QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

Vol. 16, Issue 11, 5316-5333, November 2005

This Article Abstract Full Text (PDF) All Versions of this Article: E05-01-0084v1 16/11/5316    most recent Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by James, C. G. Articles by Beier, F. Search for Related Content PubMed PubMed Citation Articles by James, C. G. Articles by Beier, F.

Microarray Analyses of Gene Expression during Chondrocyte Differentiation Identifies Novel Regulators of Hypertrophy

Claudine G. James ^* , C. Thomas G. Appleton ^* , Veronica Ulici ^* , T. Michael Underhill ^* , and Frank Beier ^*  

^* CIHR Group in Skeletal Development and Remodeling, University of Western Ontario, London, Ontario N6A 5C1, Canada; Department of Physiology and Pharmacology, University of Western Ontario, London, Ontario N6A 5C1, Canada; Department of Anatomy, Cell Biology and Physiology, University of British Columbia, Vancouver, British Columbia V6T 1Z3, Canada; and School of Dentistry, University of Western Ontario, London, Ontario N6A 5C1, Canada

Submitted February 1, 2005; Revised June 24, 2005; Accepted August 24, 2005
Monitoring Editor: Marianne Bronner-Fraser

	ABSTRACT

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION ACKNOWLEDGMENTS REFERENCES

 
Ordered chondrocyte differentiation and maturation is required for normal skeletal development, but the intracellular pathways regulating this process remain largely unclear. We used Affymetrix microarrays to examine temporal gene expression patterns during chondrogenic differentiation in a mouse micromass culture system. Robust normalization of the data identified 3300 differentially expressed probe sets, which corresponds to 1772, 481, and 249 probe sets exhibiting minimum 2-, 5-, and 10-fold changes over the time period, respectively. GeneOntology annotations for molecular function show changes in the expression of molecules involved in transcriptional regulation and signal transduction among others. The expression of identified markers was confirmed by RT-PCR, and cluster analysis revealed groups of coexpressed transcripts. One gene that was up-regulated at later stages of chondrocyte differentiation was Rgs2. Overexpression of Rgs2 in the chondrogenic cell line ATDC5 resulted in accelerated hypertrophic differentiation, thus providing functional validation of microarray data. Collectively, these analyses provide novel information on the temporal expression of molecules regulating endochondral bone development.

	INTRODUCTION

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION ACKNOWLEDGMENTS REFERENCES

 
Two processes, intramembranous ossification and endochondral ossification, govern the development of the vertebrate skeleton. Bone derived from the former process develops directly from mesenchymal progenitor cells, whereas the latter forms by way of a cartilage intermediary (Cancedda et al., 2000; Karsenty and Wagner, 2002; Eames et al., 2003). Endochondral bone formation proceeds through the formation of a cartilage intermediate that can be further subdivided into two phases: chondrogenesis and chondrocyte differentiation. The commitment of undifferentiated mesenchymal precursor cells to the chondrogenic lineage signals the onset of this process where chondroprogenitors form cellular condensations and differentiate into chondroblasts (Hall and Miyake, 1992; Shum and Nuckolls, 2002). In the second phase, subsets of newly formed chondrocytes become mitotically active, proliferate for an interlude, exit the cell cycle, hypertrophy, and terminally differentiate. End stage chondrocytes undergo apoptosis and are replaced by bone cell precursors (Akiyama et al., 2002; Kronenberg, 2003). These later phases of chondrocyte proliferation and differentiation occur in the growth plate at the cartilage-bone interface. This region therefore controls longitudinal growth of endochondral bones through the multiplication and hypertrophy of chondrocytes.

Complex regulatory and signaling networks involving cell-matrix and intercellular interactions, coupled to tightly regulated gene expression, mediate the successive stages of proliferation and differentiation that produce all cellular states observed in the growth plate (DeLise et al., 2000). Proteins of numerous molecular families have been implicated in the longitudinal growth of the skeleton including growth factors, e.g., bone morphogenetic proteins, fibroblast growth factor family members, insulin-like growth factor signaling components (Cancedda et al., 1995); extracellular matrix molecules, e.g., collagen II, aggrecan, link protein and cartilage oligomeric protein (Cheah et al., 1991; Watanabe et al., 1994; Fang et al., 2000; Tuckermann et al., 2000); and transcription factors, e.g., Sox9, Core-binding factor alpha (Cbfa1) and ATF-2. In particular, Sox9 has been shown to be required for chondrogenesis, ATF-2 controls cell cycle progression and proliferation, and Cbfa1/Runx2 is involved in hypertrophic differentiation (Reimold et al., 1996; Lefebvre and de Crombrugghe, 1998; Beier et al., 1999; Bi et al., 1999; Stricker et al., 2002).

The intricate nature of cartilage development makes strict coordination between the various chondrogenic factors imperative for the establishment and maintenance of normal growth plate physiology. Improper regulation of genes belonging to the associated functional categories has been linked to growth disturbances and pathological conditions (Ballock and O'Keefe, 2003). For example, achondrodysplasias, hypochondrodysplasias, and thanatophoric dysplasia have been associated with activating mutations in the Fgfr3 gene, encoding fibroblast growth factor receptor 3, an important modulator of growth plate function (Ornitz, 2001). In addition, the pathogenesis of osteoarthritis is thought to reiterate changes occurring during normal cartilage development (Gelse et al., 2003).

Although many of the molecular players involved in chondrogenic differentiation have been identified, a comprehensive understanding of the mechanisms governing endochondral bone formation has not been achieved. Our knowledge of essential intracellular signaling cascades is especially limited. The advent of functional genomics in combination with systems biology and integrative physiology approaches has equipped us with the tools to overcome some of the challenges associated with understanding complex developmental processes.

In this study, comprehensive gene expression profiling of the in vitro murine micromass culture system has been used to systematically investigate temporal modulation of factors that coordinate chondrogenesis and chondrocyte differentiation. These studies identified numerous genes that undergo significant changes in expression during chondrogenic differentiation. One of these genes, the regulator of G-protein signaling 2 (Rgs2) gene was selected for functional analyses that demonstrated novel functional roles of Rgs2 in hypertrophic chondrocyte differentiation. The data presented here will further our understanding of normal endochondral bone formation and consequently the impact of pathological perturbations on the developing skeleton.

