Stem Cells
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

First published online January 26, 2006
Stem Cells Vol. 24 No. 3 March 2006, pp. 642 -652
doi:10.1634/stemcells.2005-0270; www.StemCells.com
© 2006 AlphaMed Press

This Article
Right arrow Abstract Freely available
Right arrow Full Text (PDF)
Right arrow Supplemental Tables
Right arrow All Versions of this Article:
2005-0270v1
24/3/642    most recent
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow Reprints/Permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Ylöstalo, J.
Right arrow Articles by Prockop, D. J.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Ylöstalo, J.
Right arrow Articles by Prockop, D. J.
STEM CELL GENETICS AND GENOMICS

Use of Differentiating Adult Stem Cells (Marrow Stromal Cells) to Identify New Downstream Target Genes for Transcription Factors

Joni Ylöstalo, Jason R. Smith, Radhika R. Pochampally, Robert Matz, Ichiro Sekiya, Benjamin L. Larson, Jussi T. Vuoristo, Darwin J. Prockop

Center for Gene Therapy, Tulane University Health Sciences Center, New Orleans, Louisiana, USA

Key Words. Marrow stromal cells • Differentiation • Microarray • Transcription factors

Correspondence: Darwin J. Prockop, M.D., Ph.D., Center for Gene Therapy, Tulane University Health Sciences Center, 1430 Tulane Ave., New Orleans, Louisiana 70112, USA. Telephone: 504-988-7711; Fax: 504-988-7710; e-mail: dprocko{at}tulane.edu

Received on June 15, 2005; accepted for publication on October 17, 2005.


   ABSTRACT
Top
 Abstract
Introduction
 Materials and Methods
 Results
 Discussion
 Disclosures
 References
 
We developed a strategy for use of microarray data to rapidly identify new downstream targets of transcription factors known to drive differentiation by following the time courses of gene expression as a relatively homogeneous population of stem/progenitor cells are differentiated to multiple phenotypes. Microarray assays were used to follow the differentiation of human marrow stromal cells (MSCs) into chondrocytes or adipocytes in three different experimental conditions. The steps of the analysis were the following: (a) hierarchical clustering was used to define groups of similarly behaving genes in each experiment, (b) candidates for new downstream targets of transcription factors that drive differentiation were then identified as genes that were consistently co-expressed with known downstream target genes of the transcription factors, and (c) the list of candidate new target genes was refined by identifying genes whose signal intensities showed a highly significant linear regression with the signal intensities of the known targets in all the data sets. Analysis of the data identified multiple new candidates for downstream targets for SOX9, SOX5, CCAAT/enhancer binding protein (C/EBP)-{alpha}, and peroxisome proliferator-activated receptor (PPAR)-{gamma}. To validate the analysis, we demonstrated that PPAR-{gamma} protein specifically bound to the promoters of four new targets identified in the analyses. The same multistep analysis can be used to identify new downstream targets of transcription factors in other systems. Also, the same analysis should make it possible to use MSCs from bone marrow to define new mutations that alter chondogenesis or adipogenesis in patients with a variety of syndromes.


   INTRODUCTION
Top
 Abstract
 Introduction
Materials and Methods
 Results
 Discussion
 Disclosures
 References
 
Differentiation of eukaryotic cells involves the sequential expression of a large number of transcription factors and an even larger number of their downstream targets. New strategies are clearly required to acquire and interpret all the necessary data. Large amounts of data on expressed genes can now be rapidly acquired using high-density microarrays, and the assays can readily identify genes that are co-expressed in a data set. Recently, data on co-expressed genes have been used to infer that the genes are involved in common processes or have functional or causal relationships. For instance, Lee et al. [1] demonstrated that the evidence for the functional relatedness of genes increases with the number of data sets in which they are co-expressed.

The search for co-expressed genes has recently been facilitated by the development of a multiplicative model for the analysis of microarray data [2]. The model allows for rapid analysis and for estimates of the errors of expression indices [3]. The model was shown to be applicable to data across many arrays in that high and low expression values were adequately fit by the model and produced normally distributed residuals. Therefore, the model makes it possible to identify co-expressed transcripts and to globally calculate the expression indices from several experiments. The global calculation of expression indices makes it possible to introduce a proportionality criterion into the empirical definition of a pair of co-regulated genes.

Stem cells provide convenient model systems for the study of differentiation. However, most stem cells require complicating culture conditions, such as feeder layers for their maintenance and differentiation. Also, differentiation of many stem cells is a heterogeneous process in which the cells differentiate randomly into multiple cell types [4, 5]. Most of these problems can be avoided by using the adult stem cells from bone marrow, referred to as mesenchymal stem cells or marrow stromal cells (MSCs). MSCs are multipotential nonhematopoietic progenitors found in the bone marrow that can be isolated by their adherence to tissue culture plastic. They have been shown to differentiate into many cell types in vitro. In particular, differentiation of MSCs to chondrocytes and adipocytes in culture has been shown to follow time-dependent patterns of gene expression similar to chondrogenesis and adipogenesis in vivo [6–11]. MSCs are inherently free from contaminating cell types, and they can be differentiated en masse into a predictable phenotype. Therefore, they are more amenable than other cell systems to microarray approaches directed at understanding differentiation.

Chondrogenesis has been shown to be driven by a series of transcription factors [12, 13]. Some of the important transcription factors are the SOX proteins that are related to the mammalian testis-determining factor, SRY, in that they contain a similar high-mobility group DNA binding domain. SOX9 is expressed in all chondroprogenitors and differentiated chondrocytes during chondrogenesis [14, 15]. Two other SOX family members, SOX5 and SOX6, are co-expressed with SOX9 during chondrogenesis [16, 17]. The role of SOX9 was illustrated by the observation that heterozygous mutations in the gene or in flanking sequences cause the human dysmorphology syndrome, campomelic dysplasia, which is characterized by hypoplasia of most endochondral bones [18, 19]. SOX9 and SOX5 together drive expression of two genes required for synthesis of the cartilage matrix, COL2A1 and aggrecan 1 [20, 21]. Expression of SOX9 also drives expression of four other genes for cartilage matrix, COL9A1, COL9A2, COL9A3, and COL11A2 [22–25].

