Phylogenetic relationships among social parasites and their hosts in the ant tribe Tetramoriini ( Hymenoptera : Formicidae )

The phylogenetic relationships among Palaearctic species of the ant genus Tetramorium and its social parasites of the gen­ era Strongylognathus, Anergates and Teleutomyrmex, were investigated electrophoretically at 21 presumptive enzyme loci. The data set comprising 33 species was analysed with distance (UPGMA, Neighbor-joining and least squares statistics) and parsimony methods (independent allele, minimum turnover and mutation coding) in order to rule out analysis-dependent effects. Several group­ ings were consistently resolved by all procedures. Observed branching patterns support the placement of the three parasite genera and their hosts into the Palaearctic species group of Tetramorium (tribe Tetramoriini). The genus Strongylognathus forms a nionophyletic group in which the slave-makers of the S. huberi group constitute the sister group of the inquilines S. testaceus and S. karawajewi (S. testaceus group). Most species of the S. huberi group show very low genetic differentiation. However, little consensus has been found with regard to which Tetramorium species are the closest relatives of Strongylognathus. According to the electrophoretic data, social parasitism in Palaearctic tetramoriine ants has evolved independently at least twice. Though inquilinism once arose from slave-making ancestors in Strongylognathus, the extreme inquilines Anergates atratulus and Te­ leutomyrmex schneideri are clearly set apart from the Strongylognathus clade in phylogenetic analyses. Thus, extreme inquilinism cannot be regarded as the endpoint of a single parasitic lineage in the Tetramoriini. In these highly advanced inquilines, evolutionary rates at allozyme loci appear to be higher than those of their Tetramorium hosts. The results do not unambiguously reveal whether Anergates and Teleutomyrmex arose jointly or independently from Tetramorium ancestors. However, a combined analysis using all available evidence supports the former hypothesis. The finding that the Tetramorium parasites are not the closest relatives of their re­ spective host species is discussed in relation to current theories for the evolution of social parasitism.


INTRODUCTION
Social parasites of the widespread ant genus Tetramo rium are predominantly Palaearctic in distribution (Bolton, 1976), the only exception being two inquilines known from the Afrotropical zoogeographical region (Bolton, 1980).In the Palaearctis, species of the three genera of social parasites, Strongylognathus, Anergates and Teleutomyrmex, display a wide array of differing parasitic traits.Dulotic or slave-making ants kill or drive off the resident queen(s) and later rely on raiding neigh bouring host colonies for their supply of workers.Inquil ines are usually workerless parasites (but some have re tained a reduced worker caste) whose queens either coex ist with the host queen or become the only reproductives within the nest.The latter type of parasitism is here re ferred to as queen-intolerant inquilinism, but was termed pseudoinquilinism by Douwes (1990) and excluded from inquilinism by Bourke & Franks (1991).
All of the approximately 25 currently recognized spe cies of the genus Strongylognathus bear remarkable saber-shaped mandibles (Bolton, 1976).While slave making behaviour is characteristic for the members of the S. huberi group (sensu Bolton, 1976;Forel, 1905;Kutter, 1920;Sanetra & Buschinger, 1996), in the S. testaceus group (comprising only two species) inquilinism has most likely evolved via degeneration of slave raiding (e.g.Wasmann, 1905;Kutter, 1969;Acosta & Martinez, 1982).Teleutomyrmex schneideri also lives as an inquiline alongside the Tetramorium host queen, but without maintaining its own worker caste (Kutter, 1950;Stumper, 1951).Teleutomyrmex, often called the "ultimate para site" (e.g.Holldobler & Wilson, 1990), displays the most outstanding adaptations ever recorded for ant social para sites.With their highly specialized tarsal claws the flat tened Teleutomyrmex queens attach themselves to the much larger host queen, where they are efficiently tended and fed by the host workers (Kutter, 1950;Stumper, 1951;Collingwood, 1956).A clearly different strategy occurs in the workerless parasite Anergates atratulus since the infested nests are devoid of host queens (e.g.Wasmann, 1908;Wheeler, 1910).The occurrence of these varying degrees in socially parasitic behaviour raises the question of whether parasitism arose only once in the evolution of tetramoriine ants or whether it has had multiple origins.Possible transitions of different parasitic relations from one type to another would also be interest ing to discover.
Recently, some progress has been made toward under standing the evolutionary history of the Tetramoriini (Sanetra et al., 1994;Baur et al., 1996), but a thorough phylogenetic analysis has not been presented so far.A se vere problem in the use of the genera Tetramorium and Strongylognaihus for phylogenetic studies is the lack of comprehensive taxonomical revisions.Though a number of revisions dealt with Strongylognaihus in different geo graphical regions (Pisarski, 1966;Baroni Urbani, 1969;Radchenko, 1991), a satisfactory taxonomy for the entire genus has yet to be produced.The species level taxonomy in the host genus Tetramorium is also not well understood, due to the high degree of morphological similarity and considerable intraspecific variability of the species within the group.In the revisionary monographs of Bolton (1977Bolton ( , 1979Bolton ( , 1980) ) the taxa of the Old World Tropics and the New World have been studied exten sively, but the Palaearctic species of the genus (71 caespitum group sensu Bolton, 1977) remain largely unre solved.Only a few recent studies have dealt with the taxonomy and identification of Palaearctic Tetramorium in detail (Lopez, 1991a, b;Radchenko, 1992a, b;Sanetra et al., 1999).
In the last decade, the subject of social parasite evolu tion in bees, wasps and ants has received great attention (e.g.Buschinger, 1990;Bourke & Franks, 1991;Carpen ter et al., 1993;Choudhary et al., 1994;Heinze, 1991Heinze, , 1998;;Ward, 1996;Lowe & Crozier, 1997).The main topic of debate has been how close social parasites are re lated to their respective host species with regard to a hy pothesis proposed by Emery (1909), now known as Em ery's rule.Ant parasites have generally been found to be close relatives of their hosts (loose form of Emery's rule according to some authors, e.g.Ward, 1989;Bourke & Franks, 1991).Since the enormous spread of molecular methods, a number of studies have further revealed that social parasites often fall into monophyletic groups which are closely attached to their host species' groups (Choud hary et al., 1994;Baur et al., 1995Baur et al., , 1996;;Pedersen, 1996;Lowe & Crozier, 1997).Thus, the application of Emery's rule in the strict sense, that is each parasite is the sister species of its respective host, appears confined to a num ber of inquilines scattered through various groups of ants (Holldobler & Wilson, 1990;Buschinger, 1990).For in stance, according to Wilson (1984), there are nine inquiline species known in the ant genus Pheidole which seem to have evolved independently and may well be immedi ate derivatives of their host species.As subsequent speciation events might have occurred in both the host and parasite lineage, Emery's rule has recently been inter preted as implying that the sister group of the parasite clade includes all host species (Ward, 1996;Pleinze, 1998).
Another controversial issue, which is related but not identical to the phylogenetic relationships of hosts and parasites, concerns the origin of social parasitism.Social parasites may have arisen through either allopatric or sympatric speciation (e.g.Holldobler & Wilson, 1990;Buschinger, 1990;Bourke & Franks, 1991;Choudhary et al., 1994;Ward, 1996).The interspecific hypothesis sug gests that two species evolved allopatrically, and then the host-parasite relationship developed upon secondary con tact (Wilson, 1971; termed social deception hypothesis by Carpenter et al., 1993).A quite different evolutionary pathway leading to parasitism, the intraspecific hypothe sis, has been formulated by Buschinger (1990) and Bourke & Franks (1991).Under this model, social para sites evolved sympatrically from within populations of their hosts by assortative mating of intraspecific variants.According to some authors (West-Eberhardt, 1990;Bourke & Franks, 1991;Heinze, 1998) the two compet ing hypotheses may even be combined, and thus need not be regarded as mutually exclusive.Evolutionary studies of social parasitism still promise to shed new light upon this long-standing discussion, especially when dealing with the large number of ants' parasites having yet unre solved phylogenies.
In the current study we focus on the evolution of social parasites in the myrmicine ant tribe Tetramoriini.The ge netic relationships of the three parasite genera, Strongylognathus, Anergates and Teieutomyrmex, and a large number of host and non-host Tetramorium species were analysed using allozymes.We will provide the first phy logenetic trees for this group in order to find out whether parasitism has arisen multiple times or only once.We will also address if parasites and their respective hosts are sis ter species.

