Adult bone marrow stromal stem cells express germline, ectodermal, endodermal, and mesodermal genes prior to neurogenesis



Bone marrow stromal stem cells (MSCs) normally differentiate into mesenchymal derivatives but recently have also been converted into neurons, classical ectodermal cells. To begin defining underlying mechanisms, we extended our characterization of MSCs and the differentiated neurons. In addition to expected mesodermal mRNAs, populations and clonal lines of MSCs expressed germinal, endodermal, and ectodermal genes. Thus, the MSCs are apparently “multidifferentiated” in addition to being multipotent. Conversely, the differentiating neurons derived from populations and clonal lines of MSCs expressed the specific markers β-III tubulin, tau, neurofilament-M, TOAD-64, and synaptophysin de novo. The transmitter enzymes tyrosine hydroxylase and choline acetyltransferase were localized to neuronal subpopulations. Our observations suggest that MSCs are already multidifferentiated and that neural differentiation comprises quantitative modulation of gene expression rather than simple on–off switching of neural-specific genes. © 2002 Wiley-Liss, Inc.

Development involves the progressive restriction of cell fate: Early embryonic cells are totipotent (Chang and Hemmati-Brivanlou, 1998), primary germ layer cells exhibit restricted fates as gastrulation commences (Wei et al., 2000), and postnatal stem cells are thought to differentiate only into cells of their resident tissues (Armstrong and Svendsen, 2000). Stem cells represent a rare population in postnatal animals; although not totipotent, they are capable of self-renewal and differentiate into more than one specialized cell type (Kopen et al., 1999). In postnatal mammals, brain, liver, intestine, and bone marrow are now known to contain “multipotent” stem cell populations (Wei et al., 2000). We have recently differentiated adult marrow stromal stem cells, classical mesodermal derivatives, into neurons (Woodbury et al., 2000), traditional ectodermal cells, and hence reexamine adult stem cell commitment and potentiality.

Bone marrow contains two prototypical stem cell populations, those of the lymphohematopoietic lineage (HSCs) that continually repopulate the circulation, and the stromal cells (MSCs), mesodermal elements that normally give rise only to mesenchymal derivatives (Friedenstein, 1976; Caplan, 1991; Kuznetsov et al., 1997; Kadiyala et al., 1997; Deans and Moseley, 2000). Differentiation of MSCs into neural elements in vitro (Woodbury et al., 2000; Sanchez-Ramos et al., 2000; Deng et al., 2001; Kohyama et al., 2001) and in vivo (Kopen et al., 1999; Zhao et al., 2002) has raised fundamental questions about stem cell lineage, commitment, and plasticity.

Indeed, neural stem cells (NSCs), which normally differentiate into neurons and glia in vitro (Reynolds and Weiss, 1992; Lois and Alvarez-Buylla, 1993; Morshead et al., 1994; Gage et al., 1995; McKay, 1997), can differentiate into hematopoietic (Bjornson et al., 1999) and muscle (Galli et al., 2000) cells in vivo. Moreover, NSCs transplanted to the permissive environment of the blastocyst participate in the generation of tissues of all three germ layers (Clarke et al., 2000). Consequently, NSCs are not restricted to neural, or even ectodermal, fates, suggesting that stem cell potency is far more extensive than had been appreciated. The recent demonstration that marrow-derived stem cells can populate the brain and differentiate into neural cells underscores the plasticity of stem cell populations (Brazelton et al., 2000; Mezey et al., 2000).

This plasticity is exhibited by adult MSCs as well. MSCs normally differentiate into bone, cartilage, muscle, tendon, and fat, classical mesenchymal derivatives (Owen, 1988; Beresford, 1989; Young et al., 1998; Pittenger et al., 1999). We recently found that MSCs can be induced to overcome their mesenchymal fate and differentiate into neurons in vitro (Woodbury et al., 2000). Treatment with a relatively simple, fully defined medium elicited neuronal differentiation of approximately 80% of the cells. Within 5 hr, the MSCs formed characteristic refractile spherical cell bodies and extended typical long neuritic processes exhibiting terminal expansions and filopodia. The treated MSCs expressed the neuroepithelial gene product nestin, transiently, and the neuronal products neurofilament M (NF-M), tau, and Neu-N de novo and increased expression of neuron-specific enolase (NSE). These unanticipated observations suggest that MSCs may be useful in the treatment of a number of neurological diseases and raise basic questions regarding stem cell biology and mechanisms of differentiation.

