Truth Seeker
New Member
Abstract
Osteoporosis is one of the most common degenerative diseases. It is characterized by reduced bone mineral density (BMD) with an increased risk for bone fractures. There is a substantial genetic contribution to BMD, although the genetic factors involved in the pathogenesis of human osteoporosis are largely unknown. Mice with a targeted deletion of either the cannabinoid receptor type 1 (Cnr1) or type 2 (Cnr2) gene show an alteration of bone mass, and pharmacological modification of both receptors can regulate osteoclast activity and BMD. We therefore analyzed both genes in a systematic genetic association study in a human sample of postmenopausal osteoporosis patients and matched female controls. We found a significant association of single polymorphisms (P=0.0014) and haplotypes (P=0.0001) encompassing the CNR2 gene on human chromosome 1p36, whereas we found no convincing association for CNR1. These results demonstrate a role for the peripherally expressed CB2 receptor in the etiology of osteoporosis and provide an interesting novel therapeutical target for this severe and common disease.
INTRODUCTION
A significant decrease in bone mineral density (BMD) leads to an enhanced liability to bone fractures and is a major denominator of human osteoporosis. Osteoporosis is one of the most common degenerative disorders in developed countries. Although it affects both sexes and all racial groups, postmenopausal women are at the highest risk. The etiology of osteoporosis is heterogeneous; the main predictors are age, gender and a positive family history. The inheritance pattern is complex, and numerous families and twin studies have demonstrated a pronounced genetic contribution to osteoporosis and BMD with average heritability estimates in the range of 60—80% (1). Although the molecular regulation of BMD is largely unknown, variants in a few genes, including the collagen 1A1-gene (2), the vitamin D receptor gene (3) and the low density lipoprotein receptor-related protein 5 gene (4), have repeatedly been shown to be associated with bone mass. Moreover, the bone morphogenetic protein 2 gene (5) has been identified as a human osteoporosis susceptibility gene by a positional cloning approach. In addition, a few chromosomal loci including human chromosome 1p36 have been replicated in independent genome-wide linkage studies; however, the underlying susceptibility genes are unknown (6—8).
Very recently, an unexpected role of the endocannabinoid system in the regulation of bone mass has been demonstrated in mice (9,10). To date, two cannabinoid receptors, CB1 and CB2, have been identified. Both proteins are highly homologous and belong to the family of G-protein coupled seven-transmembrane domain receptors, which bind and are activated by endocannabinoids. CNR1 is predominantly expressed in the brain and peripheral neurons and is responsible for the psychotropic action of cannabinoids (11,12), whereas CNR2 is absent in the brain but is mainly expressed in immune cells (13). The neuronal CB1 receptor is encoded by the CNR1 gene on human chromosome 6q15, whereas the non-neuronal CB2 receptor is encoded by the CNR2 gene on 1p36. Mice with a targeted deletion of Cnr1 have an increased bone mass (10), whereas we could recently show that Cnr2 knockout mice have a decreased bone mass reminiscent of human osteoporosis (9). Both studies demonstrated that the pharmacological modulation of CB receptors affects bone mass in vivo, although the pharmacological results were contradictory.
Because these mouse genetic data suggest a role for the endocannabinoid system in the regulation of bone mass and because the chromosomal region 1p36, which contains the CNR2 gene, has been implicated in osteoporosis (6—8), we have therefore studied the involvement of CNR1 and/or CNR2 in the pathogenesis of complex inherited human osteoporosis by pursuing a systematic genetic case—control association study.
RESULTS
Around the human CNR1 gene, we genotyped four single nucleotide polymorphisms (SNPs) (rs806365, rs1049353, rs806377 and rs806380), for which genotyping assays were available (Table 1), covering a region of nearly 20 kb and encompassing the single coding exon of CNR1. The genotype distributions of the two samples did not deviate significantly (P≥0.05) from those expected from the Hardy—Weinberg equilibrium (HWE), as tested by the exact test. The distribution of genotypes of the four CNR1 polymorphisms is shown in Table 1. To compare the frequencies of individual SNPs between cases and controls, we applied a standard χ2 test (df=1) for allelic association with the disease phenotype. No statistically significant differences of allele or genotype distributions were found between patients and controls, but SNP rs806380 showed a borderline significant P-value of 0.057. In conclusion, our data do not support a major role of CNR1 in our sample.
The human CNR2 gene and its mouse ortholog are located on chromosomes 1p36 and 4QD3, respectively. These two genomic regions have been linked to osteoporosis and BMD by several independent genome-wide linkage analyses (6—8,14,15); however, the CNR2 gene has not been investigated as a positional candidate gene up to now. CB2 is encoded by a single exon (exon 2) with a non-coding exon 1 ∼37 kb upstream. We initially tested six SNPs (Table 2) for which genotyping assays were available in a region spanning ∼120 kb (Fig. 1A) and harboring both CNR2 exons. Again, there was no deviation of genotype distributions from the HWE, but we found a significant difference of allele frequencies between cases and controls for several SNPs (Table 3 and Fig. 1B). In contrast, seven unrelated control SNPs (intergenic and not in the vicinity of the CNR genes) from different chromosomes did not show an association with the phenotype (Table 4). Therefore, these results suggest an involvement of the CNR2 locus in human osteoporosis.
Both SNPs located within exon 2 of CNR2, i.e. rs4649124 and rs2501431, showed most significant P-values for allele (0.0026 and 0.0014, respectively) and genotype (0.0023 and 0.00073, respectively) distributions (Table 3 and Fig. 1B). These two SNPs are in tight linkage disequilibrium (LD) (D′=0.99 and r2=0.96), which is also true for most of the other SNPs within exon 2 (Fig. 1C). For rs4649124, the allele frequencies differed by 11% (allele 1: controls 52% and patients 63%). The other four SNPs, located 10 kb/30 kb 3′ and 80 kb/93 kb 5′ of CNR2, showed P-values for allelic association between 0.0042 (rs4649119, which is in strong LD to SNPs in exon 2; D′=0.98) and 0.2275 (hCV395615, which just shows a weak LD to SNPs in exon 2; D′=0.55) (Table 3 and Fig. 1B). In order to correct for multiple testing, we repeated the analysis 100 000 times with a randomly permuted assignment of the affection status. The obtained P-value (P=0.0056) of the most strongly associated SNP (rs2501431) remains significant, if we further perform a Bonferroni correction for the two investigated gene loci.
As there may be a skeletal site-specific genetic predisposition to osteoporosis [e.g. at the lumbar spine or femoral neck (FN)], we also used subsamples of our patients for further analysis. When restricting the patient sample to those with a T-score less than −2.5 at the lumbar spine (86% of the full patient sample), we still observe highly significant associations for both SNPs within the coding exon (e.g. for SNP rs4649124, P=0.005). The small size of a patient sample defined by a T-score less than −2.5 only at the FN does not allow for a valid statistical analysis. We next analyzed BMD at the lumbar spine as a quantitative trait. To this end, we tested the normal distribution of the quantitative data using the Kolmogorov—Smirnov test and the homogeneity variances with Levene's test. Congruent with the association results in the dichotomized sample, we found a highly significant difference in bone density between individuals displaying different genotypes (e.g. for SNP rs4649124, P=0.004, one-way analysis of variance) (data not shown).
