The transcriptome of fracture healing defines mechanisms of coordination of skeletal and vascular development during endochondral bone formation

Authors


Abstract

Fractures initiate one round of endochondral bone formation in which callus cells differentiate in a synchronous manner that temporally phenocopies the spatial variation of endochondral development of a growth plate. During fracture healing C57BL/6J (B6) mice initiate chondrogenesis earlier and develop more cartilage than bone, whereas C3H/HeJ (C3H) mice initiate osteogenesis earlier and develop more bone than cartilage. Comparison of the transcriptomes of fracture healing in these strains of mice identified the genes that showed differences in timing and quantitative expression and encode for the variations in endochondral bone development of the two mouse strains. The complement of strain-dependent differences in gene expression was specifically associated with ontologies related to both skeletal and vascular formation. Moreover, the differences in gene expression associated with vascular tissue formation during fracture healing were correlated with the underlying differences in development and function of the cardiovascular systems of these two strains of mice. Significant differences in gene expression associated with bone morphogenetic protein/transforming growth factor β (BMP/TGF-β) signal-transduction pathways were identified between the two strains, and a network of differentially expressed genes specific to the MAP kinase cascade was further defined as a subset of the genes of the BMP/TGF-β pathways. Other signal-transduction pathways that showed significant strain-specific differences in gene expression included the RXR/PPAR and G protein–related pathways. These data identify how bone and vascular regeneration are coordinated through expression of common sets of transcription and morphogenetic factors and suggest that there is heritable linkage between vascular and skeletal tissue development during postnatal regeneration. © 2011 American Society for Bone and Mineral Research

Introduction

Fractures are among the most common traumatic injuries in humans, and osteoporosis-related fractures are the fastest-growing health care problem of aging.1 Fracture healing is intimately dependent on vascular tissue formation, and inadequate blood supply is a major cause of delayed bony union, nonunion, and complications in many different postoperative orthopedic treatments. Studies of human tibia fractures have shown impaired rates of healing as high as 46% when the patient had concomitant vascular injuries.2 Development of lower limb ischemia in the context of fracture healing in animal models has shown delayed union,3 and there is an age-dependent delay in fracture healing that is associated with diminished vascularization in the healing tissue.4 While inadequate vascular tissue development during tissue repair and regeneration is often thought of in terms of a lack of appropriate nutrition or oxygenation,5, 6 vascular morphogenesis itself provides signals for skeletal morphogenesis and is part of the structural template that is involved in bone morphogenesis.7, 8 Thus cortical bone formation is patterned around the Haversian system, and trabecular bone formation is patterned around the vascular structures that infiltrate the empty lacunae left after chondrocyte apoptosis during endochondral bone formation. These phenomena suggest that vascular and bone tissue development share many common morphogenetic signals and environmental cues and are mechanistically coordinated during development.

It has come to be believed generally that fracture healing is a regenerative process that bones undergo as a consequence of injury because it recapitulates the mechanisms of endochondral bone formation seen during development.9–12 Fracture healing therefore represents a unique postnatal window into the biologic and molecular processes that are involved in bone development. It may be hypothesized that studies of fracture healing within different genetic backgrounds would provide a means of identifying heritable biologic and molecular processes that lead to the variations in skeletal phenotypes. Our previous studies showed that C57BL/6J (B6) and C3H/HeJ (C3H) mice recapitulate their strain-specific skeletal architectural features during the development of fracture callus. The B6 strain, which healed the fastest, initiated cartilage gene expression the earliest and had the longest period of chondrocyte maturation and hypertrophy. In contrast, the slower-healing C3H strain initiated cartilage development later than the B6 strains, had the shortest period of chondrogenic development, and showed an earlier initiation of osteogenic development. Our previous studies also have shown that these variations in endochondral bone formation mirror those seen in both the secondary growth centers and growth plates of the two strains of mice, thereby demonstrating that these aspects of endochondral bone formation are genetically inherited.13 In this investigation, a comparative temporal analysis of the transcriptomes of fracture healing in B6 and C3H strains of mice is carried out in order to identify the heritable differences in expressed genes that lead to the variations in endochondral bone formation and bone repair.

Materials and Methods

Fracture model, RNA preparation, and bone tissue assessment

Animal research was conducted in conformity with an Institutional Animal Care and Use Committee–approved protocol. C57BL/6J (B6) and C3H/HeJ (C3H) mice were purchased from Jackson Laboratories (Bar Harbor, ME, USA). All fracture studies were performed on 8- to 10-week-old male B6 and C3H mice, as described previously.13 Total RNA samples were assessed in a duplicate set of pooled samples of three to four fractured calluses harvested on days 0 (no fracture), 3, 5, 10, 14, and 21 days. Total RNA was extracted and prepared for microarray analysis as described previously.14 Growth-plate and whole-bone specimens were from 28-day-old male mice obtained from breeding stocks maintained at the Mount Sinai School of Medicine (New York, NY, USA). Histologic assessments were carried out on 10- and 14-day postfracture calluses and 28-day-old proximal femur epiphyseal growth plates as described previously.13, 15

Transcriptional profiling and analysis of gene expression data

The technical approaches that were used in these studies are as described previously in Bais and colleagues12 and are summarized in the supplemental materials. The Cluster Analysis of Gene Expression Dynamics (CAGED) program was used to analyze the gene expression data using the day 0 unfractured values as the reference group (http://genomethods.org/caged/).16 The default filter settings of the program were used, which filtered out all probes within twofold distance from the referent group in order to limit insignificant effects from distorting the cluster graphs.17 The values for each gene's expression for the C3H and B6 strains were each denoted separately, thereby allowing a gene's expression to be assigned to a given temporal cluster in a strain-specific manner. We used polynomial models of degree 4 for modeling the temporal expression profiles, which are better suited to describe short patterns.17 After the cluster analysis was completed, the clusters were further sorted into various temporal and strain-specific categories as described below.