What is the substance of MSC differentiation into neurons; what is the pattern of altered gene expression, morphology, and function? What molecular mechanisms govern the differentiation of mesodermal stem cells into classical ectodermal derivatives? How is conversion achieved in minutes to hours? What are the limits of MSC potential and plasticity; can MSCs give rise to cells of all the classical germ layers? To begin approaching these and related questions, here we extend our characterization of the MSCs and derived neurons, using reverse transcription-polymerase chain reaction (RT-PCR) and immunocytochemistry to define gene expression. We report that populations and clonal lines of MSCs express germline, ectodermal, endodermal, and mesodermal genes, allowing far broader potential than expected. In turn, neural differentiation apparently comprises quantitative modulation of gene expression rather than simple on–off switching of neural-specific genes.


MSC Isolation and Culture

rMSCs were isolated from the femurs of adult rats as previously described (Azizi et al., 1998) and maintained in Dulbecco's modified Eagle's medium (DMEM; Life Technologies, Bethesd, MD) supplemented with 20% Fetal Bovine Serum (FBS, Atlanta Biologicals, Atlanta, GA). Early-passage MSCs were subjected to flow cytometry to determine purity as described by Woodbury et al. (2000). MSCs used in these studies were extensively propagated, passaged from 15 to >20 times in vitro. The generation of clonal lines of MSCs has been described in detail (Woodbury et al., 2000). Briefly, MSCs were plated at low density in 150 mm dishes and incubated for 2 hr to allow cell attachment. Supernatant was removed, and dishes were washed with PBS to eliminate unattached cells. Dishes were examined microscopically, and well-separated single cells were identified and marked. Colonies arising from these single cells were expanded into clonal lines.

Neuronal Induction

Neuronal differentiation was performed as described elsewhere (Woodbury et al., 2000), with modification. Briefly, prior to neuronal induction, rMSCs were grown overnight in DMEM, 20% FBS, 10 ng/ml basic fibroblast growth factor (bFGF). The monolayer was rinsed twice with PBS and transferred to neuronal induction media (NIM) consisting of 100 μM BHA, 10 μM forskolin, 2% DMSO, 5 U/ml heparin, 5 nM K252a, 25 mM KCl, 2 mM valproic acid, 1× N2 supplement (Life Technologies), 10 ng/ml bFGF, 10 ng/ml platelet-derived growth factor (PDGF) in a base of DMEM. After induction, cells were maintained at 30°C without further additions. For reversion studies, NIM was removed and replaced with unsupplemented DMEM.

Osteogenic Induction

Osteogenic differentiation was performed as described by Pittenger et al. (1999). MSCs were maintained in osteogenic induction media for 13 days, with fresh media added every other day. Cells were fixed with 4% PFA and stored at 4°C until stained.

von Kossa Staining

Staining was performed as described elsewhere (Lennon et al., 2001) on MSCs subjected to osteogenic differentiation (see above).

RNA Isolation

RNA was isolated from induced and control rMSCs using Trizol reagent according to manufacturer's recommendations (Life Technologies). The resulting RNA pellet was subjected to a chloroform extraction and two ethanol precipitations. Yield was determined spectrophotometrically.

cDNA Synthesis

Two micrograms of RNA were reverse-transcribed using Superscript II reverse transcriptase (Life Technologies) in a 50 μl volume containing 1 μg oligo-dT primer, 200 μM dNTPs, and buffers supplied by the manufacturer. The reaction was carried out in a Perkin-Elmer 9600 polymerase chain reaction (PCR) machine with the following parameters: 25°C 5 min, 37°C 5 min, 42°C 60 min, 48°C 10 min. A 5 min ramp time was employed between each temperature. In control reactions the Superscript II reverse transcriptase was omitted.


cDNA target (2–5 μl) was amplified by PCR using specific primer pairs (listed below). Tfl polymerase (Epicentre) and the PCR Optimization Kit (Epicentre) were utilized following the manufacturer's recommendations. To ensure equal distribution of target, a master mix containing all components except for specific primers was generated and then aliquoted to each reaction tube. PCRs were performed in a Perkin-Elmer 9600 as follows: initial 3 min denaturing step at 92°C followed by 30–35 cycles of 94°C 5 sec, 55–65°C 10 sec, 68°C 30 sec; 74°C 30 sec. All reactions were performed in a 20 μl volume. Primers sequences and Genbank accession numbers (in parentheses) were as follows:

  • Ceruloplasmin (L33869)



  • SM22α (L41154)



  • Protamine 2 (X14674)



  • Aldolase (M63656)



  • Amyloid precursor protein (X07648)



  • NMDA Glutamate binding subunit (S61973)



  • Syntaxin (AF044581)



  • Neurofilament-M (Z12152)



  • Noggin (U79163/U31203)



  • Tau (X79321)



  • NeuroD (D82945)



