Article Text

Download PDFPDF

Extended report
Genetic variation in the GDF5 region is associated with osteoarthritis, height, hip axis length and fracture risk: the Rotterdam study
  1. R B A Vaes1,
  2. F Rivadeneira1,2,
  3. J M Kerkhof1,
  4. A Hofman2,
  5. H A P Pols1,2,
  6. A G Uitterlinden1,2,
  7. J B J van Meurs1
  1. 1
    Department of Internal Medicine, Erasmus Medical Centre, Rotterdam, The Netherlands
  2. 2
    Department of Epidemiology and Biostatistics, Erasmus Medical Centre, Rotterdam, The Netherlands
  1. Correspondence to Dr J B J van Meurs, Genetic Laboratory, Department of Internal Medicine, Erasmus Medical Centre, Room Ee571, PO Box 1738, 3000 DR, Rotterdam, The Netherlands; j.vanmeurs{at}


Background: A polymorphism (rs143383; T to C) near the GDF5 gene has been associated with height and osteoarthritis (OA), but debate exists about whether its primary biological action is directed to cartilage or bone.

Objective: To study the association between genetic variation in the GDF5 region and radiographic osteoarthritis (ROA) susceptibility, height, bone size parameters and fracture risk in a large population-based cohort of Caucasian elderly subjects.

Methods: 6365 men and women had genotype data available. ROA was defined as a Kellgren/Lawrence (K/L) score ⩾2 for hand, knee and hip joints. CTX-II levels, height, bone mineral density (BMD), bone size and fracture risk were also assessed.

Results: rs143383 and three highly correlated single nucleotide polymorphisms (SNPs) in the GDF5 region were found to be independently associated with OA, height, bone size and fracture risk in women. Women with homozygotes for the rs143383 C allele had a 37% lower risk for hand OA (p = 8×10−6) and a 28% lower risk for knee OA (p = 0.003). In addition, they were 1.1 cm taller (p = 0.001), had a larger hip axis length (HAL) (p = 4×10−4) and had a 29% increased risk of incident non-vertebral fractures (p = 0.02). No associations with hip OA or BMD were detected. No associations were found in men.

Conclusion: This population-based study shows that GDF5 gene variants are associated with hand OA, knee OA and fracture risk in elderly women. It also replicates previous association between GDF5 variation and height. Furthermore, our findings for HAL suggest that GDF5 action is primarily directed to the long bones, rather than the axial skeleton.

View Full Text

Statistics from

Growth differentiation factor 5 (GDF5) is a member of the transforming growth factor β superfamily and is related to the subfamily of bone morphogenetic proteins (BMPs).1 It is involved in several developmental processes, including chondrogenesis, skeletogenesis and joint development.2 3 In the human embryo, GDF5 is expressed in the primordial cartilage of long bones and in areas of active cartilage degradation and bone matrix formation, but shows no expression in the axial skeleton.1 GDF5 can induce ectopic cartilage and bone formation in vivo and in vitro4 and is present in adult human cartilage,5 implying that GDF5 also exerts its effects later in life.

GDF5 mutations in humans cause alterations in the appendicular skeleton. Inactivating mutations result in a broad spectrum of skeletal dysplasias, including brachydactyly type A2 and C.6 7 8 Both disorders are characterised by shortening of the digits. Activating mutations in GDF5 will lead to symphalangism and several types of acromesomelic chondrodysplasia,7 9 10 11 12 13 all characterised by ankylosis (bony fusion) of joints. Although GDF5 mutations result in many skeletal syndromes, they seem to have in common that limb and, in particular, hand, abnormalities are present.

Recently, genetic variation within the GDF5 gene (20q11.2) has been implicated in osteoarthritis (OA). A functional variant (rs143383) in the GDF5 core promoter was found to be associated with OA of the knee and hip in two Asian case–control cohorts.14 The T allele was shown to have lower transcriptional activity than the C allele and increased risk of OA.14 This association was confirmed, but with smaller effect size, in one European case–control study,5 while a Greek case–control study failed to detect any association with knee OA.15 Also, GDF5 was borderline significantly associated with hand OA in a recent analysis,16 but differences in design and outcome assessment were present between the pooled studies. Interestingly, a recent genome-wide association study identified the same genomic region as associated with height.17 While this could indicate an effect on both cartilage and bone metabolism, no studies have examined whether these associations are dependent on each other. Therefore, our study objective was to assess the effect of genetic variation in the GDF5 gene on OA susceptibility, C-telopeptide fragments of type II collagen (CTX-II) levels (a specific marker18 for cartilage degradation), height and several measures of bone health in a large population-based cohort study of elderly Caucasian subjects.

