Fossil specimens and morphological analysis

The fossil material described here was collected by Erwin Rettig and Jan Verkleij and is stored in the Fur Museum in Nederby (FUR). No permits were required for the described study, which complied with all relevant regulations. Photos of the newly described fossils were made with a Keyence VHX 600 camera system with a magnification of 50–200 under ethanol (to enhance the contrast), using stacking techniques. For each fossil, an interpretative illustration was made in Adobe Photoshop (v. 21.2.3), using both part and counterpart of the fossil and several photos of details as templates. The interpretative illustrations are coloured in different grey tones according to the visible colouration of the corresponding body parts. Solid lines indicate certain interpretation, whereas dotted lines illustrate uncertain interpretations.

The morphological terminology follows Broad et al. [21], except for the wing terminology, which is illustrated in Fig 1A for the fore wing and follows Spasojevic et al. [33] for the hind wing. Tergites and sternites are numbered in the text and abbreviated as “T1”, “T2”, etc. and “S1”, “S2” etc., respectively. Measurements for the descriptions were made in ImageJ [34]. If paratypes are available, the range of the measurements is given, and the measurement of the holotype is provided in brackets. If both the left and right version of an appendix were complete, as it often occurred in fore or hind wings, then the average over both is given in the description. In crooked fossil specimens, the body length was measured in segments. We made use of open nomenclature [35], which is often used in palaeontology, to communicate uncertainties in the taxonomic placements. A question mark is added, in our case after the subfamily or the genus, when the identification was uncertain because a key character of this group was indiscernible.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Fore wing venation shape and landmark placement.

(A) Wing vein terminology used throughout the study, shown on an idealised ichneumonid fore wing. Vein 2m-cu marked orange; on this vein, a curve with six semilandmarks was added. (B) Placement of 21 fixed landmarks.

https://doi.org/10.1371/journal.pone.0275570.g001

Taxon sampling for reference dataset

We built a large reference dataset out of extant and fossil specimens and used different subsets of it depending on the fossil species and the analysis (Table 1). For the evaluation of differences and similarities in fore wings among extant subfamilies (question i), we used 333 standardised illustrations from Townes [36–38]. This work comprises genus-level revisions of all ichneumonid subfamilies except Ichneumoninae and Hybrizontinae, with at least one habitus plate per genus, in which the wings are displayed in a standardised fashion. However, it does not include size information for the particular species in the illustrations. We thus retrieved wing or body lengths of the illustrated species from the literature using the Taxapad database [17] and used the mean if reported lengths varied within species (see S1 Table for specimen information). Townes illustrations [36–38] mainly depict female specimens, with very few exceptions. From the 42 extant subfamilies, we strategically chose nine as follows. Ctenopelmatinae, Metopiinae, Orthocentrinae, Pimplinae and Tryphoninae were included because they resemble our newly described fossil specimens the most and are thus viable candidates for their placement. Campopleginae and Cremastinae were added because of their similarity in fore wing venation and close phylogenetic relationship with one of the candidate subfamilies (see below). Finally, Ophioninae and Tersilochinae were added due to the known uniqueness of their fore wing characters [21], so we can adequately represent most of the variation found in the ichneumonid fore wing. Other closely related, but small subfamilies with only a few species were not included, nor were other subfamilies with typical body characteristics that are not found in our fossils, such as a very long and slender first tergite or a strongly pentagonal areolet in the fore wing. The species included from the nine chosen subfamilies are listed in S1 Table, and included 37 Campopleginae, 23 Cremastinae, 81 Ctenopelmatinae, 28 Metopiinae, 36 Ophioninae, 22 Orthocentrinae, 52 Pimplinae, 16 Tersilochinae, and 38 Tryphoninae species.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. An overview of the datasets and analyses.

https://doi.org/10.1371/journal.pone.0275570.t001

For the questions of classification of our new fossil specimens (questions ii & iii), we pre-selected candidate subfamilies based on body characteristics from the set of five potential candidates (Ctenopelmatinae, Metopiinae, Orthocentrinae, Pimplinae and Tryphoninae) (Table 1). For Triclistus bibori sp. nov., we ended up with a single choice (i.e., Metopiinae) based on body characteristics (for details, see taxonomic section on this species). In order to still evaluate the power of our approach for placing fossils in a subfamily, we here included two related subfamilies which have a very similar fore wing venation (Campopleginae & Ctenopelmatinae). In Crusopimpla weltii sp. nov. and Entypoma? duergari sp. nov., Pimplinae was among the candidate subfamilies for placement. In addition to the extant taxa, we there also added nine previously described fossil species from the Fur Formation (S1 Table), all of which belong to this subfamily [30] (in the tenth known species from Fur, the specimen’ fore wings were too distorted to be used in geometric morphometrics).

Landmarks for geometric morphometrics

To obtain shape information of the fore wing venation, we placed 21 fixed landmarks (LM) and one curve with six semilandmarks along vein 2m-cu (Fig 1B) on each digital image using the softwares tpsUtil (v. 1.78, [39]) and tpsDig2 (v. 2.31, [40]). The semilandmarks were chosen because the shape of 2m-cu has been mentioned previously as potentially informative for ichneumonid higher classification [21, 30]. Size information was included by setting the scale manually for each specimen. Landmarks were placed on the left fore wing in each illustrated extant taxon. In each fossil, there is at least one landmark that was placed with some uncertainty due to incomplete preservation or difficulties in the interpretation of the wings (Table 1). In all fossil wing analyses and corresponding other analyses, we excluded these uncertain landmark(s) for each fossil specimen.

Despite evidence for sexual dimorphism in wings in some ichneumonids [41, 42], we did not consider it in our study. Nearly all specimens in our reference dataset were female, and in fossils, the sex can only be clearly determined if there is a visible ovipositor or well-preserved male genital. In the case of Ebriosa flava gen. et sp. nov., Triclistus bibori sp. nov., and Lathrolestes? zlatorog sp. nov., no ovipositor is visible, leading to the assumption that they are male specimens. Rust [26] observed in 83 complete ichneumonid specimens from the Fur Formation only 17 individuals without ovipositor, identifying only one of them as a male for sure. So, the possibility remains that our fossils might be female, with a short or not well-preserved ovipositor, but if sexual dimorphism is present, a bias in our inferred affiliations cannot be ruled out.

