Gene Expression Patterns in Bone Following Mechanical Loading

The advent of high-throughput measurements of gene expression and bioinformatics analysis methods offers new ways to study gene expression patterns. The primary goal of this study was to determine the time sequence for gene expression in a bone subjected to mechanical loading during key periods of the bone-formation process, including expression of matrix-related genes, the appearance of active osteoblasts, and bone desensitization. A standard model for bone loading was employed in which the right forelimb was loaded axially for 3 minutes per day, whereas the left forearm served as a nonloaded contralateral control. We evaluated loading-induced gene expression over a time course of 4 hours to 32 days after the first loading session. Six distinct time-dependent patterns of gene expression were identified over the time course and were categorized into three primary clusters: genes upregulated early in the time course, genes upregulated during matrix formation, and genes downregulated during matrix formation. Genes then were grouped based on function and/or signaling pathways. Many gene groups known to be important in loading-induced bone formation were identified within the clusters, including AP-1-related genes in the early-response cluster, matrix-related genes in the upregulated gene clusters, and Wnt/β-catenin signaling pathway inhibitors in the downregulated gene clusters. Several novel gene groups were identified as well, including chemokine-related genes, which were upregulated early but downregulated later in the time course; solute carrier genes, which were both upregulated and downregulated; and muscle-related genes, which were primarily downregulated. © 2011 American Society for Bone and Mineral Research.


Introduction
B one responds in an anabolic manner to physiologic dynamic loading. For example, the midshaft humerus in the throwing arm of baseball pitchers and catchers showed enhanced bone mass, structure, estimated strength, and resistance to torsion compared with the nonthrowing control arm. (1) In contrast, bone mineral density (BMD) in astronauts decreased 1.0% and 1.5% in the spine and hip, respectively, per month of spaceflight. (2,3) Mechanical loading is a potent anabolic stimulus that substantially strengthens bones. Using the rat forelimb loading model, we showed that loading for only 3 minutes/day over 16 weeks doubled the energy necessary to break the ulna (4) and increased the fatigue life of the bone by 100-fold. (5) Mechanical loading uses pathways currently being investigated for new drug development, such as low density lipoprotein receptor-related protein 5 (LRP5) and sclerostin. (6) Others have proposed that the mechanical loading response is fundamental to bone biology, and many drugs work by modulating the responsiveness of bone to loading. (7) Hence, we believe that mechanical loading can be used as a paradigm for bone anabolism. This approach has the advantage of producing a local response, so the contralateral limb can be used as a control to compare with loading effects on the loaded limb.
signaling pathways and gene expression in loaded bone, (12,13) so we hypothesize that the gene expression patterns are timedependent, with early activities stimulating a chain reaction of events that directly affects bone cell behavior and ultimately leads to bone formation.
The primary goal of this study was to determine the time sequence for gene expression in a bone subjected to mechanical loading. We evaluated loading-induced gene expression over a time course of 4 hours to 32 days. We then used bioinformatics tools to cluster genes into similar expression patterns and group genes within common signaling pathways.
Mechanical loading A standard model for bone loading was employed in which the right forelimb was loaded axially for 3 minutes per day while the left forearm served as a nonloaded contralateral control. (4,14,15) Prior to loading, animals were anesthetized with 3.0% isoflurane administered at a flow rate of 1.5 L/min. Compressive load was applied as an oscillating Haversine waveform for 360 cycles at a frequency of 2 Hz using a Bose ElectroForce 3200 Series electromechanical actuator (EnduraTEC, Eden Prairie, MN, USA). The peak load achieved during loading was 13 N, which has been shown previously to be anabolic. (14) Rats were subjected to loading sessions every day with 24 hours between sessions. The study groups listed earlier are referenced to the number of days (or hours) after the first bout of bone loading was applied. At the appropriate time point, animals were anesthetized with isoflurane and euthanized by cervical dislocation.

