Twenty subjects (10 M, 10 F) were recruited from local trail running associations (age [years]: 32.8±8.3 M, 33.4±8.1 F; body height [cm]: 177.2±6.0 M, 166.3±6.9 F; body mass [kg]: 71.9±5.8 M, 61.6±6.9 F, experience in trail running [years]: 3.3±1.5 M, 4.1±1.2 F). All participants were amateur athletes habitually competing at regional to national level. The study was approved by the ethical committee of the University of Salzburg (GZ-10/2022), in conformity to the declaration of Helsinki.

Inclusion criteria were: •

Between 18 and 50 years old

After providing their informed consent, anthropometric data including body height, body mass and length of relevant body segments were recorded. On a separate day, participants completed a ∼9.1 km trail running time trial consisting of 7 laps of the same 1.3 km route (see Figure 1). Each lap presented an ascent of 60 m, resulting in 420 m of positive and negative elevation gain across the entire trial. Before the test, participants were accompanied during a complete lap of the running route to familiarize themselves with the test environment, followed by a self-selected warm up, consisting of level and incline running, as well as static and dynamic stretching exercises. Participants were instructed to complete the test in the shortest time as possible, without jeopardizing their safety in any instant of the trial. During the test, time and positional data, STP and full body kinematics were recorded.

All subjects were tested between April–August. The temperature during tests was 23.2 ± 3.7∘C, with sunny or cloudy weather conditions. No tests were performed with rain. Terrain was dirt road consisting of a stable combination of gravel and pine needles, on which athletes could safely and confidently run with no risk of falls. Sporadically, a few traits of asphalt were present as well. Due to the presence of trees, most of the route was in the shadow, thus avoiding that the athletes continuously ran under the sun.

During the trail running test, participants wore a GPS watch (Garmin Forerunner 935) and a full body motion capture system (Xsens Link, Xsens Technologies BV, Enschede, The Netherlands). The latter consisted of 15 inertial measurement units (IMUs, model MTx, size 36×24.5×10 mm, mass 10 g, sampling frequency 240 Hz). IMU sensors were located on: head, shoulders (2×), arms (2×), forearms (2×), thighs (2×), legs (2×), feet (2×), sternum and pelvis. Previous studies validated such system against gold standard marker-based methods, reporting reliable and consistent results for tasks such as running and changes of directions on both asphalt and irregular surfaces [13–16]. All subjects wore the equipment approximately 30 min before the beginning of the trial, thus having enough time to become accustomed with the suit and the wearables. Those subjects who asked for it, could run with their own watch as well, thus one at each wrist.

Since trail running trials are often characterized by speed decreasing across the entire task (i.e., positive pacing) with a end spurt in the very last stages [17], the 7th lap was excluded, as possibly not representative of the overall course. Subsequently, laps 1–3 were classified as initial stage, whilst laps 4–6 where classified as final stage. Within each lap, two UH and two DH sections were identified (see Figure 1). The whole data analysis was performed in a Python environment, after exporting joint angle and sensors raw acceleration data.

Data reduction was performed by averaging data (or summing, for the elapsed time), separately for UH and DH sections, and each race stage. In this way, for each participant, STP, joint angles and elapsed time were calculated in UH as well as in DH sections in the initial and final stage.

With respect to STP, running speed and elapsed time, separately for UH and DH sections, the assumption of normality was checked by means of the Shapiro-Wilk test, revealing normal distribution of all parameters in all conditions. Therefore, differences between initial and final stage in STP, running speed and elapsed time were assessed via paired sample t-test. Separately for UH and DH sections, differences in joint angles kinematics between race stages were assessed with Statistical Parametric Mapping [20], paired t-test. Moreover, to provide information about which sections are more strongly correlated to overall performance, the linear regression between total test time and time elapsed in UH and DH sections is reported as well. Level of significance was set to alpha = 0.05.

Significant differences were found between race stages with respect to kinematics (Figures 6–7). As for the ankle joint, in both UH and DH sections, dorsiflexion was higher in the final stage, during stance phase (p<0.05), see Figures 6A,B. In UH sections, higher ankle dorsiflexion was found immediately prior to foot strike as well (p<0.05), see Figure 6A. With respect to knee joint, in UH sections the athletes expressed higher flexion in late stance phase (p<0.05), in the final stage. In swing phase instead, for both UH and DH sections the range of motion (ROM) was lower in the final stage, as a result of similar peak flexion and lower peak extension, see Figures 6C,D. With respect to hip joint, significant differences were found during swing phase in both UH and DH sections during flexion sub-phase (p<0.05). In particular, athletes expressed lower flexion in the final stage, see Figures 6E,F. With respect to trunk lean athletes expressed higher forward tilt across the entire gait cycle in UH sections, during the last race stage (p<0.05). Conversely, no differences were found in DH sections, see Figures 7A,B. With respect to trunk axial rotation, in DH sections athletes expressed lower contralateral rotation during mid stance phase, in the final stage (p<0.05). No differences were found in UH sections, see Figures 7C,D.

