To solve similar environmental problems, unrelated organisms may either evolve different traits or converge on similar solutions. Repeated morphological patterns observed in equivalent ecological settings often disclose convergence wrought by comparable selective forces. While such outcomes support the role of selection, its operational strength is perhaps better gauged by a pattern's frequency of recurrence. East Africa's lacustrine cichlids (Kocher et al. 1993; Allender et al. 2003), Caribbean Anolis lizards (Losos et al. 1998), coral reef wrasses (Westneat et al. 2005), and Myotis bats (Ruedi & Mayer 2001) all provide exemplary cases, in which particular ecomorphs have arisen multiple times under similar environmental conditions. The independent origins of traits documented in these studies provide important evidence for the role of ecological selection pressures in evolution.

Many desert lizards also exhibit broad convergence, and certain xeric environments—for example, the loose, windblown sand of dunes—have produced remarkably similar evolutionary outcomes in several divergent taxa (Pough 1969; Pianka 1986; Arnold 1995). One conspicuous pattern shared among sand-dwelling (psammophilous) lizards is pedal specialization, often involving the elaboration of lateral phalangeal scales into fringes that facilitate running or ‘swimming’ in loose sand (Luke 1986). Dune-dwelling adaptation has taken different forms in geckos, including the evolution of toe fringes, though the six genera possessing such fringes are all primitively terrestrial forms (Bauer & Russell 1991). Within arboreal gecko lineages, the digits of psammophilous taxa have instead undergone morphological reduction, resulting in diminution or loss of the gekkotan subdigital pad—arguably one of nature's most complex pedal structures (Russell 2002).

In most geckos, climbing involves dry adhesion, a function of the subdigital pads (toe pads), whose millions of hair-like setae interact via van der Waals forces with substrate surfaces (Autumn et al. 2002; Huber et al. 2005). Adhesion requires proximity between setae and substrate, and geckos achieve this intimate contact by engaging adipose pads or vascular sinuses that subtend the setal-bearing plates (lamellae). Fine-tuned control governing the kinematics of setal attachment and detachment is accomplished by elaborate musculotendinous networks (Russell 2002). Remarkably, toe pads have evolved on multiple occasions in geckos, once in the Diplodactylidae and several times within the Gekkonidae (Russell 1979; Han et al. 2004). Conversely, taxa in both families have also lost this adhesive system, having undergone ecological shifts from arboreal to terrestrial adaptive zones (Bauer & Russell 1991).

Chondrodactylus, Colopus and Palmatogecko are terrestrial gekkonids endemic to arid southwest Africa. Although their toe pads are greatly reduced or altogether absent (figure 1), myological characters supporting their arboreal ancestry persist (Russell & Bauer 1990; Bauer & Russell 1991). Distinctive spinose scales also cover relatively short digits and/or plantar surfaces of the feet (figure 1). These pedal modifications facilitate locomotion and burrowing in dune systems of the Namib and Kalahari deserts (Haacke 1976d; Bauer & Russell 1991). Speculation is mixed regarding the advent and evolution of pedal modification in these taxa. Key in resolving competing views on terrestrial origins is a clarification of phylogenetic relationships among the three genera—henceforth termed the burrowing geckos—and the lineage to which they belong, the Pachydactylus Group (sensu Russell 1972).

Figure 1 Pachydactylus turneri (a) representing the plesiomorphic (=climbing) subdigital condition for the Pachydactylus Group—and the four species of burrowing geckos, (b) Colopus wahlbergii, (c) Chondrodactylus angulifer, (d) Palmatogecko vanzyli, and (more ...)

The Pachydactylus Group is a chiefly African assemblage comprising six genera and 73 nominal species. Spanning arboreal to ultrapsammophilous forms, this ecologically varied clade is united by an additional phalanx (hyperphalangy) in the first digit of both manus and pes (Russell 1972; Haacke 1976d). Hyperphalangy is otherwise rare in geckos; thus, monophyly of the Pachydactylus Group is widely accepted (Russell 1972; Haacke 1976d; Bauer 1990a; Kluge & Nussbaum 1995). Two subgroups are recognized: a northern, largely Mediterranean clade (Carranza et al. 2002) and a broadly distributed southern lineage (Bauer 1999). In contrast to the rather uniform northern clade (a single genus), the southern lineage exhibits extraordinary morphological variation and phyletic richness, attributed in part to the subcontinent's topographical diversity (Crowe 1990; Bauer 1999).

