Evolutionary relationships of courtship songs in the parasitic wasp genus, Cotesia (Hymenoptera: Braconidae)

Acoustic signals play an important role in premating isolation based on sexual selection within many taxa. Many male parasitic wasps produce characteristic courtship songs used by females in mate selection. In Cotesia (Hymenoptera: Braconidae: Microgastrinae), courtship songs are generated by wing fanning with repetitive pulses in stereotypical patterns. Our objectives were to sample the diversity of courtship songs within Cotesia and to identify e underlying patterns of differentiation. We compared songs among 12 of ca. 80 Cotesia species in North America, including ten species that have not been recorded previously. For Cotesia congregata, we compared songs of wasps originating from six different host-foodplant sources, two of which are considered incipient species. Songs of emergent males from wild caterpillar hosts in five different families were recorded, and pattern, frequency, and duration of song elements analyzed. Principal component analysis converted the seven elements characterized into four uncorrelated components used in a hierarchical cluster analysis and grouped species by similarity of song structure. Species songs varied significantly in duration of repeating pulse and buzz elements and/or in fundamental frequency. Cluster analysis resolved similar species groups in agreement with the most recent molecular phylogeny for Cotesia spp., indicating the potential for using courtship songs as a predictor of genetic relatedness. Courtship song analysis may aid in identifying closely related cryptic species that overlap spatially, and provide insight into the evolution of this highly diverse and agriculturally important taxon.


Introduction
Acoustic signals are used by diverse groups of insects for species recognition, fitness displays, and courtship elicitation. Songs used during insect courtship are generally stereotypical within a species and likely play a role in reproductive isolation. Moreover, courtship songs may be a useful identifying character, especially among cryptic or closely related species [1]. For example, songs of Drosophila species groups are species-specific and have been studied for evolutionary patterns [2][3][4][5]. Furthermore, courtship song analyses have been used in conjunction PLOS  songs among multiple host-foodplant complex sources. Wasps from two of these host-foodplant complex sources, Manduca sexta on tobacco ("MsT") and Ceratomia catalpae on catalpa ("CcC"), have diverged genetically and are likely incipient species [47]. These wasps display a lower male response rate to the female pheromones of the reciprocal source, slight differences in duration and frequency of some song elements, and typically produce sterile hybrid females resulting from CcC♂xMsT♀ crosses [15]. In this study we describe the courtship songs of ten additional species of Cotesia, and use clustering to explore the relationships and patterns among songs. Further, we identify song differences among select host-associated populations and incipient species of C. congregata.

Parasitic wasp collection
Cotesia spp. were primarily collected from wild caterpillar hosts at multiple sites in the United States; some C. nr. phobetri came from an ongoing laboratory colony (Table 1). Caterpillars known to be hosts of Cotesia were targeted for collection, particularly Cotesia that have published gene sequences. When possible, wasps from different sites were collected for wider population sampling. In most cases, each wasp species came from a single host species. In contrast, C. congregata were collected from six different sphingid host species feeding on different plant families ( Table 1). All permissions were obtained as necessary for field collections on both public and private land from property owners and managers. None of the species involved are listed as endangered or protected. Caterpillars, usually collected before parasitization status was known, were reared on their host plant in plastic containers under ambient laboratory conditions until parasitoid egression or pupation. Individual unattached Cotesia cocoons were placed in clear gel capsules (size 00) 2-4 days after egression. Cotesia species forming a connected cocoon mass were chilled upon adult emergence and placed in individual capsules or vials. Adults were sexed under a dissecting microscope. Wasp songs were recorded within 24 hours of emergence. Voucher samples of each species were both point pinned and stored in 95% EtOH at -20˚C. The song of one species included in analysis (Cotesia marginiventris) was obtained from a USDA-ARS sound library (https://www.ars.usda.gov/ARSUserFiles/3559/soundlibrary.html) and originally described by Sivinski & Webb [24].