	MATERIALS AND METHODS

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION ACKNOWLEDGMENTS REFERENCES

 
Murine Micromass Cultures and RNA Isolations
Micromass cultures were prepared as described (Ahrens et al., 1977; Cash et al., 1997; Weston et al., 2000; Stanton et al., 2004) with minor modifications. Forelimbs and hindlimbs from 8 to 12 litters of E11.5 CD1 mouse embryos (obtained from Charles River Laboratories, Montreal, Quebec, Canada) were removed and segmented in Puck's Saline A buffer (PSA). The isolated limb buds were dissociated in 10 mg/ml dispase (Roche Molecular Biochemicals, Indianapolis, IN) solution containing 10% fetal bovine serum (FBS)/PSA for 1.5 h at 37°C and shaken at 100 rpm. The enzymatic reaction was neutralized with 2:3 DMEM (Wisent, St.-Bruno, Quebec, Canada): F12 media containing 10% FBS (Invitrogen, Life Technologies, Burlington, Ontario, Canada). Digested limbs were subsequently passed through a 40-µm cell strainer (Falcon, Lincoln Park, NJ) to obtain a single cell suspension and were briefly centrifuged. Cells were resuspended in growth media at a concentration of 2.5 x 10⁷ cell/ml and spotted in 10 10-µl droplets per well of a six-well NUNC cell culture plate. After cells adhered to culture dishes for 1 h at 37°C in a humidified atmosphere containing 5% CO₂, 2 ml of 2:3 DMEM/F12 media containing 10% FBS, 0.5 mM glutamine, 25 U penicillin/ml, and 25 µg streptomycin/ml was added. The medium was supplemented with 0.25 mM ascorbic acid (Sigma, Oakville, Ontario, Canada) and 1 mM -glycerophosphate (Sigma) to differentiate cultures (Stanton et al., 2004). Growth media were replaced daily, and total RNA was extracted on days 3, 6 9, 12, and 15 of culture with the Qiagen RNA easy kit according to the manufacturer's instructions (Chatsworth, CA). RNA quantity and quality was subsequently confirmed with Agilent 2100 BioAnalyzer Data Review Software (Wilmington, DE) at the London Regional Genomics Center.

Staining Methods
Cells were stained for chondrogenic differentiation with Alcian blue, which binds sulfated proteoglycans. Cells were washed twice with phosphate-buffered saline (PBS) and fixed for 2 h in a 10% Formalin solution. Alcian blue was added to cells and allowed to incubate at room temperature for 2 h. Excess stain was removed and cells were rinsed twice with 70% ethanol to remove residues before visualization (Stanton et al., 2004).

Alkaline phosphatase (ALP) activity was visualized as described (Stanton et al., 2004). For micromass staining, cells were washed twice in cold PBS, and micromass cultures fixed with 10% Formalin for 2 h at room temperature. ATDC5 cells were fixed in cold 95% ethanol for 20 min at -20°C. Cells were subsequently rinsed twice and incubated in water (pH 8.0) for 15 min. A solution containing 0.1 mg/ml naphthol AS-MX phosphate (Sigma), 0.5% N,N-dimethylformamide, 2 mM MgCl₂, and 0.6 mg/ml Red Violet LB salt (Sigma) in 0.1 M Tris/HCl (pH 8.3) was added to the cells, and the cultures were placed in dark conditions for 45 min. Cells were air-dried before computer scanning and visualization.

Von Kossa staining was utilized to visualize calcium deposition in micromass cultures. Cells were washed twice with cold PBS and fixed with 10% Formalin for 2 h at room temperature. Cells were stained first for ALP before staining with 2.5% (wt/vol) silver nitrate solution for 30 min to increase contrast. After washing with water, cells were air-dried before visualization (Stanton et al., 2004).

Microarray Analysis
Expression analysis of known chondrogenic markers Sox9 and Ibsp (Bone sialoprotein [BSP]) using real-time PCR in three independent micromass trials was completed as described (Stanton et al., 2004) before microarray analysis to validate RNA quality and to verify chondrogenic differentiation in these trials. Total RNA from the three biological replicates were subsequently analyzed with the Agilent Bioanalyzer 2100 system at the London Genomics Facility to confirm RNA integrity. Hybridization proceeded according to the standard Affymetrix protocol (http://www.affymetrix.com/support/technical/manuals.affx). The MOE430A chip contains oligonucleotides for 22 690 probe sets representing 14,000 mouse genes.

Data Analysis
Microarray Suite 5.0 and GeneSpring 6.1. The gene expression data were analyzed using the Microarray Suite (MAS) 5.0 algorithm (Affymetrix, Santa Clara, CA) in which all probe sets were scaled to the target value of 150. Pivot files generated through M.A.S. analysis were imported into GeneSpring 6.1 software (Silicon Genetics, Redwood City, CA) for data mining. Data transformation values were set to <0-0.1, and per chip normalizations were set to the 50th percentile in addition to per gene normalizations to the median and to specific samples. Day 3 data from all experimental replicates were defined as the baseline array. Signal intensity values for all experimental replicates on any given time point were averaged and used for additional analysis. The starting data set represented 22,690 probe sets. Additional filtering was executed to reduce type I errors (i.e., false positives), which result from experimental procedures and probe design. Genes assigned an "Absent" call for all time points were eliminated from the data set and 16,709 probe sets remained. M.A.S. 5.0 derived algorithms assign statistically spurious expression values an "Absent" call, which signifies of a decreased likelihood that the corresponding signal intensity obtained from the analysis is a reflection of an actual expressed transcript. This data set was additionally filtered with a one-way Welch ANOVA (p-value cutoff of 0.05) and Benjamini and Hochberg False Discovery Rate testing in order to further reduce the working data set to 3334 probe sets. Probe set lists were filtered using the "Filter on Fold Change" option in GeneSpring. A minimum twofold change in gene expression defined differential expression for this data set.

GeneTraffic. Raw microarray data were also imported into GeneTraffic UNO 3.0 (Iobion Informatics, Stratagene, La Jolla, CA) and normalized according to the Robust Multi-array Analysis summary measure (Irizarry et al., 2003) in which day 3 represented the baseline culture time point. Resulting signal intensities were utilized for subsequent analysis. The 22,690 probe sets were filtered to remove all genes that were not assigned a present call in at least one culture time point. The resulting data set contained 16,256 probe sets. Gene expression values were subsequently filtered in a parallel manner to the list generated in GeneSpring after statistical analysis. Individual probe sets were selected for visualization.

Clustering. Self Organizing Maps (SOMs) were generated using the 3334 probe sets obtained from data analysis in GeneSpring 6.1. The following parameters were used: 7 rows, 6 columns, 220,000 iterations, and a neighborhood radius of 6.0. Genes without data in half of the starting conditions were not used for the analysis. K-means clustering where K = 5 of 3 of the 42 cluster sets created by SOM analysis was executed to generate groups of highly similar probe sets. Standard correlation similarity measures were used along with 10,000 iterations that converged after six iterations to create final cluster sets.

