Strong association between vertebral endplate defect and Modic change in the general population

Modic change (MC) is considered an independent risk factor for low back pain (LBP) but its aetiology remains unclear. In this cross-sectional, large-scale population-based study we sought to characterise associations between endplate defect (ED) and MC in a population sample of broad age range. The study population consisted of 831 twin volunteers (including 4155 discs and 8310 endplates) from TwinsUK. Lumbar T2-weighted MR images were coded for ED and MC. Total endplate (TEP) score was calculated at each intervertebral disc while receiver operating curves (ROC) were calculated to define critical endplate values predictive of MC. MC was detected in 32.1% of the subjects, with a significantly higher prevalence at lower lumbar levels (3.5% at L1/2-L3/4 vs. 15.9% at L4/5-L5/S1, p < 0.001). TEP score was strongly and independently associated with MC at each lumbar level (risk estimates from 1.49 to 2.44; all p ≤ 0.001) after adjustment for age, sex, BMI and twin pairing. ROC analysis showed a TEP score cut-off of 6 above which there was a significantly higher prevalence of MC. In conclusion, ED were strongly associated with MC at every lumbar level. These findings support the hypothesis that endplate defect is a major initiating factor for the cascade of events that may include disc degeneration (DD) and MC.

Globally, low back pain (LBP) is among the leading causes of disability 1 and is considered the largest single cause of disability in Western society, with lifetime prevalence of non-specific LBP estimated at 60-70% in industrialized countries 2 . With increasing patient demand in secondary and tertiary care, requesting appointments with rheumatologists, physical and rehabilitation medicine specialists and surgeons, LBP represents a highly relevant public health issue with enormous healthcare and socioeconomic costs 3 . The precise causes of such a burden have still not been clearly defined.
Modic change (MC) describes MR image signal change in subchondral and vertebral bone marrow adjacent to a vertebral body endplate. As visualized on T1-and T2-weighted MR imaging, MC can be divided into three different subtypes: Type 1, 2 and 3 [4][5][6] . MC is rarely observed adjacent to a healthy disc 7 and tends to be associated with structural change such as disc degeneration (DD) [7][8][9] , disc herniation 7,8,10 , Schmorl's nodes 7 and vertebral endplate defect 11 . Similar to association of intervertebral DD with LBP 12,13 , also MC is associated with it 14,15 , and is now considered to be an independent risk factor for LBP 16 .
Different theories have been suggested regarding the aetiology of these conditions, but the pathophysiology of MC is still not fully understood, and the role of structural bony defects as an initiating factor for MC is not confirmed. Lately, clinical studies have focused attention on endplate defects as having a possible role in the aetiopathogenesis of MC 17 Rajasekaran et al. 19 have reported that the presence of endplate defect was associated with DD in a clinical sample of 47 patients and 26 volunteers. Endplate defect evaluated on T1-weighted MR scans was classified into six categories according to severity of damage (type 1 to type 6), and total endplate (TEP) score was derived. Strong correlation between progressive grades of endplate defect and disc degeneration was shown, but unfortunately MC was not considered independently in the statistical analysis.
In a previous study from TwinsUK we have shown a strong association between endplate defect and DD in a large population based sample of volunteer twins 20 . As the association between endplate defect and DD was strong across the age spectrum, and as the endplate is positioned exactly between the intervertebral disc and the vertebral body, we sought to explore whether an additional association exists between endplate defect and MC.