Audio recordings
Males in capsules were randomly selected from each brood for recording. Individuals were placed in an open paper arena with a drop of honey as a food source to encourage them to stay in the arena. Courtship songs were induced by exposing individual males to an immobilized female of the same species. Songs were recorded using a miniature omnidirectional microphone (model 4060, DPA, Longmont, CO; 20-20,000 Hz) positioned 5-7 mm above the male and a high resolution digital audio recorder (model 702, Sound Devices, Reedsburg, WI; 48 kHz sampling rate, 24 bit resolution) in a sound isolation booth (Industrial Acoustics, Bronx, NY) at 23 ± 1.5˚C and 40-55% RH. Generally, one recording per brood was analyzed; however, recordings of different individuals were analyzed for species with fewer than four collected broods. Additional individuals of C. congregata were recorded to test for relatively small differences among host-foodplant sources. Additional C. nr. phobetri were recorded from each brood because they could not initially be identified to a known species. Duration of song elements and fundamental frequency were quantified using Raven Pro v1.3 [48]. Waveforms were high-pass filtered at 100 Hz to reduce background noise.
Songs were divided into multiple elements based on acoustic characteristics shared across species (Fig 1). "Pauses" occur between other song elements when wings are held motionless above the body and do not generate sound. "Pulses" are high amplitude elements comprising the greatest range of wing movement, referred to as "boings" in C. congregata [20]. "Terminal buzzes" immediately follow pulses and consist of continuous lower-amplitude sounds. "Pre- pulse buzzes" are steady lower amplitude sounds that precede a pulse with no pause in between. For "pulse-buzz units" we measured the duration of the pulse and buzz together. "Interpulse interval" is the time from the start of one high-amplitude pulse to the start of the next. Pulses, buzzes, and pauses make up the pulse-buzz units and interpulse interval, which were included because they make discrete units that may be important in species recognition. Song amplitude was not compared because distance from the microphone varied slightly among some species. Quantification of song elements started with the second complete pulsebuzz cycle and continued for six complete pulse-buzz cycles. Spectrograms of the entire song were produced using a short-time Fourier transform (Hann window, size = 2,000 samples, 50% overlap). Frequency spectra for song sections were calculated using fast Fourier transforms (Hann window, size = 1,000 samples, 50% overlap). Frequency of the first harmonic (fundamental frequency) was used in all comparisons.

Comparisons and analysis
Several statistical procedures were used to determine differences in courtship songs, consolidate elements, and group songs by structure. The considerable differences among some songs present a challenge for direct comparisons using standard statistical tests. For example, some elements were drastically reduced or absent in some species. For statistical tests, the repeating song elements were averaged so that each individual wasp was treated as an N of 1. Where appropriate in similar species, duration and fundamental frequency were compared using analysis of variance (ANOVA) followed by Tukey's post-hoc test. Song element durations were log transformed to meet the assumptions of ANOVA. Pulse and terminal buzz frequency were compared between adjacent elements of each song for each species and all species together using linear regression. Principal components analysis (PCA) for each individual wasp was used to condense dimensionality of the data into principal components (PCs) based on communitive explained variance. Hierarchical cluster analysis (Ward method) of mean principal components of each species was used to group wasps. The resulting dendrogram was compared to the most recent Courtship songs in Cotesia (Hymenoptera: Braconidae) molecular phylogeny of Cotesia [44] to determine if the same groups were resolved (see Table 2 for genes used and GenBank ascension numbers of species included in this study).
Subsequent PCAs were performed on six different host-foodplant complex sources of C. congregata, on the MsT and CcC incipient species, and the two geographically isolated sources of C. nr. phobetri. Sources that displayed separation in the PCAs were followed by a Welch's unequal variance t-test to compare duration and frequency of song elements. The feasibility of matching songs back to their species or population based on song elements and PCs was tested using linear discriminant analysis. All statistical and multivariate analyses were performed with JMP v11 (SAS Institute, Cary, NC).

