Introduction

For over a decade, atypical fracture of the femoral bone (AFF) has been a well-documented adverse drug reaction (ADR) associated with long-term bisphosphonate use [1]. AFF is normally preceded by weeks or months of thigh pain and is in contrast to ordinary fragility fractures related to no or minimal trauma [2]. The term ‘atypical’ refers to the deviant transverse pattern of the fracture-line revealed on plain radiographs of the affected femur [2]. Although not all AFFs occur after bisphosphonate exposure, there is a strong correlation with duration of bisphosphonate use. A more than 100-fold increase in risk is seen after 4–5 years of bisphosphonate use, and the risk declines rapidly after cessation of treatment [3,4,5].

By now, clinicians, the scientific community and patients have come to realize the many challenges associated with AFFs. Over the last decade, a 50% decrease in prescriptions of bisphosphonates for primary and secondary prevention of fragility fractures has been seen [6]. This significant decline in preventive medication is believed to be due to fear of ADRs.

A major challenge in the prevention of AFF is the overall lack of knowledge about the mechanism behind this fracture type. Theories highlight long-term buildup of micro-cracks in the bone due to an over-suppression of bone remodeling that eventually leads up to failing skeletal integrity and stress fractures [7]. Predisposing risk factors are long-term use of bisphosphonates [3], female sex [3, 8], Asian ethnicity [9], and bowing of the femur [10]. Since only a minority of bisphosphonate users develop AFF, pathophysiological theories include a predisposing genetic trait, altered collagen cross-linking, accumulation of microdamage, increased mineralization, reduced heterogeneity of mineralization, variation in rates of bone turnover, and reduced vascularity [2].

A recent systematic review found six published studies that investigated the role of genetics on AFF in a total of 44 patients [11]. The review also identified 23 cases of AFF associated with seven different monogenetic bone disorders, of which seven cases had been exposed to a bisphosphonate. There is thus some evidence of rare genetic susceptibility loci for bisphosphonate-associated AFF. If common risk variants, i.e., genetic variants occurring among at least 1%, also exist, as has been shown for many rare adverse drug reactions [12], it might be feasible to predict patients at risk through preemptive genotyping. We performed the largest case–control GWAS to date, to determine whether common genetic variants contribute to risk of bisphosphonate-associated AFF. We also performed candidate gene analyses of 29 genes that have been implicated in AFF [11].

Materials and Methods

Sample Description

The basis for case recruitment was through nation-wide spontaneous ADR reports sent from healthcare professionals to the Swedish Medical Products Agency between the years 2006 and 2015. Each patient should be at least 18 years of age and able to give informed consent. Case definition for AFF was according to the American Society for Bone and Mineral Research [2].

We collected clinical data (demographics, medical history, drug treatment history, X-ray images, and ancestry) through interviews using a standardized questionnaire, and by obtaining and reviewing medical records. Prior to genetic analysis, each case including radiographs was evaluated by at least one senior consultant in orthopedics.

Overall, 71 reported cases were available. Of these, 18 cases were not possible to include (five were deceased, five could not be reached, four declined to participate, two were not suitable to be contacted according to the reporting physician, one was not able to perform the interview, and in one case the reporting physician could not be reached). Of the remaining 53 cases, two did not pass radiograph adjudication (ordinary fragility fractures) and therefore 51 cases, all with complete fractures, were included in the study. We compared the cases with two sets of controls. In the main analysis, we utilized 4891 population controls from the Swedish Twin Registry [13], all non-related individuals. The proportion of women in this population was 46%, and birth years ranged from 1911 to 1958 (1911–1919, 0.78%; 1920–1929, 10.3%; 1930–1939, 27.7%; 1940–1949, 45.7%; 1950–1958, 15.5%). Information on diseases and drug treatments for controls was available by linkage to individual data from the Swedish National Patient Register and the Swedish Prescribed Drug Register. Complete linkage is enabled by use of the individual personal registration number provided to all Swedish citizens. To determine whether any positive GWAS findings might be due to confounding by indication, we also defined a matched control group, consisting of patients who had collected at least one prescription of a bisphosphonate and who did not have a current cancer diagnosis. This gave a total of 324 controls that had been prescribed bisphosphonates and thus resembling the same source population of individuals as the cases, i.e., bisphosphonate users. Four out of five matched controls were women, which corresponds well with the overall proportion of women/men prescribed bisphosphonates in Sweden according to the Swedish Prescribed Drug Register. None of the cases with AFF had a current diagnosis of cancer.

