MFH expression patterns correlate with those of hMSCs.
To explore the possibility that MFH derives from the transformation of hMSCs, RNA was isolated from proliferating hMSCs, hybridized on Affymetrix U133a microarrays, and compared with previously profiled sarcomas (9
). Soft-tissue sarcomas (STS) define a group of histologically and genetically diverse tumors of mesenchymal origin, with greater similarity within a given subtype than with cultured hMSCs. For this reason, we used a well-characterized panel of stem cell–specific genes (10
) as a discriminant to evaluate the potential relationship between hMSCs, MFH, and other sarcoma subtypes. Unsupervised hierarchical clustering analysis using such a stem cell gene signature revealed that hMSCs are significantly associated exclusively with MFH (Figure A). Previous transcriptional pairwise analysis had also revealed the association between hMSCs and MFH (12
). Using an in vitro approach in which hMSCs are differentiated into prototypic connective cells (i.e., fat and bone) and gene expression analysis performed (via Affymetrix U133a arrays) at multiple temporal points during differentiation, we found that MFH exhibits the closest association among the differentiating cells to undifferentiated hMSCs. Similarly, committed (i.e., lineage-specific) sarcoma cell lines (i.e., LS141 liposarcoma [LS] and SAOS2 osteosarcoma) associate with hMSCs, differentiating into either adipocytes or osteocytes, respectively, independently corroborating the hMSC-sarcoma association (Supplemental Figure 1 and Supplemental Text; supplemental material available online with this article; doi:10.1172/JCI31377DS1).
To further establish the specific relationship between sarcoma subtypes and their lineage of origin, we performed immunohistochemical analysis of lineage-specific markers on STS in a tissue microarray. As illustrated in Figure B, staining these tumors for fat-specific markers (i.e., leptin and adiponectin) resulted in intense immunostaining of LS only. As expected, only leiomyosarcomas (LMSs) strongly expressed muscle-specific markers (i.e., muscle actin and α-smooth muscle actin). Similarly, when the same lesions were studied with MSC markers (e.g., CD54 and HGF) (13
), we observed that only MFH cases displayed a positive phenotype, reinforcing its putative MSC origin. As a negative control, low-grade fibrosarcomas exhibited negative phenotypes for all biomarkers analyzed (Figure B). Although there are other classical MSC markers used for FACS analysis, they are not suited for our approach of immunohistochemistry (IHC) on sarcoma tissue (13
). Additionally, we were specifically interested in hMSC markers that were not only present and/or specific in hMSCs but also decreased their expression as hMSCs differentiated. Accordingly, we screened all known mesenchymal markers against our differentiation gene expression data to determine which, if any, decrease with differentiation (data not shown). Of all the potential MSC markers, we found that CD54 and HGF were the best candidates, and they were therefore used in this study.
DKK1 is overexpressed and Wnt/β-catenin activity is absent in MFH cells.
Previous analysis of gene expression profiles by our laboratory (9
) and others (14
) has led us to observe that genes overexpressed in specific sarcoma subtypes are representative of the cell/lineage of origin of that particular neoplasm (Figure A and Supplemental Text). We hypothesized that if hMSCs are the cells of origin of MFH, then genes overexpressed in MFH as compared with other sarcoma subtypes should be relevant to hMSC biology and function. ANOVA on gene expression profiles of the panel of STS stratified by genes overexpressed in MFH (Figure A) identified many structural proteins and/or proteins whose functions are not well established. The most prominent exception was DKK1, a secreted inhibitor of the highly conserved Wnt developmental program (6
) known to be necessary for proliferation of hMSCs (7
). Immunohistochemical studies for DKK1 on a panel of STS (Figure C) and RT-PCR (Figure B) analysis of DKK1 transcripts on RNA isolated from fresh-frozen MFHs, LS, and leiomyosarcomas demonstrated that DKK1 is more significantly expressed in MFH than other sarcoma subtypes.
DKK1 and β-catenin patterns in hMSCs and MFH.
