Identification and Functional Analysis of Candidate Genes Regulating Mesenchymal Stem Cell Self-Renewal and Multipotency


  • Lin Song,

    1. Cartilage Biology and Orthopaedics Branch, National Institute of Arthritis, and Musculoskeletal and Skin Diseases, National Institutes of Health, Department of Health and Human Services, Bethesda, Maryland, USA
    Search for more papers by this author
  • Nicole E. Webb,

    1. Cartilage Biology and Orthopaedics Branch, National Institute of Arthritis, and Musculoskeletal and Skin Diseases, National Institutes of Health, Department of Health and Human Services, Bethesda, Maryland, USA
    Search for more papers by this author
  • Yingjie Song,

    1. Cartilage Biology and Orthopaedics Branch, National Institute of Arthritis, and Musculoskeletal and Skin Diseases, National Institutes of Health, Department of Health and Human Services, Bethesda, Maryland, USA
    Search for more papers by this author
  • Rocky S. Tuan Ph.D.

    Corresponding author
    1. Cartilage Biology and Orthopaedics Branch, National Institute of Arthritis, and Musculoskeletal and Skin Diseases, National Institutes of Health, Department of Health and Human Services, Bethesda, Maryland, USA
    • Cartilage Biology and Orthopaedics Branch, National Institute of Arthritis, and Musculoskeletal and Skin Diseases, 50 South Drive, Room 1503, MSC 8022, National Institutes of Health, Bethesda, Maryland 20892-8022, USA. Telephone: 301-451-6854; Fax: 301-435-8017
    Search for more papers by this author


Adult human mesenchymal stem cells (hMSCs) possess multilineage differentiation potential, and differentiated hMSCs have recently been shown to have the ability to transdifferentiate into other lineages. However, the molecular signature of hMSCs is not well-known, and the mechanisms regulating their self-renewal, differentiation, and transdifferentiation are not completely understood. In this study, we demonstrate that fully differentiated hMSCs could dedifferentiate, a likely critical step for transdifferentiation. By comparing the global gene expression profiles of undifferentiated, differentiated, and dedifferentiation cells in three mesenchymal lineages (osteogenesis, chondrogenesis, and adipogenesis), we identified a number of “stemness” and “differentiation” genes that might be essential to maintain adult stem cell multipotency as well as to drive lineage-specific commitment. These genes include those that encode cell surface molecules, as well as components of signaling pathways. These genes may be valuable for developing methods to isolate, enrich, and purify homogeneous population of hMSCs and/or maintain and propagate hMSCs as well as guide or regulate their differentiation for gene and cell-based therapy. Using small interfering RNA gene inactivation, we demonstrate that five genes (actin filament-associated protein, frizzled 7, dickkopf 3, protein tyrosine phosphatase receptor F, and RAB3B) promote cell survival without altering cell proliferation, as well as exhibiting different effects on the commitment of hMSCs into multiple mesenchymal lineages.


Stem cells that reside in adult tissues and organs are capable of self-renewal and multipotential differentiation. Adult stem cells remain in a nonproliferative, quiescent state during most of their lifetime until stimulated by the signals triggered by tissue damage and remodeling [1, 2]. Upon stimulation, adult stem cells re-enter the cell cycle to replenish the stem cell pool, as well as to generate progenitor cells, which then give rise to a variety of differentiated cell types for tissue regeneration and homeostasis. Cytokines, growth factors, adhesion molecules, and extracellular matrix components in the stem cell microenvironment play important roles in stem cell fate determination, functioning as the driving forces for stem cells to switch from a self-renewal to a differentiation stage [3, [4], [5], [6]–7]. However, the downstream intracellular effectors in these processes are still largely unknown.

In addition to self-renewal and multipotentiality, adult stem cells possess the ability to transdifferentiate, that is, to switch their specific developmental lineage into another cell type of a different lineage, sometimes across embryonic germ layers. For example, mesenchymal stem cells (MSCs) can be induced to become nonmesodermal cells, including functional neurons, astrocytes, oligodendrocytes, and endothelial cells, by appropriate extrinsic stimuli in vitro [8, [9], [10]–11] and to become hepatocyte-like cells after xenograft [12]. Moreover, when injected into blastocysts, a subset of MSCs gave rise to almost all somatic cells in a variety of tissues and organs, such as brain, liver, retina, lung, kidney, spleen, blood, and skin, demonstrating their plastic potentiality [11]. In addition, MSCs are able to maintain their multidifferentiation potential after commitment. As recently demonstrated [13], osteoblasts, chondrocytes, and adipocytes differentiated from human mesenchymal stem cells (hMSCs) can transdifferentiate into other cell types in response to extrinsic factors, likely through genetic reprogramming. However, the molecular mechanisms behind the transdifferentiation process are poorly understood. Do stem cells switch their phenotype directly from one cell type to another? Do differentiated cells dedifferentiate first to a primitive cell type before committing to a different lineage? Are common signaling pathways employed by the different stem cells, or are there unique mechanisms controlling individual stem cells? Answers to these questions will undoubtedly enhance our understanding of how stem cells both maintain their multipotentiality and control their commitment and differentiation.

