点击显示 收起
Lilly Research Labs, Indianapolis, Indiana (L.M.H., N.H.K., T.W., P.C., S.H., F.L., D.L.H., R.R.M., E.M.A., M.S., Y.L.M., C.A.F., E.R.D., H.U.B., J.E.O.)
MBA Program/Innovation Realization Lab, Krannert School of Management, Purdue University, West Lafayette, Indiana (R.L.)
Abstract
The pharmacological preservation of bone in the ovariectomized rat by estrogen, selective estrogen receptor modulators (SERMs), and bisphosphonates has been well described. However, comprehensive molecular analysis of the effects of these pharmacologically diverse antiresorptive agents on gene expression in bone has not been performed. This study used DNA microarrays to analyze RNA from the proximal femur metaphysis of sham and ovariectomized vehicle-treated rats, and ovariectomized rats treated for 35 days with maximally efficacious doses of 17- ethinyl estradiol, the benzothiophene SERM, raloxifene, the benzopyran SERM, (S)-3-(4-hydroxyphenyl)-4-methyl-2-[4-[2-(1-piperidinyl)ethoxy]phenyl]-2H-1-benzopyran-7-ol (EM652), and the aminobisphosphonate, alendronate. Ovariectomy resulted in 644 significant probe set changes relative to sham control rats (p < 0.05), whereas E2, raloxifene, EM652, and alendronate regulated 613, 765, 652, and 737 probe sets, respectively, relative to ovariectomized control rats. An intersection of these data sets yielded 334 unique genes that were altered after ovariectomy and additionally changed by one or more antiresorptive treatment. Clustering analysis showed that the transcript profile was distinctly different for each pharmaceutical agent and that raloxifene maintained more genes at sham levels than any other treatment. In addition, E2 and alendronate suppressed a cluster of genes associated with bone formation activity below that of sham, whereas raloxifene had little effect on these genes. These data indicate stronger suppressive effects of E2 and alendronate on bone formation activity and that ovariectomy plus raloxifene resembles sham more closely than ovariectomized animals treated with E2, EM652, or alendronate.
Ovariectomy of mature rats has been shown to induce cancellous bone loss from axial and appendicular sites, as observed with postmenopausal women (Turner et al., 1994; Kimmel, 1996), and this model has been used widely to study the prevention of ovariectomy-induced bone loss (Kimmel, 1996; Sato et al., 1999). Efficacy of antiresorptive agents in ovariectomized rats has been predictive of skeletal benefit in postmenopausal women for estrogens (Turner et al., 1994; Cauley et al., 2003; Anderson et al., 2004), selective estrogen receptor modulators (SERMs), such as raloxifene and EM652 (Black et al., 1994; Sato et al., 1995; Delmas et al., 1997; Ettinger et al., 1999; Martel et al., 2000), and bisphosphonates, such as alendronate (Seedor et al., 1991; Toolan et al., 1992; Liberman et al., 1995). Estrogen, raloxifene, and alendronate have all been shown to inhibit bone resorption but they have very different mechanisms of action (Sato et al., 1999; Riggs and Parfitt, 2005).
The loss of ovarian function dramatically reduces circulating levels of estrogen and results in an increased rate of bone resorption in animals and humans. Estrogen or SERM treatment reduces skeletal turnover by maintaining the important estrogen receptor-signaling component in bone (Sato et al., 1999). SERMs, however, achieve this benefit in a tissue-specific manner and behave as estrogen agonists in bone while exerting antagonistic effects in various other estrogen target tissues, thereby avoiding the adverse effects associated with estrogen (Riggs and Hartmann, 2003). The bisphosphonates achieve a dramatic reduction in bone turnover rates by physically complexing with the bone mineral and thus inhibiting the ability of osteoclasts from resorbing the bone away (Rodan and Fleisch, 1996).
Although the histological and biomechanical changes associated with these antiresorptive therapies have been described in the ovariectomized rat, a comparison of the detailed molecular changes for each drug has not been reported. Understanding the common molecular signature that results from drug treatment could provide clues as to the most essential gene changes associated with the preservation of bone integrity after the loss of ovarian function, regardless of the mechanism by which the preservation of bone was achieved. On the other hand, observing the differences in gene expression resulting from the different treatments could yield information about the mechanisms of bone preservation resulting from estrogen receptor signaling versus that of bisphosphonate treatment.
In this report, we describe the expression profile of genes that are changed after ovariectomy and are regulated by estrogen, by two SERMs from different structural classes (the benzothiophene SERM raloxifene and the benzopyran SERM EM652), and/or by the aminobisphosphonate alendronate in the ovariectomized rat proximal femur. We sought to ascertain which genes are associated with ovariectomy 40 days after surgery when bone formation and resorption activity are known to be elevated. We also evaluated the ability of various antiresorptive agents to maintain genes that had been altered by ovariectomy near to sham levels and to determine the genes that were commonly regulated by all treatments. Finally, we sought to identify the similarities and differences among the agents that signal through the estrogen receptor (two SERMs and estrogen) in regulating ovariectomy-induced gene changes.
Materials and Methods
Animal Study Design. Six-month-old Sprague-Dawley rats (Harlan, Indianapolis, IN) were group-housed and maintained on a 12-h light/dark cycle at 22°C with access to food ad libitum (TD 89222 with 0.5% Ca and 0.4% P; Teklad, Madison, WI) and water. Rats were randomized into six groups (n = 5 rats/group): 1) sham-operated animals treated with vehicle [20% hydroxypropyl--cyclodextrin (CDX)] (Sigma, St. Louis, MO) 2) ovariectomy group treated with CDX (Ovx); 3) ovariectomized animals administered estrogen [0.1 mg/kg/day 17--ethinyl estradiol (E2; Sigma)] in CDX; 4) ovariectomized animals administered 1.0 mg/kg/day raloxifene (Eli Lilly and Co., Indianapolis, IN) in CDX; 5) ovariectomized animals administered 0.1 mg/kg/day EM652 (Eli Lilly and Co.) in CDX; and 6) ovariectomized animals administered 8 μg/kg/day alendronate (Eli Lilly and Co.) in saline. The doses were chosen because they were shown previously to be fully efficacious in rats (Schenk et al., 1986; Sato et al., 1996; Martel et al., 2000).