Data quality assessment

As using geometric morphometrics to place fossils is a rather rarely used approach, we conducted three kinds of data quality assessments. In all three assessments, we tested for a potential bias using a Procrustes analysis with the function gpagen, followed by a Procrustes ANOVA with the function procD.lm from geomorph (vers 4.0.4, [43]) in RStudio (ver. 1.2.5041, [44] R ver. 4.0.2, [45]). Additionally, the shape data was projected in morphospace with a Principal Component Analysis (PCA) to visually identify any potential biases.

First, we checked for biases regarding landmark placement by two different persons with a subset of our extant dataset (Table 1), i.e., 27 genera of the subfamily Campopleginae, with all 21 landmarks placed on the illustrations in Townes [37]. Second, we examined possible bias when using standardised drawings of wings in comparison with photos, again using a Campopleginae subset (Table 1). A bias might occur because it is not entirely clear how the illustrations in Townes [36–38, 46] were produced, although it appears that the wings were removed from the specimens, flattened and glued on cardboard in the majority if not all cases (D. Wahl, curator of Townes collection at Utah State University, Logan, pers. commun.). To make sure that the Townes illustrations can be used for our purpose, we prepared the fore wings of 14 species from the illustrations for which we also had fresh material available (see S2 Table for taxon sampling list). In the PCA we included the species that were available as photo and as illustration, as well as an additional different but congeneric species for many of them.

Third, we aimed to identify potential biases due to preservation artefacts in fossils. The left and right fore wing (right ones being mirrored), and part and counterpart in each fossil were tested for significant shape variation, using the same approach as outlined above. This comparison included the five fossils studied here, although alternative taxon samplings would be conceivable, but would rely on making a priori decisions about their final placement. Since in each fossil, there was at least one landmark with an uncertain placement, we removed LM 1, 2, 18 and 19 in all our fossil data in this analysis.

Examining informativeness of fore wings for subfamily classification

We wanted to evaluate to what extent the shape of fore wings is influenced by subfamily affiliation and/or size. For this, we used the 21 fixed LM of the extant species, performed a Procrustes analysis and calculated centroid sizes with the geomorph package [43]. To determine the allometric trends across subfamilies, a Procrustes ANOVA was implemented with the function procD.lm. A regression analysis based on the PC scores and the centroid sizes was then performed to test the shape and size correlation using MorphoJ [47]. Since there may be a link between the centroid size and the subfamily attribution (tested with a Procrustes ANOVA), we performed a pooled within-group regression. In the pooled analysis, the centroid size distribution within a predefined group (here the subfamilies) is centred to the mean of the centroid size of the groups (the mean centroid size value equals to 0) preserving the relative range of centroid size but deleting the impact of the group on the shape changes. A regression is then performed using the new centroid size configuration.

In a next step, we checked which subfamilies exhibit similar LM configuration of the wing venation and to what extent the fore wing shape venation varies within each subfamily. After minimising the distances between the LM of all extant species with a Procrustes analysis, the meanshape was calculated using the function mshape from the package geomorph [43] and the full Procrustes distances were calculated in a pairwise manner using the function procdist from the package shapes (vers. 1.2.5 [48]). We filtered the Procrustes distances, so we could calculate the mean Procrustes distance of each subfamily to another subfamily and compare the shape variation among them. The Procrustes variances per subfamily were calculated with morphol.disparity from geomorph [43].

We further analysed the shape variation of the extant taxa in a Canonical Variates Analysis (CVA) with the R package Morpho (vers. 2.9, [49]) to view the maximised amount of among-group variance, relative to within-group variance [50]. The CVA additionally computes a cross-validation test to analyse which subfamilies can be statistically differentiated from one another based on the LM configuration of the wing venation.

Fossil placement

From here, we worked with the candidate subfamilies that resemble our new fossil specimens, removing the LM that were placed with uncertainty in the respective fossil (Table 1). For a first general visualisation, we plotted our complete reference dataset, together with all fossil specimens in a PCA. The PCA is then repeated for each fossil specimen and its candidate subfamily.

To evaluate the affiliation of the fossils we performed a cross-validated CVA with the candidate subfamilies of each new fossil (Table 1), but not yet including the fossil itself, to check how well the taxonomic groups can be distinguished by LM configuration of the fore wing venation alone. The CVA results in an overall classification accuracy, including Kappa statistics (to which extent the collected data correctly represents the measured variables, ranging from -1, where the data is representing the variables worse than random, to +1, which is a perfect agreement of data and measured variables [51]. This helps to estimate the certainty when a fossil is later placed within a subfamily.

In a next step, we tested the affiliation of each fossil with the function ‘classify’ from the package Morpho [49], using the result of the conducted CVAs. This allows us to use the reference dataset as a test dataset and define the fossil specimens as new input to be assigned to a predefined taxonomic group, in our case subfamilies. If the CVA resulted in a high classification accuracy of >80% (supported by a high Kappa value of >0.7) and the fossil specimen could then be assigned to a given subfamily with a high posterior probability (pp >0.9), we used this subfamily for a narrower evaluation to tribe level or to genus group with the function groupPCA from Morpho. To do a between-group PCA (bgPCA) is compatible in this situation, since a CVA is not feasible with small sample sizes [52], as we sometimes only had a few species available per tribe or genus. The function groupPCA calculates Euclidean distances between the tribes’ averages with the fossil specimen remaining in its own group. P-values are based on permutation testing of 10’000 replications. Resemblance of the fossil specimens with extant tribes is discussed by evaluating the Euclidean distances in the descriptions of the new fossil species.

Data quality assessments

The results of the quality assessments are shown in S2 Table. In brief, we found no indication of any systematic biases in any of the three quality assessment analyses. Furthermore, the variation introduced by different persons setting the landmarks, by using drawings versus photographs of fore wings, and by fossilisation artefacts was in nearly all cases smaller than the variation between genera, suggesting that the method is suitable for higher-level classification.