Global gene expression profiling has been widely used to identify transcriptional signatures of specific stem cells and to gain insights into the signaling mechanisms regulating their differentiation program in embryonic stem cells (ESCs) [14, [15], [16], [17], [18], [19], [20], [21], [22]–23], hematopoietic stem cells (HSCs) [15, 24], neural crest stem cells (NSCs) [15, 25], and skin epithelial stem cells (SESCs) [15, 24, 26, 27]. In addition, by comparing gene expression profiles of different stem cell groups, a common pool of genes have been identified, which either serve as stem cell markers for self-renewal or maintain the uncommitted state of stem cells [16, 22, 27, 28]. Compared with the extensive studies performed with ESCs and other adult stem cells (e.g., HSCs, NSCs, and SESCs), research on the molecular mechanism(s) controlling MSC self-renewal and maintenance has lagged behind, largely because of the heterogeneous nature and lack of consensus on the defined markers for the MSCs [29, [30], [31]–32]. Attempts have been made to determine the gene expression profiles of undifferentiated MSCs from various sources [33, [34], [35], [36], [37], [38], [39]–40] and their differentiated progeny, for example, osteoblasts [41, 42], chondrocytes [43], and adipocytes [44], using serial analysis of gene expression (SAGE) and microarray analysis. Several genes have been identified to be highly expressed in undifferentiated hMSCs, including vimentin, connective tissue growth factor, collagen type I α1, and eukaryotic translation elongation factor 1 α1. Although these genes are expressed in hMSCs and thus might be considered as their molecular signature for purification and enrichment, little evidence is available on their functionality in hMSC maintenance and self-renewal, and it is not known whether they are merely housekeeping genes or actually play critical roles in preventing cells from differentiating. Another issue that remains unresolved is the regulatory mechanism controlling MSC multilineage differentiation capabilities, particularly the possibility of common signaling pathways that are shared by more than one differentiation pathways. Given the heterogeneous nature of in vitro-expanded MSCs as well as the complexity of culture conditions used in individual studies, it is almost impossible to perform comparative gene expression analysis to generate common gene lists from published data. Furthermore, the lack of functional analysis of genes in these studies has delayed in-depth assessment of the role of these reported genes.

In this study, we used an in vitro differentiation and dedifferentiation culture system using human MSCs and performed global gene expression profiling on undifferentiated hMSCs, differentiated osteoblasts, chondrocytes, and adipocytes, as well as dedifferentiated cells derived from these three distinct mesenchymal lineages. Our results demonstrated for the first time that differentiated cells could dedifferentiate into a primitive stem cell-like stage before transdifferentiating into another cell type. By comparing differentially expressed genes during differentiation and dedifferentiation processes in all three lineages, we identified a list of genes that are candidate markers of hMSCs and may function to maintain stem cells at an uncommitted state or initiate their differentiation process. We have further explored the function of five genes (actin filament-associated protein [AFAP], frizzled 7 [FZD7], dickkopf 3 [DKK3], protein tyrosine phosphatase receptor F [PTPRF], and RAB3B) in stem cell proliferation, survival, and multilineage differentiation by inactivating their expression using small interfering RNA (siRNA) technology. Our results demonstrate that all five genes promote cell survival but exhibit different effects on the commitment of hMSCs into multiple mesenchymal lineages.

Materials and Methods


Cell culture media, reagents, and antibodies were purchased from Invitrogen (Carlsbad, CA, All chemicals were obtained from Sigma-Aldrich (St. Louis, unless specified otherwise.

Isolation, Culture Expansion, Differentiation, and Dedifferentiation of hMSCs

hMSCs were isolated from bone marrow aspirate obtained with Institutional Review Board (George Washington University, Washington, DC) from patients (aged 45–84 years) undergoing elective total hip arthroplasty. Cells were culture-expanded in basal medium (BM), containing Dulbecco's modified Eagle's medium, 10% fetal bovine serum, and antibiotics [13]. hMSCs that underwent fewer than 15 population doublings were used.

Differentiation of hMSCs into three mesenchymal lineages was induced as described previously [13] with the following modifications: to induce osteogenesis, cells were plated at 5.2 × 103 cells per cm2 on monolayer and cultured in the osteogenic medium (BM supplemented with 10 nM dexamethasone, 10 mM β-glycerophosphate, 50 μg/ml ascorbate phosphate, and 10 nM 1,25 dihydroxyvitamin D3) for 7, 14, or 21 days. To induce adipogenesis, cells were plated at 4.9 × 104 cells per cm2 and cultured in adipogenic medium (BM supplemented with 1 μM dexamethasone, 1 μg/ml insulin, and 0.5 mM 3-isobutyl-1-methylxanthine) for 14 or 21 days. For chondrogenesis, cells were differentiated either in a high-density pellet culture (2.5 × 105 cells per pellet) or in a high-density cell mass in alginate beads (1.5 × 107 cells per ml) in chondrogenic medium (serum-free BM supplemented with 0.1 μM dexamethasone, 50 μg/ml ascorbate phosphate, 40 μg/ml l-proline, 100 μg/ml sodium pyruvate, 1% ITS-premix, and 10 ng/ml transforming growth factor-β3 [TGF-β3]) for 14 or 21 days. To induce dedifferentiation process, cells were removed from the induction medium, replated on monolayer, and cultured in BM for 20 days further.

To examine the effects of specific gene knockdown on hMSCs, cells were transfected with gene-specific siRNA, collected 24 hours post-transfection, and cultured accordingly as described above.

Analysis of hMSC Differentiation

Osteogenesis was detected by histochemical staining of alkaline phosphatase (ALP) activity using an enzyme kit from Sigma-Aldrich and quantitative reverse transcription-polymerase chain reaction (RT-PCR) analysis of expression levels of ALP and osteocalcin (OC). Adipogenesis was detected by the presence of neutral lipids in the cytoplasm stained with Oil Red O. To analyze chondrogenic differentiation, high-density cell pellets were fixed with 4% paraformaldehyde and cryosectioned at 8-μm thickness for histological and immunochemical staining as described previously [13]. Sulfated matrix proteoglycan was stained with Alcian Blue (pH 1.0). Collagen type II was detected by immunohistochemistry. For crystal violet staining, cells were fixed, incubated with 0.2% crystal violet in 2.5% acetic acid for 5 minutes, and rinsed with water.

Global Gene Expression Profiling

hMSCs isolated from the following patients were used for lineage-specific induction to obtain undifferentiated, differentiated, and dedifferentiated cells: three female patients aged 59, 67, and 70 for osteogenesis; four female patients aged 59, 67, 70, and 71 for adipogenesis; and three female patients aged 62, 67, and 70 for chondrogenesis.

