Eleven cyclists with recent history of LBP (LBPG) and 13 healthy cyclists (CG) were recruited from different local cycling clubs and volunteered to participate in this cross-sectional study. To be included in the study, participants had to be aged between 18 and 65 yr., have a body mass index (BMI) less than 30 kg·m−2; and pedal at least 4000 km annually for more than two consecutive years. Exclusion criteria were (1) acute traumatic body injury or surgery and (2) history of cardiovascular and respiratory diseases. Specific inclusion criteria for LBPG were (1) non-specific low back pain episodes within the last six months but symptom-free during the last 6 weeks and (2) no pharmacologic interventions for the relief of LBP in the last six months prior to the test. Non-specific LBP refers to symptomatology without an identifiable etiology linked to pathology or trauma (Krismer and van Tulder 2007) and identified by excluding other spinal disorders. The specific inclusion criterion for the CG was no history of acute or chronic LBP within the last year. Three participants (two from the CG and one from LBPG) were excluded from the analyses because excessive sweating during the cycling test negatively affected EMG signal quality. Thus, results are presented for 21 participants (10 LBPG; 11 CG) (Table 1). Informed written consent was obtained by all participants prior to their involvement in the study. Experimental protocols and procedures were approved by the Internal Review Board of the Department of Biomedical Sciences of the University of Padua (HEC-DSB/09–2023) and conformed to the standards set by the Declaration of Helsinki.

The participants completed multiple questionnaires to assess their overall health and cycling habits. First, to estimate weekly physical activity levels, participants completed the International Physical Activity Questionnaire (IPAQ-SF (Mannocci et al. 2010), which is known for its acceptable validity and reliability across various populations (Craig et al. 2003). Second, to assess the level of disability related to LBP among the LBPG, we used the Oswestry Disability Index (ODI-I, Italian Ver. 2.1a (Monticone et al. 2009)), which shows high reliability and validity, especially for minor levels of disability (Fairbank and Pynsent 2000). Last, we administrated the validated Italian version of the Chronic Pain Grade (CPG) questionnaire, which assessed both the characteristic pain intensity and the disability related to LBP (Salaffi et al. 2006). This questionnaire classifies respondents into 4 categories: Grade 0 (no pain, no disability); Grade 1 (low disability, low intensity); Grade 2 (low disability, high intensity); Grade 3 (high disability, moderately limiting); and Grade 4 (high disability, severely limiting).

Data collection was performed over a five-month period at the Nutrition and Exercise Physiology Laboratory, at the Department of Biomedical Sciences of the University of Padua, Italy. Each participant attended the laboratory for two experimental sessions separated by 7–10 days.

In the first session, participants underwent anthropometric measurements and performed the validated Carmichael Training System (CTS) Field Test (Carmichael and Ruttberg 2012), to determine their Functional Threshold Power (FTP). In the second session, participants performed the incremental cycling test with the concomitant recording of HDsEMG from lumbar ES muscles (see Testing and procedures for details).

In both sessions, participants rode their own road bicycle mounted on a smart trainer device (Elite Direto XR, Padua, Italy). Furthermore, participants were asked not to engage in any strenuous physical exercise and to avoid caffeine or energy drink consumption within 48 and 24 h before each session, respectively.

The smart trainer device allowed participants to use their own bikes and replicate their habitual riding posture. Moreover, the device is equipped with a power meter based on an optical sensor technology that measures the torsion of the trainer axis with a ± 1.5% accuracy.

The CTS Field test was used to determine participants’ FTP (i.e., the maximum power (Watt) that a cyclist can sustain over 60 min (Allen and Coggan 2010)). After 10–15 min of warm-up, characterized by an easy-to-moderate pedaling and 3 × 30 s of moderate-to-high progressions, participants completed two 8 min steps at their maximum sustainable intensity. They were instructed to maintain a cadence between 80 and 100 revolutions·min−1 (RPM) and push themselves as hard as possible without slowing down or standing up from the saddle for the 8 min “all-out” effort. A 10 min low-intensity cycling period separated the two 8-min steps. During each step, the power output was recorded using the software MyETraining (v. 1.18.3.0; Elite Srl, Padua, Italy). FTP value was calculated by subtracting the 10% from the average mean power outputs achieved during the two 8 min steps (Carmichael and Ruttberg 2012).

