A convenience sample of 131 asymptomatic volunteer participants was recruited to the Reference Database study from staff, students and visitors at the AECC University College (Bournemouth, United Kingdom) between July 2011 and July 2020. Participants were included based on the following inclusion criteria: between 21 and 80 years old, self-reported body mass index less than 30 kg/m² (to ensure image quality), free of pain on the day of testing, free of any back pain that limited normal activity for more than 1 day in the previous year, no history of abdominal surgery or spondylolisthesis, no medical radiation exposure of >8 mSv in the previous 2 years (self-identified by pre-study questionnaire detailing recent medical imaging), and not currently pregnant. Ethical approval was obtained from the United Kingdom National Research Ethics Service (SouthWest 3, 10/H0106/65) and written Informed consent was obtained from all participants prior to inclusion in the study.

For the Patient-control subgroup study, 8 patients without any obvious mechanical disruption (for example surgery or spondylolisthesis), who had been referred for QF imaging to investigate CNSLBP using the same imaging protocol, were recruited. Their imaging results were compared to those of 8 of the asymptomatic controls, following written informed consent to inclusion on the study. The controls were chosen from the Reference Database as being of similar age, sex and BMI to the patients. Their demographic information is shown in Table 1.

The design criterion for determining the sample size needed to establish a credible 95% reference interval (RI) is the ratio of the confidence interval (CI) width on the RI cut-point to the RI width. Practical values for this ratio suggested by Linnet range from 0.1 to 0.3 (Linnet, 1987). Using a conservative ratio of 0.15, with a 90% cut-off CI and a single 95% upper RI cut-point, we required 134 participants (SSS software v.1, Wiley-Blackwell, Chichester United Kingdom). To allow for tracking failure in approximately 10% of sequences, we rounded the sample size up to 148. However, assuming a non-Gaussian distribution for at least some of the reference data, we employed the non-parametric RI methodology recommended in the Clinical Laboratory Standards Institute guidelines, for which the minimum recommended sample size is 120 (CLSI, 2008). This was therefore selected as the minimum population for the Reference Database study.

The QF protocol for image acquisition and analysis procedures as been detailed in previous studies (Breen et al., 2012; Breen and Breen., 2018; du Rose et al., 2018; Breen and Breen, 2020). In brief however, participants were guided through a standard active weight-bearing flexion and return motion task. This was designed to reduce behavioural variations in participant bending, while controlling the speed and range of motion in a reproducible way. During this controlled motion, low dose fluoroscopic recordings of L2-S1 levels during continuous motion were acquired using a Siemens Arcadis Avantic digital C-arm fluoroscope (Siemens GMBH) at 15 frames per second. To achieve this, participants stood with their right-hand side next to a motion testing platform (Atlas Clinical Ltd. Lichfield, United Kingdom), which guided them through a 60° bending arc at 6°/s during both flexion and return phases (Figure 1). Participants were positioned in a comfortable upright stance with the centre of rotation of the motion platform in line with the disc space between the third and fourth lumbar vertebrae (This position was confirmed by single short pulse fluoroscopic images and the use of radiopaque markers temporarily aligned with the platform’s centre of rotation.) A sacral brace and a belt around the hips of participants were used to minimise pelvic motion and keep the spine in the field of view throughout the bending sequence. This was to ensure the best field of view for all the segments to be conveniently imaged throughout the whole range of motion.

Before the acquisition of the QF images, participants undertook 3 practice bends. These standing movements, bending to 20° flexion and return, were followed by 40-degree and 60-degree bends. This ensured that participants could perform their recorded motion confidently and smoothly.

A previously validated semi-automated tracking process was used to determine the position of each vertebra (L2, L3, L4, L5 and S1) within each image recorded during the flexion and return trials (Breen et al., 2012). This process has been shown to have an accuracy for measuring intervertebral RoM of 0.52°, (du Rose and Breen, 2016), inter-and intra-observer repeatability ranging from ICC 0.94–0.96 and SEM 0.23°–0.61° and acceptable intra-subject repeatability (ICC 0.96, MDC over 6 weeks, 60%) (Breen et al., 2006; du Rose and Breen, 2016; Breen et al., 2019b).

