A dynamic model of the windlass mechanism of the foot: evidence for early stance phase preloading of the plantar aponeurosis

The dynamic action of the plantar aponeurosis (PA) is of great biomechanical interest as it represents, together with the long plantar ligament and the spring ligament, one of the main structures maintaining the integrity of the arch of the foot, which serves also as an important energy-storing mechanism (Ker et al.,1987). The PA is a strong elastic band spanning the foot longitudinally and connecting the calcaneus to the toes. It appears as a strong and thick band from the heel to midfoot, from which point it fans out into five slips that run underneath the metatarsal heads (MHs) and attach to the plantar side of the proximal phalanx of each toe. During dorsiflexion of the toes, as occurs in late stance, the PA is stretched as it wraps around the MH. This is the so-called windlass mechanism which, in the late phase of stance, is responsible for raising the arch of the foot(Hicks, 1954) and contributing to stiffening of the foot by pulling on the heel, causing inversion at the subtalar joint and `locking' the midtarsal joint(Briggs and Tansey, 2001). Previous studies using finite elements(Cheung et al., 2004; Giddings et al., 2000) and mechanical models (Fuller,2000; Gefen, 2003)have investigated the static biomechanics of this component but so far only one paper has addressed the dynamic behaviour of the PA, using cadaveric feet(Erdemir et al., 2004). We infer that the dynamics of the PA are intrinsically correlated to the kinematics of the segments to which it is attached and with the geometry,specifically the radius and dorsoventral dimension of the MHs. Moreover we postulate that the pronounced medial arch of the normal foot suggests that higher tension forces support the medial side and, given the PA is probably the main passive structure supporting the arch, that this is reflected in the distribution of the total longitudinal tension among its mediolateral slips. Therefore the model herein proposed aims to provide a non-invasive tool to test two main hypotheses: first that the PA is preloaded at heel-strike; and second that the geometry of the MHs and the mediolateral kinematic differences of the metatarsophalangeal joints (MTPJs) affect the longitudinal tension borne by the more medial and more lateral slips of the PA. To test these hypotheses we established a multi-body inverse-kinematic model of the windlass mechanism, built on anthropometric measurements of subjects and driven by in vivo kinematic data, able to evaluate the length versustime relationship of the slips of the PA segments for the duration of the stance phase of barefoot walking.

Three subjects (age 27±7.8 years, weight 65.1±7.2 kg, height 1.78±0.15 m), with no history of lower limb major injuries or abnormalities, volunteered as subjects in this study, which used protocols approved by the Research Ethics Committee of The University of Liverpool.

Non-invasive ultrasound sonography was performed on the left foot of each subject in order to identify the origin of the PA and to estimate the radius of the MHs. Manual biometry was employed to record the main dimensions of the toes. Subsequently the subjects' feet were fitted with reflective markers using double-sided adhesive tape and kinematics were recorded at the subjects'(subjectively determined) normal walking speed when walking over a wooden walkway with an inset forceplate. The marker trajectories were filtered and imported into a multi-segment rigid-body model of the foot and inverse kinematic analysis was run to determine PA tension (PA_tension)trajectories for each trial recorded. These analyses used the anthropometric data from each subject and the kinematics of each trial to build subject-specific geometry and to compute PA elongation with time. Details of the method follow.

Sonography was performed using a linear probe (Echo Blaster 128 EXT-1Z,Telemed, Vilnius, Lithuania) and settings suitable for superficial musculoskeletal structures (5 MHz; 50 mm depth; dynamic range, 56 dB). The linear probe was fixed perpendicular to a rigid plate on which a rectangular hole, the size of the probe sensor, had been made. The plate was aligned with the sole of the subject's foot and kept in position by an elastic band wrapped around the dorsum of the foot (Fig. 1). Sagittal imaging of the PA was then performed. The PA appeared as a white strip attached to the anterior aspect of the inferior border of the calcaneus; at each dorsiflexion apparent tensing was evident by slight changes in its orientation and by the PA becoming more parallel to the probe surface(Fig. 2). Once the PA origin(PA_o) had been identified and located approximately at the centre of the ultrasound image, the position of the probe was marked on the skin by passing the tip of a pen through two holes made on the plate, 1 cm on either side of the centre of the probe. The midpoint between these two points represents the projection of the origin of the PA onto the sole of the foot(PA_p). The depth (d) of the origin of the PA_o,that is the distance of the PA origin to the probe surface, was recorded(Fig. 2).