Landmark placing by two persons

Our assessment showed that the placement of landmarks on fore wings by two different persons is highly consistent, assuming they followed the same rules of placement. The variation in shape explained by different persons is very low and there seems to be no significant difference between them (R²: 0.8%, p = 0.8), according to a Procrustes ANOVA analysis (S2 Table). Indeed, the two replicates based on the same illustration, but with landmarks placed by different people, cluster very closely together in all cases on a PCA plot (Fig 2A, blue and yellow dot within each green triangle).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Data quality assessments.

(A) PCA plot with landmarks placed by two different persons and comparison of illustrations and photos. Green triangles show the data of the same species three times, twice from an illustration, measured by two different persons, and once from a photo. Grey triangles represent species from the same genus, one species measured by two persons from an illustration and a photo of a different species, in order to additionally show some of the intra-generic variation. In those cases where our taxon sampling allowed it (Dusona, Rhimphoctona, etc.), both aspects were combined. The genus ID in the figure is composed of the first three letters of the genus (Dia–Diadegma) and the abbreviations, as well as the included taxa are listed in S2 Table. (B) Data quality assessment in fossils. PCA showing shape variation within and among the five newly described fossil species, “m” is representing the part (main part) and “c” the counterpart of the fossils, whereas the black outline is marking the right fore wings, the ones without black outlines the left fore wings.

https://doi.org/10.1371/journal.pone.0275570.g002

Photos versus illustrations

When comparing photos (black dots, Fig 2A) with illustrations (blue and yellow dots, Fig 2A) of the same species (green triangles, Fig 2A), the shape data was very similar within species, and was nearly always smaller compared to the distances between genera (Fig 2A, S1 File). The variation of shape is only little explained by the use of either photo or illustration (R²: 9%, p = 0.5), but mostly by species (R²: 88.2%, p = 0.5) (S2 Table). And even when different species within a genus were sampled (grey triangles, Fig 2A), they were usually clustering closely together compared to the overall shape variation within the subfamily. Only some of the largest genera of the subfamily, such as Dusona Cameron, 1901, Hyposoter Förster, 1869 and Bathyplectes Förster, 1869, exhibited a rather larger distance among different species, which at times exceeded that between some of the smaller genera. These genera are known for their rather high intrageneric diversity and might in fact turn out as not monophyletic in the future [53]. The similarity of the fore wing venation shape pointed out that many species could be correctly put into their genus, no matter if a photo was used or an illustration.

Fossilisation artefacts

Fossil specimens can show various artefacts, especially distortions resulting from the fossilisation process, which might not always be obvious but could nevertheless hamper geometric morphometric analyses. In our assessment with the fossil fore wings, we found the variation within specimen explained by side (left and right fore wing, R²: 0.8%, p = 0.25) and part (part and counterpart, R²: 0.5%, p = 0.57) rather minor: Left and right, as well as part and counterpart of the fore wings do not differ much within each fossil specimen (Fig 2B, S2 Table). We thus decided to continue with only the best-preserved imprint of the fore wing in the fossil specimens.

Allometry in Darwin wasp wings

The analysis of shape changes in wing venation related to size shows extremely significant correlation and thus strong signal for allometry (ANOVA: 21.9%, p < 0.001; regression: 21.9% predicted, p < 0.001, Table 2 & S4 Table). While small wings possess an enlarged pterostigma and a generally broader fore wing, larger ones have a slender pterostigma and are narrower in the anterior-posterior axis, and stand out in having the LM 11 and LM 15 very close together, and the 2cu-a longer than 3Cu (Fig 3A). Nevertheless, it is worth noting that the centroid size within the nine different subfamilies is not random. Tersilochinae and Orthocentrinae are the smallest studied ichneumonids here, while the Ophioninae are the largest ones. Indeed, shape changes associated with both, centroid size and subfamily affiliation, are extremely significant (ANOVA: 6.6%, p < 0.001). Moreover, the allometric trajectories are not similar across subfamilies (Fig 3). Then working on residuals of the regression of this analysis would delete shape information related to phylogeny. So it is necessary to pool the data by subfamily. When pooling the data by subfamily, shape changes in wing venation related to size showed an extremely significant correlation with a strong signal for allometry (Regression: 12.1% predicted, p < 0.001, S4 Table). Thus 9.8% of the shape changes may be related to both the subfamilies and the centroid size when considering the results of the regression. Relatively smaller wings are here characterised by broader cells, especially the cells 2R1 and 3Rs, while the larger ones are narrower, causing the cells 2R1, 1M+1R1, 2M and 2Cu to be elongated (Fig 3B). It is interesting to note that the shape differences in the maximum extreme shape between the unpooled and pooled analysis (Fig 3) are related to subfamily affiliation. Indeed, the characters observed in the unpooled analysis, like the 2m-cu being very close to vein 2-Rs+M, the vein 3Cu being shorter than vein 2cu-a, and the very long vein 1-R1 that almost reaches the height of LM 20, are absent or reduced from the pooled one and are typical for the subfamily Ophioninae.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Correlation of the centroid size and the wing shape in ichneumonid subfamilies.

(A) Regression analysis based on the raw data, where centroid size is plotted against the regression score, with linear regressions slopes shown by subfamily. (B) Centroid size against regression score pooled by subfamilies, with the centroid size distribution of each subfamily centred with its mean on zero. The black wings at the bottom of each plot represent shape at low and high centroid size and centroid size range, respectively. The underlying grey wing represents the mean shape of all taxa.

https://doi.org/10.1371/journal.pone.0275570.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Result of the procrustes ANOVA of shape with centroid size and subfamily affiliation.

https://doi.org/10.1371/journal.pone.0275570.t002

Shape variation among and within subfamilies

When comparing the mean Procrustes distances between the subfamilies, Tryphoninae, Ctenopelmatinae, and Campopleginae are the subfamilies that have the most similar fore wing venation to one another (Fig 4). Tersilochinae show the most distinct mean wing venation compared to others and are the most distinct from Metopiinae, Pimplinae, and Ophioninae (Fig 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Means of Procrustes distances between species pairs among and within subfamilies.

