Glucocorticoids induce specific ion-channel-mediated toxicity in human rotator cuff tendon: a mechanism underpinning the ultimately deleterious effect of steroid injection in tendinopathy?
- Benjamin John Floyd Dean,
- Sarah Louise Franklin,
- Richard J Murphy,
- Muhammad K Javaid,
- Andrew Jonathan Carr
- Nuffield Department of Orthopaedics, Rheumatology and Musculoskeletal Sciences (NDORMS), Botnar Research Centre, Institute of Musculoskeletal Sciences, Nuffield Orthopaedic Centre, Oxford, UK
- Correspondence to Dr Benjamin John Floyd Dean, Nuffield Department of Orthopaedics, Rheumatology and Musculoskeletal Sciences (NDORMS), Botnar Research Centre, Institute of Musculoskeletal Sciences, Nuffield Orthopaedic Centre, Windmill Road, Oxford, OX3 7LD, UK,
- Received 8 October 2013
- Revised 3 March 2014
- Accepted 3 March 2014
- Published Online First 27 March 2014
Background Glucocorticoid injection (GCI) and surgical rotator cuff repair are two widely used treatments for rotator cuff tendinopathy. Little is known about the way in which medical and surgical treatments affect the human rotator cuff tendon in vivo. We assessed the histological and immunohistochemical effects of these common treatments on the rotator cuff tendon.
Study design Controlled laboratory study.
Methods Supraspinatus tendon biopsies were taken before and after treatment from 12 patients undergoing GCI and 8 patients undergoing surgical rotator cuff repair. All patients were symptomatic and none of the patients undergoing local GCI had full thickness tears of the rotator cuff. The tendon tissue was then analysed using histological techniques and immunohistochemistry.
Results There was a significant increase in nuclei count and vascularity after rotator cuff repair and not after GCI (both p=0.008). Hypoxia inducible factor 1α (HIF-1α) and cell proliferation were only increased after rotator cuff repair (both p=0.03) and not GCI. The ionotropic N-methyl-d-aspartate receptor 1 (NMDAR1) glutamate receptor was only increased after GCI and not rotator cuff repair (p=0.016). An increase in glutamate was seen in both groups following treatment (both p=0.04), while an increase in the receptor metabotropic glutamate receptor 7 (mGluR7) was only seen after rotator cuff repair (p=0.016).
Conclusions The increases in cell proliferation, vascularity and HIF-1α after surgical rotator cuff repair appear consistent with a proliferative healing response, and these features are not seen after GCI. The increase in the glutamate receptor NMDAR1 after GCI raises concerns about the potential excitotoxic tendon damage that may result from this common treatment.
Shoulder and upper limb pain is the second most frequent cause of chronic musculoskeletal pain and is usually caused by abnormality of the rotator cuff tendons.1 ‘Rotator cuff tendinopathy’ most commonly affects the supraspinatus tendon as a result of the high level of stress it experiences during daily activities and sport. The cellular and molecular mechanisms underlying rotator cuff tendon degeneration have become better described in recent years, with modern theories tending to favour the aetiology of intrinsic tendon failure over extrinsic mechanical impingement.2 The neuronal changes in tendinopathy appear consistent with a failed healing response and a definite upregulation in the excitatory glutaminergic system has been shown.3 The majority of this work has been carried out in patients with Achilles and patellar tendinopathy, little is known about the role of the nervous system and the importance of neuronal mediators in rotator cuff tendinopathy.
Current treatments for rotator cuff tendinopathy include non-operative measures such as rest, analgesia, physiotherapy and glucocorticoid injections.1 In the UK over 500 000 intra-articular glucocorticoid injections are administered per year in the primary care setting alone,4 but very little is known about the way in which local glucocorticoid may affect the rotator cuff tendons in vivo. Surgical rotator cuff repair is increasingly performed in patients with symptomatic rotator cuff tendon tears.1 Metabolic changes involving the glutaminergic system have been demonstrated in humans and animal models of tendon repair,5–7 but no study has ever analysed the effects of surgical rotator cuff repair on tendon tissue in humans.
In this context our main aim was to assess the histological and immunohistochemical effects of these common treatments on the rotator cuff tendon. The secondary aim of the study was to assess whether there were any significant histological differences between the glucocorticoid group (tendinopathy) and the surgical rotator cuff repair group (full thickness tendon tears).