Description of wasp songs
Eleven species of Cotesia were collected at different sites in the United States (Table 1); one additional species was from a published recording (C. marginiventris). Some target host species collected did not yield any Cotesia. Multiple broods were collected of each species except for the uncommon C. teleae, which came from a single Antheraea polyphemus larva that produced only one living male concurrently with females. Cotesia congregata were collected from six different sphingid host species. Cotesia nr. phobetri, (currently undescribed) were supplied from a laboratory colony originating from host Grammia incorrupta (formerly G. geneura) that feed on forbs in grassland habitat and found in Redington Pass, Pima County, AZ [49]. Initially unknown specimens found as cocoon clusters independent of hosts in a recently mowed horse pasture in Gloucester County, VA, were identified to be either C. nr. phobetri or a closely related sister species based on morphology, cocoon structure, similarity of habitat, and song structure (Discussion, 8 th paragraph). Unparasitized caterpillars of Grammia virgo were found in an adjacent field, which suggest this species is the probable host. The other Cotesia species in our study were each collected from a single host species.
All twelve species of Cotesia generated songs by wing fanning that consisted of repeating high-amplitude pulses and, in most species, lower amplitude buzzes (Fig 2). Pulses were accompanied by abdominal movements. All males continued to produce songs until copulation or the female moved away; therefore, number of pulses was not used as a factor. Songs varied in duration of pulse, buzz, and pause elements (Table 3). Some songs were different Table 2. Available genes and GenBank ascension numbers of Cotesia species used in this study. Not all species have been sequenced using all four genes, and therefore are not included in the phylogeny.

Species
NADH1 mt16S rDNA n28S rDNA LW rhodopsin enough from the others as to not require statistical comparisons of individual elements, e.g., C. flavicochae produces a song with unique elements. Moreover, songs with considerably longer element durations had greater variance than songs with relatively short element durations (e.g., compare pause duration between C. teleae and C. congregata in Table 3), resulting in violation of the assumptions for ANOVA. Although an ANOVA could not be performed with all 12 species, an ANOVA was performed with four species that had song elements of similar durations and frequencies. Among all species, fundamental frequency ranged from 176 Hz in C. phobetri to 328 Hz in C. euchaetis (Table 4). All songs produced detectable harmonics up to 4-5 kHz for pulses and 1-2 kHz for buzzes (Fig 2). Analysis of individual elements was useful for comparing similar or sister species but of limited use for comparisons across all species. Courtship songs were divided into groups using a combination of element duration and frequency, and general patterns rather than focusing on individual song elements. The most common courtship song structure included pause-pulse-buzz elements repeating ca 2-3 times a second. A subset of four species with similar pause-pulse-buzz patterns was used to discern more nuanced differences. The songs differed in either duration of interpulse intervals and component elements (ANOVA: F 3,37 = 8.03, p = 0.0003; Fig 3A) or fundamental frequency (ANOVA: F 3,37 = 46.08, p < 0.0001; Fig 3B). In C. congregata, the courtship songs consisted of an initial buzz followed by repeating pulse-buzz elements with a short (20-26 ms) pause, followed by a pulse ("boing") that decays into a buzz [20]. Courtship songs of C. phobetri and C. euchaetis had similar patterns, although terminal buzzes after pulses were longer than in C. congregata. Cotesia nr. phobetri songs were similar to those of C. euchaetis but with shorter terminal buzzes. Cotesia marginiventris songs had pulse-buzzes that varied more in duration, producing a warble sound [21,24]. The song of C. marginiventris did not contain the discrete high-amplitude pulses present in C. congregata, C. phobetri, and C. euchaetis, but otherwise followed a similar pattern. Cotesia glomerata was similar but lacked discrete pauses between pulse-buzz elements and had a sudden, <50 ms, power spike at the start of each pulse that was not observed in other species (Fig 2). Three species produced courtship songs that consisted of rapid repeating pulses without long terminal buzzes. Cotesia empretiae produced a pulse train of~2 seconds that consisted of short repeating pulses with different durations. Cotesia diacrisiae produced rapid short pulses at a rate of four per second. Cotesia orobenae also produced rapid repeating pulses that were shorter in duration and had longer pauses compared to C. diacrisiae. These three species could be readily distinguished by waveform patterns (Fig 2).