Taxa and sampling
The species included in this study are listed in Table 1.Ow ing to the problematic nature of species names in the genus Tetramorium, only species which could be reliably attributed to described taxa were considered.Species rank of T. diomedeum and T. pimctatum is given in Sanetra et al. (1999).The investi gated Tetramorium species endemic to the Palaearctic belong to the morphologically well defined T. caespitum group (Bolton, 1977) and can therefore be assumed to be monophyletic.The Oriental species T. insolens, T. cf.centum, T. ianuginosum, T. kheperrct (the latter two formerly placed in the genus Triglyphothrix Forel, 1890 now being a synonym of Tetramorium), the Afrotropical T. acutisetum, T. simillimum, T. caldarium and the New World species T. hispidum were designated as out groups.
Even though similar taxonomic problems have been encoun tered in species of socially parasitic Strongylognatlms, all avail able samples were included because of the rareness of this genus.In most cases, assignment to the described taxa has been established by comparisons with the type material.Our Strong)'lognathus samples from Crimea clearly belong to S. christophi and not to S. arnoldii Radchenko, 1985, because of the shape of the volsella given as a discriminating feature in the only avail able key (Radchenko, 1991).The separation of S. testaceus and S. karawajewi was achieved by examination of the sculpturing on the head surface.
As described in Sanetra et al. (1994), T. "Czech Republic" and T. cf.semilaeve, from which only a single colony each had been available, turned out to constitute unusually small and weakly sculptured T. caespitum (see also López, 1991b).Though standing in contrast to the findings of López (1991a), T. "S.Nevada" is very likely a geographic variant of T. impurum.
Tabu-1.Species included in this study.

