Cell Isolation Induces Fate Changes of Bone Marrow Mesenchymal Cells Leading to Loss or Alternatively to Acquisition of New Differentiation Potentials


  • Ofer Shoshani,

    Corresponding author
    1. Department of Molecular Cell Biology, Israel National Center for Personalized Medicine, Weizmann Institute of Science, Rehovot, Israel
    • Correspondence: Dov Zipori, Ph.D., Department of Molecular Cell Biology, Weizmann Institute of Science, Rehovot 76100, Israel. Telephone: 972-8-934-2484; Fax: 972-8-934-4125; e-mail: dov.zipori@weizmann.ac.il; or Ofer Shoshani, Ph.D., Department of Molecular Cell Biology, Weizmann Institute of Science, Rehovot 76100, Israel. Telephone: 972-8-934-2484; Fax: 972-8-934-4125; e-mail: ofsher@hotmail.com

    Search for more papers by this author
  • Orly Ravid,

    1. Department of Molecular Cell Biology, Israel National Center for Personalized Medicine, Weizmann Institute of Science, Rehovot, Israel
    Search for more papers by this author
  • Hassan Massalha,

    1. Department of Molecular Cell Biology, Israel National Center for Personalized Medicine, Weizmann Institute of Science, Rehovot, Israel
    Search for more papers by this author
  • Alla Aharonov,

    1. Department of Molecular Cell Biology, Israel National Center for Personalized Medicine, Weizmann Institute of Science, Rehovot, Israel
    Search for more papers by this author
  • Yossi Ovadya,

    1. Department of Molecular Cell Biology, Israel National Center for Personalized Medicine, Weizmann Institute of Science, Rehovot, Israel
    Search for more papers by this author
  • Meirav Pevsner-Fischer,

    1. Department of Molecular Cell Biology, Israel National Center for Personalized Medicine, Weizmann Institute of Science, Rehovot, Israel
    Search for more papers by this author
  • Dena Leshkowitz,

    1. Bioinformatics Unit, Israel National Center for Personalized Medicine, Weizmann Institute of Science, Rehovot, Israel
    Search for more papers by this author
  • Dov Zipori

    Corresponding author
    1. Department of Molecular Cell Biology, Israel National Center for Personalized Medicine, Weizmann Institute of Science, Rehovot, Israel
    • Correspondence: Dov Zipori, Ph.D., Department of Molecular Cell Biology, Weizmann Institute of Science, Rehovot 76100, Israel. Telephone: 972-8-934-2484; Fax: 972-8-934-4125; e-mail: dov.zipori@weizmann.ac.il; or Ofer Shoshani, Ph.D., Department of Molecular Cell Biology, Weizmann Institute of Science, Rehovot 76100, Israel. Telephone: 972-8-934-2484; Fax: 972-8-934-4125; e-mail: ofsher@hotmail.com

    Search for more papers by this author


Mesenchymal stromal cell populations include a fraction, termed mesenchymal stem cells, exhibiting multipotency. Other cells within this population possess a lesser differentiation range. This was assumed to be due to a mesenchymal cellular cascade topped by a multipotent cell, which gives rise to progeny with diminishing differentiation potentials. Here, we show that mesenchymal cells, a priori exhibiting a limited differentiation potential, may gain new capacities and become multipotent following single-cell isolation. These fate changes were accompanied by upregulation of differentiation promoting genes, many of which also became H4K20me1 methylated. Early events in the process included TGFβ and Wnt modulation, and downregulation of hypoxia signaling. Indeed, hypoxic conditions inhibited the observed cell changes. Overall, cell isolation from neighboring partners caused major molecular changes and particularly, a newly established epigenetic state, ultimately leading to the acquisition of new differentiation potentials and an altered cell fate. Stem Cells 2014;32:2008–2020


Mesenchymal cells appear early during embryogenesis and persist throughout the life span of mammalians in a body-wide distribution. They were originally thought to be connective tissue supportive cells (stroma). However, it was eventually found that single mesenchymal cells may exhibit multipotent differentiation and these were therefore termed mesenchymal stem cells [1, 2]. Careful examination of cultured mesenchymal cell populations revealed their heterogeneous nature [3]. Multipotential progenitors constitute a minority whereas the rest of the cells are either oligopotent, monopotent, or are devoid of any differentiation potential. The latter are usually referred to as “fibroblasts.” A clear molecular definition of mesenchymal stem cells is lacking and the cells are currently defined retrospectively in terms of the outcome of their differentiation following cell cloning: single isolated cells capable of producing progeny that minimally differentiate into chondrocytes, adipocytes, and osteoblasts, are by this definition, mesenchymal stem cells [4]. Thus, in this manuscript, we use the term mesenchymal stromal cell (MSC) to describe MSCs, which do not necessarily harbor the above described tripotency.

The phenotype of MSCs may not be stable and seems to be strictly determined by the conditions in which the cells are derived and propagated [5]. The stem state theory suggests that cells whether early or late in the differentiation cascade are able to enter a state (=stem state) which allows the reprogramming of the cell to become a different type of cell [6]. The discovery of induced pluripotent stem cells showed that indeed cells may become pluripotent upon reprogramming following transfection with a specific gene combination [7]. However, the question remains whether such events might occur in a spontaneous manner without the requirement of artificial reprogramming. Indeed, recent studies revealed that following tissue damage, committed cells have the potential to transdifferentiate into another cell type [8], or even dedifferentiate to become bona fide stem cells [9, 10]. However, little is known about the mechanism underlying such spontaneous transitions. In an effort to examine this, we performed herein a detailed analysis of the stability of MSC differentiation potentials. The latter was evaluated both in MSC populations and at the clonal level. Our results suggest that MSCs not only lose differentiation potentials but are also capable of gaining new ones. We found that sparse culture conditions experienced by MSCs in the process of cloning causes changes in gene expression, epigenetic profile, and signaling pathways, which ultimately lead to acquisition of new differentiation potentials.

