Presented in part at the meeting of the European Calcified Tissue Society, Maastricht, The Netherlands, May 7–12, 1999.
Previous genetic linkage data suggested that a gene on chromosome 1p36.2–36.3 might be linked to low bone mineral density (BMD). Here, we examine the gene for tumor necrosis factor receptor 2 (TNFR2), a candidate gene within that interval, for association with low BMD in a group of 159 unrelated individuals. We assess two polymorphic sites within the gene, a microsatellite repeat within intron 4, and a three-nucleotide variation in the 3′ untranslated region (UTR) of the gene. The latter has five alleles of which the rarest allele is associated with low spinal BMD Z score (p = 0.008). Lowest mean spinal BMD Z scores were observed for individuals having genotypes that were heterozygous for the rarest allele. No homozygotes for the rarest allele were observed. Preliminary analysis suggests that there is a difference in the genotype frequency distribution between the group with low BMD and a control group.
SEVERAL GENES have been tested for potential association with aspects of bone health and calcium metabolism. In general, studies have identified a DNA polymorphism within the gene of interest and then tested for differences in the allele frequency distribution in study populations and for statistical association of the particular polymorphism with a bone-related parameter. A polymorphism within the first intron of the COL1A1 gene that alters a binding site for the transcription factor Sp1 has been associated with low bone density in adults and in prepubertal girls and an increased risk of hip fracture.(1–5) Several restriction fragment length polymorphisms within the vitamin D receptor (VDR) gene have shown a statistically significant association between a particular allele and some measure of bone quality.(6–9) One of the polymorphisms (FokI) is located at the translation initiation site of the VDR gene(8–10) where it causes two messenger RNA (mRNA) transcripts that differ by the presence or absence of the first initiation codon. Other genes that have been investigated to a lesser extent are parathyroid hormone,(11) transforming growth factor β1,(12) calcitonin receptor,(13–14) estrogen receptor,(15–17) insulin-like growth factor I,(18–19) interleukin-1R antagonist,(20) and interleukin-6.(21)
The data presented in this report support the addition of another gene to this growing list, the tumor necrosis factor receptor 2 (TNFR2) gene. We were led to investigate this gene by the results of a genome-wide screen for linkage with low bone mineral density (BMD) in seven families(22) and additional sib pairs.(23) The linkage results suggested that a gene or genes on chromosome 1p36.2–36.3 were linked to the low BMD trait in these two groups. The genetic interval delimited by linkage analysis was 28 cM in length, and a maximum lod score of 2.7 was obtained for femoral neck BMD with marker D1S450. One of the candidate genes within this genetic interval is the TNFR2 gene.
TNFR2 is a transmembrane receptor in cells of lymphatic origin that mediates response to the cytokines TNF-α and TNF-β in cell proliferation, cell killing, and apoptosis.(24) TNFR1 and TNFR2 are members of a superfamily of transmembrane receptors that now includes the RANK receptor for osteoclast differentiation factor.(25) The role of TNFR2 in bone metabolism is not known, but its ligands TNF-α and TNF-β are both expressed in healthy human bone biopsy specimens,(26) and TNF-α is a potent inducer of osteoclastogenesis.(27) TNF-α may mediate its activity in osteoblasts through activation of the NF-κB signal transduction pathway, by binding of the TNFR2 intracellular domain to other factors that in turn activate NF-κB.(28, 29) In a recent report,(30) murine cells of the macrophage lineage were induced to become tartrate-resistant acid phosphatase (TRAP) positive (a marker of the osteoclast phenotype) by treatment with macrophage colony-stimulating factor (M-CSF) and TNF-α. The induction was inhibited by addition to the culture of antibodies directed against the receptor for M-CSF, TNFR1, or TNFR2. This suggests a role for TNFR2 in osteoclastogenesis.
The TNFR2 gene is located on chromosome 1p36.2,(31–32) and has been well characterized because this region of chromosome 1 frequently undergoes loss of heterozygosity in tumors,(33)particularly neuroblastoma.(34–35) The TNFR2 gene spans 26 kilobases (kb) with 10 exons. Within intron 4 is a microsatellite repeat of (CA)16, (31) and in the 2200 base pair (bp) long 3′ untranslated region (UTR) there is a polymorphism detectable by single strand conformation polymorphism (SSCP) techniques.(32, 33, 36) We have characterized the second polymorphism more completely than was previously done and have analyzed the frequency distribution of alleles for both polymorphisms in three groups of Canadians. In addition, we have tested for the association of the polymorphisms with low BMD at the lumbar spine, femoral neck, trochanter, and Ward's triangle.
