All the new fossil taxa of horse flies described here are from the Lower Cretaceous Crato Formation of Brazil. The geology and paleontology of the Crato Formation, an important Gondwanan Konservat-Lagerstätten, was revised in detail by Martill et al. (2007) and Ribeiro et al. (2021). The Crato Formation is one of the stratigraphic units that constitute the Santana Group of the Araripe Basin. From the base to the top, the Santana Group consists of the Barbalha, Crato, Ipubi, and Romualdo formations. The age of the Crato Formation is considered to be upper Aptian (Lower Cretaceous) (Heimhofer and Hochuli 2010).

According to Ribeiro et al. (2021), the fossil-rich interval of the formation – the “Crato Konservat-Lagerstätte” or CKL –, consisted of a semi-arid seasonal lacustrine wetland, in which a shallow water body, including diverse aquatic fauna and flora, was succeeded up-landward by neighboring mesophytic ecotones, periodically flooded, besides outer xeric habitats. Trophic structure analysis and details of the putative food web that took place within the Crato Ecosystem are provided by Mendes et al. (2020).

The terminology used in the morphological study follows the interpretation of Cumming and Wood (2017), and Chainey (2017). The studied specimens are housed in the collections of the Instituto de Geociências da Universidade São Paulo (IGC-USP, São Paulo, Brazil) and the Centro de Ciências Naturais e Humanas, Universidade Federal do ABC (CCNH-UFABC, Santo André, Brazil).

All specimens were photographed using a Motic K700-L stereo microscope with a canon EoS Rebel T2i digital camera attached, and photos were edited using GIMP 2.10 software. Drawings were made in Inkscape 1.1. Measures and scales were made with Carl Zeiss AxioVision (Release 4.8) software.

The ingroup sampling included 20 extant horse flies from three of the traditionally recognized subfamilies (Pangoniinae, Chrysopsinae, Tabaninae). Scepsidinae was not included since it has only two very dissimilar species, and it is not recognized by some authors (e.g. Coscarón and Papavero 2009b). All species described for Cratotabanus Martins-Neto and Santos and Araripusgen. nov. were included as terminals in our analysis. Tabanipriscus transitivus Grimaldi was also included since the species has diagnostic traits of Tabanidae and also traits of the Tabanomorpha groundplan such as a two-segmented cercus and the absence of a postspiracular scale (Grimaldi 2016), which are useful to understand the early relationships of Tabanidae.

Our morphological matrix is based on previous literature (Mackerras 1954; Stuckenberg 1973; Yeates 2002; Kerr 2010; Grimaldi et al. 2011; Grimaldi 2016) and new observations. The outgroup is composed of two extant species of Rhagionidae – Rhagio mystaceus (Maquart) and Atherimorpha lamasi Santos – and three species of Athericidae – Suragina pacaraima Rafael and Henriques, Atherix ibis (Fabricius) and Dasyomma caeruleum Macquart. It includes 35 morphological characters and 31 terminal taxa, with 25 extant and 6 fossil taxa. The chosen outgroups and the codified characters are relevant to the understanding of family and subfamily phylogenetic positioning. The matrix was analyzed through parsimony and Bayesian Inference. Then, the final topologies for each of the optimality criteria were compared.

For the parsimony analysis, trees were rooted between the Rhagionidae and the Athericidae + Tabanidae, on the earliest divergence of the outgroups following Nixon and Carpenter (1993). The analysis was performed in TNT (version 1.5; Goloboff and Catalano 2016) using both equal weights and implied weighting (Goloboff 1993, 2014; Goloboff et al. 2008). For the implied weighting analyses, we used k values of 5, 15 and 150. All analyses were performed using heuristic searches with TBR branch swapping, random stepwise addition sequence, and 5000 replicates holding up to 10 trees per replication, keeping the maximum possible number of trees in memory (99999 trees).

Bayesian analyses were performed on MRBAYES 3.2.7 (Ronquist et al. 2012) under the MK model (Lewis 2001) with a lognormal distribution for modeling character rate variation. The analysis was carried out with two simultaneous runs, each with four chains. Each run contained 10 million generations of the mcmc chain, with trees sampled every 1000 generations and a burning cutoff set at 25 percent. Convergence was checked by the standard deviation of the sample spits and by examining trace plots using TRACER 1.7 (Rambaut et al. 2018).

