Circumscription of the Ganaspis brasiliensis (Ihering, 1905) species complex (Hymenoptera, Figitidae), and the description of two new species parasitizing the spotted wing drosophila, Drosophila

)


Introduction
Species that have not yet manifested morphological differences can often be separated based on molecular sequence data as well as behavioral data (Struck et al., 2018).These 'cryptic' species are a reality in modern systematics, and there are numerous examples recently published in Hymenoptera, including ants (Prebus 2021;Branstetter and Longino 2022;Schar et al. 2022), encyrtid wasps (Wang et al. 2016), eulophid wasps (Hansson and Hamback 2013), aphelinid wasps (Heraty et al. 2007), microgastrine braconids (Alex Smith et al. 2013) and parasitoids of cynipid wasps (Zhang et al. 2022), to mention some.Within biological control research, cryptic species can present a major challenge in finding the safest, most reliable, natural enemy (Rosen 1986).
Ganaspis brasiliensis (Ihering, 1905) (Hymenoptera: Figitidae: Eucoilinae) has been hypothesized to be a cryptic species complex (Nomano et al. 2017;Hopper et al. 2024).While a great deal of international exploration has been conducted in pursuit of natural enemies of the spotted-wing drosophila (SWD), Drosophila suzukii Matsumura, 1931 (Diptera: Drosophilidae) (Nomano et al. 2017;Giorgino et al. 2018;Abram et al. 2022), much work has also been conducted on which wasp populations attack this pest with the most specificity.Ganaspis brasiliensis was, early on, recognized as a common natural enemy of SWD, and was redescribed (Buffington and Forshage 2016).Since then, additional field and lab studies (summarized by Seehausen et al. 2020) have suggested G. brasiliensis could in fact be a cryptic species complex, composed of as many as five species (Nomano et al. 2017).The study by Nomano et al (2017) reported five lineages of G. brasiliensis, namely G1-G5.Since that study, G1 has been recognized as the most host-specific population of G. brasiliensis (Giorgini et al. 2018;Abram et al. 2022) and has even been cleared for release in the United States by USDA-APHIS; G3, also a natural enemy of SWD, has a slightly broader host range, and where it co-occurs with G1, a slightly less effective natural enemy of SWD; and G5 appears to be pan-tropical and not able to exploit SWD at all; other Drosophila species are hosts for G2 and G4 (Nomano et al. 2017).Altogether, these data led Seehausen et al. (2020) to set the stage for these populations of G. brasiliensis G1, G3, and G5, to be recognized as distinct species (G2 and G4 lacked sufficient sample data).Hopper et al. (2024) presented whole genome datasets that suggest G1 and G3 are in fact separate species (although G5 was not included, as it has not been recorded attacking SWD).
We utilize an integrative approach, including novel ultra-conserved element (UCE) molecular data, morphological data (based on ovipositor and scutellar characters), published host specificity studies, published crossing experiments, and distribution data to distinguish at least three lineages within the formerly recognized Ganaspis brasiliensis.As such, we herewith describe two new species: Ganaspis kimorum, new species, and Ganaspis lupini, new species.In light of these new data, Ganaspis brasiliensis is redescribed and the circumscription of this species updated.

Specimen acquisition
The source for all specimens in this study resulted from the combined rearing efforts of collaborators from around the world, as well as researchers with established Ganaspis spp.laboratory colonies for genetic research; these collaborators are listed in Suppl.material 1: table S1 as well as in the acknowledgments.As specimens were reared in the field or the lab, a subset of specimens was sent to one of us (MLB) for morphological determination and vouchering in the National Museum of Natural History, Smithsonian Institution (USNM); these were typically dry mounted and labeled, but some were retained in 95% ethanol.In some cases, large numbers of specimens were sent and kept in 95% ethanol.Outgroups included were Ganaspis hookeri (Crawford, 1913) (from a laboratory colony), two other undescribed, but morphologically distinct, species of Ganaspis, that were bred true from a laboratory colony, and specimens identified as belonging to the eucoiline genus Leptopilina Föster.