MATERIALS AND METHODS
The first group of test subjects was composed of 66 individuals from our previous studies(22–23, 37) who were unrelated to each other. They were either the founders of or had married into families previously selected for their low BMD. The mean BMD Z scores of this subgroup were not different from the group as a whole, indicating no selection bias. The second test group of 93 individuals also were unrelated to each other and to the first group and had been referred to the McGill Bone Center by their physicians for evaluation of suspected low bone density.(38) The third group was a random general clinic population (n = 141) that served as a control for allele frequency distribution in these studies. Because this group was from a general clinic, no information on bone density was available, and we assumed that bone status was more normal than that of the study group.
Blood was collected from participants and DNA was isolated by standard methods. Bone densitometry was performed on individuals in the first two groups as previously described.(22, 38) Informed consent was obtained and anonymity was maintained. The study was approved by the Institution Review Board (IRB) of each institution involved.
Microsatellite repeat in intron 4
The polymerase chain reaction (PCR) primers for detection of the (CA)16 microsatellite repeat previously described(31) were TNFR2-CAF, 5′-GTGATCTGCAAGATGAACTCAC-3′, and TNFR2-CAR, 5′-ACACCACGTCTGATGTTTCA-3′. PCR conditions were denaturation at 94.5°C for 3 minutes; followed by five cycles of 94.5°C for 30 s, 65.0°C for 30 s, 72°C for 30 s; followed by 23 cycles of 94.5°C for 30 s, 55.0°C for 30 s, 72.0°C for 30 s; followed by 72.0°C for 5 minutes in a Perkin-Elmer Thermocycler 9600. The primer TNFR2-CAF was labeled with γ-32P-adenosine triphosphate (ATP; NEN Life Science Products, Boston, MA, U.S.A.) using T4 polynucleotide kinase (Promega, Madison, WI, U.S.A.). Approximately 100–300 ng of DNA template was used with reaction components under conditions recommended by the manufacturer (Applied Biosystems, Inc., Foster City, CA, U.S.A.).
Polymorphism in 3′-UTR
PCR primers for detection of the polymorphism within the 3′-UTR of TNFR2(32) were TNFR2–7, 5′-AGGACTCTGAGGCTCTTTCT-3′, and TNFR2–8, 5′-TCACAGAGAGTCAGGGACTT-3′. PCR conditions were denaturation at 94.5°C for 3 minutes; followed by 28 cycles of 94.5°C for 30 s, 62.0°C for 30 s, 72.0°C for 30 s; followed by 72.0°C for 5 minutes. Reagents and instrumentation were as above.
The SSCP analysis performed by others(32, 33, 36) detected four distinct products of this PCR. We utilized conformation sensitive gel electrophoresis (CSGE) in order to distinguish five differentially migrating fragments of this PCR product. Gel conditions and product suppliers were as described(39) with the following modifications. Tris-taurine-EDTA running buffer was made as a 20× stock with 1.78 M Tris, 0.58 M taurine, 4.0 mM EDTA, pH 9.0, and used at 0.5×. Formamide was deionized in batches and frozen in aliquots sufficient for a single gel. Sample loading buffer was prepared each time with four parts deionized formamide and one part of loading buffer (5× = 0.1% xylene cyanol, 0.1% bromphenol blue, 10 mM EDTA, and 95% formamide). Standard sequencing gel plates and apparatus were used with 0.4-mm spacers. Samples were mixed with an equal volume of sample loading buffer, denatured for 5 minutes at 95°C, transferred to ice, and 1.5 μl immediately was loaded. Gels were run at 350 V at room temperature for 16–18 h and then dried and exposed to film. Resolution of five alleles was possible only when these conditions were adhered to.
Selected PCR fragments were sequenced according to manufacturer's recommendations using ABI 377 instrumentation and Big-Dye chemistry (Applied Biosystems) in order to determine the basis for the five alleles.