In the second session, the same experimental setup of session 1 was used with the addition of the HDsEMG electrodes to record the EMG activity and the electro-goniometer (EGN) to identify the pedal strokes (Fig. 1a).

a Experimental Set-Up: a participant is seated on his bike mounted on a smart trainer device. Two semi-disposable adhesive grids of electrodes (13 rows × 5 columns, 8 mm inter-electrode distance) positioned over the ES muscles are connected to the multichannel EMG amplifier. An electro-goniometer (EGN) positioned on the right knee was used to characterize the pedal stroke. b Experimental protocol: after a standardized warm-up phase, participants performed an incremental cycling test consisting of 4 steps of 3 min interspersed with 2 min of rest at 50% FTP

Prior to electrode placement, the skin surface was shaved, lightly scrubbed with abrasive paste (Everi, SPES Medica, Genoa, Italy) to reduce skin impedance, and cleansed with alcohol. Two high-density grids of 64 equally-spaced electrodes (13 rows (10.0 cm) × 5 (3.5 cm) columns, gold-coated, with a 1 mm diameter and 8 mm inter-electrode distance) were prepared before placement with a double-sided adhesive foam layer. Electroconductive paste (AC Cream SPES Medica, Genoa, Italy) was used to fill the holes in correspondence with the electrodes. An experienced operator (kinesiologist) attached the two grids of electrodes bilaterally over the surface of the lumbar ES muscles at a standardized position (i.e., 2 cm lateral to the lumbar spinous processes, starting from L5 level to L3 level approximately) as previously described (Barbero et al. 2012; Martinez-Valdes et al. 2019). This positioning ensures that recordings are made from the lower lumbar fascicles of the ES, specifically targeting the iliocostalis lumborum, pars lumborum and pars thoracis (Sanderson et al. 2019b). Initially, hypafix tape was used to enhance skin–electrode contact. Subsequently, to further minimize grid detachment due to sweating, a cohesive bandage (Phytop, Wuxi Jiangsu, China) was lightly applied around the trunk. This application was carefully adjusted to ensure it did not cause any constriction or discomfort for the cyclists during pedaling. The HDsEMG signals were recorded in monopolar derivation, amplified (× 150), sampled at 2048 Hz, band pass-filtered at source (10–500 Hz), and converted to digital data by a 16-bit A/D multichannel amplifier (EMG-Quattrocento, OT Bioelettronica, Turin, Italy), prior to offline analysis.

A uniaxial EGN was positioned on the right knee, with the fulcrum located proximally to the lateral condyle to collect the knee flex–extension while pedaling. The EGN signal was sampled at 2048 Hz and synchronized with HDsEMG signals by the same multichannel amplifier (EMG-Quattrocento, OT Bioelettronica, Turin, Italy).

After the positioning of the sensors, participants performed a standardized warm-up as follows: 3 min at 50% FTP, 1 min at 100% FTP, 1 min at 50% FTP, 1-min at 100% FTP and 2-min at 50% FTP. Thereafter, they performed 4 steps of 3-min at 70%, 80%, 90%, and 100% FTP, respectively. A 2-min recovery at 50% FTP was performed after each step (Fig. 1b). Participants were instructed to maintain their self-selected cycling pace, allowing them to preserve their habitual pedaling technique during the whole test. Both HDsEMG and EGN signals were recorded continuously throughout the incremental cycling test.

HDsEMG recordings were analyzed offline using Matlab R2022b (Mathworks Inc, Natick, MA, USA) and Fieldtrip toolbox (Oostenveld et al. 2011). First, monopolar HDsEMG recordings were band pass-filtered at 30–400 Hz using a second-order, zero-lag, Butterworth filter. Thereafter, each filtered recording was visually inspected and bad channels defined as those not physiologically plausible raw EMG channels (e.g., extremely noisy or flat) were identified by means of one or more of several toolbox-supported metrics (e.g., kurtosis, variance), and subsequently removed. Channel visual inspection and removal strategy were performed with the Fieldtrip toolbox (Oostenveld et al. 2011) and have been applied in several other electrophysiological studies (Nordin et al. 2020; Shirazi and Huang 2021). Those channels were then reconstructed by averaging the nearest neighboring channels. Afterward, the preprocessed signal underwent single differential computations by subtracting adjacent preprocessed monopolar signals along each column of the grid, yielding 59 bipolar signals. This approach was chosen to extract information about muscle activation aligning with the presumed orientation of the lumbar ES muscle fibers (Mawston and G. Boocock 2015). HDsEMG analysis was performed by considering only the central ~ 2-min of each %FTP step. Within these time windows and for each cycling intensity step, individual pedal strokes were identified from the EGN recordings. Specifically, we identified the top dead center (TDC) with the maximum knee flexion and the bottom dead center (BDC) with the maximum knee extensions. Thus, each right pedal stroke extends from one TDC to the subsequent TDC, while the left pedal stroke from one right BDC to the subsequent right BDC (Fig. 2a). This method served to calculate the mean cadence for each participant at each %FTP.