Histology
Nine additional adult female Lewis rats were subjected to the loading protocol for histologic analysis. These rats were euthanized 1 and 4 days after beginning loading. The shafts of the right and left forearms with intact ulnae and radii were dissected, freed of excess muscle, and fixed in 10% neutral buffered formalin (NBF) for 48 hours. The fixed forearms were decalcified in a 70:30 solution of 10% ethylenediamine tetraacetic acid (EDTA) and 4% phosphate-buffered formalin (PBF) for 4 weeks. After decalcification, forearms were embedded in paraffin and sectioned at the ulnar midshaft at 4 mm. Sections were stained with hematoxylin and eosin (H&E) and used to identify active osteoblasts on the periosteal surfaces of loaded ulnae. Active osteoblasts were counted and defined as osteoblast cell bodies that were plump and present in the layer of cells immediately adjacent to newly formed osteoid on the bone surface. The sections were photographed on a Nikon Optiphot-2 microscope (Nikon, Inc., Melville, NY, USA) using 5Â and 40Â objectives and imported into Image-Pro Plus 6.3 (MediaCybernetics, Inc., Bethesda, MD, USA) analysis software for quantification. Total perimeter of the periosteal surface was measured, and average osteoblast number per total bone perimeter (mm, Ob.N/B.Pm) was reported.

RNA isolation
The shafts of the right and left ulnae were dissected, freed of all soft tissue, and snap frozen in liquid nitrogen. The ulnae were stored at -808C until RNA isolation. RNA was extracted using Trizol (Invitrogen, Carlsbad, CA, USA) and RNeasy Mini Kits (Qiagen, Inc., Valencia, CA, USA). Frozen ulnae were placed into a mortar containing liquid nitrogen and crushed with a pestle. The crushed bone was homogenized in Trizol, incubated, and centrifuged. The supernatant was removed, and RNeasy Mini Kits were used to isolate RNA. RNA was treated with a DNA-free kit (Ambion, Austin, TX, USA) to remove any residual DNA. RNA quality and quantity were determined using a spectrophotometer (NanoDrop, Wilmington, DE, USA). A paired t test was used to compare average total RNA quantity obtained from loaded and control bones for each time group. Average RNA quantity and standard errors were reported, and a p value < .05 was considered statistically significant.
Quantitative polymerase chain reaction (qPCR) Three matched RNA samples from loaded and control ulnae for each time group were used for quantitative real-time PCR (qPCR) experiments. RNA was reverse transcribed using the SuperScript III kit with oligo(dT) primers (Invitrogen). cDNA was diluted to a concentration of 2.5 ng/mL and used in qPCR reactions. A portion of the rat collagen type 1a1 (Col1a1) gene sequence was amplified using a Taqman gene expression assay (assay ID: Rn01463848_m1, Applied Biosystems, Inc., Carlsbad, CA, USA). Serial dilutions of a single sample were amplified to calculate relative expression levels, which then were standardized to bactin expression to facilitate comparison among samples. The reactions were performed on an ABI 7900HT Fast Real-Time PCR System (Applied Biosystems). A paired t test was used to compare Col1a1 expression in loaded and control conditions. Average fold change and standard errors were reported, and p < .05 was considered statistically significant.

Exon arrays
The array methodology is summarized in a flow diagram (Fig. 1). Quality-control measures were employed to ensure that highquality RNA would be hybridized to exon arrays. A high-quality RNA sample was defined as having a minimum 260:280 ratio of 2.00. Four RNA samples had a 260:280 ratio of slightly less than 2.00, and these samples were chosen to optimize the quantity of total RNA as well as the quality. The range of 260:280 ratios of all samples used was 1.96 to 2.31.
Five matched (control and loaded) ulna RNA samples from each time group were used for exon array analysis. One exception to this was the 12-day group, where only four matched samples were used. RNA from the control and loaded ulnae from 54 animals were analyzed on separate arrays. The exon array hybridizations were carried out using the facilities of the Center for Medical Genomics (CMG) at Indiana University School of Medicine. One microgram of each sample was labeled and hybridized using the Affymetrix WT protocol [GeneChip Whole Transcript (WT) Sense Target Labeling Assay Manual, Version 4, Affymetrix, Santa Clara, CA, USA]. All processing was done in balanced batches. The exon arrays were scanned using the GeneChip Scanner 3000 using the Affymetrix GeneChip Operating System (GCOS). Data were exported for analysis in the Partek Genomics Suite (Partek, Inc., St. Louis, MO, USA).
The robust multiarray average (RMA) algorithm (16)(17)(18) was used to import raw data from the core probe sets, which represented over 8000 genes. A two-way ANOVA model [variables were loading condition (loaded or control) and animal] was used to identify differentially expressed genes, which were defined as having p < .01 and fold change beyond AE1.4. Probe sets with a signal value of less than 3 were not reliable and were excluded from analysis. Raw data and analyzed data were MIAMEcompliant (19) and were deposited in the Gene Expression Omnibus database (series accession number GSE22286).

Clustering analysis
Differentially expressed genes were further analyzed with a kmeans (ie, partitioning) clustering algorithm. The inclusion criterion for a gene to qualify for clustering analysis was that the gene must be differentially expressed at a minimum of one of the 11 time points.

