Genes encoding proteoglycans are associatedwith the risk of anterior cruciate ligament rupturesSasha Mannion,1,2 Asanda Mtintsilana,1 Michael Posthumus,1,3

Willem van der Merwe,3 Hayden Hobbs,3 Malcolm Collins,1,4 Alison V September1

▸ Additional material ispublished online only. To viewplease visit the journal online(http://dx.doi.org/10.1136/bjsports-2013-093201).1UCT/MRC Research Unit forExercise Science and SportsMedicine, Department ofHuman Biology, Faculty ofHealth Sciences, University ofCape Town, Cape Town, SouthAfrica2Division of Human Genetics,Department of ClinicalLaboratory Sciences, Faculty ofHealth Sciences, University ofCape Town, Cape Town, SouthAfrica3Sports Science OrthopaedicClinic, Cape Town, SouthAfrica4The South African MedicalResearch Council, Cape Town,South Africa

Correspondence toDr Alison V September, MRC/UCT Research Unit for ExerciseScience and Sports Medicine,University of Cape Town, POBox 115, Newlands, CapeTown 7725, South Africa;alison.september@uct.ac.za

Accepted 25 January 2014

To cite: Mannion S,Mtintsilana A,Posthumus M, et al. Br JSports Med Published OnlineFirst: [please include DayMonth Year] doi:10.1136/bjsports-2013-093201

ABSTRACTBackground Genetic variants within genes involved infibrillogenesis have previously been implicated in anteriorcruciate ligament (ACL) injury susceptibility.Proteoglycans also have important functions infibrillogenesis and maintaining the structural integrity ofligaments. Genes encoding proteoglycans are plausiblecandidates to be investigated for associations with ACLinjury susceptibility; polymorphisms within genesencoding the proteoglycans aggrecan (ACAN), biglycan(BGN), decorin (DCN), fibromodulin (FMOD) and lumican(LUM) were examined.Methods A case–control genetic association study wasconducted. 227 participants with surgically diagnosedACL ruptures (ACL group) and 234 controls without anyhistory of ACL injury were genotyped for 10polymorphisms in 5 proteoglycan genes. Inferredhaplotypes were constructed for specific regions.Results The G allele of ACAN rs1516797 wassignificantly under-represented in the controls (p=0.024;OR=0.72; 95% CI 0.55 to 0.96) compared with theACL group. For DCN rs516115, the GG genotype wassignificantly over-represented in female controls(p=0.015; OR=9.231; 95%CI 1.16 to 73.01) comparedwith the ACL group and the AA genotype wassignificantly under-represented in controls (p=0.013;OR=0.33; 95% CI 0.14 to 0.78) compared with thefemale non-contact ACL injury subgroup. Haplotypeanalyses implicated regions overlapping ACAN(rs2351491 C>T-rs1042631 T>C-rs1516797 T>G), BGN(rs1126499 C>T-rs1042103 G>A) and LUM-DCN(rs2268578 T>C-rs13312816 A>T-rs516115 A>G) inACL injury susceptibility.Conclusions These independent associations andhaplotype analyses suggest that regions within ACAN,BGN, DCN and a region spanning LUM–DCN areassociated with ACL injury susceptibility. Taking intoaccount the functions of these genes, it is reasonable topropose that genetic sequence variability within thegenes encoding proteoglycans may potentially modulatethe ligament fibril properties.

INTRODUCTIONAlthough the aetiology of the molecular mechan-isms is poorly understood, multiple intrinsic andextrinsic risk factors, including genetics, have beenassociated with anterior cruciate ligament (ACL)ruptures.1–7 DNA sequence variants within severalgenes encoding collagens, implicated in regulatingthe formation of the collagen fibril (fibrillogenesis),which is the basic building block of ligaments, havebeen associated with ACL ruptures.3–6 Similar tothe collagens, the proteoglycans aggrecan, biglycan,decorin, fibromodulin and lumican have important

structural roles in ligaments as well as play anessential role in regulating fibrillogenesis.8

Mutations within the large aggrecan (ACAN)gene cause either dominant familial osteochondritisdissecans or a recessive skeletal dysplasia inhumans.9 10 Murine tissues and mice deficient inthe small leucine-rich proteoglycans (SLRPs) bigly-can, decorin, fibromodulin or lumican have similarphysical phenotypes to humans with classicEhlers-Danlos syndrome: fibrillogenesis is compro-mised resulting in collagen fibrils of highly irregulardiameters and abnormal fibrillar organisation.11–13

Young et al14 recently investigated the extracellularmatrix (ECM) content of ruptured ACL tissue.Lower proteoglycan and glycosaminoglycan (GAG)levels were observed in ruptured human ACL tissuein comparison with the non-ruptured controls.14

Genes encoding proteoglycans are, therefore, plaus-ible candidate genes to be investigated for an asso-ciation with ACL injury risk.Variants within the ACAN and lumican (LUM)

genes, on chromosomes 15q26.1 and 12q21.3,respectively, have previously been associated withseveral multifactorial conditions.15–20 The genesencoding biglycan (BGN) on chromosome Xq28,decorin (DCN) on chromosome 12q21.33 andfibromodulin (FMOD) on chromosome 1q32, have,however, not been associated with any multifactor-ial conditions until today.This study aimed to investigate the association of

sequence variants in the ACAN, BGN, DCN,FMOD and LUM candidate genes with ACL rup-tures based on the important biological functionsof these five proteoglycan encoding genes in main-taining the structural integrity of tissues and regu-lating fibrillogenesis. More importantly, the studyaimed to identify genomic regions encompassingthese five genes which may be harbouring DNAsequence signatures relevant to our understandingof ACL injury susceptibility. In addition, we aimedto investigate whether there was any contributionof these variants to sex-linked susceptibility as hasbeen previously noted with ACL ruptures.4 6

MATERIALS AND METHODSThe reporting of this case–control genetic associ-ation is in alignment with the recommendationsoutlined by the STREGA initiative, which is anextension of the STROBE statement.21

ParticipantsA total of 461 physically active, unrelated, self-reported Caucasian participants were recruited forthis case–control genetic association study usingpreviously described inclusion and exclusion

Mannion S, et al. Br J Sports Med 2014;0:1–8. doi:10.1136/bjsports-2013-093201 1

Original article BJSM Online First, published on February 19, 2014 as 10.1136/bjsports-2013-093201

Copyright Article author (or their employer) 2014. Produced by BMJ Publishing Group Ltd under licence.

group.bmj.com on February 19, 2014 - Published by bjsm.bmj.comDownloaded from

criteria.4 These participants consisted of 227 (166 male) indivi-duals with surgically diagnosed ACL ruptures (ACL group) and234 (144 male) apparently healthy participants without anyhistory of ACL injuries (CON group). The ACL group wasrecruited from the Sports Science Orthopaedic Clinic in CapeTown, South Africa and the CON group was recruited fromsports clubs and wellness centres within Cape Town, SouthAfrica. Recruitment of participants took place between the yearsof 2006 and 2013. Of the 227 ACL participants, 126 (61.46%;94 male) sustained their injury through non-contact mechanisms(NON subgroup).22 The NON subgroup was analysed as part ofthe ACL group, and underwent additional analysis separately.

