Direct reprogramming of sarcomas

As our primary objective was to elucidate whether or not it is possible to reprogram cancer cells into a state from which terminal differentiation is achievable, and not knowing a priori: (1) whether this is feasible; and (2) which transcription factors would be necessary (and at which levels) and/or sufficient, we sought to introduce all six previously identified reprogramming transcription factors into five sarcoma cell lines. Pooled supernatant derived from lentivectors expressing cDNAs of human Oct4, Nanog, Sox2, Lin28, Klf4 and c-Myc12, 13 were used to infect human osteosarcoma cells (SAOS2, HOS, MG63), human liposarcoma cells (SW872) and human sarcomas of unknown lineage (that is, Ewing’s sarcoma SKNEP). The frequency of reprogrammed sarcoma formation ranged from 1 to 5% and the time to initial formation ranged from 18 to 42 days (Supplementary Figure 1). Our reprogramming efficiency is slightly higher than those reported for somatic cell reprogramming and may be due to pre-existing tumor suppressor loss.7 In agreement with our previous gene expression data on these cell lines,14, 15 all five ‘parental’ sarcoma cell lines expressed Myc and Klf4 (Figure 1 and Supplementary Figure 2), whereas reprogrammed sarcomas silenced the transgenes while reactivating the corresponding endogenous genes after full reprogramming (Figure 1, Supplementary Figure 3B). All reprogrammed sarcomas also expressed standard human pluripotent markers (for example, Tra-1-81 and SSEA4 (Figures 1a–e), Rex (Supplementary Figure 3), and alkaline phosphatase (Supplementary Figure 4). Reprogrammed sarcomas were also able to undergo Activin A- (50 ng/ml) Noggin- (100 ng/ml) directed differentiation into endoderm and ectoderm, respectively (Figures 1a–e). In contrast, reprogrammed cells showed no evidence of ‘spontaneous’ differentiation (that is, when cultured in the absence of Activin A or Noggin) along the ectodermal or endodermal lineages (Figures 1(a–e)–v left panel). As controls, parental cells showed no spontaneous or Activin A- or Noggin-mediated differentiation (data not shown).

Figure 1 Characterization of reprogrammed sarcomas as compared with parental sarcomas: (a–e) (i) parental cell in culture (scale bar=50 μm), (ii-top) embryoid body formation (scale bar=500 μm); (ii-bottom) reprogrammed cells in culture (scale bar=50 um); (iii) expression via RT–PCR of reprogramming transcription factors in parental (PAR) and reprogrammed (REP) sarcomas; (iv) ESC markers via immunofluorescence in reprogrammed cells, and lineage-directed differentiation (via Activin A- (50 ng/ml) or Noggin- (100 ng/ml) directed differentiation into endoderm and ectoderm, respectively; (scale bar=500 μm); (v) morphologically and via expression of lineage-specific markers (for example, Sox 17 endoderm, Pax6 ectoderm) (scale bar=50 μm). All cells were grown on matrigel. (f) Quantification of Ki67 and vimentin immunohistochemistry of xenografts from parental and reprogrammed cells as seen in Supplementary Figure 5. Full size image