To investigate the extended locus and its genetic variability more systematically, we sequenced the coding region of CNR2 in all individuals (n=388) to obtain complete information about all sequence variants. As the CNR2 coding sequence includes only a small fraction of the genomic region where significantly associated SNPs were located, we further determined the LD structure using genotypes from the Perlegen data set (16) (because of a gap in the NCBI 34 genome assembly, the current phase 1 public HapMap does not contain the CNR2 locus) (Fig. 1C). We partitioned the locus into haplotype blocks using the program hapblock (17) (USC QCB | Quantitative and Computational Biology | University of Southern California). Block partitioning required that at least 90% of the total haplotype diversity within each block was covered by common haplotypes (>5% population frequency). We selected SNPs necessary for common haplotype discrimination as tagging SNPs, altogether we genotyped additional 20 SNPs in the 300 kb extended region (Fig. 1A and B and Tables 2 and 3). The genotype data of this comprehensive set of SNPs confirmed the positive association with most significant values within a LD block including exon 2 and the 3′-untranslated region (UTR) of CNR2. The most significant association remained to be found for rs2501431 in exon 2 (Fig. 1B and Table 3). SNPs telomeric and centromeric were less or not significant.
In order to devise the genomic region containing putative disease mutations, we tested the global haplotype frequency distribution for all possible three-marker haplotypes. The smallest P-value (P=0.0001) was found for the SNP combination rs2229583—rs2501431—rs16828926 (Fig. 1B). This haplotype spans exon 2 and a small part of intron 1 of CNR2. Its P-value is an order of magnitude less than the global P-value (P=0.005) for the SNP combination rs4649124—rs3003336—rs2501431, which is located directly under the SNP association peak within the CNR2 locus. The reason for this decreased P-value became clearer when we looked at the frequency of the major haplotype in the case and the control samples (Supplementary Material, Table S1). As indicated by the strong LD between the associated SNPs, a major risk haplotype is enriched among patients. Under the SNP association peak, the major haplotype G-T-A comprising the three SNPs rs4649124—rs3003336—rs2501431 displayed an enrichment among cases comparable with the associated individual SNPs (62.1 versus 51.1%, χ2=9.63, P=0.002). When increasing the size of the genomic window towards both sides, a stronger fragmentation of the risk haplotype was observed among the controls than among cases. This fragmentation comes in parallel with the decreased P-value and makes the respective genomic region most likely to harbor the unknown disease causing mutations. The major haplotype comprising the alleles T-A-G of the SNPs rs2229583—rs2501431—rs16828926 displays a frequency of 59.3% among patients and 46.0% among controls (χ2=13.12, P=0.0003) (Supplementary Material, Table S1). This gives an odds ratio (OR) of 1.716 and an estimated etiological fraction of 24.7% for this haplotype. Further increase in the size of the genomic region led to the fragmentation of the major haplotype also in the case sample, resulting in less significant frequency differences between the two samples. Thus, the haplotype analysis confirmed and even strengthened our positive association results with the single markers and provides further evidence for a causative role of the CNR2 gene in human postmenopausal osteoporosis.
DISCUSSION
Chromosome 1p36 first emerged as susceptibility locus for low hip BMD in a genome-wide linkage study in seven extended pedigrees (6). In non-parametric analyses, the maximum Z-score for this locus was 3.51 at marker D1S450. This marker is ∼14 Mb telomeric from CNR2 (UCSC-Genome Browser), but as loci in complex genetics are in general rather broad and the borders are not exactly defined, CNR2 has to be regarded as a good candidate gene for this locus. The same locus has been confirmed in an extended study with 42 families, with a maximum multipoint LOD score of 3.53 for FN BMD (7). Also in mice, there is very good evidence from several genome-wide linkage studies that the chromosomal region around Cnr2 is implicated in the regulation of BMD. The Cnr2 gene is located on mouse chromosome 4 at 134.3 Mb very close to the marker D4Mit134 at 134.1 Mb, for which genetic localization is at 62 cM. A major quantitative trait locus (QTL, named Bmd7) for both femoral and lumbar vertebral BMD in C57BL/6 and C3H/HeJ mice strains has been mapped close to this locus. The maximum LOD score was 16.3 for femoral and 14.8 for vertebral BMD at D4Mit124 (57.4 cM), which accounts for 6.7 and 6.5% of F2 variance in femoral and vertebral BMD, respectively (14). A similar highly significant finding has been obtained in a screen for loci affecting peak bone mass in 24 recombinant inbred mouse strains with a maximum LOD score of 12.3 for D4Mit312 at 69.8 cM (15).
In contrast, the CNR1 locus on chromosome 6q14 does not seem to be a major QTL for BMD, which is in accordance with our association results. As the genomic region covering CNR1 is small (∼5.5 kb) and shows a rather high LD, the testing of four SNPs covering most of the genetic variance is probably sufficient to exclude a major role of CNR1 in our sample.
Although the physiological role of endocannabinoids and CB1 receptors in the regulation of neuronal signaling has been well studied, relatively little is known about the functions of CB2 receptors or the involvement of cannabinoid receptors in human disease processes. Indeed, our results provide the first evidence for an involvement of CB2 receptors in a human disorder. In a recent study, we have demonstrated that CB2 activation can ameliorate the development of atherosclerotic plaques in mice (18). Very interestingly, atherosclerosis, also a common and late-onset disorder, is inversely and independently correlated with BMD as shown, e.g. by a systematic population-based study in over 5200 Norwegian individuals (19). Thus, it is tempting to speculate that CNR2 polymorphisms contribute to the etiology of both disorders, thus providing one explanation for the association of atherosclerosis with osteoporosis.
Among the large number of associated polymorphisms in CNR2 are two missense variants (the double SNP rs2502992—rs2501432 and Gln63Arg; rs2229579 and His316Tyr; both amino acid exchanges are in non-conserved regions) and several SNPs within conserved regions in the 3′-UTR. These alone or in combination with other linked polymorphisms, still to be identified, may be responsible for the observed association by altering protein function and/or gene expression. On the basis of the Cnr2 knockout phenotype, and our pharmacological results (9), we would expect a decreased CB2 function in the disease-associated allele, which would also be expected to provide a protective function in atherosclerosis. Obviously, further studies will be necessary to address the functional role of CB2 receptors in these common disorders. Eventually, these studies may open novel therapeutic strategies.
MATERIALS AND METHODS
Clinical data
All 388 women were recruited and investigated at a single clinical center (Hôpital Lariboisière, Paris, France). The patients were referred for suspicion of osteoporosis. The controls were women addressed to have a BMD measurement (Viggos cohort) to evaluate their bone status. They were not referred for suspicion of osteoporosis. They were all postmenopausal Caucasian of French origin and >50 years old. All participants signed an informed consent, and the study was approved by the hospital Internal Ethical Review Board. All women underwent a routine clinical and biochemical assessment to exclude secondary causes of osteoporosis. Subjects who had plasma hormone abnormalities [thyroid stimulating hormone and parathyroid hormone (PTH)] or had received treatment known to affect BMD for more than a month were also excluded. The patients and controls underwent BMD measurements of the lumbar spine (L2—L4) or FN, determined by dual energy X-ray absorptiometry (DEXA) (LUNAR DPX-L, Madison, WI, USA). T-scores were based on a French population reference provided by the manufacturer. X-ray of the spine was performed in all women to evaluate semi-quantitatively the prevalence of vertebral osteoporotic fractures. Women were defined as having osteoporosis on the basis of a BMD T-score less than −2.5 at lumbar spine (L2—L4) or FN. Some women of the Viggos study (∼10%) had a T-score less than −2.5 and were thus classified as having osteoporosis. Therefore, our sample consisted of 168 postmenopausal osteoporotic women and 220 ethnically, age- and sex-matched controls. In the osteoporotic women, the prevalence of vertebral fractures was 46%. The control women had BMD T-score of ≥−2.5 and had no previous FN or vertebral lumbar fractures. Further details of the patient and control clinical data are given in Table 5.