Materials and Methods

Cell Culture

MSC populations and clones were derived from C57Bl mice (Harlan, Rehovot, Israel, www.harlan.com) as previously described [11]. The Weizmann Institutional Animal Care and Use Committee approved all animal experiments. Culture of cells in a low oxygen environment was done using a Heracell Tri-gas incubator (Thermo Scientific, Waltham, MA, www.thermoscientific.com).


Induction of adipogenesis, osteogenesis, and chondrogenesis of MSC populations and clones and quantification of oil red O in adipocytes were performed as previously described [12]. All differentiation assays were done at least three independent times. Photographs were taken using an Olympus IX71 microscope equipped with a DP51 camera.

Lentiviral Integration Site Analysis

DNA from mesenchymal cells infected with FUGW plasmid was digested with HindIII, separated using gel-electrophoresis, and probed using digoxigenin (DIG) labeled probes, according to Roche's DIG manual.

Flow Cytometry

Surface marker analysis was performed as previously described [12]. Green fluorescent protein-positive (GFP+) FUGW infected cells were analyzed using LSRII (Becton Dickinson, Franklin Lakes, NJ, www.bd.com). Cells stained with β-catenin antibody (Sigma, Rehovot, Israel, www.sigmaaldrich.com) were scanned using an ImagestreamX flow imager (Amnis, Seattle, WA, www.amnis.com). Sorting of cells infected with the 7TGC TCF/LEF reporter [13] was done using a Sorp-FACS AriaII (Becton Dickinson). DNA content analysis was performed as previously described [11]. In short, cells were stained with propidium iodide (Sigma) and DNA content was measured using LSRII (Becton Dickinson).

Real-Time PCR

RNA was extracted using Tri-Reagent (MRC, Cincinnati, OH, www.mrcgene.com) or NucleoSpin RNA II kit (Macherey-Nagel, Duren, Germany, www.mn-net.com), and cDNA was prepared using M-MLV Reverse Transcriptase (Promega, Madison, WI, www.promega.com) according to the manufacturer's protocols. Real-time PCR was carried out using Platinum SYBR Green qPCR Super-Mix (Invitrogen, Carlsbad, CA, www.lifetechnologies.com) and processed using ABI 7300 (Applied Biosystems, Foster City, CA, www.appliedbiosystems.com). Data were analyzed using the delta CT approach, using HPRT as a reference gene. HPRT was selected as a reference gene after testing several house-keeping genes (including GAPDH, ACTB, TBP, and 18S).

Soft Agar Assay

For growth in soft agar, 14 × 103 cells were plated in medium containing 0.35% noble agar (Sigma) in 6 cm plates on top of 0.8% agar. The plates were then incubated for 7 days in a 10% CO2 incubator at 37°C. Colonies and clusters were counted and imaged using a digital camera-equipped microscope. The results shown are representative of three repeats.

Microarray Analysis

Total RNA was extracted using the Tri-Reagent (MRC) according to the manufacturer's protocol. RNA purity was assessed with a BioAnalyzer 2100 (Agilent Technologies, Santa Clara, CA, www.agilent.com). Samples of 250 ng RNA from each sample were labeled and hybridized to Affymetrix mouse exon ST 1.0 microarrays according to manufacturer instructions. Microarrays were scanned using GeneChip scanner 3000 7G. Each experiment was conducted in triplicates from independent cultures. Statistical analysis was performed using the Partek Genomics Suite (Partek, Inc., St. Louis, Missouri 63141) software. CEL files (containing raw expression measurements) were imported to Partek GS. The data were preprocessed and normalized using the robust multichip average algorithm [14] with GC correction. To identify differentially expressed genes, one-way ANOVA was applied. Differential gene lists were created by filtering the genes based on: absolute fold change ≥1.5, p ≤ .05, and signal above background in at least one microarray (log 2 intensity ≥6). Differential genes ontology enrichments were found using Ontologizer [15]. Pathway enrichments on all genes above background were done also with gene set enrichment analysis (GSEA) [16] run with MSigDB. Log 2 gene intensities were used for heat maps drawn using Gene-E (http://www.broadinstitute.org/cancer/software/GENE-E/). Microarray data can be found at the GEO website, accession GSE48547.

Chromatin Immunoprecipitation

Cross-linked chromatin from approximately 5 × 106 cells was prepared and fragmented to sizes of 200–500 bps by 45 min sonication at maximum intensity using Bioruptor (Diagenode, Denville, NJ, www.diagenode.com). For immunoprecipitation, a rabbit anti-H4K20me1 (Abcam, Cambridge, UK, www.abcam.com) antibody or rabbit preimmune serum (control) was used (4 μg each). DNA was purified using MinElute PCR purification Kit (Qiagen, Hilden, Germany, www.qiagen.com) and used for CHIP-qPCR and CHIP-seq. Illumina sequencing of short reads (50 bp) was performed in four lanes of Hiseq 2000 using v2 clustering and sequencing reagents. Each ChIP sample yielded between 9.6 and 14.2 M reads and was aligned uniquely to the mouse genome (mm9) using bowtie [17]. Bound regions were detected using MACS [18]. Peaks for further analysis were filtered to have FDR % ≤5 (number of peaks is 11,809 for clone 4L.1.4 and 28,166 for MSC21). Peaks and coverage data (bigWig files) were uploaded to the UCSC genome browser [19]. Cistrome CEAS [20] was used to calculate average profile read near transcription start sites and to calculate enrichments in introns and genes. Intersection between the peaks of H4K20me1 in MSC21 and clone 4L.1.4 was done using bedtools [21]. Statistical evaluation of gene expression distributions was done using the R package (boxplot and Kolmogorov-Smirnov test, http://www.R-project.org/). CHIP-qPCR analysis was done using the % input method, in which signals obtained from the ChIP were divided by signals obtained from the input sample. ChIPseq data can be found at the GEO website, accession GSE48547.


