Given that Melitta willardi possesses three submarginal cells we sampled specimens from different extant subfamilies with the same arrangement of cells. All available subfamilies were included, with a maximum of 20 specimens per subfamily, and with a maximum of five specimens per species. We additionally sampled all species of Melitta available with a maximum of five specimens per species. The dataset also included eighteen extinct species [Apidae, Apinae: Anthophorula persephone Engel, 2012; Bombus randeckensis Wappler & Engel, 2012; Electrapis meliponoïdes (Buttel-Reepen, 1906); Electrapis krishnorum Engel, 2001; Electrobombus samlandensis Engel, 2001; Eufriesea melissiflora (Poinar, 1998); Melikertes stilbonotus (Engel, 2001); Melissites trigona Engel, 2001; Paleohabropoda oudardi Michez & Rasmont, 2009; Protobombus basilaris Engel, 2001; Probombus hirsutus Piton, 1940; Succinapis goeleti Engel, 2001; and Thaumastobombus andreniformis Engel, 2001; Halictidae, Halictinae: Cyrtapis anomala Cockerell, 1908; Electrolictus antiquus Engel, 2001; Halictus petrefactus Engel & Peñalver, 2006; Ocymoromelitta sorella Engel, 2002 and Ocymoromelitta florissantella (Cockerell, 1906)]. Since the holotype of Melitta willardi is a female, all individuals used in the morphometric analysis are females to avoid potential bias due to any sexual dimorphism. The assembled dataset comprised 360 specimens representing six families of Anthophila, 15 subfamilies, and 109 species (Table 1).

Taxonomic affinities of the fossil were evaluated based on wing shape. Wing venation is used widely in insect taxonomy and can provide many informative features for phylogenetic analyses and for many Late Paleozoic taxa is sometimes the only form of available data (e.g. Gumiel et al. 2003, Pretorius 2005). Moreover, use of the wings has significant advantages compared to other organs, i.e., they are relatively rigid, articulated, 2D structures that present a large number of useful landmarks formed by the homologous intersections of veins. Geometric morphometrics is a procedure which aims at quantifying and analyzing the overall shape of a structure (Bookstein 1991, Rohlf and Marcus 1993, Adams et al. 2004), and can provide a powerful tool in paleontology for discriminating taxa at different levels as well as for discussing taxonomic affinities between extinct and extant taxa (Roberts 2008, Michez et al. 2009, De Meulemeester et al. 2012, Wappler et al. 2012). The holotype of Melitta willardi does not exhibit any signs of post-mortem tectonic deformation, meaning that the venation observed is reflective of as it was in life and did not require any compensation to adjust for taphonomic or diagenetic alteration.

The right forewings of 360 female specimens were initially photographed using an Olympus SZ010 binocular coupled with a Nikon D70 camera. Photographs were gathered in one TPS file using tps-UTIL 1.56 (Rohlf 2013a). To capture the shape, two dimensional Cartesian coordinates of 18 landmarks (Figure 2) were digitized by tps-DIG 2.17 (Rohlf 2013b). Both right and symmetrized-left wings of Melitta willardi were digitized by four experimenters (AD, MD, TD, DM) to obtain an objective and robust identification. All landmark configurations were scaled, translated, and rotated against the consensus configuration by the generalized least square Procrustes superimposition method (Bookstein 1991). The superimposition was performed using R functions of the package “geomorph” (Adams and Otárola-Castillo 2013). The aligned landmark configurations were projected into the Euclidean space tangent to the curved Kendall’s shape space to aid further statistical analyses. The closeness of the tangent space to the curved shape space was tested by calculating the least-squares regression slope and the correlation coefficient between the Procrustes distances in the shape space with the Euclidean distances in the tangent space (Rohlf 1999). This variation amplitude of our dataset was calculated with tps-SMALL 1.25 (Rohlf 2013c).