Molecular genetic studies and statistical analyses
Genomic DNA was extracted by standard procedures from EDTA-blood samples, and SNPs were genotyped by polymerase chain reaction (PCR) amplification using the ABI Assays-On-Demand/Design SNP Genotyping Products or assessed by direct sequencing (Table 2). As means of quality control, we (i) always included negative controls as well as individuals who are known to have the three different genotypes in the ABI assays, (ii) typed SNP rs4649119 in duplicate and (iii) typed SNPs rs2229586, rs4649124 and rs2501431 by two independent methods (ABI assays and direct sequencing). The discrepancies between both methods were <0.4% of tested chromosomes.
Global haplotype comparison was performed using the package cocaphase (UNPHASED). This software performs likelihood ratio tests under a log-linear model of the probability that a haplotype belongs to the case or the control group, using a standard unconditional logistic regression. The expectation-maximization (EM) algorithm was used to estimate haplotype frequencies and resolve missing genotypes. Maximum-likelihood haplotype frequencies were further estimated from individual genotypes using the program snphap (https://archimedes.well.ox.ac.uk/pise/snphap.html), which implements an EM algorithm. Haplotypes were then tested against all other haplotypes for disease association by a χ2 test (df=1). Frequency estimates of the two programs were in good agreement both for common and for rare haplotypes.
RESULTS
Around the human CNR1 gene, we genotyped four single nucleotide polymorphisms (SNPs) (rs806365, rs1049353, rs806377 and rs806380), for which genotyping assays were available (Table 1), covering a region of nearly 20 kb and encompassing the single coding exon of CNR1. The genotype distributions of the two samples did not deviate significantly (P≥0.05) from those expected from the Hardy—Weinberg equilibrium (HWE), as tested by the exact test. The distribution of genotypes of the four CNR1 polymorphisms is shown in Table 1. To compare the frequencies of individual SNPs between cases and controls, we applied a standard χ2 test (df=1) for allelic association with the disease phenotype. No statistically significant differences of allele or genotype distributions were found between patients and controls, but SNP rs806380 showed a borderline significant P-value of 0.057. In conclusion, our data do not support a major role of CNR1 in our sample.
The human CNR2 gene and its mouse ortholog are located on chromosomes 1p36 and 4QD3, respectively. These two genomic regions have been linked to osteoporosis and BMD by several independent genome-wide linkage analyses (6—8,14,15); however, the CNR2 gene has not been investigated as a positional candidate gene up to now. CB2 is encoded by a single exon (exon 2) with a non-coding exon 1 ∼37 kb upstream. We initially tested six SNPs (Table 2) for which genotyping assays were available in a region spanning ∼120 kb (Fig. 1A) and harboring both CNR2 exons. Again, there was no deviation of genotype distributions from the HWE, but we found a significant difference of allele frequencies between cases and controls for several SNPs (Table 3 and Fig. 1B). In contrast, seven unrelated control SNPs (intergenic and not in the vicinity of the CNR genes) from different chromosomes did not show an association with the phenotype (Table 4). Therefore, these results suggest an involvement of the CNR2 locus in human osteoporosis.
Both SNPs located within exon 2 of CNR2, i.e. rs4649124 and rs2501431, showed most significant P-values for allele (0.0026 and 0.0014, respectively) and genotype (0.0023 and 0.00073, respectively) distributions (Table 3 and Fig. 1B). These two SNPs are in tight linkage disequilibrium (LD) (D′=0.99 and r2=0.96), which is also true for most of the other SNPs within exon 2 (Fig. 1C). For rs4649124, the allele frequencies differed by 11% (allele 1: controls 52% and patients 63%). The other four SNPs, located 10 kb/30 kb 3′ and 80 kb/93 kb 5′ of CNR2, showed P-values for allelic association between 0.0042 (rs4649119, which is in strong LD to SNPs in exon 2; D′=0.98) and 0.2275 (hCV395615, which just shows a weak LD to SNPs in exon 2; D′=0.55) (Table 3 and Fig. 1B). In order to correct for multiple testing, we repeated the analysis 100 000 times with a randomly permuted assignment of the affection status. The obtained P-value (P=0.0056) of the most strongly associated SNP (rs2501431) remains significant, if we further perform a Bonferroni correction for the two investigated gene loci.
As there may be a skeletal site-specific genetic predisposition to osteoporosis [e.g. at the lumbar spine or femoral neck (FN)], we also used subsamples of our patients for further analysis. When restricting the patient sample to those with a T-score less than −2.5 at the lumbar spine (86% of the full patient sample), we still observe highly significant associations for both SNPs within the coding exon (e.g. for SNP rs4649124, P=0.005). The small size of a patient sample defined by a T-score less than −2.5 only at the FN does not allow for a valid statistical analysis. We next analyzed BMD at the lumbar spine as a quantitative trait. To this end, we tested the normal distribution of the quantitative data using the Kolmogorov—Smirnov test and the homogeneity variances with Levene's test. Congruent with the association results in the dichotomized sample, we found a highly significant difference in bone density between individuals displaying different genotypes (e.g. for SNP rs4649124, P=0.004, one-way analysis of variance) (data not shown).
To investigate the extended locus and its genetic variability more systematically, we sequenced the coding region of CNR2 in all individuals (n=388) to obtain complete information about all sequence variants. As the CNR2 coding sequence includes only a small fraction of the genomic region where significantly associated SNPs were located, we further determined the LD structure using genotypes from the Perlegen data set (16) (because of a gap in the NCBI 34 genome assembly, the current phase 1 public HapMap does not contain the CNR2 locus) (Fig. 1C). We partitioned the locus into haplotype blocks using the program hapblock (17) (USC QCB | Quantitative and Computational Biology | University of Southern California). Block partitioning required that at least 90% of the total haplotype diversity within each block was covered by common haplotypes (>5% population frequency). We selected SNPs necessary for common haplotype discrimination as tagging SNPs, altogether we genotyped additional 20 SNPs in the 300 kb extended region (Fig. 1A and B and Tables 2 and 3). The genotype data of this comprehensive set of SNPs confirmed the positive association with most significant values within a LD block including exon 2 and the 3′-untranslated region (UTR) of CNR2. The most significant association remained to be found for rs2501431 in exon 2 (Fig. 1B and Table 3). SNPs telomeric and centromeric were less or not significant.
