Twenty-one asymptomatic participants (5 females, 16 males; age: 25.0 ± 5.3 years; height: 176.3 ± 6.6 kg; weight: 71.1 ± 10.5 kg) participated in the study. To be eligible to take part in the study, participants had to be 18–40 years old. People who reported current lower limb pain, major pathologies that affect lower limb mobility or balance, or who had undergone lower limb surgery, were ineligible to participate. The study was approved by the Internal Ethical Committee of Humanitas Research Hospital (code: CLF22/02) and participants signed a written informed consent prior to data collection.

After reporting which leg participants would use kick a ball to establish leg dominance (Melick et al. 2017), participants identified their self-selected walking speed on a treadmill in a familiarization trial. Each participant completed ten trials of at least 60 s of gait at the self-selected speed. The protocol is illustrated visually in Fig. 1. Participants performed three trials without painful stimulation to obtain stable baseline data. Then, they walked while experiencing painful stimulation in six trials, with pain locations randomized. Finally, they performed one trial without painful stimulation to assess whether gait adaptations outlasted the painful stimulation. At least 1 min of rest was provided between trials.

Protocol. Each block identifies a 60-s gait trial. Base baseline, no stimulation, Med medial stimulation, Ant anterior stimulation, Lat lateral stimulation, Post after the painful trials, no stimulation

Pain was induced in the right knee by means of electrical stimulation (Tucker et al. 2012; Gallina et al. 2021; Cabral et al. 2023). This methodology was preferred to other models such as injections of hypertonic saline solution because the pain location, intensity and duration can be easily controlled and standardized across participants. We have previously demonstrated that painful electrical stimulation reduces maximal knee extension strength, and that the decrease in force production is comparable to that of previous studies using hypertonic saline solution injections (Cabral et al. 2023). Three pairs of surface electrodes (I-Tech, 30 mm) were placed over the skin of the medial, lateral and anterior knee joint. Anteriorly, the electrodes were placed on the medial and lateral edges of the patella, approximately 6.0 ± 0.8 cm apart. This location was chosen to reproduce widespread pain behind/around the patella, commonly reported by individuals with patellofemoral pain (Collins et al. 2018). Medially and laterally, the electrodes were placed bridging the medial tibial condyle or the lateral femoral condyle in the antero-posterior direction. For both locations, the interelectrode distance was 2.1 ± 0.2 cm. The smaller interelectrode distance was necessary to avoid stimulation of muscle fibers of the distal vastus medialis and lateralis muscles. In addition, previous research (Thompson et al. 2009) showing that individuals with tibiofemoral osteoarthritis report either localized pain on the medial or lateral joint line, or regional pain around the patella, further supports the larger interelectrode distance for the anterior compared to the medial and lateral locations. Electrical stimulation was applied using an electrical stimulator (DS7, Digitimer, UK), controlled using a NIDAQ board (National Instruments, DAQ 782602-01) and Matlab (version, R2022b). The waveform was a 100-µs-long biphasic square wave, delivered at 20 Hz. Before the baseline trials, we determined the intensity of the stimulation needed to induce the target level of pain by incrementally increasing the stimulation intensity while asking the participant to report the intensity of the pain experienced on a Numerical Rating Scale (0–10 points, 0: no pain, 10: maximum pain). We applied 1-s-long stimuli starting from 0 mA, increasing in steps of 3 mA until the participant reported a pain of at least 3/10. Then, the stimulation intensity was increased in steps of 0.5 mA to identify the stimulation intensity necessary to induce a pain sensation of 4 out of 10. The stimulation intensity was then confirmed before the start of each trial while the participant stood on the treadmill. In each trial when pain was induced, the stimulation began before the participant started to walk on the treadmill. Participants reported the amount of pain 5 s after they started to walk at the self-selected speed, and at the end of the trial. At the end of each trial, participants drew the location where they felt pain on a body chart depicting the cross-section of a knee (Gallina et al. 2021). As a qualitative assessment of the temporal characteristics of the pain, participants were asked to report whether their pain increased during a specific phase of the gait cycle.