Three species produced songs with long pauses between pulses. Cotesia rubecula produced songs with pulses similar in duration and pacing to those of C. phobetri and C. congregata but lacked a terminal buzz. Cotesia teleae produced a pulse-buzz element with pauses that were about five times longer than those of the C. congregata group. The C. flaviconchae courtship song was substantially different from other Cotesia songs. It was the only species that produced songs with a buzz before the high-amplitude pulse ("pre-pulse buzz") lasting 486 ± 32 ms at 286 ± 2 Hz. The buzz-pulse repeated every 4-9 seconds, while all other songs repeated in less than a second.

Analysis of all species
Songs from all 12 Cotesia species were grouped using PCA and cluster analysis. Frequency of adjacent pulse and terminal buzz elements could not be accurately calculated in all sections of songs produced by some species due to short durations of one of these elements (e.g., C. empretiae); however, adjacent pulses and buzzes were correlated in song sections containing sufficiently long durations of both elements (r 2 = 0.65, d.f. = 770, p < 0.0001; Pulse = -31.0 + 1.1 � TBuzz; Fig 4). Therefore, frequency was consolidated into one term by using the "pulsebuzz unit" frequency in the PCA. The PCA using the seven song elements resulted in four PCs explaining 91.7% of total variance. PC1 was best represented by duration of the pulse-buzz unit, interpulse interval, and pre-pulse buzz, PC2 by frequency, PC3 almost entirely by pause duration, and PC4 by frequency, pause duration, and pulse duration ( Table 5). The factor scores of the first four PCs differed significantly among some, but not all, species (ANOVA: For example, C. flaviconchae differs significantly from all other species by PC1 and PC2, whereas species with highly similar song patterns such as C. congregata and C. phobetri may only differ by one PC (in this case PC4) but not the others. Differences in one or more PCs among species indicated groups with most species forming close clusters with some overlap (Fig 5). Species separated from the main cluster and containing relatively long duration elements had greater variance in PCs (e.g., C. flaviconchae and C. rubecula). Hierarchical cluster analysis of species using the first four PC mean factor scores resolved four main groups (Fig 6). Group 1 consists of wasps with short rapid pulses, group 2 with pulses and terminal buzzes, group 3 with long pauses between pulses, and group 4 of only C. flavichonchae with a thus far unique song pattern. Groupings from the cluster analysis generally reflect genetic groups [44], with the exception of C. rubecula which lacks the terminal buzz found in all other related species in the "rubecula" group (Fig 7).

Differentiation of C. congregata host-foodplant complex sources
Songs from the different C. congregata host-foodplant complex sources could not be distinguished by courtship songs alone. The PCA using the six song elements present in  C. congregata resulted in three PCs explaining 88.8% of total variance and four PCs explaining 99.9% of total variance. PC1 was most represented by pulse-buzz unit duration, interpulse interval, and terminal buzz duration, PC2 by pause duration and frequency, PC3 by pulse duration, and PC4 by frequency (Table 6). A PCA of the two geographic sources of MsT and the one source of CcC host-foodplant complexes produced a similar component matrix (Table 7). High overlap of every PC prevents discrimination of host-plant complex sources (Fig 8) and geographically separated populations (Fig 9), even if means of some elements and the first threes PCs differ significantly between some groups (ANOVA: p < 0.001).

Differentiation of C. nr. phobetri by location
Songs from the two sources of C. nr. phobetri could be clearly distinguished by courtship songs. The PCA using the six song elements present in C. nr. phobetri resulted in three PCs explaining 94.5% of total variance and the fourth PC explaining the remaining variance. PC1 was most represented by duration of pulse and buzz components, PC2 by pause duration and frequency, PC3 by pause duration, and PC4 by pulse duration ( Table 8). The C. nr. phobetri populations from Virginia and Arizona can be reliably distinguished by PC1 (pulse and buzz durations) but not the other PCs (pause duration and frequency) (Fig 10). Linear discriminant analysis sorts wasps by population with 100% accuracy. Mean duration of the pulse and buzz components were longer in songs of wasps originating in Arizona than Virginia (unequal variance t-test, p < 0.001), with mean pulse-buzz unit duration 0.13 s longer (t 13 = -8.6, p < 0.0001). Overall song pattern and structure remained the same in wasps from both populations and were more similar to each other than to the other wasp species analyzed (Figs 4 and 5).