All participants were required to complete a written informedconsent form according to the Declaration of Helsinki.Participants were also requested to complete a questionnaireregarding personal details, medical history, personal and familyligament and tendon injury history, as well as sports participa-tion. All participants were of self-reported Caucasian ancestry.

Sports participation of the CON and ACL groups was charac-terised into contact sports, non-contact jumping sports, non-contact non-jumping sports and skiing sports as previouslydefined,2 with slight modification.4 The most common sportsplayed by the ACL group men were rugby (45.2%) and soccer(8.4%), while the ACL group women played predominantlyhockey (26.2%) and netball (14.8%).

This study was approved by the Research Ethics Committeeof the Faculty of Health Sciences within the University of CapeTown, South Africa (reference number 164/2006).

DNA extractionApproximately 5 mL of venous blood was obtained from eachparticipant by venepuncture of a forearm vein and collectedinto an EDTA vacuum container tube. Blood samples werestored at −20°C until total DNA extraction was performed, aspreviously described by Lahiri and Nurnberger23 and modifiedby Mokone et al.24

Single nucleotide polymorphism selection and genotypingSingle nucleotide polymorphisms (SNPs; sequence variants)within each of the five candidate genes were identified using thegenome database hosted by the National Centre forBiotechnology Information (NCBI; http://ncbi.nlm.nih.gov/) aswell as SeattleSNPs (http://pga.gs.washington.edu/). SNPs wereselected based on previous associations, whether they were TagSNPs and should thus provide moderate coverage of the geneticinterval25 26 or if the SNP had a heterozygosity score greaterthan 30%.

Three SNPs (see online supplementary figure S1A) were inves-tigated in ACAN which include (1) rs2351491 23 C>T, previ-ously associated with height in individuals of African ancestry,17

(2) rs1042631 4157 T>C and (3) rs1516797 1133 T>G, withthe latter two SNPs being previously associated with lumbardisc degeneration.16 The two SNPs investigated within BGN(see online supplementary figure S1B) included rs1126499 189C>T and rs1042103 359 G>A, both of which were previouslyinvestigated for an association with congenital muscular dys-trophy but excluded as a cause of the disorder.27 DCNrs13312816 IVS1 A>T and rs516115 IVS3 A>G are TagSNPs25 26 and should therefore provide a large coverage of thegenomic region spanning approximately 18 kb of the DCN gene(see online supplementary figure S1C). Rs7543148 244 G>A,with a heterozygosity score greater than 30%, and rs108009121338 C>T, a Tag SNP, were chosen for investigation within theFMOD gene. LUM rs2268578 697 T>C, a Tag SNP,26 28 has

previously been associated with multifactorial phenotypes.26 29

Schematic diagrams of the FMOD and LUM genes are given inonline supplementary figures S1D and E, respectively. Thenomenclature used to describe the SNPs investigated is inaccordance with the NCBI (http://ncbi.nlm.nih.gov/).

TaqMan allele-discrimination assays (Applied Biosystems,Foster City, California, USA) were used to genotype participantsfor the 10 SNPs. Previously inventoried TaqMan primer sets andallele-specific MGB-labelled probes were used together with thePCR master mix, containing ampliTaq DNA polymerase Gold(Applied Biosystems, Foster City, California, USA), as per manu-facturers’ recommendations in a final reaction volume of 8 mL.The PCR reactions were performed on an Applied BiosystemsStepOnePlus Real-Time PCR system (Applied Biosystems,Foster City, California, USA) using the Applied BiosystemsStep-OnePlus Real-Time PCR software V.2.2.2 (AppliedBiosystems, Foster City, California, USA). The PCR parameterscomprised a 30 s hold step at 60°C followed by a 10 min heatactivation step at 95°C, 40 cycles of 95°C for 15 s and 60°C for1 min, ending with a 30 s hold step at 60°C. Genotypes weredetermined by endpoint fluorescence. For PCR and genotypequality control purposes, a number of positive (known geno-types) and DNA-free controls were randomly included on every96-welled PCR plate. All control samples were successfullyrepeated on every plate.

All laboratory works, DNA extraction and sample genotyping,took place at the UCT/MRC Research Unit for Exercise Science& Sports Medicine Laboratory, Faculty of Health Sciences,University of Cape Town.

StatisticsQuanto V.1.2 (http://hydra.usc.edu/gxe) was used to determinethe statistical power of the sample size. Assuming allele frequen-cies between 0.1 and 0.9 for the ‘risk’ allele of each SNP investi-gated, our sample size of 227 cases would be adequate to detectan allelic OR of 1.8 and greater at a power of 80% and a signifi-cance level of 5%.

Genotype and allele frequencies were analysed using StatisticaV.11 (StatSoft Inc, Tulsa, Oklahoma, USA) and GraphPad InStatV.5 (GraphPad software, San Diego, California, USA). The BGNgene is on the X chromosome and therefore genotype and allelefrequencies of SNPs investigated in this gene were comparedseparately between male and female participants. One-way ana-lysis of variance was used to compare continuous biologicalcharacteristics between the CON and ACL groups and betweenthe CON group and NON subgroup. Fisher’s exact and χ2 testswere used to compare categorical variables (sex and country ofbirth) between the CON group, ACL group and NON sub-group, as well as to analyse any difference in genotype andallele frequencies between the groups. Inferred haplotypes wereconstructed for the ACAN, BGN, DCN and FMOD genes usingthe specific SNPs investigated within each gene. A haplotypewas also constructed to overlap the LUM-DCN genetic interval(12q21.3-12q21.33) using the SNPs investigated within thesegenes. The Chaplin case–control haplotype inference softwareprogramme V.1.2.2 (http://www.genetics.emory.edu/labs/epstein/software/chaplin/index.html) was used to compare allele fre-quencies of the variants within each haplotype between casesand controls. CubeX: cubic exact solution (http://www.oege.org/software/cubex/)30 was used to determine which of the SNPsinvestigated are in linkage disequilibrium (LD) and thus likely tobe inherited together. Significance was accepted at p<0.05. Inorder to determine whether the genotypes obtained for each ofthe SNPs investigated were in Hardy-Weinberg equilibrium

2 Mannion S, et al. Br J Sports Med 2014;0:1–8. doi:10.1136/bjsports-2013-093201

Original article

group.bmj.com on February 19, 2014 - Published by bjsm.bmj.comDownloaded from

(HWE), the data were analysed using Genepop V.4.2 (http://genepop.curtin.edu.au/).

