Phylogeny, scaling, and the generation of extreme forces in trap-jaw ants

Arthropods are renowned for their morphological variation, and many species have evolved extreme mechanical abilities for a variety of functions such as the remarkable jumping ability of fleas, and the crushing strikes of stomatopods (Bennet-Clark and Lucey,1967; Patek et al.,2004). These extreme speeds and accelerations are often achieved with the help of specific innovations such as latches, lever arms, and spring mechanisms that help store and release high amounts of energy(Gronenberg, 1996a). It has been argued that these abilities are optimized in such a way that tradeoffs between mechanical abilities (benefits) and physiological requirements for maintaining them (costs) are balanced against the greatest required performance for that feature (Weibel and Taylor, 1998). But in nature, where animals evolve in response to a variety of selective pressures in a changing environment, optimal performance in any one context may be constrained by physical laws,developmental programs and phylogenetic history. This is particularly true if an adaptive feature or mechanism has multiple functions – optimizing it for one function may result in sub-optimal performance in another, or competing demands may leave performance in an intermediate range for a plurality of functions.

The jaw strikes of trap-jaw ants were characterized morphologically and neurobiologically in a series of papers by Gronenberg and colleagues(Gronenberg, 1995a; Gronenberg, 1995b; Gronenberg, 1996b; Gronenberg and Tautz, 1994; Just and Gronenberg, 1999) and jaw strikes of the species Odontomachus bauri Emery 1982 can reach extremely high speeds, of over 60 m s^–1(Patek et al., 2006). Beyond providing the ants with the ability to disable prey, the jaw snaps have been evolutionarily co-opted for ballistic locomotion. It has long been known that trap-jaw ants jump (Wheeler,1922), but only recently has the way they use their jaws to do so been characterized. These movements take the forms of `bouncer defense' jumps(Carlin and Gladstein, 1989),where the ants are propelled horizontally away from a threat, and `escape jumps', where the jaws are placed against or aimed at the substrate then fired, launching the ant into the air upon triggering(Patek et al., 2006). However, O. bauri is just one of approximately 60 species in the genus Odontomachus, and while all members of the genus share the same general trap-jaw morphology, there are morphological and ecological differences between species that provide the basis for comparative study.

Across the pantropically distributed genus Odontomachus, species vary considerably in their ecology (Deyrup and Cover, 2004), including nest site substrates and types of prey, as well as varying morphologically, covering a range of body sizes and mandible lengths. These differences suggest that there may be variation in the performance of the strikes among species (perhaps based on speed or chemical defenses of common prey, or the relative advantage of jumping ability in nests or foraging areas with different physical characteristics) and may provide insight into the co-option of the mandibles for locomotion as well as prey capture. Thus multi-species comparisons are informative for characterizing trap-jaw morphology and performance and, more generally, for understanding how a multi-functional system may be optimized, or constrained, relative to its various functions.

The goals of this study were to: (1) collect kinematic and morphometric data for eight species of the trap-jaw ant genus Odontomachus; (2)construct a phylogenetic hypothesis for these species; and (3) generate a model for force production based on phylogenetically corrected body size scaling equations, and compare this modeled range to the observed range across the eight species measured in this study.

In order to detect phylogenetic effects in our comparative data, and to correct for the problems of non-independence that can invalidate statistical comparisons between species (Felsenstein,1985), we developed a phylogenetic hypothesis for a sampling of species, including the eight species for which we collected strike data. We generated sequence data for 19 species including 12 Odontomachus and seven outgroup taxa from other ponerine genera. Portions of four genes were used: the mitochondrial gene for cytochrome oxidase 1 (COI), the large subunit (28S) ribosomal RNA gene, and the nuclear protein-encoding genes wingless (wg) and long-wavelength rhodopsin (LWR). Primer information is provided in Table 1. A variable-length intron in the sequenced section of rhodopsin proved difficult to align among the outgroup taxa and was included only for the Anochetusand Odontomachus species. After excluding 29 bp of ambiguously aligned 28S data we were left with 2685 bp of aligned, concatenated sequence data. Taxon information and GenBank accession numbers are given in Table 2. Final deposition of molecular voucher specimens used in this study (currently held in personal collection of C.A.S.) will be in the United States Museum of Natural History(Smithsonian).