Adipogenesis has been shown to be driven by transcription factors that include the genes for the CCAAT/enhancer binding proteins (C/EBPs), the nuclear hormone receptor peroxisome proliferator-activated receptors (PPARs), and the basic helix-loop-helix protein ADD1/SREBP1c [26–28]. C/EBP-{alpha} and PPAR-{gamma} induce changes in gene expression characteristics of mature adipocytes, and they remain elevated for the life of the adipocyte. C/EBP-ß and C/EBP-{delta} induce low levels of PPAR-{gamma} and C/EBP-{alpha}, which induce each other’s expression in a positive feedback loop that promotes and maintains the differentiated state [29, 30]. Recently, it was shown that the adipogenic action of C/EBP-{alpha} is dependent on expression of PPAR-{gamma} [31]. C/EBP-{alpha} is also a strong inhibitor of cell proliferation by directly interacting with and inhibiting cyclin-dependent kinase 2 (CDK2) and CDK4 [32]. In addition, it has been shown that PPAR-{gamma} regulates the expression of genes for lipoprotein lipase, phosphoenolpyruvate carboxykinase 1, and nuclear receptor subfamily 1H3 [33–35].

In the present study, we re-analyzed previously generated microarray data on the time courses of gene expression as human MSCs were differentiated into chondrocytes or adipocytes in three different experimental conditions [10, 11]. We used the multistep analysis summarized in Figure 1Go to identify a series of candidate genes for downstream targets for transcription factors known to drive chondrogenesis and adipogenesis.


Figure 1
View larger version (26K):
[in this window]
[in a new window]
  		Figure 1. Schematic of the strategy. (A): Genes that were co-expressed in one profile in one experiment were selected, and then the data were queried to determine whether the same genes were classified together into any profile of the other two experiments. The groups of consistently co-expressed genes that contained known downstream targets for transcription factors were selected, and the new target genes in these groups were searched for ones that had highly significant linear regressions of their signal intensities with the signal intensities of known target genes. Selected new downstream targets were subjected to promoter analysis. (B): Schematic for the identification of consistently co-expressed genes in the three experiments. Abbreviation: PPAR, peroxisome proliferator-activated receptor.

 

   MATERIALS AND METHODS
Top
 Abstract
 Introduction
 Materials and Methods
Results
 Discussion
 Disclosures
 References
 
Preparation and Differentiation of MSCs into Chondrocytes and Adipocytes
To isolate human MSCs, 2–10 ml bone marrow aspirates were taken from the iliac crest of normal adult donors after informed consent and under a protocol approved by the Institutional Review Board. Nucleated cells were isolated with a density gradient (Ficoll-Paque; Amersham Biosciences, Piscataway, NJ, http://www.amersham.com) and resuspended in complete culture medium (CCM) that consisted of {alpha}-minimal essential medium ({alpha}-MEM; Gibco-BRL, Carlsbad, CA, http://www.gibco-brl.com), 17% fetal bovine serum (FBS) (lot-selected for rapid growth of MSCs; Atlanta Biologicals, Lawrenceville, GA, http://www.atlantabio.com), 100 units per ml penicillin, 100 µg/ml streptomycin, and 2 mM L-glutamine (Gibco-BRL). All of the nucleated cells were plated in 25 ml of CCM in 145-cm2 culture dish and incubated at 37°C with 5% CO2. After 24 hours, nonadherent cells were discarded, and adherent cells were thoroughly washed twice with phosphate-buffered saline (PBS). The cells were incubated for 5–7 days in CCM, harvested at approximately 70% confluency with 0.25% trypsin and 1 mM EDTA for 5 minutes at 37°C, and replated at 6 cells per cm2 in an intercommunicating system of culture flasks (6,320 cm2; Cell Factory, Nunc, Rochester, NY, http://www.nuncbrand.com). After 12 days, the cells (passage 1) were harvested with trypsin/EDTA, suspended at 1 x 106 cells per ml in 5% DMSO and 30% FBS, and frozen in 1 ml aliquots at –80oC overnight before storage in liquid nitrogen. To expand a culture, a frozen vial of MSCs was thawed, plated in a 58-cm2 culture dish, and incubated for 4 days to recover viable cells (passage 2). The cells were harvested and diluted for further expansion by plating at initial densities of 6 or 50 cells per cm2 in 175-cm2 culture dishes. The cells were harvested after 7 days (passage 3). To select MSCs for chondrogenesis, preparations from 20 different bone marrow aspirates were differentiated into chondrocytes using the bone morphogenic protein (BMP)-6 protocol [9] and then screened for size of cartilage pellets formed and for expression of COL2A1 and COMP. All 20 preparations generated pellets that expressed COL2A1, 18 generated pellets that expressed both COL2A1 and COMP, and two generated pellets that expressed both genes and that were larger than the pellets generated by the other preparations (>2 mm in diameter). MSCs from one of the preparations that generated the largest pellet were used for chondrogenesis experiments.

For chondrocytic differentiation of MSCs [10] in the presence of BMP-6 (chondrogenesis [Ch]-6 experiment), samples were prepared by placing approximately 200,000 MSCs (passage 3) in a 15-ml polypropylene tube (Falcon, BD Biosciences, Bedford, MA, http://www.bdbiosciences.com) and centrifuging the cells to form a pellet (day 0). The pellet was cultured at 37°C with 5% CO2 in 500 µl of chondrogenic media that contained 500 ng/ml BMP-6 (R&D Systems, Minneapolis, http://www.rndsystems.com) in high-glucose Dulbecco’s modified Eagle’s medium supplemented with 10 ng/ml transforming growth factor-ß-3, 10–7 M dexamethasone, 50 µg/ml ascorbate-2-phosphate, 40 µg/ml proline, 100 µg/ml pyruvate, and 50 mg/ml ITS + Premix (6.25 µg/ml transferrin, 6.25 ng/ml selenious acid, 1.25 mg/ml bovine serum albumin, and 5.35 mg/ml linoleic acid; Becton, Dickinson and Company, Franklin Lakes, NJ, http://www.bd.com). The medium was replaced every 3–4 days for 21 days. To confirm chondrogenesis, randomly selected pellets were embedded in paraffin, cut into 5-µm sections, and stained with toluidine blue sodium borate to confirm the synthesis of proteoglycans [10]. For chondrocytic differentiation in the presence of BMP-2 (Ch-2 experiment), the same conditions were employed except the BMP-6 in the medium was replaced with 500 ng/ml BMP-2 (R&D Systems).

