Molecular phylogeny of telenomine egg parasitoids (Hymenoptera: Platygastridae s.l.: Telenominae): evolution of host shifts and implications for classification

Parasitoid wasps of the subfamily Telenominae (Hymenoptera: Platygastroidea, Platygastridae) develop as immatures within the eggs of other insects (Lepidoptera, Hemiptera, Diptera and Neuroptera). Rearing records indicate that individual species are restricted to attack hosts within only one of these four main groups. We conducted a phylogenetic analysis of the group using sequence data from multiple genes (18S, 28S, COI, EF‐1α) to assess the pattern of shifts among host groups and to test the monophyly of and relationships among genera and species‐groups. Telenominae sensu Masner—that is, including only the nominate tribe Telenomini—is not monophyletic. Representatives of the Psix group of genera (Psix Kozlov & Lê and Paratelenomus Dodd) form a monophyletic group that is sister to Gryon Haliday (Scelioninae: Gryonini) and are excluded from the subfamily. The remaining telenomines are monophyletic. The genus Phanuromyia Dodd and the crassiclava group of Telenomus Haliday, both recorded as parasitoids of planthopper eggs (Hemiptera: Auchenorrhyncha, Fulgoroidea), form a monophyletic group that is sister to all other telenomines exclusive of the Psix group. Twenty‐nine species of the crassiclava and aradi groups of Telenomus are transferred to Phanuromyia as new combinations. Basal elements of the remaining species are all in groups reared from the eggs of true bugs (Heteroptera), primarily the stink bugs (Pentatomoidea) and seed bugs (Lygaeoidea). A shift to parasitism of lepidopteran eggs evolved within a single clade, occurring either one or two times. From this clade a small group of species, the Telenomus tabanivorus group, subsequently shifted to parasitism of egg masses of true flies (Tabanidae and Stratiomyiidae). Aholcus Kieffer and Platytelenomus Dodd both belong to the clade of lepidopteran parasitoids and are considered as junior synonyms of Telenomus (new synonymy for Aholcus). The monophyletic status of the two core genera, Telenomus and Trissolcus could not be resolved using these data. The phylogenetic pattern of host shifts suggests comparisons among taxa that may be fruitful in elucidating mechanisms by which parasitoids locate their hosts, the proximate factors that determine the host range, and the changes in these factors that influence host changes.


Introduction
The mechanisms and evolution of host finding among parasitoid insects is an area of great significance given the important role such species play in regulating the populations of their hosts, both naturally and in biological control programmes. Wasps of the superfamily Platygastroidea (Hymenoptera; Fig. 1) are idiobiont endoparasitoids of the eggs of insects and spiders (Scelioninae, Teleasinae and Telenominae) or koinobiont parasitoids of the immature stages of Auchenorrhyncha (Hemiptera), Sternorrhyncha (Hemiptera) or Cecidomyiidae (Diptera) (most Platygastrinae and Sceliotrachelinae). The major groups within the superfamily show a clear pattern of host group specificity (Austin et al., 2005). However, because of the current paucity of well-supported hypotheses of phylogenetic relationship, the sequence of shifts from one host taxon to another is uncertain. The many species of the subfamily Telenominae are particularly diverse in their host relationships, having been recorded from Lepidoptera, Diptera, Neuroptera and the suborders Heteroptera and Auchenorrhyncha within the Hemiptera (summarized in Kozlov & Kononova, 1983;Johnson, 1984a). Several species of telenomines have been shown to find their egg hosts by cuing in to semiochemicals produced by the adult host, in some cases synergized by chemicals released by the plant upon which the host feeds or oviposits (for example, Colazza et al., 2009;Moraes et al., 2009;Conti et al., 2010;Arakaki et al., 2011;Peñaflor et al., 2011). No telenomine species, to our knowledge, has been reared from hosts from more than one order. Therefore, the change from, for example, Heteroptera to Lepidoptera (or vice versa), seems to be an evolutionarily 'difficult' and significant step. Insights into how it takes place may emerge from comparisons of the chemosensory capabilities of closely related species of wasps that attack the two different groups of hosts. The goal of this work is to contribute to a better understanding of relationships among telenomines to facilitate such work and, in the process, test the monophyly of its genera, species groups and the subfamily itself.
We report here the results of phylogenetic analyses of an extensive species sampling of Telenominae based on sequence data from four molecular markers: 18S and 28S rDNA, the mitochondrial protein-coding gene COI , and the nuclear protein-coding gene EF-1α F2 copy. Also included in the analyses were a wider range of species of Gryon, designed to test the putative sister-group relationship or, if possible, to test whether Telenominae are more closely related to a subset of Gryon species. The results serve as a framework for interpreting the evolution of host relationships within the subfamily and, in addition, have repercussions for the classification of Telenominae.

