Evolutionary history of burrowing asps (Lamprophiidae: Atractaspidinae) with emphasis on fang evolution and prey selection

Atractaspidines are poorly studied, fossorial snakes that are found throughout Africa and western Asia, including the Middle East. We employed concatenated gene-tree analyses and divergence dating approaches to investigate evolutionary relationships and biogeographic patterns of atractaspidines with a multi-locus data set consisting of three mitochondrial (16S, cyt b, and ND4) and two nuclear genes (c-mos and RAG1). We sampled 91 individuals from both atractaspidine genera (Atractaspis and Homoroselaps). Additionally, we used ancestral-state reconstructions to investigate fang and diet evolution within Atractaspidinae and its sister lineage (Aparallactinae). Our results indicated that current classification of atractaspidines underestimates diversity within the group. Diversification occurred predominantly between the Miocene and Pliocene. Ancestral-state reconstructions suggest that snake dentition in these taxa might be highly plastic within relatively short periods of time to facilitate adaptations to dynamic foraging and life-history strategies.

Based on scale patterns and counts, Laurent [28] assigned the known species of Atractaspis into five groups (Sections A-E). Decades later, Underwood and Kochva [18] partitioned Atractaspis into two groups based on venom gland morphology and geographic distribution: the 'bibronii' group and the 'microlepidota' group. These authors defined the 'bibronii' group as having normal-sized venom glands and a sub-Saharan distribution, and it included the following species: A. aterrima, A. bibronii, A. boulengeri, A. congica, A. corpulenta, A. dahomeyensis, A. duerdeni, A. irregularis, and A. reticulata. The 2nd 'microlepidota' group has relatively elongated venom glands and is found in western, central and eastern Africa, including the distinctive horn of Africa, the Sinai Peninsula, and much of Arabia, Israel, and the Levant. This latter group consisted of the following species: A. engaddensis, A. engdahli, A. leucomelas, A. microlepidota, A. micropholis, and A. scorteccii. Moyer and Jackson [10] reconstructed phylogenetic relationships among 14 species of Atractaspis with morphological data, incorporating Macrelaps and Homoroselaps as outgroups, based on previous studies [18]. However, the two groups of Underwood and Kochva [18] were not supported [10]. More recent molecular phylogenetic studies suggest that Homoroselaps is sister to Atractaspis, whereas Macrelaps is closely related to Amblyodipsas and Xenocalamus [8][9]27].
The diversification of burrowing asps is particularly interesting because of their unique front fangs, which are starkly different from other lamprophiids [21,[29][30][31][32]. It has been hypothesized that foraging for nestling mammalian prey was a major driver in the evolution of front fangs and "side-stabbing," which are unique to Atractaspis [31,33]. Both Atractaspis and Homoroselaps have front fangs, which differs from the rear-fang morphology that is common in their aparallactine sister group. Although Atractaspsis and Homoroselaps both contain front fangs, Atractaspis fang morphology is more similar to viperids (Atractaspis was previously and erroneously classified in the Viperidae), whereas Homoroselaps fang morphology is more similar to elapids [25,31]. Underwood and Kochva [18] suggested a Macrelaps-like ancestor for aparallactines and atractaspidines, which may have foraged above ground and fed on a wide variety of prey items. Specialization on elongated prey items (e.g., squamates and invertebrates) may have taken different evolutionary routes within aparallactines and atractaspidines, which involved morphological changes that facilitated foraging, capture, and envenomation of prey items [31]. Burrowing asps and their sister group Aparallactinae are ideal groups to study fang evolution, because they possess many fang types (i.e., rear fang, fixed front fang, and moveable front fang) [25,[29][30][31][32]. Additionally, collared snakes (aparallactines) and burrowing asps make interesting models to study fang evolution because of their dietary specializations, especially prevalent within the Aparallactinae, which feed on prey ranging from earthworms to blind snakes [25,31].
Herein, we employ phylogenetic hypotheses in conjunction with temporal biogeographic information to gain a more comprehensive understanding of the evolutionary history of Atractaspidinae. Specifically, we evaluate the following questions: Are currently recognized genera and species monophyletic? Are Atractaspis and Homoroselaps sister taxa? Are Atractaspis genetically partitioned into the 'bibronii' and 'microlepidota' groups as Underwood and Kochva [18] suggested? Additionally, we investigate patterns of diversification regarding character traits, including prey selection and fang morphology, within atractaspidines and aparallactines.

