The molecular signature of therapeutic mesenchymal stem cells exposes the architecture of the hematopoietic stem cell niche synapse

Mesenchymal stem cells (MSCs) were first isolated from bone marrow stroma and described initially as precursors for fibroblasts [1]. More recently, their multi-lineage differentiation capacity has been extensively characterized [2]. It has been previously shown that MSCs are able to differentiate into mature cells of multiple mesenchymal tissues including chondrocytes, myocytes, adipocytes and osteocytes in a process largely regulated by members of the Wnt family of proteins [3]. MSCs are also an essential component of the stromal scaffold of the bone marrow that provides physical support during hematopoiesis. Based on these findings, MSCs have been utilized to provide help for the engraftment of hematopoietic stem cells (HSCs) in vivo [4]. Stromal cells from the bone marrow, together with osteoblasts and endothelial cells, have been involved in the formation of the so called HSC niche [5]. This can be defined as a specialized microenvironment that precisely maintains a long-term storage of quiescent, slowly dividing HSCs by preventing their proliferation, differentiation or apoptosis [6, 7].

Moreover, it has been recently shown that MSCs exert a profound modulatory effect on immune cells by means of both soluble and cell contact-dependent mechanisms [8–10]. In particular, MSCs inhibit T and B cell proliferation by halting cell cycle progression [11–13] and, under some circumstances, also rescue T cells from apoptosis (Benvenuto, F. et al., unpublished observation). In a recent experiment that highlights their therapeutic potential, in vivo injection of MSCs ameliorated the course of chronic progressive experimental autoimmune encephalomyelitis (EAE), the animal model of multiple sclerosis [13]. Although MSCs are able to affect a wide array of immunological functions and, as a consequence, to control the course of EAE, the underlying mechanism by which this occurs remains largely unknown. However, it is likely that both cell to cell contact and soluble factors are involved supporting T cell inhibition in a non cognate fashion. Among the latter, transforming growth factor-β1 (TGF-β1) and hepatocyte growth factor (HGF) [8], indoleamine 2,3-dioxygenase (IDO) [14], and prostaglandin E2 (PGE₂) [15] have been proposed as consistent candidates but other studies have failed to reproduce these results [10, 16]. In order to identify a new set of genes suggestive of their biological role, we used microarray technology to examine the global transcriptional profile of MSCs extracted from the bone marrow of C57BL/6J mice. We applied bioinformatics tools to analyze our results in the context of more than 400 publicly available expression datasets and identified unique features of MSCs that are likely responsible for their observed phenotype and potential therapeutic properties. Based on our experimental results we discuss the possibility that the immunoregulatory properties of MSCs may be orchestrated by the same mechanisms that are involved in the maintenance of the HSC niche.

In this study we evaluated the global transcriptional profile of immunomodulatory MSCs and conducted a bioinformatics-based analysis to identify the possible mechanisms for their unique properties. Since the process of obtaining these cells involves a laborious, multi-step cell culturing effort, prior to the transcriptional analysis we tested our MSC enrichment protocol with functional assays that tested their immunoregulatory and therapeutic functions.

MSCs exhibit a distinctive capacity of controlling T cell proliferation and survival

We harvested MSCs from the bone marrow of C57BL/6J mice and expanded them in culture until they reached a stable CD45-CD44+Sca1+ phenotype by FACS-analysis (data not shown). In order to test their capacity for modulating T cell functions, we added MSCs to mononuclear cells (MNCs) that had been previously activated by incubation with anti-CD3 antibodies. From as early as passage 7 in culture, MSCs inhibited T cell proliferation by inducing cell cycle arrest. This was demonstrated by the observed inhibition of CFSE incorporation (Fig. 1a) and the accumulation of T cells in the G0 phase of the cycle (Fig. 1b). When T cells are exposed to IL-2 prior to TCR engagement with an anti-CD3 mAb, they undergo apoptosis, a phenomenon known as activation induced cell death (AICD). As shown in Figure 1c, MSCs were able to rescue T cells from apoptosis as they reduced the percentage of apoptotic T cells following AICD. Finally, we examined the in vivo immunosuppressive potential of MSCs by injecting 1 × 10⁶ cells into mice with MOG_35–55-induced experimental autoimmune encephalomyelitis (EAE). The i.v. administration of 1 × 10⁶ MSCs following disease onset resulted in a statistically significant amelioration of EAE clinical symptoms compared to PBS-injected control mice (Fig. 1d). These results strongly suggest that MSCs have a profound therapeutic effect on T cells not only by preventing their proliferation but also by supporting their survival. These findings are consistent with the dual function observed for bone marrow stromal cells, known to inhibit HSC proliferation and, at the same time, support their survival within the "niche" [4].