Taxonomic sampling and specimen vouchering
A total of 62 species were sequenced: 51 species for the ingroup Telenominae, and 11 species in the outgroups (File S1). The ingroup comprises representatives of six genera: Telenomus (eight species groups), Trissolcus (three species groups), Psix (three species), Paratelenomus (one species), Phanuromyia (two species) and Eumicrosoma (one species). The generic names Platytelenomus Dodd and Aholcus Kieffer are still recognized as valid by some authors (e.g. Kononova, 2008;Mineo et al., 2011). Representatives of each were also included: Platytelenomus sp1 and Telenomus dalmanni (Ratzeburg), respectively. Other telenomine genera were not sampled because they are fossil taxa (e.g. Sinoprotelenomus Zhang) or are rare, and fresh material was unavailable. Within Trissolcus all three recognized major groups (Johnson, 1984b(Johnson, , 1985a Johnson (1984a). Outgroup taxa include four genera of Scelioninae -six species of Gryon, two species of Scelio Latreille, and one species each of Oxyteleia Kieffer and Probaryconus Kieffer -and one species of Teleasinae, Trimorus caraborum (Riley). Material studied was freshly collected from around the world.
After nondestructive DNA extraction the voucher specimens were deposited in the C. A. Triplehorn Insect Collection (OSUC). Supplementary File S1 lists the specimens, voucher information and GenBank accession numbers for markers included in the analyses.

DNA extraction, amplification, and sequencing
Fresh specimens were sorted from bulk material, preserved in 95% ethanol, and maintained in the freezer at −15 • C. Nondestructive DNA extraction was performed using the DNeasy extraction protocol (Qiagen, Germantown, MD; cat. num. 69506) as modified by C.D. Zhu and J.S. Noyes (unpublished data): Individual specimens were initially softened in 70% ethanol at room temperature for 1 day. Vortexing in step 2 of the manufacturer's protocol was modified by mixing the reaction gently and incubating at 55 • C for 24 h. The mixture was stored at −20 • C for 24 h. The intact specimen was then removed from the tube and prepared for standard mounting. The reaction was incubated for 10 min at 70 • C after addition of Buffer AL. Then 200 μL of cold ethanol (96-100%) was added to the supernatant. Finally, in Step 7 the Buffer AE was warmed to 55-70 • C before addition.
Nested PCR optimizations were employed to amplify the F2 copy of EF-1α. This approach is relatively successful in reducing or eliminating unwanted products, although it dramatically increases sensitivity to contamination. The initial PCR primers were F2F and Cho10 (Danforth et al., 1999) followed by a combination of other primers from Heraty et al.  Klompen (2000) except that annealing temperatures of 54 • C, 54 • C and 58 • C were used (respectively). The amplification profile for EF-7 and EF-9 followed Simon et al. (2010). We excluded the F1 copy from the analysis because of the difficulty of amplification and the often incomplete products (as in Heraty et al., 2011). Contaminants were detected using negative controls in all rounds of PCR, as well as comparisons of the product sequences with each other and with a range of published arthropod sequences. PCR products were purified either using the QIAquick PCR purification kit (Qiagen) protocol or done prior to sequencing by Beckman Coulter Genomics (Danvers, MA). Products were sequenced in both directions and assembled using Sequencher v4.0 (Gene Codes Corporation, Ann Arbor, MI).

