraselkirkia) were either sister group to a small plesion of Xiaoheiqingella plus Yunnanpriapulus (a result consistent with the possible synonymy of these last two genera; Han et al. (2004)) (Fig. 3a,b,d) or formed a polytomy with them (Fig. 3c). The greatest differences between the trees from the different weighting schemes occurred between this ‘crown group’ of predominantly extant priapulids and the root of the tree. The Cambrian species herein invariably resolved as a grade; a succession of clades in a pectinate series along the stem. Many of the group frequencies in this region of the majority rule trees were low, and numerous alternative resolutions were possible. Nonetheless, there are a few points of agreement between the three weighting schemes. Anningvermis (Huang et al., 2004a), Corynetis (Huang et al., 2004a; Ma et al., 2010) and Louisella invariably emerged as a clade (our resolution of Anningvermis multispinosus as sister species to Corynetis brevis is consistent with the proposal that the two are synonymous (Ma et al., 2010)). These were succeeded by Ottoia (Walcott, 1911b; Conway Morris, 1977; Conway Morris & Robison, 1986) and Scolecofurca (Conway Morris, 1977) in variable order. In the two weighted analyses (Fig. 3b,c), this clade was followed by the palaeoscolecid Chalazoscolex (Conway Morris & Peel, 2010). These were invariably proximate to the fossils Maotianshania and Guanduscolex (Hu et al., 2008). Another clade recovered from all treatments of the data was Cricocosmia plus both species of Tabelliscolex (Han et al., 2003a, 2007b). Tylotites (Han et al., 2003b, 2007d) either resolved as sister to this group or to this group plus other fossil genera.

The consensus tree resulting from the implied weights analysis was the most similar to numerical analyses published hitherto (Wills, 1998b; Dong et al., 2005, 2010; Harvey et al., 2010). It also resolved the presumed outgroup taxa Kinorhyncha and Loricifera closer to the root than the trees obtained using other optimality criteria (only Ancalagon, Fieldia, Markuelia, Xystoscolex and Palaeoscolex piscatorum were more basal). To the extent that we express a preference for a phylogeny, then it is for this implied weights tree (Fig. 3c). This also constituted the rationale for the inclusion/exclusion of taxa in our morphospace analyses; all taxa variously resolved crownward of Kinorhyncha and Loricifera were included. The implied weights tree resolved Laojieella thecata (Han et al., 2006, 2007c) and (Palaeopriapulites Sicyophorus) as successive plesions to the crown group plus Xiaoheiqingella and Yunnanpriapulus. These were followed by a relatively large clade of archaeopriapulids and putative palaeoscolecids, including Anningvermis, Corynetis, Louisella, Scolecofurca, Ottoia, Chalazoscolex, Cricocosmia, Tabelliscolex and Tylotites. Maotianshania and Guanduscolex resolved successively at the base of this clade.

Empirical morphospace and disparity analyses

Despite obvious differences in the details of the three PCoA morphospaces (Figs 4-6), there were considerable similarities in the overall distributions of Cambrian and post-Cambrian taxa on the first two coordinate axes. Importantly, the two groups occupied adjacent rather than broadly overlapping regions of the spaces. However, whereas these groups were minimally linked by the MST in Wills' (1998b) morphospace (Fieldia to Priapulites), there were two links (Palaeopriapulites to both Priapulites and to Meiopriapulus) in the morphospace derived from Dong et al. (2005). In our more intensively sampled morphospace, the course of the MST was more complex still. The overall orientation of points in Fig. 6 is also rotated (and reflected) relative to that in Figs 4 and 5. Specifically, whereas in the two older ordinations, the second coordinate axis served largely to distinguish the Tubiluchidae from all other genera, it was the first coordinate axis that fulfilled this function in our new analysis.

Figure 4.

Character morphospace for the data of Wills (1998b), excluding autapomorphies. Methods described in the text. Taxa are plotted with respect to scores on the first two principal coordinates, with scores on the third coordinate indicated by size and shading of symbols. Points are joined by a minimally spanning tree, calculated from the original distance matrix (all dimensions). Cambrian taxa labelled in italics; post-Cambrian taxa in normal font.

Figure 5.