The study was initiated 5 days after ovariectomy, and compounds were administered by oral gavage (except for the subcutaneous administration of alendronate) for 35 days (40 days after surgery), after which the proximal femora were collected 24 h after dosing. Two additional studies were conducted for validation purposes and were executed exactly as the 5-week study above (except for the omission of the EM652 group). The first validation study dosed the animals for 5 weeks and used femora for histomorphometric analysis; the proximal end of the contralateral femur was subjected to RNA isolation. The second validation study dosed animals for 9 days and analyzed the RNA from the distal end of the femur. At sacrifice, anesthetized rats were subjected to cardiac puncture and asphyxiated by CO2 inhalation. Animals were fasted the night before termination of each study, and all animal procedures were reviewed by an internal animal welfare committee to ensure compliance with National Institutes of Health guidelines.
Blood samples were allowed to clot at 4°C for approximately 2 h and then centrifuged at 2000g for 10 min. Serum samples were collected and stored at -70°C for subsequent analysis of serum cholesterol. Quantitative determination of cholesterol levels was achieved by measurement of cholesterol esterase/cholesterol oxidase activity using a Roche/Hitachi 917 automated chemistry analyzer.
RNA Isolation and Northern Analysis. The epiphysis and periosteum were removed from the proximal femora, and a 3-mm section of the metaphysis was collected and directly immersed in liquid nitrogen. Samples were then stored in liquid nitrogen until subjected to RNA analysis, at which time they were mechanically homogenized in Ultraspec RNA Isolation reagent (Biotexc Laboratories, Houston, TX) according to the manufacturer's instructions. Twenty-five micrograms of total RNA was electrophoretically separated, transferred to nylon membranes, and probed with radiolabeled cDNA probes as described previously (Ma et al., 2001). Gene expression was normalized to either 18S ribosomal or cyclophilin expression.
For real-time quantitative reverse transcription PCR, an ABI Prism Sequence Detection System 5700 was used and the primer-probe sets for the genes described were obtained as Assay-on-Demand reagents (Applied Biosystems, Foster City, CA). Before cDNA synthesis, 5 μg of total RNA were DNase-treated for 30 min at 37°C (DNA-free kits; Ambion, Austin, TX). RNA was reverse-transcribed from random hexamer primers using SuperScript II reverse transcription kit (Invitrogen, Carlsbad, CA). Specific amplification reactions from the cDNAs were carried out via a two-step, real-time PCR, and relative quantities were obtained by generating a standard curve for each gene. For normalization, amplification of 18S ribosomal RNA was performed for each sample in the same PCR run.
Microarray Analysis. Affymetrix rat genome U34A microarrays were used to determine transcript abundance from total RNA samples of individual metaphyseal samples. Total RNA was labeled according to manufacturer's instructions (Affymetrix, Santa Clara, CA). In brief, all samples were cleaned using RNeasy spin columns (QIAGEN, Valencia, CA). Double-stranded cDNA was synthesized from 10 μg of clean total RNA using SuperScript II cDNA synthesis kit (Invitrogen) and the T7-(dT)24 primer containing a T7 promoter [5'-GCCAGTGAATTGTAATACGACTCACTATAGGGAGGCGG (dT)24-3'; Genset Corp, Evry, France]. Phase Lock Gel tubes (1.5 ml; Eppendorf, Westbury, NY) were used to clean cDNA after phenol/chloroform/isoamyl alcohol extraction. Biotin-labeled cRNA was synthesized from cDNA using BioArray High Yield RNA Transcript Labeling kit (Enzo Diagnostics, Farmingdale, NY) and cleaned by RNeasy spin columns (QIAGEN). Clean cRNA was fragmented by incubation at 94°C in the presence of 40 mM Tris-acetate, pH 8.1, 100 mM potassium acetate, and 30 mM magnesium acetate for 35 min. Fragmented cRNA (10 μg) was hybridized to RG_U34A arrays for 16 h at 45°C with rotation (60 rpm). Each microarray was washed and stained using an Affymetrix Fluidics Station 400 and scanned in an Affymetrix confocal GeneArray scanner. Affymetrix MAS 4.0 software was used to scale data to a target intensity of 1500 and calculate transcript abundance.
Statistical Analysis. Five animals were used in each treatment group to control for biological variation. Each biological sample was then hybridized to duplicate chips to account for the variation caused by chip performance, hybridization quality, and other differences. To test whether a gene is differentially expressed, a mixed effect model was fitted on each of the 8799 probe sets on the chip. The intensity value (Affymetrix MAS4 signal) of a particular gene was modeled as
(1)
where Ykij is the signal of the jth replicate of animal i from treatment group k, μk is the group mean of treatment k, i(k) is the animal variation (random effect) having distribution , and kij, independent of i(k), is the measurement error (chip-to-chip variation) after distribution .
Because thousands of hypotheses were tested simultaneously, the issue of multiplicity is a big concern. To eliminate the false positives, Benjamini and Hochberg's false discovery rate (FDR) was used to adjust the p values derived from the above mixed model (Benjamini and Hochberg, 1995). The algorithm of FDR calculation could be simplified as follows: Suppose that being tested are m hypotheses Hi with corresponding p value Pi, i = 1, 2,..., m. Let P(i) be the ith p value ranked from the smallest to the largest with the corresponding hypothesis noted as H(i). Let the adjusted p value be labeled as , they are calculated as
The FDR is controlled at level q if we reject all H(j), j = 1, 2,..., k, where
(2)
The analysis of variance tests were done with treatment groups being the fixed effect and animals being the random effect. The p values for the pair-wise comparisons were derived based on the corresponding t statistics.