For adipogenic differentiation (Adipo experiment), the samples were prepared by plating 100 (passage 3) MSCs in a 60-cm2 dish in CCM for 7 days with a change in medium after 3 days [11]. The medium was replaced with adipogenic medium, and the cells were cultured for an additional 21 days. The adipogenic medium consisted of CCM supplemented with 0.5 µM dexamethasone (Sigma, St. Louis, http://www.sigmaaldrich.com), 0.5 mM isobutylmethylxanthine (Sigma), and 50 µM indo-methacin (Sigma). To confirm adipogenesis, randomly selected adipogenic cultures were washed three times with PBS, fixed in 10% formalin for more than 1 hour, and stained with fresh Oil Red-O solution for 2 hours to visualize lipid vacuoles [11]. Control samples for adipogenesis were prepared by placing 100 (passage 3) MSCs in a 60-cm2 dish and incubating them in CCM with a change in medium every 3 days.

RNA Isolation and Microarray
For Ch-6 and Ch-2 experiments, total RNA was isolated from 2 million undifferentiated MSCs at day 0, from 10 cartilage pellets at day 1. Because the yields of RNA decreased as the pellets increased in size, 30 pellets were used to extract RNA for each of the samples from days 7, 14, and 21. Also, pellets incubated 7 days or longer were digested with 3 mg/ml collagenase, 1 mg/ml hyaluronidase, and 0.25% trypsin for about 3 hours at 37°C to remove matrix proteins. Total RNA was extracted by using a commercial kit (RNAqueous Kit; Ambion, Austin, TX, http://www.ambion.com). For Adipo samples and control samples, total RNA was extracted directly from pooled samples of approximately 2 million cells. Experimental procedures for microarray assays were performed according to the Affymetrix GeneChip Expression Analysis Technical Manual (Affymetrix, Santa Clara, CA, http://www.affymetrix.com). In brief, the quality of the RNA samples was first established by assays for prokaryotic genes that were spiked into the samples (Eukaryotic Poly-A RNA Control Kit; Affymetrix). Samples of approximately 5 µg of total RNA were then used to synthesize double-stranded DNA (Superscript Choice System/Gibco-BRL). The DNA was purified using phenol/chloroform extraction (Phase Lock Gel; Eppendorf Scientific, Westbury, NY, http://www.eppendorfsci.com) and concentrated by ethanol precipitation. In vitro transcription was performed to produce biotin-labeled cRNA (BioArray HighYield RNA Transcription Labeling Kit; Enzo Diagnostics, Farmingdale, NY, http://www.enzobio.com). Biotinylated cRNA was cleaned (Rneasy Mini Kit; Qiagen, Valencia, CA, http://www1.qiagen.com), fragmented to 50–200 nucleotides, and hybridized 16 hours at 45°C to a microarray (Affymetrix HG-U95Av2 or HG-U95A), which contained approximately 12,600 human genes. After washing, the array was stained with streptavidin-phycoerythrin (Molecular Probes, Inc., Eugene, OR, http://probes.invitrogen.com). The staining was amplified by biotinylated anti-streptavidin (Vector Laboratories, Burlingame, CA, http://www.vectorlabs.com) followed by streptavidin-phycoerythrin and then scanned (HP GeneArray Scanner; Affymetrix).

Image Acquisition and Filtering
The microarrays from the three experimental conditions were scanned with Microarray Suite 5.0 (MAS5.0; Affymetrix), and the images were transferred to the dChip1.3+ program [2, 3]. MAS5.0 recorded intensity values for perfect match (PM) and mismatch (MM) oligonucleotides and also assigned present (P), marginal (M), or absent (A) calls. The intensity values were normalized against the array with median overall intensity (154 in this study) by dChip1.3+. The model-based expression indices were calculated using the PM-MM algorithm, and negative values were assigned the value of zero. The genes from each experimental condition were filtered separately to obtain differentially expressed genes. A gene was considered differentially expressed over a time course if (a) the expression of a gene was scored as present in at least one time point in the experiment by MAS5.0 and (b) the variation in the expression of a gene across all the samples from one experiment (five samples) was significant as reflected by the coefficient of variation (i.e., the SD/ mean of the expression indices was greater than 0.3). After filtering, 1,912 genes were differentially expressed in the data from the Ch-6 experiment; 2,901 from the Ch-2 experiment; and 2,686 from the Adipo experiment.

Hierarchical Clustering
The dChip 1.3+ program was used to obtain standardized values from the normalized values of the differentially expressed genes by linearly scaling the values from each time course of expression to a mean of zero and an SD of one. The dChip 1.3+ program was then used for hierarchical clustering of genes and samples.

Gene Ontologies
Ten distinct profiles for the Ch-6 experiment, 20 for the Ch-2 experiment, and 13 for the Adipo experiment were defined based on the hierarchical clustering result. The genes in these profiles were analyzed for gene ontology (GO) terms to obtain information on the cellular component, biological process, and molecular function of the protein associated with the gene [36]. The dChip 1.3+ program calculated p values for each GO term using an exact hyper-geometric distribution to compare the frequencies of individual GO terms (GeneOntology Consortium, http://www.geneontology.org) within the profile with the frequencies of those terms on the entire microarray (p ≤ .01 was considered significant).

Identification of Candidate New Targets for Transcription Factors
We identified genes that were consistently co-expressed by selecting genes that were co-expressed in one profile in one experiment and also co-expressed in any profile of the other two experiments. There were more than 100 groups that contained from two to more than 150 co-expressed genes each.

To find candidate new downstream targets for the four transcription factors known to drive either chondrogenesis (SOX5 and SOX9) or adipogenesis (PPAR-{gamma} and C/EBP-{alpha}), we searched within groups of genes that were consistently co-expressed for genes that showed highly significant linear regressions (R2 ≥ 0.9) with the known target genes within the same group of consistently co-expressed genes. Known targets of SOX9 were SOX5, COL9A1, COL9A2, COL9A3, COL2A1, COL11A, and aggrecan 1. Known targets of SOX5 were COL2A1 and aggrecan 1. Known targets of C/EBP-{alpha} were PPAR-{gamma} and CDK2. Known targets of PPAR-{gamma} were LPL, phosphoenolpyruvate carboxykinase 1, and nuclear receptor subfamily 1H3.