Having identified DKK1 as a protein expressed by both MFH and hMSCs, we next sought to examine the patterns of DKK1 and β-catenin (the effector arm of the Wnt-canonical pathway; ref. 8
) expression in hMSCs in relation to MFH cells. Prockop et al. (7
) established that expression of DKK1 closely parallels proliferation of hMSCs in vitro and that Dkk1 promotes reentry of hMSCs into the cell cycle. To examine the relationship between the levels of DKK1 in hMSCs and MFH, hMSCs were grown as previously described (7
). Cell density was measured at the indicated time points (Figure E), and total cellular protein was isolated from hMSCs at an initial cell density of approximately 1 × 105
. Immunoblot analysis of DKK1 and β-catenin as a function of proliferation indicated that DKK1 is expressed specifically during the proliferative phase of hMSCs while β-catenin is expressed in an inverse pattern — absent during proliferation and present during the stationary phase (Figure E). In comparison, the MFH0022 cell line (assayed at confluence; other MFH culture time points showed similar pattern to confluence; data not shown) displayed both Dkk1 and β-catenin (Figure E), paralleling the pattern observed in proliferating hMSCs (in particular the day 8 time point) and suggesting that MFH arises specifically from the proliferating hMSC population.
Since β-catenin levels are primarily regulated by cytoplasmic-to-nuclear shuttling (8
), we examined the levels of nuclear and cytoplasmic β-catenin in both proliferating hMSCs and MFH. As hMSCs transitioned from the proliferating to the stationary phase, β-catenin accumulated first in the cytoplasm (Figure G) and then moved into the nucleus (Figure F). In contrast, while confluent and exponentially growing MFH cells expressed cytoplasmic β-catenin, no nuclear β-catenin was detectable (Figure , F and G). Finally, active β-catenin levels (via ELISA for the active form of β-catenin) measured as a function of time paralleled the nuclear β-catenin immunoblot pattern, with accumulation of β-catenin observed as hMSCs ceased proliferating and no measurable expression in MFH (see Figure E). Additionally, induction of β-catenin levels via combined treatment of MFH cells with lithium (an activator of β-catenin via inhibition of the Wnt pathway intermediate inhibitor GSK-3β) and inhibition of Dkk1 protein or RNA expression using monoclonal anti-Dkk1 antibody or siRNA-Dkk1 resulted in increased total and nuclear β-catenin accumulation (Supplemental Figure 2 and Supplemental Text).
Finally, we examined the expression of 3 known β-catenin target genes in proliferating and stationary hMSCs as well as in MFH cells. As shown in Figure H (left panel), LEF1 levels increased with increasing levels of β-catenin while c-myc and cyclin D1 levels decreased with increasing β-catenin levels. We also used chromatin immunoprecipitation to demonstrate that β-catenin directly binds the promoters of myc and cyclin D1 and that the promoter exists in a primed repressive state (Supplemental Text and Supplemental Figures 3 and 4). Furthermore, the decrease in c-myc and cyclin D1 was consistent with the observed loss of hMSC proliferation (Figure E). On the other hand, as shown in Figure H (right panel), MFH cells (assayed at confluence) expressed higher levels of c-myc and cyclin D1 than hMSCs (assayed at confluence; day 12). The expression of c-myc and cyclin D1 in MFH cells at confluence may explain their continuous proliferation. Additionally, the low levels of LEF1 in MFH confirmed the lack of Wnt signaling as inferred from lack of active (Figure E) and nuclear β-catenin (Figure F).
Taken together, these results suggest that MFH cells fail to accumulate nuclear β-catenin due to limited endogenous production, as fitting an hMSC transformed during their proliferative state where endogenous hMSC β-catenin protein levels were naturally low (Figure F).
Wnt2-canonical signaling in hMSCs mediates commitment to differentiation.
Since nuclear β-catenin levels increased as hMSCs reached confluence and since no nuclear β-catenin was detectable in MFH cells, we sought to compare the patterns of the main determinants of β-catenin activity (the Wnt proteins) in hMSCs and MFH. It has been reported that Wnt2, Wnt4, Wnt5a, Wnt11, and Wnt16 are detectable in hMSCs while the majority of other Wnt products are not (16
). To limit our exploration to the Wnt genes relevant to MFH as transformed hMSCs, we examined the relative expression of Wnt proteins in MFH as compared with other sarcoma subtypes (Supplemental Table 2). Of the 5 Wnts referred to above, Wnt2 was highly overexpressed in MFH and Wnt5a was absent as compared with other sarcoma subtypes. These results were confirmed in a panel of STS by RT-PCR (Figure A). Additionally, during in vitro hMSC proliferation, Wnt5a protein accumulated as cells ceased to grow (in agreement with previously published RNA patterns; ref. 15
) while Wnt2 levels stayed constant (Figure B). Given these observations, 2 possibilities may explain the mechanism of β-catenin accumulation: (a) Wnt5a accumulation results in canonical Wnt pathway activation and β-catenin nuclear accumulation; or (b) Wnt2 activation results in constitutive β-catenin activation, which is repressed by Dkk1, and thus β-catenin only accumulates when Dkk1 levels decrease. To explore these possibilities, total cellular protein was isolated from hMSCs at the indicated time points of in vitro proliferation after treatment with either a polyclonal antibody or with siRNA against either Wnt2 or Wnt5a under the indicated conditions (see Supplemental Figure 2 and Supplemental Text regarding the specificity of siRNA-Wnt2 and siRNa-Wnt5a) and assayed via ELISA for active β-catenin (Figure C). Only the use of either blocking antibody PoAb-Wnt2 or siRNA-Wnt2 was specifically able to suppress active β-catenin accumulation, not PoAb-Wnt5a or siRNA-Wnt5a. These data suggest that Wnt2 mediates signaling via canonical (Wnt/β-catenin) pathway in hMSCs, controlled by Dkk1 levels.