All of the following procedures were performed according to the manufacturers' instructions. Reagents were used at the recommended concentrations. Total RNA was isolated using TRIzol reagent and cleaned up with RNeasy Mini kit (Qiagen, Valencia, CA, First-strand cDNA was synthesized from 8 μg of total RNA using T7-oligo(dT) primer and SuperScript II reverse transcriptase at 42°C for 1 hour. Second-strand cDNA was then synthesized using Escherichia coli DNA ligase, DNA polymerase I, and RNase H. Double-stranded cDNA was cleaned up with the GeneChip cDNA cleanup module (Affymetrix, Santa Clara, CA, Biotin-labeled cRNA was synthesized using Enzo BioArray HighYield RNA transcript labeling kit followed by RNA cleanup. Fifteen μg of cRNA was then fragmented in 1× fragmentation buffer at 94°C for 35 minutes and kept at −20°C. Because of the limited amount of total RNA obtained from chondrogenic samples, 2 μg of total RNA was first amplified to produce aRNA using RiboAmp OA RNA amplification kit (Acturus), which was converted to double-stranded cDNA before being used for generating biotin-labeled cRNA. To make the hybridization cocktail, fragmented cRNA was mixed with control oligonucleotide B2, eukaryotic hybridization controls, herring sperm DNA, and acetylated bovine serum albumin (BSA) in 1× hybridization buffer; heated at 99°C for 5 minutes; and equilibrated to 45°C before being hybridized to the GeneChip arrays for 16 hours at 45°C. Arrays were then washed and stained using the Affymetrix Fluidics Station 450 following the user's manual. The stained arrays were scanned using the Affymetrix GeneChip scanner 3000, controlled by Affymetrix microarray suite software. Each sample was hybridized to two arrays: human genomes U133A and U133B.

Data Analysis

For GeneChip arrays, the raw intensity of individual samples was normalized and scaled among the samples using Microarray Data Management & Analysis System ( Principal component analysis (PCA) was performed using Partek Pro software (Partek, Inc., St. Charles, MO). Gene expression levels in each differentiation lineage (osteogenesis, adipogenesis, and chondrogenesis) were compared between undifferentiated and differentiated samples, as well as dedifferentiated and differentiated samples. Paired Student's t tests were performed to compare the expression change between undifferentiated and differentiated samples, as well as between dedifferentiated and differentiated samples, with a statistical significance level set at p ≤ .05. Genes that exhibited a 2.0-fold change at p ≤ .05 were filtered. Only the genes that exhibited similar expression pattern in cells obtained from all patients were selected for further analysis. Selected genes were annotated using open source DAVID 2.0 ( Selected genes were also analyzed using Ingenuity pathways analysis application (Ingenuity Systems, Mountain View, CA).

A paired Student's t test was performed with a significance level of p ≤ .05 for other sample analysis.

Quantitative RT-PCR

RNA was isolated using TRIzol reagent and cleaned up with RNeasy Mini kit (Qiagen). First-strand cDNA was synthesized using SuperScript first-strand synthesis system (Invitrogen). Five to 10 ng of cDNA was amplified and detected by using iQ SYBR Green supermix kit in an iCycler iQ real-time PCR detection system (Bio-Rad, Hercules, CA, The amount of transcript was normalized to an internal glyceraldehyde-3-phosphate dehydrogenase (GAPDH) control and averaged from triplicate samples. The following primers were used. Bone sialoprotein (BSP): 5′-ggtctctgtggtgccttctg-3′, 5′-tgctacaacactgggctatgg-3′; OC: 5′-gcctttgtgtccaagc-3′, 5′-ggaccccacatccatag-3′; lipoprotein lipase (LPL): 5′-gagatttctctgtatggcacc-3′, 5′-ctgcaaatgagacactttctc-3′; fatty acid binding protein 4 (FABP4): 5′-tgggccaggaatttgacgaagt-3′, 5′-tcaacgtcccttggcttatgct-3′; cartilage oligomeric matrix protein (COMP): 5′-tgtccccagaagagcaaccc-3′, 5′-attgtcgtcgtcgtcgtcgc-3′; metalloproteinase 13 (MMP13): 5′-aacgccagacaaatgtgaccc-3′, 5′-tccgcatcaacctgctgagg-3′; PTPRF: 5′-cgagcaaggcggagaggag-3′, 5′-tcaaggaggcaagcacaaagc-3′; AFAP: 5′-aagctgcttgagtccacctgaa-3′, 5′-actatttgacccgaaggcagca-3′; RAB3B: 5′-caagtgtgacatggaggaagag-3′, 5′-agggagagtgggctgagag-3′; FZD7: 5′-aacacgacggcaccaagacc-3′, 5′-ggcagggcacggcatagc-3′; DKK3: 5′-gctgaccaggcttcttcctacatc-3′, 5′-gcagggcactcttctccacatttc-3′; ALP: 5′-tggagcttcagaagctcaacacca-3′, 5′-atctcgttgtctgagtaccagtcc-3′; GAPDH: 5′-agggggcagagatgatgacc-3′, 5′-caaggctgagaacgggaagc-3′.

siRNA Transfection

hMSCs were seeded at 6.5 × 103 cells per cm2 (approximately 50% confluence) in antibiotic-free basal medium 24 hours prior to transfection. siRNA transfection was performed following the manufacturer's protocol. Briefly, 10 μM gene-specific siRNA oligomers (Ambion Inc.) were diluted in Opti-MEM I Reduced Serum Medium and mixed with Lipofectamine 2000 (Invitrogen). After a 20-minute incubation at room temperature, the complexes were added to the cells at a final siRNA concentration of 33 nM. The medium was replenished with antibiotic-containing medium 24 hours post-transfection. Culture medium was then changed every 3 days for the duration of the experiment. hMSCs treated with Lipofectamine 2000 only (untransfected control) and hMSCs transfected with a Silencer negative control siRNA (transfection control) were used as experimental controls.