Statistical analysis was conducted using either SPSS (IBM), or Medcalc statistical software. Unless otherwise indicated, three experimental repeats were done and analyzed using two-tailed t test.


MSC Populations with Variable Differentiation Potentials Give Rise to Clonal Isolates with Decreased or Increased Potencies

We derived independent primary MSC populations from mouse bone marrows with the aim of assessing their differentiation potential stability (Fig. 1Ai, 1Aii). Most MSC populations derived (11/15) had reduced potentials, as they did not show the classic tripotent character of mesenchymal stem cells (Supporting Information Fig. S1A), although still maintaining Sca-1 expression (Supporting Information Fig. S1B). No change in the differentiation potential of the MSC populations was detected following additional 20 population doublings. In order to evaluate the differentiation potential of single cells within the populations, single-cell clones were isolated from two primary MSCs (Fig. 1Aiii, 1Aiv), the tripotent MSC9 and the nonadipogenic bipotent MSC21 (Fig. 1B). Analysis of the differentiation potential of the clones immediately after establishment (Early clones) revealed that only 30% of those isolated from the tripotent MSC9 (Fig. 1C, MSC9 Early) was tripotent (from a total of 23 clones). This was further diminished at late passages (>20 population doublings) where only 9% maintained tripotency (Fig. 1C, MSC9 Late). Among the clones isolated from the nonadipogenic bipotent MSC21 (Supporting Information Table S1), 68% (from a total of 25 clones) showed a differentiation potential similar to that of the parent population (Fig. 1C, MSC21 Early), and this was reduced to 28% following extended culture (Fig. 1C, MSC21 Late). Surprisingly, 24% of the clones derived from MSC21 acquired the potential to differentiate into adipocytes, as 20% became tripotent, and 4% lost the chondrogenic in parallel to gain of the adipogenic potentials (Fig. 1C, MSC21 Early). This adipogenic acquisition was further increased following extended culture to a total of 60% of the clones (Fig. 1C, MSC21 Late). Some of the clones acquired a strong adipogenic tendency and even differentiated spontaneously into adipocytes. This occurred without the need for induction medium and the cells still maintained their capacity to differentiate upon induction to osteogenic or chondrogenic cells. Both MSC21 and MSC9 gave rise to clones devoid of chondrogenic potential. However, two clones isolated from MSC21 acquired a chondrogenic potential following extended culture (Supporting Information Fig. S1C). Thus, single-cell cloned mesenchymal cells originating from MSCs with limited differentiation potential, may acquire tripotency, that is, become mesenchymal stem cells by definition.

Figure 1.

MSC populations give rise to clones with decreased or increased differentiation potentials. (A): Schematic outline of the experimental design: (i) whole bone marrow from a single mouse is flushed into a culture dish, (ii) nonhematopoietic MSC populations are obtained following several passages, (iii) a single MSC is placed in a separate plate, and (iv) the single cell gives rise to an MSC clone. Both the MSC population and its derivative clones are subjected to differentiation assays at different passages. (B): MSC populations induced toward adipogenic, osteogenic, and chondrogenic lineages stained with oil red O, Alizarin red, and Alcian blue, respectively. White bars = 200 μm (×10 magnification). (C): Pie charts showing the differentiation potential of primary clones derived from the MSC populations in 1B, examined at early and late (>20 population doublings) passages (A—adipogenic, O—osteogenic, and C—chondrogenic). Full clone list derived from MSC21 is shown in Supporting Information Table S1. Total clones used for analysis were 23 clones of MSC9 and 25 clones of MSC21. (D): Frequency of adipogenic differentiation acquisition by the otherwise nonadipogenic MSC21 when seeded in 96-well plates at rising concentrations. Data are represented as mean ± SD of three experimental repeats. Abbreviation: MSC, mesenchymal stromal cell.

Sparse Culture Conditions Lead to Acquisition of Differentiation Potentials

It was then examined whether the observed changes in differentiation potentials might be cell density related, that is, a consequence of separation of the cloned cell from its neighbors. Indeed, the capacity of MSC21 to acquire adipogenic potential was found to be concentration dependent: only seeding of 100 cells/0.32 cm2, or less, resulted in positively adipogenic cells, whereas no adipocytes formed upon seeding at higher concentrations (Fig. 1D). This differentiation acquisition was not unique to MSC21 nor was it limited to the adipogenic lineage: when sparse cultures (<100 cells/0.32 cm2) of MSCs were grown, multicolony cultures with variable phenotypes were formed (Supporting Information Fig. S2A). Sparse cultures of the nonadipogenic MSC22 lead to the acquisition of adipogenic potential, similar to that of MSC21 (Supporting Information Fig. S2B). Colonies rising from sparse MSC21 or MSC22 cultures showed distinct morphologies (Supporting Information Fig. S2C), some appearing epithelial or endothelial like. Indeed, Western analysis revealed that colonies rising from sparse cultures of MSC21 or MSC21 express the epithelial marker E-cadherin and the endothelial marker vWF, while the original MSC populations lacked such expression (Supporting Information Fig. S2D). The expression of these markers was also validated on the single colony level using immunostaining (Supporting Information Fig. S2E, S2F).