In order to devise the genomic region containing putative disease mutations, we tested the global haplotype frequency distribution for all possible three-marker haplotypes. The smallest P-value (P=0.0001) was found for the SNP combination rs2229583—rs2501431—rs16828926 (Fig. 1B). This haplotype spans exon 2 and a small part of intron 1 of CNR2. Its P-value is an order of magnitude less than the global P-value (P=0.005) for the SNP combination rs4649124—rs3003336—rs2501431, which is located directly under the SNP association peak within the CNR2 locus. The reason for this decreased P-value became clearer when we looked at the frequency of the major haplotype in the case and the control samples (Supplementary Material, Table S1). As indicated by the strong LD between the associated SNPs, a major risk haplotype is enriched among patients. Under the SNP association peak, the major haplotype G-T-A comprising the three SNPs rs4649124—rs3003336—rs2501431 displayed an enrichment among cases comparable with the associated individual SNPs (62.1 versus 51.1%, χ2=9.63, P=0.002). When increasing the size of the genomic window towards both sides, a stronger fragmentation of the risk haplotype was observed among the controls than among cases. This fragmentation comes in parallel with the decreased P-value and makes the respective genomic region most likely to harbor the unknown disease causing mutations. The major haplotype comprising the alleles T-A-G of the SNPs rs2229583—rs2501431—rs16828926 displays a frequency of 59.3% among patients and 46.0% among controls (χ2=13.12, P=0.0003) (Supplementary Material, Table S1). This gives an odds ratio (OR) of 1.716 and an estimated etiological fraction of 24.7% for this haplotype. Further increase in the size of the genomic region led to the fragmentation of the major haplotype also in the case sample, resulting in less significant frequency differences between the two samples. Thus, the haplotype analysis confirmed and even strengthened our positive association results with the single markers and provides further evidence for a causative role of the CNR2 gene in human postmenopausal osteoporosis.
DISCUSSION
Chromosome 1p36 first emerged as susceptibility locus for low hip BMD in a genome-wide linkage study in seven extended pedigrees (6). In non-parametric analyses, the maximum Z-score for this locus was 3.51 at marker D1S450. This marker is ∼14 Mb telomeric from CNR2 (UCSC-Genome Browser), but as loci in complex genetics are in general rather broad and the borders are not exactly defined, CNR2 has to be regarded as a good candidate gene for this locus. The same locus has been confirmed in an extended study with 42 families, with a maximum multipoint LOD score of 3.53 for FN BMD (7). Also in mice, there is very good evidence from several genome-wide linkage studies that the chromosomal region around Cnr2 is implicated in the regulation of BMD. The Cnr2 gene is located on mouse chromosome 4 at 134.3 Mb very close to the marker D4Mit134 at 134.1 Mb, for which genetic localization is at 62 cM. A major quantitative trait locus (QTL, named Bmd7) for both femoral and lumbar vertebral BMD in C57BL/6 and C3H/HeJ mice strains has been mapped close to this locus. The maximum LOD score was 16.3 for femoral and 14.8 for vertebral BMD at D4Mit124 (57.4 cM), which accounts for 6.7 and 6.5% of F2 variance in femoral and vertebral BMD, respectively (14). A similar highly significant finding has been obtained in a screen for loci affecting peak bone mass in 24 recombinant inbred mouse strains with a maximum LOD score of 12.3 for D4Mit312 at 69.8 cM (15).
In contrast, the CNR1 locus on chromosome 6q14 does not seem to be a major QTL for BMD, which is in accordance with our association results. As the genomic region covering CNR1 is small (∼5.5 kb) and shows a rather high LD, the testing of four SNPs covering most of the genetic variance is probably sufficient to exclude a major role of CNR1 in our sample.
Although the physiological role of endocannabinoids and CB1 receptors in the regulation of neuronal signaling has been well studied, relatively little is known about the functions of CB2 receptors or the involvement of cannabinoid receptors in human disease processes. Indeed, our results provide the first evidence for an involvement of CB2 receptors in a human disorder. In a recent study, we have demonstrated that CB2 activation can ameliorate the development of atherosclerotic plaques in mice (18). Very interestingly, atherosclerosis, also a common and late-onset disorder, is inversely and independently correlated with BMD as shown, e.g. by a systematic population-based study in over 5200 Norwegian individuals (19). Thus, it is tempting to speculate that CNR2 polymorphisms contribute to the etiology of both disorders, thus providing one explanation for the association of atherosclerosis with osteoporosis.
Among the large number of associated polymorphisms in CNR2 are two missense variants (the double SNP rs2502992—rs2501432 and Gln63Arg; rs2229579 and His316Tyr; both amino acid exchanges are in non-conserved regions) and several SNPs within conserved regions in the 3′-UTR. These alone or in combination with other linked polymorphisms, still to be identified, may be responsible for the observed association by altering protein function and/or gene expression. On the basis of the Cnr2 knockout phenotype, and our pharmacological results (9), we would expect a decreased CB2 function in the disease-associated allele, which would also be expected to provide a protective function in atherosclerosis. Obviously, further studies will be necessary to address the functional role of CB2 receptors in these common disorders. Eventually, these studies may open novel therapeutic strategies.
MATERIALS AND METHODS
Clinical data
All 388 women were recruited and investigated at a single clinical center (Hôpital Lariboisière, Paris, France). The patients were referred for suspicion of osteoporosis. The controls were women addressed to have a BMD measurement (Viggos cohort) to evaluate their bone status. They were not referred for suspicion of osteoporosis. They were all postmenopausal Caucasian of French origin and >50 years old. All participants signed an informed consent, and the study was approved by the hospital Internal Ethical Review Board. All women underwent a routine clinical and biochemical assessment to exclude secondary causes of osteoporosis. Subjects who had plasma hormone abnormalities [thyroid stimulating hormone and parathyroid hormone (PTH)] or had received treatment known to affect BMD for more than a month were also excluded. The patients and controls underwent BMD measurements of the lumbar spine (L2—L4) or FN, determined by dual energy X-ray absorptiometry (DEXA) (LUNAR DPX-L, Madison, WI, USA). T-scores were based on a French population reference provided by the manufacturer. X-ray of the spine was performed in all women to evaluate semi-quantitatively the prevalence of vertebral osteoporotic fractures. Women were defined as having osteoporosis on the basis of a BMD T-score less than −2.5 at lumbar spine (L2—L4) or FN. Some women of the Viggos study (∼10%) had a T-score less than −2.5 and were thus classified as having osteoporosis. Therefore, our sample consisted of 168 postmenopausal osteoporotic women and 220 ethnically, age- and sex-matched controls. In the osteoporotic women, the prevalence of vertebral fractures was 46%. The control women had BMD T-score of ≥−2.5 and had no previous FN or vertebral lumbar fractures. Further details of the patient and control clinical data are given in Table 5.
Molecular genetic studies and statistical analyses
Genomic DNA was extracted by standard procedures from EDTA-blood samples, and SNPs were genotyped by polymerase chain reaction (PCR) amplification using the ABI Assays-On-Demand/Design SNP Genotyping Products or assessed by direct sequencing (Table 2). As means of quality control, we (i) always included negative controls as well as individuals who are known to have the three different genotypes in the ABI assays, (ii) typed SNP rs4649119 in duplicate and (iii) typed SNPs rs2229586, rs4649124 and rs2501431 by two independent methods (ABI assays and direct sequencing). The discrepancies between both methods were <0.4% of tested chromosomes.
Global haplotype comparison was performed using the package cocaphase (UNPHASED). This software performs likelihood ratio tests under a log-linear model of the probability that a haplotype belongs to the case or the control group, using a standard unconditional logistic regression. The expectation-maximization (EM) algorithm was used to estimate haplotype frequencies and resolve missing genotypes. Maximum-likelihood haplotype frequencies were further estimated from individual genotypes using the program snphap (https://archimedes.well.ox.ac.uk/pise/snphap.html), which implements an EM algorithm. Haplotypes were then tested against all other haplotypes for disease association by a χ2 test (df=1). Frequency estimates of the two programs were in good agreement both for common and for rare haplotypes.