Two different approaches then were used to assess the biologic nature of the various gene categories. In one approach, the genes were assessed using the DAVID Bioinformatics resource (http://david.abcc.ncifcrf.gov/home.jsp) described in Hosack and colleagues and Huang and colleagues.18, 19 Only genes that were assigned to biologic ontologies having an enrichment score p ≤ 0.05 were considered in this analysis. Ontologies with overlapping gene sets were aggregated into larger groups with similar biologic or molecular functions, as described previously.12 This approach was used to develop sets of genes associated with four individual tissue types (ie, skeletal, vascular, neural, and muscular) and biologic processes. The second approach used the Ingenuity Pathway Analysis software program and the Ingenuity Knowledge Base (www.ingenuity.com/). For these analyses, specific categories of genes were compared. Categories were established based on either quantitative differences in expression, qualitative differences in temporal patterns of expression, tissue-specific association, and association with specific canonical signal-transduction processes. For tissue-specific categories, we used the tissue gene sets developed with DAVID, and for signal-transduction processes, we used sets defined by the Ingenuity Knowledge Base. Comparisons were run using the log fold ratios obtained from the CAGED analysis and were uploaded along with gene identifiers into the Ingenuity Systems application to uncover networks of biologically related genes, connections between genes, and Ingenuity-defined “canonical pathways” that were significant. Only networks significantly associated with molecular or biologic functions or associated with a significantly enriched canonical pathway (p ≤ 0.05, Fisher's exact test) were considered.

RNA isolation and quantitative real-time RT-PCR

Quantitative real-time RT-PCR analysis of specific candidates was as described previously.14 Analysis of mRNA expression was carried out on replicate pools of mRNA, and individual assessments were done three times. The expression values of target genes were normalized to the specific controls, as denoted in each figure.

Cardiac and vascular tissue assessments

Heart tissues from 10- to 12-week-old C3H and B6 mice were used for cardiac tissue histologic analysis. Messenger RNA expression in early phases of heart development was assessed in hearts derived from 7-, 14-, and 28-day-old mice. For histologic assessment of the cardiac muscle, the sections from the hearts of B6 and C3H mice were fixed, embedded, and sectioned in identical sagittal planes such that all four chambers were visualized. Measurements of the hearts were determined using Image Pro Plus software (Media Cybernetics, Silver Spring, MD, USA).

Results

The transcriptome of endochondral bone formation during fracture healing

Since a fracture will initiate one round of endochondral bone formation in which the cells in the callus differentiate in a relatively synchronous manner, the temporal changes across the time course of fracture healing will phenocopy the spatial/temporal variation of the cell zones from the top to the bottom of the growth plate that define endochondral development. An overview of the phenotypic differences in endochondral bone development and structure between C3H and B6 mice is presented by a qualitative histologic comparison of selected time points of the fracture callus and the distal femur growth plate (Fig. 1AD). These comparisons show that the B6 strain initiates chondrogenesis earlier and that it lasts longer. This is exemplified by longer growth plates or the production of a larger callus that is composed of more cartilage relative to marrow and bone. Conversely, in the C3H mice, osteogenesis is initiated earlier and lasts longer. These mice have much shorter growth plates and smaller calluses that are composed of lesser amounts of cartilage and greater amounts of bone relative to the marrow and cartilage. A schematic representation of the differences in endochondral bone formation is presented in Fig. 1E. A summary of the variations in the B6 and C3H bone phenomes is presented in the supplemental materials, whereas the quantitative variations in callus structural and tissue compositions are covered extensively in Jepsen and colleagues.13

Figure 1.

Comparison of the developmental differences of the endochondral bone formation in B6 and C3H mice. (A) Histologic assessment of callus tissues at 10, 14, and 21 days after fracture. All sections were stained with Safranin-O/Fast Green, and the legend denotes the genetic strain. The arrows in these sections illustrate the differences in the cross-sectional areas of each callus. Sections taken from 21-day postfracture calluses are presented to highlight regions of fibrous tissues seen in C3H tissues. Small arrows denote the rim of newly formed cortical bone bridging the outer surface of each callus. The large arrow indicates regions of fibrous tissue that have not developed into bone (×100). (B) Variation in postnatal endochondral bone formation of 28-day-old growth plates of two strains of mice seen in distal femurs. Arrows denote differences in the overall cross-sectional area of the femur relative to the width of the growth plates (×25). (C) Comparable regions of centrally located portions of the distal epiphyseal growth plates of 28-day-old mice. Arrows illustrate the differences in the length of the growth plates in each strain (×100). (D) Comparable regions of hypertrophic chondrocytes seen in day 14 postfracture B6 callus tissues, and day 10 postfracture C3H callus tissues (C) are compared with matched areas in the growth plates of 28-day-old mice (GP) (×200). Difference in cell size is highlighted by the circles around representative cells in each section. (E) Diagrammatic representation of the temporal stages and differences in murine fracture healing in C3H and B6 mice as reported previously by Jepsen and colleagues.13 (F) Representative temporal cluster patterns. (G) Validation of cluster analysis assignment of temporal and quantitative groupings based on comparison of qRT-PCR measurements with graphic appearance of the clusters to which the genes had been assigned. Top panels show the mean expression profiles of the strain-specific clusters in which each candidate gene was contained. The scale is based on a log ratio compared with unfractured bone. The bottom graphs for each panel show the mRNA expression profiles as assessed by qRT-PCR (solid line = B6; dashed lines = C3H). The identity of each gene is denoted in the figure.