Discussion
Courtship songs of the twelve species of Cotesia presented in this study were variable yet distinguishable. Songs were characterized quantitatively by dividing songs into elements that were shared across most species. Principal components analysis was used to reduce dimensionality of correlated song elements. Species grouped using hierarchical cluster analysis corresponded to groups in the genetic phylogeny [44], with one exception. Some species that have not been sequenced can be provisionally placed into these pre-identified groups based on courtship song characteristics. Likewise, the general pattern of courtship songs may be predicted for species that have been placed within a phylogeny but were not found and recorded in this study. However, the relationships among the groups do not correspond strictly to the genetic phylogeny and the large differences among groups make their placement difficult. Differentiation within C. congregata by host-foodplant complex was not reliable; in contrast, C. nr. phobetri could be identified by source. Courtship song analysis has potential use for the systematics of this genus. Songs were generally unique to each species and could be distinguished by waveform and frequency characteristics. All songs consisted of repeating pulse and/or buzz elements generated from wing fanning, which is common among parasitic wasps. The song pattern was stereotypical for each species (Fig 2 and supplemental audio). Notably, songs of all species had overlapping ranges in the duration and frequency of some elements (Fig 5). Songs of some species with relatively longer elements (e.g., the pause element produced by C. rubecula) were more variable. The species representing the "rubecula" group produced the most similar song patterns (short pause, pulse, terminal buzz) as indicated by tight clustering in the PCA analysis (Fig 5). Songs of some species in this group differed only in the duration or frequency of one or a few elements (Fig 3). Although we attempted to capture the primary acoustic variation among species, some unique characteristics were not integrated into the PCA, e.g., the prepulse power spike produced by C. glomerata and the warble sound produced by C. marginiventris, which can be used to easily distinguish these species from others with otherwise similar song patterns. We do not rule out that some as yet unrecorded species may produce songs so similar to a sister species as to not be distinguishable, as was the case among the host-foodplant complexes of C. congregata. Moreover, playback experiments are necessary to determine whether wasps can distinguish between songs of closely related species. Importantly, some clustering of species was expected given the morphological and genetic similarity of many members of Cotesia.
Phylogenetics and taxonomy of the Microgastrinae are active areas of study [50]. Most work has been at the subfamily or genus level [51][52][53][54], on closely related species clusters or  cryptic species [47,[55][56][57], or by determining the evolutionary relationship with symbiotic viruses [58][59][60][61]. Considering the diversity of Cotesia and the difficulty of producing high-resolution phylogenies, not all of the same genes or all common species have yet been sequenced. Therefore, some species in this study cannot be placed reliably in current phylogenies (e.g., C. phobetri, C. teleae). Eight of the recorded species are included in the most recent genetic phylogeny for Cotesia reported by Michel-Salzat and Whitfield [44]. This phylogeny contains four identified groups, three of which are represented in our current study. Courtship songs of two species in the fourth group ("melanoscela") have been described in other studies [22,43]. In our study, species phylogenetically grouped together had similar courtship songs with the exception of C. rubecula. Several species not placed on the genetic phylogeny can be putatively placed within a group based on song characteristics (e.g., C. phobetri). The most basal genetic group containing C. empretiae and C. diacrisiae had songs that consisted of rapid repeated pulses although the placement of these species has low nodal support. Most songs of derived groups consisted of pulses and longer terminal buzzes.