A Gained Differentiation Potential May Disappear and Reappear as a Result of Sequential Cloning

To further analyze the differentiation potential stability of MSCs, sequential cloning experiments were performed (Fig. 2A). The resulting clonal lineage tree of MSC21 is portrayed in Supporting Information Figure S3A. This sequential cloning revealed that nonchondrogenic clones can acquire such a potential after additional cloning (Supporting Information Fig. S3B). Examination of secondary clones derived from primary clone 4 at an early passage (Supporting Information Fig. S3A, clone 4E) showed that its acquired adipogenic potential was unstable. Indeed, these secondary clones did not maintain an adipogenic potential. However, in later passages, following extended culture (Supporting Information Fig. S3A, clone 4L) the adipogenic acquisition of primary clone 4 became more stable, as the subsequent clones maintained this potency. Sequential cloning procedures revealed that the adipogenic potential can be turned on, off, and back on repeatedly (Fig. 2B, 2C and Supporting Information Fig. S3A). Clones acquiring an adipogenic potential expressed pparγ 1 and 2, as well as zfp423, genes associated with adipogenesis, at much higher levels compared to the nonadipogenic MSC21 (Fig. 2D). The adipogenic differentiation potential and adipogenic gene expression disappeared and reappeared in a correlative manner upon sequential cloning. The expression of zfp423, a known preadipocyte marker [22], was higher in clones 4L.1 and 4L.1.4 and in MSC9, which differentiate spontaneously into adipocytes. This property did not interfere with their ability to differentiate into osteocytes upon appropriate induction. Similar fate changes following cellular cloning were observed in the independent MSC20 population (Supporting Information Fig. S3C). Thus, it appears that cells introduced to repeated sparse conditions may respond either by losing or alternatively by gaining differentiation potentials.

Figure 2.

MSC differentiation potential is unstable during sequential cloning. (A): Schematic outline of the experimental design: (i–iv) as in Figure 1A (v–vi) sequential clonings. A single cell is taken from a primary MSC clone and placed in a separate plate, in similar to step iv. The dashed line illustrates this step is repeated to produce secondary, tertiary, and quaternary clones. (B): One example of the differentiation potential of MSC clones following serial cloning (red—osteogenic, blue—chondrogenic, and yellow—adipogenic). Full clonal lineage tree is shown in Supporting Information Figure S3A. (C): Oil red O staining after adipogenic differentiation of the clones presented in (B). White bars = 200 μm (×10 magnification). (D): Expression of genes involved in adipogenesis (relative to HPRT) of the clones presented in (B) in control (−) and adipogenic induced (+) conditions. Data are represented as mean ± SD (y-axis in log scale). *: significant expression increase of pparg2 after adipogenic induction (p < .05, Student's t test). #: significant increase in expression compared to MSC21 (p < .05, Student's t test). Abbreviation: MSC, mesenchymal stromal cell.

The Acquisition of Adipogenic Differentiation Potential Occurs at the Single-Cell Level

The question was then raised as to whether the acquisition of potentials occurred at the single-cell level. MSC21 was therefore infected with a lentivirus carrying GFP, which integrates into random loci within the genome, and subsequently uniquely marks individual cells, allowing lineage tracing of clones (Fig. 3A, 3B). Low infection conditions were used in order to minimize the amount of integrated lentiviral particles infecting each cell. The GFP signal served as an indicator of infection efficiency (Fig. 3C and Supporting Information Fig. S4A). Importantly, the GFP infection did not affect the frequency of adipogenic acquisition upon serial dilution (Supporting Information Fig. S4B). GFP+ clones derived from single cells, which do not a priori have an adipogenic differentiation potential, were isolated. Two such primary clones (clones C and D, Fig. 3C) were subjected to a second round of cloning to give rise to GFP+ secondary clones which were examined for their adipogenic differentiation potential. As observed before, many secondary clones acquired an adipogenic potential (16/25) and corresponding gene expression patterns (Fig. 3C and Supporting Information Fig. S4C). The secondary clones contained an identical lentiviral integration site to that of their parent clone, thus proving that the acquisition of adipogenic potential occurred in a single cell derived from a clone (Fig. 3D). To dismiss the possibility that some cells within the MSC21 population actually have an adipogenic potential, but are inhibited by surrounding nonadipogenic cells, we performed a coculture assay. When coculturing the nonadipogenic MSC21 with one of its adipogenic clones (clone 4L.1.4, Fig. 3E), at different ratios, no evidence was found for adipogenic inhibition. This was true also in cocultures of the nonadipogenic primary clone MSC20.1 and the adipogenic secondary clone MSC20.1.1 (Supporting Information Fig. S4D). To determine whether the observed changes in differentiation potential and gene expression were accompanied by changes in chromosomal content and cellular transformation we performed DNA content and soft agar colony assays (Supporting Information Fig. S5). MSC21, MSC21-GFP, and their derivative clones which acquired an adipogenic differentiation potential maintained a relatively similar DNA content, except for clone MSC21-GFP.C.4 which had increased DNA content (Supporting Information Fig. S5A, S5B). Also, none of the cells examined were able to form colonies in a soft agar colony assay (Supporting Information Fig. S5C). Thus, we conclude that there is no evidence for the involvement of cellular transformation in the process of cell fate change we observed. Overall, it is shown that mesenchymal cells undergo dramatic changes in cell phenotype solely due to their isolation from neighboring cells. Further experiments were aimed at determining the molecular events that accompany these fate shifts.