A complete list of the taxa and the data matrix used in the phylogenetic analyses are provided in the Supplementary material.

1. Ocellar triangle: (0) absent; (1) present.

2. Second palpomere lateral sulcus: (0) absent; (1) present.

3. Proboscis length: (0) longer than head height; (1) shorter than head height.

4. Labella: (0) fleshy; (1) slender.

5. Eyes dorsal margin: (0) straight; (1) curved.

6. Ocelli: (0) vestigial; (1) developed.

7. Basal callus: (0) absent; (1) present.

8. Postpedicel: (0) undifferentiated from other flagellomeres; (1) forming a basal plate derived from the fusion of proximal flagellomeres; (2) rounded, with terminal flagellomeres modified in an arista-like stylus.

9. Scape: (0) clearly longer than wide; (1) approximately as long as wide.

10. Clypeus: (0) rounded anteriorly; (1) flat anteriorly; (2) conical.

11. Metathoracic scale: (0) absent; (1) present.

12. Hind tibial spurs: (0) absent; (1) present.

13. Costal vein: (0) with normal width at base; (1) ticked at base.

14. Vein Sc: (0) bare; (1) setulose.

15. Insertion of Vein R₂₊₃: (0) close to R₁; (1) distant from R₁.

16. Shape of R₂₊₃: (0) straight most of its length, curved only at the insertion in C; (1) with one or more sinuosities on its length; (2) with a single strong concavity at the distal half.

17. Cell r₄: (0) not encompassing the wing apex; (1) encompassing the wing apex.

18. Origin of R₄: (0) anterior to the tip of M₃; (1) at the same level as the tip of M₃; (2) posterior to the tip of M₃.

19. Angle between R₄ and R₅: (0) lesser than 90 degrees; (1) 90 degrees.

20. Lower calypter: (0) reduced; (1) pronounced.

21. Basicosta: (0) bare; (1) setulose.

22. Suprametacoxal connection: (0) absent; (1) present.

23. Type of suprametacoxal connection: (0) reduced; (1) pronounced.

24. Metacoxal pit: (0) absent; (1) present.

25. Cerci, female terminalia: (0) one-segmented; (1) two-segmented.

26. Tergite IX, female terminalia: (0) divided into two separated triangular plates; (1) one single transverse bar.

27. Tergite X, female terminalia: (0) divided; (1) undivided.

28. Mushroom shaped expansion in the spermatheca: (0) absent; (1) present.

29. Gonostyli posterior extremity: (0) bipartite; (1) slender; (2) truncated; (3) rounded.

30. Tergites IX+X, male terminalia: (0) fusioned; (1) separated.

31. Sagittal division of fusioned tergites IX+X, male terminalia: (0) absent; (1) present.

32. Epandrium articulation: (0) free; (1) articulated on gonocoxites (Sinclair et al. 1994: character 12).

33. Gonocoxal apodeme length: (1) short at the most reaching anterior margin of hypandrium; (2) extending well beyond anterior margin of hypandrium (Sinclair et al. 1994; Kerr 2010).

34. Endophalic tines: (0) absent; (1) present.

35. Endoaedeagal process: (0) absent or reduced; (1) present.

The parsimony analysis using equal weights resulted in 2.630 most parsimonious trees with 56 steps, summarized in a strict consensus (Fig. 7A). Athericidae and Tabanidae were both recovered as monophyletic. The subfamilies Tabaninae and Chrysopsinae form a monophyletic group, which is congruent with molecular hypotheses (Morita et al. 2016). Tabanipriscus transitivus is sister group to all the remaining Tabanidae, while Araripus is sister group to all remaining horse flies, including Cratotabanus, which is located in a polytomy with a paraphyletic Pangoniinae. Parsimony analysis with implied weighting resulted in the same topology obtained with equal weights for all k values analyzed (5, 15, and 150).

In the Bayesian analysis (BI), chains reached convergence, with the standard deviation of the sampled splits observed to be 0.0038 and the trace plots reaching stationary distribution. Fig. 7B shows the topology and posterior probabilities (pp). The BI yielded results very similar to parsimony, however, Cratotabanus is in a polytomy at the base of a monophyletic crown group Tabanidae.