Immunofluorescence Microscopy

Transfected hMSCs were cultured in monolayer on coverslips for 7 or 14 days before fixation in 4% paraformaldehyde for 15 minutes at room temperature. Fixed cells were rinsed several times with phosphate-buffered saline (PBS), permeabilized with 0.5% Triton X-100 in deionized H2O, and rinsed in PBS. Following a 30-minute incubation with 1% BSA in PBS at room temperature, cells were then incubated in 10 μg/ml primary antibody for 1 hour at 37°C. The following antibodies were used: monoclonal mouse anti-AFAP IgG (BD Biosciences, San Diego,, goat anti-human Dkk-3 IgG (R&D Systems Inc., Minneapolis,, monoclonal rat anti-human Frizzled-7 IgG (R&D Systems), polyclonal rabbit anti-PTPRF IgG (Orbigen, Inc), and polyclonal rabbit anti-Rab3B IgG (Santa Cruz Biotechnology Inc., Santa Cruz, CA, Cells were rinsed several times in PBS and then incubated in 2 μg/ml conjugated secondary antibody for 1 hour at 37°C. The secondary antibodies used were as follows: Alexa Fluor 488-conjugated goat anti-mouse IgG (A-11029), fluorescein isothiocyanate-conjugated anti-goat IgG (F-2016) (Sigma-Aldrich), Alexa Fluor 488-conjugated goat anti-rat IgG (A-11006), and Alexa Fluor 488-conjugated donkey anti-rabbit IgG (A-21206). Following secondary antibody incubation, cells were rinsed in PBS, and nuclei were counterstained with 1 μg/ml 4′,6-diamidino-2-phenylindole, dihydrochloride (Molecular Probes Inc., Eugene, OR, and rinsed again. Coverslips were mounted in Vectashield (Vector Laboratories, Burlingame, CA,, sealed with nail polish, allowed to dry overnight at room temperature, and stored in dark at −20°C until analysis.

Western Analysis

Cells were collected by incubation with trypsin-EDTA followed by centrifugation. The cell pellet was resuspended in lysis buffer (20 mM Tris, pH7.5, 150 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate, 1 mM EDTA, and 0.1% SDS) containing proteinase inhibitors, freeze-thawed three times, and incubated at 4°C for 30 minutes. Following centrifugation at 18,000g for 15 minutes at 4°C, the supernatant containing total cell extract was collected and kept at −80°C. Protein concentration was determined by Bradford assay. Total cell protein aliquots (80 μg) were mixed with Laemmli sample buffer, boiled for 5 minutes, and separated in a 4%–20% Tris-HCl gel (Bio-Rad). Proteins were then transferred to nitrocellulose membranes using a mini Trans-blot electrophoretic transfer cell apparatus (Bio-Rad).

For immunoblotting, the nitrocellulose membrane was first incubated with the blocking buffer containing 1% fish gelatin and 0.05% Tween 20 in Tris-buffered saline (TBS) for 1 hour at room temperature. The membrane was then incubated with primary antibodies (as described above) diluted (1:100) in blocking buffer for 2 hours and rinsed three times in wash buffer (0.05% Tween 20 in TBS). Horseradish peroxidase-conjugated secondary antibodies diluted (1:1,000) in blocking buffer were added to the membrane, incubated for 1 hour, and rinsed several times with wash buffer. The membrane was then incubated with SuperSignal West Pico peroxide and luminal enhancer solutions (Pierce) for 5 minutes, exposed to the film, and developed. Films were scanned using an UMAX PowerLookIII scanner, saved as digital images, and processed using Adobe Photoshop 7.0 (Adobe Systems, Inc., San Jose, CA).

Cell Proliferation and Apoptosis Assay

Transfected cells were cultured as monolayers. At day 2, 5, 7, 9, 12, or 14, cultures were rinsed with PBS and detached with trypsin-EDTA; cells were collected by centrifugation and resuspended in BM; duplicate aliquots placed into 96-well plates; and 10 μl of Cell Counting Kit-8 solution (Dojindo Laboratories) was added to each well. After incubation for 3 hours and 15 minutes at 37°C, A450 was measured using a Victor5 Light Luminescence Counter (PerkinElmer Life Sciences, Boston,, with standards of known cell numbers prepared in the same manner used to determine cell numbers for the experimental conditions.

To detect apoptotic cells, cultures were fixed 3 days post-transfection and stained with a fluorescence-based BD ApoAlert DNA fragmentation assay kit (BD Biosciences) following the manufacturer's protocol.

Image Analysis

Light and epifluorescence microscopy were done using a Leica DM RX/E microscope (Leica, Heerbrugg, Switzerland, with appropriate filters and captured with an ORCA-ER CCD digital camera (Hamamatsu Photonics) using Openlab software (Improvision, Inc.). All images were processed using Adobe Photoshop 7.0.


Committed Adult hMSCs Regain Their Multipotency Following Dedifferentiation

Previously, we demonstrated that adult hMSCs were capable of being induced to acquire the characteristics of other cell types after they were fully committed, a phenomenon termed transdifferentiation [13]. In this study, to assess how differentiated stem cells maintain their capability to acquire a completely new phenotype, we first assessed whether differentiated hMSCs could dedifferentiate prior to transdifferentiation. As shown in Figure 1A, undifferentiated hMSCs induced in vitro for 21 days acquired the typical characteristics of mature osteoblasts, adipocytes, and chondrocytes. However, upon withdrawal of the induction stimuli, these fully differentiated cells gradually lost their characteristic markers (Fig. 1A) and appeared spindle-shaped, similar to undifferentiated hMSCs (Fig. 1B), suggesting that mature cells might have dedifferentiated and regained the properties of primitive hMSCs, that is, self-renewal and multipotency.

Figure Figure 1..

Histochemical and immunohistochemical analyses of undifferentiated, differentiated, and subsequently dedifferentiated human mesenchymal stem cells (hMSCs). (A): Osteoblast formation was induced by culturing hMSCs with dexamethasone (Dex), ascorbate, β-glycerophosphate, and VD3 and assessed on the basis of histochemically detectable, membrane-bound ALP activity (a–c). Adipocytes were induced by culturing hMSCs with BM supplemented with 1 μM dexamethasone, 1 μg/ml insulin and 0.5 mM 3-isobutyl-1-methylxanthine and identified by the presence of Oil Red O-stained neutral lipids in the cytoplasm (d–f). Chondrocyte formation of hMSCs was induced in a transforming growth factor-β3-supplemented, serum-free, high-density pellet culture and determined by the secretion of Alcian Blue-stained sulfated proteoglycan (g–i) and collagen type II-stained extracellular matrix (j–l). Differentiated cells exhibited typical phenotypes as evidenced by the specific staining, whereas dedifferentiated cells, similar to undifferentiated hMSCs, lacked these phenotypic markers. (B): Crystal violet staining of undifferentiated hMSCs (a) and dedifferentiated cells from osteoblasts (b), adipocytes (c) and chondrocytes (d). Dedifferentiated cells exhibited similar fibroblastic morphology as undifferentiated hMSCs. Bar = 10 μm. Abbreviation: COL II, collagen type II.

