Is a convergently derived muscle-activity pattern driving novel raking behaviours in teleost fishes?

The evolution of feeding strategies among aquatic vertebrates is characterised by repeated diversification of prey capture and processing behaviours possibly with accompanying shifts in the muscle-activity patterns governing such strategies. In spite of these repeated trends and the resulting intense focus on muscle-activity patterns of fish feeding, convergently derived muscle-activity patterns (MAPs) have never been documented(Wainwright and Friel, 2001; Wainwright, 2002). Experimental analyses of MAPs in basal vertebrate aquatic feeding have focused on prey-capture behaviours, whereas MAP variability across an extensive range of prey-processing behaviours remains poorly quantified(Grubich and Wainwright, 1997; Grubich, 2001). MAP evolution across prey-capture modes and among feeding behaviours remains enigmatic,despite extensive comparisons of suction-feeding taxa(Grubich, 2001), biting taxa(Friel and Wainwright, 1997; Friel and Wainwright, 1998; Alfaro and Westneat, 1999; Wainwright and Friel, 2000),and biters with ram-suction feeders(Wainwright et al., 1989; Alfaro et al., 2001). Strike MAPs are often highly variable (Wainwright et al., 1989; Alfaro et al.,2001), both at high phylogenetic (among species and above) and at phenotypic levels (among individuals).

Prey processing in a pharyngeal jaw apparatus is by far the best-studied teleost prey-processing mechanism, which has decoupled the oral jaws from mastication tasks and has convergently evolved in several derived(euteleostean) lineages (Sibbing et al.,1986; Wainwright,1989a; Wainwright, 1998b; Wainwright, 2005; Grubich and Westneat, 2006). By contrast, functionally coupled feeding mechanisms, although rare, may govern several behaviours without constraining functional versatility, such as prey capture and processing behaviours governed by the oral jaw apparatus in tetraodontids (Wainwright and Turingan,1993; Van Wassenbergh et al.,2007b). Coupled mechanisms may thus, in contrast with decoupled mechanisms, be useful functional systems for empirical tests of the evolutionary role that MAPs play during functional shifts, ultimately resulting in behavioural–ecological diversification(Alfaro and Herrel, 2001; Wainwright, 2002).

Intra-oral prey-processing mechanisms involving the hyolingual system represent such a coupled system, retaining close musculoskeletal associations with the oral jaw apparatus whilst occurring episodically across the actinopterygian radiation (Lauder,1980; Lauder and Liem,1983). The use of clearly distinguishable hyolingual prey-processing behaviours to reduce and immobilise prey make salmonid and osteoglossomorph fishes a particularly interesting case study(Sanford and Lauder, 1989; Sanford, 2001b). A novel raking behaviour, employed by all but a few members in both lineages, is governed by a novel tongue-bite apparatus (TBA), formed by opposing basihyal and oral cavity roof dentition intercalated between the oral and pharyngeal jaws (Sanford, 2001a; Sanford, 2001b; Hilton, 2001; Hilton, 2003). In addition to raking, cyclic intra-oral chewing behaviours are also used by osteoglossomorphs (Sanford and Lauder,1989; Sanford and Lauder,1990; Frost and Sanford,1999) salmonids (Sanford,2001b) and non-raking neoteleosts(Lauder, 1981).

In earlier studies, raking kinematic data from a phylogenetically diverse range of representatives from both known raking lineages were used to describe two-dimensional biomechanics (Sanford,2001a; Sanford,2001b). These mechanisms appear to involve (1) two musculoskeletal pathways driving the raking preparatory phase, and (2) three pathways driving the subsequent raking power stroke. Set in a combined model, these mechanisms predict how a raking preparatory phase `primes' the TBA via basihyal protraction, caused by the protractor hyoideus (PH), or the posterior intermandibularis (PIM) in notopterid osteoglossomorphs (see Greenwood and Lauder, 1981; Sanford and Lauder, 1989). Concurrently, the adductor mandibularis (AM) causes mandible adduction thereby immobilising prey caught inside the mouth(Sanford and Lauder,1989).

The raking power stroke is initiated with epaxial (EP)-driven cranial elevation, protracting and elevating the TBA upper jaw. Concomitantly, the basihyal (TBA lower jaw) is retracted either directly, by the sternohyoideus(SH) or indirectly, by hypaxial (HP) strain, transmitted via a novel cleithrobranchial ligament (CBL) (Sanford,2001a; Sanford,2001b; Hilton,2001). The cleithrobranchial ligament connects the pectoral girdle and basihyal, in parallel with the SH muscle(Fig. 1), and may functionally decouple this muscle during raking(Sanford and Lauder, 1989; Sanford, 2001b), while also possibly influencing more generalised strike and chewing feeding behaviours in these same taxa (Liem, 1970; Lauder, 1980; Sanford and Lauder, 1989; Sanford and Lauder, 1990; Sanford, 2001a; Sanford, 2001b). Interestingly, the SH exhibited a labile feeding MAP compared with four other feeding muscles in the osteoglossomorph Chitala, a pattern that extends widely across aquatic feeding vertebrates(Sanford and Lauder, 1989). The combination of a complex TBA morphology and EMG data suggesting a labile MAP emphasises the importance of conducting a direct functional evaluation of the SH–CBL complex.