  • GFAP primer sequences were based on those reported by Condorelli et al. (1999):




Cells were fixed in 4% paraformaldehyde and stored under PBS at 4°C until stained. Primary antibodies were polyclonal antitau, Sigma (1:1,000); polyclonal anti-TOAD-64, Chemicon (1:500); monoclonal anti-β-III tubulin, Chemicon (1:200); monoclonal antisynaptophysin, Chemicon (1:200); polyclonal anticholine acetyltransferase, Chemicon (1:1,000); monoclonal antityrosine hydroxylase, Chemicon (1:1,000). Biotinylated secondary antibodies and peroxidase ABC kit were obtained from Vector. CoCl2-enhanced diaminobenzidine (DAB) was used as the chromagen. For fluorescent microscopy, secondary antibodies labeled with Alexa fluor dyes were employed.


MSCs Express Diverse Genes

To begin characterizing the undifferentiated MSCs more fully, we examined expression of a representative spectrum of genes that included those specific for diverse lineages. Our initial microarray screen of cultured dissociates revealed that the MSCs expressed many genes associated with differentiated cells. The expressed genes were not limited to mesodermal lineages but included genes representative of all germ layers. Listed in Table I are subsets of genes expressed by undifferentiated MSCs, adult tissue where they are highly expressed, and originating germ layer.

Table I. MSCs Express Diverse Genes Representative of All Germ Layers*
Gene (accession number)SymbolLocationLayer
  • *

    Representative sampling of genes expressed by undifferentiated MSCs are accompanied by Genbank accession numbers, principal adult tissue where each gene is expressed, and originating germ layer. Expression of gene targets in the bottom seven rows was confirmed by RT-PCR amplification. The symbol used to identify targeted genes in the figures is given.

Lanosterol-14a-demethylase (U17697) LiverEndoderm
IPP isomerase (AF003835) LiverEndoderm
Leptin (D49653) AdiposeMesoderm
Myosin heavy chain (L13606) MuscleMesoderm
NMDA receptor (S39221) BrainEctoderm
Ceruloplasmin (L33869)CERULung, LiverEndoderm
SM22α (L41154)SM22Smooth muscleMesoderm
Protamine 2 (X14674)PROTSpermGerminal
Aldolase C (M63656)ALDOBrainEctoderm
Amyloid precursor protein (X07648)APPBrainEctoderm
NMDA glutamate binding subunit (S61973)GLUTBrainEctoderm
Syntaxin (AF044581)SYNBrainEctoderm

To confirm this unexpected observation, we assayed transcript levels of some of these genes (bottom seven rows in Table I) by RT-PCR, focusing on prototypical genes specific for different germ layers. In addition to expression of expected mesodermal messages, such as SM22α, RT-PCR revealed mRNA for endodermal ceruloplasmin; ectodermal syntaxin 13, aldolase C, NMDA glutamate binding subunit, and amyloid precursor protein (APP); and germline protamine 2 (Fig. 1, Table 1). Omission of RTase in controls eliminated PCR products, confirming that the signals observed in the experimental groups were derived from RNA transcripts and not contaminating genomic DNA (Fig. 1).

Figure 1.

Undifferentiated MSCs express genes representative of all germ layers. Population: RNA was harvested from primary cultures of undifferentiated MSCs and converted to cDNA, and target genes were amplified using specific primer pairs (see Table I for symbols). Amplification resulted in a single band in each reaction in which cDNA target was provided (RT +). Control lanes, in which no RTase was present during the cDNA synthesis reaction, did not yield product. Kilobase ladder (Kb) and λ/Hind III DNA (M) were employed as markers. Clonal: RNA was harvested from a clonally derived MSC population and RT-PCR performed as described above. A specific band was amplified in each reaction set.

Gene Expression in MSC Clonal Lines

Our observations indicate that primary MSCs cultured directly from the marrow express genes representative of all three germ layers and germinal tissue. It is not clear, however, whether this unanticipated result reflects multidifferentiation of individual cells or whether primary MSC cultures contain a mixed population of cells that express suites of genes associated with a single germ layer. To address this issue, we assessed expression of these representative genes in a clonal MSC line derived from a single cell (Woodbury et al., 2000). Clonally derived MSCs recapitulate the expression pattern demonstrated by primary MSC cultures, expressing genes representative of all germ layers, suggesting that the progeny of individual MSCs represents a multidifferentiated population.