Genomic DNA was extracted from one or two legs of a single adult specimen for each taxon, using the DNEasy Tissue Kit (Qiagen Inc., Valencia,California, USA). PCR amplification generally consisted of 40 cycles of 20 s at 94°C, 20 s at 48°–54°C (typically 48°C for 28S and 50–54°C for the other genes), and 50 s at 65°C, with an initial denaturation of 2 min at 94°C and a final extension of 3 min at 65°C. For most amplifications a total reaction volume of 20 μl was used,including 1 unit of HotMaster Taq (Eppendorf AG, Hamburg, Germany), 0.16 mmol l^–1 dNTP mix (Eppendorf AG, Hamburg Germany), 0.5 μmol l^–1 each primer, and 1 or 2 μl of DNA template. PCR products were cleaned and sequenced by the GATC core sequencing facility on the University of Arizona campus. Sequences were aligned manually in MacClade 4.08 (Maddison and Maddison,2005).

Phylogenetic analysis was conducted using a partitioned Bayesian approach in MrBayes 3.1.2 (Huelsenbeck and Ronquist, 2001). Protein-coding genes were partitioned by gene and codon position, with one partition for 28S and an additional partition for the rhodopsin intron (giving 11 total partitions). An exploratory MrBayes analysis was performed in which each partition was given a GTR+I+G model (nst=6, rates=invgamma) and all parameters were unlinked across partitions. Examination of the resulting parameter sampling in Tracer 1.3(Rambaut and Drummond, 2004)suggested the adequacy of a reduced model: GTR+I (nst=6, rates=propinv) for first and second codon positions and for 28S, HKY+I (nst=2, rates=propinv) for third codon positions of wingless and rhodopsin and for the rhodopsin intron, and HKY+I+G (nst=2, rates=invgamma) for third codon positions of cytochrome oxidase 1. A final analysis was performed using this modeling scheme, with variable rates across partitions (prset ratepr=variable) and all other priors left at program defaults.

Two simultaneous independent analyses were run, each with four chains and the default heating value, for a total of five million generations. The consensus tree was generated using the sumt command in MrBayes with a burn-in of one million generations, chosen post hoc after examination of parameter convergence in Tracer. Chain mixing was adequate and all parameters(including tree topology) converged rapidly. Equivalent analyses were performed on the mitochondrial and the nuclear data alone to compare results from single genome partitions.

Although the data partitioning and modeling scheme employed in this analysis is probably overparameterized, Bayesian phylogenetic inference is more robust to overparameterization than underparameterization(Huelsenbeck and Rannala,2004). In addition, the resulting topology was consistent with,though not identical to, the topology obtained by a Bayesian reversible-jump mixture model analysis of the same data set using BayesPhylogenies(Pagel and Meade, 2004), which employed two GTR + G models and no partitioning.

We collected colonies of eight species of Odontomachusrepresenting a range of body sizes and ecologies(Table 3), all of which were also included in the phylogenetic analysis. Colonies were maintained in the lab and fed a diet of mealworms, waxworms or crickets, three times per week. All data were collected as described below with the exception of O. bauri data, which were adopted from Patek et al.(Patek et al., 2006) without reanalysis and included with the other seven for comparison.

The protocol for filming of trap-jaw strikes was modified from Patek et al.(Patek et al., 2006), using a high-speed camera attached to a microscope (70 000–100 000 frames s^–1, 2–11 μs shutter speed; Ultima APX Photron, San Diego, CA, USA; Leica MZ 12.5 stereomicroscope). Ants were fixed using a drop of paraffin wax (applied to the top of the head) to the end of a rod that could be rotated to keep the jaws perpendicular to the camera's axis. Animals were hung by this translating rod in an empty beaker and stimulated to strike by touching their `trigger hairs' with a thin metal probe of known diameter(0.24 mm).