Video kinematics can provide indirect evidence of isotonic muscle contraction in simple lever systems, such as neurocranial elevation, where only one muscle (epaxial) is functionally capable of causing the movement(Sanford and Lauder, 1989; Carroll, 2004). Deeper, more complex systems require direct recording of muscle activity patterns like the TBA, where anteroventral basihyal protractors (PH/PIM) may antagonise chain-linked basihyal retractor synergists (SH and HP) situated further posteriorly (Fig. 1). The lack of a protractor pectoralis muscles in raking taxa(Greenwood and Lauder, 1981)and the presence of a CBL further increases the system complexity.

Our aim was to provide a first quantitative analysis of feeding behavioural MAPs in representative salmonid and osteoglossomorph taxa where TBA diagnostics and raking involving comparable TBA biomechanics are present, but no quantitative MAP evidence exist, and where size-matched specimens were obtainable, namely the rainbow trout, Oncorhynchus mykiss and the Australian arowana, Scleropages jardinii. We clarify the specific roles of primary TBA muscles, quantify the MAPs of raking and chewing,alternative forms of intra-oral prey processing, and compare these with strike MAPs.

We were interested in the degree of difference and stereotypy in raking MAPs, relative to other feeding behaviours within and between lineages. Specifically, we examined O. mykiss to determine whether (1) salmonid raking MAPs were different from that of strike and chew MAPs? Comparing raking MAPs from O. mykiss and S. jardinii, we then investigated whether (2) raking MAPs differ between these lineage representatives and (3)if raking may be convergently derived from ancestral feeding patterns? These analyses quantify how variability of MAPs during raking compares to changes at other organisational levels, including morphology and kinematics. A notion exists in the literature that raking kinematics are comparable to a closed-mouth strike (Sanford and Lauder,1990; Sanford,2001b). Following this notion, we hypothesise that temporal muscle-activity variables that distinguish raking from more general feeding behaviours such as the strike are convergently derived in both of the known raking lineages.

Size-matched rainbow trout [Oncorhynchus mykiss Walbaum; N=5, head lengths (HL)=50–59 mm, mean ± s.e.m.,55±0.894] were obtained from the Cold Spring Harbour fish hatchery,Long Island New York. Australian arowanas [Scleropages jardinii(Saville-Kent); N=5, HL=52–61 mm; mean 55±0.853] of a similar size (t-test; P=0.948) were purchased from local aquarium stores. All specimens were housed individually in the Hofstra University animal care facility, in accordance with applicable ethics approvals and animal-care protocols. Trout were kept at 15±2°C and arowana at 27±2°C, both in a 12 h:12 h light:dark photoperiod, as per previous studies (Ojanguren and Brana,2000; Sanford,1990; Sanford,2001b). During acclimation, individuals were fed a varied diet of crickets (Gryllus sp.), goldfish (Carassius sp.), earthworms(Lumbricus sp.) and minnows (Pimephales sp.) to avoid habituation. Experiments were commenced when individuals were well acclimated and fed vigorously under floodlight illumination with no signs of hesitation in front of the provisioner. In order to standardise the motivation level among individuals and experiments we did not feed specimens for 2 days prior to each experiment.

Electromyography (EMG) recordings were taken from five muscles that are commonly involved in vertebrate aquatic feeding(Wainwright et al., 1989; Grubich, 2001) and have been argued to be important during raking(Sanford and Lauder, 1989; Sanford, 2001a; Sanford, 2001b), namely the sternohyoideus (SH), protractor hyoideus (PH), adductor mandibularis (AM),epaxialis (EP), and hypaxialis (HP) muscles. We used a modified protocol from Wainwright et al. (Wainwright et al.,1989) and Alfaro and Westneat(Alfaro and Westneat, 1999). Hooked fine-wire electrodes were prepared by threading 1.25 m lengths of bi-filar wire (0.05 mm diameter polyethylene-coated stainless steel;California Fine Wire, Grover Beach, CA, USA) through 25 5/8 hypodermic needles. Electrodes were implanted at a 45° angle to the surface of the left side muscles of anaesthetised animals (40 p.p.m. alcoholic Eugenol, Rush& Hebble, Edinburgh, IN, USA) (Munday and Wilson, 1997). During implanting, the hooks were anchored percutaneously into the muscle bellies, parallel to the muscle fibre orientation. Electrode wires were anchored to a mid-dorsal suture on the specimen, joined with glue while the electrode connector ends were crimped into din-25 adaptor pins, and floats were taped to the electrode wires to prevent tangling of specimens. Surgery lasted ∼20 min, and specimens recovered totally ∼45 min after surgery onset.