Approvals and permissions
Permission for DFH, MB and EG to collect snakes in Uganda was obtained from the Uganda Wildlife Authority (UWA-permit no. 2888 issued on August 1, 2014, permit no. 29279 issued on August 11, 2015) Table 1) that followed appropriate legal guidelines and regulations for collection and loans of specimens.

Laboratory protocols
Genomic DNA was isolated from alcohol-preserved muscle or liver tissue samples with the Qiagen DNeasy tissue kit (Qiagen Inc., Valencia, CA, USA). Primers used herein are shown in Table 2. We used 25 μL PCR reactions with gene-specific primers with an initial denaturation step of 95˚C for 2 min, followed by denaturation at 95˚C for 35 seconds (s), annealing at 50˚C for 35 s, and extension at 72˚C for 95 s with 4 s added to the extension per cycle for 32 (mitochondrial genes) or 34 (nuclear gene) cycles. Amplification products were visualized on a 1.5% agarose gel stained with SYBR Safe DNA gel stain (Invitrogen Corporation, Carlsbad, CA, USA). Sequencing reactions were purified with CleanSeq magnetic bead solution (Agencourt Phylogenetics, fang evolution, and prey selection of snakes in the subfamily Atractaspidinae Bioscience, La Jolla, CA) and sequenced with an ABI 3130xl automated sequencer at the University of Texas at El Paso (UTEP) Genomic Analysis Core Facility.

Sequence alignment and phylogenetic analyses
Phylogenetic analyses were conducted for our individual and five-gene concatenated data sets. Data were interpreted using the program SeqMan [36]. An initial alignment for each gene was produced in MUSCLE [37] in the program Mesquite v3.10 [38], and manual adjustments were made in MacClade v4.08 [39]. The Maximum Likelihood (ML) analyses of single gene and concatenated data sets were conducted using the GTRGAMMA model in RAxML v8.2.9 via the Cipres Science Gateway v3.3 [40]. All parameters were estimated, and a random starting tree was used. Support values for clades inferred by ML analyses were assessed with the rapid bootstrap algorithm with 1,000 replicates [40]. We also conducted Bayesian inference (BI) analyses with MrBayes v3.2.6 via the Cipres Science Gateway [40]. The model included 13 data partitions: independent partitions for each codon position of the protein-coding genes ND4, cyt b, c-mos, and RAG1, and a single partition for the mitochondrial gene 16S. Phylogenies were constructed based on concatenated data, which included 16S and the four protein-coding genes listed above. Concatenated data sets were partitioned identically for ML and BI analyses. The program PartitionFinder v1.1.1 [41][42] was used to find the model of evolution that was most consistent with our data for BI analyses. Bayesian analyses were conducted with random starting trees, run for 20,000,000 generations, and sampled every 1000 generations. Phylogenies were visualized using FigTree v1.3.1 [43].

Divergence dating
The program BEAST v1.8.3 via Cipres Science Gateway [40] was used to estimate divergence times across atractaspidine phylogenetic estimates. The five-gene data set was used to estimate divergence dates in BEAST. Substitution and clock models were unlinked for all partitions; trees were unlinked across the nuclear loci, but were linked for the two mitochondrial partitions because these evolve as a single unit. We implemented an uncorrelated log-normal relaxed clock model with a Yule tree prior. Two independent analyses were run for 100 million generations, sampling every 10,000 generations. Primary calibration points were obtained from Head et al. [44] and a secondary calibration point was obtained from Kelly et al. [7] including: the split between Scolecophidia and all other snakes (120-92 mya); split between Caenophidia and its nearest sister taxon, Booidea (72.1-66 mya); split between Colubroidea and its nearest sister taxon (Acrochordus + Xenodermatidae) (72.1-50.5 mya); the divergence of Colubridae + Elapoidea (30.9 ± 0.1 mya); and the split between Crotalinae and Viperinae

Ancestral-state reconstructions
To understand the evolution of fang morphology and diet selection in atractaspidines, we reconstructed the pattern of character changes on the ML phylogeny herein. For ancestralstate reconstructions, we included all samples of aparallactines and atractaspidines available to us in order to better characterize fang and diet characters. All ancestral-state reconstructions were conducted by tracing characters over trees in Mesquite v3.10 [38]. We scored taxa using descriptions from the literature [25,[30][31][45][46][47][48][49][50][51][52][53][54][55], and from our own data. We evaluated the following characters for fang morphology and diet selection: A. Fang morphology: (0) no fang, (1) rear fang, (2) fixed front fang, (3) moveable front fang, and (4) rear-front fang intermediate (anterior half of the maxilla, but not the anteriormost tooth); B. prey selection (0) rodents, (1) rodents, snakes, fossorial lizards, and amphibians, (2) snakes, (3) amphisbaenians, (4) snakes and fossorial lizards, (5) invertebrates, and (6) fish and amphibians. A ML approach was used for both analyses, because it accounts for and estimates probabilities of all possible character states at each node, thus providing an estimate of uncertainty [56]. A Markov K-state oneparameter model (Mk-1; [57]) that considers all changes as equally probable was implemented in our ancestral-state reconstructions. States were assigned to nodes if their probabilities exceeded a decision threshold; otherwise nodes were recovered as equivocal.