The resulting topologies of both optimality criteria employed are similar. For discussion of the evolution of traits, we plotted morphological characters in one of the most parsimonious trees obtained under equal weights. This tree was chosen as it summarizes results obtained under both Parsimony and Bayesian analyses. Clades recovered by different optimality criteria were indicated in this topology (Fig. 8).

Both the monophyly of Tabanidae and its sister group relationship with Athericidae are well established in the literature (e.g. Yeates 2002; Wiegmann et al. 2011). Under both of the optimality criteria here employed, the sister group relationship between the two families is well supported. In the most parsimonious tree chosen as representative of our hypothesis, the clade Athericidae + Tabanidae is supported by four unequivocal synapomorphies: the fusion of tergites IX and X in the male [ch. 30(0)], a well-known characteristic of both families (Stuckenberg 1973; Woodley 1989); the presence of a suprametacoxal connection [ch. 24(1)]; the epandrium articulated freely with the gonocoxites [ch. 32(1)]; and gonocoxal apodemes extended beyond the hypandrium [ch. 33(1)].

In our analysis, Tabanidae is monophyletic, and the subfamilies Tabaninae and Chrysopsinae form a clade within Tabanidae. Chrysopsinae, however, appears as paraphyletic (Fig. 7). The monophyly of Chrysopsinae + Tabaninae is sustained by the presence of a conspicuous basal callus [ch. 7(1)]; the postpedicel fused with proximal flagellomeres in a basal plate [ch. 8(1)]; tergite IX of female divided in two triangular plates [ch. 26(0)]; and the sagittal division of fused tergites IX+X in the male terminalia [ch. 31(1)].

The Pangoniinae are recovered in a polytomy at the base of the Tabanidae crown-group (Fig. 8). In Morita et al. (2016), the Pangoniinae are monophyletic and sister group to the remaining subfamilies, with the Chrysopsinae paraphyletic relative to Tabaninae. The polytomy in the base of Tabanidae obtained in our results is possibly due to the reduced sample of each subfamily and the missing data introduced by fossil taxa. That said, it is not the objective of this paper to evaluate the internal relationship of Tabanidae, but, instead, to construct a framework to understand the evolution and phylogenetic positioning of fossil taxa using morphological characters. The fossil groups included in this analysis and their phylogenetic relationships are discussed in the sections below.

The genus Tabanipriscus was described by Grimaldi (2016) based on a specimen from the Burmese amber. Despite not belonging to the Brazilian Cretaceous, we decided to include this genus in our analysis because it has similar traits to some Rhagionidae, Athericidae, and Tabanidae. According to Grimaldi (2016), Tabanipriscus could be a Tabanidae stem-group or sister group to Athericidae + Tabanidae. Our phylogenetic analysis supports the first hypothesis. In both optimality criteria employed, Tabanipriscus is sister to the clade Araripus + Cratotabanus + crown-group Tabanidae (Fig. 7). In the cladogram of Fig. 8, Tabanipriscus shares with the other Tabanidae the proximal flagellomeres not fused into a postpedicel [ch. 8(0)], which is considered the plesiomorphic condition within Brachycera (Stuckenberg 1999). In Tabanidae, this condition is kept in most of the Pangoniinae genera, with different levels of fusion occurring in several groups, notably in the Tabaninae and Chrysopsinae. Other characters that support Tabanipriscus + Tabanidae include a vein C thickened at its basal portion [ch. 13(1)]; and R₂₊₃ inserted in the costal vein far from R₁ [ch. 15(1)]. Both characters are recognized as diagnostic for Tabanidae (e.g. Stuckenberg 1973; Yeates 2002; Chainey 2017), with no known exceptions among living taxa. Tabanipriscus also lacks a metacoxal pit [ch. 22(1)] (Grimaldi 2016), considered diagnostic for Athericidae but absent from the athericid Dasyomma Macquart and all of the Tabanidae.

