Duty Factor Reflects Lower Limb Kinematics of Running

The aim was to identify the differences in lower limb kinematics used by high (DFhigh) and low (DFlow) duty factor (DF) runners, particularly their sagittal plane (hip, knee, and ankle) joint angles and pelvis and foot segment angles during stance. Fifty-nine runners were divided in two DF groups based on their mean DF measured across a range of speeds. Temporal characteristics and whole-body three-dimensional kinematics of the running step were recorded from treadmill runs at 8, 10, 12, 14, 16, and 18 km/h. Across speeds, DFhigh runners, which limit vertical displacement of the COM and promote forward propulsion, exhibited more lower limb flexion than DFlow during the ground contact time and were rearfoot strikers. On the contrary, DFlow runners used a more extended lower limb than DFhigh due to a stiffer leg and were midfoot and forefoot strikers. Therefore, two different lower limb kinematic mechanisms are involved in running and the one of an individual is reflected by the DF.

An emergent area of research is focusing on the individual runner to try to identify how the person's structure and functional abilities influence performance, running economy, or running-related injuries [12]. For instance, self-organizing map [13] or hierarchical cluster analysis [14] approaches identified different running gait strategies within a given sample of runner. Based on a subjective evaluation of the spontaneous overall running pattern, other researchers categorized runners in two groups with kinematic differences, called terrestrial and aerial runners [15,16]. Using the duty factor (DF; an objective parameter representing the proportion of time spent in contact with the ground during a running stride), Lussiana et al. (2019) observed global biomechanical differences when classifying runners in two groups called DF low (a low mean DF (DF) averaged over different running speeds) and DF high (a high DF) [17]. In their analysis, the authors divided the running cycle in several subtimings, namely, braking (t b ) and pushing (t p ) times, which are defined from footstrike (FS) to mid-stance (MS) and from MS to toe-off (TO), respectively, and elevation (t e ) and dropping (t d ) times, defined as the time from TO to mid-flight (MF) and from MF to FS of the contralateral foot, respectively. MS and MF events were calculated to divide t c and flight time (t f ), respectively. The authors observed that the two subcomponents of the contact phase (t b and t p ) were statistically longer for DF high than for DF low and that both t e and t d were statistically shorter for DF high than for DF low . In addition, the authors showed that the vertical displacement of the center of mass (COM) during the elevation phase (∆z COM,e ) was statistically greater for DF low than for DF high , but no statistical differences were obtained for the vertical displacement of the COM during braking (∆z COM,b ), pushing (∆z COM,p ), and dropping (∆z COM,d ) phases.
Similar subgroupings of participants were obtained based on the DF measured at 12 km/h than based on a subjective evaluation of the spontaneous overall running pattern [18]. The two groups of runners have the same running economy, but with a different running pattern: aerial and DF low favor a long t f together with a midfoot to forefoot strike pattern. On the contrary, terrestrial and DF high favor a long t c associated with a rearfoot strike pattern [17,19]. Moreover, grouping runners based on their DF reflect their global running pattern, which takes into account not only the FS pattern but also vertical oscillation of the head, antero-posterior motion of the elbows, vertical pelvis position at ground contact, and antero-posterior foot position at ground contact. Similarly to what was observed when grouping runners based on their global running pattern, no difference in running economy has been reported while using more local and finer parameters, such as the FS angle [20] or strike pattern [21][22][23].
Beyond metabolic cost, different FS patterns are providing different impact loading rates. Using a forefoot strike pattern does not eliminate the passive impact peak at the ground contact [22]. This passive impact peak usually occurs at about 25 ms after initial ground contact and this event marks the middle of the passive peak impact attenuation (IA) phase [24][25][26][27][28][29][30][31][32]. Both passive and active impact peaks need to be attenuated and different strategies for impact attenuation have been observed [33,34]. Forefoot strikers were shown to have a greater reliance on soft tissues (active mechanisms) [34,35] whereas rearfoot runners were shown to more prominently attenuate the impact using passive mechanisms [24,34], but also using active mechanisms [36,37]. Indeed, the heel pad of the foot and skeletal tissue are absorbing most of the shock when rearfoot striking [24,34]. In addition, the eccentric action of the knee extensor [37] and dorsiflexor [36] muscles during t c were also shown to play a role. As for running with a forefoot strike pattern, the eccentric action of the plantar flexor muscles during t c , acting as a shock absorber, is required [25,38,39]. These impact attenuation strategies between forefoot and rearfoot strike patterns induce different loadings on the lower limb and different three-dimensional (3D) stress patterns in the ankle, knee, and hip joint [40][41][42], and different sagittal plane (hip, knee, and ankle) joint angles and pelvis and foot segment angles during stance [42,43], but with no global advantage of one strike pattern over the other [40].
Following verbal instructions directing the runners towards a more compliant running stride, rearfoot strikers were able to adopt a higher degree of knee and hip flexions during t c [44]. Similarly, "Groucho" running [45] shows global body flexion while "Pose" running [46] uses an extended posture albeit the greater knee flexion at initial contact. Therefore, these observations suggest the existence of global running patterns associated with specific lower limb kinematics.
Hence, the classification of runners based on their DF showed the existence of two different self-optimized global running patterns [17]. However, nothing is known about their lower limb kinematics, particularly their sagittal plane (hip, knee, and ankle) joint angles and pelvis and foot segment angles during stance. Therefore, the purpose of this study was to identify the lower limb kinematics for both DF high and DF low . As DF high and DF low favor long t c and long t f (therefore shorter relative t c ), respectively, we hypothesized that these two DF groups would rely on two different lower limb kinematic mechanisms. More specifically, we hypothesized that DF high would use more lower limb flexion during t c than DF low .