The probe, once removed from the supporting plate, was then placed on the plantar side of each MH and aligned with each (slightly flexed) toe. The curvature of the plantar surface of the MH can be seen clearly in Fig. 3. The cephalic radius of the MH can be approximated by a circle passing through three points on the curvature. Three different measurements, for a total of nine points, were taken for each MH. A Matlab routine (version 7.5, The MathWorks, Natick, MA,USA) then calculated the radius of the circle passing through each triplet of points; the radius of the given MH was taken as the average of the three measurements.

Six motion-tracking cameras (ProReflex 1000 MCU, Qualysis, Gothenburg,Sweden) were used to track the kinematics of the foot during the stance phase of walking.

Ground reaction forces were recorded (forceplate model 9281C, Kistler Instruments, Hook, Hampshire, UK) and used to identify stance-phase timing. The subjects' left feet were fitted with 12 (7 mm diameter) reflective markers, attached using double-sided adhesive tape, to track the movement of the calcaneus, the metatarsus and the five toes(Table 1; Fig. 4). The marker set is essentially a modification of the protocol designed by Leardini and colleagues(Leardini et al., 2007), in which four markers were added to the heads of the proximal phalanges of toe 2 to 5. Although three non-collinear points are sufficient to track the movement of a rigid body in space, it was decided to track the movement of the metatarsus using four markers (FMH, FMB, VMH and VMB) as a rigid body assumption seemed inappropriate given the motion of the five metatarsals.

A static calibration of the markers in situ was taken with the foot in neutral position on the ground. This posture was achieved by placing the target foot anterior to the contralateral foot so that the heel was approximately aligned to the contralateral first MTPJ. The subject shifted body weight so that it was carried primarily on the contralateral foot. In this position the knee of the instrumented foot was slightly flexed and the hip of the contralateral leg laterally tilted(Fig. 5). This position minimized the load borne by the foot, so that the PA would as far as possible be fully relaxed, while maintaining a quasi-normal body posture. In order to identify the sagittal plane of the hallux, and only for this static calibration, an additional marker (FMH2) was placed over the distalmost central point of the first metatarsal. This calibration allowed determination of the resting lengths of the PA slips, the MTPJ neutral (0 deg.) angles, and the location and orientation of the rigid-body segments at the start of the dynamic analysis.

Subsequently a reflective marker was placed on the sole of the foot at PA_p, the projection of PA_o on the sole of the foot. Another static calibration was made with the subject's foot suspended over the ground in such a way that the four reflective markers on the rearfoot could be imaged by the cameras. A Matlab routine was then used to calculate the 3D location of the PA_o on the calcaneus when the foot was on the ground in neutral position.

The total length of each toe was measured as the distance between the most distal central point on the dorsal side of each metatarsal and tip of the toe;proximal phalanx length (l) as the distance between the former and the reflective marker on that toe. Phalanx width and thickness at the first interphalangeal joint were measured with calipers. These measurements were used to create toe geometry and to help locate the axis of rotation of the MTPJs.

A commercial software package for multibody dynamic analysis (MSC.ADAMS Ver.2005.2.0, MSC Software Corporation, Santa Ana, CA, USA) was used to build the model geometry, which was customized for each subject's morphology. The size and position of the ellipsoids approximating the calcaneal and metatarsal segments were parameterized using the marker location with the foot in the neutral position. The toes were modelled according to the manual biometric measurements described above. The marker locations when the foot was in the neutral position and the sonographic measurements helped establish a reference frame for each MTPJ. The ellipsoid representing each toe was assumed to be parallel to the hallux, the longitudinal axis of which belonged to the sagittal plane through FMH2 and PM and intersected the origin of the MTPJ reference frame (Fig. 6).

As Briggs (Briggs, 2005)points out, the MTPJs are actually `sloppy hinge joints' providing flexion/extension but also a small amount of abduction/adduction. Here they were modelled as ideal hinge joints, the rotation axis of which was assumed to be parallel to the z-axis and to pass through the origin of the MTPJ reference frame. This location and orientation were used as the estimated starting values for an optimization routine which determined the MTPJ centres and orientations. Just as is the case for many other joints in the human body,the MTPJ axis is not stationary. Therefore an optimization approach was chosen to compute the location and orientation of a `compromise' axis at each MH, the mean of the momentary axes of rotation (see Appendix).