RESULTSParticipant characteristicsThere were significantly more men in the ACL group (p=0.008)and NON (p=0.013) subgroup in comparison with the CONgroup (table 1). Participants within the CON and ACL groups,and within the NON subgroup, were similarly matched for crit-ical body mass index (BMI) (BMI at recruitment for CONgroup; p=0.107 and 0.141, respectively) and country of birth(COB; p=0.769 and 0.814, respectively). The ACL group(p=0.019) and NON subgroup (p=0.028) were significantlyyounger than the CON group. When covaried for the differ-ences in sex and age, the ACL group (p=0.002) and NON sub-group (p=0.011) still weighed significantly more than the CONgroup. When covaried for sex, height did not differ significantlybetween groups (p=0.231 and 0.153).

Age and weight of the ACL group at the time of recruitmentwas 4.6±8.9 years (n=225) older and 2.0±12.1 kg (n=221)heavier than at the time of the first ACL rupture. In the NONsubgroup, the age and weight at the time of recruitment was 4.1±7.3 years (n=126) older and 2.5±11.0 kg (n=124) heavierthan at the time of first ACL rupture.

The female ACL and CON groups were matched for partici-pation in non-contact non-jumping sports (p=0.764) and skiingsports (p=0.526; data not shown). The male ACL and CONgroups were matched for non-contact non-jumping sports(p=0.138). Significantly more participants (men and women)within the ACL group participated in contact sports (p=0.009and 0.001, respectively) and non-contact jumping sports(p=0.003 and < 0.001, respectively) in comparison with con-trols. Significantly more ACL group men also participated inskiing sports in comparison with controls (p<0.001).

With the exception of a significant ACAN rs1042631 geno-type effect on sex (p=0.033), there were no other genotypeeffects on participant characteristics (see online supplementarytable S1).

ACAN geneThere were no significant differences in the genotype and allelefrequency distributions between the CON and ACL groups atthe ACAN rs2351491 (p=0.547 and 0.415) and rs1042631(p=0.168 and 0.064) SNPs. Similar results were noted when

stratified by mechanism of injury (NON subgroup) and sex (seeonline supplementary table S2). There was a trend (p=0.059)for the TT genotype to be over-represented in the CON group(51.1%, n=119) when compared with the ACL group (42.3%,n=96) for rs1516797. Interestingly, the G allele of rs1516797was significantly under-represented in the CON group (27.5%,n=128; p=0.024; OR=0.72; 95% CI 0.55 to 0.96) in com-parison with the ACL group (34.4%, n=156). No significantdifferences (p=0.325) in the allele frequencies for rs1516797were however noted between the CON group and NON sub-group. The genotype and allele frequency distributions for allthree SNPs were similar between the male and female partici-pants for all groups (CON, ACL and NON). All the groupswere in HWE for all three ACAN SNPs.

Only six of the possible eight haplotypes constructed fromthe three ACAN variants (rs2351491 C>T—rs1042631 T>C—rs1516797 T>G) had a frequency greater than 2%. The haplo-type containing alleles T–C–Twas significantly over-represented(p=0.001; LR=10.30) in the CON group (N=233, 43.55%) incomparison with the ACL group (N=225, 32.74%), while T–C–G was significantly under-represented (p=0.005; LR=7.79)in the CON group (N=233, 20.83%) in comparison with theACL group (N=225, 29.08%; figure 1A). ACAN rs2351491and rs1042631 were found to be in complete LD (D0=1.0),while rs1042631 and rs1516797 were not in LD (D0=−0.784).

BGN geneThere were no significant differences in the genotype frequencydistributions between the CON and ACL groups, or betweenthe CON group and NON subgroup at the BGN rs1126499 orrs1042103 loci for either men (rs1126499: CON vs ACL,p=0.533; CON vs NON, p=0.672; rs1042103: CON vs ACL,p=0.383; CON vs NON, p=0.580) or women (rs1126499:CON vs ACL, p=0.105; CON vs NON, p=0.278; rs1042103:CON vs ACL, p=0.226; CON vs NON, p=0.584; see onlinesupplementary table S2). Similarly, no significant differences inallele frequencies were noted. However, there was a trend forthe BGN rs1126499T allele to be under-represented (p=0.068)in the female CON group (48.3%, n=85) when compared withthe female ACL group (59.0%, n=72). All the female groupswere in HWE for both BGN SNPs.

There were no significant differences in the distribution ofthe inferred haplotypes constructed from the BGN variants(rs1126499 C>T—rs1042103 G>A) when only the male

Table 1 Characteristics of the asymptomatic control group (CON), the ACL rupture group (ACL), and the ACL subgroup with a non-contact(NON) mechanism of injury

CON (N=234) ACL (N=227) p Value * NON (N=126) p Value †

Sex (% males) 61.5 (234) 73.1 (227) 0.008 74.6 (126) 0.013Age (years) 29.3±11.3 (228) 26.8±11.0 (198) 0.019 26.6±10.5 (120) 0.028Weight (kg) 74.0±14.7 (229) 80.2±16.9 (206) <0.001 (0.002) 79.8±15.8 (123) <0.001 (0.011)Height (cm) 175.0±9.4 (228) 177.5±9.3 (206) 0.006 (0.231) 178.0±9.1 (122) 0.004 (0.153)BMI (kg/cm2) 23.8±4.1 (228) 24.6±5.8 (206) 0.107 24.6±5.4 (122) 0.141COB (% SA) 86.2 (225) 82.7 (208) 0.769 82.9 (123) 0.814

Data reported as mean±SD, except for sex and COB which are presented as frequency (%).The number of participants with available data for each variable is reported in parentheses.Age, weight and BMI are self-reported values at the time of first ACL rupture for the ACL group and NON subgroup, and at time of recruitment for the CON group.Unadjusted p values are shown, with the exception of weight and height which have adjusted p values in parentheses. Weight has been adjusted for age and sex, and height has beenadjusted for sex only.Significant p values are noted in italics.*CON vs ACL.†CON vs NON.ACL, anterior cruciate ligament; BMI, body mass index; COB, country of birth; SA, South African.