Figure 3.

MSC differentiation potentials change at the single-cell level. (A): Schematic outline of the experimental design: (i) MSC population is infected with a lentivirus. (ii) Single GFP+ cells are placed in a separate plate. (iii) GFP+ clones are examined for adipogenic potential. (iv) GFP+ single cells taken from nonadipogenic primary clones are placed in a separate plate. (v) The adipogenic potential of the secondary clones is examined. The viral integration site within secondary clones which acquired adipogenic potential is compared to their primary parent clone. (B): Schematic demonstrating targets of the DIG labeled probe (88 bp long) designed against the LTR regions of the lentiviral insert (FUGW vector). A 556 bp internal control and an unknown length target composed of viral and host genomic DNA dependent on gDNA HindIII location downstream of the viral insertion site are expected. Horizontal black lines—viral insert DNA, vertical black lines show LTR regions on both sides of insert, green—host cell gDNA, vertical red lines—HindIII restriction sites within the viral insert (dashed line marks host's gDNA proximal HindIII restriction site downstream of viral insert), blue—probe recognition sites within viral LTRs. (C): Adipogenic induction of MSC21-GFP and its primary (C and D) and secondary (C.4 and D.9) clones stained with oil red O. MSC21-GFP is 12% GFP positive and the clones shown are 100% GFP positive (verified by fluorescence-activated cell sorting and microscopy). Fluorescent photographs were taken at identical exposure times. White bars = 200 μm (×10 magnification). (D): Southern blot analysis of MSC21-GFP primary and secondary clones shown in (C) probed specifically as explained in (B). The viral vector FUGW was cut with HindIII as a ladder control (FUGW HindIII). A HindIII restriction map of pFUGW is displayed (from the nebcutter website). Arrows mark viral integration sites (upper arrow—unique site of clones D and D.9, middle arrow—unique site of clones C and C.4, lower arrow—viral control). (E): Quantification of oil red O staining in the nonadipogenic MSC21 and its adipogenic tertiary clone MSC21.4L.1.4, at different cell ratios. Photographs on top are representatives of each cell ratio. Data are represented as mean ± SD. *, p-value < .0001 (Student's t test). Abbreviations: GFP, green fluorescent protein; MSC, mesenchymal stromal cell.

MSC Clones Exhibit Altered Gene Expression Patterns and H4K20me1 Methylation Profiles

The gene expression profiles of the nonadipogenic MSC21 and its descendent tripotent tertiary clone 4L.1.4 were examined by DNA microarray analysis, and differentially expressed genes (p < .05, fold-change > 1.5) were analyzed for ontology enrichments (Fig. 4A). As expected, adipose cell differentiation groups were significantly different (Fig. 4B). Additional major differences were observed in cell fate determining genes, such as genes involved in the Wnt signaling (Fig. 4B, 4C). This signaling pathway is known to be involved in general stem cell fate determination [23], and in differentiation processes of MSCs [24]. It thus seemed reasonable that Wnt signaling might also be involved in the observed acquisition of potentials. Previously, the Wnt signaling pathway was shown to be modulated by histone modification H4K20me1 [25]. ChIP-seq analysis comparing the H4K20me1 profile of MSC21 and clone 4L.1.4 showed that H4K20me1 peaks concentrated in gene regions, preferentially at the 5′ intronic regions (Supporting Information Fig. S6A, S6B), as previously shown [26]. Histone modification H4K20me1 was found to be enriched in epithelial-mesenchymal transition (EMT)/mesenchymal-epithelial transition (MET) and MSC-related genes, and as expected also in Wnt-related genes (Fig. 4C). Specifically, many genes regulating adipogenic differentiation (i.e., lpl, pparγ, cebpa, and cebpb) that were highly expressed in clone 4L.1.4 were also H4K20me1 methylated. The expression of genes uniquely H4K20me1 methylated in either MSC21 (2,468 genes) or in clone 4L.1.4 (139 genes) was higher in MSC21 (p < .0005) or clone 4L.1.4 (p < .0005), respectively (Fig. 4D). In general, genes methylated in both cells were highly expressed, and genes not methylated were less expressed. These results were further validated (Supporting Information Fig. S6C, S6D) by analyzing the expression of several genes found to have differential peaks, including: lpl, cebpa, rspo2, and wisp1 (Fig. 4E). Additional analysis revealed that the expression of imprinted genes is significantly different between the population and the clone (p = .0001599, hyper geometric test), as among the 100 imprinted genes which are present in the array, 36 were differentially expressed (p < .05). The differential expression of 15 imprinted genes was at least twofold (significance of difference between the population and the clone: p = 1.336E-6, hyper geometric test). The changes in H4K20me1 methylation and gene expression occurred at the clonal level, as clone 4L had a similar profile to that of MSC21, both being different from clone 4L.1.4 (Supporting Information Fig. S6C, S6D). Overall, a comparison of the original MSC population to clones derived from it and propagated in vitro showed that the cell fate changes observed are accompanied by changes in H4K20me1 methylation and expression of fate determining genes.

Figure 4.