Genome-Wide Array Data and Analyses

DNA was extracted from peripheral venous blood. Cases were genotyped with the Illumina Infinium OmniExpressExome 1 M array, and controls were genotyped with the Illumina HumanOmniExpress 700 K array. Genotype calls were generated using the Genome Studio software from Illumina and the Genome Reference Consortium human assembly GRCh37.

Genotyping quality control (QC) and data management was performed using PLINK v1.9 [14]. The resulting merged data included 604,238 SNPs post QC. Imputation was performed using the Sanger imputation server [15]. The pipeline with Eagle2 (v2.0.5) prephasing [16] and PBWT imputation [17] was used with the haplotype reference consortium panel as reference (v1.1) [15]. The total number of SNPs after imputation and QC was 7,585,874. All cases and controls were within the European cluster according to genetic principal component analysis (PCA), except for one case of Chilenean origin (Supplemental Fig. 1). Additional details on QC, PCA and imputation can be found in the Supplement.

Logistic regression on a genome-wide level was performed using PLINK v1.9 [14]. All genome-wide analyses were adjusted for the first four principal components. SNP effects were modeled only as additive and the conventional genome-wide significance threshold p < 5 × 10−8 was used to correct for multiple testing [18]. Results are presented as Manhattan plots. QQ-plots are presented in Supplemental Figs. 2 and 3.

Candidate Gene Analyses

In addition to genome-wide analyses, we performed candidate gene analyses in the imputed data set for a panel of 29 genes that have been implicated in AFF (Table 1) [11]. We examined a panel consisting of 8709 SNPs distributed in these genes. We both tested all 51 cases vs all 4891 controls and all 51 cases vs the 324 matched controls. Adjustment for multiple testing was done with Bonferroni correction (0.05/8709 ≈ 5.74 × 10−6).

Table 1 Candidate genes tested in the study

Power Calculation

Given a genome-wide significance level of p < 5 × 10−8 and using an additive genetic model, our sample size was powered to detect common genetic variants with effect sizes of clinical utility [19]. We had approximately 80% power to detect an odds ratio (OR) of 3–4 for variants with a minor allele frequency (MAF) of 40%, and 80% power to detect an OR of 4–5 for variants with a MAF of 20% (Supplemental Figs. 4 and 5). Given the significance level of p < 5.74 × 10−6 in the candidate gene analyses, we had 80% power to detect an OR of about 3 for variants with a MAF of 40%, and 80% power to detect an OR of about 4 for variants with a MAF of 20% (Supplemental Figs. 6 and 7).

Results

Characteristics of the 51 cases (48 women and 3 men) of bisphosphonate-associated AFF and the 324 matched controls are shown in Table 2. Most of the cases were of Swedish ethnicity (n = 47), while one each was of Finnish, Norwegian, British or Chilean origin.

Table 2 Characteristics of cases of bisphosphonate-associated atypical femoral fractures and matched controls

Genome-Wide Association Analyses—Cases Versus All Population Controls

Bisphosphonate-associated AFF was significantly associated with four isolated single nucleotide polymorphisms (SNP) (Fig. 1a; Table 3). The first SNP was rs7729897, which is located in an intergenic region upstream of the NR3C1 gene (nuclear receptor subfamily 3 group C member 1) on chromosome 5, OR 10.27 [95% confidence interval (CI) 4.95, 21.31] p = 4.00 × 10−10. The NR3C1 gene encodes a glucocorticoid receptor, which functions as a transcription factor that activates glucocorticoid responsive genes, and as a regulator of other transcription factors [20]. Variants of this gene have been associated with decreased bone mineral density in patients with endogenous hypercortisolism [21, 22].