Difficulties with obtaining size-matched specimens meant that several experiments had to be conducted on each specimen (see below). Owing to the large specimen size, muscles of interest were large, easily identifiable and clearly delineated. Thus, instead of verifying electrode placement in dissections following experiments, we conducted repeated-practice implants on freshly thawed specimens from each species, and used dissections to verify electrode placements in these. During experiments, prey, consisting of live goldfish, approximating lateromedial gape width of the specimen sampled(35–40 mm total length TL), were released with the predator,which typically caught the prey instantaneously using a rapid ram-suction feeding strike. Electrical signals produced by muscle activity during prey capture and processing were sampled at 4 kHz, amplified 1000 times (A-M systems, differential AC amplifier, model 1700, Everett, WA, USA), and conditioned with band-pass at 1000–100 Hz with a 60 Hz (notch) filter engaged. EMG signals, and a manual trigger code (+5 V) used to label the behaviours during recording, were digitized using a PowerLab 16/30 system linked to a PC running Chart v.5.4 for Windows (PowerLab, Colorado Springs,CO, USA). After each feeding trial, signals were comprehensively logged for,e.g. behaviour type, prey orientation and predator motivation, using the comments tool in Chart. In order to focus on the level of inter-lineage stereotypy in behaviourally related MAPs we used replicate events(N=5) from more individuals (N=5) than is commonly used in MAP studies, thus allowing more accurate quantification of intra-specific motor-pattern variability (e.g. Wainwright et al., 1989). Although we observed the predator feeding on several prey items aggressively before any effects of satiation were evident,we only selected video sequences from the first three feedings of each experiment for analysis.

We selected EMG signals for analysis based on evaluations of predator voracity and other relevant information retrieved during data collections. From rectified EMG signals of the five muscles sampled, we measured a total of 20 different MAP variables, which can be divided into five groups: (1) muscle activity duration (ms); (2) relative MAP onset (ms), relative to EP muscle onset time (reference muscle); (3) mean amplitude (signal intensity; in mV),corrected with the maximum recorded activity (spike) from that muscle for that implant across behaviours; and (4), integrated area (signal energy), i.e.`area under the curve', but above the signal baseline, in mV^*s,corrected with the same maximum amplitude spike; and finally, (5) total duration of behaviour i.e. the duration from onset timing of first to offset timing of last muscle activity during one behavioural bout was also quantified(Table 1) but not included in further analyses.

Recruitment variables were analysed with and without spikes and using binning of 10 ms time intervals for a subset of the data, all of which resulted in no significant differences (ANOVA; P=0.12). Thus our original data was used without removal of spikes or binning. Parametric analyses were initially performed on corrected and non-corrected variables,and integrated area was the only variable group for which the resulting PC-factor scores and component loadings did not differ. To retain potentially important biological information in the dataset, final parametric analyses used uncorrected integrated area values.

The contribution by sternohyoideus muscle contraction to feeding kinematics is poorly quantifiable using conventional (high-speed video) methods, given the medial placement of this muscle and the hyoid apparatus relative to the mandible, opercular series and suspensorium. Also, inferences about muscle strain cannot reliably be made from electromyography(Carroll, 2004). Therefore, in two O. mykiss we sutured two 2 mm piezoelectric crystals (Sonometrics Corp., London, ON, Canada) to the ventral coracoid and urohyal keels(Fig. 1), at the sternohyoideus origin and insertion ends, respectively. The keels provide an accurate measure of SH length while minimising invasive surgery. As a proxy of SH length, this configuration of crystal placement would also provide evidence of dynamic tension on the CBL exerted during activity in the PH (anterior) and HP(posterior). During EMG electrode implantation, the mid-ventral region surrounding the sternohyoideus muscle was palpated to locate the bony elements immediately under the integument, to which the crystals were sutured through integument and bone. The sturdiness of crystal attachment was verified by vigorous expansive and compressive head manipulations on the anesthetised fish. Sonomicrometry data was inspected in SonoView (Sonometrics Corp.,London, ON, Canada) and the analogue sonometric signals were simultaneously recorded on the EMG PowerLab system, where they were calibrated and converted to mm (Fig. 2).

