Molecular Characterization of Spontaneous Mesenchymal Stem Cell Transformation

Background We previously reported the in vitro spontaneous transformation of human mesenchymal stem cells (MSC) generating a population with tumorigenic potential, that we termed transformed mesenchymal cells (TMC). Methodology/Principal Findings Here we have characterized the molecular changes associated with TMC generation. Using microarrays techniques we identified a set of altered pathways and a greater number of downregulated than upregulated genes during MSC transformation, in part due to the expression of many untranslated RNAs in MSC. Microarray results were validated by qRT-PCR and protein detection. Conclusions/Significance In our model, the transformation process takes place through two sequential steps; first MSC bypass senescence by upregulating c-myc and repressing p16 levels. The cells then bypass cell crisis with acquisition of telomerase activity, Ink4a/Arf locus deletion and Rb hyperphosphorylation. Other transformation-associated changes include modulation of mitochondrial metabolism, DNA damage-repair proteins and cell cycle regulators. In this work we have characterized the molecular mechanisms implicated in TMC generation and we propose a two-stage model by which a human MSC becomes a tumor cell.


INTRODUCTION
The development of a solid tumor is considered a multi-step process in which several molecular checkpoints must be altered to generate a tumor from a normal cell [1]. The acquired capabilities of tumor cells include their ability to proliferate continuously ignoring apoptosis or growth-inhibitory signals, generating their own mitogenic signals. In advanced phases of tumor development, a neoangiogenesis process takes place and finally tumor cells acquire the capacity of tissue invasion and metastasize to other organs. Generally, it is admitted that most tumors acquire these characteristics through genome instability, telomere stabilization and disruption of regulatory circuits [2].
A recent theory suggests the existence of cancer stem cells (CSC), a subpopulation of cells with tumorigenic potential that is lacked in the rest of the cells within this tumor. CSC were reported for some tumor types including breast and lung cancer, leukemia and glioblastoma [3,4]. However, there is a great ignorance about how the ''acquired capabilities'' of tumor cells would take place; directly on adult stem cells, or on differentiated cells that suffer a dedifferentiation process. In this regard, CSC share several features with adult stem cells such as self-renewal ability, asymmetric division, and differentiation potential [5].
Adult human mesenchymal stem cell spontaneous immortalization and transformation were recently reported by our group [6], supporting the hypothesis of the stem cell origin of CSC. Independent laboratories have confirmed these data, reporting similar results using MSC derived from human or murine bone marrow [7][8][9][10][11]. In this regard, we have previously characterized the cellular sequence of steps necessary to transform a human MSC into a tumorigenic cell [6]. Following approximately 20 population doublings in vitro, mesenchymal stem cell cultures enter a senescence phase, but are able to bypass it at a high frequency. These cells then continue to divide until they reach a crisis phase. Only some samples are able to escape from this crisis phase spontaneously, but those that do have undergone tumorigenic transformation generating TMC.
However, until now genetic alterations implied in spontaneous MSC transformation are little known. Some groups have studied molecular pathways involved in the artificial transformation of MSC transduced with oncogenes [12,13]. In this study, we have characterized the molecular mechanisms implicated in TMC generation and we propose a two-stage model by which a human MSC becomes in vitro a tumor cell.

Comparative gene expression analysis of MSC transformation by microarray analysis
To analyze molecular differences associated with TMC generation, we performed microarray studies using mRNA from pre-and post-senescence MSC, and from TMC. From a general perspective, data analysis showed that the greatest changes were associated with TMC generation, as TMC functions were more different than postsenescence MSC, compared to pre-senescence MSC (Table 1). Although in a minor intensity, post-senescence MSC have the same altered functions that TMC. In both cases the principal category affected is ''cancer'' (Table 1). However, the main pathways deregulated in both, post-senescence MSC and TMC, are related to stress, toxic events and mitochondrial metabolism ( Table 2). On the other hand, there was more down-than upregulated RNA transcripts associated with TMC generation ( Figure S1). The main differences in mRNA expression profiles between pre-and postsenescence MSC are shown in Table 3. Table 4 shows differences between pre-senescence MSC and TMC.