The kinematic data were used to calculate speed, acceleration and the lag time (if any) between the first jaw to close and the second. Custom software developed by the authors (available as Supplemental Items S1, S2 and S3 at http://www.life.uiuc.edu/suarez/datasets.html)in MATLAB (v. R2007a, Mathworks, Natick, MA, USA) was used to track the mandible movements and calculate their angular and tangential speeds and accelerations. An optimization technique was used whereby the root mean square(RMS) error was minimized with reference to the coordinates of the center of rotation and the size of each mandible. The mathematical challenge was to fit a circle to a sequence of traced points; the circle would be the mandible tip trajectory, and its center would be the center of rotation of the movement.

The code was composed of two parts: the first traced the paths of the jaws,and the second calculated speed and acceleration. Information from jaw-snap films was input into the tracing module, including resolution (in dpi), size(width and height in pixels), frame rate (frames s^–1) and the magnification factor of the microscope. Then, each frame in the sequence was displayed as a MATLAB figure and the position of the mandible tips was recorded in each frame. Also, the approximate position of the mandible base was recorded for the first and last frames. These data were then stored as two matrices of coordinates, one containing the mandible tip coordinates for each frame and the other containing the mandible base location; the latter were x and y coordinates averaged from the two sets. The data were then saved and loaded into the calculations program.

Measurement error for digitization of strikes was estimating by retracking and recalculating a representative two-jaw strike from each of the frame rates used (70 000 frames s^–1, 90 000 frames s^–1,and 100 000 frames s^–1) five times, yielding a total of 12 single-jaw strikes for each re-tracked video segment. Percentage difference from the mean was then averaged across all 12 strikes at each frame rate.

Differentiation of point-tracking data to produce velocity and acceleration values has been considered problematic, particularly for acceleration data, as it requires second order differentiation and is likely to amplify tracking error (Walker, 1998). Subsequent `choosing' of points of greatest acceleration, as we have done here, might be expected to systematically overestimate mean maximum acceleration values. We evaluated four combinations of methods for alternative calculation of maximum velocity and maximum acceleration of a subset of the data to determine whether our results could be improved by filtering. Both cubic and quintic splines were fitted to the data, and tracking sequences were differentiated using both two-point (the control, or baseline differentiation method) and three-point differentiation methods, yielding six means (linear,cubic and quintic spline fits, each with two differentiation methods). We chose to use unfiltered, two-point differentiated data, as the spline-fit data tended to slightly overestimate maxima, which did not solve our overestimation problem, and the three-point differentiations resulted in unrealistically low estimates (as much as 31% less) for acceleration, whether or not a spline curve was fitted to the data points. Plots comparing effects of the filtering techniques explored, can be seen in Supplemental Item 4.

We filmed four to six workers from each species and up to six strikes per worker. Total strikes recorded and analyzed per species ranged from 13 (O. chelifer) to 25 (O. cephalotes). Following jaw-snap recordings,individual worker ants were killed by freezing and stored in a –20°C freezer. To minimize changes in mass caused by drying, ants were stored in air-tight vials and all mass measurements were made within 10 days of freezing. We measured the following for each ant: whole-body mass, head length(clypeus to apex), and head width [including the eyes; after Hölldobler and Wilson (Hölldobler and Wilson,1990)]. We then dissected out the mandibles of each ant and measured them individually for mass and length. Linear measurements were made using a Semprex Micro-DRO digital stage micrometer (0.005 mm resolution;Semprex Corporation, San Diego, CA, USA) connected to a Leica MZ 12.5 stereomicroscope, and masses were measured using a UMX2 microbalance with 0.1μg resolution (Mettler-Toledo, Columbus, OH, USA).

Size measurements were log₁₀ transformed, and TFSI [test for serial independence, as specified by Abouheif(Abouheif, 1999)] analyses were performed using our phylogenetic hypothesis in the software PI v. 2.0(Reeve and Abouheif, 2003) to determine whether any of the following (log transformed) measurements showed significant phylogenetic signal: head width, jaw length, body mass and jaw mass. Similarly, values for speed, acceleration, raw and normalized force were subject to the TFSI test to determine whether further statistical tests would be influenced by statistical non-independence due to phylogeny; ANOVA and post-hoc testing were only performed on species means that did not show significant phylogenetic signal in the TFSI test.