We computed means and standard errors from all EMG variables across species-specific behaviours (trout strike, chew and rake; arowana strike and rake; Table 1), presented in box plots of species-specific behaviours(Fig. 3), but parametric analyses used derived, non-averaged data. This study aimed at answering three major questions regarding raking behaviours and the underlying muscle-activity patterns. We segregated our statistical analyses accordingly, using a parametric statistical design incorporating a principal component analyses(PCA) on the covariance matrices (eigenvalues >1), with a MANOVA on PC factor scores for the entire dataset. Subsequently, ANOVAs on factor scores for each axis and Bonferroni-corrected post-hoc tests were used to determine which behaviours, if any, differed.

Scatter plots of the informative PC axes resulting from the 19 derived MAP variables (total MAP duration not analysed) illustrated the distribution of species-specific behaviours relative to MAP similarities and differences. In biplots of informative PC axes, the most influential component loadings (for covariance matrices, values over 15), were scaled to PC axis length and superimposed in order to identify the most important variables in segregating species-specific behaviours across available multivariate MAP space.

To test the overall hypothesis that the MAPs governing the three behaviours, strike, chew and rake, are different in O. mykiss we ran a PCA to identify informative axes of variation (eigenvalues >1), followed by a MANOVA. To ascertain which behaviours differed significantly, we used ANOVAs on PC factor scores for each principal component axis followed by post-hoc tests using Bonferroni-corrected comparisons with behaviour as the fixed main effect and the interaction term (behaviour×individual)as the denominator in calculating the mean-square error(Zar, 1999). This design identifies significant PC axes and behavioural differences along them, whilst accounting for the effects caused by individual variability.

To test whether raking MAPs differ between lineages, we analysed the EMG data from O. mykiss and S. jardinii using a PCA analysis and a nested MANOVA design with species as the fixed main effect [following Zar(Zar, 1999)].

In order to determine whether raking MAPs overall are convergently derived from the remaining feeding behaviours utilised by the species studied, we ran a PCA on the covariance matrix of the complete dataset, excluding chewing data from S. jardinii, which did not reach the required sample size because of infrequent use of this behaviour. Evolutionary differences in MAPs were distinguished from significant individual and species effects, using a mixed-model MANOVA with species as a fixed, and individual within species as a random effect.

The duration of rakes (160±32 ms; mean ± s.e.m.) and chews(176±35 ms) in O. mykiss were not significantly different, and both lasted significantly shorter than the average strike (ANOVA; P<0.001) (Table 1). Statistically different MAPs characterise the three feeding behaviours(MANOVA: Wilk's λ=0.039, F_8,138=70.221, P<0.001) and a PCA on the 19 derived EMG variables from all O. mykiss feeding behaviours is summarised in Table 2. Timing and duration variables are overriding in distinguishing raking from other feeding behaviours, compared with amplitude and integrated area variables. Specifically, early protractor hyoideus (PH) and adductor mandibularis (AM)onset-timing characterises raking MAPs(Fig. 3A–C, Table 1).

The PCA recovered four informative PC axes and ANOVAs on the PC factor scores, with Bonferroni-corrected post-hoc tests run separately on each axis, showed that the raking MAP differs from strike and chewing along PC1 only, whereas the two latter behaviours were not statistically different(Table 2). A scatter plot of PC1 and 2 (Fig. 4A) illustrates that the O. mykiss raking MAP occupies a distinct region of multivariate MAP space, primarily driven by an early onset in AM and PH activity during raking, while differences in AM and PH activity durations separate strike and chew MAPs.

Raking is governed by a much more stereotypical MAP than other feeding behaviours, as illustrated by the tight cluster of raking PC factor scores(Fig. 4A), and by a very restricted variation in MAP among individuals along PC1 relative to chewing and in particular strike MAPs (Fig. 4B). The strike MAP factor scores displaced towards high positive values along PC1 belong to one individual of a statistically similar size to the four other individuals. Running analysis without strikes for that individual gave very similar percentages of variance explanation along PC axes. Duration of SH activity was the only MAP variable driving significant variation among feeding behaviours along PC3(Table 2).

Although SH mean amplitude did not influence the PCA for O. mykiss, differences between rakes and chews were evident (ANOVA, P<0.05), in correspondence with qualitative SH amplitude differences (Fig. 2) between the different behaviours (Table 2). Measurements of SH length changes obtained viasonomicrometry indicated that during rakes with low SH amplitude(Fig. 2, rake 1); the SH–CBL complex was stretched by approx. 2% of SH resting length –reliably indicating the absence of SH shortening. Meanwhile, in rakes with higher SH amplitude (Fig. 2,rake 2), initial SH stretching was subsequently recovered via SH contraction.