Materials and methods
Supraspinatus tendon biopsies were taken from 8 patients undergoing rotator cuff repair and from 12 patients undergoing subacromial glucocorticoid injection for rotator cuff tendinopathy. All patients were referred to a specialist upper limb out patient service and were aged between 35 and 65 (mean age=53 years and SD±8 years). All patients in the rotator cuff repair group had symptomatic degenerative full thickness supraspinatus tears, diagnosed by ultrasound and confirmed at surgery. Tear size was based on the accepted classification of Post et al.8 Measuring the tear at its longest anteroposterior diameter, tear size was classified as follows: small ≤1, medium >1 and ≤3; large >3 and ≤5; and massive >5 cm. All patients in the glucocorticoid injection group were diagnosed with rotator cuff tendinopathy by the clinical evaluation of the senior consultant surgeon (AJC), and treatment was with glucocorticoid injection. All patients had subacromial impingement pain, ‘painful arc’ and positive Hawkins’ and Jobe’s tests. Structural integrity of the rotator cuff was assessed by ultrasound in all patients. A patient completed questionnaire for the Oxford Shoulder Score (OSS),9 a well validated and widely used clinical outcome measure, was undertaken at both the week 0 and week 7 time point in both treatment groups.
Patients were excluded if there was a history of previous shoulder surgery, any other significant shoulder pathology not involving the rotator cuff (osteoarthritis, frozen shoulder, instability or previous fracture of the humerus/clavicle/scapula), more than three previous glucocorticoid injections, a glucocorticoid injection within 6 weeks of the treatment intervention and systemic steroid use. The diagnosis of other significant shoulder pathology was made on the basis of clinical history, clinical examination and radiological findings by the treating consultant surgeon (AJC).
Glucocorticoid injection consisted of one ultrasound-guided injection of 40 mg Depo-Medrone and 4 mL of 2% lignocaine into the subacromial bursa performed by one trained surgeon in the outpatient department using aseptic precautions. Rotator cuff repair was performed by a single consultant surgeon (AJC) using an arthroscopically assisted mini-open approach.
Sampling, sectioning and fixation
A percutaneous ultrasound-guided core biopsy technique was used to acquire the supraspinatus tendon tissue samples in both treatment groups. This validated technique has been developed at our institution and is described in more detail elsewhere.10 All longitudinal tendon biopsies were taken at a consistent location approximately 5–10 mm posterior to the rotator interval under ultrasound guidance. A single biopsy was obtained at each time point with the approximate size of each biopsy being 2 mm×2 mm×10 mm. In the glucocorticoid injection group biopsies were taken immediately prior to and 7 weeks following glucocorticoid injection. In the rotator cuff repair group biopsies were taken from the newly created footprint at the time of surgery and 7 weeks following rotator cuff repair. All tendon repairs were intact at the 7-week time point as assessed ultrasonographically. There were no local or systemic complications as a result of the biopsy procedure.
All fresh samples were immediately fixed in 10% buffered formalin and allowed sufficient time for the fixative to penetrate the tissue. On full fixation the tissue samples were processed using a Shandon Pathcentre Tissue Processor (Thermo Scientific, Cheshire, UK) and embedded in paraffin wax.
Tissue was sectioned at 4 µm using a rotary RM-2135 microtome (Leica Microsystems Ltd, Bucks, UK), and collected onto adhesive glass slides.
Histology and immunohistochemical procedures
An H&E stain was performed on samples from every patient to assess the general structural condition of the tissue samples and to confirm they were tendon (figure 1). Prior to antibody staining, tissue sections were taken through deparaffinisation and target retrieval steps using an automated PT Link (Dako, Denmark). All antibody staining was performed using the EnVision FLEX visualisation system with an Autostainer Link 48 (Dako, Denmark). Antibody binding was visualised using the FLEX DAB (3,3′-diaminobenzidine) substrate working solution, and haematoxylin was used as a counterstain. Details of antibodies used and their working concentrations, as well as all staining protocols, can be found in online supplementary appendix 1. Human brain tissue sections (Neuromics, Minnesota, USA) were used as positive controls for each neuronal antibody staining run. Isotype control staining was also performed. Every tissue sample had one tissue section stained per antibody. The staining for each antibody was carried out simultaneously on the same run for all the sections.