Outgroups
Tetramorium lanuginosum Mayr, 1870 Tetramorium cf.centum Bolton, 1977 Tetramorium kheperra (Bolton, 1976) 1987 (Poldi, in litt.), but is not included in this analysis.71 "Crete 2" is most likely 71 diomecleum, of which the typical form occurs in southern Italy (see also Sanetra et ah, 1999).Dif ferences between typical 71 semilaeve and T. "Tenerife" are slight at best and the latter form probably does not deserve taxo nomic recognition.It still remains an open question, however, as to whether 71 semilaeve from the western (nominotypical form) and the eastern Mediterranean (T."Crete/Rhodes" in Sanetra et ah, 1994) are different species (see Sanetra et ah, 1999).Field collections were carried out mainly in Central Europe, the Mediterranean region and in the western part of Asia (see Appendix 1).Nest samples were aspirated and transported back to the laboratory alive, where most of the material was preserved at -80°C.Additionally, some nest fragments were maintained alive.Three to five individuals from each colony were analysed for each enzyme.However, some of the enzymes could not be scored in colonies where too few individuals had been collected, as was the case in some of the socially parasitic and outgroup species.

Data analysis
The genetic variability within species was evaluated using the parameters expected mean heterozygosity per locus (Hcxp), mean number of alleles per locus (A) and percentage of polymorphic loci (PT O ) following the 99% criterion.Thus, a locus was consid ered polymorphic if the frequency of the most common allele did not exceed 0.99.Standard errors were obtained by jackknif ing over loci (Shoemaker et ah, 1992).Allele frequencies were obtained from the sampled genotypes by weighting the frequen cies within colonies equally.
Coding of allozyme data for cladistic parsimony analyses was accomplished using three different strategies, because there has been some debate as to the best way to analyse these types of data (e.g.Wiens, 1995;Swofford et al., 1996).In the first proce dure, electromorphs were coded as input for Hennig86 in a presence/absence fashion (Mickevich & Johnson, 1976), with alleles treated as characters under a "corrective weighting scheme" that applied the same weight to each locus (Moran et ah, 1990).Following the recommendations of Maddison et ah (1984) and Nixon & Carpenter (1993), outgroup (T.insolens and T. hispidum omitted because of a large number of missing values, see Table 2) and ingroup were simultaneously resolved to seek a globally most parsimonious solution.
For the procedures described in the following a hypothetical ancestral taxon was synthesized by outgroup analysis (Kluge & Farris, 1969;Watrous & Wheeler, 1981;Maddison et ah, 1984) using the information from all available outgroup taxa.Ances tral state assessments were obtained by the generalized principle that electromorphs found in both the ingroup and the outgroup are plesiomorphic (e.g.Richardson et ah, 1986;Baverstock et ah, 1979;Patton & Avise, 1983).
In the second approach, character states were ordered into transformation series (or character state trees) under the assump tions of the minimum allele turnover model (Mickevich & Mitter, 1981, 1983); a few unconnectable states were left unordered (see also Carpenter et ah, 1993).New states were allowed as an cestral condition in branching character state trees (Mardulyn & Pasteéis, 1994).Polarization of these character states followed the taxonomic outgroup and functional outgroup criteria of Wa trous & Wheeler (1981), as expanded by Farris (1982).Branch ing character state trees (see Appendix 2) were recoded and prepared as input matrices for Hennig86 using ordinal coding (Mickevich & Weller, 1990).
The mutation model for coding allozyme data in conjunction with the "quadraphenic evaluation procedure" (Murphy, 1993) was used in the last approach.A more effective application re sulted from combined analysis of the electrophoretic data and some non-molecular characters, which are 1.mandibles elongate and saber-shaped, 2. mandibles reduced and edentate, 3. sting strongly reduced, non functional, 4. adelphogamy, males pupoidal and reduced in number, 5. parasitism.
Trees were constructed from these data matrices with the Hennig86 commands mhennig*\bb*\ which find exact solutions on most occasions (Farris, 1988;Platnick, 1989).In some in stances successive weighting (Farris, 1969;Carpenter, 1988) was applied to optimize the results under the criterion of mini mum homoplasy (Retention-Index from Farris, 1989).The Jack knife monophyly index (JMI) developed by Siddall (1995) was performed with Lanyon.exe in Random Cladistics (Siddall, 1994).JMI values were utilised to infer which clades are more (or less) stable than others within and among the most parsimo nious trees and in the consensus tree.Character states were opti mized and mapped onto the trees with accelerated optimization as implemented in Clados (Version 1.6.1;Nixon, 1998).This option prefers reversals over parallelism if the two are "equally parsimonious".

Allozyme variation
Allozyme phenotypes of putative heterozygotes were in accordance with those expected on the basis of the quar ternary structure of the enzymes given by Murphy et al. (1996) except for Xdh (see Sanetra et al., 1994).The products of 21 presumptive gene loci were obtained for the taxa summarized in Table 1.Of these loci, three (.Acoh-2, G6pdh, Mdh-2) were either monomorphic for all ingroup samples or the electromorphs could not be reli ably scored.Allele frequency data for the remaining 18 polymorphic loci are presented in Table 2.Among these loci, Acoh-1, Gpi, Hk-1/2 and Pgm-1/2 displayed the highest levels of polymorphism, whereas others, like Mdh-1 and G3pdh-1, showed only occasional variation.Electromorphs extremely different in their electrophoretic mobility were mostly restricted to Anergates, Teleutomyrmex and/or the outgroup taxa, sometimes even occurring at otherwise monomorphic loci.Several species of Tetramorium can be distinguished by diagnostic electromorphs which are most likely due to fixed allelic differences (see Table 3).At certain loci (es pecially Gpi, G3pdh-1, Mdh-1), all T. diotnedeum and T. ferox individuals had the typical banding patterns of het erozygotes, a phenomenon which deserves closer exami nation (see also Sanetra et al., 1999).These diagnostic enzyme variants proved to be a powerful tool for the reli able identification of Tetramorium worker samples that were often very difficult to classify as species morpho logically (see also Lopez, 1991a;Sanetra et al., 1994Sanetra et al., , 1999)).Immature stages of the parasitic genera Strongylognathus, Anergates and Teleutomyrmex may also be de tected among their host brood by using some of the loci reported in this study.In the late summer the larvae of A. atratuhis can be tentatively identified in the field because of their coloration and shape.However, in many cases these larvae cannot be easily reared until pupation, thus making electrophoretic tests effective for species confir mation.