For scaling relationships, head width was used as a proxy for body size, as it is a standard measurement in the ant literature, and is a better predictor of body mass across the subfamily Ponerinae(Kaspari and Weiser, 1999)than head length, which we verified for our test species with RMA regression using RMA for Java (Bohonak and Van der Linde, 2004; Sokal and Rohlf,1981), as r² for RMA regression of body mass vs head width=0.99, whereas r² for body mass vs head length=0.98. Except where otherwise cited, statistics were performed using Statistica software (version 6.0, StatSoft Inc., Tulsa, OK,USA), and plots were produced using Excel 2003 (Microsoft Inc., Seattle, WA,USA).

Because species values are not statistically independent, we used the method of independent contrasts(Felsenstein, 1985) as implemented in the program PDAP in the Mesquite comparative analysis package(Midford et al., 2005; Maddison and Maddison, 2006)to develop the scaling equations for jaw length, jaw mass and body mass, and to produce the regression line for angular acceleration (alpha) and head width. Continuous data for head width, jaw length, jaw mass, body mass, were log₁₀ transformed and input into PDAP along with the topology and branch length data. With this information, PDAP provides hypothetical values for ancestral nodes and normalizes them to produce contrast values. The procedure of Garland et al. (Garland et al., 1999) was implemented to produce scaling equations for size parameters and to plot angular acceleration against head width. Linear ordinary-least-squares regressions with the intercepts set to the origin were performed on the normalized contrast values to calculate the slopes for the scaling equations. Biologically meaningful intercepts for scaling equations were calculated by substituting the mean values from the root nodes (which serve as estimates of the ancestral conditions) for the independent and dependent variable from each equation, the IC-corrected slope, and solving for the intercept value. Contrast values and resulting slopes were checked using independent contrasts derived in the Macintosh program CAIC(Purvis and Rambaut,1995).

To predict values for maximum force perpendicular to the jaw surface across a range of ant sizes based solely on scaling relationships, we parameterized Eqn. 2 using the scaling equations for jaw length and jaw mass (Eqn. 3a, 3b and Eqn. 4a, 4b, see Results section) and angular acceleration, as functions of head width (Eqn. 6a, 6b). This curve was also parameterized with scaling equations produced by phylogenetically uncorrected OLS regression on the species means for comparison between force production scalings that account for phylogeny, and those that do not.

The phylogenetic hypothesis with relative branch lengths developed for all exemplar species from combined data is shown in Fig. 1. Odontomachusis monophyletic in the combined tree, with Anochetus, another genus of trap-jaw ants, as the most probable sister group. Nuclear-only and mitochondrial-only trees (not shown) differed only in the rooting of the Odontomachus clade and in the relative position of a single taxon(O. ruginodis), but the single-gene trees were supported by posterior probabilities 26% lower than our preferred, combined-data tree. Branch lengths for internal nodes in the Odontomachus clade appear to be short relative to those for terminal taxa in this group, and both topology and branch-length information from the combined-data tree were included in the subsequent comparative analyses.

Five types of strike were seen: one jaw only (left or right), and two-jaw strikes with the left jaw leading, right jaw leading, or simultaneous closure. Leading jaw was defined as the jaw that achieved maximum acceleration toward the midline first; closures were considered simultaneous when the two maxima occurred in the same video frame. No significant pattern was seen within or between species in terms of a preference for a leading jaw in two-jaw strikes(34 were left-right, 38 were right-left; χ²=0.22, P=0.64), whereas simultaneous closure occurred in seven of 79 two-jaw strikes. Single-jaw strikes appeared to favor the left side (38 left-only vs 18 right-only strikes, χ²=7.14, P=0.007). All species examined included individuals that made both leading-right and leading-left strikes, and despite low strike numbers per individual, 11 of 25 individuals exhibited both types of strike, and the remaining 14 were evenly divided between `left-dominant' and `right-dominant' individuals.