Sequence alignment
Sequences of COI (approximately 857 bp) were aligned initially using MUSCLE (Edgar, 2004) with default settings and then adjusted by eye. The total of 1124 bp of EF-1α were aligned by eye against a reference taxon (Archaeoteleia mellea -GenBank:GQ410731.1) and reported intron positions (Danforth & Ji, 1998;Heraty et al., 2011). Alignments of both protein-coding genes were translated to amino acids to verify the homology using Mesquite (Maddison & Maddison, 2011). Secondary structural alignments were implemented for ribosomal RNA sequences of 18S and 28S for a total of 1783 bp. The by-eye alignment conventions followed Kjer (1995) with slight modifications (Gillespie, 2004). The core alignments follow published secondary structure models of Hymenoptera (as in Gillespie et al., 2005;Murphy et al., 2007;Heraty et al., 2011). Ambiguous homology positions were defined across all taxa based on the structural criteria of Kjer (1995). These ambiguous regions were excluded from the final analyses. The final matrix alignment for analyses contains 3767 bp (COI : 857, 18S : 1003, 28S : 782, EF-1α: 1125). The alignment summary, gene partitions, nucleotide composition and percentage of parsimony-informative sites are presented in online supporting documentation.

Data coding and partitioning
Regier & Zwick (2011) discussed the role of heterogeneity in base composition and associated synonymous nucleotide change in protein-coding genes in reducing the performance of phylogenetic analyses, and they proposed mechanisms to attempt to compensate for this effect. Here we employ three strategies to reduce the effect of synonymous changes. The third codon position (nt3) is typically the source of most synonymous change and, therefore, one option is to eliminate (or completely degenerate) these data from the analysis. Additionally, synonymous changes in the first codon position are possible for sequences coding for arginine and leucine. The DEGEN approach (Regier & Zwick, 2011) degenerates synonymous nucleotide changes in both the first and third codon positions. The Perl code for DEGEN coding was modified to use the invertebrate mitochondrial genetic code for COI . The third strategy is to segregate data from nt3 in a separate partition.

Phylogenetic analyses
Three phylogenetic estimation strategies were implemented to infer relationships: maximum parsimony, maximum likelihood and Bayesian approaches. Maximum parsimony analyses were conducted in TNT v1.1 (Goloboff et al., 2008). Heuristic tree search algorithms were implemented using the New Technology Search by default setting with some modifications followed by tree bisection-reconnection (TBR) branch swapping. Some parameters were adjusted: using a sectorial search, treating gaps as missing data, tree drifting of four cycles, tree fusing of five rounds and best score hit of 20-25 times. The branch support was estimated using the Parsimony Jackknife (JK; Farris et al., 1996) executed with the setting 36% probability removal, 'emulate JAC' resampling, 1000 replications, 'random addition sequences' = 1, and 'hold trees' = 2 (Freudenstein et al., 2004;Klompen et al., 2007). The parsimony jackknife was selected over the alternative, bootstrapping, because it is less sensitive to missing and invariant characters (Farris et al., 1996;Freudenstein et al., 2004).
Maximum likelihood analyses were executed using RAxML v7.2.6 (Stamatakis, 2006;Stamatakis et al., 2008). By using GTRGAMMA for the model of nucleotide substitution (-m), individual α-shape parameters, GTR-rates and empirical base frequencies were estimated and optimized for each partition. RAxML analyses were implemented using a rapid bootstrap and search for the best-scoring ML tree in one single program run. Heuristic hill-climbing tree searches were performed generated from 1000 distinct randomized maximum parsimony starting trees and computed for 1000 repetitions. An additional ten runs were made using random starting seeds: all resulted in the same topology and likelihood scores.