The means of the pairwise Procrustes distances for members of each subfamily pair are given in the left portion of the graph, sorted in ascending order. Whiskers indicate the standard error for each subfamily pair. On the right, the mean intra-subfamily pairwise distances are shown as barplots, with dashed lines extending to the left portion of the graph to allow direct comparison of mean intra- and inter-subfamily distances. The smallest distance between subfamiles is observed between Tryphoninae and Ctenopelmatinae and is lower than the three largest observed intra-subfamilial distances (Ophioninae, Cremastinae and Metopiinae). The largest distance is the one between Ophioninae and Tersilochinae and is about twice as high as the distances within Ophioninae. Horizontal lines and abbreviations of subfamilies are coloured according to the colour code above.

https://doi.org/10.1371/journal.pone.0275570.g004

The highest mean of Procrustes distances of pairs within subfamilies is reached by Metopiinae and Cremastinae (Fig 4, horizontal lines). The mean within those two subfamilies is higher than in many among-subfamily distances, while the mean within-subfamilies is rather consistent otherwise (S3 Table). And although the mean of Procrustes distances within-subfamily is not exactly the same as Procrustes variance, the results are very similar with the highest Procrustes variance found in Metopiinae (0.01089), closely followed by Cremastinae (0.0173) (S3 Table). Lowest Procrustes variance, and therefore the most consistent fore wing shape is found in Tersilochinae (0.00445) and Campopleginae (0.00482).

Shape variation related to subfamily

The shape changes accounting the subfamily affiliation are high with 25.8% (p < 0.001, Table 2). The classification results based on the pooled data and the raw data (S4 Table) are equally extremely efficient (0.3% of reclassification differences). Thus, in order to simplify the process of producing our results, we will perform our analyses only using the raw data. We conducted a cross-validated CVA of the reference taxa (Table 3A), where each individual was tested against the remaining individuals grouped in subfamilies, resulting in an overall classification accuracy of 87.9%. Ophioninae species were 100% assigned correctly, followed by over 90% correctly classified Campopleginae, Tersilochinae and Pimplinae. The subfamily Tryphoninae was the one with the fewest correct assignments. Over 20% of the Tryphoninae were falsely placed in Ctenopelmatinae, but only around 5% of Ctenopelmatinae were wrongly placed within Tryphoninae. Because about 60% (CV1 and CV2) of the shape variation split mainly Ophioninae and Tersilochinae (S1 Fig), and none of them represents a possible target subfamily of the later tested fossil specimens, we repeated the CVA without those two subfamilies (Table 3B, Fig 5).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Canonical variation analysis with seven extant subfamilies.

Examples of the shape at the maximum and minimum of the first four canonical axes. (A) Plot of the first two CV axes. Metopius Panzer, 1806 species are marked with a black contour. (B) Plot of the third and fourth CV axis. If subfamilies are more or less separate from others, we coloured the max/min shape of the wing accordingly. Grey wing outlines represent the general shape if several subfamilies overlap. The outlines on the wing illustrations are interpretative and oriented on the specimen that has the minimum/maximum CV score on the respective axis.

https://doi.org/10.1371/journal.pone.0275570.g005

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Cross-validated classification results in % from their CV scores.

https://doi.org/10.1371/journal.pone.0275570.t003

Without Ophioninae and Tersilochinae, the CV1 (38.1% explained variance) separates almost all Pimplinae and Orthocentrinae, the two subfamilies of Pimpliformes in our dataset, from the other groups, which all belong to Ophioniformes. There is some overlap though with some Tryphoninae, which is probably the sister group of the other ophioniform subfamilies. Pimplinae and Orthocentrinae can be distinguished from the other subfamilies mainly by an evenly outwards-bowed 2m-cu vein and the length of vein 4Cu being longer as vein 5Cu (Fig 5A). Regarding CV1, species of Cremastinae for example represent the opposite shape of Pimplinae and Orthocentrinae, where vein 4Cu is shorter than vein 5Cu, and the 2m-cu is bowed inwards. Along CV2, 26.6% of the variation is captured, and it represents the width of the fore wing to some extent, again the curve of the 2m-cu vein, and areolet size and shape (Fig 5A). Most Metopiinae split here from the rest by having a very slender wing shape in the anterior-posterior axis, a small and slender pterostigma, narrowed cells and the 2m-cu vein slightly inward bowed and sinuous. Metopiinae also have vein 3Cu longer than vein 2cu-a and species in the genus Metopius possess a relatively large areolet (Fig 5). In CV3 (15.2% explained variance), all Cremastinae except Belesica pictipennis Tosquinet, 1896 split from the other subfamilies, mainly due to the straight and interstitial placement of vein 1cu-a to 1M + 1Rs, where 1Cu is very short or absent (also visible in the mean shape in Fig 6), and a slightly more elongate overall fore wing. In CV4, the Orthocentrinae also mostly group separately, overlapping some Ctenopelmatinae. The wing venation separates Orthocentrinae by having shorter 2R1 and 2Cu cells, a broader pterostigma and 3Cu about the same length as 2cu-a. CV5 (7.7% explained variance) separates Campopleginae from many other taxa, due to their relatively short 2Cu cell, a longer 3Cu cell, as well as the anteriorly inwards bowed 2m-cu vein (S2 Fig). CV6 (4.2% explained variance) shows a slight shape difference in some Tryphoninae that have a slightly wider fore wing, resulting in some broader 2M, 3M, and 2R1 cells (S2 Fig). Other than that, taxa in Tryphoninae seem difficult to separate entirely from taxa of Ctenopelmatinae in the CVA or the comparison of their mean shapes (Fig 6).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Fore wing mean shapes of subfamilies and fore wings of the five new fossil species.

Mean shapes of all nine subfamilies, where Metopiinae were split into two groups: the genus Metopius with a very distinct fore wing venation, and the rest of Metopiinae. The mean shape of areolets are represented open (eg. Tersilochinae) or closed (eg. Campopleginae) based on the majority of included species. Group specific venation shapes are coloured in the subfamily colour code when mentioned in the results or discussion. Names of the new fossil specimens are underlined with the colour of the respective subfamily they were placed in.

https://doi.org/10.1371/journal.pone.0275570.g006

Subfamily placement of the newly described fossil species

The plots of PC1 against PC2 (Fig 7) of the candidate subfamilies and the new fossil specimens show first tendencies of subfamily affiliation. Some specimens grouped better within a subfamily e.g. Crusopimpla weltii sp. nov. in Pimplinae (Fig 7B), while others grouped less explicitly, e.g. Lathrolestes? zlatorog sp. nov. (Fig 7E) in between Ctenopelmatinae and Tryphoninae.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. PCA analyses of fore wing shape including extant and fossil taxa.

