We examined all of the fossils described in the literature as bumble bees or as closely allied extinct genera (Table 1), corresponding to 15 specimens representing 14 described species. For all species, we tried to locate the type material to check against the original description and to better illustrate the fossil, if needed. We contacted the potential repositories of the fossils and were able to locate 13 specimens for review (Fig. 1). Information about museum repositories is included in the “Results” section. Overall, we gathered pictures and/or drawings of the forewings of 13 specimens representing 12 of 14 species (Table 1).

The morphological terminology follows that of Engel (2001) and Michener (2007), while the higher classification (i.e., subfamily, tribe) follows that of Michener (2007) (i.e., seven families: Andrenidae, Apidae, Colletidae, Halictidae, Megachilidae, Melittidae, and Stenotritidae). For bumble bees, we used the subgeneric system of Williams et al. (2008) where 15 subgenera were proposed. A complete list of extant species with their nomenclature is available at the following link (updated from Williams 1998): (http://www.nhm.ac.uk/research-curation/research/projects/bombus/groups.html).

Fossils of bumble bees have been described from eleven deposits from the Late Eocene to the Upper Miocene: Brembridge Marls, Florissant, BesKonak, Latah, Bílina Mine, Krottensee, Randeck Maar, Shandong, Botchi River, Euboea, and La Cerdanya (Table 1).

The Insect Bed of the Bembridge Marls from the Late Eocene (i.e., 36.0 Ma) is located on the Isle of Wight (UK). Two bee fossils were recorded from the deposit: the presumed bombine Oligobombus cuspidatus Antropov, 2014 and specimen NHMUK In.10012 (Megachilidae, incertae sedis) (Antropov et al. 2014).

The Florissant shale of Colorado (USA) (Swisher and Prothero 1990), situated near the Eocene-Oligocene boundary is approximately 34.0 Ma in age (Epis and Chapin 1974; Evanoff et al. 2001; Boyle et al. 2008; Mustoe 2008; Veatch and Meyer 2008). It produced a large number of the known bee fossils, most of which were described in the early part of the 20^th Century, with 36 specimens representing 34 species in 19 genera (Zeuner and Manning 1976; Michez et al. 2012). One possible bumble bee fossil was recorded: Calyptapis florissantensis Cockerell, 1906.

The deposits of BesKonak are from the Lower Miocene (Aquitanian, i.e., 22.5 Ma) and located in Anatolia, north of Ankara Province, Turkey (Paichelier et al. 1978). The only bombine fossils discovered in the deposits of BesKonak are Oligoapis beskonakensis Nel & Petrulevičius, 2003 and Paraelectrobombus patriciae Nel & Petrulevičius, 2003.

The Latah Formation encompasses the Lower to Middle Miocene (i.e., 21.3–12.1 Ma) of eastern Washington and northwestern Idaho (USA) (Berry 1929; Kirkham and Melville 1929; Gray and Kittleman 1967; Lewis 1969; Robinson 1991; Derkey et al. 2003). The only known bee fossils from this deposit are Bombus proavus Cockerell, 1931 and an undetermined megachiline specimen (Cockerell 1931; Engel 2004).

The deposits of the Most Formation at Bílina Mine date from the Lower Miocene (i.e., 20.0 Ma), in northern Bohemia (Czech Republic) (Kvaček 1998; Prokop and Nel 2000; Prokop et al. 2003; Kvaček et al. 2004; Knor et al. 2012). Two bee fossils have been reported from the deposits of the Most Formation: undetermined specimens of Apis and the bumble bee B. trophonius Prokop, Dehon, Michez & Engel, 2017 (Prokop et al. 2003; Prokop et al. 2017; Engel pers. obs.).

Krottensee, also in the Czech Republic, dates from the Lower Miocene (i.e., 18.0–17.0 Ma), and is also referred to as Mokřina (Bůžek et al. 1996; Mlíkovský 1996; Rojík 2004). A single bumble bee has been recovered from the deposits, B. crassipes Novák, 1878 (Novák 1878; Krzemiński and Prokop 2011).

The Randeck Maar deposits of the Lower-Middle Miocene (i.e.,18.0–16.0 Ma) are located in southwestern Germany, southeast of Stuttgart at the escarpment of the Swabian Alps (Heizmann 1983), and is the largest ancient Maar in that region (Köppen and Geiger 1928; Gregor 1986; Krautter and Schweigert 1991; Schweigert 1998; Lutz et al. 2000; Kottek et al. 2006). This fossil Lagerstätte contains exceptionally well-preserved flora and fauna (e.g., Armbruster 1938; Gregor 1986; Schawaller 1986; Ansorge and Kohring 1995; Kotthoff 2005; Kotthoff and Schmid 2005; Kotthoff et al. 2011). Several prominent bee fossils have been reported from Randeck Maar – Apis armbrusteri Zeuner, 1931 (Kotthoff et al. 2011), B. randeckensis Wappler & Engel, 2012 (Wappler et al. 2012), and Halictus schemppi (Armbuster, 1938) – and while those of Bombus and Halictus are each from single specimens, a plethora of honey bee workers have been recorded (Kotthoff et al. 2011).