Morphology
Microcomputed tomography (CT) scans of specimens were produced using GE Phoenix V|Tome|X systems at the General Electric Sensing & Inspection Technologies in Scan Carlos, CA and University of Florida's Nanoscale Research Facility. X-ray tube voltage and current, detector capture time, voxel resolution, and projection number were optimized for each specimen (S1 File). The radiographs were converted into tomograms with Phoenix Datos| R, and then rendered in three dimensions with volumetric rendering suite VGStudioMax 3.2 (http:// www.volumegraphics.com). Tomogram stacks and 3D mesh files for all scans are available on Morphosource.org (S1 File). . Individuals with missing data were included in the concatenated sequence analyses, because placement of individuals that are missing a significant amount of sequence data can be inferred in a phylogeny, given an appropriate amount of informative characters [8,[58][59][60]. Furthermore, Jiang et al. [61] showed that excluding genes with missing data often decreases accuracy relative to including those same genes, and they found no evidence that missing data consistently bias branch length estimates.

Concatenated gene tree analyses
The following models of nucleotide substitution were selected by PartitionFinder for BI analyses: 16S (GTR+G), ND4 1 st codon position (GTR+G), ND4 2 nd codon position (TVM+G), and ND4 3 rd codon position (HKY+I+G); cyt b 1 st codon position (TVM+G), cyt b 2 nd codon position (HKY+I+G) and cyt b 3 rd codon position (GTR+G); c-mos and RAG1 1 st , 2 nd and 3 rd codon positions (HKY+I). Preferred topologies for the ML and BI analyses were identical, with similar, strong support values for most clades (Fig 2), and single-gene mtDNA analyses recovered similar topologies (not shown). The ML analysis likelihood score was -46340.867388. The relationships of Elapidae, Lamprophiinae, Micrelaps, and Psammophylax with respect to the ingroup Atractaspidinae, were not strongly supported in ML and BI analyses. However, Atractaspidinae was recovered in a strongly supported clade. Atractaspis and Homoroselaps were strongly supported as sister taxa (Fig 2). The genus Homoroselaps was recovered as a monophyletic group, and H. lacteus was partitioned into several well-supported clades. There were several strongly supported clades within Atractaspis:  A. microlepidota, A. micropholis, A. watsoni, and A. sp. Atractaspis andersonii did not form a monophyletic group, because one of the samples from Oman (AF471127) was recovered as sister to a clade of A. engaddensis with strong support (Fig 2). The western African species A. aterrima was recovered with strong support as sister to a clade containing A. reticulata heterochilus and A. boulengeri. Atractaspis corpulenta kivuensis samples from eastern DRC were strongly supported as sister to A. corpulenta from northwestern Republic of Congo (near Gabon, the type locality). A well-supported clade of Atractaspis irregularis samples was partitioned by strongly supported central (A. cf. irregularis) and western African (A. irregularis) subclades. Atractaspis duerdeni was recovered within a well-supported A. bibronii complex. Atractaspis bibronii rostrata samples were partitioned into two highly divergent clades from southeastern DRC and Tanzania/Mozambique.