Rotations were extracted from the positions for each of the tracked vertebrae (L2, L3, L4, L5 and S1, Figure 2) in each of the QF images throughout the flexion and return movement. Changes in the intervertebral angle from the starting position at each level (L2-L3, L3-L4, L4-L5, and L5-S1) over time were then computed. The motion outputs were separated into two phases, the flexion phase, and the return to neutral phase. To standardise the representation of motion across all participants, the L2-S1 angle was normalized to a percentage of its range of motion (RoM). Thus, during the flexion phase, standing was defined as 0% RoM and maximum flexion as 100% RoM, while in the return phase, 0% RoM was defined as maximum flexion and 100% RoM as being returned to the original reference position.

Changes in intervertebral angles were then interpolated to obtain each intervertebral motion segment’s rotation for every 1% increment of the L2–S1 RoM. The segmental contribution of each intervertebral level as a percentage of the change in L2–S1 angle was then computed at every increment.

For the Reference Database study, the share of intervertebral segmental motion was calculated for all participants for each level throughout the bending task. Statistical Parametric Mapping (SPM) was then used to compare the whole kinematic time-series between levels’ contributions to motion for both the flexion and return sequences (Friston et al., 2007). SPM analysis is an open-source spm1d package (available from www.spm1d.org) based on Random Field Theory, and has been validated for 1D data (Adler et al., 2007; Pataky et al., 2016; Pataky, 2016). Following normality testing, custom Python programs (Python version 3.8) were used to conduct parametric two-tailed, two-sample t-tests across the time series. Statistical significance occurs when the SPM curves cross the critical threshold node at any time, taking into account that each time point is related to those on either side (Friston et al., 2007; Papi et al., 2020). Where multiple adjacent points of the SPM curves exceeded the critical threshold, the associated p-values were calculated using Random Field Theory.

For the Patient-control secondary analysis, SPM analysis was conducted using non-parametric two tailed t-tests. This compared segmental contributions to bending between patients and controls throughout the motion. Previous measures of segmental contribution have been shown to have high observer reliability and acceptable intrasubject repeatability over 6 weeks (Breen et al., 2019b; To et al., 2020).

The maximum intervertebral ranges throughout flexion and return motion (means) for the Reference Database study group and the Patient-control sub-study group are shown in Table 2. Maximum change in L5-S1 RoM was significantly less than the controls in the Patient-control sub-study.

Figure 3 shows statistically significant differences in contributions to bending during the motion, both between and within levels. Each intervertebral level had its own characteristic motion signature across the Reference Database study population, with significant differences (p < 0.05, noted from the lack of overlap of the 95%CI bands) between each level’s contribution throughout most of the motion. It is also notable that these paths are in a state of constant change as the motion progresses, although all levels exhibit more uniform motion sharing in the return phase than in the flexion phase. In addition, there is a negative contribution to motion of L5-S1 at the beginning of flexion (Figure 3A). This is expected as participants attempt to move their hips back to keep the centre of mass over the feet.

The SPM analysis reported in Figure 4 reveals these differences to be highly significant (p < 0.001) between levels for almost all data points across the motion for both flexion and return in the Reference Database study cohort, confirming the presence of discrete motion paths for each motion segment. During the outward flexion phase of movement, the superior lumbar motion segment of each pair (Figure 4) consistently contributed more to the range of motion, exceeding the critical value for 50–99% of the task. In addition, the L2-3 vs. L3-4 motion segment combination also had a supra-threshold cluster at the end of flexion where the inferior motion segment contributed more (p = 0.008).

During the return to upright position phase of the task, in 3 of the 6 inter joint combinations, the inferior motion segment of the pair constantly contributed significantly more to the return phase of bending (p < 0.001). The exceptions were “L3-4 vs. L5-S1” and “L4-5 vs. L5-S1”, where the superior motion segment contributed a greater amount (p < 0001), and at “L2-3 vs. L5-S1”, where initially (between 5–40% of the RoM) L5-S1 contributed more (p < 0.001). In the late stages of bending (at approximately 100% of RoM), L2-3 contributed more (p < 0.001).

The motion contributions in the secondary analysis are shown in Figure 5. These subjectively demonstrate differences in the motion sharing patterns between patients and controls, especially at L5-S1. Verification of these differences can be seen in the non-parametric SPM analysis provided in Supplementary Material III.