Intraspecillc variability
The average percentage of polymorphic loci (P99 ± SE) was 27.9 ± 2.1 for the Palaearctic Tetramorium species, 14.7 ± 1.7 for the genus Strongylognathus and 16.7 ± 1.8 for the extreme inquilines Anergates and Teleutomyrmex.The highest mean number of alleles per locus (with a maximum of four alleles) occurred in Tetramorium impu rum (1.76 ± 0.056) and T. caespitum (1.71 ± 0.050), whereas fewer alleles were detectable in Strongy lognathus (1.16 ± 0.023), Anergates (1.29 ± 0.033) and Teleutomyrmex (1.05 ± 0.015).The estimates of heterozy gosity (Hexp ± SE) ranged from 0.064 ± 0.008 for the parasitic species to 0.093 ± 0.008 for the Tetramorium species.The P99 values of the social parasites were sig nificantly lower than those in the free-living Tetramorium species, as judged by the fact that the 95% confidence in tervals did not overlap.Confidence intervals of Hexp showed only a marginal overlap of 0.003.

Phenetic and phylogenetic inferences
UPGMA cluster analyses using four different genetic distances yielded very similar results.In fact, the use of Nei's distance and Roger's distance led to exactly the same topology.In a phenogram constructed with Nei's D (Fig. 1), the extreme inquilines Anergates and Teleuto myrmex attain D values >1 and are thus distinctly set apart from the remaining Palaearctic species.On the other hand, among Strongylognathus and Tetramorium species, pairwise genetic distance ranges from 0.19-0.92(on aver age 0.56) indicating closer relationships of these genera.Bootstrap analysis shows that all Strongylognathus spe cies are grouped with relatively low support, whereas the S. huberi and the S. testaceus group within Strongy lognathus are given higher support values.The nearest cluster attached to Strongylognathus comprises all Te tramorium host and non-host species sampled from the western Palaearctic.
Trees obtained from pairwise chord distances employ ing the algorithms of Fitch (Fig. 2) and Neighbor had to pologies virtually identical to one another.The lowest percent standard deviation (% SD) of branch lengths used as criterion for goodness of fit is 8.85 for the Fitch-Margoliash method and 10.61 for Neighbor-joining.The results of these rate-independent analyses share the gen eral pattern with the UPGMA phenogram, specifically that Anergates and Teleutomyrmex branch off separately near the basal node of the tree.The genus Strongy lognathus forms a distinct species assemblage in which  the S. huberi and the S. testaceus group are sister groups.Most of the species in the S. huberi group are genetically very similar (see branch lengths in Figs 1-2).In contrast to the phenogram obtained by UPGMA, Strongylognathus appears within the main cluster of Palaearctic Tetramorium, suggesting the group (chefketi + (forte + marocanum)) as its closest free-living relatives (Fig. 2).The branch including the two most commonly used host species, (caespitum + impurum), is placed near to Strongylognalhus.However, the relative positions of the Tetramorium clusters varied considerably (they occurred in less than 50% of bootstrap replicates), as indicated by missing values at the respective nodes.
Maximum parsimony analysis of discrete characters using independent allele coding resulted in six equally parsimonious cladograms.One of these was arbitrarily chosen to demonstrate optimization of cladogram charac ters (Fig. 4).The strict consensus tree (Fig. 3) shows sev eral features that have already occurred in some of the previously discussed trees, such as the monophyly of Strongylognathus and the relatively distant, but separate, positions of Anergates and Teleutomyrmex.Interrelation ships among Palaearctic Tetramorium species are only partially resolved into four major groups, which are equally closely attached to the Strongylognathus clade.Thus, according to this evolutionary model the sister group of Strongylognathus may either include all Te tramorium ingroup species or may be composed of differ ent subsets of those four groups (for a possible example see Fig. 4).The relative clade stability was estimated with the Jackknife Monophyly Index (JMI), which assigns a value to each clade according to its frequency of occur rence in pseudoreplicates.A comparison of single taxon removal (Siddall, 1995) further revealed that T. biskrense is the most problematic taxon, as its exclusion reduces the number of equally parsimonious solutions to only two trees.Then, a (chejketi + (forte + marocanum)) clade forms the sister group to Strongylognathus, as previously seen in the Fitch-Margoliash tree.
Parsimony analysis with minimum turnover coding re sulted in three cladograms, which were also stable to suc cessive weighting.The consensus tree (Fig. 5) is far bet ter resolved than that with independent allele coding (Fig. 3).On the former cladogram, Anergates and Teleutomyr mex are placed as sister genera, an interesting result which is supported by a relatively high JMI value.The nearby position of this clade to the ingroup species (T.semilaeve + T. diomedeum) is also remarkable.In accor dance with all other trees, Strongylognathus is monophyletic with a clear distinction between the S. testaceus and the S. huberi group.The species pair (T.moravicum + T. rhenanum) appears most closely attached to the Strongylognathus clade.However, JMI values indicate that the clades forming sister taxa to the parasites are less stable than those including the parasitic species.