The Middle Miocene sediments of the Shanwang Formation (17.0–15.2 Ma) are located in Linqu County, Shandong Province, China (Yang et al. 2007). Many insects have been listed from this deposit, including bees (Megachilidae, Apidae), and specifically the bumble bees B. anacolus Zhang, B. luianus Zhang, and B. dilectus Zhang (Zhang 1990; Zhang et al. 1994).

The Botchi Formation is from the Upper Miocene (i.e., 11.2–7.1 Ma) and is located on the left bank of the Botchi River in Russia (Khabarovsk Region) (Akhmetjev 1973). This formation has yielded various plants, fishes, Crustacea, and insects, including B. vetustus Rasnitsyn & Michener, 1991.

The deposit of Kumi (Euboea, Greece) is from the Middle-Upper Miocene (i.e., 11.2–7.1 Ma). Insects from the orders Coleoptera, Diptera, and Hymenoptera were discovered in the Kumi deposit, and these included B. pristinus Unger, 1867.

The Spanish deposit of La Cerdanya corresponds to Upper Miocene lacustrine beds (i.e., 10.0 Ma) located in Spain (Lleida, Bellver-en-Cerdaña) (Diéguez et al. 1996; Jiménez-Moreno et al. 2010). The flora and entomofauna are quite abundant and diverse (Peñalver-Molla et al. 1999; Arillo 2001) with a rather high occurrence of bees, although nearly all specimens belong to Apis (Nel et al. 1999). Bombus cerdanyensis Dehon, De Meulemeester & Engel, 2014 was described from this deposit (Dehon et al. 2014).

We performed geometric morphometric analyses of the forewing shape in order to assess the taxonomic affinities of 12 bumble bee fossil species (13 specimens) showing well-preserved forewings (Fig. 1). This tool is useful in insect taxonomy for discriminating and diagnosing taxa at different levels (e.g., Pretorius 2005; Petit et al. 2006; Francoy et al. 2009, 2012; Sadeghi et al. 2009; Perrard et al. 2014; Van Cann et al. 2015), as well as in paleontology for assessing taxonomic affinities of fossils with contemporary and extinct taxa (e.g., Kennedy et al. 2009; Michez et al. 2009a; De Meulemeester et al. 2012; Wappler et al. 2012; Dehon et al. 2014, 2017; Dewulf et al. 2014; Perrard et al. 2016; Prokop et al. 2017). Several studies have demonstrated the utility of forewing shape analyses for diagnosing subgenera, species, and populations of bumble bees, depending on rearing conditions (e.g., Aytekin et al. 2007; Wappler et al. 2012; Barkan and Aytekin 2013; Gérard et al. 2018).

We used three different datasets to assess the taxonomic affinities of the fossils at different taxonomic levels. All three datasets represent a sampling of contemporary and extinct tribes with three submarginal cells, were largely assembled and analyzed in previous studies (i.e., Dehon et al. (2017) for the first dataset and Prokop et al. (2017) for the second and third datasets). We only (i) modified the classification of the species in the different datasets based on Bossert et al. (2019) and (ii) added the 13 fossils in each of these datasets. The first dataset included a comprehensive sampling of bee tribes in order to ensure correct tribal placement of the 13 fossils. This dataset consisted of 50 tribes representing 226 species and 979 specimens (refer to Dehon et al. (2017) for full details; Suppl material 1: Table S1). It also uncovered a group of six tribes (i.e., Ancylaini, Electrapini, Emphorini, Euglossini, Melikertini, and Tetrapediini) showing similar wing shapes to Bombini. We then used a second dataset with more extensive sampling within Bombini and these six similar tribes. This second dataset was assembled and tested by Prokop et al. (2017). It includes 841 specimens and represents all 15 subgenera and 210 species of extant bumble bees (80% of the total species diversity) as well as tribes Ancylaini, Electrapini, Emphorini, Euglossini, Melikertini, and Tetrapediini, representing a further 18 genera, 43 species, and 132 specimens altogether (Suppl material 2: Table S2). Finally, fossils confirmed to belong to Bombini using the first and second datasets were compared to a third dataset that only consists of the bumble bee specimens of the second dataset in order to assess the taxonomic affinities of the specimens with extant subgenera of Bombus. Dehon et al. (2017) (Suppl materials 3–5: Tables S3–S5) and Prokop et al. (2017) (Suppl materials 6–7: Tables S6–S7) demonstrated reliability for these datasets in classifying bee specimens based on forewing shape similarity relative to the reference datasets of forewings. Hence, the cross-validation allows us to be confident in the discrimination.