The apparent phylogenetic signal of courtship songs allows for predictions of song structure before recording. For example, most species in the "rubecula" group have a pause-pulse-buzz pattern of similar duration (Fig 7). Species such as C. schizurae and C. electrae likely have songs similar in pattern to C. congregata and not more different than the more distantly related  Courtship songs in Cotesia (Hymenoptera: Braconidae) C. euchaetis or C. marginiventris. In both the genetic phylogeny and acoustic dendrogram, C. glomerata was placed as a sister group to most of the "rubecula" group, with its main distinction being the pre-pulse power spike. In the "kariyai" group, C. cyaniridis is predicted to have a song with a buzz that leads directly into a pulse with a long interpulse duration. The song may differ in details of timing and frequency, but otherwise should sound similar to the closely related C. flaviconchae (Fig 7). These predictions are supported by the recent recording of two male C. schizurae (host Schizura unicornis; 37.7549, -77.3458) from the same brood. This species, closely related to C. congregata, was targeted for collection but not initially found. The courtship song of C. schizurae (S13 Audio) is similar enough in structure and duration of pulses, buzzes, and pauses to that of C. congregata to be firmly placed as a closely related species. Although courtship songs can be used to construct species groups, predicting evolutionary relationships among these groups is more challenging due to the lack of data for intermediate species. Among other insect groups, phylogenetic signals have been reported for the diverse songs of psyllids, but divergence may occur more rapidly among sympatric species [62]. Predicting the courtship songs of other acoustically active insect groups is challenging due to rapid diversification of song characteristics unrelated to genetic distance (e.g., the Drosophila willistoni species complex [3] and Chrysoperla lacewings [63]).
Cotesia rubecula is the only species that deviates from expectation. It is genetically placed with species with pulse-terminal buzz patterns (e.g., C. congregata) but lacks a discrete terminal buzz (Fig 2). The most parsimonious explanation is that C. rubecula secondarily lost the long terminal buzz and replaced it with a long pause. The time between pulses is similar to those within its genetic group. Alternatively, C. rubecula may not belong in this group, although high nodal support for its inclusion makes this possibility less likely (Fig 7). Notably, C. rubecula has other characteristics that differentiate it from most species recorded in the "rubecula" and "glomerata" groups-it is solitary and the largest Cotesia species recorded. Determining whether a solitary or gregarious life history or relative size influences courtship song characteristics would require recording a broader range of species, particularly more solitary species (the only other being C. marginiventris in this study). Moreover, C. rubecula and C. glomerata are the only species collected that utilize the same host, Pieris rapae on Brassicaceae. Their songs may display greater divergence in part due to character displacement, which has been demonstrated for the songs of few insects [64]; however, extensive surveying suggests competitive exclusion between these two species over most of the range in the United States [65].
Cotesia teleae has a song that challenges direct placement into a group. The pattern of the pulse-buzz unit is similar in many ways to those of the "rubecula" group; however, it has a short pulse with a high energy terminal buzz that loses amplitude at the end (Fig 2). Most distinctly, there are long pauses between pulses. Possibly, it belongs with the "rubecula" groupthe cluster analysis supports a relation with C. rubecula-but its placement remains less certain than for other species without either genetic information or another species with a similar song pattern. The song of C. teleae is also the only one analyzed using a single male. A single parasitized polyphemus caterpillar (Antheraea. polyphemus) yielded few adult wasps. The brood began egression in October with most wasp larvae going into diapause, which was not broken in the lab and yielded only a single male concurrently with females. Considering that courtship songs are conserved within species and this male was healthy, the recorded song is presumably a reliable representation of this species. A second male emerged without a living conspecific female present and would not initiate courtship when presented with other species. Attempts to find a second brood over multiple years failed. Since additional samples of C. teleae are improbable, it was included in this study.
The "melanoscela" group, containing C. sesamiae and C. flavipes, is the only major group not included in this analysis. These two species, widely used as biocontrol agents of stemborer pests, are not native to North America and could not be acquired for this study. Similar to other Cotesia spp., their songs consist of repeating pulse, buzz, and pause elements with a frequency of 222-290 Hz [22,43] (note: our terminology differs from that used in these references). The overall pattern of higher-amplitude pulse decaying into a longer terminal buzz has some structural similarities to songs of C. marginiventris and others in the "rubecula" group; however, the time between high amplitude elements (termed buzz 1 in [22,43]) is nearly a full second in C. flavipes and C. sesamiae, which is considerably longer than most of the other Cotesia species recorded. The species of the "melanoscela" group would form a distinct cluster based on reported song characteristics, and may be placed close to the "rubecula" group based on overall pattern.