Wnt2 mediates commitment while Wnt5a mediates a viability checkpoint in hMSCs.
To further explore the role of Wnt2 signaling in hMSC differentiation, hMSCs were pretreated with either PoAb-Wnt2 or PoAb-Wnt5a, then induced to differentiate via further treatment with adipocytic differentiation medium (ADM) or osteogenic differentiation medium (ODM), and finally assessed for early markers of mesenchymal differentiation (EMofMD) (Figure A) (see Supplemental Figure 5 and Supplemental Text for further details). Treatment of cells with PoAb-Wnt2 but not PoAb-Wnt5a led to the inability of hMSCs to express EMofMD, suggesting that Wnt2/β-catenin activity is necessary for commitment to the mesenchymal differentiation program. (In the Supplemental Text and in Supplemental Figure 5, we discuss and illustrate that MFH cells treated with lithium and PoAb-Dkk1 [which we show results in nuclear β-catenin accumulation; Supplemental Figure 2] display detectable expression of EMofMD.)
Wnt5a regulates a commitment viability checkpoint.
In contrast, hMSCs pretreated with PoAb-Wnt5a express EMofMD (Figure A), suggesting that Wnt5a does not affect commitment of hMSCs to differentiation. However, further growth of hMSCs following PoAb-Wnt5a treatment resulted in apoptotic-appearing cells as compared with control cells (Figure B). Propidium iodide staining and DNA content detection demonstrated an accumulation of a large sub-G1 peak consistent with apoptosis (Figure B). This was not observed in either hMSCs proliferating as controls or in hMSCs treated with PoAb-Wnt2. Based on this set of observations, we concluded that Wnt2 (via Wnt/β-catenin) signaling is necessary for commitment to differentiation while Wnt5a signaling does not promote commitment to differentiation (as judged by expression of EMofMD) but may mediate an antiapoptotic signal.
Wnt5a signals through JNK in hMSCs.
Since Wnt5a has been previously reported to mediate noncanonical Wnt signaling, we sought to examine the main noncanonical Wnt pathways active during hMSC proliferation and MFH sarcomagenesis. As discussed in a previous review (17
), Wnt/noncanonical is mediated by 3 main pathways: Wnt-PKC/calmodulin, Wnt-Rho GTPase (Rho/ROK/cdd42), and Wnt/JNK. Total cellular protein was isolated at multiple time points during hMSC in vitro proliferation and in confluent MFH cells after continuous culturing both in the absence and presence of an inhibitor to each pathway (hispidin, inhibitor of PKC, ref. 18
; Y-27632, inhibitor of ROK, ref. 19
; SP600125, inhibitor of JNK, ref. 20
). The activity of each pathway was assayed as described in Methods. Only Wnt/JNK changed during hMSC in vitro proliferation (Figure A), increasing approximately 4-fold as compared with baseline activity at plating. This activity was specifically blocked by JNK inhibitor SP600125.
To assess whether or not the rise in Wnt/JNK activity was Wnt5a mediated, Wnt/JNK signaling activity was measured in proliferating hMSCs either transfected with siRNA-Wnt5a or grown in medium supplemented with PoAb-Wnt5a. Under these conditions, no Wnt/JNK accumulation was observed, similar to the pattern of hMSCs grown in media supplemented with SP600125. As a control, hMSCs transfected with siRNA-Wnt2 or grown in medium supplemented with PoAb-Wnt2 showed no effect on Wnt/JNK activity (Figure B). These studies demonstrate that Wnt5a accumulation during hMSC growth activates Wnt/JNK activity, a well-characterized antiapoptotic pathway (21
). In contrast, MFH cells at confluence do not show significant Wnt/JNK activity (Figure , A and B), mirroring the pattern seen when comparing β-catenin activity in MFH and hMSCs (Figure , D and E).