The five metatarsal bones are normally considered as a single unit in kinematics (Carson et al.,2001; Leardini et al.,2007) as there is little motion between the metatarsal bones. Referring to Fig. 6, the metatarsus is shown as a rigid ellipsoid, included only for visualization purposes. The actual interaction modelled was between the PA slips and the MHs represented as cylinders (which can be considered as moving rigidly with the metatarsus). The ultrasound measurements provided the radii for the cylinders approximating the MHs. The cylinders were set at the location of the MTPJ centre, with their longitudinal axis parallel to the z-axis of the MTPJ reference frame.

The PA was modelled as five springs, attached proximally to the calcaneus and distally to the inferior border of the proximal phalanx of each toe. More precisely the central spring was attached at PA_o and the origins of the other four springs were uniformly distributed around the central origin over a total width of 20 mm. In order to calculate the tension exerted by each slip of the PA, bulk stiffness values were assigned in accordance with Kitaoka et al. (Kitaoka et al., 1994),whose study showed little difference in the stiffness of different longitudinal bands of the PA from medial to lateral, any differences being dependent on the technique used to subdivide the PA more than on genuinely different zone-related material properties. Therefore, the mean stiffness value of 203.7 N mm^–2 found in that study was divided by five, the number of longitudinal springs in parallel forming the PA structure.

Distally four small spheres allowed the elastic elements to wrap around the MH (Fig. 6); frictionless contact forces were created between the spheres and cylinders. These spheres simulate the action of the `plantar plates', fibrocartilaginous plates which lie over the plantar aspect of the MTPJ. The fibre composition and orientation of these structures suggests that their main functions are to transmit the tensile load of the PA to the proximal phalanges and to sustain compressive weight, so supporting the articular surface of the MTPJ(Gregg et al., 2007; Mohana-Borges et al., 2003,see Fig. 7).

Anatomically the plantar plates provide insertions for the PA and are linked transversely by the deep transverse intermetatarsal ligaments(Briggs, 2005). The mediolateral stabilization action of these ligaments was simulated in the model by adding two elastic `ligaments' to each side of the contact spheres. This allowed the PA slips to move backwards and forwards underneath each MTPJ while preventing them from losing contact with the MH during dynamic analysis. Fig. 6 shows the full model as created for one of the subjects used in this study.

Following a residual analysis (Jackson,1979), the trajectories of the reflective markers were filtered using a zero-phase-shifting low-pass filter, with a cut-off frequency of 12 Hz. Force plate timing information was used to select the range of data representing the stance phase. Model geometry was set for each subject according to the location of the twelve markers with the foot in a neutral position. For each set of walking kinematics the analysis was performed in two steps. First the foot segments were translated from their neutral positions to the first frame of the stance phase (heel-strike). This neutral position was the position of the segments at the start of the analysis(Fig. 5), guaranteeing that the PA `springs' were fully relaxed before the inverse kinematic analysis was run. New trajectories were created through linear interpolation for each trial and each marker in order to get the segments into position for the second part of the analysis. Subsequently the temporal profiles of the marker trajectories,as recorded in vivo, were used to drive the segments in the actual inverse kinematic analysis of the stance phase. This two-step dynamic analysis was run for each trial of each subject (see supplementary material Movie 1). At the end of each analysis the deformation versus time curves of the five PA elastic slips, the contact forces between spheres and MH, and the MTPJ flexion/extension angles were exported for further scrutiny.

Ultrasonography permitted measurement of the depth of the PA origin,PA_o. The mean depth d of PA_o for the three subjects was 14.7±0.4 mm (mean±s.d.). The value of dwas used, together with the location of PA_p relative to the reference frame of the calcaneus, to determine the 3D location of PA_o. The distance between PA_o and the insertion on the plantar side of the MH, when the foot is in neutral position, represented the resting lengths of each slip of the PA(Table 3). Furthermore, by applying linear ultrasound measurements to the plantar side of each MH, we were able to estimate the cephalic radius of the metatarsals in the parasaggital plane. The average radii of the five MHs as assessed by sonography are shown in Table 2.