We next assessed the in vivo tumorigenicity of the reprogrammed sarcomas. One million parental sarcoma cells and one million of the reprogrammed sarcomas were injected into the left and right anterior tibialis muscle of NOD-SCID-gamma (NSG) mice, respectively, and observed. As per our previous reports14, 15 in which we described the properties of these cell lines in detail, parental sarcomas formed tumors at the 8–10-week point and all mice were killed when the largest of the two forming tumors (either parental or reprogrammed sarcomas) reached 1 cm in greatest dimension. In all cases (eight mice per sarcoma-type) both the parental and their reprogrammed counterparts formed tumors, but the parental sarcomas formed tumors at a faster rate than their reprogrammed counterparts. Xenografts that arose from reprogrammed sarcomas (a) were approximately half the size of the xenografts formed from parental cell lines; (b) were associated with necrosis, which was not observed in the parental cells; (c) were less densely packed with tumor cells as compared with parental controls; (d) stained less frequently for Ki67 than parental cell lines; and (e) expressed less vimentin (a definitive mesenchymal marker) than their parental counterparts (Figure 1f and Supplementary Figures 5 and 6). As we have previously observed15 most sarcoma cell lines (including the ones used here) form undifferentiated sarcomas when grown as xenografts regardless of initial histopathology. Although pathologically the reprogrammed sarcomas still best resemble undifferentiated sarcomas, it is important to note that the histopathological pattern shown by reprogrammed sarcomas (that is, less vimentin expression than their reprogrammed counterparts without morphological indications of differentiation or expression of other lineage markers—pan-cytokeratin/epithelial, S100/melanocyte and neurophysin/neuroendocrine were all negative in both parental and reprogrammed cells (data not shown)) is pathologically consistent with a shift towards a less-differentiated state. Despite all this evidence, we cannot absolutely rule out the possibility that a transition from a high-grade sarcoma to a malignant teratoma or embryonal carcinoma has occurred. However, given that no benign mature tissue elements were observed, we can conclude that a benign teratoma (as would have been expected for reprogrammed somatic cells) did not form. ‘Teratoma’ formation from solid human cancer cells has been alluded to previously,8 but reported only as xenograft formation with no histology or immunohistochemistry presented. Thus, it remains unclear whether or not human solid tumor cells can be reprogrammed into cells capable of dedifferentiating and forming benign teratomas. Regardless, our results clearly show that direct reprogramming decreases the aggressiveness of cancer (as measured by growth/size/cellularity) as compared with its parental counterpart and additionally induces a loss of mesenchymal lineage markers, suggesting a concurrent loss of differentiation.

Reprogrammed-sarcomas differentiate into multiple mature connective tissue phenotypes

We next assessed whether reprogrammed sarcomas can be terminally differentiated either into the lineage in which they were initially blocked (for example, osteosarcoma into osteocytes) or into an alternate lineage (for example, osteosarcomas into adipocytes, a lineage switch) by growing all reprogrammed sarcomas in either osteogenic or adipogenic differentiation medium. As terminal differentiation is defined as both expression of the mature phenotype and terminal cell division (that is, cessation of proliferation) we assessed both these properties in all five reprogrammed sarcomas and in two differentiation lineages. As seen in Figure 2, reprogrammed sarcomas (but not the parental cells) in the presence of either adipogenic or osteogenic differentiation medium express the terminal phenotype of that differentiation lineage (that is, either bone formation as measured by calcium mineralization via Alizarin Red S or fat formation as measured by lipid accumulation via Oil-Red-O, considered ‘gold standards’16, 17, 18 for terminal differentiation of embryonic and mesenchymal stem cells (MSCs) into bone and fat, respectively). In both cases the mature phenotype is accompanied by cessation of proliferation as measured by loss of Ki67 staining (Figure 2) and by standard proliferation assays (Supplementary Figure 7) in only the differentiating reprogrammed sarcomas. Interestingly, reprogrammed sarcomas do not regain Ki67 staining or show signs of further growth following 4 weeks in replacement of differentiation medium with maintenance medium (data not shown). In agreement with the latter, no tumors formed upon subcutaneous inoculation of the entire cell culture obtained from differentiated reprogrammed sarcomas into NSG mice and followed for up to 16 weeks (Supplementary Figure 8). Taken together, we conclude that terminal differentiation of reprogrammed sarcomas can abrogate the tumorigenicity of the parental sarcoma cells.