Molecular differences in cell cycle-related proteins
We had reported differences in cell cycle progression in pre-and post-senescence MSC and TMC, including more rapid cell division in TMC [6]. Here we used microarray techniques to explore the cell cycle-related molecular differences in postsenescence MSC and TMC compared to pre-senescence MSC. Compared to pre-senescence MSC, post-senescence MSC showed few differences in cell cycle-related proteins. In contrast, TMC samples showed significant mRNA modulations; Cdk1 and Cdk4 as well as cyclins B1 and D2 were upregulated, whereas cyclin D1 was downregulated ( Figure 1A).
We evaluated microarray experiments analyzed by qRT-PCR the expression of some of these genes. No expression difference was found between pre-and post-senescent MSC, while TMC overexpressed Cdk2 and Cdk6 ( Figure 1B). In qRT-PCR experiments we also studied met-TMC, a cell line derived of lung metastases generated after s.c. inoculation of TMC in immunodeficient mice (Rubio et al, unpublished results).
To determine whether the differences in mRNA levels gave rise to altered protein expression, we compared these samples by western blot. In TMC, cyclin B1, Cdk2 and Cdk6 were upregulated, whereas Cdk1 and Cdk4 remained constant. Cyclin D1 was downregulated from pre-senescence MSC to TMC ( Figure 1C).

Upregulation of DNA repair pathways in transformed MSC
As DNA repair mechanisms are responsible for the bypass of senescence and crisis, as well as for tumor maintenance and progression [14], we analyzed the major proteins linked to these processes. We tested proteins involved in pathways including nonhomologous end joining (NHEJ), base excision repair (BER), nucleotide excision repair (NER), mismatch repair (MMR), and homologous recombination (HR).
Microarray analysis showed no mRNA significant differences between pre-and post-senescence MSC (Figure 2A). However protein analysis showed that DNA-PKcs was repressed in postsenescence MSC compared to pre-senescence MSC and a downregulation of Rad51 after senescence bypass, although expression was restored and upregulated after crisis bypass in TMC ( Figure 2C).
In contrast, mRNA levels were modulated differentially between TMC and pre-senescence MSC for several DNA repair-associated proteins. Specifically, DNA-PKcs and PCNA were upregulated in TMC, and XPA was downregulated compared to pre-senescence MSC. Although DNA polymerase m and ERCC3 upregulation were statistically significant, differences in mRNA levels appeared to be minor, based on x-fold change and Z-score values (Figure 2A).
We performed qRT-PCR experiments to validate microarray results. No modulation of expression of DNA repair-related genes was detected between pre-and post-senescence MSC. In contrast, several of these genes were overexpressed in TMC, such as DNA-PKcs, DNA polymerase m, RAD51 or ERCC4 ( Figure 2B).  We evaluated the correlation between RNA and protein levels by western blot analysis. Proteins were elevated in nearly all DNA repair pathways in TMC compared to pre-senescence MSC. These effects were protein-rather than pathway-specific. MMR was the only pathway that showed no differences between MSC and TMC, although we analyzed only PCNA in this pathway ( Figure 2C).

Changes in tumor suppressors and oncogenes expression related with MSC transformation
To explore the role of tumor suppressor gene inactivation in MSC senescence and crisis bypass, we analyzed p16, p21 and p53. In a microarray screening assay, we found no changes in p21 or p53 levels, whereas MSC expressed less p16 mRNA after senescence bypass ( Figure 3A), concurring with reports of a role for p16 in senescence induction [15]. p16 downregulation was more marked when we compared TMC and pre-senescence MSC mRNA levels, and we found slight but significant p21 downregulation. We detected no differences in p53 mRNA ( Figure 3A).
To validate the microarray results, we used qRT-PCR to compare mRNA levels among the three cell types. p16 mRNA levels were downregulated in post-senescence MSC, to approximately 60% of the pre-senescence level. We observed no p16 mRNA amplification in TMC samples ( Figure 3B). p16 protein is downregulated in post-senescence MSC, and absent in TMC [6], concurring with these results. We analyzed whether complete p16 repression in TMC was caused by promoter methylation, as reported for this tumor suppressor [16,17] studying the methyl-ation status. Methylation analysis of the Ink4a/Arf promoter showed no methylation (not shown). We used specific DNA-PCR to detect homozygous deletion of p14, p15 and p16, and found no amplification ( Figure 3C-E), indicating that the Ink4A/Arf locus is deleted specifically in TMC.
Microarray analysis suggested slight p21 downregulation in TMC compared to pre-senescence MSC, with no difference between pre-and post-senescence MSC. We found no changes in p21 levels in western blot analysis of these populations ( Figure 3F). Although p53 pathway defects have been associated with the generation of many tumor types [18], this pathway appeared to be functional in our stem cell transformation model, as p53 was upregulated after UV irradiation and was phosphorylated in all samples tested ( Figure 3G). Basal p53 protein levels were also higher in TMC than in pre-or post-senescence MSC.
Finally, we assayed Rb levels and phosphorylation in pre-and post-senescent MSC, two TMC samples and a met-TMC sample. Rb protein levels were upregulated in TMC and met-TMC compared to pre-and post-senescence MSC. Rb phosphorylation increased progressively through the sequence of steps in TMC generation, and was slightly downregulated in met-TMC compared to TMC ( Figure 3H).
We compared mRNA regulation differences in the oncogenes cmyc and telomerase in post-senescence MSC and TMC with presenescence MSC. We found no differences in oncogene mRNA transcript levels in these populations ( Figure 3A). Nonetheless, in a previous publication we have shown that c-myc protein overexpression is linked to senescence bypass and is maintained in TMC;