Mannion S, et al. Br J Sports Med 2014;0:1–8. doi:10.1136/bjsports-2013-093201 3

Original article

group.bmj.com on February 19, 2014 - Published by bjsm.bmj.comDownloaded from

participants were compared between the CON and ACL groups(figure 1B). However, when the female participants were com-pared, the BGN C–G inferred haplotype was significantly over-represented (p=0.027) in the CON group (N=88, 39.64%) incomparison with the ACL group (N=61, 27.15%; figure 1C).BGN rs1126499 and rs1042103 were found to be in low LD(D0=0.321).

DCN geneNo significant differences in genotype frequencies were notedfor the DCN rs13312816 (p=0.221) or rs516115 (p=0.926)SNPs when all participants (men and women) were analysed, orwhen only the male participants (rs13312816: p=0.256;rs516115: p=0.334) were analysed between the CON and ACLgroups (see online supplementary table S2); similarly, no

significant differences in allele frequencies were noted.Furthermore, no significant difference in the genotype distribu-tion of DCN rs13312816 was noted in women between theACL and CON groups (p=0.214). However, the GG genotypeof rs516115 was significantly over-represented in the CONgroup (13.3%, p=0.015; OR=9.23; 95% CI 1.17 to 73.01)when compared with the ACL group (1.6%), as well as beingsignificantly over-represented in the CON group (13.3%,p=0.035; OR=10.35; 95% CI 0.60 to 180.20) compared withthe NON subgroup (0.0%), where the GG genotype was absentwhen only female participants were compared. In contrast, theAA genotype was under-represented in the CON group (38.9%,p=0.065) in comparison with the ACL group (54.2%) and sig-nificantly under-represented in the CON group (38.9%,p=0.013; OR=0.33; 95% CI 0.14 to 0.78) compared with the

Figure 1 Frequency distribution of the (A) ACAN (rs2351491 C>T—rs1042631 T>C—rs1516797 T>G), (B and C) BGN (rs1126499 C>T—rs1042103 G>A), (D and E) DCN (rs13312816 A>T—rs516115 A>G) and (F) LUM–DCN (rs2268578 T>C—rs13312816 A>T—rs516115 A>G)inferred haplotypes among the control (CON, white bars) and anterior cruciate ligament rupture (ACL, black bars) groups for all participants, and formen and women separately in the BGN and DCN genes. The p values of significantly different distributions are noted. The total number ofparticipants with available genotype data within each group is indicated in parenthesis on the graph.

4 Mannion S, et al. Br J Sports Med 2014;0:1–8. doi:10.1136/bjsports-2013-093201

Original article

group.bmj.com on February 19, 2014 - Published by bjsm.bmj.comDownloaded from

NON subgroup (65.6%) when only female participants wereanalysed. In addition, the G allele of rs516115 was significantlyover-represented in the CON group (37.2%) when comparedwith the ACL group (23.8%, p=0.014; OR=1.90; 95% CI1.14 to 3.18) and the NON subgroup (17.2%, p=0.003;OR=2.86; 95% CI 1.40 to 5.85) when only female participantswere analysed. There were no significant differences in the allelefrequency distributions of rs13312816 between the threegroups for the female participants. All the DCN variants werein HWE for all groups.

Only three of the possible four inferred haplotypes con-structed for the two DCN variants (rs13312816 A>T—rs516115 A>G) had a frequency greater than 0%. There wereno significant differences in the distribution of the DCNinferred haplotypes when only male participants were comparedbetween the CON and ACL groups (figure 1D). When thefemale participants were analysed, there was a trend for the T–A haplotype to be under-represented in the CON group(N=90, 62.8%) in comparison with the ACL group (N=61,76.2%; figure 1E). The rs13312816 and rs516115 SNPs werefound to be in compete LD (D0=1.0).

FMOD and LUM genesNo significant differences in genotype frequencies were notedbetween the CON and ACL groups when the variants in theFMOD (rs7543148: p=0.458; rs10800912: p=0.616) andLUM (rs2268578: p=0.598) genes were analysed. Similarly, nosignificant differences in allele frequencies were noted betweenthese groups (see online supplementary table S2). Likewise, nosignificant differences in the genotype and allele frequency dis-tributions were noted when the data were stratified by mechan-ism of injury or sex.

There were no significant differences in the distribution ofthe inferred haplotypes for FMOD (rs7543148 G>A—rs10800912 C>T) between the CON and ACL groups (data notshown). Only six of the possible eight haplotypes constructedfor the 56 kb genetic interval overlapping the LUM and DCNgenes (rs2268578 T>C—rs13312816 A>T—rs516115 A>G)had a frequency greater than 2%. The T–A–G inferred haplo-type was significantly over-represented (p=0.038) in the CONgroup (N=234, 9.16%) in comparison with the ACL group(N=227, 7.26%; figure 1F). LUM rs2268578 and DCNrs13312816 were in high LD (D0=0.927) and DCNrs13312816 and rs516115 were in complete LD (D0=1.000).

DISCUSSIONProteoglycans, such as aggrecan, have major structural roles inligaments, and the SLRPs biglycan, decorin, fibromodulin andlumican are critical in regulating ECM remodelling and collagenfibrillogenesis through their interactions with the collagennetwork, the major structural component of ligaments andtendons.31–33 In light of the essential role of proteoglycans infibrillogenesis, and the previous associations of sequence var-iants within genes implicated in fibrillogenesis (COL1A1,COL5A1 and COL12A1) with ACL injury risk,3–6 this studyaimed to investigate 10 variants within 5 genes encoding pro-teoglycans (ACAN rs2351491, rs1042631, rs1516797; BGNrs1126499, rs1042103; DCN rs13312816, rs516115; FMODrs7543148, rs10800912 and LUM rs2268578) for an associ-ation with the risk of ACL injuries. The main findings of thisstudy include: (1) ACAN rs1516797 was independently asso-ciated with the risk of ACL injury in all participants; (2) DCNrs516115 was independently associated with the risk of injuryin female participants (sex-linked association) and (3) haplotype

analyses further implicated regions overlapping four of the pro-teoglycan encoding genes (ACAN, BGN and LUM-DCN) withACL injury susceptibility. This study is the first report of geneticassociations between the genes encoding proteoglycans and ACLinjury susceptibility.