For the analyses including all atractaspidine and aparallactine samples available to us (Fig 3), preferred topologies for the ML and BI analyses were identical, with similar, strong support values for most clades (Fig 3). The ML analysis likelihood score was -73090.650849. The concatenated ML and BI analyses recovered similar topologies to those from Portillo et al. [62] and Fig 2.

Divergence dating
Topologies from the BEAST (Fig 4) analyses were mostly consistent with the results from our concatenated tree analyses (Figs 2 and 3). BEAST results recovered A. corpulenta corpulenta/A. corpulenta kivuensis as sister to A. congica/A. dahomeyensis with strong support (Figs 2-4). Additionally, the relationship between Atractaspis irregularis and A. corpulenta/A. congica/A. dahomeyensis was strongly supported in BEAST analyses (Fig 4). Results from dating analyses suggested atractaspidines split from aparallactines during the early Oligocene around 29 mya (24.8-31.4 mya, 95% highest posterior densities [HPD]) ( Table 3, Fig 4), which is similar to the results (34 mya) of Portillo et al. [62]. Subsequently, Atractaspis split from Homoroselaps in the mid-Oligocene, and most radiation events within each of the major clades associated with these genera occurred during the mid-to late Miocene and Pliocene (Fig 4). Specific dates with ranges are specified in Table 3.

Ancestral-state reconstructions
X-ray computer tomography of collared snakes and burrowing asps can be seen in Figs 3 and 5. Likelihood reconstructions of atractaspidine ancestral fang morphology inferred a rear fang condition for the ancestral condition of all lamprophiids (96.7%) (Fig 6[A]). Subsequently, the Subfamily Lamprophiinae lost a venom delivery fang condition. The common ancestor of aparallactines and atractaspidines was inferred to have a rear fang condition (97.8%). The analyses suggested a rear fang ancestor (72.5%) for the clade containing Homoroselaps and Atractaspis. The ancestor to Atractaspis was inferred to have a moveable front fang condition (97.4%). Results recovered a fixed front fang condition for the ancestor of all Homoroselaps (99.8%). The ancestor to all aparallactines was inferred to have a rear fang condition (99.6%), and this remained consistent throughout most aparallactine nodes with the exception of Polemon (rear/front fang intermediate, 97.8%) and Aparallactus modestus (no specialized fang, 99.7%).
For the analyses with diet data, likelihood reconstructions inferred a generalist diet of rodents, reptiles, and amphibians for the ancestral condition of all lamprophiids (99.7%) (Fig 6  [B]). Several lamprophiines (Lycodonomorphus) subsequently adopted a more specialized diet of amphibians, reptiles, and fish. The common ancestor for aparallactines and atractaspidines was inferred to have a generalist diet of rodents, reptiles, and amphibians (92.4%). Results   Table 3). https://doi.org/10.1371/journal.pone.0214889.g004 Phylogenetics, fang evolution, and prey selection of snakes in the subfamily Atractaspidinae (amphisbaenians, 98.8%). Results suggested a generalist diet for Atractaspidinae (92.3%). The ancestor of Homoroselaps was inferred to have a diet consisting of mostly lizards and snakes (99.9%), whereas the ancestor of Atractaspis was inferred to have a broader diet of rodents, reptiles, and amphibians (99.2%).