Bioinformatics Analysis. Principal component analysis was employed to reduce the dimensionality of the data and to assess any animal-to-animal variability by taking advantage of coregulation among a large number of genes while retaining as much variation as possible. The reduction is achieved by transforming the data into a new set of independent variables, the principal components. The principal components are ordered in such a way that the first few retain most of the variation present in all original genes. The analysis was performed in R using the principal component analysis function with standard data (Venables and Ripley, 1997). The first three principal components with at least 60% variation were exported into DecisionSite (Spotfire, Somerville, MA) to generate sample scatter plots for visually examining the data structure.
Differentially expressed genes were identified as significant if p < 0.05 and the median signal was larger than 500 for each animal within one or more treatment groups. Hierarchical clustering analysis (HCA) was done in DecisionSite in Euclidean space by the complete linkage method. Heat map visualization of clusters formed was generated using range-scaled expression values. To compensate for local minima often seen in HCA, self-organizing maps (SOM) were also employed to cluster genes. SOM analysis was carried out in DecisionSite with a grid size of 4 x 4 with default parameters. Clusters from SOM analysis were compared with those from HCA and similar clusters were merged into one.
To determine the statistical significance of the numbers of genes maintained at sham level by drug treatment, the proportion of genes in a given gene ontology (Ashburner et al., 2000) term relative to the total number of genes that had changed by ovariectomy was calculated. For each of five groups, the proportion of genes in a given term relative to the total number of genes in that group was calculated. Over-representation significance, when k representors are present in a sample size n, was calculated based on the hypergeometric cumulative distribution function:
(3)
where N is the number of genes that had changed with ovariectomy, M is the number of genes in the gene ontology term of interest, n is the number of genes in the treatment group, and k is the number of genes in the treatment group that are in the gene ontology term of interest. N, M, n, and k are integers such that 0 M N, 0 < n N. These calculations were done on software written by Eli Lilly and Co.; however, comparable software is publicly available (Boyle et al., 2004).
Bone Parameters. Quantitative computed tomography of the distal femur and biomechanical analyses of the proximal femur from the 5-week validation study were performed as described previously (Sato et al., 1997).
Gene Annotation. Target sequences for each chip were downloaded from the Affymetrix web site and then compared with the NCBI genome builds, with UniGene, and with RefSeq transcripts with BLAT. Annotations from these sources were used to map the probe sets both to LocusLink IDs and to full-length sequence IDs. The LocusLink IDs were mapped into the HumanPSD database (Hodges et al., 2002) via indices provided by that database. Note that this database aspires to contain the full protein complement of mouse and rat as well as human. For probe sets without a LocusLink based mapping, the full-length sequences were compared with the protein sequences in HumanPSD. Alignments with at least 100 amino acids of 100% identity and/or a BLAST e value of better than 1e-20 were recorded. Multiple identifications at the same reliability level were suppressed as potential conflicts. Functional information, including gene ontology classification, gene names, and descriptions were retrieved from HumanPSD.
Results
Ovariectomy and treatment by all compounds were well tolerated, with no clinical issues observed. Study outcomes (body weight and serum cholesterol levels) of each group used to collect femoral metaphyseal RNA for the microarray are detailed in Table 1. Ovariectomy increased serum cholesterol and body weight in vehicle control rats (Ovx) relative to sham. Compounds signaling through the estrogen receptor demonstrated a decrease in body weight as expected; it is known that estrogen receptor agonism reduces the ovariectomy-induced hyperphagia associated with estrogen loss (Geary, 2001; Meli et al., 2004). E2 and raloxifene treatment significantly reduced both body weight and cholesterol below Ovx and sham, whereas alendronate had no effect relative to Ovx (Black et al., 1994; Frolik et al., 1996). EM652 significantly lowered cholesterol below that of sham and Ovx, but the lowering of body weight did not achieve statistical significance in our study; however, this has been demonstrated by others (Martel et al., 2000).
Bioinformatics Analyses of Gene Changes. To explore hidden patterns and to visually identify coregulated genes, unsupervised clustering algorithms were used for data analysis. Principal component analysis evaluated the entire microarray data from all animals and all treatments without any filtering of the data. A subtle pattern emerged in the overall expression profile that could be used to distinguish various treatment groups. The duplicate microarray chips from each alendronate and E2 animal appeared to cluster together, whereas the EM652, raloxifene, and Ovx chip profiles seemed to be more closely associated. The sham group chips clustered together and were positioned intermediate between the other two groupings. It should be noted that duplicate chips from one animal in the EM652 group did not cluster with the remainder of its treatment group (Fig. 1).
To understand which genes relevant to ovariectomy-induced bone loss were driving these associations, we limited our subsequent analysis to those 644 probe sets that were significantly changed (p < 0.05) by Ovx relative to sham. Genes changed after each treatment regimen relative to Ovx control rats were determined for each antiresorptive agent: E2 (613 probe sets), raloxifene (765), EM652 (652), and alendronate (737) and compared with the list of ovariectomy-induced gene changes. The intersection of these data generated 380 probe sets that represent 334 unique genes changed by ovariectomy that are additionally modulated by one or more of the antiresorptive agents (see Supplemental data). The median of the intensity data for each treatment group was then subjected to HCA. The heat map visualization of the data set (Fig. 2) illustrates that gene changes associated with Ovx were almost equally distributed between induction or repression from sham levels, and each of the antiresorptive therapies had a unique expression pattern compared with the others.
Given the complexity of the data, we further used SOMs to identify unique patterns of expression for each treatment. SOM analysis (Fig. 3) identified unique patterns of gene expression that could be assigned into four broad categories: 1) genes altered by ovariectomy but restored to or maintained near sham levels by all antiresorptive treatments (clusters 5, 6, 15, and 16); 2) genes uniquely kept near sham levels only by raloxifene whereas the other therapies had no effect (clusters 8, 9, 13, and 14); 3) genes uniquely returned to or below that of sham by all of the agents except raloxifene and partially by EM652 (cluster 1); and 4) genes uniquely kept near sham levels only by E2 (clusters 2 and 11). The genes that were commonly regulated by all the antiresorptives are listed in Table 2. E2 and raloxifene uniquely regulated genes back toward sham level (i.e., no other treatment did so) and are listed as "unique" genes in Tables 3 and 4. In contrast, there were no specific clusters of genes that were solely regulated by EM652 or by alendronate.