Relationships within the Tetramoriini
Phylogenetic relationships among tetramoriine ants de rived from different tree building methods consistently re vealed that the species of the parasite genus Strongylognathus fall into one clade.This is not a sur prise because the characteristic saber-shaped mandibles, by which the genus is easily distinguished amongst the Tetramoriini, appear as an autapomorphy relative to Tetramorium.Thus, the electrophoretic evidence confirms that these peculiar mandibles are not merely an adaptive convergence to parasitism but constitute a reliable guide to phylogeny.Nonetheless there have been a few ac counts indirectly suggesting the polyphyly of Strongy lognathus.In a previous allozyme study with fewer characters, a single origin of Strongylognathus was not recovered (Sanetra et al., 1994).Baroni Urbani (1969) as serted that the Asian S. koreanus arose independently from the species of the western Palaearctic.However, from a morphological viewpoint there seems to be no jus tification to treat S. koreanus differently from the other species of the S. huberi group.Thus, the monophyiy of the genus Strongylognathus seems hardly disputable, as it has been corroborated by both biochemical and morpho logical characters.
The sister group relationship between the S. huberi and the S. testaceus group is of particular interest because two different types of parasitic relations occur within these groups.Obligate dulosis is predominant among species of the S. huberi group.The only known exception is S. kratochvili, which is supposed to live as an inquiline (Kratochvil, 1940), while the two species of the S. testaceus group evidently are true inquilines which tolerate the host queen (Sanetra et al., 1999;Sanetra & Buschinger, un published;Kipyatkov, pers. connn.).From the available information it seems clear that the inquilines in Strongy lognathus are derived from slave-making ancestors (e.g.Wasmann, 1905;Kutter, 1969;Acosta & Martinez, 1982), although this evolutionary pathway usually leads to queen-intolerant inquilinism (Buschinger, 1990).The transition from slave-making to queen-tolerant inquilines is unique among social parasites, and has previously been questioned for various reasons (Buschinger, 1986).Given the internal phylogeny of Strongylognathus presented in this study, the presumably dulotic ancestor (see above) must have split very early into one lineage maintaining dulotic behaviour and another giving rise to inquilinism.Accordingly, if the supposition that S. h-atochvili is an in quiline were true, inquilinism must have arisen two times independently in the evolutionary history of Strongy lognathus.The comparably high phylogenetic age of the inquilinous S. testaceus group is supported by morpho logical peculiarities and the wide range of the species S. testaceus.In comparison, the morphologically and geneti cally homogeneous species of the S. huberi group show the patchy geographic occurrence typical for social para sites.
(which means a wider spectrum of host species, see Table 4), especially in the S. testaceus group, suggests that at least some host shifts have occurred.
Since Strongylognathus has been established as a clade nested within the genus Tetramorium, the latter becomes paraphyletic.Hence, the question arises if the hosts and social parasites should be taxonomically separated.Many authors (e.g.Wilson, 1984;Carpenter et al., 1993;Ward, 1996) prefer to include hosts and parasites into a single genus when phylogenetic relationships appear as they do in this case.However, if all species included in this study were treated as a single genus, the reader not familiar with the Tetramoriini would have difficulties in keeping track of which species are the parasites and which are free-living.An aggravating problem in this case is that Strongylognathus Mayr, 1853 would have priority over Tetramorium Mayr, 1855, and the approximately 400 Te tramorium species then would have to be transferred to Strongylognathus.With respect to nomenclatural stability, wc suggest that Strongylognathus be retained as TAiii.it 4. Parasitic life habits and host species in the studied Palaearctic Tetramoriini.