Therapeutic MSCs show a unique gene expression signature

We extracted total RNA from 7 cultured MSC samples deriving from different mice and passages (see Methods) with proven immunomodulatory ability, and analyzed their gene expression profiles by hybridization onto Affymetrix Mouse Genome 430 2.0 arrays. We then compared these results to those reported in 486 publicly available platform-compatible microarrays exploring other 6 murine tissues (brain, heart, skeletal muscle, liver, kidney, lung) and 6 cell types (dendritic cells, DCs, embryonic stem cells, ESCs, embryonic fibroblasts, MEFs, neural stem cells, NSCs, HSCs, T cells). We first used principal component analysis (PCA), a dimensionality reduction technique, with the goal of testing whether the expression variance within tissues was smaller than that across tissues. In this exploratory analysis we were able to observe a recognizable segregation of samples according to the source of their RNA. The distribution of samples in PCA space was particularly pronounced for MSCs, which were clearly separated from most other tissues and cell types (Fig. 2a). Furthermore, the spread observed within principal components of MSCs was markedly lower than that observed for arrays derived from brain or ESCs. We hypothesize that although part of the observed pattern is due to sample heterogeneity (arrays from brain were obtained from several different laboratories and under many different conditions), a true biological effect is primarily responsible for the clear separation among groups of arrays. A two-way hierarchical clustering analysis confirmed the PCA results and allowed us to identify groups of genes characteristic of each tissue. In this analysis, MSC signature was readily observed when we compared the expression profile of ontogenically related cell types (Fig. 2b). Since MSCs are direct precursors of fibroblasts, it would be expected that they exhibit some degree of similarity. Interestingly, both analyses located samples from these two cell types next to each other, suggesting an overall similarity in their expression patterns. Remarkably, the unsupervised hierarchical clustering of samples resulted in a nearly perfect sorting of samples by germinal layer of origin (Fig. 2b). In addition to this, the striking similarity between ESCs and NSCs at the transcriptional level is consistent with previous reports [17]. These findings provide further evidence that the genes selected for validation are discriminatory, and that the in-silico analysis is sensitive enough to detect the characteristic signature of each cell type. Furthermore, the discriminatory power of these profiles seems to rely on a relatively small subset of genes.

Transcripts differentially expressed by MSCs belong to the HSC niche

In order to identify the characteristic signature of MSCs we performed two types of analyses. First, we defined differentially expressed transcripts as those expressed in all tissues but showing statistically significant differences in pair-wise comparison against MSCs. Our second analysis focused on cell "specific" genes, that is those that were detected in one but not in any other cell type. We found 403 differentially expressed genes between MSCs and any given tissue or cell type in our panel [See Additional file 1]. Among these, galectin-1 (Lgals1) was the highest expressed transcript in MSCs, a finding that correlates well with previous reports [18, 19]. We next sought to verify whether high expression of Lgals1 was also a characteristic of MSCs at the protein level. Indeed, the intracellular presence of Lgals1 colocalizing with MSC cytoskeleton was detected by immunostaining and visualized by confocal microscopy (Fig. 3). We then confirmed that Lgals1 is also secreted by MSCs as demonstrated by its presence in MSC supernatants measured by ELISA (Fig. 3). These results provide further support for a biological role of the previously observed transcriptional signature of MSCs. Among the transcripts with the highest signal intensities, we were able to detect several hematopoiesis related molecules. These transcripts code mostly for secreted proteins that play a role within the HSC niche such as fibronectin-1 (Fn1), osteopontin (Spp1), chemokine C-X-C motif ligand-12 (Cxcl12), thrombospondin-1 (Thbs1), thrombospondin-2 (Thbs2), transforming growth factor-β2 (Tgfb2), angiopoietin-1 (Angpt1), insulin-like growth factor binding protein-4 (Igfbp4), fibroblast growth factor-7 (Fgf7), secreted frizzled-related protein-1 (Sfrp1), secreted frizzled-related protein-2 (Sfrp2), dickkopf-3 (Dkk3), vascular cell adhesion molecule-1 (Vcam1), and bone morphogenetic protein receptor type 1a (Bmpr1a). The expression of 4 representative genes from this list was validated by quantitative RT-PCR using commercially obtained RNA representing most tissues previously analyzed by microarray (Fig. 4). In general, there was good agreement between the expression of the tested genes in MSCs and other tissues and cell types (R² = 0.7977), thus confirming our observations. Unexpectedly, the expression of these genes was shared not only by MSCs and MEFs, but also by DCs.