Gene Ontologies. Probe set lists resulting from the comparison of genes expressed on day 3 versus 15 of culture filtered using a twofold cutoff were assigned a molecular function with the fatiGO program (Al-Shahrour et al., 2004; http://fatigo.bioinfo.cnio.es/), using the level two filtering parameter. Probe sets annotated by the Gene Ontology Consortium were selected for analysis. Genes assigned to individual categories listed were calculated based on the proportion of annotated probe sets included in all lists.

RT-PCR
cDNA from the micromass time course was generated with the First-Strand cDNA Synthesis system for RT-PCR using SuperScript II RNase H-Reverse Transcriptase (Invitrogen Life Technologies), random hexamers (Invitrogen Life Technologies) and p(dT) 12-18 (Invitrogen Life Technologies). RT products were generated according to the manufacturer's specifications using 1 µg of RNA and a 50-min incubation period at 42°C. A 15-min cDNA incubation at 65°C terminated the reaction. PCR was executed in 50-µl reaction volumes containing 1 µl RT products and 0.5 µl of AmpliTaq Gold polymerase (Perkin Elmer-Cetus, Boston, MA). Amplification conditions for primers began with a 1.5-min denaturation step at 95°C, with annealing temperatures ranging from 50 to 63°C with a 72°C extension phase. PCR reactions occurred over 25-32 cycles, and the reaction was terminated with a 6-min final extension phase at 72°C. PCR products were visualized with UV light after electrophoresis of a 1% agarose gel containing ethidium bromide. For primer sequences, see Table 3.

In Situ Hybridization
PCR products from RT-PCR amplification of murine Rgs2 mRNA (see Table 3 for primer sequences) were cloned into pGEM T-Easy vector (Promega, Madison, WI), and clones were validated by sequencing. Linearized pGEM-Rgs2 vector was used to generate sense (via T7 RNA Polymerase) and antisense (via SP6 RNA Polymerase) DIG-labeled riboprobes using 10X DIG Labeling Mix (Roche).

Limbs from embryonic day 15.5 mice (E15.5) were fixed overnight with 4% paraformaldehyde (PFA) in PBS, pH 7.4, at 4°C, washed overnight with PBS (4°C), dehydrated, and embedded in paraffin. Sections of 10 µm were cut, mounted on 3-aminopropyltriethoxy-silane-treated (positively charged) glass slides, dewaxed in xylenes, and rehydrated. After washing in PBS, they were digested with 10 µg/ml proteinase K, fixed in 4% PFA, and acetylated with 0.25% acetic anhydride. The sections were hybridized overnight at 55°C with DIG-labeled riboprobes for Rgs2 (sense or antisense). After hybridization, the sections were washed in 2x SSC, 1x SSC, and 0.5x SSC at 50°C. Riboprobes were digested with RNase A and washed once with tris-buffered saline (TBS; pH 7.5). Anti-DIG antibody conjugated to ALP (Roche) was used as the primary antibody. The blocking and detection of the DIG-labeled sections using NBT-BCIP colorimetric reaction was carried out according to the instructions of the manufacturer (Roche).

Generation of Stable Transfectants and ATDC-5 Cell Culture
ATDC5 cells were cultured and transfected as described (Wang et al., 2004). For transfections, cells were seeded at 2 x 10⁴ cells/ml per well of six-well tissue culture plates and individually transfected with expression vectors for human RGS2 (Guthrie cDNA Resource Center, http://www.cdna.org/) and empty expression vector pcDNA3.1+ (Invitrogen) using Fugene6 (Roche) according to the manufacturer's specifications. Pools of transfected cells were selected with 800 mg/ml Geneticin. Differentiation was induced by addition of ITS as described (Wang et al., 2004).

Western Blotting
ATDC5 cells were centrifuged at 1000 x g in ice-cold PBS and resuspended in ice-cold RIPA lysis buffer (150 mM NaCl, 50 mM Tris-HCl, pH 7.5, 1% Triton X-100, 1% deoxycholate, 0.1% SDS, 2 mM EDTA, 500 mM sodium fluoride, 100 mM sodium orthovanadate). Cell lysates were subsequently sonicated and the protein was quantified by bicinchoninic acid (BCA) assay (Sigma) according to the manufacturer's specifications. To confirm expression of HA-tagged RGS2, 60 µg of protein was loaded and size-fractionated on a 10% SDS-PAGE and blotted overnight onto a nitrocellulose membrane (PROTRAN, Schleicher and Schuell Bioscience, Keene, NH). Nonspecific sites were blocked in a 5% solution of nonfat milk powder in TBS. The nitrocellulose membrane was incubated with high-affinity rat monoclonal anti-HA IgG (Roche) followed by incubation with goat anti-rat secondary IgG antibody with horseradish peroxidase conjugate (Santa Cruz Biotechnology, Santa Cruz, CA) using 1:1000 and 1:5000 dilutions, respectively. Lysates containing HA-tagged constructs were visualized with enhanced chemiluminescence Advance Western blot detection system (Amersham Biosciences, Piscataway, NJ) and Alphaimager 2200.

View larger version (68K): [in this window] [in a new window]   Figure 1. Qualitative characterization of micromass cultures. Images of micromass cultures undergoing chondrogenesis and chondrocyte differentiation on days 3, 6, 9, 12, and 15 of culture are represented. Cells were stained with Alcian blue to show synthesis of sulfated glycosaminoglycans, for ALP activity and with von Kossa for mineralization. All markers increase throughout the culture period.

Real-Time PCR
Real-Time PCR was performed to quantitatively assess RNA samples according to (Stanton et al., 2004). Primer and probe sequences for TaqMan GAPDH control reagents were used as an internal control, because this gene demonstrated less variability and greater reproducibility in our system (compared with other standardization probes such as 18s RNA). The GAPDH primer probe set was 5'-GAAGGTGAAGGTCGGAGTC-3' for the forward primer; 5'-GAAGATGGTGATGGGATTTC-3' for the reverse primer and JOE-CAAGCTTCCCGTTCTCAGCC-TAMRA for the probe. Specific target primers and probes for Fgfr3 and Ihh with primers and probes generated from Applied Biosystems TaqMan Assays-on-Demand consisting of two unlabeled primers and FAM (6-carboxyfluorescein) dye-labeled TaqMan MGB probe in addition to primer and probe sets for Sox9, Col2a1, and BSP as described (Stanton et al., 2004). Individual primer pairs along with their corresponding probes and template RNA were combined with TaqMan one-step mastermix kit (Applied Biosystems) up to a total volume of 15 µl. The ABI Prism 7900 HT sequence detector (Perkin Elmer-Cetus) was used to detect the amplified target sequences. The primers were annealed at 60°C for 40 PCR cycles. Experimental values represent reaction mixtures completed in triplicate for each time point, and data are a compilation of a minimum of three experimental trials. Negative controls include the mastermix without template RNA. Real-time expression values were calculated using the relative standard curve method. Standard curves were generated for both the target of interest and the endogenous control (GAPDH) by measuring the cycle number at which exponential amplification occurred in a dilution series of samples with known concentrations (plotted as the log concentration). The level of target and endogenous control transcripts were calculated by solving for the x-intercept of the slope of the standard curve line. Normalized target values were subsequently generated by taking the antilog of the x-intercept and finally taking the quotient of the amounts of target and endogenous control