Pachydactylus is Africa's most species-rich gekkonid genus (40+spp.), occurring throughout the continent's southern sector (Bauer 1999). Its broad distribution encompasses the geographic ranges of Chondrodactylus, Colopus, Palmatogecko and Rhoptropus—all the remaining genera in the southern Pachydactylus Group. Haacke (1976d) postulated close systematic affinities among Chondrodactylus, Colopus and Palmatogecko, hypothesizing that they formed the sister group of Pachydactylus. Similarly, the burrowing geckos formed a monophyletic group that was sister to Rhoptropus+Pachydactylus in a cladistic analysis of African–Madagascan gekkonid genera (Kluge & Nussbaum 1995). Others (Joger 1985; Bauer 1990a) have suggested that the burrowing geckos may have arisen within Pachydactylus. Bauer & Good (1996) hypothesized that Rhoptropus is sister taxon to a monophyletic Pachydactylus sensu lato, i.e. Pachydactylus+Chondrodactylus, Colopus and Palmatogecko. Clearly, a consensus on evolutionary relationships among these southern genera is yet to be established.

We report DNA sequences for one nuclear and three mitochondrial loci, from which we derive a phylogenetic hypothesis for the southern Pachydactylus Group. We use this molecular phylogeny to establish the systematic status of Chondrodactylus, Colopus and Palmatogecko and address the question: does pedal modification in the burrowing geckos trace to a single origin, or did it arise independently and repeatedly through convergence?

Sequences were aligned using Clustal x 1.81 (Thompson et al. 1997), and 12S and 16S alignments were examined in detail with regard to indel variation, exploring gap placements for a series of gap opening and extension costs. Regions of rRNA sequence for which inferred nucleotide position homologies varied across gap parameters were considered alignment-ambiguous and excluded from further analysis.

Table 1 Summary statistics for parsimony and Bayesian analyses.

We also analysed the sequence data using maximum parsimony (MP), allowing comparisons to clades (and their support) identified by Bayesian inference. Prior to MP analysis, we conducted an incongruence length difference test, implemented in PAUP^* 4.0 (Swofford 2002) as a partition homogeneity test, to detect possible incongruence among genes. This test (100 random addition sequences of taxa; 500 replicates) did not contradict the congruence of the four gene sequence partitions (P=0.970), which were combined for MP analysis. Parsimony trees were generated by a heuristic search in PAUP^*, with tree bisection-reconnection branch swapping, MULPARS (save all equally most parsimonious trees) and random addition of sequences (1000 replicates). Nodal support for MP trees was estimated by a bootstrap analysis involving 1000 pseudoreplicates.

The three Bayesian runs yielded statistically equivalent log-likelihood scores and identical consensus trees, supporting their convergence. Figure 2 depicts this consensus phylogram in which Rhoptropus is sister to the other southern group species (pP =1.0), and all three burrowing genera lie embedded within Pachydactylus (pP =1.0). The basal clade of Pachydactylus contains the two tropical African species, Pachydactylus tetensis and Pachydactylus tuberculosus. The remaining species of Pachydactylus, together with the three burrowing genera, form an assemblage (pP =1.0) with a largely arid zone distribution across southern Africa. Moreover, each burrowing genus is placed in a separate clade, revealing three distinct lineage associations within the southern Pachydactylus radiation (figure 2). Chondrodactylus is sister taxon (pP =1.0) to the bibronii complex, a group of three large-bodied species with well-developed toe pads. Within the remaining clade of Pachydactylus, Palmatogecko receives strong support (pP =1.0) as sister taxon to Pachydactylus austeni and, elsewhere, Colopus is less well supported (pP =0.88) as sister species to Pachydactylus kochii.