Character morphospace for the data of Dong et al. (2005), excluding autapomorphies. Methods described in the text. Taxa are plotted with respect to scores on the first two principal coordinates, with scores on the third coordinate indicated by size and shading of symbols. Points are joined by a minimally spanning tree, calculated from the original distance matrix (all dimensions). Cambrian taxa labelled in italics; post-Cambrian taxa in normal font.

Figure 6.

Character morphospace for the data presented herein, excluding autapomorphies. Methods described in the text. Taxa are plotted with respect to scores on the first two principal coordinates, with scores on the third coordinate indicated by size and shading of symbols. Points are joined by a minimally spanning tree, calculated from the original distance matrix (all dimensions). Cambrian taxa labelled in italics; post-Cambrian taxa in normal font.

For all three of the data sets analysed, indices of disparity based on mean pairwise Manhattan and Euclidean distances were greater for post-Cambrian taxa than their Cambrian counterparts (Table 1). Indices of disparity derived from our ordinations (sums of ranges and variances) were based on sufficient coordinates to capture 90% of the variance within each data set (11, 12 and 28 coordinates for Wills (1998b), Dong et al. (2005) and the data presented here, respectively). For both of these indices over all sample sizes, median and mean post-Cambrian disparity indices were always greater than median and mean Cambrian disparity indices (Table 1). Differences were only deemed significant, however, when the median bootstrapped index for the smaller (post-Cambrian in the case of Wills (1998b) and Dong et al. (2005)) sample lay outside the 90% confidence interval for the larger (Cambrian) sample (Fig. 7). For the Wills (1998b) data set, this only occurred for the sum of variances and then only at sample sizes above four. Similarly, the data of Dong et al. (2005) revealed significant differences only for the sum of variances at sample sizes above two. By contrast, there were significant differences in both metrics for the new and larger data set presented here. Sums of ranges were significant above samples of 14, whereas sums of variances were significant above samples of six.

Figure 7.

The effects of taxonomic rarefaction upon two indices of disparity for the three data sets discussed in this paper. Top row shows sum of ranges; bottom row shows sum of variances. Left-hand panels illustrate data from Wills (1998b) using the first 9 principal coordinates. Middle panels illustrate data from Dong et al. (2005) using the first 11 principal coordinates. Right-hand panels illustrate data from the present paper using the first 28 principal coordinates. Solid lines denote sample medians based on 1,000 bootstrap resamplings; broken lines denote estimates of upper and lower 5% confidence intervals. Filled (black) circles indicate Cambrian taxa; open circles indicate post-Cambrian taxa. See text for details.

Table 1. Indices of disparity for Cambrian and post-Cambrian priapulids, archaeopriapulids and palaeoscolecids
Data setPartitionNumber of taxaMean pairwise Eulidean distanceMean pairwise Manhattan distanceHypervolume measures
Measured at maximum sample sizeMedian bootstrapped to smaller sample size
Sum of rangesSum of variancesSum of rangesSum of variances
Wills (1998ab)Cambrian93.0110.7929.669.7124.158.72



Dong et al. (2005)Cambrian122.9511.4140.0513.9632.2513.07



Data hereinCambrian203.0413.06101.5130.6590.0338.16




The intensity of clustering for Cambrian and post-Cambrian taxa on the first two coordinate axes can be visualized in the contour density plots (Fig. 8). However, a quantitative assessment of clustering requires an analysis on all (or most) axes simultaneously. Results from the NNA are given in Table 2. The data sets of Wills (1998b) and Dong et al. (2005) had similar general properties: specifically, Cambrian and post-Cambrian taxa considered in isolation had negatively contagious distributions (i.e. were not clustered because math formula > 0), whereas all taxa considered simultaneously were positively contagious (i.e. were clustered with math formula < 0). However, for the data of Wills (1998b), the global clustering was not significant (P = 0.644), whereas for Dong et al. (2005), it was (P = 0.039). This reflects the increased taxon sampling of the latter data set, although statistical power was weakened by the low sample size/dimensionality ratios. Results from our new data set contrasted with those from both of its predecessors. Both Cambrian and post-Cambrian partitions in isolation, as well as the entire data set, showed positive contagion and clustering of points; math formula < 0 with significant clustering in the case of Cambrian taxa (P = 0.046) and all taxa together (P = 0.006). Hence, the sampling strategies that deliberately coded just one representative from each genus were destined to find those exemplars scattered widely through the empirical morphospaces in each time bin (Cambrian or post-Cambrian), almost by definition. The positive clustering of the global distributions therefore reflected the comparative separation of Cambrian and post-Cambrian genera (which formed more local clusters). By sampling predominantly at the level of species, our new data set revealed positive clustering in both time bins (significant for the Cambrian), not just in the entire sample.