To the best of our knowledge this is the first attempt to reproduce the mechanics of the windlass mechanism in a full 3D model during the stance phase duration of barefoot walking. The temporal loading pattern of the PA(Fig. 8) was very similar to that measured by Erdemir et al. (Erdemir et al., 2004) using cadaveric feet loaded by a gait simulator. In contrast to their finding, however, in this study the PA appeared to be pre-loaded at heel-strike. It seems sensible to assume that some tension may be present in the PA at early stance phase. The simultaneous action of the ankle dorsiflexors and toe extensors, which prevent foot-slap and dorsiflex the toes at the MTPJ, and the plantarflexion moment applied to the calcaneus by the vertical ground reaction forces could account for some pre-stretching of the PA. A MTPJ dorsiflexion angle of about 30 deg. was measured for the three subjects thus confirming the action of the toe dorsiflexors at and prior to heel-strike (Table 4).

Indeed, a recent study has proposed that early stance preloading of the PA may be beneficial to propulsion during walking(Pataky et al., 2008). While the present study strongly suggests that such preloading exists, without further experimental and/or modelling studies, we can only speculate as to the possible advantages of such preloading. Loading the PA at heel-strike is likely to reduce the crimp present in unloaded collagenous tissues(Butler et al., 1978), thereby resulting in earlier arch stiffening and helping to ensure that, as the propulsive phase begins, a greater proportion of force is transferred by the foot to the ground.

Furthermore the detailed kinematics and anatomical description of the windlass mechanism made it possible to estimate the elongation of the five slips of the PA during the whole duration of stance. For all three subjects the medial slip of the PA was the most stretched and the three central slips showed similar patterns of elongation and maximum tension. With the exception of subject 3, whose lateral-slip elongation was comparable to that of the three central slips, the lateral slip was the least stretched(Fig. 9).

From heel-strike to about 30% of the stance phase the length of the PA was rather constant (see Fig. 8). After a peak at heel-strike, the action of tibialis anterior and of toe extensors decreases (Winter and Yack,1987), and the toes plantarflex, to achieve full contact with the ground, and thus relax the PA. PA elongation lasted from 30% of stance until midstance, as the body weight was borne fully above the longitudinal arch. After midstance, a sharper increase in PA elongation occurred, with a peak at about 80% of stance. The push-off phase is of course characterized by the action of the powerful ankle plantarflexors pulling on the calcaneus via the Achilles' tendon. Through the windlass mechanism the PA slips are elongated, being constrained to wrap around the MHs, while the toes, now firmly planted on the ground, dorsiflex at the MTPJs.

The maximum strains recorded for each PA slip in each subject, that is the ratio between maximum elongation and resting length(Table 3), ranged between 3.5%and 6%, less than those reported by Gefen(Gefen, 2003) on the basis of radiographic fluoroscopy measurements on two subjects (around 12%) but in accordance with in vitro cadaveric studies which measured strain of up to 6.5% in the PA (Kogler et al.,1999). In another in vitro study(Carlson et al., 2000) the combination of MTPJ dorsiflexion and Achilles' tendon tension caused a maximum strain in the PA of 3.5%. This seems to further support the validity of the current dynamic model.

The maximum PA_tension, summed across all five slips, was approximately 1.5 BW (averaged across subjects). This was higher than that measured by Erdemir and colleagues (Erdemir et al., 2004) of about 1 BW, but lower than 1.9 BW estimated by Giddins and colleagues (Giddins et al., 2000). A possible explanation for these differences, in addition to the different methodologies employed in these studies, might be related to the adopted estimate of the PA stiffness,determined from mechanical tests on cadaveric feet(Kitaoka et al., 1994). The common mean value of PA stiffness used for all three subjects (203.7 N mm^–1) is already affected by significant variance (50 N mm^–1) which could imply significant differences in the mechanical properties of the PA tissue among different subjects. Lower maximum tensions in the PA would have been obtained by adopting Gefen's(Gefen, 2003) stiffness value of 170±45 N mm^–1. However, Kitaoka's value was adopted in this study as Gefen's value was derived through indirect measurement(fluoroscopy) of a much smaller sample. Moreover the assumption of linear elastic behaviour is an approximation of the actual material properties of the PA (Ker et al., 1987). As already mentioned above, the collagen in the PA tissue could be responsible for a certain degree of crimp, a phenomenon characterizing the stress–strain relationship of collagen-containing tissues such as tendons and ligaments (Butler et al.,1978). As a consequence of crimp, the tension in the tissue is low until the strain has passed a certain threshold (toe-region). For these reasons the elongation and therefore the maximum tension exerted in the PA, as predicted by this model, may have been slightly overestimated.