Participants
The study was conducted during two months between September and October 2017, which permitted to test 59 participants, out of which 42 were males (age: 36.9 ± 9.5 y, height: 1.77 ± 0.06 m, leg length: 0.84 ± 0.04 m, body mass: 72.1 ± 10.0 kg, running time: 4.6 ± 2.3 h/week, and running distance: 43.8 ± 23.2 km/week) and 17 were females (age: 36.1 ± 8.2 y, height: 1.65 ± 0.07 m, leg length: 0.79 ± 0.05 m, body mass: 58.7 ± 6.9 kg, running time: 3.8 ± 1.9 h/week, and running distance: 34.9 ± 16.5 km/week). All participants had been running regularly for at least two years at the time of their study participation. Inclusion criteria were good self-reported general health and no current or recent (<3 months) lower-extremity injury. The university's institutional review board approved the study protocol prior to participant recruitment (CPP: 2014-A00336-41), which was conducted in accordance with international ethical standards, and adhered to the last Declaration of Helsinki of the World Medical Association.

Experimental Procedure
Each participant completed one experimental session in the laboratory. Running bouts were always performed under similar environmental conditions (23 ± 2 • C and 45 ± 7% relative humidity). All participants were advised to avoid strenuous exercise the day before the test. After providing written informed consent, retroreflective markers were positioned on participants (described in Subsec. Data Collection) to assess their running biomechanics. As for each participant, first, a 5-s standing static trial using a standard anatomical position was recorded on a treadmill (Medic 2850, TMS, Champs sur Yonne, France) for calibration purposes. Then, a 10-min warm-up run was performed on the same treadmill. The running speed could be freely chosen and modified during the 10-min run by the participant. This was followed, after a 1-min break, by six 30-s runs at 8, 10, 12, 14, 16, and 18 km/h (with 1-min recovery periods on the treadmill between each run). 3D kinematic data were collected during the static trial and the last 15 s (22 ± 2 running strides) of the running trials. All participants were familiar with running on a treadmill as part of their usual training program and wore their habitual running shoes during testing.

Running Gait Classification
The 59 participants were ranked according to their DF, i.e., the average over six runs performed at 8, 10, 12, 14, 16, and 18 km/h, and classified in two groups. The number of participants being odd, one group had an extra runner. The runner with the 30th DF corresponded to the median runner and has been attributed to the DF high group, but attributing him to the DF low group or removing him from the study would not have had an impact on the results. Runners with the 30 highest DF were attributed to the DF high group. The DF low group was composed of the runners with the 29 lowest DF. As Lussiana et al. (2019) observed a significantly larger DF for DF high than DF low at all speeds examined, DF is a reliable parameter to rank participants.

Data Collection
Whole-body 3D kinematic data were collected at 179 Hz using eight infrared Oqus 500+ cameras and the Qualisys Track Manager software version 2018.1, build 4100 (Qualisys AB, Göteborg, Sweden). The laboratory coordinate system was oriented such that the x-, y-, and z-axis denoted the medio-lateral (pointing towards the right side of the body), postero-anterior, and infero-superior axis, respectively. Forty-five and 41 retro-reflective markers of 12 mm diameter were used for static and running trials, respectively. They were affixed to the skin and shoes of individuals over anatomical landmarks using double-sided tape following standard guidelines from the Project Automation Framework Running package [47], and as already reported in Lussiana et al. (2019).
The 3D marker data were exported in .c3d format and processed in Visual3D Professional software version 6.01.12 (C-Motion Inc., Germantown, MD, USA). More explicitly, the 3D marker data were interpolated using a third-order polynomial least-square fit algorithm (using three frames of data before and after the "gap" to calculate the coefficients of the polynomial), allowing a maximum of 20 frames for gap filling, and subsequently low-pass filtered at 20 Hz using a fourth-order Butterworth filter.

