Gene and SNP selection

Seven candidate genes ( CYP2C8 , COL1A1 , RANK , OPN , MMP2 , OPG and TNF ) were selected based on their potential roles in osteoclastogenesis, osteoclast differentiation, and bone resorption and/or previous association with bone mineral density (BMD) or osteoporosis. COL1A1 gene is considered an important functional candidate gene for the pathogenesis of osteoporosis because the type I collagen is an important protein of bone and a mutation in this gene results in osteogenesis imperfecta, characterized by reduced BMD and increased bone fragility . Thus genetic variants of COL1A1 may play an important role in osteoporosis or osteoporotic fractures. A study showed that the G>T polymorphism of COL1A1 gene, was associated with osteoporosis . The gene for matrix metalloproteinase 2 ( MMP2 ) has been suggested as a candidate gene for the development of BONJ because BP treatment is associated with atrial fibrillation and MMP2 is the only gene known to be associated with bone abnormalities and atrial fibrillation . RANK and osteoprotegerin ( OPG ), a receptor and decoy receptor, respectively, for RANK ligand, are members of the tumour necrosis factor receptor super family that plays a central role in osteoclast development. Studies of the mouse counterpart suggest that RANK directly mediates the osteoprotegerin ligand ( OPGL )-induced osteoclastogenesis in osteoclast precursor cells. Koh et al. identified two novel SNPs (+34,863 G>A and +35,928 insertion/deletion C) in RANK that are possibly associated with low BMD in postmenopausal women . Another study by Hsu et al. showed significant positive association between BMD in men and polymorphism in exon 7 (Ala192Val) of RANK . OPG is a known regulator of bone remodelling which, together with RANK , is central to osteoclastogenesis and activation of bone resorption. Osteopontin ( OPN ), which is produced as adhesive glycophosphoprotein in different tissues such as, bone, teeth and kidney, is involved in extracellular matrix organization, bone metabolism, angiogenesis, immune regulation and is essential for osteoclast function. Two OPN promoter SNPs (rs11730582 and rs28357094) that have been shown to affect the transcription level of OPN were selected for this study .

The SNPs selected are either reported in the literature to be associated with risk for osteoporosis or bone remodelling or nonsynonymous SNPs that are considered to be functional with minor allele frequency of >5% or showed more significance than other SNPs in a study (e.g. CYP 2C8) . The 10 SNPs included in this study are: COL1A1 (rs1800012), RANK (rs12458117), CYP2C8 (rs1934980 and rs1934951, the 2 strongest SNPs from previous report), MMP2 (rs243865), OPN (rs11730582 and rs28357094), OPG (rs2073618 and rs3102735) and TNF (rs1800629).

Genotyping

Genomic DNA was extracted from lymphocytes in whole blood using a commercially available kit (Qiagen DNA Blood Isolation Kit, Qiagen, Valencia, CA, USA). Polymorphisms were genotyped by Taqman ^® genotyping method or pyrosequencing . The Applied Biosystems 7900 HT SNP genotyping platform was used for the Taqman ^® assay. The polymerase chain reaction (PCR) primers and probes for COL1A1 SNP G/T rs1800012 (ID C___7477170_30), RANK SNP A/G rs12458117 (ID C__31393804_10), CYP2C8 SNP C/T rs1934980 (ID C___361427_10), CYP2C8 SNP A/G rs1934951 (ID C___361409_1_) and MMP2 SNP C/T rs243865 (ID C_3225943_10) were purchased from Applied Biosystems (Applied Biosystems, Foster City, USA). Reactions of 5 μl each in 384-well plate were prepared and the assays were performed and analysed according to the manufacturer’s recommendations. Pyrosequencing (Biotage, Uppsala, Sweden) was performed on the rest of the SNPs using a PSQ HS96A SNP reagent kit (Biotage AB, Uppsala, Sweden) (primers available upon request). Genotype accuracy was verified by genotyping 5–10% randomly selected duplicate samples for each SNP.

Statistical analysis

Baseline characteristics were compared using Student’s t -test or the χ ² -test as appropriate. Allele frequencies and deviations from Hardy–Weinberg equilibrium were assessed within each race group using the χ ² -test with one degree of freedom. A dominant mode of inheritance was assumed in single SNP association analyses. The event rates were compared between carriers and non-carriers using χ ² tests. Logistic regression was used to assess the odds ratios (OR) for BONJ, using wild-type homozygous as the reference group.

To study the potential combined effect of all five SNPs that showed a trend for statistical significance ( OPG _rs2073618, COL1A1 _rs1800012, RANK _rs12458117, MMP2 _rs243865 and OPN _rs11730582), the authors constructed genotype scores on the basis of the number of unfavourable alleles (those associated with higher risk for BONJ) that were carried by each subject for each of the five SNPs. The genotype score for each patient was calculated as the sum of genotype scores for all five SNPs, each with a possible value of 0 (homozygous for common allele), 1 (heterozygous) or 2 (homozygous for rare allele). Based on the distribution of the genotype scores, the genotype scores were dichotomized into < or ≥5 (at least one rare allele for each of the five SNPs on average). The crude incidence rates of BONJ were calculated according to strata of genotype scores of < or ≥5. The genotype scoring method has been widely used to evaluate multiple gene effects on complex diseases . In the multivariate logistic regression, covariates such as age, gender, race/ethnicity, smoking history, diabetes, type of BP treatment and number of prior treatments were included in a stepwise model selection analysis (with selection criteria of p = 0.1 for entry and p = 0.05 for staying). Only the significant risk factors for BONJ were included in the final multivariate logistic model.

Power analysis revealed that with sample size of 12 cases and 66 controls, assuming additive mode of inheritance, at alpha of 0.05, there is >80% power to detect ORs of >5 with minor allele frequency of 10% and ORs of >4 with minor allele frequency of 20%. Pharmacogenetic effect sizes tend to be much larger than ORs commonly seen in disease association studies . Thus, whilst the ORs used for these power calculations may not be realistic for disease association studies, the preliminary data, and numerous lines of evidence in other pharmacogenetic studies, suggest these are reasonable ORs for the pharmacogenetic questions being addressed. All statistical analyses were performed using SAS version 9.1 (SAS Institute, Cary, NC, USA). A p value of <0.05 was considered statistically significant.