Biogeography
Atractaspidines are distributed throughout sub-Saharan Africa except for three species of Atractaspis that are found in western Asia/Middle East (Atractaspis andersonii, A. engaddensis, and A. microlepidota) [25,[29][30][31]. Based on our results, the most likely scenario for Atractaspis is an African origin with a vicariance or dispersal event into the western Asia/Middle East region in the late Miocene (Fig 4). Atractaspis from western Asia/Middle East and Africa last shared a common ancestor during the late Miocene around 12.1 mya (7.8-17.6). Other studies of African-western Asian/Middle Eastern complexes (e.g., Echis and Uromastyx) recovered similar dates during the late Miocene, with the Red Sea proving to be a strong biogeographic barrier [63][64][65][66][67][68][69]. However, lineages of Varanus from Africa and the Middle East split from each other 6.9 mya [70], and African and Middle Eastern Bitis arietans last shared a common ancestor around 4 mya [64]. These dating estimates suggest that there were multiple dispersal events, which were taxon specific. Many Middle Eastern amphibians and reptiles have Atractaspis began to diversify around the mid-Oligocene simultaneously with many aparallactine genera [62]. Many of the modern species split from recent common ancestors during the mid-to late Miocene (Table 3, Fig 4). The late Miocene was characterized by considerable xeric conditions, which led to the expansion of savannas globally [72][73]. Other studies on Central and East African herpetofauna, including squamates (Adolfus, Atheris, Boaedon, Naja, Kinyongia, and Panaspis) and frogs (Amietia, Leptopelis, and Ptychadena), have shown similar trends of species diversification during the late Miocene [3-5, 62, 74-78]. Phylogenetics, fang evolution, and prey selection of snakes in the subfamily Atractaspidinae The diversification of several western and central African Atractaspis was most likely a consequence of increasingly xeric conditions during the Miocene, when forest and other moist habitats were fragmented [72]. These Atractaspis were likely isolated in fragmented patches of forest during the mid-to late Miocene. Atractaspis irregularis is partitioned clearly by western African and central African lineages that diverged in the mid-Miocene, similar to Aparallactus modestus [62]. At this time, southern African and Middle Eastern Atractaspis also diversified. Atractaspis from the Near and Middle East (A. andersonii, A. engaddensis, and A. microlepidota) and southern Africa (A. bibronii and A. duerdeni) are not tropical forest species, and they inhabit deserts or semi-desert savannas and dry woodland [30,[79][80]. This adaptation to more xeric and open habitats would have allowed Near and Middle Eastern, and southern African Atractaspis, to disperse into these habitats during the dry conditions of the mid-to late Miocene. Studies on mammals and birds show most diversification events during the Pliocene [81][82][83][84], which is consistent with the timing of diversification for Atractaspis aterrima, A. congica, A. dahomeyensis, and populations of South African A. bibronii (Fig 4). Phylogenetics, fang evolution, and prey selection of snakes in the subfamily Atractaspidinae In contrast to Aparallactus jacksonii, Atractaspis bibronii rostrata showed no clear genetic partitioning between populations in the Nguru, Usambara, and Udzungwa Mountains [62]. Aparallactus jacksonii clearly exhibited deep divergence between an extreme northern Tanzanian population, and a population from the Nguru Mountains. These two populations diverged from each other during the late Miocene, suggesting that the habitats of this taxon were fragmented with increased aridity [62]. Other vertebrate taxa that have shown substantial divergences between populations found in extreme northern Tanzania (Usambara, Taita, and Pare Mountains) and those slightly south (Uluguru, Ukaguru, Nguru, and Malundwe Mountains), include the reed frog Hyperolius puncticulatus, the green barbet (Stactolaema olivacea), and the streaky canary (Serinus striolatus) [82,85]. But like Atractaspis bibronii rostrata, the hyperoliid reed frog Hyperolius spinigularis and the aparallactine Aparallactus guentheri showed no clear biogeographic patterns between populations in different areas of the Eastern Arc Mountains. These results support the hypothesis that the evolutionary history of species from the Eastern Arc Mountains is lineage specific [85]. Atractaspis bibronii rostrata inhabit low-elevation woodlands and grasslands, and transitional habitats, rather than montane forest (i.e., Aparallactus jacksonii) [25]. This would allow taxa such as Atractaspis bibronii rostrata to continuously disperse between the different mountains of the Eastern Arcs, despite increased aridity. Additionally, ecological niche requirements may also explain the different biogeographic patterns seen in Aparallactus jacksonii and Atractaspis bibronii rostrata. Atractaspis bibronii has a generalist diet (mammals, squamates, and amphibians) and could have exploited more habitats than Aparallactus jacksonii, which is a centipede specialist [25].

Evolutionary relationships and taxonomy of Atractaspidinae
Our results indicate that both Atractaspis and Homoroselaps are strongly supported as monophyletic sister taxa. Results from Figueroa et al. [27] recovered a monophyletic group containing aparallactines and atractaspidines, but their results did not recover a monophyletic Atractaspis (A. irregularis was recovered as sister to aparallactines + atractaspidines). This sample was excluded from our analyses, because the only sequence available for this taxon was from BDNF, a gene not used herein. The results from Figueroa et al. [27] may be an artifact of sample size of atractaspidines, or incomplete lineage sorting of the BDNF nuclear gene. Results from our study indicate that A. irregularis is a monophyletic lineage within a strongly supported, monophyletic Atractaspis.