Most strikes were two-jaw closures with one jaw beginning to close one or more frames after the first. The majority of two-jaw strikes (72 of 79 total strikes) included temporal offset, but due to extremely rapid acceleration of the second jaw, the second jaw often `caught up' with the first, and the two jaws scissored past each other at or very near the midline of the ant's head. However, the jaws closed simultaneously in the remaining seven strikes, or so close to simultaneous that the tiny offset could not be resolved. Distribution in lag time between maximum jaw accelerations for all strikes can be seen in Fig. 2, with a mode of 30–40 μs. Distribution is unimodal with a long right-hand tail representing strikes with long between-jaw lags, where one jaw closes completely before the other begins to close.

Across all species, the log–log regression slopes for independent contrast values for jaw length, jaw mass and body mass against head width(Fig. 3) did not allow rejection of the null hypothesis of isometry. Jaw length (P=0.0053,mean slope=1.12, 95% confidence interval for slope=0.61–1.63, r²=0.75) scaled to the first power with body lengths,whereas the slope of ∼3 for jaw mass (P=0.0018 man slope=3.26,95% CI=2.07–4.45, r²=0.82) and body mass(P<0.001, mean slope=2.94, 95% CI=2.03–3.85, r²=0.87) indicated isometry between total body mass and head width (Fig. 2), as mass scales to the third power of linear size. The r² values were lower for jaw length than for length–mass plots (0.75 vs0.82 and 0.87 for jaw mass and body mass, respectively), indicating that across species, jaw length may be more variable than body mass and jaw mass as a function of head width.

Mean maximum radial jaw speeds differed significantly among species(Fig. 4), ranging from 35.9±7.7 m s^–1 in O. chelifer to 48.8±8.9 m s^–1 in O. clarus desertorum,bracketing previously reported values for O. bauri. The three fastest-striking species (O. haematodus, O. clarus desertorum and O. erythrocephalus) differed significantly from the three slowest(O. ruginodis, O. cephalotes and O. chelifer), whereas O. brunneus did not differ significantly from either of these groups. Mean maximum jaw speed did not correlate significantly with head width or body mass (P>0.05), and phylogenetic signal was not significant for jaw speed using the TFSI test (P=0.21). Measurement error due to digitizing for speeds and accelerations averaged ±6% and ±11%,respectively.

Re-running the TSFI analysis following calculation of independent contrast results showed that phylogenetic signal was no longer significant when independent contrast values for acceleration were used (P=0.27).

Based on species means for maximum acceleration, jaw length and jaw mass,mean maximum single-jaw forces (Fig. 6A) ranged from 22±8 mN in O. haematodus to 85±26 mN in O. chelifer.

Predicting force production based on a single scaling parameter (head width) yielded a curve showing maximum single-jaw force production increasing a range of head widths. This was done by parameterizing Eqn. 2 by substituting Eqns 3a, 4a and 6a for M, R and α. Plotting model predictions for a range of head widths yielded the curve shown in Fig. 6B, with maximum jaw force continually increasing as a function of head width.

General predictions from modeling force production based on scaling equations (both phylogenetically corrected and uncorrected) were then compared to the forces estimated from the original species data(Fig. 7). Comparing species'measurement-based maximum force values with general size-based model predictions (as in Fig. 6B)showed a mean absolute difference of 12% when phylogeny was accounted for, and 11% when compared to the phylogenetically uncorrected model(Fig. 7). Size-based force predictions from phylogenetically corrected scalings differed from those made with uncorrected (or `star phylogeny') scalings by an average of 3%.

The kinematic data presented here show a large range of jaw force-generation abilities in the genus Odontomachus: a nearly fourfold difference between the largest and smallest species, scaling with a related range of sizes but varying considerably (±12%) from strictly size-based expectations. In the context of the phylogeny, this variation gives us clues to which features may be most evolutionarily labile, giving rise to relatively high- and low-force producing species. Comparing phylogenetically corrected and star-phylogeny models suggests that differences in performance relative to the model may be due to relatively recent selection pressures.