Global Gene Expression Analysis During Stem Cell Differentiation and Dedifferentiation

If the dedifferentiated cells had regained the potential of hMSCs, it was reasonable to predict that they would share a similar gene expression profile to primitive stem cells. Thus, the differentiation and dedifferentiation culture system employed here provided a useful platform to identify genes that control stem cell self-renewal and differentiation. Global gene expression levels were determined and compared in undifferentiated hMSCs, differentiated osteoblasts, chondrocytes, and adipocytes, as well as dedifferentiated cells from osteoblasts, chondrocytes, and adipocytes. Genes were categorized into two groups according to their expression patterns (Fig. 2A). “Differentiation genes” included genes that showed increased expression during differentiation but decreased expression during dedifferentiation. On the other hand, “stemness genes” included genes that were downregulated during differentiation but upregulated during dedifferentiation. As expected, dedifferentiated cells exhibited a gene expression profile similar to that of undifferentiated hMSCs, and both groups of cells were distinct from differentiated cells, as analyzed by PCA (Fig. 2B). Within the differentiated groups, cells from the individual lineages exhibited different expression profiles, indicating the intrinsic uniqueness of each cell type.

Figure Figure 2..

Identification of genes regulating stem cell self-renewal and multilineage differentiation. (A): Schematic diagram depicting the experimental design. Genes were selected based on their expression change during differentiation and dedifferentiation processes. (B): Principal component analysis of global gene expression data. Differentiated cells (i.e., AD, OS, and CH) exhibited a significantly different global gene expression profile compared with undifferentiated human mesenchymal stem cells (hMSCs) and dedifferentiated cells (De-AD, De-CH, and De-OS). On the other hand, undifferentiated and dedifferentiated cells shared a similar global gene expression profile. (C): Venn diagrams showing the genes that changed their expression levels during differentiation and dedifferentiation processes by at least twofold (p < .05). (D): Real-time reverse transcription-polymerase chain reaction analysis of expression levels of five selected genes in undifferentiated hMSCs, differentiated as well as dedifferentiated OS, AD, and CH. Three genes, AFAP, PTPRF, and RAB3B, decreased their expressions during differentiation and increased their expressions during dedifferentiation in all three lineages. The other two genes, DKK3 and FZD7, exhibited an expression pattern similar to that of the others during osteogenesis and adipogenesis but a different expression pattern during chondrogenesis. Abbreviations: AD, adipocytes; AFAP, actin filament-associated protein; CH, chondrocytes; De-AD, dedifferentiated adipocytes; De-CH, dedifferentiated chondrocytes; De-OS, dedifferentiated osteoblasts; DKK3, dickkopf 3; hMSC, human mesenchymal stem cell; MSC, mesenchymal stem cell; OS, osteoblasts; PTPRF, protein tyrosine phosphatase receptor F.

As shown in Figure 2, there are 460 genes in the stemness genes group, with more than twofold significant decrease during differentiation and increase during dedifferentiation, in at least one lineage. Among these, 62 genes exhibited similar expression patterns in two lineages, and 11 genes in all three lineages (supplemental online Table 2). In the differentiation genes group, 456 genes were upregulated during differentiation and downregulated during dedifferentiation significantly, by more than twofold in at least one lineage, with 12 genes shared by all three lineages and 40 genes by two lineages (supplemental online Table 1). Six genes from differentiation genes group that are typical markers of individual lineage were selected for quantitative RT-PCR analysis, including BSP and OC for osteoblasts, LPL and FABP4 for adipocytes, and COMP and MMP13 for chondrocytes. Although they confirmed the GeneChip data, the quantitative PCR results revealed much higher fold changes in gene expression during both differentiation and dedifferentiation processes in all six genes (supplemental online Table 3). In addition, the fold change between dedifferentiated and undifferentiated cells was close to one, suggesting that the two cell populations are similar at the transcription level as indicated by PCA. Taken together, these results confirmed our hypothesis and proved the effectiveness of our system to identify candidate genes controlling stem cell self-renewal and multipotency.

In general, 37% of differentially expressed genes belong to one of eight major canonical pathways: integrin signaling (RHOJ, ITGA10, FN1, RAP2B, ITGA7, RHOU, COL4A1, ITGA11, LAMB3, LAMA4, PIK3R1, AKT3, MAPK3, COL4A2, LAMA2), IGF-1 signaling (YWHAH, IGF1, IGFBP5, IGFBP4, IGFBP3, IGFBP7, PIK3R1, AKT3, PRKAR2B, MAPK3, FOXO1A), G-protein-coupled receptor signaling (NFKBIA, PDE3A, PDE1C, RGS2, EDNRB, PIK3R1, AKT3, PRKAR2B, RGS4, MAPK3, AGTR1), IL-6 signaling (NFKBIA, HSPB1, IL6, IL1R1, interleukin 1 receptor type II [IL1R2], LBP, JAK2, TNFAIP6, MAPK3, MAP2K3), insulin receptor signaling (PPP1CB, JAK2, PIK3R1, PTPRF, AKT3, PRKAR2B, MAPK3, SGK, FOXO1A, NCK1), pyrimidine metabolism (UPP1, TXNRD1, POLE4, REV3L, ENTPD1, NME1, TRUB2, NT5E, DUT, DPYSL3), nuclear factor-γB signaling (NFKBIA, TNFSF11, NGFB, IL1R2, IRAK3, PIK3R1, AKT3, IL1R1, TLR2), and Wnt/β-catenin signaling (SFRP4, ACVR2, FZD1, FZD7, WNT5A, LEF1, TCF3, AKT3, CDH2, DKK3).