Figure 6 compares the motion sharing patterns for all 8 patients and 8 controls in the secondary analysis. There was little difference between patients and controls in terms of motion sharing at most intervertebral levels, although these have been found to differentiate patients from controls in passive recumbent studies (Breen and Breen, 2018). However, SPM analyses reveals that there are statistically significant differences between the groups’ motion share at L2-L3 during the return to neutral phase of the task (p < 0.001) and at the end range of L5-S1 motion (Flexion p = 0.012 and Return p = 0.004) (Figure 6).

To the best of the authors’ knowledge this study reports the largest database of continuous intersegmental lumbar spine kinematics during weight bearing in-vivo flexion and return, providing normative reference values for making patient-specific kinematic comparisons, for informing dynamic FE loading models, and to help identify biomarkers for CNSLBP (Zanjani-Pour et al., 2018; Breen and Breen, 2020). The Reference Database study used an established standardised protocol to measure the intersegmental contributions to motion from L2-S1 during weight bearing flexion and return bending–unlike most conventional recording of lumbar flexion, which depends on participant co-ordination for its consistency. Using this protocol, continuous change in mean proportional motion share was observed during both the flexion and return phases and revealed significant differences in the motion paths between levels. In addition, while the similarity between this study’s measures of lumbar segmental contributions and previous measures of lumbo-pelvic rhythms are interesting, these are not the same and should not be confused.

It would be appropriate to compare this database with previous fluoroscopy studies, however no one study has applied all the criteria required. We could find only four that attempted to employ completely continuous motion analysis (Okawa et al., 1998; Harada et al., 2000; Nagel et al., 2014; Aiyangar et al., 2015). This may, in part, account for the failure of studies that reported only quasi-static intersegmental motion to detect variations in the contributions of individual segments during bending (Wong et al., 2006). Only three used proportional motion (Teyhen et al., 2007; Nagel et al., 2014; Aiyangar et al., 2015), and none applied the degree of standardisation of participant motion during imaging used in the present study (Breen et al., 2012). Return phase motion (which is not represented as flexion in reverse) was reported in only 4 (Okawa et al., 1998; Harada et al., 2000; Teyhen et al., 2007; Aiyangar et al., 2015), while only 5 measured all levels from L2-S1 (Takayanagi et al., 2001; Lee et al., 2002; Wong et al., 2006; Ahmadi et al., 2009; Aiyangar et al., 2015). However, when comparing the segmental contributions derived from moderate or maximal flexion studies with continuous intervertebral motion studies, the distribution of sharing was found to be similar (Breen and Breen, 2020). Thus, the quasi-static spine kinematics literature, as reviewed by Widmer et al. (2019) exhibits a degree of consistency with more recent continuous motion studies in terms of lumbar intervertebral motion sharing.

The above considerations, plus the large number of participants in the Reference Database study, may account for the remarkably consistent motion sharing patterns during both outward and return continuous motion, despite some heterogeneity in participant characteristics. Although the age range in our sample was wide (21–70 years), body weight had an upper reference range of only 96 Kg, while weights of up to 119 Kg have been shown to be associated with substantially increased L5-S1 compression in flexed postures (Hajihosseinali et al., 2015). This may affect the segmental contribution at that level and was also noted in relation to RoM in the Widmer et al. (2019) review and in modelling studies by Zander et al. (2002). However, segmental contributions, once thought to be RoM-dependent, did not exhibit this in our Reference Database study, nor in other studies that included all segments from L2 to S1 (Miyasaka et al., 2000; Ahmadi et al., 2009; Aiyangar et al., 2015). Contribution patterns were also distinctly different in flexion and return, as one would expect with different phasing of trunk muscle activation (Ouaaid et al., 2013).

The differences between patient and control subgroups in the return phase also seem to complement those previously found in weight bearing studies that combined outward and return motion (Breen and Breen, 2020). Moreover, the standardised image acquisition protocol would seem to make it unlikely that abnormal patterns are attributable to artefact rather than motion pathology. However, it also raises the possibility that other individual factors, such as lumbar geometry, may have an influence, making clinical assessments based on motion sharing patterns alone inadvisable.