The two test groups were compared by analysis of variance (ANOVA) for age, body mass index and BMD Z scores at the spine, femoral neck, trochanter, and Ward's triangle and then combined for all additional analyses. The frequency distribution of alleles for the two polymorphisms within the TNFR2 gene was analyzed by χ2. Association of bone density at the spine (L2-L4), femoral neck, trochanter, and Ward's triangle with genotype for each of the two polymorphisms was evaluated by ANOVA. We used the BMD Z scores at each of the anatomical sites as the dependent variables because they were already corrected for age, gender, and weight. Statistical significance was achieved at p < 0.025, using the Bonferroni correction for multiple testing. Additional tests were performed to help explain the observations: sampling without replacement and the Tukey-Kramer post hoc test. Power analysis was performed to determine the lowest sample number required to observe a significant (p = 0.025) effect of genotype on BMD. The statistical software package employed was JMP (version 3.1; SAS Institute, Inc., Cary, NC, U.S.A.)
We have assayed two polymorphisms within the TNFR2 gene in three groups of individuals. Test group 1 was comprised of individuals from families used for linkage analysis of the low BMD trait(22, 23, 37) who were unrelated to each other. Test group 2 was made up of unrelated individuals from an osteoporosis clinic that had been tested previously for VDR genotype.(38) The third group was composed of unrelated patients from a general clinic population with no testing for or knowledge of their bone status. This third group was used as reference for allele frequency comparisons. The composition and mean bone density Z scores for the first two test groups are compared in Table 1. Only age was significantly different among the groups (p = 0.011), because of the younger mean age for the males in the second group. However, ANOVA of age as the dependent variable with genotype for the two polymorphisms as class variables indicated that age was unaffected by genotype. Therefore, the results of genotype testing in the two groups were pooled for subsequent analyses.
Table Table 1. Results of ANOVA Among the Males and Females Comprising the Two Test Groups for Age, Body Mass Index, and Bone Density8
The microsatellite repeat within intron 4 of the TNFR2 gene was reported to have five alleles(31) ranging in size from 267 to 277 bp. We observed these five plus a sixth allele (275 bp) in both the test and the control groups with a statistically significant difference in the frequency distribution between the two groups (p < 0.025; Table 2). A total of 18 different genotypes would be expected from six alleles assuming Hardy-Weinberg equilibrium. The observed frequency distribution of genotypes of the intron 4 microsatellite repeat for both the test and the control groups did not differ significantly from expected (test group χ2 = 20.54 and degree of freedom (df) = 19, NS; control group χ2 = 10.08 and df = 19, NS). Comparison of the genotype distributions between the test and control groups also was not significant (χ2 = 22.46 and df = 19, NS)
Table Table 2. Allele Frequency Distribution for the Two TNFR2 Polymorphisms in Test Group (n = 159) Compared With Control Group (n = 141)8
The polymorphism within the 3′-UTR of the TNFR2 gene was previously characterized as having four alleles that were identified by different banding patterns under electrophoretic conditions that detect single strand conformation differences.(32, 33, 36) We employed a different conformation sensitive gel technique(39) that resolved five alleles, each as a single band (Fig. 1). The allele frequency distribution could not be compared with the published data(36) because the correspondence of bands between the two gel systems was not tested. However, the allele frequency distribution was compared between the test and the control groups in this study, with no statistically significant differences (Table 2). The frequency distribution of the 15 genotypes was in Hardy-Weinberg equilibrium in both groups (test group χ2 = 8.44 and df = 14, NS; control group χ2 = 22.57 and df = 14, NS). However, comparison of the genotype distributions for the 3′-UTR polymorphism between the test and the control groups was statistically significant (χ2 = 48.70, df = 14, and p < 0.001).
The nature of the UTR polymorphism was established by sequencing of PCR products from individuals who had been scored as homozygous for alleles 2, 3, 4, and 5 or heterozygous for combinations of allele 1 and the other four alleles (Fig. 2). No 1/1 homozygotes were observed. The polymorphism is actually a combination of three polymorphic sites within a span of 27 nucleotides, affecting nucleotides 593, 598, and 620 of exon 10 of the TNFR2 gene (Genbank Accession U52165). The five alleles are defined as summarized in Table 3. There are eight possible alleles from these three polymorphic sites, but we have observed only five.