Image analysis and statistics
The assessment of the microscopic appearance of the slides stained for H&E was undertaken using ImageJ (public domain software, National Institutes of Health, Bethesda, Maryland, USA: http://rsb.info.nih.gov/ij/). The nuclei count programme was validated against the counts of two blinded observers using Bland Altman plots and the calculation of the Intraclass Correlation Coefficient (ICC 0.971).
The assessment of the microscopic appearance of the slides immunostained with DAB was undertaken by a single blinded investigator. A Nikon Inverted Microscope using Axiovision software was used to capture images starting at one corner of the tissue moving systematically in a horizontal–vertical–horizontal manner until the tissue section was exhausted using ×100 magnification with oil immersion. A minimum of 10 images were captured for each tissue section. Each antibody set was imaged using the same exact light intensity at the same sitting. The images were then analysed using an ImageJ algorithm in order to quantify the percentage of DAB staining present in each image. For each antibody stain DAB threshold levels were manually determined and kept constant for each antibody dataset to ensure that results were comparable. This mode of image analysis using colour convolution has been well described in several other studies11–13 which have shown that measure of percentage area stained correlates well with the subjective grading systems used by experienced histopathologists.14 ,15 The percentage area stained was adjusted for cell number using the validated cell counting programme for antibodies which resulted in significant staining of the resident fibroblast population. Positive vessel counts were carried out manually.
Statistical analysis was carried using SPSS (IBM Corp. Released 2011. IBM SPSS Statistics for Windows, V.20.0. Armonk, New York, USA: IBM Corp.). Statistical significance was set at a level of p<0.05 unless otherwise stated. Histograms for all data sets were analysed, confirming that all the immunohistochemical data was non-parametric in distribution. The Wilcoxon signed-rank test was used to test for differences between the paired samples, while the Mann-Whitney U test was used for testing for differences between unpaired samples.
Metrics: reliability and validity
Interobserver and intraobserver reliability for this method have been tested using the ICC. These compared the intrarater and inter-rater observations. The ICCs relating to inter-rater measurement were between 0.724 and 0.909 for three full patient sets (minimum 70 patients). The intraobserver ICC was 0.830 after 1 month. We compared the quantitative analysis results with blinded subjective analysis as to the presence (1) or absence (0) of DAB staining as the use of subjective grading is a well validated current method of assessing staining.16 Significant differences were found between groups (p<0.0001) for all antibody datasets analysed, confirming the criterion validity of the method. The quantitative results for CD34 were also correlated against the results of the well-validated method of manual vessel counts yielding Spearman correlation coefficients of 0.853 and 0.839. The quantitative results for Proliferating Cell Nuclear Antigen (PCNA) and p53 revealed a strong correlation between percentage area stained per cell and percentage positive cells as counted manually (Spearman r of 0.943 and 0.967, respectively).
The mean age of the rotator cuff repair group was 58 years (SD±6) and the mean age of the glucocorticoid injection group was 49 years (SD±5). The length of functional and pain symptoms ranged from 8 months to 10 years (table 1). In terms of sporting activity one of the glucocorticoid injection group was a professional tennis coach, while five others in the glucocorticoid injection group had a sporting pastime (golf, horse riding, kayaking, swimming and darts). One patient in the rotator cuff repair group had a sporting pastime (squash). Prior to surgery, and 7 weeks of follow-up, each patient completed a questionnaire for the OSS,9 a well validated and widely used clinical outcome measure. There were no significant changes in OSS after rotator cuff repair or glucocorticoid injection compared to baseline. The demographics of the study cohort are summarised in table 1.
Basic histology and cell viability
There was a significant increase in nuclei count, the SD of nuclei count and vascularity (mean vessel count) after rotator cuff repair and not after glucocorticoid injection (p=0.008, 0.002 and 0.008 respectively). Hypoxia inducible fatcor 1α (HIF-1α) and cell proliferation were only increased after rotator cuff repair and not after glucocorticoid injection (p=0.03 for both results; figures 2 and 3). Protein 53 (p53) was significantly increased after both treatments (p=0.03 for both). In terms of differences between the groups at baseline 0 weeks, vascularity (mean vessel count) was significantly reduced in the rotator cuff repair group versus the glucocorticoid injection group (p=0.03). Figure 2 depicts the quantified protein expression changes seen relating to PCNA and HIF-1α. Table 2 demonstrates the basic histological characteristics and quantified immunohistochemistry of tissue viability. Figures 1 and 3 show photomicrographs relating to basic histology (H&Es) and cell viability (PCNA and HIF-1α).