The potential effect of sexual dimorphism on subgeneric assignment using wing morphometry was tested by Wappler et al. (2012) for the subgenus Bombus s. str. For this subgenus, the results showed that sexual dimorphism had limited impact on subgeneric assignment. We tested it on four additional subgenera (based on 82 specimens from 12 species of four subgenera: Bombias, Cullumanobombus, Melanobombus, and Mendacibombus); the identification of the subgenera based on wing shape was again highly supported (Suppl material 7: Table S7). Therefore, to limit intraspecific variability in our dataset, we sampled female specimens only. We selected females because Bombini are mostly social species and workers (i.e., females) are the most abundant caste. Moreover, most of the known fossil specimens are females, although the holotype of B. vetustus is a male as evidenced by the lack of a corbicula, male flagellomeres, etc. (Rasnitsyn and Michener 1991).

Left forewings were photographed using an Olympus SZH10 microscope combined with a Nikon D200 camera. Photographs were then uploaded in the software tpsUTIL 1.69 (Rohlf 2013a). The forewing shape was captured by digitizing two-dimensional Cartesian coordinates of 18 landmarks on the wing veins and cells (Fig. 4) with the software tpsDIG version 2.27 (Rohlf 2013b). Position of the landmarks was based on Owen (2012) and other studies like De Meulemeester et al. (2012), Wappler et al. (2012), Dewulf et al. (2014), Dehon et al. (2014, 2017), Gérard et al. (2015), and Prokop et al. (2017). The two-dimensional configurations of the landmarks were superimposed using the GLS Procrustes superimposition in the software R version 3.0.2 (Rohlf and Slice 1990; Bookstein 1991; Adams and Otárola-Castillo 2013; R Development Core Team 2013). The closeness of the tangent space to the curved shape space was analyzed by calculating the least-squares regression slope and the correlation coefficient between the Procrustes distances (in the shape space) and the Euclidean distances (in the tangent space) (Rohlf 1999). This was calculated using the software tpsSMALL v1.25 (Rohlf 2013c).

Variation of shape in the dataset was explored with PCA analyses to visualize clustering and detect outliers (Fig. 5). Discrimination of the wing shape of the different taxa was assessed by Linear Discriminant Analyses (LDA) of the projected aligned configuration of landmarks like in Prokop et al. (2017). We performed three LDAs with the first dataset (Dehon et al. 2017, Suppl material 1: Table S1) with different levels a priori grouping (i.e., the groups are known a priori by the analysis): family, subfamily and tribe (LDA 1–3, Tables 1, Suppl materials 3–5). We did a fourth LDA analysis with the second dataset (i.e., bumble bees + six similar tribes, Suppl material 2: Table S2) with tribe level as a priori grouping (LDA 4, Table 1, Suppl material 6). Finally, we used a comprehensive sampling of extant bumble bees (i.e., third dataset, Suppl material 2: Table S2) for a fifth LDA considering the subgenus level as a priori grouping (LDA 5, Tables 1, Suppl material 7: Table S7). The LDA effectiveness was assessed by the percentages of individuals correctly classified to their original taxon (hit-ratio, HR) in a leave-one-out (LOO) cross-validation procedure based on the posterior probabilities of assignment (Suppl materials 3–7: Tables S3–S7). Given the observed scores of an “unknown”, the posterior probability (pp) equals the probability of the unit to belong to one group compared to all others. The unit is consequently assigned to the group for which the posterior probability is the highest (Huberty and Olejnik 2006). All discriminant analyses were performed using the R software (R Development Core Team 2013).

Taxonomic affinities of the fossils were assessed based on the score in the predictive discriminant space of shapes. Aligned coordinates of the specimens from the three datasets (including the fossils) were used to calculate the same five LDA as presented in the previous section. Assignment of the fossils was estimated by calculating the Mahalanobis Distance between each fossil and group mean of each taxon and then assigning it to the nearest group in the discriminant shape of the LDA (Suppl materials 7–12: Tables S7–S12). Principal Component Analyses (PCA) were also computed to visualize shape affinities between the fossils and the extant groups in the second dataset (Fig. 5). Mahalanobis Distance is well suited for dealing with large datasets of close-relative taxa (Claude 2008).