Parasite
Type of parasitism Host species References for host species S. huberi dulosis .
Anergates and Teleutomyrmex, though radically different from Tetramorium in anatomical characters, provide simi lar cases in that their phylogenetic positions render the genus of their hosts paraphyletic.The placement of Anergates and Teleutomyrmex dif fered according to which evolutionary model was used.While distance methods show that Anergates and Teleuto myrmex split off separately near the base of the tree, par simony analyses using a hypothetical ancestor suggest a common origin within the remaining ingroup species.There is other evidence advocating the occurrence of common ancestry (see also character mapping in Fig. 6).First, two shared-derived base substitutions in the ITS-1 and 5.8S rRNA sequences have been reported by Baur et al. (1996).Secondly, both Anergates and Teleutomyrmex have developed a very similar form of workerless parasit ism including manifold adaptations, such as physogastry, reduced mouthparts and specialized males (Kutter, 1969;Bolton, 1976).Although the latter apply well to the "inquiline syndrome" (Wilson, 1971), which refers to a spe cific set of characteristics convergently acquired by many inquilines, this observation cannot be taken as a simple generalization about parasite evolution.Admittedly, each of the above-mentioned characters alone is of limited value, but in its entirety all available evidence supports a common evolutionary origin of Anergates and Teleuto myrmex.
Regardless of the method, the phylogenetic hypothesis in which Anergates and Teleutomyrmex are descendants of dulotic or inquilinous Strongylognathus ancestors is never supported.Therefore, in contrast to the suggestions of some earlier authors (e.g.Wasmann, 1905Wasmann, , 1909;;Kut ter, 1969), extremely degenerate inquilinism does not rep resent the final step in the evolution of dulotic tetramoriine parasites, but rather evolved directly from free-living Tetramorium species.This finding also re flects the general scarcity of transitions found among dif ferent types of parasitic relations in ants (Buschinger, 1990).For instance, secondary inquilinism in Strongylognathus refers to such a progression.If both Anergates and Teleutomyrmex indeed had a common ancestor, an evolutionary relationship between queen-tolerant and queen-intolerant inquilinism would be evident.Buschin ger (1990) suggested that Anergates might represent an inquiline having secondarily specialized on the exploita tion of already orphaned Tetramorium colonies.It seems unlikely, but not impossible, that queen-tolerance in Te leutomyrmex is derived from that latter strategy.In the formicixcnine genus Doronomyrmex it is also unknown which parasite type is ancestral (Bourke & Franks, 1991), queen-intolerant inquilinism in D. goesswaldi or queentolerant inquilinism in D. pads and D. kutteri.
The extreme inquilines Anergates and Teleutomyrmex are somewhat enigmatic parasites because current theo ries in evolution do not seem consistent with at least some of the molecular data.Evolutionary rates of the enzyme loci appear very high compared to their Tetramorium hosts.This is particularly surprising in light of the low di vergence of ITS-1 sequences in this group (Baur et al., 1996).A number of studies have shown that rates of mo lecular evolution vary considerably among hosts and parasites belonging to different taxonomic groups (e.g.Baverstock et al., 1985;Flafner et al., 1994), but are sup posed to be stable among closely related species with similar life histories (see Hillis et al., 1996).Therefore, extreme specialization of parasites and high dependence on their hosts may affect patterns of molecular evolution.One would expect that certain selection pressures on parasites might be reduced, because they spend nearly their whole life cycle in the stable conditions of a hosts' nest.Under these circumstances increased accumulation of mutations could occur in some parts of the genome.This idea is indeed supported by the large genetic dis tances relative to Tetramorium, according to which the origin of Anergates and Teleutomyrmex would date back to a very early stage in tetramoriine evolution (see Figs 1-2).On the other hand, geographic allozyme variation and heterozygosities are considerably lower than in the free-living Tetramorium species which suggests that posi tive selection has occurred.The adaptive value of allozymes has been demonstrated in several groups of insects (e.g.Barnes & Laurie-Ahlberg, 1986;Watt et ah, 1986) and at least correlated phenotypic effects are known to occur within ant colonies (Keller & Ross, 1993).The phenomenon of relatively little geographic variation also applies to the striking constancy of morphological fea tures, especially in Anergates which is distributed throughout the western Palaearctic (Bolton, 1976).Again, this sharply contrasts with the troublesome morphological variability in their Tetramorium hosts.
Multiple origins of parasitism in the tetramoriine phylogeny could clearly be demonstrated to be independent of the kind of analysis, whereas relationships among Te tramorium species remained largely ambiguous.The lat ter finding was obtained by comparing many different methods and the use of resampling techniques.Four main species assemblages of Tetramorium were nonetheless grouped together very frequently (as an example see the consensus tree in Fig. 3), which are (chefketi + (forte + marocanum)), (punctatum + (brevicorne + meridionale), (caespitum + impurum) sometimes together with (moravicum + rhenanum), and (diomedeum + semilaeve) two times attached to ferox.The most straightforward reason for this lack of accuracy is that there are not enough in formative characters to provide complete resolution of trees (Swofford et ah, 1996).Computer simulations re cently reported by Hillis (1996) suggest that including large numbers of taxa in an analysis may be another way to enhance phylogenetic accuracy.Additional taxa have the effect of shortening the branches (according to the problem of long branch attraction stated by Felsenstein, 1978) which makes the phylogenetic signal better distin guishable from the noise of homoplasy (Hillis, 1996;Pur vis & Quicke, 1997).
The presently known host range of Tetramorium para sites in Europe and adjacent regions is summarized in Ta ble 4. It shows a prevalence of T. caespitum and T. impurum, but there are still significant gaps in the docu mentation of host use which render the establishment of a complete list of host-parasite associations impossible.In addition, due to common misidentifications of Tetramo rium species, several questionable records have been pre sented.A comprehensive revisionary work on the Tetra morium fauna of the Palaearctis is much needed since only a few partial studies of varying quality have been published (Lopez 1991a;Radchenko 1992a, b;Sanetra et ah, 1999).Many of the described taxa currently residing in species or subspecies rank (listed by Bolton, 1995) are likely to constitute no more than intraspecific forms.For instance, the proximity of the morphologically similar T. forte and T. marocanum in phylogenetic trees suggests their taxonomic synonymy.By contrast, the extensive survey of enzyme polymorphism in the genus Tetramo rium led to the detection of some differentiated subpopu lations (not included in this study) which might represent yet unrecognized, cryptic species (for examples see Sane tra et al., 1999).