Functional characterization
Groups of genes then were defined based on gene and associated protein function using Ingenuity Pathways Analysis (IPA, Ingenuity Systems, Redwood City, CA, USA, www.ingenuity. com). IPA uses information about gene relationships from the literature to characterize gene sets, create gene networks, and identify important signaling pathways in gene expression data.

Results
Histology H&E-stained sections through the ulnar midshaft were used to count osteoblasts and look for evidence of bone formation (Fig. 2). There were no osteoblasts on the periosteal surface of control ulnae at either 1 or 4 days after loading. Some osteoblasts were present on the periosteal surface of a single loaded bone on 1 day, but no active osteoblasts were observed in any other loaded bones on 1 day. The average Ob.N/B.Pm was 2.22 AE 4.97 for the loaded bones 1 day after loading. The most remarkable results were in loaded bones in the 4-day group. Osteoid was observed on the periosteal surface of loaded bones, which indicated that new bone was being formed. In addition, active osteoblasts were present and the average Ob.N/B.Pm was 33.1 AE 6.6 for loaded bones at 4 days. Osteoclasts were not observed in any of the bone sections.

Total RNA
The total amount of RNA isolated from loaded bones was significantly greater than the total amount of RNA isolated from control bones for all time points except 2 days (Fig. 3). Total RNA increased substantially from 4 to 8 days in the loaded bones, and then the amount declined from 12 to 32 days.
Col1a1 expression qPCR was used to measure Col1a1, a major bone matrix gene, to characterize matrix synthesis. (20)(21)(22) Figure 4 shows that Col1a1 gene expression began to increase in loaded ulnae at 4 days, when active osteoblasts were present on the periosteal bone surface (Fig. 2). Col1a1 expression peaked at 12 days in loaded bones and declined toward baseline levels at later time points. Col1a1 expression was not changed in control ulnae. Since Col1a1 expression corresponded with the observation of active osteoblasts, we presume that osteoblasts were the primary contributors to the qPCR signal for Col1a1. Furthermore, the Col1a1 expression time course was very similar to the time course of osteoblast recruitment and bone formation observed by others. (8)(9)(10) Clustering analysis The primary goal of the clustering analysis was to determine the patterns of gene expression in bones subjected to mechanical loading over a time sequence of 4 hours to 32 days. We identified 1051 genes that were differentially expressed at a minimum of one of the 11 time points. This gene list was analyzed using k-means clustering. Six distinct patterns of differential gene expression were identified over the time course (Fig. 5): an earlyresponse cluster (blue line) in which the genes were upregulated early but not late in the time course, three matrix-formation (up) clusters (red lines) that followed the pattern of matrix synthesis illustrated by expression of the Col1a1 gene (Fig. 4), and two matrix-formation (down) clusters (green lines) that were downregulated during matrix formation. There were 88 genes in the early-response cluster, and genes in this cluster were upregulated early on, primarily at 4 hours. There were 23 genes in the matrix-formation (up) high-magnitude cluster, 182 genes in the matrix-formation (up) medium-magnitude cluster, and 308 genes in the matrix-formation (up) low-magnitude cluster. Genes in these three clusters were upregulated during matrix formation, and expression reached peak levels between 12 and 16 days. There were 124 genes in the matrix-formation (down) high-magnitude cluster and 326 genes in the matrix-formation (down) low-magnitude cluster. Genes in these two clusters were downregulated during matrix formation, and differential expression was greatest at 16 days. Within each cluster of genes, we identified groups of genes that have common functions or are part of common pathways using IPA. The complete list of 1051 genes is available in Supplemental Table S1.
Early-response cluster Selected gene groups from the early-response cluster are listed in Table 1 and included AP-1, calcium signaling, chemokine, cytokine, and matrix. Of these, the chemokine group that was upregulated within 12 hours of loading was novel for bone formation. Genes in the chemokine group included a binding

Matrix-formation (up) clusters
Expression of genes in the matrix-formation (up) clusters peaked during the synthetic phase of bone formation. Several important gene groups were identified and include apoptosis, calcium signaling, cytokine, growth factor, ion channel, matrix, muscle, neurotransmitter, solute carrier, transforming growth factor b (TGF-b) signaling, and Wnt/b-catenin signaling ( Table 2). The matrix group included many of the expected genes associated with bone formation, such as Alpl, Bglap, and Col1a2. Probes targeting the Col1a1 gene were not detected on the exon array and therefore Col1a1 is not reported in Table 2, but using qPCR, we found that its expression followed a similar pattern (Fig. 4). The solute carrier group represents a novel finding with respect to bone mechanotransduction. This group includes transporters for amino acids and many ions. Presumably, such transport is necessary to facilitate the synthetic activity of osteoblasts.