	RESULTS

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION ACKNOWLEDGMENTS REFERENCES

 
Validation of Micromass Culture System
We used micromass cultures as an established in vitro model system for chondrogenesis and chondrocyte differentiation of mouse mesenchymal cells (Cormier et al., 2003; Stanton et al., 2004; Zhang et al., 2004). Cultures were qualitatively assessed with Alcian blue, ALP and von Kossa staining on days 3, 6, 9,12, and 15 of the time course. In all cases, staining intensity increases over time confirming the progression of the cultures along the chondrogenic lineage (Figure 1). Quantitative Real-Time PCR analysis of the micromass cultures revealed that expression of well-characterized markers of chondrocyte differentiation, namely Sox9 and Ibsp, follow expected expression patterns (Chen et al., 1992; Bi et al., 1999; unpublished data). Sox9 mRNA showed peak expression during days 3 and 6 of the culture period, after which transcript levels decreased by 50% in parallel with the maturation of differentiated chondrocytes. The opposite pattern was observed for Ibsp transcripts, which showed very low expression until days 12 and 15 of culture when cartilage cells terminally differentiate and up-regulate Ibsp mRNA levels 1200-fold.

View larger version (40K): [in this window] [in a new window]   Figure 2. Affymetrix microarray analysis of chondrogenesis and chondrocyte differentiation. Total RNA from the micromass time course was hybridized to Affymetrix MOE430A chips containing 14,000 murine genes. Data from three biological replicates are shown. Preliminary analysis in M.A.S 5.0 and GeneSpring 6.1, in which all probe sets were subject to various filters (see Materials and Methods), resulted in the creation of data set consisting of 3334 probes sets. Expression profiles for Sox9 and Ibsp are highlighted in red and green, respectively (A). Expression patterns for additional characterized cartilage markers such as Aggrecan (Agc1), cartilage-derived retinoic acid-sensitive protein (Cdrap), Sox9, Link protein (Crtl1), and Collagen II (Col2a1) are also shown (Ba). Expression patterns of IGF signaling constituents are shown (Bb). Transcripts involved in extracellular matrix signaling including matrilin (Matn1), perlecan (Hspg2), collagen 11a1 (Col11a1), osteopontin (Opn1), tissue inhibitor of metalloproteinase 3 (Timp3), collagen 9a2 (Col9a2), and osteonectin (Spock1) are shown (Bc). Genes such as insulinlike growth factor binding protein 6 (Igfbp6) and matrix metalloprotease 13 (Mmp13), Ibsp, a disintegrin and metalloprotease domain 23 (Adam23), and cartilage oligomeric protein (Comp) exhibit greater than fivefold changes in gene expression and are not shown for scaling reasons. (B). All values reflect fold changes in gene expression relative to day 3 of micromass culture.

The insulinlike growth factor (IGF) signaling system is one of the major regulators of endochondral ossification (Schmid, 1995). We analyzed expression of IGF signaling components in our data sets. Most notably, mRNA encoding IGF1 is markedly up-regulated at later stages of differentiation, along with transcripts for IGF-binding protein (IGFBPs) 2, 4, and 6, whereas IGF2 displays the opposite behavior (Figure 2Bb, Table 1). In particular, IGFBP6 is up-regulated 11-fold on day 15 versus day 3 of culture and shows a very similar expression pattern to matrix metalloprotease 13 (Mmp13), Ibsp, Cartilage Oligomeric Protein (Comp) and the newly implicated a disintegrin and metalloprotease domain 23 (Adam23), which are examples for end-stage cartilage markers that exhibit greater than fivefold upregulation in gene expression in our array (IGFBP6 could not be shown in Figure 2 for scaling reasons).

View larger version (29K): [in this window] [in a new window]   Figure 3. Validation of differentially expressed genes by RT-PCR. Temporal expression patterns of Cartilage Link Protein 1 (Crtl1; A), nuclear cap binding protein (Ncbp2; B), and myogenesis differentiation factor 1 (Myod-1; C) were confirmed by RT-PCR using -actin as a loading control. The microarray expression profile shown as fold change in expression is shown on the left with the corresponding RT-PCR to the right.

Identification of Trends in Gene Expression
The distribution of differentially expressed probe sets was assessed for the duration of the time course. Specifically, the number of differentially expressed transcripts between sequential time points was compared using various fold change cutoffs (Figure 4A). The largest changes in gene expression were observed between days 3 and 6 of the time course at all fold change cutoffs. Conversely, the smallest number of genes changed between days 12 and 15 of culture.

View larger version (38K): [in this window] [in a new window]   Figure 4. Distribution of differentially expressed probe sets and corresponding molecular classification according to GO terms. Differential gene expression analyzed in succession shows the number of probe sets exhibiting changes as the developmental time line progresses. Probe sets were filtered according to 2-, 5-, and 10-fold change cutoffs (A). Day 3 versus day 6, and day 12 versus day 15 represent the largest and smallest proportion of changes between culture days, respectively. Genes exhibiting minimum two-fold changes in gene expression between day 3 and day 15 were assigned molecular function and categorized according to GO annotations assigned in the FatiGO program (B). The proportion of genes that are up- or down-regulated in the gene list derived from twofold changes in gene expression between days 3 and 15 of micromass culture (C).

Probe set lists for transcripts differentially regulated between day 6, 9, 12, or 15 of culture and the day 3 baseline were examined using 2-, 5-, and 10-fold change cutoffs. Approximately 1772, 481, and 249 probe sets were differentially expressed by 2-, 5-, and 10-fold, respectively, between days 3 and 15 of culture, which coincides with the largest change in gene expression between two nonsequential culture days. Table 2 shows genes demonstrating at least 10-fold changes in expression levels between days 3 and 15. This is consistent with the different stages of cartilage formation: namely chondrogenesis and the terminal differentiation of chondrocytes. Closer examination of these lists involved functional categorization of differentially expressed transcripts in the twofold change category (1772 probe sets). Classification according to GeneOntology (GO) annotations showed that the majority of annotated transcripts were involved in catalysis (39%), signal transduction (23%), and transcriptional regulation (11%). Approximately 10% of transcripts included were not assigned a functional classification (shown as "other"; Figure 4B). Similar distributions were observed for genes with a 5- and 10-fold cutoff (unpublished data).