As in previous work on O. bauri(Gronenberg and Tautz, 1994; Patek et al., 2006), both mandibles rarely snapped shut synchronously. The lag between jaws followed the same general pattern previously demonstrated by Patek et al.(Patek et al., 2006), where lag time between individual pairs averaged ∼40 μs; however, the synchronous closing in a small number of snaps (seven total strikes, in three species – O. haematodus, O. clarus desertorum and O. cephalotes) suggests the time-lag does not represent a minimum time for neural conduction from one mandible to the other. It is possible that the `no lag' strikes are triggered differently from the strikes exhibiting the lag,perhaps by having both jaws stimulated simultaneously, assuming that most strikes result from a stimulation of the trigger-hairs on one side of the cocked mandibles and require conduction to the other jaw for firing of both jaws.

Alternately, there may be an adaptive explanation for a lag between mandibles if temporally off-set strikes either help prevent damage to the jaws if the target is missed or create greater force at impact with the second mandible as the target gets displaced towards the midline by the first. Jaw lag might also be expected to contribute to the jump trajectories of the ants,possibly introducing a rotation about the ants' head-to-vent axis, or tending to throw the animal sideways rather than vertically. However, without a model that translates jaw speed and acceleration into jump performance, and video data from jump sequences that can resolve distances, angles and speeds of individual jaws as they contact substrates during the acceleration phase of jumps, this hypothesis cannot be tested. The lack of such a model also limits our ability to make predictions about jump performance with the current dataset, as existing models for jumping are based on acceleration during extension of jointed legs (e.g. Alexander,1995) rather than rapid rotation of opposing fixed-end jaws against a substrate.

Morphological variation across the eight species of Odontomachusexamined here showed the simplest pattern of differentiation(Wheeler, 1991; Wilson, 1953), where worker variation follows a continuous, linear isometric or allometric curve. Without a larger sampling regime, it is impossible to reject among-species variation along slopes that conform to the simplest submodel of continuous linear variation, that of isometry. This contrasts with the allometric, clearly differentiated morphological castes(Wilson, 1976), found in some species of polymorphic ants.

Under any scaling model, maximum force, as a product of jaw mass, jaw length and angular acceleration, would be tightly linked to mandible mass and length. As seen here, in even the slowest-accelerating Odontomachusexamined (O. chelifer), large values for mass and length compensated for reduced acceleration, resulting in a fourfold greater force generation than seen in the smallest species (O. ruginodis), despite the latter having the highest mean maximum angular acceleration of the species studied. Species mean values generally track model predictions well, with variation from model predictions falling within standard deviations for all eight species. Despite the positive relationship between maximum force and size,there appears to be no clustering of species at the high end of the range of sizes seen, nor is there any obvious trend toward larger size in more derived species in the phylogeny.

It is worth noting that when not performing full lock-and-release strikes, Odontomachus ants have been shown to have some of the slowest jaw movements of any ants, as their adductor muscles, though quite large, are composed almost entirely of long-sarcomere, slow-contracting fibers(Gronenberg et al., 1997). Most ants have a mixture of long- and short-sarcomere fibers in their jaw adductors, and their jaw movements may be five to ten times faster than non-power-amplified Odontomachus jaw closures(Gronenberg et al., 1997; Paul and Gronenberg, 2002). The low speeds of normal jaw movements do not appear to be a problem for these ants, as the workers are generally monomorphic and can perform all nest tasks(carrying food, moving larvae and eggs, moving nesting materials) using their oversized, slow-contracting jaws.

Of the species studied, O. chelifer is the clear champion in terms of force production (Fig. 6A). Laboratory observations (A.V.S. and J.C.S., unpublished data) show that these robust ants do indeed deliver devastating strikes, such that they seldom, if ever, use their stings in attacks on prey animals – a single strike is usually enough to disable the prey item. This is in contrast to smaller species, which generally strike and subsequently sting to disable prey.

With continuous, log-linear size variation and multiple species with workers considerably smaller than the largest Odontomachus species,it appears that optimal size for a particular species is not dependent on maximum force production. More likely, in such an isometric context, maximum size is balanced against the developmental and physiological costs of growing and carrying (and loading) oversized adductor muscles and jaws. Alternative hypotheses need to be examined including those relating to `optimal speeds'for capturing elusive prey such as springtails(Brown and Wilson, 1959), or`ecological release' relative to jaw performance – wherein there is no natural enemy or prey item requiring such extreme speed or force production,so that individual size is determined by other selective pressures, such as food availability or optimal size relative to available nesting sites. In other ant lineages where trap-jaw morphologies have evolved independently,including taxa in the Myrmicinae(Gronenberg, 1996b) and Formicinae (Moffett, 1985)subfamilies, we might expect to see similar species-scaled differences in performance, although isometric scaling cannot be assumed for these.