Figure 8.

Bivariate contour density plots of taxa with respect to the first two principal coordinates. Coloured contours indicate the probability distributions of taxa (for purposes of visualization only). Cambrian taxa denoted using circles; post-Cambrian taxa using triangles. See text for details.

Table 2. Nearest neighbour analysis of priapulid morphospaces. Mean math formula is the mean clustering intensity, SE the standard error. P-value indicates the significance of a Mann–Whitney U test for differences from zero clustering intensity (random distribution)
Data setPartitionNumber of taxaNumber of dimensionsMean math formulaSEP-value








et al.












The principal point–based approach provided results in line with those from nearest-neighbour analyses (Fig. 9). This method investigated how efficiently a restricted set of k points could emulate the distribution of the empirical sample. If all of the points in the sample were superimposed, a single point (= 1) would efficiently summarize this. If the sample were to be distributed into three tight clusters, then three points would suffice (= 3), with little or no advantage conferred by additional points. For a spatially homogeneous distribution, small numbers of k will struggle to capture the structure of the empirical distribution, such that the representation will improve as k increases and approaches the empirical sample size. The efficacy of the reduced point pattern at each value of k (measured in terms of the SMSD) is compared with that simulated from a multivariate homogeneous distribution within the same spatial bounds. For the smallest data set (Wills, 1998b), both the entire sample of taxa and its partitions (Cambrian and post-Cambrian) were consistent with a homogenous point distribution at all numbers of principal points. For the data set of Dong et al. (2005), both the Cambrian and post-Cambrian subsamples were similarly consistent with a homogeneous distribution. However, the entire data set from these authors showed significant nonhomogeneity with reference to 11 or more principal points. The data presented herein showed an even more significantly clustered distribution, not only for all species considered together (significant above six principal points) but also for Cambrian and post-Cambrian subsamples (significant above 10 and 14 principal points, respectively). The distribution of all species was therefore significantly clustered over the wide range of spatial scales, whereas Cambrian and post-Cambrian species (considered independently) were consistent with a homogeneous distribution at the largest scale, but tightly clustered more locally.

Figure 9.

Sample mean squared deviation (SMSD) analysis of priapulid morphospaces. Open circles indicate the observed SMSD with an increasing number of principal points (see text). Solid line corresponds to the expected SMSD curve for a multivariate homogeneous distribution of the same number of points and the same spatial bounds as the observed distribution. Dashed lines are lower and upper bounds of the 95% confidence interval around this null expectation. Top row shows all taxa analysed together; middle and bottom rows show Cambrian and post-Cambrian taxa analysed separately.


We temper all of our phylogenetic conclusions with the observation that trees of priapulids and their relatives are notoriously labile (even with minimal perturbations of the data set), and subject to the translocation of branches at all levels (Dong et al., 2010). Indeed, priapulids were cited as a case study group exemplifying the frequently marked effects of removing single terminals in first-order jack-knifing experiments (Cobbett et al., 2007). Moreover, different weighting schemes also yielded different relationships, particularly within the fossil stem, and with regard to the group hypothesized to be most closely allied to extant crown clades. There are principally three possible reasons for this phenomenon. Firstly, poor preservation in many fossil scalidophorans introduces significant amounts of ‘missing data’. Depending upon its distribution, this can result in taxa that can be resolved equally parsimoniously in many locations throughout the tree (i.e. with low leaf stability). Secondly, Dong et al. (2010) emphasized the limitations of the published data: many archaeopriapulids require additional microstructural description and analysis, with a particular deficit in the character-rich introvert (e.g. the arrangement and fine structure of scalids). This has similar effects to the incompleteness of fossils and is mediated through the same mechanism. The third possibility is that levels of homoplasy among priapulids are sufficiently high to obfuscate relationships and make trees intrinsically unstable. Something of this nature has been reported in trilobites in particular and arthropods more generally. It is a pattern consistent with a rapid radiation and the early exploration and saturation of character space (Wagner, 2000; Wagner et al., 2006).