Biomechanical Parameters
From the marker set, a full-body biomechanical model with six degrees of freedom and 15 rigid segments was constructed. Segments included the head, upper arms, lower arms, hands, thorax, pelvis, thighs, shanks, and feet. In Visual3D software, segments were treated as geometric objects. Segments were assigned inertial properties and center of mass locations based on their shape [48] and attributed relative mass based on standard regression equations [49]. Kinematic parameters were calculated using rigid-body analysis and whole-body COM location was calculated from the parameters of all 15 segments (COM was directly provided by Visual3D). Lower limb joint angles (hip, knee, and ankle) were defined as the orientation of the distal segment relative to the proximal one [50] and computed using an x-y-z Cardan sequence [51,52]. For example, the knee joint angle was computed as the orientation of the shank relative to the thigh. Using the proximal segment as the reference segment makes the x-y-z Cardan sequence equivalent to the joint coordinate system [52,53], leading to rotations with functional and anatomical meaning (flexion-extension, abduction-adduction, and internal-external rotation). Segment angles (pelvis and foot) were defined as the orientation of a specific segment relative to the laboratory coordinate system and were also computed using an x-y-z Cardan sequence. These segment angles have also functional meaning that are retroversion-anteversion, negative drop-positive drop, and external-internal rotation for the pelvis and forefoot-rearfoot angle, pronation-supination, and toe-out-toe-in for the foot.
Running events were derived from the kinematic data. More explicitly, mid-toe and mid-foot landmarks were created midway between the first and fifth toe markers, and the heel marker and mid-toe landmark, respectively. The mid-toe landmark was rescaled by subtracting its respective global minimum. Moreover, heel and mid-toe accelerations were calculated as the second derivative of the heel marker and mid-toe landmark, respectively. As for each running trial, FS was defined as the first acceleration spike between the heel marker and mid-toe landmark acceleration spikes. TO was defined as the instance when the mid-toe reached 1 cm on ascent. t c and swing time (t s ) were defined as the time from FS to TO and from TO to FS of the same foot, respectively, and t f as the time from TO of a given foot to FS of the contralateral foot. MS was defined as the instant when the COM reached its lowest vertical position during t c while MF was defined as the instant when the COM reached its highest vertical position during t f . The end of the IA phase was set at 50 ms after FS for every runner due to several reasons: (1) the passive impact peak of forefoot strikers is not identifiable without a force platform, (2) runners showing a passive impact peak have a large variability in the timing of this peak around 25 ms, and (3) this peak was shown to occur around 25 ms for speeds ranging from 3 to 5.5 m/s [28]. Passive peak attenuation time (t a ) was defined as the time from FS to IA, i.e., 50 ms. All events were verified to ensure correct identification and were manually adjusted when required. Values for t c , t a , t b , t p , t f , t e , t d , and t s were calculated based on FS, IA, MS, TO, and MF events, and DF was calculated as follows [54] All COM positions and displacements were expressed as a percentage of participant's height. Temporal characteristics and COM positions and displacements were given as the average between right and left values.
The joint and segment angles (θ) were calculated at FS, IA, MS, and TO together with their range of motions during the attenuation phase (∆θ a ; from FS to IA), braking phase (∆θ b ; from FS to MS), and pushing phase (∆θ p ; from MS to TO). Joint and segment angles were given as the average between right and left joint and segment angles, respectively. Segment angles were expressed as a difference with respect to segment angles in the static upright stance.
As for joint angles, we only kept the first Cardan angle, i.e., the rotation about the laterally directed x-axis (flexion-extension) because of possible errors due to kinematic crosstalk [55][56][57]. As for segment angles, the three Cardan angles were considered. Nonetheless, segment angles in the frontal and transversal planes were presented in Supplementary Materials, as they do not represent flexion-extension of the lower limb. The x-axis being directed laterally for the right-side segments and medially for the left-side segments, the left limb segment angles corresponding to a rotation about the antero-posterior y-axis and infero-superior z-axis were multiplied by -1 in order to maintain the same anatomical polarity between right and left segment angles and to ensure a correct averaging process.
The "running wheel", defined as the vertical position versus antero-posterior position of the COM of the foot segment (the motion of the foot COM in the sagittal plane), was calculated during the running stride. These positions were expressed as a percentage of participant's leg length. The leg length was defined as the average between right and left magnitudes of the 3D vector going from the hip joint to the ankle joint and calculated from the standing static trial.