Figure 2 (a–e) Comparison of sarcomas and reprogrammed sarcomas in terms of achieving terminal differentiation as measured by acquisition of the terminal phenotype (both bone via Alizarin Red S staining for calcium deposition and fat via Oil-Red-O staining for lipid accumulation) and cessation of proliferation via loss of Ki67. (f) MSCs differentiated as controls into bone and fat. (Oil-Red-O images are magnified 10-fold to accentuate lipid formation associated with each cell). Scale bars=10 μm. Full size image

Reprogrammed sarcoma differentiate into hematopoietic cells

We further sought to examine whether the terminal differentiation potential of the reprogrammed sarcomas was restricted to connective tissue lineages or whether reprogrammed sarcomas had broader differentiation potential. Thus, we differentiated reprogrammed sarcomas along the erythroid lineage by employing an ESC to erythroid differentiation protocol.19, 20 In each case, the cells undergo a significant fold reduction in cell size: starting with parental sarcoma cells having fibroblastic appearance measuring approximately 25 μm in diameter (Figure 3a, first panel); to reprogrammed sarcoma cells having a more rounded appearance and measuring approximately 20 μm in diameter (Figure 3a, second panel); and to differentiated erythroid cells measuring approximately 8 μm in greatest dimension (Figure 3a, most right panel). The volume reduction from a diameter change of 25–8 μm, assuming a fully spherical shape (and therefore a function of the radius cubed), suggests a 30-fold reduction in volume. Volume reduction is accompanied by obvious morphological changes from fibroblastic to hematopoietic appearing cells accompanied with loss of cell adhesion (Figures 3a). Additionally, characteristic pro-erythroblastic cells were seen, many on the verge of enucleation (Figure 3b). Finally, we performed a hematopoietic cell surface antigen analysis as a function of time. Our data indicate that CD71 and CD36 both peak during the mid-phase of our differentiation protocol, a point of differentiation at which the majority of cells are CD235(GPA)-negative, consistent with that previously shown for normal ESC to erythroid differentiation.19, 20 This specific profile is most consistent with erythroid colony formation units (CFU-Es; CD36+/CD71+/CD235a(GPA)−).21 However, upon further differentiation both CD71 and CD36 decline and CD235a(GPA)-positive cells appear (Figure 3c). We further validated these results by performing immunofluorescence for both CD71 and CD235a both at day 14 and at day 28 of the erythroid differentiation time course. As seen in Figure 3d at day 14 most cells are CD71-positive, but lose CD71 expression by day 28. In contrast, only rare CD235a(GPA)-expressing cells can be observed at day 14; however, at day 28 three distinct populations of cells can be identified: CD235a+/DAPI+ (green circle); CD235-/DAPI+ (blue circle); and CD235a+/DAPI− (yellow circle). The latter indicates enucleated mature red blood cells. Taken together, our differentiation results using reprogrammed sarcomas and specifically the prevalence of CFU-Es with rare mature cells are identical to those for ESCs cultured along the erythrocytic lineage differentiation.19, 20, 22 Additionally, as above, no tumors formed upon intramuscular or subcutaneous inoculation of the entire cell culture obtained from erythroid differentiated reprogrammed sarcomas into NSG mice and followed for 16 weeks (Supplementary Figure 9).

Figure 3 (a) Morphological comparison of sarcomas, reprogrammed sarcomas and reprogrammed sarcomas at days 7, 14, 21 and 28 of an erythroid differentiation protocol. Scale bars=10 μM. (b) Individual cells on the verge of enucleation from day 28 erythroid differentiation cultures. Scale bars=5 μM. The original magnification for all panels in a were × 200; and × 1000 for b. (c) Percentage of cells from HOS, SW872, SKNEP, Rep-HOS, Rep-SW872 and Rep-SKNEP cultures undergoing erythroid differentiation expressing CD71, CD235a (GPA), CD36 and CD34. Error bars=SD. (d) CD71 and CD235(GPA) immunofluorescence in reprogrammed SW872 cells after 14 and 28 days of erythroid differentiation. Green circle=CD235a+/DAPI+; blue circle=CD235-/DAPI+; yellow circle=CD235a+/DAPI−. Scale bars=10 μm. Full size image