Specimen illustration and observation
Representative specimens were imaged using the Macropod®™ multiple-focus imaging system to illustrate diagnostic characters; single montage images were produced from image stacks with the program Zerene Stacker®™.Scanning electron micrographs were generated using a Hitachi®™ TM3000 desktop scanning electron microscope; specimens were coated in 25-30 nm gold-palladium alloy (Cressington®™ 108 auto sputtercoater), using 'analysis' voltage, running in 'compo' mode.Diagnoses focus on easily recognized gross morphologies, and species/genera that can be confused with G. brasiliensis are diagnosed.Terminology for all descriptive characters, as well as phylogenetic characters, follow Buffington and Forshage (2016).

DNA extraction
DNA was extracted using the Qiagen DNeasy® Blood and Tissue Kit (Qiagen, Valencia, CA, U.S.A.).DNA extractions were performed by either placing an entire individual (male or female) or their metasoma, in a 2 mL tube with 0.5 mm diameter glass lysis beads (BioSpec Products, Bartlesville, OK, U.S.A.).The samples were placed in a -20 °C freezer for ~10 minutes and then placed in a TissueLyser II (Qiagen Inc., USA) for 30 s at 30 Hz to disrupt the tissue and facilitate the lysis process.Cell lysis was performed overnight with 20 µL of Proteinase K in a dry bath shaker at 56 °C and at 500 rpm.The recommendations of the manufacturer were followed for the extraction process except that the cleaned, extracted DNA from the spin-collection columns was eluted with two (rather than one) washes, each consisting of 55 µL of nuclease-free water, differing from the manufacturer's recommendation of 200 µL of AE buffer.DNA extractions were quantified using 2 µL of DNA template in a Qubit 4 Fluorometer and with the 1X dsDNA High Sensitivity (HS) assay Kit (Thermo Fisher Scientific, Inc.).DNA extraction concentration ranged from 0.001-7.20 ng/µL (mean = 0.943 ng/µL).

Generation of UCE data: Library preparation
Prior to library preparation, 1-50 ng of DNA template was sheared to an average fragment length of 300-600 bp using a Qsonica Q800R2 Sonicator (Qsonica LLC, Newton, CT, U.S.A.) for 60 s with amplitude set at 25 and the pulse set at 10. Libraries were prepared on 96-well plates on a DynaMag™-96 side magnet (Invitrogen, Thermo Fischer Scientific, Waltham, MA, U.S.A.) and using the Kapa Hyper Prep Library Kit (Roche Diagnostics Corporation, Indianapolis, IN, U.S.A.) as described in Faircloth et al. (2015) with the iTru Adapter protocol.We implemented all magnetic bead clean-up steps (Fisher et al. 2011) as described in Faircloth et al. (2015) and used dual-indexing TruSeq adapters (Faircloth andGlenn 2012, Glenn et al. 2019) for ligation.The ligation step was followed by PCR-amplification of 15 µL of the library product using 25 µL of KAPA HiFi ReadyMix (Roche Diagnostics Corporation, Indianapolis, IN, U.S.A.), 2.5 µL of each of Illumina TruSeq (i5 and i7) primers, and 5 µL nuclease-free ddH 2 O.The following thermal cycler program was executed: 98 °C for 45 s; 13 cycles of 98 °C for 15 s, 60 °C for 30 s, 72 °C for 60 s; and final extension at 72 °C for 5 m.Following PCR, we purified DNA products using 1.1× Kapa Pure beads and rehydrated the purified product in 22 µL of Elution Buffer (pH = 8).Individual libraries were quantified using 2 µL of library product in a Qubit 4 Fluorometer using the 1X dsDNA Broad Range (BR) assay Kit (Thermo Fisher Scientific, Inc.).Post-PCR libraries have concentrations ranging from 3.05-114.9ng/µL.