Gene Expression During Neuronal Differentiation

Although the expression of diverse gene products characteristic of distinct lineages by “undifferentiated” MSCs was unexpected, it may reflect extensive plasticity intrinsic to this stem cell population. Differentiation to specialized cell types, in turn, might be expected to alter differentially transcription of subsets of these messages. To examine this contention, we subjected MSCs to the neuronal differentiation protocol (Woodbury et al., 2000) and harvested RNA after 48 hr. Exposure to NIM caused a dramatic change in morphology of treated MSCs (Fig. 2). Before treatment, MSCs displayed a flat, fibroblastic morphology, with little evidence of refractility. At 48 hr posttreatment, MSC-derived neurons were rounded, exhibited highly refractile cell bodies, and displayed prominent process-like extensions. Neuronal differentiation significantly altered the pattern of gene expression of the foregoing prototypical target genes (Fig. 3). The transcription of germline protamine 2 was reduced to undetectable levels. Similarly, the expression of endodermal ceruloplasmin dramatically decreased. Perhaps unexpectedly, NMDA glutamate binding subunit mRNA levels also decreased. Expression of the neural genes aldolase C and syntaxin 13 exhibited modest decreases, as did muscle-specific SM22α. Among the genes surveyed, only APP expression remained unchanged as the MSCs assumed neuronal morphologic characteristics, allowing this signal to serve as an internal control, confirming that equal amounts of cDNA were present in each reaction.

Figure 2.

Morphological differentiation of MSCs. Images of MSCs were captured before treatment (0 hrs) and after 48 hr (48 hrs) of exposure to NIM. MSCs rapidly transform from fibroblastic to neuronal morphologies. Images are representative of typical morphologies before and after treatment but are not from the same field.

Figure 3.

Neuronal differentiation of MSCs alters the pattern of gene expression. RNA was harvested from undifferentiated MSCs (S) or from MSC-derived neurons (N) 48 hr after neuronal induction. RT-PCR was performed to amplify gene products representing specific germ layers. Neuronal induction decreased expression of ceruloplasmin, NMDA glutamate binding subunit, and protamine 2 significantly, whereas expression of aldolase C, SM22α, and syntaxin was less affected. The level of expression of APP was not changed by neuronal induction. λ/Hind III DNA was used as a molecular weight marker (M).

Expression of Neuroglial Genes

Having established the utility of RT-PCR in defining the unexpected expression of diverse differentiated genes in the “undifferentiated” MSCs and in neuronal differentiation, we examined transcriptional regulation of genes specific to the neuroglial lineage. APP served as a control, ensuring that equal amounts of target cDNA were used in each reaction (Fig. 4A). We examined glial fibrillary acidic protein (GFAP), the classical astrocyte marker (Eng et al., 1971), which is also expressed in neuroglial precursor cells (Laywell et al., 2000; Doetsch et al., 1999). Uncommitted MSCs expressed GFAP (Fig. 4A). With neuronal differentiation, the gene product was detectable at 24 hr, but at 48 hr postneuronal induction was no longer discernible, which is consistent with neuronal, but not glial, differentiation. Nevertheless, expression of GFAP by the MSCs is consistent with a growing body of evidence indicating that neural precursors in vivo express neuronal and glial markers and can differentiate to either lineage.

Figure 4.

Neuronal differentiation alters the expression of precursor and mature neuronal messages. A: RT-PCR analysis was utilized to assess expression of neuroglial genes during the initial 48 hr of neuronal differentiation. APP expression in undifferentiated MSCs (S) and MSC-derived neurons at 24 hr (N 24) and 48 hr (N 48) postinduction remained constant. GFAP and neuroD (NeD) mRNA levels were elevated in MSCs and decreased as neuronal differentiation proceeded. Message for the neuronal markers NF-M and tau were undetectable in MSCs but increased with ongoing neuronal differentiation. Tau message was present in multiple isoforms (bracketed bands). λ/Hind III DNA served as an MW marker (M). B: RNA was harvested from undifferentiated MSCs maintained in SFM and from MSC-derived neurons maintained in NIM for 24 hr. RT-PCR performed with primers specific for NF-M yielded a single band from MSC-derived neurons. A very faint band corresponding to the NF-M product was also detected in cells maintained in SFM. RT-PCR of RNA derived from the cerebellum (Cb) of adult rats yielded an NF-M product of the same size, demonstrating specificity of the reaction. Equal amplification of noggin (inset) indicates equivalence of cDNA target. λ/Hind III DNA was used as an MW marker (M). C: RT-PCR followed by high-resolution electrophoresis revealed three distinct bands corresponding to known splice variants of tau message that were present in MSC-derived neurons at 48 hr postinduction (arrowheads). No signal was discernible in undifferentiated MSCs. Sizes (bp) of low-MW fragments of kilobase ladder (Kb) are indicated.