Relationships
The results are largely congruent among the three inference techniques. The parsimony results are, in general, less wellresolved and less well-supported. The level of support for clades discussed below in different analyses are presented in File S3. Figure 2 illustrates the results of the maximum likelihood analysis of the single partition DEGEN matrix.
The representatives of the two genera of the Psix group -Psix and Paratelenomus -are individually monophyletic, and the two genera are also consistently grouped together in all analyses (parsimony jackknife support 99-100%, maximum likelihood bootstraps 100%, Bayesian posterior support 100%). Although Paratelenomus is represented by a single species, the samples span its geographic distribution from Ghana to Thailand. Together, Psix , Paratelenomus and Gryon are monophyletic (bootstraps from maximum likelihood analyses varying from 68 to 75%, Bayesian posterior probability 99%).
Telenomus (Aholcus) dalmanni (Ratzeburg) and Platytelenomus always group with the species of the Te. californicus complex. The position of Eumicrosoma is unstable: it emerged either with the californicus complex (most analyses) or as sister to Trissolcus ogyges Dodd (Bayesian analysis, DEGEN coding, four partitions).
Neither Telenomus nor Trissolcus are resolved as monophyletic groups, even with the exclusion of the crassiclava species group from the former. Representatives of the thyantae group of Trissolcus emerge near the base of the Telenomus + Trissolcus + Eumicrosoma clade, but none of the three species groups of Trissolcus are monophyletic. The species assigned to the podisi and floridanus groups of Telenomus do not group together. The floridanus group is particularly dispersed, with one species, Te. FLO1, consistently placed at the base of the Telenomus + Trissolcus + Eumicrosoma clade (but with low support values).

Scope of Telenominae
Telenominae has been one of the best-defined groups within the superfamily Platygastroidea since the subfamily was first formally recognized by C.G. Thomson (1860). It has been characterized by the combination of the following characters: metasoma with wide laterotergites loosely attached to the sterna; the absence of laterosternites on the metasoma; the second metasomal segment is significantly longer than any other; and the number of antennomeres in the female is reduced (10 or 11) in comparison to the male (12). Kozlov's (1970) inclusion of Tiphodytini and Aradophagini was based solely on the broad laterotergites in these two (then) monobasic tribes, but these species possess laterosternites, and the antenna is 12-segmented in both sexes (Masner, 1972;Masner & Huggert, 1979). Masner's (1976) restriction of the concept of the subfamily to the nominate tribe restored a clear definition of the taxon.
We believe that the results presented here strongly indicate that the subfamily, even in this restricted definition, is not monophyletic. We did not anticipate this outcome because the genera of the Psix group have the four synapomorphies of Telenominae. However, the position of these genera within the subfamily has been problematic. They all share a distinctive and plausibly apomorphic character: the central keel on the frons forks ventrally to pass on either side of the antennal insertions. Additionally, they all have strongly developed fanlike striae on the frons arising near the anterior mandibular articulation (very rarely present and only weakly developed in a few other telenomines), have a different structure of the metascutellum (the dorsellar lip of Johnson & Masner, 1985), and lack differentiated sublateral setae on the first metasomal   tergite. Mikó et al. (2007) discovered that the two groups differ in that the pleural apodeme of the mesothorax is fused with the anterior margin of the speculum in Telenomus and Trissolcus, whereas in Psix and Paratelenomus the apodeme extends to the mesopleural articulation. Finally, the Psix group of genera lack two previously unreported, apparent synapomorphies of the other telenomines. First, the metapostnotum is delimited posteriorly by a deep sulcus that arises anteriorly from the propodeal spiracle and continues posteriorly and medially beneath the metascutellum (see, e.g., figs 18, 48 in Johnson, 1984a). Reexamination of newly collected specimens of Mudigere confirms this character, contradicting the statement in Johnson (1988a). Second, the labrum is large, visible externally, articulated with the external ventral margin of the narrow clypeus (i.e. not on the internal surface within the buccal cavity), and is almost always separated from the clypeus by a distinct suture. The molecular data not only reaffirm the unity of the Psix group, but indicate that they are more closely related to Gryon than to other telenomines. Thus, the defining characters of the subfamily must be reinterpreted as having evolved independently in the Psix group of genera and in the Telenominae s.s. Most of these features are also found in other lineages within the superfamily. The reduction in number of female antennomeres is also found in the Gryonini genus Maruzza Mineo, and commonly occurs in other parts of the subfamily Scelioninae. Wide laterotergites are characteristic of species of Tiphodytes and Aradophagus, as well as males of Baeus Haliday. The size dominance of the second metasomal segment is also found among some species of Gryon, in more distantly related genera such as Yunkara Galloway and Baeus, as well as in the subfamilies Platygastrinae and Sceliotrachelinae. The last remaining character is the loss of laterosternites in the metasoma. This is, we believe, one strong piece of morphological evidence supporting the monophyly of Telenominae sensu Masner because it occurs so rarely elsewhere in Platygastroidea. However, it stands in stark contrast to the molecular evidence we present here.
We conclude that the taxonomic concept of Telenominae must be modified. Two options are available, consistent with our desire that any formally recognized taxon be monophyletic. One is to expand the concept of Telenominae to include the genus Gryon and, by implication, closely related genera. The second is to exclude Psix and Paratelenomus -and by extension the closely related Mudigere and Nirupama -from the subfamily. We prefer the latter option. From a practical point of view, to include Gryon in Telenominae would put us in the position of having no morphological characters by which to recognize the subfamily. Furthermore, the design of this study may be inadequate to support such a conclusion. The analyses lead to the conclusion that the Psix group is more closely related to species of Gryon than to other species of Telenominae, but do not necessarily demonstrate that Gryon, out of all other genera of Platygastroidea, is the group to which it most closely related overall. We therefore prefer the option of a further contraction and refinement of the subfamily Telenominae. In this concept the subfamily can be defined on the basis of the presence of broad laterotergites, absence of laterosternites, sexual heterogeneity in antennomere number, as well as the presence of externally demarcated metapostnotum and labrum. We consider the Psix group of genera to belong to the subfamily Scelioninae.