Library pooling and target enrichment of libraries
Post-PCR libraries were pooled at equimolar concentrations into 25 pools, each containing 6-12 libraries.Pool concentration was adjusted to ~71.5 ng/µL by drying the sample in a vacuum centrifuge for 45-60 min or until all liquid was evaporated at 60 °C, and then by re-suspending the pool in nuclease-free water at the estimated volume.We then used 2 µL of the resuspended product to measure the pool final concentration in a Qubit 4 Fluorometer with the 1X dsDNA BR assay Kit.The final concentration of the pre-enrichment pools was 63.6-91.4ng/µL.The pool was enriched by using the myBaits (Arbor Biosciences, Ann Harbor, MI, U.S.A.) UCE Hymenoptera bait set ("Hymenoptera 2.5Kv2P") targeting 2590 conserved loci in Hymenoptera (Branstetter et al. 2017) at an incubation temperature of 65 °C for 24 h in a thermal cycler.Enrichment, bead-cleaning, and PCR reaction procedures partially followed the Arbor Biosciences v5.0.1 (https://arborbiosci.com/mybaits-manual/) protocol and Branstetter et al. (2021) and Hanisch et al. (2022).The resulting reaction was purified using 1.0X Kapa Pure beads (Roche Diagnostics Corporation, Indianapolis, IN, U.S.A.) and the enriched pool was then rehydrated in 22 µL elution buffer.The final two enriched pools were submitted to Admera Health Biopharma Services (NJ, U.S.A.) for quality control and sequencing of two lanes on an Illumina HiSeq2500 instrument.A summary of the raw data, contigs, UCE loci recovered, and other assembly statistics for each sample is presented in Suppl.material 1: table S2.Most extractions and UCE laboratory work were conducted in the Laboratories of Analytical Biology (LAB) facilities of the National Museum of Natural History, Smithsonian Institution.New raw sequences generated as part of this study are deposited in the NCBI Sequence Read Archive (SRA) under BioProject number PRJNA1088885 and under accession No. SAMN40504884-SAMN40505120.
To identify and remove outlier or poorly aligned sequence fragments we used the Python tool SPRUCEUP (Borowiec 2019) on the Ganasbra237t_60p alignment.Parameters in the configuration file were set up to the uncorrected p-distance for computing the distances, window size = 20 bp and overlap = 15 bp, a lognormal distribution to identify outlier distances, and a global cutoff of 0.997.As a result, SPRUCEUP removed 56,045 (0.02%) outlier nucleotide-site state assignments (i.e., matrix cell values (Suppl.material 1: table S3).We then used the resulting SPRUCEUP-trimmed alignment and repeated the unpartitioned analysis using the same parameters as above, including 1,000 replicates of the SH-like approximation likelihood-ratio test (-alrt 1000) (Guindon et al. 2010).Upon inspection of the resulting tree and the png files generated by SPRUCEUP, additional, manual cutoffs were performed for 23 taxa (see Suppl.material 1: table S3 for cutoff values).
We partitioned the resulting trimmed alignment using the Sliding Window Site Characteristics based in Entropy method (SWSC-EN; Tagliacollo and Lanfear 2018), in which each UCE locus is divided into three regions (a core and two flanking regions).The SWSC-EN algorithm identified 4,137 subsets.We then identified the best partitioning scheme by merging the resulting subsets using ModelFinder (Kalyaanamoorthy et al. 2017) as implemented in IQ-TREE (Minh et al. 2020).For the merging step, we used the -m MF+MERGE command, the fast relaxed -rclusterf algorithm (set to 10; Lanfear et al. 2017) and compared the top 10% of the resulting partitioning schemes using the corrected Akaike information criterion (AICc), restricting the evaluated models to those implemented in RAxML by using the command -mset raxml.The best-fit partitioning scheme (0.997_lognorm_man_Ganasbra237t_60p_SWSCEN) consisted of 1,629 partitions.
We tested for model violation based on assumptions of stationarity and homogeneity by performing a test of symmetry (Naser-Khdour et al. 2019) as implemented in IQ-TREE 2.1.3(Chernomor et al. 2016;Minh et al. 2020) on the partitioned dataset above.We removed bad partitions by using the -symtest-remove-bad option with a P-value cutoff set as the default (P = 0.05).The test of symmetry identified and removed 536 out of the 4,137 subsets generated by SWSC-EN.The resulting best-fit partitioning scheme (0.997_lognorms_cutoff_man_Ganaspis237-60p_partitions.nex.good_SYMTEST) consisted of 1,422 partitions.
We performed further maximum-likelihood (ML) analyses on the trimmed alignment with the different partitioning schemes using IQ-TREE multicore v.2.1.3 (Chernomor et al. 2016;Minh et al. 2020), estimating branch support with the ultrafast bootstrap (Hoang et al. 2018) and the SH-like approximation likelihood ratio test (Guindon et al. 2010) set at 1,000 replicates, with other settings set at default values on the 0.997_lognorms_cutoff_man_Ganaspis237-60p_partitions.nex.good_SYMTESTalignment.Statistics for all the data matrices generated for this study are summarized in Suppl.material 1: table S4 and all trees generated for this study are in supplementary information (https://figshare.com/s/93d692506c0fe68a7ddd).
We employed the species delimitation plugin v.1.4.5 (Masters et al. 2011) in Geneious Prime (www.biomatters.com)to summarize the average pairwise tree distance (using the tree in Fig. 6) among members of a clade (Intra Dist) and the average pairwise tree distance among a clade and its closest clade (Inter Dist).Results are summarized in Table 1 and Suppl. material 1: table S5.