Genbank Acc No
Gene name x-fold change p-value z-score    moreover, TMC also have telomerase activity which is not found in pre-or post-senescence MSC [6].

DISCUSSION
Mesenchymal stem cells can be easily isolated and expanded in culture to generate large numbers of cells for cellular therapies. Human MSC in early passage are safe although stressful conditions (as they are cultured for a long time) can turn them in immortal and occasionally they became tumorogenic [6]. Further research is necessary to understand this process in order to develop better protocols for culture adult stem cells, as it has been demanded recently [19]. Here, we describe several molecular alterations in our spontaneous human MSC transformation model that affect cell cycle regulation, oncogene expression, mitochondrial metabolism, DNA repair mechanisms and inactivation of tumor suppressor genes.
TMC versus pre-senescence MSC array analysis showed that functions with a higher significance are related, as expected, with transformation, genetic disorders and cell death (Table 1). We described previously that post-senescence MSC are non-tumorigenic and their cellular behaviour in culture was very similar to pre-senescence MSC [6]. Interestingly, post-senescence MSC versus pre-senescence MSC array analysis also showed the same functions altered than TMC, although in smaller grade (Table 1), suggesting a pre-tumoral state of pre-senescence MSC.
We observed expression of many untranslated RNAs in MSC concurring with reports which show a large and ''silent'' mRNA pool in stem cells [20], this could be the reason why, following MSC transformation, we identified more downregulated than upregulated genes in arrays experiments ( Figure S1). Comparison of mRNA and protein expression in pre-and post-senescence MSC and in TMC showed variation in RNA and protein regulation. Cyclin D2, Cdk1 and PCNA mRNA were upregulated in TMC compared to MSC, although their protein levels did not change; whereas c-Myc, Cdk2, Cdk6, DNA ligase IV and DNA polymerases mRNA levels remained stable but their protein levels were upregulated. Translational control could thus be important for adult stem cells, and retention of large numbers of silenced transcripts might allow rapid stem cell differentiation to other lineages in response to appropriate stimuli. These data also indicates the limitations of results based on RNA-exclusive analysis of MSC.
Telomerase activity has been found in almost all human tumors but not in adjacent normal cells [21] and maintenance of telomere stability is required for the long-term proliferation of tumor cells [22]. The escape from cellular senescence and thus becoming immortal by activating telomerase is required by most tumor cells for their ongoing proliferation [23]. In our model, during TMC generation these cells acquire a detectable telomerase activity [6]. Telomerase promotes MSC immortalization and, in conjunction with additional events, produces cell transformation [12,13,24]. These additional events usually implied an oncogene deregulation.
One of the most important oncogenes involved in MSC transformation is c-myc. In our spontaneous model, senescent and post-senescence MSC, as well as TMC, overexpress c-myc [6]. Consistent with our previous results, data from other groups have shown that c-myc seems to be essential to spontaneously transform MSC [7,9,10]  resulting in c-myc stabilization [25], suggesting that c-myc might be necessary to transform MSC.
We explored DNA repair mechanisms to elucidate their role in MSC transformation. Post-senescent MSC showed downregulation of DNA-PKcs, ERCC3 and Rad51 proteins, each of which is associated to a distinct DNA repair pathway. Extremely restricted clonal selection takes place during cell crisis, and only cells with functional DNA repair mechanisms would continue to grow. TMC have a higher metabolic rate and divide more rapidly than pre-or post-senescence MSC, with a consequent increase in DNA damage. Proteins that participate in DNA repair are upregulated in TMC compared to MSC; this, together with telomere length maintenance, could permit cell survival, despite oxidative damage to DNA and be responsible for TMC karyotype stabilization. Recently it has been published the dependency on oxidative phosphorylation during MSC transformation [11]. We have not detected statistically significant changes of these genes in our microarray experiments (Table S1), although potential pathways leading to changes in postsenescence MSC and TMC revealed change in stress, toxic events and mitochondrial metabolism pathways ( Table 2). The definitive role of mitochondrial respiration on spontaneous MSC transformation remains to be investigated.