Aggrecan is a large structural proteoglycan that stabilises thecollagen network by having a highly fixed negative chargewhich results in water retention (figure 2). This proteoglycanforms large aggregates through its interactions with hyaluronanin the ECM. Although aggrecan is believed to indirectly interactwith the collagen fibril, the exact interaction between aggrecanand the collagen network remains unclear.34 Aggrecan is com-posed of a protein core comprising three globular domains, G1,G2 and G3, each of which have a specific function.35–37 Thepresent study found that participants with the rs1516797 Gallele had an increased risk of rupturing their ACL (p=0.024;OR=0.72; 95% CI 0.55 to 0.96). The biological function ofthis T>G substitution in intron 12 of ACAN is unknown.

Analysis of the ACAN haplotype (rs2351491 C>T—rs1042631 T>C—rs1516797 T>G) further highlights thepotential role of aggrecan in the pathobiology of ACL injurysusceptibility. The T–C–G haplotype was associated with anincreased risk of ACL ruptures (p=0.005) while the T–C–Thaplotype was associated with a decreased risk of ACL ruptures(p=0.001). These haplotypes, which overlap the G3 domain,are therefore suggesting that this genomic interval, whichincludes ACAN rs1516797 T>G, may be harbouring DNAsequence signatures which may possibly alter aggrecans’ role inthe collagen network, and should thus be further explored.Interestingly, disease-causing mutations in proximity tors1516797 T>G have been associated with inherited forms ofskeletal dysplasia, a connective tissue disorder.9 10

This study also provided preliminary evidence suggesting thatSLRPs such as biglycan and decorin may play a role in thepathobiology of ACL injuries. Disease-causing mutations withinthe DCN gene have previously been associated with connectivetissue disorders.38 39 The SLRPs function predominantly in col-lagen fibrillogenesis, but have also been implicated in regulatingcell growth and matrix remodelling.31 33 40–43

BGN was chosen as a candidate for investigation because it ison the X chromosome; sex is an intrinsic risk factor for ACLruptures44 and previous ACL injury genetic association studieshave observed sex–genetic interactions.4 6 Although neither ofthe two SNPs investigated within the BGN gene were independ-ently associated with ACL injury risk, the C–G BGN haplotype(rs1126499 C>T—rs1042103 G>A) was associated with adecreased risk of ACL injury in female participants (p=0.027).This haplotype analysis suggests that the region overlappingBGN is modulating ACL injury susceptibility and the BGN geneshould be further explored to identify the causal regulatoryDNA sequence signatures.

Biglycan regulates collagen fibrillogenesis and structure byinteracting with collagen fibrils (figure 2)31 45; this allows bigly-can to maintain the structure of the ECM in connectivetissues.46 In addition, biglycan also interacts with growth factorssuch as transforming growth factor β (TGFβ), indicating thisproteoglycans’ involvement in modulating growth factor avail-ability to cells and its role in regulating matrix turnover.42

Previous genetic association studies have also observed a similarsex-specific selective advantage, as identified in this study, withvariants localised to the COL5A1 30-UTR which is also impli-cated in fibrillogenesis.4

The core protein of decorin binds to collagen to regulate col-lagen fibrillogenesis (figure 2).40 This study noted that DCN

Mannion S, et al. Br J Sports Med 2014;0:1–8. doi:10.1136/bjsports-2013-093201 5

Original article

group.bmj.com on February 19, 2014 - Published by bjsm.bmj.comDownloaded from

rs516115 A>G was implicated in ACL injury risk in female par-ticipants, with the GG genotype specifically associated with a10.4-fold decreased risk of injury (p=0.035; OR=10.35; 95%CI 0.60 to 180.20) and the AA genotype associated with anincreased risk of injury (p=0.013; OR=0.33; 95% CI 0.14 to0.78) when the CON group and NON subgroup were com-pared. The associations were further illustrated when the A andG alleles were significantly over-represented in the ACL andCON groups, respectively (p=0.014), and mirrored when thedata were stratified by mechanism of injury (p=0.003). Thefunctional significance of this variant is unknown but one canhypothesise that variations in the core protein may affect theinteraction of decorin with TGFβ or the collagen fibril,33 47

thereby possibly altering the process of fibrillogenesis and affect-ing the mechanical properties of the ligament.

Currently, the authors are not aware of any published evi-dence suggesting a selective biological advantage for the specificregions overlapping BGN and DCN to have a sex-specific associ-ation. Although biglycan is on the X chromosome, there is con-flicting evidence as to whether this gene undergoesX-inactivation and whether BGN has a Y chromosome homo-logue.48 The difference, or lack thereof, in dosage and expres-sion of BGN between male and female participants has not beenfully explored, and may play a role in the altered risk of ACLinjury in female participants. Ovarian hormone levels, particu-larly oestrogen, modulate the synthesis and degradation ofSLRPs such as biglycan and decorin.49 50 Receptor sites for oes-trogen and progesterone have been found in the ACL,51 52 andsex hormones are suggested to affect ACL structure and com-position.51 The regulation of SLRPs by oestrogen53–55 mayaccount for the sex-specific association of these proteoglycangenes with the risk of ACL injury in women.

Haplotype analysis, the investigation of a set of variants inher-ited together, is often more informative in detecting an associ-ation compared with analysing individual variants alone.56 No

independent associations were noted for LUM; however, thehaplotype encompassing the LUM–DCN (rs2268578 T>C—DCN rs13312816 A>T—rs516115 A>G) genes implicated theT–A–G allele combination with reduced ACL injury risk(p=0.038). One can speculate that this genomic interval over-lapping LUM and DCN influences ACL injury susceptibility byeffecting fibrillogenesis. Therefore, it is critical that this genomicregion is further interrogated to identify functional DNAsequences.