Fig. 1
figure 1

a Manhattan plot of the genome-wide association analysis—cases vs all controls. b Manhattan plot of the genome-wide association analysis—cases vs matched controls. Analyses of 51 cases of bisphosphonate-associated atypical femoral fractures versus a all 4891 population controls, and b 324 matched controls. There were 7,585,874 SNPs after imputation, and adjustment was made for genetic principal components 1–4. The red line shows the threshold for genome-wide significance of 5 × 10−8. a Four SNPs were statistically significant when cases were compared with all 4891 controls. The top SNP was rs7729897, located in an intergenic region upstream of the NR3C1 gene (nuclear receptor subfamily 3 group C member 1) on chromosome 5, odds ratio (OR) 10.27 [95% confidence interval (CI) 4.95, 21.31] p = 4.00 × 10−10. There was also a significant association with rs11465606 positioned in an intronic region within the IL18R1 gene (interleukin 18 receptor 1) on chromosome 2, OR 6.15 [95% CI 3.32, 11.37], p = 7.13 × 10−9. A third significant association was with rs145787127, which is located in an intron region of the NTN1 (netrin 1) gene on chromosome 17, OR 7.37 [95% CI 3.63, 14.93], p = 3.08 × 10−8. The fourth significant association was with rs144094653, located close to the pseudogene TUBB8P5 (tubulin beta 8 class VIII pseudogene 5 on chromosome 12, OR 7.68 [95% CI 3.70, 15.91], p = 4.20 × 10−8. SNP single nucleotide polymorphism. b There were no statistically significant findings when cases were compared with matched controls. SNP single nucleotide polymorphism

Table 3 Top genome-wide associations with bisphosphonate-associated atypical femoral fractures

The second SNP was rs11465606 positioned in an intron within the IL18R1 gene (interleukin 18 receptor 1) on chromosome 2, OR 6.15 [95% CI 3.32, 11.37], p = 7.13 × 10−9. The third SNP was rs145787127, which is located in an intron of the NTN1 (netrin 1) gene on chromosome 17, OR 7.37 [95% CI 3.63, 14.93], p = 3.08 × 10−8. Genetic variation within NTN1 has been linked to osteoporosis [23]. The last SNP was rs144094653, located close to the pseudogene TUBB8P5 (tubulin beta 8 class VIII pseudogene 5 on chromosome 12, OR 7.68 [95% CI 3.70, 15.91], p = 4.20 × 10−8.

Genome-Wide Association Analyses—Cases Versus Controls with Bisphosphonate Use

No statistically significant association with gene status was revealed when cases of bisphosphonate-associated AFF were compared with matched controls (Fig. 1b; Table 4).

Table 4 Top genome-wide associations with bisphosphonate-associated atypical femoral fractures—cases vs matched controls

Candidate Gene Analyses—Cases Versus All Population Controls

When cases of bisphosphonate-associated AFF were compared with all population controls, there were no statistically significant associations (Fig. 2a; Table 5; Supplemental Table 1).

Fig. 2
figure 2

a Manhattan plot of the candidate gene analyses—cases vs all 4891 controls. b Manhattan plot of the candidate gene analyses—cases vs matched controls. Analyses of 51 cases of bisphosphonate-associated atypical femoral fractures versus a all 4891 controls, and b 324 matched controls. Adjustment was made for genetic principal components 1–4. The red line shows the threshold for statistical significance of 5.74 × 10−6. There were no statistically significant associations in either analysis

Table 5 Top candidate gene associations with bisphosphonate-associated atypical femoral fractures

Candidate Gene Analyses—Cases Versus Matched Controls

When cases of bisphosphonate-associated AFF were compared with matched controls, no statistically significant associations were revealed (Fig. 2b; Table 6; Supplemental Table 2).

Table 6 Top candidate gene associations with bisphosphonate-associated atypical femoral fractures

Discussion

We were hoping to find a strong common genetic susceptibility trait for AFF to predict patients at high risk of this ADR. Our results indicate that there is no common genetic variant that can be used for this purpose. The only significant finding on a genome-wide level was with four SNPs when cases were compared with population controls, but these were uncommon SNPs, all of which were single hits, meaning that these associations are likely false positives [24, 25], although two may theoretically be related to the treatment indication (NR3C1 and NTN1). None of these specific SNPs have, however, previously been implicated in AFF or osteoporosis [11, 26,27,28]. After reducing the risk of confounding by indication with the use of a comparison to bisphosphonate-treated controls, no statistically significant association remained.