(A) PC1 against PC2 of seven subfamilies, including fossil Pimplinae and the newly described fossils. All 21 fixed and six semilandmarks were used, and the landmarks that were missing in the fossil specimens were estimated here. (B–F) PC1 and PC2 of the fossil species together with their candidate subfamilies. Landmarks of the reference subfamilies were adjusted to fit the respective fossil specimen (Table 1).

https://doi.org/10.1371/journal.pone.0275570.g007

The candidate subfamilies are tested in a separate CVA analysis without the fossil to see how often their members are assigned correctly in the given comparison (Table 4, light grey). Subsequently, the fossil specimen is added and assigned to an extant subfamily (Table 4, dark grey), showing the posterior probability for each candidate subfamily. In Crusopimpla weltii sp. nov. (FUR #11550) and Entypoma? duergari sp. nov. (FUR # 11264) the classification result of the CVA and the posterior probability of the fossil classification were considered clear enough (classification accuracy of CVA >80%, Kappa value >0.7, posterior classification >0.9), to continue the bgPCA with narrower groups of the resulting subfamily, Pimplinae in the former, Orthocentrinae in the latter case. In Ebriosa flava gen. et sp. nov. (FUR #11112) and Lathrolestes? zlatorog sp. nov. (FUR #10651), where the two very similarly looking subfamilies Ctenopelmatinae and Tryphoninae were considered candidate subfamilies, the classification results are less clear and for both those fossil specimens the bgPCA was conducted with tribes of both candidate subfamilies. An additional argument for testing both candidate subfamilies in Ebriosa flava gen. et sp. nov. (FUR #11112) is that the resulting classification disagreed with the fore wing characteristics of two bullae in vein 2m-cu, which is not caught in the geometric morphometrics analysis. The specimen of Triclistus bibori sp. nov. (FUR #13809), who had three uncertain and therefore removed LM (Table 1), was classified as Ctenopelmatinae (Table 4). In cell 2Cu, also represented by the here missing LM 18 and LM 19, differences are visible between the mean shapes of Ctenopelmatinae and Metopiinae (Fig 6), so we additionally included two nicely preserved paratypes in the classification analyses, where only LM 1 was uncertainly placed (Table 1). Both those paratypes (FUR #11215 and FUR #13809) were classified as Metopiinae (Table 4). For Triclistus bibori sp. nov. we therefore used first, tribes of all candidate subfamilies (genus Metopius and other Metopiinae split) and second, all genera of Metopiinae in the bgPCA, since there are no defined tribes. The resemblance of the fossil species fore wing venation, according to the Euclidean distances of the groups averages from the bgPCA, are included in the fossil species descriptions below.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Cross-validated classification results and probabilities of the fossil assignment for classification.

https://doi.org/10.1371/journal.pone.0275570.t004

Systematic palaeontology

Hymenoptera Linnaeus, 1758

Ichneumonidae Latreille, 1802

Pimplinae Wesmael, 1844

Crusopimpla Kopylov, Spasojevic & Klopfstein, 2018

Crusopimpla weltii Viertler, Spasojevic & Klopfstein, sp. nov. (Fig 8)

urn:lsid:zoobank.org:act:3F7CFB00-9C3F-4F0B-91D4-154A2D2B3612

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 8. Holotype of Crusopimpla weltii sp. nov. (FUR #11550).

(A) Photo of the fossil part. (B) Interpretative drawing, part and counterpart were used as templates. Scale bar: 2mm.

https://doi.org/10.1371/journal.pone.0275570.g008

Diagnosis.

The genus differs from other Tryphoninae and Ctenopelmatinae genera in having a strongly curved M&Rs vein in the fore wing, a character not present either in any other ichneumonids. In addition, the genus is characterised by a long and wide pterostigma, a 1R1 vein that is shorter than the pterostigma length, an evenly bowed 2m-cu with 2 bullae, an at least partly areolated propodeum, and a broad and gradually widening T1 with the spiracle in the anterior half and latero-median carinae reaching beyond the middle.

Ebriosa flava Viertler, Spasojevic & Klopfstein, sp. nov. (Fig 9)

urn:lsid:zoobank.org:act:92A2E9BC-1712-4289-8295-7DC088E1B982

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 9. Holotype of Ebriosa flava gen. et sp. nov. (FUR #11112).

(A) Photo of the fossil part. (B) Interpretative drawing, fossil part and counterpart were used as templates. (C) Strongly bowed M&Rs vein of the left fore wing, which is the name giving character. (D) Fore wing of the paratype #13812. Scale bars: (B) 2mm, (C) 1mm.

https://doi.org/10.1371/journal.pone.0275570.g009

Description.

Body 6.1–6.9 mm (6.1 mm). Head, mesoscutum and wing venation orange to brownish. Pterostigma with dark brown spot in centre. Antenna, rest of mesosoma and entire metasoma yellowish.

Head. Rather short. Flagellum 5.3 mm, with around 31 segments, evenly thick throughout entire length; basal segments not clearly visible, appear at least twice as long as wide; median and apical segments slightly longer than wide. Traces of occipital carina dorsolaterally visible.

Mesosoma. Short and stout. Notauli seem present at least in the very front, but length and shape further back not discernible. Propodeum higher than long, with traces of carination. Legs not clearly visible, appear slender.