To begin assessing mechanisms regulating neuronal differentiation, we examined neuroD, a transcription factor transiently expressed in neuronal precursor cells, known to regulate neuronal fate decisions (Lee, 1997; Morrow et al., 1999). Robust expression of neuroD was detected in the uncommitted MSCs, suggesting that these undifferentiated cells were already “primed” for neural differentiation (Fig. 4A). With differentiation, neuroD progressively decreased and was markedly diminished by 48 hr, which is consistent with transient expression of the transcription factor in differentiating neurons (Lee et al., 1995).

We used NF-M and tau as neuron-specific marker prototypes to examine changes in transcription associated with neuronal differentiation. We had previously found that the neuron-specific intermediate filament NF-M is expressed by MSC-derived neurons but not by undifferentiated MSCs (Woodbury et al., 2000). Here, we employed NF-M expression as a temporal, quantitative index of neuronal differentiation to help place the present observations in context. NF-M expression is associated with initiation of neuritogenesis, neural process outgrowth, and assumption of the characteristic mature neuronal morphology (Carden et al., 1987). Consistent with our previous findings, NF-M mRNA was undetectable in uncommitted MSCs but was present after 48 hr of neuronal induction. Similarly, tau message is undetectable in MSCs but is increasingly expressed with ongoing neuronal differentiation (Fig. 4A).

To examine further the changes in gene expression associated with neuronal differentiation, we investigated NF-M and tau expression in more detail. MSCs maintained in serum-free medium (SFM) for 24 hr exhibited extremely low, but detectable, levels of NF-M mRNA by RT-PCR, consistent with adoption of neuronal morphologies by a small number of cells under these conditions (Fig. 4B). However, incubation with the neural induction medium for 24 hr greatly enhanced NF-M message levels, consistent with differentiation of the vast majority of MSCs into neurons (Fig. 4B). As a positive control, we detected NF-M in the cerebellum of 32-day-old rats (Cb), generating a single band of the expected size. The encephalizing gene noggin (Smith and Harland, 1992) was unchanged 24 hr after neural differentiation, establishing the specificity of the increase in NF-M message.

Detailed analysis of the tau transcripts revealed that the known tau mRNA isoforms, generated by alternative splicing of exons 2 and 3 (Goedert et al., 1991), were present in the MSC-derived neurons but were not detectable in undifferentiated MSCs (Fig. 4C). Collectively, these data indicate that incubation in NIM decreases the expression of neural precursor messages (GFAP, neuroD) while simultaneously increasing the expression of specific neuronal markers (NF-M, tau), consistent with ongoing neuronal differentiation. In aggregate, our observations indicate that assumption of neuronal morphologies by MSCs is accompanied by a dramatic increase in the prototypical neuronal genes NF-M and tau, complex modulation of other neuronal genes, and decreased transcription of germline and mesodermal genes.

Tau was also detectable at the protein level by immunocytochemistry, confirming our PCR data. MSC-derived neurons, maintained in NIM for 10 days, were probed with antitau antibody. There was significant heterogeneity in the level of tau expression in the neurons even after 10 days, which often correlated with the degree of neuronal morphologic differentiation. For example, intensely tau-positive neurons (Fig. 5, arrow heads) with long varicose (Fig. 5, arrows) processes were evident, whereas neighboring cells exhibiting immature, transitional morphologic features displayed weak staining (Fig. 5A).

Figure 5.

MSC-derived neurons express neuronal protein markers 10 days after induction. MSC-derived neurons were fixed and processed immunocytochemically for the expression of neuronal markers. A: An intensely tau-positive neuron (arrowhead) displays contracted cell body and long processes studded with varicosity-like swellings (arrows); neighboring cells lack both neuronal morphologies and strong tau staining. B: All MSC-derived neurons stained for β-III-tubulin expression. C: Synaptophysin is detected throughout the cell body, with intense staining in the varicosity-like swellings (arrows). inset: RT-PCR for synaptophysin was performed on RNA harvested from undifferentiated MSCs (S) and MSC-derived neurons 24 (N24) and 48 (N48) hr postinduction. A signal indicating the presence of synaptophysin message is not detected in MSCs but becomes increasingly evident with progressing neuronal differentiation. D: No staining is seen when the primary antibody is omitted from the reaction. E: Cells exhibiting neuronal morphologies show staining for ChAT (arrow heads), whereas flat stromal-like cells (arrow) do not. A perinuclear staining pattern is evident. F: A subpopulation of cells expresses TH (arrowheads), whereas most MSC-derived neurons (arrows) do not express this protein. Magnification ×200. Images in B–F were acquired through a blue filter to enhance contrast.