Matrix-formation (down) clusters
Genes in the matrix-formation (down) clusters were downregulated during the synthetic phase of bone formation. Several important gene groups were identified in these clusters and include calcium signaling, cell cycle, chemokine, cytokine, growth factor, ion channel, matrix, neurotransmitter, TGF-b signaling, ubiquitin, and inhibitors of Wnt/b-catenin signaling ( Table 3). In particular, there were large groups of muscle and solute carrier genes that were downregulated. Another important finding in this cluster was that genes encoding inhibitors of both TGF-b (Grem1 and Chrdl1) and Wnt/b-catenin signaling (Sost and Sfrp4) were downregulated, supporting observations that negative regulation of inhibitors was important to enhance signaling through these pathways. (21)

Discussion
We have identified six distinct patterns of gene expression, and these were categorized into three primary clusters of genes: Genes in the early-response cluster (n ¼ 88) were upregulated primarily at 4 hours. Expression of genes in the matrix-formation (up) clusters peaked between 12 and 16 days, and this cluster was subdivided into three magnitudes: high (n ¼ 23), medium (n ¼ 182), and low (n ¼ 308). Expression of genes in the matrix-formation (down) clusters peaked at 16 days, and this cluster was subdivided into two magnitudes: high (n ¼ 124) and low (n ¼ 326). Annexin A2 S100a4 S100 calcium-binding protein A4 S100a10 S100 calcium-binding protein A10 Chemokine Ccbp2 Chemokine-binding protein 2 Ccl2 Chemokine (C-C motif) ligand 2 Ccl7 Chemokine (C-C motif) ligand 7 Cxcl1 Chemokine (C-X-C motif) ligand 1 Cxcl13 Chemokine (C-X-C motif) ligand 13 Cytokine Il1rl1 Interleukin 1 receptor-like 1 Il1rl2 Interleukin 1 receptor-like 2 Osm Oncostatin M Osmr Oncostatin M receptor Socs3 Suppressor of cytokine signaling 3 Stat3 Signal transducer and activator of transcription 3 Tnfrsf12a Tumor necrosis factor receptor superfamily, member 12A Matrix Adamts1 ADAM metallopeptidase with thrombospondin type 1 motif Ecm1 Extracellular matrix protein 1 Serpina3n Serine (or cysteine) peptidase inhibitor, clade A, member 3N Serpine1 Serpin peptidase inhibitor, clade E, member 1 Tfpi2 Tissue factor pathway inhibitor 2 Other Ccnd2 Cyclin d2 Clic1 Chloride intracellular channel 1 Gpr1 G protein-coupled receptor 1 Kcne4 Potassium voltage-gated channel, Isk-related subfamily, gene 4 Lep Leptin Sdc4 Syndecan 4  early-response, matrix-formation (up), and matrix-formation (down). Genes within these clusters were grouped based on function and/or signaling pathways using IPA. Many gene groups known to be important in loading-induced bone formation were identified, as were several novel gene groups whose function with respect to bone formation is not known.