Differences in the proportion of transcripts showing increased or decreased expression patterns were also assigned to various functional groups. The most pronounced difference occurred in the signal transduction, transcription, and motor activity categories where 1.8-fold more up-regulated transcripts, 1.5-fold more down-regulated transcripts, and 4.4-fold more down-regulated transcripts, respectively, were identified (Figure 4C). These patterns are consistent with gene expression profiles observed for vascular endothelial growth factors and matrix metalloproteases in the case of the signal transduction, Sox9 and other patterning molecules in the case of transcription, and muscle markers such as myosin in the case of motor activity. The identification of gene expression trends provides insight into the distribution of functional classes that temporally modulate chondrocyte maturation as well as additional support for the micromass culture system as an appropriate in vitro model for chondrocyte differentiation.

Clustering Analyses
Genes exhibiting similar expression pattern may be involved in similar biological processes and may therefore be regulated by similar upstream mechanisms (Marcotte et al., 1999). Thus, we attempted to identify coexpressed genes by cluster analysis. Two clustering methods were used to complete the analysis: SOMs and K-means clustering (see Materials and Methods for details). SOMs were executed in GeneSpring using the initial data set consisting of all significantly expressed transcripts. Of the 42 resulting SOMs, three patterns were analyzed in detail. Cluster 1 contained 140 probe sets that are up-regulated toward the end of the developmental program, likely representing transcripts involved in hypertrophic differentiation (Figure 5A, left panel). Mmp13, a known marker of late-stage hypertrophic differentiation, is found in this cluster (Tuckermann et al., 2000). Cluster 2 contained 58 transcripts that peak around day 6 and day 9 of culture and are subsequently down-regulated (represented by Crtl1; Figure 5B, left panel). Cluster 3 contained 119 probe sets that are down-regulated during the time course (e.g., Myod1; Figure 5C, left panel). Additional probe sets demonstrating these expression profiles for each cluster are shown in Figure 5 (right panels).

View larger version (37K): [in this window] [in a new window]   Figure 5. Self-organizing maps and k-means clustering of microarray data. SOM analysis of 3334 probe sets from GeneSpring analysis and subsequent k-means clustering of three SOM derived clusters. Cluster 1 contains genes that show up-regulation toward day 15 of the micromass time course (A, left panel). Genes in this cluster include Mmp13, Ibsp, Comp, and fatty acid binding protein 4 (Fabp4; A, right panel). Standard correlation was used to measure similarity between expression profiles. Cluster 2 contains genes following an expression pattern similar to that of Cartilage link protein (Crtl1; B, left panel), such as Angiotensin II receptor type 2 (Atgr2), collagen 14 (Col14a1), and Col2a1 (B, right panel). Cluster 3 represents the expression pattern of genes down-regulated over time (C, left panel), including cardiac actin (Actc1), brain expressed X-linked 2 (Bex2), Tncc, Myod1, troponin (Tnnt1), and alpha actinin (Actn2; C, right panel). Red lines show up-regulation and blue lines show down-regulation of gene expression.

Functional Validation of Microarray Results: Rgs2 Regulates Chondrocyte Differentiation
One pervasive expression pattern identified with cluster analysis of the microarray data set included genes that are up-regulated as micromass cultures approach the end of their development program. We postulated that genes following this particular expression pattern might promote chondrocyte maturation. Rgs2 was among the probe sets that exhibit this expression pattern. Rgs2 transcripts increased by twofold as the micromass cultures progressed from day 3 to day 12 of culture and dropped slightly thereafter (Figure 6A). Semiquantitative RT-PCR confirmed that Rgs2 was indeed markedly up-regulated during micromass culture; however, the drop in Rgs2 expression at day 15 of culture was not observed in these experiments, potentially due to slight variations in the amounts of RNA or cDNA as shown by Actin RT-PCR (Figure 6B). To clarify the expression pattern of Rgs2 in growth plate cartilage in vivo, we performed in situ hybridization on tibia sections from E15.5 mice (Figure 6C). Rgs2 expression was weak in resting chondrocytes, increased markedly in proliferating and in prehypertrophic chondrocytes, and decreased upon full hypertrophic differentiation. These in vivo data thus confirm the expression patterns observed in microarray analyses of micromass cultures.

View larger version (40K): [in this window] [in a new window]   Figure 6. Confirmation of Rgs2 expression in murine chondrocyte differentiation. Microarray expression pattern for Rgs2 in micromass cultures show transcript accumulation from days 3 to 12 of culture, with a subsequent drop in expression (A). Rgs2 mRNA up-regulation during micromass differentiation was confirmed by RT-PCR using -actin as a loading control (B). Rgs2 mRNA expression in the tibia of a E15.5 mouse was analyzed by in situ hybridization (C). Sense probes resulted in no signal, whereas Rgs2 antisense probes demonstrated weak or no expression in resting chondrocytes (R), strong expression of Rgs2 mRNA in proliferating (P) and prehypertrophic chondrocytes, and lower expression in fully hypertrophic chondrocytes (H). Confirmation of RGS2 overexpression in chondrogenic ATDC-5 cells by RT-PCR and Western blotting (C). Ladder (L), pcDNA 3.1+ vector (V), and a 657-bp Rgs2 amplicon are shown (left panel). Cell lysates isolated from ATDC-5 cells overexpressing HA-tagged RGS2 were separated by SDS-PAGE and blotted onto nitrocellulose membrane, which was subsequently probed with an anti-HA antibody (right panel).

We stably transfected an HA-tagged human RGS2 expression vector into the chondrogenic cell line ATDC5 (Atsumi et al., 1990) and differentiated them for 18 d to further investigate the role of RGS2 expression in chondrocyte differentiation. Overexpression of RGS2 was confirmed by 1) RT-PCR using primers designed against the N-terminal HA-tag and the RGS2 coding region, and 2) Western blotting using an anti-HA antibody. Both RGS2 mRNA and protein were elevated compared with cells expressing the control vector and generated a 657-base pairs amplicon and 32-kDa protein, respectively (Figure 6D).