We examined additional neuronal gene products that are expressed in more mature neurons or that are associated with functional neuronal communication. At 10 days, β-III tubulin, an intermediate filament characteristic of mature neurons (Menezes and Luskin, 1994), was present in virtually all cells (Fig. 5B). In contrast to these neuronal markers, MBP, a marker for mature oligodendrocytes, was not detectable in MSC-derived neurons (data not shown).

Genes Associated With Neurotransmission

To begin assessing the developing ability for functional communication, we initially examined synaptophysin, which is associated with synaptic vesicles and transmission. Analysis by RT-PCR indicated that synaptophysin mRNA was not present in undifferentiated MSCs but was detectable after 24 hr of neuronal differentiation and continued to increase thereafter (Fig. 5C, inset). The protein was detected in cell bodies as well as varicose, putative transmitter release sites along processes, reflecting an immature pattern of distribution (Fig. 5C). Control reactions in which the primary antibody was omitted were devoid of staining (Fig. 5D).

To assess further the developing capability for communicative function, we examined the expression of neurotransmitter enzymes. At 10 days, a large population of the neurons expressed choline acetyltransferase (ChAT), which catalyzes the synthesis of the excitatory transmitter acetylcholine (Fig. 5E). A similar percentage of ChAT-positive cells was observed when a monoclonal ChAT antibody from a different commercial source was used for staining, validating this staining pattern (data not shown). Interestingly, the majority of neurons (>85%) derived from the multipotent P19 embryonal carcinoma cell line are also cholinergic (Parnas and Linial, 1995). A smaller subpopulation of MSC-derived neurons expressed tyrosine hydroxylase, the rate-limiting enzyme in the synthesis of catecholamines, dopamine, norepinephrine, and epinephrine (Fig. 5F).

Clonal Analysis of Derived Neurons

To assess further the differentiation potential of MSCs, we investigated the expression of neuronal markers in a clonal population derived from a single MSC. We had reported previously that clonal MSCs subjected to the differentiation protocol display neuronal morphologies and express high levels of NSE (Woodbury et al., 2000). We demonstrate here that clonal MSC-derived neurons expressed the neuronal markers tau (Fig. 6A), TOAD-64 (Fig. 6B), β-III tubulin (Fig. 6C), and ChAT (Fig. 6D). The arrowhead in Figure 6C indicates a β-III-tubulin+ MSC-derived neuron, and the arrow identifies the nucleus of an unstained MSC. Similarly, a ChAT+ MSC-derived neuron and an unstained nonneuronal cell are identified by an arrowhead and arrow, respectively in Figure 6D. In aggregate, these observations indicate that the MSC-derived neurons were developing the structural apparatus for synaptic communication and the machinery for transmitter signal biosynthesis. The expression of enzymes for different transmitters suggests that the MSC-derived neurons are capable of differentiating into multiple subtypes.

Figure 6.

Clonally derived MSCs exhibit neuronal or osteoblastic phenotypes after differentiation. Neuronal differentiation of clonally derived MSCs express tau (A) and TOAD-64. (B) C: A β-III tubulin-positive MSC-derived neurons is indicated by the arrowhead. Arrow identifies the nucleus of an unstained cell. D: Most clonal MSC-derived neurons express ChAT (arrowhead). Arrow indicates an unstained cell not displaying neuronal morphologies. Magnification ×250. E,F: von Kossa staining of clonally derived MSCs maintained in osteogenic media for 13 days. Dark staining of the nodules demonstrates deposition of mineralized matrix characteristic of bone formation.

To assess the ability of this clonal line to differentiate into mesenchymal derivatives, we grew these MSCs in osteogenic media for 13 days and probed for deposition of mineralized matrix. Under these conditions MSCs displayed cuboidal morphologies consistent with osteoblastic differentiation. Nodules of differentiated cells formed, producing mineralized matrix as demonstrated by von Kossa staining (Fig. 6E,F).

Multipotentiality of MSC Clones

Clonal lines retained the ability to differentiate into osteoblasts, exhibiting a cuboidal morphology and elaborating a mineralized matrix in osteogenic media (Fig. 6). Thus, clonally derived MSCs exhibit multipotency, expressing ectodermal neuronal or mesenchymal osteoblastic traits.