Underwood and Kochva [18] recognized two groups within Atractaspis: (1) the 'bibronii' group (represented in our study by A. aterrima, A. bibronii, A. boulengeri, A. congica, A. corpulenta, A. dahomeyensis, A. irregularis, and A. reticulata), characterized by a single posterior supralabial, three anterior infralabials, normal-sized venom glands, and a sub-Saharan distribution; and (2) the 'microlepidota' group (represented in our study by A. andersonii, A. engaddensis, A. microlepidota, and A. micropholis), characterized by two anterior temporals, highly elongated venom glands, and a North African/Near and Middle Eastern distribution. Whereas our study did not include genetic samples of all known species of Atractaspis, results herein (Fig 2) support partitioning of the genus into two groups sensu Underwood and Kochva [18]. Our results indicated a clear partition between a 'Middle Eastern + African' clade (including A. watsoni, a species that was not included by Underwood and Kochva [18]) and a 'sub-Saharan African' clade (Figs 2 and 4). These results strengthen the notion that venom gland size and length in Atractaspis are homologous. Our support for the 'microlepidota' group is consistent with the "Section A" (A. andersonii, A. fallax, A. leucomelas, A. microlepidota, and A. micropholis) of Laurent [28] and the A. micropholis/A. microlepidota/A. watsoni clade recovered by Moyer and Jackson [10]. However, our phylogeny (Fig 2) contrasts with the remaining "sections" of Laurent [28], most relationships depicted in the morphological phylogeny of Moyer and Jackson [10], and the molecular phylogenies of Pyron et al. [8][9] and Vidal et al. [22].
Based on relatively long branch lengths, several lineages of Atractaspis seem to be cryptic complexes of species. Because of the extensive geographic distribution of A. bibronii in central, eastern and southern Africa, it is unsurprising to find several highly divergent lineages that likely represent cryptic species. Given the proximity (ca. 167-333 km) of our Tanzanian localities of A. bibronii rostrata (Nguru, Usambara, and Udzungwa Mountains) to the insular type locality for this taxon (Zanzibar, Tanzania), the morphological similarity between our voucher specimens and the types [86], and the relatively long branch length and reciprocal monophyly of this clade compared to topotypic South African A. bibronii (Fig 2), it is likely that the former taxon is a valid species. However, additional comparisons to type specimens are needed to clarify the taxonomic status of populations in this clade, including samples from Haut-Katanga Province in southeastern DRC.
Our phylogenetic results indicated that several other species, including A. andersonii, A. boulengeri, A. congica, A. corpulenta, A. dahomeyensis, and A. irregularis likely represent more than a single species. For example, topotypic Angolan samples of A. congica are deeply divergent from our eastern DRC sample (Fig 2), which is likely attributable to A. congica orientalis [46]. Like Polemon fulvicollis fulvicollis (Gabon) and P. fulvicollis laurenti (DRC) [62], Gabonese Atractaspis corpulenta and eastern DRC populations of A. corpulenta kivuensis also showed marked genetic divergences between each other (Fig 2). The well-supported clade of A. irregularis from western Africa likely includes topotypic populations, because they straddle the type locality (Accra, Ghana) [87], whereas our Albertine Rift samples are likely attributable to one of the taxon's many synonyms. One of these, Atractaspis bipostocularis from Mt. Kenya, was named for its two postocular scales, which distinguishes it from the single postocular of topotypic A. irregularis [88]. Because Mt. Kenya is located east of the Kenyan Rift, a major biogeographic barrier to several species of squamates [78], and moreover, all voucher specimens of A. cf. irregularis from the Albertine Rift have a single postocular (EG pers. obs.), A. bipostocularis is likely a distinct species that is endemic to the central Kenyan highlands. Other synonyms of A. irregularis that have one postocular and type localities in or near the Albertine Rift are likely attributable to our well-supported clade of A. cf. irregularis (Fig 2 in [87]), and include the following taxa: A. conradsi Sternfeld, 1908 Laurent, 1945 (type locality: Bunia, DRC [46]). Additional sampling and morphological analyses are in progress that will help clarify the correct taxonomy for these lineages. Because of the relative lack of fieldwork in Central Africa in recent decades [92][93] and the relatively rare encounters of these snakes above ground (EG, pers. obs.), it is likely that genetic samples from the above topotypic populations will remain elusive for many years.