To identify consensus sequences for transcription factors in promoters, we examined 10 kb of the promoters with LocusLink (http://www.ncbi.nlm.nih.gov/LocusLink/) number and the Traser database (http://genome-www6.stanford.edu/cgi-bin/Traser/traser). The upstream sequences were searched for putative binding sites for SOX9, using the SOX9 function in the ConSite Web site (http://mordor.cgb.ki.se/cgi-bin/CONSITE/consite). Similarly, the SOX5 targets were examined for binding sites using the Sox5 function (mouse gene). C/EBP-{alpha} targets were searched for binding sites using the cEBP function. PPAR-{gamma} targets were examined for binding sites using the PPAR-{gamma} and PPAR-{gamma}-RXR-{alpha}-l (complex) functions.

Electrophoretic Mobility Shift Assays
A known target (LPL) and four new targets (DAT1, serum amyloid A1, glycogenin 1, and FABP4) for PPAR-{gamma} were selected for transcription factor binding confirmation with electrophoretic mobility shit assay (EMSA). The highest-scored binding site for PPAR-{gamma}-RXR-{alpha}-1 was selected for each gene, and 35-bp oligonucleotides spanning these sites were designed. The oligonucleotides for LPL were (AAATTTTTCCGTCTGC-CCTTTCCCCCTCTTCTCGTTGGCA), for DAT1 were (CA-GTTGCCAGCGAGGGGTAACAGATCATACAGTTGGAGGG), for serum amyloid A1 were (TTAAATAAATCCTCCTCCTTT-GACCTTCGCATGTATTCAG), for glycogenin 2 were (TGTATG-CATGAATTCACCTTTCACCCATTCATGCACTATG), and for FABP4 were (ACACACACAAAATAAGGTCGAAGTTTA-TCTCAAAATAATT). The oligonucleotides were radiolabeled using the manufacturer’s suggested protocols (Starfire Probe Labeling Kit; IDT DNA Technologies, Coralville, IA, http://www.idtdna.com) and [32P]-ATP (Amersham Biosciences). The radioactivity was measured by scintillation counter, and approximately 100,000 counts were used per reaction. The binding assay was performed using the manufacturer’s suggested protocols for NuShift PPAR-{alpha} (Active Motif, Carlsbad, CA, http://www.activemotif.com). In brief, the probes were incubated with nuclear extracts from THP-1 (human monocytic leukemia) cell line. Cold (unlabeled) oligonucleotide was added at a 1:10 ratio, and antibody for PPAR-{alpha} was used in supershift assays. After a 20-minute incubation at room temperature, the reaction was separated by electrophoresis on 6% polyacrylamide gels with 2% glycerol in 89 mM Tris-borate/2 mm EDTA buffer (pH 8.5). The gels were vacuum-dried and exposed to x-ray films for varying times. The antibody to PPAR-{alpha} was polyclonal, and the lanes with supershift bands were either merged from a different exposures of the same gel or digitally processed (Adobe Photoshop CS; Adobe, San Jose, CA, http://www.adobe.com) to enhance presentation of the bands.


   RESULTS
Top
 Abstract
 Introduction
 Materials and Methods
 Results
Discussion
 Disclosures
 References
 
Three Experimental Conditions for Differentiation of Human MSCs
Human MSCs were differentiated under three conditions: (a) culture as micropellets in a serum-free medium containing BMP-6 and other components previously shown to promote chondrogenic differentiation (Ch-6 experiment), (b) culture as micropellets under the same conditions except that the BMP-6 was replaced by BMP-2 (Ch-2 experiment), and (c) cultures in which adherent log-phase cells were transferred to adipogenic medium (Adipo experiment). Control samples were cultured under the same conditions as the Adipo experiment except that the medium was CCM throughout. Previous reports demonstrated that under Ch-6 conditions, the cells formed firm pellets of chondrocytes and cartilage matrix that stained for proteoglycans surrounded by a thin layer of fibrous tissue [9, 10]. The differentiation was confirmed by reverse transcriptase polymerase chain reaction (PCR) assay that demonstrated time-dependent increases in expression of COL2A1, COL10A1, COMP, Link, SOX4, SOX5, SOX6, SOX9, PTHrP, PTHrP-R, BMP-2, and BMP-6 [9, 10]. Under the Ch-2 conditions, the results were the same as in the Ch-6 experiments except that the cartilage pellets were about twice as large (3.4 mg ± 0.32 SD vs. 1.8 mg ± 0.07 SD), indicating that BMP-2 enhanced chondrogenesis. Under the Adipo conditions, the cells became filled with lipid vacuoles that stained with Oil Red-O and expressed a series of genes characteristic of adipocytes [11]. The differentiation was confirmed by reverse transcriptase PCR assays that demonstrated time-dependent increases in the expression of the genes for PPAR-{gamma}, RXR-{alpha}, C/EBP-{alpha}, ACS, LPL, and FABP4 [11]. The genes that changed in levels of expression were generally similar to those identified by Hung et al. [37], who used microarray assays to compare human MSCs before and after differentiation into adipocytes, but because a cDNA microchip was used instead of the oligonucleotide microchip employed here, a detailed comparison was not feasible.

For each of the three experiments, total RNA was extracted on days 0, 1, 7, 14, and 21 after transfer of the cells to the differentiating conditions. The RNA was then assayed with microarrays, and the data were filtered to identify genes that demonstrated significant differences in expression in each experiment (see Materials and Methods).

Demonstration That the Three Experiments Provided Distinct Data Sets
To determine that the three experimental conditions provided distinct sets of expression data, the normalized values of the signal intensities of the differentially expressed genes were hierarchically clustered using dChip 1.3+ [2, 3]. The data consisted of a total of 33 microarray assays: (a) four replicates of the five time points from the Ch-6 experiment, (b) five time points from the Ch-2 experiment, (c) five time points from the Adipo experiment, and (d) three additional control samples for days 7, 14, and 21.

As expected, the day-0 samples from all three experiments clustered together (Fig. 2Go). In the two chondrogenesis experiments, data from day 1 and day 7 clustered together, but the data from day 14 and day 21 appeared in separate clusters. Therefore, the results indicated that there were differences in the final stages of differentiation with substitution of BMP-6 with BMP-2 in Ch-2 experiments. For the Adipo experiment, the data from day 1 clustered with the merged super-sample of three day-0 samples. However, the Adipo day-7, -14, and -21 data clustered separately from the other samples.