As for UH sections, the hypothesis of lower SL in the final stage was confirmed. As SF did not change, lower running speed in the final stage is uniquely caused by lower SL, consistently with the findings of Townshend et al. [2], who reported that running speed depends more on SL rather than on SF. The hypothesis of longer CT in the final stage was confirmed as well. In this regard, previous works reported a negative relationship between running speed and CT [7], as well as higher CT in fatigued conditions [8]. In UH conditions, muscles must perform primarily positive net mechanical work to increase the potential energy of the center of mass (COM), during propulsion phase. Previous studies showed that, in UH conditions, the main contributor to positive net mechanical work is hip, followed by ankle and knee [23]. Despite differences in hip joint kinematics were found in swing phase only (Figure 6E), in the final stage we found lower ankle plantar flexion and lower knee extension during propulsion phase (Figures 6A,C), as well as during late swing phase, immediately prior to foot strike. Such differences result into lower ROM during propulsion phase, due to lower ankle plantar flexion and knee extension, in the final stage. Previous investigations reported that higher running speed in UH running is associated to higher ROM and higher energy generation, at the ankle joint [24]. Given this kinematic-kinetic link and the lower ROM found at the ankle in the present study, our results suggest lower energy generation at the ankle joint during propulsion phase, in the final stage. This would make propulsion less efficient. A possible reason is muscle fatigue at the ankle plantar flexor muscles. Such hypothesis would be consistent with previous investigations, reporting that such muscle group is particularly stressed, during trail running, showing both reduced central activation and strength loss [25]. With respect to the knee joint, no clear relationship has been found between kinematics and kinetics [24]. Nonetheless, lower ROM during propulsion phase (i.e., knee extension) suggests possible muscle fatigue at the knee extensors. In this regard, Giandolini et al. [26] investigated strength loss of plantar flexors and knee extensors with respect to exercise duration, considering both level running and trail running data. The authors reported comparable trend between level and trail running, suggesting that strength loss at plantar flexors and knee extensors is mainly related to exercise duration, rather than elevation gain. Based on such findings and considering the duration of the present trail running trial, strength loss at plantar flexors and knee extensors would be ∼ 15%.

Results confirmed the hypothesis of shorter FT in later race stages. Such findings are consistent with previous studies, where a positive relationship between running speed and FT was reported [7]. From a kinetic perspective, FT is determined by the impulse of ground reaction force (GRF) in the normal direction, i.e., the area underneath the normal GRF curve during CT. To maintain the same area (i.e., the same FT) during shorter CT, larger normal GRF are needed, and vice versa. Therefore, the combination of longer CT and shorter FT in our data clearly indicate the production of lower GRF and lower GRF impulse in the normal direction, in the final stage. Such findings corroborate our aforementioned hypothesis of less efficient propulsion phase during stance, in particular at the ankle joint. Provided that during flight phase body behaves like a ballistic object, a shorter FT, in turn results into lower SL. Finally, the combination of similar SF and longer CT explains the higher DF found in the final stage. With respect to upper body, trunk was significantly more tilted forward across the entire gait cycle (see Figure 7A). Forward lean has been reported to occur in presence of fatigue at the paraspinal muscles [8]. Reviewing the effects of acute fatigue in overground level running, it was reported that higher forward lean combined with reduced hip extension during swing phase, as found in the present study, may be a strategy to reduce stress on lower limbs joints [27], since higher trunk forward tilt has been indicated as a strategy to reduce stress at the joints and the risk of injury [28]. Nonetheless, such posture has also been reported as detrimental to running economy [29]. Given the increasing popularity of trail running, future works may investigate whether such findings apply to incline running as well. With respect to trunk axial rotation, no differences were found between race stages (see Figure 7C).