Although the workers of most Odontomachus species show little variation in size within a single colony, some species do have workers within a colony that exhibit a range of sizes (e.g. O. cephalotes, a Northern Australian species). Detailed study of species with broad intra-specific distribution of worker sizes, including characterization of behavior of individuals by size and age, will help determine how trap-jaw phenotypes are tuned by the social environment, development and evolutionary history. More generally, greater within-species sampling and narrowly focused study of species that may deviate from the log-linear relations presented here will be valuable in understanding the selective pressures contributing to diversity (Biewener, 2003) in trap-jaw ants.

The predictions of this paper should also be tested via direct measurements of force production across these (and other) Odontomachus species. The behavioral ecology, including prey and natural enemy types, and relative frequency and ecological correlates of jaw usage (jumps vs strikes) remains largely unknown, and may help explain the preponderance of relatively small species.

The Odontomachus phylogeny developed here, with its relatively short internal branches, suggests the possibility that this genus diversified quickly, with fewer subsequent speciation events following an original radiation, or an increase in extinction rate leaving relatively long terminal branches. Alternatively, our sampling regime may have been broad enough and evenly distributed enough to create relatively long branches as an artifact. In either case, it approximates the `star phylogeny' assumed in use of non-phylogenetically corrected species data, and is less likely to be confounded by an uneven distribution of recently and less-recently diverged species (Garland et al., 1999; Price, 1997). Despite this,the results of the TFSI tests demonstrated significant phylogenetic signal in key parameters expected to influence force production, particularly jaw acceleration, arguing for incorporation of statistical methods correcting for phylogeny.

We found only small differences between jaw-strike forces predicted by the phylogenetically corrected and uncorrected models for force production. However, there is still significant value added when the data are viewed in the context of the phylogeny, both from first principles and in terms of the quality of results for purposes of additional hypothesis generation and testing. First, without a phylogeny, there is no a priori way to know what the effect of accounting for branching patterns and branch lengths would be, and the assumption that it will not influence the outcome has been shown to be incorrect in numerous studies (e.g. Nunn and Barton, 2000; Zani, 2000; Smith and Cheverud, 2004). Second, given that the data appear to contain phylogenetic signal according to the TFSI tests, but that accounting for that signal does not necessarily improve predictions of force-generation performance for the actual terminal taxa, we can make inferences about evolution of the trap-jaw system that would otherwise be difficult to support. In the present study, the situation where phylogenetic signal may exist but does not account for the differences in performance between taxa may be a case like that presented by Price(Price, 1997) where a variable character has been under recent selection in the individual species'environments, and the contrast data, representing relatively deep divergences,can be overwhelmed by recent adjustments in the character – in this case, body size, with performance scaling in simple isometry with changes in size.

The authors thank Brian Fisher of the California Academy of Sciences, Chris Smith of University of Illinois and Mark Deyrup of the Archbold Field Station for assistance in collecting and maintaining ant colonies, and Joe Baio and Tawny Mata for assistance with filming strikes and data processing. We also thank the following individuals for loan of specimens for phylogenetic analyses: Lloyd Davis, David Donoso, Kim Franklin, Jurgen Liebig, David Maddison, John Mangold, Wendy Moore, Maruyama Munetoshi, Chris Smith and Alex Wild. For permission to collect and import ants, we thank the Ministry of Environment and Energy (Permit 122-2004-OFAU) of Costa Rica, the Ministerio de Salud y Ambiente (Permit 20202/05) of Argentina, the Administracion de Parques Nacionales (Permit 002870-2) of Argentina, James Cook University, Australia for the loan of O. cephalotes specimens and the United States Department of Agriculture (APHIS import permit 69963). This work was supported by a seed grant from the Beckman Institute for Advanced Science and Technology.