Figure 2
View larger version (14K):
[in this window]
[in a new window]
  		Figure 2. Hierarchial clustering of all the samples using differentially expressed genes from three experiments (Ch-6, Ch-2, and Adipo) and a control experiment with marrow stromal cells (Con). Abbreviation: Ch, chondrogenesis.

 
Genes Co-Expressed in the Same Time-Dependent Profiles
Hierarchical clustering of the data from each of the three experiments indicated that the patterns of gene expression could be described by a limited number of time-dependent profiles. Ten distinct profiles were defined for the Ch-6 experiment, 20 for the Ch-2 experiment, and 13 for the Adipo experiment (Fig. 3Go). The profiles assigned to each of the three experiments accounted for more than 97% of the filtered genes. In each experiment, the profile with the largest number of genes had a peak of expression on day 0 followed by decreases thereafter (profile A in Ch-6, profile a in Ch-2, and profiles 1 and 2 in Adipo). Also, a large number of genes fell within profiles that showed progressive increases during the terminal stages of differentiation (profile H in Ch-6, profiles l and m in Ch-2, and profile 9 in Adipo).


Figure 3
View larger version (51K):
[in this window]
[in a new window]
  		Figure 3. Hierarchial clustering of differentially expressed genes. Red represents expression level above mean expression of a gene across all samples, white represents expression at the mean level, and blue represents expression lower than the mean. Co-expressed genes were defined from the clustering picture, and average profiles are shown (identified as A-J for Ch-6, a-t for Ch-2, and 1–13 for Adipo) on the left of the clustering picture. In the picture, a row represents a gene, and each column represents a sample from the time course (days 0, 1, 7, 14, and 21). Abbreviation: Ch, chondrogenesis.

 
Gene Ontologies
The dChip 1.3+ program [2, 3] was used to carry out an analysis of GO terms on co-expressed genes [36]. Of the 12,600 total genes on the microarray, 7,699 genes were annotated in dChip. The number of differentially expressed genes varied from 1,912 to 2,901 in the three experiments, but the fraction of genes for transcription factors stayed relatively constant at 5.3% to 5.49% of the total. For the Ch-6 (supplemental online Table 1) and Ch-2 (supplemental online Table 2) experiments, the genes that decreased after day 0 were primarily cell cycle and nuclear division genes (profile A in Ch-6 and profile a in Ch-2). The results were consistent with previous observations that MSCs in the micropellets rapidly cease dividing [10]. For adipogenesis (supplemental online Table 3), one cluster of similar genes decreased between day 1 and day 7 (Adipo, profile 2). The results were consistent with the observation that proliferation continued for a few days in the mid-log phase MSCs used to initiate adipogenesis [11]. As expected, the genes expressed at increasing levels in the terminal stages of chondrogenesis were significantly enriched for genes involved in skeletal development and extracellular matrix (profile H in Ch-6 and profiles l and m in Ch-2). The genes expressed in the terminal stages of adipogenesis were significantly enriched for genes involved in lipid metabolism (profile 9 in Adipo).

Identification of Candidates for New Downstream Targets for Known Transcription Factors
We then identified genes that were consistently co-expressed in all three data sets by selecting genes that were co-expressed in one profile in one experiment and determined whether the same genes were co-expressed in any profile of the other two experiments (Fig. 1Go). The results identified more than 100 clusters of consistently co-expressed genes that contained from two to more than 150 genes each (supplemental online Table 4).

Next, we searched the data for known downstream targets of transcription factors previously shown to drive either chondrogenesis or adipogenesis. For chondrogenesis, we chose downstream targets for the transcription factor SOX9, which was previously shown to have seven downstream targets: SOX5, aggrecan 1, COL2A1 COL9A1, COL9A2, COL9A3, and COL11A2 (Fig. 4AGo). To identify new targets, we searched within clusters of consistently co-expressed genes for genes that showed highly significant linear regression (R2 ≥ 0.9, and significant slope, Fisher’s exact test, p < .005) with seven known target genes. Seventy-one new candidate target genes were identified (supplemental online Table 5). We then used the same approach to search for new downstream targets for SOX5 by searching for linear regression in expression with COL2A1 and aggrecan 1, known downstream targets of SOX5 (Fig. 4AGo). Ten new candidate targets were found (supplemental online Table 5). We used a similar strategy to search for new downstream targets for transcription factors that drive adipogenesis. First we searched for genes that showed linear regression in expression with two known downstream targets of C/EBP-{alpha}: PPAR-{gamma} and CDK2 (Fig. 4BGo). Ten new candidates were found (supplemental online Table 6). Then we searched for new targets for PPAR-{gamma} by testing for linear regressions with three of its known downstream targets: lipoprotein lipase, phosphoenol-pyruvate carboxykinase 1, and nuclear receptor subfamily 1H3 (Fig. 4BGo). Five new targets were found supplemental online Table 6). Examples of the linear regression data with known downstream targets of SOX9, C/EBP-{alpha}, and PPAR-{gamma} with selected new candidate downstream targets are shown in Figure 5Go.


Figure 4
View larger version (25K):
[in this window]
[in a new window]
  		Figure 4. Schematic for identifying downstream targets. (A): Transcription factors that drive chondrogenesis, their known downstream targets, and possible new targets identified here. Symbols: H-l-Ø, Profile H from Ch-6 differentiation, Profile l from Ch-2 differentiation, and no Profile (not differentially expressed) from Adipo differentiation; H-r-Ø, Profile H from Ch-6, Profile r from Ch-2, and no Profile from Adipo; Ø-m-Ø, Profile m from Ch-2, and no Profiles from Ch-6 or Adipo. Lines originating from the known targets represent the new targets with the number indicating the multitude of the new targets consistently co-expressed with the known target. (B): Transcription factors that drive adipogenesis, their known downstream targets, and possible new targets identified here. Symbols: H-Ø-9, Profile H from Ch-6 differentiation, Profile 9 from Adipo differentiation, and no Profile from Ch-2 differentiation; A-a-3, Profile A from Ch-6, Profile a from Ch-2, and Profile 3 from Adipo; Ø-n-9, Profile n from Ch-2, Profile 9 from Adipo, and no Profile from Ch-6; I-f-Ø, Profile I from Ch-6, Profile f from Ch-2, and no Profile from Adipo; H-l-9, Profile H from Ch-6, Profile l from Ch-2, and Profile 9 from Adipo. Abbreviations: CDK, cyclin-dependent kinase; C/EBP, CCAAT/enhancer binding protein; Ch, chondrogenesis; PEP, phosphoenolpyruvate carboxykinase; PPAR, peroxisome proliferator-activated receptor.