Fig. 12 reveals a good linear relationship between maximal PA_tension and the contact force applied to the MH (Figs 7 and 11). While expected from pulley mechanics, when coupled with the estimated magnitude of the PA_tension of 1.5 BW, this finding suggests that the windlass mechanism may be important in maintaining MTPJ integrity. If our interpretation of the mechanism is correct, then one would predict decreased MTPJ integrity with reduced and/or absent PA_tension.

This hypothesis finds support in a study by Sharkey et al.(Sharkey et al., 1999) who measured a shift of pressure, from toes to the MH, in cadaveric feet, when bands of the PA were sequentially released. In the same study a significant increase was reported in strain and bending in the second metatarsal, probably also affecting all metatarsals to some degree. An increase in the dorsiflexion of the first metatarsal has also been measured by Harton et al.(Harton et al., 2002) in patients who had undergone plantar fascioctomy. From the diagram in Fig. 7 it is possible to see how the PA acts to support the vertical forces on the forefoot, and, through the insertion on the plantar aspect of the proximal phalanx, the horizontal component of the tension in the PA acts to pull the proximal phalanx back towards the MH, exerting a plantarflexion moment on the metatarsals which helps to counterbalance the dorsiflexion moment due to vertical ground reaction forces. Therefore the posterior shift of pressure measured in the cited in vitro studies following release of the PA appears to be a consequence of an unbalanced force distribution in the forefoot, where the release of the PA would prevent proper support of the MTPJ and decrease the plantarflexion moment applied to the metatarsals through the PA insertion at the proximal phalanges.

As shown by Gibbon and Long (Gibbon and Long, 1999), ultrasound imaging can thus be successfully used to identify the PA origin. Here the combination of sonography and stereophotogrammetry has been used to estimate the 3D location of the origin of the PA in relation to three bony landmarks on the calcaneus. Sonography has again been applied to measure the approximate cephalic radius of the metatarsals. The mean intersubject values are well in accordance with the range of values measured by Largey and colleagues(Largey et al., 2007) on a larger set of bones using CT scanning.

If medial and lateral MH motion is adequately described, the relative motion of the central MHs is less well tracked by the chosen marker set. Magnetic resonance studies of a large group of volunteers have shown statistical differences in the vertical displacement of the second to fourth MH, compared with the first and fifth MH, when changing from a supine position to a position requiring weight bearing by the forefoot(Weishaupt et al., 2002). This could lead to underestimation of the elongation of the three central `springs'of the PA running underneath these MHs.

The optimization analysis located the axes of the MTPJs near the geometric centre of the MHs (see Table 5). Although this result has been supported at least for non-pathological first rays (Shereff et al., 1986) no publications have addressed this subject for the lateral MTPJs. The maximum dorsiflexion angles of the MTPJs at push-off(Table 4) were within the anatomical range of motion (65 deg.) and decreased from medial to lateral. This was in accordance with normal foot kinematics at push-off, and a probable consequence of the lateral metatarsals lifting off the ground sooner than the first ray, in the propulsive phase (Root et al., 1977).

Intrinsic torsion of the metatarsals has been measured through CT scanning on cadaveric feet (Largey et al.,2007), and shows that the anatomical orientation of the MHs is not parallel to the ground, but rather slightly inverted or everted. The use of non-invasive sonographic imaging in this study did not allow assessment of the degree of inversion/eversion of the MHs of each subject: therefore the cylinders modelling the MHs were represented as parallel to the ground when the foot is in neutral position, and this is clearly a simplification. However, a sensitivity analysis run on this parameter has shown that this approximation affects PA elongation only minimally.

By demonstrating that the windlass mechanism results in early stance-phase loading of the PA, and linking this effect to the radius of the MH, this study also raises interesting evolutionary issues meriting further systematic study. The MHs of chimpanzees and probably other non-human great apes are dorsoventrally shallower than those of humans(Susman and Brain, 1988)[discussed by Aiello and Dean (Aiello and Dean, 1990)] and thus must have a smaller radius. It follows that in their occasional bipedalism the early stance-phase preloading of the PA seen in this study would either not occur or be relatively low. This might in part account for the tail-off in vertical ground reaction forces which differentiates bipedal walking of all other great apes (chimpanzees, bonobos,gorillas and orangutans) from that of humans(Crompton et al., 2008).