Proteoglycans such as aggrecan, biglycan, decorin, fibromodu-lin and lumican play an important role in fibrillogenesis, pos-sibly through their myriad of direct/indirect interactions withvarious proteins, including the collagen network (figure 2) andcell-signalling molecules within the ECM. Altering the proper-ties of the collagen fibril will most likely alter the biomechanicaland functional properties of the ligament and one can thereforehypothesise that this modulation will impact on ACL injuryrisk.57 It was thus not surprising that our novel results are impli-cating sequence variants within four proteoglycan genes withACL injury susceptibility. Although there is no immediate clin-ical translation from this study, the results are suggesting thatinterindividual variations in the collagen network and fibrilassembly might be an important molecular mechanism contrib-uting to the aetiology of ACL ruptures, similar to Achilles tendi-nopathy.58 To improve our understanding of ACL pathogenesisand susceptibility, it is imperative that we start elucidating thenet effect of the intricate interactions of proteoglycans, such asaggrecan, in regulating the structural properties of the ECM,including the collagen fibril. The elucidation of these interac-tions is vital for the development of possible therapeutic inter-ventions targeting proteoglycans specifically.59

The cases and controls investigated in this study were notmatched for the confounding variables weight and height, andthis is, therefore, a limitation of the study. It has been suggestedthat heavier individuals are more likely to sustain an ACL

Figure 2 A schematic diagram representing the association of aggrecan and the small leucine-rich proteoglycans (SLRPs) biglycan, decorin,fibromodulin and lumican, as well as the glycosaminoglycans (GAGs), with the collagen network. Multiple aggrecan core proteins bind hyaluronanvia a link protein to form large aggregates; hyaluronan is able to directly interact with the collagen fibril.

6 Mannion S, et al. Br J Sports Med 2014;0:1–8. doi:10.1136/bjsports-2013-093201

Original article

group.bmj.com on February 19, 2014 - Published by bjsm.bmj.comDownloaded from

rupture as adiposity is a contributing risk factor to the develop-ment of musculoskeletal soft tissue injuries.60 61 Although carewas taken to recruit controls matched for sports participation, alimitation is that the cases and controls were not matchedexactly for exposure to contact and non-contact jumping sportswhich are of high risk for ACL injuries.

The pathway-based approach followed in this study providesevidence that highlights the potentially important biological rolethat proteoglycans ACAN, BGN, DCN and/or LUM may have inmodulating ACL injury susceptibility. These findings should,however, be repeated in independent populations to confirm theassociations described. These results suggest a need to furtherinterrogate the genomic intervals encompassing the proteogly-can genes with regard to ACL injury risk using functional ana-lyses, as well as to identify the reasons for the multiplesex-specific associations observed.

What are the new findings?

▸ The genes encoding the proteoglycans aggrecan and decorinare independently associated with the risk of anteriorcruciate ligament (ACL) injuries (decorin in females only).

▸ This study implicates specific regions within fourproteoglycan genes in ACL pathogenesis.

▸ We propose that specific regions within ACAN, BGN, DCNand LUM potentially contain functional sequencesmodulating ligament fibril properties, affecting ACL injurysusceptibility.

How might it impact on clinical practice in the nearfuture?

▸ There are no immediate clinical translations; however, theseresults implicate proteoglycans in anterior cruciate ligament(ACL) pathogenesis, making them potential therapeutic targets.

▸ Results of this study add to the growing body of evidence,suggesting that interindividual variations in collagen fibrilassembly might be an important molecular mechanism inthe aetiology of ACL ruptures.

▸ Genetic risk factors could, in future, be included intomultifactorial models to assess an individuals’ ACL injurysusceptibility.

Acknowledgements The authors would like to thank Ms M Rahim and Ms MHay for assisting with recruiting participants. Dr D O’Cuinneagain is also thanked forhis assistance with participant recruitment and diagnosis.

Contributors SM contributed to participant recruitment, laboratory work,genotyping, analyses and manuscript preparation. AM contributed to the laboratorywork and genotyping. MP was responsible for manuscript preparation andparticipant recruiting. WvdM and HH contributed to participant recruitment,diagnosis and manuscript editing. MC was responsible for project development,analysis and manuscript preparation. AVS was responsible for genotyping QC, projectdevelopment, analyses and manuscript preparation.

Funding This study was supported in part by funds from the University of CapeTown, and the South African Medical Research Council. SM was supported by theNational Research Foundation (NRF) of South Africa. MP was supported by theThembakazi Trust.

Competing interests MC and AVS have filed patents on the application ofspecific sequence variations (not included in this manuscript) related to riskassessment of Achilles tendinopathy and anterior cruciate ligament injuries.

Patient consent Obtained.

Ethics approval Ethics approval was obtained from Human Research EthicsCommittee, University of Cape Town, South Africa.

Provenance and peer review Not commissioned; externally peer reviewed.

REFERENCES1 Harner CD, Paulos LE, Greenwald AE, et al. Detailed analysis of patients with

bilateral anterior cruciate ligament injuries. Am J Sports Med 1994;22:37–43.2 Flynn RK, Pedersen CL, Birmingham TB, et al. The familial predisposition toward

tearing the anterior cruciate ligament: a case control study. Am J Sports Med2005;33:23–8.

3 Khoschnau S, Melhus H, Jacobson A, et al. Type I collagen alpha1 Sp1polymorphism and the risk of cruciate ligament ruptures or shoulder dislocations.Am J Sports Med 2008;36:2432–6.

4 Posthumus M, September AV, O’Cuinneagain D, et al. The COL5A1 gene isassociated with increased risk of anterior cruciate ligament ruptures in femaleparticipants. Am J Sports Med 2009;37:2234–40.

5 Posthumus M, September AV, Keegan M, et al. Genetic risk factors for anteriorcruciate ligament ruptures: COL1A1 gene variant. Br J Sports Med 2009;43:352–6.

6 Posthumus M, September AV, O’Cuinneagain D, et al. The association between theCOL12A1 gene and anterior cruciate ligament ruptures. Br J Sports Med2010;44:1160–5.

7 Posthumus M, Collins M, van der Merwe L, et al. Matrix metalloproteinase geneson chromosome 11q22 and the risk of anterior cruciate ligament (ACL) rupture.Scand J Med Sci Sports 2012;22:523–33.

8 Chen S, Birk DE. The regulatory roles of small leucine-rich proteoglycans inextracellular matrix assembly. FEBS J 2013;280:2120–37.

9 Stattin E, Wiklund F, Lindblom K, et al. A missense mutation in the aggrecan C-typelectin domain disrupts extracellular matrix interactions and causes dominant familialosteochondritis dissecans. Am J Hum Genet 2010;86:126–37.

10 Tompson SW, Merriman B, Funari VA, et al. A recessive skeletal dysplasia, SEMDaggrecan type, results from a missense mutation affecting the C-type lectin domainof aggrecan. Am J Hum Genet 2009;84:72–9.