On examination of chondrocyte phenotype, we found that RGS2 overexpression causes increased glycosaminoglycan synthesis (as shown by a more rapid increase in the intensity of Alcian blue staining; Figure 7A) and ALP activity, both in staining (Figure 7B) and enzymatic assays (Figure 7C). Real-time PCR analysis of the culture for stage-specific cartilage markers reveal no significant differences in the levels of Sox9, Col2a1, and Ihh transcripts in cells overexpressing Rgs2 compared with control cultures (Figure 8, A, B, and D). The expression of two other markers of chondrocyte differentiation, Fgfr3 and Ibsp, however, is increased by 2- and 10-fold, respectively in these cultures (Figure 8, C and E). The accelerated increase in glycosaminoglycan synthesis, ALP activity, and Fgfr3 and Ibsp expression suggest that overexpression of RGS2 accelerates the rate of chondrocyte differentiation and modulates the expression of certain markers of the chondrocyte phenotype, thus confirming the roles postulated from the microarray expression pattern.

View larger version (41K): [in this window] [in a new window]   Figure 7. Functional characterization of RGS2 in chondrocyte differentiation. Differentiating ATDC-5 cells stably overexpressing RGS2 were stained with Alcian blue (A) and ALP on days 3, 12, and 18 of culture (B). ALP activity was subsequently quantified by enzyme assay. Stainings were repeated three times on independent trials, and ALP activity values represent an average of six independent trials. Significant differences from the vector control (***) were determined by p < 0.001. RGS2 overexpression results in accelerated and increased induction of glycosaminoglycan synthesis and ALP activity.

View larger version (17K): [in this window] [in a new window]   Figure 8. Effects of RGS2 overexpression on chondrocyte gene expression. The relative abundance of Sox9 (A), Col2a1 (B), Fgfr3 (C), Ihh (D), and Ibsp (E) transcripts in ATDC-5 cells overexpressing RGS2 was quantified with real-time PCR. Expression of Sox9 and Col2a1 were assessed after 6 d of culture and the expression of all subsequent markers assessed on day 15 of culture. Results represent means and standard deviations from three independent trials completed in triplicate. Expression values are normalized to a Gapdh internal control. Significant differences in expression from control (*) were determined by p < 0.001.

	DISCUSSION

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION ACKNOWLEDGMENTS REFERENCES

In addition to demonstrating the expected expression patterns of established differentiation markers, microarray data were validated using real-time and RT-PCR analyses for several genes, as well as by in situ hybridization of tissue sections for Rgs2. All these results demonstrate that expression patterns observed in the microarray experiments of micromass cultures reflect authentic expression in vitro and in vivo, at least for the vast majority of genes.

The functional distribution of GO-annotated probe sets identified using different fold change cutoffs was similar. Strongly represented categories included probe sets involved in catalysis and signal transduction. A large proportion of unannotated genes were also identified, suggesting that additional novel chondrogenic markers or markers that have not been implicated in chondrocyte differentiation exist and warrant further investigation. This analysis presents an additional avenue from which candidate genes can be categorized and analyzed. Comparisons between genes expressed on day 3 and day 15 of micromass culture that were filtered with a minimum twofold change cutoff produced a list containing 1772 genes. Numerous gene families that have been implicated in different cellular processes were shown, some of which were unique to a particular expression pattern (up- or down-regulated). For example, several myogenic markers including Myod1, myosins, myogenin, and troponins demonstrate strongest down-regulation over time in micromass culture.

Analysis of the filtered data set shows that changes in gene expression accumulate from day 3 to day 15 of culture, so that the largest changes are observed when data sets from these culture days were compared, irrespective of the fold change cutoff used. This trend is consistent with the changes observed in the development of an organ system in which cells undergo a transition from a state of developmental plasticity to a state of determination, specification, and ultimately terminal differentiation (Loebel et al., 2003). The largest successive changes in gene expression occur between days 3 and 6 of the culture period, which is consistent with chondrogenic differentiation. Changes observed between the later developmental stages were conversely smaller.

A caveat of microarray analysis is the generation of data sets with high dimensionality, which in turn generates a proportion type I errors (Dudoit and Fridlyand, 2002; Reiner et al., 2003). Microarray analysis is also limited in its ability to quantify the expression levels observed for any given chip (Sekiya et al., 2002; Barash et al., 2004). These two features pose a problem with interpreting the significance of expression data. Currently, studies are underway to bypass these obstacles; however, a consensus has yet to be reached regarding the selection of optimal normalization algorithms in particular (Love et al., 2002). M.A.S. 5.0 algorithm from Affymetrix, Model Based Expression Index (Li and Wong, 2001) and Robust Multi-array analysis (RMA; Irizarry et al., 2003) are examples of currently used summary measures. In our case we processed the data according to MAS 5.0 algorithms and subsequently filtered the data in both GeneSpring 6.1 and in GeneTraffic 3.0, which used RMA. A range of filtering parameters (i.e., 2-, 5- and 10-fold change cutoffs) was also used to identify the distribution and the degree to which the expression of certain genes change.

Higher stringency normalization and filtering reduces the frequency of false positive data, because noise is dependent on the observed signal intensity (Tu et al., 2002; Cole et al., 2003). We must establish a balance between excluding biologically meaningful data by using restrictive analysis criteria and using permissive parameters, which could likewise reduce the biological value of the data by increasing the number of artifacts. For example, normalization and analysis of our data set in GeneSpring identified the temporal expression pattern of Rgs2, whereas changes in Rgs2 mRNA expression were not deemed significant using the more stringent RMA normalization in GeneTraffic. However, confirmation of the Rgs2 expression with RT-PCR suggests that normalization and subsequent filtering of the data in GeneSpring produces biologically relevant data sets. Similar findings were also obtained with numerous other genes (unpublished data).

At the same time, however, the larger gene lists generated in GeneSpring make the selection of a manageable number of candidate molecules for functional characterization a much more complex task. It is important to note that the gene expression patterns observed in the data set are similar between both analysis methods, i.e., GeneSpring versus GeneTraffic. Furthermore, the probe sets identified by parallel filtering methods show the expected inverse correlation between the number of identified probe sets and the fold change cutoff used. For example, the myogenic markers identified in this list are consistently down-regulated in both programs even though GeneSpring analysis proves to be less stringent. These findings highlight the fine balance between the specificity and sensitivity of normalization algorithms and filtering conditions, and the importance of implementing algorithms that coincide with the biological questions of interest. Combining robust statistical algorithms with various filtering stringencies provides a broader spectrum of changes from which integrative hypotheses may be derived while maintaining both the statistical confidence and biological relevance of the analyzed expression data.