Relationships within the predominantly extant clade of ((Priapulidae, Halicryptidae), Maccabeidae) were the most stable and impervious to the precise mode of analysis. All of the flat-weighted, implied-weighted and stratigraphically filtered trees agreed upon the same general structure. Acanthopriapulus resolved in a polytomy with Priapulus, which was sister group to Priapulopsis. This was then followed by the plesion Priapulites, then Halicryptus and Maccabeus as successive sister groups. This was entirely consistent with the genus-level analyses of Wills (1998b), Dong et al. (2005, 2010) and Harvey et al. (2010), as well as most taxonomies of the group. The successively reweighted trees differed only in their placement of Priapulites within a paraphyletic Priapulopsis. Tubiluchus and Tubiluchidae (Tubiluchus and Meiopriapulus) also resolved as clades in all of our trees. The Chinese fossil Paratubiluchus bicaudatus (Han et al., 2004; Huang et al., 2006b) resolved as sister plesion to the extant Tubiluchidae in the implied weights analysis (Fig. 3c), but elsewhere in the other optimizations.

Even allowing for the inclusion of Paratubiluchus and Priapulites within the crown group, the extant priapulids did not constitute a clade. Rather they were rendered paraphyletic by the intrusion of the small Cambrian clade Selkirkia + Paraselkirkia. In the implied weights analysis, the extant taxa also formed a polytomy with a Cambrian clade of Xiaoheiqingella + Yunnanpriapulus (Han et al., 2004; Huang et al., 2004b). Harvey et al. (2010) also resolved these last two genera in a polytomy with the extant taxa. In all of our other trees (Fig. 3a,b,d), Xiaoheiqingella plus Yunnanpriapulus constituted the sister plesion of the crown group. Huang et al. (2004b) presented a simplified cladogram with Xiaoheiqingella and Yunnanpriapulus as successive plesions to all extant priapulids except the Tubiluchidae, citing the shared large and bulbous introvert among other characters. When Harvey et al. (2010) constrained their data to resolve all palaeoscolecid genera as the sister group to other ecdysozoans, Xiaoheiqingella and Yunnanpriapulus were unambiguously placed as sister group to ((Priapulidae, Halicryptidae), Maccabeidae), thereby making extant priapulids paraphyletic (although this arrangement is clearly suboptimal by definition). We therefore concur with the general conclusions of Huang et al. (2004b) that Xiaoheiqingella and Yunnanpriapulus are plausible close allies of the living priapulids. A close relationship between Selkirkia + Paraselkirkia and extant priapulids as found in all trees is less precedented, although the position of Selkirkia has been somewhat labile in previous analyses. The overall body form of both fossil genera differs considerably from that in living priapulids. However, Priapulidae, Halicryptidae, Maccabeidae and Tubiluchidae also have very distinctive sets of apomorphies, and such differences cannot preclude a phylogenetic relationship.

Unusually for such analyses, we resolved Palaeoscolex piscatorum within the Nematoida irrespective of our weighting scheme. This is a position less inconsistent with that envisaged by Conway Morris & Peel (2010), and closer (at least) to the base of the Ecdysozoa. Harvey et al. (2010) placed palaeoscolecids sensu stricto in a clade with other elongate taxa informally assigned to or allied with the palaeoscolecids (i.e. Tabelliscolex, Cricocosmia, Tylotites). More recently, Dong et al. (2010) resolved them in a variety of locations, but most stably in a clade with Scolecofurca, Louisella and Ottoia, and followed by two small clades of Cricocosmia + Tabelliscolex and Tylotites + Markuelia. More broadly, we found that archaeopriapulids (putative Cambrian priapulids) and palaeoscolecids sensu lato are paraphyletic relative to one another.