Finally, although data from more subjects would increase the statistical significance of the current results, we must highlight that the three current subjects all showed very similar trends. Furthermore, the intensive nature of the current data collection procedure prohibited us from collecting a larger sample of data; small samples are typical of biomechanical testing of this nature (Gefen, 2003; Giddins et al., 2000). Regardless, the current results should be considered with reference to the limited number of subjects employed in the study.

Sonography and manual biometry have enabled a biorealistic representation of the elements relevant to the analysis of the windlass mechanism, and stereophotogrammetry has enabled us to analyse their in vivodynamics. Preloading of the PA in early stance and the role of the PA in sustaining the MTPJ in late stance have been highlighted. For the first time it has been possible to estimate the distribution of tension in the PA between its five longitudinal slips. The maximum strain in the PA slips decreased from medial to lateral, confirming the role of the PA in sustaining the medial longitudinal arch. These results are novel and emphasize both the importance and complexity of PA dynamics during barefoot walking.

The location and the orientation of the five hinge joints, in local reference frames, modelling the MTPJs were identified by means of optimization. The reference frame of the first MTPJ had its origin in(PM_x–l,FMH2_y–(R+r), PM_z)and the x-axis belonged to the saggital plane through PM and FMH2(pointing anteriorly), where: PM_x and PM_z are the x- and z-coordinates of the marker on the proximal phalanx, respectively; FMH2_y is the y-coordinate of the marker on the head of the first metatarsal; R and r are the radii of the metatarsal head(Table 2) and of the reflective marker (3.5 mm), respectively; and l is the proximal phalanx length. The origins of the reference frames of the lateral MTPJs were established according to the convention described above by replacing PM with the marker on the proximal phalanx articulating at that MTPJ (e.g. SPH for the MTPJ of the second toe) and by estimating the y-coordinate of the head of the metatarsal concerned through linear interpolation between FMH2_y and VMH_y, where VMH_y is the y-coordinate of the marker on the fifth metatarsal head (see Figs 4 and 6).

The (x,y) coordinates of the centre of each MTPJ and the axis orientation angle α in the coronal plane, coincident with the yz plane of the joint reference frame, and the axis orientation angleβ in the transverse plane, coincident with the xz plane of the joint reference frame, were parameterized. When (x,y)=(0,0) and(α,β)=(0 deg.,0 deg.) the MTPJ rotation axis is coincident with the z-axis of the joint reference frame. This was the MTPJ location at model construction, which was close to the geometric centre of each MH and represented the initial conditions for the optimization analysis.

For each MTPJ, the trajectories of the four markers on the metatarsus and of the marker on the proximal phalanx of the toe during one walking trial were used as input. The optimization problem was set using the four parameters of each MTPJ axis as independent variables and the mean-squared error between the recorded in vivo marker kinematics and their counterparts on the multisegment model as the objective function.

ADAMS sequential quadratic programming (OPTDES-SQP) optimization algorithm minimized Eqn A1 for each MTPJ independently.

Table 5 shows the optimal locations of each MTPJ centre and axis orientation relative to the MTPJ reference frame. Good intersubject consistency demonstrates the robustness of the current approach.

LIST OF ABBREVIATIONS

 
• BW

  body weight

 
• CF

  contact force between PA slip and metatarsal head, vertical component

 
• d

  depth of the origin of the plantar aponeurosis

 
• f

  error function

 
• HS

  heel-strike

 
• l

  length of the proximal phalanx

 
• MH

  metatarsal head

 
• MTPJ

  metatarsophalangeal joint

 
• PA

  plantar aponeurosis

 
• PA_o

  origin of the plantar aponeurosis

 
• PA_p

  projection of PA_o on sole of the foot

 
• PA_tension

  tension exerted by the PA slips

 
• PO

  push-off

 
• R

  radius of the metatarsal head

 
• r

  radius of the reflective marker

 
• r.m.s.

  root mean square

 
• RMSE

  root mean squared error

 
• TO

  toe-off