In contrast to O. mykiss, the mean rake duration in S. jardinii (151±30 ms; mean ± s.e.m.) was significantly longer than the mean strike duration (110±22 ms; t-test, P<0.01; Table 1). Chewing is a relatively uncommon behaviour in S. jardinii and is therefore not treated here. Raking MAPs in O. mykiss and S. jardinii were statistically different (MANOVA: Wilk's λ=0.17, F_4,5=55.119, P<0.001), and a principal component analysis recovered four PC axes, summarised in Table 3. A subsequent ANOVA on the PC factor scores (Table 3)determined that raking MAPs between the two species differ along PC1–3,and a scatter plot of axes 1 and 2 (Fig. 5) show raking behaviour separation across multivariate MAP space.

Raking in both O. mykiss and S. jardinii is characterised by early onset in PH and AM activity, while an extended duration of activity in these muscles during S. jardinii raking is the most significant discriminator of inter-specific raking MAPs. Inter-specific differences in raking behaviours are also driven by earlier onset and longer lasting activity duration in S. jardinii, of the ventral muscles driving pectoral girdle retraction (HP) and direct basihyal retraction (SH) during the raking power stroke (Table 1; Fig. 3). During S. jardinii raking, activity in these muscles is also characterised by relatively low, although not significantly different mean amplitude and integrated area means compared to O. mykiss(Fig. 3).

Rakes are shorter than strikes in O. mykiss but longer than strikes in S. jardinii (t-test, P<0.01; Table 1), whereas mean rake duration in O. mykiss and S. jardinii was very similar(t-test, P=0.403; Table 1). Mean duration of SH and HP muscle activity in S. jardinii is longer during rakes than during strikes, whereas the opposite is the case in O. mykiss (Table 1, Fig. 3). Given these inter-specific MAP-duration differences, the divergent inter-specific relationships in SH and HP muscle duration between rakes and strikes were confirmed by calculating total behaviour-duration-corrected muscle activity means, which did not differ from non-corrected means(Table 1).

All rakes investigated had a different onset profile than the more generalised feeding behaviours (strike and chew; Fig. 3). Raking was primarily driven by early mean onset times in the PH and AM muscles relative to the onset of EP activity. In more generalised behaviours both PH and AM muscles lag behind the EP. A principal component analysis on the covariance matrix of 19 transformed data variables for all inter-specific feeding behaviours analysed explained 91.5% of the total variation in the dataset. Three principal components with Eigenvalues over 2.0 were recovered, explaining 53.1%, 21.3% and 10.9% of the variation, respectively. Raking behavioural MAPs occupy a completely distinct region of multivariate space from strike and chewing behaviours, the latter polygons of which largely overlap(Fig. 6). The principal variables driving this noteworthy displacement of raking MAPs in both taxa are early onset timings of AM and PH musculature, which occurs in both taxa(Fig. 3).

Furthermore, the variables separating O. mykiss and S. jardinii raking centre on differences in activity duration in the AM and PH musculature, both lasting shorter during O. mykiss raking, while remaining active during the raking power-stroke MAP of S. jardinii. The areas of species-specific behaviour polygons indicate the relative degree of stereotypy in species-specific behaviours(Fig. 6): raking is clearly governed by a restricted degree of MAP variability in both taxa, while, for example, MAPs of trout strikes are particularly variable. However, whereas raking is particularly stereotypical in O. mykiss, early preparatory phase variation often occurs in S. jardinii, as an initial burst of activity in all muscles, prior to PH activity onset (rake B, Fig. 2).

Clear distinctions were found between the muscle activity pattern (MAP) of raking, and the more similar, and generalised, strike and chewing behaviours. The most divergent MAP traits were early activity onset during raking in the protractor hyoideus (PH) and adductor mandibularis (AM) muscles. This corresponds well with early onset of jaw adduction and basihyal protraction characteristic of the raking preparatory phase(Sanford and Lauder, 1989; Frost and Sanford, 1999; Sanford, 2001a; Sanford, 2001b). Although Sanford and Lauder (Sanford and Lauder,1989) distinguished raking from strike and chewing in the basal osteoglossomorph Chitala, MAPs of the latter two behaviours were also significantly different. However, this finding could be explained by differences in EMG protocol used by Sanford and Lauder(Sanford and Lauder, 1989). Differences in EMG protocol prevent direct comparison of our results with Chitala, as Sanford and Lauder(Sanford and Lauder, 1989)used maximum amplitudes, while omitting amplitude correction for implant. Nevertheless, it is now clear that raking MAPs in representative taxa from both known raking lineages thus far studied differ from strike and chewing MAPs.