Patients and methods

Study population

Subjects were participants of the Rotterdam Study, a prospective population-based cohort study of people aged 55 years and over. Further details on the Rotterdam Study can be found elsewhere.19

The Rotterdam Study was approved by the medical ethics committee of the Erasmus University Medical School and written informed consent was obtained from every participant. In total, 6365 participants had genotype information available for the rs143383. Of those, 6114 subjects had data on height and weight, 3050 subjects had hand OA data, 2764 subjects had knee OA data and 3044 subjects had data on hip OA. For all other single nucleotide polymorphisms (SNPs), between 5974 and 5703 participants had genotype information available.


The genetic variants were genotyped using genomic DNA extracted from peripheral venous blood samples according to standard procedures.20 For the rs143383, the Assay-on-Demand service (, accessed 2 July 2009) was used to set up a Taqman allelic discrimination assay. PCR cycling reaction was performed according to the manufacturers protocol. To confirm the accuracy of genotyping, 332 (5%) randomly selected samples were re-genotyped with the same method. Less than 1% call inconsistencies were observed.

All other analysed SNPs were genotyped using the version 3 Illumina Infinium II HumanHap550 SNP chip array. Genotyping procedures were followed according to the manufacturers protocol (Illumina, San Diego, California, USA). Details on performance in our cohort can be found elsewhere.21 We extracted 236 SNPs from the array, spanning a 3000 kb region on chromosome 20 (32 000 kb–35 000 kb) and selected 40 SNPs surrounding rs143383 which were used to evaluate linkage disequilibrium (LD).

Outcome assessment

Anthropometric measurements

Height and weight were measured in standing position with indoor clothing without shoes. Body mass index (BMI; kg/m2) was calculated accordingly.

Bone mineral density and bone size measures

At baseline, bone mineral density (BMD, in g/cm2) was measured at the femoral neck. Average lumbar spine area (cm2) and BMD were measured over the L2–L4 of the lumbar spine by dual energy x-ray absorptiometry (Lunar DPX-L densitometer, Lunar Corp, USA) as described previously.22 Data on BMD was available for 5827 participants on femoral neck and lumbar spine. Thoracolumbar spine height (cm) extending from T7 to L4 was calculated by summing vertebral heights assessed radiographically in 2550 subjects after a mean (SD) follow-up of 6.4 (0.4) years.

The hip axis length (HAL) was measured after a mean (SD) follow-up of 11.1 (0.6) years after baseline using a Prodigy fan-beam densitometer (GE-LUNAR Corp, USA) in 2538 participants. The distance from the lower base of the greater trochanter, through the femoral neck, up to the inner pelvic brim is measured and its length correlates with the length of the long bones in the appendicular skeleton.

Radiological OA assessment

The radiographic scoring has been described in detail, previously.23 24 25 26 Radiographs were scored for the presence of radiographic OA of the hip, knee and hand according to the Kellgren/Lawrence (K/L) score.27 28 Knee and hip ROA were defined as a K/L score ⩾2 of one or both joints. Hand OA was defined as the presence of a K/L score ⩾2 in two out of three hand joint groups (distal interphalangeal, proximal interphalangeal, first carpometacarpal/trapezioscaphoid) of one or both hands. For all joint sites, the joint space width and presence of osteophytes were graded separately, as described in more detail elsewhere.18 24 25 26

Biomarker of cartilage degradation

CTX-II concentrations (ng/l) were measured in urine for a subset of the total population (n = 1235) using an enzyme-linked immunosorbent assay (Osteometer BioTech A/S, Herlev, Denmark). The CTX-II concentrations were standardised for total urine creatinine (mmol/l) as described previously.18

Incident fracture assessment

Follow-up for incident fractures started either on 1 January 1991, or at the time of entering the study cohort. For this study, follow-up lasted until 1 January 2002, or earlier if the participant died. All collected fractures were verified by reviewing discharge reports and letters from medical specialists. A more detailed description can be found elsewhere.29

Thoracolumbar radiographs of the spine were obtained at baseline and were available for 3469 subjects. All radiographs were scored for the presence of vertebral fracture by the McCloskey/Kanis method as described earlier.30 31

Statistical analysis

Hardy–Weinberg equilibrium of the polymorphisms was tested using a χ2 test.

All analyses were performed stratified according to gender. Differences between genotype groups in means of baseline measurements were compared using analysis of (co)variance and linear regression. For OA and vertebral fractures, odds ratios (ORs) were calculated using logistic regression analysis and were adjusted for age and BMI in a secondary analysis.

For analysis of non-vertebral fracture risk, we applied a multivariable Cox proportional hazards model.

For the genetic analyses of OA end points (fig 1) we used PLINK version 1.01 (, accessed 2 July 2009), developed by Shaun Purcell.32 We used a linear regression model under an additive genetic model. Haploview version 4.0 was used for measuring LD.