H4K20me1 modifications accompany changes in gene expression between an MSC population and its derived clone. (A): Schematic outline of the experimental design: (i) a single cell from the nonadipogenic population MSC21 is placed in a separate plate to establish a primary clone. (ii) This is repeated (dashed line) until the adipogenic tertiary clone MSC21.4L.1.4 is obtained. Then, both cultures are analyzed using DNA microarray and ChIP-seq (H4K20me1). (B): Relevant enriched gene ontologies in upregulated genes within clone 4L.1.4 compared to parent MSC21 (using Ontologizer). (C): Heat-map showing significant differences between MSC21 and clone 4L.1.4 in the expression of key gene groups which govern MSC fate, and their corresponding H4K20me1 methylation. Data presented are row relative microarray expression. Black boxes indicate significant H4K20me1 peaks overlapping genes. (D): Box-plot showing the gene expression (log scale) distribution of four groups: (I) genes with H4K20me1 methylation peaks in both MSC21 and clone 4L.1.4, (II) genes with no peaks in either, (III) genes with unique peaks in clone 4L.1.4, and (IV) genes with unique peaks in MSC21. The gene expression distribution of MSC21 and clone 4L.1.4 was compared between each gene group. Genes with unique peaks in clone 4L.1.4 were significantly upregulated in clone 4L.1.4 (*, p < .0005). Genes with unique peaks in MSC21 were significantly upregulated in MSC21 (**, p < .0005) Kolmogorov-Smirnov test. No differences in average expression were found in the two other gene groups (n.s.—not significant). Error bars represent minimum and maximum values. (E): Representative ChIP-seq regions viewed by the UCSC genome viewer showing the read coverage at significant H4K20me1 methylation bound peaks in clone 4L.1.4 (lpl, cebpa, and rspo2) and in MSC21 (wisp1). Abbreviation: MSC, mesenchymal stromal cell.

Early Events Leading to Cell Fate Changes in Sparsely Growing Mesenchymal Cells

To elucidate the early molecular events occurring shortly after detachment of cells from their neighbors, the gene expression profiles of sparsely seeded MSC21 cultures were compared to those of dense populations after 1, 2, and 3 days of culture (Fig. 5A). The genes analyzed listed in Figure 4C above as distinguishing between parental and cloned cell populations, including the Wnt-related genes, were also found to be differentially expressed in sparse versus dense conditions (Supporting Information Fig. S7A). Comparison of our data to published datasets using GSEA showed that the expression of MSC and stroma gene signatures was significantly enriched in the dense cultures (Fig. 5B). Also, genes that were previously found to be regulated by TGFβ1 were significantly expressed in the dense culture. In contrast, genes shown to be highly repressed by CpG and H3K27me3 methylations were increased in the sparse cultures. This indicates that sparse culture conditions dramatically alter the gene expression program of MSCs early following seeding. Additional changes were found in the expression of small nucleolar RNAs, which were previously implicated in fate determination of MSCs (Supporting Information Fig. S7B) [27]. Specifically, snora44 expression was high at the early stages of sparse cultures (Supporting Information Fig. S7C). In accordance with the above findings suggesting the modulation of the Wnt signaling during sparse culture, we examined the localization of its canonical mediator. Indeed, sparsely seeded cells showed a slight increase in nuclear and a significant decrease in membrane β-catenin localization, compared to dense cultures, indicating that the canonical Wnt signaling is activated (Fig. 5C–5E and Supporting Information Fig. S7D).

Figure 5.

Sparse conditions trigger multiple molecular changes including β-catenin translocalization. (A): Schematic outline of the experimental design. (i) A confluent MSC21 plate was passaged (1:4) into three plates and cells collected at consecutive days (days 1, 2, and 3 after seeding, day 3 was almost 100% confluent) for gene expression analysis. (ii) Also, from confluent MSC21, 15,000 cells were taken and placed in a separate plate to produce sparse cultures (multiple such plates were seeded). Following 1, 2, and 3 days of sparse culture, cells were collected for gene expression analysis. Both dense and sparse cultures were analyzed for β-catenin localization using imagestream fluorescence-activated cell sorting analysis and traditional immunostaining. (B): Gene set enrichment analysis comparing the dense-sparse microarray data to four published datasets: dense cultures upregulate (from a total of 405 datasets with FDR q-val < 0.05 which were significantly similar): (I) genes expressed in adipose derived CD31− stem cells (Boquest), (II) genes downregulated in MEF cells upon stimulation with TGFB1 (Plasari), and (III) genes discriminating stromal cells that can support hematopoietic stem cells from those that cannot. Sparse cultures upregulate (from a total of 130 datasets with FDR q-val < 0.05 which were significantly similar) genes with high-CpG-density promoters (HCP) and bearing H3K27me3 in MEF cells. (C): Representative flow cytometry (imagestream) analysis of β-catenin localization within MSC21 cells cultured in dense and sparse conditions (D2—2 days after seeding). Colocalization of DAPI and β-catenin (similarity score) indicated nuclear localization of β-catenin. β-Catenin signal relative to the cell center (max contour) was used to determine membranal localization (%—indicates gate used to compare membrane localization of β-catenin). (D): Summary of imagestream experiments as shown in (C). D1 and D2 represent sparse cultures 1 day and 2 days after seeding, respectively. Data are represented as mean ± SD (*, p < .05, paired t test). Representative photographs of nuclear localized (high similarity) and membrane localized (max contour) β-catenin (b-cat) are shown. (E): Immunofluorescense images of β-catenin localization in dense and sparse cells in culture 1 day after seeding (merge of DAPI-blue and β-catenin-Cy3). Dotted ellipses mark cell nucleus. White bars = 200 μm (×10 magnification). Abbreviations: MSC, mesenchymal stromal cell; NES, normalized enrichment score.

