Participants

Thirty-nine women were recruited via advertisement and social media. The participants were carefully selected based on an initial questionnaire to assess inclusion and exclusion criteria. All participants were required to be in the age range of 18–40 years, to not have taken part in regular strength training (≤1 strength session per week) in the past 6 months prior to the intervention, and to be free of injury or condition preventing resistance training. Users of oral contraceptives (OC, n = 20) were included if they had been using combined OC pills (as opposed to progestogen-only pills) of the 2nd or 3rd generations for at least 6 months prior to the intervention, and no other form of hormonal contraception. Participants not using OC (NOC, n = 19) were included if they had not used any hormonal birth control in the past 6 months, had a menstrual cycle within a typical range of 28–32 days, and were not undergoing pregnancy. All participants signed a written informed consent. The study protocol was approved by the institutional ethics committee (No. 24-260917) and conducted in accordance with the Declaration of Helsinki.

Experimental design

The study was part of a larger project investigating the influence of OC on training outcomes, which involved different muscle groups and modalities of resistance training. However, these aspects did not interfere with the present training protocol, which focused on hamstring muscles. The training period lasted for approximately 12 weeks, corresponding to three menstrual cycles for the NOC group or to three pill cycles for the OC group. The tracking of menstrual and pill cycles was done using a smartphone application (Clue, Biowink, GmbH) and ovulation tests (Babyplan, BPL Diagnostics As, Oslo, Norway) in NOC participants. All participants trained twice a week, with supervision provided for the first two sessions and at least two additional sessions over the course of the training period.

Participants visited the lab on two occasions before the onset of training and once after the training period (Fig. 1). The initial visit was dedicated to collecting data from a subset of participants (n = 8) to assess the inter-day reliability of ultrasound-based measurements, and to familiarize (all) the participants with strength tests. All main outcome variables were measured during the baseline and the post-training sessions (2nd and 3rd visits respectively). All measurements were performed on the right leg.

Fig. 1Eccentric resistance training

The study incorporated eccentric knee extension and hip flexion exercises to train the primary functions of the hamstrings, as three of the four hamstring muscles are bi-articular and serve as powerful hip extensors. These exercises were included since each of them was found to induce fascicle lengthening in the biceps femoris long head (Bourne et al. 2017). For the Nordic hamstring exercises, the ankles of the participants were held in place to the ground with wall fixtures or with loaded barbells and their knees slightly elevated on a ~15-cm pad. The load progression was primarily regulated with a resistance band, whereby loading was increased by using bands with a lower stiffness or by removing the band entirely. This assistance and the knee elevation allowed participants to complete Nordic hamstring exercises over the full range of motion (i.e., up to 0° of knee flexion). Participants progressing to a sufficient strength level to perform the exercises without any assistance were instructed to increase the number of repetitions per set. For the single-leg eccentric hip flexions, a Roman chair was used to secure the ankle of the working leg in position. Load progression was achieved by having participants hold a weight either by their chest or over their heads. The number of repetitions was guided by the rate of perceived exertion (RPE), with failure subjectively assessed by either the training instructor or the participants themselves. We used repetition in reserve (RIR) and RPE-based training to ensure all participants exerted the same level of effort. Resistance training based on RIR has been demonstrated to be an effective way to autoregulate resistance training, allowing for individual adjustments based on daily readiness and performance levels (Larsen et al. 2021). Progress was also assessed by either increasing the load or total number of repetitions performed weekly within each menstrual/OC cycle (Table 1). Furthermore, two sets of Nordic hamstring exercises and eccentric hip flexions exercises were added to increase the overall training volume, to optimize gains in knee flexor strength and fascicle length (Severo-Silveira et al. 2021).

Table 1 Resistance training programSerum hormones and monitoring of menstrual cycles

Serum estradiol, progesterone, follicle-stimulating hormone (FSH) and luteinizing hormone (LH) were measured in all participants, at time points corresponding to the follicular phases (FP) and luteal phases (LP), during the 1st and 3rd menstrual cycles (Fig. 1). To standardize sampling to the diurnal peak of circulating estrogen (Bao et al. 2003), fasted blood was drawn from an antecubital vein in the early follicular phase, 3–6 days after the onset of menstruation (NOC), or 3–6 days in the withdrawal phase (OC), and in the mid-luteal phase, 20–24 after the onset of menstruation (NOC) or after the active pill phase, (OC) of the first and third cycles of training. Blood samples were first centrifuged and stored at +4 °C and sent to a private laboratory where serum levels of target hormones were measured using immunochemiluminometric assays (ADVIA Centaur XPT Immunoassay System, Siemens Healthineers) within 8 h. The analytical coefficients of variation for estradiol, progesterone, FSH, LH and SHBG were 7.9, 3.5, 3.1, 2.4 and 3.8%, respectively (values provided by the analysis laboratory). The detection thresholds were 0.04 nmol/L, 3.00 nmol/L, 0.3 IU/L, and 0.07 IU/L, respectively, for estradiol, progesterone, FSH and LH.

Muscle morphology and shear modulus

For the ultrasound-based measurements (Fig. 2), the participants were positioned supine on the backrest of a dynamometer chair. Their right leg was elevated and secured on the dynamometer attachment. The hip and knee angles were set at 90˚ and 45˚ respectively, with 0˚ corresponding to full extension in both cases. This scanning position imposed a passive tension on the hamstrings, and was chosen as it may be more sensitive to inter-individual differences in shear wave-based measurements (Avrillon et al. 2020) and to acute changes in this variable (Lacourpaille et al. 2017) compared to measurements taken in slack muscles. To identify scanning areas relative to mid-femur length, the femur length was measured between the greater trochanter and the lateral epicondyle. Additionally, matching the scanning areas in subsequent sessions was ascertained by visually matching echoic features (pattern of connective tissue, blood vessels) from initial scans. All ultrasound scans requiring manual analysis were anonymized.