Figure 1

(A) Unadjusted p values for hand (diamonds), knee (squares) and hip (triangles) osteoarthritis (OA) for single nucleotide polymorphisms (SNPs) in a 3000 kb region surrounding GDF5. The horizontal line indicates a p value of 0.05. (B) The 407 kb zoomed region surrounding GDF5 with unadjusted p values for hand (diamonds), knee (squares) and hip (triangles) OA. Again, the horizontal line indicates a p value of 0.05. (C) The 407 kb zoomed region surrounding GDF5 where the genes (horizontal bars) and tested SNPs (vertical lines) are depicted. The arrow indicates the rs143383.

All other analyses were performed using SPSS for Windows. Corrections for multiple testing were not applied, because our detailed analysis was hypothesis based. In addition, many of the studied variables are correlated. We considered a two-sided p value ⩽0.05 statistically significant.


Association of GDF5 region with OA

We examined polymorphisms in a 407 kb genomic region surrounding GDF5 for association with OA of the hand, knee and hip using an allelic model. Figure 1 shows a highly significant association for the rs143383 (p = 3×10−6) with hand OA in women. Three SNPs (rs4911494, rs6088813 and rs6087705) in high LD with the rs143383 (multimarker r2 = 0.93; D′ = 0.98) showed similar p values (2×10−5, 1×10−5 and 1×10−5, respectively). Also, we observed two significant associations outside the GDF5 gene arising from the matrix metalloproteinase 24 gene (online supplementary data).

No association was detected in men. Since possible functionality of the rs143383 was demonstrated previously, we focused all further analyses on this SNP.

OA associations

The minor allele frequency for the rs143383 was 39% and genotype distribution followed Hardy–Weinberg equilibrium proportions (p = 0.72). Table 1 shows the association of the rs143383 with OA at several joint sites. In a dominant model, C carriers had a 22% decreased risk (p = 0.002) for hand OA and a 21% decrease in risk (p = 0.01) for knee OA compared with the TT genotype. This decrease was fully attributable to women, with a 37% (p = 8×10−6) and a 28% (p = 0.003) decreased risk respectively, while no significant effect was observed in men. These risk estimates did not change considerably after adjustment for age and BMI. In women, the population attributable risk (PAR) for the rs143383 is 11% for hand OA and 8.5% for knee OA. A small joint space width and presence of osteophytes (osteoarthritic characteristics), were both less prominent in women carrying the C allele. No association between GDF5 genotype and hip OA was seen in men or women.

Table 1

Association of rs143383 with different osteoarthritis phenotypes

Also, an association was detected with CTX-II levels in women, but not in men. C-allele carriers had significantly lower CTX-II levels than women with the TT genotype (p = 0.03, table 1).

Associations with anthropometric measurements

The C allele was associated with taller stature, especially in women (table 2). In women, each C-allele copy was associated with a height increase of 0.55 cm (ptrend = 0.001), with a similar, but non-significant, trend in men. The explained variance for height in women and men is 0.3% and 0.1%, respectively. No associations were detected for weight or BMI.

Table 2

Baseline characteristics by rs143383 genotype

Association with fracture and bone parameters

Women with the CC genotype had a significant increase in non-vertebral fractures (table 3) when compared with the TT genotype (HR = 1.29, 95% CI 1.05 to 1.58, p = 0.02). The effect of the polymorphism on fracture risk was consistent after stratification for height (low/high), age (low/high), presence of hand OA and presence of knee OA (data not shown). We did not detect an association with non-vertebral fracture risk in men. No associations were detected with vertebral fracture risk.

Table 3

Non-vertebral fractures by rs143383 genotype

Table 4 shows that the HAL was significantly larger in women with the CC genotype (ptrend = 4×10−4) than in those with the TT or TC genotype. The lumbar spine area in men and women showed borderline significant differences, but disappeared after adjustment for height and weight. No associations independent of height and weight were detected with BMD or thoracolumbar spine height, in either men or women.

Table 4

Bone parameters by rs143383 genotype


In this study we examined the role of a functional SNP5 14 (rs143383) in the GDF5 gene and its surrounding SNPs with bone- and cartilage-related traits within a large population-based cohort of elderly Caucasian subjects in the Netherlands. We observed significant associations of GDF5 polymorphisms with OA susceptibility in the knee and hand joints, incident fracture risk, height and hip axis length in elderly women. All these associations were independent of each other.