 

Figure 5
View larger version (26K):
[in this window]
[in a new window]
  		Figure 5. Graphs of representative linear correlations between the known downstream targets of SOX9, C/EBP-{alpha}, and PPAR-{gamma}, and candidate new downstream targets. Signal intensities of the new targets were plotted against the signal intensities of the known targets (all 15 samples from three experiments). Abbreviations: C/EBP, CCAAT/enhancer binding protein; PPAR, peroxisome proliferator-activated receptor.

 
Validation of Four of the New Downstream Targets
To validate the analysis, we selected a few of the candidates for new downstream targets and used the ConSite program to search 10 kb of their promoter sequences for binding sites for the known transcription factors: SOX9, SOX5, C/EBP-{alpha}, and PPAR-{gamma}. As indicated in Table 1Go, a large number of the selected new targets had binding sites in their promoters that had both a high relative score and a high absolute score based on the model used in the program. The scores of a number of the candidate new targets were equal to or greater than the scores for binding sites in the promoters of the known downstream targets of the same transcription factors.


View this table:
[in this window]
[in a new window]
  		Table 1. Selected downstream targets with transcription factor binding sites

 
To validate the analysis more directly, we carried out EMSAs with nuclear extracts from cells known to express PPAR-{gamma} and oligonucleotides with sequences for binding sites for PPAR-{gamma} in the promoters of one known downstream target (lipoprotein lipase, LPL) and four candidate new downstream targets: neuronal transcription factor DAT1 (DAT1), serum amyloid A1, glycogenin 2, and fatty acid binding protein 4 (FABP4). As indicated in Figure 6Go, the results established that the known downstream target and all four candidate downstream targets had promoter sequences that specifically bound PPAR-{gamma}.


Figure 6
View larger version (58K):
[in this window]
[in a new window]
  		Figure 6. Electrophoretic mobility shift assays. The oligonucleotides spanning PPAR-{gamma} binding sites were incubated with nuclear extracts from THP-1 (human monocytic leukemia) cell line. Cold oligonucleotide was added at a 1:10 ratio, and antibody for PPAR-{gamma} was used for supershift assays. LPL is a known target for PPAR-{gamma}, and DAT1, FABP4, and serum amyloid A1 are new targets based on the transcriptome analysis. Antibody to PPAR-{gamma} used here is polyclonal, and the lanes with supershift bands were digitally processed (Adobe Photoshop CS; Adobe) to enhance viewability of existing bands.

 

   DISCUSSION
Top
 Abstract
 Introduction
 Materials and Methods
 Results
 Discussion
Disclosures
 References
 
The analytical strategy employed here is similar to the strategy Lee et al. [1] recently developed to analyze multiple data sets of archived data to identify co-expressed genes. Lee et al. [1] addressed the ambiguities inherent in hierarchical clustering by defining co-expressed genes according to their correlation within a data set and then counting the number of data sets in which the two genes are co-expressed. Using this method, the authors developed a general organizing principle similar to hierarchical clustering that does not have the same drawbacks. We decided to apply a similar approach to data obtained from the time courses of differentiation of adult stem cells into two distinct cellular phenotypes.

The advantage of the analytical strategy employed here was that it enabled us to develop simple and increasing restrictive tests to identify co-expressed genes and then to use the lists of co-expressed genes to identify new downstream targets of transcription factors previously shown to drive chondrogenesis and adipogenesis (Fig. 1Go). The initial test for co-expressed genes was to identify genes that were expressed in the same time-dependent profile in each experiment. The second test was to identify consistently co-expressed genes that were co-expressed across all three experiments. Use of data from both adipogenic and chondrogenic differentiation in the second test greatly increased the rigor of the strategy because it excluded genes that were fortuitously co-expressed under one set of conditions for differentiation. At this point, the vast literature on chondrogenesis and adipogenesis was used to search the list of consistently co-expressed genes for known downstream targets of transcription factors previously shown to drive chondrogenesis and adipogenesis. Then, new candidates for downstream targets were identified in the data as genes that were consistently co-expressed with the known downstream targets of the same transcription factors. The list of new candidates for downstream targets was further refined by requiring a highly significant linear regression of their signal intensities with the signal intensities the known downstream targets of the transcription factors. To validate the analysis, we demonstrated that a number of the new candidate downstream targets had binding sites in the promoters for the transcription factors and that the promoters of four of them specifically bound the putative transcription factor. Although EMSA data do not in themselves provide definitive data, the results are highly suggestive that the strategy did in fact identify new targets for the transcription factors.

The strategy employed here can probably be applied to other systems of differentiation in mammalian cells. It offers a means of identifying downstream targets of transcription factors much more rapidly than the classic methods of mutation analysis or of tracking back from analysis of promoters to the transcription factors. Also, as pointed out previously [38], transcriptome profiling with or without the additional tests employed here can detect genes not essential for a pathway but that provide a "buffering capacity" for loss of associated genes. In addition, it supplements classic strategies and may detect additional later-stage functions of genes whose mutations are lethal early in development. For example, the strategy may make it possible to detect the late-stage functions of genes such as tolloid, which produces lethal pattern defects in embryonic Drosophila, Xenopus, and zebra fish [39–41] but which also provides a critical enzyme for processing of procollagen later in vertebrate development [42].

The application of the strategy to follow chondrogenesis using human MSCs may have a special application to detect and study new mutations causing cartilage disorders such as chondrodysplasias and osteoarthritis [43]. Samples of cartilage can rarely be obtained from patients, and the tissue is not amenable to genetic manipulations. In contrast, human MSCs are readily generated from small amounts of bone marrow aspirates and are readily differentiated into cartilage under the conditions used. Therefore, examining chrondrogenesis with MSCs from patients with the strategy presented here might be highly informative. For example, functional mutations in SOX9, SOX5, or PPAR-{gamma} should be detectable by a failure to detect mRNAs for the downstream genes in micromass cultures of MSCs. The data might also help explain some of the clinical heterogeneity seen with some mutations such as the highly diverse forms of campomelic dysplasia caused by mutations in SOX9 [18, 19, 44, 45].