Phanuromyia
Within Telenominae s.s. the data support the merging of Phanuromyia together with the species of the crassiclava group of Telenomus. Phanuromyia languished in obscurity for nearly 90 years after it was originally described (Dodd, 1914a) because it had been defined only on the basis of the exserted ovipositor (a highly variable character even among individuals of the same species) and the type species, P. rufobasalis, had not been studied. Johnson & Musetti (2003) offered additional characters (claval formula, head shape, absence of a hyperoccipital carina, orientation of the malar sulcus and size of the gena) that could be used to recognize and define Phanuromyia, specifically to distinguish it from the crassiclava group. Mineo (2006) subsequently rejected this on the grounds that these characters ' . . . are generally used to advocate a species group of either Telenomus or other scelionid wasps. ' We cannot adopt the implication that some characters are a priori acceptable to define genera whereas others are only of value in defining a group of species. The critical issues are whether a new taxon is at least plausibly monophyletic and what effect the recognition of the new taxon would have on the monophyly of other recognized genera. Phanuromyia, now in an expanded concept to include the crassiclava group, is demonstrated in these analyses to be a monophyletic group, and its recognition does not result in other existing genera becoming paraphyletic. To follow Mineo's position of synonymizing Phanuromyia with Telenomus, while maintaining monophyly, would require the inclusion of all of the genus Trissolcus as well as any related genera. We believe that the requirement of monophyly is the only objective criterion by which to choose among alternative classifications. As such, it seems that the best option at this point is to treat Phanuromyia as a valid genus. We could treat the crassiclava group as an independent genus, but with this level of taxon sampling, it seems more prudent to recognize only a single genus.
Most species of Phanuromyia may be morphologically recognized among Telenominae s.s. by the presence of a welldeveloped sternaulus on the mesepisternum. Typically this is indicated as a line of deep foveae (the episternal foveae of Johnson, 1984a), but sometimes only as a crease in the sclerite. In this genus the sternaulus arises anteriorly near the dorsal apex of the acetabular carina and continues obliquely dorsally toward the mesepisternal pit. The great majority of species may also be recognized as Phanuromyia by the convex frons (frontal depression limited to a small area just dorsal of the antennal insertions), the presence of two parallel dorsoventral lines of setae below the median ocellus, eyes glabrous or with very short setation, second metasomal tergite with shallowly incised reticulate sculpture posterior to the basal longitudinal costae.