With respect to DH section instead, results confirmed the hypothesis of lower SL in the final stage (see Figure 5B). However, such decrease did not result into a lower running speed, as SF values showed a slight increase, in the final stage (see Figure 5A). Similarly to UH sections, it appears that running speed is more strongly regulated by SL rather than SF, as running speed and SL present the same trend. Results confirmed the hypothesis of longer CT in the final stage, despite the effect was not large, from a quantitative standpoint (see Figure 5C). FT, on the other hand, showed an evident decrease in the final stage (see Figure 5C). With respect to initial stage, the decrease in FT was larger than the increase in CT in the final stage, resulting in a lower stride period, which explains the slight tendency to higher SF in the final stage. With respect to kinematics, athletes showed a systematic tendency to higher dorsiflexion at the ankle joint in the final stage, which becomes significant during the propulsion sub-phase of stance (see Figure 6B). This seems to corroborate the hypothesis of muscle fatigue at the ankle plantar flexors. Interestingly, in the final stage, both hip and knee were significantly less flexed in early swing phase and contralateral stance phase, despite peak flexion values during swing did not differ between race stages (see Figures 6D,F). In previous works it was suggested that different kinematics of the swing leg may contribute to change its inertial properties and, ultimately, impact performance. In fact, lower knee flexion “has been shown to hinder the hip flexors from propelling the leg quickly forward due to the associated changes in the moment of inertia of the swing leg” [7, 29]. In trail running, higher hip and knee flexion during swing phase have been reported, when running at higher speed [9]. Next to the above considerations about inertial properties of the swing leg, it was suggested that these kinematic differences may result in a more forward displacement of the COM of the swing leg. Ultimately, this would contribute to maximize the forward momentum of the full body, thus optimizing the propulsion phase of contralateral stance leg. This would result into higher horizontal speed at toe off which, combined with the longer FT, would explain the higher SL in the initial rage stage. When considering the knee joint, the peak extension during the swing phase was lower in the final stage, as observed for UH sections. This finding supports the hypothesis of muscle fatigue at the knee extensors as well. With respect to upper body, trunk axial rotation between race stages was different approximately at 10-20% and 60-70% of the gait cycle (see Figure 7D). These ranges coincide with those where the contralateral leg is in the swing phase and its hip flexion differs between race stages. In particular, during initial stage athletes expressed higher hip flexion of the (contralateral) swing leg combined with higher trunk ipsilateral rotation, whilst the opposite is true in the final stage. Most likely, this is a strategy to maintain null the full body angular momentum about the vertical axis, with upper body and lower body rotating in opposite directions. If a mismatch occurred, it would result in a non straight running trajectory. These results are consistent with those reported by Genitrini et al. [9], where athletes running at lower speeds expressed lower hip and knee joint flexion during swing, as well as lower ROM for trunk axial rotation. With respect to trunk lean, no differences were found between race stages (see Figure 7B). In last stance, longer CT, shorter FT and similar SF resulted into a higher duty factor in the final stage.

The present study demonstrates how, comparing UH and DH sections, almost identical changes in performance-related variables (STP) can be elicited by dissimilar changes in technique-related variables (joint angle kinematics). This information is of relevance for sportspeople, given the increasing popularity of wearable devices to monitor training and performance. In fact, sensors such as foot pods or similar usually provide running metrics such as speed and STP, but no information about kinematics. Considering incline running, when observing a similar relationship between speed and STP in both UH and DH conditions, users would likely assume that this are caused by as much similar changes in running technique. The present work demonstrates that this is not the case. In particular, when comparing race stages, ankle joint kinematics differs in both stance and swing phases in UH sections, but in the stance phase only in DH sections. Knee joint kinematics differs in both stance and swing phases in UH sections, but in swing phase only in DH sections. Between race stages and with respect to upper body, in UH sections differences were found for the trunk in the sagittal plane, whilst in DH sections differences were found in the transverse plane. Therefore, sportspeople may target specific and different aspects of running technique, when running on different inclines.

A limitation of the present study is represented by the walking gait cycles: a small part of the decrease in performance in the final stage is explained by the fact that subjects were walking instead of running, and not uniquely by the differences in running biomechanics. Nonetheless, we believe that the criteria of excluding those subjects for whom the number walking gait cycles was dissimilar (i.e., more than 10%) between race stage may offset such limitation. Also, the population tested in the present study consists of amateur athletes, therefore, it is possible that the present biomechanical adaptations differ from those occurring in elite trail runners. Moreover, in trail running, psychological factors such as self motivation and willingness to take risks on technical terrain play a critical role. Therefore, future investigations may combine biomechanical tests with questionnaires focusing on the mental condition of the athletes. In last stance, it is well established that wearable sensors, usable in the field, do not provide the same accuracy as gold standard methods in laboratory conditions. To limit as much as possible this aspect, sensors were carefully fixed based on the instructions of the manufacturer. Also, as sensors may move during the test, we calibrated both before and after the test, so to have a a posteriori calibration, in case of low data quality due to small movements of the sensors with respect to the beginning of the test.