Source, Graphs and Figures: Cannabinoid receptor type 2 gene is associated with human osteoporosis
Osteoporosis is one of the most common degenerative diseases. It is characterized by reduced bone mineral density (BMD) with an increased risk for bone fractures. There is a substantial genetic contribution to BMD, although the genetic factors involved in the pathogenesis of human osteoporosis are largely unknown. Mice with a targeted deletion of either the cannabinoid receptor type 1 (Cnr1) or type 2 (Cnr2) gene show an alteration of bone mass, and pharmacological modification of both receptors can regulate osteoclast activity and BMD. We therefore analyzed both genes in a systematic genetic association study in a human sample of postmenopausal osteoporosis patients and matched female controls. We found a significant association of single polymorphisms (P=0.0014) and haplotypes (P=0.0001) encompassing the CNR2 gene on human chromosome 1p36, whereas we found no convincing association for CNR1. These results demonstrate a role for the peripherally expressed CB2 receptor in the etiology of osteoporosis and provide an interesting novel therapeutical target for this severe and common disease.
INTRODUCTION
A significant decrease in bone mineral density (BMD) leads to an enhanced liability to bone fractures and is a major denominator of human osteoporosis. Osteoporosis is one of the most common degenerative disorders in developed countries. Although it affects both sexes and all racial groups, postmenopausal women are at the highest risk. The etiology of osteoporosis is heterogeneous; the main predictors are age, gender and a positive family history. The inheritance pattern is complex, and numerous families and twin studies have demonstrated a pronounced genetic contribution to osteoporosis and BMD with average heritability estimates in the range of 60—80% (1). Although the molecular regulation of BMD is largely unknown, variants in a few genes, including the collagen 1A1-gene (2), the vitamin D receptor gene (3) and the low density lipoprotein receptor-related protein 5 gene (4), have repeatedly been shown to be associated with bone mass. Moreover, the bone morphogenetic protein 2 gene (5) has been identified as a human osteoporosis susceptibility gene by a positional cloning approach. In addition, a few chromosomal loci including human chromosome 1p36 have been replicated in independent genome-wide linkage studies; however, the underlying susceptibility genes are unknown (6—8).
Very recently, an unexpected role of the endocannabinoid system in the regulation of bone mass has been demonstrated in mice (9,10). To date, two cannabinoid receptors, CB1 and CB2, have been identified. Both proteins are highly homologous and belong to the family of G-protein coupled seven-transmembrane domain receptors, which bind and are activated by endocannabinoids. CNR1 is predominantly expressed in the brain and peripheral neurons and is responsible for the psychotropic action of cannabinoids (11,12), whereas CNR2 is absent in the brain but is mainly expressed in immune cells (13). The neuronal CB1 receptor is encoded by the CNR1 gene on human chromosome 6q15, whereas the non-neuronal CB2 receptor is encoded by the CNR2 gene on 1p36. Mice with a targeted deletion of Cnr1 have an increased bone mass (10), whereas we could recently show that Cnr2 knockout mice have a decreased bone mass reminiscent of human osteoporosis (9). Both studies demonstrated that the pharmacological modulation of CB receptors affects bone mass in vivo, although the pharmacological results were contradictory.
Because these mouse genetic data suggest a role for the endocannabinoid system in the regulation of bone mass and because the chromosomal region 1p36, which contains the CNR2 gene, has been implicated in osteoporosis (6—8), we have therefore studied the involvement of CNR1 and/or CNR2 in the pathogenesis of complex inherited human osteoporosis by pursuing a systematic genetic case—control association study.
RESULTS
Around the human CNR1 gene, we genotyped four single nucleotide polymorphisms (SNPs) (rs806365, rs1049353, rs806377 and rs806380), for which genotyping assays were available (Table 1), covering a region of nearly 20 kb and encompassing the single coding exon of CNR1. The genotype distributions of the two samples did not deviate significantly (P≥0.05) from those expected from the Hardy—Weinberg equilibrium (HWE), as tested by the exact test. The distribution of genotypes of the four CNR1 polymorphisms is shown in Table 1. To compare the frequencies of individual SNPs between cases and controls, we applied a standard χ2 test (df=1) for allelic association with the disease phenotype. No statistically significant differences of allele or genotype distributions were found between patients and controls, but SNP rs806380 showed a borderline significant P-value of 0.057. In conclusion, our data do not support a major role of CNR1 in our sample.
The human CNR2 gene and its mouse ortholog are located on chromosomes 1p36 and 4QD3, respectively. These two genomic regions have been linked to osteoporosis and BMD by several independent genome-wide linkage analyses (6—8,14,15); however, the CNR2 gene has not been investigated as a positional candidate gene up to now. CB2 is encoded by a single exon (exon 2) with a non-coding exon 1 ∼37 kb upstream. We initially tested six SNPs (Table 2) for which genotyping assays were available in a region spanning ∼120 kb (Fig. 1A) and harboring both CNR2 exons. Again, there was no deviation of genotype distributions from the HWE, but we found a significant difference of allele frequencies between cases and controls for several SNPs (Table 3 and Fig. 1B). In contrast, seven unrelated control SNPs (intergenic and not in the vicinity of the CNR genes) from different chromosomes did not show an association with the phenotype (Table 4). Therefore, these results suggest an involvement of the CNR2 locus in human osteoporosis.
Both SNPs located within exon 2 of CNR2, i.e. rs4649124 and rs2501431, showed most significant P-values for allele (0.0026 and 0.0014, respectively) and genotype (0.0023 and 0.00073, respectively) distributions (Table 3 and Fig. 1B). These two SNPs are in tight linkage disequilibrium (LD) (D′=0.99 and r2=0.96), which is also true for most of the other SNPs within exon 2 (Fig. 1C). For rs4649124, the allele frequencies differed by 11% (allele 1: controls 52% and patients 63%). The other four SNPs, located 10 kb/30 kb 3′ and 80 kb/93 kb 5′ of CNR2, showed P-values for allelic association between 0.0042 (rs4649119, which is in strong LD to SNPs in exon 2; D′=0.98) and 0.2275 (hCV395615, which just shows a weak LD to SNPs in exon 2; D′=0.55) (Table 3 and Fig. 1B). In order to correct for multiple testing, we repeated the analysis 100 000 times with a randomly permuted assignment of the affection status. The obtained P-value (P=0.0056) of the most strongly associated SNP (rs2501431) remains significant, if we further perform a Bonferroni correction for the two investigated gene loci.
As there may be a skeletal site-specific genetic predisposition to osteoporosis [e.g. at the lumbar spine or femoral neck (FN)], we also used subsamples of our patients for further analysis. When restricting the patient sample to those with a T-score less than −2.5 at the lumbar spine (86% of the full patient sample), we still observe highly significant associations for both SNPs within the coding exon (e.g. for SNP rs4649124, P=0.005). The small size of a patient sample defined by a T-score less than −2.5 only at the FN does not allow for a valid statistical analysis. We next analyzed BMD at the lumbar spine as a quantitative trait. To this end, we tested the normal distribution of the quantitative data using the Kolmogorov—Smirnov test and the homogeneity variances with Levene's test. Congruent with the association results in the dichotomized sample, we found a highly significant difference in bone density between individuals displaying different genotypes (e.g. for SNP rs4649124, P=0.004, one-way analysis of variance) (data not shown).