Evolution of dietary preference and fang morphology
Burrowing asps and collared snakes have unique ecologies, particularly in terms of dietary preferences. Atractaspis in particular have very distinctive fangs (solenoglyphous fangs, similar to viperids) that have made their taxonomic history complicated (e.g., previously classified as viperids) [25,31,94]. The fangs of Homoroselaps resemble fangs of elapids more than vipers. In contrast, aparallactines tend to have rear fangs (Figs 3 and 6) [18,25,[29][30]. Our ancestralstate reconstruction analysis of fang morphology suggested a rear fang ancestor for all collared snakes and burrowing asps (Aparallactinae and Atractaspidinae). Most lamprophiids are either rear fanged or lack fangs [25]. Our analyses also recovered dietary generalization as an ancestral-state for atractaspidines and aparallactines. Both of these conditions support the hypothesis proposed by Underwood and Kochva [18], which postulated that collared snakes and burrowing asps likely had a Macrelaps-like ancestor (large and rear fanged) that foraged above ground or in burrows of other organisms, and these taxa subsequently evolved into more specialized forms with specialized diets. Several aparallactines are dietary specialists [25,31], that feed on the following: Aparallactus specialize on centipedes and possibly other invertebrates like earthworms; Chilorhinophis and Amblyodipsas consume snakes and other small, fossorial reptiles; Polemon are ophiophagous [25,31,95], but may occasionally consume other squamate prey items; Macrelaps consume reptiles, amphibians, and rarely mammals [25]; and Xenocalamus consume amphisbaenians [25,31].
Unlike several aparallactines, Atractaspis are dietary generalists that consume a diverse variety of squamates, rodents (particularly nestling rodents), and occasionally amphibians [25,31,33,52,[96][97][98][99][100]. The venom glands of Atractaspis are anatomically distinct from those of other front-fanged snakes such as viperids and elapids, because atractaspidines lack a distinct accessory gland and the presence of mucous-secreting cells at the end of each serous tubule [32,[101][102][103]. Similar to two other front-fanged snake groups (Elapidae and Viperidae), elongated venom glands have evolved within Atractaspis from western and northern African, and western Asia/Middle East species. These glands may be up to 12 cm long in A. engaddensis and 30 cm long in A. microlepidota [32]. Phylogenetically, Atractaspis is clearly partitioned according to venom gland length and geographic distribution (Figs 1 and 2). The purpose of these anatomical adaptations are unclear, although it is possible that they evolved to influence venom yield, as in Calliophis bivirgatus (Elapidae) [32]. The unique viper-like front fangs of Atractaspis may have evolved to facilitate the predation of rodent nestlings or squamates in tight burrows. Preying on animals in tight burrows limits mobility of the predator, because the body of the prey item can serve as a physical barrier, stopping the predator from further pursuit. Many lizards can detach their tails if a predator grabs the tails from behind. Shine et al. [31] postulated that it would be advantageous for a predator to push past the tail and envenomate or seize the prey by the body, a scenario ideal for Atractaspis. Deufel and Cundall [33] hypothesized that the evolution of the front fang in Atractaspis was likely the result of the following advantages: (1) greater envenomation efficiency resulting from the longer fangs; (2) closed mouth venom delivery system, allowing envenomation during head contact with any part of the prey; (3) capacity to quickly envenomate and release prey; and (4) potential for effective defense against adult rodents. Most prey consumed by Atractaspis (amphisbaenians, fossorial skinks, typhlopid snakes) [25] are also consumed by other atractaspidines and aparallactines, including Amblyodipsas, Chilorhinophis, Homoroselaps, Macrelaps, Polemon, and Xenocalamus [25,31,97]. These observations suggest that squamate prey are consumed across all atractaspidine and aparallactine genera, and therefore, they may not be the only selective force driving the evolution of the unique fang in Atractaspis. However, rodents and other mammals are not commonly preyed on by other burrowing asps and collared snakes [31,104]. Deufel and Cundall [33] stated that it is unlikely that mammalian prey alone drove the evolution of a moveable front fang in Atractaspis, but the success and wide distribution of this genus may be partially attributed to mammalian prey. Unlike aparallactines, Atractaspis can quickly envenomate and dispatch all rodents in a nest [33]. A rear fang condition would require the snake to bite, hold and chew on every prey item, which is undoubtedly a more energetically costly form of envenomation compared with the predatory behavior of Atractaspis. Interestingly, in a feeding experiment, Atractaspis never attempted to ingest snake prey until the prey stopped reacting to fang pricks [33]. This observation suggests that Atractaspis will not risk injury until prey are completely immobilized. The unique fang and predatory behavior of Atractaspis has its functional trade-offs; Atractaspis lack large mandibular and maxillary teeth that allow snakes to quickly consume prey [33], and therefore, they take longer to ingest prey items. Because Atractaspis forage, kill, and consume prey in the soil and below the surface, there were likely no negative selective pressures acting against slow ingestion of prey. Because they are fossorial, Atractaspis may be relatively safe from predators while feeding, which is when non-fossorial snakes may be vulnerable to predation or attacks from other animals [25,33].