We show a convincing association with hand OA, probably arising from our powered design and accurate phenotype definition. We detected a highly significant decreased risk in female C-allele carriers, when compared with women with the TT genotype. Since GDF5 mutations are known to cause specific abnormalities in the development of the phalanges in humans,6 7 8 9 10 11 12 13 an association with hand OA is not surprising. Recently, GDF5 was found to be borderline significantly associated with hand OA in a pooled analysis from three cohorts using in total 955 cases and 2742 controls.16 As noted by these authors, there were significant differences in study design and hand OA definitions between the pooled studies, lowering the power to detect an effect.

Our findings on CTX-II levels, which were lower in women carrying a C allele than in women with the TT genotype (p = 0.03), further support the association with OA and, more specifically, cartilage degradation. In addition, both joint space narrowing and osteophytosis were less prominent in women carrying the C allele, which indicates an effect of GDF5 on both cartilage and bone.

This study replicates previous findings on knee OA susceptibility, but not on hip OA.5 14 This difference could arise owing to differences in study design, because both previous studies were hospital based, whereas our study was population based with a less severe (radiographic) hip phenotype. Additionally, the minimum detectable OR for hip OA in women was 0.62 (α = 0.05, power = 0.80) in a dominant model, which is a larger effect size than any other observed association in our cohort.

We also replicated previous findings on height differences,17 and found that HAL was significantly larger in women with the CC genotype (ptrend = 4×10−4). We observed only a weak association with lumbar area (ptrend = 0.02), which depended on height and weight and no association with thoracolumbar spine height by x-ray examination. Thus, the increased height is primarily attributable to differences in the long bones and not to differences in vertebral size. Our findings are consistent with previous observations, where GDF5 expression could not be localised in the axial skeleton.1 Earlier studies showed differential expression of the different alleles of the rs143383 in an expression vector14 and in RNA from arthroplasty patients,5 suggesting a difference in bioavailability of the GDF5 protein. Nevertheless, we cannot exclude the possibility that other SNPs in high LD with the rs143383 variant are driving these associations.

To our knowledge, this is the first study showing an independent association between GDF5 genotypes and fracture risk in women. Neither adjustment nor stratification for height and presence of OA had an effect on the association with fracture risk. This implies that the association is not driven by any of the other observed associations (ie, height, hand OA, knee OA), and genetic variation in the GDF5 gene seems to contribute independently to an increased fracture risk. Our analyses did not show an effect of GDF5 gene variation on BMD, thus BMD differences do not explain these differences in fracture risk.

Although we do not have a comprehensive biological explanation for the increase in fracture risk by rs143383 genotype, it is possible that the quality of bone is different between the genotype groups, which is not reflected in the BMD. Alternatively, since an increase in HAL is an established risk factor for hip fractures,33 34 the increased fracture risk can partly be attributed to HAL differences. An inverse relation between OA and osteoporosis is well established35 and our observations provide molecular support for this inverse relationship. However, since this is the first study to detect an association with fracture risk, replication in an independent population would aid in generalising these results.

We detected association in women, but not in men. This might be due to a lower power in men, since OA and fractures occur less often in men and our cohort consists of fewer men than women. Alternatively, it might be that sex-specific hormones, like androgens and oestrogens, might be involved in GDF5 regulation. GDF5 is closely related to the BMP subfamily, especially BMP-5 and BMP-6,1 and interaction between BMP-6 and oestrogens has been reported.36 37 38

Major strengths of our study include its size, accurate OA definition and the ability to assess the effect of a large number of traits, related to both OA and osteoporosis, in the same cohort. Furthermore, our study is the first to quantify the risk attributable to this genetic variation. The explained variance of height differences in women is 0.3%, which does not imply a major role for the rs143383 in determination of height. A PAR of 11% for the rs143383 in women with hand OA and 8.5% in women with knee OA, however, shows that the risk for OA attributable to this polymorphism is substantial in our population.

In conclusion, this study in elderly Dutch men and women provides evidence that, at least in women, genetic variation in the GDF5 gene is implicated in the pathogenesis of OA in different joint sites and contributes to differences in fracture risk. In addition, our study shows that GDF5 variation contributes to stature variation through an effect on the long bones.


We thank all participants of the Rotterdam study and the general practitioners, pharmacists and the many field workers at the research centre in Ommoord, Rotterdam, the Netherlands. We thank P Arp, P Jhamai, M Verbiest, Dr M Moorhouse and M Verkerk for technical assistance. We are grateful to Professor H Valkenburg, Drs E Odding, A Bergink, M Reijman and S Dahaghin for scoring the radiographs.


View Abstract

Supplementary materials


  • Additional data are published online only at

  • Funding This project was funded by EU-support: TREAT-OA (FP7-200800), GEFOS (FP7-201865) and the Netherlands Scientific Organisation - Large Investments (NWO Groot Investments; 175.01.2005.011).

  • Competing interests None.

  • Ethics approval Approved by the medical ethics committee of the Erasmus University Medical School.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.