A chromosome 5 alteration and a (3;11) translocation are recurrent, stable features of in vitro cultured TMC [6]. The telomerase gene map to human chromosome 5, suggests that it is activated by internal amplification of this chromosome in TMC. Chromosome 11 alterations are recurrent in tumors [26]. Although we did not detect a target gene in the 3;11 translocation in our model, genes involved in cell transformation are likely to be located in this region [27][28][29].
As tumor suppressor genes are major targets in neoplastic transformation, we analyzed their expression in these cells. The tumor suppressor Rb is implicated in several cancer types [18]. In our model of MSC transformation, Rb protein levels are upregulated progressively, and Rb is inactivated by a phosphorylation mechanism in TMC, as described [30]. In addition of Rb, loss of p53 function is common in many tumor types [18], but this pathway appeared to be functional in our model, as p53 was upregulated and phosphorylated in UV-irradiated cells. We observed higher basal p53 levels in TMC than in MSC, even when they had not been exposed to UV irradiation. In TMC, p16 mRNA and protein were entirely absent, and the Ink4a/Arf locus had been deleted. The increase in basal p53 may thus be due to stabilization by the ubiquitin protein ligase MDM2, due to the lack of p16 [31]. Identical results, p16 locus deletion and normal p53 activity, was detected in telomerase-immortalized human MSC [32]. The results suggest that p16 inhibition is essential for TMC generation, as is the case for human malignancies including glioblastoma, melanoma, pancreatic adenocarcinoma, non-small cell lung cancer, bladder carcinoma and oropharyngeal cancer, where this tumor suppressor is frequently lost [33]. Finally, we propose a two-stage model in which a mesenchymal stem cell becomes a tumor cell (Figure 4). The first step, the senescence bypass or M1 phase, is associated with c-myc overexpression and p16 repression; many DNA repair proteins are subsequently downregulated. Telomere shortening provokes the cell crisis phase, or M2, in which cells undergo stringent selection. TMC then upregulate many DNA repair proteins, which may be necessary for crisis bypass. Finally, escape from crisis is associated with telomere stabilization, Rb hyperphosphorylation and p16 deletion that seems to be essential to promote transformation [33,34]. TMC also upregulate many DNA repair proteins, which may be necessary for crisis bypass. These levels are maintained in TMC and could permit cell survival, despite oxidative damage to DNA.
The essential steps in TMC generation described here are basically in agreement with results of other authors working in MSC transformation [7][8][9][10][11]32] and these alterations are very similar to molecular changes associated with transformation of other cell types. In epithelial cells, spontaneous immortalization of human keratinocytes exhibited a small number of chromosomal aberrations, reduced p 16INK4a mRNA, elevated telomerase activity and functionally normal p53 [35]. Immortalization of human prostate cells by cmyc stabilizes telomere length through up-regulation of TERT expression and lack Rb/p16 INK4a checkpoint, being easily trans-formed [36]. In mesodermic cells, fibroblast cell lines immortalized either spontaneously or radio-chemically induced maintaining their telomerase activity, displayed loss of expression of p16 INK4a and hyperphosphorylation of Rb [37]. Telomerase-immortalized human fibroblast revealed overexpression of the c-myc and Bmi-1 oncogenes, as well as loss of p14 ARF expression [38,39], while overexpression of cmyc immortalizes freshly isolated human foreskin fibroblasts displayed a marked decrease in expression of p14 ARF [40].
In sum, all these evidences strongly suggest that cells with a mesodermal origin could require a common sequence of oncogenic events to become a tumor cells. How these processes are coordinated or associated with the critical cell evolution/ selection revealed in the culture [6] remains to be studied in deep. In addition, the cause/consequence relationship of this molecular signature with the recently characterized mesenchymal to epithelial transition (Rubio, D. et al, in press) or other potentially involved mechanisms remains also to be determined.