Status of Aholcus
Whereas females over 90% of telenomine species have 11-segmented antennae, 70 species of Telenomus have been described in which this number is reduced to 10. Kieffer (1913) created the genus Aholcus for a single species from Kenya, A. monticola Kieffer. The same year, Dodd (1913) erected the genus Neotelenomus for five Australian species. In both cases, the only character differing from Telenomus was the number of female antennomeres. Nixon (1935) synonymized Aholcus with Telenomus, but later reversed his finding and treated it as a valid subgenus (Nixon, 1937). In the latter paper he synonymized Neotelenomus with Telenomus (Aholcus). Since then, Aholcus has been used as a valid name, both as a subgenus and, occasionally, as a genus in 20 publications (including Nixon, 1937) for 31 newly described species. Johnson (1984a) suggested that Aholcus may be polyphyletic and in subsequent papers has not adopted the rudimentary subgeneric classification (currently there are only two subgenera recognized).
Aholcus is represented in our study by the common Holarctic species Telenomus dalmanni (Ratzeburg), a parasitoid of the eggs of the rusty tussock moth, Orgyia antiqua (Linnaeus) (Lepidoptera: Lymantriidae). In all of the data coding, partitioning and analysis schemes employed, this species consistently clusters with those of the californicus complex of Telenomus. Our taxon sample did not include T. monticola (Kieffer), so our results do not directly indicate how this species -the type species of Aholcus -is related to other telenomines. We can conclude, though, that recognition of a genus defined solely on the basis of the reduced number of antennomeres in the female sex would make Telenomus paraphyletic. Furthermore, the male genitalia of T. dalmanni (see Johnson, 1984a) are very distinctive, having a spectacularly long aedeagal lobe. This feature is not characteristic of all Aholcus species (see illustrations in Nixon, 1935Nixon, , 1937, but is found in some species of Telenomus s.s. in which the females may have either 10-or 11-segmented antennae (e.g., . This strongly suggests that the reduction in antennomere number has occurred in parallel among different parts of Telenomus in the broad sense and that Aholcus is a polyphyletic taxon. Although we have no molecular data for Telenomus monticola itself, its relationship to the lepidopteran parasitoids is corroborated by a previously unpublished morphological character. Specifically, the ventral portion of the occipital carina is deflected anteriorly around the posterior articulation of the mandible and is visible laterally (see fig. 31 in Johnson, 1984a). In conclusion, we consider Aholcus Kieffer, 1913 to be a junior synonym of Telenomus Haliday (syn.n), 1833 and reject its use as a subgenus.

Status of Platytelenomus
The name Platytelenomus is less commonly encountered than Aholcus, but its history is similar in many respects. The genus was originally described by Dodd (1914b) for an Australian species in which the body is strongly flattened dorsoventrally. Johnson (1988b) proposed that the same name had been used for two distinct groups of species: one, represented by Dodd's type species, P. planus, has elongate clavomeres and corresponds to the floridanus group of Telenomus; the second has transverse clavomeres and parasitizes the flattened eggs of grass-feeding Lepidoptera (hence the flattened body). The species used in this study, on the basis of morphology, falls in the latter group. Consistent with Johnson (1988b), this species clusters with the other moth parasitoids of the californicus complex, and not with the species of the floridanus group. Therefore, we reaffirm the conclusion of Johnson (1988b) and consider Platytelenomus Dodd, 1914 to be a junior synonym of Telenomus Haliday, 1833.

Position of Eumicrosoma
This name encompasses a small group of species found worldwide that parasitize the eggs of blissine bugs. The genus is quite distinctively defined by its dorsoventrally depressed body, elongate lanceolate wings, elongate marginal and truncate postmarginal vein in the forewing, and opisthognathous mandibles. Johnson (1984a) noted the similarity between Eumicrosoma and the floridanus group of Telenomus, and implied that the former was merely a morphologically extreme form of the latter. Our results do not support the hypothesis that Eumicrosoma is most closely related to species of the floridanus group. The position of Eumicrosoma is unstable: the genus arises near the californicus complex in some analyses or with the Australian species Trissolcus ogyges (Dodd) in others. In fact, the floridanus group itself fails to emerge as a monophyletic group; the only species with a stable position in the phylogeny -with low support -is Telenomus FLO1 at the base of the clade of Trissolcus + Telenomus + Eumicrosoma. In view of the lack of strong support for resolution in this part of the inferred phylogeny, we prefer to adopt a conservative approach and continue to treat Eumicrosoma as a separate, valid genus.