The basic differences in the tissue-level developmental patterns of endochondral bone formation that were seen between the two strains in Fig. 1A–D were translated to the level of gene expression by carrying out a genome-wide gene-expression profiling study. An approach in which the genes are arranged into clusters defined by differences in the timing, magnitude, and direction of change in their expression was used for this study. Representative examples of the six types of basic temporal patterns of gene expression (ie, early, middle, late, biphasic, continuous, and down) that the clustering approach defined are depicted in Fig. 1F. The patterns of all 87 clusters and how they are arranged into these basic temporal groupings and the biologic and molecular functions that are associated with each of the temporal groupings are summarized in the supplemental materials. Since during the clustering process the expression of each gene is assigned to either the B6 or C3H strain, this approach allows us to identify the genes that track across fracture healing within the two strains in either comparable or differing temporal and quantitative manners. Thus, by identifying a given cluster in which a gene tracks in one strain compared with the cluster in which it tracks in the other strain, genes were sorted by strain into categories of genes that have similar or differing temporal and quantitative expression profiles.

The strain-specific cluster patterns of four candidate genes into which the gene was assigned is compared with the temporal pattern of expression that was obtained when the gene's expression was assayed by qRT-PCR. We chose as our candidates well-known central transcription (osterix and Sox9) and morphogenetic factors (PTHrP and IIH) that control the progression of endochondral bone formation. The top of each panel presents the clusters in which these genes were separately sorted in the B6 and C3H strains, and the bottom of each panel presents the qRT-PCR expression profile of the individual gene's expression. It is important to note in examining these comparisons that the array cluster patterns in which a gene appears represent the average of the hundreds of genes that were sorted into that cluster and does represent the individual gene's temporal expression pattern. This set of comparisons is presented as an example of how the genes were sorted into temporal and quantitative categories and provides a validation that the clustering approach accurately sorts the genes into patterns of expression that reflect the underlying biologic differences reflective of the mechanisms that control endochondral bone formation (Fig. 1G).

The analysis of Sox9, the master transcriptional regulator of chondrogenesis,20 is compared with osterix (SP7), one of the master transcriptional regulators of osteogenesis.21 The expression patterns of these two genes are shown because they represent categories in which the temporal patterns of the genes are shifted to an earlier time in one strain or the other, and they accurately reflect differences in gene expression that can be related to the biologic differences in endochondral bone formation. As can be seen, Sox9 expression is initiated earlier and has a broader temporal profile of induced expression in the B6 strain relative to the C3H strain, and this strain of mouse develops more cartilage than bone. In contrast, osterix shows an earlier peak in its expression in the C3H strain, and this mouse develops more bone than cartilage. We use the morphogenetic proteins parathyroid hormone–related peptide (PTHrP) and Indian hedgehog (IHH), which are known to coordinate and control the tempo and extent of the chondrocyte and osteogenic differentiation, respectively,22 as examples to validate that the clustering method appropriately presents quantitative differences that match up with biologically known functions for these genes. In this case, PTHrP in the B6 strain is found in a cluster that shows much higher level of expression than in the C3H strain, but in this case, both clusters show comparable temporal patterns. In contrast, IHH is localized in both a cluster that is more highly expressed and shows a narrower peak of expression for the C3H strain relative to the B6 strain.

Strain-dependent differences in cardiovascular development are phenocopied during peripheral vascular regeneration in fracture healing

We next used comparisons of the clusters into which genes were sorted as well as differences in gene quantitative expression to identify the global set of genes that were differentially expressed within the callus tissues that were strain-specific. Three categories of gene expression based on variations in quantitative expression or qualitative temporal or directional (up in one strain and down in the other strain) variation of expression were generated by comparison of the clustering analyses. It is these differences in gene expression that encode for the transcriptomic changes that contribute to the phenotypic differences in endochondral bone formation between the B6 and C3H strains (Table 1). The categories associated with genes that showed differential quantitative expression in one strain compared with the other strain were based on whether the gene met the default filtering criteria of the clustering analysis of twofold distance from its reference group (unfractured bone) in one strain and not the other. This means that if the gene showed a greater than twofold difference in expression from the unfractured bone in both strains, it was considered to be commonly expressed during fracture healing. If the gene was expressed twofold higher than its reference in one strain but did not show this difference in the other strain, then it would be considered to be uniquely expressed in the one strain over the other strain.

Table 1. Percentage Expressed Genes by Distribution in Ontology and Differential Expressiona
Composite ontologyCommonly expressed categoryUniquely expressed categoryb
Up 1748dDown 1578B6 variable,c 617 earlier, 286 invertedeC3H variable, 378 earlier, 114 invertedB6 up 4933B6 down 847C3H up 308C3H down 869
  • a

    Total number of genes associated with a given composite ontology is based on the manual annotation of genes associated with biologic processes that have common functions that were identified in DAVID and meeting p < 0.05 by Fisher's exact test.