The new genus described here, Araripusgen. nov., is sister to Cratotabanus + crown Tabanidae in both optimality criteria employed (Fig. 7). The monophyly of the clade Araripusgen. nov. + Tabanidae is sustained by the cell r₄ encompassing the wing apex [ch. 17(1)]; and the split of R₄ and R₅ posterior to the base of M₃ [ch. 18(2)] (Fig. 6B). Both characters are present in all of the extant Tabanidae, while the monophyly of the clade Cratotabanus + crown Tabanidae is sustained by the angle of 90 degrees at R₄ [ch. 19(1)], a character with known variation within the family. A well-defined ocellar triangle may not be distinguished in Araripus crassitibialisgen. nov et sp. nov., and there is a dark structure near the vertex that could be a frontal callosity, but its position at the frons does not match the correspondent structures in extant horse flies. Araripusgen. nov. also has an inflated tibia, a characteristic that appears several times in both Tabaninae and Chrysopsinae subfamilies (Coscarón and Papavero 2009a; Chainey et al. 2017) but unknown in the Pangoniinae or in any other fossil previously described. Given its multiple appearances in species of different genera within Tabanidae, we did not include this character in our matrix.

The genus Cratotabanus has currently four described species, two of them – C. stonemyomorphus and C. cearensissp.nov. – from Brazilian Crato formation. The general form of the body and wing shape leaves little room for doubt about the identity of Cratotabanus as a Tabanidae (Bechly 2007; Grimaldi 2016). Nevertheless, some of the described species have characters considered plesiomorphies of Tabanomorpha, which raises interesting questions about the phylogenetic position of Cratotabanus. In our parsimony analysis (Fig. 7A), the genus is recovered within a polytomy at the base of Tabanidae with species of the Pangoniinae subfamily, while the Bayesian Inference supports the species currently assigned to the genus in a polytomy basal to a monophyletic crown-group Tabanidae (Figs 7B and 8). The lack of preservation of diagnostic characters poses a challenge for understanding the evolution of Cratotabanus, since most characters that could be apomorphic for the genus are preserved in no more than two or three species. Still, our results confirm the relationship of these species with Tabanidae.

Three species of Cratotabanus (C. asiaticus, C. stonemyomorphus, and C. cearensissp. nov.) lack a basal callus [ch. 7(0)], a character present in only a few Pangoniinae but conspicuous in the Tabaninae and Chrysopsinae (e.g. Coscarón and Papavero 2009a). Cratotabanus newjerseyensis, the only species with preserved antennae, has seven flagellomeres, with a postpedicel undifferentiated from the remaining flagellomeres (Grimaldi et al. 2011). This character state is considered plesiomorphic for Tabanomorpha and present in most species of Pangoniinae with different levels of fusion in different groups (e.g. Veprius Rondani and Zophina Philip). The proximal segments of the flagellum form a basal plate of four to five fused flagellomeres in the Chrysopsinae and Tabaninae [ch. 8(1)]. The basicosta of C. asiaticus is bare (Grimaldi 2016: fig. 15A), a character state plesiomorphic in Tabanomorpha and in Tabanidae, while a basicosta with setae as dense as those in the costal vein is an apomorphy of the Tabanini tribe [ch. 21(1)].

Cratotabanus retains three important plesiomorphic characters of Tabanormopha – the two-segmented cercus [ch. 25(1)]; an undivided tergite X [ch. 27(1)]; and the absence of a metathoracic postspiracular scale [ch. 11(0)]. The undivided tergite X in the female is seen in at least two species (C. asiaticus and C. cearensissp. nov.) [ch. 27(1)]. In most Tabanidae, the tergite X in the female is divided, with the exception of the uncommon Pangoniinae genera Archeomyotes Philip and Coscarón, Austromyans Philip and Coscarón, Fairchildimyia Philip and Coscarón (Philip and Coscarón 1971) and some species of Philoliche Wiedemann (Stuckenberg 1973). Although not included in our analysis, these species are morphologically different among themselves and from Cratotabanus, indicating that a unique origin of the fusion of the tergite X in these groups is unlikely.

As Tabanipriscus and Araripusgen. nov., the cerci of Cratotabanus are two-segmented. A one-segmented cercus is synapomorphy of Athericidae + Tabanidae while the plesiomorphic two-segmented condition is present in most families of Tabanomorpha. As with the undivided tergite X, our results suggest that a cercus with two segments is plesiomorphic in Tabanidae and that the cercus with one segment appears independently in Athericidae and the crown Tabanidae.

The postspiracular region is only visible in C. asiaticus (Grimaldi 2016). An absent or very reduced postspiracular scale (Grimaldi 2016) could be an apomorphy for the extinct genus, but further evidence is necessary to confirm this hypothesis.