The N-methyl-d-aspartate receptor 1 (NMDAR1) glutamate receptor was increased after glucocorticoid injection and not after rotator cuff repair (p=0.016; figures 2 and 3). An increase in glutamate (p=0.04 both groups) and the receptor metabotropic glutamate receptor 1 (mGluR1; p=0.016 for rotator cuff repair and p=0.04 for glucocorticoid injection) was seen in both groups following treatment. An increase in the receptor mGluR7 was only seen after rotator cuff repair (p=0.016). The substance P receptor ((neurokinin receptor 1 (NK-1)) and the nerve growth factor (NGF) receptors (P75 low affinity NGF receptor (p75) and tyrosine receptor kinase A (TrkA)) were all significantly increased after rotator cuff repair and not after glucocorticoid injection (p=0.03, 0.014 and 0.03, respectively). The neural marker Protein Gene Product 9.5 (PGP9.5) was also significantly increased after rotator cuff repair and not after glucocorticoid injection (p<0.001). NMDAR1, mGluR3 and mGluR4 were all significantly increased in the week 0 rotator cuff repair group versus the week 0 glucocorticoid injection group (p=0.006, 0.005 and 0.005, respectively). In descriptive terms, a large number of the markers were predominantly expressed by the resident tenocyte population (NMDAR1 and mGluRs). The neurotrophin NGF and the neuropeptide substance P and their receptors (TrkA, p75 and NK-1) were predominantly expressed around blood vessels and not by the resident fibroblasts. The markers for new nerve growth and general nerves (GAP43 and PGP9.5, respectively) were largely expressed around vascular structures, but also to some degree by the resident fibroblasts. The numerical results relating to the different neuronal antibody groups are detailed in table 3. Figure 3 shows photomicrographs relating to NMDAR1 and glutamate. There were no obvious differences between men and women in terms of both baseline histology and their histological response to treatment.
This study describes the profound histological and immunohistochemical changes involving the glutaminergic system that occur in the supraspinatus tendon of patients undergoing rotator cuff repair or glucocorticoid injection. The most important finding is that NMDAR1 was significantly increased in the glucocorticoid injection group and not in the rotator cuff repair group. This is in context of the proproliferative and proangiogenic changes that were seen after rotator cuff repair and not after glucocorticoid injection. The increase in glutamate, alongside the increase in cell proliferation and vascularity, seen after rotator cuff repair in our study is consistent with previous animal studies.5 ,6 It is interesting that despite the increase in glutamate there is no significant increase in NMDAR1 after rotator cuff repair. In light of the emerging clinical evidence demonstrating poorer outcomes and a higher risk of recurrence after glucocorticoid injection in tendinopathy,17 our study provides evidence of a potential mechanism by which glucocorticoid may bring about tendon damage and possibly poorer long-term clinical outcomes.
Glucocorticoid and a plausible mechanism of harm in tendinopathy
The clinical evidence for the use of glucocorticoid in shoulder pain is not convincing, many trials have shown only short-term benefits with no long-term gains.18 ,19 Emerging evidence points to poorer long-term outcomes associated with glucocorticoid injection in the treatment of tendinopathy.17 The mechanisms of action of glucocorticoids are multiple, highly complex and incompletely understood20; one important pathway involves the activation of specific cytoplasmic glucocorticoid receptors which then migrate to the cell nucleus to affect gene transcription. Generally glucocorticoids are thought to be anti-inflammatory but the reality may not be so simple.21 Our understanding of the tendon tissue effects of glucocorticoid injection relies almost exclusively on in vitro and animal in vivo work. Several studies have shown that glucocorticoid is antiproliferative, reduces type I collagen formation and has cytotoxic effects on tendon cells.22 The effects of glucocorticoid on the mechanical properties of tendon are conflicting.22 Only two previous studies have analysed the effects of glucocorticoid injection on in vivo human tendon. The study by Lee and Ling23 on the Achilles demonstrated reduced collagen organisation and increased collagen necrosis following glucocorticoid injection. While the more recent study by Poulsen et al24 showed that glucocorticoid injection-induced fibroblast senescence in human rotator cuff tendon. In vitro studies have demonstrated that apoptosis and oxidative stress are increased in rotator cuff tendinopathy.2
Recent research suggests that this glucocorticoid-induced cell damage and cell death may be mediated via NMDAR,25 ,26 while glucocorticoids have been shown to mediate the stress-induced extracellular accumulation of glutamate.27 The exacerbation of neuropathic pain via glucocorticoid receptor and NMDAR activation has been demonstrated in an animal model. Glucocorticoid appears to increase the susceptibility of cells to oxidative stress and induce apoptosis.28 It is possible that cell death seen in tendinopathy is glutamate-induced and mediated by p53.29 The transcription factor p53 is important in tendon healing30 and the increase after rotator cuff repair is not unexpected. However the increase in p53 seen after glucocorticoid injection may be less innocent and may be a marker of glucocorticoid-induced cell senescence and death.24 The increase in HIF-1α after rotator cuff repair may be a protective response and it is interesting that this is not seen after glucocorticoid injection.31