Figure 2 Bayesian inference phylogram for the southern Pachydactylus Group. Tree branches are coded blue for Pachydactylus and red for the burrowing genera (also illustrated). Specific epithets are those of Pachydactylus. Grey branches indicate posterior probabilities (more ...)

MP analysis of 827 informative nucleotides, with 669 sites contributed by the mtDNA partition and 158 by RAG-1, yielded two equally parsimonious trees (TL=6333; CI =0.273; RI =0.351). The MP consensus tree is largely congruent with the Bayesian tree (figure 3). Bootstrap support for the clade of southern African Pachydactylus+Chondrodactylus, Colopus and Palmatogecko is reasonably strong (84%; 97%, based solely on the RAG-1 partition), and each burrowing genus falls within the respective clades of Pachydactylus recovered by Bayesian inference. Several other well-supported clades identified in both analyses corroborate previously recognized alpha-level groups in Pachydactylus, the taxonomic implications of, which will be detailed elsewhere (Bauer & Lamb 2005).

Figure 3 Comparison of Bayesian and MP trees, the latter a strict consensus of two equally parsimonious trees. Specific epithets are those of Pachydactylus; burrowing genera are listed in bold. Grey branches indicate posterior probabilities (pP)≥0.96 and (more ...)

Table 2 Results of SH tests comparing the Bayesian inference tree with alternative topologies. (Δ−lnL is the difference in ln-likelihood between the alternative topology as stated and the Bayesian tree depicted in figure 2. Each SH test (more ...)

Geckos have experienced incontestable evolutionary success in xeric ecosystems, forming one of the more significant components of desert vertebrate fauna worldwide (Pianka 1986). The majority possesses toe pads and are rock outcrop (rupicolous) specialists; relatively few desert-dwelling geckos are truly terrestrial, of which fewer still represent pad-bearing lineages (Bauer & Russell 1991). The rarity of such transformation, in conjunction with systematic inferences for the monophyly of Colopus, Chondrodactylus and Palmatogecko (Haacke 1976d; Kluge & Nussbaum 1995), is consistent with interpretations that their digital modifications for burrowing were derived but once from a climbing ancestor. Using multi-genic sequence data and effectively complete taxon sampling, we have generated the first explicit phylogenetic hypothesis for the southern Pachydactylus Group. Highlights of this robust phylogeny include: (i) the unequivocal incorporation of Chondrodactylus, Colopus and Palmatogecko within Pachydactylus, and (ii) their placements in distinct, separate clades. From these findings, we argue that subdigital transformations observed among the burrowing geckos evolved independently. In effect, the convergence on a pedal design for locomotion and burrowing in sand reflects repeated divergence in function within lower-level clades of Pachydactylus.

The reconstruction of character evolution in Pachydactylus revealed a major shift from general to strictly rupicolous climbing (figure 4), with the more basal members of the genus being arboreal opportunists (i.e. using both trees and rocky surfaces). Most Pachydactylus are indeed rupicolous, yet within this rock-specialist assemblage, the transition from climbing to terrestrial forms has occurred a minimum of eight times (figure 4). Interestingly, five of the eight terrestrial lineages occupy sandy substrates. A trend revealing lower lamellar counts among the more derived lineages must be interpreted with caution, given correlation between body size and number of lamellae in geckos (Bauer 1990b). This caveat notwithstanding, lamellar counts for terrestrial species represent the lowest numbers throughout the phylogeny. Such is the case for the sister species to Palmatogecko (Pachydactylus austeni; n=3–4) and Colopus (P. kochii; n=3). Further, both P. kochii and P. austeni qualify as psammophiles, the latter species essentially replacing Palmatogecko on Namib dunes south of the Holgat River (Haacke 1976a).

Figure 4 Reconstruction of the evolution of substrate preferences and digital lamellar counts on the Bayesian tree for the southern Pachydactylus Group.