Lower limb kinematics was collected using an optical motion capture system (SMART-DX, BTS, Italy) using a sampling rate at 100 Hz. Four clusters of 5 retro-reflective markers were fixed to the lateral surface of the right thigh and shank using inextensible bands (Temporiti et al. 2020) to minimize skin motion artifacts. Eleven retro-reflective markers were placed on anatomical landmarks of the pelvis and right leg following Helen Hayes protocol (Kadaba et al. 1990), and the anthropometric measures needed to calculate kinematics were collected.

Lower limb kinematics was calculated using SmartTracker (BTS Bioengineering, Italy), and further processed using Matlab (R2021a, Mathworks, USA). For each trial, 60 s of gait at steady self-selected speed were analyzed. Individual gait cycles were identified based on the location of the marker on the heel. Hip, knee and ankle joint angles were calculated as Euler angles (XYZ sequence), low-pass filtered at 5 Hz (Butterworth, 2nd order), then averaged across all gait cycles. We also calculated the number of strides per minute (cadence), the time between two consecutive heel strikes (gait cycle duration), the antero-posterior displacement of the heel marker between toe off and heel strike (stride length), and the time between heel strike and toe off expressed as percentage of the gait cycle duration (stance phase duration).

For the calculation of the helical axes, raw marker data were low-pass filtered at 5 Hz (Butterworth, 2nd order). Markers clusters were used to identify tights and shanks as rigid bodies, and their mutual positions were described every 10 degrees of motion along the sagittal plane as a composition of a rotation around and translation along a fixed axis using the methodology described previously (Temporiti et al. 2020; Adamo et al. 2022). Knee helical axes dispersion was described using the mean distance and mean angle parameters in order to quantify helical axes displacement and orientation during the gait cycle. Mean distance represents the distance among helical axes providing a measure of joint rotation center displacement, whereas mean angle consists of a measure of helical axes orientation and quantifies the plane of motion variability during the joint movements. Mean distance and mean angle are presented for the sagittal plane which was the plane of motion that demonstrated highest reliability (intraclass correlation coefficients between 0.70 and 0.90; (Adamo et al. 2022)). Each gait cycle was divided into the following phases: (1) early stance, from 95% of the previous gait cycle to 10% of the subsequent gait cycle; (2) late stance, from 10 to 40% of gait cycle; (3) early swing, from 40 to 70% of gait cycle; and (4) Late swing, from 70 to 95% of gait cycle (Temporiti et al. 2020). For both joint angles and helical axes, estimates obtained in multiple repetitions of the same condition were averaged, resulting in a single value for baseline, anterior, medial, lateral, and after pain conditions.

Pain drawings were digitized as.jpg images and imported into Matlab. The colored pixels were discriminated from the white background and the black outline of the knee using the K-means algorithm. Since each pain condition was repeated twice, any pixel colored in at least one of the two maps was considered part of the colored area. The size of the colored area was calculated as the percentage of colored pixels with respect to the number of pixels contained within the knee cross-section outline.

Stimulation intensity and pain ratings were averaged across repetitions with the same pain location. Qualitative reports of whether pain was more intense in a specific phase of the gait cycle were classified in the following categories: ‘constant’, ‘stance’, ‘swing’, ‘toe off’, ‘heel strike’; each one indicates higher pain during that specific gait phase compared to the rest of the gait cycle (except ‘constant’).

All statistical analyses were performed using SPSS (version 28.0, IBM, Chicago). Parametric (one-way repeated measure ANOVAs) or non-parametric statistics (Friedman tests) with the condition as intra-subject variable were used based on data distribution (Shapiro–Wilk test). When Mauchly’s test identified violation of the sphericity assumption, a Greenhouse–Geisser correction was applied. In post-hoc comparisons, p values were Bonferroni-corrected for the number of comparisons. Depending on the distribution, data are presented either as mean and standard deviation, or as median and interquartile range. Stimulation intensity, pain intensity at the start and end of the trial, and all kinematic outcomes were compared across pain locations using ANOVAs for repeated measures or the Friedman tests. Pairwise comparisons were used as post-hoc tests. For each condition, pain intensity at the start and end of the trial was compared using a paired Wilcoxon signed-rank test.