   DISCLOSURES
Top
 Abstract
 Introduction
 Materials and Methods
 Results
 Discussion
 Disclosures
References
 
D.J.P. indicates a financial interest in FibroGen.


   ACKNOWLEDGMENTS
Top
 Abstract
 Introduction
 Materials and Methods
 Results
 Discussion
 Disclosures
 References
 
This work was supported in part by NIH (AR47796 and AR45323), the Oberkotter Foundation, W.M. Keck Foundation, HCA the Health Care Company, and the Louisiana Gene Therapy Research Consortium. J.R.S. is supported by a Louisiana Board of Regents Support Fund Fellowship. J.Y. and J.R.S. contributed equally to this work.


   REFERENCES
Top
 Abstract
 Introduction
 Materials and Methods
 Results
 Discussion
 Disclosures
 References
 

  1. Lee HK, Hsu AK, Sajdak J et al. Coexpression analysis of human genes across many microarray data sets. Genome Res 2004;14:1085–1094.[Abstract/Free Full Text]

  2. Li C, Wong WH. Model-based analysis of oligonucleotide arrays: Expression index computation and outlier detection. Proc Natl Acad Sci U S A 2001;98:31–36.[Abstract/Free Full Text]

  3. Li C, Wong WH. DNA-chip analyzer (dChip). In: Parmigiani G, Garrett ES, Irizarry R et al., eds. The Analysis of Gene Expression Data: Methods and Software. New York: Springer-Verlag New York, Inc., 2003.

  4. Loebel DA, Watson CM, De Young RA et al. Lineage choice and differentiation in mouse embryos and embryonic stem cells. Dev Biol 2003;264:1–14.[CrossRef][Medline]

  5. Pevny L, Rao MS. The stem-cell menagerie. Trends Neurosci 2003;26:351–359.[CrossRef][Medline]

  6. Mackay AM, Beck SC, Murphy JM et al. Chondrogenic differentiation of cultured human mesenchymal stem cells from marrow. Tissue Eng 1998;4:415–428.[Medline]

  7. Owen M, Friedenstein AJ. Stromal stem cells: Marrow-derived osteogenic precursors. Ciba Found Symp 1988;136:42–60.[Medline]

  8. Prockop DJ. Marrow stromal cells as stem cells for nonhematopoietic tissues. Science 1997;276:71–74.[Abstract/Free Full Text]

  9. Sekiya I, Colter DC, Prockop DJ. BMP-6 enhances chondrogenesis in a subpopulation of human marrow stromal cells. Biochem Biophys Res Commun 2001;284:411–418.[CrossRef][Medline]

  10. Sekiya I, Vuoristo JT, Larson BL et al. In vitro cartilage formation by human adult stem cells from bone marrow stroma defines the sequence of cellular and molecular events during chondrogenesis. Proc Natl Acad Sci U S A 2002;99:4397–4402.[Abstract/Free Full Text]

  11. Sekiya I, Larson BL, Vuoristo JT et al. Adipogenic differentiation of human adult stem cells from bone marrow stroma (MSCs). J Bone Miner Res 2004;19:256–264.[CrossRef][Medline]

  12. Erlebacher A, Filvaroff EH, Gitelman SE et al. Toward a molecular understanding of skeletal development. Cell 1995;80:371–378.[CrossRef][Medline]

  13. de Crombrugghe B, Lefebvre V, Behringer RR et al. Transcriptional mechanisms of chondrocyte differentiation. Matrix Biol 2000;19:389–394.[CrossRef][Medline]

  14. Ng LJ, Wheatley S, Muscat GE et al. SOX9 binds DNA, activates transcription, and coexpresses with type II collagen during chondrogenesis in the mouse. Dev Biol 1997;183:108–121.[CrossRef][Medline]

  15. Zhao Q, Eberspaecher H, Lefebvre V et al. Parallel expression of Sox9 and Col2a1 in cells undergoing chondrogenesis. Dev Dyn 1997;209:377–386.[CrossRef][Medline]

  16. Lefebvre V, Li P, de Crombrugghe B. A new long form of Sox5 (L-Sox5), Sox6 and Sox9 are coexpressed in chondrogenesis and cooperatively activate the type II collagen gene. EMBO J 1998;17:5718–5733.[CrossRef][Medline]

  17. Lefebvre V, Behringer RR, de Crombrugghe B. L-Sox5, Sox6 and Sox9 control essential steps of the chondrocyte differentiation pathway. Osteoarthritis Cartilage 2001;9(suppl A):S69–S75.[Medline]

  18. Wagner T, Wirth J, Meyer J et al. Autosomal sex reversal and campomelic dysplasia are caused by mutations in and around the SRY-related gene SOX9. Cell 1994;79:1111–1120.[CrossRef][Medline]

  19. Foster JW, Dominguez-Steglich MA, Guioli S et al. Campomelic dysplasia and autosomal sex reversal caused by mutations in an SRY-related gene. Nature 1994;372:525–530.[CrossRef][Medline]

  20. Bell DM, Leung KK, Wheatley SC et al. SOX9 directly regulates the type-II collagen gene. Nat Genet 1997;16:174–178.[CrossRef][Medline]

  21. Lefebvre V, Huang W, Harley VR et al. SOX9 is a potent activator of the chondrocyte-specific enhancer of the pro alpha1(II) collagen gene. Mol Cell Biol 1997;17:2336–2346.[Abstract]

  22. Bridgewater LC, Lefebvre V, de Crombrugghe B. Chondrocyte-specific enhancer elements in the Col11a2 gene resemble the Col2a1 tissue-specific enhancer. J Biol Chem 1998;273:14998–15006.[Abstract/Free Full Text]

  23. de Crombrugghe B, Lefebvre V, Nakashima K. Regulatory mechanisms in the pathways of cartilage and bone formation. Curr Opin Cell Biol 2001;13:721–727.[CrossRef][Medline]

  24. Xie WF, Zhang X, Sakano S et al. Trans-activation of the mouse cartilage-derived retinoic acid-sensitive protein gene by Sox9. J Bone Miner Res 1999;14:757–763.[CrossRef][Medline]