At this time we are therefore left to models based on pharmacological and clinical considerations to minimize the risk of AFF. The prevailing pathophysiological theory of AFF is that bisphosphonates lead to over-suppression of bone remodeling [29]. Because bisphosphonates preferentially suppress the targeted repair mechanism, increased numbers of micro-cracks and reduced heterogeneity of the bone can be seen in bone tissue from animals and humans [7, 30,31,32]. The combination of these can lead to accumulation of micro-cracks during normal loading and propagation to larger cracks, eventually resulting in complete AFF. Studies have shown that the risk of developing an AFF is on average 50-fold greater for a bisphosphonate user compared to a nonuser, and more than 100-fold greater after 4–5 years of treatment [3, 5, 33]. In contrast, discontinuation of the drug will lead to a steep decline in the risk for developing an AFF [3]. In addition, different bisphosphonates might vary in terms of risk [3, 5, 34]. Hence, treatment duration and choice of bisphosphonate could be subject to manipulation in order to gain maximum treatment benefit while reducing the risk of AFF.

Many attempts have been made to identify risk factors that may predispose bisphosphonate users to AFF. A potential genetic influence has been suggested as a possible explanation to why only a minority of bisphosphonate users develop AFF. For instance, studies have revealed that polymorphisms in the gene encoding farnesyl diphosphate synthase (FDPS) may affect bone mineral density and bone turnover following bisphosphonate treatment in some patients, while not in others [35,36,37,38]. A possible genetic cause is also supported by studies that have demonstrated a difference in risk of AFF based on ethnicity, with Asians being at higher risk. A recent study by Lo et al. revealed a hazard ratio of 6.6 for females of Asian ethnicity compared with Caucasian women [9]. In addition, theories of a possible genetic trait have been long existing for other bisphosphonate ADRs that manifest in the skeleton [39].

There are several limitations to this study. First, matching of controls was done using bisphosphonate exposure as a proxy for osteoporosis as the Swedish Patient Register mainly includes information on diagnoses from hospital care. We were thus unable to identify controls who were prescribed a bisphosphonate for osteoporosis prevention. Secondly, although this is the largest genetic study of bisphosphonate-associated AFF to date, the number of included cases is still low. This means that the power to detect weakly associated common variants and strongly associated rare variants is low. It is also possible that several variants, inherited independently of one another, are required to infer a risk of AFF, in which case they will go undetected. To elucidate this would require a larger study and whole genome or exome sequencing, which was beyond the scope of this study. Lastly, there are suggestions that the association between bisphosphonate use and AFF is mainly driven by a genetic predisposition [11]. However, since 4–5 years of bisphosphonate use in Swedish women is associated with a 125-fold increase in risk of AFF [3], the potential underlying causal genetic risk allele(-s) should have a firm relation with both AFF and bisphosphonate use to entirely extenuate the exponential increase in risk with duration of bisphosphonate use. Noteworthily, a more moderately strong effect modification between bisphosphonates and genetic predisposition might still exist, but the current study is too small to disentangle such genetic modifying effects.

That several genetic loci, perhaps varying between individuals, might explain at least some cases of bisphosphonate-associated AFF has been proposed by some studies, although methodological issues and other limitations makes it difficult to conclude whether the findings are of relevance for a larger population of individuals with bisphosphonate-associated AFF. In the study by Pérez-Núñez et al. that compared 13 women with AFF and 268 female controls, 21 loci were more frequent in the fracture group [40]. Most patients accumulated two or more allelic variants, and the number of variants was different between patients with fractures and the controls, suggesting that several genes may be involved. The study was, however, limited by the fact that the controls were a mix of normal and osteoporotic women, and that only 12 of the 13 cases had been exposed to bisphosphonates. In another study, Roca-Ayats et al. performed whole-exome sequencing in three sisters who had all developed AFF following bisphosphonate treatment, and compared with three unrelated patients with bisphosphonate-associated AFF [41]. They detected 37 rare nonsynonymous mutations in 34 genes, but the results are questionable due to lack of validation and a small sample size. In a further study, Funck-Brentano et al. performed sequencing of four genes amongst two patients with bisphosphonate-associated AFF and found genetic variants in one, a rare heterozygous mutation in COL1A2 (c.213G > A; p.Arg708GIn) [42]. Limitations of this study include the small sample size. While these findings suggest a polygenic model in which an accumulation of susceptibility variants may lead to a predisposition to bisphosphonate-associated AFF, larger studies are required to provide solid evidence.

Conclusion

With this genome-wide association and candidate gene study, we were unable to find evidence of common genetic traits predisposition for bisphosphonate-associated AFF. This does not rule out the possibility of weakly associated genetic traits or the presence of rare genetic variants that confer a risk. Further studies of larger sample size as well as whole-exome or whole-genome sequencing studies are warranted.