The differences between patients and controls in the secondary analysis may reflect differences in the respective roles of the deep multifidus and erector spinae muscle groups in people with CNSLBP (Wallwork et al., 2009) and/or passive tissue restraint. Two trends are particularly apparent in this study of weight bearing motion. Firstly, L5-S1 shares less motion in patients, albeit non-significantly until the end of range. This is also reflected in the reduced RoM of L5-S1. Secondly, L2-3 shares significantly more motion in patients during the return phase, although this is only apparent in the mid-ranges and would not be measurable when merely investigating segmental range. Alterations in the readiness of lumbar joints to move in CNSLBP patients is also reflected clinically in the Kinesiopathological Model of low back pain, and is considered to be an important factor in rehabilitation (Van Dillen et al., 2013).

This is the largest dataset available to date to present normative values for continuous segmental contributions to motion in the lumbar spine using variables that have been shown to distinguish patients with CNSLBP from asymptomatic controls (Ahmadi et al., 2009; Breen and Breen, 2020). Moreover, the Patient-control sub-study provides further evidence of a kinematic biomarker for nonspecific back pain. However, standardising the motion protocol involves a trade-off between natural motion and the repeatability necessary to make patient-specific comparisons. In terms of the latter, the methodology used has undergone extensive validation in terms of precision and validity (Breen et al., 2006; Breen and Breen, 2016; Breen et al., 2019b) and has previously been used in preliminary dynamic loading studies using FE modelling (Zanjani-Pour et al., 2018). Thus, further subject-specific estimates of joint loading using dynamic imaging may be expected to improve the sensitivity of subject-specific model-based lumbar spinal loading estimates (Byrne et al., 2020). However, like many other biomechanical studies that compare patients to controls, the sample size of our sub-study was small, which is a limitation that may be mitigated by further replication. In addition, while evidence suggests that magnitude of loading (beyond body weight), in vivo, does not have any significant effect on individual segmental contribution to motion (Aiyangar et al., 2015), biomechanical modelling should exercise caution if using this database to model unloaded or excessive loading states.

It would also have been useful for future biomechanical modelling studies if it had been possible to include the whole lumbar spine, but this could not be done owing to the limited image intensifier diameter. This is a problem with most current intensifiers which will be overcome as flat panel machines become more plentiful.

Further studies are needed, not only to replicate the present study’s findings, but also to explore the effects of other variables, as well as coronal plane motion and passive recumbent motion, where body mass and muscular contractions are mitigated. However, there is still considerable scope for elaboration of motion sharing studies of weight bearing flexion and return. For example, variations in pelvic tilt may be an important source of heterogeneity in light of the variations in the motion segment flexibility that the QF procedure aims to measure (Retailleau and Colloud, 2020).

The specialised motion frame apparatus used in the current work, in addition to standardising the velocity and range of bending, also partially stabilises the sacrum. This increases to varying degrees, the contribution of the lumbar spine to the flexion motion, regardless of the degree of lordosis or sacral inclination. Although the degree of this restraint is not standardised and depends on the individual’s natural lumbo-pelvic contribution to bending, this does not seem to disrupt the consistency of the resulting motion sharing patterns. Nevertheless, there is likely to be some relationship between lumbar sagittal shape and the motion contribution, albeit within the boundaries of the normative ranges of variation. This should be explored. Given that spine shape has been shown to influence a person’s preference for squatting or stooping during lifting tasks, it would be useful to determine the relationships between spine shape and dynamic loading stresses at individual levels based on their contributions to flexion and return motion (Pavlova et al., 2018).

It would also be useful to explore other kinematic indices in terms of motion contributions, as the present database provides only rotational data, and there is evidence that the translational component, although small, also affects inter-segment rotational stiffness (Affolter et al., 2020). However, in a previous study, we did not find it to differentiate nonspecific back pain patients from controls (Breen et al., 2018).

Finally, it is now timely to explore possible relationships between the motion variants that seem to be associated with CNSLBP and possible sources of nociception. As these may not necessarily involve disco-ligamentous micro-strain, it may be useful to explore muscular metabolic pain as a mechanism by including blood flow studies with those of motion contributions during bending.