Statistical Analysis
Assuming a medium effect size of 0.5 [58], an α error of 0.05, and a power of 0.8, standard sample size calculations have led to the requirement of 54 participants, however the 59 participants were kept to slightly increase statistical power [59]. Descriptive statistics are presented using mean ± standard deviation. After inspecting residual plots, no obvious deviations from homoscedasticity, homogeneity of variances, or normality were present. Unpaired two-sided Student's t-tests were used to compare participant characteristics between DF groups. Two-way (DF groups × running speed) repeated measures ANOVA with Mauchly's correction for sphericity and employing Holm procedures for pair-wise post-hoc comparisons were used to investigate whether biomechanical parameters differed between DF high and DF low groups, while accounting for the effect of running speed. Cohen's d effect size was calculated when a significant DF groups main effect was observed [58], and classified as small, moderate, and large when d values were larger than 0.2, 0.5, and 0.8, respectively [58]. Statistical analysis was performed using Jamovi (version 1.0.8, (Computer Software), Retrieved from https://www.jamovi.org) and R 3.5.0 (The R Foundation for Statistical Computing, Vienna, Austria) with a level of significance set at p ≤ 0.05.

Participant Characteristics between DF high and DF low Groups
The baseline characteristics of DF high and DF low together with pelvis and foot segment angles in the static upright stance are given in Table 1 and were similar between groups. The increase of the running speed from 8 to 18 km/h was accompanied with a decrease of the DF from 35.5 ± 2.4% to 26.6 ± 1.5% for DF high while DF low decreased their DF from 29.9 ± 2.2% to 23.1 ± 2.4%. 12 30.6 ± 2.2 26.0 ± 2.5 NA Note: data are means ± standard deviation. DF: average of the duty factors obtained from the data collected at several running speeds, M: male, F: female, and NA: not applicable.

Temporal Characteristics and Center of Mass Displacement
The two subcomponents of the contact phase (t b and t p ) were longer for DF high than for DF low (DF groups main effect, p < 0.001; t b : moderate effect size, d = 0.62 and t p : large effect size, d = 0.82; Table 2 and Figure 1). The opposite was observed for t e and t d (DF groups main effect, p < 0.001; large effect size, d ≥ 1.37; Table 2 and Figure 1), with greater values for DF low than for DF high . Running speed affected all temporal parameters, with a decrease of t b and t p and an increase of t e and t d with increasing running speed (speed main effect, p < 0.001; Table 2). Interaction effects were observed for all temporal parameters, indicating a decrease of t b and t p and an increase of t e and t d (all p ≤ 0.01; Table 2) with the increase of running speed for both DF groups.    Figure 1). There was a significant decrease of |∆z COM,b | and ∆z COM,p and a significant increase of ∆z COM,e and |∆z COM,d | while running speed increased for both DF groups (running speed main effect, p < 0.001; Table 2).

Hip, Knee, and Ankle Joint Angles
DF high had a larger hip flexion than DF low at FS, IA, and MS (DF groups main effect, p ≤ 0.002; large effect size, d ≥ 0.81; Table 3 and Figure 2A). More hip extension during t p was observed for DF high as depicted by the larger |∆θ p | (DF groups main effect, p < 0.001; medium effect size, d = 0.75; Table 3 and Figure 2A). Running speed affected all hip angle parameters, with an increase of the hip flexion at FS, IA, and MS, a decrease at TO, and more hip extension during all subphases of the stance phase with increasing running speed (larger |∆θ|; speed main effect, p < 0.001; Table 3). An interaction effect between DF groups and speeds was observed for ∆θ p , indicating a more pronounced hip extension during t p (larger |∆θ p |) with the increase of running speed for both DF groups (p = 0.012; Table 3).
The knee flexion was larger at MS for DF high than for DF low (DF groups main effect, p = 0.019; medium effect size, d = 0.59; Table 4 and Figure 2B). DF high exhibited more flexion of the knee during t b (larger ∆θ b ) and more extension during t p (larger |∆θ p |; DF groups main effect, p ≤ 0.019; medium effect size, d ≥ 0.53; Table 4 and Figure 2B). Running speed affected all knee angle parameters, with an increase of the knee flexion at the four running events, more flexion of the knee during t a and t b (larger ∆θ a and ∆θ b ), and more extension during t p with increasing running speed (larger |∆θ p |, speed main effect, p ≤ 0.001; Table 4). The interaction effect indicated an increase of the knee flexion at IA, more flexion of the knee during t a (larger ∆θ a ), and more extension during t p (larger |∆θ p |) with the increase of running speed for both DF groups (p ≤ 0.029; Table 4). Table 3. Hip extension-flexion joint angle at specific events of the running stance (θ FS , θ IA , θ MS , and θ TO ) and their range of motions during the different running stance phases (∆θ a , ∆θ b , and ∆θ p ) for the high (DF high ) and low (DF low ) duty factor running groups at the different running speeds.  Note: data are means ± standard deviation. Angles at foot strike (θ FS ), impact attenuation (θ IA ), mid-stance (θ MS ), and toe-off (θ TO ) events together with range of motions during the attenuation (∆θ a ), braking (∆θ b ), and pushing (∆θ p ) phases are presented. Significant differences (p < 0.05) identified by the two-way repeated measures ANOVA are indicated in bold. No significant difference between DF groups at a given speed was determined by Holm post-hoc tests. FS: foot strike, IA: impact attenuation, MS: mid-stance, TO: toe-off, and MF: mid-flight.
Appl. Sci. 2020, 10, x FOR PEER REVIEW 10 of 19 groups at a given speed was determined by Holm post-hoc tests. FS: foot strike, IA: impact attenuation, MS: mid-stance, TO: toe-off, and MF: mid-flight.