The identified stemness genes and differentiation genes are involved in different cellular processes and functions. For example, a large percentage of stemness genes are involved in signaling pathways, such as IGF-1 signaling (YWHAH, IGF1, IGFBP4, IGFBP5, IGFBP3, AKT3, MAPK3), JAK/Stat signaling (STAT1, STAT4, SOCS2, SOCS5), TGF-β signaling (INHBA, SMURF2, SMAD3, SERPINE1, ACVR2), and Wnt/β-catenin signaling (SFRP4, WNT5A, FZD7, CDH2, TCF3, DKK3). On the other hand, the differentiation genes group contains genes involved largely in metabolism. For instance, GCLC, GLUL, and GSS in glutamate metabolism and GPX3, ANPEP, GCLC, and GSS in glutathione metabolism. Genes in nuclear factor κB (NF-κB) signaling (NFKBIA, IL1R2, IRAK3, PIK3R1, TLR2) and death receptor signaling (TNFSF1, BIRC3) are also significantly represented in the differentiation genes group. Among the genes that shared expression pattern in more than two lineages, those involved in organ morphology, renal and urological disease, amino acid metabolism, dental disease, organismal survival, and free radical scavenging are highly represented in the differentiation genes group, whereas the stemness genes group contains genes that are primarily involved in cell morphology, cancer, cell-to-cell signaling and interaction, cellular growth and proliferation, nervous system development and function, tissue development, and tumor morphology.

Ninety-one genes in the stemness genes group encode proteins that are cell surface proteins and/or receptors, including 20 genes shared by at least two lineages (Table 1). Except for a few genes whose functions are unknown, the majority of these genes function in defined cellular processes, such as metabolism (ATPase and solute carrier proteins), carcinogenesis and metastasis (Tetraspanin family members), cell growth and survival and senescence (AXL, TNFRSF10D), development (NOTCH2, NUMB, JAG1), and signal transduction (PTPRF, FZD7, ICAM1).

Table Table 1.. Selected cell surface proteins and receptors
original image

Functional Analysis of Genes Involved in Stem Cell Self-Renewal and Multipotency

As proof of principle, five genes from the stemness genes group were selected based on their unique expression pattern in individual lineage and their cellular function. By reducing their expression level using siRNA, their effects on hMSC expansion and multilineage differentiation were accessed. Initially identified from GeneChip data, PTPRF exhibited similar expression patterns in all three lineages, AFAP and RAB3B in two lineages, and FZD7 and DKK3 in only one lineage. However, quantitative RT-PCR analysis demonstrated that AFAP and RAB3B also shared a pattern similar to that of PTPRF in all three lineages (Fig. 2D; supplemental online Table 3), whereas FZD7 and DKK3 appeared to behave differently in chondrogenic lineage than in osteogenic and adipogenic lineages (Fig. 2D; supplemental online Table 3). Using gene-specific siRNA, we successfully reduced the gene expression level by 60%–80% compared with transfection controls (Fig. 3A). Protein levels were also reduced for all five genes as confirmed by Western analysis (Fig. 3B) and immunofluorescence staining (Fig. 3C), and inactivation of these genes was sustained for at least 7 days post-siRNA transfection.

Figure Figure 3..

Reduction of gene expression using siRNA transfection. (A): Real-time reverse transcription-polymerase chain reaction analysis of transcript levels of five selected genes 7 days post-transfection. On average, expression of selected genes was significantly reduced by 60%–80% of untransfected and transfection controls. Values are mean ± SD (n = 3). *, p < .05. (B): Western analysis of protein levels 7 days post-siRNA transfection. (C): Immunofluorescence staining of proteins in human mesenchymal stem cells post-transfection. Protein reduction was sustained 7 days post-siRNA transfection. Bar = 10 μm. Abbreviations: AFAP, actin filament-associated protein; DKK3, dickkopf 3; FZD7, frizzled 7; PTPRF, protein tyrosine phosphatase receptor F.

When cultured in basal medium, hMSCs with reduced level of AFAP, PTPRF, RAB3B, FZD7, or DKK3 exhibited a slower growth rate up to 9 days post-transfection, with a more dramatic difference observed in cells with inactivated DKK3 and PTPRF, but little difference after 9 days, except in DKK3-inactivated cells, compared with controls (Fig. 4A). These results suggest that these genes have little effect on stem cell proliferation when functioning alone. In addition, reduction of each of these five genes dramatically increased the percentage of cells undergoing apoptosis (Fig. 4C), suggesting that these genes function as cell survival protectors.

Figure Figure 4..

Effect of gene reduction by siRNA transfection on human mesenchymal stem cell proliferation and cellular viability. (A): Growth curves showing cell proliferation over 14 days post-siRNA transfection. Cells exhibited a slightly slower rate of proliferation from day 1 to day 8 but a similar rate after day 8, compared with both untransfected control and transfection control. Values are mean ± SD (n = 3). (B): Gene knockdown by siRNA transfection increased the percentage of apoptotic cells 3 days post-transfection. Mean values are presented (n = 2). Abbreviations: AFAP, actin filament-associated protein; DKK3, dickkopf 3; FZD7, frizzled 7; PTPRF, protein tyrosine phosphatase receptor F.

hMSCs with reduced expression of these five genes were challenged to undergo differentiation into three mesenchymal lineages. As shown in Figure 5, cells with inactivated AFAP, DKK3, FZD7, PTPRF, or RAB3B were able to differentiate into osteoblasts with enhanced ALP activity (Fig. 5A). Reduction of AFAP, FZD7, or RAB3B enhanced ALP transcription significantly (Fig. 5B). Furthermore, inactivation of FZD7 significantly increased OC expression, whereas others appeared to have little difference compared with controls (Fig. 5C). Unlike the uniform enhancement on osteogenesis, inactivation of each of these five genes exhibited a unique impact on chondrogenesis. As shown in Figure 6, reduction of DKK3 increased the production of Alcian Blue-stained sulfated proteoglycan but decreased the collagen type II. Inactivation of AFAP and RAB3B had an indiscernible effect on sulfated proteoglycan but a dramatic increase on collagen type II deposition. On the other hand, reduction of either FZD7 or PTPRF decreased both proteoglycan and collagen type II synthesis. Contrary to the enhancement of osteogenesis and varied effects on chondrogenesis, inactivation of any of these five genes suppressed adipogenesis, demonstrated by the reduction in the number of Oil Red O-positive adipocytes (Fig. 6).