The assignment of each fossil was assessed in each dataset. When using the first dataset, all fossils were assigned to Apidae, more specifically to Apinae (except for the second specimen of C. florissantensis described by Cockerell (1908) and B. vetustus, both assigned to Eucerinae) and to Bombini (except for B. dilectus assigned to Tetrapediini, and Oligobombus cuspidatus and both specimens of C. florissantensis, all three assigned to Electrapini) (see LDA 1–3, Tables 1, Suppl materials 9–11). We then specifically assessed the assignment of each fossils based on the second and the third dataset. Oligobombus cuspidatus and both specimens of C. florissantensis were close to the shape space of contemporary Bombini, while being placed outside of contemporary Bombini and fossil Electrapini (Dataset 2, LDA 4). Among bombine subgenera, Bombias was most similar in forewing shape to Oligobombus and Calyptapis (Dataset 3, LDA 5). Based on our discriminant analyses, Paraelectrobombus patriciae was also similar to extant Bombini while being just outside of its shape space (Dataset 2, LDA 4). Bombus beskonakensis clustered within the shape space of extant Bombini (Dataset 2, LDA 4) and was similar to the subgenus Mendacibombus but outside the modern shape space of that subgenus (Dataset 3, LDA 5). The assignment of B. trophonius based on wing shape was already assessed in Prokop et al. (2017), this fossil clustered within contemporary Cullumanobombus (Dataset 3, LDA 5). The forewing shape of B. randeckensis was previously analyzed by Wappler et al. (2012) who proposed that the specimen was close to the subgenus Bombus s. str. Our analyses found a close similarity with Cullumanobombus (Dataset 3, LDA 5), while in Wappler et al. (2012) this subgenus was the fourth most similar subgenus to the fossil. This discrepancy may be explained by the fact that we used a larger and more diverse dataset, or possibly also because forewings were digitized by different experimenters in both studies. Bombus luianus clustered within Bombini based on its forewing shape (Dataset 2, LDA 4). Moreover, its forewing shape was similar to, but outside of modern Melanobombus, suggesting this fossil might be sister to extant Melanobombus (Dataset 3, LDA 5). Bombus dilectus did not cluster within the shape space of Bombini, but the tribe was the most similar (Dataset 2, LDA 4), but it is possible the published drawings are not entirely accurate. The most similar subgenus was the subgenus Bombias (Dataset 2, LDA 4), one of the most basal subgenera of the genus (Fig. 5). Bombus anacolus clustered just outside of crown-group Bombini while being quite similar to this tribe based on its forewing shape (Dataset 2, LDA 4), but again the published drawing is rather poor. Nonetheless, its forewing shape was similar to modern Mendacibombus (Dataset 3, LDA 5). Bombus vetustus, which is a male, was most similar to the tribe Bombini based on forewing shape but was placed outside of the shape space of modern Bombini (Dataset 2, LDA 4). Despite this, the most similar subgenus was Bombias (Dataset 3, LDA 5). There is only one specimen available of B. pristinus, which has incomplete wings. However, all landmarks are available except for number 16, whose position could be accurately estimated by the extension of cu-a and portion of vein A. We decided therefore to apply the same LDA analyses to the specimen, with the 18 landmarks. Our results found that this fossil clustered inside the shape space of Bombini (Dataset 2, LDA 4) and was similar to the subgenus Cullumanobombus (Dataset 3, LDA 5), although it would be worth reanalyzing this specimen with a dataset encompassing males from extant species in order to have greater confidence. Finally, the wing shape of B. cerdanyensis was first analyzed in Dehon et al. (2014). This specimen has incomplete wings; nonetheless, all landmarks are available except for numbers 17 and 18, but the position of the latter could be accurately estimated by the extension of cu-a and portion of vein A. Assignment of the fossil in the discriminant space did not allow a reliable subgeneric attribution. Herein, B. cerdanyensis is assigned to Melanobombus (Dataset 3, LDA 5). The subgeneric assignment of each fossil within Bombus s. l. through geometric morphometric analyses is summarized in Table 1.

In the following account of fossil bombine species, we have organized the taxa by general age, proceeding from oldest to youngest.

Subfamily Apinae Latreille

Clade Corbiculata Engel

Stem-group Bombini Latreille

Late Eocene

Genus Oligobombus Antropov, 2014

Type species

Oligobombus cuspidatus Antropov, 2014, by original designation.

Diagnosis

Sex unknown. Forewing distinctly pointed apically (apparently taphonomically altered); three submarginal cells of approximately equal sizes; marginal cell elongate, longer than distance between its apex and forewing tip, with apex roundly truncate; forewing distal membrane papillate; pterostigma short, with margin within marginal cell straight, approximately 4.0 times as long as prestigma; r-rs arising from distal part of pterostigma after its midpoint; 1rs-m straight; 2rs-m with posterior half curved apically; angle between 1rs-m and part of M inside third submarginal cell obtuse; first submarginal cell with an oblique translucent vein rs and not wider than second submarginal cell; second submarginal cell shorter than third marginal cell; third submarginal cell widest; 1m-cu slightly curved anteriorly, reaching second submarginal cell in its midpoint; 2m-cu curved anteriorly, reaching M basad 2rs-m; distance between anterior ends of 1m-cu and 2m-cu exceeding their length; basal vein slightly basad cu-a. See Antropov et al. (2014) for original diagnosis.

Oligobombus cuspidatus Antropov, 2014

Holotype

Sex unknown. NHMUK In.17349 (part and counterpart), Smith collection of the Natural History Museum (NHM, London, UK). Type specimen has been located and revised (Figs 1A, 3A).

Figure 1. 

Representative fossil bumble bees A Oligobombus cuspidatus (photograph by Antropov et al. (2014)) B Holotype of Calyptapis florissantensis (photograph by Manuel Dehon) C C. florissantensis (photograph by Talia S. Karim) D Bombus (Paraelectrobombus) patriciae (photograph by Gaëlle Doitteau) E B. (Mendacibombus) beskonakensis (photograph by Gaëlle Doitteau).

Type strata and locality

Late Eocene (i.e., 36.0 Ma), Insect Bed of the Bembridge Marls from the Isle of Wight, UK.

Diagnosis

Owing to monotypy, the diagnosis for the species is identical to that of the genus (vide supra).