Pelvis and Foot Segment Angles
DFhigh had a larger pelvis anteversion than DFlow at FS (DF groups main effect, p = 0.026; medium effect size, d = 0.53; Table 6 and Figure 3A). More pelvis retroversion during ta and tb (larger |∆ a| and |∆ b|) were observed for DFhigh (DF groups main effect, p ≤ 0.002; medium effect size, d ≥ 0.62; Table 6 and Figure 3A). Running speed affected all pelvis anteversion-retroversion parameters, with an increase of the pelvis anteversion at FS, IA, MS, and TO, less retroversion during ta and tb (smaller |∆ a| and |∆ b|), and less anteversion during tp with increasing running speed (smaller |∆ p|, speed main effect, p ≤ 0.038; Table 6). An interaction effect between DF groups and speeds was observed for ∆ a, indicating less pronounced pelvis retroversion during ta (smaller |∆ a|) with the increase of running speed for both DF groups (p = 0.001; Table 6). Table 6. Retroversion-anteversion of the pelvis segment angle at specific events of the running stance ( FS, IA, MS, and TO) and their range of motions during the different running stance phases (∆ a, Figure 2. Extension-flexion of the joint angles (θ) during the running stance for the high (DF high ) and low (DF low ) duty factor running groups while running at 14 km/h for (A) hip, (B) knee, and (C) ankle joint angle. The running speed is representative of all speed. IA: impact attenuation and MS: mid-stance.
A larger plantar flexion was observed at FS for DF low than for DF high (DF groups main effect, p < 0.001; large effect size, d = 1.39; Table 5 and Figure 2C). DF low exhibited more dorsiflexion during t a and t b (larger ∆θ a and ∆θ b ) than DF high (DF groups main effect, p < 0.001; large effect size, d ≥ 1.24; Table 5 and Figure 2C). The attenuation phase was composed of a plantar flexion followed by a dorsiflexion for DF high whereas it only consisted in an ankle dorsiflexion for DF low ( Figure 2C). Running speed affected all ankle angle parameters except ∆θ b , with an increase of the plantar flexion at FS and MS and of the dorsiflexion at IA and TO, together with more dorsiflexion during t a (larger ∆θ a ) and less plantar flexion during t p with increasing running speed (smaller |∆θ p |; speed main effect, p ≤ 0.046; Table 5). Table 5. Ankle extension-flexion joint angle at specific events of the running stance (θ FS , θ IA , θ MS , and θ TO ) and their range of motions during the different running stance phases (∆θ a , ∆θ b , and ∆θ p ) for the high (DF high ) and low (DF low ) duty factor running groups at the different running speeds. Note: data are means ± standard deviation. Angles at foot strike (θ FS ), impact attenuation (θ IA ), mid-stance (θ MS ), and toe-off (θ TO ) events together with range of motions during the attenuation (∆θ a ), braking (∆θ b ), and pushing (∆θ p ) phases are presented. Significant differences (p < 0.05) identified by the two-way repeated measures ANOVA are indicated in bold. No significant difference between DF groups at a given speed was determined by Holm post-hoc tests. FS: foot strike, IA: impact attenuation, MS: mid-stance, TO: toe-off, and MF: mid-flight.