Generation of DNA Barcoding (COI)
Employing the Phyluce script phyluce_assembly_match_contigs_to_barcode and a COI sequence as reference ('Ganaspis brasiliensis' downloaded from GenBank accession No. MN013168.1),we extracted from the UCE data bycatch (Ströher et al. 2017) the cytochrome oxidase I (COI) gene fragment for a subset of samples, including from a non-type specimen of the original description of Ganaspis brasiliensis not included in the UCE data analyses above.The contig.slicefiles were then inspected in Geneious Prime v.2024.0.4 (www.biomatters.com)and mapped to the reference sequence (MN013168.1)using the BBMap v.1.0(Bushnell 2014) plugin.COI sequences, of some specimens, are given at the end of the description of each species, and they have also been deposited in GenBank under accession No. PP599368-PP599375 (see Suppl.material 1: table S6).

UCE sequencing and matrix assembly
In average, we recovered 1,369 UCE loci (range: 672-1,670) with a mean length of 878 bp (range: 341-1,801 bp).The final alignment included 237 terminals, 1,379 UCE loci, and 1,221,982 bp of sequence data, of which 523,179 were parsimonyinformative sites.The alignment was composed mostly of samples in the genus Ganaspis and two samples belonging to the genus Leptopilina Föster, 1862, as a distant outgroup.The test of symmetry conducted in IQTREE 2.1.3(Chernomor et al. 2016;Minh et al. 2020) identified and removed 536 and 565 bad partitions depending on the SPRUCEUP manual cut-off employed (Suppl.material 1: table S3).The resulting 'good' alignments consisted of 964,889 and 960,213 bp of sequence data and 427,640 and 427,499, respectively.For additional assembly and additional statistics, see Suppl.material 1: table S4.All trees generated in this study are deposited in supplemental information (https://figshare.com/s/93d692506c0fe68a7ddd).