Results from this study indicate that the rear-fang condition can cover a wide variety of dietary specializations. But this condition is not ubiquitous among aparallactines. Aparallactus modestus clearly lacks enlarged fangs (Figs 5 and 6), but previous studies have found venom glands in this taxon [105]. Additionally, the venom gland of A. modestus is reported to differ from the venom gland of A. capensis, but further details of the discrepancies were not discussed [32,105,106]. Interestingly, this species may prey on earthworms rather than centipedes (II pers. obs. [30]), explaining the loss of a rear-fang condition, which is present in all other Aparallactus species used for this study, including A. niger, the sister species to A. modestus (Figs 5 and 6).
Polemon fangs are not easily classified. The fangs of Polemon are located on the anterior half of the maxilla, rather than the more typical posterior end (Figs 5 and 6). These fangs are large and deeply grooved, and resemble a fixed front-fang condition, but yet they are positioned behind one or two smaller maxillary teeth. The ophiophagous diet of Polemon likely influenced the evolution of a front-fang condition in this genus. Polemon are known to prey on large and formidable snake prey, which can rival the predator in size [35,48,95,107]. With large, deeply grooved fangs positioned on the anterior side of the maxilla, Polemon can quickly envenomate and kill relatively large and powerful prey (snakes) more effectively than they would with a rear-fang condition like Aparallactus. Snakes with rear fangs must typically chew in a forward orientation until the rear fang can penetrate the flesh of the prey item [25]. Several front-fanged, elapid genera prey heavily on snakes (e.g., Micrurus and Ophiophagus). The front-fang condition may be a favorable trait to feed on snakes, in order to immobilize and kill more quickly.
In Xenocalamus, similar selective pressures (e.g., tight burrow foraging) that led to the evolution of fang and predatory behaviors in Atractaspis, may have led to the evolution of its unique quill-shaped snout [31]. Unlike Amblyodipsas polylepis, Xenocalamus possess relatively large maxillary teeth that gradually increase in size from the anterior to posterior side of the maxilla (Figs 3 and 5). This trait seems advantageous to improve their grasp of amphisbaenian prey.
It is not surprising that the rear fang and dietary generalist conditions were recovered as the ancestral-state condition for both atractaspidines and aparallactines, considering many lamprophiids are dietary generalists [25,30]. Collared snakes and burrowing asps seem to have experienced the opposite of niche conservatism as results herein indicated that foraging behaviors and diet have heavily and rapidly influenced the evolution of fang morphology, dietary specializations, and snout shape. In collared snakes (aparallactines), dietary specializations seem to have shaped variation (and loss) of fangs and snout shape, particularly for Aparallactus, Polemon, and Xenocalamus. These genera tend to have more specialized diets than Macrelaps, Chilorhinophis and Amblyodipsas, all of which possess more typical rear fangs (Figs 3 and 5) [25,[30][31]. A fundamental controversy in snake evolution is whether front and rear fangs share the same evolutionary and developmental origin. Burrowing asps and collared snakes possess all known types of snake dentition (no fang, rear fang, fixed front fang, and moveable front fang). Our results lend credence to the hypothesis that rear fangs and front fangs share a common origin [94]. Our results also indicated that snake dentition, specifically alethinophidian groups such as atractaspidines and aparallactines, may be highly plastic within relatively short periods of time to facilitate foraging and life history strategies.