Methods
Study population. The study population was part of the TwinsUK register of King's College London (www. twinsuk.ac.uk) 12 . Clinical information on sex, body mass index (BMI), and lumbar DD had been collected previously. All subjects have signed an informed consent form. The study was approved by St. Thomas' Hospital Ethics Committee and is in accordance with Declaration of Helsinki.
Magnetic resonance imaging and grading. Details of the MR scans, and grading of MRI findings, have been described previously 21 . Briefly, T2-weighted MR scans were performed on volunteer subjects unselected for LBP, using a Siemens (Munich, Germany) 1.0-T superconducting magnet. Sagittal T2-weighted images were   19 and total endplate scores for each disc were constructed by summing the endplate defect score of both rostral and caudal endplates in each functional spine unit 19 . Inter-rater agreement for the coding of endplate defect was calculated using Cohen's weighted kappa and Pearson's correlation. Agreement phase: An initial training phase was held in which an inter-rater agreement on endplate defect detection and grading of ≥0.85 was reached on at least 100 subjects and 1000 endplates, as reported previously 20 . Disagreements on coding scores were settled by discussion and consensus.
Grading of Modic change. The coding of MC was performed previously and has been described in detail elsewhere 22 . Briefly, a single observer (JHM) assessed all images without any prior knowledge of the clinical status. Inter-rater reliability was calculated when the second reader (SW) assessed randomly selected subset of images (n = 50). For the purpose of this study, MC was coded as absent (grade 0) or present (grade 1) for vertebral bone marrow adjacent to each endplate separately. MC detected only in one sagittal slice was not included. MC was considered to affect a lumbar disc if at least one of its endplates were affected.
Grading of disc degeneration. Disc degeneration has been evaluated earlier using Pfirrmann classification 23 . The classification consists of progressive evaluation of disc degeneration, assessed from grade 1 (homogeneous disc with bright hyperintense white signal intensity and normal disc height) to grade 5 (inhomogeneous disc with hypotense black signal intensity and no more detectable difference between the nucleus and annulus). In accordance with previous studies 19,20 , grade 4 or greater was considered an indicator of a degenerate disc.

Statistical analysis.
Inter-rater agreement for the coding of endplate defects was calculated using Pearson's correlation and Cohen's weighted kappa as reported previously 20 . Critical scores were calculated for MC on the same lumbar level using Receiver Operating Curves (ROC)-analysis. Cumulative link mixed models and linear mixed models were used to analyse association between endplate defects and MC, adjusting for covariates and family structure. Kaplan-Meier survival analysis and Cox proportional hazards models analysis were carried out to infer any age-dependent relationship between endplate defect and MC. All calculations were performed in R using packages "psych", "survival", "ordinal", "lme4", and "OptimalCutpoints". Table 2 summarises the twin sample and its characteristics. In total, there were 831 subjects (mean age 54 ± 8 years, 95.8% female) in the study. Male subjects in the sample were similar to females as regards age and prevalence of MC, but they had significantly higher TEP scores in the lumbar spine compared to females (median [IQR] 29 [20,36] vs 22 [19,28], p = 0.001 respectively).    Table 3. In summary, grade 2 was the most prevalent grade and grade 6 was significantly more prevalent at the lower lumbar levels than upper (p < 0.001). MC was present in 267 (32.1%) subjects. Of all the endplates being evaluated, we found 646 (7.8%) MC in total. Figure 2 shows the prevalence of MC by individual endplate levels. The prevalence of MC was significantly higher at the lower (L4-S1) compared to upper (L1-L4) lumbar levels (p < 0.001).

Results
Relationship between TEP score and MC. The prevalence of MC showed a significant correlation with increasing TEP score. Using multivariate linear mixed models to allow for covariates, TEP score was found to be strongly and independently associated with MC (risk estimates from 1.49 to 2.44 and p-values p < 0.001, at all lumbar levels, Table 4). Associations with other factors in the final multivariate models (including age, sex and BMI) were significant only at a few lumbar levels. ROC analysis indicated a critical TEP score of 6, above which there was a significantly higher likelihood of MC. Considering each lumbar level, similar cut-off points were    found throughout the lumbar spine; TEP score of seven for L2/L3, and six for each of L1/L2, L3/L4, L4/L5, and L5/S1. Of all MC, 287 (81.3%) were at lumbar levels with TEP score ≥6 (Table 5). Regarding more precise endplate risk evaluation, the probability of having MC adjacent to the endplate having TEP score ≥6 was 6.4% at L1/L2, 12.2% at L2/L3, 21.9% at L3/L4, 48.5% at L4/L5, and 45.1% at L5/S1. Pearson's correlation between TEP score and MC was statistically significant for every disc level. Considering the whole lumbar spine, if TEP score was ≥6 at any level, the probability of having MC at any level was 81.3%. At each  Figure) and <6 (denoted 'Teps -No' in the Figure). The probability is significantly increased in TEP score positive age subgroups at each disc level. The probabilities increase with age, and the influence of TEP score on MC is least at L5/S1. HR = hazard ratio. lumbar level, there was a highly statistical significance in MC (p < 0.001) between those with and without TEP score ≥6 (except for L2/L3 TEP score ≥7).
A survival analysis paired with Cox proportional hazards model analysis provided the probabilities of having MC in the presence of TEP score <6 or ≥6, by age subgroups (Fig. 3). The probabilities increased significantly in TEP score ≥6 for all age strata.