Courtship song analysis can be used to match unidentified wasps to species, particularly those with similar morphology. Two cases occurred during this study in which parasitoid cocoons were found separated from their host. Unknown Cotesia cocoons found on a garden tomato plant were identified as C. orobenae upon recording. Presumably, the cross-stripped cabbageworm (Evergestis rimosalis) hosts had decimated nearby cabbages and had then migrated to the tomato before parasitoid egression. The hosts were absent, leaving only the wasp cocoons remaining on the leaves (C. orobenae cocoons typically do not remain attached to the host). In the second case, three loose bundles of parasitoid cocoons were found with no host in a mowed horse pasture in Virginia, USA. After recording the adults, the species acoustically matched those grouped with C. congregata but not any currently recorded species. Subsequently, cocoons of C. nr. phobetri that originated from Arizona were received. These wasps produced courtship songs that very closely resembled those of the unknown Virginia wasps, and also were similar in morphology, cocoon structure, and habitat. This is the first known record of C. nr. phobetri outside of Arizona. Considering that this species utilizes a common caterpillar genus, Grammia, as hosts in a common habitat type, they may be widespread in the United States.
Courtship song elements may differ even among closely related species or host-associated populations. For example, allopatric populations of C. sesamiae and C. flavipes utilizing different hosts had courtship songs that differed in element duration and frequency [43]. Likewise, C. congregata originating from hosts M. sexta on tobacco (MsT) and Ce. catalpae on catalpa (CcC) differed significantly in pulse and pause durations though the differences were not enough to reliably distinguish all individuals [15]. We expanded this earlier finding by using four additional host-foodplant sources of C. congregata in Virginia and an additional population of MsT wasps from Indiana (Table 1). These six total sphingid host species represent two subfamilies so thus were phylogenetically diverse [66]. Mean song element duration (pulse and pause) and PCs differ among MsT and CcC wasps; however, the degree of range overlap with the additional sources prevents reliable discrimination by source (Figs 8 and 9). The slight differences may indicate recent reproductive isolation that over time may become discrete differences under sexual selection or genetic drift. Breeding crosses using these additional sources of C. congregata indicate a pattern of asymmetric hybrid female sterility with either MsT or CcC wasps, suggesting only two primary lineages (Bredlau et al., in submission).
In contrast, geographically separated populations of C. nr. phobetri differ in song element durations (Fig 10), even though they are similar enough to be recognized as the same species (S7 and S8 Audio). We cannot determine whether the Virginia and Arizona populations represent sister species, host-associated races, or isolated populations without additional information on reproductive compatibility and range. These populations are separated by 3,150 km and thus may be expected to have some differences in song elements regardless of species status. Another possibility is that laboratory rearing of the Arizona population for three years could have resulted in slight changes in courtship song elements, as reported for other braconid wasps [25]. Collecting wild C. nr. phobetri at multiple sites would be required to make that assessment. The other geographically separated samples came from C. glomerata and C. rubecula; however, not enough individuals were recorded to discern acoustic differences within these species.
Relatively small differences in songs among closely related species indicate a phylogenetic signal that may have useful applications for systematics. Sexual selection likely plays a role in the differentiation of some songs. However, within the majority of the "rubecula" group, songs consist of similar pause-pulse-buzz patterns more indicative of a slow build-up of differences rather than active sexual selection. Furthermore, song differences in reproductively isolated incipient species of C. congregata are slight and cannot be used to reliably identify host-foodplant complex sources. Song differentiation of reproductively isolated species may change over time via genetic drift in the absence of strong sexual selection, producing minor changes in elements among closely related species. In this scenario, songs may seem arbitrarily different among species with relatively small changes, yet still be conserved within species. Courtship shortly after emergence on the natal host-plant may play a role in limiting contact among sympatric, closely related species, thereby reducing selective pressure on courtship song differentiation. Likewise, other factors such as host-plant learning and adaptations to host immune systems may play a greater role in parasitic wasp speciation, leading to differing rates of song differentiation. In contrast, species clusters of Drosophila are reported to have large differences in courtship songs, suggesting strong sexual selection leading to differentiation before other traits [2][3][4]67]. Playback experiments using the D. buzzatii species cluster demonstrate that females are more likely to accept males with a conspecific song, supporting the role of sexual selection [68]. Likewise, cryptic species complexes of lacewings [8] and sand flies [10] can be reliably distinguished by courtship song patterns.