Morphological study
Morphological examination of the scutellum and ovipositor revealed subtle but consistent differences among specimens examined (Fig. 3).For the scutellum, the lateral aspects were carinate/striate in tropical specimens (G.brasiliensis), but totally smooth in more temperate specimens (G.kimorum, G. lupini).Further, members of G. kimorum were discovered to have a reduced ovipositor clip (Fig. 4).Research on the genome (Hopper et al. 2024), host specificity (Girod et al. 2018b;Wang et al. 2018;Seehausen et al. 2020;Daane et al. 2021), reproductive isolation (Seehausen et al. 2020;Hopper et al. 2024), and DNA barcode region (Nomano et al. 2017) are consistent with the morphological characters and UCE phylogenomic data shown here (Fig. 6).Together these data support the description of G. kimorum and G. lupini as species new to science.Fig. 6 shows the phylogram of these species; major sources of specimens are highlighted.Table 1 summarizes the various lines of study for species delimitation.
Diagnosis.Scutellum large, terminating anterior to the end of the scutellum; convex in lateral view, bulging slightly.Marginal cell closed in forewing.Posterior edge of metapleuron uninterrupted.Segments of female antennal clava very moderately enlarged and concolorous with other flagellomeres (Buffington and Forshage 2016).In the Ganaspis brasiliensis species group, the plate covers at least half of the scutellum, when viewed dorsally; in lateral view, the scutellar plate is clearly convex, even bulging, anterior to the glandular release pit.In other Ganaspis species, the scutellar plate may be as large or smaller, covering less than half of the scutellum when viewed dorsally; the plate, in lateral view, is flat or gently convex, and if large, not bulging.The marginal cell and claval characters are quite variable in other Ganaspis species.
Description.Coloration with head, mesosoma, and metasoma black to dark brown; legs uniformly light brown.Sculpture on vertex, lateral surface of pronotum, mesosoma, and metasoma absent, surface entirely smooth (Fig. 1).Length 1.5-1.75mm.Head.In anterior view, rounded, approximately as high as broad; in lateral view, more transverse, not protruding.Pubescence on head sparse, nearly glabrous.Sculpture along lateral margin of occiput absent (Fig. 2A).Gena (measured from compound eye to posterolateral margin of head) short, ratio of length of gena to length of compound eye in dorsal view <0.3 mm.Sculpture of gena absent, smooth.Lateral margin of occiput evenly rounded, not well defined (Fig. 2A).Occiput (except extreme lateral margin) smooth.Carina issuing from lateral margin of postocciput absent.Ocelli small, ratio of maximum diameter of a lateral ocellus to shortest distance between lateral ocelli 0.2-0.4mm.Anterior ocellus far from posterior ocelli, clearly anterior to anterior margins of posterior ocelli.Relative position of antennal sockets close to ocelli, ratio of vertical distance between inner margin of antennal foramen and ventral margin of clypeus to vertical distance between anterior ocellus and antennal rim <2.0.Median keel of face absent.Vertical carina adjacent to ventral margin of antennal socket absent.Facial sculpture absent, surface smooth.Facial impression absent, face flat.Antennal scrobe absent.Anterior tentorial pits small.Longitudinal axis of posterior tentorial pits oblique.Vertical delineations on lower face absent.Ventral clypeal margin laterally, close to anterior mandibular articulation, straight.Ventral clypeal margin medially straight, not projecting.Clypeus smooth, evenly rounded.Malar space adjacent to anterior articulation of mandible evenly rounded, smooth.Malar sulcus present.Eye close to ocelli, ratio of distance between compound eye and posterior mandibular articulation to distance between posterior ocellus and compound eye >1.2 mm.Compound eyes, in dorsal view, distinctly protruding from the surface of the head, particularly laterally.Pubescence on compound eyes absent.Orbital furrows absent.Lateral frontal carina of face absent.Dorsal aspect of vertex smooth.Posterior aspect of vertex smooth.Hair punctures on lateral aspect of vertex absent.Posterior surface of head deeply impressed around post-occiput.
Labial-maxillary complex.Apical segment of maxillary palp with pubescence, consisting only of erect setae.First segment of labial palp shorter than apical segment.Labial palp composed of two segments.Apical seta on apical segment of maxillary palp longer than twice length of second longest apical seta.Erect setae medially on apical segment of maxillary palp absent.Maxillary palp composed of four segments.Last two segments of maxillary palp (in normal repose) curved inwards.Distal margin of subapical segment of maxillary palp slanting inwards, apical segment bending inwards.Apical segment of maxillary palp more than 1.5× as long as preceding segment.
Pubescence on outer surface of metatarsal claw sparse, consisting of few setae.Outer surface of metatarsal claw almost entirely smooth.Base of metatarsal claw lammelate, with translucent cuticular flange.
Wings.Pubescence of fore wing present, long, dense on most of surface (Fig. 2E).Apical margin of female fore wing rounded.Rs+M of forewing completely defined (Fig. 2E).Vein R1 tubular along at least basal part of anterior margin of marginal cell.Mesal end of Rs+M vein situated closer to posterior margin of wing, directed towards posterior end of basalis (Fig. 2E).Basal abscissa of R1 (the abscissa between 2r and the wing margin) of fore wing as broad as adjacent wing veins.Coloration of wing absent, entire wing hyaline (Fig. 2E).Marginal cell of fore wing membranous.Areolet absent.Hair fringe along apical margin of fore wing present, very short.