Early-response cluster
The early-response cluster was characterized primarily by genes that showed increased expression 4 hours after a single loading session. It is known that gene expression of components in the AP-1 transcription factor complex increases with mechanical loading. Fosl1 and Junb genes were both upregulated shortly after loading in our study and those of others, (23,24) and Fosb gene expression increased after loading and may be associated with osteoblast differentiation and bone formation. (25) Others have observed increased protein expression of FOS and JUN family members after loading in association with osteoblast proliferation. (26) JUND is one of the primary AP-1 signals when osteoblasts are fully differentiated during matrix production and mineralization, and JUND inhibits the S phase of the cell cycle. (26) Our results are consistent with these findings because JunD gene expression increased during the matrix-formation phase, which corresponded to the observation of differentiated osteoblasts on the bone surface. A group of chemokines was identified in the early-response cluster and presents a potentially novel finding. Chemokines act as chemotactic molecules for a number of different cell types. Chemokines are known to be important in many biologic functions, including development, immune function, and wound repair. (27) Many cytokines and chemokines that function in the immune system regulate osteoblasts and osteoclasts, (28) and osteoblasts are known to produce both chemokines and their receptors. (29) The chemokine-related genes identified in our study included a binding protein (Ccbp2), C-C motif ligands (Ccl2 and Ccl7), and C-X-C motif ligands (Cxcl1 and Cxcl13). CXCL13 upregulates alkaline phosphatase activity and also induces release of b-N-acetylhexosaminidase, which is involved in bone remodeling. (30) There also was a group of chemokine-related genes that were downregulated during matrix formation, which included a C-C motif ligand (Ccl11), C-C motif receptors (Ccr1, Ccr2, and Ccr3), a C-X-C motif ligand (Cxcl12), and a C-X-C motif receptor (Cxcr5). Ligands that bind CCR1 are chemoattractants that promote recruitment of marrow cells that can develop into osteoclasts. (31) CCR2 is expressed on osteoclasts, (32) and CCR2 knockout mice have high bone mass, impaired osteoclast function, and are resistant to ovariectomy-induced bone loss. (33) The receptor for CXCL13, CXCR5, is expressed in osteoblasts. (30) We postulate that chemokines in the early-response cluster may be active shortly after a mechanical loading event to recruit osteoblasts and/or osteoblast precursors to a site of bone formation. However, later in the synthetic phase, additional osteoblasts are not necessary, so chemokines may be downregulated. It is also possible that differential expression of chemokines is related to the interaction of the bone cells and immune cells. CXCL12 plays a role in fetal bone marrow colonization, (34,35) and interaction of CXCL12 and its receptor, CXCR4, is necessary for retention of hematopoietic stem cells (HSCs) in adult bone marrow. (36,37) CXCL12/CXCR4 signaling modulates bone resorption, (38) osteoblast proliferation, (39) and priming of hematopoietic progenitor cells. (40,41) In addition, CXCL13/CXCR5 signaling is known to affect several aspects of Bcell functionality. (42,43) A group of cytokines was present in the early-response cluster as well. In bone, cytokines are usually associated with osteoclasts and increased bone resorption. However, osteoclasts were not observed in any of the histologic sections we analyzed. Oncostatin M (OSM) is a cytokine that was shown recently to inhibit Sost gene expression in primary calvarial osteoblasts, and OSM treatment of calvariae in vivo enhanced bone formation. (44) In our study, Osm expression was highest at 4 hours and could have contributed to decreased Sost expression during the synthetic phase.