Pelvis and Foot Segment Angles
DF high had a larger pelvis anteversion than DF low at FS (DF groups main effect, p = 0.026; medium effect size, d = 0.53; Table 6 and Figure 3A). More pelvis retroversion during t a and t b (larger |∆θ a | and |∆θ b |) were observed for DF high (DF groups main effect, p ≤ 0.002; medium effect size, d ≥ 0.62; Table 6 and Figure 3A). Running speed affected all pelvis anteversion-retroversion parameters, with an increase of the pelvis anteversion at FS, IA, MS, and TO, less retroversion during t a and t b (smaller |∆θ a | and |∆θ b |), and less anteversion during t p with increasing running speed (smaller |∆θ p |, speed main effect, p ≤ 0.038; Table 6). An interaction effect between DF groups and speeds was observed for ∆θ a , indicating less pronounced pelvis retroversion during t a (smaller |∆θ a |) with the increase of running speed for both DF groups (p = 0.001; Table 6). Table 6. Retroversion-anteversion of the pelvis segment angle at specific events of the running stance (θ FS , θ IA , θ MS , and θ TO ) and their range of motions during the different running stance phases (∆θ a , ∆θ b , and ∆θ p ) for the high (DF high ) and low (DF low ) duty factor running groups at the different running speeds. Note: data are means ± standard deviation. Segment angles at foot strike (θ FS ), impact attenuation (θ IA ), mid-stance (θ MS ), and toe-off (θ TO ) events together with range of motions during the attenuation (∆θ a ), braking (∆θ b ), and pushing (∆θ p ) phases are presented. Segment angles are expressed as a difference with respect to segment angles in static upright stance. Significant differences (p < 0.05) identified by the two-way repeated measures ANOVA are indicated in bold. * Significant difference between DF groups at a given speed as determined by Holm post-hoc tests. FS: foot strike, IA: impact attenuation, MS: mid-stance, TO: toe-off, MF: mid-flight, and NA: not-applicable.
( IA), mid-stance ( MS), and toe-off ( TO) events together with range of motions during the attenuation (∆ a), braking (∆ b), and pushing (∆ p) phases are presented. Segment angles are expressed as a difference with respect to segment angles in static upright stance. Significant differences (p < 0.05) identified by the two-way repeated measures ANOVA are indicated in bold. * Significant difference between DF groups at a given speed as determined by Holm post-hoc tests. FS: foot strike, IA: impact attenuation, MS: mid-stance, TO: toe-off, MF: mid-flight, and NA: not applicable.  DF high had a more pronounced rearfoot angle at FS and IA, a smaller forefoot angle at MS, and a larger forefoot angle at TO than DF low (DF groups main effect, p ≤ 0.028; θ FS and θ IA : large effect size, d ≥ 1.10; θ MS and θ TO : medium effect size, d ≥ 0.60; Table 7 and Figure 3B). More forefoot angle during t a , t b , and t p (larger |∆θ a |, |∆θ b |, and |∆θ p |) were observed for DF high (DF groups main effect, p < 0.001; large effect size, d ≥ 1.04; Table 7 and Figure 3B). Running speed affected all forefoot-rearfoot angle parameters, with an increase of the rearfoot angle at FS, an increase of the forefoot angle at IA, MS, and TO, and more forefoot angle during t a , t b , and t p with increasing running speed (larger |∆θ a |, |∆θ b |, and |∆θ p |; speed main effect, p ≤ 0.001; Table 7). The interaction effect indicated a slight increase of the forefoot angle at IA with the increase of running speed for both DF groups (p = 0.001; Table 7). Table 7. Forefoot-rearfoot of the foot segment angle at specific events of the running stance (θ FS , θ IA , θ MS , and θ TO ) and their range of motions during the different running stance phases (∆θ a , ∆θ b , and ∆θ p ) for the high (DF high ) and low (DF low ) duty factor running groups at the different running speeds. Note: data are means ± standard deviation. Segment angles at foot strike (θ FS ), impact attenuation (θ IA ), mid-stance (θ MS ), and toe-off (θ TO ) events together with range of motions during the attenuation (∆θ a ), braking (∆θ b ), and pushing (∆θ p ) phases are presented. Segment angles are expressed as a difference with respect to segment angles in static upright stance. Significant differences (p < 0.05) identified by the two-way repeated measures ANOVA are indicated in bold. * Significant difference between DF groups at a given speed as determined by Holm post-hoc tests. FS: foot strike, IA: impact attenuation, MS: mid-stance, TO: toe-off, MF: mid-flight, and NA: not applicable.

Running Wheel
The running wheel at 14 km/h was more vertical for DF low (maximum of the vertical position of the center of mass of the foot segment, i.e., z foot = 50%) than for DF high (maximum of z foot = 40%; Figure 4). The antero-posterior range of motion of the running wheel was similar between both DF groups at 14 km/h (range of motion = 107%; Figure 4). However, DF high had a smaller minimum posterior position of the foot at 14 km/h (minimum of the antero-posterior position of the center of mass of the foot segment, i.e., y foot = -96%) compared to DF low (minimum of y foot = -102%; Figure 4). The compensation happened right before the foot struck the ground with a larger maximum anterior position of the foot for DF low (maximum z foot = 9%) than for DF high (maximum y foot = 5%; Figure 4) at 14 km/h. Similar effects were observed at all speeds with a more pronounced difference at a higher speed.