Wings. Fore wing 5–5.5 mm (5 mm). Areolet closed, quadrate-oblique, 3rs-m 1.5x longer than 2rs-m, 4M very short. 2m-cu evenly curved outwards, with two bullae that cover together around 35% of total 2m-cu length. M&RS vein strongly curved. 4Cu about 1.4x 5Cu. 4Rs straight. Ramulus absent. 1cu-a interstitial to 1M + 1Rs and intersection thickened; 1cu-a emerges in a right angle. Pterostigma 2.7x as long as wide, 1.1x length of vein 1R1. Cell 2R1 3.2x as long as wide. 2Cu 0.85x 1M + 1Rs, 1.45x r-rs. 1m-cu&2-Rs+M vein angled. 3Cu clearly longer than 2cu-a. Hind wing, 1RS 0.8x rs-m. 1Cu slightly longer than cu-a.

Metasoma. Dorsoventrally compressed. T1 gradually widening and 1.14x as long as wide posteriorly; spiracle in the anterior half; traces of latero-median carinae reach more than half length of T1, parallel to slightly diverging. T2 0.58 as long as posteriorly wide. T2–T4 constantly widening, with and after T5 tergites narrower again, T7 appears very short.

Orthocentrinae Förster, 1869

Entypoma? Förster, 1869

Entypoma? duergari Viertler, Spasojevic & Klopfstein, sp. nov. (Fig 10)

urn:lsid:zoobank.org:act:5D05BA68-9012-4B64-B9CD-31A7BBBFCEC6

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 10. Holotype of Entypoma? duergari sp. nov. (FUR #11264).

(A) Photo of the fossil part. (B) Interpretative drawing, fossil part and counterpart were used as templates. Scale bar: (B) 2mm.

https://doi.org/10.1371/journal.pone.0275570.g010

Description.

Body 4.4–4.8 mm (4.4 mm). Head, mesosoma and T1 dark brown. Antenna, legs, T2 and T3 seem yellow/orange. T 4–8 and ovipositor sheaths brown.

Head. Short and rather small. Flagellum 2.5–3 mm (2.5 mm), number of segments unclear but at least 16; evenly thick throughout whole length. Temple appears narrow.

Mesosoma. Rather short and stout. Notauli either absent or weak; if present then not extending beyond middle of mesoscutum. Mesopleural furrow more or less straight. Metapleuron about as long as wide. Pleural carina present; unclear if reduced posteriorly or complete. Propodeum evenly rounded in profile, about as long as high; with longitudinal carinae present, other carinae unclear. Legs seem slender. Hind leg with coxa of normal dimension, slightly longer than wide; femur 3.2x as long as wide; tibia dimensions normal; tarsal claws appear simple.

Wings. Fore wing 2.9–3.7 mm (2.9 mm). Areolet closed, quadrate-oblique, 4M short, 2+3M and 3rs-m similar in length. 2m-cu slightly and evenly curved outwards, with two bullae, which are covering about 36% of total 2m-cu length. 4Cu 1.8x 5Cu. 4Rs straight. Ramulus absent. 1cu-a interstitial to 1M + 1Rs, with 1cu-a bowed distally. Pterostigma about 2.5x as long as wide; 0 .8x 1R1. Cell 2R1 2.3x as long as wide. 2Cu 0.83x 1M + Rs, 1.1x r-rs. 1M + Rs 1.3x r-rs. 1m-cu&2-Rs+M angled. 3Cu clearly longer than 2cu-a. Hind wing with M + Cu slightly curved distally; 1Cu longer than cu-a, last abscissa of RS and CU seem mostly transparent.

Metasoma. Depressed or cylindrical. Sclerotized part of S1 seems not to reach spiracle, apparently free from T1. Laterotergite 1 indistinct. Laterotergites 2, 3 and 4 rather broad, about 0.4–0.5x as wide as long; appear to be separated from dorsal part of tergite by a crease. T1 2x as long as posteriorly wide, box-like; with spiracle in anterior half; T2–T7 similar in length, very short and wide. Ovipositor around 0.38x length of metasoma; ovipositor and sheaths seem parallel-sided; shape of tip unclear.

Ctenopelmatinae? Förster, 1869

Perilissini? Thomson, 1883

Lathrolestes? Förster, 1869

Lathrolestes? zlatorog Viertler, Spasojevic & Klopfstein, sp. nov. (Fig 11)

urn:lsid:zoobank.org:act:508DC4A4-0549-483C-8C43-FF92DDA53DBE

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 11. Holotype of Lathrolestes? zlatorog sp. nov. (FUR #10651).

(A) Photo of the fossil part. (B) Interpretative drawing, fossil part and counterpart were used as templates. Scale bar: (B) 5mm.

https://doi.org/10.1371/journal.pone.0275570.g011

Description.

Body 6.8 mm. Head and mesosoma dark brown. Mandibles, antenna, legs and metasoma bright orange. Wing venation dark brown.

Head. Mandible bidentate; lower tooth a little wider and longer than upper tooth; both tips appear evenly rounded. Malar space about 0.8x mandibular width. Face including clypeus flat in profile. Flagellum 8 mm, with at least 30 segments, evenly thick throughout whole length, central flagellar segments about 1.2x as long as wide. Temple seems short.

Mesosoma. Pronotum about 0.6x as long as high. Notauli either absent or very weak. Epicnemial carina laterally and ventrally present. Propodeum rounded in profile; about as long as high; lateromedian and lateral longitudinal carinae present; anterior and posterior transverse carinae indiscernible; hind margin appears simple. Hind leg with coxa of normal proportion; femur 3.2x as long as wide; tibia normal, evenly tapering, with one tibial spur visible but the second one potentially hidden; tarsal claw appears simple.

Wings. Fore wing 7.4 mm. Areolet closed, quadrate-oblique, with 4M about half the length of 2Rs. Upper half of 2m-cu straight, then bowed outwards, with one large bulla covering 45% of whole 2m-cu vein. 4Cu 1.5x of 5Cu. 4Rs almost straight. Ramulus absent. 1cu-a postfurcal to 1M + 1Rs, with 1Cu about as vein width of 1cu-a; 1cu-a emerging distally. Pterostigma 2.9x as long as wide. Cell 2R1 2.6x as long as wide. 5M tubular throughout whole length. 2Cu 0.7x 1M + 1Rs and 0.9x r-rs. 1m-cu&2-Rs+M arched. 3Cu almost double the length of 2cu-a. Hind wing only partially visible, with 1Rs 1.5x rs-m.