In Dong et al.'s study (2005), the Cambrian Palaeopriapulites (Hou et al., 2004; Han et al., 2007a) and Sicyophorus (Han et al., 2007a; Dornbos & Chen, 2008) (considered synonymous by Huang et al. (2004b)) were placed low in a Cambrian clade (just above Selkirkia). In the recent analyses of Harvey et al. (2010), these two genera were resolved low in the phylogeny (just above Kinorhyncha and Loricifera), further contributing to the paraphyly of Cambrian genera in their trees. Dong et al. (2010) variously placed them as sister group to a clade of ((Priapulidae, Halicryptidae), Maccabeidae), or to a clade of all extant Priapulida plus Xiaoheiqingella and Yunnanpriapulus in some weighted analyses. This large shift in their apparent relationships is largely mediated by the inclusion of Markuelia in the latter data set. In our new data set, Palaeopriapulites and Sicyophorus emerged as sister species close to the crown group in our implied weights analysis (Fig. 3c) (as well as in the trees of Dong et al. (2005, 2010) and Harvey et al. (2010)) or in a small polytomy with Paratubiluchus bicaudatus as part of a much larger basal clade in our other analyses. The position of Markuelia was variable in the analyses of Dong et al. (2010). In all of our analyses, it resolved close to the base of the tree, and above Xystoscolex.

As would be expected, the structures of the character morphospaces reflected the overall structures of the corresponding trees. The three successive morphospaces (derived from Wills, 1998ab; Dong et al., 2005 and the data herein) had a similar overall structure, at least when the first three principal coordinates were considered (Figs 4-6 and 8). Most broadly, Cambrian and post-Cambrian taxa occupied adjacent rather than overlapping regions of their morphospaces. Moreover, Cambrian taxa invariably occupied a smaller region of morphospace than their post-Cambrian counterparts. These differences were not significant for range-based indices in the earlier data sets, but significant at larger sample sizes for the variance-based indices. For the expanded data set presented here, there were significant differences in both indices, and over a much wider range of sample sizes.

We also note that the overall intensity of local clustering and discontinuity of morphospace occupation both increase as research time progresses (Figs 8 and 9). More specifically, the predominant archaeopriapulid and palaeoscolecid regions of all plots become more densely occupied. This increased clustering of points (particularly and especially many Cambrian forms) may partially explain the increased sensitivity of our bootstrapping tests to differences between Cambrian and post-Cambrian taxa at decreasing sample sizes.

It is not surprising that the discovery and description of new fossil taxa (along with better and more detailed studies of both fossil and extant species) has influenced some generalities and many details of the morphospace analyses. However, the general conclusions drawn from the first such study (Wills, 1998b) appear robust to these additions and changes.

In general, it must be expected that fossils will offer less complete morphological data than extant taxa. We note, however, that many archaeopriapulids and palaeoscolecids are preserved in fine detail and have been subjected to SEM and ultrastructural studies (Han et al., 2007b,d; Hu et al., 2008; Conway Morris & Peel, 2010; Topper et al., 2010). Similarly, not all extant species have been investigated at high resolution, neither have they been described from multiple or even complete specimens in some cases (Amor, 1975). In general, however, fossil taxa do contribute more ‘missing’ data cells to cladistic matrices, and this is the case for the present data set.

More broadly, there is good empirical evidence that fossil taxa are about as likely to influence inferred phylogenetic relationships as their extant counterparts. There is some suggestion (Dong et al., 2010; Fig. 2) that the inclusion/exclusion of fossil archaeopriapulid taxa has a greater impact upon apparent relationships than the inclusion/exclusion of extant taxa. However, these differences are not statistically significant (P > 0.25 in all Mann–Whitney tests for all metrics) and are more likely to be the result of shorter and less well-supported branches within archaeopriapulid and palaeoscolecid clades than of ‘missing’ data per se. Missing entries are not necessarily a problem in morphological cladistic analyses because relationships are inferred from the data that are available, not from those that are unavailable (see Mounce & Wills, 2011). A concentration of missing entries in particular taxa is more likely to result in poor resolution and an obfuscation of relationships in some or all regions of the tree, but the magnitude of such effects is contingent on the precise amount, position and nature of the missing cells (Wiens, 2003, 2005, 2006). More precisely, the quality of the data that are available is more important than the amount of data that are not. Missing entries do not, in and of themselves, predispose taxa to resolve artefactually in any location (e.g. basal or derived) in the tree. What may have a biasing effect is the erroneous assumption that characters that cannot be coded in fossils (because taphonomic conditions preclude their preservation) were actually absent in the animal when alive. Insofar as absences are often more characteristic of the ancestral condition (a common occurrence when characters are primarily coded with reference to variation in the ingroup), this may cause fossils to resolve artefactually closer to the root of the tree. This phenomenon of ‘stemward slippage’ has been reported for vertebrates (Sansom et al., 2010) and is currently being investigated for other phyla including priapulids. Detailed, empirical taphonomic data are the key to testing such hypotheses.