Table Table 3. Alleles of the TNFR2 UTR Polymorphism as Defined by Nucleotide Composition at Three Variable Sites. Numbering is Based on Genbank Accession U521658
Analysis of variance was performed on BMD Z scores at the spine, femoral neck, trochanter, and Ward's triangle with respect to genotype at each polymorphic site (Table 4). No statistically significant relationship was observed (i.e., p < 0.025). However, ordering the UTR genotypes with respect to mean spinal BMD Z scores suggested that allele 1 might be associated with low BMD (Table 5). The probability of allele 1 occurring in the four groups with lowest mean spinal BMD was determined using a test of sampling without replacement (p < 0.001). A more conservative analysis of the genotype data using Tukey-Kramer post hoc pairwise comparisons indicated that the mean Z score for spinal BMD of individuals having genotype 1/2 for the UTR polymorphism was significantly different than those with genotype 2/2. The mean femoral neck BMD Z scores are shown for comparison and indicate a similar trend in that the two lowest mean Z scores occur in groups having allele 1.
Table Table 5. Genotypes for the TNFR2 UTR Polymorphism with Frequency and Mean Spinal or Femoral Neck BMD Z Scores for the Study Group8
Table Table 4. Results of ANOVA of BMD Z Scores with Genotype for Each of the Two TNFR2 Polymorphisms as the Grouping Variables in the Study Group (n = 159)8
The results of sampling without replacement and the Tukey-Kramer tests suggested that allele 1 was significantly associated with low spinal BMD. Comparison of mean spinal BMD Z scores for presence/absence of allele 1 revealed a statistically significant relationship between spinal BMD and TNFR2 UTR genotype (p = 0.008). There was no difference between the two genotype groups for BMD at the other three anatomical sites (Table 6).
Table Table 6. Mean BMD Z Scores (±SE) for TNFR2 3′-UTR Genotype with the Grouping Variable as Presence or Absence of Allele 18
The frequency of allele 1 in the three major ethnic backgrounds represented in our test groups, western European (French and British), southern European (Greek and Italian), and eastern European (Ashkenazi Jewish), was not significantly different. However, a statistically significant trend was observed (p = 0.035, χ2 = 6.685, and df = 2) with allele 1 being least frequent in the French and British component of our test group and most frequent in the Ashkenazi Jewish group.
The power of our study to detect a significant difference of p = 0.025 (σ = 1.416; δ = 0.476; the values observed with the current data set) between spinal BMD and TNFR2 UTR genotype was 0.667. The least significant number (LSN) required to achieve p = 0.025 was 225 individuals. Similar analysis for the femoral neck data indicated a power of 0.697 and LSN = 208. The other two hip sites gave similar results.
Bone quality is a complex trait that has significant genetic and environmental input. Estimates of the genetic component range from 40 to 80% of the total variation in bone density, with environmental factors comprising the remainder. Identifying the constellation of genes that contribute to the genetic component of bone quality, quantitating that contribution, and estimating risk of fracture based on knowledge of variation in these genes is the ultimate goal of research in the field of osteoporosis genetics. Considering the entire genome, there are tens, perhaps even hundreds of genes that can affect bone quality, whether it be the architecture, the density, the metabolism, or the structural properties of the tissue.
One approach to identifying genes important to bone quality has been to concentrate on bone density as the most easily measured and most predictive aspect of bone integrity and to use bone density as a quantitative trait in genetic linkage analysis. Linkage analysis is a powerful, robust tool in the elimination of portions of the genome and preliminary identification of other regions of the genome that might be involved in the trait of interest, that is, bone density. For example, Johnson et al.(40) have identified a locus on chromosome 11q21 that is linked to high bone density in one family. Results of the genome screen we reported(22, 23) suggested that chromosome 1p36 might contain a gene or genes that were related to low bone density in our test population. Because the test population was not large enough, nor were any of the families of sufficient size to allow a pursuit of the candidate region of chromosome 1p36 by positional cloning, we opted for the candidate gene approach to identification of the important genes within the interval.