Prior to the assignment of the fossil, shape variation within the reference dataset and discrimination of the different taxa was assessed by Linear Discriminant Analyses (LDA) of the projected aligned configurations of landmarks, with subfamily levels as a priori grouping by using the software R version 3.0.2 (2013, http://www.R-project.org/). The effectiveness of the LDA for discriminating subfamilies was assessed by the percentages of individuals correctly classified to their original taxon (hit-ratio, HR) in a leave-one-out cross-validation procedure based on the posterior probabilities of assignment. 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).

Taxonomic affinities of the fossil were assessed based on their score in the predictive discriminant space of shapes. After superimposition of the 368 landmark configurations (i.e. corresponding to the reference dataset and the fossil), aligned coordinates of the 360 specimens from the reference dataset were used to calculate the LDA. We included a posteriori the eight aligned landmark configurations of Melitta willardi in the computed LDA space as “unknown” specimens and calculated their score. Assignments of the fossil configurations were estimated by calculating the Mahalanobis Distance (MD) between “unknowns” and group mean of each subfamily. We also calculated posterior probabilities of assignment to confirm the assignment to one taxon.

In order to assess the taxonomic affinities of Melitta willardi with the family Andrenidae, PCA was computed to visualize shape affinities between the fossil and andrenid subfamilies.

The regression coefficient between the Procrustes distances and the Euclidean distances is close to 1 (0.9999). This means that the linear tangent space closely approximates the shape space, thereby permitting us to be confident in the variation amplitude of our dataset.

In LDA space with subfamily a priori grouping, discrimination of the 15 groups are effective, with a cross-validated HR of 98.61% (e.g., 5 misclassified specimens), and 10 of the 15 subfamilies that account for a HR of 100% (Table 2). Other subfamilies have a HR between 90% and 99%. Due to sampling size within groups, the HR drastically drop down when a single specimen is misclassified. This is the case for the five groups with HR lower than 100%. Cross-validation assignment (Table 2) allows us to be confident in the group discrimination at subfamily level.

All of the 109 specimens of Melitta were correctly classified to their original taxon (Melittinae) in the leave-one-out cross-validation procedure. However, the eight landmark configurations of Melitta willardi are assigned to Andreninae (MDs = 2.72 – 4.84; PPs = 0.9999 – 1). Taxonomic affinities of the fossil were also assessed based on non-supervised analyses within Andrenidae. In the morphometric space defined by the PCA, the fossil is undoubtedly clustered with the subfamily Andreninae (Figure 3).

The wings of Andrenopteryx gen. n. have three submarginal cells, suggesting that the genus does not probably belong to subfamilies such as Xeromelissinae, Hylaeinae, Euryglossinae (all Colletidae), Dasypodainae (Melittidae), Megachilinae (Megachilidae), or various tribes among the Apidae (i.e., Allodapini, Ammobatini, Ammobatoidini, Biastini, Boreallodapini, Caenoprosopidini, Ctenoplectrini, Neolarrini, and Townsendiellini). Furthermore, Andrenopteryx gen. n. clearly possesses pollen-collecting structures, suggesting that the fossil was probably not cleptoparasitic and accordingly those genera may also be excluded (cleptoparasitic genera occur in various families, see Michener 2007). The GM analysis of the wing shape of Andrenopteryx gen. n. suggests that this fossil belongs to the Andrenidae (see previous results, vide supra). Nevertheless, diagnostic features of Andrenidae such as the two subantennal sulci and the short to long pointed glossa are not preserved in the only available specimen of the species.

Assuming that its clustering among Andrenidae is an accurate reflection of its relationships, among andrenids the three submarginals cells excludes placement among most of the Panurginae. The species has a long marginal cell with an acutely curved apex that lies along the costal margin as in Andreninae, while the other subfamilies have a marginal cell with a truncate apex (Michener 2007). The holotype clearly possesses a scopa which is limited to the metafemur and metabasitarsus and without a flocculus, unlike the diverse extant genus Andrena Fabricius. The metabasitarsus is more than half as long as the metatibia, in stark contrast to the form present in Megandrena Cockerell. Lastly, Andrenopteryx gen. n. does not have an enlarged inner tooth on the pretarsal claws, and therefore is distinct from the southern South American genus Orphana Vachal. Thus, while Cockerell’s species certainly is best placed in the Andreninae it seems generically distinct and this has served as the basis for our decision to describe a new genus.