Discussion
In this study, in accordance with our hypothesis, DF high was relying on more lower limb flexion during t c than DF low despite a similar knee flexion at FS. In addition, DF high was striking the ground using the rearfoot while the midfoot was used in DF low . These two DF groups allow one to distinguish between two lower limb kinematic mechanisms in running.
The DF values in our sample of runners (Table 1) were similar to those previously reported in the literature at similar running speeds and agree with the fact that DF values of running locomotion should be under 50.0% [17,54,60]. DF low exhibited significantly shorter t b and t p but longer t e and t d associated with larger ∆z COM,e and ∆z COM,d than DF high ( Table 2). In addition, with increasing running speed, DF high was more prone to decrease t b and t p and to increase t e and t d than DF low ( Table 2). Increasing the running speed leads to the decrease of t c and increase of t f . Therefore, as DF high runners start from larger t c and smaller t f at 8 km/h than DF low and as t c and t f cannot indefinitely decrease and increase, respectively, the observed results naturally follow. These findings are in agreement with the observations of Lussiana et al. (2019). The running wheel was more vertical for DF low than DF high while the antero-posterior range of motion was similar between groups ( Figure 4). However, DF low was shown to use a more anterior running wheel than DF high while DF high a more posterior one than DF low (Figure 4). These different running wheels are reflecting the different temporal characteristics and COM displacements between the DF groups. As DF low depicted larger t d and ∆z COM,d than DF high , their vertical speed and thus vertical kinetic energy was higher at impact. In addition, the smaller t b and t p and therefore t c observed in DF low are indications that these runners have a presumably higher stiffness of the lower limb than DF high [61]. These observations led to the idea that DF low relies on the reuse of elastic energy (optimization of the spring-mass model) while DF high limits vertical displacement of the COM and promote forward propulsion (pulley system) to minimize running economy [17]. Similar conclusions were drawn when classifying runners based on a subjective evaluation of the spontaneous overall running pattern [19]. Besides a different strategy to minimize running economy between these two DF groups, these differences in running biomechanics also suggest different lower limb kinematic mechanisms.
Following the classification of FS pattern proposed by Altman and Davis (2012), DF high exhibited a rearfoot strike pattern [62] at FS while DF low was midfoot strikers ( Table 7). The fact that DF low did not show a clear forefoot strike pattern is probably related to the fact almost 90% of the population is striking the ground with the rearfoot [63] and that the inclusion criteria of this study were not specifically based on the strike pattern. Nevertheless, taken as a group, DF low runners were certainly using more their midfoot and forefoot than DF high (Table 7). Accordingly, a significantly more pronounced forefoot angle was observed at IA and MS for DF low than DF high , suggesting that DF high is relying more on the rearfoot during t a and t b than DF low . This is also partly explaining why the foot segment angle is not exactly zero at MS. Indeed, the body, and especially the foot, is applying a higher force to the ground during the ground contact phase of the running stride than when standing in a static upright stance, leading to more compression in the rear (DF high ) or fore (DF low ) part of the shoe insole and thus providing a non-zero angle. In addition, DF high depicted a significantly larger forefoot angle at TO than DF low , which was associated with a significantly larger |∆θ p | (Table 7). These findings further suggest that DF high promotes forward propulsion by pushing more with the toes during t p .
DF low exhibited significantly more plantar flexion at FS and more dorsiflexion during t a and t b than DF high (Table 5). A visual observation of Figure 2C depicts that a plantar flexion takes place at the very first instants of the stance for DF high , right before the usual dorsiflexion. These results suggest that DF low relies on the midfoot to forefoot and the eccentric action of the plantar-flexor muscles to attenuate the impact. On the contrary, DF high seems to additionally rely on the eccentric action of the dorsiflexor muscles, as already pointed out by Breine et al. (2019) for rearfoot strikers. The findings of this study mirror previous observations of rearfoot and forefoot strikers. Indeed, Derrick et al. (1998) observed that rearfoot strikers were absorbing most of the impact shock using the heel and skeletal tissue whereas forefoot strikers used an eccentric action of the plantar flexor muscles to absorb the shock [25,38,39]. Additionally, a rearfoot strike pattern was shown to cause a significant impact force on the heel, whereas peak plantar pressure was centered at the forefoot when forefoot striking [41]. Similarly, Dickinson et al. (1985) mentioned that for forefoot strikers, instead of being primarily an energy producer, the ankle takes the additional task of absorbing energy during t b and Hasegawa et al. (2007) pointed out the fact that shorter t c and increased plantar flexion angles require a greater amount of energy to be absorbed at the ankle [64].