Origin of social parasitism
The results of this study do not support Emery's rule in a strict sense because neither Strongylognathus nor Aner gates nor Teleutomyrmex are most closely related to a sin gle host species.Strongylognathus shows close phyloge netic affinities to the Palaearctic Tetramorium species, but exact sister group relationships are not well resolved.Yet, closer relations of Strongylognathus to T. caespitum, T. moravicum and T. forte (and some related species) are supported by their frequent occurrence in neighbouring branches.Recent DNA studies had already revealed An ergates and Teleutomyrmex to be relatively closely re lated to their tetramoriine hosts (Baur et al., 1996).In the present study they were placed as successive outgroups of the remaining Palaearctic species when distance methods and parsimony with independent allele coding were used.In contrast, parsimony analyses involving a hypothetical ancestor suggest relationships to T. diomedeum and T. semilaeve, which are not among their regular host species.These considerations, however, are complicated by the fact that tetramoriine parasites usually have two or more hosts (see Table 4).The results thus do not show if all host species of each parasite lineage are included in the respective sister groups.
In the Palaearctic Tetramoriini there exist at least two independent origins of social parasitism, which could have been the result of different types of speciation (e.g.Holldobler & Wilson, 1990).The interspecific hypothesis of parasite evolution' suggests colonization and exploita tion of congeners that have arisen allopatrically (Wilson, 1971;Holldobler & Wilson, 1990).In contrast, the intra specific hypothesis assumes direct development of social parasites from their host stock, invoking the highly de bated evolutionary process of sympatric speciation (Buschinger 1986(Buschinger , 1990;;Bourke & Franks, 1991).An in teresting synthesis of both these hypotheses is that intra specific parasites could somehow start to use a different species as a host and subsequently attain reproductive iso lation from its non-parasitic parent species in sympatry (Bourke & Franks, 1991;Heinze, 1998; see also West-Eberhard, 1986, 1990).This evolutionary concept bears much resemblance to sympatric speciation after host shifts in phytophagous insects (Bush 1975(Bush , 1994;;Ridley, 1996).It should be noticed, however, that the interspe cific hypothesis as originally proposed by Wilson (1971) relies on the most questionable premise that a species can switch to parasitism without phyletic change.
We now explicitly discuss the three above-mentioned hypotheses separately from the general considerations about phylogenetic relationships, because, as recently pointed out by Lowe & Crozier (1997), these relation ships alone need not be decisive regarding the evolution of social parasitism.To paraphrase, whether or not Em ery's rule applies in a strict sense does not necessarily in dicate the underlying mode of speciation, as has been claimed by some authors (Carpenter et al., 1993;Choudhary et al., 1994).However, the intraspecific hypothesis requires the nearest non-parasitic outgroup of the parasite(s) to be a clade that includes the host species (Buschinger, 1990;Ward, 1996).This relaxed interpreta tion of Emery's rule has recently become equated with the strict version of the rule (Ward, 1996;Heinze, 1998), but the desired unambiguous definition of Emery's rule throughout the literature is made very difficult by the syn onymous usage of both versions.For clarity, it might be better to distinguish between sister species or sister groups in host-parasite relationships.
Because we cannot exclude the possibility that the tetramoriine host and parasite lineages are sister groups, there is no reason to reject the intraspecific hypothesis.The logic of Carpenter et al. (1993) andChoudhary et al. (1994) who state that monophyly of a group of parasites is inconsistent with sympatric speciation cannot be ac cepted (see also Lowe & Crozier, 1997).Particularly, the inquilines Anergates and Teleutomyrmex may have evolved from intraspecific parasites of their ancestral host species, since inbreeding, nest-mating and polygyny are expected under the sympatric speciation model (Buschin ger, 1990).Bourke & Franks (1991) also concluded that intraspecific parasitism and sympatric speciation played an important role in the evolution of inquiline ants, but some cases of Emery's rule not applying (sister group of the parasite lacks the host species) have been proposed for extremely rare inquilines in the genera Cataglyphis (Agosti, 1994), Pseudomyrmex (Ward, 1989, 1996) and Leptothorcix (Heinze, 1998).Though the data on host use of these parasites are very scant, they were taken as the basis for questioning the general occurrence of sympatric speciation in ants (Ward, 1996).However, host-parasite relationships not obeying Emery's rule can also be ex plained by host transfer with subsequent sympatric spe ciation (West-Eberhard, 1990;Bourke & Franks, 1991;Heinze, 1998).Yet this most realistic evolutionary model, which is a synthesis of both competing hypotheses, has received only little attention in recent discussions on the subject (e.g.Choudhary et al., 1994;Ward, 1996).