Modulation of Wnt Signaling due to Sparse Culture Conditions

A direct examination of Wnt signaling using a GFP TCF/LEF reporter [13] inserted into MSC21 and two of its clones (4L and 4L.2.2) was then performed (Fig. 6A). This analysis revealed that under dense conditions no Wnt activation occurs (Fig. 6B, Dense-control). After treatment with LiCl, GFP+ cells with activated Wnt appeared, as expected (Fig. 6B, Dense-LiCl). In contrast, some sparse cells became GFP+ without further treatment (Fig. 6B, Sparse-control). This GFP signal was not maintained during culture, and disappeared following a few days. Once the sparse cultures reached confluence, some colonies became GFP+ (Fig. 6B, sparse → dense-control and Supporting Information Fig. S8A). Many of the GFP+ colonies formed had a distinct morphology, some appearing epithelial-like (Supporting Information Fig. S8B), suggesting a MET-like process had occurred. Treatment with TGFβ2 did not prevent the appearance of GFP+ isolated cells in sparse cultures (Fig. 6B, Sparse-TGFβ2), however, once reaching confluence, no GFP+ colonies were visible (Fig. 6B, sparse → dense-TGFβ2). This suggests that any possible MET-like events were inhibited. Seeding of GFP− cells gave rise to GFP+ colonies, whereas selection of GFP+ cells did not increase the amount of GFP+ cells above ∼10% in subsequent sparse cultures (Fig. 6C). Statistical analysis shows that the outcome of the initial sparse → dense procedure determined the probability of the cell to become GFP+ in the subsequent sparse → dense procedure (Fig. 6D). Expression of wisp1, one target of Wnt signaling activation was elevated in GFP+ cells (Supporting Information Fig. S8C). In contrast, the expression of lpl and cebpa was higher in GFP−, as expected from previous publications indicating that Wnt inhibits adipogenesis [28]. MSC marker expression analysis showed that GFP+ cells significantly expressed less nestin, sca-1, and thy1 and more nt5e than GFP− cells (Fig. 6E). Specifically, repeated seeding in sparse conditions resulted in a reduction of thy1 and sca-1 expression, regardless of GFP expression (Supporting Information Fig. S8D). Taken together, this data reveal that cell isolation leads to Wnt activation in morphologically distinct colonies with altered MSC marker profile.

Figure 6.

Sparse conditions lead to activation of Wnt signaling and alteration of MSC marker expression. (A): Schematic outline of the experimental design. MSC21 population and two of its clones, 4L and 4L.2.2, were infected with the 7TGC reporter (the infection was monitored using the constitutive mcherry signal). From dense infected cultures, 15,000 cells were seeded in separate plates to establish sparse cultures (i). Upon reaching confluence, these originally sparse cultures were photographed using a fluorescent microscope and separated using FACS for GFP+ and GFP− cells, each was reseeded as new sparse cultures. This was then repeated again. (B): Activation of WNT signaling at different conditions as detected in MSCs infected with the 7TGC reporter. Representative images of cells in dense, sparse and sparse → dense (sparsely seeded cells reaching confluence) conditions are shown. White bars = 200 μm (×10 magnification), blue bars = 100 μm (×20 magnification), orange bars = 500 μm (×4 magnification). (C): Analysis of GFP-positive 7TGC infected cells (indicating WNT activation) using flow cytometry. Clones 4L and 4L.2.2 were infected with the 7TGC reporter and %GFP+ at dense and sparse → dense conditions was determined. Then, GFP positive and negative cells were separated from the sparse → dense cultures by FACS (sort I), reseeded at sparse conditions and allowed to regain confluence (3 weeks in culture). This was followed by an additional sorting (sort II) stage and culture until reaching confluence. %GFP+ was determined in each step. Data are represented in log-scale as mean ± SD. (D): Two-way ANOVA analysis performed on the data presented in (C). The results show that there is no significant interaction between sort I and II in terms of GFP probability. This means that the probability of the cells to be GFP+ after sort II primarily depends whether they were GFP+ after sort I. (E): MSC marker expression (relative to HPRT) in cells with active WNT (GFP+) compared to cells with inactive WNT (GFP−). Values are ratios of GFP+/GFP− and are represented as mean ± SD (statistical difference between GFP+ and GFP− of each gene: *, p < .005; **, p = .0005; ***, p < .05; n.s.—not significant, paired t test). Abbreviations: FACS, fluorescence-activated cell sorting; GFP, green fluorescent protein.

Hypoxia Inhibits the Fate Shifts in Mesenchymal Cell Populations

Besides the obvious lack of cell-to-cell contact in sparse cultures, we found that dense and sparse cultures also differ by their sensation of oxygen levels. Dense cultures showed a hypoxic gene expression signature compared to sparse cultures, suggesting they experience relative hypoxia (Fig. 7A). As hypoxic conditions were reported to affect the proliferation and differentiation of stem cells through Wnt signaling [29], we examined the effect of different oxygen levels on activation of Wnt in mesenchymal cells. Under hypoxic conditions, cell proliferation remained constant even when seeding low cell numbers (Supporting Information Fig. S8E), in contrast to a cell density dependent decrease in the proliferation rate under normoxic conditions. Also, sparse cultures at hypoxic conditions did not give rise to Wnt active GFP+ colonies (Fig. 7B, 7C). Finally, a two-third reduction in the frequency of adipogenic potential acquisition in hypoxic compared to normoxic conditions was observed (Fig. 7D). These findings imply that hypoxic conditions prevent cell fate changes of isolated mesenchymal cells, while allowing maintenance of the proliferative capacity.

Figure 7.