Metasoma. Depressed until T4, rest unclear. T1 about 1.25x longer than wide; evenly tapering to front; spiracle in anterior half; with trace of dorso-lateral carina. T2–T4 about 1.2x wider than long.

Metopiinae Förster, 1869

Triclistus Förster,1869

Triclistus bibori Viertler, Spasojevic & Klopfstein, sp. nov. (Fig 12)

urn:lsid:zoobank.org:act:544F7D34-E61A-4862-9247-40612C518DEE

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 12. Holotype of Triclistus bibori sp. nov. (FUR #13809).

(A) Photo of the fossil part. (B) Interpretative drawing, fossil part and counterpart were used as templates. (C) Photo and drawn outlines of the fore tarsus, with the tarsomeres 2–4 very short and stout. (D) Head with visible groove between antennae. Scale bars: (B) 4mm, (C) 1mm, (D) 0.5mm.

https://doi.org/10.1371/journal.pone.0275570.g012

Robust data quality revealed

Our data quality evaluations showed negligible differences when different people place landmarks, which confirms previous investigations of this aspect in other studies [54, 55]. We could also demonstrate that the differences between forewing drawings from the literature and photographs of flattened wings were minor, at least if the former were obtained in a standardised fashion similar to the one used by Townes [36–38, 46]. This also suggests that photographs of different species could easily be added to our dataset in the future. This is at least the case if the dataset is used for classification purposes above the genus level, as some of our replicates using different congeneric species resulted in a partial overlap of intra- versus inter-generic distances (Fig 2A). This was especially the case in some large genera and might partly be due to non-monophyly of the involved genera; indeed, for instance the species included in Bathyplectes used to be classified in two different genera in the past [17]. In any case, while the limits of our approach for classifying taxa at or below genus level requires further scrutiny, it appears very robust for higher-level classification [10, 11].

Fossilisation artefacts appear to only have a minor effect on the performance of the method, at least if the wings appear reasonably flat, which is the case in a large portion of compression fossils. In our analysis, the fore wings of part and counterpart, as well as the left and right side of a compression fossil, which often showed a slightly different quality of preservation, were very similar. This indicates that it does not really matter which wing is used in the geometric morphometric analysis, except if it is obviously distorted or damaged. This insight is especially crucial for palaeontology, as preservation of compression fossils can vary widely [32].

Wing size and shape correlation

Fore wing shape is strongly correlated with centroid size in our dataset (Fig 3), with the aspect ratio (length to width) increasing with size. This allometric relationship means that smaller species tend to have relatively wider, and larger species relatively more narrow wings, which might ensure a minimum surface area of the wing conferring sufficient uplift even in small species. Our finding is in agreement with previous studies that found that fore wing length is a good proxy for specimen size [56, 57], while fore wing width did not seem to be strongly correlated to size, at least in Apoidea [58]. However, the relationship of size and aspect ratio varies among insect groups, which might be due to differences in other aerodynamic properties of the wings [59].

The wing cells of Darwin wasps tend to change in parallel with the entire wing, becoming more narrow in larger species. This effect is strongest for the central and apical cells (Fig 3). This leads to a rather even distribution of wing veins over the entire length and width of the wing in larger specimens, which might stabilise the wing in larger species. A similar pattern, including the increasing pterostigma width in smaller specimens, has been observed in other Hymenoptera as well [59, 60], although considerable variation has been observed in other insect orders [59].

We did not observe any reduction of wing veins in small-bodied species in our dataset, as it is observed in many other groups of Hymenoptera [61]. Indeed, the wing venation of Darwin wasps is surprisingly constant, with most veins present even in the smallest representatives of the family [21]. However, there are two exceptions, Hybrizontinae and Neorhacodinae, both rather small-bodied and species-poor subfamilies that were not included in our dataset. In both cases, the reduction mostly concerns the apical part of the wing, and the remaining veins furthermore shift towards the base [21].

Interestingly, the allometry effect varies among some of the subfamilies, despite some of them sharing the same size range, e.g. Cremastinae and Tryphoninae. Ophioninae stand out in this respect by exhibiting a rather large fore wing size range, but only minor allometric effects. Most Ophioninae species are nocturnal, and wing features of nocturnal Hymenoptera also seem exceptional in other lineages [15, 62], although the reason for this observation remains unknown. Irrespective of their body size, Ophioninae wings all appear rather similar to those found in previous analyses of larger hymenopterans, which possess a slender pterostigma and more elongate distal cells, whose functions appear related to flying performance [62, 63]. Overall, the varying correlations of fore wing size and venation shape observed among many subfamilies might be caused by more than one factor. Evolutionary history, geographical components, hosts, selective pressures in response to mimicry, and body shape variation all are factors that might influence shape and size differences [56, 64]. To disentangle the reasons for the different allometry patterns among subfamilies, one needs to be able to rule out environmental or genetic variation that affect size, or to know the extent of their influence in order to correct for it [65].

Variation within and between extant subfamilies

According to the Procrustes variance within the subfamilies, the most extensive shape variation is found in Metopiinae and Cremastinae, whereas the least variance was found in Tersilochinae and Campopleginae (Fig 4, S4 Table). Why fore wings of Cremastinae are the most variable among the included subfamilies is unclear, given that the subfamily is rather well-defined morphologically, with the exception of the Belesica group of genera, and has consistently been recovered as monophyletic in phylogenetic analyses [20, 66]. Their wings not only vary in cell dimensions, but also in areolet shape and pterostigma dimensions, although their short and straight 2m-cu is very consistent. Ecology as a possible explanation for this large fore wing variation seems unlikely, since they all do share similar habitats and hosts. The hosts might be concealed differently, so the available space for their pupation might differ, but this is highly speculative. In Metopiinae, the variation is introduced mostly by the genus Metopius, which is represented by seven species in our dataset and exhibits a much narrower fore wing and a clearly different areolet shape from the other Metopiinae genera (Fig 7). Metopius species are also distinct in other body features, like having a shield-like face, only one mesotibial spur and being generally larger than most other Metopiinae genera [67]. They also are exceptional among other members of the subfamily in attacking free-ranging Lepidoptera, instead of hosts that are hidden in silk webbing or leaf-rolls [21], so the difference in wing venation might be due to this environmental factor or the size difference.