Fig. 2

Experimental setup and representative ultrasound scans collected for the measurement of muscle thickness (1st row), fascicle length (2nd row) and muscle shear modulus (3rd row). BFlh: biceps femoris long head muscle, ST: semitendinosus muscle, SM: semimembranosus muscle

Muscle thickness and fascicle length were measured using B mode ultrasound (L10-2/38 mm, Mach 30, Hologic SuperSonic Imagine, Aix-en-Provence, France). Transversal scans of the biceps femoris long head (BF), semitendinosus (ST), and semimembranosus (SM) muscles were collected at 50% of femur length to measure thickness. Muscle thickness was obtained offline by manually outlining superficial and deeper aponeuroses and by measuring the mean distance between these structures (Fiji ImageJ distribution, Schindelin et al. 2012). While muscle thickness does not capture the same information as the volume or cross-sectional area in terms of force production capacity, it is a sensitive marker of hypertrophy (Franchi et al. 2018) and pilot testing suggested that it would be measured more reliably for this population. Muscle thickness measured from two different days in a sub-sample of participants (n = 8) indicated moderate to good (Koo and Li 2016) reliability, with a 95% intraclass correlation coefficient (ICC(3, 1)) of 0.81, 0.88 and 0.70, for the BFlh, ST and SM muscles, respectively. The raw typical error (TE) was 1.5, 1.2 and 2.3 mm and the coefficient of variation (CV) was 3.4, 4.6 and 4.9% respectively, for the BFlh, ST and SM muscles. The minimum detectable change at 95% confidence interval (MDC), computed as standard error of measurement × √2 × 1.96, was 4.2, 3.3 and 6.5 mm, respectively, for the BFlh, ST and SM muscles.

Architecture scans were collected from the BF muscle. With an initial probe position at approximately 7-cm proximal to the femur mid-line, the probe was moved distally over the muscle to collect a panoramic scan of complete fascicles. Fascicle length was measured offline (Fiji ImageJ distribution). In each image, three fascicles were manually drawn between aponeuroses by following regional gradients of visible segments. The resultant segmented lines were smoothed with a spline fitting and the mean length of the curves was retained as the final value. The fascicle length of the BFlh measured from two different days in a sub-sample of participants indicated an excellent (Koo and Li 2016) reliability, with an ICC(3, 1) of 0.91, a TE of 0.4 mm, an MDC of 1 mm, and a CV of 1.8%.

Shear wave elastography scans were collected from BFlh, ST and SM muscles with the same equipment as for B mode scans (presets: penetration mode, smoothing level 5, persistence off, sampling rate: 1 Hz). The scans were acquired with the probe orientated along the longitudinal axis of the muscles, at 50% of femur length. Ten frames were sampled for each muscle, and shear wave color maps were analyzed offline using a custom script (Python, swepy, Seynnes 2023). Shear wave data were averaged across frames and used as an indicator of instantaneous material property, by computing shear modulus from wave velocity data (Hug et al. 2015). To avoid bias due to pixel saturation, trials in which saturated pixels (≥500 kPa) exceeded 5% (averaged across frames) were excluded. Muscle shear modulus measured from two different days in a sub-sample of participants (n = 7) indicated good (Koo and Li 2016) reliability for the BFlh and SM muscles, with ICC(3, 1) coefficients of 0.89 and 0.86, respectively, a TE of 4.0 and 4.6 kPa, an MDC of 11.0 and 12.8 kPa and a CV of 6.7 and 8.5%, respectively. The ICC for the ST muscle was however moderate (coefficient = 0.73), with higher TE (7.6 kPa), MDC (21.0 kPa) and CV (21.2%).

Isometric and eccentric moment of the knee flexor muscles

To monitor changes in maximal strength of the hamstring muscles, we measured the knee flexor moment resulting from isometric maximal voluntary contraction (MVC) and from eccentric MVC (Humac Norm, Computer Sports Medicine Inc.; Massachusetts, USA). The tests were performed in a seated position with the back angle set at 70° of flexion (0° being when the hip is fully extended). Each test started with a specific warm-up consisting of 10 sub-maximal contractions. For the isometric test, the knee joint angle was set at 50°. The eccentric tests were performed over a knee range of motion of 10°–90° (0° being when the knee is fully extended), at a velocity of 30°/s. Each test was attempted three times. However, if any attempt yielded a result differing by more than 10% of the moment from the best attempt, the attempt was repeated. Each test/attempt was separated by a 2-min rest period.

Statistical analysis

Differences between groups at baseline were tested with a paired Student’s t test. Between-group differences in cycle-averaged hormonal levels were tested with a Mann–Whitney U test, as these variables were not normally distributed (as indicated with a Shapiro–Wilk test). For all main outcome variables, a repeated mixed-factor ANOVA (with training and group as factors) was used to determine whether the training intervention resulted in significant changes and whether there was an interaction with OC consumption. The effect sizes were reported as partial eta squared (\(\eta_}}^\)). When a significant interaction effect was found, a post hoc comparison was conducted using the Bonferroni correction to control for Type I error and the effect size was measured as the Cohen’s d. The analyses were performed in JASP 0.19.0 (2024). The α level was set at p < 0.05 and data are reported as mean ± standard deviation.