Discussion
This is the first large, population-based study examining the association between vertebral endplate defect and MC in the lumbar spine. Results showed that the presence of vertebral endplate defect in the lumbar spine was strongly associated with MC in the bone marrow adjacent to the endplate. When analysed by ROC, we found that a TEP ≥6 strongly predicted risk of MC.
Interestingly, our previous findings from the same sample from TwinsUK registry showed TEP ≥ 5 to be predictive for disc degeneration 20 which may indicate that endplate defect, disc degeneration and MC are strongly associated with each other and it is highly likely that the same pathological process is involved. Moreover, our results are also in accord with the findings from another experiment in which endplate defect was correlated with DD in a clinical sample of 47 patients and 26 volunteers 19 . A cut-off score of 6 was found to be predictive of DD, after which a rapid response with a great increase of likelihood of DD for each unit increase in TEP score was shown, exactly as in our study with MC.
In a clinical study with 1-year follow-up, MC type 1 was associated with both decreasing the disc height and increasing the size of endplate defects 17 . In another longitudinal clinical study sample DD, endplate defects and MC were found to be significantly associated with each other. Endplate defect grade ≥4 was found to be a risk factor for both DD and MC progression, with MC being the last MR feature to develop in this process 18 . All these results seem to build on the same line of thoughts (Fig. 4). With regards to the motion segment in the spine, the cartilaginous endplate is considered to be a mechanically vulnerable structure 24 , which helps to equalize loading between the vertebral body and the disc 25 . At surgery, herniated disc material is reported to be found to contain cartilage endplate more frequently in those patients also manifesting MC 26 . Moreover, cartilage endplate stripping increases the permeability of the endplate and thus facilitates passage between the vertebral body and the intervertebral disc 26 .
That endplate defect may lead to both MC and DD can be explained by the physical location of the endplate; being positioned between the disc and vertebral body. Biomechanical studies suggest that microfractures at the vertebral endplate 27 allow inflammatory cytokines to pass from the intervertebral disc to the bone marrow to initiate MC 28 , while any physical defect in endplate may allow transport of pro-inflammatory mediators from the disc to the vertebral body, leading to the oedema characteristically seen in MC 29 .
It is also plausible that an endplate defect creates the basis for low-grade virulent anaerobic bacteria (Propionibacterium acnes, P. acnes) to enter the disc from the vertebral body and give rise to slowly developing infection 28 . In this context, intervertebral disc infected with anaerobic bacteria have been shown to be more likely to develop adjacent vertebral body MC than those without 30 .
Biomechanically, endplate defect has been shown to lead to decompression of the adjacent intervertebral disc nucleus, which in turn could drive the degeneration of the disc through several mechanisms by (i) increase of shear forces acting on the annulus 31 ; (ii) decrease of stability at that spinal level 32 inducing disruptive changes in the annulus; (iii) altered matrix synthesis 33 ; and (iv) vascularisation of the nucleus pulposus with autoimmune changes 30 .
There are some limitations we would like to highlight in this study. The TwinsUK sample shows a marked female predominance, for historical reasons. For that reason, every trait was examined separately by gender to determine how best to account for differences in the analysis. TwinsUK participants have been shown to be representative of singletons for a wide range of lifestyle and demographic traits 34 . For historical and funding reasons, only T2-weighted scans were available and, therefore, the evaluation of MC types was not possible. This study is cross-sectional by design, and our results support the hypothesis that endplate defect may be the initiating factor for MC and intervertebral disc degeneration. As a suggestion for further improvement we suggest designing a longitudinal study of endplates, DD and MC to unequivocably confirm the direction of such causality. Another relevant strength is the thorough training phase undergone by the two evaluators (JHM and MR) leading to almost perfect inter-rater agreement for endplate defect: kappa value of 0.86, Pearson correlation 0.864, with correlation and kappa being employed in conjunction to uncover nonrandom examiner error, as in Hunt 35 .

Conclusions
To conclude, this study performed on a large population-based sample confirmed that endplate defect is strongly and independently associated with DD and MC, and that this relationship is evident in adults across the age spectrum and at all lumbar levels. As the same conclusions were reported for ED and DD in our previous study in the same population using the same methods, we imply that causal effects can occur between ED, DD and MC. A longitudinal study of twins' MR spine scans and endplate is currently under way to better confirm the order of events.