Courtship songs of Cotesia spp. are structurally complex and highly variable compared to those produced by other braconids. For example, courtship songs of other microgastrines vary from the consistent wing fanning sounds that increase in amplitude ca. every 2 seconds produced by Glyptapantales flavicoxis [28] (in a genus closely related to Cotesia [44]) to the short repeating pulse trains of Microplitis croceipes that may merge into a warble [24]. More distantly related braconids, such as those in the Opiinae [24][25][26], typically produce short repeating pulses similar in structure to the song of C. diacrisiae. Wasps in the Aphidiinae [27] and Euphorinae (Bredlau, unpublished) produce longer pulses (200-220 ms) with longer pauses between pulses (200-370 ms) that do not trail into terminal buzzes, similar in structure to the song of C. rubecula. The song of at least one ichneumonid, Compoletis sonorensis, also produces a song consisting of pulses (215 ms) and long pauses (215 ms) (Bredlau, unpublished). Among parasitic wasps, constant wing fanning or repeating pulses are the most common patterns. Considering that all Cotesia songs consist of pulses in one form or another, the ancestral song most likely also consisted of pulses or consistent wing fanning varying in amplitude that later developed pauses before the high amplitude components. Similarity of songs among species such as C. rubecula and those in other subfamilies likely evolved independently. For example, convergence of song structure has been reported for allopatric cryptic species of Chrysoperla lacewings [69].
The analysis presented in this study has several limitations. Even the relatively simple songs of parasitic wasps contain multiple acoustic elements and frequencies, often with different degrees of variance depending on song structure. Principal component analysis is useful for reducing the dimensionality of complex datasets to uncorrelated variables, and in identifying elements that contain the greatest variance. Moreover, PCA is a common method of data exploration widely understood by biologists and has been used in the comparison of songs in diverse taxa including birds (e.g. [70,71]) and insects [4,8,10,63,72]. We used PCA as a means to reduce the acoustic data for comparison and included the number of PCs that adequately explained aspects of the courtship songs in the cluster analysis. However, limitations such as uneven scaling of long vs short elements and time vs frequency were not accounted for because their biological significance is unknown. No one element could adequately capture the differences among songs; furthermore, some elements calculated were intentionally redundant. Alternative methods to a cluster analysis that consider the probability of a given acoustic tree among all possible trees should be considered with additional data. Additionally, the sequencing of all sampled wasps with additional genes or recordings of those already sequenced will permit a more thorough examination of song trait evolution.
This comprehensive study of courtship song diversity within a genus of parasitic wasps, Cotesia, implicates a wide diversity of song patterns that can be divided into groups based on duration and frequency of song elements. The basal song most likely consisted of regular pulses generated by high-amplitude wing strokes, as seen in other members of subfamily Microgastrinae [24]. This song diverged into the several distinct patterns among the major groups of Cotesia. Many wasps not yet recorded can likely be placed into these groups based on a combination of song structure and morphology. The unique structure of songs for each species can potentially be used for species recognition and as a reproductive barrier between cryptic species; however, the influence of sexual selection is uncertain. Despite measurable differences among species, the songs among C. congregata host-foodplant complexes cannot be reliable distinguished, suggesting that song differentiation does not proceed without other reproductive barriers. In total, fifteen Cotesia species have been recorded out of the estimated 1,000 species globally [50]. Considering the size of this genus, other entirely new song patterns may yet be discovered. When combined with additional genetic data, courtship song analysis should prove useful in determining the systematics and evolutionary history of groups of parasitic wasps, particularly in this highly diverse and agriculturally important taxon.