Ganaspis brasiliensis (Ihering, 1905)
Figs 1A, 3A, B, 4A Diagnosis.Separated from G. kimorum and G. lupini by the sculpturing on the side of the scutellum.In G. brasiliensis, this area has distinct dorso-ventral carinae enclosing one or a few cells (Fig. 3A).In G. kimorum and G. lupini, this area is totally smooth (arrows, Fig. 3C, E).Further separated from G. kimorum by the well-developed ovipositor clip that is expanded across more the half the width of the ovipositor valve (Fig. 4A).
Cytochrome c oxidase subunit I (COI) Barcode region.
Biology.Kionobiont endoparasitoid of Drosophila melanogaster, D. simulans, and other Drosophila in decaying fruit.Original description and some label data suggesting Anastrepha (Tephritidae) as a host are most likely to represent erroneous associations (cf.Buffington and Forshage 2016).
Description.As in description for G. brasiliensis species complex, but with the lateral aspect of the scutellum completely smooth; ovipositor clip large, extending beyond the halfway point across the fused ovipositor valve.Previous studies referencing 'Gb G3' or 'G3' refer to this species (Nomano et al. 2017;Giorgini et al. 2018;Abram et al. 2022).

Discussion
The logic behind this comprehensive phylogenetic species delimitation study is based on observations that COI can sometimes be error-prone in species discrimination in Eucoilinae, as well as other insect groups (Brower 2006;Lohse 2009;Goldstein and LaSalle 2011;Collins and Cruickshank 2012), sometimes mediated by the presence of Wolbachia (Jiggins 2003;Klopfstein et al. 2016;Cariou et al. 2017).When we compared our trees with the basic topology of Nomano et al. (2017), which relied on COI and ITS2 for discriminating the 'G-species', we find some disagreement with respect to phylogeny.The former G3 (now G. lupini) is recovered as the sister-group to G. brasiliensis, and G1 (now G. kimorum) is recovered as sister-group to the clade containing G. brasiliensis+G.lupini.As our dataset is more comprehensive than that of Nomano et al. (2017), both in terms of taxon sampling and number of loci, we suggest the tree pre-sented here is a more accurate interpretation of the evolution of this group.While the use of UCE markers coupled with nex-gen sequencing technology has made generating larger amounts of data much easier and more affordable, there is still a place for mitochondrial 'barcode' data with respect to determining these cryptic species.We strongly encourage newly generated barcode data to be only compared to barcode data here in this paper, as well as barcode data in the Drosophila parasitoid database DROP (Lue et al. 2021) where sequences are backed by authoritatively identified voucher specimens.
The dataset here has yielded more nuanced results concerning in-group relationships than previous studies of these taxa.For instance, Ganaspis lupini has three distinct subclades within the species; we have decided to retain these three clades as members of the same species.Within Ganaspis brasiliensis, even more subclades can be discerned, some of which may eventually be split out into additional species.The ratio of the intra vs inter pairwise tree distances (Table 1 and Suppl.material 1: table S5) suggests that members of both G. brasiliensis and G. lupini seem to be more diverse compared to members within of G. kimorum.However, unlike the situation between G. brasiliensis, G. lupini and G. kimorum, where morphological differences were noted, though difficult to observe, there are no such morphological differences among the clades of G. brasiliensis.Hence, for the present time, we are considering all these subclades to be members of G. brasiliensis.It would have been very interesting to consider UCE data from types of G. brasiliensis, but unfortunately, attempts at amplifying extracts from the type series yield low-quality UCE data.
Biogeographically, G. brasiliensis appears to be a pan-tropical species, and seemingly no specimens of this species have been collected outside the subtropics (with Hawai'i being the most 'temperate' locality), while Ganaspis lupini and G. kimorum appear to be temperate species.
This appears to be the first study to utilize the ovipositor clip for species-level discrimination.Prior to this study, the clip was formally described (van Lenteren et al. 1998), followed by Buffington (2007) where the presence/absence of the clip across Figitidae was examined.Buffington (2007) hypothesized that the absence of the clip in Aspicerinae and Anacharitinae was linked to a 'quick-strike' oviposition strategy, as the hosts they attacked (Syrphidae and Hemerobiidae, respectively) are themselves aggressive hosts to be attacking.Further, the Charipinae are hyperparasitoids of aphidophagous braconids and chalcidoids, where there is no need for host restraint.Finally, the leaf-miner specialists, among the eucoiline Zaeucoilini and Diglyphosematini, also lack the clip as their host is a 'captive audience' that cannot readily escape parasitization.Together, these data suggest the ovipositor clip, in the appropriate circumstance, is an asset; in other circumstances, a liability.