The speciose genera: Telenomus and Trissolcus
The genera that today go by the names Telenomus and Trissolcus were first distinguished by C.G. Thomson (1860) as Phanurus Thomson and Telenomus, respectively. Most species were recognized by comparison of the sculpture of the frons: absent in Telenomus, present in Trissolcus; eye setation: long in Telenomus, short or absent in Trissolcus; development of notauli: almost always absent in Telenomus, commonly found in Trissolcus; shape of the second metasomal tergite: longer than wide in Telenomus, wider than long in Trissolcus; and number of female clavomeres: most Telenomus with five or fewer, Trissolcus with six. In general, species of Trissolcus are robust and stout-bodied, whereas many Telenomus species are elongate and gracile. None of these is without exception, but in concert they leave few species that cannot be unambiguously placed under one name or the other. The monophyly of these taxa, however, has never been demonstrated. The morphological diversity among species that has made it so difficult to unambiguously diagnose genera is also reflected in the attempts to delimit groups of species within them. We began this study not with the expectation that either Telenomus or Trissolcus would turn out to be monophyletic, but, rather, with the hope that the relationships among the species groups would provide the scaffold upon which to build a better understanding of the phylogeny of the subfamily.
Our data fail to settle decisively the question of the monophyly of either of the two core genera of the subfamily, Trissolcus and Telenomus. Further, though, none of the three species groups of Trissolcus were recovered in these analyses. At least one of these, the flavipes group, is very homogeneous and distinctive in morphology and, therefore, seems likely on those grounds to be monophyletic. The basal position of two of the species of the thyantae group is consistent with our expectations based on structure -for example, presence of notauli, presence of sublateral setae on the first metasomal tergite, and presence of setae on the posteroventral metapleuron.
We were able to recover some of the expected structure among the Telenomus species represented. Particularly, the placement of the crassiclava group distinctly apart from other Telenomus is significant. Among the remaining groups, the podisi and floridanus groups were not found to be monophyletic. As discussed in the section on Eumicrosoma, species of the floridanus group are scattered throughout the phylogeny. The only Old World species of the podisi group in the study, Te. sechellensis Kieffer, is always more closely related to the californicus complex than with others of its putative group. This result is corroborated by morphological evidence, specifically the same character of the occipital carina described above for Te. monticola. This feature is found in all Old World Telenomus parasitoids of stink bugs as well as in all species of the californicus complex, and it is lacking in all New World members of the podisi group. Consistent with the hypotheses in Johnson (1984a) the species of the californicus complex grouped together in a single clade, from which emerges the tabanivorus species group. These are all united by the structure of the metascutellum (evenly produced, coarsely sculptured dorsally, longitudinally striate ventrally) and the lower course of the occipital carina. The laricis and longicornis groups are also monophyletic, and most species are fairly easily recognized on the basis of morphology. The laricis group is characterized by the greatly enlarged gena and elongate body; those of the longicornis group have a strongly pigmented basal vein on the forewing, distinctive male genitalia with an elongate, spatulate aedeagal lobe, and a distinctively convex mesosoma.