Figure Figure 5..

Osteogenic differentiation potential of human mesenchymal stem cells was altered by siRNA-mediated reduction of gene expression. (A): Alkaline phosphatase (ALP) staining of transfected cells cultured in the absence or presence of osteogenic induction medium for 14 days. Compared with controls, transfected cells showed enhanced ALP staining at the cellular level in osteogenic medium (bottom row). In addition, the number of ALP-positive cells increased in four transfected cell populations (AFAP, FZD7, PTPRF, and RAB3B), even in the absence of osteogenic stimuli (top row). (B): ALP expression levels analyzed by quantitative reverse transcription-polymerase chain reaction (RT-PCR). Reduction of AFAP, FZD7, and RAB3B significantly increased ALP transcription. (C): Osteocalcin (OC) expression levels analyzed by quantitative RT-PCR. Reduction of FZD7 expression level increased OC transcription significantly, whereas other genes had little effect on OC level. For (B) and (C), all values are mean ± SD (n = 3). *, p < .05. Bar = 10 μm. Abbreviations: AFAP, actin filament-associated protein; DKK3, dickkopf 3; FZD7, frizzled 7; PTPRF, protein tyrosine phosphatase receptor F; TC, transfected control; UC, untransfected control.

Figure Figure 6..

Chondrogenic and adipogenic differentiation of human mesenchymal stem cells (hMSCs) was altered by siRNA gene reduction. Chondrogenesis of hMSCs was assessed on the basis of Alcian Blue staining of sulfated proteoglycan matrix and collagen type II immunostaining after cells were induced in chondrogenic medium for 21 days post-transfection. Sections from two individual pellets are shown for each staining. The level of sulfated proteoglycan was increased in cells transfected with DKK3 siRNA and decreased in those transfected with FZD7 or PTPRF siRNA. The sulfated proteoglycan level was not significantly changed in cells transfected with AFAP or RAB3B siRNA. On the other hand, reduction of AFAP or RAB3B increased collagen type II production, whereas reduction of DKK3, FZD7, or PTPRF decreased it. Adipogenesis was detected by the presence of Oil Red O-stained neutral lipids in the cytoplasm. All transfected cells exhibited fewer adipocytes after a 21-day induction. Bar = 10 μm. Abbreviations: AFAP, actin filament-associated protein; DKK3, dickkopf 3; FZD7, frizzled 7; PTPRF, protein tyrosine phosphatase receptor F; TC, transfected control; UC, untransfected control.


Transdifferentiation refers to the process in which stem cells of a certain lineage differentiate into cell types of a different lineage across embryonic germ layers or the process in which fully differentiated cells switch their phenotype and acquire characteristics of other cell types within or beyond their original lineages. For example, Park et al. [45] demonstrated that mature human adipocytes were able to dedifferentiate and redifferentiate into adipocytes and osteoblasts at the clonal level. This dedifferentiation phenomena was further confirmed by Tagami et al. [46]. By using electron microscopy and quantitative PCR, they showed that upon the withdrawal of stimuli, adipocytes derived from hMSCs can dedifferentiate into fibroblast-like stem cells at both transcriptional and structural levels. In a recent study, we have shown that fully differentiated adult hMSCs (e.g., osteoblasts, chondrocytes, and adipocytes) were able to change their differentiation programs to a different cell type in response to extrinsic stimuli by virtue of retaining their multipotentiality [13]. We hypothesized that this event involves the dedifferentiation of differentiated hMSCs into a cell type similar to the original hMSCs. In the present study, we have demonstrated that adult hMSCs could dedifferentiate into a primitive stem cell-like population, upon the withdrawal of extrinsic stimulation after their osteogenic, adipogenic, or chondrogenic differentiation. Consistent with our hypothesis, dedifferentiated hMSCs shared a similar transcriptional profile as primitive hMSCs. Although it is possible that the similarity of gene expression profiles observed between undifferentiated and dedifferentiated hMSCs was a result of expansion of nondifferentiated hMSCs during the dedifferentiation process, it is unlikely to be the case here. First, we [13] and others [45, 46] have demonstrated that majority of the homogeneous mature osteoblasts and adipocytes can dedifferentiate in the absence of continuous induction. Secondly, there was a dramatic phenotypic loss of differentiated cells but no massive cell death in the cell population during the dedifferentiation step, suggesting that the fibroblast-like cells observed after dedifferentiation process were derived largely, if not completely, from previously differentiated cells. Taken together, these results imply that dedifferentiation is very likely a necessary step for differentiated stem cells to pursue transdifferentiation, similar to the events occurring during limb and eye regeneration in other vertebrates [47]. That differentiation and dedifferentiation can be triggered upon the addition or withdrawal of inducing factors suggests that the stemness and differentiation states of stem cells must be exquisitely regulated, likely at both transcriptional and translational levels. However, the exact molecular mechanism behind transdifferentiation is unclear. It has been proposed that uncommitted adult stem cells maintain their multipotentiality by keeping a reservoir of genes representing diverse lineages actively expressed and that certain groups of genes are selectively suppressed upon stimulation before the cells commit to a given characteristic phenotype [48]. Similarly, differentiated cells could dedifferentiate, losing their committed phenotype by suppressing the lineage-specific genes while activating those that are responsible for stem cell maintenance. Since the dedifferentiation process is a functional reversal of differentiation, we can reasonably argue that genes maintaining the stemness of cells would be downregulated during differentiation and elevated during dedifferentiation.