Gene Changes Resulting from Ovariectomy. At 40 days after ovariectomy, the expected increase in expression from sham level associated with bone formation genes [such as collagen type I 2 (Col1a2 + 2.4-fold), collagen type V 1 (Col5a1 + 1.4-fold), osteocalcin (Ocn + 1.9-fold), osteonectin (Sparc + 2.1-fold), and decorin (+2.3-fold)] were observed by microarray. Increases in osteoclastic genes (i.e., cathepsin K, tartrate resistant acid phosphatase, and calcitonin receptor) were either not expressed above background levels or were not significantly changed by ovariectomy. This absence was not surprising; the 5-week time point was well beyond the peak of osteoclastic activity (2 weeks), and the trabecular bone volume at the metaphyseal site was reduced by 40% (Wronski et al., 1989). However, because the sample region is rich in osteoblasts, the osteoblastic marker of bone resorption (RANK ligand) was measured to demonstrate that signature molecular changes could be observed after ovariectomy by microarray. Because probe sets for RANK ligand were not included on the chip, its expression was analyzed by Northern blot and found to be elevated (+2-fold) with ovariectomy (data not shown). These data are consistent with bone histomorphometry of the Ovx rat that showed rapid increase in bone formation rates (BFRs) in the metaphysis, along with up-regulation of osteoclastic activity (Turner et al., 1994). More than 300 genes were coregulated with these formation and resorption genes after ovariectomy. These ovariectomy-induced gene changes are implicated in a variety of molecular functions and biological processes that are summarized by noting their gene ontology descriptor as outlined under Materials and Methods. Table 5 details the gene ontology assignment of gene changes altered by Ovx relative to sham.
Genes Regulated by All Antiresorptive Agents. Seventy genes listed in Table 2 were commonly regulated by all antiresorptives relative to Ovx after 5 weeks of treatment as identified by SOM analysis. These 70 genes were maintained near sham levels with all treatments and were either elevated (n = 42) or suppressed (n = 28) by ovariectomy. The majority of these changes were small, but 44% of the genes altered by ovariectomy were changed 1.5-fold or more from sham. Genes that were changed by less than 1.5-fold seemed to be stable changes in that the FDR was less than 0.3 for 77% of the genes. Genes affecting matrix production and mineralization that were significantly elevated by ovariectomy [collagen type V2 (+2.7-fold), serine proteinase inhibitor clade H (Serpinh1 + 1.7-fold), lysyl oxidase (+1.8-fold), and collagen type XI1 (+3-fold), Sparc (+3.0 fold), and tissue inhibitor of metalloproteinase 1 (+2.1-fold)] were suppressed near to sham levels by all test compounds. On the other hand, each drug increased a subset of genes that were suppressed by ovariectomy, such as fibroblast growth factor 9 (-2-fold), mucin (-1.7-fold), K-cadherin (-1.9-fold), and retinoic acid receptor (-5.8-fold). The molecular function and biological processes potentially affected by these gene changes were evaluated by assessing the gene ontology terms associated with each gene as shown in Table 5. Statistical testing of these common gene changes relative to those changed by Ovx was performed to determine which categories were more significantly affected; these are listed in Table 6. Three genes regulated commonly by all treatments were validated in a separate animal study [lysyl oxidase, Serpinh1, and procollagen C-proteinase enhancer protein (Pcolce)]. Comparisons of their regulation relative to ovariectomized control rats in the original microarray study and in a second independent study are presented in Table 7.
Genes Uniquely Regulated by Raloxifene. Raloxifene maintained more genes near sham level than any other treatment. In addition to the commonly regulated genes (Table 2), raloxifene uniquely returned 67 genes (Table 3) near to sham level, whereas the other three drug treatments had no effect on these genes (Fig. 2, clusters 2 and 3). Sixty-three percent of the unique raloxifene genes were suppressed by ovariectomy and subsequently increased by raloxifene treatment; 41% of these were altered by at least 1.5-fold from Ovx. The gene ontology categorization of these genes are listed in Table 5 and the results of statistical testing of gene ontology terms most significantly maintained near to sham levels by raloxifene are listed in Table 6.
Two of the raloxifene unique genes with larger -fold change values were evaluated in an independent 5-week validation study. Carbonic anhydrase 4 (Ca4) and cytochrome P450 family 27 subfamily A polypeptide 1 (Cyp27a1) were reduced by ovariectomy on the microarray and then uniquely increased by raloxifene. Ovariectomy did reduce the levels of both of these genes in the follow-up study, but only the Cyp27a1 reduction was significantly (p < 0.05) lowered. Neither the Cyp27a1 nor the Ca4 expression changes after antiresorptive treatment achieved statistical significance (Table 7).
Genes Uniquely Regulated by Estrogen. Estrogen also uniquely maintained a small subset of genes near sham level in addition to the 70 commonly regulated genes. Genes that were altered by ovariectomy and were returned toward sham levels solely by E2 treatment are listed in Table 4 and highlighted in clusters 4 and 5 of Fig. 2. The changes were nearly equivalently divided between genes that were increased and those decreased by Ovx and uniquely regulated to sham levels by E2. Fifty percent of these E2 unique genes were changed by 1.5-fold or more relative to Ovx. The gene ontology classification of these genes (Tables 5 and 6) revealed that E2 significantly maintained genes near sham levels that were associated with perception of a stimulus (mechanical and pain stimuli), feeding behavior, and antigen binding.