Hypoxia abrogates the effects of sparse conditions on Wnt activation and adipogenic acquisition. (A): Gene set enrichment analysis showing that dense cultures upregulate genes highly expressed in hypoxia (Mense_Hypoxia_Up), and sparse cultures upregulate genes repressed in hypoxic conditions (Manalo_Hypoxia_Dn). NES—normalized enrichment score. (B): Lack of activation of WNT signaling in cells infected with the 7TGC reporter and grown from sparse to dense cultures under hypoxic conditions (3% O2). Representative images of cells are shown. White bars = 200 μm (×10 magnification). (C): Expression of GFP (indicating WNT activation) in clones 4L and 4L.2.2 after seeding at sparse conditions and reaching confluence is blocked under hypoxic conditions (GFP levels are similar to that of dense cultures as shown in Fig. 6C). Data (% cells) are represented in log-scale as mean ± SD. (D): Frequency of adipogenic differentiation acquisition of nonadipogenic MSC21 seeded in 96-well plates (10 cells/well) at hypoxic (3% O2) and normoxic (20% O2) conditions. Data are represented as mean ± SD (*, p = .0268, paired t test). (E): Summarizing schematic. Dense MSCs are under steady-state conditions (high TGFβ and low available O2 levels in the medium). Once isolated, the cell's environment is changed (low TGFβ and high available O2 levels in the medium), leading to modulation of the Wnt signaling pathway. Upon re-establishment of confluence, the cells show an altered fate illustrated by the formation of H4K20me1 marks in differentiation-related genes which enable their expression and lead to acquired differentiation potentials. Abbreviations: GFP, green fluorescent protein; MSC, mesenchymal stromal cell.


In this study, we show that cloning or seeding at low density of cells may lead to decreased or increased differentiation potentials. This finding indicates that defining mesenchymal stemness by examining differentiation potentials following single-cell cloning is misleading, and that this approach should be re-evaluated. Cell progeny raised from single or sparse cell cultures exhibited dramatic changes in their gene expression profile including changes in TGFβ and Wnt signaling, critical to mesenchyme fate [30]. These provide a possible explanation to the observed acquisition of differentiation potentials. Hypoxic conditions, thought to represent the environment of stem cell niches in vivo, blocked the fate transitions imposed by cell isolation. It is possible therefore that the observed gain of differentiation potentials in vitro represents events that may also occur in vivo under stress, when progenitors are driven out of their niches (Fig. 7E). We found evidence that implicates the Wnt signaling pathway in the process of cell fate change. Sparse culture conditions promoted Wnt activation at different stages of the culture. Initially, some single cells were found to have activated Wnt in the initial days of sparse culturing. As Wnt is known to inhibit adipogenesis [28], our finding that Wnt is activated in the cell fate process and could lead to acquired adipogenic potential might seem contradicting. However, our observations clearly show that the timing of Wnt activation is of importance, and thus it is possible that this early Wnt activation might be responsible for the redirection of the cell fate, allowing future adipogenic differentiation in established clones which have inactive Wnt. Following this early activation, Wnt was inactive and was found active again in some colonies upon reaching confluence, but never in adipogenic colonies. The fact that such colonies exhibited a different epithelial-like morphology might suggest that Wnt plays a different role at this stage, possibly related to a MET-like process.

Isolation of cells from their neighbors imposed epigenetic changes, since normally repressed genes were found active in sparse cultures, and imprinted gene expression differed significantly between the population and its clone. Specifically, H4K20me1 methylation, involved in Wnt signaling [25], was found to be modulated, thus creating a new epigenetic state in clonal isolates. These results might explain previously described cases in which differentiation potentials of cells were increased [8–10, 31]. The observed instability of the MSC phenotype may point to the in vivo function of these cells [32]. It is possible that migrating MSCs, transiting from one in vivo local to another, are affected by the changes in the microenvironment and accordingly change their differentiation potential or other properties such as the expression of cell surface markers. One microenvironment factor discovered by our in vitro analysis was oxygen. It appears that the concentration of the cell cultures has a profound impact on the availability of oxygen to the cells, as indicated by the fact that dense cells showed a hypoxic gene signature compared to the sparse cultures. Oxygen levels also improved cell proliferation, and inhibited the ability to acquire new differentiation potentials (e.g., adipogenic potential). The fact that Wnt was inhibited in hypoxic conditions suggests it might serve as a link between oxygen response and cell fate alteration. Indeed, previous studies showed that Wnt might be activated [29], or inhibited [33, 34], by hypoxic conditions, indicating that its modulation might be context dependent. Therefore, the nature of the O2-Wnt-Cell fate link is an important matter which should be further clarified in future studies.


The results demonstrated that MSCs with a limited differentiation potential are able to undergo cell fate changes when cultured in sparse conditions. This leads to loss of potentials in some instances, and surprisingly also to gain of potentials up to the point of acquisition of multipotency. Also, single-cell cloning is sufficient to completely alter the nature of the cell which becomes divergent from the parental population. Apparently, the Wnt signaling pathway plays an important role in the process, leading to epigenetic modulation of H4K20me1 and to expression of differentiation-related genes. This was inhibited once the cells were cultured under hypoxic conditions. Such fate changes might occur in vivo under appropriate environmental settings.


This study was supported by The Leona M. and Harry B. Helmsely Charitable Trust. D.Z. is an incumbent of the Joe and Celia Weinstein Professorial Chair at the Weizmann Institute of Science. O.S. is currently affiliated with the Ludwig Institute for Cancer Research, University of California, San Diego, CA.

Author Contributions

O.S.: conception and design, manuscript writing, collection of data, data analysis and interpretation, bioinformatics analysis, and final approval of manuscript; D.Z.: conception and design, manuscript writing, and final approval of manuscript; O.R., H.M., A.A, Y.O., and M.P.-F.: collection of data and final approval of manuscript; D.L.: bioinformatics analysis and final approval of manuscript.

Disclosure of Potential Conflicts of Interest

The authors indicate no potential conflicts of interest.