When estimating morphological disparity, missing entries can have other effects, depending upon their treatment. One possibility is to recode unordered characters into nonadditive binary, and replace all ‘?'s with column (character) means as a preparation for principal components analysis (Briggs et al., 1992; Wills et al., 1994; Wills, 2001). Missing entries will then move OTUs closer to the centroid of the entire character space, such that a hypothetical and entirely unknown taxon would be plotted at the origin or centre of gravity of the morphospace. This is conservative to the extent that the bounds of the morphospace can never be defined on the basis of the absence of data. However, the apparent disparity of a subgroup of taxa (e.g. Cambrian or post-Cambrian) could potentially be inflated by missing entries, especially where the majority of those taxa are plotted distant from the global centroid. Here, we used an alternative approach, calculating the generalized Euclidean distance (Wills, 1998a) between each pair of OTUs on the basis of those characters scored for both, and using this to make a correction for characters not scored for both (Wills, 1998a, 2001). The resulting distance matrix between taxa was then ordinated using principal coordinates analysis. Missing entries will not cause taxa to move preferentially towards the global centroid: rather they will reduce the accuracy with which they are placed in the morphospace. The introduction of missing entries can cause an OTU to move in any direction relative to a completely known one. However, it will resolve as most proximate to the OTU with which it shares the greatest similarity in coded characters. The second manner of dealing with missing entries is therefore more closely analogous to the treatment of missing entries by parsimony-based phylogeny programmes.

The data missing from fossils are actually of two distinct types. The type of missing entries (‘?'s) discussed above are data that we know we do not have. However, there may also have been characters in the living animals that were not preserved as fossils and that are not exemplified by any extant forms. Volatile soft tissues and fine ultrastructural details unique to extinct species might never be recorded in the fossil record. These are data that we do not know that we do not have. Hence, there is a hypothetical asymmetry between the manner in which characters are coded in fossil and extant organisms, with living species offering the more complete frame of reference. In disparity studies, this may mean that we miss certain characters that only varied between extinct species, and that we thereby underestimate their morphological variation (as well as the differences between extinct and extant taxa). In phylogenetics, we may miss characters shared by all, or some subset of, extinct taxa. This could have the effect of rendering genuine fossil clades as paraphyletic or polyphyletic. Simulations and empirical case studies are needed to test this for priapulids and more generally (Sansom et al., 2010).

More broadly still, extinct taxa are more likely to be omitted from analyses altogether than are their extant counterparts. Again, this can be because fossils are known to be highly incomplete or are poorly described (and therefore likely to obfuscate phylogenetic reconstructions), or because fossils are unknown (not yet discovered or never preserved in the first place). The effects of these omissions upon inferred relationships are difficult to predict, except where incomplete fossils are demonstrably uninformative, bringing no novel character combinations to bear (Wilkinson, 1995). If fossil taxa were omitted from the sample at random, then their absence might be expected to reduce range-based measures of disparity to the extent that these are contingent upon sample size. Procedures that minimize the effects of sample size differences – such as rarefaction or the use of variance-based indices – would be anticipated to largely obviate this problem. However, it is probable that taxa are not omitted at random: poorly known fossils might easily be concentrated in particular clades, in particular localities or at particular times. Again, the effects of such biased sampling upon disparity indices are difficult to predict in specific cases (other than that range-based indices are likely to be reduced), and there is no reason to suppose that they will consistently operate in any particular direction.