Gene Ontology (GO) statistical analysis of genes differentially and uniquely expressed by MSCs

In order to integrate expression profiles into a functional framework we performed a GO-based statistical analysis. When we analyzed the 403 differentially expressed genes, we found that extracellular matrix and extracellular were among the first three statistically significant GO Cell Component categories, while cell adhesion was among the first five most enriched Biological Process classes (Fig. 5). This suggests that the most abundantly expressed transcripts were mainly secreted proteins, a significant part of which are involved in cell-cell contact. Conversely, when we analyzed the list of "specific" genes, they appeared to be primarily located in the nucleoplasm and mainly involved in development and transcription processes (Fig. 5). This strongly suggests that the list of "specific" genes is composed mainly by low level-expressed transcription factors and molecules involved in development (i.e. Wnt pathway).

Overall, this report provides a transcriptional map of immunomodulatory MSCs through global RNA profiling and bioinformatics-based comparative analysis with several other mouse tissues and cell types. By analyzing arrays from the same platform (Affymetrix 430) we were able to perform statistical analyses combining results across different laboratories. Thus, we overcame the limitation of comparisons across different platforms, generally regarded as problematic from a biological perspective because the measurements may represent different physical quantities [21]. Conversely, the approach used in this study offers a more precise assessment of the true degree of differential expression and helps refine the list of candidate genes warranting biological validation.

By applying principal component and hierarchical clustering analyses, we identified a distinctive gene expression profile for MSCs. MSC gene expression profile appeared in fact to be characteristic and substantially homogeneous. One of the most significant findings of this analysis was the nearly perfect sorting of tissues by germinal layer of origin utilizing an unsupervised classification. As expected, being MEFs derived from MSCs, their expression profile was highly similar. Furthermore, previous studies have shown that cultured MEFs maintain some of the multipotency of MSCs [22]. Surprisingly, MSCs and DCs displayed a similar profile as well, a finding further confirmed by qRT-PCR analysis. We speculate this could be due to the fact that both MSCs and DCs are involved in the determination of bystander cell fate, respectively HSCs and T cells, by cell-cell contact-dependent mechanisms.

When focusing on the list of differentially expressed genes we found that many of them had been reported in previous transcriptional studies of MSCs, thus validating the sensitivity and reproducibility of our approach [18, 23]. Among MSC specific genes, several proteins belonging to the Wnt signaling pathway or Wnt-regulated genes (Sfrp1, Dvl2, Axin1, Tcf3, Rarg, Cspg2, Efnb2) were detected. Wnt proteins have been previously reported to promote myogenesis [24], inhibit chondrogenesis [25] and adipogenesis [26] and to have a dual effect on osteogenesis [27]. Some of the most abundantly expressed genes are also inhibitors (Sfrp1, Sfrp2, Dkk3) or act downstream (Thbs1, Thbs2, Cyr61, Sema3a, Sfrp2) of the Wnt signaling pathway [20]. Among the differentially expressed genes we were also able to find most recently described molecules involved in stem cell renewal such as Dkk3 [28] and Serpinf1 (PEDF) [29]. Altogether, these results are consistent with the involvement of Wnt signaling in mesenchymal lineage specification.

We also used GO-based functional classification to characterize both differentially expressed and MSC specific genes. This analysis confirmed our primary observation that differentially expressed transcripts were significantly enriched in molecules present in the extracellular space and involved in cell adhesion process. On the other hand, MSC specific genes were mainly nucleoplasm-located transcription factors involved in development. Importantly, the observation that these transcription factors are expressed at low levels is consistent with most recent findings about the biology of ESCs. These cells have been recently shown to express low levels of transcription factors by means of the polycomb group proteins and by specific chromatin modification patterns thus keeping them poised for activation during development [30]. It can be speculated that a similar mechanism may be working in further committed stem cells such as MSCs.

The combined transcriptional and comparative analyses presented here allowed us to identify several secreted molecules that may be critical in determining cell functions. Several of these molecules have been previously reported to play a pivotal role in the constitution of the HSC niche synapse [5] or as being expressed by hematopoiesis-supportive stromal cells [31] [See additional file 4]. For example, angiopoietin-1 (Angpt1) is considered one of the major regulators of blood vessel stability [32]and, in contrast with angiopoietin-2 (Angpt2), it seems to exert a potent anti-inflammatory effect [33]. Moreover, Angpt1 and its receptor endothelial-specific receptor tyrosine kinase(Tek) are known key components of the HSC niche, namely of the cell-cell signaling between osteoblasts and HSCs. Furthermore, Angpt1 is likely to be one of the most important molecules in the induction of the "quiescent" state of the HSCs, by enhancing their tight adhesion to osteoblasts, inhibiting cell division, arresting cells in the G0 phase of the cell cycle and promoting cell survival under myelosuppressive stress [34].