To investigate the extended locus and its genetic variability more systematically, we sequenced the coding region of CNR2 in all individuals (n=388) to obtain complete information about all sequence variants. As the CNR2 coding sequence includes only a small fraction of the genomic region where significantly associated SNPs were located, we further determined the LD structure using genotypes from the Perlegen data set (16) (because of a gap in the NCBI 34 genome assembly, the current phase 1 public HapMap does not contain the CNR2 locus) (Fig. 1C). We partitioned the locus into haplotype blocks using the program hapblock (17) (USC QCB | Quantitative and Computational Biology | University of Southern California). Block partitioning required that at least 90% of the total haplotype diversity within each block was covered by common haplotypes (>5% population frequency). We selected SNPs necessary for common haplotype discrimination as tagging SNPs, altogether we genotyped additional 20 SNPs in the 300 kb extended region (Fig. 1A and B and Tables 2 and 3). The genotype data of this comprehensive set of SNPs confirmed the positive association with most significant values within a LD block including exon 2 and the 3′-untranslated region (UTR) of CNR2. The most significant association remained to be found for rs2501431 in exon 2 (Fig. 1B and Table 3). SNPs telomeric and centromeric were less or not significant.
In order to devise the genomic region containing putative disease mutations, we tested the global haplotype frequency distribution for all possible three-marker haplotypes. The smallest P-value (P=0.0001) was found for the SNP combination rs2229583—rs2501431—rs16828926 (Fig. 1B). This haplotype spans exon 2 and a small part of intron 1 of CNR2. Its P-value is an order of magnitude less than the global P-value (P=0.005) for the SNP combination rs4649124—rs3003336—rs2501431, which is located directly under the SNP association peak within the CNR2 locus. The reason for this decreased P-value became clearer when we looked at the frequency of the major haplotype in the case and the control samples (Supplementary Material, Table S1). As indicated by the strong LD between the associated SNPs, a major risk haplotype is enriched among patients. Under the SNP association peak, the major haplotype G-T-A comprising the three SNPs rs4649124—rs3003336—rs2501431 displayed an enrichment among cases comparable with the associated individual SNPs (62.1 versus 51.1%, χ2=9.63, P=0.002). When increasing the size of the genomic window towards both sides, a stronger fragmentation of the risk haplotype was observed among the controls than among cases. This fragmentation comes in parallel with the decreased P-value and makes the respective genomic region most likely to harbor the unknown disease causing mutations. The major haplotype comprising the alleles T-A-G of the SNPs rs2229583—rs2501431—rs16828926 displays a frequency of 59.3% among patients and 46.0% among controls (χ2=13.12, P=0.0003) (Supplementary Material, Table S1). This gives an odds ratio (OR) of 1.716 and an estimated etiological fraction of 24.7% for this haplotype. Further increase in the size of the genomic region led to the fragmentation of the major haplotype also in the case sample, resulting in less significant frequency differences between the two samples. Thus, the haplotype analysis confirmed and even strengthened our positive association results with the single markers and provides further evidence for a causative role of the CNR2 gene in human postmenopausal osteoporosis.
DISCUSSION
Chromosome 1p36 first emerged as susceptibility locus for low hip BMD in a genome-wide linkage study in seven extended pedigrees (6). In non-parametric analyses, the maximum Z-score for this locus was 3.51 at marker D1S450. This marker is ∼14 Mb telomeric from CNR2 (UCSC-Genome Browser), but as loci in complex genetics are in general rather broad and the borders are not exactly defined, CNR2 has to be regarded as a good candidate gene for this locus. The same locus has been confirmed in an extended study with 42 families, with a maximum multipoint LOD score of 3.53 for FN BMD (7). Also in mice, there is very good evidence from several genome-wide linkage studies that the chromosomal region around Cnr2 is implicated in the regulation of BMD. The Cnr2 gene is located on mouse chromosome 4 at 134.3 Mb very close to the marker D4Mit134 at 134.1 Mb, for which genetic localization is at 62 cM. A major quantitative trait locus (QTL, named Bmd7) for both femoral and lumbar vertebral BMD in C57BL/6 and C3H/HeJ mice strains has been mapped close to this locus. The maximum LOD score was 16.3 for femoral and 14.8 for vertebral BMD at D4Mit124 (57.4 cM), which accounts for 6.7 and 6.5% of F2 variance in femoral and vertebral BMD, respectively (14). A similar highly significant finding has been obtained in a screen for loci affecting peak bone mass in 24 recombinant inbred mouse strains with a maximum LOD score of 12.3 for D4Mit312 at 69.8 cM (15).
In contrast, the CNR1 locus on chromosome 6q14 does not seem to be a major QTL for BMD, which is in accordance with our association results. As the genomic region covering CNR1 is small (∼5.5 kb) and shows a rather high LD, the testing of four SNPs covering most of the genetic variance is probably sufficient to exclude a major role of CNR1 in our sample.
Although the physiological role of endocannabinoids and CB1 receptors in the regulation of neuronal signaling has been well studied, relatively little is known about the functions of CB2 receptors or the involvement of cannabinoid receptors in human disease processes. Indeed, our results provide the first evidence for an involvement of CB2 receptors in a human disorder. In a recent study, we have demonstrated that CB2 activation can ameliorate the development of atherosclerotic plaques in mice (18). Very interestingly, atherosclerosis, also a common and late-onset disorder, is inversely and independently correlated with BMD as shown, e.g. by a systematic population-based study in over 5200 Norwegian individuals (19). Thus, it is tempting to speculate that CNR2 polymorphisms contribute to the etiology of both disorders, thus providing one explanation for the association of atherosclerosis with osteoporosis.
Among the large number of associated polymorphisms in CNR2 are two missense variants (the double SNP rs2502992—rs2501432 and Gln63Arg; rs2229579 and His316Tyr; both amino acid exchanges are in non-conserved regions) and several SNPs within conserved regions in the 3′-UTR. These alone or in combination with other linked polymorphisms, still to be identified, may be responsible for the observed association by altering protein function and/or gene expression. On the basis of the Cnr2 knockout phenotype, and our pharmacological results (9), we would expect a decreased CB2 function in the disease-associated allele, which would also be expected to provide a protective function in atherosclerosis. Obviously, further studies will be necessary to address the functional role of CB2 receptors in these common disorders. Eventually, these studies may open novel therapeutic strategies.
MATERIALS AND METHODS
Clinical data
All 388 women were recruited and investigated at a single clinical center (Hôpital Lariboisière, Paris, France). The patients were referred for suspicion of osteoporosis. The controls were women addressed to have a BMD measurement (Viggos cohort) to evaluate their bone status. They were not referred for suspicion of osteoporosis. They were all postmenopausal Caucasian of French origin and >50 years old. All participants signed an informed consent, and the study was approved by the hospital Internal Ethical Review Board. All women underwent a routine clinical and biochemical assessment to exclude secondary causes of osteoporosis. Subjects who had plasma hormone abnormalities [thyroid stimulating hormone and parathyroid hormone (PTH)] or had received treatment known to affect BMD for more than a month were also excluded. The patients and controls underwent BMD measurements of the lumbar spine (L2—L4) or FN, determined by dual energy X-ray absorptiometry (DEXA) (LUNAR DPX-L, Madison, WI, USA). T-scores were based on a French population reference provided by the manufacturer. X-ray of the spine was performed in all women to evaluate semi-quantitatively the prevalence of vertebral osteoporotic fractures. Women were defined as having osteoporosis on the basis of a BMD T-score less than −2.5 at lumbar spine (L2—L4) or FN. Some women of the Viggos study (∼10%) had a T-score less than −2.5 and were thus classified as having osteoporosis. Therefore, our sample consisted of 168 postmenopausal osteoporotic women and 220 ethnically, age- and sex-matched controls. In the osteoporotic women, the prevalence of vertebral fractures was 46%. The control women had BMD T-score of ≥−2.5 and had no previous FN or vertebral lumbar fractures. Further details of the patient and control clinical data are given in Table 5.