  • b

    Selection as quantitatively unique based on CAGED analysis of expressed genes showing twofold or greater baseline unfractured bones.

  • c

    Manual annotation based on comparisons of the graphic appearance of the clusters in which a given gene was assigned.

  • d

    Number of expressed genes in each category.

  • e

    Strain assignment for a gene shown to have inverted expression was based on having an upregulated value in that strain.

    fThe percentage of expressed genes in a single composite ontology showing differential temporal or quantitative expression in the horizontal rows.

    g(Values) The percentage of expressed genes in the individual composite ontology groups for a given differential temporal or quantitative category in a vertical column. Genes that appear in more than one ontology were given fractional numeric values when assessing percentages.

Skeletogenesis48%f(4%)g28%(7%)9%(7%)12%(1%)3%(7%)
Vasculogenesis32%(3%)18%(5%)1%(2%)41%(2%)7%(5%)1%(3%)
Neurogenesis29%(5%)15%(10%)9%(17%)37%(4%)3%(5%)5%(20%)2%(1%)
Myogenesis31%(2%)8%(2%)54%(2%)7%(4%)
Signaling41%(14%)9%(10%)10%(35%)26%(5%)4%(10%)1%(10%)8%(8%)
Ion transport29%(9%)11%(12%)58%(10%)3%(7%)
Proliferation21%(4%)33%(10%)7%(5%)3%(7%)23%(3%)1%(2%)12%(8%)
Immune response28%(4%)29%(6%)4%(2%)3%(4%)20%(2%)2%(2%)15%(6%)
Motility23%(13%)14%(12%)8%(15%)1%(5%)44%(13%)1%(13%)9%(16%)
Development5%(2%)35%(22%)2%(3%)3%(13%)42%(9%)10%(33%)2%(20%)1%(2%)
Metabolism11%(9%)24%(31%)1%(2%)1%(8%)48%(22%)1%(5%)14%(37%)
Catabolism40%(4%)17%(3%)1%(2%)25%(1%)3%(3%)14%(5%)
Transcription7%(2%)9%(5%)<1%(1%)74%(13%) 9%(9%)
Apoptosis30%(3%)2%(<1%)47%(3%)8%(7%)13%(4%)
Stress response30%(5%)28%(7%)5%(3%)29%(3%)2%(3%)3%(12%)2%(1%)
Membrane secretions22%(3%)11%(2%)5%(2%)56%(4%)5%(2%)
Cell adhesion molecules69%(4%)5%(1%)17%(4%)8%(13%)
Collagen formation100%(1%)
Proteolysis69%(1%)31%(<1%)
Regulation33%(7%)3%(1%)23%(16%)<1%(1%)28%(3%)9%(14%)<1%(2%)2%(1%)

The first row in Table 1 summarizes the overall distribution of genes between these two categories, with 85% of the genes that were uniquely expressed in B6 being upregulated and 73% of the genes that were unique to C3H being downregulated. Since a gene's differential expression is based on its expression relative to unfractured bone, the basic variations in the complement of genes that are uniquely expressed in either the B6 or the C3H callus tissues resides in the strain-dependent differences in the overall compositions of the callus tissues relative to the starting bone. As described earlier, and as shown previously,13 B6 fracture calluses are substantially larger and composed of a higher percentage of cartilage tissue relative to the starting bone tissue. The callus tissues therefore will have a larger number of cartilage genes that are expressed in sufficient quantity to reach the twofold cutoff that arbitrarily scores the gene as being differentially expressed. In contrast, the starting marrow space relative to cortical bone is much smaller in C3H mice, and as the callus forms, its composition becomes primarily new bone tissues. Thus, in this strain, hematopoietic and myeloid tissues show a larger numbers of genes that are uniquely downregulated in the callus tissues relative to the starting bone tissues.

We also used qualitative differences in the timing of a gene's expression to identify genes that are functionally related to the differing patterns of endochondral bone development in the two strains of mice. The variable expression category of expressed genes was identified as the subset of genes in the common category that showed temporal or directional differences in expression. It represents genes with expression profiles in one strain that were assigned to a cluster with one temporal pattern and assigned in the other strain to a cluster with a different temporal pattern. Osterix and Sox9 are two such examples, as shown earlier. In the B6 strain, Sox9 is associated with a cluster that shows a pattern of earlier expression than seen for the C3H strain. On the other hand, osterix is expressed in a cluster that shows a pattern of expression that shows that it is expressed at an earlier time after injury in the C3H strain than the cluster to which it was assigned in the B6 strain. The second group of genes included in each of the variable categories contains genes that were upregulated during callus formation in one strain and downregulated in the other strain.