Wing characteristics like the shape and length of the cell 2R1, the emerging position of 1cu-a, the shape of the areolet and many more have previously been used in identification keys [21, 68]. It is therefore surprising that this is the first geometric morphometric approach to distinguish ichneumonid subfamilies. While other geometric morphometric studies successfully discriminated taxa at lower taxonomic ranks [69–71], we here demonstrate that fore wing venation is sufficient to distinguish many extant ichneumonid subfamilies (Figs 4 and 5), with shape differences usually larger between subfamilies than within them (Fig 4, S4 Table). Many subfamilies can thus be reclassified successfully in the CVA, although some remain difficult to distinguish, especially Ctenopelmatinae and Tryphoninae (Table 3). The classification accuracy could probably be increased by adding more species to each subfamily, considering that there are more than 25’000 described ichneumonid species [17]. Subfamilies like Pimplinae and Ctenopelmatinae are probably not even monophyletic [20, 21, 72], although at least in Pimplinae, this did not have a noteworthy impact on the success of their classification; this might be linked to the clade’s wing venation being rather plesiomorphic in many aspects [30]. In Ctenopelmatinae, the classification success was often low, especially when compared with Tryphoninae (Table 4), and it might improve once the subfamily is split into monophyletic entities. The similarity between Ctenopelmatinae and Tryphoninae was already observed in many body characteristics before [21, 68], even though they are not very closely related within Ophioniformes; they are thus an excellent example that morphoclustering can differ from phylogeny because of plesiomorphic characters and divergent rates of evolution [73]. While Tryphoninae is probably the sister group of the rest of Ophioniformes [20], Ctenopelmatinae might just have maintained the plesiomorphic fore wing venation in parallel. Plesiomorphic trait retention, natural and sexual selection, genetic drift, and abiotic factors might change wing venation faster than visible in molecular data, as seen in fossil halictid bees [11]. Convergence could be another explanation, and it underlines the importance of not confusing morphoclustering and phylogenetics. Ctenopelmatinae and Tryphoninae share the same hosts, sawfly larvae [37, 68], which makes an ecological influence on morphology probable. But this would need further investigations, and the complexity of influencing factors needs to be kept in mind. Despite such effects, the high classification success observed in our study at the tribal and subfamily level suggests that there is a lot of phylogenetic information in wing venation waiting to be harvested even for deep phylogenetic studies in Darwin wasps.

Fossil classification aided by geometric morphometrics

Our study shows that in ichneumonids, geometric morphometrics of fore wings results in robust leads to possible affiliations of fossil species to extant subfamilies. Quantifying the shape similarity of fossil specimens with extant taxonomic groups by use of geometric morphometrics is a rather new application, but it has already proven successful in various taxa. For example, in a fossil agamid lizard (Amphibolurinae), the jaw was used to compare it with extant genera [74]. In bees and wasps, the fore wings have repeatedly been used to assign fossils to extant groups, for example in fossil bumble bees (Apidae) to tribes or subgenera [13, 75], halictid bees (Halictidae) to tribes [11], and in ensign wasps (Evaniidae) to genera [16]. Using geometric morphometrics to inform fossil classification is certainly a very promising approach and will gain further traction in the near future.

However, we also encountered some potential pitfalls of the method. For instance, in Triclistus bibori sp. nov., the classification analysis gave ambiguous results, which might be due to the fact that three landmarks were missing in the holotype. This highlights the necessity of the fore wings to be as complete as possible, since too much information could get lost with only a few missing landmarks. Alternatively, the problem might also be explained by the relatively small number of extant Metopiinae taxa in our analysis (N = 28); this would be difficult to rectify, as the subfamily does not include many more genera. Finally, our ambiguous classification result might also be an indication of an especially fast evolution of wing venation in Metopiinae, which is somewhat supported by the large difference between Metopius and the remaining genera in the subfamily. Indeed such a speed-up of evolution could be expected to result in a pronounced difference between fossil and extant taxa, but this remains to be shown and would need further investigation.

Another potential pitfall relates to the fact that we attempted to classify the new fossil taxa into existing groups, in our case extant subfamilies, or even further to extant tribes if possible. Such an approach might however be unjustified for really old fossils, especially if there was a high turnover of higher taxa through time. In such cases, the fossils might instead belong to a stem lineage and not a crown group, or even to an extinct clade that has not yet been discovered and named. Indeed, if the Euclidean distances in the between-group PCA were very large between the fossil specimen and the given taxonomic groups, it might be due to the fact that we did not include the closest relatives yet, which might be already extinct. For example, the distances between Ebriosa flava gen. et sp. nov. and many tribes of Ctenopelmatinae or Tryphoninae are larger than the distance between the tribes of those two subfamilies. The exceptional fore wing venation of Ebriosa flava sp. nov., with the strongly curved M&Rs vein, raises the possibility of it belonging to an extinct tribe, a stem-lineage of either Tryphoninae or Ctenopelmatinae, or even to a now extinct subfamily that branched off even earlier. Classifying the many remaining undescribed ichneumonid fossils from the Fur Formation and integrating fossils from other formations of similar age, such as the Menat Formation in France [76, 77] might help shed light on alternative groupings the fossil might belong to. Also, integrating the relevant wing vein characteristics into a proper phylogenetic analysis would help distinguish between a stem position of a fossil in the tree of extant taxa and its belonging to a now extinct lineage, a problem that cannot be easily solved by similarity-based methods.

Phylogeny-based methods in general would provide the means of making the most of the information inherent in insect wing venation, while avoiding some of the pitfalls of similarity-based methods. However, it remains somewhat unclear how to best extract relevant characters for phylogenetic analyses from geometric morphometric datasets [78, 79]. Methods to estimate the phylogenetic signal in GM data, on the other hand, are already rather well established [80], and they could be used to separate the prevalent noise inherent in morphometric data from the signal due to deep phylogenetic relationships. Such approaches could also be used to aid placing fossil specimens, at least if an adequate phylogeny of the group is at hand.