Epigenetic silencing of oncogenes

Our results suggest that direct reprogramming can supersede the specific genetic and/or epigenetic changes that resulted in the lineage-specific differentiation block and subsequent tumorigenesis. To explore the relationship between reprogramming and the epigenetics of oncogenes we focused on myc as: (a) it is expressed at baseline in all of our sarcoma cell lines (Figure 1); (b) it was further introduced as part of the reprogramming process and its expression did not change upon reprogramming (Figure 1); (c) it is an established oncogene prognosticating poor outcomes in sarcoma patients;23 and (d) myc levels were found to decline dramatically after reprogrammed sarcomas were differentiated, not merely reprogrammed (Figure 4a). We have confirmed these semi-quantitative RT–PCR results using quantitative PCR (Supplementary Figure 10). Thus, our data show that differentiation following direct reprogramming is accompanied by myc downregulation and suggested us that epigenetic changes at the myc promoter may change during the reprogramming process in order to become amenable to transcriptional downregulation during differentiation.

Figure 4 (a) Myc RT–PCR in parental and reprogrammed sarcoma cells following 3 weeks in either maintenance (control), adipogenic (fat) or osteogenic (bone) differentiation medium. ChIP of the myc promoter using anti-H3K4triMe (b) or anti-HEK27triMe (c) antibodies in either maintenance (control), adipogenic (fat) or osteogenic (bone) differentiation medium. (d) DNA promoter methylation analysis showing all statistically differentially methylated promoters (t-test P<0.05). (e) Percentage of tumor suppressor gene (TSGs) and oncogene promoters accounting for all differentially expressed promoters (t-test P<0.05) plotted as a function of fold change. (f) Unsupervised hierarchical clustering of sarcomas and their reprogrammed counterparts using all statistically differentially methylated promoters (t-test P<0.05). Full size image

We assessed the epigenetic status of the myc promoter in parental sarcomas and reprogrammed sarcomas using chromatin immunoprecipitation (ChIP) to assess for active, H3K4triMe and repressive, H3K27triMe marks.24, 25 We observed that in parental sarcoma cell lines the myc promoter is H3K4triMe-modified consistent with an active state, whereas in reprogrammed sarcomas the myc promoter is H3K4triMe- and H3K27triMe-modified (Figure 4b) consistent with a bivalent state,24, 25 characteristic of many stem cell genes that exist in a ‘poised’ state25, 26, 27 that is amenable to being rapidly activated or repressed by removal of the appropriate modification. Furthermore, the H3K4triMe active mark is indeed lost as reprogrammed sarcoma cells undergo adipogenic and osteogenic differentiation (Figure 4b). Thus, our data show that reprogramming sarcomas results in an epigenetic change in the myc promoter in which it becomes permissive to being silenced, but only becomes silenced with subsequent differentiation. It is noteworthy that the reprogramming epigenetic patterns of the myc promoter as observed in reprogrammed sarcoma cells before and after differentiation are identical to those observed in normal MSCs before induction of and during osteogenic and adipocytic differentiation.24

