*2.1. Subjects*

Ten young male adults volunteered for this study (mean ± standard deviation; age: 24.10 ± 1.79 years, height: 175.30 ± 5.12 cm, and mass: 69.40 ± 8.15 kg). All subjects were familiar with treadmill walking and had no neuromuscular, cardiovascular, or orthopedic diseases. All subjects were provided with the informed consent form for the experiment, and the experiment was approved by the Ethics Committee of Beijing Sport University (No. 2019007H (2019.01–2021.01)).

#### *2.2. Experimental Protocol*

Subjects walked on a force treadmill (Bertec, Columbus, OH, USA), on which the walking grade and walking speed could be adjusted accordingly. Each subject was familiarized with walking on the treadmill for about 5 min with the walking speed set to 1.11 m/s and the walking grade set to 0◦. The speed of 1.11 m/s (4 km/h) was selected according to the American College of Sports Medicine [22] and the literature on uphill and downhill walking [23]. Each subject was required to walk for 2 min on the treadmill with three mass backpack loads (10, 20, and 30 kg) during four grades (level (0◦), 3, 5, and 10◦ grades). The level and slope walking on the treadmill without loads (0 kg) were also completed by each subject as a reference. Subjects were asked to perform all the load mass tests at one certain grade in order to shorten the experiment time. The load masses were in random order during one grade test, and the grade testes were also in random order. Subjects were asked to walk at one grade for 3 min with one load during one trial. Subjects rested for 3 min between different load mass trials and 30 min between different grade tests. The experimenters adjusted the load mass and the walking grade while the participants rested. The speed of the treadmill was set to 1.11 m/s in every trial. The temporal stride kinematics and surface electromyographic (EMG) during the final 30 s of each trial were recorded.

The temporal stride kinematics were calculated according to the ground force data recorded by the force platform in the treadmill. Two force platforms were embedded in the treadmill and recorded the ground interactive forces and moments between foot and ground. The software in the optical motion-capture system (Motion, Columbus, OH, USA) [26] calculated the center of force operation based on the forces and moments.

The EMG signals were collected by the wireless EMG system (Delsys TrignoTM Wireless EMG System (Natick, MA, USA)). The muscles selected included the GM, RF, HA, anterior tibialis (AT), and GA, which are the main activated muscles during human lower extremity movements [27]. The pre-amplified single differential electrodes (Trigno, Delsys, Natick, MA, USA) were placed on the muscle bellies after preparing the skin with alcohol (the schematic diagram of human lower limb muscles in this study is shown in Figure 1). The surface EMG (sEMG) signal sensors were fixed with double-sided tapes and bandages to prevent displacement between the sensors and muscles during walking.

This optical motion capture system (Motion, Columbus, OH, USA) can integrate the Delsys device and the Bertec force platforms by a data acquisition card (DAQ, National Instruments, Austin, TX, USA); thus, the sEMG signal and kinetic data can be collected simultaneously. The sampling frequency was selected as 1200 Hz.

**Figure 1.** The schematic diagram of the human lower limb muscles in this study. The position of the two white points on one muscle represented the position of the electromyographic (EMG) electrodes attached.

#### *2.3. Data Analysis*

#### 2.3.1. The Temporal Stride Parameters

The kinetic data were obtained by the output of the software provided by the optical motion-capture system (Cortex-64, Santa Rosa, CA, USA), which was linked with the force treadmill system (Bertec, Columbus, OH, USA). Raw kinetic data were smoothed using a 4th-order Butterworth filter with a cutoff frequency of 10 Hz [28]. We then used vertical ground reaction force data and a threshold of 20 N (on the basis of the standard deviation of the vertical ground reaction force signal during leg swing) [29] to determine the heel strike and toe-off for each leg and computed the temporal characteristics of each trial using custom software (MathWorks Inc., Natick, MA, USA). The step length was determined as the distance between the point where the left heel strike occurred and the point where the right heel strike occurred in the walking direction, and the step width was determined as the distance between the two heel strike points in the walking transverse distance [30].

#### 2.3.2. The EMG Analysis

The raw EMG signals were filtered by a bidirectional Butterworth band-pass filter with cutoff values of 20 and 500 Hz [31,32] in a custom script written in MATLAB (MathWorks Inc., Natick, MA, USA). The signals were full-wave rectified and filtered by a low-pass filter at 10 Hz [33,34]. The filtered EMG signals were then used to calculate the root mean square (RMS EMG) with a window of 10 ms to describe the muscle activation during movement [33]. In addition, the filtered EMG signals were used to calculate the mean of the EMG (MEMG) to describe the work of one muscle during movement. Since the gait cycle was usually normalized to 0~100%, the sEMG data acquired synchronously should be also interpolated into 0~100% to describe how the muscles were activated during the gait [34,35]. Therefore, all the RMS EMGs were interpolated into 101 points corresponding to the gait cycle.

#### *2.4. Statistical Analysis*

We calculated the mean values of step length, step width, and RMS EMG, as well as MEMG, over ten consecutive strides. We then normalized the temporal stride parameters and determined the EMG parameters by calculating the ratio of the values during slope walking with loads to the values during level walking without loads. We used a two-factor (grade × load mass) analysis of variance for repeated measures to test the significant effects of the grade and load masses. When there was a significant effect (*p* < 0.05), Bonferroni corrected post hoc comparisons (adjusted *p* < 0.0072 (0.05/7, two dependent stride variables and five sEMG variables) [18,22–24,36] were carried out to evaluate the

differences between the grades and load masses. All statistical analyses were conducted using SPSS software (IBM SPSS Statistics 22).