The complement of genes that are associated with the differentially expressed categories defines the biologic and molecular differences of endochondral bone formation between the two strains of mice. In order to deduce what biologic and molecular processes are associated with the differentially expressed genes, the functional gene ontologies into which the genes are associated was determined. Our first analysis was done with Database for Annotation, Visualization and Integrated Discovery (DAVID) software (http://david.abcc.ncifcrf.gov/home.jsp). A summary of the percentage distribution of genes between the various biologic and molecular ontologies both within a category that was differentially expressed in the two strains of mice (vertical columns) and within a given ontology across categories (horizontal rows) is presented in Table 1. As an example, 48% of the genes associated with skeletogenesis show similar temporal and quantitative profiles of expression; however, 40% of the these genes are either expressed in a variable or unique manner in the B6 strain, whereas only 12% of these genes are expressed in a variable or unique manner in the C3H strain. Looking down the very first column of genes that are expressed both quantitatively and qualitatively in a similar fashion in the fracture calluses of both strains, those associated with skeletogenesis represents 4% of all expressed genes, whereas genes associated with signal transduction (13%) and motility (14%) represent the most prevalent types of biologic functions associated with fracture healing. This analysis showed that gene ontologies associated with four different tissues (ie, skeletal, neural, vascular, and muscular) were observed in the fracture callus. The ontologies associated with the biologic functions showing the largest percentage change that were expressed either at an earlier time or at a quantifiably different level in the B6 strain compared with the C3H strain were those which were associated with neurogenesis, vasculogenesis, myogenesis, ion transport, proteolysis, and transcription. As an example, approximately 70% of the genes associated with ion transport and myogenesis showed exclusive unique or temporally earlier expression in the B6 strain, whereas approximately 65% of those associated with vasculogenesis were either uniquely expressed or showed an earlier induction in temporal expression in the B6 strain compared with the C3H strain.

The striking changes in temporal and quantitative expression of genes associated with myogenesis, which would be associated with smooth muscle cells found in vessels and vasculogenesis, led us to examine the strain-dependent differences in the expression of a panel of candidate genes associated with vascular tissue formation within the callus (Fig. 2A), as we had done for the bone-specific genes. Genes for the two primary vascular endothelial growth factor (VEGF) receptor ligands found in bone, placental growth factor (PlGF) and vascular endothelia growth factor c (VEGFc), and two genes that are expressed predominantly by vascular endothelial cells (CD31) and VEGF receptor 2 (Vegfr2) were assayed. In addition, two of the central transcription factors known to regulate the progression of both chondrocyte differentiation23–25 and cardiac morphogenesis, Hand2 and FoxC2,26, 27 were examined. These results all showed that every one of these mRNAs was both expressed earlier and/or showed greater induction in the B6 strain compared with the C3H strain.

Figure 2.

Comparison of the differences in vascular-related gene expression seen during endochondral bone formation and their relationship to systemic cardiovascular development in B6 and C3H mice. (A) Expression profiles of a selection of mRNAs associated with vascular tissue development across the time course of fracture healing. Top panels show the mean expression profiles of the strain-specific clusters in which each candidate gene was contained. The scale is based on a log ratio compared with unfractured bone. The bottom graphs for each panel show the mRNA expression profiles as assessed by qRT-PCR (solid line = B6; dashed lines = C3H). The identity of each gene is denoted in the figure. (B) Gross appearance in size of 8- to 10-week-old hearts from B6 and C3H mice. (C) Histological assessment of heart tissues of 8- to 10-week-old heart tissues. Lines within the heart tissues show the primary measurements taken from each tissue. Graphics below the histological sections show the averaged measurements taken from each heart. (D) Mean weight of 8- to 10-week-old hearts from B6 and C3H mice (N = 8). (E) Variation in the relative expression of three separate developmentally related mRNA within heart tissues at days 14 and 28 postnatally. Values are expressed as fold difference from day 7. Values are from the average of five hearts.

These large differences in the expression of multiple genes associated with peripheral vascular tissue formation in the fracture callus that takes place during fracture healing in the B6 strain relative to the C3H strain led us to hypothesize that these differences would be associated with systemic differences in vascular development. In order to test this hypothesis, we examined the hearts of the two strains of mice (Fig. 2B–D). As can be seen, the gross size of the hearts from C3H mice was smaller by approximately 40%. Histomorphometric analysis showed that the hearts from C3H mice had smaller left ventricles, smaller circumferences, and shorter lengths. In the final panels of this figure, the mRNA expression levels for the FoxC2, Hand2, and Vegfr2 were examined in heart tissues over their first month of postnatal development. Similar quantitative differences were seen for the expression of these genes in the heart tissue as in the callus, with the exception of Hand2, which was expressed initially at a higher level in C3H versus B6 mice but then at lower levels by 28 days. Together these data demonstrate that the expression of these genes during skeletal tissue regeneration is predictive of the systemic genetic variation in cardiac tissue development. A summary of the current phenome data related to growth and morphogenesis of the heart and variations in cardiovascular function of the B6 and C3H strains is presented in the supplemental materials and validated our findings.

The gene sets that were identified from our ontology assessment in Table 1, which were associated with the development of skeletal and vascular tissues, were compared in order to identify genes that would be common to the development of both skeletal and vascular tissues because it is these genes that potentially lead to the coordinated differences in development of vascular and skeletal tissues in the two strains of mice. A selected number of these genes that were both uniquely or variably regulated in the two strains of mice and that are associated with both skeletal and vascular tissues are summarized in Table 2. The selection of this subset of genes for presentation in this table was based on prior candidate experimental analysis in mice or genome-wide association study (GWAS) analysis in humans that showed that they were associated with either cardiovascular or skeletal diseases and/or development of both tissues. A complete listing of all the genes showing overlapping expression between both tissues is presented in the supplemental materials. In addition, we also present in the supplemental materials the construction of two networks of gene interactions that would be observed for the skeletal gene ontology sets that were differentially expressed.