Suppression of Bone Formation Genes by Alendronate and Estrogen but Not Raloxifene. Cluster analysis of the microarray data identified an additional group of genes that were elevated by Ovx and were suppressed to or below that of sham levels by all treatments except raloxifene (Fig. 2 cluster 1, Fig. 3 cluster 1). The identities of the genes in this cluster were those associated with osteoblastic activity [several collagens, osteocalcin, Sparc, and biglycan (Bgn)]. E2 and alendronate suppressed the expression of these genes below that of sham; however, EM652 only partially suppressed their expression (Fig. 4). Three additional genes clustered together with these osteoblastic genes: immunoglobulin superfamily member 4 (Igsf4 or syncam), homeobox protein C8 (Hox8c or Hox3a), and fibroblast growth factor receptor 1 (Fgfr1) implying that they may also serve an important role in osteoblastic activity because they are coregulated with the collagens Ocn, Bgn, and Sparc.
Additional animal studies were conducted, and representative genes from this cluster were analyzed to determine the consistency of treatment effects in independent assays. In addition, the femora were subjected to biomechanical testing in the first validation study that was an exact replicate of the original 5-week array study. Biomechanical testing on the femoral neck (the site used for array analysis) from this 5-week validation study showed no differences in the strength of the bone from Ovx by any drug treatment (Fig. 5A); however, the sham group was significantly stronger than Ovx (p < 0.05). At a different bone site (the distal femur), pharmacological efficacy was observed for all treatments (p < 0.001) with the maintenance of bone mineral density (BMD) at sham levels (Fig. 5B).
The expression of the osteoblast activity genes in the proximal femur was reevaluated in the 5-week validation study and was observed to parallel the differential expression pattern observed on the array. Although the -fold change from control values were not identical between the array and the 5-week Taqman validation experiment, the rank order at which these agents suppressed the bone formation activity genes was the same. Alendronate was the most suppressive of all agents tested and consistently suppressed the expression of the bone formation genes (i.e., Ocn, collagens, Sparc, and Bgn) below that of sham in the array and the repeated study. Estrogen was less suppressive than alendronate but suppressed the expression of these genes more than raloxifene. Raloxifene lowered the expression of some of these genes in the validation study more than had been observed on the microarray; however, levels were not different from sham in nearly each case (Table 7).
A second validation study evaluated the expression profile of bone formation genes in the distal femur at an earlier time point (9 days) before an effect on BMD would be observed (Sato et al., 1994). The increased expression of Col 5a1, Sparc, and Bgn induced after 5 weeks of ovariectomy were not yet altered at this earlier time point. However, Ocn and Col1a2 were significantly elevated by ovariectomy, and the suppression of Ocn by E2 and alendronate was also observed by 9 days in the distal femur (data not shown).
Discussion
Ovariectomized rats greater than 5 months of age have been shown to reproducibly lose cancellous bone from axial and appendicular skeletal sites as a result of estrogen deficiency, not unlike postmenopausal women (Turner et al., 1994). Upon pharmacologic administration of maximally efficacious doses of E2, raloxifene (Sato et al., 1995, 1996), and alendronate (Toolan et al., 1992), a similar preservation of BMD can be achieved in femora of ovariectomized rats. Given the differences in the mechanism of action by which estrogens/SERMs and alendronate sustain BMD after ovariectomy (Sato et al., 1999), a gene array analysis was initiated in an effort to elucidate possible differences of these compounds on skeletal physiology.
After 5 weeks of treatment, the overall expression profile of all genes on the microarray suggested that E2 and alendronate were most similar, whereas the gene expression profiles of EM652 and raloxifene seemed to be distinctly different from E2 (Fig. 1). This was an unexpected finding, in that we had hypothesized that because of the similarity in mechanism of action (i.e., estrogen receptor agonism), the SERMs and E2 would more closely resemble each other. However, array analyses showed that EM652 and raloxifene were quite different from E2 in their gene expression profile in bone. To understand more fully which genes were driving this association between the compounds, we looked in more detail at only those genes (334 unique genes or 380 probe sets), which were significantly modulated by ovariectomy and were maintained near sham levels by any of the drug treatments.
Ovariectomy induced the expected changes in bone formation and resorption genes such as Ocn, Sparc, Col1a2 and Col5a1, and Rank ligand. Because osteoclastic activity is elevated after ovariectomy, one would expect to observe increases in osteoclastic genes such as calcitonin receptor, cathepsin K, and tartrate-resistant acid phosphatase. However, these genes were either not significantly altered by ovariectomy or were so lowly expressed they did not meet our exclusion criteria. The likely explanation for the lack of detection of these gene changes is the kinetics of bone loss in the ovariectomized rat model. The 5-week time point at which we collected femoral RNA was well beyond the peak of osteoclastic activity, and the cancellous bone volume was reduced by nearly 40% (Wronski et al., 1989). In unpublished data from our laboratory, we have measured increases in these osteoclast genes after 12 days of ovariectomy, but by 5 weeks, these changes were no longer observed, probably because of the lack of bone surface in the ovariectomized group, which was lost by this time.
As expected, there was a subset of genes (n = 70) commonly regulated toward sham levels by all antiresorptive agents irrespective of their differences in mechanism of action, representing 20% of the total ovariectomy-induced and drug-responsive genes. The regulation of these common genes probably represents a core set of molecular changes necessary to maintain bone mass after ovariectomy in the rat and occur before any biomechanical effect in this bone site (Fig. 5). Validation in an independent experiment of three genes commonly regulated by all treatments (Pcolce, lysyl oxidase, and Serpinh1) confirmed a similar down-regulation by all agents tested (Table 7). In addition to the 70 commonly regulated genes, only two of the treatments (raloxifene and E2) could modulate subsets of genes uniquely by their treatment. There were not any genes uniquely maintained near sham level by EM652 or alendronate in that at least one of the other agents also regulated the genes associated with these treatments. Validation of two raloxifene unique genes (Cyp27a1 and Ca4) in a repeated 5-week experiment was not successful. The change in expression by ovariectomy was statistically lowered for only one of the genes (Cyp27a1), and subsequent increases in expression by raloxifene did not achieve statistical significance with either gene (Table 7), highlighting the importance of validating gene expression in independent assays.