As for histone modifications, changes in DNA methylation are characteristic of late nuclear reprogramming28, 29 and tumorigenesis. Cancer has been described as globally DNA hypomethylated, oncogene promoter DNA hypomethylated and tumor suppressor promoter DNA hypermethylated.30 Interestingly, nuclear reprogramming of somatic cells is associated with similar patterns;31 however, DNA methylation patterns in reprogrammed cancer cells have not been examined. We examined global DNA methylation changes in sarcomas and their reprogrammed counterparts using Infinium HumanMethylation27 arrays (Supplementary Table 2). Cancer reprogramming was accompanied by global DNA demethylation (Figure 4d). In fact, of the 216 CpG islands (corresponding to 205 gene promoters) demonstrating a 15% change in DNA promoter CpG methylation (P<0.05, paired t-test; Supplementary Table 3) only one promoter (cg17133183; 1.1968464 up) corresponding to CRABP1 (cellular retinoic acid binding protein 1) showed an increase in methylation. Interestingly, of these 205 gene promoters showing significant changes in CpG methylation only 29 unique genes (37 Affymetrix IDs, Santa Clara, CA, USA; Supplementary Table 4) showed significant changes (P<0.05, paired t-test) at the gene expression level (please see next section). In all, 25 of the 37 (67%) Affymetrix IDs showed an increase in gene expression levels (Supplementary Table 4 and Supplementary Figure 11), suggesting that promoter hypomethylation accompanying reprogramming results in an increase in overall gene expression of those genes. We further confirmed the gene expression changes obtained via microarray expression data by performing quantitative RT–PCR for 26 of the 29 genes. With only occasional exceptions in some cell lines (that is, MPZL1, IMP4 and FBLIM1) 23 of the 26 genes showed an increase in gene expression in reprogrammed cells as compared with their parental counterparts (Supplementary Figure 12), including three tumor suppressor genes (TSGs) (DNTM3a, HOOK3, and JAK2; and one oncogene TOML1; please see next section), thus re-affirming that a decrease in promoter DNA methylation correlates to an increase in gene expression (when gene expression changes are observed). However, given that 205 promoters were identified as hypomethylated and only 23 genes showed a definitive increase in gene expression, our data suggest that (as demonstrated for the myc promoter above) changes in DNA methylation are not sufficient to result in changes in gene expression for the majority of genes.

As reprogramming of sarcomas resulted in global hypomethylation and restoration of differentiation potential, we were particularly curious as to how reprogramming affects promoter DNA methylation of oncogenes and TSGs. We obtained large gene lists of both oncogenes and TSGs (downloaded from http://cbio.mskcc.org/CancerGenes;32 Supplementary Tables 5 and 6) and overlapped these lists with all 1241 promoters whose degree of methylation was significantly changed by direct reprogramming (P<0.05, paired t-test; no restriction on fold change, Supplementary Table 7 as opposed to the analysis above which was restricted to a 15% fold change shown as Supplementary Table 3). Using this approach, 82 TSGs and 32 oncogenes were identified (Supplementary Tables 8 and 9) accounting for 7.1% and 3.2%, respectively, of the genes affected by DNA methylation. As expected because the vast majority of genes were hypomethylated, all oncogenes and TSGs affected by DNA methylation were also hypomethylated as a result of reprogramming. The observation that oncogenes were hypomethylated is not surprising as during somatic reprogramming and tumorigenesis, oncogenes are known to be hypomethylated. However, in contrast to somatic reprogramming and tumorigenesis, which would have predicted that TSGs should have been silenced (via DNA hypermethylation), our observations that TSGs are also hypomethylated suggests that there may be significant differences in epigenetic control between reprogrammed cancer and somatic cells. Furthermore, as many of our cell lines are deleted for specific TSGs (Supplementary Table 1), we feel that this data may underrepresent the actual difference in epigenetic control during reprogramming between somatic and cancer cells at tumor suppressor loci. To pursue the association between reprogramming and TSG and oncogene DNA hypomethylation, we examined the relationship between the frequency of oncogenes and TSGs as a function of DNA methylation fold change. As seen in Figure 4e, the percentage of TSGs and oncogenes accounting for all genes that undergo a given fold change (P<0.05 paired t-test), increase as the fold change threshold increases. Note that for both the oncogene and TSG curves, R2 (the coefficient of determination; that is, the proportion of variability in a data set that is accounted for by the statistical model and capable of predicting future outcomes) approaches the ideal of 1. Or in other words, the subset of genes with the greatest change in gene promoter DNA methylation between the sarcomas and their reprogrammed counterparts actually occurs at oncogenes and tumor suppressors. This is the first report of a global DNA promoter methylation analysis comparing human cancers to their reprogrammed counterparts.