Table 2. Selected Gene Set of Major Candidates With Differential Expression in B6 Versus C3H in Both Vascular and Skeletal Tissues
Clustera B6Cluster C3HUnigene numberNameTissuebCategoryc
  • a

    Clusters in which the gene is localized are denoted separately.

  • b

    Tissue category of genes ontology associations are denoted as n = neural, v = vascular, s = skeletal, and m = muscular.

  • c

    Category is based on characterization as in Table 1: C = common, same temporal pattern; Be = expressed in B6 earlier than in C3H; Ce = expressed earlier in C3H than in B6; Bo = expressed uniquely in B6 only; Co = expressed uniquely in B6 only; I = inverted expression in the two strains of mice

  • d

    Genes that appear more than once in the table represent splice variants.

 17Mm.258589Map3k7s, vI
1120Mm.4406Mmp9s, vCe
1020Mm.256509Msx1dn, v, sCe
 2Mm.256509Msx1n, v, sCo
1144Mm.1752Sox5s, vCe
6118Mm.295194Tbx1n, v, sCe
4320Mm.291928Catnbn, v, sBe
18 Mm.7106Bmpr2s, n, vBo
612Mm.14092Foxc2n, v, sBe
21 Mm.261588Hand2n, v, sBo
211Mm.261588Hand2n, v, sBe
5861Mm.26954Hoxa11s, vBo
21 Mm.258589Map3k7s, vI
18 Mm.256509Msx1n, v, sBo
211Mm.294225Pthr2S, n, vBe
22 Mm.1752Sox5s, vBo
61 Mm.1752Sox5s, vBo
33 Mm.295194Tbx1n, v, sBo
18 Mm.85544Tcfap2an, v, sBo
61 Mm.85544Tcfap2an, v, sBo
25 Mm.18213Tgfb2n, v, sBo
43 Mm.6458TimelesssnvBo
2120Mm.10153Twsg1n, sBe

Strain-specific differences in gene expression was coordinated between various canonical signaling pathways that control morphogenetic and metabolic growth

We used the Ingenuity Pathway Analysis software program (Ingenuity Systems, Inc. Redwood City, CA, USA) to identify the canonical pathways that had significantly expressed numbers of genes in each strain of mice for the common, unique, and variably expressed categories. The identified pathways were arranged into groups associated either with canonically defined functions specific to a cell or tissue type and to groups of related canonical signal-transduction processes (Fig. 3). Two of the cell- or tissue-specific groups that showed statistical differences in fracture healing between the two strains of mice were related to innate (macrophage) and adaptive (T and B cell) immune functions of fracture healing and were associated with differences in gene expression in C3H mice. The other two groups of cell- or tissue-related canonical pathways that showed alterations in gene expression were associated with embryonic stem cells and the development of cardiac tissues, and these were shown to be representative of differences in gene expression in the B6 mice.

Figure 3.

Canonical pathways that show temporally shifted or unique quantitative expression between the B6 and C3H strains across the time course of fracture healing. Bar graphs generated by Ingenuity Pathway Analysis software (IPA) analysis summarize the various canonical pathways associated with each of the temporal groupings. Red line indicates the p < 0.05 cutoff value for Fisher's exact test for numbers of genes needed to meet significance per indicated pathway. No disease or toxicologic functions were included for this analysis, and only the pathways showing significance for at least one of the three categories of gene groups are presented in each of the figures. The legends indicate the associated temporal grouping with which each pathway is associated.

A comparison of the canonical signal-transduction pathways showed specific alterations in gene expression associated with nitrous oxide (NO), rennin-angiotensin, adrenergic, and relaxin pathways, which are all commonly linked through their increased expression of Gαs in the B6 strain. Alterations in expression of genes associated with these pathways are consistent with phenotypic characteristics of the higher blood pressures and larger hearts that are seen in the B6 strain. The second large functional group of canonical pathways that showed significant numbers of expressed genes with variation in expression in the B6 strain compared with the C3H strain was the steroid superfamily of interacting factors, which included retinoid X receptor (RXR), vitamin D receptor (VDR), thyroid hormone receptor (TR), and their associated interactions with peroxisome proliferator-activated receptor (PPAR) canonical signaling pathways. The common difference that was seen across these pathways was increased expression of RXRα in the B6 strain of mice relative to its levels of expression in the C3H strain. The final group of canonical pathways that showed significant numbers of genes that were differentially expressed between the B6 and C3H strains was associated with the bone morphogenetic protein/transforming growth factor β (BMP/TGF-β) superfamily of interacting factors that carried out signal transduction through the various TGF-β and BMP receptors. Multiple genes that are linked to the function of these pathways showed selective increased or decreased expression in the B6 and C3H strains. It is also interesting to note that BMP signaling processes were central elements in tissue- and cell-related canonical pathways associated with stem cells and cardiac tissue development. An overview of all the canonical signal-transduction pathways that had significantly different levels of gene expression between the strains of mice identified two broad categories of growth regulation that were altered between the two strains: (1) local morphogenetic regulation associated with BMPs, TGF-β, and Wnt signaling and (2) systemic metabolic regulation associated with the RXR, TR, VDR and PPAR steroid superfamily and G protein–coupled signaling, which is neural/endocrine in nature.