  25. Zhang P, Jimenez SA, Stokes DG. Regulation of human COL9A1 gene expression. Activation of the proximal promoter region by SOX9. J Biol Chem 2003;278:117–123.[Abstract/Free Full Text]

  26. Cao Z, Umek RM, McKnight SL. Regulated expression of three C/EBP isoforms during adipose conversion of 3T3–L1 cells. Genes Dev 1991; 5:1538–1552.[Abstract/Free Full Text]

  27. Rosen ED, Walkey CJ, Puigserver P et al. Transcriptional regulation of adipogenesis. Genes Dev 2000;14:1293–1307.[Free Full Text]

  28. Yeh WC, Cao Z, Classon M et al. Cascade regulation of terminal adipocyte differentiation by three members of the C/EBP family of leucine zipper proteins. Genes Dev 1995;9:168–181.[Abstract/Free Full Text]

  29. Lin FT, Lane MD. CCAAT/enhancer binding protein alpha is sufficient to initiate the 3T3–L1 adipocyte differentiation program. Proc Natl Acad Sci U S A 1994;91:8757–8761.[Abstract/Free Full Text]

  30. Tontonoz P, Hu E, Spiegelman BM. Stimulation of adipogenesis in fibroblasts by PPAR gamma 2, a lipid-activated transcription factor. Cell 1994;79:1147–1156.[CrossRef][Medline]

  31. Rosen ED, Hsu CH, Wang X et al. C/EBPalpha induces adipogenesis through PPARgamma: A unified pathway. Genes Dev 2002;16:22–26.[Abstract/Free Full Text]

  32. Wang H, Iakova P, Wilde M et al. C/EBPalpha arrests cell proliferation through direct inhibition of Cdk2 and Cdk4. Mol Cell 2001;8:817–828.[CrossRef][Medline]

  33. Chawla A, Boisvert WA, Lee CH et al. A PPAR gamma-LXR-ABCA1 pathway in macrophages is involved in cholesterol efflux and atherogenesis. Mol Cell 2001;7:161–171.[CrossRef][Medline]

  34. Lapsys NM, Kriketos AD, Lim-Fraser M et al. Expression of genes involved in lipid metabolism correlate with peroxisome proliferator-activated receptor gamma expression in human skeletal muscle. J Clin Endocrinol Metab 2000;85:4293–4297.[Abstract/Free Full Text]

  35. Tontonoz P, Hu E, Devine J et al. PPAR gamma 2 regulates adipose expression of the phosphoenolpyruvate carboxykinase gene. Mol Cell Biol 1995;15:351–357.[Abstract]

  36. Ashburner M, Ball CA, Blake JA et al. Gene ontology: Tool for the unification of biology. The Gene Ontology Consortium. Nat Genet 2000;25:25–29.[CrossRef][Medline]

  37. Hung SC, Chang CF, Ma HL et al. Gene expression profiles of early adipogenesis in human mesenchymal stem cells. Gene 2004;340:141–150.[CrossRef][Medline]

  38. Smith JJ, Marelli M, Christmas RH et al. Transcriptome profiling to identify genes involved in peroxisome assembly and function. J Cell Biol 2002;158:259–271.[Abstract/Free Full Text]

  39. Marques G, Musacchio M, Shimell MJ et al. Production of a DPP activity gradient in the early Drosophila embryo through the opposing actions of the SOG and TLD proteins. Cell 1997;91:417–426.[CrossRef][Medline]

  40. Piccolo S, Sasai Y, Lu B et al. Dorsoventral patterning in Xenopus: Inhibition of ventral signals by direct binding of chordin to BMP-4. Cell 1996;86:589–598.[CrossRef][Medline]

  41. Blader P, Rastegar S, Fischer N et al. Cleavage of the BMP-4 antagonist chordin by zebrafish tolloid. Science 1997;278:1937–1940.[Abstract/Free Full Text]

  42. Prockop DJ, Sieron AL, Li SW. Procollagen N-proteinase and procollagen C-proteinase. Two unusual metalloproteinases that are essential for procollagen processing probably have important roles in development and cell signaling. Matrix Biol 1998;16:399–408.[CrossRef][Medline]

  43. Reginato AM, Olsen BR. The role of structural genes in the pathogenesis of osteoarthritic disorders. Arthritis Res 2002;4:337–345.[CrossRef][Medline]

  44. Moog U, Jansen NJ, Scherer G et al. Acampomelic campomelic syndrome. Am J Med Genet 2001;104:239–245.[CrossRef][Medline]

  45. Wirth J, Wagner T, Meyer J et al. Translocation breakpoints in three patients with campomelic dysplasia and autosomal sex reversal map more than 130 kb from SOX9. Hum Genet 1996;97:186–193.[CrossRef][Medline]




This article has been cited by other articles:


Home page
 Stem CellsHome page
G. J. Block, S. Ohkouchi, F. Fung, J. Frenkel, C. Gregory, R. Pochampally, G. DiMattia, D. E. Sullivan, and D. J. Prockop
Multipotent Stromal Cells Are Activated to Reduce Apoptosis in Part by Upregulation and Secretion of Stanniocalcin-1
Stem Cells, March 1, 2009; 27(3): 670 - 681.
[Abstract] [Full Text] [PDF]

Home page
 Cancer Res.Home page
C. D. Scharer, C. D. McCabe, M. Ali-Seyed, M. F. Berger, M. L. Bulyk, and C. S. Moreno
Genome-Wide Promoter Analysis of the SOX4 Transcriptional Network in Prostate Cancer Cells
Cancer Res., January 15, 2009; 69(2): 709 - 717.
[Abstract] [Full Text] [PDF]

Home page
 J. Biol. Chem.Home page
K. Piper Hanley, F. Oakley, S. Sugden, D. I. Wilson, D. A. Mann, and N. A. Hanley
Ectopic SOX9 Mediates Extracellular Matrix Deposition Characteristic of Organ Fibrosis
J. Biol. Chem., May 16, 2008; 283(20): 14063 - 14071.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow Full Text (PDF)
Right arrow Supplemental Tables
Right arrow All Versions of this Article:
2005-0270v1
24/3/642    most recent
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow Reprints/Permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Ylöstalo, J.
Right arrow Articles by Prockop, D. J.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Ylöstalo, J.
Right arrow Articles by Prockop, D. J.

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
STEM CELLS THE ONCOLOGIST CME ALPHAMED PRESS JOURNALS