In addition to lower lamellar counts, the terrestrial species of arid southwest regions (P. austeni, P. kochii, Pachydactylus labialis, Pachydactylus mariquensis, Pachydactylus punctatus, Pachydactylus sherzi) show concomitant reductions in: (i) pad area; (ii) size of the distal phalanges; and (iii) size of the dorsal interossei muscles, which permit digital hyperextension (Russell 1976). During hyperextension, the mechanism responsible for setal disengagement, the digits are peeled from the substrate distally to proximally, detaching setal fields to release the pad and foot. This elaborate mechanism remains operational in terrestrial Pachydactylus, but with a different purpose: their reduced subdigital pads are held hyperextended during locomotion on sand to prevent clogging of the setal fields (Bauer & Russell 1991).

The burrowing geckos carry this morphotypic trend further: Colopus and Palmatogecko vanzyli show nearly complete pad loss, and Chondrodactylus and Palmatogecko rangei are entirely padless (figure 1). As in terrestrial Pachydactylus, both Colopus and Palmatogecko vanzyli employ hyperextension to hold their reduced digital pads perpendicular to sandy substrates during locomotion (Haacke 1976a, 1976b). The distal phalanges of Chondrodactylus are extremely reduced (figure 1c₂) and permanently hyperextended (Russell 1976). Thus, the burrowing geckos culminate the morphotypic series towards pad loss in secondary terrestriality among independent lineages of the southern Pachydactylus Group.

Pachydactylus is an old (17–21Myr ago) gekkonid lineage (Lamb & Bauer 2002; Carranza et al. 2002), and opportunities for ecological shifts from arboreal to terrestrial adaptive zones in southern Africa date to the Middle Miocene (Tyson & Partridge 2000). The subcontinent's mesic woodlands of the Lower Miocene (Bamford 2000) experienced substantive climatic change with the inception of the Benguela Current, which has drastically reduced precipitation along southwest Africa since the Middle Miocene (Siesser 1980). The marked reduction in Atlantic moisture also promoted aridity across central southern Africa, particularly in the Kalahari (Stokes et al. 1997). Attenuation of woodland habitat under increasingly xeric conditions may account for the prevalence of rupicoly in Pachydactylus.

Initial formations of the Namib and Kalahari dunes were contemporaneous with the subcontinent's palaeoclimatic change. Kalahari sands date from the Middle Miocene to Pleistocene (Stokes et al. 1997), and deposits underlying modern dune systems in the Namib represent episodic accumulations since the Miocene (Pickford et al. 1995; Lancaster 2000). Protracted aridity and recurrent dune formation are considered key selective agents in the evolution of Namib psammophilous taxa (Robinson & Seely 1980; Seely & Griffin 1986), including endemic lacertid and gerrhosaurid lizards (Lamb & Bauer 2003; Lamb et al. 2003). Southwest Africa's palaeoaridity apparently provided novel opportunities for Pachydactylus as well, allowing ancestral climbing lineages to exploit progressively extreme desert habitats.

Identifying the environmental ties of a certain morphological feature is not always straightforward, as relevant parameters may include biotic variables, abiotic variables, or combinations thereof. Colopus, Chondrodactylus and Palmatogecko all inhabit desert sands, where the interaction between a particular, harsh physical environment and locomotor performance is readily apparent. Gekkotan toepad design is such that setae are passively self-cleaning with respect to small particles (Hansen & Autumn 2005), but mean sand grain size in the Namib, Kalahari and other deserts far exceeds this practical upper size limit (Bauer & Russell 1991). As noted previously, sand's clogging effect on setal fields provides a cogent mechanistic explanation for both lamellar reduction and sustained digital hyperextension during locomotion. The loss of toe pad function notwithstanding, independent gains of digital form in sand dune environments may more persuasively convey convergence. To wit, all four burrowing geckos possess spinose scales on the ventral aspect of the digits and/or plantar surfaces of the feet (figure 1); these scales are short, erect structures, ‘thereby acquiring appearance of a pyramid with a hexagonal base’ (Haacke 1976d). This peculiar scale condition does not occur in the other terrestrial Pachydactylus, including respective sister species to Colopus and Palmatogecko.