Description

Part consists in middle and apical parts of right forewing; counterpart consists of middle part of right forewing; forewing distal membrane papillate; complete venation preserved; total forewing length 13.3 mm, maximum width 4.0 mm as preserved; basal vein length 2.3 mm, relatively straight and basad cu-a; cu-a length 0.3 mm; marginal cell length 4.0 mm, width 0.9 mm, apex roundly truncate; prestigma 0.2 mm; pterostigma length 0.8 mm; 1^st abscissa of Rs straight; 2^nd abscissa of Rs almost straight; 3Rs length approximately same as r-rs; 4Rs slightly longer than 3Rs; M+Rs length 1.2 mm; three submarginal cells; first submarginal cell length 1.5 mm (as measured from origin of Rs+M to juncture of r-rs and Rs), width 0.6 mm (as measured from Rs+M to pterostigma); second submarginal cell length 1.3 mm (as measured from juncture of Rs+M and M to juncture of Rs and 1rs-m), width 0.7 mm (as measured from midpoint on M between 1m-cu and 1rs-m to juncture of r-rs and Rs); third submarginal cell length 1.4 mm (as measured from juncture of 1rs-m and M to juncture of M and 2rs-m), width 1.0 mm (as measured from juncture of M and 2m-cu to juncture of 2rs-m and Rs); 1rs-m straight; 2rs-m posterior half curved apically; 1m-cu anterior half curved apically, reaching M approximately at midpoint between 2^nd abscissa of Rs and 1rs-m; 2m-cu basad 2rs-m. See Antropov et al. (2014) for original description.

Comments

There is only one specimen, the holotype NHMUK In.17349, consisting of a part and counterpart. Antropov et al. (2014) described the specimen and considered it as possibly a member of Bombini. According to the original author, the forewing shape displays mixed features of Bombini, Electrapini, Electrobombini, Euglossini, and Melikertini (e.g., the forewing distal membrane being papillate is characteristic of Bombini, Electrobombini, and Euglossini, the shape of vein Rs displays mixed features reminiscent of the corbiculate tribes Bombini, Electrobombini, Electrapini (i.e., Thaumastobombus Engel, 2001), Euglossini, and Melikertini (i.e., Melikertes Engel, 1998 and Succinapis Engel, 2001), the submarginal cells are reminiscent of Electrapini, Electrobombini, and Euglossini, 1m-cu is reminiscent of Electrapini and Electrobombini, 2m-cu is reminiscent of Electrapini, Euglossini, and Melikertini). All in all, the specimen has a forewing venation with features that can be found in different extinct and extant tribes of Corbiculata, but that taken together do not occur in any of them. According to the Antropov et al. (2014), the fossil forewing venation is generally similar to extant species of Bombini, but the lack of features from the pro-, meso-, and metasoma prevents identification of its exact taxonomic affinities. Based on the general morphology and forewing shape affinities, Oligobombus is perhaps a stem-group bombine and we consider it as such for the moment. Further material and additional characters, ideally analyzed in a cladistic framework, are needed to corroborate this placement, or the species could have phylogenetic affinities with Electrobombini or Electrapini.

Lower Miocene

Subgenus Cullumanobombus Vogt, 1911

Bombus (Cullumanobombus) trophonius Prokop, Dehon, Michez & Engel, 2017

Holotype

Female. ZD0003 (coll. Bílina mine). Type specimen has been located and revised (Figs 2A, 3F).

Type strata and locality

Lower Miocene (i.e., 20.0 Ma), Clayey Superseam Horizon, Bílina mine, Czech Republic.

Diagnosis

The fossil has a wing pattern most similar to B. (Cullumanobombus) rufocinctus Cresson (Milliron 1973; Williams et al. 2014). Moreover, both species display a similar combination of 3Rs about as long as r-rs but shorter than 4Rs, a basal vein basad 1cu-a, a vein 2Rs arched posteriorly but not as greatly prolonged proximally as in several other species of Cullumanobombus (e.g., Milliron 1971), and a vein 1m-cu entering second submarginal cell near midpoint. However, the convex pterostigmal border within the marginal cell, less apically narrowed marginal cell, and less arched 2rs-m minimally serve to distinguish the fossil species from B. rufocinctus. See Prokop et al. (2003) and Prokop et al. (2017) for original diagnosis.