Osteopontin (Spp1) is a highly acidic phosphoprotein with pleiotropic effects, including regulation of inflammation, cell adhesion and angiogenesis [35]. In addition, it has recently been demonstrated to be a pivotal molecule in limiting the size of the HSC pool [36]. Similarly, it has also been shown that the thrombin-cleaved form of Spp1 promotes the quiescence of HSCs by exerting a profound suppression of proliferation of HSCs without inducing apoptosis [37]. Interestingly, Spp1-deficient mice developed a milder form of EAE, and activated T cells from these animals produced less IFNγ and more IL-10 than their wild type counterparts consistently with the role of a pro-inflammatory molecule [38]. Although an elevated Spp1 expression in therapeutic MSCs is in apparent contradiction with its previously reported proinflammatory role, its pleiotropism and ubiquitousness suggest that it may interact with different receptors at different times and through different regulatory networks, thus preventing a straightforward characterization of its biological functions.

Thrombospondin (Thbs)-1 and -2 are extracellular matrix proteins that share several effects on cell growth, adhesion, survival and differentiation. In particular, Thbs1, secreted by DCs, has been demonstrated to inhibit both TCR signaling and T cell proliferation, mainly acting through the CD47 receptor [39]. In addition, Thbs1 has also been described as part of the HSC niche [40]. Both Thbs1 and, to a lesser extent, Thbs2 have been studied as inhibitors of endothelial growth [41]. On the other hand, Thbs2 has been described as an autocrine inhibitor of proliferation secreted by MSCs which acts through cell-cycle inhibition without the induction of apoptosis [42].

Fibronectin-1 (Fn1) is another major component of the bone marrow stroma. The interaction of Fn1 with β1-integrins, VLA-5 in particular, has been shown to be a negative regulator of HSC proliferation by preventing them from entering the cell cycle [43].

Galectin-1 (Lgals1) is an endogenous lectin characterized by its affinity for beta-galactosides which has been demonstrated to exert an immunosuppressive effect on T cells through both apoptotic and non-apoptotic mechanisms [44]. This molecule has been shown to be expressed by hematopoietic-related tissues [23] and to inhibit HSC growth particularly at high concentrations [45]. Most recently, Lgals1 has been shown to play a role within the NSC niche [46]. We were able to confirm MSC expression of this molecule both by ELISA and immunocytochemistry assays.

Finally, semaphorins are molecules that were originally identified as mediators of axon guidance but have also been described as part of the immunological synapse [47]. Semaphorin 3A (Sema3a), in particular, has also been recently demonstrated to mediate T cell proliferation inhibition by arresting T cells in the G0/G1 phase of the cell cycle [48].

Here we have shown that MSCs arrest T cells in the G0 phase of the cell cycle. These results are in line with previous reports showing that MSCs induce T cell [12], B cell [11] and dendritic cell division arrest anergy [49]. Such effect not only does not depend on the induction of apoptosis on immune cells [13] but is accompanied by an anti-apoptotic effect on target cells, when they are overstimulated. Although this dual in-vitro effect of MSCs on T cells appears counterintuitive as the mechanism responsible for ameliorating EAE, we speculate that MSCs may exert their immunoregulatory action by virtue of their anti-proliferative effect, while their anti-apoptotic action is mostly beneficial for the survival and protection of neural cells. Recent data seem to support this hypothesis [50].

Our data clearly shows that immunomodulatory MSCs abundantly express several molecules also secreted by osteoblasts in the context of the HSC niche synapse where they may keep HSCs quiescent while supporting their survival. In addition, some of these molecules belong to the protein class of matricellular proteins (Spp1, Thbs1, Thbs2, Sparc), namely extracellular matrix proteins that, rather than serving a primary structural role, mediate both tissue morphogenesis and homeostasis modulating cell-matrix and cell-cell interactions and other important processes such as cell growth, angiogenesis and inflammation. This observation, together with the similarities we found between MSC and DC transcriptomes, highlights the relevance of macromolecular assemblies at the interface between MSCs and surrounding cells. We hypothesize that, the interactions between MSCs and HSCs resemble those described for the immunological synapse thus leading to quiescence both of proliferating HSCs and immune cells. Therefore, given the analogies between the effect exerted by osteoblasts on HSCs and by MSCs on immune cells, we propose that the same bystander mechanisms may play a role in both types of cell-cell interaction (Fig. 6). At the same time, interaction between MSCs, their extracellular matrix components and differentiating somatic cells is an established mechanism within stem cell niches throughout the body [51]. Based on our experimental results, we also hypothesize that MSCs represent fundamental niche cells within the bone marrow and potentially other stem cell niches as well. In particular, when administered in vivo, they could participate in the formation and modulation of the recently named "atypical ectopic perivascular niches" in the context of inflammatory disorders of the CNS [52]. Overall, our data shed further light into the biology of MSCs by proposing several candidate molecules as potential mediators both of the HSC niche-supportive and immunomodulatory activities of MSCs.