The pattern observed in the Ganaspis brasiliensis species complex can be interpreted along the same lines.In the case of both G. brasiliensis and G. lupini, the ovipositor clip has retained its typical size, spanning the width and depth of the fused valve of the ovipositor (Figs 4,5).By contrast, G. kimorum has a much-reduced ovipositor clip, so much reduced that the last author had to mount some 40 ovipositors before the nature of the reduced clip could actually be observed (in some cases, there appeared to be no ovipositor clip present) (Fig. 4).Both G. brasiliensis and G. lupini are attacking hosts within an already decomposing substrate, allowing the would-be host ample room for escape.Further, the skin of the fruit, if still intact, would be very soft, and the clip itself would not be engaged by it.In the case of G. kimorum, which attacks its host in ripe fruit, the host larvae would have much less space for escape.Further, as the skin of the fruit is still intact, this more rigid barrier may in fact cause complications for the insertion of the ovipositor, as the clip itself may in fact snag on the fruit skin during insertion.The reduction of the clip into a much more streamlined silhouette, we hypothesize, helps the ovipositor insert into the fruit more effectively and not engage the fruit skin.And, as the host larvae has a reduced chance of escape inside of fresh fruit, this shallower ovipositor clip remains effective at securing the host (Fig. 5).
What is the future of species delimitation using UCE data?We have demonstrated here that these data are quite effective at discriminating among morphologically virtually identical but biologically distinct species.We may very well be observing the immediate after-effects of speciation, where morphological characters have not yet manifested themselves to be observed, but clearly, biological and genetic characters distinguishing these species are present.And if this is the case, the much larger amount of sequence data per specimen that UCE methodology provides is rather critical.As the cost of this technique continues to decline, we predict this technique will certainly be considered more closely in the future.
Perhaps a more difficult question to consider is: what is the future of Ganaspis taxonomy?Ganaspis currently has only 49 nominal species, of which 17 actually belong in other genera and are awaiting new combinations, while 8 are nomina inquirenda, the types of which have never been studied by modern researchers, leaving 24 described Ganaspis species.But then there are an additional 46 Ganaspis species that are currently classified in other genera and await new combinations in Ganaspis (unpublished data).On top of this there are numerous undescribed species, including a remarkable number of 'BINs' in BOLD (Sosa-Calvo and Buffington, pers.obsv.)But these considerations are all based on current circumscription where it seems very likely that Ganaspis is at least a paraphyletic assemblage of all the "typical Ganaspini" without certain striking apomorphies which define related genera such as Areaspis Lin, 1988, Didyctium Riley, 1879, Discaspis Lin, 1988, Endecameris Yoshimoto, 1963, Gastraspis Lin, 1988and Hexacola Föster, 1869(cf. the keys in van Noort et al. 2014;Buffington and Forshage 2015).But there is also a possibility that this represents a morphology that several lineages have been converging into based on similar life histories (the similarity between different Ganaspis and Leptopilina species attacking similar drosophilid hosts is rather striking, considering that Leptopilina belongs to an entirely different group within Eucoilinae (the tribe Eucoilini)).Indeed, in all published phylogenetic analyses where it has been tested, Ganaspis has come out non-monophyletic (Fontal-Cazalla et al. 2002;Buffington et al. 2007;Blaimer et al. 2020).Whether the genus is indeed a paraphyletic grouping of the more plesiomorphic crown-group Ganaspini, or rather a polyphyletic assemblage of wasps having converged on a similar morphology, is very difficult to say, and is perhaps not even a very meaningful question to ask before a more rigid circumscription of the genus has been attained.A global review of the genus is desperately needed to give it a meaningful circumscription and to identify monophyletic species groups which can be properly revised.Perhaps the new UCE methodology can assist, especially since UCE can be generated effectively from older museum specimens.
In conclusion, working out the limits of these three species of Ganaspis, like in other cases of cryptic species complexes, has required a great deal of behavioral study, genetic study, and morphological study, and benefitted from the reciprocal illumination they have offered.This has also involved researchers from around the world, conducting very careful work documenting these species, as well as the centralization of voucher specimens such that various lines of evidence can be directly, and quickly, compared.Thus, this work represents a celebration of international collaboration between research groups in different countries with different specializations for an integrated solution to independently noted problems.