Description

Wings and integument black as preserved; forewing total length 14.6 mm; maximum width 5.10 mm; basal vein weakly arched basally, comparatively straight along length, basad cu-a by about vein width, in line with 1Rs; M+Rs originating anteriad, 1Rs slightly shorter than r-rs; pterostigma short, slightly longer than wide, tapering inside of marginal cell, border inside marginal cell convex, prestigma nearly as long as pterostigma; marginal cell length 5.1 mm, width 1.1 mm, free portion slightly shorter than portion bordering submarginal cells, apex rounded and offset from anterior wing margin by much more than vein width, not appendiculate; 2Rs strongly arched basally and slightly arched outward; r-rs about as long as 3Rs; 4Rs slightly longer than 3Rs; three submarginal cells of approximately same sizes, albeit third slightly larger than first or second; first submarginal cell length 0.9 mm (as measured from origin of M+Rs to juncture of r-rs and Rs), width 1.0 mm (as measured from Rs+M to pterostigma); second submarginal cell length 1.3 mm (as measured from juncture of Rs+M and M to juncture of Rs and 1rs-m), width 0.9 mm (as measured from midpoint on M between 1m-cu and 1rs-m to juncture of r-rs and Rs); third submarginal cell length 1.6 mm (as measured from juncture of 1rs-m and M to juncture of M and 2rs-m), width 1.2 mm (as measured from juncture of M and 2m-cu to juncture of 2rs-m and Rs); 1rs-m straight; 2rs-m arched distally in posterior half; 1m-cu distinctly angulate anteriorly near M, entering second submarginal cell near cell’s midlength; 2m-cu slightly arched apically, meeting third submarginal cell at cell’s apical fifth of length. Hind wing length 9.4 mm, width 2.6 mm. Preserved portion of mesosoma and legs difficult to describe, although portion of metatibial corbicula preserved (basal quarter to third), and sclerites with numerous, long setae. See Prokop et al. (2003) and Prokop et al. (2017) for original description.

Figure 2. 

Representative fossil bumble bees A Bombus (Cullumanobombus) trophonius (photograph by Jakup Prokop) B B. (Cullumanobombus) randeckensis (photograph by Torsten Wappler) C B. vetustus (photograph by Alexandr P. Rasnitsyn) D B. (Cullumanobombus) pristinus (photograph by Irene Zorn and Monika Brüggeman-Ledolter) E B. (Melanobombus) cerdanyensis (photograph by Thibaut De Meulemeester).

Figure 3. 

Forewing drawings of the fossil bumble bees studied herein. Some forewings were mirrored to enable comparison across all specimens A Oligobombus cuspidatus (mirrored) B Holotype of Calyptapis florissantensis (mirrored) C C. florissantensis D Bombus (Paraelectrobombus) patriciae (mirrored) E B. (Mendacibombus) beskonakensis F B. (Cullumanobombus) trophonius (mirrored) G B. (Cullumanobombus) randeckensis (mirrored) H B. vetustus I B. (Cullumanobombus) pristinus (mirrored) J B. (Melanobombus) cerdanyensis.

Comments

The specimen was first reported as Bombus sp. in Prokop et al. (2003). Prokop et al. (2017) demonstrated that the fossil clustered within contemporary Cullumanobombus and formally described the species. Although the majority of contemporary species of Cullumanobombus are found in the New World and a few species in the Old World, Hines (2008) estimated that the subgenus originated around 20.0–15.0 Ma in the Palearctic. Our result, as well as that of Prokop et al. (2017), is consistent with Hines (2008) as the fossil specimen was found in the Lower Miocene (i.e., 20.0 Ma) deposits of Bílina Mine in northern Bohemia (Czech Republic).

Bombus (Cullumanobombus) randeckensis Wappler and Engel in Wappler et al. (2012)

Holotype

Sex unknown. The fossil consists of an isolated forewing. SMNS 68000/28 (old Armbruster collection No. A5119). Conserved in the Staatliches Museum für Naturkunde, Stuttgart, Germany. Type specimen has been located and revised (Figs 2B, 3G).

Type strata and locality

Randeck Maar, southeast of Stuttgart, Swabian Alb; Early Miocene, i.e., 16.0–18.0 Ma (Burdigalian, Karpatian, MN 5).

Diagnosis

Bombiform bee; infuscate area in marginal cell extends entire length of anterior half of marginal cell; forewing venation strictly similar to that of an extant bumble bee, with transector visible on both forewings. See Wappler et al. (2012) for original diagnosis.

Description

Forewing length 14.3 mm, maximum width 5.0 mm; marginal cell length 3.9 mm; basal vein almost straight, slightly curved in its base, slightly basad cu-a; vein cu-a straight; three submarginal cells; first submarginal cell length 1.7 mm (as measured from origin of M+Rs to juncture of r-rs and Rs), width 0.8 mm (as measured from M+Rs to pterostigma); second submarginal cell width 0.7 mm (as measured from midpoint on M between 1m-cu and 1rs-m to juncture of r-rs and Rs); third submarginal cell length 1.3 mm (as measured from juncture of 1rs-m and M to juncture of M and 2rs-m), width 1.1 mm (as measured from juncture of M and 2m-cu to juncture of 2rs-m and Rs); height of second medial cell 1.1 mm (as measured from Cu1 to juncture of 1m-cu and M); 1^st abscissa of Rs almost straight; 2^nd abscissa of Rs with anterior half curved apically; r-rs almost straight; M+Rs straight and longer than r-rs; 3Rs almost as long as r-rs; 4Rs slight smaller than M+Rs; 1rs-m almost straight; 2rs-m with posterior half curved apically; 1m-cu curved apically in last anterior part, reaching second submarginal cell before midpoint; 2m-cu slightly curved, reaching M basad to 2rs-m. See Wappler et al. (2012) for original description.

