Episodic radiations in the fly tree of life

The unexpected yet robustly supported sister group to Drosophilidae (including Drosophila) consists of two enigmatic and highly derived families of insect parasites, Braulidae and Cryptochetidae, within the superfamily Ephydroidea (Fig. 1). In simple terms, the closest relatives of Drosophila, the organism that has contributed the most to our understanding of modern genetics, are, rather than being “typical,” particularly unusual. Braulidae (bee lice), unrecognizable as flies to the common eye, are tiny, flattened, wingless creatures that live within beehives and attach themselves to adult honey bees. Cryptochetidae are endoparasites of scale insects as larvae and have been used successfully as biocontrol agents (22). The family Drosophilidae itself displays a diversity of life histories, including, like Braulidae, inquilines in honey bee hives and, like Cryptochetidae, parasites of Hemiptera (6). Before this study, there was little agreement on the position of Braulidae and Cryptochetidae within Schizophora, and the undecipherable evolutionary relationships among numerous families of small acalyptrate flies made the identification of a sister group to Drosophilidae challenging. Our dense sampling of family-level lineages provides an opportunity to refine evolutionary comparisons between drosophilids and these newly identified sister groups.

The relative placement of the lower cyclorrhaphan families Phoridae (scuttle flies) and Syrphidae (bee mimics known as hover flies) has important implications for understanding evolutionary development within Diptera (23), yet some morphological and molecular analyses yield conflicting phylogenies regarding these two families (24, 25). Although Episyrphus balteatus (Syrphidae) is a cyclorrhaphan, it exhibits features of development that are similar to those of other insects; however, unlike other cyclorrhaphans, Episyrphus possesses an anterodorsal serosa anlage [middorsal in Drosophila melanogaster, Musca domestica, and Megaselia abdita (Phoridae)], which expresses hunchback (absent in Drosophila, Musca, and Megaselia) and exhibits a strong influence of caudal on the anteroposterior axis (23), all features observed in noncyclorrhaphan insects. If syrphids are closer relatives to higher flies than phorids, as our molecular tree denotes, it indicates either parallel evolution in Megaselia and schizophorans or a reversal to an ancestral mode of development in Episyrphus. Furthermore, a paraphyletic relationship of phorids and syrphids would support the hypothesis that their shared special mode of extraembryonic development (dorsal amnion closure) (26) evolved in the stem lineage of Cyclorrhapha and preceded the origin of the schizophoran amnioserosa.

To test this hypothesis, we used a relatively recent phylogenomic marker: small, noncoding, regulatory micro-RNAs (miRNAs). miRNAs exhibit a striking phylogenetic pattern of conservation across the metazoan tree of life, suggesting the accumulation and maintenance of miRNA families throughout organismal evolution (27). Phylogenetic conservation is also evident in presence/absence comparisons of the expressed miRNA complement in the sequenced genomes of closely related Drosophila species (28). We constructed small RNA libraries for two tier 1 taxa, Episyrphus balteatus (Syrphidae) and Megaselia abdita (Phoridae), and used 454 pyrosequencing. We analyzed the resulting reads with miRMiner (29) to identify known miRNAs and to compare them with known complements of the insects Tribolium (Coleoptera) and Bombyx (Lepidoptera) and the dipterans Aedes aegypti, Anopheles gambiae, and Culex quinquefasciatus (Culicidae) as well as several Drosophila species. The phylogenetic distribution of our newly identified fly miRNAs is concordant with expected fly relationships from our multigene datasets, including the monophyly of Cyclorrhapha and the placement of Syrphidae as a closer relative to Schizophora than Phoridae based on the possession of two unique miRNA families (miR-956 and miR-971) and the possession of a paralogue (miR-318) of the miR-3/309 family and a paralogue (miR-304) of the miR-216/283 family (Fig. 2). Our data support the finding that either Megaselia and schizophorans have undergone parallel evolution or Episyrphus has undergone loss of several typical cyclorrhaphan developmental features. We also confirm the phylogenetic utility of miRNAs within Diptera.

An important finding from all our analyses is that the phylogenetic history of flies has been episodic. Phylogenetic analyses of fewer taxa (tier 1; Fig. S2) and full taxon sets (tier 1/tier 2; Fig. 1 and Fig. S1) reveal three regions of rapid radiations of fly lineages (Fig. 3), characterized by short branch lengths on reconstructed trees that result from both conflicting and limited evidence assignable to relatively brief periods of recent common ancestry. The identified radiations correspond to the earliest fly lineages (including mosquitos); lower Brachycera (including horse flies); and higher flies, schizophoran Cyclorrhapha (including Drosophila and house flies). Our molecular-based time-calibrated phylogeny of dipteran families (Fig. 3, Fig. S3, and Table S3) corroborates the fossil- and molecule-based hypothesis that fly diversification has been driven by episodic bursts of rapid radiation (30). From a Permian origin (31), early aquatic lineages of extant lower Diptera radiated quickly in the Triassic. Later, primarily terrestrial lineages of lower Brachycera, many of which are flower visitors with long proboscides for nectar feeding, radiated in the early Jurassic soon after the recently hypothesized origin of angiosperms (32). Most recently, we see the radiation of the cyclorrhaphan clade Schizophora. A long gap exists between the earliest known Cyclorrhapha in the Early Cretaceous (Opetiala, ∼127 Ma) (6) and the earliest confirmed record of the now-dominant Schizophora in the early Paleocene (Phytomyzites, 64 Ma) (33). Many acalyptrate families are present in the mid-Eocene Baltic amber deposits (∼42 Ma), however (34), implying an explosive radiation of the schizophoran relatives of Drosophila in the early Tertiary (65–40 Ma). The schizophoran radiation, which accounts for more than a third of extant fly diversity and 3% of all animal diversity, is (together with macrolepidopteran moths) the largest insect radiation in the Tertiary (6, 9). This vigorous burst of diversification, likely occurring near the K-T boundary, was coincident with the development of the ptilinal sac, an improved escape mechanism for the fly from its puparium.