Tree building methods
Distance methods have long been used for inferring phylogenies from electrophoretic data (e.g.Buth, 1984;Berlocher, 1984;Heinze, 1998), but their efficiency at ac complishing this task has been questioned for a variety of reasons (e.g.Farris, 1981;Carpenter, 1990).We em ployed UPGMA, Neighbor-joining and least squares sta tistics, whose powers in phylogenetic reconstruction must be viewed differently.The UPGMA algorithm, better re ferred to as phenetic clustering, is unequivocally the least preferable of these methods (e.g.Saitou & Nei, 1987;Carpenter, 1990), because the underlying assumption of a molecular clock has been seriously undermined (e.g.Avise & Aquadro, 1982;Hillis et al., 1996).Neighbor joining and least squares statistics do not rely on rate con stancy (Felsenstein, 1984(Felsenstein, , 1997)).The latter of these methods is considered the most effective at recovering simulated phylogenies (Kuhner & Felsenstein, 1994).
Several authors have stated that transforming character data to genetic distances results in a considerable loss of information (e.g.Farris, 1981;Patton & Avise, 1983;Swofford et al., 1996).Alternative methods of character analysis based directly on the distribution of electrornorphs (e.g.Mickevich & Johnson, 1976;Mickevich & Mitter, 1981;Richardson et al., 1986) usually reconstruct trees under the criterion of maximum parsimony.Though conceptually simple, parsimony analyses belong to a group of theoretically more defensible techniques (e.g.Farris, 1981Farris, , 1983;;Carpenter, 1990).However, these may be statistically inconsistent when parallel changes outnumber non-parallel ones (Felsenstein, 1978).For ana lysing electrophoretic data, the most serious problem for character-based methods is their high susceptibility to missing rare alleles by sampling error (e.g.Swofford & Berlocher, 1987;Swofford et al., 1996).
The results of this study may be viewed differently with respect to the above-mentioned principles of phylogenetic inference.Conclusions derived from distance methods could be misleading in the case of Anergates and Teleutomyrmex, because of considerable evolutionary rate varia tion within the group.The large amount of autapomorphic characters in these extreme inquilines has a strong influ ence on the resulting distance trees, but does not affect parsimony analyses.In turn, synapomorphies are of much lower consequence for distance analyses.
Coding of polymorphic characters, such as allozymes and ontogenetic variants, has always been problematic and has therefore become a major subject of debate (e.g.Mickevich & Mitter, 1981, 1983;Buth, 1984;Swofford & Berlocher, 1987;Mabee & Humphries, 1993;Murphy, 1993;Wiens, 1995).The frequency parsimony algorithm proposed by Swofford & Berlocher (1987) was not used in this study, because many objections have been put for ward to the assumption that allele frequencies can be treated as heritable characters (see Crother, 1990;Murphy, 1993).We also did not use multistate locus cod ing, because it offered much ambiguity in similar applica tions (e.g.Carpenter et al., 1993;Wiens, 1995) and the assumption of non-additive transformation is unlikely for allelic arrays (e.g.Mabee & Humphries, 1993;Swofford et al., 1996).From the coding formats applied to our data set, independent allele coding led to several unresolved poly tom ies in the consensus tree, while in a comparable study of Carpenter et al. (1993) this coding method pro vided best resolution of the taxa examined.However, the characters are logically linked in their loci and inferred ancestors on the cladograms are assigned no alleles for some loci making this approach biologically unrealistic (e.g.Mickevich & Mitter, 1981;Swofford & Berlocher, 1987;Murphy, 1993).Minimum turnover coding (being similar to the "scaled" method of Mabee & Humphries, 1993 but here expanded according to Mardulyn & Pas teéis, 1994) performed relatively best in terms of resolu tion of the consensus tree.
The main problem common to all the previously men tioned coding procedures is that they equate gains with losses in terms of evolutionary steps.Murphy (1993) pro posed a coding method that considers only new mutation events as well as a stepwise evaluation procedure for ana lysing allozyme data.An important characteristic of the latter method is that clades can be divided into different categories with respect to their reliability.In the order of reliability, branchings are derived from shared mutation events, shared losses of alleles or from unordered parsi mony analyses.We used this conceptual framework in combination with all other evidence available, involving morphology and life-habits.It has been shown that the principle of using all relevant information in a combined analysis should be the strategy of choice, as opposed to comparing independently generated trees (total evidence versus taxonomic congruence; see Kluge, 1989Kluge, , 1998;;Nixon & Carpenter, 1996).The result based on simulta neous analysis of allozymes and non-molecular characters with Murphy's (1993) method is therefore a particularly interesting representation of the tetramoriine data, be cause the underlying assumptions are the most realistic in an evolutionary sense.
In this study various tree building methods were em ployed to analyse a single data set (see also e.g.Avise et al., 1994;Clabaugh et al., 1996;Van der Bank & Kramer, 1996).Given the controversial discussions about different tree building methods in the literature, we deem it insuffi cient to rely on a single method only (see also Hillis et al. 1996).To avoid this shortcoming we compared the results from alternative procedures in order to rule out analysisdependent effects, and we regard phylogenetic arrange ments consistently resolved by all methods as the most reliable.Even if conservative summaries of phylogenetic relationships as presented here often fail to resolve many of the taxa examined, such approaches should neverthe less become more frequently applied.

*
Very rare, could be due to hybridization with T. impurum.

Fig. 5 .
Fig. 5. Strict consensus tree of three cladograms (1210, ci 54, ri 66) resulting from parsimony analyses using minimum turn over coding and a hypothetical ancestor.Jackknife monophyiy indices (Siddall, 1995) are shown at each node to indicate rela tive stability of clades.

Table 2 .
Allele frequencies at the polymorphic loci investigated (values <0.005 are reported exactly, while all others are rounded).