Molecular genetic studies and statistical analyses
Genomic DNA was extracted by standard procedures from EDTA-blood samples, and SNPs were genotyped by polymerase chain reaction (PCR) amplification using the ABI Assays-On-Demand/Design SNP Genotyping Products or assessed by direct sequencing (Table 2). As means of quality control, we (i) always included negative controls as well as individuals who are known to have the three different genotypes in the ABI assays, (ii) typed SNP rs4649119 in duplicate and (iii) typed SNPs rs2229586, rs4649124 and rs2501431 by two independent methods (ABI assays and direct sequencing). The discrepancies between both methods were <0.4% of tested chromosomes.
Global haplotype comparison was performed using the package cocaphase (UNPHASED). This software performs likelihood ratio tests under a log-linear model of the probability that a haplotype belongs to the case or the control group, using a standard unconditional logistic regression. The expectation-maximization (EM) algorithm was used to estimate haplotype frequencies and resolve missing genotypes. Maximum-likelihood haplotype frequencies were further estimated from individual genotypes using the program snphap (https://archimedes.well.ox.ac.uk/pise/snphap.html), which implements an EM algorithm. Haplotypes were then tested against all other haplotypes for disease association by a χ2 test (df=1). Frequency estimates of the two programs were in good agreement both for common and for rare haplotypes.
RESULTS
Around the human CNR1 gene, we genotyped four single nucleotide polymorphisms (SNPs) (rs806365, rs1049353, rs806377 and rs806380), for which genotyping assays were available (Table 1), covering a region of nearly 20 kb and encompassing the single coding exon of CNR1. The genotype distributions of the two samples did not deviate significantly (P≥0.05) from those expected from the Hardy—Weinberg equilibrium (HWE), as tested by the exact test. The distribution of genotypes of the four CNR1 polymorphisms is shown in Table 1. To compare the frequencies of individual SNPs between cases and controls, we applied a standard χ2 test (df=1) for allelic association with the disease phenotype. No statistically significant differences of allele or genotype distributions were found between patients and controls, but SNP rs806380 showed a borderline significant P-value of 0.057. In conclusion, our data do not support a major role of CNR1 in our sample.
The human CNR2 gene and its mouse ortholog are located on chromosomes 1p36 and 4QD3, respectively. These two genomic regions have been linked to osteoporosis and BMD by several independent genome-wide linkage analyses (6—8,14,15); however, the CNR2 gene has not been investigated as a positional candidate gene up to now. CB2 is encoded by a single exon (exon 2) with a non-coding exon 1 ∼37 kb upstream. We initially tested six SNPs (Table 2) for which genotyping assays were available in a region spanning ∼120 kb (Fig. 1A) and harboring both CNR2 exons. Again, there was no deviation of genotype distributions from the HWE, but we found a significant difference of allele frequencies between cases and controls for several SNPs (Table 3 and Fig. 1B). In contrast, seven unrelated control SNPs (intergenic and not in the vicinity of the CNR genes) from different chromosomes did not show an association with the phenotype (Table 4). Therefore, these results suggest an involvement of the CNR2 locus in human osteoporosis.
Both SNPs located within exon 2 of CNR2, i.e. rs4649124 and rs2501431, showed most significant P-values for allele (0.0026 and 0.0014, respectively) and genotype (0.0023 and 0.00073, respectively) distributions (Table 3 and Fig. 1B). These two SNPs are in tight linkage disequilibrium (LD) (D′=0.99 and r2=0.96), which is also true for most of the other SNPs within exon 2 (Fig. 1C). For rs4649124, the allele frequencies differed by 11% (allele 1: controls 52% and patients 63%). The other four SNPs, located 10 kb/30 kb 3′ and 80 kb/93 kb 5′ of CNR2, showed P-values for allelic association between 0.0042 (rs4649119, which is in strong LD to SNPs in exon 2; D′=0.98) and 0.2275 (hCV395615, which just shows a weak LD to SNPs in exon 2; D′=0.55) (Table 3 and Fig. 1B). In order to correct for multiple testing, we repeated the analysis 100 000 times with a randomly permuted assignment of the affection status. The obtained P-value (P=0.0056) of the most strongly associated SNP (rs2501431) remains significant, if we further perform a Bonferroni correction for the two investigated gene loci.
As there may be a skeletal site-specific genetic predisposition to osteoporosis [e.g. at the lumbar spine or femoral neck (FN)], we also used subsamples of our patients for further analysis. When restricting the patient sample to those with a T-score less than −2.5 at the lumbar spine (86% of the full patient sample), we still observe highly significant associations for both SNPs within the coding exon (e.g. for SNP rs4649124, P=0.005). The small size of a patient sample defined by a T-score less than −2.5 only at the FN does not allow for a valid statistical analysis. We next analyzed BMD at the lumbar spine as a quantitative trait. To this end, we tested the normal distribution of the quantitative data using the Kolmogorov—Smirnov test and the homogeneity variances with Levene's test. Congruent with the association results in the dichotomized sample, we found a highly significant difference in bone density between individuals displaying different genotypes (e.g. for SNP rs4649124, P=0.004, one-way analysis of variance) (data not shown).
To investigate the extended locus and its genetic variability more systematically, we sequenced the coding region of CNR2 in all individuals (n=388) to obtain complete information about all sequence variants. As the CNR2 coding sequence includes only a small fraction of the genomic region where significantly associated SNPs were located, we further determined the LD structure using genotypes from the Perlegen data set (16) (because of a gap in the NCBI 34 genome assembly, the current phase 1 public HapMap does not contain the CNR2 locus) (Fig. 1C). We partitioned the locus into haplotype blocks using the program hapblock (17) (USC QCB | Quantitative and Computational Biology | University of Southern California). Block partitioning required that at least 90% of the total haplotype diversity within each block was covered by common haplotypes (>5% population frequency). We selected SNPs necessary for common haplotype discrimination as tagging SNPs, altogether we genotyped additional 20 SNPs in the 300 kb extended region (Fig. 1A and B and Tables 2 and 3). The genotype data of this comprehensive set of SNPs confirmed the positive association with most significant values within a LD block including exon 2 and the 3′-untranslated region (UTR) of CNR2. The most significant association remained to be found for rs2501431 in exon 2 (Fig. 1B and Table 3). SNPs telomeric and centromeric were less or not significant.