Although raloxifene treatment was the most efficacious agent at maintaining gene expression near sham levels in the microarray, there was a small cluster of genes that raloxifene did not suppress or only slightly lowered after their ovariectomy-induced increase. The genes which remained elevated by raloxifene, were identified as those associated with bone formation activity (Ocn, Col1a1, Col1a2, Col5a2, Bgn, and Sparc). E2 and alendronate suppressed the expression of these osteoblastic genes below that of sham, whereas EM652 treatment resulted in an intermediate expression level (Fig. 4). A replicated study (5-week treatment) confirmed that alendronate treatment was the most suppressive of the expression of these formation genes, whereas raloxifene was the least (Table 7). An additional validation study revealed that the suppression of Ocn by E2 and alendronate occurred as early as 9 days of treatment (data not shown). Although the present study evaluated only RNA changes, the significantly enhanced suppression of Ocn protein by alendronate versus that of raloxifene has been observed in the serum of postmenopausal women after 1 year of treatment (Johnell et al., 2002). These data are consistent with previously published histomorphometric analyses in postmenopausal women in which alendronate reduced bone formation activity (BFR per unit bone surface and activation frequency) by 90% (Chavassieux et al., 1997), whereas estrogen had milder suppressive effects (Lufkin et al., 1992; Weinstein et al., 2003). Raloxifene had no effect on BFR per unit of bone volume or on the formation period in clinical samples (Ott et al., 2002). These array data suggest a possible explanation as to why the combined effects of alendronate and PTH were found to be less efficacious than PTH alone (Black et al., 2003; Finkelstein et al., 2003) in clinical studies; but when PTH was given in combination with raloxifene, an additive skeletal effect was observed (Cosman et al., 2004; Deal et al., 2004). A possible explanation is that suppression by alendronate of osteoblastic function (as evidenced by the lack of expression of osteoblastic genes) antagonized PTH efficacy, whereas raloxifene treatment did not impair osteoblast activity and thus can complement PTH in the skeleton.
Because Hox3a, Fgfr1, and Syncam clustered tightly (Fig. 4) with several known bone formation activity genes (Ocn, Bgn, and collagens) there is great interest in pursuing a deeper understanding of their relevance to bone formation in the ovariectomized rat. Fgfr1 is critical to embryonic craniofacial development (Rice et al., 2003) and has been shown to be elevated robustly during the formation of fracture callus in rat femur (Nakajima et al., 2001), and activation mutations in this gene lead to increased bone formation in the crania of both human and mice (Zhou et al., 2001). Hox3a, a transcriptional repressor of the group 8 Hox family, is also known to be involved with osteoblast and cartilage differentiation (Yueh et al., 1998; Yang et al., 2000) and mouse skeletal patterning (van den Akker et al., 2001). Our data support the importance of Fgfr1 and Hox3a expression in the adult rat proximal femur, and we have now demonstrated their regulation during ovariectomy-induced bone formation. Syncam (Igsf4) has been described as an important intracellular adhesion protein and as a possible tumor suppressor in lung (Masuda et al., 2002). Although the roles of Igsf4 in bone formation have never been studied, the tight association of its regulation with that of known bone formation genes in this study implies that it plays a role in bone formation of the adult rat skeleton.
The data in the current study demonstrate that although SERMs may be estrogen-like in their effects on BMD in the bone, the molecular events after prolonged treatment with raloxifene or EM652 show large differences in RNA expression profiles compared with that of E2. Whether these expression changes are the result of differences in levels of cellular activity, function, RNA stabilization or transcription, or cell number is unknown; any of these parameters could be changed as a result of prolonged antiresorptive treatment that achieves changes in bone strength and architecture. Further study is necessary to understand whether some of the differences in gene expression observed in this study could contribute to architectural or biomechanical qualities that further differentiate these antiresorptive therapies. Because it has been demonstrated that BMD cannot completely predict fracture efficacy of a compound (Riggs and Melton, 2002; Sarkar et al., 2004), there is great interest in furthering our understanding of what parameters define bone quality and what molecular events could help to predict bone fracture. Profiling the RNA changes associated with antiresorptive therapy demonstrate that even though these drugs have similar efficacy at maintaining BMD in the ovariectomized rat (Fig. 5), the underlying molecular events that collectively result in improved bone strength and density seem to be complex. This study helps to uncover some of the possible mechanistic changes resulting from various drug treatments and provides new avenues for investigation of the role these differences may play in determining bone quality. Gene expression changes may help to complement current methods of assessing bone quality (i.e., bone density measurements) so that better indicators of fracture risk could be developed.
Acknowledgements
We thank John Calley and Amar Kumar for excellent bioinformatics support; Harlan Cole, Rick Cain, Ellen Rowley, and Pam Shetler for in vivo expertise; and Allan Schmidt for QCT and biomechanical assessments.
Footnotes
This study was funded by Lilly Research Laboratories.
L.M.H. and R.L. contributed equally to this manuscript.
ABBREVIATIONS: SERM, selective estrogen receptor modulator; E2, 17--ethinyl estradiol; EM652, (S)-3-(4-hydroxyphenyl)-4-methyl-2-[4-[2-(1-piperidinyl)ethoxy]phenyl]-2H-1-benzopyran-7-ol; CDX, hydroxypropyl--cyclodextrin; PCR, polymerase chain reaction; FDR, false discovery rate; HCA, hierarchical clustering analysis; SOM, self-organizing map; Ovx, ovariectomized vehicle control rats; BFR, bone formation rate; Cyp27a1, cytochrome P450 family 27 subfamily A polypeptide 1; Ca4, Carbonic anhydrase 4; BMD, bone mineral density; PTH, parathyroid hormone; Col1a2, collagen type I 2; Col5a1, collagen type V 1; Ocn, osteocalcin; Sparc, osteonectin; Serpinh1, serpine proteinase inhibitor clade H; Pcolce, procollagen C-proteinase enhancer protein; Ca4, carbonic anhydrase 4; Cyp 27a1, cytochrome P450 family 27 subfamily A polypeptide 1; Bgn, biglycan; Igsf4, immunoglobulin superfamily member 4; Hox3a, homeobox protein C8; Fgfr1, fibroblast growth factor receptor 1.
The online version of this article (available at http://molpharm.aspetjournals.org) contains supplemental material.