MATERIALS AND METHODS
Isolation of MSC and cell culture MSC were isolated as described [6]. Briefly, adipose tissue from non-oncogenic patients were digested with collagenase P and  cultured (37uC, 5% CO 2 ) in MSC medium (DMEM plus 10% FCS, 2 mM glutamine, 50 mg/ml gentamycin) and passaged when they reached 85% confluence. TMC and met-TMC were cultured under the same conditions.

Microarray labeling
Total RNA was isolated from four biological replicates of pre-and post-senescence MSC and from TMC using TriReagent Solution (Sigma) following manufacturer's instructions. RNAs were purified with MegaClear (Ambion) and integrity confirmed using the Agilent 2100 Bioanalyzer (Agilent Technologies). Total RNA (1.5 mg each) was amplified using the Amino Allyl MessageAmp aRNA kit (Ambion); we obtained 15-60 mg of amino-allyl amplified RNA (aRNA). Mean aRNA size was 1500 nucleotides, as measured using the Agilent 2100 Bioanalyzer. For each sample, 2.5 mg of aRNA was labeled with one aliquot of Cy3 or Cy5 Mono NHS Ester (CyDye Post-labeling Reactive Dye, Amersham) and purified with the Amino Allyl MessageAmp aRNA kit. Cy3 and Cy5 incorporation were measured using 1 ml probe in a Nanodrop spectrophotometer (Nanodrop Technologies). For each hybridizations, 80-100 pmol of Cy3 and Cy5 probes were mixed, dried by speed-vacuum, and resuspended in 9 ml RNase-free water. Labeled aRNA was fragmented by adding 1 ml of 106 fragmentation buffer (Ambion) and incubating (70uC, 15 min). The reaction was terminated with 1 ml stop solution (Ambion).

Slide treatment and hybridization
Slides containing 22,102 annotated genes corresponding to the Human 70-mer oligo library (V2.2) (Qiagen-Operon) were obtained from the Genomics and Microarray Laboratory (Cincinnati University). Information on printing and the oligo set can be found at http://microarray.uc.edu. Slides were prehybridized (42uC, 45-60 min) in 66 SSC, 0.5% SDS and 1% BSA, then rinsed 10 times with distilled water. Fragmented Cy5 and Cy3 aRNA probes were mixed (80-100 pmol of each label) with 10 mg PolyA (Sigma) and 5 mg Human Cot-DNA (Invitrogen) and dried in a speed-vacuum. Each probe mix was resuspended in 60 ml of hybridization buffer (50% formamide, 66 SSC, 0.5% SDS, 56 Denhardt's solution). Probes were denatured (95uC, 5 min) and applied to the slide using a LifterSlip (Erie Scientific). Slides were incubated (48uC, 16 h) in hybridization chambers (Array-It; Telechem International) in a water bath. After incubation, slides were washed twice with 0.56 SSC, 0.1% SDS (5 min each), three times with 0.56SSC (5 min) and finally in 0.056SSC (5 min), then dried by centrifugation (563 g, 1 min). Images from Cy3 and Cy5 channels were equilibrated and captured with an Axon 4000B scanner and spots quantified using GenePix 5.1 software.
Four independent biological replicates were ''dye swapped'' and studied (8 hybridizations). Data were analyzed using Almazen software. Each replicate was lowess-normalized and the log ratios merged with the corresponding standard deviation and z-score. We obtained adjusted p-values using limma by FDR [40]. Differentially expressed genes were selected by filtering signal intensity (.64), z-score (.3.5 or ,23.5) and p-value (,0.01).