In order to devise the genomic region containing putative disease mutations, we tested the global haplotype frequency distribution for all possible three-marker haplotypes. The smallest P-value (P=0.0001) was found for the SNP combination rs2229583—rs2501431—rs16828926 (Fig. 1B). This haplotype spans exon 2 and a small part of intron 1 of CNR2. Its P-value is an order of magnitude less than the global P-value (P=0.005) for the SNP combination rs4649124—rs3003336—rs2501431, which is located directly under the SNP association peak within the CNR2 locus. The reason for this decreased P-value became clearer when we looked at the frequency of the major haplotype in the case and the control samples (Supplementary Material, Table S1). As indicated by the strong LD between the associated SNPs, a major risk haplotype is enriched among patients. Under the SNP association peak, the major haplotype G-T-A comprising the three SNPs rs4649124—rs3003336—rs2501431 displayed an enrichment among cases comparable with the associated individual SNPs (62.1 versus 51.1%, χ2=9.63, P=0.002). When increasing the size of the genomic window towards both sides, a stronger fragmentation of the risk haplotype was observed among the controls than among cases. This fragmentation comes in parallel with the decreased P-value and makes the respective genomic region most likely to harbor the unknown disease causing mutations. The major haplotype comprising the alleles T-A-G of the SNPs rs2229583—rs2501431—rs16828926 displays a frequency of 59.3% among patients and 46.0% among controls (χ2=13.12, P=0.0003) (Supplementary Material, Table S1). This gives an odds ratio (OR) of 1.716 and an estimated etiological fraction of 24.7% for this haplotype. Further increase in the size of the genomic region led to the fragmentation of the major haplotype also in the case sample, resulting in less significant frequency differences between the two samples. Thus, the haplotype analysis confirmed and even strengthened our positive association results with the single markers and provides further evidence for a causative role of the CNR2 gene in human postmenopausal osteoporosis.
DISCUSSION
Chromosome 1p36 first emerged as susceptibility locus for low hip BMD in a genome-wide linkage study in seven extended pedigrees (6). In non-parametric analyses, the maximum Z-score for this locus was 3.51 at marker D1S450. This marker is ∼14 Mb telomeric from CNR2 (UCSC-Genome Browser), but as loci in complex genetics are in general rather broad and the borders are not exactly defined, CNR2 has to be regarded as a good candidate gene for this locus. The same locus has been confirmed in an extended study with 42 families, with a maximum multipoint LOD score of 3.53 for FN BMD (7). Also in mice, there is very good evidence from several genome-wide linkage studies that the chromosomal region around Cnr2 is implicated in the regulation of BMD. The Cnr2 gene is located on mouse chromosome 4 at 134.3 Mb very close to the marker D4Mit134 at 134.1 Mb, for which genetic localization is at 62 cM. A major quantitative trait locus (QTL, named Bmd7) for both femoral and lumbar vertebral BMD in C57BL/6 and C3H/HeJ mice strains has been mapped close to this locus. The maximum LOD score was 16.3 for femoral and 14.8 for vertebral BMD at D4Mit124 (57.4 cM), which accounts for 6.7 and 6.5% of F2 variance in femoral and vertebral BMD, respectively (14). A similar highly significant finding has been obtained in a screen for loci affecting peak bone mass in 24 recombinant inbred mouse strains with a maximum LOD score of 12.3 for D4Mit312 at 69.8 cM (15).
In contrast, the CNR1 locus on chromosome 6q14 does not seem to be a major QTL for BMD, which is in accordance with our association results. As the genomic region covering CNR1 is small (∼5.5 kb) and shows a rather high LD, the testing of four SNPs covering most of the genetic variance is probably sufficient to exclude a major role of CNR1 in our sample.
Although the physiological role of endocannabinoids and CB1 receptors in the regulation of neuronal signaling has been well studied, relatively little is known about the functions of CB2 receptors or the involvement of cannabinoid receptors in human disease processes. Indeed, our results provide the first evidence for an involvement of CB2 receptors in a human disorder. In a recent study, we have demonstrated that CB2 activation can ameliorate the development of atherosclerotic plaques in mice (18). Very interestingly, atherosclerosis, also a common and late-onset disorder, is inversely and independently correlated with BMD as shown, e.g. by a systematic population-based study in over 5200 Norwegian individuals (19). Thus, it is tempting to speculate that CNR2 polymorphisms contribute to the etiology of both disorders, thus providing one explanation for the association of atherosclerosis with osteoporosis.
Among the large number of associated polymorphisms in CNR2 are two missense variants (the double SNP rs2502992—rs2501432 and Gln63Arg; rs2229579 and His316Tyr; both amino acid exchanges are in non-conserved regions) and several SNPs within conserved regions in the 3′-UTR. These alone or in combination with other linked polymorphisms, still to be identified, may be responsible for the observed association by altering protein function and/or gene expression. On the basis of the Cnr2 knockout phenotype, and our pharmacological results (9), we would expect a decreased CB2 function in the disease-associated allele, which would also be expected to provide a protective function in atherosclerosis. Obviously, further studies will be necessary to address the functional role of CB2 receptors in these common disorders. Eventually, these studies may open novel therapeutic strategies.
MATERIALS AND METHODS
Clinical data
All 388 women were recruited and investigated at a single clinical center (Hôpital Lariboisière, Paris, France). The patients were referred for suspicion of osteoporosis. The controls were women addressed to have a BMD measurement (Viggos cohort) to evaluate their bone status. They were not referred for suspicion of osteoporosis. They were all postmenopausal Caucasian of French origin and >50 years old. All participants signed an informed consent, and the study was approved by the hospital Internal Ethical Review Board. All women underwent a routine clinical and biochemical assessment to exclude secondary causes of osteoporosis. Subjects who had plasma hormone abnormalities [thyroid stimulating hormone and parathyroid hormone (PTH)] or had received treatment known to affect BMD for more than a month were also excluded. The patients and controls underwent BMD measurements of the lumbar spine (L2—L4) or FN, determined by dual energy X-ray absorptiometry (DEXA) (LUNAR DPX-L, Madison, WI, USA). T-scores were based on a French population reference provided by the manufacturer. X-ray of the spine was performed in all women to evaluate semi-quantitatively the prevalence of vertebral osteoporotic fractures. Women were defined as having osteoporosis on the basis of a BMD T-score less than −2.5 at lumbar spine (L2—L4) or FN. Some women of the Viggos study (∼10%) had a T-score less than −2.5 and were thus classified as having osteoporosis. Therefore, our sample consisted of 168 postmenopausal osteoporotic women and 220 ethnically, age- and sex-matched controls. In the osteoporotic women, the prevalence of vertebral fractures was 46%. The control women had BMD T-score of ≥−2.5 and had no previous FN or vertebral lumbar fractures. Further details of the patient and control clinical data are given in Table 5.
Molecular genetic studies and statistical analyses
Genomic DNA was extracted by standard procedures from EDTA-blood samples, and SNPs were genotyped by polymerase chain reaction (PCR) amplification using the ABI Assays-On-Demand/Design SNP Genotyping Products or assessed by direct sequencing (Table 2). As means of quality control, we (i) always included negative controls as well as individuals who are known to have the three different genotypes in the ABI assays, (ii) typed SNP rs4649119 in duplicate and (iii) typed SNPs rs2229586, rs4649124 and rs2501431 by two independent methods (ABI assays and direct sequencing). The discrepancies between both methods were <0.4% of tested chromosomes.
Global haplotype comparison was performed using the package cocaphase (UNPHASED). This software performs likelihood ratio tests under a log-linear model of the probability that a haplotype belongs to the case or the control group, using a standard unconditional logistic regression. The expectation-maximization (EM) algorithm was used to estimate haplotype frequencies and resolve missing genotypes. Maximum-likelihood haplotype frequencies were further estimated from individual genotypes using the program snphap (https://archimedes.well.ox.ac.uk/pise/snphap.html), which implements an EM algorithm. Haplotypes were then tested against all other haplotypes for disease association by a χ2 test (df=1). Frequency estimates of the two programs were in good agreement both for common and for rare haplotypes.
Source, Graphs and Figures: Cannabinoid receptor type 2 gene is associated with human osteoporosis