In this study we focused on one of the candidate genes, TNFR2, and analyzed its potential importance to bone density from an association rather than a linkage viewpoint. Defining association requires that there be polymorphisms within the gene of interest, and that the polymorphism be in linkage disequilibrium with the variation in the gene that leads to its variability in function. Two good examples of this are the genes COL1A1 and VDR both of which have sequence variants that may be functionally important.(1–10, 41)
We have analyzed two polymorphisms here, a (CA)16 repeat within intron 4 and a three-nucleotide variation within the 3′-UTR of the TNFR2 gene, in three groups of unrelated individuals. The first group was composed of unrelated individuals of seven large families(22, 37) or 39 small families(23) on whom linkage analysis had been performed. The second group(38) was composed of unrelated individuals who had been referred to the McGill University Bone Centre by their primary care physicians. These two groups were combined because there were no statistically significant differences between them except for age, which was not related to genotype. A third group, which served as a control, was chosen at random from a general clinic population from the same geographical area as the two previous groups, and this group provided a reference population for allele and genotype frequency data comparisons.
The allele frequency distribution for the microsatellite repeat polymorphism was different between the study and test groups (p = 0.025), but the allele frequency distribution of the 3′-UTR polymorphism was not different. However, the genotype frequency distribution between the two groups was significantly different for the 3′-UTR (p = 0.001). We interpret this with caution because although the control group was selected from the same geographical area as the test group, no information on the ethnic composition of the control group was available. Therefore, we cannot rule out population admixture as the cause of the differences in allele and genotype frequency distributions.
Analysis of variance of genotype at either polymorphic site of TNFR2 with BMD at the spine, the femoral neck, the trochanter, or Ward's triangle was not significant (Table 4). However, ordering of the genotypes with respect to mean spinal BMD Z scores suggested that presence of allele 1 of the 3′-UTR polymorphism might be associated with low BMD. This was the case when the data were analyzed as presence versus absence of allele 1; the spinal BMD Z scores were significantly lower in the group having allele 1 (p = 0.008).
The 3′-UTR polymorphism is an unusual variant that involves three nucleotide sites within a span of 27 bp. There are eight possible allelic combinations of the two nucleotides at each site, but we have observed only five in the nearly 300 individuals we have genotyped in this study. Allele 1 is the rarest allele in the test and control groups, and it is associated with low spinal BMD in the test group. It is the only allele among the five that has a guanine at position 598. The 3′-UTR of the TNFR2 gene is 2200 bases long, and position 598 is located 193 bases after the stop codon. It has been suggested by others that sequence elements in the 3′-UTR of several cytokine genes, including IL-6(21) may affect mRNA secondary structure with consequences to mRNA stability, and thus regulate gene expression posttranscriptionally.(42, 43)
There are two additional polymorphic elements in the TNFR2 3′-UTR (31): a trinucleotide repeat and an Alu sequence. We have not assessed our study subjects for these variants, nor do we know of any other single nucleotide polymorphisms (SNPs) within the gene. However, further study of all of the types of variants within TNFR2 will help to better define the character of the allele responsible for low bone density.
Although we were led to investigate the TNFR2 gene by the observation of suggested linkage to femoral neck BMD, we have not observed statistically significant association of the TNFR2 gene polymorphisms with the femoral neck. Conversely, results of linkage analysis of this chromosomal interval to the spinal BMD trait gave a lod score of +0.26,(23) and yet we have observed a statistically significant association of TNFR2 genotype with spinal BMD. The following are several possible explanations that might resolve this apparent discrepancy:
(1) There are two different loci within this interval of chromosome 1, one that is associated with spinal BMD and the other which is linked to femoral neck BMD. Other loci in the interval that could be critical are several other members of the TNFR superfamily(44) and PLOD(45) the gene for lysyl hydroxylase, a collagen-modifying enzyme.
(2) There is a single locus within the interval, but it is not the TNFR2 locus.
(3) The power to detect association in this study was only 70% and therefore may have been underpowered to detect association at the femoral neck.
(4) A single locus, perhaps the TNFR2 locus, is linked to and associated with a third trait to which spinal and hip BMD are differentially related.
(5) There is a single locus, perhaps TNFR2, that is both linked to and associated with BMD, but the effects of variants at this locus are of differing extents in various individuals of the study group. One way to resolve the issue is to perform tests of linkage and association, such as the transmission disequilibrium test, in affected sib pairs not only for the TNFR2 locus, but for other loci within the chromosomal interval.
The authors thank Dr. Arupa Ganguly (University of Pennsylvania) for helpful discussions and technical assistance with sequencing and Dr. Michael O'Connor (Drexel University) for helpful discussions regarding statistical analysis. This work was supported by NIH RO1 grant HD36606–02.