Figure 1 .
Figure 1.Type-specimens of the species belonging to the G. brasiliensis species complex A brasiliensis, lectotype B kimorum, holotype C lupini, holotype.

Figure 2 .
Figure 2. External anatomy of Ganaspis kimorum, a member of the Ganaspis brasiliensis species complex A head and mesosoma, lateral view B head and mesosoma, dorsal view C femal antennae D metapectalpropodeal complex, lateral view E fore-and hindwings, female F propodeum, dorso-lateral view.

Figure 4 .
Figure 4. Comparison of the ovipositor clip between G. lupini (A, C) and G. kimorum (B, D).A G. lupini ovipositor tip, lateral view B G. kimorum, ovipositor tip, lateral view C G. lupini ventral of ovipositor showing ovipositor clip and membrane D G. kimorum ventral of ovipositor showing ovipositor clip and membrane.Black arrows indicate the location of the ovipositor clip.

Figure 5 .
Figure 5. Functional morphology of the ovipositor clip.

Figure 6 .
Figure 6.Phylogeny of the Ganaspis brasiliensis species complex based on the SYMTEST analysis.Specific populations, mentioned in previous studies, are highlighted by arrows, as well as where the holotypes of the two new species are located.The adventive population in British Columbia is also noted.

Table 1 .
Summary of differences among species in the Ganaspis brasiliensis species complex.