Our results were consistent with previous findings (Wills, 1998b): Cambrian priapulids, archaeopriapulids and palaeoscolecids are only marginally less disparate than their extant counterparts. However, over the last 15 years, many important new fossil finds have been made, and many more intensive morphological studies of extant species have been undertaken. Our larger and more inclusive species-level analysis now reveals this modest difference to be a statistically more significant one using range-based as well as variance-based indices. Cambrian and living taxa still occupy broadly overlapping regions of their empirical morphospace, but several recently described Cambrian fossils appear closer to living species. What implications do these new findings have for our understanding of the Cambrian explosion? Most significantly, indices of morphological disparity (unlike morphological phylogenies) are robust to modest differences in taxon and character sampling, lending our results a new rigour. Hence, the near-extant levels of disparity embodied by the Cambrian worms indicate that their appearance in the fossil record must have been preceded by large amounts of morphological change during their divergence from a presumed common ancestor. Although debates on the time available for this divergence are set to continue (Valentine et al., 1996; Erwin, 2007; Erwin et al., 2012), the enormous magnitude of the explosion it represents is clear.


  1. The detailed relationships between priapulids, archaeopriapulids and palaeoscolecids remain unclear. The inclusion of data from additional fossil and extant species confirms the general findings of several previous analyses (Cobbett et al., 2007; Dong, 2007; Dong et al., 2010), most notably that relationships deep in the tree are easily perturbed by relatively small changes to the taxon and character sample.
  2. Priapulidae consistently emerges as a clade, within which relationships are [(Priapulus + Acanthopriapulus) Priapulopsis + Priapulites]. Priapulidae is sister group to Halicryptidae, and this entire clade is then sister group to Maccabeidae. Tubiluchidae also invariably emerges as a clade, but internal relationships depend upon the weighting scheme used. The fossil Paratubiluchus may constitute a plesion to Tubiluchidae. In addition to this, a variety of small fossil clades intervene such that Tubiluchidae never resolve in an exclusive clade with Priapulidae, Halicryptidae and Maccabeidae.
  3. Species variously referred to the archaeopriapulids and palaeoscolecids sensu lato form small clades that are generally interspersed within an otherwise pectinate series of stem group plesions. Relationships within this region of the tree are especially labile with modest changes to the taxon sample or weighting scheme. More broadly, we find that archaeopriapulids (putative Cambrian priapulids) and palaeoscolecids sensu lato are paraphyletic relative to one another.
  4. The overall patterns of character morphospace occupation derived from the three data sets are similar, despite significantly increasing taxon sampling over research time. More specifically, Cambrian and post-Cambrian taxa occupy adjacent rather than broadly overlapping regions of the spaces. This is consistent with the phylogenetic resolution of a crown group including few fossils, and a paraphyletic series of plesions. Despite this, Priapulidae, Halicryptidae, Macccabeidae and Tubiluchidae are all phenetically more proximate to fossil genera than they are to one another. The Tubiluchidae are nearest neighbours to the Chinese Paratubiluchus bicaudatus (Han et al., 2004) (itself potentially the plesion of the latter family), whereas Halicryptidae is phenetically closest to Yunnanpriapulus halteroformis (Han et al., 2004; Huang et al., 2004b).
  5. Extant priapulids (plus the Carboniferous Priapulites) are more disparate than Cambrian archaeopriapulids and palaeoscolecids. This is true for all three data sets investigated here and for all indices of morphological disparity (once sample size differences are taken into account). For the data sets of Wills (1998b) and Dong et al. (2005), these differences are only significant for variance-based indices. In our new data set, improved taxon sampling reveals significant differences for all indices. Unlike the results of cladistic analyses (which are extremely sensitive to the details of taxon and character sampling), the results of disparity analyses are much more robust. The intensity of local clustering and discontinuity with which our morphospaces are occupied increases with research time, thus providing greater insights into the structure of the morphospace. Cambrian and post-Cambrian time bins now display positive clustering of their constituent species. Their nonoverlapping distributions also yield a globally heterogeneous pattern when the data for all species are pooled.


MAW, SG and MH all thank the Leverhulme Trust (Grant F/00 351/Z) for funding. MR acknowledges support from the NERC (Advanced Research Fellowship NE/F014872/1). We are very grateful to Gareth Dyke and two anonymous referees for their constructive and helpful comments on drafts of this manuscript.