From a functional perspective, the differences between behaviour-specific MAPs closely match the biomechanical prerequisites of the different behaviours. Raking entails preparatory, power-stroke and recovery phases(Sanford, 2001a; Sanford, 2001b), much like the basal-most vertebrate aquatic prey-capture mechanism, the suction-feeding strike that also involves such consecutive phases(Liem, 1978; Lauder, 1985; Shaffer and Lauder, 1985; Lauder and Reilly, 1994; Reilly, 1995; Motta et al., 1997). During the raking preparatory phase, PH-driven basihyal protraction `primes' the TBA,while the prey is retained and compressed in the oral cavity by AM-driven oral jaw occlusion, preventing its escape. Oral jaw occlusion may also optimise the trajectory of subsequent basihyal retraction during the power stroke(Sanford and Lauder, 1989). Conversely, early AM and PH activity onset during strike or chewing behaviours would prevent prey capture and mastication, whereas both may be involved in slow oral jaw occlusion post-strike, provided the line of action of the PH has swung dorsal to the lower jaw joint(Lauder, 1981). Certainly,this occurs in S. jardinii during raking, and maintained PH activity may thus be contributing to the firm mandible occlusion seen throughout the raking power stroke in this, and other osteoglossomorph taxa examined to date(Sanford and Lauder, 1989; Sanford, 2001a). Overall,raking contrasts with strike and chewing behaviours, both in terms of biomechanical and neuromotor physiological characteristics, whilst minimising the proximal constraints on prey acquisition and offering an alternate avenue of prey processing.

Clearly, AM and PH activity variables, whilst primarily governing the raking preparatory phase, play overriding roles both in distinguishing raking MAPs from other feeding behaviours, and also distinguishes inter-lineage raking MAPs in this study. Future analyses of raking MAPs from a broader species range should involve parametric comparisons both with and without PH and AM variables, to establish if more subtle MAP differences exist in the feeding muscles more directly involved in the raking power stroke.

Raking MAPs remained subtly distinct between O. mykiss and S. jardinii, primarily because of shorter durations of AM and PH activity in O. mykiss. The significantly longer PH activity duration in S. jardinii provides at least a partial explanation of the reduced posterior pectoral girdle excursion compared to O. mykiss. Maintained PH contraction from the preparatory phase into the power stroke(Fig. 3), i.e. overlapping with basihyal retractor musculature activity, may result in tension in this protraction component of the muscular sling suspending the basihyal, thus influencing force transmission from the power-stroke musculature to the basihyal. Alternatively, oral jaw occlusion can result in dorsally directed compressive forces of the basihyal. The prominent anterodorsal oral-cavity incline in S. jardinii means that prolonged PH tension, in synchrony with SH and HP contraction (Fig. 3), can result in a dorsally directed crushing action of the basihyal toothplate against the dorsal toothplates in S. jardinii. Interestingly, toothplate morphology in S. jardinii, more closely resembles crushing pharyngeal plates than the caniniform basihyal fangs in O. mykiss, providing morphological evidence that crushing may be the primary osteoglossid TBA function. In O. mykiss, evidence of a posteriorly directed shearing action of the basihyal fangs during raking is provided by the fact that SH and HP activity commences well after AM and PH activity, and lasts for a significantly shorter time during raking than during strikes.

Raking kinematics in S. jardinii, and likely in other osteoglossids, involves a complex preparatory phase compared to salmonids,with a bipartite pattern of activity associated with intra-oral prey manipulation prior to raking (rake B in Fig. 2). In 254 rake EMGs, from five S. jardinii, such bipartite muscle activity was not temporally distinguishable from the main rake EMG in 104 cases (41%), primarily occurring in three individuals during particularly aggressive trains of prey-processing behaviour. In general, a synchronous onset of the SH and HP early in the raking MAP of S. jardinii supports the observation that the cranium is noticeably depressed in osteoglossids during the preparatory phase(Sanford and Lauder, 1990). Also, the late onset of EP relative to SH and HP contrasts with the rake MAP in O. mykiss, suggesting that in S. jardinii basihyal compression–retraction is occurring before the cranium is elevated. Overall, this MAP complexity corresponds well with the kinematic patterns observed in S. jardinii (N.K. and C.P.J.S., unpublished data) and its close osteoglossid sister taxa during raking(Sanford and Lauder, 1990). In osteoglossids, neurocranial and pectoral girdle excursions were restricted,whilst oral jaw occlusion and hyoid adduction was more pronounced compared to existing salmonid raking data (Sanford,2001b). In osteoglossids, this unique combination of raking kinematics probably results in a composite compression and grinding action of the basihyal against the millstone-shaped oral roof. This composite action clearly diverges drastically from the unidirectional prey shredding commonly hypothesised to result from basihyal kinesis during raking in previous studies. Thus, sonomicrometry will be particularly useful when investigating the functional differences in deep basihyal kinesis across morphologically divergent raking taxa.