The in vitro differentiation and dedifferentiation system developed here provides a useful platform to identify genes that might play crucial roles in stem cell self-renewal, maintenance, and multilineage differentiation. Particularly, by comparing the differentially expressed genes in three mesenchymal lineages, we were able to select a group of genes common in multiple cell types, which likely serve as markers of uncommitted hMSCs or function as regulatory factors for differentiation. One interesting gene in the differentiation genes group is IL1R2, which is involved in multiple signaling pathways (NF-κB, p38 MAPK, PPAR, and IL-6 signaling) [49, [50], [51]–52]. By acting as a decoy receptor inhibiting IL1A and IL1B activities, IL1R2 could inactivate MAP2K3 and p38 MAPK kinase activity, which in turn suppress cytokine production and apoptosis. It could also decrease NF-κB transcriptional activity, leading to the reduction of IL-6. Consistently, expression of IL-6 was downregulated during differentiation, further indicating that IL1R2-mediated NF-κB signaling might be a crucial regulatory pathway in stem cell differentiation [53, [54], [55]–56]. At present, the exact molecular mechanism of NF-κB action in stem cell differentiation is not known and requires further investigation.

Our culturing and screening system generated a list of genes that were identified previously to be enriched in undifferentiated hMSCs, including Thy-1, epithelial protein lost in neoplasm β, biglycan, dickkopf 3, decorin, thrombospondin1, steroid-sensitive gene 1, CD73, and inhibin β A [34, [35], [36], [37], [38], [39]–40]. In addition, among the stemness genes identified, several genes are also highly expressed in other stem cells, including protein tyrosine receptor type F [19], enolase 2, RAB3B, and THY-1 in ESCs [20, 21, 23]; frizzled 7, dickkopf 3, and biglycan in both ESCs [20, [21], [22]–23] and skin epithelial stem cells (SCs) [26, 27]; and brain-derived neurotrophic factor in epithelial SCs [26]. This implies that similar regulatory mechanisms might exist to control stem cell maintenance and commitment regardless of their origin or pluripotency. However, we did not detect the typical ESC markers (e.g., OCT-4, NANOG, SOX4, or FGF4) in MSCs. Lack of these markers in MSCs could be due to their subdetection expression levels or might reflect their difference from pluripotent ESCs.

Adult hMSCs are usually isolated based on their adhesion to plastic, which results in a morphologically, phenotypically, and functionally heterogeneous population of cells [30, [31]–32]. A number of cell surface markers have been successfully used to enrich and purify a morphologically homogeneous population [29, 57]. However, the fact that these markers are also expressed by fibroblasts proved that they are not sufficient to distinguish hMSCs from other cell types [38]. In addition, even the highly purified hMSC colonies presented different clonogenic and differentiation potentials in vitro [57]. Taken together, it is apparent that more markers, either surface or genetic, need to be discovered for purifying hMSCs. Lack of defined markers for MSCs also hinders their further characterization and in-depth functional analysis. Thus, one major goal of the current study was to identify molecular markers, particularly cell surface antigens, to facilitate the enrichment and purification of homogeneous MSC population. Previous studies using SAGE have generated a set of genes as potential markers for undifferentiated human MSCs [34, 35, 37]. However, very few cell surface markers were identified. In contrast, in this study, 91 genes were identified to encode surface antigens. Some genes encode well-known cell surface receptors for stem cells, such as THY-1 and CD151, whereas others appear unique in MSCs. Because of their kinetic expression profiles during differentiation and dedifferentiation, cell surface receptors identified from at least two independent lineages are likely the alternative candidate markers to enrich or purify a homogeneous population of human MSCs. Further investigation is required to confirm the feasibility of using these markers.

Stem cells in adult connective tissues, such as MSCs, are assumed to maintain a mitotically quiescent state in their native environment. Even upon stimulation by injury or remodeling, MSCs are expected to undergo tightly controlled proliferation and differentiation, as extensive cell division or differentiation may result in oncogenic conditions or excess tissue that interferes with normal physiological function, respectively. Similar regulation may also operate in cultured MSCs in vitro. The complexity of balancing cell survival, proliferation, and differentiation naturally requires the functioning and cross- talk of multiple signaling pathways. We have identified here several signaling factors that are highly expressed in undifferentiated hMSCs, including those in phosphatidylinositol-3-kinase (PI3K) signaling (PTPRF, AFAP, and RAB3B) and Wnt/β-catenin signaling (FZD7 and DKK3). As expected, all five genes exhibited similar effects on cell apoptosis and proliferation, protecting MSCs from extensive cell division as well as from cell death. PTPRF, AFAP, and RAB3B likely function through PI3K and AKT or ERK1/2 to balance cell growth or apoptosis, whereas β-catenin/LEF/TCF mediate the effects of FZD7 and DKK3. On the other hand, these five genes play different roles in lineage-specific commitment. Consistent with their expression pattern during differentiation and dedifferentiation, all five genes suppress osteogenesis. In addition, as expected, AFAP and RAB3B inhibit chondrogenesis, whereas DKK3 and FZD7 promote chondrocyte formation. However, PTPRF appears to enhance chondrogenesis, and all five genes seem to stimulate adipogenic commitment, which is inconsistent with their reduced expression during differentiation and our prediction. The mechanisms by which these genes act to effect these changes are not known, nor is the relationship of gene expression level and its normal function. Since adipogenesis is a cell-density dependent process, one possible explanation is that increased cell death caused by gene inactivation could reduce cell density, which in turn suppresses adipocyte formation. Furthermore, it is very likely that more than one signaling pathways are required during chondrogenesis of hMSCs, which cross-talk and function in a collaborative and temporal matter.

In conclusion, our study demonstrates the self-renewal, maintenance, and multilineage commitment of adult MSCs are tightly regulated by a variety of signaling pathways in a collaborative matter. The cell surface molecules identified here provide candidate markers for the enrichment and purification of genuine MSCs, as well as their identification in situ. A homogeneous population of MSCs will greatly facilitate the functional analysis of the molecular mechanisms controlling these cells. Moreover, the elucidation of the signaling pathways will enhance our ability to maintain, propagate, and expand functional MSCs in vitro; to obtain multipotential cells by inducing dedifferentiation in committed cells; and to guide MSC differentiation into specific lineage(s) for cell therapy and tissue engineering applications.


The authors indicate no potential conflicts of interest.


We thank Drs. Hong-wei Sun and Shen Huang for assistance in GeneChip data analysis. The research was funded in part by the Intramural Research Program of National Institute of Arthritis & Musculoskeletal & Skin Diseases, NIH (Z01 AR41113).