11 Corsi A, Xu T, Chen XD, et al. Phenotypic effects of biglycan deficiency are linked tocollagen fibril abnormalities, are synergized by decorin deficiency, and mimicEhlers-Danlos-like changes in bone and other connective tissues. J Bone Miner Res2002;17:1180–9.

12 Ameye L, Young MF. Mice deficient in small leucine-rich proteoglycans: novel invivo models for osteoporosis, osteoarthritis, Ehlers-Danlos syndrome, musculardystrophy, and corneal diseases. Glycobiology 2002;12:107R–16R.

13 Danielson KG, Baribault H, Holmes DF, et al. Targeted disruption of decorin leadsto abnormal collagen fibril morphology and skin fragility. J Cell Biol1997;136:729–43.

14 Young K, Samiric T, Feller J, et al. Extracellular matrix content of ruptured anteriorcruciate ligament tissue. Knee 2011;18:242–6.

15 Weedon MN, Lango H, Lindgren CM, et al. Genome-wide association analysisidentifies 20 loci that influence adult height. Nat Genet 2008;40:575–83.

16 Videman T, Saarela J, Kaprio J, et al. Associations of 25 structural, degradative, andinflammatory candidate genes with lumbar disc desiccation, bulging, and heightnarrowing. Arthritis Rheum 2009;60:470–81.

17 N’Diaye A, Chen GK, Palmer CD, et al. Identification, replication, and fine-mappingof Loci associated with adult height in individuals of african ancestry. PLoS Genet2011;7:e1002298.

18 Yip SP, Leung KH, Ng PW, et al. Evaluation of proteoglycan gene polymorphisms asrisk factors in the genetic susceptibility to high myopia. Invest Ophthalmol Vis Sci2011;52:6396–403.

19 Chen ZT, Wang IJ, Shih Y, et al. The association of haplotype at the lumican genewith high myopia susceptibility in Taiwanese patients. Ophthalmology2009;116:1920–7.

20 Zhang F, Zhu T, Zhou Z, et al. Association of lumican gene with susceptibility topathological myopia in the northern han ethnic chinese. J Ophthalmol2009;2009:514306.

21 Little J, Higgins JPT, Ioannidis JPA, et al. Strengthening the reporting of geneticassociation studies (STREGA)—an extension of the strengthening the reporting ofobservational studies in epidemiology (STROBE) statement. J Clin Epidemiol2009;62:597–608.e4.

22 Marshall S, Padua D, McGrath M. Incidence of ACL injuries. In: Hewett T, Schultz S,Griffin L. eds. Understanding and preventing noncontact ACL injuries. Champaign,IL: Human Kinetics, 2007:5–30.

23 Lahiri DK, Nurnberger JI Jr. A rapid non-enzymatic method for the preparation ofHMW DNA from blood for RFLP studies. Nucleic Acids Res 1991;19:5444.

Mannion S, et al. Br J Sports Med 2014;0:1–8. doi:10.1136/bjsports-2013-093201 7

Original article

group.bmj.com on February 19, 2014 - Published by bjsm.bmj.comDownloaded from

24 Mokone GG, Gajjar M, September AV, et al. The guanine-thymine dinucleotiderepeat polymorphism within the tenascin-C gene is associated with Achilles tendoninjuries. Am J Sports Med 2005;33:1016–21.

25 van Diemen CC, Postma DS, Vonk JM, et al. Decorin and TGF-beta1 polymorphismsand development of COPD in a general population. Respir Res 2006;7:89.

26 Kelemen LE, Couch FJ, Ahmed S, et al. Genetic variation in stromal proteins decorinand lumican with breast cancer: investigations in two case-control studies. BreastCancer Res 2008;10:R98.

27 Peat RA, Gécz J, Fallon JR, et al. Exclusion of biglycan mutations in a cohort ofpatients with neuromuscular disorders. Neuromuscul Disord 2008;18:606–9.

28 Dimasi DP, Burdon KP, Hewitt AW, et al. Candidate gene study to investigate thegenetic determinants of normal variation in central corneal thickness. Mol Vis2010;16:562–9.

29 Couch FJ, Wang X, McWilliams RR, et al. Association of breast cancer susceptibilityvariants with risk of pancreatic cancer. Cancer Epidemiol Biomarkers Prev2009;18:3044–8.

30 Gaunt T, Rodríguez S, Day I. Cubic exact solutions for the estimation of pairwisehaplotype frequencies: implications for linkage disequilibrium analyses and a webtool ‘CubeX’. BMC Bioinformatics 2007;8:428.

31 Chiu R, Li W, Herber RP, et al. Effects of biglycan on physico-chemical properties ofligament-mineralized tissue attachment sites. Arch Oral Biol 2012;57:177–87.

32 Kresse H, Schonherr E. Proteoglycans of the extracellular matrix and growth control.J Cell Physiol 2001;189:266–74.

33 Scott JE. Proteoglycan-fibrillar collagen interactions. Biochem J 1988;252:313–23.34 Heinegard D. Proteoglycans and more—from molecules to biology. Int J Exp Pathol

2009;90:575–86.35 Heinegard D, Hascall VC. Aggregation of cartilage proteoglycans. 3. Characteristics

of the proteins isolated from trypsin digests of aggregates. J Biol Chem1974;249:4250–6.

36 Kiani C, Chen L, Lee V, et al. Identification of the motifs and amino acids inaggrecan G1 and G2 domains involved in product secretion. Biochemistry2003;42:7226–37.

37 Aspberg A, Adam S, Kostka G, et al. Fibulin-1 is a ligand for the C-type lectindomains of aggrecan and versican. J Biol Chem 1999;274:20444–9.

38 Bredrup C, Knappskog PM, Majewski J, et al. Congenital stromal dystrophy of thecornea caused by a mutation in the decorin gene. Invest Ophthalmol Vis Sci2005;46:420–6.

39 Rodahl E, Van Ginderdeuren R, Knappskog PM, et al. A second decorin frame shiftmutation in a family with congenital stromal corneal dystrophy. Am J Ophthalmol2006;142:520–1.

40 Reed CC, Iozzo RV. The role of decorin in collagen fibrillogenesis and skinhomeostasis. Glycoconj J 2002;19:249–55.

41 Jepsen KJ, Wu F, Peragallo JH, et al. A syndrome of joint laxity and impairedtendon integrity in lumican- and fibromodulin-deficient mice. J Biol Chem2002;277:35532–40.