When focusing on the entire lower limb, a larger range of hip extension and a greater increase in the range of motion of hip extension with increasing running speed during t p was observed for DF high than for DF low (Table 4). Besides, DF high demonstrated more hip flexion at FS, IA, and MS (Table 3) but a similar range of motion during t a and t b than DF low . In addition, the pelvis showed a higher range of retroversion during t a and t b and no change of retroversion with increasing running speed (Table 6) despite a statistically higher anteversion at FS for DF high than DF low . The lower vertical kinetic energy at impact (shorter t d and smaller ∆z COM,d ) and thus a softer (less stiff) lower limb allows a higher pelvis mobility in DF high than DF low and to rotate in the three anatomical planes (see Supplementary Materials) while DF low requires, due to the higher vertical kinetic energy at FS, a stiffer lower limb than DF high , which suggests limiting the rotation in the sagittal plane. Furthermore, less knee flexion was observed for DF low than DF high at MS but a similar one at FS, which led to a larger range of motion of the knee joint angle during t b and t p for DF high than DF low (Table 4). Therefore, a larger leg compression and decompression was observed for DF high than DF low despite the similar knee joint angle at FS. In addition, with increasing running speed, DF high depicted a larger increase of knee flexion at IA than DF low and a larger increase in the range of motion of the knee joint during t a and t p (Table 4). These findings are potentially representing a more pronounced eccentric action of the quadriceps muscle for DF high than DF low during t b . These results are in line with the observations of Gerritsen et al. (1995), which showed that the eccentric action of the knee extensor muscles was playing a role to attenuate the impact when rearfoot striking. Additionally, Knorz et al. (2017) have shown that different FS patterns change the distribution of loadings in a runner's body. The authors concluded that a forefoot strike pattern was generally associated with a greater vertical maximum peak force but significantly lower loading rates in the ankle, knee, and hip joints, which is in line with the lower flexion and compression of the lower limb observed for DF low than DF high . Similar studies were performed by other researchers and demonstrated that a forefoot strike pattern reduces knee loads whereas a rearfoot strike pattern reduces loads at the Achilles tendon [65][66][67] and ankle joint [68], again agreeing with less lower limb flexion for DF low than DF high .
Running speed affected most of the biomechanical parameters. t c decreased while t f increased with the increase of the running speed, leading to higher vertical oscillations of the COM during t f but smaller ones during t c for both DF groups. Similar observations were obtained by Lussiana et al. (2019). These results suggest that the increase of the running speed promotes the DF low temporal characteristics and COM displacements rather than the DF high ones, i.e., a faster running speed favors t f and corresponding COM displacements. However, the hip and knee flexions and the ankle dorsiflexion were increased during t c with an increase of the running speed for both DF groups. Moreover, a more rearfoot strike pattern was observed for both DF high and DF low with increasing running speed. An increase of the FS angle was also observed for aerial and terrestrial runners with the increase of the running speed [16]. These results suggest more lower limb flexion with increasing running speed. Nevertheless, even if higher running speeds seem to produce a favor to t f associated with more lower limb flexion during t c , differences between DF low and DF high are present at all speeds, with DF low showing a higher favor to t f and a more extended lower limb during t c than DF high .
Overall, these findings suggest the presence of two different lower limb kinematic mechanisms, which are directly linked to the runner's DF. These two mechanisms could be related to two different impact attenuation strategies. The existence of different type of shock attenuation strategies was already proposed by Novacheck (1998). Indeed, the author related these different strategies to the different forms of impact passive peak on the vertical ground reaction force [69]. A similar impact attenuation strategy than the DF high one was observed for "Groucho" running [45], while the shock absorption for "Pose" running [46] resembles the one of DF low . These two different strategies seem to rely on different muscles, i.e., plantar flexor muscles for DF low and dorsiflexor and quadriceps muscles for DF high . Altogether, it can be speculated that the impact attenuation strategy is also self-selected and is another part of the subconsciously driven self-optimization, a central element in the development of an economical and safe running gait [1,[9][10][11].
A few limitations to the present study exist. The ground reaction forces were not measured, thus making it not possible to directly link the calculated running kinematics (temporal characteristics, COM displacement, and joint and segment angles of the running step) to the impact loadings on the lower limb. Furthermore, whole-body 3D kinematic data was collected in order to report how the body globally structures itself while running. However, this "global" choice limited the precision of our 3D measurements when considering segment angles (as depicted by the large standard deviations). Therefore, further studies should focus on specific segment (and joint) angles to potentially observe more precise details. Moreover, no sex distinction was taken into account within the full-body biomechanical model. More specifically, a male-specific set of body segment parameters was used in Visual3D [48,49], which could have impacted the outcomes, such as the COM, and could have confounded conclusions. In addition, no sex distinction was taken into account within our DF groups. Thus, future work should focus on the impact of sex on global running pattern. For instance, one could wonder if, within a DF group, females depict similar lower limb flexion than males. Finally, FS and TO events were defined from an algorithm which has not yet been validated. However, all events were verified to ensure correct identification and were manually adjusted when required. Nonetheless, a future study focusing on the validation of this algorithm should be conducted to avoid manual verification of all events.

Conclusions
To conclude, this study observed that two lower limb kinematic mechanisms were involved in running and the one of an individual is reflected by the DF. DF low runners were shown to exhibit a more extended lower limb than DF high due to a stiffer leg, which requires less range of motion at the knee and hip joints, and to be midfoot and forefoot strikers. On the contrary, DF high runners, which limit vertical displacement of the COM and promote forward propulsion, exhibited more lower limb flexion during t c than DF low and were rearfoot strikers. Funding: This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.