Comments

The fossil was discovered in the Lower Miocene (i.e., 18.0–16.0 Ma) deposits of Randeck Maar, Germany, an age and locality in general accord with the estimate that Cullumanobombus originated between 20.0–15.0 Ma in the Old World. Based on the forewing shape affinities and the general morphological assessment, B. randeckensis is likely an extinct species of Cullumanobombus, like B. trophonius.

As shown in Michez et al. (2009b), De Meulemeester et al. (2012), Wappler et al. (2012), Dehon et al. (2014, 2017), Dewulf et al. (2014), and Prokop et al. (2017), geometric morphometric analyses of forewing shape provide a robust tool for assessing the taxonomic affinities of bee fossils with contemporary taxa and insights into bee evolution. We additionally demonstrate herein that the Hit Ratios for subgeneric level assessments were high for the genus Bombus. However, we need to combine geometric morphometrics of forewing shape with morphological features (e.g., pilosity, leg morphology, head and mouthpart characters, etc.) to get more powerful results, but such morphological characters are either limited or lacking with the current impression fossils.

Figure 4. 

Left forewing of Bombus (Bombus) terrestris (Linnaeus, 1758) with the 18 landmark points indicated on the veins to describe the shape (photograph by Michaël Terzo). The names of the veins and cells can be found in Dehon et al. (2017).

When using the first dataset with tribe a priori grouping, the most similar tribe to C. florissantensis (i.e., both specimens) and O. cuspidatus is Electrapini, while Tetrapediini is the second most similar tribe to C. florissantensis and Tetrapediini is the second most similar tribe to O. cuspidatus (LDA 3; Suppl material 10: Table S10). When using the second dataset, the most similar tribe to the holotype of C. florissantensis is Bombini, while the most similar tribe to the specimen described by Cockerell (1908) is Electrapini (the second most similar tribe being Bombini). The most similar tribe to O. cuspidatus is Bombini when using the second dataset (LDA 4; Suppl material 12: Table S12). Those results might be explained by the fact that when using the first dataset, only 20 specimens (i.e., four species, with five specimens per species) were chosen to represent Bombini, compared to 841 specimens (i.e., 210 species) in the second dataset and thereby more fully encompasses the breadth of morphospace represented among modern bombines. Furthermore, the species used to represent Bombini in the first dataset do not represent early-branching subgenera, which were found to be the most similar subgenera to C. florissantensis and O. cuspidatus in the third dataset (i.e., Bombias) (LDA 5; Suppl material 13: Table S13). The similarity of O. cuspidatus with Electrapini when using the first dataset is concordant with Antropov et al. (2014), who considered it as a possible member of Bombini with mixed features of Bombini and other corbiculate tribes (i.e., the extinct Electrapini, Electrobombini, Melikertini, and the extant Euglossini). This fossil may represent an extinct stem-group to Bombini and it would explain the different results obtained with the different datasets; those similarities across Bombini and the other tribes representing symplesiomorphic features encompassing the clade. Based on morphological features (i.e., presence of a corbicula, the forewing venation similar to Bombus s. l.) and forewing shape similarities, C. florissantensis could also belong to a stem-group bombine (Cockerell 1906, 1908; Zeuner and Manning 1976). Based on the available evidence, the conservative position is to consider both species as possible stem-group Bombini. It would be highly desirable to verify this hypothesis using cladistic analyses of new morphological characters in the future.

Our results generally support the timing of divergence of extant species proposed by Williams (1985) and Hines (2008) (Fig. 5), noting that the meager record available for Bombini means such corroboration is minimal at best, albeit non-contradictory. The record of fossil bumble bees is sufficiently scant that at its best we can conclude that the available record does not contradict prior estimates, and falls in line with those for the subgenera Cullumanobombus, Melanobombus, and Mendacibombus. Unfortunately, fossils of most lineages within Bombus and certainly from more numerous and refined slices of time are simple lacking, meaning that the current record of fossil bumble bees lacks resolution for determining the timing of most diversification events (e.g., most fossils are clustered within a few deposits representing widely disparate slices in the Oligocene and Miocene). Nonetheless, none of the specimens from Eocene and Oligocene deposits were assigned within the shape space of any contemporary subgenus of Bombus, which is not surprising when looking at more completely preserved bees from, for example, the Eocene amber deposits (Baltic, Cambay, Rovno) (e.g., Engel 2001). On the other hand, most specimens coming from Miocene deposits were assigned within the contemporary shape space of Bombus s. l., and for some of them within contemporary subgenera (i.e., Cullumanobombus, Melanobombus, and Mendacibombus), again a pattern consistent with more completely preserved bees from other Miocene deposits (e.g., Dominican, Mexican amber) (e.g., Engel et al. 2012). This pattern mirrors the hypothesis that there were significant changes in the bee fauna between Eocene and Oligocene epochs and again at the Paleogene-Neogene transition (e.g., Engel 2001, 2004, 2019a, b). Contemporary species of Melanobombus and Mendacibombus are restricted to the Old World. On the other hand, most species of Cullumanobombus (excluding B. cullumanus, B. semenoviellus, B. unicus) occur in the New World (Williams 1998; Williams et al. 2014). Cullumanobombus and Melanobombus are estimated to have originated between 20.0–15.0 Ma, while the basal subgenus Mendacibombus has been estimated to have originated around 34.0–30.0 Ma (Hines 2008). All fossils having affinities to extant subgenera have an age that is posterior to the stem age of subgenera (Hines 2008), and therefore our assignments are coincident with the estimated origin and divergence times of Bombus s. l. and extant subgenera (Hines 2008).