Glutamate and its key role in painful tendinopathy
Glutamate is a vital amino acid which has been implicated in many cellular processes including aerobic and anaerobic cell metabolism, collagen synthesis, neurotransmission and excitotoxic cell death.32–34 The finding of raised glutamate levels and an upregulated glutaminergic system in human tendinopathy has been well documented in recent years3 ,35; the upregulation of NMDAR1 in painful tendinopathy has also been demonstrated.3 Increased glutamate levels have also been correlated with pain intensity in other musculoskeletal conditions such as trapezius myalgia.36 ,37 The underlying mechanism behind these glutaminergic changes does however remain unclear. It is likely that glutaminergic homoeostasis is important in tendon, involving complex balances between the variety of both inotropic and metabotropic receptors. Our findings of increased NMDAR1 and metabotropic receptors mGluRs3/4 in the rotator cuff repair group week 0 versus the glucocorticoid injection group week 0 is consistent with this, representing the glutaminergic changes occurring in the increased structural tendon failure of the rotator cuff repair group.
The majority of work relating to glutamate has been carried out regarding its role in central nervous system. Glutamate is a key factor in the mechanism of hypoxic-ischaemic neuronal cell death. Glutamate induced neurotoxicity may be largely mediated by a toxic influx of extracellular calcium and this may be blocked by the antagonism of the NMDA subtype of glutamate receptor.33 The NMDA-receptor activated ion channel is a major route by which glutamate induces the toxic calcium influx. It is known that group 1 metabotropic receptors (mGluR1/5) increase NMDAR activity, while group 2 (mGluR2/3) and group 3 receptors (mGluR4/6/7/8) decrease NMDAR activity.38 Thus the complex interplay between the different components of the glutaminergic system is pivotal in determining the overall tissue response. Age, vascular risk factors and metabolic disease are all strongly associated with rotator cuff tendinopathy.39 ,40 This adds weight to the idea that age-related metabolic change is critical in the pathogenesis of rotator cuff tendinopathy. It may be hypothesised that the increased age-related oxidative stress41 results in a decline in metabolic function,42 and that this results in a glutamate-induced excitotoxicity of tendon.
Proliferative tendon healing is seen after rotator cuff repair
Generally speaking operative treatment is reserved for when non-operative measures have failed.1 There are a variety of operative strategies available including subacromial decompression, rotator cuff repair and several types of shoulder arthroplasty. Both patient and surgeon factors play an important role in deciding on an appropriate treatment strategy for the individual. Rotator cuff repair is a highly successful procedure in terms of clinical outcomes but rerupture rates remain high.43 ,44 Several studies have shown that the clinical outcome is significantly poorer if the rotator cuff repair fails.43 ,45 Tendon healing occurs with sequential inflammatory, proliferative and remodelling phases.46 Fibroblast proliferation, angiogenesis and nerve in-growth are all important in the healing process47; this is consistent with our findings of increased vascularity and PGP9.5 expression after rotator cuff repair. The immunopositive reactions of some fibroblasts to PGP9.5 is of interest and may relate to a subpopulation of type B synoviocyte-like cells that have been repeatedly seen in the horse48; of note PGP9.5-immunopositivity has previously been demonstrated in the human fibroblasts.49 The increase in the NK-1 and TrkA receptors, seen specifically around vascular structures, points towards the importance of both substance P and NGF in the process of neurovascular ingrowth.50 The p75 receptor was strongly expressed by both vascular structures and the resident tenocytes; this is consistent with the work of Bagge et al51 which demonstrated the presence of tenocytes immunopositive to the p75 receptor in the human Achilles tendon. The metabolic activity of a healing tendon after acute injury is hugely increased6 and fibroblasts are metabolically adapted for survival in this hypoxic environment.52 The elevated lactate metabolite present in the acutely traumatised tendon is also a potent stimulant for collagen synthesis.53 The upregulation of metabolites including glutamate demonstrated after Achilles tendon repair in humans7 is consistent with this increased metabolic activity and this appears likely to be important in the increased cell proliferation present in healing tendon. The relative quantified tissue changes are generally greater after rotator cuff repair than glucocorticoid injection. This is unsurprising given the relative extent of the two treatments, with rotator cuff repair involving major mechanical change following the surgical reconstruction of the supraspinatus footprint.