Pathways analysis
By using Ingenuity Pathways Analysis (IPA), potential pathways leading to changes in MSC-postsenescence and TMC were created. This web-delivered application reveals relevant networks by comparing gene expression data with known pathways and interactions between genes (http://www.ingenuity.com). The filtered expression data set for MSC-postsenescence and TMC regulated genes were uploaded as tab-delimited text into IPA for generating biological networks. Each gene identifier was mapped to its corresponding object in the Ingenuity Pathways Knowledge Base. This software assigned a score for all networks that were ranked on the probability that a collection of genes equal to or greater than the number in a network could be achieved by chance alone (a score of 2 represents a 99% confidence level, and 3 a 99.9%). Biological functions are then calculated an assigned to each network Quantitative real-time PCR (qRT-PCR) cDNA was generated from 100 ng of total RNA using the High Capacity cDNA Archive Kit (Applied Biosystems) in a 10 ml final reaction volume. Real-time PCR reactions were performed in triplicate using two dilutions (1/50, 1/500; 3 ml/well) of each cDNA, 16 TaqMan Assay-On-Demand (Hs00233365_m1, Cdkn2a; Hs00195591_m1, snail; Hs00161904_m1, slug) or primers described in Table S2, 16 SYBR Green PCR Master Mix or 16 TaqMan Universal PCR Master Mix (Applied Biosystems) in a volume of 8 ml in 384-well optical plates, or using Universal ProbeLibrary (Roche). PCR reactions were run on an ABI Prism 7900HT (Applied Biosystems) and SDS v2.2 software was used to analyze the results with the Comparative Ct Method (DDCt).

p53 activation assay
We induced p53 upregulation and activation in UV-irradiated pre-and post-senescence MSC, TMC, and met-TMC (15 JU/ m 2 ). Extracts were collected 18 h after irradiation and used in western blot with anti-p53 or -phospho Ser15-p53 antibodies. atubulin was used as control.
Analysis of p16ink4a, p15ink4b and p14ARF CpG island methylation status We determined DNA methylation patterns in the CpG islands of p16ink4a, p15ink4b and p14ARF tumor suppressor genes by chemical conversion of unmethylated, but not methylated, cytosine to uracil, followed by methyl-specific PCR (MSP) amplification using primers specific for methylated or modified unmethylated DNA [41,42]. Placental DNA treated in vitro with Sss I methyltransferase was used as positive control, and DNA from normal lymphocytes as negative control for methylated alleles. Each PCR sample (12 ml) was separated in non-denaturing 6% polyacrylamide gels, ethidium bromide-stained, and visualized with UV illumination. Promoter methylation status of these genes was verified by bisulfite genomic sequencing of CpG islands. Both strands were sequenced. Primers for bisulfite genomic sequencing and methylation-specific PCR were designed according to genomic sequences around presumed transcription start sites of the genes studied. Primer sequences and PCR conditions for methylation analysis are available on request.

Bisulfite treatment
Genomic DNA was EcoRI-digested to shear DNA and achieve complete chemical conversion after bisulfite treatment. Sodium bisulfite conversion of genomic DNA (1 mg) was performed as described [41,42], with modifications. Briefly, NaOH was added to denature DNA (0.3 M final concentration) and incubated (15 min, 37uC). Fresh bisulfite solution (2.5 M sodium metabisulfite and 125 mM hydroquinone, pH 5.0) was added to each sample, and incubation continued (16 h, 50uC, in the dark). Modified DNA was purified using Wizard DNA purification resin (Promega) and eluted in water at 60uC. After desulfonation with NaOH (0.3 M final concentration; 10 min, 37uC), isolation was continued with 0.3 volume of 10.5 M ammonium acetate, followed by incubation (5 min, RT). Modified DNA was precipitated using 2.5 volumes of 100% ethanol and glycogen (5 mg/ml) as a carrier. The pellet was washed with 70% ethanol, dried, and eluted in distilled water.

Homozygous deletion analysis
We analyzed fragments of the p16INK4a-E1a, E2, p14ARF-E1b, and p15INK4b genes as described [16] to detect homozygous deletion in TMC. Comparative multiplex PCR was performed as described [43] to analyze each gene locus, using the b-actin fragment as internal control. Normal lymphocytes (NL) were used as negative control of tumor suppressor gene methylation, HCT116 (colorectal cancer line) as positive control of Ink4a/Arf locus methylation and MDA-MB231 (mammary adenocarcinoma) as control of Ink4a/Arf locus deletion. Figure S1 Comparison of mRNA differences between pre-and post-senescence MSC and in TMC. Microarray analysis pattern of overall mRNA differences between pre-and post-senescence MSC (A), and pre-senescence MSC and TMC (B). MA plots are shown, being A: log-ratio of two expression intensities vs. M: the mean log-expression of the two. Found at: doi:10.1371/journal.pone.0001398.s001 (4.09 MB TIF)