Recapitulation of hMSC Wnt signaling patterns in MFH cells leads to controlled differentiation of MFH cells into mature connective tissue lineages. We have shown that Dkk1 is overexpressed in MFH and that it controls a canonical Wnt2 signaling pathway required for commitment to mesenchymal differentiation while Wnt5a/JNK noncanonical signaling regulates a commitment-viability checkpoint. Based on these findings, we sought to further investigate whether establishing both Wnt2/canonical signaling and Wnt5a/noncanonical signaling in MFH cells may be sufficient to recapitulate the overall pattern of Wnt signaling observed in confluent hMSCs and thus allow for the controlled differentiation of MFH cells into mature connective tissue lineages.
Accordingly, MFH0022 cells were cultured in the presence of both recombinant Wnt2 (rWnt2) and rWnt5a for 72 hours prior to changing medium, and further grown in either ADM or ODM without rWnt5a or rWnt2. Treatment of MFH cells with either ADM or ODM did not result in any significant accumulation of markers of either adipocytic (Figure A) or osteogenic (Figure B) differentiation. However, MFH cells pretreated with rWnt2 and rWnt5a readily accumulated markers of fat and mineralized calcium (Figure , C and D, respectively). Notably, the continual exposure of rWnt2 and rWnt5a after the addition of either ADM or ODM to MFH resulted in significantly less accumulation of differentiated cells. Similar treatment with rWnt2 alone resulted in cellular death while treatment with rWnt5a alone led to no evidence of differentiated cells (data not shown).
Differentiation and transformation via Wnt signaling.
Inhibition in hMSCs of Wnt-mediated commitment results in transformation and MFH morphology.
Having shown that activation of both canonical and noncanonical Wnt signaling is sufficient to commit MFH cells to differentiation, we next sought to test the hypothesis that inhibition of both Wnt signaling pathways in hMSCs may result in tumorigenesis. Since our previous observations suggest that Wnt5a (via Wnt/JNK signaling) is not involved in differentiation but rather mediates a commitment-viability checkpoint, we hypothesized that inhibition of canonical Wnt signaling would be sufficient to block commitment of hMSCs, which may serve as a tumorigenic event. Furthermore, if our initial gene expression analysis experiments were correct in identifying hMSCs as the progenitors of MFH, then the transformed phenotype of hMSCs should resemble MFH. To test this hypothesis, hMSCs and hMSCs previously immortalized with SV40 large T antigen (22
) were treated with human recombinant Dkk1 (hrDkk1) for 2 weeks at a concentration of 50 ng/ml (based on a previous determination of Dkk1 media levels in expanding hMSCs) (7
). As controls, hMSCs and SV40-hMSCs were grown in the absence of Dkk1 supplementation. All cultures were allowed to grow to confluence and then diluted and replated at a low density. Cells were grown for an additional 2 weeks in the presence and/or absence of hrDkk1 or until the detection of colony formation. hMSCs and SV40-hMSCs grown in the presence of hrDkk1 showed tumorigenic colony formation (Figure , G and H, respectively). This result was most evident in SV40-hMSCs and not seen in controls. No spontaneous transformation in either hMSCs or SV40-hMSCs grown in the absence of hrDKK1 was observed (data not shown).
To assess the in vivo tumorigenicity of these cells, an SV40-hMSC tumorigenic clone (SV40-hMSC-TC) was isolated and expanded, and 4 × 106
cells were injected subcutaneously into the flanks of nude mice. Parental SV40-hMSC cells were injected for controls. Within 3 weeks, SV40-hMSC-TC formed tumors with average volumes of 454 mm3
in comparison with 24 mm3
for SV40-hMSC parental cells (Figure I). Since the current gold standard for diagnosis of MFH is based on tumor morphology, a histological analysis of the formed tumors was performed (3
). Histological analysis of SV40-hMSC-TC tumors revealed their high grade and undifferentiated nature (Figure , J and K) and their mesenchymal origin, according to IHC studies (vimentin positive; S100, SMA, cytokeratin negative; data not shown). Finally, these tumors were reviewed in a blinded manner by a sarcoma pathologist and morphologically recapitulated MFH histopathology. The only apparent difference between the SV40-hMSC-TC xenograft tumors and human MFH was that the xenograft tumors did not display the full degree of pleomorphism commonly observed in MFH.