Our cluster analyses revealed numerous groups of coexpresssed genes, three of which were analyzed in more detail. The probe sets identified in cluster 1 contained Mmp13 as well as probe sets for genes exhibiting a minimum 10-fold up-regulation from day 3 to 15 of micromass culture. Specifically, this cluster included other matrix molecules such as Comp and factors that have not been well characterized in the context of chondrogenic differentiation. Similarly, a group of CC and CXC chemokine receptors (CCRs) were identified in cluster 3 (Table 3). Silvestri et al. (2003) established the role of chemokine receptors in the maintenance of healthy cartilage by maintaining the balance between catabolic and anabolic processes in the tissues. Alaaeddine et al. (2001) demonstrated that chemokine receptor genes (CCRs) are up-regulated in human osteoarthritic. It is interesting to note that, contrary to the pattern observed for CCRs, several of their ligand chemokines exhibited a minimum 10-fold increase in gene expression during micromass differentiation in our data sets. This family of molecules has been implicated in promoting chondrocyte hypertrophy (Merz et al., 2003). Further studies should elucidate the roles chemokine receptors play in both normal and pathological cartilage conditions. Novel factors were also identified with the cluster analysis (Table 4). For example, the angiotensin type II receptor (Agtr2) (Ma et al., 1998) exhibited an expression pattern similar to cartilage link protein and was found in cluster 2.

The activation of signaling cascades essential to endochondral bone growth and remodeling occurs partially through the coordinate action of G-protein-coupled receptors (GPCRs) and heterotrimeric G-proteins, for example, the parathyroid hormone and parathyroid hormone-related peptide receptor (PTH/PTHrP-R) and G subunit proteins (Bringhurst et al., 1993; Bowler et al., 1998; Chung et al., 1998; Inoue and Matsumoto, 2000). Signaling through GPCRs is regulated by the activities of regulators of G-protein signaling (RGS) proteins, which are a family of signaling molecules implicated in regulating the rate at which G-proteins hydrolyze bound GTP (Hepler, 1999; De Vries et al., 2000; Hepler, 2003; Ishii and Kurachi, 2003). RGS2 has been implicated in the down-regulation of PTH-mediated signaling in osteoblasts (Ko et al., 2001; Thirunavukkarasu et al., 2002). Our data suggest that the gene encoding RGS2 is also expressed in chondrocytes and functions in the regulation of chondrocyte differentiation in ATDC5 cells. RGS2 overexpression in these cultures advanced the production of glycosaminoglycans and ALP. Parallel increases in the expression of the chondrogenic marker genes Fgfr3 and Ibsp, which to our knowledge have not yet been described as RGS2 target genes, were also observed. However, other chondrocyte markers, such as Sox9 and Indian hedgehog, were not affected by RGS2 overexpression, suggesting that the effects of RGS2 are specific to certain aspects of chondrocyte maturation. Future studies will include in depth analysis of the mechanism of RGS2 regulation of chondrocyte differentiation. Nevertheless, these studies provide an example how our microarray analyses result in the identification of novel candidate regulators of chondrocyte differentiation and in the subsequent experimental validation of the hypothesized biological roles of these candidates.

The integration of an in vitro limb culture system and high throughput microarray analysis has provided a valuable tool for identifying global gene expression profiles of markers and potential regulators of chondrogenic differentiation. Functional studies confirmed the biological relevance of the Rgs2 expression pattern identified in our microarray analyses. This research facilitates the development of novel complex, testable hypotheses regarding potential regulators of chondrocyte development. These data thus provide the basis for improved understanding of cartilage development, homeostasis, and disease.

Note added in proof. All data sets have been submitted to the NCBI/GEO database (entry number GSE2154 [NCBI GEO] ).

	ACKNOWLEDGMENTS

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION ACKNOWLEDGMENTS REFERENCES

 
Work in the laboratory of F.B. is supported by grants from the Canadian Institutes of Health Research, the Arthritis Society, the Canadian Arthritis Network, the Canada Research Chair Foundation, the National Science and Engineering Research Council, and the Hospital for Sick Children Foundation. C.G.J. is the recipient of an Ontario Graduate Scholarship in Science and Technology and a Canadian Institutes of Health Research Doctoral Award, and C.T.G.A. was supported by a stipend from the Ontario Genomics Institute.

	Footnotes

 
This article was published online ahead of print in MBC in Press (http://www.molbiolcell.org/cgi/doi/10.1091/mbc.E05-01-0084) on August 31, 2005.

Address correspondence to: Frank Beier (fbeier{at}uwo.ca).

	REFERENCES

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION ACKNOWLEDGMENTS REFERENCES

 
Ahrens, P. B., Solursh, M., and Reiter, R. S. ((1977). ). Stage-related capacity for limb chondrogenesis in cell culture. Dev. Biol. 60, , 69-82.[CrossRef][Medline]

Akiyama, H., Chaboissier, M. C., Martin, J. F., Schedl, A., and de Crombrugghe, B. ((2002). ). The transcription factor Sox9 has essential roles in successive steps of the chondrocyte differentiation pathway and is required for expression of Sox5 and Sox6. Genes Dev. 16, , 2813-2828.[Abstract/Free Full Text]

Al-Shahrour, F., Diaz-Uriarte, R., and Dopazo, J. ((2004). ). FatiGO: a web tool for finding significant associations of gene ontology terms with groups of genes. Bioinformatics 20, , 578-580.[Abstract/Free Full Text]

Alaaeddine, N., Olee, T., Hashimoto, S., Creighton-Achermann, L., and Lotz, M. ((2001). ). Production of the chemokine RANTES by articular chondrocytes and role in cartilage degradation. Arthritis Rheum. 44, , 1633-1643.[CrossRef][Medline]

Atsumi, S., Umezawa, K., Iinuma, H., Naganawa, H., Nakamura, H., Iitaka, Y., and Takeuchi, T. ((1990). ). Production, isolation and structure determination of a novel beta-glucosidase inhibitor, cyclophellitol, from Phellinus sp. J. Antibiot. (Tokyo) 43, , 49-53.[Medline]

Ballock, R. T., and O'Keefe, R. J. ((2003). ). Physiology and pathophysiology of the growth plate. Birth Defects Res. Part C Embryo Today 69, , 123-143.

Barash, Y., Dehan, E., Krupsky, M., Franklin, W., Geraci, M., Friedman, N., and Kaminski, N. ((2004). ). Comparative analysis of algorithms for signal quantitation from oligonucleotide microarrays. Bioinformatics 20, , 839-846.[Abstract/Free Full Text]

Beier, F., Lee, R. J., Taylor, A. C., Pestell, R. G., and LuValle, P. ((1999). ). Identification of the cyclin D1 gene as a target of activating transcription factor 2 in chondrocytes. Proc. Natl. Acad. Sci. USA 96, , 1433-1438.[Abstract/Free Full Text]

Bi, W., Deng, J. M., Zhang, Z., Behringer, R. R., and de Crombrugghe, B. ((1999). ). Sox9 is required for cartilage formation. Nat. Genet. 22, , 85-89.[CrossRef][Medline]

Bowler, W. B., Gallagher, J. A., and Bilbe, G. ((1998). ). G-protein coupled receptors in bone. Front Biosci. 3, , D769-D780.