a Example of 59 single differential HDsEMG signals recorded from the right erector spinae (ES) muscle for one representative participant during the test. The black line indicates the electro-goniometer (EGN) signal used for the identification of the pedal strokes. TDC and BDC indicate, respectively, the top dead center and the bottom dead center. Each right pedal stroke was defined from one TDC to the subsequent TDC. Note the periodic burst of activation of the right ES for each corresponding pedal stroke; b Representative RMS maps of average EMG amplitude recorded from ES of one cyclist without LBP (on the top) and for one cyclist with recent history of LBP (on the bottom) at increasing cycling intensities (% FTP). The RMS maps are normalized for average RMS value at 50% FTP. Note the higher activity of the ES muscle in the LBP cyclist. Areas with dark red correspond to higher RMS amplitude. The white circles indicate the position of the barycenter (Y-bar) of the ES activity

Subsequently, the root-mean-square amplitude (RMS) was calculated from the 59 bipolar signals for each pedal stroke within each %FTP step. The obtained RMS values were then averaged over the 59 signals and for all pedaling strokes, yielding an average RMS value (RMSMEAN) for each %FTP step. To enable the comparison among individuals, the RMS values were normalized to the average RMS value expressed in the first minute at 50% FTP. This normalization against a submaximal contraction is considered preferable for individuals affected by LBP that might face challenges in achieving maximal muscle activation (Ng et al. 2002) and has demonstrated high sensitivity in detecting alterations in muscle activation during cycling (Martinez-Valdes et al. 2016) and rowing (Martinez-Valdes et al. 2019) as the intensity increased. Thereafter, electromyographic activation maps (RMS maps) were extracted for each %FTP step, enabling a visuospatial distribution of ES activity (Fig. 2 b). The RMS maps provide insights into the intensity of muscle activity across different points of the acquisition grid. Furthermore, the modified entropy was computed as previously described (Farina et al. 2008) from the normalized RMS values to characterize the complexity of the EMG signal (i.e., degree of homogeneity or heterogeneity in muscle activation). Specifically, higher values correspond to a more homogeneous distribution of muscle activity and refer to a pattern of activation that is less localized and more equally distributed along the muscles. Conversely, a more heterogeneous signal suggests a more localized and less-uniform muscle activation. Finally, the muscle activity’s barycenter was calculated, yielding an indication of the average location of muscle activity. Particularly, the Y-axis coordinate of each RMS map’s barycenter (Y-bar) was analyzed to assess the cranio-caudal displacement of the barycenter at increasing cycling intensities. All channels were considered for this calculation.

Further statistical inference was performed on the averaged left- and right-side ES muscle activity, resulting in a single value per subject and %FTP step for RMSMEAN modified entropy, and Y-bar coordinates. This decision was made upon the rationale that participants in the LBPG had non-specific LBP (i.e., not localized to a specific side or both sides), and an additional statistical test failed to detect significant differences between left and right RMSMEAN, modified entropy, and Y-bar coordinates in LBPG and CG.

An a priori power analysis calculation was performed using the G*Power V.3.1.9.4 software (Heinrich Heine University, Dusseldorf, Germany) to determine the required sample size. RMSMEAN was specified as the primary outcome based on previous studies indicating alteration of RMS amplitude with increased load during specific tasks (Falla et al. 2014; Martinez-Valdes et al. 2019). By setting the α risk at 0.05, the statistical power at 0.8, and the effect size f at 0.35, it has been estimated that 20 participants were required to detect significant differences in the RMSMEAN. To account for potential loss of data due to signal quality or participant withdrawal, a total of 24 participants were recruited.

The normality of data distribution and the sphericity hypothesis were tested with the Shapiro–Wilk and Mauchly tests, respectively. When the sphericity assumption was violated, the Greenhouse–Geisser correction was applied. Moreover, a Levene’s test was used to check for the homogeneity of variances. Baseline anthropometric characteristics, ODI-I, and IPAQ were compared between groups by independent t-tests. For each EMG parameter (RMSMEAN, modified entropy, and Y-bar coordinates), interaction and main effects were checked with a 2-way mixed ANOVA for repeated measures with group (LBPG vs. CG) and FTP intensities (70%, 80%, 90%, and 100% FTP) as between and within factors, respectively. Partial eta-squared (ηp2) was calculated to measure the amount of variance of a dependent variable attributable to a given independent variable, considering the influence of the other independent variables present in the model. A ηp2 less than 0.06 indicates a small effect, between 0.07 and 0.14 a medium effect, and greater than 0.14 a large effect (Cohen 1988). If a significant group by FTP intensities interactions was found, the Bonferroni post hoc analysis was run for multiple comparisons. Statistical significance was set at p < 0.05. All statistical tests were performed with the software package JASP V. 0.16.4.0 (JASP Team, Amsterdam, the Netherlands).