References
Anderson GL, Limacher M, Assaf AR, Bassford T, Beresford SA, Black H, Bonds D, Brunner R, Brzyski R, Caan B, et al. (2004) Effects of conjugated equine estrogen in postmenopausal women with hysterectomy: the Women's Health Initiative randomized controlled trial. J Am Med Assoc 291: 1701-1712.
Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, Cherry JM, Davis AP, Dolinski K, Dwight SS, Eppig JT, et al. (2000) Gene ontology: tool for the unification of biology. The gene ontology Consortium. Nat Genet 25: 25-29.
Benjamini Y and Hochberg Y (1995) Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc 57: 289-300.
Black DM, Greenspan SL, Ensrud KE, Palermo L, McGowan JA, Lang TF, Garnero P, Bouxsein ML, Bilezikian JP, Rosen CJ, et al. (2003) The effects of parathyroid hormone and alendronate alone or in combination in postmenopausal osteoporosis. N Engl J Med 349: 1207-1215.
Black LJ, Sato M, Rowley ER, Magee DE, Bekele A, Williams DC, Cullinan GJ, Bendele R, Kauffman RF, Bensch WR, et al. (1994) Raloxifene (LY139481 HCI) prevents bone loss and reduces serum cholesterol without causing uterine hypertrophy in ovariectomized rats. J Clin Investig 93: 63-69.
Boyle E, Weng S, Gollub J, Jin H, Botstein D, Cherry JM, and Sherlock G (2004) GO:TermFinder—open source software for accessing gene ontology information and finding significantly enriched gene ontology terms associated with a list of genes. Bioinformatics 20: 3710-3715.
Cauley JA, Robbins J, Chen Z, Cummings SR, Jackson RD, LaCroix AZ, LeBoff M, Lewis CE, McGowan J, Neuner J, et al. (2003) Effects of estrogen plus progestin on risk of fracture and bone mineral density: the Women's Health Initiative randomized trial. J Am Med Assoc 290: 1729-1738.
Chavassieux PM, Arlot ME, Reda C, Wei L, Yates AJ, and Meunier PJ (1997) Histomorphometric assessment of the long-term effects of alendronate on bone quality and remodeling in patients with osteoporosis. J Clin Investig 100: 1475-1480.
Cosman F, Nieves J, Barbuto N, Zion M, and Lindsay R (2004) Parathyroid hormone added to raloxifene and subsequent maintenance of BMD gain with raloxifene alone. J Bone Miner Res 19: S98.
Deal C, Omizo M, Schwartz EN, Eriksen EF, Cantor P, Wang J, Glass EV, Myers SL, and Krege JH (2004) Raloxifene in combination with teriparatide reduces teriparatide-induced stimulation of bone resorption but not formation in postmenopausal women with osteoporosis. J Bone Miner Res 19: S44.
Delmas PD, Bjarnason NH, Mitlak BH, Ravoux AC, Shah AS, Huster WJ, Draper M, and Christiansen C (1997) Effects of raloxifene on bone mineral density, serum cholesterol concentrations and uterine endometrium in postmenopausal women. N Engl J Med 337: 1641-1647.
Ettinger B, Black DM, Mitlak BH, Knickerbocker RK, Nickelsen T, Genant HK, Christiansen C, Delmas PD, Zanchetta JR, Stakkestad J, et al. (1999) Reduction of vertebral fracture risk in postmenopausal women with osteoporosis treated with raloxifene: results from a 3-year randomized clinical trial. Multiple Outcomes of Raloxifene Evaluation (MORE) Investigators. J Am Med Assoc 282: 637-645.
Finkelstein JS, Hayes A, Hunzelman JL, Wyland JJ, Lee H, and Neer RM (2003) The effects of parathyroid hormone, alendronate, or both in men with osteoporosis. N Engl J Med 349: 1216-1226.
Frolik CA, Bryant HU, Black EC, Magee DE, and Chandrasekhar S (1996) Time-dependent changes in biochemical bone markers and serum cholesterol in ovariectomized rats: effects of raloxifene HCl, tamoxifen, estrogen and alendronate. Bone 18: 621-627.
Geary N (2001) Estradiol, CCK and satiation. Peptides 22: 1251-1263.
Hodges PE, Carrico PM, Hogan JD, O'Neill KE, Owen JJ, Mangan M, Davis BP, Brooks JE, and Garrels JI (2002) Annotating the human proteome: the Human Proteome Survey Database (HumanPSD) and an in-depth target database for G protein-coupled receptors (GPCR-PD) from Incyte Genomics. Nucleic Acids Res 30: 137-141.
Johnell O, Scheele WH, Lu Y, Reginster JY, Need AG, and Seeman E (2002) Additive effects of raloxifene and alendronate on bone density and biochemical markers of bone remodeling in postmenopausal women with osteoporosis. J Clin Endocrinol Metab 87: 985-992.
Kimmel D (ed) (1996) Animal Models for in Vivo Experimentation in Osteoporosis Research. Academic Press, San Diego.
Liberman UA, Weiss SR, Broll J, Minne HW, Quan H, Bell NH, Rodriguez-Portales J, Downs RW Jr, Dequeker J, and Favus M (1995) Effect of oral alendronate on bone mineral density and the incidence of fractures in postmenopausal osteoporosis. The Alendronate Phase III Osteoporosis Treatment Study Group. N Engl J Med 333: 1437-1443.
Lufkin EG, Wahner HW, O'Fallon WM, Hodgson SF, Kotowicz MA, Lane AW, Judd HL, Caplan RH, and Riggs BL (1992) Treatment of postmenopausal osteoporosis with transdermal estrogen. Ann Intern Med 117: 1-9.
Ma YL, Cain RL, Halladay DL, Yang X, Zeng Q, Miles RR, Chandrasekhar S, Martin TJ, and Onyia JE (2001) Catabolic effects of continuous human PTH (1-38) in vivo is associated with sustained stimulation of RANKL and inhibition of osteoprotegerin and gene-associated bone formation. Endocrinology 142: 4047-4054.