Because of the prominence of the BMP/TGF-β signaling pathways to both vascular and skeletal tissue development, we focused specifically on this pathway. A complete summary of all the expressed genes that are associated with both these pathways is presented in the supplemental materials. Figure 4A presents a schematic of this pathway and the variations in expression of specific genes that make up this pathway, and Fig. 4B shows the specific network of gene interactions that was constructed from the subgroup of genes that were differentially regulated in C3H mice and associated with BMP/TGF-β signal-transduction pathways. It is interesting to note that the majority of the individual genes of the network of differentially regulated genes in C3H mice all were part of the MAP kinase signaling cascade. In this regard, TGF-β receptor morphogenetic signaling can mediate many of its actions through p38/JNK arms of the MAP kinase cascade through Tak1 expression, which was specifically upregulated in B6 mice while showing downregulated expression in C3H mice. On the other hand, aspects of the differential regulation also were seen that was mediated through the actions of the ERK arm of MAP kinase cascade through SOS1/2, which also was variably regulated in the two strains of mice. At the transcriptional level, Smad2/3, Smad6, Smad7, and Smad4 all showed increased levels of expression in B6 mice, whereas in C3H mice, reduced levels of c-Jun were observed.

Figure 4.

Composite representations of the strain specific differences in gene expression of the BMP/TGF-β signal-transduction pathways. (A) The composite pathway was generated by comparison of the mean quantitative expression of each gene that is indicated in the pathways across the time course of fracture healing for the two strains of mice. The nature of each pathway is denoted in the figure and is based on the current pathway descriptions in the Ingenuity Analysis software database. The pathway shown for the BMP/TGF-β canonical signaling shows both sets of receptors for the separate arms of the pathway. The direction (up, red, or down, green) and intensity of the color relate to the mean quantitative change in the expression for individual genes in the pathway and are indicated in the figure. Direction of gene changes for individual strains is denoted with B6 depicted on the left and C3H depicted on the right halves of each gene symbol. Only genes that met the original twofold difference in expression are color coded for a pathway. White indicates genes that were extrapolated to be part of the pathway of interactions. (B) Network of expressed genes that were developed from the subset of expressed genes associated with BMP/TGF-β canonical signaling that were differentially regulated in the C3H mice. Key for the individual symbols in the pathways is indicated in the legend in the figure.

Discussion

Since the central nongenetic factor of development, aging, and many chronic diseases is time, we reasoned that by defining both qualitative differences in timing and quantitative differences in gene expression, we would be able to identify many of the gene expression differences that encode for the developmental differences in endochondral bone formation in the two strains of mice. This approach selectively identified the coordinated and specific variations in expression of many of the currently known transcription and morphogenetic factors associated with inherited deficiencies in skeletal tissues.28 Our data defined how multiple transcription factors and multiple signal-transduction pathways are coordinated and temporally regulated during endochondral bone formation. Modeling of these pathways and the interactions of the groups of genes that were strain-specific allowed us to infer interactions between transcription factors and signal-transduction pathways that coordinate the formation of vascular and skeletal tissues. Finally, while we did not test directly how specific genetic variations in the two strains are linked to actual phenome-level differences, we could identify how differences in gene expression related the known loss of Toll4 activity in the C3H strain affected macrophage- and immune cell–associated gene expression29 and were reflected in the patterns of canonical pathways associated with marcophages and adaptive (T and B cell) immune functions.

A second aspect of our approach involved relating the patterns of dynamic gene expression in the healing fracture callus to tissue-associated gene ontologies. This allowed us to infer how different tissues coordinated their development and contributed to the phenotypic differences in the skeletal organs of these two strains of mice. In particular, these studies showed that strain-dependent differences in cardiovascular development were recapitulated during peripheral vascular regeneration in fracture healing. Furthermore, these studies demonstrated that multiple signal-transduction pathways responsible for systemic cardiovascular function and regulation of blood pressure exhibited coordinated changes in their gene expression that were strain-specific. While these data did not define whether local differences in either vascular or skeletal gene expression were causally related to variations in each other's patterns of development, our previous studies have shown functional linkage in the development of these two tissues.30, 31 While these prior studies show that there are reciprocal regulatory mechanisms between these tissues that coordinate their development, this study suggests that heritable variation in the function of these molecular mediators may influence the development of both tissues.

While many previous studies have used genetic approaches to identify specific loci that are associated with the altered skeletal development in these two strains of mice, this is the first demonstration that alterations in BMP signaling contribute in a major way to skeletal differences between these two strains. The demonstration that the variations in expressed components of BMP/TGF-β signaling cascades were central elements associated with strain-specific differences in both endochondral bone formation and cardiovascular tissue development suggests that it is one of the common mechanisms by which vascular and skeletal tissue development is coordinated. This suggestion is supported by emerging data showing that BMP/TGF-β signaling carries out essential functions in both cardiovascular tissue formation32–34 and peripheral vascular tissue development and repair.35, 36