Bringhurst, F. R., Juppner, H., Guo, J., Urena, P., Potts, J. T., Jr., Kronenberg, H. M., Abou-Samra, A. B., and Segre, G. V. ((1993). ). Cloned, stably expressed parathyroid hormone (PTH)/PTH-related peptide receptors activate multiple messenger signals and biological responses in LLC-PK1 kidney cells. Endocrinology 132, , 2090-2098.[Abstract]

Cancedda, R., Castagnola, P., Cancedda, F. D., Dozin, B., and Quarto, R. ((2000). ). Developmental control of chondrogenesis and osteogenesis. Int. J. Dev. Biol. 44, , 707-714.[Medline]

Cancedda, R., Descalzi Cancedda, F., and Castagnola, P. ((1995). ). Chondrocyte differentiation. Int. Rev. Cytol. 159, , 265-358.[Medline]

Cash, D. E., Bock, C. B., Schughart, K., Linney, E., and Underhill, T. M. ((1997). ). Retinoic acid receptor alpha function in vertebrate limb skeletogenesis: a modulator of chondrogenesis. J. Cell Biol. 136, , 445-457.[Abstract/Free Full Text]

Cheah, K. S., Lau, E. T., Au, P. K., and Tam, P. P. ((1991). ). Expression of the mouse alpha 1(II) collagen gene is not restricted to cartilage during development. Development 111, , 945-953.[Abstract/Free Full Text]

Chen, J., Shapiro, H. S., and Sodek, J. ((1992). ). Development expression of bone sialoprotein mRNA in rat mineralized connective tissues. J Bone Miner. Res. 7, , 987-997.[Medline]

Chung, U. I., Lanske, B., Lee, K., Li, E., and Kronenberg, H. ((1998). ). The parathyroid hormone/parathyroid hormone-related peptide receptor coordinates endochondral bone development by directly controlling chondrocyte differentiation. Proc. Natl. Acad. Sci. USA 95, , 13030-13035.[Abstract/Free Full Text]

Cole, S.W.G., Galic, Z., and Zack, J. A. ((2003). ). Controlling false-negative errors in microarray differential expression analysis: a PRIM approach. Bioinformatics 19, , 1808-1816.[Abstract/Free Full Text]

Cormier, S. A., Mello, M. A., and Kappen, C. ((2003). ). Normal proliferation and differentiation of Hoxc-8 transgenic chondrocytes in vitro. BMC Dev. Biol. 3, , 4.[CrossRef][Medline]

De Vries, L., Zheng, B., Fischer, T., Elenko, E., and Farquhar, M. G. ((2000). ). The regulator of G protein signaling family. Annu. Rev. Pharmacol. Toxicol. 40, , 235-271.[CrossRef][Medline]

DeLise, A. M., Fischer, L., and Tuan, R. S. ((2000). ). Cellular interactions and signaling in cartilage development. Osteoarthritis Cartilage 8, , 309-334.[CrossRef][Medline]

Dietz, U. H., and Sandell, L. J. ((1996). ). Cloning of a retinoic acid-sensitive mRNA expressed in cartilage and during chondrogenesis. J. Biol. Chem. 271, , 3311-3316.[Abstract/Free Full Text]

Dudoit, S., and Fridlyand, J. ((2002). ). A prediction-based resampling method for estimating the number of clusters in a dataset. Genome Biol. 3, , RESEARCH0036.

Eames, B. F., de la Fuente, L., and Helms, J. A. ((2003). ). Molecular ontogeny of the skeleton. Birth Defects Res. Part C Embryo Today 69, , 93-101.[CrossRef]

Eyre, D. R. ((2004). ). Collagens and cartilage matrix homeostasis. Clin. Orthop. 427, (suppl), S118-S122.

Fang, C., Carlson, C. S., Leslie, M. P., Tulli, H., Stolerman, E., Perris, R., Ni, L., and Di Cesare, P. E. ((2000). ). Molecular cloning, sequencing, and tissue and developmental expression of mouse cartilage oligomeric matrix protein (COMP). J. Orthop. Res. 18, , 593-603.[CrossRef][Medline]

Gelse, K., Soder, S., Eger, W., Diemtar, T., and Aigner, T. ((2003). ). Osteophyte development—molecular characterization of differentiation stages. Osteoarthritis Cartilage 11, , 141-148.[CrossRef][Medline]

Hall, B. K., and Miyake, T. ((1992). ). The membranous skeleton: the role of cell condensations in vertebrate skeletogenesis. Anat. Embryol. (Berl.) 186, , 107-124.[Medline]

Hepler, J. R. ((1999). ). Emerging roles for RGS proteins in cell signaling. Trends Pharmacol. Sci. 20, , 376-382.[CrossRef][Medline]

Hepler, J. R. ((2003). ). RGS protein and G protein interactions: a little help from their friends. Mol. Pharmacol. 64, , 547-549.[Free Full Text]

Inoue, D., and Matsumoto, T. ((2000). ). Parathyroid hormone-related peptide and bone: pathological and physiological aspects. Biomed. Pharmacother. 54, (Suppl 1), 32s-41s.

Irizarry, R. A., Hobbs, B., Collin, F., Beazer-Barclay, Y. D., Antonellis, K. J., Scherf, U., and Speed, T. P. ((2003). ). Exploration, normalization, and summaries of high density oligonucleotide array probe level data. Biostatistics 4, , 249-264.[Abstract]

Ishii, M., and Kurachi, Y. ((2003). ). Physiological actions of regulators of G-protein signaling (RGS) proteins. Life Sci. 74, , 163-171.[CrossRef][Medline]

Karsenty, G., and Wagner, E. F. ((2002). ). Reaching a genetic and molecular understanding of skeletal development. Dev. Cell 2, , 389-406.[CrossRef][Medline]

Kim, I. S., Otto, F., Zabel, B., and Mundlos, S. ((1999). ). Regulation of chondrocyte differentiation by Cbfa1. Mech. Dev. 80, , 159-170.[CrossRef][Medline]

Knudson, C. B., and Knudson, W. ((2001). ). Cartilage proteoglycans. Semin. Cell Dev. Biol. 12</