Inter-specific raking MAP differences include SH and HP variables,prompting the question of whether functional decoupling is linked to different architectures of a novel cleithrobranchial ligament (CBL) found in the midline, embedded in the SH muscle. Specifically, the CBL origin-insertion trajectory differs between species, with a linear CBL architecture linking the pectoral girdle to the third hypobranchial in S. jardinii (N.K. and C.P.J.S., personal observation). This CBL may be an efficient pathway for transmission of HP strain to basihyal retraction, while permitting decoupling of the SH to modulate basihyal power-stroke kinesis. Considering the labile MAP of SH muscles in basal aquatic vertebrates(Carroll, 2004; Van Wassenbergh et al.,2007a), convergent CBL formations may in fact represent evolutionary enhancements to an existing musculoskeletal pathway.

The peculiar curved CBL in O. mykiss, inserting partially onto the third hypobranchial tip and predominately onto the dorsocaudal margin of the dorsal hypohyal (N.K. and C.P.J.S., personal observation), may perform differently to the CBL of S. jardinii in transmission of HP strain. This interpretation is corroborated by our synchronous sonomicrometry and EMG recordings: manipulation of cleared and stained specimens revealed that approx. 2% SH stretching, the commonly observed SH stretch magnitude in our sonomicrometry data for O. mykiss(Fig. 2), resulted in a straight CBL, making it functionally capable of direct HP strain transmission to retract the basihyal. Among such diverse suction-feeding taxa as centrarchids (Carroll, 2004)and clariids (van Wassenbergh et al.,2005) SH stretching is known to occur during strikes, except in taxa with a massive SH (van Wassenbergh et al., 2007a). SH activity variability extends across all feeding behaviours in a broad range of basal aquatic-feeding vertebrates(Wainwright et al., 1989),basal teleosts (Lauder, 1980);scarines (Alfaro et al., 2001),caracoids (Lauder, 1981),centrarchids (Carroll, 2004; Carroll et al., 2004; Carroll and Wainwright, 2006),clariids (Van Wassenbergh et al.,2005; Wassenbergh et al., 2006; Wassenbergh et al., 2007a;Wassenbergh et al., 2007c) and balistids(Wainwright and Turingan,1993). Accordingly, a curved CBL may represent a trade-off in TBA coupler-linkage architecture to facilitate more efficient suction feeding (by allowing the SH to stretch) in salmonids compared with S. jardinii.

Understanding the nature and diversity of this potential decoupling pathway, and the relative contribution from three complementary biomechanical mechanisms to TBA function in salmonids and osteoglossomorphs will be key in future studies. These should focus on quantifying the relative contribution from morphological, mechanical and functional diversity in the convergent evolution of novel raking feeding behaviours. Synchronised use of EMG,sonomicrometry and high-speed video is a powerful tool, which will help to quantify the effect of having a CBL decoupling the SH during raking. Whatever the role of the CBL, the observation that such a ligament is present only in TBA-bearing taxa (Sanford and Lauder,1989) suggests that its role is primarily driving the control of prey-processing behaviours associated with the TBA rather than oral jaw-associated prey acquisition behaviours.

Overall, muscle activity total duration varied much less during raking than during strikes in O. mykiss, whereas mean rake durations for both species fell within a surprisingly consistent interval of 150–160±31 ms. In the multivariate analysis there was remarkably little variation in rake MAPs among O. mykiss individuals(Fig. 4B), whereas the scatter of rake EMGs across PC axes 1 and 2 was more pronounced in S. jardinii (Fig. 5). However, pronounced variation in O. mykiss strikes relative to other behaviours (Fig. 4) corresponds well with this taxon utilising a generalist prey-capture behaviour(Liem, 1980).

Clearly, a high degree of behavioural stereotypy governs raking in O. mykiss, either because of architectural properties of the cleithrobranchial ligament, or, as suggested by the highly conservative temporal MAPs, resulting from a more tightly controlled central oscillator mechanism than in S. jardinii. The mechanisms facilitating and regulating stereotypy in basal vertebrates remain unknown, as fusimotor feedback loops have not been demonstrated (Ross et al., 2007). Still, the identification of proto-muscle spindles in a salmonid jaw adductor muscle(Maeda et al., 1978) suggests an evolutionary origin of neuromotor control governing stereotypical prey-processing mechanisms among pre-tetrapodean gnathostomes. The restricted raking MAP spread, and clear lack of recruitment differences suggests that the functional basis of this behaviour somehow lacks the prerequisites for modulation. It may also be that inter-specific differences exist in the central neural oscillators controlling raking MAPs, which would explain the rake-specific convergence in MAP timing traits (see below). Indeed, raking kinematics in the salmonid Salvelinus fontinalis were devoid of modulation in response to biomechanical prey-type differences(Sanford, 2001b), whereas a kinematic prey-type effect in response to comparable prey types was detected in the basal osteoglossomorph Chitala(Sanford and Lauder, 1990). Without investigation of additional taxa it remains unknown whether raking modulation is the ancestral or derived condition. One promising avenue for future work will be to ascertain if MAP stereotypy, or lack thereof, is linked to modulation in derived taxa.