Martel C, Picard S, Richard V, Belanger A, Labrie C, and Labrie F (2000) Prevention of bone loss by EM-800 and raloxifene in the ovariectomized rat. J Steroid Biochem Mol Biol 74: 45-56.
Masuda M, Yageta M, Fukuhara H, Kuramochi M, Maruyama T, Nomoto A, and Murakami Y (2002) The tumor suppressor protein TSLC1 is involved in cell-cell adhesion. J Biol Chem 277: 31014-31019.
Meli R, Pacilio M, Raso GM, Esposito E, Coppola A, Nasti A, Di Carlo C, Nappi C, and Di Carlo R (2004) Estrogen and raloxifene modulate leptin and its receptor in hypothalamus and adipose tissue from ovariectomized rats. Endocrinology 145: 3115-3121.
Nakajima A, Nakajima F, Shimizu S, Ogasawara A, Wanaka A, Moriya H, Einhorn TA, and Yamazaki M (2001) Spatial and temporal gene expression for fibroblast growth factor type I receptor (FGFR1) during fracture healing in the rat. Bone 29: 458-466.
Ott SM, Oleksik A, Lu Y, Harper K, and Lips P (2002) Bone histomorphometric and biochemical marker results of a 2-year placebo-controlled trial of raloxifene in postmenopausal women. J Bone Miner Res 17: 341-348.
Rice DP, Rice R, and Thesleff I (2003) Fgfr mRNA isoforms in craniofacial bone development. Bone 33: 14-27.
Riggs BL and Hartmann LC (2003) Selective estrogen-receptor modulators -mechanisms of action and application to clinical practice [published erratum appears in N Engl J Med 2003 348:1192]. N Engl J Med 348: 618-629.
Riggs BL and Melton LJ 3rd (2002) Bone turnover matters: the raloxifene treatment paradox of dramatic decreases in vertebral fractures without commensurate increases in bone density. J Bone Miner Res 17: 11-14.
Riggs BL and Parfitt AM (2005) Drugs used to treat osteoporosis: the critical need for a uniform nomenclature based on their action on bone remodeling. J Bone Miner Res 20: 177-184.
Rodan GA and Fleisch HA (1996) Bisphosphonates: mechanisms of action. J Clin Investig 97: 2692-2696.
Sarkar S, Reginster JY, Crans GG, Diez-Perez A, Pinette KV, and Delmas PD (2004) Relationship between changes in biochemical markers of bone turnover and BMD to predict vertebral fracture risk. J Bone Miner Res 19: 394-401.
Sato M, Grese TA, Dodge JA, Bryant HU, and Turner CH (1999) Emerging therapies for the prevention or treatment of postmenopausal osteoporosis. J Med Chem 42: 1-24.
Sato M, Kim J, Short LL, Slemenda CW, and Bryant HU (1995) Longitudinal and cross-sectional analysis of raloxifene effects on tibiae from ovariectomized aged rats. J Pharmacol Exp Ther 272: 1252-1259.
Sato M, McClintock C, Kim J, Turner CH, Bryant HU, Magee D, and Slemenda CW (1994) Dual-energy x-ray absorptiometry of raloxifene effects on the lumbar vertebrae and femora of ovariectomized rats. J Bone Miner Res 9: 715-724.
Sato M, Rippy MK, and Bryant HU (1996) Raloxifene, tamoxifen, nafoxidine, or estrogen effects on reproductive and nonreproductive tissues in ovariectomized rats. FASEB J 10: 905-912.
Sato M, Zeng GQ, and Turner CH (1997) Biosynthetic human parathyroid hormone (1-34) effects on bone quality in aged ovariectomized rats. Endocrinology 138: 4330-4337.
Schenk R, Eggli P, Fleisch H, and Rosini S (1986) Quantitative morphometric evaluation of the inhibitory activity of new aminobisphosphonates on bone resorption in the rat. Calcif Tissue Int 38: 342-349.
Seedor JG, Quartuccio HA, and Thompson DD (1991) The bisphosphonate alendronate (MK-217) inhibits bone loss due to ovariectomy in rats. J Bone Miner Res 6: 339-346.
Toolan BC, Shea M, Myers ER, Borchers RE, Seedor JG, Quartuccio H, Rodan G, and Hayes WC (1992) Effects of 4-amino-1-hydroxybutylidene bisphosphonate on bone biomechanics in rats. J Bone Miner Res 7: 1399-1406.
Turner RT, Riggs BL, and Spelsberg TC (1994) Skeletal effects of estrogen. Endocrine Rev 15: 275-300.
van den Akker E, Fromental-Ramain C, de Graaff W, Le Mouellic H, Brulet P, Chambon P, and Deschamps J (2001) Axial skeletal patterning in mice lacking all paralogous group 8 Hox genes. Development 128: 1911-1921.
Venables W and Ripley BD (1997) Modern Applied Statistics with S-PLUS. Springer-Verlag, New York.
Weinstein RS, Parfitt AM, Marcus R, Greenwald M, Crans G, and Muchmore DB (2003) Effects of raloxifene, hormone replacement therapy and placebo on bone turnover in postmenopausal women. Osteoporos Int 14: 814-822.
Wronski TJ, Dann LM, Scott KS, and Cintron M (1989) Long-term effects of ovariectomy and aging on the rat skeleton. Calcif Tissue Int 45: 360-366.
Yang X, Ji X, Shi X, and Cao X (2000) Smad1 domains interacting with Hoxc-8 induce osteoblast differentiation. J Biol Chem 275: 1065-1072.
Yueh YG, Gardner DP, and Kappen C (1998) Evidence for regulation of cartilage differentiation by the homeobox gene Hoxc-8. Proc Natl Acad Sci USA 95: 9956-9961.
Zhou YX, Xu X, Chen L, Li C, Brodie SG, and Deng CX (2001) A Pro250Arg substitution in mouse Fgfr1 causes increased expression of Cbfa1 and premature fusion of calvarial sutures. Hum Mol Genet 9: 2001-2008.