An analysis of the specific network of genes associated with BMP/TGF-β canonical signal-transduction pathways showed that the MAP kinase cascade is one of the primary components of these pathways that showed differential expression between the strains. More specifically, the variable quantitative expression in the B6 and C3H mouse strains of Tak1 and Sos1 suggests that the proteins transcribed by these two genes, respectively, are central mediators in integrating the p38/JNK and ERK1/2 arms of the MAP kinase pathway. The observed phenotypic differences in the expression of Tak1 between the two strains of mice is consistent with recent studies that have shown that loss of Tak1 in chondrocytes leads to decreased proliferation and failure to undergo hypertrophic development in both late embryonic and early postnatal cartilage differentiation.37, 38 Our studies are also consistent with the finding showing that Tak1 regulates MAP kinase p38 signaling through activation of the BMP/SMAD signaling pathway. It is of further note, that Tak1 mediates BMP/TGF-β signaling functions during vascular development,39 during pathologic cardiac hypertrophy,40 and during cardiac remodeling after infarction.41 Sos1 was the other gene that we observed that was consistently upregulated across multiple pathways that is known to interact through the RAS-activated arm of the MAP kinase pathway via Erk1/2. Identification of elevated levels of Sos1 in the B6 strain is intriguing because a definitive functional role for Sos1 has not been shown in skeletal tissues. Clinically, however, individuals with mutations in SOS1, which lead to Noonan syndrome, present with congenital heart defects, craniofacial anomalies, and extremely short statures,42, 43 all of which are generally similar features to perturbations in BMP and/or MAP kinase signaling.

The other canonical set of signaling pathways that appeared to share components that are differentially expressed between the two strains was associated with the steroid superfamily of interacting factors that included RXR, VDR, TR, and their associated interactions with PPAR(s). In particular, RXRα was shown to be upregulated in B6 mice relative to C3H mice, along with a number of other factors associated with retinol signaling. Differences in this pathway are of interest because RXR, is a central coregulator of PPARγ activity,44, 45 and PPARγ has been shown to be involved in directing mesenchymal cell differentiation between osteogenic and adipogenic lineages.46, 47 Interestingly, while both strains of mice showed increased adipose tissue development in the marrow space in response to rosiglitazone, microarray analysis showed that the C3H strain selectively downregulated cell cycle–associated genes.48 This is noteworthy as well because the core circadian regulatory mechanism now has been shown to be intimately associated with regulation of adipose tissue development, has been shown to be functional in murine and human mesenchymal stem cells (MSCs),49–51 and has been shown to be variably regulated in these two strains of mice.52

In summary, two major conclusions may be drawn from our studies. The first is that genetic differences in the expression of genes with roles in cardiovascular tissue development emerged during postnatal peripheral vascular tissue formation that takes place during skeletal tissue regeneration. The second is that the development of vascular and skeletal tissues is regulated through the coordination of a common set of signal-transduction pathways that as a group appear to integrate tissue response to systemic metabolic conditions with those which control local tissue morphogenesis. These changes include those seen in gene expression for proteins that mediate canonical signal transduction associated with the BMP/TGF-β and PPAR/RXR pathways. From a systems point of view, integrating these two types of signal-transduction processes provides a mechanistic structure by which available nutrient and oxygen resources would be calibrated to the energy requirements needed for morphogenetic growth that is associated with development, postnatal tissue injury–induced regeneration, or postnatal homeostatic maintenance. In total, these data suggest that the primary differences between the B6 and C3H inbred mouse strains resides in variations in BMP/TGF-β and RXR/PPAR signal-transduction processes. A summary providing a unified overview of how we hypothesize the local and systemic growth-controlling mechanisms to interact and to be coordinated is shown in Fig. 5. It may be speculated, then, that the development of chronic pathologic processes that affect either cardiovascular or skeletal tissues arises from a long-term imbalance between local morphogenetic regulation, which directs patterns of cellular differentiation needed for tissue regeneration and maintenance of tissue homeostasis, and regulation that responds to and controls systemic metabolism. While not generally thought of as having related etiologies, these conditions share many of the same concordant risk factors and environmental modifiers, including their association with aging, diabetes, smoking, low physical activity, inflammation, and obesity.53 These conclusions may have considerable relevance to the development and treatment of complex polygenetic pathologies such as osteoporosis, cardiovascular disease, and fracture nonunion.

Figure 5.

Systems-oriented division of the primary canonical signaling pathways that are expressed during fracture healing into pathways that mediate metabolic regulation and are responsive to the nutrient and re/dox environment and those which mediate development, regeneration, and homeostatic maintenance and are responsive to injury and exogenous challenge. The receptor tyrosine kinases (RTK) and innate/immune signal-transduction pathways are shown overlapping metabolic and local morphogenetic regulation with preference to either metabolic or morphogenetic growth regulation. The MAP kinase cascade is depicted as the central point of convergence that is common to all these pathways. Normal growth, injury-induced regeneration, and homeostatic maintenance are achieved through balanced response to the two sets of signaling processes.

Disclosures

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

Acknowledgements

We would like to acknowledge the following people: Ms Jody McLean for carrying out the microarray data reduction normalization and CAGED analysis, Dr Temple Smith for his contributions in the microarray chip design, and Drs Marc Lenburg, Paola Sebastiani, and David Karasik for their critical reading and comments on the manuscript. This study was supported by grants from the National Institute of Arthritis and Musculoskeletal and Skin Diseases (PO1AR049920, AR44927) toTAE and LCG and the Department of Defense (DAMD17-03-1-0576) to LCG and KJJ.

All primary and annotated data sets are available upon request from Dr. Gerstenfeld.

Authors' roles: Dr. Gerstenfeld made substantial contributions to conception and design, acquisition of data, analysis, and interpretation of data and writing of the manuscript; Dr. Fitch and Ms. Grimes participated in acquisition and data analysis and drafting the manuscript or revising of the manuscript. Drs. Jepsen and Einhorn had roles in data interpretation and drafting and/or revising of the manuscript.

Ancillary