Electromyography data from raking in O. mykiss and S. jardinii provide strong evidence that a convergently derived MAP is governing raking in these evolutionarily distinct lineages. The close apposition of raking MAPs in the principal component plots, relative to alternative feeding behaviours in both taxa(Fig. 6), is driven by a 100%consistent convergence in early AM and PH activity-onset timing, being the principal factors discriminating raking MAPs from strike and chew MAPs overall(Fig. 3). Although the resulting MAP is a novelty in motor-pattern research, these convergent variables primarily govern the raking preparatory phase. Raking MAPs displayed onset-timing sequences (PH>AM>EP> SH>HP) with consistency indices of 73% (O. mykiss; N=106 rakes) and 75% (S. jardinii; N=130 rakes). The remaining onset timing sequence only differed for the SH, HP and EP between species. A proportion of this variation, at least in S. jardinii, could be explained by very low raking activity levels in SH and HP (Fig. 3).

Several other lines of evidence support the notion of a convergently derived raking MAP, at the organisational levels of cranial morphology, muscle activity and kinematics. Firstly, the morphological key criteria for a TBA are presence of a cleithrobranchial ligament and derived basihyal dentition, none of which are present in basal teleosts(Sanford and Lauder, 1989), or in the basal salmonid taxa Coregonus and Thymallus(C.P.J.S., personal observation). Although fish oral cavities commonly are adorned with dermal teeth, basal teleosts, such as Amia, Lepisosteusand Polypterus do not posses basihyal fangs with opposing vomerine or parasphenoid tooth plates, nor do they have a CBL(Sanford and Lauder, 1989). Secondly, previously presented muscle-activity data for these basal taxa provided no evidence of early activity onset in the jaw adductor and basihyal protractor musculature, either during prey capture, or during intra-oral prey processing (chewing and prey transport)(Lauder, 1980; Sanford and Lauder, 1989). Thirdly, every other described feeding behaviour involving the oral jaw apparatus is characterized by a pronounced expansive phase involving mandible abduction. By contrast, raking is a unique feeding behaviour that involves early and maintained strong oral jaw adduction throughout the behaviour.

It is commonly assumed that convergent morphologies that face similar functional demands are generally associated with convergent motor-activity patterns (MAPs); however, there has been no documented evidence of this. Indeed, Sanford (Sanford,2001a) presented kinematic evidence that would seem to contradict this assertion in the TBA of two closely related notopterid knifefishes. Thus,our finding of convergently derived raking MAPs in representatives of two different teleostean lineages suggests that MAPs, even in coupled systems, are labile and have not posed a limitation in the origin and diversification of novel feeding behaviours. Nevertheless, our findings further suggest that once a novel MAP is established it can be highly stereotypical.

The MAPs of five cranial muscles differed markedly between raking, and the strike and chewing behaviours in O. mykiss, whereas inter-specific comparisons of raking MAPs between O. mykiss and S. jardiniirevealed more subtle, albeit significant, differences. In both taxa,intraspecific MAP variability was much less during raking than other behaviours. Early activity onset in PH and AM musculature, driving basihyal protraction and lower jaw occlusion, respectively, are convergently derived traits, only found in raking MAPs. This is strong evidence of a convergently derived motor pattern driving raking in at least these taxon representatives from evolutionarily unrelated lineages. However, both muscles exert their primary role during the raking preparatory phase, whereas the `power-stroke'muscles exhibit more labile onset-timing patterns during raking.

LIST OF ABBREVIATIONS

 
• AM

  adductor mandibularis

 
• CBL

  cleithrobranchial ligament

 
• EMG

  electromyography

 
• EP

  epaxialis

 
• HL

  head length

 
• HP

  hypaxialis

 
• MAP

  muscle-activity pattern

 
• PH

  protractor hyoideus

 
• PIM

  posterior intermandibularis

 
• SH

  sternohyoideus

 
• TBA

  tongue-bite apparatus

 
• TL

  total length

Thanks to Cold Spring Harbour Fish Hatchery for sourcing specimens, to M. Ajemian, M. McGuire, P. Wiehl, M. Kats, S. Watts and R. Bershader for help with experiments, to the Hofstra University Animal Care Facility staff for specimen maintenance, and to P. Doherty, A. Camp and two anonymous reviewers for comments that greatly improved the manuscript quality. This work was funded by the National Science Foundation (IOB#0444891, DBI#0420440).