Morphologic Reversion and Gene Expression

Neuronal differentiation of MSCs exhibited notable plasticity at its earliest stages, when it was partially reversible. Cells exposed to NIM assumed characteristic neuronal morphologies, displaying refractile cell bodies and long processes by 24 hr postinduction (Fig. 7A). Withdrawal of NIM from MSC-derived neurons elicited process retraction and reversion of morphology within 24 hr to flat cells, which shared characteristics with uncommitted MSCs but generally displayed a more stellate morphology (Fig. 7B). To assess changes in gene expression associated with reversion, MSCs were differentiated to neurons by exposure to NIM for 24 hr. At this time, NIM was removed from half of the cells and replaced with SFM while the remaining cells were maintained in NIM. Twenty-four hours later (48 hr in total), we harvested RNA from neuronal and reverted cells and assessed changes in gene expression by RT-PCR. We again assayed expression of the archetypal targets from various germ layers expressed at high levels in uncommitted MSCs. Reversion from the neuronal to the MSC phenotype was associated with striking changes in expression of a subset of the target genes. Germline protamine 2 mRNA, which decreased dramatically with neuronal conversion, was highly expressed in the reverted MSCs. Similarly, endodermal ceruloplasmin message was detectable in MSCs, whereas depressed levels existed in the neurons. Strikingly, neuroD mRNA was highly expressed in the reverted MSCs, presumably indicating the potential for redifferentiation into neurons. GFAP message was also expressed, consistent with maintenance of a primitive neuroglial precursor state. In sum, the reverted MSCs appear to express multipotentiality, consistent with the plasticity displayed at the time of differentiation. In contrast, expression of several genes was unchanged by reversion to MSCs; message levels for APP, muscle-specific SM22α (Fig. 7C), syntaxin, and aldolase C (data not shown) remained the same.

Figure 7.

Neuronal differentiation is reversible morphologically and transcriptionally. A: MSC-derived neurons at 24 hr postinduction display neuronal morphologies with highly refractile cell bodies and long process-like extensions. B: Same cells as in A 24 hr after NIM withdrawl. Cell bodies have flattened, and processes have been withdrawn, yielding cells with stromal morphologies. Magnification ×200. C: Morphologic reversion is accompanied by changes in gene expression, identified by RT-PCR. Ceruloplasmin, protamine 2, GFAP, and neuroD are expressed at higher levels in reverted cells (R 48) than in MSC-derived neurons (N 48) at 48 hr postinduction. The level of expression of APP and SM22α is unaltered by the reversion process.


The recent demonstration that adult rat and human MSCs, classical mesodermal cells, can differentiate into neurons within 5 hr in culture has raised fundamental questions regarding commitment, lineage restriction, and differentiation itself (Woodbury et al., 2000). To define the process of stem cell differentiation and begin approaching underlying mechanisms, we have more extensively characterized MSCs and developing neurons at the population and clonal levels, defining expression patterns for representative genes of different lineages and correlating expression with morphologic maturation. Our observations indicate that the “undifferentiated” MSCs express germline, endodermal, and ectodermal genes as well as the expected mesodermal genes. Neuronal differentiation of the MSCs involves complex modulation of these different gene sets rather than simple on–off switching of neural and nonneural genes.

Multidifferentiation of MSCs

Far from being undifferentiated, blank slates, the MSCs actively transcribe genes specific for all the classical embryonic germ layers. As expected, the stromal cells express prototypical mesodermal genes, including SM22α, myosin, and leptin. In addition, however, MSCs express protamine 2, which is germline-specific (Domenjoud et al., 1991), indicative of an early, uncommitted state. Simultaneously, endodermal ceruloplasmin, expressed at high levels in liver and lung (Fleming and Gitlin, 1990), is transcribed in uncommitted MSCs. Likewise, ectodermal syntaxin 13, highly enriched in the brain (Advanti et al., 1998), and brain-specific aldolase C (Mukai et al., 1991) are expressed by MSCs. NMDA glutamate binding protein (Kumar et al., 1991) and APP (Shivers et al., 1988) represent additional neural genes transcribed by MSCs. These observations are consistent with our previous finding that the MSCs express low levels of NSE (Woodbury et al., 2000).

It may be concluded that MSCs are not “undifferentiated” but rather “multidifferentiated.” Recently, Labat and coworkers (2000) have proposed the existence of a monocytoid ectomesenchymal stem/progenitor cell that expresses both neural and mesenchymal gene products. Previous work by Enver and colleagues has demonstrated that lymphohematopoietic MSCs express genes characteristic of multiple hematopoietic lineages prior to unilineage commitment (Hu et al., 1997; Cross and Enver, 1997; Enver and Greaves, 1998). Moreover, MSCs coexpress genes specific for a number of mesenchymal lineages, including adipocytes, osteoblasts, fibroblasts, and muscle (Seshi et al., 2000). The present work extends these observations, indicating that populations and clonal lines of MSCs transcribe germline, endodermal, and ectodermal genes in addition to mesodermal genes.