Wing shape analyses were successfully employed in previous studies to discriminate extant bee taxa at various classificatory levels, from subspecies to tribes (e.g., Kandemir et al. 2011, De Meulemeester et al. 2012). In addition, these analyses are sufficient to confidently associate bee fossils with extant groups (e.g., Michez et al. 2009, De Meulemeester et al. 2012, Wappler et al. 2012), and this lends increased confidence to the affinities of Andrenopteryx willardi as outlined above.

Cockerell (1909) mentioned some features that for him indicated that his fossil species was referable to Melitta. He noted the three submarginal cells, the particular form of the pterostigma, the scopa confined to the metafemur and metabasitarsus, and the absence of a flocculus. However, a majority of these similarities are unfounded and not indicative of Melitta, and in fact some are more suggestive of Andreninae. First, the proportions of the submarginal cells are more similar to Andrenidae than any melittid. Second, the second submarginal cell does not receive the 1m-cu well before its midpoint, and this is true for both the left and right forewings. Thus, even based on the evidence available to Cockerell and from his description the fossil should not be placed within Melitta, and it is peculiar to us why he made such a taxonomic decision.

Based on the discovery that Cockerell’s fossil Melitta is more likely an andrenine, some previous hypotheses regarding the biogeography of North American bees require reconsideration. Michez and Eardley (2007) speculated the presence of Melitta in North America during the Oligocene based on Cockerell’s (1909) assertion of the taxonomic identity of Andrenopteryx willardi, and Dellicour et al. (2014) demonstrated that North American species of Melitta form a derived clade within the genus. There is now no evidence for Michez and Eardley’s scenario. The origin of Melitta could be more recent than previously hypothesized and Dellicour et al.’s North American clade could have entered and diversified on the continent during the Neogene. In contrast, the record of Andreninae in North America during the Oligocene is now corroborated by the present fossil. There are additional records of putative andrenines from Florissant, such as Lithandrena saxorum Cockerell, 1906, Pelandrena reducta Cockerell 1909, and five additional species Cockerell placed in Andrena (Michez et al. 2012), but these are in need of re-evaluation. It is hoped that these species may also be subjected to morphometric analyses and their relationships clarified.

The bees of the Florissant shale have been ignored for a long time (Engel 2002). It was Cockerell’s intention to document the whole fossil fauna and flora from Colorado and this partly drove his efforts to document the known bee remains from these deposits (Cockerell 1927, Engel 2002). Cockerell, who largely relied on a hand lens to study specimens, often based his hypotheses regarding the placement of particular fossils on the their general habitus, or relied on a suite of traits recognized nowadays as not indicative of those same families, subfamilies, and even genera. For example, many of the traits concerning wing shape such as the relative positions of the rs-m or m-cu crossveins are quite variable within individual families. Subsequent to Cockerell, Zeuner and Manning (1976) tried to evaluate the Florissant fossil bees, but they based their work solely from Cockerell’s original descriptions and did not examine type material. Zeuner and Manning’s monograph is further compromised by the fact that both authors died before the work was completed, leaving behind only notes that were subsequently cobbled together to form the publication, and this explains its poor quality and limited utility (Engel 2002). The only other works were brief accounts by Engel (2001b, 2002), who attempted to re-evaluate Cockerell’s Florissant halictines as well as newly discovered material, and to describe a new large carpenter bee. Outside of this, the Florissant fossil bee fauna has remained dormant and given that it is one of the most diverse and specimen-rich deposits for the Anthophila, it is all the more imperative that its species be properly evaluated in a modern context. We hope that this brief treatment of one such species will inspire more investigations into the fossil bees from Colorado.