Cladistics (1996) 12:41±64 THE HIGHER CLASSIFICATION OF THE ORDER EMBIOPTERA: A CLADISTIC ANALYSIS Claudia A. Szumik Department of Entomology, American Museum of Natural History, Central Park West at 79th Street, New York, 10024, U.S.A. and CONICET, Facultad de Ciencias Naturales e Instituto Miguel Lillo, Miguel Lillo 205, 4000 S.M. de TucumaÂn, Argentina Received for publication 24 April 1994; accepted 24 July 1995 Abstract Ð A matrix of 41 Embiid taxa (representing the 8 formally recognized families of the Order) and 36 characters were cladistically analysed as a first attempt for understanding the higher classification of the Order Embioptera. The resulting trees were rooted with Clothodidae as the sister group of the other Embioptera. The results suggest that the current classification contains several artificial groups. With the rooting used, only Anisembiidae and Australembiidae are monophyletic. Embiidae is polyphyletic, as Australembiidae+ Notoligotomidae, Enveja (incertae sedis) and Oligotomidae+Teratembiidae appear within Embiidae, and the ªembiidº Microembia appears within Notoligotomidae. Oligotomidae is paraphyletic in terms of Teratembiidae. Four of the genera included in the analysis are paraphyletic: Mesembia, Chelicerca (in terms of Dactylocerca and Pelorembia), Aposthonia (in terms of Oligotoma), and Metoligotoma (in terms of Australembia). Pelorembia and Dactylocerca are synonymized with Chelicerca. 1996 The Willi Hennig Society Introduction The insect order Embioptera is a monophyletic group defined by several morphological and behavioral characters. The most striking apomorphy of these subsocial insects is their ability to produce silk in all instars from unicellular glands in the swollen first basitarsus (Barth, 1954; Alberti et al., 1976; Nagashima et al., 1991) and to use it to construct tunnels under bark, stones or soil. Other synapo- morphies of the order are the prognathous head, presence of a gula, absence of ocelli, poorly developed compound eyes, three segmented tarsi, thickened hind femora, bisegmented cerci, secondary intromittent organs, ovipositor absent and wingless females (Hennig, 1981). In a cladistic analysis of insect orders (Wheeler et al., in prep.), Embioptera appeared as sister group of the Plecoptera by virtue of sharing tarsal plantulae, reduced phalomeres, a trocantin-episternal sulcus, separ- ate coxopleuron and premental lobes. Most publications on the systematics of the order Embioptera are descriptions of new species and new records. Of all published research on this group, only three papers (Davis, 1938, 1940b; Ross, 1970) refer to their higher classification. Two of these contributions (Davis, 1938, 1940b) were based on systematic analy- ses, which, interestingly, used methods closely resembling parsimony methods used today. Davis (1938) chose 21 terminal taxa as representatives of the main lineages of Embioptera. He used seven multistate characters, coding the characters with numbers and the states with letters. Based on the distribution of character states observed in his representative taxa, he proposed a scheme (Fig. 1) that had 0748-3007/96/010041+24/$18.00/0 1996 The Willi Hennig Society 42 C. A. SZUMIK Fig. 1. Tree of Davis (reproduced with permission from Proc. Linn. Soc. N.S.W. (1938) 65: 265, fig. 120). ORDER EMBIOPTERA: A CLADISTIC ANALYSIS 43 ªbeen drawn up in the simplest possible way; convergence has not been invoked unless the opposite course does violence to any known facts.º (1938:266). Davis indicated on his cladogram all the changes (even the homoplastic ones), thereby allowing a complete reconstruction of his data set (Appendix 3). Although Davis included some terminal taxa at the nodes of the tree, he pointed out that that was not to be interpreted as a proposal of actual ancestry, but instead, ªimplied that the hypothetical ancestor resembled the existing form namedÐat least for those characters here dealt withÐand that the existing type has changed little in these characters from this ancestor.º (1938:262). In a subsequent analysis, the same author (Davis, 1940b), using 55 taxa and 15 characters, presented four trees (each for a different geographic area), but this time specifying only the changes between hypothetical ancestors and terminal taxa. Ross (1970) proposed an entirely new classification. He divided the order into 4 subgroups and 14 families and rejected all previous classifications on the following grounds (1970:164): ªNo attempts will be made at this time to discuss the history of Embioptera classification. The great number of new species and higher categor- ies known to the writer make all old classifications obsoleteº. It is true that, after Davis' work, the number of described genera and species of embiids greatly increased (mostly thanks to Ross' excellent contributions). However, the discovery of new taxa is certainly not in itself a reason to refute any previous classification, particularly when the evidence cited in support for some of the groups