42 Iozzo RV. The family of the small leucine-rich proteoglycans: key regulators of matrixassembly and cellular growth. Crit Rev Biochem Mol Biol 1997;32:141–74.

43 Schaefer L, Schaefer RM. Proteoglycans: from structural compounds to signalingmolecules. Cell Tissue Res 2010;339:237–46.

44 Anderson AF, Dome DC, Gautam S, et al. Correlation of anthropometricmeasurements, strength, anterior cruciate ligament size, and intercondylar notchcharacteristics to sex differences in anterior cruciate ligament tear rates. Am J SportsMed 2001;29:58–66.

45 Raspanti M, Viola M, Forlino A, et al. Glycosaminoglycans show a specific periodicinteraction with type I collagen fibrils. J Struct Biol 2008;164:134–9.

46 Tenorio DM, Santos MF, Zorn TM. Distribution of biglycan and decorin in rat dentaltissue. Braz J Med Biol Res 2003;36:1061–5.

47 Hildebrand A, Romaris M, Rasmussen LM, et al. Interaction of the small interstitialproteoglycans biglycan, decorin and fibromodulin with transforming growth factorbeta. Biochem J 1994;302(Pt 2):527–34.

48 Geerkens C, Vetter U, Just W, et al. The X-chromosomal human biglycan gene BGNis subject to X inactivation but is transcribed like an X-Y homologous gene. HumGenet 1995;96:44–52.

49 Salgado RM, Favaro RR, Martin SS, et al. The estrous cycle modulates smallleucine-rich proteoglycans expression in mouse uterine tissues. Anat Rec (Hoboken)2009;292:138–53.

50 Hong SH, Nah HY, Lee JY, et al. Analysis of estrogen-regulated genes in mouseuterus using cDNA microarray and laser capture microdissection. J Endocrinol2004;181:157–67.

51 Liu SH, al-Shaikh R, Panossian V, et al. Primary immunolocalization of estrogen andprogesterone target cells in the human anterior cruciate ligament. J Orthop Res1996;14:526–33.

52 Silvers HJ, Mandelbaum BR. Prevention of anterior cruciate ligament injury in thefemale athlete. Br J Sports Med 2007;41(Suppl 1):i52–9.

53 Rodrigo MC, Martin DS, Eyster KM. Estrogen decreases biglycan mRNA expressionin resistance blood vessels. Am J Physiol Regul Integr Comp Physiol 2003;285:R754–61.

54 Salgado RM, Favaro RR, Zorn TM. Modulation of small leucine-rich proteoglycans(SLRPs) expression in the mouse uterus by estradiol and progesterone. Reprod BiolEndocrinol 2011;9:22.

55 Markiewicz M, Znoyko S, Stawski L, et al. A role for estrogen receptor-alpha andestrogen receptor-beta in collagen biosynthesis in mouse skin. J Invest Dermatol2013;133:120–7.

56 Pettersson FH, Anderson CA, Clarke GM, et al. Marker selection for geneticcase-control association studies. Nat Protoc 2009;4:743–52.

57 Collins M, Posthumus M. Type V collagen genotype and exercise-related phenotyperelationships: a novel hypothesis. Exerc Sport Sci Rev 2011;39:191–8.

58 Hay M, Patricios J, Collins R, et al. Association of type XI collagen genes withchronic Achilles tendinopathy in independent populations from South Africa andAustralia. Br J Sports Med 2013;47:569–74.

59 Frey H, Schroeder N, Manon-Jensen T, et al. Biological interplay betweenproteoglycans and their innate immune receptors in inflammation. FEBS J2013;280:2165–79.

60 Griffin LY, Albohm MJ, Arendt EA, et al. Understanding and preventing noncontactanterior cruciate ligament injuries: a review of the Hunt Valley II meeting, January2005. Am J Sports Med 2006;34:1512–32.

61 Posthumus M, Collins M, September AV, et al. The intrinsic risk factors for ACLruptures: an evidence-based review. Phys Sportsmed 2011;39:62–73.

8 Mannion S, et al. Br J Sports Med 2014;0:1–8. doi:10.1136/bjsports-2013-093201

Original article

group.bmj.com on February 19, 2014 - Published by bjsm.bmj.comDownloaded from

doi: 10.1136/bjsports-2013-093201 published online February 19, 2014Br J Sports Med

 Sasha Mannion, Asanda Mtintsilana, Michael Posthumus, et al. ligament rupturesassociated with the risk of anterior cruciate Genes encoding proteoglycans are

http://bjsm.bmj.com/content/early/2014/02/19/bjsports-2013-093201.1.full.htmlUpdated information and services can be found at:

These include:

Data Supplement http://bjsm.bmj.com/content/suppl/2014/02/19/bjsports-2013-093201.DC1.html

"Supplementary Data"

References http://bjsm.bmj.com/content/early/2014/02/19/bjsports-2013-093201.1.full.html#ref-list-1

This article cites 60 articles, 22 of which can be accessed free at:

P<P Published online February 19, 2014 in advance of the print journal.

serviceEmail alerting

the box at the top right corner of the online article.Receive free email alerts when new articles cite this article. Sign up in

CollectionsTopic

(13 articles)Ligament rupture   � Articles on similar topics can be found in the following collections

(DOIs) and date of initial publication. publication. Citations to Advance online articles must include the digital object identifier citable and establish publication priority; they are indexed by PubMed from initialtypeset, but have not not yet appeared in the paper journal. Advance online articles are Advance online articles have been peer reviewed, accepted for publication, edited and

http://group.bmj.com/group/rights-licensing/permissionsTo request permissions go to:

http://journals.bmj.com/cgi/reprintformTo order reprints go to:

http://group.bmj.com/subscribe/To subscribe to BMJ go to:

group.bmj.com on February 19, 2014 - Published by bjsm.bmj.comDownloaded from

Notes

(DOIs) and date of initial publication. publication. Citations to Advance online articles must include the digital object identifier citable and establish publication priority; they are indexed by PubMed from initialtypeset, but have not not yet appeared in the paper journal. Advance online articles are Advance online articles have been peer reviewed, accepted for publication, edited and

http://group.bmj.com/group/rights-licensing/permissionsTo request permissions go to:

http://journals.bmj.com/cgi/reprintformTo order reprints go to:

http://group.bmj.com/subscribe/To subscribe to BMJ go to:

group.bmj.com on February 19, 2014 - Published by bjsm.bmj.comDownloaded from