Lastly, we performed unsupervised hierarchical clustering (a bio-informatic approach that allows self-organization based on similarities in gene expression using predefined genes) using the 216 CpG island set demonstrating a 15% change in DNA promoter CpG methylation (P<0.05, paired t-test; Supplementary Table 3) on the five sarcomas, their reprogrammed counterparts and human MSCs (hMSCs). As seen in Figure 4f: (1) all reprogrammed sarcomas (with the exception of Rep-HOS) cluster together; (2) all parental sarcomas cluster together; (3) parental HOS and reprogrammed HOS cluster together as well (most likely as many of the genes hypomethylated during reprogramming appear already hypomethylated in HOS); (4) methylation is higher in each parental sarcomas (red on the heat map) as compared with its reprogrammed counterpart (blue on the heat map) in agreement with the data presented as averages in Figure 4d; and (5) potentially most interesting is that reprogrammed sarcomas exhibit a methylation state that is much closer to hMSCs than to sarcomas (note that many of the genes hypomethylated during reprogramming are hypomethylated in hMSCs, blue on the heat map). The latter suggests that reprogramming of sarcomas results in an epigenetic state (at least partially) similar to hMSCs.

Degree of differentiation reversion following direct reprogramming

One of the many unsettled issues in the field of reprogramming is the degree of differentiation reversion. How far back does a somatic cell need to be de-differentiated before it is possible to re-differentiate it along an alternate lineage? What about a cancer cell? The presumption is that reprogramming recapitulates the ESC state as shown by acquisition of functional pluripotency and ESC markers. To assess whether this was similarly true for reprogrammed cancer cells, we performed gene expression analysis of our sarcoma and reprogrammed sarcomas and compared them with known gene expression patterns for ESCs (GSE6561), MSCs (GSE6460), fibroblasts (GSE9865), reprogrammed fibroblasts (GSE9709), partially reprogrammed fibroblasts (GSE9865) and reprogrammed sarcomas. We limited our analysis to a discriminatory 50 gene set,33 previously used to differentiate between mouse ESCs, mouse embryonic fibroblasts (MEFs) and reprogrammed fibroblasts (Supplementary Table 10). Performing unsupervised hierarchical clustering using the above gene set reveals that reprogrammed sarcomas associates most closely with sarcomas and partially reprogrammed fibroblasts (Figure 5a) and not with the fully reprogrammed state (as defined via the clustering group of ESCs and reprogrammed fibroblasts).

Figure 5 (a) Unsupervised hierarchical clustering using a gene set representing ANOVA between ESCs, reprogrammed fibroblasts, partially reprogrammed fibroblasts, reprogrammed sarcomas, sarcomas, fibroblasts and MSCs. (b) Scatter plot analysis showing differences in gene expression between indicated cell types. red box=comparison between sarcomas and rep-sarcomas, blue box=comparison between ESCs and sarcomas; green box=comparison between ESCs and rep-fibroblasts. (c) Unsupervised hierarchical clustering of sarcomas and their reprogrammed counterparts using all statistically differentially expressed genes between sarcomas and their reprogrammed counterparts (paired t-test P 0.05; 1.5-fold change). (d) Schematic representation of a primary component analysis (accounting for 100% of all genes representative of an ESC-MSC-fibroblast analysis; please see text for details) showing relative relations of indicated samples as a ‘differentiation’ axis. (e) Model of degree of cancer reprogramming and subsequent differentiation (please see Discussion for details). Full size image