The spinose scales play a functional role in burrowing as well as locomotion and are best understood in the context of the geckos' microhabitat preferences. None of the burrowing geckos are dune slipface specialists (though Palmatogecko rangei can readily traverse slipfaces). They tend to use other microhabitats: Palmatogecko rangei frequents windward sides of dunes; Palmatogecko vanzyli (formerly Kaokogecko vanzyli) occupies dry river courses and sandy gravel plains adjacent dunes; Chondrodactylus angulifer and Colopus wahlbergii prefer sandy flats or interdune valleys (Haacke 1976a,–c). The substrate is generally more compact in these microhabitats, and the spinose ventral scales function as wedges to loosen consolidated sand during excavation (Haacke 1976a,–c; Bauer & Russell 1991). The digging kinematics in Palmatogecko rangei—despite the presence of spinose plantar scales—is an altogether different approach involving interdigital webbing. The contour of the webbing's edge acts as a spade to break the sand surface while the webbing itself forms an efficient scoop (Russell & Bauer 1990).

Russell (1979) argued that the spinose scales also serve to exclude sand grains, and measurements of interspine distances for the burrowing geckos are sufficiently narrow to exclude sand grains of local modal size (Bauer & Russell 1991). Experimental evidence suggests that such exclusion prevents clogging of the plantar surfaces, particularly in moist sand (Bauer & Russell 1991). Nocturnal advective fogs routinely moisten the Namib (Robinson & Seely 1980) and are coincident with activity periods of the burrowing geckos (Haacke 1976d).

Reduced toe pads and spinose digital scales characterize two other psammophilous geckos: Tarentola chazaliae (formerly the monotypic Geckonia), in the northern Pachydactylus Group (Carranza et al. 2002), and Diplodactylus damaeus, an Australian diplodactylid (Russell 1979). Subdigital spinose scales have also evolved in primitively padless psammophilous geckos inhabiting coastal deserts with predictable precipitation (Bauer & Russell 1991).

The burrowing geckos share additional characters with other terrestrial geckos, including a cylindrical body, long slender limbs, elevated posture and restriction of autotomy to the tail base (Werner & Broza 1969; Haacke 1976d; Bauer & Russell 1995). All of these traits are largely uncoupled from phylogenetic history, both within the Pachydactylus Group and among geckos in general. Thus, their repeated evolution (pedal traits, particularly) in similar dune environments corroborates selection's role in shaping observed morphological similarities. Further, the straightforward nature of the presumed selective agent, i.e. physical attributes of sandy substrates, provides a clear, direct association with pedal performance and, we argue, ecomorphological convergence among the burrowing geckos.

The ultimate goal of ecomorphological assessment involves recovering the precise evolutionary path that confers parallelism or convergence. Remarkably, recent population genetic studies have identified alleles at specific loci responsible for parallel character evolution in fruit flies (Gompel & Carroll 2003; Sucena et al. 2003) and stickleback fish (Colosimo et al. 2005), both model organisms whose genomic databases allow mapping such genes. Comparable genomic data are currently unavailable for geckos. Nonetheless, it is worth noting that the multiple origins of gekkotan subdigital pads trace to a common developmental transition of the outer epidermal (Oberhaütchen) layer, whose minute rugosities aid in skin shedding (Maderson 1970; Russell 2002). This layer is distinctly elaborate in geckos, bearing numerous sculpted spinules (Maderson 1970), and setae have been determined to be homologous Oberhaütchen outgrowths, exapted for the novel role of adhesion (Russell 2002). As such, lamellar formation likely entails a common developmental pathway that, if under regulatory control of one or few genes, may be readily, repeatedly reversed.

We thank authorities from Namibia, South Africa, and Zimbabwe for permits to collect and export specimens examined in this study. We gratefully acknowledge R. D. Babb, W. R. Branch, D. G. Broadley, D. Good, J. Marais and P. E. Moler for assistance in the field. M. Barts, M. Griffin and G. Watkins-Colwell kindly provided additional specimens. J. E. Bond assisted with analytical procedures, and R. D. Babb rendered the line drawings in figure 1. Comments from J. E. Bond, K. Summers, J. Stiller and two anonymous reviewers greatly improved the manuscript. Research was supported by National Science Foundation (NSF) grant DEB-9707568.