in the new classification is weak. This lack of justification is extreme in several cases. For example, all information given by Ross for his new ªSuborder Bº (said to include Enveja bequaerti and undescribed species from Africa) is that it includes ª. very large, often colorful embiids with distinctive features in almost every structureº (1970:169). A new ªFamily Cº is proposed, but no information for the constituent genera and species is given, except that they are undescribed, they live in Peru and Colombia, and they have been discovered recently (1970:168). Previous work by Davis and Ross on the higher classification of Embioptera invites, for different reasons, a reanalysis of the problem. In the case of Davis' work, it is tempting to examine to what extent results obtained in one of the first papers using cladistic techniques would hold 50 years later. In the case of Ross' work, it is interesting to see to what extent a deep general knowledge of the group is sufficient in itself to erect groups which, given the lack of explicit justification and methodology, appear to be based on no more than intuition. Methods and Materials CLADISTIC ANALYSIS Hennig86 (Farris, 1988) was used to calculate most parsimonious tree(s) under equal weights for all characters, with the commands mh* bb*. However, as argued by Farris (1969, 1983) and most recently by Goloboff (1993a), cladistic analysis depends on the weights assigned to the characters. Therefore, the program Pee-Wee (Goloboff, 1993b) was used to weight the characters and calculate tree(s) that give the highest weight for the characters. The quantity to be maximized is referred to as the ªfitº of the characters, as the weight of a character is a function 44 C. A. SZUMIK of its homoplasy or fit to a tree; the fit for each character is measured as a concave function of its number of extra steps (see Goloboff, 1993a, for justification and discussion). In this analysis, the concavity value of Pee-Wee was set at 5 and all characters were given a prior weight of 10. Character fits were then calculated as 50/5+R+ES, where R is the number of steps implied by polymorphisms in terminal taxa and ES is the number of extra steps on the tree. The total fit of all the charac- ters is reported as rescaled fit (expressed as percentage) and was calculated using the formula 1−(max.fit−fit)/(max.fit−min.fit) (this formula is analogous to that of the retention index; see Goloboff's 1993b documentation). Among all possible trees, the tree(s) with the highest total fit are chosen. The command mult* was used to search trees of highest fit, performing 50 repli- cations of a random addition sequence Wagner tree each followed by tree bisec- tion reconnection branch-swapping. The command jump* was used in an attempt to find additional fittest trees, swapping in trees suboptimal by a fit difference of up to 0´2. To compare the results with previous classifications the commands force, max/ and cmp of Pee-Wee were used. Force was used to define groups to be constrained for monophyly and max/ to search the fittest trees containing those groups. Then, cmp was used to check the difference in steps and fit of each character between the constrained and the best-fitting tree(s). To give some notion of how strong the support is for each clade in the tree, the commands swap, mv and cmp were used; these commands used in combination find the differences in fit and length when a clade or taxon is moved to another part of the cladogram. The program Clados (Nixon, 1992) was used to produce tree diagrams. Ambigu- ous optimizations were not considered as support for any clade (i.e. the ambigu- ous- option of Pee-Wee was used). The consensus tree does not show a single most parsimonious optimization, but instead those synapomorphies shared by 500 dichotomous parsimonious resolutions. This was accomplished with the command apo of Pee-Wee (Goloboff, 1993b), which can be used to find change common to sets of trees. Many more dichotomous parsimonious resolutions exist; these 500 were arbitrarily chosen. TAXA EXAMINED The trees were rooted with Clothodidae as the sister group to all other Embiop- tera, following the concepts of rooting and polarity most recently described in Nixon and Carpenter (1993). Clothodidae has been `'definedº (Davis, 1938, 1940b; Ross, 1970, 1984) as the family comprising the ªsimplestº Embioptera. All other embiids differ from clothodids and are united into a monophyletic group by being medium to small in size, having weak and incomplete wing venation, modi- fied head and thoracic structures and asymmetrical male terminalia. The trees were rooted arbitrarily on Clothoda with Clothodidae as paraphyletic, but the root might as well have been placed on Antipaluria or Clothodidae. To make an analysis of the higher classification, one should include representa- tives of the different groups used by Davis and Ross. However, six of the 14 families mentioned by Ross (1970) are made up exclusively of undescribed genera and species.