### Experimental Design

Twenty-nine runners (19 males, 10 females) of varying levels of aerobic fitness (age: 34 ± 10 [mean ± SD] years; height: 1.74 ± 0.09 m; body mass: 68.3 ± 12.2 kg; \({\dot{\text{V}}\text{O}}_{{{\text{2max}}}}\): 56.6 ± 8.9 ml min^{−1} kg^{−1}) performed four sessions: a level running incremental test followed by three sessions with running bouts of 4 min each, at constant speeds (8, 10, 12 or 14 km h^{−1}) and slopes (− 20%, − 10%, − 5%, 0, + 5%, + 10%, + 15% or + 20%) with 1 week resting periods in-between the three sessions. Seven to eight running bouts were performed in each of the three sessions in a randomized sequence to perform a total of 25 conditions, but with a similar exercise load between sessions. Each running bout was followed by 2 min of rest for intensities lower than the first ventilatory threshold, and 5 min of rest for higher intensities. Conditions where the participants were not able to reach a steady state were aborted. All running sessions were performed on an instrumented treadmill (T-170-FMT, Arsalis, Belgium) in Lausanne (Switzerland). Participants were recruited in the local population and represented a wide range of aerobic fitness, running between one and five times a week. They were all familiar with treadmill running but were not trail specialists. All subjects signed an informed consent. The protocol was approved by the local ethical committee (CER-VD 2015–00,006) and conducted according to the declaration of Helsinki.

### Physiological Parameters

Gas exchanges were measured breath by breath (Quark CPET, Cosmed, Rome, Italy) and volume and gas calibrations were checked before each session. Heart rate was continuously measured (Polar Electro, Kempele, Finland). Blood lactate concentration was assessed from finger blood samples (Lactate Scout+, EKF Diagnostics, Leipzig, Germany) before the maximal incremental running test and after 1 and 3 min of recovery. For the constant-speed running bouts sessions, the blood samples were collected after 3 min of recovery.

### Maximal Incremental Level Running Test

All participants performed an incremental running test until exhaustion. The first stage began at 8 km h^{−1} for 4 min and then increased by 1 km h^{−1} every min.

### Energy Cost of Running Trials

As indicators of running economy, ECR was normalized by body weight (BW) and averaged over 30 s for each running condition by dividing the mean \({\dot{\text{V}}\text{O}}_{{2}}\) by the velocity and multiplied by the energy equivalent of O_{2} estimated by respiratory exchange ratio to be expressed in J kg^{−1} m^{−1} [16].

### Biomechanical Parameters

Ground reaction forces in the vertical (VGRF), forward, and lateral components were continuously recorded with an instrumented single-belt treadmill (T170—FMT-MED; Arsalis, Belgium) at a sampling rate of 1000 Hz. The calculation of the various biomechanical parameters has been described previously [16, 17]. The VLR was calculated as the average VGRF slope between 20 and 80% of the point where the VGRF slope is less than 15 BW s^{−1} [18]. The VLR was then averaged over the 30 s of recordings corresponding to 70–90 consecutive steps.

### Statistical Analysis

Due to the lack of several conditions, linear mixed models (LMM) were fitted with a random intercept effect for subjects, Fixed effects were analyzed for treadmill slopes and running speeds as ordinal variables (Jamovi 1.2, Sydney; Australia). Significance of fixed effects was evaluated with an analysis of variance (Restricted Maximum Likelihood—REML—estimation). As VLR, ECR and \({\dot{\text{V}}\text{O}}_{{2}}\) were used as dependent variables, Bonferroni’s correction was applied on the alpha level to account for repeated univariate testing. After fitting the LMM, the residuals were checked for normality using the Kolmogorov Smirnov test and post-hoc comparisons were performed within groups with Bonferroni’s multiplicity correction.

Pearson’s product moment correlation coefficients (*r*) were used to assess the intensity of the relations between variables using Statistica (13.5, Tulsa, Oklahoma, USA). For all these analyses, *p* < 0.05 was considered statistically significant. All data are expressed as mean ± standard deviation (SD).