Study strengths and limitations
No previous study has analysed the histological effects of any treatment on the rotator cuff and only one study has analysed the effect of glucocorticoid injection on in vivo human tendon (the Achilles).23 One of this study's great strengths is therefore its novelty. The use of paired tendon samples from these two treatment groups has increased the power of the study by enabling the comparison of the effects of glucocorticoid injection with those of the tendon healing response after rotator cuff repair. The tissue analysis has been undertaken in as objective a manner as possible by the use of automated measurement and the strict blinding of observers. The number of patients in our study is adequate to determine tissue differences, as we have demonstrated. However the study is not designed to test the effects of glucocorticoid injection and rotator cuff repair on patients’ clinical outcomes. The lack of clinical improvement at 7 weeks in both treatment groups is likely to be an expected finding, as the benefits of glucocorticoid injection are typically short lasting, while the benefits of rotator cuff repair take longer to become clinically apparent.
A recognised weakness with immunohistochemistry is its semiquantitative nature. Despite trying to make all the methods of semiquantitative analysis as objective and unbiased as possible, the results are still semiquantitative and must be interpreted as such. It should be noted that it has not been possible to describe all the exact details regarding the locations of the various immunohistochemical reaction patterns. Only 3 of the 12 patients in the glucocorticoid injection treatment group were truly steroid ‘naïve’ (ie, had received no previous glucocorticoid injection). This reflects how widespread the use of glucocorticoid injection is in the UK, with the vast majority of patients in the UK being treated with glucocorticoid injection before referral to secondary care.4 The tissue samples were taken at a single time point 7 weeks following treatment and therefore the persistence of these observations is unknown.
Rotator cuff repair and glucocorticoid injection both result in significant histological and immunohistochemical changes to tendon which involves the excitatory glutaminergic system. There are distinct differences between the tissue response to both treatments, most notably the increase in NMDAR1 which occurs after glucocorticoid injection. These novel in vivo findings add weight to the recent evidence demonstrating that glucocorticoid may have harmful effects on tendon.
What are the new findings?
There are significant tissue changes in tendon tissue after both rotator cuff repair and glucocorticoid injection.
The increase in cell proliferation, vascularity and hypoxia inducible factor 1α after rotator cuff repair appears consistent with a proliferative healing response, and these features are not seen after glucocorticoid injection.
The increase in the glutamate receptor N-methyl-d-aspartate receptor 1 after glucocorticoid injection raises concerns about the potential excitotoxic damage that may result from this common treatment.
How might it impact on clinical practice in the near future?
The study's results provide a plausible mechanism by which glucocorticoid injection may harm tendon which should be considered by any clinician using glucocorticoid injection in a close proximity to the tendon.
The pharmacological modification of the glutaminergic system appears to have great potential both in terms of developing novel analgesic agents and in terms of augmenting tendon healing.
The authors would like to acknowledge the great contributions from research nurses Kim Wheway and Bridget Watkins in terms of tissue collection and of providing care and support for the study participants.
Contributors BJFD contributed to acquisition of the data, interpretation of the data, drafting of the article and revising and final approval of the article. SLF contributed to acquisition of the data, interpretation of the data, revising and final approval of the article. RJM contributed to conception and design, acquisition of the data and final approval of the article. MKJ contributed to interpretation of the data, revision and the final approval of the article. AJC contributed to concept and design, interpretation of the data, revision and final approval of the article. BJFD is as the guarantor.
Funding This work is funded by the Musculoskeletal Biomedical Research Unit of the National Institute for Health Research (BD, MM, EL, TO and AC), the Jean Shanks Foundation (BD) and Orthopaedic Research UK (BD).
Competing interests None.
Ethics approval Ethical approval for this study was granted by the local research ethics committee (Oxfordshire REC B, ref: 09/H0605/111).
Provenance and peer review Not commissioned; externally peer reviewed.