Matrix-formation (up) and matrix-formation (down) clusters
Most genes exhibited differential expression during the synthetic phase of bone formation and belonged to the matrix-formation (up) and matrix-formation (down) clusters. The matrix-formation (up) clusters were comprised of three groups that exhibited different degrees of upregulation in response to loading, and the matrix-formation (down) clusters were comprised of two groups that were downregulated in response to loading. Matrix genes dominated the upregulated clusters, whereas muscle-related genes dominated the downregulated clusters. Several gene groups were expressed in both up-and downregulated clusters, including solute carriers, Wnt/b-catenin signaling, and TGF-b signaling.
The largest group of upregulated genes identified in the synthetic phase was associated with extracellular matrix and had 31 members, including proteoglycans, collagens, genes related to collagen synthesis, osteoblast markers, matrix-related proteins, and matrix metallopeptidases (MMPs). These groups of genes are well known to be upregulated with mechanical loading and bone formation. Probes targeting the Col1a1 gene were not detected on the exon array platform, so Col1a1 expression was measured using qPCR. Col1a1 expression in loaded ulnae followed a similar pattern as the genes in the matrix-formation (up) clusters. A few matrix genes were part of the matrix-formation (up) high-magnitude cluster, which had 23 total members and reached peak upregulation of approximately 4-fold at 12 days. Lox is important for collagen cross-linking, and Cthrc1 and Lepre1 are associated with collagen synthesis. In addition, this cluster included two proteoglycans, Acan and Vcan; a metallopeptidase inhibitor, Timp1; and two minor collagens, Col5a1 and Col11a2.
The second most abundant gene group in the matrixformation (up) clusters was comprised of solute carriers, which have not been previously associated with mechanical loading. Twenty solute carriers were upregulated by loading, and examples of the types of molecules transported include amino acids, glucose, and various ions. In addition, 15 solute carriers were downregulated during matrix formation. Solute carriers may be involved in supporting matrix synthesis by adjusting ion L Chemokine (C-C motif) receptor 1 Ccr2 L Chemokine (C-C motif) receptor 2 Ccr3 L Chemokine (C-C motif) receptor 3 Cxcl12 L Chemokine (C-X-C motif) ligand 12 Cxcl14 H Chemokine (C-X-C motif) ligand 14 Cxcr5 L Chemokine (C-X-C motif) receptor 5 L Radical S-adenosyl methionine domain containing 2 levels and bringing amino acids and other necessary ingredients into the osteoblasts to facilitate matrix protein production. In addition to gene groups, we identified signaling pathways important in loading-induced bone formation. The Wnt/bcatenin signaling pathway is known to play an integral role in mechanotransduction and enhances the sensitivity of osteoblasts and osteocytes to loading. (45) The matrix-formation (up) clusters included several genes from the canonical Wnt pathway and also included two inhibitors, Kremen1 and Wif1. Additionally, two Wnt inhibitors, Sost and Sfrp4, were downregulated during matrix formation. It was shown recently that Sost gene expression decreased after in vivo bone loading and increased with hind limb unloading. (15) Thus, decreased expression of Sost is associated with increased Wnt signaling and bone formation. SFRP4 has not been as well studied. The Dkk1 gene encodes another Wnt antagonist that has been shown previously to decrease with loading, (15) but probes targeting the Dkk1 gene were not detected on the exon array platform used in our study. We also observed increased Osm expression at early time points, and this has been shown to contribute to decreased Sost expression in bone. (34) LRP5 is a Wnt coreceptor, and Wnt signaling through LRP5 is required for mechanically induced bone formation. (45) In addition, it has been suggested that LRP5 may inhibit gut-derived serotonin synthesis, providing an endocrine regulatory mechanism to increase bone mass. (46) Muscle genes were the largest group of downregulated genes during matrix formation. This group was comprised mainly of genes encoding muscle structural components and included members of the actin, myosin, and troponin gene families. It is important to note that all muscle was stripped from the ulnae during dissection, so RNA was isolated from bone and marrow only with no contamination from muscle. Therefore, these genes were expressed in cells within the bone. Others have identified muscle genes that were differentially expressed between osteoblasts and osteocytes. (47) Interestingly, 36 muscle-related genes were downregulated in osteoblasts with respect to osteocytes, including many of the same genes that were downregulated with loading in our study (Acta1, Dmd, Myocd, Myl1, Myplf, Tnni2, Tnnt3, and Tpm2). (47) We speculate that our finding that muscle-related genes are downregulated during bone formation is congruent with the finding that musclerelated genes are downregulated in osteoblasts compared with osteocytes. Although the involvement of muscle-related genes and proteins in bone biology is not well understood, it is clear that they are highly regulated in bone cells.
The TGF-b pathway is important in loading-induced bone formation, (48,49) and many genes in the TGF-b superfamily were upregulated during matrix formation. Two TGF-b superfamily inhibitor genes (Chrdl1 and Grem1) were downregulated. Overall, it seems that signaling through the TGF-b pathway was enhanced mostly by increasing expression of TGF-b inducers and decreasing expression of TGF-b inhibitors of bone formation.
A limitation of our approach is that because we isolated RNA from the entire ulnar shaft, we cannot discern regional changes in bone formation in response to loading. It is well documented that mechanical loading elicits a regional bone-formation response in the ulna loading model. (4,14,50) Most of the bone forms in the ulna midshaft on the periosteal surface, and greater bone formation is observed on the medial surface than on the lateral surface. (14) In addition, since bone is deposited primarily on the medial and lateral periosteal surfaces in the ulnar midshaft, the geometric properties I max and I min change with loading and are location-dependent. I min increased significantly in the ulnar midshaft, whereas I max was moderately increased with loading distally. (4) In addition, we have completed a computational model that was programmed to form bone in regions that had greatest strain energy density. (50) This computational model almost perfectly mimicked the locations of new bone formation measured in the ulna, thus showing that there is a predictable pattern of bone formation in response to loading.
In conclusion, we determined the time sequence of gene expression in a bone subjected to mechanical loading, identified groups of genes that shared a time-dependent gene expression pattern, and determined which functions or pathways the genes had in common. We identified six clusters of genes that exhibited unique expression patterns in response to loading over a time course that captured key periods of the bone-formation process, including expression of matrix-related genes, the appearance of active osteoblasts, and bone desensitization. Many gene groups known to be important in loading-induced bone formation were identified within the clusters, including matrix, Wnt/b-catenin, and TGF-b. Several novel gene groups were identified as well, including chemokine, solute carrier, and muscle, whose functions with respect to bone formation are not known.

Disclosures
All the authors state that they have no conflicts of interest.