Figure 5. 

Ordination of the fossils along the three first axes of the PCA (PC1 = 44.29%, PC2 = 11.37%, PCA3 = 10.05%) in subgeneric dataset of Bombus s. l.

Other fossil specimens that were assigned to extant subgenera of Bombus s. l. are in accordance with the estimated stem age of those groups. In occurrence, our analyses are concordant with the ages of Cullumanobombus and Melanobombus (i.e., between 20.00–15.00 Ma), as well as Mendacibombus (i.e., between 34.0–30.0 Ma) (Hines 2008). These species highlight that Bombini had diversified significantly by the Miocene and that these limited fossil data are concordant with dating estimates. Continued paleontological exploration will only further refine our understanding based on direct evidence for dating bumble bee evolution.

Corbiculata are the most represented bees in the fossil record, especially in terms of number of specimens found in amber deposits, with workers of certain stingless bees (Meliponini) numbering into the tens of thousands of individuals (Michez et al. 2012; Engel and Michener 2013b; Engel pers. obs.). Corbiculata appeared in the Late Cretaceous based on the occurrence of a crown-group meliponine in Maastrichtian-aged Raritan amber (Michener and Grimaldi 1988; Engel 2000). This indicates that the divergence events among the tribes, at least among their stem groups, extend back to at least the latest Cretaceous. The three extinct tribes of corbiculate bees Electrapini, Electrobombini, and Melikertini are known from the Eocene (Baltic amber, Cambay amber, various impression fossil deposits such as Messel and Eckfeld), and some of these were assuredly advanced eusocial like Apini and Meliponini based on the presence of morphologically specialized workers (Engel 2001). Dehon et al. (2014) described a new corbiculate species, Euglossopteryx biesmeijeri De Meulemeester, Michez & Engel, 2014 discovered in the Parachute Creek Member of the Green River Formation (Utah, USA), and had phenetic similarities in wing shape to Euglossini, but it remains to be determined whether this was symplesiomorphic similarity or indicative of a cladistic relationship.

During the Paleocene-Eocene (the Paleocene-Eocene Thermal Maximum and Eocene Thermal Optimum), the concentration of greenhouse gases and the mean global temperature was higher than at present, with poles with little to no ice (Zachos et al. 2001; Royer 2006). The Early Eocene was marked by the EECO (Early Eocene Climatic Optimum) 51–53 million years ago, with a high pCO₂ and the global temperature reaching a long-term maximum. This was likely caused in part by differences in volcanic emissions, particularly high during parts of the Paleocene-Eocene periods (i.e., 40.0–60.0 Ma) (Walker et al. 1981). The PETM (Paleocene Eocene Thermal Maximum, i.e., 55.0 Ma) is the most prominent and best-studied hyperthermal episode, during which the global temperature increased by more than 5°C in less than 10,000 years (Zachos et al. 2001, 2008). A global cooling that most certainly caused the large-scale extinction of many plant and animal species marked the Eocene-Oligocene transition (i.e., 34.0 Ma) (Zachos et al. 2008; Hansen et al. 2013). Although still speculative at this time, it could be hypothesized that the latter event was related to the extinction of the group of Bombini to which Calyptapis florissantensis and Oligobombus cuspidatus might have belonged. Similarly, Dehon et al. (2014) suggested that E. biesmeijeri could also be consistent with the hypothesis that global climates, particularly cooling and drying events, were somehow related to the loss of corbiculate diversity (Engel 2001, 2002; Engel et al. 2013a), and that this was perhaps a global phenomenon impacting similar bee lineages in the New World (Dehon et al. 2014). It is noticeable that extant bumble bees appear especially sensitive to hyperthermal crises (Kerr et al. 2015; Rasmont et al. 2015). These climatic events resulted in significantly floral turnovers and these florist changes certainly could have influenced ancient lineages of bees (Collinson 1992). Those events of successive changes in temperature might have played a role in the appearance or extinction of species studied herein.

Figure 6. 

Hypothesis of bumble bee evolution according to the branching dates of Hines (2008) alone with the subgeneric system of Williams et al. (2008). Fossils are mapped onto the clade according to our hypotheses based on our wing morphometry/shape results. Geometric morphometric analyses should be considered as a heuristic tool given the absence of other forms of pertinent data (e.g., absence of information on mandibular form, pretarsal structure, genitalic characters, etc.). A = Alpinobombus. B = Bombus s.str. LF = clade with mostly long-faced species. P = Pyrobombus. SF = clade with mostly short-faced species.