The diversification of these three predicted rapid radiations, based on likelihood birth–death models, is characterized primarily by very low extinction rates (discussed in ref. 35), with speciation rates actually lower than immediately adjacent lineages (Fig. S3 and Table S4). In other words, flies appear to become diverse by failing to go extinct rather than through the rapid emergence of new forms. This finding is in contrast to other diverse groups that experience patterns of high turnover (3). The schizophoran radiation, in particular, shows strong evidence for accelerated diversification (Table S4), a finding that is extremely robust to details of node calibration and analysis method. Individual fly families, however, exhibit significant variation in diversification rate.

If diversification was constant across groups or through time, a plot of species diversity vs. age would show that older clades are consistently more diverse, both overall and within major taxonomic divisions. This is not the case with flies (Fig. S4). One reason why diversity and clade age may become decoupled is if diversification slows as old clades saturate available resources, such as might occur in ecologically driven radiation. Moreover, families identified in this plot as having exceptionally low diversification rates are, in many cases, sister groups to major fly clades. Note that the widespread persistence of such “living fossil” lineages is improbable based on most current models of diversification, wherein clade longevity is expected to be strongly correlated with species diversity (36). Taken together, these results suggest an important and complex role for extinction rate variation in generating broad-scale patterns of fly diversity. The generality and macroevolutionary significance of these patterns should be further explored, especially in light of the recognized need for more realistic models to describe diversification at deep phylogenetic levels (37).

Data collection was completed in two tiers. In tier 1, 42 dipteran taxa and three holometabolous outgroups were sampled for sequence data from 12 nuclear protein-coding genes, 18S and 28S ribosomal DNA, and complete mitochondrial genomes (30 kb; Table S1). In tier 2, 202 taxa, comprising at least one species from 149 of the ∼157 recognized families and including three additional outgroups, were sampled for 5 nuclear genes (7 kb; Table S1). Laboratory procedures for sequence collection of nuclear genes and mitochondrial genomes can be found in SI Materials and Methods.

From numerous studies, we established a list of 457 putatively informative morphological characters to cover the anatomical diversity of Diptera. Team consultation reduced this to 371 external and internal morphological characters for larvae (93), pupae (11), and adults (267, including 55 head, 54 wing, 31 female genitalia, and 49 male genitalia). We scored 42 tier 1 dipteran taxa and 5 holometabolous outgroup taxa. More information regarding the collection of morphological data can be found in SI Materials and Methods.

Likelihood (using RAxML) (39) and Bayesian (using MrBayes) (40, 41) analyses were completed using variations of taxon sampling and character inclusion and partitioning, the results of which can be found in Figs. S1 and S2 and Table S2. For the ML analysis shown in Fig. 1, molecular data from tier 1 and tier 2 were combined (202 taxa, 7–30 kb); ribosomal, mitochondrial, and nuclear protein-coding genes were partitioned separately; and third positions and regions of ambiguous homology were excluded. In RAxML, 1,000 independent runs from random starting trees were performed to find the highest scoring replicate (9). ML node support was calculated by acquiring bootstrap values from heuristic searches of 1,000 resampled datasets, using the rapid bootstrap feature of RAxML (42). Taxa were tested for instability using a custom R script, and unstable taxa were removed to observe the effects on support values. More information regarding phylogenetic analyses can be found in SI Materials and Methods.

We constructed miRNA libraries for two fly species from the dataset of tier 1, Episyrphus balteatus (Syrphidae) and Megaselia abdita (Phoridae). Libraries were constructed as described by Wheeler et al. (29). The small RNA libraries were sequenced using the Roche GS-FLX pyrosequencer of the North Carolina State University Genome Sciences Laboratory, and the resulting sequences were analyzed with miRMiner (29). In all, we identified 137,059 miRNA sequences in our sample that were assignable to 1 of 159 known miRNA families.

Divergence time estimates were calculated for the ingroup-only combined dataset using the penalized likelihood method (43) in r8s, version 1.71, with fossil constraints included. Diversification analyses were completed using the MEDUSA program (3). Additional details regarding diversification analyses can be found in SI Materials and Methods, Figs. S3 and S4, and Table S3 and Table S4.