Having observed that reprogrammed sarcomas associates most closely with sarcomas and partially reprogrammed fibroblasts, we further enquired as to which genes (out of the entire genome) best discriminate between these states. To simultaneously analyze gene expression of multiple groups we used analysis of variance (ANOVA) to identify a subset of genes that was able to discriminate among ESCs, MSCs, fibroblasts, reprogrammed fibroblasts, partially reprogrammed fibroblasts and reprogrammed sarcomas (Supplementary Table 11), and examined the differences in gene expression data sets between each two of the five groups. As seen in Figure 5b (red outlined), sarcomas and reprogrammed sarcomas share the most homology (that is, expressing the least divergence from the diagonal), whereas all comparisons with the ESC (for example, ESCs compared with sarcomas, blue outline) and the four other subtypes shows large deviations from the diagonal. It is worth pointing out that ESCs and reprogrammed fibroblasts (green outline) are fairly convergent (that is, showing a lesser degree of divergence from the diagonal than any other comparison with the ESC state). In fact, a paired t-test (P 0.05; 1.5-fold change) comparing sarcomas to their reprogrammed counterparts identified only 125 differentially expressed unique genes (169 Affymetrix IDs; Supplementary Table 12). However, unlike the unsupervised hierarchical clustering based on DNA methylation (Figure 4f), which demonstrated global hypomethylation in the reprogrammed state, unsupervised hierarchical clustering based on this gene set did not result in an obvious one sided shift (that is, towards hypomethylation and thus a predicted increase in global gene expression). In fact, this analysis identified two distinct gene subgroups (Figure 5c) in which approximately half of the 125 genes increase their expression (red-colored) and half decrease their expression (blue-colored) as a function of reprogramming. This is the first report of a global gene expression analysis comparing human cancers to their counterparts reprogrammed via defined factors. Although several cancer cell lines5, 6, 7, 8 have been previously reprogrammed by different means, the degree of differentiation reversion following reprogramming has never been directly assessed. Our data here (from Figure 5a) suggests that cancer reprogramming does not result in the pluripotent state via gene expression; albeit still allowing for attainment of terminal differentiation into multiple previously blocked lineages.

Principal component analysis (PCA) identifies genes that are uniformly changing or uniquely different in various known subgroups, thus allowing those subgroups of genes to uniquely identify certain states. Extrapolation of gene expression change between states and the ordering of states to reflect normal biology allows the creation of a roadmap on to which other states may be mapped. We have previously used PCA to evaluate degrees of cancer differentiation in comparison with normal differentiating cells or cells representative of various differentiation stages,24 analogous to that recently described for ESCs and induced pluripotent stem cells (iPSCs).34 The latter analysis hypothesizes that a highly representative component of the PCA would align the samples along a differentiation axis. To first establish such a differentiation axis we first performed ANOVA using gene expression profiles from ESCs, MSCs and fibroblasts (from above) to identify differentially expressed genes between the three states (Supplementary Table 13), which we hypothesized, represent three stages along a mesodermal differentiation spectrum. We then performed PCA using gene expression profiles for ESCs, rep-fibroblasts, rep-sarcomas, MSCs, sarcomas and fibroblasts. To be further unbiased we allowed for only one component to the PCA to be identified, thus ensuring that all differentially identified genes between the three states would be used to assign the above groups to the differentiation axis. In Figure 5d, we observe that(1) ESCs and fibroblasts were at opposite ends of a putative differentiation spectrum; (2) reprogrammed fibroblasts were close to ESCs; (3) sarcomas fall in between MSCs and fibroblasts (as per our previous data in which we assigned differentiation stages to sarcomas using gene expression profiling);24, 35 and (4) reprogrammed sarcomas fall on the ESC side of MSCs in between MSCs and partially reprogrammed fibroblasts (Figure 5d). Taken together, this data suggest that reprogramming sarcomas does indeed dedifferentiate them to a pre-MSC state (closer to ESCs; red arrow Figure 5d). However, in direct comparison with the differentiation change undergone by reprogrammed fibroblasts, which are reprogrammed to the ESC state (blue arrow, Figure 5d), the differentiation change undergone by sarcomas only reverts them to a pre-MSC state. Taken together, our data would suggest that sarcomas are reprogrammed to a differentiation state that best resembles partially reprogrammed fibroblasts. Yet, it is important to note that this partial reprogramming appears to be enough to re-establish the ability to achieve terminal differentiation in multiple lineages. This observation leads us to suggest that small, albeit specific changes in epigenetics and genetics might be sufficient in order to change non-differentiable tumor into one that is amenable to terminal differentiation (Figure 5e and see discussion).