Our observations imply, consequently, that genes specific for multiple lineages are accessible for transcription in the MSCs, allowing for diverse differentiative fates. Indeed, bone marrow cells have already been shown to give rise to skeletal muscle, hepatocytes, glia, and neurons in addition to the aforementioned mesenchymal derivatives (see Morrison, 2001, for review, and references therein). One might anticipate that MSCs are capable of generating a far larger spectrum of cell types. In the case of neuronal differentiation, prior multidifferentiation may help to elucidate aspects of the process and begin to approach underlying mechanisms. One striking feature of MSC neuronal differentiation is rapidity: Within 5 hr of exposure to the induction medium, the cells assume typical neuronal morphological features and express a variety of neuron-specific genes (Woodbury et al., 2000; present study). The prior expression of neuronal genes by the MSCs may explain this rapid response. Presumably, quantitative alteration in genes already being transcribed overcomes the need for elaboration of new transcription factors or histone acetylation, for example. Although multidifferentiaion is not evidence of multipotentiality in any cell type, it may explain the rapid response to differentiative signals displayed by MSCs.

Expression of NeuroD by MSCs

Similarly, the expression of neuroD by the MSCs may also account for the speed of differentiation and may provide insight into regulatory mechanisms. NeuroD family members, bHLH transcription factors, are transiently expressed in neuronal precursors and initiate neuronal differentiation (Lee, 1997). These factors appear to function as master regulators of mammalian neurogenesis, in that transfection of murine embryonic carcinoma cells with neuroD2 transcripts initiates neural differentiation in nonneural cells (Farah et al., 2000). In the neural retina, neuroD plays a role in multiple developmental functions, including retinal cell fate determination, differentiation, and neuron survival (Morrow et al., 1999). In this model system, neuroD induces withdrawal from the cell cycle, regulates neuronal vs. glial cell fate decisions, and favors amacrine over bipolar differentiation.

The expression of neuroD by MSCs and its decrease with neuronal differentiation is consistent with a role in stromal conversion to neurons, a contention that we are presently examining experimentally. Moreover, the appearance of neuroD in neurons that have reverted to the MSC phenotype is particularly provocative, suggesting that the neuronal potential is an intrinsic property even of stromal stem cells “dedifferentiated” from neurons.

MSCs as Neuroglial Precursors

Unexpectedly, the undifferentiated MSCs expressed glial as well as neuronal genes. GFAP, the traditional astrocytic marker, is expressed in the MSCs but decreased with neuronal differentiation. The gene product was detectable 24 hr after neural induction, but by 48 hr was undetectable, consistent with neuronal but not glial differentiation. Consequently, neuronal differentiation of MSCs exhibits commonalities with neuronal differentiation in the normal adult rodent brain in which neurons derive from neuroglial precursors that express GFAP as well as neuronal characters (Doetsch et al., 1999; Laywell et al., 2000). Although the MSCs expressed the astrocyte marker GFAP, myelin basic protein (MBP), a marker specific for mature oligodendrocytes, was not detected. These observations complement our previous work, demonstrating that the primitive intermediate filament nestin, characteristic of neuroepithelial precursors, is expressed in MSC-derived neurons at 5 hr but decreases progressively and is undetectable 6 days after neuronal differentiation, mimicking normal neuronal differentiation in vivo (Woodbury et al., 2000). We tentatively conclude that neuronal differentiation from MSCs exhibits many sequential features of normal neuronal differentiation in vivo.

Plasticity of MSCs and Neurons

Plasticity is apparently maintained for a period of time after MSCs differentiate into neurons. Removal of the neural inducing medium, after initial conversion, resulted in the striking reversion of neurons to the MSC phenotype within 24 hr. Neuritic processes rapidly contracted and disappeared, cell bodies lost refractility and flattened, and the cells assumed a typical MSC morphology. With reversion, germline protamine 2 was reexpressed after becoming undetectable in the neurons. Endodermal ceruloplasmin increased significantly in the reverted MSCs. Most dramatically, in the present context, neuroD mRNA was apparent in the reverted MSCs but had disappeared 48 hr after neuronal differentiation, suggesting that the revertants retained the potential for redifferentiation. The reverted MSCs also expressed GFAP mRNA, which disappeared in the neurons, suggesting that the MSCs have, indeed, regressed to a primitive neuroglial precursor state. Our observations raise the possibility, more generally, that fate determination and differentiation are not necessarily irrevocable and unidirectional but, rather, may be multidirectional under appropriate circumstances.


The authors thank Drs. S. Campos and A. Wood for supplying microarray data, Drs. E.J. Schwarz and D.J. Prockop for providing the rMSCs used for some of these studies, and G. Munoz-Elias for many helpful discussions. The authors thank Mr. A. Zofchak for his generous support.