Host relationships
Biological data have been acquired for 16 of the 51 ingroup species (31%). Thus, our assumptions of the host relationships and specificity should be viewed with some caution. Nevertheless, the number of host records for telenomines overall is very substantial. With that caveat in mind, it appears that the combined clade of Gryon, the Psix group, and Telenominae s.s. evolved from an ancestral species that probably parasitized the eggs of terrestrial Heteroptera. Hosts in the Pentatomoidea, Coreoidea, Reduvioidea and Lygaeoidea are shared among these three taxa. Paratelenomus has only been recorded as a parasitoid of turtle bugs (Pentatomoidea: Plataspidae), whereas species of Psix have been reared from a number of species of Pentatomoidea as well as Coreoidea. For the basal groups of Telenominae s.s., with the exception of Phanuromyia, the most common families of hosts are the stink bugs and their allies in Pentatomoidea, particularly the Pentatomidae and Scutelleridae. This is true of all three of the species groups of Trissolcus, as well as the Telenomus podisi group (represented by Te. sechellensis, Te. grenadensis and Te. podisi ). Species of the floridanus group of Telenomus, parasites of Lygaeidae, also arise at or near the base of the sister group of Phanuromyia. These include Te. nysivorus, Te. consimilis and three unidentified species. The single species of Eumicrosoma in this study is from Thailand and its host is as yet unknown.
However, three species of the genus, E. beneficum Gahan, E. bicolor (Ashmead) and E. blissae (Maki) are all well known as parasitoids of chinch bug eggs (Blissus, Ischnodemus: Heteroptera: Lygaeidae). Gryon is a very large genus of Scelioninae (331 valid species). A few species have been recorded as having been reared from eggs of Lepidoptera, but the vast majority parasitizes the same groups of Heteroptera as these telenomines (with the exception of Blissidae).
The parasitism of the eggs of auchenorrhynchous bugs -planthoppers of the families Fulgoridae and Issidae are the only hosts so far recorded -appears to have evolved but once, in Phanuromyia (interpreted as including the T. crassiclava and aradi groups). Telenomus has often been described as a group composed primarily of parasitoids of the eggs of moths and butterflies (e.g. Masner, 1976). Certainly a good number of species have this biology, and these are commonly encountered. Yet the host relationships of the genus are much more diverse. Parasitism of the eggs of Lepidoptera has only been elsewhere recorded in the Platygastroidea for two species of Gryon. Clearly, the shift(s) to this pool of hosts was evolutionarily significant, at least for diversification within Telenomus. On the basis of our results this shift occurred once or twice, depending on the interpretation of the position and putative host relationships of Telenomus FLO3 within the clade corresponding to the Te. californicus complex. Then, from a lepidopteran parasitoid, there was a subsequent shift in the Te. tabanivorus group to attack the eggs of some groups of horse flies (Diptera: Tabanidae) and soldier flies (Diptera: Stratiomyiidae). In a separate lineage are the laricis group of Telenomus; one representative of the phymatae group, Te. dolichocerus (Ashmead); and the longicornis group of Telenomus. The first of these, the laricis group, has few host records, but these are limited to the plant bugs (Miridae), including the economically important tarnished plant bug, Lygus lineolaris (Palisot de Beauvois). Telenomus dolichocerus is a parasitoid of Reduviidae. The hosts of the longicornis group are as yet unknown. However, the species are somehow associated with water: they are commonly found alongside ponds and streams and have even been collected in floating emergence traps. Marshall (2006) asserted that one species in this group parasitizes the eggs of Rhagovelia obesa Uhler (Heteroptera: Veliidae), but supporting data have not been published, and we have never seen specimens so labelled. Certainly, given the position of the longicornis group, in a clade of heteropteran parasitoids, either Gerridae or Veliidae would be good candidates for hosts.
In summary, the sequence of host shifts that best fits the results presented here begins from a ground-plan host of some group of terrestrial Heteroptera. From there, one clade, Phanuromyia, moved on to the Auchenorrhyncha. Its sister group went in two different directions, with one subgroup (the californicus complex) shifting first to Lepidoptera and then to Diptera, whereas a second group shifted to Miridae (the laricis group), Reduviidae and a group of aquatic or semiaquatic insects, probably Heteroptera.
Although we have discussed host relationships in terms of taxonomic groups, obviously the parasitoids are using other cues in order to locate suitable eggs to parasitize. Host names merely serve as proxies for the real sensory modalities involved, such as long-or short-range chemicals (e.g. Colazza et al., 2007Colazza et al., , 2009Peri et al., 2007;Moraes et al., 2009;Salerno et al., 2009), tactile cues (e.g. Laumann et al., 2007), or possibly vision (Laumann et al., 2011). Determination of the breadth of host range and shifts from one set of hosts to another would seem involve either an overlap in the cues directly or indirectly produced by a host species, or a change in the capacity or responses of parasitoids to these stimuli. Research on the proximate mechanisms of host-finding in telenomines has been focused on species that are parasitoids of important pests such as Trissolcus basalis, Telenomus podisi , Te. busseolae (Gahan) and Te. remus Nixon. A phylogenetic hypothesis is sorely needed to be able to frame questions about the evolution of host relationships and to appropriately integrate the findings from these several species. The results presented here are a first step in that direction, providing support for some hypotheses and highlighting those areas in need of more intense future work.

Supporting Information
Additional Supporting Information may be found in the online version of this article under the DOI reference: 10.1111/syen.12032 File S1. List of taxa and sequenced markers for phylogenetic analysis of Telenominae: GenBank accession numbers and identifiers for specimens sequenced.